1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][a-zA-Z$._][a-zA-Z$._0-9]*``'. Identifiers which require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves.
83 #. Unnamed values are represented as an unsigned numeric value with
84 their prefix. For example, ``%12``, ``@2``, ``%44``.
85 #. Constants, which are described in the section Constants_ below.
87 LLVM requires that values start with a prefix for two reasons: Compilers
88 don't need to worry about name clashes with reserved words, and the set
89 of reserved words may be expanded in the future without penalty.
90 Additionally, unnamed identifiers allow a compiler to quickly come up
91 with a temporary variable without having to avoid symbol table
94 Reserved words in LLVM are very similar to reserved words in other
95 languages. There are keywords for different opcodes ('``add``',
96 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
97 '``i32``', etc...), and others. These reserved words cannot conflict
98 with variable names, because none of them start with a prefix character
101 Here is an example of LLVM code to multiply the integer variable
108 %result = mul i32 %X, 8
110 After strength reduction:
114 %result = shl i32 %X, 3
120 %0 = add i32 %X, %X ; yields {i32}:%0
121 %1 = add i32 %0, %0 ; yields {i32}:%1
122 %result = add i32 %1, %1
124 This last way of multiplying ``%X`` by 8 illustrates several important
125 lexical features of LLVM:
127 #. Comments are delimited with a '``;``' and go until the end of line.
128 #. Unnamed temporaries are created when the result of a computation is
129 not assigned to a named value.
130 #. Unnamed temporaries are numbered sequentially (using a per-function
131 incrementing counter, starting with 0).
133 It also shows a convention that we follow in this document. When
134 demonstrating instructions, we will follow an instruction with a comment
135 that defines the type and name of value produced.
143 LLVM programs are composed of ``Module``'s, each of which is a
144 translation unit of the input programs. Each module consists of
145 functions, global variables, and symbol table entries. Modules may be
146 combined together with the LLVM linker, which merges function (and
147 global variable) definitions, resolves forward declarations, and merges
148 symbol table entries. Here is an example of the "hello world" module:
152 ; Declare the string constant as a global constant.
153 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
155 ; External declaration of the puts function
156 declare i32 @puts(i8* nocapture) nounwind
158 ; Definition of main function
159 define i32 @main() { ; i32()*
160 ; Convert [13 x i8]* to i8 *...
161 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
163 ; Call puts function to write out the string to stdout.
164 call i32 @puts(i8* %cast210)
169 !1 = metadata !{i32 42}
172 This example is made up of a :ref:`global variable <globalvars>` named
173 "``.str``", an external declaration of the "``puts``" function, a
174 :ref:`function definition <functionstructure>` for "``main``" and
175 :ref:`named metadata <namedmetadatastructure>` "``foo``".
177 In general, a module is made up of a list of global values (where both
178 functions and global variables are global values). Global values are
179 represented by a pointer to a memory location (in this case, a pointer
180 to an array of char, and a pointer to a function), and have one of the
181 following :ref:`linkage types <linkage>`.
188 All Global Variables and Functions have one of the following types of
192 Global values with "``private``" linkage are only directly
193 accessible by objects in the current module. In particular, linking
194 code into a module with an private global value may cause the
195 private to be renamed as necessary to avoid collisions. Because the
196 symbol is private to the module, all references can be updated. This
197 doesn't show up in any symbol table in the object file.
199 Similar to ``private``, but the symbol is passed through the
200 assembler and evaluated by the linker. Unlike normal strong symbols,
201 they are removed by the linker from the final linked image
202 (executable or dynamic library).
203 ``linker_private_weak``
204 Similar to "``linker_private``", but the symbol is weak. Note that
205 ``linker_private_weak`` symbols are subject to coalescing by the
206 linker. The symbols are removed by the linker from the final linked
207 image (executable or dynamic library).
209 Similar to private, but the value shows as a local symbol
210 (``STB_LOCAL`` in the case of ELF) in the object file. This
211 corresponds to the notion of the '``static``' keyword in C.
212 ``available_externally``
213 Globals with "``available_externally``" linkage are never emitted
214 into the object file corresponding to the LLVM module. They exist to
215 allow inlining and other optimizations to take place given knowledge
216 of the definition of the global, which is known to be somewhere
217 outside the module. Globals with ``available_externally`` linkage
218 are allowed to be discarded at will, and are otherwise the same as
219 ``linkonce_odr``. This linkage type is only allowed on definitions,
222 Globals with "``linkonce``" linkage are merged with other globals of
223 the same name when linkage occurs. This can be used to implement
224 some forms of inline functions, templates, or other code which must
225 be generated in each translation unit that uses it, but where the
226 body may be overridden with a more definitive definition later.
227 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
228 that ``linkonce`` linkage does not actually allow the optimizer to
229 inline the body of this function into callers because it doesn't
230 know if this definition of the function is the definitive definition
231 within the program or whether it will be overridden by a stronger
232 definition. To enable inlining and other optimizations, use
233 "``linkonce_odr``" linkage.
235 "``weak``" linkage has the same merging semantics as ``linkonce``
236 linkage, except that unreferenced globals with ``weak`` linkage may
237 not be discarded. This is used for globals that are declared "weak"
240 "``common``" linkage is most similar to "``weak``" linkage, but they
241 are used for tentative definitions in C, such as "``int X;``" at
242 global scope. Symbols with "``common``" linkage are merged in the
243 same way as ``weak symbols``, and they may not be deleted if
244 unreferenced. ``common`` symbols may not have an explicit section,
245 must have a zero initializer, and may not be marked
246 ':ref:`constant <globalvars>`'. Functions and aliases may not have
249 .. _linkage_appending:
252 "``appending``" linkage may only be applied to global variables of
253 pointer to array type. When two global variables with appending
254 linkage are linked together, the two global arrays are appended
255 together. This is the LLVM, typesafe, equivalent of having the
256 system linker append together "sections" with identical names when
259 The semantics of this linkage follow the ELF object file model: the
260 symbol is weak until linked, if not linked, the symbol becomes null
261 instead of being an undefined reference.
262 ``linkonce_odr``, ``weak_odr``
263 Some languages allow differing globals to be merged, such as two
264 functions with different semantics. Other languages, such as
265 ``C++``, ensure that only equivalent globals are ever merged (the
266 "one definition rule" --- "ODR"). Such languages can use the
267 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
268 global will only be merged with equivalent globals. These linkage
269 types are otherwise the same as their non-``odr`` versions.
270 ``linkonce_odr_auto_hide``
271 Similar to "``linkonce_odr``", but nothing in the translation unit
272 takes the address of this definition. For instance, functions that
273 had an inline definition, but the compiler decided not to inline it.
274 ``linkonce_odr_auto_hide`` may have only ``default`` visibility. The
275 symbols are removed by the linker from the final linked image
276 (executable or dynamic library).
278 If none of the above identifiers are used, the global is externally
279 visible, meaning that it participates in linkage and can be used to
280 resolve external symbol references.
282 The next two types of linkage are targeted for Microsoft Windows
283 platform only. They are designed to support importing (exporting)
284 symbols from (to) DLLs (Dynamic Link Libraries).
287 "``dllimport``" linkage causes the compiler to reference a function
288 or variable via a global pointer to a pointer that is set up by the
289 DLL exporting the symbol. On Microsoft Windows targets, the pointer
290 name is formed by combining ``__imp_`` and the function or variable
293 "``dllexport``" linkage causes the compiler to provide a global
294 pointer to a pointer in a DLL, so that it can be referenced with the
295 ``dllimport`` attribute. On Microsoft Windows targets, the pointer
296 name is formed by combining ``__imp_`` and the function or variable
299 For example, since the "``.LC0``" variable is defined to be internal, if
300 another module defined a "``.LC0``" variable and was linked with this
301 one, one of the two would be renamed, preventing a collision. Since
302 "``main``" and "``puts``" are external (i.e., lacking any linkage
303 declarations), they are accessible outside of the current module.
305 It is illegal for a function *declaration* to have any linkage type
306 other than ``external``, ``dllimport`` or ``extern_weak``.
308 Aliases can have only ``external``, ``internal``, ``weak`` or
309 ``weak_odr`` linkages.
316 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
317 :ref:`invokes <i_invoke>` can all have an optional calling convention
318 specified for the call. The calling convention of any pair of dynamic
319 caller/callee must match, or the behavior of the program is undefined.
320 The following calling conventions are supported by LLVM, and more may be
323 "``ccc``" - The C calling convention
324 This calling convention (the default if no other calling convention
325 is specified) matches the target C calling conventions. This calling
326 convention supports varargs function calls and tolerates some
327 mismatch in the declared prototype and implemented declaration of
328 the function (as does normal C).
329 "``fastcc``" - The fast calling convention
330 This calling convention attempts to make calls as fast as possible
331 (e.g. by passing things in registers). This calling convention
332 allows the target to use whatever tricks it wants to produce fast
333 code for the target, without having to conform to an externally
334 specified ABI (Application Binary Interface). `Tail calls can only
335 be optimized when this, the GHC or the HiPE convention is
336 used. <CodeGenerator.html#id80>`_ This calling convention does not
337 support varargs and requires the prototype of all callees to exactly
338 match the prototype of the function definition.
339 "``coldcc``" - The cold calling convention
340 This calling convention attempts to make code in the caller as
341 efficient as possible under the assumption that the call is not
342 commonly executed. As such, these calls often preserve all registers
343 so that the call does not break any live ranges in the caller side.
344 This calling convention does not support varargs and requires the
345 prototype of all callees to exactly match the prototype of the
347 "``cc 10``" - GHC convention
348 This calling convention has been implemented specifically for use by
349 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
350 It passes everything in registers, going to extremes to achieve this
351 by disabling callee save registers. This calling convention should
352 not be used lightly but only for specific situations such as an
353 alternative to the *register pinning* performance technique often
354 used when implementing functional programming languages. At the
355 moment only X86 supports this convention and it has the following
358 - On *X86-32* only supports up to 4 bit type parameters. No
359 floating point types are supported.
360 - On *X86-64* only supports up to 10 bit type parameters and 6
361 floating point parameters.
363 This calling convention supports `tail call
364 optimization <CodeGenerator.html#id80>`_ but requires both the
365 caller and callee are using it.
366 "``cc 11``" - The HiPE calling convention
367 This calling convention has been implemented specifically for use by
368 the `High-Performance Erlang
369 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
370 native code compiler of the `Ericsson's Open Source Erlang/OTP
371 system <http://www.erlang.org/download.shtml>`_. It uses more
372 registers for argument passing than the ordinary C calling
373 convention and defines no callee-saved registers. The calling
374 convention properly supports `tail call
375 optimization <CodeGenerator.html#id80>`_ but requires that both the
376 caller and the callee use it. It uses a *register pinning*
377 mechanism, similar to GHC's convention, for keeping frequently
378 accessed runtime components pinned to specific hardware registers.
379 At the moment only X86 supports this convention (both 32 and 64
381 "``cc <n>``" - Numbered convention
382 Any calling convention may be specified by number, allowing
383 target-specific calling conventions to be used. Target specific
384 calling conventions start at 64.
386 More calling conventions can be added/defined on an as-needed basis, to
387 support Pascal conventions or any other well-known target-independent
393 All Global Variables and Functions have one of the following visibility
396 "``default``" - Default style
397 On targets that use the ELF object file format, default visibility
398 means that the declaration is visible to other modules and, in
399 shared libraries, means that the declared entity may be overridden.
400 On Darwin, default visibility means that the declaration is visible
401 to other modules. Default visibility corresponds to "external
402 linkage" in the language.
403 "``hidden``" - Hidden style
404 Two declarations of an object with hidden visibility refer to the
405 same object if they are in the same shared object. Usually, hidden
406 visibility indicates that the symbol will not be placed into the
407 dynamic symbol table, so no other module (executable or shared
408 library) can reference it directly.
409 "``protected``" - Protected style
410 On ELF, protected visibility indicates that the symbol will be
411 placed in the dynamic symbol table, but that references within the
412 defining module will bind to the local symbol. That is, the symbol
413 cannot be overridden by another module.
418 LLVM IR allows you to specify name aliases for certain types. This can
419 make it easier to read the IR and make the IR more condensed
420 (particularly when recursive types are involved). An example of a name
425 %mytype = type { %mytype*, i32 }
427 You may give a name to any :ref:`type <typesystem>` except
428 ":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
429 expected with the syntax "%mytype".
431 Note that type names are aliases for the structural type that they
432 indicate, and that you can therefore specify multiple names for the same
433 type. This often leads to confusing behavior when dumping out a .ll
434 file. Since LLVM IR uses structural typing, the name is not part of the
435 type. When printing out LLVM IR, the printer will pick *one name* to
436 render all types of a particular shape. This means that if you have code
437 where two different source types end up having the same LLVM type, that
438 the dumper will sometimes print the "wrong" or unexpected type. This is
439 an important design point and isn't going to change.
446 Global variables define regions of memory allocated at compilation time
447 instead of run-time. Global variables may optionally be initialized, may
448 have an explicit section to be placed in, and may have an optional
449 explicit alignment specified.
451 A variable may be defined as ``thread_local``, which means that it will
452 not be shared by threads (each thread will have a separated copy of the
453 variable). Not all targets support thread-local variables. Optionally, a
454 TLS model may be specified:
457 For variables that are only used within the current shared library.
459 For variables in modules that will not be loaded dynamically.
461 For variables defined in the executable and only used within it.
463 The models correspond to the ELF TLS models; see `ELF Handling For
464 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
465 more information on under which circumstances the different models may
466 be used. The target may choose a different TLS model if the specified
467 model is not supported, or if a better choice of model can be made.
469 A variable may be defined as a global ``constant``, which indicates that
470 the contents of the variable will **never** be modified (enabling better
471 optimization, allowing the global data to be placed in the read-only
472 section of an executable, etc). Note that variables that need runtime
473 initialization cannot be marked ``constant`` as there is a store to the
476 LLVM explicitly allows *declarations* of global variables to be marked
477 constant, even if the final definition of the global is not. This
478 capability can be used to enable slightly better optimization of the
479 program, but requires the language definition to guarantee that
480 optimizations based on the 'constantness' are valid for the translation
481 units that do not include the definition.
483 As SSA values, global variables define pointer values that are in scope
484 (i.e. they dominate) all basic blocks in the program. Global variables
485 always define a pointer to their "content" type because they describe a
486 region of memory, and all memory objects in LLVM are accessed through
489 Global variables can be marked with ``unnamed_addr`` which indicates
490 that the address is not significant, only the content. Constants marked
491 like this can be merged with other constants if they have the same
492 initializer. Note that a constant with significant address *can* be
493 merged with a ``unnamed_addr`` constant, the result being a constant
494 whose address is significant.
496 A global variable may be declared to reside in a target-specific
497 numbered address space. For targets that support them, address spaces
498 may affect how optimizations are performed and/or what target
499 instructions are used to access the variable. The default address space
500 is zero. The address space qualifier must precede any other attributes.
502 LLVM allows an explicit section to be specified for globals. If the
503 target supports it, it will emit globals to the section specified.
505 By default, global initializers are optimized by assuming that global
506 variables defined within the module are not modified from their
507 initial values before the start of the global initializer. This is
508 true even for variables potentially accessible from outside the
509 module, including those with external linkage or appearing in
510 ``@llvm.used``. This assumption may be suppressed by marking the
511 variable with ``externally_initialized``.
513 An explicit alignment may be specified for a global, which must be a
514 power of 2. If not present, or if the alignment is set to zero, the
515 alignment of the global is set by the target to whatever it feels
516 convenient. If an explicit alignment is specified, the global is forced
517 to have exactly that alignment. Targets and optimizers are not allowed
518 to over-align the global if the global has an assigned section. In this
519 case, the extra alignment could be observable: for example, code could
520 assume that the globals are densely packed in their section and try to
521 iterate over them as an array, alignment padding would break this
524 For example, the following defines a global in a numbered address space
525 with an initializer, section, and alignment:
529 @G = addrspace(5) constant float 1.0, section "foo", align 4
531 The following example defines a thread-local global with the
532 ``initialexec`` TLS model:
536 @G = thread_local(initialexec) global i32 0, align 4
538 .. _functionstructure:
543 LLVM function definitions consist of the "``define``" keyword, an
544 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
545 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
546 an optional ``unnamed_addr`` attribute, a return type, an optional
547 :ref:`parameter attribute <paramattrs>` for the return type, a function
548 name, a (possibly empty) argument list (each with optional :ref:`parameter
549 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
550 an optional section, an optional alignment, an optional :ref:`garbage
551 collector name <gc>`, an opening curly brace, a list of basic blocks,
552 and a closing curly brace.
554 LLVM function declarations consist of the "``declare``" keyword, an
555 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
556 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
557 an optional ``unnamed_addr`` attribute, a return type, an optional
558 :ref:`parameter attribute <paramattrs>` for the return type, a function
559 name, a possibly empty list of arguments, an optional alignment, and an
560 optional :ref:`garbage collector name <gc>`.
562 A function definition contains a list of basic blocks, forming the CFG
563 (Control Flow Graph) for the function. Each basic block may optionally
564 start with a label (giving the basic block a symbol table entry),
565 contains a list of instructions, and ends with a
566 :ref:`terminator <terminators>` instruction (such as a branch or function
567 return). If explicit label is not provided, a block is assigned an
568 implicit numbered label, using a next value from the same counter as used
569 for unnamed temporaries (:ref:`see above<identifiers>`). For example, if a
570 function entry block does not have explicit label, it will be assigned
571 label "%0", then first unnamed temporary in that block will be "%1", etc.
573 The first basic block in a function is special in two ways: it is
574 immediately executed on entrance to the function, and it is not allowed
575 to have predecessor basic blocks (i.e. there can not be any branches to
576 the entry block of a function). Because the block can have no
577 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
579 LLVM allows an explicit section to be specified for functions. If the
580 target supports it, it will emit functions to the section specified.
582 An explicit alignment may be specified for a function. If not present,
583 or if the alignment is set to zero, the alignment of the function is set
584 by the target to whatever it feels convenient. If an explicit alignment
585 is specified, the function is forced to have at least that much
586 alignment. All alignments must be a power of 2.
588 If the ``unnamed_addr`` attribute is given, the address is know to not
589 be significant and two identical functions can be merged.
593 define [linkage] [visibility]
595 <ResultType> @<FunctionName> ([argument list])
596 [fn Attrs] [section "name"] [align N]
602 Aliases act as "second name" for the aliasee value (which can be either
603 function, global variable, another alias or bitcast of global value).
604 Aliases may have an optional :ref:`linkage type <linkage>`, and an optional
605 :ref:`visibility style <visibility>`.
609 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
611 .. _namedmetadatastructure:
616 Named metadata is a collection of metadata. :ref:`Metadata
617 nodes <metadata>` (but not metadata strings) are the only valid
618 operands for a named metadata.
622 ; Some unnamed metadata nodes, which are referenced by the named metadata.
623 !0 = metadata !{metadata !"zero"}
624 !1 = metadata !{metadata !"one"}
625 !2 = metadata !{metadata !"two"}
627 !name = !{!0, !1, !2}
634 The return type and each parameter of a function type may have a set of
635 *parameter attributes* associated with them. Parameter attributes are
636 used to communicate additional information about the result or
637 parameters of a function. Parameter attributes are considered to be part
638 of the function, not of the function type, so functions with different
639 parameter attributes can have the same function type.
641 Parameter attributes are simple keywords that follow the type specified.
642 If multiple parameter attributes are needed, they are space separated.
647 declare i32 @printf(i8* noalias nocapture, ...)
648 declare i32 @atoi(i8 zeroext)
649 declare signext i8 @returns_signed_char()
651 Note that any attributes for the function result (``nounwind``,
652 ``readonly``) come immediately after the argument list.
654 Currently, only the following parameter attributes are defined:
657 This indicates to the code generator that the parameter or return
658 value should be zero-extended to the extent required by the target's
659 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
660 the caller (for a parameter) or the callee (for a return value).
662 This indicates to the code generator that the parameter or return
663 value should be sign-extended to the extent required by the target's
664 ABI (which is usually 32-bits) by the caller (for a parameter) or
665 the callee (for a return value).
667 This indicates that this parameter or return value should be treated
668 in a special target-dependent fashion during while emitting code for
669 a function call or return (usually, by putting it in a register as
670 opposed to memory, though some targets use it to distinguish between
671 two different kinds of registers). Use of this attribute is
674 This indicates that the pointer parameter should really be passed by
675 value to the function. The attribute implies that a hidden copy of
676 the pointee is made between the caller and the callee, so the callee
677 is unable to modify the value in the caller. This attribute is only
678 valid on LLVM pointer arguments. It is generally used to pass
679 structs and arrays by value, but is also valid on pointers to
680 scalars. The copy is considered to belong to the caller not the
681 callee (for example, ``readonly`` functions should not write to
682 ``byval`` parameters). This is not a valid attribute for return
685 The byval attribute also supports specifying an alignment with the
686 align attribute. It indicates the alignment of the stack slot to
687 form and the known alignment of the pointer specified to the call
688 site. If the alignment is not specified, then the code generator
689 makes a target-specific assumption.
692 This indicates that the pointer parameter specifies the address of a
693 structure that is the return value of the function in the source
694 program. This pointer must be guaranteed by the caller to be valid:
695 loads and stores to the structure may be assumed by the callee
696 not to trap and to be properly aligned. This may only be applied to
697 the first parameter. This is not a valid attribute for return
700 This indicates that pointer values :ref:`based <pointeraliasing>` on
701 the argument or return value do not alias pointer values which are
702 not *based* on it, ignoring certain "irrelevant" dependencies. For a
703 call to the parent function, dependencies between memory references
704 from before or after the call and from those during the call are
705 "irrelevant" to the ``noalias`` keyword for the arguments and return
706 value used in that call. The caller shares the responsibility with
707 the callee for ensuring that these requirements are met. For further
708 details, please see the discussion of the NoAlias response in `alias
709 analysis <AliasAnalysis.html#MustMayNo>`_.
711 Note that this definition of ``noalias`` is intentionally similar
712 to the definition of ``restrict`` in C99 for function arguments,
713 though it is slightly weaker.
715 For function return values, C99's ``restrict`` is not meaningful,
716 while LLVM's ``noalias`` is.
718 This indicates that the callee does not make any copies of the
719 pointer that outlive the callee itself. This is not a valid
720 attribute for return values.
725 This indicates that the pointer parameter can be excised using the
726 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
727 attribute for return values and can only be applied to one parameter.
730 This indicates that the value of the function always returns the value
731 of the parameter as its return value. This is an optimization hint to
732 the code generator when generating the caller, allowing tail call
733 optimization and omission of register saves and restores in some cases;
734 it is not checked or enforced when generating the callee. The parameter
735 and the function return type must be valid operands for the
736 :ref:`bitcast instruction <i_bitcast>`. This is not a valid attribute for
737 return values and can only be applied to one parameter.
741 Garbage Collector Names
742 -----------------------
744 Each function may specify a garbage collector name, which is simply a
749 define void @f() gc "name" { ... }
751 The compiler declares the supported values of *name*. Specifying a
752 collector which will cause the compiler to alter its output in order to
753 support the named garbage collection algorithm.
760 Attribute groups are groups of attributes that are referenced by objects within
761 the IR. They are important for keeping ``.ll`` files readable, because a lot of
762 functions will use the same set of attributes. In the degenerative case of a
763 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
764 group will capture the important command line flags used to build that file.
766 An attribute group is a module-level object. To use an attribute group, an
767 object references the attribute group's ID (e.g. ``#37``). An object may refer
768 to more than one attribute group. In that situation, the attributes from the
769 different groups are merged.
771 Here is an example of attribute groups for a function that should always be
772 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
776 ; Target-independent attributes:
777 attributes #0 = { alwaysinline alignstack=4 }
779 ; Target-dependent attributes:
780 attributes #1 = { "no-sse" }
782 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
783 define void @f() #0 #1 { ... }
790 Function attributes are set to communicate additional information about
791 a function. Function attributes are considered to be part of the
792 function, not of the function type, so functions with different function
793 attributes can have the same function type.
795 Function attributes are simple keywords that follow the type specified.
796 If multiple attributes are needed, they are space separated. For
801 define void @f() noinline { ... }
802 define void @f() alwaysinline { ... }
803 define void @f() alwaysinline optsize { ... }
804 define void @f() optsize { ... }
807 This attribute indicates that, when emitting the prologue and
808 epilogue, the backend should forcibly align the stack pointer.
809 Specify the desired alignment, which must be a power of two, in
812 This attribute indicates that the inliner should attempt to inline
813 this function into callers whenever possible, ignoring any active
814 inlining size threshold for this caller.
816 This attribute indicates that this function is rarely called. When
817 computing edge weights, basic blocks post-dominated by a cold
818 function call are also considered to be cold; and, thus, given low
821 This attribute suppresses lazy symbol binding for the function. This
822 may make calls to the function faster, at the cost of extra program
823 startup time if the function is not called during program startup.
825 This attribute indicates that the source code contained a hint that
826 inlining this function is desirable (such as the "inline" keyword in
827 C/C++). It is just a hint; it imposes no requirements on the
830 This attribute disables prologue / epilogue emission for the
831 function. This can have very system-specific consequences.
833 This indicates that the callee function at a call site is not
834 recognized as a built-in function. LLVM will retain the original call
835 and not replace it with equivalent code based on the semantics of the
836 built-in function. This is only valid at call sites, not on function
837 declarations or definitions.
839 This attribute indicates that calls to the function cannot be
840 duplicated. A call to a ``noduplicate`` function may be moved
841 within its parent function, but may not be duplicated within
844 A function containing a ``noduplicate`` call may still
845 be an inlining candidate, provided that the call is not
846 duplicated by inlining. That implies that the function has
847 internal linkage and only has one call site, so the original
848 call is dead after inlining.
850 This attributes disables implicit floating point instructions.
852 This attribute indicates that the inliner should never inline this
853 function in any situation. This attribute may not be used together
854 with the ``alwaysinline`` attribute.
856 This attribute indicates that the code generator should not use a
857 red zone, even if the target-specific ABI normally permits it.
859 This function attribute indicates that the function never returns
860 normally. This produces undefined behavior at runtime if the
861 function ever does dynamically return.
863 This function attribute indicates that the function never returns
864 with an unwind or exceptional control flow. If the function does
865 unwind, its runtime behavior is undefined.
867 This attribute suggests that optimization passes and code generator
868 passes make choices that keep the code size of this function low,
869 and otherwise do optimizations specifically to reduce code size.
871 This attribute indicates that the function computes its result (or
872 decides to unwind an exception) based strictly on its arguments,
873 without dereferencing any pointer arguments or otherwise accessing
874 any mutable state (e.g. memory, control registers, etc) visible to
875 caller functions. It does not write through any pointer arguments
876 (including ``byval`` arguments) and never changes any state visible
877 to callers. This means that it cannot unwind exceptions by calling
878 the ``C++`` exception throwing methods.
880 This attribute indicates that the function does not write through
881 any pointer arguments (including ``byval`` arguments) or otherwise
882 modify any state (e.g. memory, control registers, etc) visible to
883 caller functions. It may dereference pointer arguments and read
884 state that may be set in the caller. A readonly function always
885 returns the same value (or unwinds an exception identically) when
886 called with the same set of arguments and global state. It cannot
887 unwind an exception by calling the ``C++`` exception throwing
890 This attribute indicates that this function can return twice. The C
891 ``setjmp`` is an example of such a function. The compiler disables
892 some optimizations (like tail calls) in the caller of these
895 This attribute indicates that AddressSanitizer checks
896 (dynamic address safety analysis) are enabled for this function.
898 This attribute indicates that MemorySanitizer checks (dynamic detection
899 of accesses to uninitialized memory) are enabled for this function.
901 This attribute indicates that ThreadSanitizer checks
902 (dynamic thread safety analysis) are enabled for this function.
904 This attribute indicates that the function should emit a stack
905 smashing protector. It is in the form of a "canary" --- a random value
906 placed on the stack before the local variables that's checked upon
907 return from the function to see if it has been overwritten. A
908 heuristic is used to determine if a function needs stack protectors
909 or not. The heuristic used will enable protectors for functions with:
911 - Character arrays larger than ``ssp-buffer-size`` (default 8).
912 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
913 - Calls to alloca() with variable sizes or constant sizes greater than
916 If a function that has an ``ssp`` attribute is inlined into a
917 function that doesn't have an ``ssp`` attribute, then the resulting
918 function will have an ``ssp`` attribute.
920 This attribute indicates that the function should *always* emit a
921 stack smashing protector. This overrides the ``ssp`` function
924 If a function that has an ``sspreq`` attribute is inlined into a
925 function that doesn't have an ``sspreq`` attribute or which has an
926 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
927 an ``sspreq`` attribute.
929 This attribute indicates that the function should emit a stack smashing
930 protector. This attribute causes a strong heuristic to be used when
931 determining if a function needs stack protectors. The strong heuristic
932 will enable protectors for functions with:
934 - Arrays of any size and type
935 - Aggregates containing an array of any size and type.
937 - Local variables that have had their address taken.
939 This overrides the ``ssp`` function attribute.
941 If a function that has an ``sspstrong`` attribute is inlined into a
942 function that doesn't have an ``sspstrong`` attribute, then the
943 resulting function will have an ``sspstrong`` attribute.
945 This attribute indicates that the ABI being targeted requires that
946 an unwind table entry be produce for this function even if we can
947 show that no exceptions passes by it. This is normally the case for
948 the ELF x86-64 abi, but it can be disabled for some compilation
953 Module-Level Inline Assembly
954 ----------------------------
956 Modules may contain "module-level inline asm" blocks, which corresponds
957 to the GCC "file scope inline asm" blocks. These blocks are internally
958 concatenated by LLVM and treated as a single unit, but may be separated
959 in the ``.ll`` file if desired. The syntax is very simple:
963 module asm "inline asm code goes here"
964 module asm "more can go here"
966 The strings can contain any character by escaping non-printable
967 characters. The escape sequence used is simply "\\xx" where "xx" is the
968 two digit hex code for the number.
970 The inline asm code is simply printed to the machine code .s file when
971 assembly code is generated.
976 A module may specify a target specific data layout string that specifies
977 how data is to be laid out in memory. The syntax for the data layout is
982 target datalayout = "layout specification"
984 The *layout specification* consists of a list of specifications
985 separated by the minus sign character ('-'). Each specification starts
986 with a letter and may include other information after the letter to
987 define some aspect of the data layout. The specifications accepted are
991 Specifies that the target lays out data in big-endian form. That is,
992 the bits with the most significance have the lowest address
995 Specifies that the target lays out data in little-endian form. That
996 is, the bits with the least significance have the lowest address
999 Specifies the natural alignment of the stack in bits. Alignment
1000 promotion of stack variables is limited to the natural stack
1001 alignment to avoid dynamic stack realignment. The stack alignment
1002 must be a multiple of 8-bits. If omitted, the natural stack
1003 alignment defaults to "unspecified", which does not prevent any
1004 alignment promotions.
1005 ``p[n]:<size>:<abi>:<pref>``
1006 This specifies the *size* of a pointer and its ``<abi>`` and
1007 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1008 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
1009 preceding ``:`` should be omitted too. The address space, ``n`` is
1010 optional, and if not specified, denotes the default address space 0.
1011 The value of ``n`` must be in the range [1,2^23).
1012 ``i<size>:<abi>:<pref>``
1013 This specifies the alignment for an integer type of a given bit
1014 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1015 ``v<size>:<abi>:<pref>``
1016 This specifies the alignment for a vector type of a given bit
1018 ``f<size>:<abi>:<pref>``
1019 This specifies the alignment for a floating point type of a given bit
1020 ``<size>``. Only values of ``<size>`` that are supported by the target
1021 will work. 32 (float) and 64 (double) are supported on all targets; 80
1022 or 128 (different flavors of long double) are also supported on some
1024 ``a<size>:<abi>:<pref>``
1025 This specifies the alignment for an aggregate type of a given bit
1027 ``s<size>:<abi>:<pref>``
1028 This specifies the alignment for a stack object of a given bit
1030 ``n<size1>:<size2>:<size3>...``
1031 This specifies a set of native integer widths for the target CPU in
1032 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1033 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1034 this set are considered to support most general arithmetic operations
1037 When constructing the data layout for a given target, LLVM starts with a
1038 default set of specifications which are then (possibly) overridden by
1039 the specifications in the ``datalayout`` keyword. The default
1040 specifications are given in this list:
1042 - ``E`` - big endian
1043 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment
1044 - ``S0`` - natural stack alignment is unspecified
1045 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1046 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1047 - ``i16:16:16`` - i16 is 16-bit aligned
1048 - ``i32:32:32`` - i32 is 32-bit aligned
1049 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1050 alignment of 64-bits
1051 - ``f16:16:16`` - half is 16-bit aligned
1052 - ``f32:32:32`` - float is 32-bit aligned
1053 - ``f64:64:64`` - double is 64-bit aligned
1054 - ``f128:128:128`` - quad is 128-bit aligned
1055 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1056 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1057 - ``a0:0:64`` - aggregates are 64-bit aligned
1059 When LLVM is determining the alignment for a given type, it uses the
1062 #. If the type sought is an exact match for one of the specifications,
1063 that specification is used.
1064 #. If no match is found, and the type sought is an integer type, then
1065 the smallest integer type that is larger than the bitwidth of the
1066 sought type is used. If none of the specifications are larger than
1067 the bitwidth then the largest integer type is used. For example,
1068 given the default specifications above, the i7 type will use the
1069 alignment of i8 (next largest) while both i65 and i256 will use the
1070 alignment of i64 (largest specified).
1071 #. If no match is found, and the type sought is a vector type, then the
1072 largest vector type that is smaller than the sought vector type will
1073 be used as a fall back. This happens because <128 x double> can be
1074 implemented in terms of 64 <2 x double>, for example.
1076 The function of the data layout string may not be what you expect.
1077 Notably, this is not a specification from the frontend of what alignment
1078 the code generator should use.
1080 Instead, if specified, the target data layout is required to match what
1081 the ultimate *code generator* expects. This string is used by the
1082 mid-level optimizers to improve code, and this only works if it matches
1083 what the ultimate code generator uses. If you would like to generate IR
1084 that does not embed this target-specific detail into the IR, then you
1085 don't have to specify the string. This will disable some optimizations
1086 that require precise layout information, but this also prevents those
1087 optimizations from introducing target specificity into the IR.
1089 .. _pointeraliasing:
1091 Pointer Aliasing Rules
1092 ----------------------
1094 Any memory access must be done through a pointer value associated with
1095 an address range of the memory access, otherwise the behavior is
1096 undefined. Pointer values are associated with address ranges according
1097 to the following rules:
1099 - A pointer value is associated with the addresses associated with any
1100 value it is *based* on.
1101 - An address of a global variable is associated with the address range
1102 of the variable's storage.
1103 - The result value of an allocation instruction is associated with the
1104 address range of the allocated storage.
1105 - A null pointer in the default address-space is associated with no
1107 - An integer constant other than zero or a pointer value returned from
1108 a function not defined within LLVM may be associated with address
1109 ranges allocated through mechanisms other than those provided by
1110 LLVM. Such ranges shall not overlap with any ranges of addresses
1111 allocated by mechanisms provided by LLVM.
1113 A pointer value is *based* on another pointer value according to the
1116 - A pointer value formed from a ``getelementptr`` operation is *based*
1117 on the first operand of the ``getelementptr``.
1118 - The result value of a ``bitcast`` is *based* on the operand of the
1120 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1121 values that contribute (directly or indirectly) to the computation of
1122 the pointer's value.
1123 - The "*based* on" relationship is transitive.
1125 Note that this definition of *"based"* is intentionally similar to the
1126 definition of *"based"* in C99, though it is slightly weaker.
1128 LLVM IR does not associate types with memory. The result type of a
1129 ``load`` merely indicates the size and alignment of the memory from
1130 which to load, as well as the interpretation of the value. The first
1131 operand type of a ``store`` similarly only indicates the size and
1132 alignment of the store.
1134 Consequently, type-based alias analysis, aka TBAA, aka
1135 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1136 :ref:`Metadata <metadata>` may be used to encode additional information
1137 which specialized optimization passes may use to implement type-based
1142 Volatile Memory Accesses
1143 ------------------------
1145 Certain memory accesses, such as :ref:`load <i_load>`'s,
1146 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1147 marked ``volatile``. The optimizers must not change the number of
1148 volatile operations or change their order of execution relative to other
1149 volatile operations. The optimizers *may* change the order of volatile
1150 operations relative to non-volatile operations. This is not Java's
1151 "volatile" and has no cross-thread synchronization behavior.
1153 IR-level volatile loads and stores cannot safely be optimized into
1154 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1155 flagged volatile. Likewise, the backend should never split or merge
1156 target-legal volatile load/store instructions.
1158 .. admonition:: Rationale
1160 Platforms may rely on volatile loads and stores of natively supported
1161 data width to be executed as single instruction. For example, in C
1162 this holds for an l-value of volatile primitive type with native
1163 hardware support, but not necessarily for aggregate types. The
1164 frontend upholds these expectations, which are intentionally
1165 unspecified in the IR. The rules above ensure that IR transformation
1166 do not violate the frontend's contract with the language.
1170 Memory Model for Concurrent Operations
1171 --------------------------------------
1173 The LLVM IR does not define any way to start parallel threads of
1174 execution or to register signal handlers. Nonetheless, there are
1175 platform-specific ways to create them, and we define LLVM IR's behavior
1176 in their presence. This model is inspired by the C++0x memory model.
1178 For a more informal introduction to this model, see the :doc:`Atomics`.
1180 We define a *happens-before* partial order as the least partial order
1183 - Is a superset of single-thread program order, and
1184 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1185 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1186 techniques, like pthread locks, thread creation, thread joining,
1187 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1188 Constraints <ordering>`).
1190 Note that program order does not introduce *happens-before* edges
1191 between a thread and signals executing inside that thread.
1193 Every (defined) read operation (load instructions, memcpy, atomic
1194 loads/read-modify-writes, etc.) R reads a series of bytes written by
1195 (defined) write operations (store instructions, atomic
1196 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1197 section, initialized globals are considered to have a write of the
1198 initializer which is atomic and happens before any other read or write
1199 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1200 may see any write to the same byte, except:
1202 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1203 write\ :sub:`2` happens before R\ :sub:`byte`, then
1204 R\ :sub:`byte` does not see write\ :sub:`1`.
1205 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1206 R\ :sub:`byte` does not see write\ :sub:`3`.
1208 Given that definition, R\ :sub:`byte` is defined as follows:
1210 - If R is volatile, the result is target-dependent. (Volatile is
1211 supposed to give guarantees which can support ``sig_atomic_t`` in
1212 C/C++, and may be used for accesses to addresses which do not behave
1213 like normal memory. It does not generally provide cross-thread
1215 - Otherwise, if there is no write to the same byte that happens before
1216 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1217 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1218 R\ :sub:`byte` returns the value written by that write.
1219 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1220 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1221 Memory Ordering Constraints <ordering>` section for additional
1222 constraints on how the choice is made.
1223 - Otherwise R\ :sub:`byte` returns ``undef``.
1225 R returns the value composed of the series of bytes it read. This
1226 implies that some bytes within the value may be ``undef`` **without**
1227 the entire value being ``undef``. Note that this only defines the
1228 semantics of the operation; it doesn't mean that targets will emit more
1229 than one instruction to read the series of bytes.
1231 Note that in cases where none of the atomic intrinsics are used, this
1232 model places only one restriction on IR transformations on top of what
1233 is required for single-threaded execution: introducing a store to a byte
1234 which might not otherwise be stored is not allowed in general.
1235 (Specifically, in the case where another thread might write to and read
1236 from an address, introducing a store can change a load that may see
1237 exactly one write into a load that may see multiple writes.)
1241 Atomic Memory Ordering Constraints
1242 ----------------------------------
1244 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1245 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1246 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1247 an ordering parameter that determines which other atomic instructions on
1248 the same address they *synchronize with*. These semantics are borrowed
1249 from Java and C++0x, but are somewhat more colloquial. If these
1250 descriptions aren't precise enough, check those specs (see spec
1251 references in the :doc:`atomics guide <Atomics>`).
1252 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1253 differently since they don't take an address. See that instruction's
1254 documentation for details.
1256 For a simpler introduction to the ordering constraints, see the
1260 The set of values that can be read is governed by the happens-before
1261 partial order. A value cannot be read unless some operation wrote
1262 it. This is intended to provide a guarantee strong enough to model
1263 Java's non-volatile shared variables. This ordering cannot be
1264 specified for read-modify-write operations; it is not strong enough
1265 to make them atomic in any interesting way.
1267 In addition to the guarantees of ``unordered``, there is a single
1268 total order for modifications by ``monotonic`` operations on each
1269 address. All modification orders must be compatible with the
1270 happens-before order. There is no guarantee that the modification
1271 orders can be combined to a global total order for the whole program
1272 (and this often will not be possible). The read in an atomic
1273 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1274 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1275 order immediately before the value it writes. If one atomic read
1276 happens before another atomic read of the same address, the later
1277 read must see the same value or a later value in the address's
1278 modification order. This disallows reordering of ``monotonic`` (or
1279 stronger) operations on the same address. If an address is written
1280 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1281 read that address repeatedly, the other threads must eventually see
1282 the write. This corresponds to the C++0x/C1x
1283 ``memory_order_relaxed``.
1285 In addition to the guarantees of ``monotonic``, a
1286 *synchronizes-with* edge may be formed with a ``release`` operation.
1287 This is intended to model C++'s ``memory_order_acquire``.
1289 In addition to the guarantees of ``monotonic``, if this operation
1290 writes a value which is subsequently read by an ``acquire``
1291 operation, it *synchronizes-with* that operation. (This isn't a
1292 complete description; see the C++0x definition of a release
1293 sequence.) This corresponds to the C++0x/C1x
1294 ``memory_order_release``.
1295 ``acq_rel`` (acquire+release)
1296 Acts as both an ``acquire`` and ``release`` operation on its
1297 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1298 ``seq_cst`` (sequentially consistent)
1299 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1300 operation which only reads, ``release`` for an operation which only
1301 writes), there is a global total order on all
1302 sequentially-consistent operations on all addresses, which is
1303 consistent with the *happens-before* partial order and with the
1304 modification orders of all the affected addresses. Each
1305 sequentially-consistent read sees the last preceding write to the
1306 same address in this global order. This corresponds to the C++0x/C1x
1307 ``memory_order_seq_cst`` and Java volatile.
1311 If an atomic operation is marked ``singlethread``, it only *synchronizes
1312 with* or participates in modification and seq\_cst total orderings with
1313 other operations running in the same thread (for example, in signal
1321 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1322 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1323 :ref:`frem <i_frem>`) have the following flags that can set to enable
1324 otherwise unsafe floating point operations
1327 No NaNs - Allow optimizations to assume the arguments and result are not
1328 NaN. Such optimizations are required to retain defined behavior over
1329 NaNs, but the value of the result is undefined.
1332 No Infs - Allow optimizations to assume the arguments and result are not
1333 +/-Inf. Such optimizations are required to retain defined behavior over
1334 +/-Inf, but the value of the result is undefined.
1337 No Signed Zeros - Allow optimizations to treat the sign of a zero
1338 argument or result as insignificant.
1341 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1342 argument rather than perform division.
1345 Fast - Allow algebraically equivalent transformations that may
1346 dramatically change results in floating point (e.g. reassociate). This
1347 flag implies all the others.
1354 The LLVM type system is one of the most important features of the
1355 intermediate representation. Being typed enables a number of
1356 optimizations to be performed on the intermediate representation
1357 directly, without having to do extra analyses on the side before the
1358 transformation. A strong type system makes it easier to read the
1359 generated code and enables novel analyses and transformations that are
1360 not feasible to perform on normal three address code representations.
1362 Type Classifications
1363 --------------------
1365 The types fall into a few useful classifications:
1374 * - :ref:`integer <t_integer>`
1375 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1378 * - :ref:`floating point <t_floating>`
1379 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1387 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1388 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1389 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1390 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1392 * - :ref:`primitive <t_primitive>`
1393 - :ref:`label <t_label>`,
1394 :ref:`void <t_void>`,
1395 :ref:`integer <t_integer>`,
1396 :ref:`floating point <t_floating>`,
1397 :ref:`x86mmx <t_x86mmx>`,
1398 :ref:`metadata <t_metadata>`.
1400 * - :ref:`derived <t_derived>`
1401 - :ref:`array <t_array>`,
1402 :ref:`function <t_function>`,
1403 :ref:`pointer <t_pointer>`,
1404 :ref:`structure <t_struct>`,
1405 :ref:`vector <t_vector>`,
1406 :ref:`opaque <t_opaque>`.
1408 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1409 Values of these types are the only ones which can be produced by
1417 The primitive types are the fundamental building blocks of the LLVM
1428 The integer type is a very simple type that simply specifies an
1429 arbitrary bit width for the integer type desired. Any bit width from 1
1430 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1439 The number of bits the integer will occupy is specified by the ``N``
1445 +----------------+------------------------------------------------+
1446 | ``i1`` | a single-bit integer. |
1447 +----------------+------------------------------------------------+
1448 | ``i32`` | a 32-bit integer. |
1449 +----------------+------------------------------------------------+
1450 | ``i1942652`` | a really big integer of over 1 million bits. |
1451 +----------------+------------------------------------------------+
1455 Floating Point Types
1456 ^^^^^^^^^^^^^^^^^^^^
1465 - 16-bit floating point value
1468 - 32-bit floating point value
1471 - 64-bit floating point value
1474 - 128-bit floating point value (112-bit mantissa)
1477 - 80-bit floating point value (X87)
1480 - 128-bit floating point value (two 64-bits)
1490 The x86mmx type represents a value held in an MMX register on an x86
1491 machine. The operations allowed on it are quite limited: parameters and
1492 return values, load and store, and bitcast. User-specified MMX
1493 instructions are represented as intrinsic or asm calls with arguments
1494 and/or results of this type. There are no arrays, vectors or constants
1512 The void type does not represent any value and has no size.
1529 The label type represents code labels.
1546 The metadata type represents embedded metadata. No derived types may be
1547 created from metadata except for :ref:`function <t_function>` arguments.
1561 The real power in LLVM comes from the derived types in the system. This
1562 is what allows a programmer to represent arrays, functions, pointers,
1563 and other useful types. Each of these types contain one or more element
1564 types which may be a primitive type, or another derived type. For
1565 example, it is possible to have a two dimensional array, using an array
1566 as the element type of another array.
1573 Aggregate Types are a subset of derived types that can contain multiple
1574 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1575 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1586 The array type is a very simple derived type that arranges elements
1587 sequentially in memory. The array type requires a size (number of
1588 elements) and an underlying data type.
1595 [<# elements> x <elementtype>]
1597 The number of elements is a constant integer value; ``elementtype`` may
1598 be any type with a size.
1603 +------------------+--------------------------------------+
1604 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1605 +------------------+--------------------------------------+
1606 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1607 +------------------+--------------------------------------+
1608 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1609 +------------------+--------------------------------------+
1611 Here are some examples of multidimensional arrays:
1613 +-----------------------------+----------------------------------------------------------+
1614 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1615 +-----------------------------+----------------------------------------------------------+
1616 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1617 +-----------------------------+----------------------------------------------------------+
1618 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1619 +-----------------------------+----------------------------------------------------------+
1621 There is no restriction on indexing beyond the end of the array implied
1622 by a static type (though there are restrictions on indexing beyond the
1623 bounds of an allocated object in some cases). This means that
1624 single-dimension 'variable sized array' addressing can be implemented in
1625 LLVM with a zero length array type. An implementation of 'pascal style
1626 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1637 The function type can be thought of as a function signature. It consists
1638 of a return type and a list of formal parameter types. The return type
1639 of a function type is a first class type or a void type.
1646 <returntype> (<parameter list>)
1648 ...where '``<parameter list>``' is a comma-separated list of type
1649 specifiers. Optionally, the parameter list may include a type ``...``,
1650 which indicates that the function takes a variable number of arguments.
1651 Variable argument functions can access their arguments with the
1652 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1653 '``<returntype>``' is any type except :ref:`label <t_label>`.
1658 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1659 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1660 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1661 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1662 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1663 | ``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. |
1664 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1665 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1666 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1676 The structure type is used to represent a collection of data members
1677 together in memory. The elements of a structure may be any type that has
1680 Structures in memory are accessed using '``load``' and '``store``' by
1681 getting a pointer to a field with the '``getelementptr``' instruction.
1682 Structures in registers are accessed using the '``extractvalue``' and
1683 '``insertvalue``' instructions.
1685 Structures may optionally be "packed" structures, which indicate that
1686 the alignment of the struct is one byte, and that there is no padding
1687 between the elements. In non-packed structs, padding between field types
1688 is inserted as defined by the DataLayout string in the module, which is
1689 required to match what the underlying code generator expects.
1691 Structures can either be "literal" or "identified". A literal structure
1692 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1693 identified types are always defined at the top level with a name.
1694 Literal types are uniqued by their contents and can never be recursive
1695 or opaque since there is no way to write one. Identified types can be
1696 recursive, can be opaqued, and are never uniqued.
1703 %T1 = type { <type list> } ; Identified normal struct type
1704 %T2 = type <{ <type list> }> ; Identified packed struct type
1709 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1710 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1711 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1712 | ``{ 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``. |
1713 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1714 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1715 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1719 Opaque Structure Types
1720 ^^^^^^^^^^^^^^^^^^^^^^
1725 Opaque structure types are used to represent named structure types that
1726 do not have a body specified. This corresponds (for example) to the C
1727 notion of a forward declared structure.
1740 +--------------+-------------------+
1741 | ``opaque`` | An opaque type. |
1742 +--------------+-------------------+
1752 The pointer type is used to specify memory locations. Pointers are
1753 commonly used to reference objects in memory.
1755 Pointer types may have an optional address space attribute defining the
1756 numbered address space where the pointed-to object resides. The default
1757 address space is number zero. The semantics of non-zero address spaces
1758 are target-specific.
1760 Note that LLVM does not permit pointers to void (``void*``) nor does it
1761 permit pointers to labels (``label*``). Use ``i8*`` instead.
1773 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1774 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1775 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1776 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1777 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1778 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1779 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1789 A vector type is a simple derived type that represents a vector of
1790 elements. Vector types are used when multiple primitive data are
1791 operated in parallel using a single instruction (SIMD). A vector type
1792 requires a size (number of elements) and an underlying primitive data
1793 type. Vector types are considered :ref:`first class <t_firstclass>`.
1800 < <# elements> x <elementtype> >
1802 The number of elements is a constant integer value larger than 0;
1803 elementtype may be any integer or floating point type, or a pointer to
1804 these types. Vectors of size zero are not allowed.
1809 +-------------------+--------------------------------------------------+
1810 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1811 +-------------------+--------------------------------------------------+
1812 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1813 +-------------------+--------------------------------------------------+
1814 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1815 +-------------------+--------------------------------------------------+
1816 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1817 +-------------------+--------------------------------------------------+
1822 LLVM has several different basic types of constants. This section
1823 describes them all and their syntax.
1828 **Boolean constants**
1829 The two strings '``true``' and '``false``' are both valid constants
1831 **Integer constants**
1832 Standard integers (such as '4') are constants of the
1833 :ref:`integer <t_integer>` type. Negative numbers may be used with
1835 **Floating point constants**
1836 Floating point constants use standard decimal notation (e.g.
1837 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1838 hexadecimal notation (see below). The assembler requires the exact
1839 decimal value of a floating-point constant. For example, the
1840 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1841 decimal in binary. Floating point constants must have a :ref:`floating
1842 point <t_floating>` type.
1843 **Null pointer constants**
1844 The identifier '``null``' is recognized as a null pointer constant
1845 and must be of :ref:`pointer type <t_pointer>`.
1847 The one non-intuitive notation for constants is the hexadecimal form of
1848 floating point constants. For example, the form
1849 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1850 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1851 constants are required (and the only time that they are generated by the
1852 disassembler) is when a floating point constant must be emitted but it
1853 cannot be represented as a decimal floating point number in a reasonable
1854 number of digits. For example, NaN's, infinities, and other special
1855 values are represented in their IEEE hexadecimal format so that assembly
1856 and disassembly do not cause any bits to change in the constants.
1858 When using the hexadecimal form, constants of types half, float, and
1859 double are represented using the 16-digit form shown above (which
1860 matches the IEEE754 representation for double); half and float values
1861 must, however, be exactly representable as IEEE 754 half and single
1862 precision, respectively. Hexadecimal format is always used for long
1863 double, and there are three forms of long double. The 80-bit format used
1864 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1865 128-bit format used by PowerPC (two adjacent doubles) is represented by
1866 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1867 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
1868 will only work if they match the long double format on your target.
1869 The IEEE 16-bit format (half precision) is represented by ``0xH``
1870 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
1871 (sign bit at the left).
1873 There are no constants of type x86mmx.
1878 Complex constants are a (potentially recursive) combination of simple
1879 constants and smaller complex constants.
1881 **Structure constants**
1882 Structure constants are represented with notation similar to
1883 structure type definitions (a comma separated list of elements,
1884 surrounded by braces (``{}``)). For example:
1885 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1886 "``@G = external global i32``". Structure constants must have
1887 :ref:`structure type <t_struct>`, and the number and types of elements
1888 must match those specified by the type.
1890 Array constants are represented with notation similar to array type
1891 definitions (a comma separated list of elements, surrounded by
1892 square brackets (``[]``)). For example:
1893 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1894 :ref:`array type <t_array>`, and the number and types of elements must
1895 match those specified by the type.
1896 **Vector constants**
1897 Vector constants are represented with notation similar to vector
1898 type definitions (a comma separated list of elements, surrounded by
1899 less-than/greater-than's (``<>``)). For example:
1900 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1901 must have :ref:`vector type <t_vector>`, and the number and types of
1902 elements must match those specified by the type.
1903 **Zero initialization**
1904 The string '``zeroinitializer``' can be used to zero initialize a
1905 value to zero of *any* type, including scalar and
1906 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1907 having to print large zero initializers (e.g. for large arrays) and
1908 is always exactly equivalent to using explicit zero initializers.
1910 A metadata node is a structure-like constant with :ref:`metadata
1911 type <t_metadata>`. For example:
1912 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
1913 constants that are meant to be interpreted as part of the
1914 instruction stream, metadata is a place to attach additional
1915 information such as debug info.
1917 Global Variable and Function Addresses
1918 --------------------------------------
1920 The addresses of :ref:`global variables <globalvars>` and
1921 :ref:`functions <functionstructure>` are always implicitly valid
1922 (link-time) constants. These constants are explicitly referenced when
1923 the :ref:`identifier for the global <identifiers>` is used and always have
1924 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
1927 .. code-block:: llvm
1931 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1938 The string '``undef``' can be used anywhere a constant is expected, and
1939 indicates that the user of the value may receive an unspecified
1940 bit-pattern. Undefined values may be of any type (other than '``label``'
1941 or '``void``') and be used anywhere a constant is permitted.
1943 Undefined values are useful because they indicate to the compiler that
1944 the program is well defined no matter what value is used. This gives the
1945 compiler more freedom to optimize. Here are some examples of
1946 (potentially surprising) transformations that are valid (in pseudo IR):
1948 .. code-block:: llvm
1958 This is safe because all of the output bits are affected by the undef
1959 bits. Any output bit can have a zero or one depending on the input bits.
1961 .. code-block:: llvm
1972 These logical operations have bits that are not always affected by the
1973 input. For example, if ``%X`` has a zero bit, then the output of the
1974 '``and``' operation will always be a zero for that bit, no matter what
1975 the corresponding bit from the '``undef``' is. As such, it is unsafe to
1976 optimize or assume that the result of the '``and``' is '``undef``'.
1977 However, it is safe to assume that all bits of the '``undef``' could be
1978 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
1979 all the bits of the '``undef``' operand to the '``or``' could be set,
1980 allowing the '``or``' to be folded to -1.
1982 .. code-block:: llvm
1984 %A = select undef, %X, %Y
1985 %B = select undef, 42, %Y
1986 %C = select %X, %Y, undef
1996 This set of examples shows that undefined '``select``' (and conditional
1997 branch) conditions can go *either way*, but they have to come from one
1998 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
1999 both known to have a clear low bit, then ``%A`` would have to have a
2000 cleared low bit. However, in the ``%C`` example, the optimizer is
2001 allowed to assume that the '``undef``' operand could be the same as
2002 ``%Y``, allowing the whole '``select``' to be eliminated.
2004 .. code-block:: llvm
2006 %A = xor undef, undef
2023 This example points out that two '``undef``' operands are not
2024 necessarily the same. This can be surprising to people (and also matches
2025 C semantics) where they assume that "``X^X``" is always zero, even if
2026 ``X`` is undefined. This isn't true for a number of reasons, but the
2027 short answer is that an '``undef``' "variable" can arbitrarily change
2028 its value over its "live range". This is true because the variable
2029 doesn't actually *have a live range*. Instead, the value is logically
2030 read from arbitrary registers that happen to be around when needed, so
2031 the value is not necessarily consistent over time. In fact, ``%A`` and
2032 ``%C`` need to have the same semantics or the core LLVM "replace all
2033 uses with" concept would not hold.
2035 .. code-block:: llvm
2043 These examples show the crucial difference between an *undefined value*
2044 and *undefined behavior*. An undefined value (like '``undef``') is
2045 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2046 operation can be constant folded to '``undef``', because the '``undef``'
2047 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2048 However, in the second example, we can make a more aggressive
2049 assumption: because the ``undef`` is allowed to be an arbitrary value,
2050 we are allowed to assume that it could be zero. Since a divide by zero
2051 has *undefined behavior*, we are allowed to assume that the operation
2052 does not execute at all. This allows us to delete the divide and all
2053 code after it. Because the undefined operation "can't happen", the
2054 optimizer can assume that it occurs in dead code.
2056 .. code-block:: llvm
2058 a: store undef -> %X
2059 b: store %X -> undef
2064 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2065 value can be assumed to not have any effect; we can assume that the
2066 value is overwritten with bits that happen to match what was already
2067 there. However, a store *to* an undefined location could clobber
2068 arbitrary memory, therefore, it has undefined behavior.
2075 Poison values are similar to :ref:`undef values <undefvalues>`, however
2076 they also represent the fact that an instruction or constant expression
2077 which cannot evoke side effects has nevertheless detected a condition
2078 which results in undefined behavior.
2080 There is currently no way of representing a poison value in the IR; they
2081 only exist when produced by operations such as :ref:`add <i_add>` with
2084 Poison value behavior is defined in terms of value *dependence*:
2086 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2087 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2088 their dynamic predecessor basic block.
2089 - Function arguments depend on the corresponding actual argument values
2090 in the dynamic callers of their functions.
2091 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2092 instructions that dynamically transfer control back to them.
2093 - :ref:`Invoke <i_invoke>` instructions depend on the
2094 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2095 call instructions that dynamically transfer control back to them.
2096 - Non-volatile loads and stores depend on the most recent stores to all
2097 of the referenced memory addresses, following the order in the IR
2098 (including loads and stores implied by intrinsics such as
2099 :ref:`@llvm.memcpy <int_memcpy>`.)
2100 - An instruction with externally visible side effects depends on the
2101 most recent preceding instruction with externally visible side
2102 effects, following the order in the IR. (This includes :ref:`volatile
2103 operations <volatile>`.)
2104 - An instruction *control-depends* on a :ref:`terminator
2105 instruction <terminators>` if the terminator instruction has
2106 multiple successors and the instruction is always executed when
2107 control transfers to one of the successors, and may not be executed
2108 when control is transferred to another.
2109 - Additionally, an instruction also *control-depends* on a terminator
2110 instruction if the set of instructions it otherwise depends on would
2111 be different if the terminator had transferred control to a different
2113 - Dependence is transitive.
2115 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2116 with the additional affect that any instruction which has a *dependence*
2117 on a poison value has undefined behavior.
2119 Here are some examples:
2121 .. code-block:: llvm
2124 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2125 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2126 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2127 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2129 store i32 %poison, i32* @g ; Poison value stored to memory.
2130 %poison2 = load i32* @g ; Poison value loaded back from memory.
2132 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2134 %narrowaddr = bitcast i32* @g to i16*
2135 %wideaddr = bitcast i32* @g to i64*
2136 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2137 %poison4 = load i64* %wideaddr ; Returns a poison value.
2139 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2140 br i1 %cmp, label %true, label %end ; Branch to either destination.
2143 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2144 ; it has undefined behavior.
2148 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2149 ; Both edges into this PHI are
2150 ; control-dependent on %cmp, so this
2151 ; always results in a poison value.
2153 store volatile i32 0, i32* @g ; This would depend on the store in %true
2154 ; if %cmp is true, or the store in %entry
2155 ; otherwise, so this is undefined behavior.
2157 br i1 %cmp, label %second_true, label %second_end
2158 ; The same branch again, but this time the
2159 ; true block doesn't have side effects.
2166 store volatile i32 0, i32* @g ; This time, the instruction always depends
2167 ; on the store in %end. Also, it is
2168 ; control-equivalent to %end, so this is
2169 ; well-defined (ignoring earlier undefined
2170 ; behavior in this example).
2174 Addresses of Basic Blocks
2175 -------------------------
2177 ``blockaddress(@function, %block)``
2179 The '``blockaddress``' constant computes the address of the specified
2180 basic block in the specified function, and always has an ``i8*`` type.
2181 Taking the address of the entry block is illegal.
2183 This value only has defined behavior when used as an operand to the
2184 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2185 against null. Pointer equality tests between labels addresses results in
2186 undefined behavior --- though, again, comparison against null is ok, and
2187 no label is equal to the null pointer. This may be passed around as an
2188 opaque pointer sized value as long as the bits are not inspected. This
2189 allows ``ptrtoint`` and arithmetic to be performed on these values so
2190 long as the original value is reconstituted before the ``indirectbr``
2193 Finally, some targets may provide defined semantics when using the value
2194 as the operand to an inline assembly, but that is target specific.
2196 Constant Expressions
2197 --------------------
2199 Constant expressions are used to allow expressions involving other
2200 constants to be used as constants. Constant expressions may be of any
2201 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2202 that does not have side effects (e.g. load and call are not supported).
2203 The following is the syntax for constant expressions:
2205 ``trunc (CST to TYPE)``
2206 Truncate a constant to another type. The bit size of CST must be
2207 larger than the bit size of TYPE. Both types must be integers.
2208 ``zext (CST to TYPE)``
2209 Zero extend a constant to another type. The bit size of CST must be
2210 smaller than the bit size of TYPE. Both types must be integers.
2211 ``sext (CST to TYPE)``
2212 Sign extend a constant to another type. The bit size of CST must be
2213 smaller than the bit size of TYPE. Both types must be integers.
2214 ``fptrunc (CST to TYPE)``
2215 Truncate a floating point constant to another floating point type.
2216 The size of CST must be larger than the size of TYPE. Both types
2217 must be floating point.
2218 ``fpext (CST to TYPE)``
2219 Floating point extend a constant to another type. The size of CST
2220 must be smaller or equal to the size of TYPE. Both types must be
2222 ``fptoui (CST to TYPE)``
2223 Convert a floating point constant to the corresponding unsigned
2224 integer constant. TYPE must be a scalar or vector integer type. CST
2225 must be of scalar or vector floating point type. Both CST and TYPE
2226 must be scalars, or vectors of the same number of elements. If the
2227 value won't fit in the integer type, the results are undefined.
2228 ``fptosi (CST to TYPE)``
2229 Convert a floating point constant to the corresponding signed
2230 integer constant. TYPE must be a scalar or vector integer type. CST
2231 must be of scalar or vector floating point type. Both CST and TYPE
2232 must be scalars, or vectors of the same number of elements. If the
2233 value won't fit in the integer type, the results are undefined.
2234 ``uitofp (CST to TYPE)``
2235 Convert an unsigned integer constant to the corresponding floating
2236 point constant. TYPE must be a scalar or vector floating point type.
2237 CST must be of scalar or vector integer type. Both CST and TYPE must
2238 be scalars, or vectors of the same number of elements. If the value
2239 won't fit in the floating point type, the results are undefined.
2240 ``sitofp (CST to TYPE)``
2241 Convert a signed integer constant to the corresponding floating
2242 point constant. TYPE must be a scalar or vector floating point type.
2243 CST must be of scalar or vector integer type. Both CST and TYPE must
2244 be scalars, or vectors of the same number of elements. If the value
2245 won't fit in the floating point type, the results are undefined.
2246 ``ptrtoint (CST to TYPE)``
2247 Convert a pointer typed constant to the corresponding integer
2248 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2249 pointer type. The ``CST`` value is zero extended, truncated, or
2250 unchanged to make it fit in ``TYPE``.
2251 ``inttoptr (CST to TYPE)``
2252 Convert an integer constant to a pointer constant. TYPE must be a
2253 pointer type. CST must be of integer type. The CST value is zero
2254 extended, truncated, or unchanged to make it fit in a pointer size.
2255 This one is *really* dangerous!
2256 ``bitcast (CST to TYPE)``
2257 Convert a constant, CST, to another TYPE. The constraints of the
2258 operands are the same as those for the :ref:`bitcast
2259 instruction <i_bitcast>`.
2260 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2261 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2262 constants. As with the :ref:`getelementptr <i_getelementptr>`
2263 instruction, the index list may have zero or more indexes, which are
2264 required to make sense for the type of "CSTPTR".
2265 ``select (COND, VAL1, VAL2)``
2266 Perform the :ref:`select operation <i_select>` on constants.
2267 ``icmp COND (VAL1, VAL2)``
2268 Performs the :ref:`icmp operation <i_icmp>` on constants.
2269 ``fcmp COND (VAL1, VAL2)``
2270 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2271 ``extractelement (VAL, IDX)``
2272 Perform the :ref:`extractelement operation <i_extractelement>` on
2274 ``insertelement (VAL, ELT, IDX)``
2275 Perform the :ref:`insertelement operation <i_insertelement>` on
2277 ``shufflevector (VEC1, VEC2, IDXMASK)``
2278 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2280 ``extractvalue (VAL, IDX0, IDX1, ...)``
2281 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2282 constants. The index list is interpreted in a similar manner as
2283 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2284 least one index value must be specified.
2285 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2286 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2287 The index list is interpreted in a similar manner as indices in a
2288 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2289 value must be specified.
2290 ``OPCODE (LHS, RHS)``
2291 Perform the specified operation of the LHS and RHS constants. OPCODE
2292 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2293 binary <bitwiseops>` operations. The constraints on operands are
2294 the same as those for the corresponding instruction (e.g. no bitwise
2295 operations on floating point values are allowed).
2300 Inline Assembler Expressions
2301 ----------------------------
2303 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2304 Inline Assembly <moduleasm>`) through the use of a special value. This
2305 value represents the inline assembler as a string (containing the
2306 instructions to emit), a list of operand constraints (stored as a
2307 string), a flag that indicates whether or not the inline asm expression
2308 has side effects, and a flag indicating whether the function containing
2309 the asm needs to align its stack conservatively. An example inline
2310 assembler expression is:
2312 .. code-block:: llvm
2314 i32 (i32) asm "bswap $0", "=r,r"
2316 Inline assembler expressions may **only** be used as the callee operand
2317 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2318 Thus, typically we have:
2320 .. code-block:: llvm
2322 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2324 Inline asms with side effects not visible in the constraint list must be
2325 marked as having side effects. This is done through the use of the
2326 '``sideeffect``' keyword, like so:
2328 .. code-block:: llvm
2330 call void asm sideeffect "eieio", ""()
2332 In some cases inline asms will contain code that will not work unless
2333 the stack is aligned in some way, such as calls or SSE instructions on
2334 x86, yet will not contain code that does that alignment within the asm.
2335 The compiler should make conservative assumptions about what the asm
2336 might contain and should generate its usual stack alignment code in the
2337 prologue if the '``alignstack``' keyword is present:
2339 .. code-block:: llvm
2341 call void asm alignstack "eieio", ""()
2343 Inline asms also support using non-standard assembly dialects. The
2344 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2345 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2346 the only supported dialects. An example is:
2348 .. code-block:: llvm
2350 call void asm inteldialect "eieio", ""()
2352 If multiple keywords appear the '``sideeffect``' keyword must come
2353 first, the '``alignstack``' keyword second and the '``inteldialect``'
2359 The call instructions that wrap inline asm nodes may have a
2360 "``!srcloc``" MDNode attached to it that contains a list of constant
2361 integers. If present, the code generator will use the integer as the
2362 location cookie value when report errors through the ``LLVMContext``
2363 error reporting mechanisms. This allows a front-end to correlate backend
2364 errors that occur with inline asm back to the source code that produced
2367 .. code-block:: llvm
2369 call void asm sideeffect "something bad", ""(), !srcloc !42
2371 !42 = !{ i32 1234567 }
2373 It is up to the front-end to make sense of the magic numbers it places
2374 in the IR. If the MDNode contains multiple constants, the code generator
2375 will use the one that corresponds to the line of the asm that the error
2380 Metadata Nodes and Metadata Strings
2381 -----------------------------------
2383 LLVM IR allows metadata to be attached to instructions in the program
2384 that can convey extra information about the code to the optimizers and
2385 code generator. One example application of metadata is source-level
2386 debug information. There are two metadata primitives: strings and nodes.
2387 All metadata has the ``metadata`` type and is identified in syntax by a
2388 preceding exclamation point ('``!``').
2390 A metadata string is a string surrounded by double quotes. It can
2391 contain any character by escaping non-printable characters with
2392 "``\xx``" where "``xx``" is the two digit hex code. For example:
2395 Metadata nodes are represented with notation similar to structure
2396 constants (a comma separated list of elements, surrounded by braces and
2397 preceded by an exclamation point). Metadata nodes can have any values as
2398 their operand. For example:
2400 .. code-block:: llvm
2402 !{ metadata !"test\00", i32 10}
2404 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2405 metadata nodes, which can be looked up in the module symbol table. For
2408 .. code-block:: llvm
2410 !foo = metadata !{!4, !3}
2412 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2413 function is using two metadata arguments:
2415 .. code-block:: llvm
2417 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2419 Metadata can be attached with an instruction. Here metadata ``!21`` is
2420 attached to the ``add`` instruction using the ``!dbg`` identifier:
2422 .. code-block:: llvm
2424 %indvar.next = add i64 %indvar, 1, !dbg !21
2426 More information about specific metadata nodes recognized by the
2427 optimizers and code generator is found below.
2432 In LLVM IR, memory does not have types, so LLVM's own type system is not
2433 suitable for doing TBAA. Instead, metadata is added to the IR to
2434 describe a type system of a higher level language. This can be used to
2435 implement typical C/C++ TBAA, but it can also be used to implement
2436 custom alias analysis behavior for other languages.
2438 The current metadata format is very simple. TBAA metadata nodes have up
2439 to three fields, e.g.:
2441 .. code-block:: llvm
2443 !0 = metadata !{ metadata !"an example type tree" }
2444 !1 = metadata !{ metadata !"int", metadata !0 }
2445 !2 = metadata !{ metadata !"float", metadata !0 }
2446 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2448 The first field is an identity field. It can be any value, usually a
2449 metadata string, which uniquely identifies the type. The most important
2450 name in the tree is the name of the root node. Two trees with different
2451 root node names are entirely disjoint, even if they have leaves with
2454 The second field identifies the type's parent node in the tree, or is
2455 null or omitted for a root node. A type is considered to alias all of
2456 its descendants and all of its ancestors in the tree. Also, a type is
2457 considered to alias all types in other trees, so that bitcode produced
2458 from multiple front-ends is handled conservatively.
2460 If the third field is present, it's an integer which if equal to 1
2461 indicates that the type is "constant" (meaning
2462 ``pointsToConstantMemory`` should return true; see `other useful
2463 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2465 '``tbaa.struct``' Metadata
2466 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2468 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2469 aggregate assignment operations in C and similar languages, however it
2470 is defined to copy a contiguous region of memory, which is more than
2471 strictly necessary for aggregate types which contain holes due to
2472 padding. Also, it doesn't contain any TBAA information about the fields
2475 ``!tbaa.struct`` metadata can describe which memory subregions in a
2476 memcpy are padding and what the TBAA tags of the struct are.
2478 The current metadata format is very simple. ``!tbaa.struct`` metadata
2479 nodes are a list of operands which are in conceptual groups of three.
2480 For each group of three, the first operand gives the byte offset of a
2481 field in bytes, the second gives its size in bytes, and the third gives
2484 .. code-block:: llvm
2486 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2488 This describes a struct with two fields. The first is at offset 0 bytes
2489 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2490 and has size 4 bytes and has tbaa tag !2.
2492 Note that the fields need not be contiguous. In this example, there is a
2493 4 byte gap between the two fields. This gap represents padding which
2494 does not carry useful data and need not be preserved.
2496 '``fpmath``' Metadata
2497 ^^^^^^^^^^^^^^^^^^^^^
2499 ``fpmath`` metadata may be attached to any instruction of floating point
2500 type. It can be used to express the maximum acceptable error in the
2501 result of that instruction, in ULPs, thus potentially allowing the
2502 compiler to use a more efficient but less accurate method of computing
2503 it. ULP is defined as follows:
2505 If ``x`` is a real number that lies between two finite consecutive
2506 floating-point numbers ``a`` and ``b``, without being equal to one
2507 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2508 distance between the two non-equal finite floating-point numbers
2509 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2511 The metadata node shall consist of a single positive floating point
2512 number representing the maximum relative error, for example:
2514 .. code-block:: llvm
2516 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2518 '``range``' Metadata
2519 ^^^^^^^^^^^^^^^^^^^^
2521 ``range`` metadata may be attached only to loads of integer types. It
2522 expresses the possible ranges the loaded value is in. The ranges are
2523 represented with a flattened list of integers. The loaded value is known
2524 to be in the union of the ranges defined by each consecutive pair. Each
2525 pair has the following properties:
2527 - The type must match the type loaded by the instruction.
2528 - The pair ``a,b`` represents the range ``[a,b)``.
2529 - Both ``a`` and ``b`` are constants.
2530 - The range is allowed to wrap.
2531 - The range should not represent the full or empty set. That is,
2534 In addition, the pairs must be in signed order of the lower bound and
2535 they must be non-contiguous.
2539 .. code-block:: llvm
2541 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2542 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2543 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2544 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2546 !0 = metadata !{ i8 0, i8 2 }
2547 !1 = metadata !{ i8 255, i8 2 }
2548 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2549 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2554 It is sometimes useful to attach information to loop constructs. Currently,
2555 loop metadata is implemented as metadata attached to the branch instruction
2556 in the loop latch block. This type of metadata refer to a metadata node that is
2557 guaranteed to be separate for each loop. The loop identifier metadata is
2558 specified with the name ``llvm.loop``.
2560 The loop identifier metadata is implemented using a metadata that refers to
2561 itself to avoid merging it with any other identifier metadata, e.g.,
2562 during module linkage or function inlining. That is, each loop should refer
2563 to their own identification metadata even if they reside in separate functions.
2564 The following example contains loop identifier metadata for two separate loop
2567 .. code-block:: llvm
2569 !0 = metadata !{ metadata !0 }
2570 !1 = metadata !{ metadata !1 }
2572 The loop identifier metadata can be used to specify additional per-loop
2573 metadata. Any operands after the first operand can be treated as user-defined
2574 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2575 by the loop vectorizer to indicate how many times to unroll the loop:
2577 .. code-block:: llvm
2579 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2581 !0 = metadata !{ metadata !0, metadata !1 }
2582 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2587 Metadata types used to annotate memory accesses with information helpful
2588 for optimizations are prefixed with ``llvm.mem``.
2590 '``llvm.mem.parallel_loop_access``' Metadata
2591 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2593 For a loop to be parallel, in addition to using
2594 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2595 also all of the memory accessing instructions in the loop body need to be
2596 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2597 is at least one memory accessing instruction not marked with the metadata,
2598 the loop must be considered a sequential loop. This causes parallel loops to be
2599 converted to sequential loops due to optimization passes that are unaware of
2600 the parallel semantics and that insert new memory instructions to the loop
2603 Example of a loop that is considered parallel due to its correct use of
2604 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2605 metadata types that refer to the same loop identifier metadata.
2607 .. code-block:: llvm
2611 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2613 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2615 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2619 !0 = metadata !{ metadata !0 }
2621 It is also possible to have nested parallel loops. In that case the
2622 memory accesses refer to a list of loop identifier metadata nodes instead of
2623 the loop identifier metadata node directly:
2625 .. code-block:: llvm
2632 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2634 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2636 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2640 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2642 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2644 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2646 outer.for.end: ; preds = %for.body
2648 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2649 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2650 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2652 '``llvm.vectorizer``'
2653 ^^^^^^^^^^^^^^^^^^^^^
2655 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2656 vectorization parameters such as vectorization factor and unroll factor.
2658 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2659 loop identification metadata.
2661 '``llvm.vectorizer.unroll``' Metadata
2662 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2664 This metadata instructs the loop vectorizer to unroll the specified
2665 loop exactly ``N`` times.
2667 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2668 operand is an integer specifying the unroll factor. For example:
2670 .. code-block:: llvm
2672 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2674 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2677 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2678 determined automatically.
2680 '``llvm.vectorizer.width``' Metadata
2681 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2683 This metadata sets the target width of the vectorizer to ``N``. Without
2684 this metadata, the vectorizer will choose a width automatically.
2685 Regardless of this metadata, the vectorizer will only vectorize loops if
2686 it believes it is valid to do so.
2688 The first operand is the string ``llvm.vectorizer.width`` and the second
2689 operand is an integer specifying the width. For example:
2691 .. code-block:: llvm
2693 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2695 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2698 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2701 Module Flags Metadata
2702 =====================
2704 Information about the module as a whole is difficult to convey to LLVM's
2705 subsystems. The LLVM IR isn't sufficient to transmit this information.
2706 The ``llvm.module.flags`` named metadata exists in order to facilitate
2707 this. These flags are in the form of key / value pairs --- much like a
2708 dictionary --- making it easy for any subsystem who cares about a flag to
2711 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2712 Each triplet has the following form:
2714 - The first element is a *behavior* flag, which specifies the behavior
2715 when two (or more) modules are merged together, and it encounters two
2716 (or more) metadata with the same ID. The supported behaviors are
2718 - The second element is a metadata string that is a unique ID for the
2719 metadata. Each module may only have one flag entry for each unique ID (not
2720 including entries with the **Require** behavior).
2721 - The third element is the value of the flag.
2723 When two (or more) modules are merged together, the resulting
2724 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2725 each unique metadata ID string, there will be exactly one entry in the merged
2726 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2727 be determined by the merge behavior flag, as described below. The only exception
2728 is that entries with the *Require* behavior are always preserved.
2730 The following behaviors are supported:
2741 Emits an error if two values disagree, otherwise the resulting value
2742 is that of the operands.
2746 Emits a warning if two values disagree. The result value will be the
2747 operand for the flag from the first module being linked.
2751 Adds a requirement that another module flag be present and have a
2752 specified value after linking is performed. The value must be a
2753 metadata pair, where the first element of the pair is the ID of the
2754 module flag to be restricted, and the second element of the pair is
2755 the value the module flag should be restricted to. This behavior can
2756 be used to restrict the allowable results (via triggering of an
2757 error) of linking IDs with the **Override** behavior.
2761 Uses the specified value, regardless of the behavior or value of the
2762 other module. If both modules specify **Override**, but the values
2763 differ, an error will be emitted.
2767 Appends the two values, which are required to be metadata nodes.
2771 Appends the two values, which are required to be metadata
2772 nodes. However, duplicate entries in the second list are dropped
2773 during the append operation.
2775 It is an error for a particular unique flag ID to have multiple behaviors,
2776 except in the case of **Require** (which adds restrictions on another metadata
2777 value) or **Override**.
2779 An example of module flags:
2781 .. code-block:: llvm
2783 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2784 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2785 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2786 !3 = metadata !{ i32 3, metadata !"qux",
2788 metadata !"foo", i32 1
2791 !llvm.module.flags = !{ !0, !1, !2, !3 }
2793 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2794 if two or more ``!"foo"`` flags are seen is to emit an error if their
2795 values are not equal.
2797 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2798 behavior if two or more ``!"bar"`` flags are seen is to use the value
2801 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2802 behavior if two or more ``!"qux"`` flags are seen is to emit a
2803 warning if their values are not equal.
2805 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2809 metadata !{ metadata !"foo", i32 1 }
2811 The behavior is to emit an error if the ``llvm.module.flags`` does not
2812 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2815 Objective-C Garbage Collection Module Flags Metadata
2816 ----------------------------------------------------
2818 On the Mach-O platform, Objective-C stores metadata about garbage
2819 collection in a special section called "image info". The metadata
2820 consists of a version number and a bitmask specifying what types of
2821 garbage collection are supported (if any) by the file. If two or more
2822 modules are linked together their garbage collection metadata needs to
2823 be merged rather than appended together.
2825 The Objective-C garbage collection module flags metadata consists of the
2826 following key-value pairs:
2835 * - ``Objective-C Version``
2836 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2838 * - ``Objective-C Image Info Version``
2839 - **[Required]** --- The version of the image info section. Currently
2842 * - ``Objective-C Image Info Section``
2843 - **[Required]** --- The section to place the metadata. Valid values are
2844 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2845 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2846 Objective-C ABI version 2.
2848 * - ``Objective-C Garbage Collection``
2849 - **[Required]** --- Specifies whether garbage collection is supported or
2850 not. Valid values are 0, for no garbage collection, and 2, for garbage
2851 collection supported.
2853 * - ``Objective-C GC Only``
2854 - **[Optional]** --- Specifies that only garbage collection is supported.
2855 If present, its value must be 6. This flag requires that the
2856 ``Objective-C Garbage Collection`` flag have the value 2.
2858 Some important flag interactions:
2860 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2861 merged with a module with ``Objective-C Garbage Collection`` set to
2862 2, then the resulting module has the
2863 ``Objective-C Garbage Collection`` flag set to 0.
2864 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2865 merged with a module with ``Objective-C GC Only`` set to 6.
2867 Automatic Linker Flags Module Flags Metadata
2868 --------------------------------------------
2870 Some targets support embedding flags to the linker inside individual object
2871 files. Typically this is used in conjunction with language extensions which
2872 allow source files to explicitly declare the libraries they depend on, and have
2873 these automatically be transmitted to the linker via object files.
2875 These flags are encoded in the IR using metadata in the module flags section,
2876 using the ``Linker Options`` key. The merge behavior for this flag is required
2877 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2878 node which should be a list of other metadata nodes, each of which should be a
2879 list of metadata strings defining linker options.
2881 For example, the following metadata section specifies two separate sets of
2882 linker options, presumably to link against ``libz`` and the ``Cocoa``
2885 !0 = metadata !{ i32 6, metadata !"Linker Options",
2887 metadata !{ metadata !"-lz" },
2888 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2889 !llvm.module.flags = !{ !0 }
2891 The metadata encoding as lists of lists of options, as opposed to a collapsed
2892 list of options, is chosen so that the IR encoding can use multiple option
2893 strings to specify e.g., a single library, while still having that specifier be
2894 preserved as an atomic element that can be recognized by a target specific
2895 assembly writer or object file emitter.
2897 Each individual option is required to be either a valid option for the target's
2898 linker, or an option that is reserved by the target specific assembly writer or
2899 object file emitter. No other aspect of these options is defined by the IR.
2901 Intrinsic Global Variables
2902 ==========================
2904 LLVM has a number of "magic" global variables that contain data that
2905 affect code generation or other IR semantics. These are documented here.
2906 All globals of this sort should have a section specified as
2907 "``llvm.metadata``". This section and all globals that start with
2908 "``llvm.``" are reserved for use by LLVM.
2910 The '``llvm.used``' Global Variable
2911 -----------------------------------
2913 The ``@llvm.used`` global is an array which has
2914 :ref:`appending linkage <linkage_appending>`. This array contains a list of
2915 pointers to global variables, functions and aliases which may optionally have a
2916 pointer cast formed of bitcast or getelementptr. For example, a legal
2919 .. code-block:: llvm
2924 @llvm.used = appending global [2 x i8*] [
2926 i8* bitcast (i32* @Y to i8*)
2927 ], section "llvm.metadata"
2929 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
2930 and linker are required to treat the symbol as if there is a reference to the
2931 symbol that it cannot see. For example, if a variable has internal linkage and
2932 no references other than that from the ``@llvm.used`` list, it cannot be
2933 deleted. This is commonly used to represent references from inline asms and
2934 other things the compiler cannot "see", and corresponds to
2935 "``attribute((used))``" in GNU C.
2937 On some targets, the code generator must emit a directive to the
2938 assembler or object file to prevent the assembler and linker from
2939 molesting the symbol.
2941 The '``llvm.compiler.used``' Global Variable
2942 --------------------------------------------
2944 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
2945 directive, except that it only prevents the compiler from touching the
2946 symbol. On targets that support it, this allows an intelligent linker to
2947 optimize references to the symbol without being impeded as it would be
2950 This is a rare construct that should only be used in rare circumstances,
2951 and should not be exposed to source languages.
2953 The '``llvm.global_ctors``' Global Variable
2954 -------------------------------------------
2956 .. code-block:: llvm
2958 %0 = type { i32, void ()* }
2959 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2961 The ``@llvm.global_ctors`` array contains a list of constructor
2962 functions and associated priorities. The functions referenced by this
2963 array will be called in ascending order of priority (i.e. lowest first)
2964 when the module is loaded. The order of functions with the same priority
2967 The '``llvm.global_dtors``' Global Variable
2968 -------------------------------------------
2970 .. code-block:: llvm
2972 %0 = type { i32, void ()* }
2973 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
2975 The ``@llvm.global_dtors`` array contains a list of destructor functions
2976 and associated priorities. The functions referenced by this array will
2977 be called in descending order of priority (i.e. highest first) when the
2978 module is loaded. The order of functions with the same priority is not
2981 Instruction Reference
2982 =====================
2984 The LLVM instruction set consists of several different classifications
2985 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
2986 instructions <binaryops>`, :ref:`bitwise binary
2987 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
2988 :ref:`other instructions <otherops>`.
2992 Terminator Instructions
2993 -----------------------
2995 As mentioned :ref:`previously <functionstructure>`, every basic block in a
2996 program ends with a "Terminator" instruction, which indicates which
2997 block should be executed after the current block is finished. These
2998 terminator instructions typically yield a '``void``' value: they produce
2999 control flow, not values (the one exception being the
3000 ':ref:`invoke <i_invoke>`' instruction).
3002 The terminator instructions are: ':ref:`ret <i_ret>`',
3003 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3004 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3005 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3009 '``ret``' Instruction
3010 ^^^^^^^^^^^^^^^^^^^^^
3017 ret <type> <value> ; Return a value from a non-void function
3018 ret void ; Return from void function
3023 The '``ret``' instruction is used to return control flow (and optionally
3024 a value) from a function back to the caller.
3026 There are two forms of the '``ret``' instruction: one that returns a
3027 value and then causes control flow, and one that just causes control
3033 The '``ret``' instruction optionally accepts a single argument, the
3034 return value. The type of the return value must be a ':ref:`first
3035 class <t_firstclass>`' type.
3037 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3038 return type and contains a '``ret``' instruction with no return value or
3039 a return value with a type that does not match its type, or if it has a
3040 void return type and contains a '``ret``' instruction with a return
3046 When the '``ret``' instruction is executed, control flow returns back to
3047 the calling function's context. If the caller is a
3048 ":ref:`call <i_call>`" instruction, execution continues at the
3049 instruction after the call. If the caller was an
3050 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3051 beginning of the "normal" destination block. If the instruction returns
3052 a value, that value shall set the call or invoke instruction's return
3058 .. code-block:: llvm
3060 ret i32 5 ; Return an integer value of 5
3061 ret void ; Return from a void function
3062 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3066 '``br``' Instruction
3067 ^^^^^^^^^^^^^^^^^^^^
3074 br i1 <cond>, label <iftrue>, label <iffalse>
3075 br label <dest> ; Unconditional branch
3080 The '``br``' instruction is used to cause control flow to transfer to a
3081 different basic block in the current function. There are two forms of
3082 this instruction, corresponding to a conditional branch and an
3083 unconditional branch.
3088 The conditional branch form of the '``br``' instruction takes a single
3089 '``i1``' value and two '``label``' values. The unconditional form of the
3090 '``br``' instruction takes a single '``label``' value as a target.
3095 Upon execution of a conditional '``br``' instruction, the '``i1``'
3096 argument is evaluated. If the value is ``true``, control flows to the
3097 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3098 to the '``iffalse``' ``label`` argument.
3103 .. code-block:: llvm
3106 %cond = icmp eq i32 %a, %b
3107 br i1 %cond, label %IfEqual, label %IfUnequal
3115 '``switch``' Instruction
3116 ^^^^^^^^^^^^^^^^^^^^^^^^
3123 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3128 The '``switch``' instruction is used to transfer control flow to one of
3129 several different places. It is a generalization of the '``br``'
3130 instruction, allowing a branch to occur to one of many possible
3136 The '``switch``' instruction uses three parameters: an integer
3137 comparison value '``value``', a default '``label``' destination, and an
3138 array of pairs of comparison value constants and '``label``'s. The table
3139 is not allowed to contain duplicate constant entries.
3144 The ``switch`` instruction specifies a table of values and destinations.
3145 When the '``switch``' instruction is executed, this table is searched
3146 for the given value. If the value is found, control flow is transferred
3147 to the corresponding destination; otherwise, control flow is transferred
3148 to the default destination.
3153 Depending on properties of the target machine and the particular
3154 ``switch`` instruction, this instruction may be code generated in
3155 different ways. For example, it could be generated as a series of
3156 chained conditional branches or with a lookup table.
3161 .. code-block:: llvm
3163 ; Emulate a conditional br instruction
3164 %Val = zext i1 %value to i32
3165 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3167 ; Emulate an unconditional br instruction
3168 switch i32 0, label %dest [ ]
3170 ; Implement a jump table:
3171 switch i32 %val, label %otherwise [ i32 0, label %onzero
3173 i32 2, label %ontwo ]
3177 '``indirectbr``' Instruction
3178 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3185 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3190 The '``indirectbr``' instruction implements an indirect branch to a
3191 label within the current function, whose address is specified by
3192 "``address``". Address must be derived from a
3193 :ref:`blockaddress <blockaddress>` constant.
3198 The '``address``' argument is the address of the label to jump to. The
3199 rest of the arguments indicate the full set of possible destinations
3200 that the address may point to. Blocks are allowed to occur multiple
3201 times in the destination list, though this isn't particularly useful.
3203 This destination list is required so that dataflow analysis has an
3204 accurate understanding of the CFG.
3209 Control transfers to the block specified in the address argument. All
3210 possible destination blocks must be listed in the label list, otherwise
3211 this instruction has undefined behavior. This implies that jumps to
3212 labels defined in other functions have undefined behavior as well.
3217 This is typically implemented with a jump through a register.
3222 .. code-block:: llvm
3224 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3228 '``invoke``' Instruction
3229 ^^^^^^^^^^^^^^^^^^^^^^^^
3236 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3237 to label <normal label> unwind label <exception label>
3242 The '``invoke``' instruction causes control to transfer to a specified
3243 function, with the possibility of control flow transfer to either the
3244 '``normal``' label or the '``exception``' label. If the callee function
3245 returns with the "``ret``" instruction, control flow will return to the
3246 "normal" label. If the callee (or any indirect callees) returns via the
3247 ":ref:`resume <i_resume>`" instruction or other exception handling
3248 mechanism, control is interrupted and continued at the dynamically
3249 nearest "exception" label.
3251 The '``exception``' label is a `landing
3252 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3253 '``exception``' label is required to have the
3254 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3255 information about the behavior of the program after unwinding happens,
3256 as its first non-PHI instruction. The restrictions on the
3257 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3258 instruction, so that the important information contained within the
3259 "``landingpad``" instruction can't be lost through normal code motion.
3264 This instruction requires several arguments:
3266 #. The optional "cconv" marker indicates which :ref:`calling
3267 convention <callingconv>` the call should use. If none is
3268 specified, the call defaults to using C calling conventions.
3269 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3270 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3272 #. '``ptr to function ty``': shall be the signature of the pointer to
3273 function value being invoked. In most cases, this is a direct
3274 function invocation, but indirect ``invoke``'s are just as possible,
3275 branching off an arbitrary pointer to function value.
3276 #. '``function ptr val``': An LLVM value containing a pointer to a
3277 function to be invoked.
3278 #. '``function args``': argument list whose types match the function
3279 signature argument types and parameter attributes. All arguments must
3280 be of :ref:`first class <t_firstclass>` type. If the function signature
3281 indicates the function accepts a variable number of arguments, the
3282 extra arguments can be specified.
3283 #. '``normal label``': the label reached when the called function
3284 executes a '``ret``' instruction.
3285 #. '``exception label``': the label reached when a callee returns via
3286 the :ref:`resume <i_resume>` instruction or other exception handling
3288 #. The optional :ref:`function attributes <fnattrs>` list. Only
3289 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3290 attributes are valid here.
3295 This instruction is designed to operate as a standard '``call``'
3296 instruction in most regards. The primary difference is that it
3297 establishes an association with a label, which is used by the runtime
3298 library to unwind the stack.
3300 This instruction is used in languages with destructors to ensure that
3301 proper cleanup is performed in the case of either a ``longjmp`` or a
3302 thrown exception. Additionally, this is important for implementation of
3303 '``catch``' clauses in high-level languages that support them.
3305 For the purposes of the SSA form, the definition of the value returned
3306 by the '``invoke``' instruction is deemed to occur on the edge from the
3307 current block to the "normal" label. If the callee unwinds then no
3308 return value is available.
3313 .. code-block:: llvm
3315 %retval = invoke i32 @Test(i32 15) to label %Continue
3316 unwind label %TestCleanup ; {i32}:retval set
3317 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3318 unwind label %TestCleanup ; {i32}:retval set
3322 '``resume``' Instruction
3323 ^^^^^^^^^^^^^^^^^^^^^^^^
3330 resume <type> <value>
3335 The '``resume``' instruction is a terminator instruction that has no
3341 The '``resume``' instruction requires one argument, which must have the
3342 same type as the result of any '``landingpad``' instruction in the same
3348 The '``resume``' instruction resumes propagation of an existing
3349 (in-flight) exception whose unwinding was interrupted with a
3350 :ref:`landingpad <i_landingpad>` instruction.
3355 .. code-block:: llvm
3357 resume { i8*, i32 } %exn
3361 '``unreachable``' Instruction
3362 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3374 The '``unreachable``' instruction has no defined semantics. This
3375 instruction is used to inform the optimizer that a particular portion of
3376 the code is not reachable. This can be used to indicate that the code
3377 after a no-return function cannot be reached, and other facts.
3382 The '``unreachable``' instruction has no defined semantics.
3389 Binary operators are used to do most of the computation in a program.
3390 They require two operands of the same type, execute an operation on
3391 them, and produce a single value. The operands might represent multiple
3392 data, as is the case with the :ref:`vector <t_vector>` data type. The
3393 result value has the same type as its operands.
3395 There are several different binary operators:
3399 '``add``' Instruction
3400 ^^^^^^^^^^^^^^^^^^^^^
3407 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3408 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3409 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3410 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3415 The '``add``' instruction returns the sum of its two operands.
3420 The two arguments to the '``add``' instruction must be
3421 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3422 arguments must have identical types.
3427 The value produced is the integer sum of the two operands.
3429 If the sum has unsigned overflow, the result returned is the
3430 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3433 Because LLVM integers use a two's complement representation, this
3434 instruction is appropriate for both signed and unsigned integers.
3436 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3437 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3438 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3439 unsigned and/or signed overflow, respectively, occurs.
3444 .. code-block:: llvm
3446 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3450 '``fadd``' Instruction
3451 ^^^^^^^^^^^^^^^^^^^^^^
3458 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3463 The '``fadd``' instruction returns the sum of its two operands.
3468 The two arguments to the '``fadd``' instruction must be :ref:`floating
3469 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3470 Both arguments must have identical types.
3475 The value produced is the floating point sum of the two operands. This
3476 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3477 which are optimization hints to enable otherwise unsafe floating point
3483 .. code-block:: llvm
3485 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3487 '``sub``' Instruction
3488 ^^^^^^^^^^^^^^^^^^^^^
3495 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3496 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3497 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3498 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3503 The '``sub``' instruction returns the difference of its two operands.
3505 Note that the '``sub``' instruction is used to represent the '``neg``'
3506 instruction present in most other intermediate representations.
3511 The two arguments to the '``sub``' instruction must be
3512 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3513 arguments must have identical types.
3518 The value produced is the integer difference of the two operands.
3520 If the difference has unsigned overflow, the result returned is the
3521 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3524 Because LLVM integers use a two's complement representation, this
3525 instruction is appropriate for both signed and unsigned integers.
3527 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3528 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3529 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3530 unsigned and/or signed overflow, respectively, occurs.
3535 .. code-block:: llvm
3537 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3538 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3542 '``fsub``' Instruction
3543 ^^^^^^^^^^^^^^^^^^^^^^
3550 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3555 The '``fsub``' instruction returns the difference of its two operands.
3557 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3558 instruction present in most other intermediate representations.
3563 The two arguments to the '``fsub``' instruction must be :ref:`floating
3564 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3565 Both arguments must have identical types.
3570 The value produced is the floating point difference of the two operands.
3571 This instruction can also take any number of :ref:`fast-math
3572 flags <fastmath>`, which are optimization hints to enable otherwise
3573 unsafe floating point optimizations:
3578 .. code-block:: llvm
3580 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3581 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3583 '``mul``' Instruction
3584 ^^^^^^^^^^^^^^^^^^^^^
3591 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3592 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3593 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3594 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3599 The '``mul``' instruction returns the product of its two operands.
3604 The two arguments to the '``mul``' instruction must be
3605 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3606 arguments must have identical types.
3611 The value produced is the integer product of the two operands.
3613 If the result of the multiplication has unsigned overflow, the result
3614 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3615 bit width of the result.
3617 Because LLVM integers use a two's complement representation, and the
3618 result is the same width as the operands, this instruction returns the
3619 correct result for both signed and unsigned integers. If a full product
3620 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3621 sign-extended or zero-extended as appropriate to the width of the full
3624 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3625 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3626 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3627 unsigned and/or signed overflow, respectively, occurs.
3632 .. code-block:: llvm
3634 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3638 '``fmul``' Instruction
3639 ^^^^^^^^^^^^^^^^^^^^^^
3646 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3651 The '``fmul``' instruction returns the product of its two operands.
3656 The two arguments to the '``fmul``' instruction must be :ref:`floating
3657 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3658 Both arguments must have identical types.
3663 The value produced is the floating point product of the two operands.
3664 This instruction can also take any number of :ref:`fast-math
3665 flags <fastmath>`, which are optimization hints to enable otherwise
3666 unsafe floating point optimizations:
3671 .. code-block:: llvm
3673 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3675 '``udiv``' Instruction
3676 ^^^^^^^^^^^^^^^^^^^^^^
3683 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3684 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3689 The '``udiv``' instruction returns the quotient of its two operands.
3694 The two arguments to the '``udiv``' instruction must be
3695 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3696 arguments must have identical types.
3701 The value produced is the unsigned integer quotient of the two operands.
3703 Note that unsigned integer division and signed integer division are
3704 distinct operations; for signed integer division, use '``sdiv``'.
3706 Division by zero leads to undefined behavior.
3708 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3709 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3710 such, "((a udiv exact b) mul b) == a").
3715 .. code-block:: llvm
3717 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3719 '``sdiv``' Instruction
3720 ^^^^^^^^^^^^^^^^^^^^^^
3727 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3728 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3733 The '``sdiv``' instruction returns the quotient of its two operands.
3738 The two arguments to the '``sdiv``' instruction must be
3739 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3740 arguments must have identical types.
3745 The value produced is the signed integer quotient of the two operands
3746 rounded towards zero.
3748 Note that signed integer division and unsigned integer division are
3749 distinct operations; for unsigned integer division, use '``udiv``'.
3751 Division by zero leads to undefined behavior. Overflow also leads to
3752 undefined behavior; this is a rare case, but can occur, for example, by
3753 doing a 32-bit division of -2147483648 by -1.
3755 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3756 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3761 .. code-block:: llvm
3763 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3767 '``fdiv``' Instruction
3768 ^^^^^^^^^^^^^^^^^^^^^^
3775 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3780 The '``fdiv``' instruction returns the quotient of its two operands.
3785 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3786 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3787 Both arguments must have identical types.
3792 The value produced is the floating point quotient of the two operands.
3793 This instruction can also take any number of :ref:`fast-math
3794 flags <fastmath>`, which are optimization hints to enable otherwise
3795 unsafe floating point optimizations:
3800 .. code-block:: llvm
3802 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3804 '``urem``' Instruction
3805 ^^^^^^^^^^^^^^^^^^^^^^
3812 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3817 The '``urem``' instruction returns the remainder from the unsigned
3818 division of its two arguments.
3823 The two arguments to the '``urem``' instruction must be
3824 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3825 arguments must have identical types.
3830 This instruction returns the unsigned integer *remainder* of a division.
3831 This instruction always performs an unsigned division to get the
3834 Note that unsigned integer remainder and signed integer remainder are
3835 distinct operations; for signed integer remainder, use '``srem``'.
3837 Taking the remainder of a division by zero leads to undefined behavior.
3842 .. code-block:: llvm
3844 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3846 '``srem``' Instruction
3847 ^^^^^^^^^^^^^^^^^^^^^^
3854 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3859 The '``srem``' instruction returns the remainder from the signed
3860 division of its two operands. This instruction can also take
3861 :ref:`vector <t_vector>` versions of the values in which case the elements
3867 The two arguments to the '``srem``' instruction must be
3868 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3869 arguments must have identical types.
3874 This instruction returns the *remainder* of a division (where the result
3875 is either zero or has the same sign as the dividend, ``op1``), not the
3876 *modulo* operator (where the result is either zero or has the same sign
3877 as the divisor, ``op2``) of a value. For more information about the
3878 difference, see `The Math
3879 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3880 table of how this is implemented in various languages, please see
3882 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3884 Note that signed integer remainder and unsigned integer remainder are
3885 distinct operations; for unsigned integer remainder, use '``urem``'.
3887 Taking the remainder of a division by zero leads to undefined behavior.
3888 Overflow also leads to undefined behavior; this is a rare case, but can
3889 occur, for example, by taking the remainder of a 32-bit division of
3890 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3891 rule lets srem be implemented using instructions that return both the
3892 result of the division and the remainder.)
3897 .. code-block:: llvm
3899 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
3903 '``frem``' Instruction
3904 ^^^^^^^^^^^^^^^^^^^^^^
3911 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3916 The '``frem``' instruction returns the remainder from the division of
3922 The two arguments to the '``frem``' instruction must be :ref:`floating
3923 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3924 Both arguments must have identical types.
3929 This instruction returns the *remainder* of a division. The remainder
3930 has the same sign as the dividend. This instruction can also take any
3931 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
3932 to enable otherwise unsafe floating point optimizations:
3937 .. code-block:: llvm
3939 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
3943 Bitwise Binary Operations
3944 -------------------------
3946 Bitwise binary operators are used to do various forms of bit-twiddling
3947 in a program. They are generally very efficient instructions and can
3948 commonly be strength reduced from other instructions. They require two
3949 operands of the same type, execute an operation on them, and produce a
3950 single value. The resulting value is the same type as its operands.
3952 '``shl``' Instruction
3953 ^^^^^^^^^^^^^^^^^^^^^
3960 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
3961 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
3962 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
3963 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3968 The '``shl``' instruction returns the first operand shifted to the left
3969 a specified number of bits.
3974 Both arguments to the '``shl``' 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 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
3982 where ``n`` is the width of the result. If ``op2`` is (statically or
3983 dynamically) negative or equal to or larger than the number of bits in
3984 ``op1``, the result is undefined. If the arguments are vectors, each
3985 vector element of ``op1`` is shifted by the corresponding shift amount
3988 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
3989 value <poisonvalues>` if it shifts out any non-zero bits. If the
3990 ``nsw`` keyword is present, then the shift produces a :ref:`poison
3991 value <poisonvalues>` if it shifts out any bits that disagree with the
3992 resultant sign bit. As such, NUW/NSW have the same semantics as they
3993 would if the shift were expressed as a mul instruction with the same
3994 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
3999 .. code-block:: llvm
4001 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4002 <result> = shl i32 4, 2 ; yields {i32}: 16
4003 <result> = shl i32 1, 10 ; yields {i32}: 1024
4004 <result> = shl i32 1, 32 ; undefined
4005 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4007 '``lshr``' Instruction
4008 ^^^^^^^^^^^^^^^^^^^^^^
4015 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4016 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4021 The '``lshr``' instruction (logical shift right) returns the first
4022 operand shifted to the right a specified number of bits with zero fill.
4027 Both arguments to the '``lshr``' instruction must be the same
4028 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4029 '``op2``' is treated as an unsigned value.
4034 This instruction always performs a logical shift right operation. The
4035 most significant bits of the result will be filled with zero bits after
4036 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4037 than the number of bits in ``op1``, the result is undefined. If the
4038 arguments are vectors, each vector element of ``op1`` is shifted by the
4039 corresponding shift amount in ``op2``.
4041 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4042 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4048 .. code-block:: llvm
4050 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4051 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4052 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4053 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4054 <result> = lshr i32 1, 32 ; undefined
4055 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4057 '``ashr``' Instruction
4058 ^^^^^^^^^^^^^^^^^^^^^^
4065 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4066 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4071 The '``ashr``' instruction (arithmetic shift right) returns the first
4072 operand shifted to the right a specified number of bits with sign
4078 Both arguments to the '``ashr``' instruction must be the same
4079 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4080 '``op2``' is treated as an unsigned value.
4085 This instruction always performs an arithmetic shift right operation,
4086 The most significant bits of the result will be filled with the sign bit
4087 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4088 than the number of bits in ``op1``, the result is undefined. If the
4089 arguments are vectors, each vector element of ``op1`` is shifted by the
4090 corresponding shift amount in ``op2``.
4092 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4093 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4099 .. code-block:: llvm
4101 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4102 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4103 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4104 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4105 <result> = ashr i32 1, 32 ; undefined
4106 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4108 '``and``' Instruction
4109 ^^^^^^^^^^^^^^^^^^^^^
4116 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4121 The '``and``' instruction returns the bitwise logical and of its two
4127 The two arguments to the '``and``' instruction must be
4128 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4129 arguments must have identical types.
4134 The truth table used for the '``and``' instruction is:
4151 .. code-block:: llvm
4153 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4154 <result> = and i32 15, 40 ; yields {i32}:result = 8
4155 <result> = and i32 4, 8 ; yields {i32}:result = 0
4157 '``or``' Instruction
4158 ^^^^^^^^^^^^^^^^^^^^
4165 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4170 The '``or``' instruction returns the bitwise logical inclusive or of its
4176 The two arguments to the '``or``' instruction must be
4177 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4178 arguments must have identical types.
4183 The truth table used for the '``or``' instruction is:
4202 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4203 <result> = or i32 15, 40 ; yields {i32}:result = 47
4204 <result> = or i32 4, 8 ; yields {i32}:result = 12
4206 '``xor``' Instruction
4207 ^^^^^^^^^^^^^^^^^^^^^
4214 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4219 The '``xor``' instruction returns the bitwise logical exclusive or of
4220 its two operands. The ``xor`` is used to implement the "one's
4221 complement" operation, which is the "~" operator in C.
4226 The two arguments to the '``xor``' instruction must be
4227 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4228 arguments must have identical types.
4233 The truth table used for the '``xor``' instruction is:
4250 .. code-block:: llvm
4252 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4253 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4254 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4255 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4260 LLVM supports several instructions to represent vector operations in a
4261 target-independent manner. These instructions cover the element-access
4262 and vector-specific operations needed to process vectors effectively.
4263 While LLVM does directly support these vector operations, many
4264 sophisticated algorithms will want to use target-specific intrinsics to
4265 take full advantage of a specific target.
4267 .. _i_extractelement:
4269 '``extractelement``' Instruction
4270 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4277 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4282 The '``extractelement``' instruction extracts a single scalar element
4283 from a vector at a specified index.
4288 The first operand of an '``extractelement``' instruction is a value of
4289 :ref:`vector <t_vector>` type. The second operand is an index indicating
4290 the position from which to extract the element. The index may be a
4296 The result is a scalar of the same type as the element type of ``val``.
4297 Its value is the value at position ``idx`` of ``val``. If ``idx``
4298 exceeds the length of ``val``, the results are undefined.
4303 .. code-block:: llvm
4305 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4307 .. _i_insertelement:
4309 '``insertelement``' Instruction
4310 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4317 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4322 The '``insertelement``' instruction inserts a scalar element into a
4323 vector at a specified index.
4328 The first operand of an '``insertelement``' instruction is a value of
4329 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4330 type must equal the element type of the first operand. The third operand
4331 is an index indicating the position at which to insert the value. The
4332 index may be a variable.
4337 The result is a vector of the same type as ``val``. Its element values
4338 are those of ``val`` except at position ``idx``, where it gets the value
4339 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4345 .. code-block:: llvm
4347 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4349 .. _i_shufflevector:
4351 '``shufflevector``' Instruction
4352 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4359 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4364 The '``shufflevector``' instruction constructs a permutation of elements
4365 from two input vectors, returning a vector with the same element type as
4366 the input and length that is the same as the shuffle mask.
4371 The first two operands of a '``shufflevector``' instruction are vectors
4372 with the same type. The third argument is a shuffle mask whose element
4373 type is always 'i32'. The result of the instruction is a vector whose
4374 length is the same as the shuffle mask and whose element type is the
4375 same as the element type of the first two operands.
4377 The shuffle mask operand is required to be a constant vector with either
4378 constant integer or undef values.
4383 The elements of the two input vectors are numbered from left to right
4384 across both of the vectors. The shuffle mask operand specifies, for each
4385 element of the result vector, which element of the two input vectors the
4386 result element gets. The element selector may be undef (meaning "don't
4387 care") and the second operand may be undef if performing a shuffle from
4393 .. code-block:: llvm
4395 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4396 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4397 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4398 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4399 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4400 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4401 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4402 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4404 Aggregate Operations
4405 --------------------
4407 LLVM supports several instructions for working with
4408 :ref:`aggregate <t_aggregate>` values.
4412 '``extractvalue``' Instruction
4413 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4420 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4425 The '``extractvalue``' instruction extracts the value of a member field
4426 from an :ref:`aggregate <t_aggregate>` value.
4431 The first operand of an '``extractvalue``' instruction is a value of
4432 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4433 constant indices to specify which value to extract in a similar manner
4434 as indices in a '``getelementptr``' instruction.
4436 The major differences to ``getelementptr`` indexing are:
4438 - Since the value being indexed is not a pointer, the first index is
4439 omitted and assumed to be zero.
4440 - At least one index must be specified.
4441 - Not only struct indices but also array indices must be in bounds.
4446 The result is the value at the position in the aggregate specified by
4452 .. code-block:: llvm
4454 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4458 '``insertvalue``' Instruction
4459 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4466 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4471 The '``insertvalue``' instruction inserts a value into a member field in
4472 an :ref:`aggregate <t_aggregate>` value.
4477 The first operand of an '``insertvalue``' instruction is a value of
4478 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4479 a first-class value to insert. The following operands are constant
4480 indices indicating the position at which to insert the value in a
4481 similar manner as indices in a '``extractvalue``' instruction. The value
4482 to insert must have the same type as the value identified by the
4488 The result is an aggregate of the same type as ``val``. Its value is
4489 that of ``val`` except that the value at the position specified by the
4490 indices is that of ``elt``.
4495 .. code-block:: llvm
4497 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4498 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4499 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4503 Memory Access and Addressing Operations
4504 ---------------------------------------
4506 A key design point of an SSA-based representation is how it represents
4507 memory. In LLVM, no memory locations are in SSA form, which makes things
4508 very simple. This section describes how to read, write, and allocate
4513 '``alloca``' Instruction
4514 ^^^^^^^^^^^^^^^^^^^^^^^^
4521 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4526 The '``alloca``' instruction allocates memory on the stack frame of the
4527 currently executing function, to be automatically released when this
4528 function returns to its caller. The object is always allocated in the
4529 generic address space (address space zero).
4534 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4535 bytes of memory on the runtime stack, returning a pointer of the
4536 appropriate type to the program. If "NumElements" is specified, it is
4537 the number of elements allocated, otherwise "NumElements" is defaulted
4538 to be one. If a constant alignment is specified, the value result of the
4539 allocation is guaranteed to be aligned to at least that boundary. If not
4540 specified, or if zero, the target can choose to align the allocation on
4541 any convenient boundary compatible with the type.
4543 '``type``' may be any sized type.
4548 Memory is allocated; a pointer is returned. The operation is undefined
4549 if there is insufficient stack space for the allocation. '``alloca``'d
4550 memory is automatically released when the function returns. The
4551 '``alloca``' instruction is commonly used to represent automatic
4552 variables that must have an address available. When the function returns
4553 (either with the ``ret`` or ``resume`` instructions), the memory is
4554 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4555 The order in which memory is allocated (ie., which way the stack grows)
4561 .. code-block:: llvm
4563 %ptr = alloca i32 ; yields {i32*}:ptr
4564 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4565 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4566 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4570 '``load``' Instruction
4571 ^^^^^^^^^^^^^^^^^^^^^^
4578 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4579 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4580 !<index> = !{ i32 1 }
4585 The '``load``' instruction is used to read from memory.
4590 The argument to the ``load`` instruction specifies the memory address
4591 from which to load. The pointer must point to a :ref:`first
4592 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4593 then the optimizer is not allowed to modify the number or order of
4594 execution of this ``load`` with other :ref:`volatile
4595 operations <volatile>`.
4597 If the ``load`` is marked as ``atomic``, it takes an extra
4598 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4599 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4600 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4601 when they may see multiple atomic stores. The type of the pointee must
4602 be an integer type whose bit width is a power of two greater than or
4603 equal to eight and less than or equal to a target-specific size limit.
4604 ``align`` must be explicitly specified on atomic loads, and the load has
4605 undefined behavior if the alignment is not set to a value which is at
4606 least the size in bytes of the pointee. ``!nontemporal`` does not have
4607 any defined semantics for atomic loads.
4609 The optional constant ``align`` argument specifies the alignment of the
4610 operation (that is, the alignment of the memory address). A value of 0
4611 or an omitted ``align`` argument means that the operation has the ABI
4612 alignment for the target. It is the responsibility of the code emitter
4613 to ensure that the alignment information is correct. Overestimating the
4614 alignment results in undefined behavior. Underestimating the alignment
4615 may produce less efficient code. An alignment of 1 is always safe.
4617 The optional ``!nontemporal`` metadata must reference a single
4618 metatadata name ``<index>`` corresponding to a metadata node with one
4619 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4620 metatadata on the instruction tells the optimizer and code generator
4621 that this load is not expected to be reused in the cache. The code
4622 generator may select special instructions to save cache bandwidth, such
4623 as the ``MOVNT`` instruction on x86.
4625 The optional ``!invariant.load`` metadata must reference a single
4626 metatadata name ``<index>`` corresponding to a metadata node with no
4627 entries. The existence of the ``!invariant.load`` metatadata on the
4628 instruction tells the optimizer and code generator that this load
4629 address points to memory which does not change value during program
4630 execution. The optimizer may then move this load around, for example, by
4631 hoisting it out of loops using loop invariant code motion.
4636 The location of memory pointed to is loaded. If the value being loaded
4637 is of scalar type then the number of bytes read does not exceed the
4638 minimum number of bytes needed to hold all bits of the type. For
4639 example, loading an ``i24`` reads at most three bytes. When loading a
4640 value of a type like ``i20`` with a size that is not an integral number
4641 of bytes, the result is undefined if the value was not originally
4642 written using a store of the same type.
4647 .. code-block:: llvm
4649 %ptr = alloca i32 ; yields {i32*}:ptr
4650 store i32 3, i32* %ptr ; yields {void}
4651 %val = load i32* %ptr ; yields {i32}:val = i32 3
4655 '``store``' Instruction
4656 ^^^^^^^^^^^^^^^^^^^^^^^
4663 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4664 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4669 The '``store``' instruction is used to write to memory.
4674 There are two arguments to the ``store`` instruction: a value to store
4675 and an address at which to store it. The type of the ``<pointer>``
4676 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4677 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4678 then the optimizer is not allowed to modify the number or order of
4679 execution of this ``store`` with other :ref:`volatile
4680 operations <volatile>`.
4682 If the ``store`` is marked as ``atomic``, it takes an extra
4683 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4684 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4685 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4686 when they may see multiple atomic stores. The type of the pointee must
4687 be an integer type whose bit width is a power of two greater than or
4688 equal to eight and less than or equal to a target-specific size limit.
4689 ``align`` must be explicitly specified on atomic stores, and the store
4690 has undefined behavior if the alignment is not set to a value which is
4691 at least the size in bytes of the pointee. ``!nontemporal`` does not
4692 have any defined semantics for atomic stores.
4694 The optional constant ``align`` argument specifies the alignment of the
4695 operation (that is, the alignment of the memory address). A value of 0
4696 or an omitted ``align`` argument means that the operation has the ABI
4697 alignment for the target. It is the responsibility of the code emitter
4698 to ensure that the alignment information is correct. Overestimating the
4699 alignment results in undefined behavior. Underestimating the
4700 alignment may produce less efficient code. An alignment of 1 is always
4703 The optional ``!nontemporal`` metadata must reference a single metatadata
4704 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4705 value 1. The existence of the ``!nontemporal`` metatadata on the instruction
4706 tells the optimizer and code generator that this load is not expected to
4707 be reused in the cache. The code generator may select special
4708 instructions to save cache bandwidth, such as the MOVNT instruction on
4714 The contents of memory are updated to contain ``<value>`` at the
4715 location specified by the ``<pointer>`` operand. If ``<value>`` is
4716 of scalar type then the number of bytes written does not exceed the
4717 minimum number of bytes needed to hold all bits of the type. For
4718 example, storing an ``i24`` writes at most three bytes. When writing a
4719 value of a type like ``i20`` with a size that is not an integral number
4720 of bytes, it is unspecified what happens to the extra bits that do not
4721 belong to the type, but they will typically be overwritten.
4726 .. code-block:: llvm
4728 %ptr = alloca i32 ; yields {i32*}:ptr
4729 store i32 3, i32* %ptr ; yields {void}
4730 %val = load i32* %ptr ; yields {i32}:val = i32 3
4734 '``fence``' Instruction
4735 ^^^^^^^^^^^^^^^^^^^^^^^
4742 fence [singlethread] <ordering> ; yields {void}
4747 The '``fence``' instruction is used to introduce happens-before edges
4753 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4754 defines what *synchronizes-with* edges they add. They can only be given
4755 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4760 A fence A which has (at least) ``release`` ordering semantics
4761 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4762 semantics if and only if there exist atomic operations X and Y, both
4763 operating on some atomic object M, such that A is sequenced before X, X
4764 modifies M (either directly or through some side effect of a sequence
4765 headed by X), Y is sequenced before B, and Y observes M. This provides a
4766 *happens-before* dependency between A and B. Rather than an explicit
4767 ``fence``, one (but not both) of the atomic operations X or Y might
4768 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4769 still *synchronize-with* the explicit ``fence`` and establish the
4770 *happens-before* edge.
4772 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4773 ``acquire`` and ``release`` semantics specified above, participates in
4774 the global program order of other ``seq_cst`` operations and/or fences.
4776 The optional ":ref:`singlethread <singlethread>`" argument specifies
4777 that the fence only synchronizes with other fences in the same thread.
4778 (This is useful for interacting with signal handlers.)
4783 .. code-block:: llvm
4785 fence acquire ; yields {void}
4786 fence singlethread seq_cst ; yields {void}
4790 '``cmpxchg``' Instruction
4791 ^^^^^^^^^^^^^^^^^^^^^^^^^
4798 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4803 The '``cmpxchg``' instruction is used to atomically modify memory. It
4804 loads a value in memory and compares it to a given value. If they are
4805 equal, it stores a new value into the memory.
4810 There are three arguments to the '``cmpxchg``' instruction: an address
4811 to operate on, a value to compare to the value currently be at that
4812 address, and a new value to place at that address if the compared values
4813 are equal. The type of '<cmp>' must be an integer type whose bit width
4814 is a power of two greater than or equal to eight and less than or equal
4815 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4816 type, and the type of '<pointer>' must be a pointer to that type. If the
4817 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4818 to modify the number or order of execution of this ``cmpxchg`` with
4819 other :ref:`volatile operations <volatile>`.
4821 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4822 synchronizes with other atomic operations.
4824 The optional "``singlethread``" argument declares that the ``cmpxchg``
4825 is only atomic with respect to code (usually signal handlers) running in
4826 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4827 respect to all other code in the system.
4829 The pointer passed into cmpxchg must have alignment greater than or
4830 equal to the size in memory of the operand.
4835 The contents of memory at the location specified by the '``<pointer>``'
4836 operand is read and compared to '``<cmp>``'; if the read value is the
4837 equal, '``<new>``' is written. The original value at the location is
4840 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4841 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4842 atomic load with an ordering parameter determined by dropping any
4843 ``release`` part of the ``cmpxchg``'s ordering.
4848 .. code-block:: llvm
4851 %orig = atomic load i32* %ptr unordered ; yields {i32}
4855 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4856 %squared = mul i32 %cmp, %cmp
4857 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4858 %success = icmp eq i32 %cmp, %old
4859 br i1 %success, label %done, label %loop
4866 '``atomicrmw``' Instruction
4867 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4874 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4879 The '``atomicrmw``' instruction is used to atomically modify memory.
4884 There are three arguments to the '``atomicrmw``' instruction: an
4885 operation to apply, an address whose value to modify, an argument to the
4886 operation. The operation must be one of the following keywords:
4900 The type of '<value>' must be an integer type whose bit width is a power
4901 of two greater than or equal to eight and less than or equal to a
4902 target-specific size limit. The type of the '``<pointer>``' operand must
4903 be a pointer to that type. If the ``atomicrmw`` is marked as
4904 ``volatile``, then the optimizer is not allowed to modify the number or
4905 order of execution of this ``atomicrmw`` with other :ref:`volatile
4906 operations <volatile>`.
4911 The contents of memory at the location specified by the '``<pointer>``'
4912 operand are atomically read, modified, and written back. The original
4913 value at the location is returned. The modification is specified by the
4916 - xchg: ``*ptr = val``
4917 - add: ``*ptr = *ptr + val``
4918 - sub: ``*ptr = *ptr - val``
4919 - and: ``*ptr = *ptr & val``
4920 - nand: ``*ptr = ~(*ptr & val)``
4921 - or: ``*ptr = *ptr | val``
4922 - xor: ``*ptr = *ptr ^ val``
4923 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
4924 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
4925 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
4927 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
4933 .. code-block:: llvm
4935 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
4937 .. _i_getelementptr:
4939 '``getelementptr``' Instruction
4940 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4947 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4948 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4949 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
4954 The '``getelementptr``' instruction is used to get the address of a
4955 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
4956 address calculation only and does not access memory.
4961 The first argument is always a pointer or a vector of pointers, and
4962 forms the basis of the calculation. The remaining arguments are indices
4963 that indicate which of the elements of the aggregate object are indexed.
4964 The interpretation of each index is dependent on the type being indexed
4965 into. The first index always indexes the pointer value given as the
4966 first argument, the second index indexes a value of the type pointed to
4967 (not necessarily the value directly pointed to, since the first index
4968 can be non-zero), etc. The first type indexed into must be a pointer
4969 value, subsequent types can be arrays, vectors, and structs. Note that
4970 subsequent types being indexed into can never be pointers, since that
4971 would require loading the pointer before continuing calculation.
4973 The type of each index argument depends on the type it is indexing into.
4974 When indexing into a (optionally packed) structure, only ``i32`` integer
4975 **constants** are allowed (when using a vector of indices they must all
4976 be the **same** ``i32`` integer constant). When indexing into an array,
4977 pointer or vector, integers of any width are allowed, and they are not
4978 required to be constant. These integers are treated as signed values
4981 For example, let's consider a C code fragment and how it gets compiled
4997 int *foo(struct ST *s) {
4998 return &s[1].Z.B[5][13];
5001 The LLVM code generated by Clang is:
5003 .. code-block:: llvm
5005 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5006 %struct.ST = type { i32, double, %struct.RT }
5008 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5010 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5017 In the example above, the first index is indexing into the
5018 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5019 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5020 indexes into the third element of the structure, yielding a
5021 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5022 structure. The third index indexes into the second element of the
5023 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5024 dimensions of the array are subscripted into, yielding an '``i32``'
5025 type. The '``getelementptr``' instruction returns a pointer to this
5026 element, thus computing a value of '``i32*``' type.
5028 Note that it is perfectly legal to index partially through a structure,
5029 returning a pointer to an inner element. Because of this, the LLVM code
5030 for the given testcase is equivalent to:
5032 .. code-block:: llvm
5034 define i32* @foo(%struct.ST* %s) {
5035 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5036 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5037 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5038 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5039 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5043 If the ``inbounds`` keyword is present, the result value of the
5044 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5045 pointer is not an *in bounds* address of an allocated object, or if any
5046 of the addresses that would be formed by successive addition of the
5047 offsets implied by the indices to the base address with infinitely
5048 precise signed arithmetic are not an *in bounds* address of that
5049 allocated object. The *in bounds* addresses for an allocated object are
5050 all the addresses that point into the object, plus the address one byte
5051 past the end. In cases where the base is a vector of pointers the
5052 ``inbounds`` keyword applies to each of the computations element-wise.
5054 If the ``inbounds`` keyword is not present, the offsets are added to the
5055 base address with silently-wrapping two's complement arithmetic. If the
5056 offsets have a different width from the pointer, they are sign-extended
5057 or truncated to the width of the pointer. The result value of the
5058 ``getelementptr`` may be outside the object pointed to by the base
5059 pointer. The result value may not necessarily be used to access memory
5060 though, even if it happens to point into allocated storage. See the
5061 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5064 The getelementptr instruction is often confusing. For some more insight
5065 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5070 .. code-block:: llvm
5072 ; yields [12 x i8]*:aptr
5073 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5075 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5077 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5079 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5081 In cases where the pointer argument is a vector of pointers, each index
5082 must be a vector with the same number of elements. For example:
5084 .. code-block:: llvm
5086 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5088 Conversion Operations
5089 ---------------------
5091 The instructions in this category are the conversion instructions
5092 (casting) which all take a single operand and a type. They perform
5093 various bit conversions on the operand.
5095 '``trunc .. to``' Instruction
5096 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5103 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5108 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5113 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5114 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5115 of the same number of integers. The bit size of the ``value`` must be
5116 larger than the bit size of the destination type, ``ty2``. Equal sized
5117 types are not allowed.
5122 The '``trunc``' instruction truncates the high order bits in ``value``
5123 and converts the remaining bits to ``ty2``. Since the source size must
5124 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5125 It will always truncate bits.
5130 .. code-block:: llvm
5132 %X = trunc i32 257 to i8 ; yields i8:1
5133 %Y = trunc i32 123 to i1 ; yields i1:true
5134 %Z = trunc i32 122 to i1 ; yields i1:false
5135 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5137 '``zext .. to``' Instruction
5138 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5145 <result> = zext <ty> <value> to <ty2> ; yields ty2
5150 The '``zext``' instruction zero extends its operand to type ``ty2``.
5155 The '``zext``' instruction takes a value to cast, and a type to cast it
5156 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5157 the same number of integers. The bit size of the ``value`` must be
5158 smaller than the bit size of the destination type, ``ty2``.
5163 The ``zext`` fills the high order bits of the ``value`` with zero bits
5164 until it reaches the size of the destination type, ``ty2``.
5166 When zero extending from i1, the result will always be either 0 or 1.
5171 .. code-block:: llvm
5173 %X = zext i32 257 to i64 ; yields i64:257
5174 %Y = zext i1 true to i32 ; yields i32:1
5175 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5177 '``sext .. to``' Instruction
5178 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5185 <result> = sext <ty> <value> to <ty2> ; yields ty2
5190 The '``sext``' sign extends ``value`` to the type ``ty2``.
5195 The '``sext``' instruction takes a value to cast, and a type to cast it
5196 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5197 the same number of integers. The bit size of the ``value`` must be
5198 smaller than the bit size of the destination type, ``ty2``.
5203 The '``sext``' instruction performs a sign extension by copying the sign
5204 bit (highest order bit) of the ``value`` until it reaches the bit size
5205 of the type ``ty2``.
5207 When sign extending from i1, the extension always results in -1 or 0.
5212 .. code-block:: llvm
5214 %X = sext i8 -1 to i16 ; yields i16 :65535
5215 %Y = sext i1 true to i32 ; yields i32:-1
5216 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5218 '``fptrunc .. to``' Instruction
5219 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5226 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5231 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5236 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5237 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5238 The size of ``value`` must be larger than the size of ``ty2``. This
5239 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5244 The '``fptrunc``' instruction truncates a ``value`` from a larger
5245 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5246 point <t_floating>` type. If the value cannot fit within the
5247 destination type, ``ty2``, then the results are undefined.
5252 .. code-block:: llvm
5254 %X = fptrunc double 123.0 to float ; yields float:123.0
5255 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5257 '``fpext .. to``' Instruction
5258 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5265 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5270 The '``fpext``' extends a floating point ``value`` to a larger floating
5276 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5277 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5278 to. The source type must be smaller than the destination type.
5283 The '``fpext``' instruction extends the ``value`` from a smaller
5284 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5285 point <t_floating>` type. The ``fpext`` cannot be used to make a
5286 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5287 *no-op cast* for a floating point cast.
5292 .. code-block:: llvm
5294 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5295 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5297 '``fptoui .. to``' Instruction
5298 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5305 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5310 The '``fptoui``' converts a floating point ``value`` to its unsigned
5311 integer equivalent of type ``ty2``.
5316 The '``fptoui``' instruction takes a value to cast, which must be a
5317 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5318 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5319 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5320 type with the same number of elements as ``ty``
5325 The '``fptoui``' instruction converts its :ref:`floating
5326 point <t_floating>` operand into the nearest (rounding towards zero)
5327 unsigned integer value. If the value cannot fit in ``ty2``, the results
5333 .. code-block:: llvm
5335 %X = fptoui double 123.0 to i32 ; yields i32:123
5336 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5337 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5339 '``fptosi .. to``' Instruction
5340 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5347 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5352 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5353 ``value`` to type ``ty2``.
5358 The '``fptosi``' instruction takes a value to cast, which must be a
5359 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5360 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5361 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5362 type with the same number of elements as ``ty``
5367 The '``fptosi``' instruction converts its :ref:`floating
5368 point <t_floating>` operand into the nearest (rounding towards zero)
5369 signed integer value. If the value cannot fit in ``ty2``, the results
5375 .. code-block:: llvm
5377 %X = fptosi double -123.0 to i32 ; yields i32:-123
5378 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5379 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5381 '``uitofp .. to``' Instruction
5382 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5389 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5394 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5395 and converts that value to the ``ty2`` type.
5400 The '``uitofp``' instruction takes a value to cast, which must be a
5401 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5402 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5403 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5404 type with the same number of elements as ``ty``
5409 The '``uitofp``' instruction interprets its operand as an unsigned
5410 integer quantity and converts it to the corresponding floating point
5411 value. If the value cannot fit in the floating point value, the results
5417 .. code-block:: llvm
5419 %X = uitofp i32 257 to float ; yields float:257.0
5420 %Y = uitofp i8 -1 to double ; yields double:255.0
5422 '``sitofp .. to``' Instruction
5423 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5430 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5435 The '``sitofp``' instruction regards ``value`` as a signed integer and
5436 converts that value to the ``ty2`` type.
5441 The '``sitofp``' instruction takes a value to cast, which must be a
5442 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5443 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5444 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5445 type with the same number of elements as ``ty``
5450 The '``sitofp``' instruction interprets its operand as a signed integer
5451 quantity and converts it to the corresponding floating point value. If
5452 the value cannot fit in the floating point value, the results are
5458 .. code-block:: llvm
5460 %X = sitofp i32 257 to float ; yields float:257.0
5461 %Y = sitofp i8 -1 to double ; yields double:-1.0
5465 '``ptrtoint .. to``' Instruction
5466 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5473 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5478 The '``ptrtoint``' instruction converts the pointer or a vector of
5479 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5484 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5485 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5486 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5487 a vector of integers type.
5492 The '``ptrtoint``' instruction converts ``value`` to integer type
5493 ``ty2`` by interpreting the pointer value as an integer and either
5494 truncating or zero extending that value to the size of the integer type.
5495 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5496 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5497 the same size, then nothing is done (*no-op cast*) other than a type
5503 .. code-block:: llvm
5505 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5506 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5507 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5511 '``inttoptr .. to``' Instruction
5512 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5519 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5524 The '``inttoptr``' instruction converts an integer ``value`` to a
5525 pointer type, ``ty2``.
5530 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5531 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5537 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5538 applying either a zero extension or a truncation depending on the size
5539 of the integer ``value``. If ``value`` is larger than the size of a
5540 pointer then a truncation is done. If ``value`` is smaller than the size
5541 of a pointer then a zero extension is done. If they are the same size,
5542 nothing is done (*no-op cast*).
5547 .. code-block:: llvm
5549 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5550 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5551 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5552 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5556 '``bitcast .. to``' Instruction
5557 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5564 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5569 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5575 The '``bitcast``' instruction takes a value to cast, which must be a
5576 non-aggregate first class value, and a type to cast it to, which must
5577 also be a non-aggregate :ref:`first class <t_firstclass>` type. The bit
5578 sizes of ``value`` and the destination type, ``ty2``, must be identical.
5579 If the source type is a pointer, the destination type must also be a
5580 pointer. This instruction supports bitwise conversion of vectors to
5581 integers and to vectors of other types (as long as they have the same
5587 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It is
5588 always a *no-op cast* because no bits change with this conversion. The
5589 conversion is done as if the ``value`` had been stored to memory and
5590 read back as type ``ty2``. Pointer (or vector of pointers) types may
5591 only be converted to other pointer (or vector of pointers) types with
5592 this instruction. To convert pointers to other types, use the
5593 :ref:`inttoptr <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions
5599 .. code-block:: llvm
5601 %X = bitcast i8 255 to i8 ; yields i8 :-1
5602 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5603 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5604 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5611 The instructions in this category are the "miscellaneous" instructions,
5612 which defy better classification.
5616 '``icmp``' Instruction
5617 ^^^^^^^^^^^^^^^^^^^^^^
5624 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5629 The '``icmp``' instruction returns a boolean value or a vector of
5630 boolean values based on comparison of its two integer, integer vector,
5631 pointer, or pointer vector operands.
5636 The '``icmp``' instruction takes three operands. The first operand is
5637 the condition code indicating the kind of comparison to perform. It is
5638 not a value, just a keyword. The possible condition code are:
5641 #. ``ne``: not equal
5642 #. ``ugt``: unsigned greater than
5643 #. ``uge``: unsigned greater or equal
5644 #. ``ult``: unsigned less than
5645 #. ``ule``: unsigned less or equal
5646 #. ``sgt``: signed greater than
5647 #. ``sge``: signed greater or equal
5648 #. ``slt``: signed less than
5649 #. ``sle``: signed less or equal
5651 The remaining two arguments must be :ref:`integer <t_integer>` or
5652 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5653 must also be identical types.
5658 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5659 code given as ``cond``. The comparison performed always yields either an
5660 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5662 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5663 otherwise. No sign interpretation is necessary or performed.
5664 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5665 otherwise. No sign interpretation is necessary or performed.
5666 #. ``ugt``: interprets the operands as unsigned values and yields
5667 ``true`` if ``op1`` is greater than ``op2``.
5668 #. ``uge``: interprets the operands as unsigned values and yields
5669 ``true`` if ``op1`` is greater than or equal to ``op2``.
5670 #. ``ult``: interprets the operands as unsigned values and yields
5671 ``true`` if ``op1`` is less than ``op2``.
5672 #. ``ule``: interprets the operands as unsigned values and yields
5673 ``true`` if ``op1`` is less than or equal to ``op2``.
5674 #. ``sgt``: interprets the operands as signed values and yields ``true``
5675 if ``op1`` is greater than ``op2``.
5676 #. ``sge``: interprets the operands as signed values and yields ``true``
5677 if ``op1`` is greater than or equal to ``op2``.
5678 #. ``slt``: interprets the operands as signed values and yields ``true``
5679 if ``op1`` is less than ``op2``.
5680 #. ``sle``: interprets the operands as signed values and yields ``true``
5681 if ``op1`` is less than or equal to ``op2``.
5683 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5684 are compared as if they were integers.
5686 If the operands are integer vectors, then they are compared element by
5687 element. The result is an ``i1`` vector with the same number of elements
5688 as the values being compared. Otherwise, the result is an ``i1``.
5693 .. code-block:: llvm
5695 <result> = icmp eq i32 4, 5 ; yields: result=false
5696 <result> = icmp ne float* %X, %X ; yields: result=false
5697 <result> = icmp ult i16 4, 5 ; yields: result=true
5698 <result> = icmp sgt i16 4, 5 ; yields: result=false
5699 <result> = icmp ule i16 -4, 5 ; yields: result=false
5700 <result> = icmp sge i16 4, 5 ; yields: result=false
5702 Note that the code generator does not yet support vector types with the
5703 ``icmp`` instruction.
5707 '``fcmp``' Instruction
5708 ^^^^^^^^^^^^^^^^^^^^^^
5715 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5720 The '``fcmp``' instruction returns a boolean value or vector of boolean
5721 values based on comparison of its operands.
5723 If the operands are floating point scalars, then the result type is a
5724 boolean (:ref:`i1 <t_integer>`).
5726 If the operands are floating point vectors, then the result type is a
5727 vector of boolean with the same number of elements as the operands being
5733 The '``fcmp``' instruction takes three operands. The first operand is
5734 the condition code indicating the kind of comparison to perform. It is
5735 not a value, just a keyword. The possible condition code are:
5737 #. ``false``: no comparison, always returns false
5738 #. ``oeq``: ordered and equal
5739 #. ``ogt``: ordered and greater than
5740 #. ``oge``: ordered and greater than or equal
5741 #. ``olt``: ordered and less than
5742 #. ``ole``: ordered and less than or equal
5743 #. ``one``: ordered and not equal
5744 #. ``ord``: ordered (no nans)
5745 #. ``ueq``: unordered or equal
5746 #. ``ugt``: unordered or greater than
5747 #. ``uge``: unordered or greater than or equal
5748 #. ``ult``: unordered or less than
5749 #. ``ule``: unordered or less than or equal
5750 #. ``une``: unordered or not equal
5751 #. ``uno``: unordered (either nans)
5752 #. ``true``: no comparison, always returns true
5754 *Ordered* means that neither operand is a QNAN while *unordered* means
5755 that either operand may be a QNAN.
5757 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5758 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5759 type. They must have identical types.
5764 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5765 condition code given as ``cond``. If the operands are vectors, then the
5766 vectors are compared element by element. Each comparison performed
5767 always yields an :ref:`i1 <t_integer>` result, as follows:
5769 #. ``false``: always yields ``false``, regardless of operands.
5770 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5771 is equal to ``op2``.
5772 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5773 is greater than ``op2``.
5774 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5775 is greater than or equal to ``op2``.
5776 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5777 is less than ``op2``.
5778 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5779 is less than or equal to ``op2``.
5780 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5781 is not equal to ``op2``.
5782 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5783 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5785 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5786 greater than ``op2``.
5787 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5788 greater than or equal to ``op2``.
5789 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5791 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5792 less than or equal to ``op2``.
5793 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5794 not equal to ``op2``.
5795 #. ``uno``: yields ``true`` if either operand is a QNAN.
5796 #. ``true``: always yields ``true``, regardless of operands.
5801 .. code-block:: llvm
5803 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5804 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5805 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5806 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5808 Note that the code generator does not yet support vector types with the
5809 ``fcmp`` instruction.
5813 '``phi``' Instruction
5814 ^^^^^^^^^^^^^^^^^^^^^
5821 <result> = phi <ty> [ <val0>, <label0>], ...
5826 The '``phi``' instruction is used to implement the φ node in the SSA
5827 graph representing the function.
5832 The type of the incoming values is specified with the first type field.
5833 After this, the '``phi``' instruction takes a list of pairs as
5834 arguments, with one pair for each predecessor basic block of the current
5835 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5836 the value arguments to the PHI node. Only labels may be used as the
5839 There must be no non-phi instructions between the start of a basic block
5840 and the PHI instructions: i.e. PHI instructions must be first in a basic
5843 For the purposes of the SSA form, the use of each incoming value is
5844 deemed to occur on the edge from the corresponding predecessor block to
5845 the current block (but after any definition of an '``invoke``'
5846 instruction's return value on the same edge).
5851 At runtime, the '``phi``' instruction logically takes on the value
5852 specified by the pair corresponding to the predecessor basic block that
5853 executed just prior to the current block.
5858 .. code-block:: llvm
5860 Loop: ; Infinite loop that counts from 0 on up...
5861 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5862 %nextindvar = add i32 %indvar, 1
5867 '``select``' Instruction
5868 ^^^^^^^^^^^^^^^^^^^^^^^^
5875 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5877 selty is either i1 or {<N x i1>}
5882 The '``select``' instruction is used to choose one value based on a
5883 condition, without branching.
5888 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
5889 values indicating the condition, and two values of the same :ref:`first
5890 class <t_firstclass>` type. If the val1/val2 are vectors and the
5891 condition is a scalar, then entire vectors are selected, not individual
5897 If the condition is an i1 and it evaluates to 1, the instruction returns
5898 the first value argument; otherwise, it returns the second value
5901 If the condition is a vector of i1, then the value arguments must be
5902 vectors of the same size, and the selection is done element by element.
5907 .. code-block:: llvm
5909 %X = select i1 true, i8 17, i8 42 ; yields i8:17
5913 '``call``' Instruction
5914 ^^^^^^^^^^^^^^^^^^^^^^
5921 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
5926 The '``call``' instruction represents a simple function call.
5931 This instruction requires several arguments:
5933 #. The optional "tail" marker indicates that the callee function does
5934 not access any allocas or varargs in the caller. Note that calls may
5935 be marked "tail" even if they do not occur before a
5936 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
5937 function call is eligible for tail call optimization, but `might not
5938 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
5939 The code generator may optimize calls marked "tail" with either 1)
5940 automatic `sibling call
5941 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
5942 callee have matching signatures, or 2) forced tail call optimization
5943 when the following extra requirements are met:
5945 - Caller and callee both have the calling convention ``fastcc``.
5946 - The call is in tail position (ret immediately follows call and ret
5947 uses value of call or is void).
5948 - Option ``-tailcallopt`` is enabled, or
5949 ``llvm::GuaranteedTailCallOpt`` is ``true``.
5950 - `Platform specific constraints are
5951 met. <CodeGenerator.html#tailcallopt>`_
5953 #. The optional "cconv" marker indicates which :ref:`calling
5954 convention <callingconv>` the call should use. If none is
5955 specified, the call defaults to using C calling conventions. The
5956 calling convention of the call must match the calling convention of
5957 the target function, or else the behavior is undefined.
5958 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5959 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5961 #. '``ty``': the type of the call instruction itself which is also the
5962 type of the return value. Functions that return no value are marked
5964 #. '``fnty``': shall be the signature of the pointer to function value
5965 being invoked. The argument types must match the types implied by
5966 this signature. This type can be omitted if the function is not
5967 varargs and if the function type does not return a pointer to a
5969 #. '``fnptrval``': An LLVM value containing a pointer to a function to
5970 be invoked. In most cases, this is a direct function invocation, but
5971 indirect ``call``'s are just as possible, calling an arbitrary pointer
5973 #. '``function args``': argument list whose types match the function
5974 signature argument types and parameter attributes. All arguments must
5975 be of :ref:`first class <t_firstclass>` type. If the function signature
5976 indicates the function accepts a variable number of arguments, the
5977 extra arguments can be specified.
5978 #. The optional :ref:`function attributes <fnattrs>` list. Only
5979 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5980 attributes are valid here.
5985 The '``call``' instruction is used to cause control flow to transfer to
5986 a specified function, with its incoming arguments bound to the specified
5987 values. Upon a '``ret``' instruction in the called function, control
5988 flow continues with the instruction after the function call, and the
5989 return value of the function is bound to the result argument.
5994 .. code-block:: llvm
5996 %retval = call i32 @test(i32 %argc)
5997 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
5998 %X = tail call i32 @foo() ; yields i32
5999 %Y = tail call fastcc i32 @foo() ; yields i32
6000 call void %foo(i8 97 signext)
6002 %struct.A = type { i32, i8 }
6003 %r = call %struct.A @foo() ; yields { 32, i8 }
6004 %gr = extractvalue %struct.A %r, 0 ; yields i32
6005 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6006 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6007 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6009 llvm treats calls to some functions with names and arguments that match
6010 the standard C99 library as being the C99 library functions, and may
6011 perform optimizations or generate code for them under that assumption.
6012 This is something we'd like to change in the future to provide better
6013 support for freestanding environments and non-C-based languages.
6017 '``va_arg``' Instruction
6018 ^^^^^^^^^^^^^^^^^^^^^^^^
6025 <resultval> = va_arg <va_list*> <arglist>, <argty>
6030 The '``va_arg``' instruction is used to access arguments passed through
6031 the "variable argument" area of a function call. It is used to implement
6032 the ``va_arg`` macro in C.
6037 This instruction takes a ``va_list*`` value and the type of the
6038 argument. It returns a value of the specified argument type and
6039 increments the ``va_list`` to point to the next argument. The actual
6040 type of ``va_list`` is target specific.
6045 The '``va_arg``' instruction loads an argument of the specified type
6046 from the specified ``va_list`` and causes the ``va_list`` to point to
6047 the next argument. For more information, see the variable argument
6048 handling :ref:`Intrinsic Functions <int_varargs>`.
6050 It is legal for this instruction to be called in a function which does
6051 not take a variable number of arguments, for example, the ``vfprintf``
6054 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6055 function <intrinsics>` because it takes a type as an argument.
6060 See the :ref:`variable argument processing <int_varargs>` section.
6062 Note that the code generator does not yet fully support va\_arg on many
6063 targets. Also, it does not currently support va\_arg with aggregate
6064 types on any target.
6068 '``landingpad``' Instruction
6069 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6076 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6077 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6079 <clause> := catch <type> <value>
6080 <clause> := filter <array constant type> <array constant>
6085 The '``landingpad``' instruction is used by `LLVM's exception handling
6086 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6087 is a landing pad --- one where the exception lands, and corresponds to the
6088 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6089 defines values supplied by the personality function (``pers_fn``) upon
6090 re-entry to the function. The ``resultval`` has the type ``resultty``.
6095 This instruction takes a ``pers_fn`` value. This is the personality
6096 function associated with the unwinding mechanism. The optional
6097 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6099 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6100 contains the global variable representing the "type" that may be caught
6101 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6102 clause takes an array constant as its argument. Use
6103 "``[0 x i8**] undef``" for a filter which cannot throw. The
6104 '``landingpad``' instruction must contain *at least* one ``clause`` or
6105 the ``cleanup`` flag.
6110 The '``landingpad``' instruction defines the values which are set by the
6111 personality function (``pers_fn``) upon re-entry to the function, and
6112 therefore the "result type" of the ``landingpad`` instruction. As with
6113 calling conventions, how the personality function results are
6114 represented in LLVM IR is target specific.
6116 The clauses are applied in order from top to bottom. If two
6117 ``landingpad`` instructions are merged together through inlining, the
6118 clauses from the calling function are appended to the list of clauses.
6119 When the call stack is being unwound due to an exception being thrown,
6120 the exception is compared against each ``clause`` in turn. If it doesn't
6121 match any of the clauses, and the ``cleanup`` flag is not set, then
6122 unwinding continues further up the call stack.
6124 The ``landingpad`` instruction has several restrictions:
6126 - A landing pad block is a basic block which is the unwind destination
6127 of an '``invoke``' instruction.
6128 - A landing pad block must have a '``landingpad``' instruction as its
6129 first non-PHI instruction.
6130 - There can be only one '``landingpad``' instruction within the landing
6132 - A basic block that is not a landing pad block may not include a
6133 '``landingpad``' instruction.
6134 - All '``landingpad``' instructions in a function must have the same
6135 personality function.
6140 .. code-block:: llvm
6142 ;; A landing pad which can catch an integer.
6143 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6145 ;; A landing pad that is a cleanup.
6146 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6148 ;; A landing pad which can catch an integer and can only throw a double.
6149 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6151 filter [1 x i8**] [@_ZTId]
6158 LLVM supports the notion of an "intrinsic function". These functions
6159 have well known names and semantics and are required to follow certain
6160 restrictions. Overall, these intrinsics represent an extension mechanism
6161 for the LLVM language that does not require changing all of the
6162 transformations in LLVM when adding to the language (or the bitcode
6163 reader/writer, the parser, etc...).
6165 Intrinsic function names must all start with an "``llvm.``" prefix. This
6166 prefix is reserved in LLVM for intrinsic names; thus, function names may
6167 not begin with this prefix. Intrinsic functions must always be external
6168 functions: you cannot define the body of intrinsic functions. Intrinsic
6169 functions may only be used in call or invoke instructions: it is illegal
6170 to take the address of an intrinsic function. Additionally, because
6171 intrinsic functions are part of the LLVM language, it is required if any
6172 are added that they be documented here.
6174 Some intrinsic functions can be overloaded, i.e., the intrinsic
6175 represents a family of functions that perform the same operation but on
6176 different data types. Because LLVM can represent over 8 million
6177 different integer types, overloading is used commonly to allow an
6178 intrinsic function to operate on any integer type. One or more of the
6179 argument types or the result type can be overloaded to accept any
6180 integer type. Argument types may also be defined as exactly matching a
6181 previous argument's type or the result type. This allows an intrinsic
6182 function which accepts multiple arguments, but needs all of them to be
6183 of the same type, to only be overloaded with respect to a single
6184 argument or the result.
6186 Overloaded intrinsics will have the names of its overloaded argument
6187 types encoded into its function name, each preceded by a period. Only
6188 those types which are overloaded result in a name suffix. Arguments
6189 whose type is matched against another type do not. For example, the
6190 ``llvm.ctpop`` function can take an integer of any width and returns an
6191 integer of exactly the same integer width. This leads to a family of
6192 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6193 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6194 overloaded, and only one type suffix is required. Because the argument's
6195 type is matched against the return type, it does not require its own
6198 To learn how to add an intrinsic function, please see the `Extending
6199 LLVM Guide <ExtendingLLVM.html>`_.
6203 Variable Argument Handling Intrinsics
6204 -------------------------------------
6206 Variable argument support is defined in LLVM with the
6207 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6208 functions. These functions are related to the similarly named macros
6209 defined in the ``<stdarg.h>`` header file.
6211 All of these functions operate on arguments that use a target-specific
6212 value type "``va_list``". The LLVM assembly language reference manual
6213 does not define what this type is, so all transformations should be
6214 prepared to handle these functions regardless of the type used.
6216 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6217 variable argument handling intrinsic functions are used.
6219 .. code-block:: llvm
6221 define i32 @test(i32 %X, ...) {
6222 ; Initialize variable argument processing
6224 %ap2 = bitcast i8** %ap to i8*
6225 call void @llvm.va_start(i8* %ap2)
6227 ; Read a single integer argument
6228 %tmp = va_arg i8** %ap, i32
6230 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6232 %aq2 = bitcast i8** %aq to i8*
6233 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6234 call void @llvm.va_end(i8* %aq2)
6236 ; Stop processing of arguments.
6237 call void @llvm.va_end(i8* %ap2)
6241 declare void @llvm.va_start(i8*)
6242 declare void @llvm.va_copy(i8*, i8*)
6243 declare void @llvm.va_end(i8*)
6247 '``llvm.va_start``' Intrinsic
6248 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6255 declare void %llvm.va_start(i8* <arglist>)
6260 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6261 subsequent use by ``va_arg``.
6266 The argument is a pointer to a ``va_list`` element to initialize.
6271 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6272 available in C. In a target-dependent way, it initializes the
6273 ``va_list`` element to which the argument points, so that the next call
6274 to ``va_arg`` will produce the first variable argument passed to the
6275 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6276 to know the last argument of the function as the compiler can figure
6279 '``llvm.va_end``' Intrinsic
6280 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6287 declare void @llvm.va_end(i8* <arglist>)
6292 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6293 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6298 The argument is a pointer to a ``va_list`` to destroy.
6303 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6304 available in C. In a target-dependent way, it destroys the ``va_list``
6305 element to which the argument points. Calls to
6306 :ref:`llvm.va_start <int_va_start>` and
6307 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6312 '``llvm.va_copy``' Intrinsic
6313 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6320 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6325 The '``llvm.va_copy``' intrinsic copies the current argument position
6326 from the source argument list to the destination argument list.
6331 The first argument is a pointer to a ``va_list`` element to initialize.
6332 The second argument is a pointer to a ``va_list`` element to copy from.
6337 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6338 available in C. In a target-dependent way, it copies the source
6339 ``va_list`` element into the destination ``va_list`` element. This
6340 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6341 arbitrarily complex and require, for example, memory allocation.
6343 Accurate Garbage Collection Intrinsics
6344 --------------------------------------
6346 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6347 (GC) requires the implementation and generation of these intrinsics.
6348 These intrinsics allow identification of :ref:`GC roots on the
6349 stack <int_gcroot>`, as well as garbage collector implementations that
6350 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6351 Front-ends for type-safe garbage collected languages should generate
6352 these intrinsics to make use of the LLVM garbage collectors. For more
6353 details, see `Accurate Garbage Collection with
6354 LLVM <GarbageCollection.html>`_.
6356 The garbage collection intrinsics only operate on objects in the generic
6357 address space (address space zero).
6361 '``llvm.gcroot``' Intrinsic
6362 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6369 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6374 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6375 the code generator, and allows some metadata to be associated with it.
6380 The first argument specifies the address of a stack object that contains
6381 the root pointer. The second pointer (which must be either a constant or
6382 a global value address) contains the meta-data to be associated with the
6388 At runtime, a call to this intrinsic stores a null pointer into the
6389 "ptrloc" location. At compile-time, the code generator generates
6390 information to allow the runtime to find the pointer at GC safe points.
6391 The '``llvm.gcroot``' intrinsic may only be used in a function which
6392 :ref:`specifies a GC algorithm <gc>`.
6396 '``llvm.gcread``' Intrinsic
6397 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6404 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6409 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6410 locations, allowing garbage collector implementations that require read
6416 The second argument is the address to read from, which should be an
6417 address allocated from the garbage collector. The first object is a
6418 pointer to the start of the referenced object, if needed by the language
6419 runtime (otherwise null).
6424 The '``llvm.gcread``' intrinsic has the same semantics as a load
6425 instruction, but may be replaced with substantially more complex code by
6426 the garbage collector runtime, as needed. The '``llvm.gcread``'
6427 intrinsic may only be used in a function which :ref:`specifies a GC
6432 '``llvm.gcwrite``' Intrinsic
6433 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6440 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6445 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6446 locations, allowing garbage collector implementations that require write
6447 barriers (such as generational or reference counting collectors).
6452 The first argument is the reference to store, the second is the start of
6453 the object to store it to, and the third is the address of the field of
6454 Obj to store to. If the runtime does not require a pointer to the
6455 object, Obj may be null.
6460 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6461 instruction, but may be replaced with substantially more complex code by
6462 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6463 intrinsic may only be used in a function which :ref:`specifies a GC
6466 Code Generator Intrinsics
6467 -------------------------
6469 These intrinsics are provided by LLVM to expose special features that
6470 may only be implemented with code generator support.
6472 '``llvm.returnaddress``' Intrinsic
6473 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6480 declare i8 *@llvm.returnaddress(i32 <level>)
6485 The '``llvm.returnaddress``' intrinsic attempts to compute a
6486 target-specific value indicating the return address of the current
6487 function or one of its callers.
6492 The argument to this intrinsic indicates which function to return the
6493 address for. Zero indicates the calling function, one indicates its
6494 caller, etc. The argument is **required** to be a constant integer
6500 The '``llvm.returnaddress``' intrinsic either returns a pointer
6501 indicating the return address of the specified call frame, or zero if it
6502 cannot be identified. The value returned by this intrinsic is likely to
6503 be incorrect or 0 for arguments other than zero, so it should only be
6504 used for debugging purposes.
6506 Note that calling this intrinsic does not prevent function inlining or
6507 other aggressive transformations, so the value returned may not be that
6508 of the obvious source-language caller.
6510 '``llvm.frameaddress``' Intrinsic
6511 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6518 declare i8* @llvm.frameaddress(i32 <level>)
6523 The '``llvm.frameaddress``' intrinsic attempts to return the
6524 target-specific frame pointer value for the specified stack frame.
6529 The argument to this intrinsic indicates which function to return the
6530 frame pointer for. Zero indicates the calling function, one indicates
6531 its caller, etc. The argument is **required** to be a constant integer
6537 The '``llvm.frameaddress``' intrinsic either returns a pointer
6538 indicating the frame address of the specified call frame, or zero if it
6539 cannot be identified. The value returned by this intrinsic is likely to
6540 be incorrect or 0 for arguments other than zero, so it should only be
6541 used for debugging purposes.
6543 Note that calling this intrinsic does not prevent function inlining or
6544 other aggressive transformations, so the value returned may not be that
6545 of the obvious source-language caller.
6549 '``llvm.stacksave``' Intrinsic
6550 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6557 declare i8* @llvm.stacksave()
6562 The '``llvm.stacksave``' intrinsic is used to remember the current state
6563 of the function stack, for use with
6564 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6565 implementing language features like scoped automatic variable sized
6571 This intrinsic returns a opaque pointer value that can be passed to
6572 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6573 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6574 ``llvm.stacksave``, it effectively restores the state of the stack to
6575 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6576 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6577 were allocated after the ``llvm.stacksave`` was executed.
6579 .. _int_stackrestore:
6581 '``llvm.stackrestore``' Intrinsic
6582 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6589 declare void @llvm.stackrestore(i8* %ptr)
6594 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6595 the function stack to the state it was in when the corresponding
6596 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6597 useful for implementing language features like scoped automatic variable
6598 sized arrays in C99.
6603 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6605 '``llvm.prefetch``' Intrinsic
6606 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6613 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6618 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6619 insert a prefetch instruction if supported; otherwise, it is a noop.
6620 Prefetches have no effect on the behavior of the program but can change
6621 its performance characteristics.
6626 ``address`` is the address to be prefetched, ``rw`` is the specifier
6627 determining if the fetch should be for a read (0) or write (1), and
6628 ``locality`` is a temporal locality specifier ranging from (0) - no
6629 locality, to (3) - extremely local keep in cache. The ``cache type``
6630 specifies whether the prefetch is performed on the data (1) or
6631 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6632 arguments must be constant integers.
6637 This intrinsic does not modify the behavior of the program. In
6638 particular, prefetches cannot trap and do not produce a value. On
6639 targets that support this intrinsic, the prefetch can provide hints to
6640 the processor cache for better performance.
6642 '``llvm.pcmarker``' Intrinsic
6643 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6650 declare void @llvm.pcmarker(i32 <id>)
6655 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6656 Counter (PC) in a region of code to simulators and other tools. The
6657 method is target specific, but it is expected that the marker will use
6658 exported symbols to transmit the PC of the marker. The marker makes no
6659 guarantees that it will remain with any specific instruction after
6660 optimizations. It is possible that the presence of a marker will inhibit
6661 optimizations. The intended use is to be inserted after optimizations to
6662 allow correlations of simulation runs.
6667 ``id`` is a numerical id identifying the marker.
6672 This intrinsic does not modify the behavior of the program. Backends
6673 that do not support this intrinsic may ignore it.
6675 '``llvm.readcyclecounter``' Intrinsic
6676 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6683 declare i64 @llvm.readcyclecounter()
6688 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6689 counter register (or similar low latency, high accuracy clocks) on those
6690 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6691 should map to RPCC. As the backing counters overflow quickly (on the
6692 order of 9 seconds on alpha), this should only be used for small
6698 When directly supported, reading the cycle counter should not modify any
6699 memory. Implementations are allowed to either return a application
6700 specific value or a system wide value. On backends without support, this
6701 is lowered to a constant 0.
6703 Note that runtime support may be conditional on the privilege-level code is
6704 running at and the host platform.
6706 Standard C Library Intrinsics
6707 -----------------------------
6709 LLVM provides intrinsics for a few important standard C library
6710 functions. These intrinsics allow source-language front-ends to pass
6711 information about the alignment of the pointer arguments to the code
6712 generator, providing opportunity for more efficient code generation.
6716 '``llvm.memcpy``' Intrinsic
6717 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6722 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6723 integer bit width and for different address spaces. Not all targets
6724 support all bit widths however.
6728 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6729 i32 <len>, i32 <align>, i1 <isvolatile>)
6730 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6731 i64 <len>, i32 <align>, i1 <isvolatile>)
6736 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6737 source location to the destination location.
6739 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6740 intrinsics do not return a value, takes extra alignment/isvolatile
6741 arguments and the pointers can be in specified address spaces.
6746 The first argument is a pointer to the destination, the second is a
6747 pointer to the source. The third argument is an integer argument
6748 specifying the number of bytes to copy, the fourth argument is the
6749 alignment of the source and destination locations, and the fifth is a
6750 boolean indicating a volatile access.
6752 If the call to this intrinsic has an alignment value that is not 0 or 1,
6753 then the caller guarantees that both the source and destination pointers
6754 are aligned to that boundary.
6756 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6757 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6758 very cleanly specified and it is unwise to depend on it.
6763 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6764 source location to the destination location, which are not allowed to
6765 overlap. It copies "len" bytes of memory over. If the argument is known
6766 to be aligned to some boundary, this can be specified as the fourth
6767 argument, otherwise it should be set to 0 or 1.
6769 '``llvm.memmove``' Intrinsic
6770 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6775 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6776 bit width and for different address space. Not all targets support all
6781 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6782 i32 <len>, i32 <align>, i1 <isvolatile>)
6783 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6784 i64 <len>, i32 <align>, i1 <isvolatile>)
6789 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6790 source location to the destination location. It is similar to the
6791 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6794 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6795 intrinsics do not return a value, takes extra alignment/isvolatile
6796 arguments and the pointers can be in specified address spaces.
6801 The first argument is a pointer to the destination, the second is a
6802 pointer to the source. The third argument is an integer argument
6803 specifying the number of bytes to copy, the fourth argument is the
6804 alignment of the source and destination locations, and the fifth is a
6805 boolean indicating a volatile access.
6807 If the call to this intrinsic has an alignment value that is not 0 or 1,
6808 then the caller guarantees that the source and destination pointers are
6809 aligned to that boundary.
6811 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6812 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6813 not very cleanly specified and it is unwise to depend on it.
6818 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6819 source location to the destination location, which may overlap. It
6820 copies "len" bytes of memory over. If the argument is known to be
6821 aligned to some boundary, this can be specified as the fourth argument,
6822 otherwise it should be set to 0 or 1.
6824 '``llvm.memset.*``' Intrinsics
6825 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6830 This is an overloaded intrinsic. You can use llvm.memset on any integer
6831 bit width and for different address spaces. However, not all targets
6832 support all bit widths.
6836 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6837 i32 <len>, i32 <align>, i1 <isvolatile>)
6838 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6839 i64 <len>, i32 <align>, i1 <isvolatile>)
6844 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6845 particular byte value.
6847 Note that, unlike the standard libc function, the ``llvm.memset``
6848 intrinsic does not return a value and takes extra alignment/volatile
6849 arguments. Also, the destination can be in an arbitrary address space.
6854 The first argument is a pointer to the destination to fill, the second
6855 is the byte value with which to fill it, the third argument is an
6856 integer argument specifying the number of bytes to fill, and the fourth
6857 argument is the known alignment of the destination location.
6859 If the call to this intrinsic has an alignment value that is not 0 or 1,
6860 then the caller guarantees that the destination pointer is aligned to
6863 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6864 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6865 very cleanly specified and it is unwise to depend on it.
6870 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6871 at the destination location. If the argument is known to be aligned to
6872 some boundary, this can be specified as the fourth argument, otherwise
6873 it should be set to 0 or 1.
6875 '``llvm.sqrt.*``' Intrinsic
6876 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6881 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6882 floating point or vector of floating point type. Not all targets support
6887 declare float @llvm.sqrt.f32(float %Val)
6888 declare double @llvm.sqrt.f64(double %Val)
6889 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6890 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6891 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6896 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
6897 returning the same value as the libm '``sqrt``' functions would. Unlike
6898 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
6899 negative numbers other than -0.0 (which allows for better optimization,
6900 because there is no need to worry about errno being set).
6901 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
6906 The argument and return value are floating point numbers of the same
6912 This function returns the sqrt of the specified operand if it is a
6913 nonnegative floating point number.
6915 '``llvm.powi.*``' Intrinsic
6916 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6921 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
6922 floating point or vector of floating point type. Not all targets support
6927 declare float @llvm.powi.f32(float %Val, i32 %power)
6928 declare double @llvm.powi.f64(double %Val, i32 %power)
6929 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6930 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6931 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6936 The '``llvm.powi.*``' intrinsics return the first operand raised to the
6937 specified (positive or negative) power. The order of evaluation of
6938 multiplications is not defined. When a vector of floating point type is
6939 used, the second argument remains a scalar integer value.
6944 The second argument is an integer power, and the first is a value to
6945 raise to that power.
6950 This function returns the first value raised to the second power with an
6951 unspecified sequence of rounding operations.
6953 '``llvm.sin.*``' Intrinsic
6954 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6959 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
6960 floating point or vector of floating point type. Not all targets support
6965 declare float @llvm.sin.f32(float %Val)
6966 declare double @llvm.sin.f64(double %Val)
6967 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6968 declare fp128 @llvm.sin.f128(fp128 %Val)
6969 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6974 The '``llvm.sin.*``' intrinsics return the sine of the operand.
6979 The argument and return value are floating point numbers of the same
6985 This function returns the sine of the specified operand, returning the
6986 same values as the libm ``sin`` functions would, and handles error
6987 conditions in the same way.
6989 '``llvm.cos.*``' Intrinsic
6990 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6995 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
6996 floating point or vector of floating point type. Not all targets support
7001 declare float @llvm.cos.f32(float %Val)
7002 declare double @llvm.cos.f64(double %Val)
7003 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7004 declare fp128 @llvm.cos.f128(fp128 %Val)
7005 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7010 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7015 The argument and return value are floating point numbers of the same
7021 This function returns the cosine of the specified operand, returning the
7022 same values as the libm ``cos`` functions would, and handles error
7023 conditions in the same way.
7025 '``llvm.pow.*``' Intrinsic
7026 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7031 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7032 floating point or vector of floating point type. Not all targets support
7037 declare float @llvm.pow.f32(float %Val, float %Power)
7038 declare double @llvm.pow.f64(double %Val, double %Power)
7039 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7040 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7041 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7046 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7047 specified (positive or negative) power.
7052 The second argument is a floating point power, and the first is a value
7053 to raise to that power.
7058 This function returns the first value raised to the second power,
7059 returning the same values as the libm ``pow`` functions would, and
7060 handles error conditions in the same way.
7062 '``llvm.exp.*``' Intrinsic
7063 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7068 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7069 floating point or vector of floating point type. Not all targets support
7074 declare float @llvm.exp.f32(float %Val)
7075 declare double @llvm.exp.f64(double %Val)
7076 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7077 declare fp128 @llvm.exp.f128(fp128 %Val)
7078 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7083 The '``llvm.exp.*``' intrinsics perform the exp function.
7088 The argument and return value are floating point numbers of the same
7094 This function returns the same values as the libm ``exp`` functions
7095 would, and handles error conditions in the same way.
7097 '``llvm.exp2.*``' Intrinsic
7098 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7103 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7104 floating point or vector of floating point type. Not all targets support
7109 declare float @llvm.exp2.f32(float %Val)
7110 declare double @llvm.exp2.f64(double %Val)
7111 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7112 declare fp128 @llvm.exp2.f128(fp128 %Val)
7113 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7118 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7123 The argument and return value are floating point numbers of the same
7129 This function returns the same values as the libm ``exp2`` functions
7130 would, and handles error conditions in the same way.
7132 '``llvm.log.*``' Intrinsic
7133 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7138 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7139 floating point or vector of floating point type. Not all targets support
7144 declare float @llvm.log.f32(float %Val)
7145 declare double @llvm.log.f64(double %Val)
7146 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7147 declare fp128 @llvm.log.f128(fp128 %Val)
7148 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7153 The '``llvm.log.*``' intrinsics perform the log function.
7158 The argument and return value are floating point numbers of the same
7164 This function returns the same values as the libm ``log`` functions
7165 would, and handles error conditions in the same way.
7167 '``llvm.log10.*``' Intrinsic
7168 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7173 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7174 floating point or vector of floating point type. Not all targets support
7179 declare float @llvm.log10.f32(float %Val)
7180 declare double @llvm.log10.f64(double %Val)
7181 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7182 declare fp128 @llvm.log10.f128(fp128 %Val)
7183 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7188 The '``llvm.log10.*``' intrinsics perform the log10 function.
7193 The argument and return value are floating point numbers of the same
7199 This function returns the same values as the libm ``log10`` functions
7200 would, and handles error conditions in the same way.
7202 '``llvm.log2.*``' Intrinsic
7203 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7208 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7209 floating point or vector of floating point type. Not all targets support
7214 declare float @llvm.log2.f32(float %Val)
7215 declare double @llvm.log2.f64(double %Val)
7216 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7217 declare fp128 @llvm.log2.f128(fp128 %Val)
7218 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7223 The '``llvm.log2.*``' intrinsics perform the log2 function.
7228 The argument and return value are floating point numbers of the same
7234 This function returns the same values as the libm ``log2`` functions
7235 would, and handles error conditions in the same way.
7237 '``llvm.fma.*``' Intrinsic
7238 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7243 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7244 floating point or vector of floating point type. Not all targets support
7249 declare float @llvm.fma.f32(float %a, float %b, float %c)
7250 declare double @llvm.fma.f64(double %a, double %b, double %c)
7251 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7252 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7253 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7258 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7264 The argument and return value are floating point numbers of the same
7270 This function returns the same values as the libm ``fma`` functions
7273 '``llvm.fabs.*``' Intrinsic
7274 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7279 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7280 floating point or vector of floating point type. Not all targets support
7285 declare float @llvm.fabs.f32(float %Val)
7286 declare double @llvm.fabs.f64(double %Val)
7287 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7288 declare fp128 @llvm.fabs.f128(fp128 %Val)
7289 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7294 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7300 The argument and return value are floating point numbers of the same
7306 This function returns the same values as the libm ``fabs`` functions
7307 would, and handles error conditions in the same way.
7309 '``llvm.floor.*``' Intrinsic
7310 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7315 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7316 floating point or vector of floating point type. Not all targets support
7321 declare float @llvm.floor.f32(float %Val)
7322 declare double @llvm.floor.f64(double %Val)
7323 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7324 declare fp128 @llvm.floor.f128(fp128 %Val)
7325 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7330 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7335 The argument and return value are floating point numbers of the same
7341 This function returns the same values as the libm ``floor`` functions
7342 would, and handles error conditions in the same way.
7344 '``llvm.ceil.*``' Intrinsic
7345 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7350 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7351 floating point or vector of floating point type. Not all targets support
7356 declare float @llvm.ceil.f32(float %Val)
7357 declare double @llvm.ceil.f64(double %Val)
7358 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7359 declare fp128 @llvm.ceil.f128(fp128 %Val)
7360 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7365 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7370 The argument and return value are floating point numbers of the same
7376 This function returns the same values as the libm ``ceil`` functions
7377 would, and handles error conditions in the same way.
7379 '``llvm.trunc.*``' Intrinsic
7380 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7385 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7386 floating point or vector of floating point type. Not all targets support
7391 declare float @llvm.trunc.f32(float %Val)
7392 declare double @llvm.trunc.f64(double %Val)
7393 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7394 declare fp128 @llvm.trunc.f128(fp128 %Val)
7395 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7400 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7401 nearest integer not larger in magnitude than the operand.
7406 The argument and return value are floating point numbers of the same
7412 This function returns the same values as the libm ``trunc`` functions
7413 would, and handles error conditions in the same way.
7415 '``llvm.rint.*``' Intrinsic
7416 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7421 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7422 floating point or vector of floating point type. Not all targets support
7427 declare float @llvm.rint.f32(float %Val)
7428 declare double @llvm.rint.f64(double %Val)
7429 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7430 declare fp128 @llvm.rint.f128(fp128 %Val)
7431 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7436 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7437 nearest integer. It may raise an inexact floating-point exception if the
7438 operand isn't an integer.
7443 The argument and return value are floating point numbers of the same
7449 This function returns the same values as the libm ``rint`` functions
7450 would, and handles error conditions in the same way.
7452 '``llvm.nearbyint.*``' Intrinsic
7453 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7458 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7459 floating point or vector of floating point type. Not all targets support
7464 declare float @llvm.nearbyint.f32(float %Val)
7465 declare double @llvm.nearbyint.f64(double %Val)
7466 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7467 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7468 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7473 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7479 The argument and return value are floating point numbers of the same
7485 This function returns the same values as the libm ``nearbyint``
7486 functions would, and handles error conditions in the same way.
7488 Bit Manipulation Intrinsics
7489 ---------------------------
7491 LLVM provides intrinsics for a few important bit manipulation
7492 operations. These allow efficient code generation for some algorithms.
7494 '``llvm.bswap.*``' Intrinsics
7495 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7500 This is an overloaded intrinsic function. You can use bswap on any
7501 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7505 declare i16 @llvm.bswap.i16(i16 <id>)
7506 declare i32 @llvm.bswap.i32(i32 <id>)
7507 declare i64 @llvm.bswap.i64(i64 <id>)
7512 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7513 values with an even number of bytes (positive multiple of 16 bits).
7514 These are useful for performing operations on data that is not in the
7515 target's native byte order.
7520 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7521 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7522 intrinsic returns an i32 value that has the four bytes of the input i32
7523 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7524 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7525 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7526 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7529 '``llvm.ctpop.*``' Intrinsic
7530 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7535 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7536 bit width, or on any vector with integer elements. Not all targets
7537 support all bit widths or vector types, however.
7541 declare i8 @llvm.ctpop.i8(i8 <src>)
7542 declare i16 @llvm.ctpop.i16(i16 <src>)
7543 declare i32 @llvm.ctpop.i32(i32 <src>)
7544 declare i64 @llvm.ctpop.i64(i64 <src>)
7545 declare i256 @llvm.ctpop.i256(i256 <src>)
7546 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7551 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7557 The only argument is the value to be counted. The argument may be of any
7558 integer type, or a vector with integer elements. The return type must
7559 match the argument type.
7564 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7565 each element of a vector.
7567 '``llvm.ctlz.*``' Intrinsic
7568 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7573 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7574 integer bit width, or any vector whose elements are integers. Not all
7575 targets support all bit widths or vector types, however.
7579 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7580 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7581 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7582 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7583 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7584 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7589 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7590 leading zeros in a variable.
7595 The first argument is the value to be counted. This argument may be of
7596 any integer type, or a vectory with integer element type. The return
7597 type must match the first argument type.
7599 The second argument must be a constant and is a flag to indicate whether
7600 the intrinsic should ensure that a zero as the first argument produces a
7601 defined result. Historically some architectures did not provide a
7602 defined result for zero values as efficiently, and many algorithms are
7603 now predicated on avoiding zero-value inputs.
7608 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7609 zeros in a variable, or within each element of the vector. If
7610 ``src == 0`` then the result is the size in bits of the type of ``src``
7611 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7612 ``llvm.ctlz(i32 2) = 30``.
7614 '``llvm.cttz.*``' Intrinsic
7615 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7620 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7621 integer bit width, or any vector of integer elements. Not all targets
7622 support all bit widths or vector types, however.
7626 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7627 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7628 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7629 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7630 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7631 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7636 The '``llvm.cttz``' family of intrinsic functions counts the number of
7642 The first argument is the value to be counted. This argument may be of
7643 any integer type, or a vectory with integer element type. The return
7644 type must match the first argument type.
7646 The second argument must be a constant and is a flag to indicate whether
7647 the intrinsic should ensure that a zero as the first argument produces a
7648 defined result. Historically some architectures did not provide a
7649 defined result for zero values as efficiently, and many algorithms are
7650 now predicated on avoiding zero-value inputs.
7655 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7656 zeros in a variable, or within each element of a vector. If ``src == 0``
7657 then the result is the size in bits of the type of ``src`` if
7658 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7659 ``llvm.cttz(2) = 1``.
7661 Arithmetic with Overflow Intrinsics
7662 -----------------------------------
7664 LLVM provides intrinsics for some arithmetic with overflow operations.
7666 '``llvm.sadd.with.overflow.*``' Intrinsics
7667 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7672 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7673 on any integer bit width.
7677 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7678 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7679 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7684 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7685 a signed addition of the two arguments, and indicate whether an overflow
7686 occurred during the signed summation.
7691 The arguments (%a and %b) and the first element of the result structure
7692 may be of integer types of any bit width, but they must have the same
7693 bit width. The second element of the result structure must be of type
7694 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7700 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7701 a signed addition of the two variables. They return a structure --- the
7702 first element of which is the signed summation, and the second element
7703 of which is a bit specifying if the signed summation resulted in an
7709 .. code-block:: llvm
7711 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7712 %sum = extractvalue {i32, i1} %res, 0
7713 %obit = extractvalue {i32, i1} %res, 1
7714 br i1 %obit, label %overflow, label %normal
7716 '``llvm.uadd.with.overflow.*``' Intrinsics
7717 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7722 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7723 on any integer bit width.
7727 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7728 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7729 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7734 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7735 an unsigned addition of the two arguments, and indicate whether a carry
7736 occurred during the unsigned summation.
7741 The arguments (%a and %b) and the first element of the result structure
7742 may be of integer types of any bit width, but they must have the same
7743 bit width. The second element of the result structure must be of type
7744 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7750 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7751 an unsigned addition of the two arguments. They return a structure --- the
7752 first element of which is the sum, and the second element of which is a
7753 bit specifying if the unsigned summation resulted in a carry.
7758 .. code-block:: llvm
7760 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7761 %sum = extractvalue {i32, i1} %res, 0
7762 %obit = extractvalue {i32, i1} %res, 1
7763 br i1 %obit, label %carry, label %normal
7765 '``llvm.ssub.with.overflow.*``' Intrinsics
7766 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7771 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7772 on any integer bit width.
7776 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7777 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7778 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7783 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7784 a signed subtraction of the two arguments, and indicate whether an
7785 overflow occurred during the signed subtraction.
7790 The arguments (%a and %b) and the first element of the result structure
7791 may be of integer types of any bit width, but they must have the same
7792 bit width. The second element of the result structure must be of type
7793 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7799 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7800 a signed subtraction of the two arguments. They return a structure --- the
7801 first element of which is the subtraction, and the second element of
7802 which is a bit specifying if the signed subtraction resulted in an
7808 .. code-block:: llvm
7810 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7811 %sum = extractvalue {i32, i1} %res, 0
7812 %obit = extractvalue {i32, i1} %res, 1
7813 br i1 %obit, label %overflow, label %normal
7815 '``llvm.usub.with.overflow.*``' Intrinsics
7816 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7821 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
7822 on any integer bit width.
7826 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7827 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7828 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7833 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7834 an unsigned subtraction of the two arguments, and indicate whether an
7835 overflow occurred during the unsigned subtraction.
7840 The arguments (%a and %b) and the first element of the result structure
7841 may be of integer types of any bit width, but they must have the same
7842 bit width. The second element of the result structure must be of type
7843 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7849 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7850 an unsigned subtraction of the two arguments. They return a structure ---
7851 the first element of which is the subtraction, and the second element of
7852 which is a bit specifying if the unsigned subtraction resulted in an
7858 .. code-block:: llvm
7860 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7861 %sum = extractvalue {i32, i1} %res, 0
7862 %obit = extractvalue {i32, i1} %res, 1
7863 br i1 %obit, label %overflow, label %normal
7865 '``llvm.smul.with.overflow.*``' Intrinsics
7866 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7871 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
7872 on any integer bit width.
7876 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7877 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7878 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7883 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7884 a signed multiplication of the two arguments, and indicate whether an
7885 overflow occurred during the signed multiplication.
7890 The arguments (%a and %b) and the first element of the result structure
7891 may be of integer types of any bit width, but they must have the same
7892 bit width. The second element of the result structure must be of type
7893 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7899 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7900 a signed multiplication of the two arguments. They return a structure ---
7901 the first element of which is the multiplication, and the second element
7902 of which is a bit specifying if the signed multiplication resulted in an
7908 .. code-block:: llvm
7910 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7911 %sum = extractvalue {i32, i1} %res, 0
7912 %obit = extractvalue {i32, i1} %res, 1
7913 br i1 %obit, label %overflow, label %normal
7915 '``llvm.umul.with.overflow.*``' Intrinsics
7916 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7921 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
7922 on any integer bit width.
7926 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7927 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7928 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7933 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7934 a unsigned multiplication of the two arguments, and indicate whether an
7935 overflow occurred during the unsigned multiplication.
7940 The arguments (%a and %b) and the first element of the result structure
7941 may be of integer types of any bit width, but they must have the same
7942 bit width. The second element of the result structure must be of type
7943 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7949 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7950 an unsigned multiplication of the two arguments. They return a structure ---
7951 the first element of which is the multiplication, and the second
7952 element of which is a bit specifying if the unsigned multiplication
7953 resulted in an overflow.
7958 .. code-block:: llvm
7960 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7961 %sum = extractvalue {i32, i1} %res, 0
7962 %obit = extractvalue {i32, i1} %res, 1
7963 br i1 %obit, label %overflow, label %normal
7965 Specialised Arithmetic Intrinsics
7966 ---------------------------------
7968 '``llvm.fmuladd.*``' Intrinsic
7969 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7976 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
7977 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
7982 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
7983 expressions that can be fused if the code generator determines that (a) the
7984 target instruction set has support for a fused operation, and (b) that the
7985 fused operation is more efficient than the equivalent, separate pair of mul
7986 and add instructions.
7991 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
7992 multiplicands, a and b, and an addend c.
8001 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8003 is equivalent to the expression a \* b + c, except that rounding will
8004 not be performed between the multiplication and addition steps if the
8005 code generator fuses the operations. Fusion is not guaranteed, even if
8006 the target platform supports it. If a fused multiply-add is required the
8007 corresponding llvm.fma.\* intrinsic function should be used instead.
8012 .. code-block:: llvm
8014 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8016 Half Precision Floating Point Intrinsics
8017 ----------------------------------------
8019 For most target platforms, half precision floating point is a
8020 storage-only format. This means that it is a dense encoding (in memory)
8021 but does not support computation in the format.
8023 This means that code must first load the half-precision floating point
8024 value as an i16, then convert it to float with
8025 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8026 then be performed on the float value (including extending to double
8027 etc). To store the value back to memory, it is first converted to float
8028 if needed, then converted to i16 with
8029 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8032 .. _int_convert_to_fp16:
8034 '``llvm.convert.to.fp16``' Intrinsic
8035 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8042 declare i16 @llvm.convert.to.fp16(f32 %a)
8047 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8048 from single precision floating point format to half precision floating
8054 The intrinsic function contains single argument - the value to be
8060 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8061 from single precision floating point format to half precision floating
8062 point format. The return value is an ``i16`` which contains the
8068 .. code-block:: llvm
8070 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8071 store i16 %res, i16* @x, align 2
8073 .. _int_convert_from_fp16:
8075 '``llvm.convert.from.fp16``' Intrinsic
8076 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8083 declare f32 @llvm.convert.from.fp16(i16 %a)
8088 The '``llvm.convert.from.fp16``' intrinsic function performs a
8089 conversion from half precision floating point format to single precision
8090 floating point format.
8095 The intrinsic function contains single argument - the value to be
8101 The '``llvm.convert.from.fp16``' intrinsic function performs a
8102 conversion from half single precision floating point format to single
8103 precision floating point format. The input half-float value is
8104 represented by an ``i16`` value.
8109 .. code-block:: llvm
8111 %a = load i16* @x, align 2
8112 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8117 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8118 prefix), are described in the `LLVM Source Level
8119 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8122 Exception Handling Intrinsics
8123 -----------------------------
8125 The LLVM exception handling intrinsics (which all start with
8126 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8127 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8131 Trampoline Intrinsics
8132 ---------------------
8134 These intrinsics make it possible to excise one parameter, marked with
8135 the :ref:`nest <nest>` attribute, from a function. The result is a
8136 callable function pointer lacking the nest parameter - the caller does
8137 not need to provide a value for it. Instead, the value to use is stored
8138 in advance in a "trampoline", a block of memory usually allocated on the
8139 stack, which also contains code to splice the nest value into the
8140 argument list. This is used to implement the GCC nested function address
8143 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8144 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8145 It can be created as follows:
8147 .. code-block:: llvm
8149 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8150 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8151 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8152 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8153 %fp = bitcast i8* %p to i32 (i32, i32)*
8155 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8156 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8160 '``llvm.init.trampoline``' Intrinsic
8161 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8168 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8173 This fills the memory pointed to by ``tramp`` with executable code,
8174 turning it into a trampoline.
8179 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8180 pointers. The ``tramp`` argument must point to a sufficiently large and
8181 sufficiently aligned block of memory; this memory is written to by the
8182 intrinsic. Note that the size and the alignment are target-specific -
8183 LLVM currently provides no portable way of determining them, so a
8184 front-end that generates this intrinsic needs to have some
8185 target-specific knowledge. The ``func`` argument must hold a function
8186 bitcast to an ``i8*``.
8191 The block of memory pointed to by ``tramp`` is filled with target
8192 dependent code, turning it into a function. Then ``tramp`` needs to be
8193 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8194 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8195 function's signature is the same as that of ``func`` with any arguments
8196 marked with the ``nest`` attribute removed. At most one such ``nest``
8197 argument is allowed, and it must be of pointer type. Calling the new
8198 function is equivalent to calling ``func`` with the same argument list,
8199 but with ``nval`` used for the missing ``nest`` argument. If, after
8200 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8201 modified, then the effect of any later call to the returned function
8202 pointer is undefined.
8206 '``llvm.adjust.trampoline``' Intrinsic
8207 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8214 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8219 This performs any required machine-specific adjustment to the address of
8220 a trampoline (passed as ``tramp``).
8225 ``tramp`` must point to a block of memory which already has trampoline
8226 code filled in by a previous call to
8227 :ref:`llvm.init.trampoline <int_it>`.
8232 On some architectures the address of the code to be executed needs to be
8233 different to the address where the trampoline is actually stored. This
8234 intrinsic returns the executable address corresponding to ``tramp``
8235 after performing the required machine specific adjustments. The pointer
8236 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8241 This class of intrinsics exists to information about the lifetime of
8242 memory objects and ranges where variables are immutable.
8244 '``llvm.lifetime.start``' Intrinsic
8245 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8252 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8257 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8263 The first argument is a constant integer representing the size of the
8264 object, or -1 if it is variable sized. The second argument is a pointer
8270 This intrinsic indicates that before this point in the code, the value
8271 of the memory pointed to by ``ptr`` is dead. This means that it is known
8272 to never be used and has an undefined value. A load from the pointer
8273 that precedes this intrinsic can be replaced with ``'undef'``.
8275 '``llvm.lifetime.end``' Intrinsic
8276 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8283 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8288 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8294 The first argument is a constant integer representing the size of the
8295 object, or -1 if it is variable sized. The second argument is a pointer
8301 This intrinsic indicates that after this point in the code, the value of
8302 the memory pointed to by ``ptr`` is dead. This means that it is known to
8303 never be used and has an undefined value. Any stores into the memory
8304 object following this intrinsic may be removed as dead.
8306 '``llvm.invariant.start``' Intrinsic
8307 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8314 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8319 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8320 a memory object will not change.
8325 The first argument is a constant integer representing the size of the
8326 object, or -1 if it is variable sized. The second argument is a pointer
8332 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8333 the return value, the referenced memory location is constant and
8336 '``llvm.invariant.end``' Intrinsic
8337 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8344 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8349 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8350 memory object are mutable.
8355 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8356 The second argument is a constant integer representing the size of the
8357 object, or -1 if it is variable sized and the third argument is a
8358 pointer to the object.
8363 This intrinsic indicates that the memory is mutable again.
8368 This class of intrinsics is designed to be generic and has no specific
8371 '``llvm.var.annotation``' Intrinsic
8372 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8379 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8384 The '``llvm.var.annotation``' intrinsic.
8389 The first argument is a pointer to a value, the second is a pointer to a
8390 global string, the third is a pointer to a global string which is the
8391 source file name, and the last argument is the line number.
8396 This intrinsic allows annotation of local variables with arbitrary
8397 strings. This can be useful for special purpose optimizations that want
8398 to look for these annotations. These have no other defined use; they are
8399 ignored by code generation and optimization.
8401 '``llvm.ptr.annotation.*``' Intrinsic
8402 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8407 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8408 pointer to an integer of any width. *NOTE* you must specify an address space for
8409 the pointer. The identifier for the default address space is the integer
8414 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8415 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8416 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8417 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8418 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8423 The '``llvm.ptr.annotation``' intrinsic.
8428 The first argument is a pointer to an integer value of arbitrary bitwidth
8429 (result of some expression), the second is a pointer to a global string, the
8430 third is a pointer to a global string which is the source file name, and the
8431 last argument is the line number. It returns the value of the first argument.
8436 This intrinsic allows annotation of a pointer to an integer with arbitrary
8437 strings. This can be useful for special purpose optimizations that want to look
8438 for these annotations. These have no other defined use; they are ignored by code
8439 generation and optimization.
8441 '``llvm.annotation.*``' Intrinsic
8442 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8447 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8448 any integer bit width.
8452 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8453 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8454 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8455 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8456 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8461 The '``llvm.annotation``' intrinsic.
8466 The first argument is an integer value (result of some expression), the
8467 second is a pointer to a global string, the third is a pointer to a
8468 global string which is the source file name, and the last argument is
8469 the line number. It returns the value of the first argument.
8474 This intrinsic allows annotations to be put on arbitrary expressions
8475 with arbitrary strings. This can be useful for special purpose
8476 optimizations that want to look for these annotations. These have no
8477 other defined use; they are ignored by code generation and optimization.
8479 '``llvm.trap``' Intrinsic
8480 ^^^^^^^^^^^^^^^^^^^^^^^^^
8487 declare void @llvm.trap() noreturn nounwind
8492 The '``llvm.trap``' intrinsic.
8502 This intrinsic is lowered to the target dependent trap instruction. If
8503 the target does not have a trap instruction, this intrinsic will be
8504 lowered to a call of the ``abort()`` function.
8506 '``llvm.debugtrap``' Intrinsic
8507 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8514 declare void @llvm.debugtrap() nounwind
8519 The '``llvm.debugtrap``' intrinsic.
8529 This intrinsic is lowered to code which is intended to cause an
8530 execution trap with the intention of requesting the attention of a
8533 '``llvm.stackprotector``' Intrinsic
8534 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8541 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8546 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8547 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8548 is placed on the stack before local variables.
8553 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8554 The first argument is the value loaded from the stack guard
8555 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8556 enough space to hold the value of the guard.
8561 This intrinsic causes the prologue/epilogue inserter to force the
8562 position of the ``AllocaInst`` stack slot to be before local variables
8563 on the stack. This is to ensure that if a local variable on the stack is
8564 overwritten, it will destroy the value of the guard. When the function
8565 exits, the guard on the stack is checked against the original guard. If
8566 they are different, then the program aborts by calling the
8567 ``__stack_chk_fail()`` function.
8569 '``llvm.objectsize``' Intrinsic
8570 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8577 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8578 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8583 The ``llvm.objectsize`` intrinsic is designed to provide information to
8584 the optimizers to determine at compile time whether a) an operation
8585 (like memcpy) will overflow a buffer that corresponds to an object, or
8586 b) that a runtime check for overflow isn't necessary. An object in this
8587 context means an allocation of a specific class, structure, array, or
8593 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8594 argument is a pointer to or into the ``object``. The second argument is
8595 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8596 or -1 (if false) when the object size is unknown. The second argument
8597 only accepts constants.
8602 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8603 the size of the object concerned. If the size cannot be determined at
8604 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8605 on the ``min`` argument).
8607 '``llvm.expect``' Intrinsic
8608 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8615 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8616 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8621 The ``llvm.expect`` intrinsic provides information about expected (the
8622 most probable) value of ``val``, which can be used by optimizers.
8627 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8628 a value. The second argument is an expected value, this needs to be a
8629 constant value, variables are not allowed.
8634 This intrinsic is lowered to the ``val``.
8636 '``llvm.donothing``' Intrinsic
8637 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8644 declare void @llvm.donothing() nounwind readnone
8649 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8650 only intrinsic that can be called with an invoke instruction.
8660 This intrinsic does nothing, and it's removed by optimizers and ignored