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
390 .. _visibilitystyles:
395 All Global Variables and Functions have one of the following visibility
398 "``default``" - Default style
399 On targets that use the ELF object file format, default visibility
400 means that the declaration is visible to other modules and, in
401 shared libraries, means that the declared entity may be overridden.
402 On Darwin, default visibility means that the declaration is visible
403 to other modules. Default visibility corresponds to "external
404 linkage" in the language.
405 "``hidden``" - Hidden style
406 Two declarations of an object with hidden visibility refer to the
407 same object if they are in the same shared object. Usually, hidden
408 visibility indicates that the symbol will not be placed into the
409 dynamic symbol table, so no other module (executable or shared
410 library) can reference it directly.
411 "``protected``" - Protected style
412 On ELF, protected visibility indicates that the symbol will be
413 placed in the dynamic symbol table, but that references within the
414 defining module will bind to the local symbol. That is, the symbol
415 cannot be overridden by another module.
422 LLVM IR allows you to specify name aliases for certain types. This can
423 make it easier to read the IR and make the IR more condensed
424 (particularly when recursive types are involved). An example of a name
429 %mytype = type { %mytype*, i32 }
431 You may give a name to any :ref:`type <typesystem>` except
432 ":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
433 expected with the syntax "%mytype".
435 Note that type names are aliases for the structural type that they
436 indicate, and that you can therefore specify multiple names for the same
437 type. This often leads to confusing behavior when dumping out a .ll
438 file. Since LLVM IR uses structural typing, the name is not part of the
439 type. When printing out LLVM IR, the printer will pick *one name* to
440 render all types of a particular shape. This means that if you have code
441 where two different source types end up having the same LLVM type, that
442 the dumper will sometimes print the "wrong" or unexpected type. This is
443 an important design point and isn't going to change.
450 Global variables define regions of memory allocated at compilation time
451 instead of run-time. Global variables may optionally be initialized, may
452 have an explicit section to be placed in, and may have an optional
453 explicit alignment specified.
455 A variable may be defined as ``thread_local``, which means that it will
456 not be shared by threads (each thread will have a separated copy of the
457 variable). Not all targets support thread-local variables. Optionally, a
458 TLS model may be specified:
461 For variables that are only used within the current shared library.
463 For variables in modules that will not be loaded dynamically.
465 For variables defined in the executable and only used within it.
467 The models correspond to the ELF TLS models; see `ELF Handling For
468 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
469 more information on under which circumstances the different models may
470 be used. The target may choose a different TLS model if the specified
471 model is not supported, or if a better choice of model can be made.
473 A variable may be defined as a global ``constant``, which indicates that
474 the contents of the variable will **never** be modified (enabling better
475 optimization, allowing the global data to be placed in the read-only
476 section of an executable, etc). Note that variables that need runtime
477 initialization cannot be marked ``constant`` as there is a store to the
480 LLVM explicitly allows *declarations* of global variables to be marked
481 constant, even if the final definition of the global is not. This
482 capability can be used to enable slightly better optimization of the
483 program, but requires the language definition to guarantee that
484 optimizations based on the 'constantness' are valid for the translation
485 units that do not include the definition.
487 As SSA values, global variables define pointer values that are in scope
488 (i.e. they dominate) all basic blocks in the program. Global variables
489 always define a pointer to their "content" type because they describe a
490 region of memory, and all memory objects in LLVM are accessed through
493 Global variables can be marked with ``unnamed_addr`` which indicates
494 that the address is not significant, only the content. Constants marked
495 like this can be merged with other constants if they have the same
496 initializer. Note that a constant with significant address *can* be
497 merged with a ``unnamed_addr`` constant, the result being a constant
498 whose address is significant.
500 A global variable may be declared to reside in a target-specific
501 numbered address space. For targets that support them, address spaces
502 may affect how optimizations are performed and/or what target
503 instructions are used to access the variable. The default address space
504 is zero. The address space qualifier must precede any other attributes.
506 LLVM allows an explicit section to be specified for globals. If the
507 target supports it, it will emit globals to the section specified.
509 By default, global initializers are optimized by assuming that global
510 variables defined within the module are not modified from their
511 initial values before the start of the global initializer. This is
512 true even for variables potentially accessible from outside the
513 module, including those with external linkage or appearing in
514 ``@llvm.used``. This assumption may be suppressed by marking the
515 variable with ``externally_initialized``.
517 An explicit alignment may be specified for a global, which must be a
518 power of 2. If not present, or if the alignment is set to zero, the
519 alignment of the global is set by the target to whatever it feels
520 convenient. If an explicit alignment is specified, the global is forced
521 to have exactly that alignment. Targets and optimizers are not allowed
522 to over-align the global if the global has an assigned section. In this
523 case, the extra alignment could be observable: for example, code could
524 assume that the globals are densely packed in their section and try to
525 iterate over them as an array, alignment padding would break this
528 For example, the following defines a global in a numbered address space
529 with an initializer, section, and alignment:
533 @G = addrspace(5) constant float 1.0, section "foo", align 4
535 The following example defines a thread-local global with the
536 ``initialexec`` TLS model:
540 @G = thread_local(initialexec) global i32 0, align 4
542 .. _functionstructure:
547 LLVM function definitions consist of the "``define``" keyword, an
548 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
549 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
550 an optional ``unnamed_addr`` attribute, a return type, an optional
551 :ref:`parameter attribute <paramattrs>` for the return type, a function
552 name, a (possibly empty) argument list (each with optional :ref:`parameter
553 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
554 an optional section, an optional alignment, an optional :ref:`garbage
555 collector name <gc>`, an opening curly brace, a list of basic blocks,
556 and a closing curly brace.
558 LLVM function declarations consist of the "``declare``" keyword, an
559 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
560 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
561 an optional ``unnamed_addr`` attribute, a return type, an optional
562 :ref:`parameter attribute <paramattrs>` for the return type, a function
563 name, a possibly empty list of arguments, an optional alignment, and an
564 optional :ref:`garbage collector name <gc>`.
566 A function definition contains a list of basic blocks, forming the CFG
567 (Control Flow Graph) for the function. Each basic block may optionally
568 start with a label (giving the basic block a symbol table entry),
569 contains a list of instructions, and ends with a
570 :ref:`terminator <terminators>` instruction (such as a branch or function
571 return). If explicit label is not provided, a block is assigned an
572 implicit numbered label, using a next value from the same counter as used
573 for unnamed temporaries (:ref:`see above<identifiers>`). For example, if a
574 function entry block does not have explicit label, it will be assigned
575 label "%0", then first unnamed temporary in that block will be "%1", etc.
577 The first basic block in a function is special in two ways: it is
578 immediately executed on entrance to the function, and it is not allowed
579 to have predecessor basic blocks (i.e. there can not be any branches to
580 the entry block of a function). Because the block can have no
581 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
583 LLVM allows an explicit section to be specified for functions. If the
584 target supports it, it will emit functions to the section specified.
586 An explicit alignment may be specified for a function. If not present,
587 or if the alignment is set to zero, the alignment of the function is set
588 by the target to whatever it feels convenient. If an explicit alignment
589 is specified, the function is forced to have at least that much
590 alignment. All alignments must be a power of 2.
592 If the ``unnamed_addr`` attribute is given, the address is know to not
593 be significant and two identical functions can be merged.
597 define [linkage] [visibility]
599 <ResultType> @<FunctionName> ([argument list])
600 [fn Attrs] [section "name"] [align N]
608 Aliases act as "second name" for the aliasee value (which can be either
609 function, global variable, another alias or bitcast of global value).
610 Aliases may have an optional :ref:`linkage type <linkage>`, and an optional
611 :ref:`visibility style <visibility>`.
615 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
617 .. _namedmetadatastructure:
622 Named metadata is a collection of metadata. :ref:`Metadata
623 nodes <metadata>` (but not metadata strings) are the only valid
624 operands for a named metadata.
628 ; Some unnamed metadata nodes, which are referenced by the named metadata.
629 !0 = metadata !{metadata !"zero"}
630 !1 = metadata !{metadata !"one"}
631 !2 = metadata !{metadata !"two"}
633 !name = !{!0, !1, !2}
640 The return type and each parameter of a function type may have a set of
641 *parameter attributes* associated with them. Parameter attributes are
642 used to communicate additional information about the result or
643 parameters of a function. Parameter attributes are considered to be part
644 of the function, not of the function type, so functions with different
645 parameter attributes can have the same function type.
647 Parameter attributes are simple keywords that follow the type specified.
648 If multiple parameter attributes are needed, they are space separated.
653 declare i32 @printf(i8* noalias nocapture, ...)
654 declare i32 @atoi(i8 zeroext)
655 declare signext i8 @returns_signed_char()
657 Note that any attributes for the function result (``nounwind``,
658 ``readonly``) come immediately after the argument list.
660 Currently, only the following parameter attributes are defined:
663 This indicates to the code generator that the parameter or return
664 value should be zero-extended to the extent required by the target's
665 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
666 the caller (for a parameter) or the callee (for a return value).
668 This indicates to the code generator that the parameter or return
669 value should be sign-extended to the extent required by the target's
670 ABI (which is usually 32-bits) by the caller (for a parameter) or
671 the callee (for a return value).
673 This indicates that this parameter or return value should be treated
674 in a special target-dependent fashion during while emitting code for
675 a function call or return (usually, by putting it in a register as
676 opposed to memory, though some targets use it to distinguish between
677 two different kinds of registers). Use of this attribute is
680 This indicates that the pointer parameter should really be passed by
681 value to the function. The attribute implies that a hidden copy of
682 the pointee is made between the caller and the callee, so the callee
683 is unable to modify the value in the caller. This attribute is only
684 valid on LLVM pointer arguments. It is generally used to pass
685 structs and arrays by value, but is also valid on pointers to
686 scalars. The copy is considered to belong to the caller not the
687 callee (for example, ``readonly`` functions should not write to
688 ``byval`` parameters). This is not a valid attribute for return
691 The byval attribute also supports specifying an alignment with the
692 align attribute. It indicates the alignment of the stack slot to
693 form and the known alignment of the pointer specified to the call
694 site. If the alignment is not specified, then the code generator
695 makes a target-specific assumption.
698 This indicates that the pointer parameter specifies the address of a
699 structure that is the return value of the function in the source
700 program. This pointer must be guaranteed by the caller to be valid:
701 loads and stores to the structure may be assumed by the callee
702 not to trap and to be properly aligned. This may only be applied to
703 the first parameter. This is not a valid attribute for return
706 This indicates that pointer values :ref:`based <pointeraliasing>` on
707 the argument or return value do not alias pointer values which are
708 not *based* on it, ignoring certain "irrelevant" dependencies. For a
709 call to the parent function, dependencies between memory references
710 from before or after the call and from those during the call are
711 "irrelevant" to the ``noalias`` keyword for the arguments and return
712 value used in that call. The caller shares the responsibility with
713 the callee for ensuring that these requirements are met. For further
714 details, please see the discussion of the NoAlias response in `alias
715 analysis <AliasAnalysis.html#MustMayNo>`_.
717 Note that this definition of ``noalias`` is intentionally similar
718 to the definition of ``restrict`` in C99 for function arguments,
719 though it is slightly weaker.
721 For function return values, C99's ``restrict`` is not meaningful,
722 while LLVM's ``noalias`` is.
724 This indicates that the callee does not make any copies of the
725 pointer that outlive the callee itself. This is not a valid
726 attribute for return values.
731 This indicates that the pointer parameter can be excised using the
732 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
733 attribute for return values and can only be applied to one parameter.
736 This indicates that the function always returns the argument as its return
737 value. This is an optimization hint to the code generator when generating
738 the caller, allowing tail call optimization and omission of register saves
739 and restores in some cases; it is not checked or enforced when generating
740 the callee. The parameter and the function return type must be valid
741 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
742 valid attribute for return values and can only be applied to one parameter.
746 Garbage Collector Names
747 -----------------------
749 Each function may specify a garbage collector name, which is simply a
754 define void @f() gc "name" { ... }
756 The compiler declares the supported values of *name*. Specifying a
757 collector which will cause the compiler to alter its output in order to
758 support the named garbage collection algorithm.
765 Attribute groups are groups of attributes that are referenced by objects within
766 the IR. They are important for keeping ``.ll`` files readable, because a lot of
767 functions will use the same set of attributes. In the degenerative case of a
768 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
769 group will capture the important command line flags used to build that file.
771 An attribute group is a module-level object. To use an attribute group, an
772 object references the attribute group's ID (e.g. ``#37``). An object may refer
773 to more than one attribute group. In that situation, the attributes from the
774 different groups are merged.
776 Here is an example of attribute groups for a function that should always be
777 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
781 ; Target-independent attributes:
782 attributes #0 = { alwaysinline alignstack=4 }
784 ; Target-dependent attributes:
785 attributes #1 = { "no-sse" }
787 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
788 define void @f() #0 #1 { ... }
795 Function attributes are set to communicate additional information about
796 a function. Function attributes are considered to be part of the
797 function, not of the function type, so functions with different function
798 attributes can have the same function type.
800 Function attributes are simple keywords that follow the type specified.
801 If multiple attributes are needed, they are space separated. For
806 define void @f() noinline { ... }
807 define void @f() alwaysinline { ... }
808 define void @f() alwaysinline optsize { ... }
809 define void @f() optsize { ... }
812 This attribute indicates that, when emitting the prologue and
813 epilogue, the backend should forcibly align the stack pointer.
814 Specify the desired alignment, which must be a power of two, in
817 This attribute indicates that the inliner should attempt to inline
818 this function into callers whenever possible, ignoring any active
819 inlining size threshold for this caller.
821 This indicates that the callee function at a call site should be
822 recognized as a built-in function, even though the function's declaration
823 uses the ``nobuiltin`` attribute. This is only valid at call sites for
824 direct calls to functions which are declared with the ``nobuiltin``
827 This attribute indicates that this function is rarely called. When
828 computing edge weights, basic blocks post-dominated by a cold
829 function call are also considered to be cold; and, thus, given low
832 This attribute suppresses lazy symbol binding for the function. This
833 may make calls to the function faster, at the cost of extra program
834 startup time if the function is not called during program startup.
836 This attribute indicates that the source code contained a hint that
837 inlining this function is desirable (such as the "inline" keyword in
838 C/C++). It is just a hint; it imposes no requirements on the
841 This attribute disables prologue / epilogue emission for the
842 function. This can have very system-specific consequences.
844 This indicates that the callee function at a call site is not recognized as
845 a built-in function. LLVM will retain the original call and not replace it
846 with equivalent code based on the semantics of the built-in function, unless
847 the call site uses the ``builtin`` attribute. This is valid at call sites
848 and on function declarations and definitions.
850 This attribute indicates that calls to the function cannot be
851 duplicated. A call to a ``noduplicate`` function may be moved
852 within its parent function, but may not be duplicated within
855 A function containing a ``noduplicate`` call may still
856 be an inlining candidate, provided that the call is not
857 duplicated by inlining. That implies that the function has
858 internal linkage and only has one call site, so the original
859 call is dead after inlining.
861 This attributes disables implicit floating point instructions.
863 This attribute indicates that the inliner should never inline this
864 function in any situation. This attribute may not be used together
865 with the ``alwaysinline`` attribute.
867 This attribute indicates that the code generator should not use a
868 red zone, even if the target-specific ABI normally permits it.
870 This function attribute indicates that the function never returns
871 normally. This produces undefined behavior at runtime if the
872 function ever does dynamically return.
874 This function attribute indicates that the function never returns
875 with an unwind or exceptional control flow. If the function does
876 unwind, its runtime behavior is undefined.
878 This attribute suggests that optimization passes and code generator
879 passes make choices that keep the code size of this function low,
880 and otherwise do optimizations specifically to reduce code size.
882 On a function, this attribute indicates that the function computes its
883 result (or decides to unwind an exception) based strictly on its arguments,
884 without dereferencing any pointer arguments or otherwise accessing
885 any mutable state (e.g. memory, control registers, etc) visible to
886 caller functions. It does not write through any pointer arguments
887 (including ``byval`` arguments) and never changes any state visible
888 to callers. This means that it cannot unwind exceptions by calling
889 the ``C++`` exception throwing methods.
891 On an argument, this attribute indicates that the function does not
892 dereference that pointer argument, even though it may read or write the
893 memory that the pointer points to if accessed through other pointers.
895 On a function, this attribute indicates that the function does not write
896 through any pointer arguments (including ``byval`` arguments) or otherwise
897 modify any state (e.g. memory, control registers, etc) visible to
898 caller functions. It may dereference pointer arguments and read
899 state that may be set in the caller. A readonly function always
900 returns the same value (or unwinds an exception identically) when
901 called with the same set of arguments and global state. It cannot
902 unwind an exception by calling the ``C++`` exception throwing
905 On an argument, this attribute indicates that the function does not write
906 through this pointer argument, even though it may write to the memory that
907 the pointer points to.
909 This attribute indicates that this function can return twice. The C
910 ``setjmp`` is an example of such a function. The compiler disables
911 some optimizations (like tail calls) in the caller of these
914 This attribute indicates that AddressSanitizer checks
915 (dynamic address safety analysis) are enabled for this function.
917 This attribute indicates that MemorySanitizer checks (dynamic detection
918 of accesses to uninitialized memory) are enabled for this function.
920 This attribute indicates that ThreadSanitizer checks
921 (dynamic thread safety analysis) are enabled for this function.
923 This attribute indicates that the function should emit a stack
924 smashing protector. It is in the form of a "canary" --- a random value
925 placed on the stack before the local variables that's checked upon
926 return from the function to see if it has been overwritten. A
927 heuristic is used to determine if a function needs stack protectors
928 or not. The heuristic used will enable protectors for functions with:
930 - Character arrays larger than ``ssp-buffer-size`` (default 8).
931 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
932 - Calls to alloca() with variable sizes or constant sizes greater than
935 If a function that has an ``ssp`` attribute is inlined into a
936 function that doesn't have an ``ssp`` attribute, then the resulting
937 function will have an ``ssp`` attribute.
939 This attribute indicates that the function should *always* emit a
940 stack smashing protector. This overrides the ``ssp`` function
943 If a function that has an ``sspreq`` attribute is inlined into a
944 function that doesn't have an ``sspreq`` attribute or which has an
945 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
946 an ``sspreq`` attribute.
948 This attribute indicates that the function should emit a stack smashing
949 protector. This attribute causes a strong heuristic to be used when
950 determining if a function needs stack protectors. The strong heuristic
951 will enable protectors for functions with:
953 - Arrays of any size and type
954 - Aggregates containing an array of any size and type.
956 - Local variables that have had their address taken.
958 This overrides the ``ssp`` function attribute.
960 If a function that has an ``sspstrong`` attribute is inlined into a
961 function that doesn't have an ``sspstrong`` attribute, then the
962 resulting function will have an ``sspstrong`` attribute.
964 This attribute indicates that the ABI being targeted requires that
965 an unwind table entry be produce for this function even if we can
966 show that no exceptions passes by it. This is normally the case for
967 the ELF x86-64 abi, but it can be disabled for some compilation
972 Module-Level Inline Assembly
973 ----------------------------
975 Modules may contain "module-level inline asm" blocks, which corresponds
976 to the GCC "file scope inline asm" blocks. These blocks are internally
977 concatenated by LLVM and treated as a single unit, but may be separated
978 in the ``.ll`` file if desired. The syntax is very simple:
982 module asm "inline asm code goes here"
983 module asm "more can go here"
985 The strings can contain any character by escaping non-printable
986 characters. The escape sequence used is simply "\\xx" where "xx" is the
987 two digit hex code for the number.
989 The inline asm code is simply printed to the machine code .s file when
990 assembly code is generated.
992 .. _langref_datalayout:
997 A module may specify a target specific data layout string that specifies
998 how data is to be laid out in memory. The syntax for the data layout is
1001 .. code-block:: llvm
1003 target datalayout = "layout specification"
1005 The *layout specification* consists of a list of specifications
1006 separated by the minus sign character ('-'). Each specification starts
1007 with a letter and may include other information after the letter to
1008 define some aspect of the data layout. The specifications accepted are
1012 Specifies that the target lays out data in big-endian form. That is,
1013 the bits with the most significance have the lowest address
1016 Specifies that the target lays out data in little-endian form. That
1017 is, the bits with the least significance have the lowest address
1020 Specifies the natural alignment of the stack in bits. Alignment
1021 promotion of stack variables is limited to the natural stack
1022 alignment to avoid dynamic stack realignment. The stack alignment
1023 must be a multiple of 8-bits. If omitted, the natural stack
1024 alignment defaults to "unspecified", which does not prevent any
1025 alignment promotions.
1026 ``p[n]:<size>:<abi>:<pref>``
1027 This specifies the *size* of a pointer and its ``<abi>`` and
1028 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1029 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
1030 preceding ``:`` should be omitted too. The address space, ``n`` is
1031 optional, and if not specified, denotes the default address space 0.
1032 The value of ``n`` must be in the range [1,2^23).
1033 ``i<size>:<abi>:<pref>``
1034 This specifies the alignment for an integer type of a given bit
1035 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1036 ``v<size>:<abi>:<pref>``
1037 This specifies the alignment for a vector type of a given bit
1039 ``f<size>:<abi>:<pref>``
1040 This specifies the alignment for a floating point type of a given bit
1041 ``<size>``. Only values of ``<size>`` that are supported by the target
1042 will work. 32 (float) and 64 (double) are supported on all targets; 80
1043 or 128 (different flavors of long double) are also supported on some
1045 ``a<size>:<abi>:<pref>``
1046 This specifies the alignment for an aggregate type of a given bit
1048 ``s<size>:<abi>:<pref>``
1049 This specifies the alignment for a stack object of a given bit
1051 ``n<size1>:<size2>:<size3>...``
1052 This specifies a set of native integer widths for the target CPU in
1053 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1054 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1055 this set are considered to support most general arithmetic operations
1058 When constructing the data layout for a given target, LLVM starts with a
1059 default set of specifications which are then (possibly) overridden by
1060 the specifications in the ``datalayout`` keyword. The default
1061 specifications are given in this list:
1063 - ``E`` - big endian
1064 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment
1065 - ``S0`` - natural stack alignment is unspecified
1066 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1067 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1068 - ``i16:16:16`` - i16 is 16-bit aligned
1069 - ``i32:32:32`` - i32 is 32-bit aligned
1070 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1071 alignment of 64-bits
1072 - ``f16:16:16`` - half is 16-bit aligned
1073 - ``f32:32:32`` - float is 32-bit aligned
1074 - ``f64:64:64`` - double is 64-bit aligned
1075 - ``f128:128:128`` - quad is 128-bit aligned
1076 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1077 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1078 - ``a0:0:64`` - aggregates are 64-bit aligned
1080 When LLVM is determining the alignment for a given type, it uses the
1083 #. If the type sought is an exact match for one of the specifications,
1084 that specification is used.
1085 #. If no match is found, and the type sought is an integer type, then
1086 the smallest integer type that is larger than the bitwidth of the
1087 sought type is used. If none of the specifications are larger than
1088 the bitwidth then the largest integer type is used. For example,
1089 given the default specifications above, the i7 type will use the
1090 alignment of i8 (next largest) while both i65 and i256 will use the
1091 alignment of i64 (largest specified).
1092 #. If no match is found, and the type sought is a vector type, then the
1093 largest vector type that is smaller than the sought vector type will
1094 be used as a fall back. This happens because <128 x double> can be
1095 implemented in terms of 64 <2 x double>, for example.
1097 The function of the data layout string may not be what you expect.
1098 Notably, this is not a specification from the frontend of what alignment
1099 the code generator should use.
1101 Instead, if specified, the target data layout is required to match what
1102 the ultimate *code generator* expects. This string is used by the
1103 mid-level optimizers to improve code, and this only works if it matches
1104 what the ultimate code generator uses. If you would like to generate IR
1105 that does not embed this target-specific detail into the IR, then you
1106 don't have to specify the string. This will disable some optimizations
1107 that require precise layout information, but this also prevents those
1108 optimizations from introducing target specificity into the IR.
1110 .. _pointeraliasing:
1112 Pointer Aliasing Rules
1113 ----------------------
1115 Any memory access must be done through a pointer value associated with
1116 an address range of the memory access, otherwise the behavior is
1117 undefined. Pointer values are associated with address ranges according
1118 to the following rules:
1120 - A pointer value is associated with the addresses associated with any
1121 value it is *based* on.
1122 - An address of a global variable is associated with the address range
1123 of the variable's storage.
1124 - The result value of an allocation instruction is associated with the
1125 address range of the allocated storage.
1126 - A null pointer in the default address-space is associated with no
1128 - An integer constant other than zero or a pointer value returned from
1129 a function not defined within LLVM may be associated with address
1130 ranges allocated through mechanisms other than those provided by
1131 LLVM. Such ranges shall not overlap with any ranges of addresses
1132 allocated by mechanisms provided by LLVM.
1134 A pointer value is *based* on another pointer value according to the
1137 - A pointer value formed from a ``getelementptr`` operation is *based*
1138 on the first operand of the ``getelementptr``.
1139 - The result value of a ``bitcast`` is *based* on the operand of the
1141 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1142 values that contribute (directly or indirectly) to the computation of
1143 the pointer's value.
1144 - The "*based* on" relationship is transitive.
1146 Note that this definition of *"based"* is intentionally similar to the
1147 definition of *"based"* in C99, though it is slightly weaker.
1149 LLVM IR does not associate types with memory. The result type of a
1150 ``load`` merely indicates the size and alignment of the memory from
1151 which to load, as well as the interpretation of the value. The first
1152 operand type of a ``store`` similarly only indicates the size and
1153 alignment of the store.
1155 Consequently, type-based alias analysis, aka TBAA, aka
1156 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1157 :ref:`Metadata <metadata>` may be used to encode additional information
1158 which specialized optimization passes may use to implement type-based
1163 Volatile Memory Accesses
1164 ------------------------
1166 Certain memory accesses, such as :ref:`load <i_load>`'s,
1167 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1168 marked ``volatile``. The optimizers must not change the number of
1169 volatile operations or change their order of execution relative to other
1170 volatile operations. The optimizers *may* change the order of volatile
1171 operations relative to non-volatile operations. This is not Java's
1172 "volatile" and has no cross-thread synchronization behavior.
1174 IR-level volatile loads and stores cannot safely be optimized into
1175 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1176 flagged volatile. Likewise, the backend should never split or merge
1177 target-legal volatile load/store instructions.
1179 .. admonition:: Rationale
1181 Platforms may rely on volatile loads and stores of natively supported
1182 data width to be executed as single instruction. For example, in C
1183 this holds for an l-value of volatile primitive type with native
1184 hardware support, but not necessarily for aggregate types. The
1185 frontend upholds these expectations, which are intentionally
1186 unspecified in the IR. The rules above ensure that IR transformation
1187 do not violate the frontend's contract with the language.
1191 Memory Model for Concurrent Operations
1192 --------------------------------------
1194 The LLVM IR does not define any way to start parallel threads of
1195 execution or to register signal handlers. Nonetheless, there are
1196 platform-specific ways to create them, and we define LLVM IR's behavior
1197 in their presence. This model is inspired by the C++0x memory model.
1199 For a more informal introduction to this model, see the :doc:`Atomics`.
1201 We define a *happens-before* partial order as the least partial order
1204 - Is a superset of single-thread program order, and
1205 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1206 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1207 techniques, like pthread locks, thread creation, thread joining,
1208 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1209 Constraints <ordering>`).
1211 Note that program order does not introduce *happens-before* edges
1212 between a thread and signals executing inside that thread.
1214 Every (defined) read operation (load instructions, memcpy, atomic
1215 loads/read-modify-writes, etc.) R reads a series of bytes written by
1216 (defined) write operations (store instructions, atomic
1217 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1218 section, initialized globals are considered to have a write of the
1219 initializer which is atomic and happens before any other read or write
1220 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1221 may see any write to the same byte, except:
1223 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1224 write\ :sub:`2` happens before R\ :sub:`byte`, then
1225 R\ :sub:`byte` does not see write\ :sub:`1`.
1226 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1227 R\ :sub:`byte` does not see write\ :sub:`3`.
1229 Given that definition, R\ :sub:`byte` is defined as follows:
1231 - If R is volatile, the result is target-dependent. (Volatile is
1232 supposed to give guarantees which can support ``sig_atomic_t`` in
1233 C/C++, and may be used for accesses to addresses which do not behave
1234 like normal memory. It does not generally provide cross-thread
1236 - Otherwise, if there is no write to the same byte that happens before
1237 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1238 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1239 R\ :sub:`byte` returns the value written by that write.
1240 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1241 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1242 Memory Ordering Constraints <ordering>` section for additional
1243 constraints on how the choice is made.
1244 - Otherwise R\ :sub:`byte` returns ``undef``.
1246 R returns the value composed of the series of bytes it read. This
1247 implies that some bytes within the value may be ``undef`` **without**
1248 the entire value being ``undef``. Note that this only defines the
1249 semantics of the operation; it doesn't mean that targets will emit more
1250 than one instruction to read the series of bytes.
1252 Note that in cases where none of the atomic intrinsics are used, this
1253 model places only one restriction on IR transformations on top of what
1254 is required for single-threaded execution: introducing a store to a byte
1255 which might not otherwise be stored is not allowed in general.
1256 (Specifically, in the case where another thread might write to and read
1257 from an address, introducing a store can change a load that may see
1258 exactly one write into a load that may see multiple writes.)
1262 Atomic Memory Ordering Constraints
1263 ----------------------------------
1265 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1266 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1267 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1268 an ordering parameter that determines which other atomic instructions on
1269 the same address they *synchronize with*. These semantics are borrowed
1270 from Java and C++0x, but are somewhat more colloquial. If these
1271 descriptions aren't precise enough, check those specs (see spec
1272 references in the :doc:`atomics guide <Atomics>`).
1273 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1274 differently since they don't take an address. See that instruction's
1275 documentation for details.
1277 For a simpler introduction to the ordering constraints, see the
1281 The set of values that can be read is governed by the happens-before
1282 partial order. A value cannot be read unless some operation wrote
1283 it. This is intended to provide a guarantee strong enough to model
1284 Java's non-volatile shared variables. This ordering cannot be
1285 specified for read-modify-write operations; it is not strong enough
1286 to make them atomic in any interesting way.
1288 In addition to the guarantees of ``unordered``, there is a single
1289 total order for modifications by ``monotonic`` operations on each
1290 address. All modification orders must be compatible with the
1291 happens-before order. There is no guarantee that the modification
1292 orders can be combined to a global total order for the whole program
1293 (and this often will not be possible). The read in an atomic
1294 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1295 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1296 order immediately before the value it writes. If one atomic read
1297 happens before another atomic read of the same address, the later
1298 read must see the same value or a later value in the address's
1299 modification order. This disallows reordering of ``monotonic`` (or
1300 stronger) operations on the same address. If an address is written
1301 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1302 read that address repeatedly, the other threads must eventually see
1303 the write. This corresponds to the C++0x/C1x
1304 ``memory_order_relaxed``.
1306 In addition to the guarantees of ``monotonic``, a
1307 *synchronizes-with* edge may be formed with a ``release`` operation.
1308 This is intended to model C++'s ``memory_order_acquire``.
1310 In addition to the guarantees of ``monotonic``, if this operation
1311 writes a value which is subsequently read by an ``acquire``
1312 operation, it *synchronizes-with* that operation. (This isn't a
1313 complete description; see the C++0x definition of a release
1314 sequence.) This corresponds to the C++0x/C1x
1315 ``memory_order_release``.
1316 ``acq_rel`` (acquire+release)
1317 Acts as both an ``acquire`` and ``release`` operation on its
1318 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1319 ``seq_cst`` (sequentially consistent)
1320 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1321 operation which only reads, ``release`` for an operation which only
1322 writes), there is a global total order on all
1323 sequentially-consistent operations on all addresses, which is
1324 consistent with the *happens-before* partial order and with the
1325 modification orders of all the affected addresses. Each
1326 sequentially-consistent read sees the last preceding write to the
1327 same address in this global order. This corresponds to the C++0x/C1x
1328 ``memory_order_seq_cst`` and Java volatile.
1332 If an atomic operation is marked ``singlethread``, it only *synchronizes
1333 with* or participates in modification and seq\_cst total orderings with
1334 other operations running in the same thread (for example, in signal
1342 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1343 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1344 :ref:`frem <i_frem>`) have the following flags that can set to enable
1345 otherwise unsafe floating point operations
1348 No NaNs - Allow optimizations to assume the arguments and result are not
1349 NaN. Such optimizations are required to retain defined behavior over
1350 NaNs, but the value of the result is undefined.
1353 No Infs - Allow optimizations to assume the arguments and result are not
1354 +/-Inf. Such optimizations are required to retain defined behavior over
1355 +/-Inf, but the value of the result is undefined.
1358 No Signed Zeros - Allow optimizations to treat the sign of a zero
1359 argument or result as insignificant.
1362 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1363 argument rather than perform division.
1366 Fast - Allow algebraically equivalent transformations that may
1367 dramatically change results in floating point (e.g. reassociate). This
1368 flag implies all the others.
1375 The LLVM type system is one of the most important features of the
1376 intermediate representation. Being typed enables a number of
1377 optimizations to be performed on the intermediate representation
1378 directly, without having to do extra analyses on the side before the
1379 transformation. A strong type system makes it easier to read the
1380 generated code and enables novel analyses and transformations that are
1381 not feasible to perform on normal three address code representations.
1383 .. _typeclassifications:
1385 Type Classifications
1386 --------------------
1388 The types fall into a few useful classifications:
1397 * - :ref:`integer <t_integer>`
1398 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1401 * - :ref:`floating point <t_floating>`
1402 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1410 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1411 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1412 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1413 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1415 * - :ref:`primitive <t_primitive>`
1416 - :ref:`label <t_label>`,
1417 :ref:`void <t_void>`,
1418 :ref:`integer <t_integer>`,
1419 :ref:`floating point <t_floating>`,
1420 :ref:`x86mmx <t_x86mmx>`,
1421 :ref:`metadata <t_metadata>`.
1423 * - :ref:`derived <t_derived>`
1424 - :ref:`array <t_array>`,
1425 :ref:`function <t_function>`,
1426 :ref:`pointer <t_pointer>`,
1427 :ref:`structure <t_struct>`,
1428 :ref:`vector <t_vector>`,
1429 :ref:`opaque <t_opaque>`.
1431 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1432 Values of these types are the only ones which can be produced by
1440 The primitive types are the fundamental building blocks of the LLVM
1451 The integer type is a very simple type that simply specifies an
1452 arbitrary bit width for the integer type desired. Any bit width from 1
1453 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1462 The number of bits the integer will occupy is specified by the ``N``
1468 +----------------+------------------------------------------------+
1469 | ``i1`` | a single-bit integer. |
1470 +----------------+------------------------------------------------+
1471 | ``i32`` | a 32-bit integer. |
1472 +----------------+------------------------------------------------+
1473 | ``i1942652`` | a really big integer of over 1 million bits. |
1474 +----------------+------------------------------------------------+
1478 Floating Point Types
1479 ^^^^^^^^^^^^^^^^^^^^
1488 - 16-bit floating point value
1491 - 32-bit floating point value
1494 - 64-bit floating point value
1497 - 128-bit floating point value (112-bit mantissa)
1500 - 80-bit floating point value (X87)
1503 - 128-bit floating point value (two 64-bits)
1513 The x86mmx type represents a value held in an MMX register on an x86
1514 machine. The operations allowed on it are quite limited: parameters and
1515 return values, load and store, and bitcast. User-specified MMX
1516 instructions are represented as intrinsic or asm calls with arguments
1517 and/or results of this type. There are no arrays, vectors or constants
1535 The void type does not represent any value and has no size.
1552 The label type represents code labels.
1569 The metadata type represents embedded metadata. No derived types may be
1570 created from metadata except for :ref:`function <t_function>` arguments.
1584 The real power in LLVM comes from the derived types in the system. This
1585 is what allows a programmer to represent arrays, functions, pointers,
1586 and other useful types. Each of these types contain one or more element
1587 types which may be a primitive type, or another derived type. For
1588 example, it is possible to have a two dimensional array, using an array
1589 as the element type of another array.
1596 Aggregate Types are a subset of derived types that can contain multiple
1597 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1598 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1609 The array type is a very simple derived type that arranges elements
1610 sequentially in memory. The array type requires a size (number of
1611 elements) and an underlying data type.
1618 [<# elements> x <elementtype>]
1620 The number of elements is a constant integer value; ``elementtype`` may
1621 be any type with a size.
1626 +------------------+--------------------------------------+
1627 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1628 +------------------+--------------------------------------+
1629 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1630 +------------------+--------------------------------------+
1631 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1632 +------------------+--------------------------------------+
1634 Here are some examples of multidimensional arrays:
1636 +-----------------------------+----------------------------------------------------------+
1637 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1638 +-----------------------------+----------------------------------------------------------+
1639 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1640 +-----------------------------+----------------------------------------------------------+
1641 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1642 +-----------------------------+----------------------------------------------------------+
1644 There is no restriction on indexing beyond the end of the array implied
1645 by a static type (though there are restrictions on indexing beyond the
1646 bounds of an allocated object in some cases). This means that
1647 single-dimension 'variable sized array' addressing can be implemented in
1648 LLVM with a zero length array type. An implementation of 'pascal style
1649 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1660 The function type can be thought of as a function signature. It consists
1661 of a return type and a list of formal parameter types. The return type
1662 of a function type is a first class type or a void type.
1669 <returntype> (<parameter list>)
1671 ...where '``<parameter list>``' is a comma-separated list of type
1672 specifiers. Optionally, the parameter list may include a type ``...``,
1673 which indicates that the function takes a variable number of arguments.
1674 Variable argument functions can access their arguments with the
1675 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1676 '``<returntype>``' is any type except :ref:`label <t_label>`.
1681 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1682 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1683 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1684 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1685 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1686 | ``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. |
1687 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1688 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1689 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1699 The structure type is used to represent a collection of data members
1700 together in memory. The elements of a structure may be any type that has
1703 Structures in memory are accessed using '``load``' and '``store``' by
1704 getting a pointer to a field with the '``getelementptr``' instruction.
1705 Structures in registers are accessed using the '``extractvalue``' and
1706 '``insertvalue``' instructions.
1708 Structures may optionally be "packed" structures, which indicate that
1709 the alignment of the struct is one byte, and that there is no padding
1710 between the elements. In non-packed structs, padding between field types
1711 is inserted as defined by the DataLayout string in the module, which is
1712 required to match what the underlying code generator expects.
1714 Structures can either be "literal" or "identified". A literal structure
1715 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1716 identified types are always defined at the top level with a name.
1717 Literal types are uniqued by their contents and can never be recursive
1718 or opaque since there is no way to write one. Identified types can be
1719 recursive, can be opaqued, and are never uniqued.
1726 %T1 = type { <type list> } ; Identified normal struct type
1727 %T2 = type <{ <type list> }> ; Identified packed struct type
1732 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1733 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1734 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1735 | ``{ 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``. |
1736 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1737 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1738 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1742 Opaque Structure Types
1743 ^^^^^^^^^^^^^^^^^^^^^^
1748 Opaque structure types are used to represent named structure types that
1749 do not have a body specified. This corresponds (for example) to the C
1750 notion of a forward declared structure.
1763 +--------------+-------------------+
1764 | ``opaque`` | An opaque type. |
1765 +--------------+-------------------+
1775 The pointer type is used to specify memory locations. Pointers are
1776 commonly used to reference objects in memory.
1778 Pointer types may have an optional address space attribute defining the
1779 numbered address space where the pointed-to object resides. The default
1780 address space is number zero. The semantics of non-zero address spaces
1781 are target-specific.
1783 Note that LLVM does not permit pointers to void (``void*``) nor does it
1784 permit pointers to labels (``label*``). Use ``i8*`` instead.
1796 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1797 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1798 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1799 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1800 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1801 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1802 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1812 A vector type is a simple derived type that represents a vector of
1813 elements. Vector types are used when multiple primitive data are
1814 operated in parallel using a single instruction (SIMD). A vector type
1815 requires a size (number of elements) and an underlying primitive data
1816 type. Vector types are considered :ref:`first class <t_firstclass>`.
1823 < <# elements> x <elementtype> >
1825 The number of elements is a constant integer value larger than 0;
1826 elementtype may be any integer or floating point type, or a pointer to
1827 these types. Vectors of size zero are not allowed.
1832 +-------------------+--------------------------------------------------+
1833 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1834 +-------------------+--------------------------------------------------+
1835 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1836 +-------------------+--------------------------------------------------+
1837 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1838 +-------------------+--------------------------------------------------+
1839 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1840 +-------------------+--------------------------------------------------+
1845 LLVM has several different basic types of constants. This section
1846 describes them all and their syntax.
1851 **Boolean constants**
1852 The two strings '``true``' and '``false``' are both valid constants
1854 **Integer constants**
1855 Standard integers (such as '4') are constants of the
1856 :ref:`integer <t_integer>` type. Negative numbers may be used with
1858 **Floating point constants**
1859 Floating point constants use standard decimal notation (e.g.
1860 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1861 hexadecimal notation (see below). The assembler requires the exact
1862 decimal value of a floating-point constant. For example, the
1863 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1864 decimal in binary. Floating point constants must have a :ref:`floating
1865 point <t_floating>` type.
1866 **Null pointer constants**
1867 The identifier '``null``' is recognized as a null pointer constant
1868 and must be of :ref:`pointer type <t_pointer>`.
1870 The one non-intuitive notation for constants is the hexadecimal form of
1871 floating point constants. For example, the form
1872 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1873 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1874 constants are required (and the only time that they are generated by the
1875 disassembler) is when a floating point constant must be emitted but it
1876 cannot be represented as a decimal floating point number in a reasonable
1877 number of digits. For example, NaN's, infinities, and other special
1878 values are represented in their IEEE hexadecimal format so that assembly
1879 and disassembly do not cause any bits to change in the constants.
1881 When using the hexadecimal form, constants of types half, float, and
1882 double are represented using the 16-digit form shown above (which
1883 matches the IEEE754 representation for double); half and float values
1884 must, however, be exactly representable as IEEE 754 half and single
1885 precision, respectively. Hexadecimal format is always used for long
1886 double, and there are three forms of long double. The 80-bit format used
1887 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1888 128-bit format used by PowerPC (two adjacent doubles) is represented by
1889 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1890 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
1891 will only work if they match the long double format on your target.
1892 The IEEE 16-bit format (half precision) is represented by ``0xH``
1893 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
1894 (sign bit at the left).
1896 There are no constants of type x86mmx.
1898 .. _complexconstants:
1903 Complex constants are a (potentially recursive) combination of simple
1904 constants and smaller complex constants.
1906 **Structure constants**
1907 Structure constants are represented with notation similar to
1908 structure type definitions (a comma separated list of elements,
1909 surrounded by braces (``{}``)). For example:
1910 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1911 "``@G = external global i32``". Structure constants must have
1912 :ref:`structure type <t_struct>`, and the number and types of elements
1913 must match those specified by the type.
1915 Array constants are represented with notation similar to array type
1916 definitions (a comma separated list of elements, surrounded by
1917 square brackets (``[]``)). For example:
1918 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1919 :ref:`array type <t_array>`, and the number and types of elements must
1920 match those specified by the type.
1921 **Vector constants**
1922 Vector constants are represented with notation similar to vector
1923 type definitions (a comma separated list of elements, surrounded by
1924 less-than/greater-than's (``<>``)). For example:
1925 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1926 must have :ref:`vector type <t_vector>`, and the number and types of
1927 elements must match those specified by the type.
1928 **Zero initialization**
1929 The string '``zeroinitializer``' can be used to zero initialize a
1930 value to zero of *any* type, including scalar and
1931 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1932 having to print large zero initializers (e.g. for large arrays) and
1933 is always exactly equivalent to using explicit zero initializers.
1935 A metadata node is a structure-like constant with :ref:`metadata
1936 type <t_metadata>`. For example:
1937 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
1938 constants that are meant to be interpreted as part of the
1939 instruction stream, metadata is a place to attach additional
1940 information such as debug info.
1942 Global Variable and Function Addresses
1943 --------------------------------------
1945 The addresses of :ref:`global variables <globalvars>` and
1946 :ref:`functions <functionstructure>` are always implicitly valid
1947 (link-time) constants. These constants are explicitly referenced when
1948 the :ref:`identifier for the global <identifiers>` is used and always have
1949 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
1952 .. code-block:: llvm
1956 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1963 The string '``undef``' can be used anywhere a constant is expected, and
1964 indicates that the user of the value may receive an unspecified
1965 bit-pattern. Undefined values may be of any type (other than '``label``'
1966 or '``void``') and be used anywhere a constant is permitted.
1968 Undefined values are useful because they indicate to the compiler that
1969 the program is well defined no matter what value is used. This gives the
1970 compiler more freedom to optimize. Here are some examples of
1971 (potentially surprising) transformations that are valid (in pseudo IR):
1973 .. code-block:: llvm
1983 This is safe because all of the output bits are affected by the undef
1984 bits. Any output bit can have a zero or one depending on the input bits.
1986 .. code-block:: llvm
1997 These logical operations have bits that are not always affected by the
1998 input. For example, if ``%X`` has a zero bit, then the output of the
1999 '``and``' operation will always be a zero for that bit, no matter what
2000 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2001 optimize or assume that the result of the '``and``' is '``undef``'.
2002 However, it is safe to assume that all bits of the '``undef``' could be
2003 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2004 all the bits of the '``undef``' operand to the '``or``' could be set,
2005 allowing the '``or``' to be folded to -1.
2007 .. code-block:: llvm
2009 %A = select undef, %X, %Y
2010 %B = select undef, 42, %Y
2011 %C = select %X, %Y, undef
2021 This set of examples shows that undefined '``select``' (and conditional
2022 branch) conditions can go *either way*, but they have to come from one
2023 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2024 both known to have a clear low bit, then ``%A`` would have to have a
2025 cleared low bit. However, in the ``%C`` example, the optimizer is
2026 allowed to assume that the '``undef``' operand could be the same as
2027 ``%Y``, allowing the whole '``select``' to be eliminated.
2029 .. code-block:: llvm
2031 %A = xor undef, undef
2048 This example points out that two '``undef``' operands are not
2049 necessarily the same. This can be surprising to people (and also matches
2050 C semantics) where they assume that "``X^X``" is always zero, even if
2051 ``X`` is undefined. This isn't true for a number of reasons, but the
2052 short answer is that an '``undef``' "variable" can arbitrarily change
2053 its value over its "live range". This is true because the variable
2054 doesn't actually *have a live range*. Instead, the value is logically
2055 read from arbitrary registers that happen to be around when needed, so
2056 the value is not necessarily consistent over time. In fact, ``%A`` and
2057 ``%C`` need to have the same semantics or the core LLVM "replace all
2058 uses with" concept would not hold.
2060 .. code-block:: llvm
2068 These examples show the crucial difference between an *undefined value*
2069 and *undefined behavior*. An undefined value (like '``undef``') is
2070 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2071 operation can be constant folded to '``undef``', because the '``undef``'
2072 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2073 However, in the second example, we can make a more aggressive
2074 assumption: because the ``undef`` is allowed to be an arbitrary value,
2075 we are allowed to assume that it could be zero. Since a divide by zero
2076 has *undefined behavior*, we are allowed to assume that the operation
2077 does not execute at all. This allows us to delete the divide and all
2078 code after it. Because the undefined operation "can't happen", the
2079 optimizer can assume that it occurs in dead code.
2081 .. code-block:: llvm
2083 a: store undef -> %X
2084 b: store %X -> undef
2089 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2090 value can be assumed to not have any effect; we can assume that the
2091 value is overwritten with bits that happen to match what was already
2092 there. However, a store *to* an undefined location could clobber
2093 arbitrary memory, therefore, it has undefined behavior.
2100 Poison values are similar to :ref:`undef values <undefvalues>`, however
2101 they also represent the fact that an instruction or constant expression
2102 which cannot evoke side effects has nevertheless detected a condition
2103 which results in undefined behavior.
2105 There is currently no way of representing a poison value in the IR; they
2106 only exist when produced by operations such as :ref:`add <i_add>` with
2109 Poison value behavior is defined in terms of value *dependence*:
2111 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2112 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2113 their dynamic predecessor basic block.
2114 - Function arguments depend on the corresponding actual argument values
2115 in the dynamic callers of their functions.
2116 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2117 instructions that dynamically transfer control back to them.
2118 - :ref:`Invoke <i_invoke>` instructions depend on the
2119 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2120 call instructions that dynamically transfer control back to them.
2121 - Non-volatile loads and stores depend on the most recent stores to all
2122 of the referenced memory addresses, following the order in the IR
2123 (including loads and stores implied by intrinsics such as
2124 :ref:`@llvm.memcpy <int_memcpy>`.)
2125 - An instruction with externally visible side effects depends on the
2126 most recent preceding instruction with externally visible side
2127 effects, following the order in the IR. (This includes :ref:`volatile
2128 operations <volatile>`.)
2129 - An instruction *control-depends* on a :ref:`terminator
2130 instruction <terminators>` if the terminator instruction has
2131 multiple successors and the instruction is always executed when
2132 control transfers to one of the successors, and may not be executed
2133 when control is transferred to another.
2134 - Additionally, an instruction also *control-depends* on a terminator
2135 instruction if the set of instructions it otherwise depends on would
2136 be different if the terminator had transferred control to a different
2138 - Dependence is transitive.
2140 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2141 with the additional affect that any instruction which has a *dependence*
2142 on a poison value has undefined behavior.
2144 Here are some examples:
2146 .. code-block:: llvm
2149 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2150 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2151 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2152 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2154 store i32 %poison, i32* @g ; Poison value stored to memory.
2155 %poison2 = load i32* @g ; Poison value loaded back from memory.
2157 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2159 %narrowaddr = bitcast i32* @g to i16*
2160 %wideaddr = bitcast i32* @g to i64*
2161 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2162 %poison4 = load i64* %wideaddr ; Returns a poison value.
2164 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2165 br i1 %cmp, label %true, label %end ; Branch to either destination.
2168 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2169 ; it has undefined behavior.
2173 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2174 ; Both edges into this PHI are
2175 ; control-dependent on %cmp, so this
2176 ; always results in a poison value.
2178 store volatile i32 0, i32* @g ; This would depend on the store in %true
2179 ; if %cmp is true, or the store in %entry
2180 ; otherwise, so this is undefined behavior.
2182 br i1 %cmp, label %second_true, label %second_end
2183 ; The same branch again, but this time the
2184 ; true block doesn't have side effects.
2191 store volatile i32 0, i32* @g ; This time, the instruction always depends
2192 ; on the store in %end. Also, it is
2193 ; control-equivalent to %end, so this is
2194 ; well-defined (ignoring earlier undefined
2195 ; behavior in this example).
2199 Addresses of Basic Blocks
2200 -------------------------
2202 ``blockaddress(@function, %block)``
2204 The '``blockaddress``' constant computes the address of the specified
2205 basic block in the specified function, and always has an ``i8*`` type.
2206 Taking the address of the entry block is illegal.
2208 This value only has defined behavior when used as an operand to the
2209 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2210 against null. Pointer equality tests between labels addresses results in
2211 undefined behavior --- though, again, comparison against null is ok, and
2212 no label is equal to the null pointer. This may be passed around as an
2213 opaque pointer sized value as long as the bits are not inspected. This
2214 allows ``ptrtoint`` and arithmetic to be performed on these values so
2215 long as the original value is reconstituted before the ``indirectbr``
2218 Finally, some targets may provide defined semantics when using the value
2219 as the operand to an inline assembly, but that is target specific.
2223 Constant Expressions
2224 --------------------
2226 Constant expressions are used to allow expressions involving other
2227 constants to be used as constants. Constant expressions may be of any
2228 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2229 that does not have side effects (e.g. load and call are not supported).
2230 The following is the syntax for constant expressions:
2232 ``trunc (CST to TYPE)``
2233 Truncate a constant to another type. The bit size of CST must be
2234 larger than the bit size of TYPE. Both types must be integers.
2235 ``zext (CST to TYPE)``
2236 Zero extend a constant to another type. The bit size of CST must be
2237 smaller than the bit size of TYPE. Both types must be integers.
2238 ``sext (CST to TYPE)``
2239 Sign extend a constant to another type. The bit size of CST must be
2240 smaller than the bit size of TYPE. Both types must be integers.
2241 ``fptrunc (CST to TYPE)``
2242 Truncate a floating point constant to another floating point type.
2243 The size of CST must be larger than the size of TYPE. Both types
2244 must be floating point.
2245 ``fpext (CST to TYPE)``
2246 Floating point extend a constant to another type. The size of CST
2247 must be smaller or equal to the size of TYPE. Both types must be
2249 ``fptoui (CST to TYPE)``
2250 Convert a floating point constant to the corresponding unsigned
2251 integer constant. TYPE must be a scalar or vector integer type. CST
2252 must be of scalar or vector floating point type. Both CST and TYPE
2253 must be scalars, or vectors of the same number of elements. If the
2254 value won't fit in the integer type, the results are undefined.
2255 ``fptosi (CST to TYPE)``
2256 Convert a floating point constant to the corresponding signed
2257 integer constant. TYPE must be a scalar or vector integer type. CST
2258 must be of scalar or vector floating point type. Both CST and TYPE
2259 must be scalars, or vectors of the same number of elements. If the
2260 value won't fit in the integer type, the results are undefined.
2261 ``uitofp (CST to TYPE)``
2262 Convert an unsigned integer constant to the corresponding floating
2263 point constant. TYPE must be a scalar or vector floating point type.
2264 CST must be of scalar or vector integer type. Both CST and TYPE must
2265 be scalars, or vectors of the same number of elements. If the value
2266 won't fit in the floating point type, the results are undefined.
2267 ``sitofp (CST to TYPE)``
2268 Convert a signed integer constant to the corresponding floating
2269 point constant. TYPE must be a scalar or vector floating point type.
2270 CST must be of scalar or vector integer type. Both CST and TYPE must
2271 be scalars, or vectors of the same number of elements. If the value
2272 won't fit in the floating point type, the results are undefined.
2273 ``ptrtoint (CST to TYPE)``
2274 Convert a pointer typed constant to the corresponding integer
2275 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2276 pointer type. The ``CST`` value is zero extended, truncated, or
2277 unchanged to make it fit in ``TYPE``.
2278 ``inttoptr (CST to TYPE)``
2279 Convert an integer constant to a pointer constant. TYPE must be a
2280 pointer type. CST must be of integer type. The CST value is zero
2281 extended, truncated, or unchanged to make it fit in a pointer size.
2282 This one is *really* dangerous!
2283 ``bitcast (CST to TYPE)``
2284 Convert a constant, CST, to another TYPE. The constraints of the
2285 operands are the same as those for the :ref:`bitcast
2286 instruction <i_bitcast>`.
2287 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2288 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2289 constants. As with the :ref:`getelementptr <i_getelementptr>`
2290 instruction, the index list may have zero or more indexes, which are
2291 required to make sense for the type of "CSTPTR".
2292 ``select (COND, VAL1, VAL2)``
2293 Perform the :ref:`select operation <i_select>` on constants.
2294 ``icmp COND (VAL1, VAL2)``
2295 Performs the :ref:`icmp operation <i_icmp>` on constants.
2296 ``fcmp COND (VAL1, VAL2)``
2297 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2298 ``extractelement (VAL, IDX)``
2299 Perform the :ref:`extractelement operation <i_extractelement>` on
2301 ``insertelement (VAL, ELT, IDX)``
2302 Perform the :ref:`insertelement operation <i_insertelement>` on
2304 ``shufflevector (VEC1, VEC2, IDXMASK)``
2305 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2307 ``extractvalue (VAL, IDX0, IDX1, ...)``
2308 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2309 constants. The index list is interpreted in a similar manner as
2310 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2311 least one index value must be specified.
2312 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2313 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2314 The index list is interpreted in a similar manner as indices in a
2315 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2316 value must be specified.
2317 ``OPCODE (LHS, RHS)``
2318 Perform the specified operation of the LHS and RHS constants. OPCODE
2319 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2320 binary <bitwiseops>` operations. The constraints on operands are
2321 the same as those for the corresponding instruction (e.g. no bitwise
2322 operations on floating point values are allowed).
2329 Inline Assembler Expressions
2330 ----------------------------
2332 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2333 Inline Assembly <moduleasm>`) through the use of a special value. This
2334 value represents the inline assembler as a string (containing the
2335 instructions to emit), a list of operand constraints (stored as a
2336 string), a flag that indicates whether or not the inline asm expression
2337 has side effects, and a flag indicating whether the function containing
2338 the asm needs to align its stack conservatively. An example inline
2339 assembler expression is:
2341 .. code-block:: llvm
2343 i32 (i32) asm "bswap $0", "=r,r"
2345 Inline assembler expressions may **only** be used as the callee operand
2346 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2347 Thus, typically we have:
2349 .. code-block:: llvm
2351 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2353 Inline asms with side effects not visible in the constraint list must be
2354 marked as having side effects. This is done through the use of the
2355 '``sideeffect``' keyword, like so:
2357 .. code-block:: llvm
2359 call void asm sideeffect "eieio", ""()
2361 In some cases inline asms will contain code that will not work unless
2362 the stack is aligned in some way, such as calls or SSE instructions on
2363 x86, yet will not contain code that does that alignment within the asm.
2364 The compiler should make conservative assumptions about what the asm
2365 might contain and should generate its usual stack alignment code in the
2366 prologue if the '``alignstack``' keyword is present:
2368 .. code-block:: llvm
2370 call void asm alignstack "eieio", ""()
2372 Inline asms also support using non-standard assembly dialects. The
2373 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2374 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2375 the only supported dialects. An example is:
2377 .. code-block:: llvm
2379 call void asm inteldialect "eieio", ""()
2381 If multiple keywords appear the '``sideeffect``' keyword must come
2382 first, the '``alignstack``' keyword second and the '``inteldialect``'
2388 The call instructions that wrap inline asm nodes may have a
2389 "``!srcloc``" MDNode attached to it that contains a list of constant
2390 integers. If present, the code generator will use the integer as the
2391 location cookie value when report errors through the ``LLVMContext``
2392 error reporting mechanisms. This allows a front-end to correlate backend
2393 errors that occur with inline asm back to the source code that produced
2396 .. code-block:: llvm
2398 call void asm sideeffect "something bad", ""(), !srcloc !42
2400 !42 = !{ i32 1234567 }
2402 It is up to the front-end to make sense of the magic numbers it places
2403 in the IR. If the MDNode contains multiple constants, the code generator
2404 will use the one that corresponds to the line of the asm that the error
2409 Metadata Nodes and Metadata Strings
2410 -----------------------------------
2412 LLVM IR allows metadata to be attached to instructions in the program
2413 that can convey extra information about the code to the optimizers and
2414 code generator. One example application of metadata is source-level
2415 debug information. There are two metadata primitives: strings and nodes.
2416 All metadata has the ``metadata`` type and is identified in syntax by a
2417 preceding exclamation point ('``!``').
2419 A metadata string is a string surrounded by double quotes. It can
2420 contain any character by escaping non-printable characters with
2421 "``\xx``" where "``xx``" is the two digit hex code. For example:
2424 Metadata nodes are represented with notation similar to structure
2425 constants (a comma separated list of elements, surrounded by braces and
2426 preceded by an exclamation point). Metadata nodes can have any values as
2427 their operand. For example:
2429 .. code-block:: llvm
2431 !{ metadata !"test\00", i32 10}
2433 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2434 metadata nodes, which can be looked up in the module symbol table. For
2437 .. code-block:: llvm
2439 !foo = metadata !{!4, !3}
2441 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2442 function is using two metadata arguments:
2444 .. code-block:: llvm
2446 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2448 Metadata can be attached with an instruction. Here metadata ``!21`` is
2449 attached to the ``add`` instruction using the ``!dbg`` identifier:
2451 .. code-block:: llvm
2453 %indvar.next = add i64 %indvar, 1, !dbg !21
2455 More information about specific metadata nodes recognized by the
2456 optimizers and code generator is found below.
2461 In LLVM IR, memory does not have types, so LLVM's own type system is not
2462 suitable for doing TBAA. Instead, metadata is added to the IR to
2463 describe a type system of a higher level language. This can be used to
2464 implement typical C/C++ TBAA, but it can also be used to implement
2465 custom alias analysis behavior for other languages.
2467 The current metadata format is very simple. TBAA metadata nodes have up
2468 to three fields, e.g.:
2470 .. code-block:: llvm
2472 !0 = metadata !{ metadata !"an example type tree" }
2473 !1 = metadata !{ metadata !"int", metadata !0 }
2474 !2 = metadata !{ metadata !"float", metadata !0 }
2475 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2477 The first field is an identity field. It can be any value, usually a
2478 metadata string, which uniquely identifies the type. The most important
2479 name in the tree is the name of the root node. Two trees with different
2480 root node names are entirely disjoint, even if they have leaves with
2483 The second field identifies the type's parent node in the tree, or is
2484 null or omitted for a root node. A type is considered to alias all of
2485 its descendants and all of its ancestors in the tree. Also, a type is
2486 considered to alias all types in other trees, so that bitcode produced
2487 from multiple front-ends is handled conservatively.
2489 If the third field is present, it's an integer which if equal to 1
2490 indicates that the type is "constant" (meaning
2491 ``pointsToConstantMemory`` should return true; see `other useful
2492 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2494 '``tbaa.struct``' Metadata
2495 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2497 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2498 aggregate assignment operations in C and similar languages, however it
2499 is defined to copy a contiguous region of memory, which is more than
2500 strictly necessary for aggregate types which contain holes due to
2501 padding. Also, it doesn't contain any TBAA information about the fields
2504 ``!tbaa.struct`` metadata can describe which memory subregions in a
2505 memcpy are padding and what the TBAA tags of the struct are.
2507 The current metadata format is very simple. ``!tbaa.struct`` metadata
2508 nodes are a list of operands which are in conceptual groups of three.
2509 For each group of three, the first operand gives the byte offset of a
2510 field in bytes, the second gives its size in bytes, and the third gives
2513 .. code-block:: llvm
2515 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2517 This describes a struct with two fields. The first is at offset 0 bytes
2518 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2519 and has size 4 bytes and has tbaa tag !2.
2521 Note that the fields need not be contiguous. In this example, there is a
2522 4 byte gap between the two fields. This gap represents padding which
2523 does not carry useful data and need not be preserved.
2525 '``fpmath``' Metadata
2526 ^^^^^^^^^^^^^^^^^^^^^
2528 ``fpmath`` metadata may be attached to any instruction of floating point
2529 type. It can be used to express the maximum acceptable error in the
2530 result of that instruction, in ULPs, thus potentially allowing the
2531 compiler to use a more efficient but less accurate method of computing
2532 it. ULP is defined as follows:
2534 If ``x`` is a real number that lies between two finite consecutive
2535 floating-point numbers ``a`` and ``b``, without being equal to one
2536 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2537 distance between the two non-equal finite floating-point numbers
2538 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2540 The metadata node shall consist of a single positive floating point
2541 number representing the maximum relative error, for example:
2543 .. code-block:: llvm
2545 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2547 '``range``' Metadata
2548 ^^^^^^^^^^^^^^^^^^^^
2550 ``range`` metadata may be attached only to loads of integer types. It
2551 expresses the possible ranges the loaded value is in. The ranges are
2552 represented with a flattened list of integers. The loaded value is known
2553 to be in the union of the ranges defined by each consecutive pair. Each
2554 pair has the following properties:
2556 - The type must match the type loaded by the instruction.
2557 - The pair ``a,b`` represents the range ``[a,b)``.
2558 - Both ``a`` and ``b`` are constants.
2559 - The range is allowed to wrap.
2560 - The range should not represent the full or empty set. That is,
2563 In addition, the pairs must be in signed order of the lower bound and
2564 they must be non-contiguous.
2568 .. code-block:: llvm
2570 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2571 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2572 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2573 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2575 !0 = metadata !{ i8 0, i8 2 }
2576 !1 = metadata !{ i8 255, i8 2 }
2577 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2578 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2583 It is sometimes useful to attach information to loop constructs. Currently,
2584 loop metadata is implemented as metadata attached to the branch instruction
2585 in the loop latch block. This type of metadata refer to a metadata node that is
2586 guaranteed to be separate for each loop. The loop identifier metadata is
2587 specified with the name ``llvm.loop``.
2589 The loop identifier metadata is implemented using a metadata that refers to
2590 itself to avoid merging it with any other identifier metadata, e.g.,
2591 during module linkage or function inlining. That is, each loop should refer
2592 to their own identification metadata even if they reside in separate functions.
2593 The following example contains loop identifier metadata for two separate loop
2596 .. code-block:: llvm
2598 !0 = metadata !{ metadata !0 }
2599 !1 = metadata !{ metadata !1 }
2601 The loop identifier metadata can be used to specify additional per-loop
2602 metadata. Any operands after the first operand can be treated as user-defined
2603 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2604 by the loop vectorizer to indicate how many times to unroll the loop:
2606 .. code-block:: llvm
2608 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2610 !0 = metadata !{ metadata !0, metadata !1 }
2611 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2616 Metadata types used to annotate memory accesses with information helpful
2617 for optimizations are prefixed with ``llvm.mem``.
2619 '``llvm.mem.parallel_loop_access``' Metadata
2620 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2622 For a loop to be parallel, in addition to using
2623 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2624 also all of the memory accessing instructions in the loop body need to be
2625 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2626 is at least one memory accessing instruction not marked with the metadata,
2627 the loop must be considered a sequential loop. This causes parallel loops to be
2628 converted to sequential loops due to optimization passes that are unaware of
2629 the parallel semantics and that insert new memory instructions to the loop
2632 Example of a loop that is considered parallel due to its correct use of
2633 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2634 metadata types that refer to the same loop identifier metadata.
2636 .. code-block:: llvm
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 %for.end, label %for.body, !llvm.loop !0
2648 !0 = metadata !{ metadata !0 }
2650 It is also possible to have nested parallel loops. In that case the
2651 memory accesses refer to a list of loop identifier metadata nodes instead of
2652 the loop identifier metadata node directly:
2654 .. code-block:: llvm
2661 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2663 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2665 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2669 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2671 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2673 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2675 outer.for.end: ; preds = %for.body
2677 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2678 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2679 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2681 '``llvm.vectorizer``'
2682 ^^^^^^^^^^^^^^^^^^^^^
2684 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2685 vectorization parameters such as vectorization factor and unroll factor.
2687 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2688 loop identification metadata.
2690 '``llvm.vectorizer.unroll``' Metadata
2691 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2693 This metadata instructs the loop vectorizer to unroll the specified
2694 loop exactly ``N`` times.
2696 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2697 operand is an integer specifying the unroll factor. For example:
2699 .. code-block:: llvm
2701 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2703 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2706 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2707 determined automatically.
2709 '``llvm.vectorizer.width``' Metadata
2710 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2712 This metadata sets the target width of the vectorizer to ``N``. Without
2713 this metadata, the vectorizer will choose a width automatically.
2714 Regardless of this metadata, the vectorizer will only vectorize loops if
2715 it believes it is valid to do so.
2717 The first operand is the string ``llvm.vectorizer.width`` and the second
2718 operand is an integer specifying the width. For example:
2720 .. code-block:: llvm
2722 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2724 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2727 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2730 Module Flags Metadata
2731 =====================
2733 Information about the module as a whole is difficult to convey to LLVM's
2734 subsystems. The LLVM IR isn't sufficient to transmit this information.
2735 The ``llvm.module.flags`` named metadata exists in order to facilitate
2736 this. These flags are in the form of key / value pairs --- much like a
2737 dictionary --- making it easy for any subsystem who cares about a flag to
2740 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2741 Each triplet has the following form:
2743 - The first element is a *behavior* flag, which specifies the behavior
2744 when two (or more) modules are merged together, and it encounters two
2745 (or more) metadata with the same ID. The supported behaviors are
2747 - The second element is a metadata string that is a unique ID for the
2748 metadata. Each module may only have one flag entry for each unique ID (not
2749 including entries with the **Require** behavior).
2750 - The third element is the value of the flag.
2752 When two (or more) modules are merged together, the resulting
2753 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2754 each unique metadata ID string, there will be exactly one entry in the merged
2755 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2756 be determined by the merge behavior flag, as described below. The only exception
2757 is that entries with the *Require* behavior are always preserved.
2759 The following behaviors are supported:
2770 Emits an error if two values disagree, otherwise the resulting value
2771 is that of the operands.
2775 Emits a warning if two values disagree. The result value will be the
2776 operand for the flag from the first module being linked.
2780 Adds a requirement that another module flag be present and have a
2781 specified value after linking is performed. The value must be a
2782 metadata pair, where the first element of the pair is the ID of the
2783 module flag to be restricted, and the second element of the pair is
2784 the value the module flag should be restricted to. This behavior can
2785 be used to restrict the allowable results (via triggering of an
2786 error) of linking IDs with the **Override** behavior.
2790 Uses the specified value, regardless of the behavior or value of the
2791 other module. If both modules specify **Override**, but the values
2792 differ, an error will be emitted.
2796 Appends the two values, which are required to be metadata nodes.
2800 Appends the two values, which are required to be metadata
2801 nodes. However, duplicate entries in the second list are dropped
2802 during the append operation.
2804 It is an error for a particular unique flag ID to have multiple behaviors,
2805 except in the case of **Require** (which adds restrictions on another metadata
2806 value) or **Override**.
2808 An example of module flags:
2810 .. code-block:: llvm
2812 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2813 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2814 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2815 !3 = metadata !{ i32 3, metadata !"qux",
2817 metadata !"foo", i32 1
2820 !llvm.module.flags = !{ !0, !1, !2, !3 }
2822 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2823 if two or more ``!"foo"`` flags are seen is to emit an error if their
2824 values are not equal.
2826 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2827 behavior if two or more ``!"bar"`` flags are seen is to use the value
2830 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2831 behavior if two or more ``!"qux"`` flags are seen is to emit a
2832 warning if their values are not equal.
2834 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2838 metadata !{ metadata !"foo", i32 1 }
2840 The behavior is to emit an error if the ``llvm.module.flags`` does not
2841 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2844 Objective-C Garbage Collection Module Flags Metadata
2845 ----------------------------------------------------
2847 On the Mach-O platform, Objective-C stores metadata about garbage
2848 collection in a special section called "image info". The metadata
2849 consists of a version number and a bitmask specifying what types of
2850 garbage collection are supported (if any) by the file. If two or more
2851 modules are linked together their garbage collection metadata needs to
2852 be merged rather than appended together.
2854 The Objective-C garbage collection module flags metadata consists of the
2855 following key-value pairs:
2864 * - ``Objective-C Version``
2865 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2867 * - ``Objective-C Image Info Version``
2868 - **[Required]** --- The version of the image info section. Currently
2871 * - ``Objective-C Image Info Section``
2872 - **[Required]** --- The section to place the metadata. Valid values are
2873 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2874 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2875 Objective-C ABI version 2.
2877 * - ``Objective-C Garbage Collection``
2878 - **[Required]** --- Specifies whether garbage collection is supported or
2879 not. Valid values are 0, for no garbage collection, and 2, for garbage
2880 collection supported.
2882 * - ``Objective-C GC Only``
2883 - **[Optional]** --- Specifies that only garbage collection is supported.
2884 If present, its value must be 6. This flag requires that the
2885 ``Objective-C Garbage Collection`` flag have the value 2.
2887 Some important flag interactions:
2889 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2890 merged with a module with ``Objective-C Garbage Collection`` set to
2891 2, then the resulting module has the
2892 ``Objective-C Garbage Collection`` flag set to 0.
2893 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2894 merged with a module with ``Objective-C GC Only`` set to 6.
2896 Automatic Linker Flags Module Flags Metadata
2897 --------------------------------------------
2899 Some targets support embedding flags to the linker inside individual object
2900 files. Typically this is used in conjunction with language extensions which
2901 allow source files to explicitly declare the libraries they depend on, and have
2902 these automatically be transmitted to the linker via object files.
2904 These flags are encoded in the IR using metadata in the module flags section,
2905 using the ``Linker Options`` key. The merge behavior for this flag is required
2906 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2907 node which should be a list of other metadata nodes, each of which should be a
2908 list of metadata strings defining linker options.
2910 For example, the following metadata section specifies two separate sets of
2911 linker options, presumably to link against ``libz`` and the ``Cocoa``
2914 !0 = metadata !{ i32 6, metadata !"Linker Options",
2916 metadata !{ metadata !"-lz" },
2917 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2918 !llvm.module.flags = !{ !0 }
2920 The metadata encoding as lists of lists of options, as opposed to a collapsed
2921 list of options, is chosen so that the IR encoding can use multiple option
2922 strings to specify e.g., a single library, while still having that specifier be
2923 preserved as an atomic element that can be recognized by a target specific
2924 assembly writer or object file emitter.
2926 Each individual option is required to be either a valid option for the target's
2927 linker, or an option that is reserved by the target specific assembly writer or
2928 object file emitter. No other aspect of these options is defined by the IR.
2930 .. _intrinsicglobalvariables:
2932 Intrinsic Global Variables
2933 ==========================
2935 LLVM has a number of "magic" global variables that contain data that
2936 affect code generation or other IR semantics. These are documented here.
2937 All globals of this sort should have a section specified as
2938 "``llvm.metadata``". This section and all globals that start with
2939 "``llvm.``" are reserved for use by LLVM.
2943 The '``llvm.used``' Global Variable
2944 -----------------------------------
2946 The ``@llvm.used`` global is an array which has
2947 :ref:`appending linkage <linkage_appending>`. This array contains a list of
2948 pointers to named global variables, functions and aliases which may optionally
2949 have a pointer cast formed of bitcast or getelementptr. For example, a legal
2952 .. code-block:: llvm
2957 @llvm.used = appending global [2 x i8*] [
2959 i8* bitcast (i32* @Y to i8*)
2960 ], section "llvm.metadata"
2962 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
2963 and linker are required to treat the symbol as if there is a reference to the
2964 symbol that it cannot see (which is why they have to be named). For example, if
2965 a variable has internal linkage and no references other than that from the
2966 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
2967 references from inline asms and other things the compiler cannot "see", and
2968 corresponds to "``attribute((used))``" in GNU C.
2970 On some targets, the code generator must emit a directive to the
2971 assembler or object file to prevent the assembler and linker from
2972 molesting the symbol.
2974 .. _gv_llvmcompilerused:
2976 The '``llvm.compiler.used``' Global Variable
2977 --------------------------------------------
2979 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
2980 directive, except that it only prevents the compiler from touching the
2981 symbol. On targets that support it, this allows an intelligent linker to
2982 optimize references to the symbol without being impeded as it would be
2985 This is a rare construct that should only be used in rare circumstances,
2986 and should not be exposed to source languages.
2988 .. _gv_llvmglobalctors:
2990 The '``llvm.global_ctors``' Global Variable
2991 -------------------------------------------
2993 .. code-block:: llvm
2995 %0 = type { i32, void ()* }
2996 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2998 The ``@llvm.global_ctors`` array contains a list of constructor
2999 functions and associated priorities. The functions referenced by this
3000 array will be called in ascending order of priority (i.e. lowest first)
3001 when the module is loaded. The order of functions with the same priority
3004 .. _llvmglobaldtors:
3006 The '``llvm.global_dtors``' Global Variable
3007 -------------------------------------------
3009 .. code-block:: llvm
3011 %0 = type { i32, void ()* }
3012 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3014 The ``@llvm.global_dtors`` array contains a list of destructor functions
3015 and associated priorities. The functions referenced by this array will
3016 be called in descending order of priority (i.e. highest first) when the
3017 module is loaded. The order of functions with the same priority is not
3020 Instruction Reference
3021 =====================
3023 The LLVM instruction set consists of several different classifications
3024 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3025 instructions <binaryops>`, :ref:`bitwise binary
3026 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3027 :ref:`other instructions <otherops>`.
3031 Terminator Instructions
3032 -----------------------
3034 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3035 program ends with a "Terminator" instruction, which indicates which
3036 block should be executed after the current block is finished. These
3037 terminator instructions typically yield a '``void``' value: they produce
3038 control flow, not values (the one exception being the
3039 ':ref:`invoke <i_invoke>`' instruction).
3041 The terminator instructions are: ':ref:`ret <i_ret>`',
3042 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3043 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3044 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3048 '``ret``' Instruction
3049 ^^^^^^^^^^^^^^^^^^^^^
3056 ret <type> <value> ; Return a value from a non-void function
3057 ret void ; Return from void function
3062 The '``ret``' instruction is used to return control flow (and optionally
3063 a value) from a function back to the caller.
3065 There are two forms of the '``ret``' instruction: one that returns a
3066 value and then causes control flow, and one that just causes control
3072 The '``ret``' instruction optionally accepts a single argument, the
3073 return value. The type of the return value must be a ':ref:`first
3074 class <t_firstclass>`' type.
3076 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3077 return type and contains a '``ret``' instruction with no return value or
3078 a return value with a type that does not match its type, or if it has a
3079 void return type and contains a '``ret``' instruction with a return
3085 When the '``ret``' instruction is executed, control flow returns back to
3086 the calling function's context. If the caller is a
3087 ":ref:`call <i_call>`" instruction, execution continues at the
3088 instruction after the call. If the caller was an
3089 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3090 beginning of the "normal" destination block. If the instruction returns
3091 a value, that value shall set the call or invoke instruction's return
3097 .. code-block:: llvm
3099 ret i32 5 ; Return an integer value of 5
3100 ret void ; Return from a void function
3101 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3105 '``br``' Instruction
3106 ^^^^^^^^^^^^^^^^^^^^
3113 br i1 <cond>, label <iftrue>, label <iffalse>
3114 br label <dest> ; Unconditional branch
3119 The '``br``' instruction is used to cause control flow to transfer to a
3120 different basic block in the current function. There are two forms of
3121 this instruction, corresponding to a conditional branch and an
3122 unconditional branch.
3127 The conditional branch form of the '``br``' instruction takes a single
3128 '``i1``' value and two '``label``' values. The unconditional form of the
3129 '``br``' instruction takes a single '``label``' value as a target.
3134 Upon execution of a conditional '``br``' instruction, the '``i1``'
3135 argument is evaluated. If the value is ``true``, control flows to the
3136 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3137 to the '``iffalse``' ``label`` argument.
3142 .. code-block:: llvm
3145 %cond = icmp eq i32 %a, %b
3146 br i1 %cond, label %IfEqual, label %IfUnequal
3154 '``switch``' Instruction
3155 ^^^^^^^^^^^^^^^^^^^^^^^^
3162 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3167 The '``switch``' instruction is used to transfer control flow to one of
3168 several different places. It is a generalization of the '``br``'
3169 instruction, allowing a branch to occur to one of many possible
3175 The '``switch``' instruction uses three parameters: an integer
3176 comparison value '``value``', a default '``label``' destination, and an
3177 array of pairs of comparison value constants and '``label``'s. The table
3178 is not allowed to contain duplicate constant entries.
3183 The ``switch`` instruction specifies a table of values and destinations.
3184 When the '``switch``' instruction is executed, this table is searched
3185 for the given value. If the value is found, control flow is transferred
3186 to the corresponding destination; otherwise, control flow is transferred
3187 to the default destination.
3192 Depending on properties of the target machine and the particular
3193 ``switch`` instruction, this instruction may be code generated in
3194 different ways. For example, it could be generated as a series of
3195 chained conditional branches or with a lookup table.
3200 .. code-block:: llvm
3202 ; Emulate a conditional br instruction
3203 %Val = zext i1 %value to i32
3204 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3206 ; Emulate an unconditional br instruction
3207 switch i32 0, label %dest [ ]
3209 ; Implement a jump table:
3210 switch i32 %val, label %otherwise [ i32 0, label %onzero
3212 i32 2, label %ontwo ]
3216 '``indirectbr``' Instruction
3217 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3224 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3229 The '``indirectbr``' instruction implements an indirect branch to a
3230 label within the current function, whose address is specified by
3231 "``address``". Address must be derived from a
3232 :ref:`blockaddress <blockaddress>` constant.
3237 The '``address``' argument is the address of the label to jump to. The
3238 rest of the arguments indicate the full set of possible destinations
3239 that the address may point to. Blocks are allowed to occur multiple
3240 times in the destination list, though this isn't particularly useful.
3242 This destination list is required so that dataflow analysis has an
3243 accurate understanding of the CFG.
3248 Control transfers to the block specified in the address argument. All
3249 possible destination blocks must be listed in the label list, otherwise
3250 this instruction has undefined behavior. This implies that jumps to
3251 labels defined in other functions have undefined behavior as well.
3256 This is typically implemented with a jump through a register.
3261 .. code-block:: llvm
3263 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3267 '``invoke``' Instruction
3268 ^^^^^^^^^^^^^^^^^^^^^^^^
3275 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3276 to label <normal label> unwind label <exception label>
3281 The '``invoke``' instruction causes control to transfer to a specified
3282 function, with the possibility of control flow transfer to either the
3283 '``normal``' label or the '``exception``' label. If the callee function
3284 returns with the "``ret``" instruction, control flow will return to the
3285 "normal" label. If the callee (or any indirect callees) returns via the
3286 ":ref:`resume <i_resume>`" instruction or other exception handling
3287 mechanism, control is interrupted and continued at the dynamically
3288 nearest "exception" label.
3290 The '``exception``' label is a `landing
3291 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3292 '``exception``' label is required to have the
3293 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3294 information about the behavior of the program after unwinding happens,
3295 as its first non-PHI instruction. The restrictions on the
3296 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3297 instruction, so that the important information contained within the
3298 "``landingpad``" instruction can't be lost through normal code motion.
3303 This instruction requires several arguments:
3305 #. The optional "cconv" marker indicates which :ref:`calling
3306 convention <callingconv>` the call should use. If none is
3307 specified, the call defaults to using C calling conventions.
3308 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3309 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3311 #. '``ptr to function ty``': shall be the signature of the pointer to
3312 function value being invoked. In most cases, this is a direct
3313 function invocation, but indirect ``invoke``'s are just as possible,
3314 branching off an arbitrary pointer to function value.
3315 #. '``function ptr val``': An LLVM value containing a pointer to a
3316 function to be invoked.
3317 #. '``function args``': argument list whose types match the function
3318 signature argument types and parameter attributes. All arguments must
3319 be of :ref:`first class <t_firstclass>` type. If the function signature
3320 indicates the function accepts a variable number of arguments, the
3321 extra arguments can be specified.
3322 #. '``normal label``': the label reached when the called function
3323 executes a '``ret``' instruction.
3324 #. '``exception label``': the label reached when a callee returns via
3325 the :ref:`resume <i_resume>` instruction or other exception handling
3327 #. The optional :ref:`function attributes <fnattrs>` list. Only
3328 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3329 attributes are valid here.
3334 This instruction is designed to operate as a standard '``call``'
3335 instruction in most regards. The primary difference is that it
3336 establishes an association with a label, which is used by the runtime
3337 library to unwind the stack.
3339 This instruction is used in languages with destructors to ensure that
3340 proper cleanup is performed in the case of either a ``longjmp`` or a
3341 thrown exception. Additionally, this is important for implementation of
3342 '``catch``' clauses in high-level languages that support them.
3344 For the purposes of the SSA form, the definition of the value returned
3345 by the '``invoke``' instruction is deemed to occur on the edge from the
3346 current block to the "normal" label. If the callee unwinds then no
3347 return value is available.
3352 .. code-block:: llvm
3354 %retval = invoke i32 @Test(i32 15) to label %Continue
3355 unwind label %TestCleanup ; {i32}:retval set
3356 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3357 unwind label %TestCleanup ; {i32}:retval set
3361 '``resume``' Instruction
3362 ^^^^^^^^^^^^^^^^^^^^^^^^
3369 resume <type> <value>
3374 The '``resume``' instruction is a terminator instruction that has no
3380 The '``resume``' instruction requires one argument, which must have the
3381 same type as the result of any '``landingpad``' instruction in the same
3387 The '``resume``' instruction resumes propagation of an existing
3388 (in-flight) exception whose unwinding was interrupted with a
3389 :ref:`landingpad <i_landingpad>` instruction.
3394 .. code-block:: llvm
3396 resume { i8*, i32 } %exn
3400 '``unreachable``' Instruction
3401 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3413 The '``unreachable``' instruction has no defined semantics. This
3414 instruction is used to inform the optimizer that a particular portion of
3415 the code is not reachable. This can be used to indicate that the code
3416 after a no-return function cannot be reached, and other facts.
3421 The '``unreachable``' instruction has no defined semantics.
3428 Binary operators are used to do most of the computation in a program.
3429 They require two operands of the same type, execute an operation on
3430 them, and produce a single value. The operands might represent multiple
3431 data, as is the case with the :ref:`vector <t_vector>` data type. The
3432 result value has the same type as its operands.
3434 There are several different binary operators:
3438 '``add``' Instruction
3439 ^^^^^^^^^^^^^^^^^^^^^
3446 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3447 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3448 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3449 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3454 The '``add``' instruction returns the sum of its two operands.
3459 The two arguments to the '``add``' instruction must be
3460 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3461 arguments must have identical types.
3466 The value produced is the integer sum of the two operands.
3468 If the sum has unsigned overflow, the result returned is the
3469 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3472 Because LLVM integers use a two's complement representation, this
3473 instruction is appropriate for both signed and unsigned integers.
3475 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3476 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3477 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3478 unsigned and/or signed overflow, respectively, occurs.
3483 .. code-block:: llvm
3485 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3489 '``fadd``' Instruction
3490 ^^^^^^^^^^^^^^^^^^^^^^
3497 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3502 The '``fadd``' instruction returns the sum of its two operands.
3507 The two arguments to the '``fadd``' instruction must be :ref:`floating
3508 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3509 Both arguments must have identical types.
3514 The value produced is the floating point sum of the two operands. This
3515 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3516 which are optimization hints to enable otherwise unsafe floating point
3522 .. code-block:: llvm
3524 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3526 '``sub``' Instruction
3527 ^^^^^^^^^^^^^^^^^^^^^
3534 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3535 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3536 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3537 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3542 The '``sub``' instruction returns the difference of its two operands.
3544 Note that the '``sub``' instruction is used to represent the '``neg``'
3545 instruction present in most other intermediate representations.
3550 The two arguments to the '``sub``' instruction must be
3551 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3552 arguments must have identical types.
3557 The value produced is the integer difference of the two operands.
3559 If the difference has unsigned overflow, the result returned is the
3560 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3563 Because LLVM integers use a two's complement representation, this
3564 instruction is appropriate for both signed and unsigned integers.
3566 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3567 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3568 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3569 unsigned and/or signed overflow, respectively, occurs.
3574 .. code-block:: llvm
3576 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3577 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3581 '``fsub``' Instruction
3582 ^^^^^^^^^^^^^^^^^^^^^^
3589 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3594 The '``fsub``' instruction returns the difference of its two operands.
3596 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3597 instruction present in most other intermediate representations.
3602 The two arguments to the '``fsub``' instruction must be :ref:`floating
3603 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3604 Both arguments must have identical types.
3609 The value produced is the floating point difference of the two operands.
3610 This instruction can also take any number of :ref:`fast-math
3611 flags <fastmath>`, which are optimization hints to enable otherwise
3612 unsafe floating point optimizations:
3617 .. code-block:: llvm
3619 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3620 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3622 '``mul``' Instruction
3623 ^^^^^^^^^^^^^^^^^^^^^
3630 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3631 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3632 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3633 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3638 The '``mul``' instruction returns the product of its two operands.
3643 The two arguments to the '``mul``' instruction must be
3644 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3645 arguments must have identical types.
3650 The value produced is the integer product of the two operands.
3652 If the result of the multiplication has unsigned overflow, the result
3653 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3654 bit width of the result.
3656 Because LLVM integers use a two's complement representation, and the
3657 result is the same width as the operands, this instruction returns the
3658 correct result for both signed and unsigned integers. If a full product
3659 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3660 sign-extended or zero-extended as appropriate to the width of the full
3663 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3664 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3665 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3666 unsigned and/or signed overflow, respectively, occurs.
3671 .. code-block:: llvm
3673 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3677 '``fmul``' Instruction
3678 ^^^^^^^^^^^^^^^^^^^^^^
3685 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3690 The '``fmul``' instruction returns the product of its two operands.
3695 The two arguments to the '``fmul``' instruction must be :ref:`floating
3696 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3697 Both arguments must have identical types.
3702 The value produced is the floating point product of the two operands.
3703 This instruction can also take any number of :ref:`fast-math
3704 flags <fastmath>`, which are optimization hints to enable otherwise
3705 unsafe floating point optimizations:
3710 .. code-block:: llvm
3712 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3714 '``udiv``' Instruction
3715 ^^^^^^^^^^^^^^^^^^^^^^
3722 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3723 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3728 The '``udiv``' instruction returns the quotient of its two operands.
3733 The two arguments to the '``udiv``' instruction must be
3734 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3735 arguments must have identical types.
3740 The value produced is the unsigned integer quotient of the two operands.
3742 Note that unsigned integer division and signed integer division are
3743 distinct operations; for signed integer division, use '``sdiv``'.
3745 Division by zero leads to undefined behavior.
3747 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3748 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3749 such, "((a udiv exact b) mul b) == a").
3754 .. code-block:: llvm
3756 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3758 '``sdiv``' Instruction
3759 ^^^^^^^^^^^^^^^^^^^^^^
3766 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3767 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3772 The '``sdiv``' instruction returns the quotient of its two operands.
3777 The two arguments to the '``sdiv``' instruction must be
3778 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3779 arguments must have identical types.
3784 The value produced is the signed integer quotient of the two operands
3785 rounded towards zero.
3787 Note that signed integer division and unsigned integer division are
3788 distinct operations; for unsigned integer division, use '``udiv``'.
3790 Division by zero leads to undefined behavior. Overflow also leads to
3791 undefined behavior; this is a rare case, but can occur, for example, by
3792 doing a 32-bit division of -2147483648 by -1.
3794 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3795 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3800 .. code-block:: llvm
3802 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3806 '``fdiv``' Instruction
3807 ^^^^^^^^^^^^^^^^^^^^^^
3814 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3819 The '``fdiv``' instruction returns the quotient of its two operands.
3824 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3825 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3826 Both arguments must have identical types.
3831 The value produced is the floating point quotient of the two operands.
3832 This instruction can also take any number of :ref:`fast-math
3833 flags <fastmath>`, which are optimization hints to enable otherwise
3834 unsafe floating point optimizations:
3839 .. code-block:: llvm
3841 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3843 '``urem``' Instruction
3844 ^^^^^^^^^^^^^^^^^^^^^^
3851 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3856 The '``urem``' instruction returns the remainder from the unsigned
3857 division of its two arguments.
3862 The two arguments to the '``urem``' instruction must be
3863 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3864 arguments must have identical types.
3869 This instruction returns the unsigned integer *remainder* of a division.
3870 This instruction always performs an unsigned division to get the
3873 Note that unsigned integer remainder and signed integer remainder are
3874 distinct operations; for signed integer remainder, use '``srem``'.
3876 Taking the remainder of a division by zero leads to undefined behavior.
3881 .. code-block:: llvm
3883 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3885 '``srem``' Instruction
3886 ^^^^^^^^^^^^^^^^^^^^^^
3893 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3898 The '``srem``' instruction returns the remainder from the signed
3899 division of its two operands. This instruction can also take
3900 :ref:`vector <t_vector>` versions of the values in which case the elements
3906 The two arguments to the '``srem``' instruction must be
3907 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3908 arguments must have identical types.
3913 This instruction returns the *remainder* of a division (where the result
3914 is either zero or has the same sign as the dividend, ``op1``), not the
3915 *modulo* operator (where the result is either zero or has the same sign
3916 as the divisor, ``op2``) of a value. For more information about the
3917 difference, see `The Math
3918 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3919 table of how this is implemented in various languages, please see
3921 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3923 Note that signed integer remainder and unsigned integer remainder are
3924 distinct operations; for unsigned integer remainder, use '``urem``'.
3926 Taking the remainder of a division by zero leads to undefined behavior.
3927 Overflow also leads to undefined behavior; this is a rare case, but can
3928 occur, for example, by taking the remainder of a 32-bit division of
3929 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3930 rule lets srem be implemented using instructions that return both the
3931 result of the division and the remainder.)
3936 .. code-block:: llvm
3938 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
3942 '``frem``' Instruction
3943 ^^^^^^^^^^^^^^^^^^^^^^
3950 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3955 The '``frem``' instruction returns the remainder from the division of
3961 The two arguments to the '``frem``' instruction must be :ref:`floating
3962 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3963 Both arguments must have identical types.
3968 This instruction returns the *remainder* of a division. The remainder
3969 has the same sign as the dividend. This instruction can also take any
3970 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
3971 to enable otherwise unsafe floating point optimizations:
3976 .. code-block:: llvm
3978 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
3982 Bitwise Binary Operations
3983 -------------------------
3985 Bitwise binary operators are used to do various forms of bit-twiddling
3986 in a program. They are generally very efficient instructions and can
3987 commonly be strength reduced from other instructions. They require two
3988 operands of the same type, execute an operation on them, and produce a
3989 single value. The resulting value is the same type as its operands.
3991 '``shl``' Instruction
3992 ^^^^^^^^^^^^^^^^^^^^^
3999 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4000 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4001 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4002 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4007 The '``shl``' instruction returns the first operand shifted to the left
4008 a specified number of bits.
4013 Both arguments to the '``shl``' instruction must be the same
4014 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4015 '``op2``' is treated as an unsigned value.
4020 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4021 where ``n`` is the width of the result. If ``op2`` is (statically or
4022 dynamically) negative or equal to or larger than the number of bits in
4023 ``op1``, the result is undefined. If the arguments are vectors, each
4024 vector element of ``op1`` is shifted by the corresponding shift amount
4027 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4028 value <poisonvalues>` if it shifts out any non-zero bits. If the
4029 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4030 value <poisonvalues>` if it shifts out any bits that disagree with the
4031 resultant sign bit. As such, NUW/NSW have the same semantics as they
4032 would if the shift were expressed as a mul instruction with the same
4033 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4038 .. code-block:: llvm
4040 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4041 <result> = shl i32 4, 2 ; yields {i32}: 16
4042 <result> = shl i32 1, 10 ; yields {i32}: 1024
4043 <result> = shl i32 1, 32 ; undefined
4044 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4046 '``lshr``' Instruction
4047 ^^^^^^^^^^^^^^^^^^^^^^
4054 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4055 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4060 The '``lshr``' instruction (logical shift right) returns the first
4061 operand shifted to the right a specified number of bits with zero fill.
4066 Both arguments to the '``lshr``' instruction must be the same
4067 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4068 '``op2``' is treated as an unsigned value.
4073 This instruction always performs a logical shift right operation. The
4074 most significant bits of the result will be filled with zero bits after
4075 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4076 than the number of bits in ``op1``, the result is undefined. If the
4077 arguments are vectors, each vector element of ``op1`` is shifted by the
4078 corresponding shift amount in ``op2``.
4080 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4081 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4087 .. code-block:: llvm
4089 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4090 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4091 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4092 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4093 <result> = lshr i32 1, 32 ; undefined
4094 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4096 '``ashr``' Instruction
4097 ^^^^^^^^^^^^^^^^^^^^^^
4104 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4105 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4110 The '``ashr``' instruction (arithmetic shift right) returns the first
4111 operand shifted to the right a specified number of bits with sign
4117 Both arguments to the '``ashr``' instruction must be the same
4118 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4119 '``op2``' is treated as an unsigned value.
4124 This instruction always performs an arithmetic shift right operation,
4125 The most significant bits of the result will be filled with the sign bit
4126 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4127 than the number of bits in ``op1``, the result is undefined. If the
4128 arguments are vectors, each vector element of ``op1`` is shifted by the
4129 corresponding shift amount in ``op2``.
4131 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4132 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4138 .. code-block:: llvm
4140 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4141 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4142 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4143 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4144 <result> = ashr i32 1, 32 ; undefined
4145 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4147 '``and``' Instruction
4148 ^^^^^^^^^^^^^^^^^^^^^
4155 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4160 The '``and``' instruction returns the bitwise logical and of its two
4166 The two arguments to the '``and``' instruction must be
4167 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4168 arguments must have identical types.
4173 The truth table used for the '``and``' instruction is:
4190 .. code-block:: llvm
4192 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4193 <result> = and i32 15, 40 ; yields {i32}:result = 8
4194 <result> = and i32 4, 8 ; yields {i32}:result = 0
4196 '``or``' Instruction
4197 ^^^^^^^^^^^^^^^^^^^^
4204 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4209 The '``or``' instruction returns the bitwise logical inclusive or of its
4215 The two arguments to the '``or``' instruction must be
4216 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4217 arguments must have identical types.
4222 The truth table used for the '``or``' instruction is:
4241 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4242 <result> = or i32 15, 40 ; yields {i32}:result = 47
4243 <result> = or i32 4, 8 ; yields {i32}:result = 12
4245 '``xor``' Instruction
4246 ^^^^^^^^^^^^^^^^^^^^^
4253 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4258 The '``xor``' instruction returns the bitwise logical exclusive or of
4259 its two operands. The ``xor`` is used to implement the "one's
4260 complement" operation, which is the "~" operator in C.
4265 The two arguments to the '``xor``' instruction must be
4266 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4267 arguments must have identical types.
4272 The truth table used for the '``xor``' instruction is:
4289 .. code-block:: llvm
4291 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4292 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4293 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4294 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4299 LLVM supports several instructions to represent vector operations in a
4300 target-independent manner. These instructions cover the element-access
4301 and vector-specific operations needed to process vectors effectively.
4302 While LLVM does directly support these vector operations, many
4303 sophisticated algorithms will want to use target-specific intrinsics to
4304 take full advantage of a specific target.
4306 .. _i_extractelement:
4308 '``extractelement``' Instruction
4309 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4316 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4321 The '``extractelement``' instruction extracts a single scalar element
4322 from a vector at a specified index.
4327 The first operand of an '``extractelement``' instruction is a value of
4328 :ref:`vector <t_vector>` type. The second operand is an index indicating
4329 the position from which to extract the element. The index may be a
4335 The result is a scalar of the same type as the element type of ``val``.
4336 Its value is the value at position ``idx`` of ``val``. If ``idx``
4337 exceeds the length of ``val``, the results are undefined.
4342 .. code-block:: llvm
4344 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4346 .. _i_insertelement:
4348 '``insertelement``' Instruction
4349 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4356 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4361 The '``insertelement``' instruction inserts a scalar element into a
4362 vector at a specified index.
4367 The first operand of an '``insertelement``' instruction is a value of
4368 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4369 type must equal the element type of the first operand. The third operand
4370 is an index indicating the position at which to insert the value. The
4371 index may be a variable.
4376 The result is a vector of the same type as ``val``. Its element values
4377 are those of ``val`` except at position ``idx``, where it gets the value
4378 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4384 .. code-block:: llvm
4386 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4388 .. _i_shufflevector:
4390 '``shufflevector``' Instruction
4391 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4398 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4403 The '``shufflevector``' instruction constructs a permutation of elements
4404 from two input vectors, returning a vector with the same element type as
4405 the input and length that is the same as the shuffle mask.
4410 The first two operands of a '``shufflevector``' instruction are vectors
4411 with the same type. The third argument is a shuffle mask whose element
4412 type is always 'i32'. The result of the instruction is a vector whose
4413 length is the same as the shuffle mask and whose element type is the
4414 same as the element type of the first two operands.
4416 The shuffle mask operand is required to be a constant vector with either
4417 constant integer or undef values.
4422 The elements of the two input vectors are numbered from left to right
4423 across both of the vectors. The shuffle mask operand specifies, for each
4424 element of the result vector, which element of the two input vectors the
4425 result element gets. The element selector may be undef (meaning "don't
4426 care") and the second operand may be undef if performing a shuffle from
4432 .. code-block:: llvm
4434 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4435 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4436 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4437 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4438 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4439 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4440 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4441 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4443 Aggregate Operations
4444 --------------------
4446 LLVM supports several instructions for working with
4447 :ref:`aggregate <t_aggregate>` values.
4451 '``extractvalue``' Instruction
4452 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4459 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4464 The '``extractvalue``' instruction extracts the value of a member field
4465 from an :ref:`aggregate <t_aggregate>` value.
4470 The first operand of an '``extractvalue``' instruction is a value of
4471 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4472 constant indices to specify which value to extract in a similar manner
4473 as indices in a '``getelementptr``' instruction.
4475 The major differences to ``getelementptr`` indexing are:
4477 - Since the value being indexed is not a pointer, the first index is
4478 omitted and assumed to be zero.
4479 - At least one index must be specified.
4480 - Not only struct indices but also array indices must be in bounds.
4485 The result is the value at the position in the aggregate specified by
4491 .. code-block:: llvm
4493 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4497 '``insertvalue``' Instruction
4498 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4505 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4510 The '``insertvalue``' instruction inserts a value into a member field in
4511 an :ref:`aggregate <t_aggregate>` value.
4516 The first operand of an '``insertvalue``' instruction is a value of
4517 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4518 a first-class value to insert. The following operands are constant
4519 indices indicating the position at which to insert the value in a
4520 similar manner as indices in a '``extractvalue``' instruction. The value
4521 to insert must have the same type as the value identified by the
4527 The result is an aggregate of the same type as ``val``. Its value is
4528 that of ``val`` except that the value at the position specified by the
4529 indices is that of ``elt``.
4534 .. code-block:: llvm
4536 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4537 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4538 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4542 Memory Access and Addressing Operations
4543 ---------------------------------------
4545 A key design point of an SSA-based representation is how it represents
4546 memory. In LLVM, no memory locations are in SSA form, which makes things
4547 very simple. This section describes how to read, write, and allocate
4552 '``alloca``' Instruction
4553 ^^^^^^^^^^^^^^^^^^^^^^^^
4560 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4565 The '``alloca``' instruction allocates memory on the stack frame of the
4566 currently executing function, to be automatically released when this
4567 function returns to its caller. The object is always allocated in the
4568 generic address space (address space zero).
4573 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4574 bytes of memory on the runtime stack, returning a pointer of the
4575 appropriate type to the program. If "NumElements" is specified, it is
4576 the number of elements allocated, otherwise "NumElements" is defaulted
4577 to be one. If a constant alignment is specified, the value result of the
4578 allocation is guaranteed to be aligned to at least that boundary. If not
4579 specified, or if zero, the target can choose to align the allocation on
4580 any convenient boundary compatible with the type.
4582 '``type``' may be any sized type.
4587 Memory is allocated; a pointer is returned. The operation is undefined
4588 if there is insufficient stack space for the allocation. '``alloca``'d
4589 memory is automatically released when the function returns. The
4590 '``alloca``' instruction is commonly used to represent automatic
4591 variables that must have an address available. When the function returns
4592 (either with the ``ret`` or ``resume`` instructions), the memory is
4593 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4594 The order in which memory is allocated (ie., which way the stack grows)
4600 .. code-block:: llvm
4602 %ptr = alloca i32 ; yields {i32*}:ptr
4603 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4604 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4605 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4609 '``load``' Instruction
4610 ^^^^^^^^^^^^^^^^^^^^^^
4617 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4618 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4619 !<index> = !{ i32 1 }
4624 The '``load``' instruction is used to read from memory.
4629 The argument to the ``load`` instruction specifies the memory address
4630 from which to load. The pointer must point to a :ref:`first
4631 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4632 then the optimizer is not allowed to modify the number or order of
4633 execution of this ``load`` with other :ref:`volatile
4634 operations <volatile>`.
4636 If the ``load`` is marked as ``atomic``, it takes an extra
4637 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4638 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4639 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4640 when they may see multiple atomic stores. The type of the pointee must
4641 be an integer type whose bit width is a power of two greater than or
4642 equal to eight and less than or equal to a target-specific size limit.
4643 ``align`` must be explicitly specified on atomic loads, and the load has
4644 undefined behavior if the alignment is not set to a value which is at
4645 least the size in bytes of the pointee. ``!nontemporal`` does not have
4646 any defined semantics for atomic loads.
4648 The optional constant ``align`` argument specifies the alignment of the
4649 operation (that is, the alignment of the memory address). A value of 0
4650 or an omitted ``align`` argument means that the operation has the ABI
4651 alignment for the target. It is the responsibility of the code emitter
4652 to ensure that the alignment information is correct. Overestimating the
4653 alignment results in undefined behavior. Underestimating the alignment
4654 may produce less efficient code. An alignment of 1 is always safe.
4656 The optional ``!nontemporal`` metadata must reference a single
4657 metadata name ``<index>`` corresponding to a metadata node with one
4658 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4659 metadata on the instruction tells the optimizer and code generator
4660 that this load is not expected to be reused in the cache. The code
4661 generator may select special instructions to save cache bandwidth, such
4662 as the ``MOVNT`` instruction on x86.
4664 The optional ``!invariant.load`` metadata must reference a single
4665 metadata name ``<index>`` corresponding to a metadata node with no
4666 entries. The existence of the ``!invariant.load`` metadata on the
4667 instruction tells the optimizer and code generator that this load
4668 address points to memory which does not change value during program
4669 execution. The optimizer may then move this load around, for example, by
4670 hoisting it out of loops using loop invariant code motion.
4675 The location of memory pointed to is loaded. If the value being loaded
4676 is of scalar type then the number of bytes read does not exceed the
4677 minimum number of bytes needed to hold all bits of the type. For
4678 example, loading an ``i24`` reads at most three bytes. When loading a
4679 value of a type like ``i20`` with a size that is not an integral number
4680 of bytes, the result is undefined if the value was not originally
4681 written using a store of the same type.
4686 .. code-block:: llvm
4688 %ptr = alloca i32 ; yields {i32*}:ptr
4689 store i32 3, i32* %ptr ; yields {void}
4690 %val = load i32* %ptr ; yields {i32}:val = i32 3
4694 '``store``' Instruction
4695 ^^^^^^^^^^^^^^^^^^^^^^^
4702 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4703 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4708 The '``store``' instruction is used to write to memory.
4713 There are two arguments to the ``store`` instruction: a value to store
4714 and an address at which to store it. The type of the ``<pointer>``
4715 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4716 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4717 then the optimizer is not allowed to modify the number or order of
4718 execution of this ``store`` with other :ref:`volatile
4719 operations <volatile>`.
4721 If the ``store`` is marked as ``atomic``, it takes an extra
4722 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4723 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4724 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4725 when they may see multiple atomic stores. The type of the pointee must
4726 be an integer type whose bit width is a power of two greater than or
4727 equal to eight and less than or equal to a target-specific size limit.
4728 ``align`` must be explicitly specified on atomic stores, and the store
4729 has undefined behavior if the alignment is not set to a value which is
4730 at least the size in bytes of the pointee. ``!nontemporal`` does not
4731 have any defined semantics for atomic stores.
4733 The optional constant ``align`` argument specifies the alignment of the
4734 operation (that is, the alignment of the memory address). A value of 0
4735 or an omitted ``align`` argument means that the operation has the ABI
4736 alignment for the target. It is the responsibility of the code emitter
4737 to ensure that the alignment information is correct. Overestimating the
4738 alignment results in undefined behavior. Underestimating the
4739 alignment may produce less efficient code. An alignment of 1 is always
4742 The optional ``!nontemporal`` metadata must reference a single metadata
4743 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4744 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4745 tells the optimizer and code generator that this load is not expected to
4746 be reused in the cache. The code generator may select special
4747 instructions to save cache bandwidth, such as the MOVNT instruction on
4753 The contents of memory are updated to contain ``<value>`` at the
4754 location specified by the ``<pointer>`` operand. If ``<value>`` is
4755 of scalar type then the number of bytes written does not exceed the
4756 minimum number of bytes needed to hold all bits of the type. For
4757 example, storing an ``i24`` writes at most three bytes. When writing a
4758 value of a type like ``i20`` with a size that is not an integral number
4759 of bytes, it is unspecified what happens to the extra bits that do not
4760 belong to the type, but they will typically be overwritten.
4765 .. code-block:: llvm
4767 %ptr = alloca i32 ; yields {i32*}:ptr
4768 store i32 3, i32* %ptr ; yields {void}
4769 %val = load i32* %ptr ; yields {i32}:val = i32 3
4773 '``fence``' Instruction
4774 ^^^^^^^^^^^^^^^^^^^^^^^
4781 fence [singlethread] <ordering> ; yields {void}
4786 The '``fence``' instruction is used to introduce happens-before edges
4792 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4793 defines what *synchronizes-with* edges they add. They can only be given
4794 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4799 A fence A which has (at least) ``release`` ordering semantics
4800 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4801 semantics if and only if there exist atomic operations X and Y, both
4802 operating on some atomic object M, such that A is sequenced before X, X
4803 modifies M (either directly or through some side effect of a sequence
4804 headed by X), Y is sequenced before B, and Y observes M. This provides a
4805 *happens-before* dependency between A and B. Rather than an explicit
4806 ``fence``, one (but not both) of the atomic operations X or Y might
4807 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4808 still *synchronize-with* the explicit ``fence`` and establish the
4809 *happens-before* edge.
4811 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4812 ``acquire`` and ``release`` semantics specified above, participates in
4813 the global program order of other ``seq_cst`` operations and/or fences.
4815 The optional ":ref:`singlethread <singlethread>`" argument specifies
4816 that the fence only synchronizes with other fences in the same thread.
4817 (This is useful for interacting with signal handlers.)
4822 .. code-block:: llvm
4824 fence acquire ; yields {void}
4825 fence singlethread seq_cst ; yields {void}
4829 '``cmpxchg``' Instruction
4830 ^^^^^^^^^^^^^^^^^^^^^^^^^
4837 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4842 The '``cmpxchg``' instruction is used to atomically modify memory. It
4843 loads a value in memory and compares it to a given value. If they are
4844 equal, it stores a new value into the memory.
4849 There are three arguments to the '``cmpxchg``' instruction: an address
4850 to operate on, a value to compare to the value currently be at that
4851 address, and a new value to place at that address if the compared values
4852 are equal. The type of '<cmp>' must be an integer type whose bit width
4853 is a power of two greater than or equal to eight and less than or equal
4854 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4855 type, and the type of '<pointer>' must be a pointer to that type. If the
4856 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4857 to modify the number or order of execution of this ``cmpxchg`` with
4858 other :ref:`volatile operations <volatile>`.
4860 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4861 synchronizes with other atomic operations.
4863 The optional "``singlethread``" argument declares that the ``cmpxchg``
4864 is only atomic with respect to code (usually signal handlers) running in
4865 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4866 respect to all other code in the system.
4868 The pointer passed into cmpxchg must have alignment greater than or
4869 equal to the size in memory of the operand.
4874 The contents of memory at the location specified by the '``<pointer>``'
4875 operand is read and compared to '``<cmp>``'; if the read value is the
4876 equal, '``<new>``' is written. The original value at the location is
4879 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4880 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4881 atomic load with an ordering parameter determined by dropping any
4882 ``release`` part of the ``cmpxchg``'s ordering.
4887 .. code-block:: llvm
4890 %orig = atomic load i32* %ptr unordered ; yields {i32}
4894 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4895 %squared = mul i32 %cmp, %cmp
4896 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4897 %success = icmp eq i32 %cmp, %old
4898 br i1 %success, label %done, label %loop
4905 '``atomicrmw``' Instruction
4906 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4913 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4918 The '``atomicrmw``' instruction is used to atomically modify memory.
4923 There are three arguments to the '``atomicrmw``' instruction: an
4924 operation to apply, an address whose value to modify, an argument to the
4925 operation. The operation must be one of the following keywords:
4939 The type of '<value>' must be an integer type whose bit width is a power
4940 of two greater than or equal to eight and less than or equal to a
4941 target-specific size limit. The type of the '``<pointer>``' operand must
4942 be a pointer to that type. If the ``atomicrmw`` is marked as
4943 ``volatile``, then the optimizer is not allowed to modify the number or
4944 order of execution of this ``atomicrmw`` with other :ref:`volatile
4945 operations <volatile>`.
4950 The contents of memory at the location specified by the '``<pointer>``'
4951 operand are atomically read, modified, and written back. The original
4952 value at the location is returned. The modification is specified by the
4955 - xchg: ``*ptr = val``
4956 - add: ``*ptr = *ptr + val``
4957 - sub: ``*ptr = *ptr - val``
4958 - and: ``*ptr = *ptr & val``
4959 - nand: ``*ptr = ~(*ptr & val)``
4960 - or: ``*ptr = *ptr | val``
4961 - xor: ``*ptr = *ptr ^ val``
4962 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
4963 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
4964 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
4966 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
4972 .. code-block:: llvm
4974 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
4976 .. _i_getelementptr:
4978 '``getelementptr``' Instruction
4979 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4986 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4987 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4988 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
4993 The '``getelementptr``' instruction is used to get the address of a
4994 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
4995 address calculation only and does not access memory.
5000 The first argument is always a pointer or a vector of pointers, and
5001 forms the basis of the calculation. The remaining arguments are indices
5002 that indicate which of the elements of the aggregate object are indexed.
5003 The interpretation of each index is dependent on the type being indexed
5004 into. The first index always indexes the pointer value given as the
5005 first argument, the second index indexes a value of the type pointed to
5006 (not necessarily the value directly pointed to, since the first index
5007 can be non-zero), etc. The first type indexed into must be a pointer
5008 value, subsequent types can be arrays, vectors, and structs. Note that
5009 subsequent types being indexed into can never be pointers, since that
5010 would require loading the pointer before continuing calculation.
5012 The type of each index argument depends on the type it is indexing into.
5013 When indexing into a (optionally packed) structure, only ``i32`` integer
5014 **constants** are allowed (when using a vector of indices they must all
5015 be the **same** ``i32`` integer constant). When indexing into an array,
5016 pointer or vector, integers of any width are allowed, and they are not
5017 required to be constant. These integers are treated as signed values
5020 For example, let's consider a C code fragment and how it gets compiled
5036 int *foo(struct ST *s) {
5037 return &s[1].Z.B[5][13];
5040 The LLVM code generated by Clang is:
5042 .. code-block:: llvm
5044 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5045 %struct.ST = type { i32, double, %struct.RT }
5047 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5049 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5056 In the example above, the first index is indexing into the
5057 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5058 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5059 indexes into the third element of the structure, yielding a
5060 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5061 structure. The third index indexes into the second element of the
5062 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5063 dimensions of the array are subscripted into, yielding an '``i32``'
5064 type. The '``getelementptr``' instruction returns a pointer to this
5065 element, thus computing a value of '``i32*``' type.
5067 Note that it is perfectly legal to index partially through a structure,
5068 returning a pointer to an inner element. Because of this, the LLVM code
5069 for the given testcase is equivalent to:
5071 .. code-block:: llvm
5073 define i32* @foo(%struct.ST* %s) {
5074 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5075 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5076 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5077 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5078 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5082 If the ``inbounds`` keyword is present, the result value of the
5083 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5084 pointer is not an *in bounds* address of an allocated object, or if any
5085 of the addresses that would be formed by successive addition of the
5086 offsets implied by the indices to the base address with infinitely
5087 precise signed arithmetic are not an *in bounds* address of that
5088 allocated object. The *in bounds* addresses for an allocated object are
5089 all the addresses that point into the object, plus the address one byte
5090 past the end. In cases where the base is a vector of pointers the
5091 ``inbounds`` keyword applies to each of the computations element-wise.
5093 If the ``inbounds`` keyword is not present, the offsets are added to the
5094 base address with silently-wrapping two's complement arithmetic. If the
5095 offsets have a different width from the pointer, they are sign-extended
5096 or truncated to the width of the pointer. The result value of the
5097 ``getelementptr`` may be outside the object pointed to by the base
5098 pointer. The result value may not necessarily be used to access memory
5099 though, even if it happens to point into allocated storage. See the
5100 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5103 The getelementptr instruction is often confusing. For some more insight
5104 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5109 .. code-block:: llvm
5111 ; yields [12 x i8]*:aptr
5112 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5114 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5116 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5118 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5120 In cases where the pointer argument is a vector of pointers, each index
5121 must be a vector with the same number of elements. For example:
5123 .. code-block:: llvm
5125 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5127 Conversion Operations
5128 ---------------------
5130 The instructions in this category are the conversion instructions
5131 (casting) which all take a single operand and a type. They perform
5132 various bit conversions on the operand.
5134 '``trunc .. to``' Instruction
5135 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5142 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5147 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5152 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5153 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5154 of the same number of integers. The bit size of the ``value`` must be
5155 larger than the bit size of the destination type, ``ty2``. Equal sized
5156 types are not allowed.
5161 The '``trunc``' instruction truncates the high order bits in ``value``
5162 and converts the remaining bits to ``ty2``. Since the source size must
5163 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5164 It will always truncate bits.
5169 .. code-block:: llvm
5171 %X = trunc i32 257 to i8 ; yields i8:1
5172 %Y = trunc i32 123 to i1 ; yields i1:true
5173 %Z = trunc i32 122 to i1 ; yields i1:false
5174 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5176 '``zext .. to``' Instruction
5177 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5184 <result> = zext <ty> <value> to <ty2> ; yields ty2
5189 The '``zext``' instruction zero extends its operand to type ``ty2``.
5194 The '``zext``' instruction takes a value to cast, and a type to cast it
5195 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5196 the same number of integers. The bit size of the ``value`` must be
5197 smaller than the bit size of the destination type, ``ty2``.
5202 The ``zext`` fills the high order bits of the ``value`` with zero bits
5203 until it reaches the size of the destination type, ``ty2``.
5205 When zero extending from i1, the result will always be either 0 or 1.
5210 .. code-block:: llvm
5212 %X = zext i32 257 to i64 ; yields i64:257
5213 %Y = zext i1 true to i32 ; yields i32:1
5214 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5216 '``sext .. to``' Instruction
5217 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5224 <result> = sext <ty> <value> to <ty2> ; yields ty2
5229 The '``sext``' sign extends ``value`` to the type ``ty2``.
5234 The '``sext``' instruction takes a value to cast, and a type to cast it
5235 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5236 the same number of integers. The bit size of the ``value`` must be
5237 smaller than the bit size of the destination type, ``ty2``.
5242 The '``sext``' instruction performs a sign extension by copying the sign
5243 bit (highest order bit) of the ``value`` until it reaches the bit size
5244 of the type ``ty2``.
5246 When sign extending from i1, the extension always results in -1 or 0.
5251 .. code-block:: llvm
5253 %X = sext i8 -1 to i16 ; yields i16 :65535
5254 %Y = sext i1 true to i32 ; yields i32:-1
5255 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5257 '``fptrunc .. to``' Instruction
5258 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5265 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5270 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5275 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5276 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5277 The size of ``value`` must be larger than the size of ``ty2``. This
5278 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5283 The '``fptrunc``' instruction truncates a ``value`` from a larger
5284 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5285 point <t_floating>` type. If the value cannot fit within the
5286 destination type, ``ty2``, then the results are undefined.
5291 .. code-block:: llvm
5293 %X = fptrunc double 123.0 to float ; yields float:123.0
5294 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5296 '``fpext .. to``' Instruction
5297 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5304 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5309 The '``fpext``' extends a floating point ``value`` to a larger floating
5315 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5316 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5317 to. The source type must be smaller than the destination type.
5322 The '``fpext``' instruction extends the ``value`` from a smaller
5323 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5324 point <t_floating>` type. The ``fpext`` cannot be used to make a
5325 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5326 *no-op cast* for a floating point cast.
5331 .. code-block:: llvm
5333 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5334 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5336 '``fptoui .. to``' Instruction
5337 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5344 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5349 The '``fptoui``' converts a floating point ``value`` to its unsigned
5350 integer equivalent of type ``ty2``.
5355 The '``fptoui``' instruction takes a value to cast, which must be a
5356 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5357 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5358 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5359 type with the same number of elements as ``ty``
5364 The '``fptoui``' instruction converts its :ref:`floating
5365 point <t_floating>` operand into the nearest (rounding towards zero)
5366 unsigned integer value. If the value cannot fit in ``ty2``, the results
5372 .. code-block:: llvm
5374 %X = fptoui double 123.0 to i32 ; yields i32:123
5375 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5376 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5378 '``fptosi .. to``' Instruction
5379 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5386 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5391 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5392 ``value`` to type ``ty2``.
5397 The '``fptosi``' instruction takes a value to cast, which must be a
5398 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5399 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5400 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5401 type with the same number of elements as ``ty``
5406 The '``fptosi``' instruction converts its :ref:`floating
5407 point <t_floating>` operand into the nearest (rounding towards zero)
5408 signed integer value. If the value cannot fit in ``ty2``, the results
5414 .. code-block:: llvm
5416 %X = fptosi double -123.0 to i32 ; yields i32:-123
5417 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5418 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5420 '``uitofp .. to``' Instruction
5421 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5428 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5433 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5434 and converts that value to the ``ty2`` type.
5439 The '``uitofp``' instruction takes a value to cast, which must be a
5440 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5441 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5442 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5443 type with the same number of elements as ``ty``
5448 The '``uitofp``' instruction interprets its operand as an unsigned
5449 integer quantity and converts it to the corresponding floating point
5450 value. If the value cannot fit in the floating point value, the results
5456 .. code-block:: llvm
5458 %X = uitofp i32 257 to float ; yields float:257.0
5459 %Y = uitofp i8 -1 to double ; yields double:255.0
5461 '``sitofp .. to``' Instruction
5462 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5469 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5474 The '``sitofp``' instruction regards ``value`` as a signed integer and
5475 converts that value to the ``ty2`` type.
5480 The '``sitofp``' instruction takes a value to cast, which must be a
5481 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5482 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5483 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5484 type with the same number of elements as ``ty``
5489 The '``sitofp``' instruction interprets its operand as a signed integer
5490 quantity and converts it to the corresponding floating point value. If
5491 the value cannot fit in the floating point value, the results are
5497 .. code-block:: llvm
5499 %X = sitofp i32 257 to float ; yields float:257.0
5500 %Y = sitofp i8 -1 to double ; yields double:-1.0
5504 '``ptrtoint .. to``' Instruction
5505 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5512 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5517 The '``ptrtoint``' instruction converts the pointer or a vector of
5518 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5523 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5524 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5525 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5526 a vector of integers type.
5531 The '``ptrtoint``' instruction converts ``value`` to integer type
5532 ``ty2`` by interpreting the pointer value as an integer and either
5533 truncating or zero extending that value to the size of the integer type.
5534 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5535 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5536 the same size, then nothing is done (*no-op cast*) other than a type
5542 .. code-block:: llvm
5544 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5545 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5546 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5550 '``inttoptr .. to``' Instruction
5551 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5558 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5563 The '``inttoptr``' instruction converts an integer ``value`` to a
5564 pointer type, ``ty2``.
5569 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5570 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5576 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5577 applying either a zero extension or a truncation depending on the size
5578 of the integer ``value``. If ``value`` is larger than the size of a
5579 pointer then a truncation is done. If ``value`` is smaller than the size
5580 of a pointer then a zero extension is done. If they are the same size,
5581 nothing is done (*no-op cast*).
5586 .. code-block:: llvm
5588 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5589 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5590 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5591 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5595 '``bitcast .. to``' Instruction
5596 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5603 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5608 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5614 The '``bitcast``' instruction takes a value to cast, which must be a
5615 non-aggregate first class value, and a type to cast it to, which must
5616 also be a non-aggregate :ref:`first class <t_firstclass>` type. The bit
5617 sizes of ``value`` and the destination type, ``ty2``, must be identical.
5618 If the source type is a pointer, the destination type must also be a
5619 pointer. This instruction supports bitwise conversion of vectors to
5620 integers and to vectors of other types (as long as they have the same
5626 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It is
5627 always a *no-op cast* because no bits change with this conversion. The
5628 conversion is done as if the ``value`` had been stored to memory and
5629 read back as type ``ty2``. Pointer (or vector of pointers) types may
5630 only be converted to other pointer (or vector of pointers) types with
5631 this instruction. To convert pointers to other types, use the
5632 :ref:`inttoptr <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions
5638 .. code-block:: llvm
5640 %X = bitcast i8 255 to i8 ; yields i8 :-1
5641 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5642 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5643 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5650 The instructions in this category are the "miscellaneous" instructions,
5651 which defy better classification.
5655 '``icmp``' Instruction
5656 ^^^^^^^^^^^^^^^^^^^^^^
5663 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5668 The '``icmp``' instruction returns a boolean value or a vector of
5669 boolean values based on comparison of its two integer, integer vector,
5670 pointer, or pointer vector operands.
5675 The '``icmp``' instruction takes three operands. The first operand is
5676 the condition code indicating the kind of comparison to perform. It is
5677 not a value, just a keyword. The possible condition code are:
5680 #. ``ne``: not equal
5681 #. ``ugt``: unsigned greater than
5682 #. ``uge``: unsigned greater or equal
5683 #. ``ult``: unsigned less than
5684 #. ``ule``: unsigned less or equal
5685 #. ``sgt``: signed greater than
5686 #. ``sge``: signed greater or equal
5687 #. ``slt``: signed less than
5688 #. ``sle``: signed less or equal
5690 The remaining two arguments must be :ref:`integer <t_integer>` or
5691 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5692 must also be identical types.
5697 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5698 code given as ``cond``. The comparison performed always yields either an
5699 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5701 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5702 otherwise. No sign interpretation is necessary or performed.
5703 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5704 otherwise. No sign interpretation is necessary or performed.
5705 #. ``ugt``: interprets the operands as unsigned values and yields
5706 ``true`` if ``op1`` is greater than ``op2``.
5707 #. ``uge``: interprets the operands as unsigned values and yields
5708 ``true`` if ``op1`` is greater than or equal to ``op2``.
5709 #. ``ult``: interprets the operands as unsigned values and yields
5710 ``true`` if ``op1`` is less than ``op2``.
5711 #. ``ule``: interprets the operands as unsigned values and yields
5712 ``true`` if ``op1`` is less than or equal to ``op2``.
5713 #. ``sgt``: interprets the operands as signed values and yields ``true``
5714 if ``op1`` is greater than ``op2``.
5715 #. ``sge``: interprets the operands as signed values and yields ``true``
5716 if ``op1`` is greater than or equal to ``op2``.
5717 #. ``slt``: interprets the operands as signed values and yields ``true``
5718 if ``op1`` is less than ``op2``.
5719 #. ``sle``: interprets the operands as signed values and yields ``true``
5720 if ``op1`` is less than or equal to ``op2``.
5722 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5723 are compared as if they were integers.
5725 If the operands are integer vectors, then they are compared element by
5726 element. The result is an ``i1`` vector with the same number of elements
5727 as the values being compared. Otherwise, the result is an ``i1``.
5732 .. code-block:: llvm
5734 <result> = icmp eq i32 4, 5 ; yields: result=false
5735 <result> = icmp ne float* %X, %X ; yields: result=false
5736 <result> = icmp ult i16 4, 5 ; yields: result=true
5737 <result> = icmp sgt i16 4, 5 ; yields: result=false
5738 <result> = icmp ule i16 -4, 5 ; yields: result=false
5739 <result> = icmp sge i16 4, 5 ; yields: result=false
5741 Note that the code generator does not yet support vector types with the
5742 ``icmp`` instruction.
5746 '``fcmp``' Instruction
5747 ^^^^^^^^^^^^^^^^^^^^^^
5754 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5759 The '``fcmp``' instruction returns a boolean value or vector of boolean
5760 values based on comparison of its operands.
5762 If the operands are floating point scalars, then the result type is a
5763 boolean (:ref:`i1 <t_integer>`).
5765 If the operands are floating point vectors, then the result type is a
5766 vector of boolean with the same number of elements as the operands being
5772 The '``fcmp``' instruction takes three operands. The first operand is
5773 the condition code indicating the kind of comparison to perform. It is
5774 not a value, just a keyword. The possible condition code are:
5776 #. ``false``: no comparison, always returns false
5777 #. ``oeq``: ordered and equal
5778 #. ``ogt``: ordered and greater than
5779 #. ``oge``: ordered and greater than or equal
5780 #. ``olt``: ordered and less than
5781 #. ``ole``: ordered and less than or equal
5782 #. ``one``: ordered and not equal
5783 #. ``ord``: ordered (no nans)
5784 #. ``ueq``: unordered or equal
5785 #. ``ugt``: unordered or greater than
5786 #. ``uge``: unordered or greater than or equal
5787 #. ``ult``: unordered or less than
5788 #. ``ule``: unordered or less than or equal
5789 #. ``une``: unordered or not equal
5790 #. ``uno``: unordered (either nans)
5791 #. ``true``: no comparison, always returns true
5793 *Ordered* means that neither operand is a QNAN while *unordered* means
5794 that either operand may be a QNAN.
5796 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5797 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5798 type. They must have identical types.
5803 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5804 condition code given as ``cond``. If the operands are vectors, then the
5805 vectors are compared element by element. Each comparison performed
5806 always yields an :ref:`i1 <t_integer>` result, as follows:
5808 #. ``false``: always yields ``false``, regardless of operands.
5809 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5810 is equal to ``op2``.
5811 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5812 is greater than ``op2``.
5813 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5814 is greater than or equal to ``op2``.
5815 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5816 is less than ``op2``.
5817 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5818 is less than or equal to ``op2``.
5819 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5820 is not equal to ``op2``.
5821 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5822 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5824 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5825 greater than ``op2``.
5826 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5827 greater than or equal to ``op2``.
5828 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5830 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5831 less than or equal to ``op2``.
5832 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5833 not equal to ``op2``.
5834 #. ``uno``: yields ``true`` if either operand is a QNAN.
5835 #. ``true``: always yields ``true``, regardless of operands.
5840 .. code-block:: llvm
5842 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5843 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5844 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5845 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5847 Note that the code generator does not yet support vector types with the
5848 ``fcmp`` instruction.
5852 '``phi``' Instruction
5853 ^^^^^^^^^^^^^^^^^^^^^
5860 <result> = phi <ty> [ <val0>, <label0>], ...
5865 The '``phi``' instruction is used to implement the φ node in the SSA
5866 graph representing the function.
5871 The type of the incoming values is specified with the first type field.
5872 After this, the '``phi``' instruction takes a list of pairs as
5873 arguments, with one pair for each predecessor basic block of the current
5874 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5875 the value arguments to the PHI node. Only labels may be used as the
5878 There must be no non-phi instructions between the start of a basic block
5879 and the PHI instructions: i.e. PHI instructions must be first in a basic
5882 For the purposes of the SSA form, the use of each incoming value is
5883 deemed to occur on the edge from the corresponding predecessor block to
5884 the current block (but after any definition of an '``invoke``'
5885 instruction's return value on the same edge).
5890 At runtime, the '``phi``' instruction logically takes on the value
5891 specified by the pair corresponding to the predecessor basic block that
5892 executed just prior to the current block.
5897 .. code-block:: llvm
5899 Loop: ; Infinite loop that counts from 0 on up...
5900 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5901 %nextindvar = add i32 %indvar, 1
5906 '``select``' Instruction
5907 ^^^^^^^^^^^^^^^^^^^^^^^^
5914 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5916 selty is either i1 or {<N x i1>}
5921 The '``select``' instruction is used to choose one value based on a
5922 condition, without branching.
5927 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
5928 values indicating the condition, and two values of the same :ref:`first
5929 class <t_firstclass>` type. If the val1/val2 are vectors and the
5930 condition is a scalar, then entire vectors are selected, not individual
5936 If the condition is an i1 and it evaluates to 1, the instruction returns
5937 the first value argument; otherwise, it returns the second value
5940 If the condition is a vector of i1, then the value arguments must be
5941 vectors of the same size, and the selection is done element by element.
5946 .. code-block:: llvm
5948 %X = select i1 true, i8 17, i8 42 ; yields i8:17
5952 '``call``' Instruction
5953 ^^^^^^^^^^^^^^^^^^^^^^
5960 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
5965 The '``call``' instruction represents a simple function call.
5970 This instruction requires several arguments:
5972 #. The optional "tail" marker indicates that the callee function does
5973 not access any allocas or varargs in the caller. Note that calls may
5974 be marked "tail" even if they do not occur before a
5975 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
5976 function call is eligible for tail call optimization, but `might not
5977 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
5978 The code generator may optimize calls marked "tail" with either 1)
5979 automatic `sibling call
5980 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
5981 callee have matching signatures, or 2) forced tail call optimization
5982 when the following extra requirements are met:
5984 - Caller and callee both have the calling convention ``fastcc``.
5985 - The call is in tail position (ret immediately follows call and ret
5986 uses value of call or is void).
5987 - Option ``-tailcallopt`` is enabled, or
5988 ``llvm::GuaranteedTailCallOpt`` is ``true``.
5989 - `Platform specific constraints are
5990 met. <CodeGenerator.html#tailcallopt>`_
5992 #. The optional "cconv" marker indicates which :ref:`calling
5993 convention <callingconv>` the call should use. If none is
5994 specified, the call defaults to using C calling conventions. The
5995 calling convention of the call must match the calling convention of
5996 the target function, or else the behavior is undefined.
5997 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5998 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6000 #. '``ty``': the type of the call instruction itself which is also the
6001 type of the return value. Functions that return no value are marked
6003 #. '``fnty``': shall be the signature of the pointer to function value
6004 being invoked. The argument types must match the types implied by
6005 this signature. This type can be omitted if the function is not
6006 varargs and if the function type does not return a pointer to a
6008 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6009 be invoked. In most cases, this is a direct function invocation, but
6010 indirect ``call``'s are just as possible, calling an arbitrary pointer
6012 #. '``function args``': argument list whose types match the function
6013 signature argument types and parameter attributes. All arguments must
6014 be of :ref:`first class <t_firstclass>` type. If the function signature
6015 indicates the function accepts a variable number of arguments, the
6016 extra arguments can be specified.
6017 #. The optional :ref:`function attributes <fnattrs>` list. Only
6018 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6019 attributes are valid here.
6024 The '``call``' instruction is used to cause control flow to transfer to
6025 a specified function, with its incoming arguments bound to the specified
6026 values. Upon a '``ret``' instruction in the called function, control
6027 flow continues with the instruction after the function call, and the
6028 return value of the function is bound to the result argument.
6033 .. code-block:: llvm
6035 %retval = call i32 @test(i32 %argc)
6036 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6037 %X = tail call i32 @foo() ; yields i32
6038 %Y = tail call fastcc i32 @foo() ; yields i32
6039 call void %foo(i8 97 signext)
6041 %struct.A = type { i32, i8 }
6042 %r = call %struct.A @foo() ; yields { 32, i8 }
6043 %gr = extractvalue %struct.A %r, 0 ; yields i32
6044 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6045 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6046 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6048 llvm treats calls to some functions with names and arguments that match
6049 the standard C99 library as being the C99 library functions, and may
6050 perform optimizations or generate code for them under that assumption.
6051 This is something we'd like to change in the future to provide better
6052 support for freestanding environments and non-C-based languages.
6056 '``va_arg``' Instruction
6057 ^^^^^^^^^^^^^^^^^^^^^^^^
6064 <resultval> = va_arg <va_list*> <arglist>, <argty>
6069 The '``va_arg``' instruction is used to access arguments passed through
6070 the "variable argument" area of a function call. It is used to implement
6071 the ``va_arg`` macro in C.
6076 This instruction takes a ``va_list*`` value and the type of the
6077 argument. It returns a value of the specified argument type and
6078 increments the ``va_list`` to point to the next argument. The actual
6079 type of ``va_list`` is target specific.
6084 The '``va_arg``' instruction loads an argument of the specified type
6085 from the specified ``va_list`` and causes the ``va_list`` to point to
6086 the next argument. For more information, see the variable argument
6087 handling :ref:`Intrinsic Functions <int_varargs>`.
6089 It is legal for this instruction to be called in a function which does
6090 not take a variable number of arguments, for example, the ``vfprintf``
6093 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6094 function <intrinsics>` because it takes a type as an argument.
6099 See the :ref:`variable argument processing <int_varargs>` section.
6101 Note that the code generator does not yet fully support va\_arg on many
6102 targets. Also, it does not currently support va\_arg with aggregate
6103 types on any target.
6107 '``landingpad``' Instruction
6108 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6115 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6116 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6118 <clause> := catch <type> <value>
6119 <clause> := filter <array constant type> <array constant>
6124 The '``landingpad``' instruction is used by `LLVM's exception handling
6125 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6126 is a landing pad --- one where the exception lands, and corresponds to the
6127 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6128 defines values supplied by the personality function (``pers_fn``) upon
6129 re-entry to the function. The ``resultval`` has the type ``resultty``.
6134 This instruction takes a ``pers_fn`` value. This is the personality
6135 function associated with the unwinding mechanism. The optional
6136 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6138 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6139 contains the global variable representing the "type" that may be caught
6140 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6141 clause takes an array constant as its argument. Use
6142 "``[0 x i8**] undef``" for a filter which cannot throw. The
6143 '``landingpad``' instruction must contain *at least* one ``clause`` or
6144 the ``cleanup`` flag.
6149 The '``landingpad``' instruction defines the values which are set by the
6150 personality function (``pers_fn``) upon re-entry to the function, and
6151 therefore the "result type" of the ``landingpad`` instruction. As with
6152 calling conventions, how the personality function results are
6153 represented in LLVM IR is target specific.
6155 The clauses are applied in order from top to bottom. If two
6156 ``landingpad`` instructions are merged together through inlining, the
6157 clauses from the calling function are appended to the list of clauses.
6158 When the call stack is being unwound due to an exception being thrown,
6159 the exception is compared against each ``clause`` in turn. If it doesn't
6160 match any of the clauses, and the ``cleanup`` flag is not set, then
6161 unwinding continues further up the call stack.
6163 The ``landingpad`` instruction has several restrictions:
6165 - A landing pad block is a basic block which is the unwind destination
6166 of an '``invoke``' instruction.
6167 - A landing pad block must have a '``landingpad``' instruction as its
6168 first non-PHI instruction.
6169 - There can be only one '``landingpad``' instruction within the landing
6171 - A basic block that is not a landing pad block may not include a
6172 '``landingpad``' instruction.
6173 - All '``landingpad``' instructions in a function must have the same
6174 personality function.
6179 .. code-block:: llvm
6181 ;; A landing pad which can catch an integer.
6182 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6184 ;; A landing pad that is a cleanup.
6185 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6187 ;; A landing pad which can catch an integer and can only throw a double.
6188 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6190 filter [1 x i8**] [@_ZTId]
6197 LLVM supports the notion of an "intrinsic function". These functions
6198 have well known names and semantics and are required to follow certain
6199 restrictions. Overall, these intrinsics represent an extension mechanism
6200 for the LLVM language that does not require changing all of the
6201 transformations in LLVM when adding to the language (or the bitcode
6202 reader/writer, the parser, etc...).
6204 Intrinsic function names must all start with an "``llvm.``" prefix. This
6205 prefix is reserved in LLVM for intrinsic names; thus, function names may
6206 not begin with this prefix. Intrinsic functions must always be external
6207 functions: you cannot define the body of intrinsic functions. Intrinsic
6208 functions may only be used in call or invoke instructions: it is illegal
6209 to take the address of an intrinsic function. Additionally, because
6210 intrinsic functions are part of the LLVM language, it is required if any
6211 are added that they be documented here.
6213 Some intrinsic functions can be overloaded, i.e., the intrinsic
6214 represents a family of functions that perform the same operation but on
6215 different data types. Because LLVM can represent over 8 million
6216 different integer types, overloading is used commonly to allow an
6217 intrinsic function to operate on any integer type. One or more of the
6218 argument types or the result type can be overloaded to accept any
6219 integer type. Argument types may also be defined as exactly matching a
6220 previous argument's type or the result type. This allows an intrinsic
6221 function which accepts multiple arguments, but needs all of them to be
6222 of the same type, to only be overloaded with respect to a single
6223 argument or the result.
6225 Overloaded intrinsics will have the names of its overloaded argument
6226 types encoded into its function name, each preceded by a period. Only
6227 those types which are overloaded result in a name suffix. Arguments
6228 whose type is matched against another type do not. For example, the
6229 ``llvm.ctpop`` function can take an integer of any width and returns an
6230 integer of exactly the same integer width. This leads to a family of
6231 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6232 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6233 overloaded, and only one type suffix is required. Because the argument's
6234 type is matched against the return type, it does not require its own
6237 To learn how to add an intrinsic function, please see the `Extending
6238 LLVM Guide <ExtendingLLVM.html>`_.
6242 Variable Argument Handling Intrinsics
6243 -------------------------------------
6245 Variable argument support is defined in LLVM with the
6246 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6247 functions. These functions are related to the similarly named macros
6248 defined in the ``<stdarg.h>`` header file.
6250 All of these functions operate on arguments that use a target-specific
6251 value type "``va_list``". The LLVM assembly language reference manual
6252 does not define what this type is, so all transformations should be
6253 prepared to handle these functions regardless of the type used.
6255 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6256 variable argument handling intrinsic functions are used.
6258 .. code-block:: llvm
6260 define i32 @test(i32 %X, ...) {
6261 ; Initialize variable argument processing
6263 %ap2 = bitcast i8** %ap to i8*
6264 call void @llvm.va_start(i8* %ap2)
6266 ; Read a single integer argument
6267 %tmp = va_arg i8** %ap, i32
6269 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6271 %aq2 = bitcast i8** %aq to i8*
6272 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6273 call void @llvm.va_end(i8* %aq2)
6275 ; Stop processing of arguments.
6276 call void @llvm.va_end(i8* %ap2)
6280 declare void @llvm.va_start(i8*)
6281 declare void @llvm.va_copy(i8*, i8*)
6282 declare void @llvm.va_end(i8*)
6286 '``llvm.va_start``' Intrinsic
6287 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6294 declare void %llvm.va_start(i8* <arglist>)
6299 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6300 subsequent use by ``va_arg``.
6305 The argument is a pointer to a ``va_list`` element to initialize.
6310 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6311 available in C. In a target-dependent way, it initializes the
6312 ``va_list`` element to which the argument points, so that the next call
6313 to ``va_arg`` will produce the first variable argument passed to the
6314 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6315 to know the last argument of the function as the compiler can figure
6318 '``llvm.va_end``' Intrinsic
6319 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6326 declare void @llvm.va_end(i8* <arglist>)
6331 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6332 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6337 The argument is a pointer to a ``va_list`` to destroy.
6342 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6343 available in C. In a target-dependent way, it destroys the ``va_list``
6344 element to which the argument points. Calls to
6345 :ref:`llvm.va_start <int_va_start>` and
6346 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6351 '``llvm.va_copy``' Intrinsic
6352 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6359 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6364 The '``llvm.va_copy``' intrinsic copies the current argument position
6365 from the source argument list to the destination argument list.
6370 The first argument is a pointer to a ``va_list`` element to initialize.
6371 The second argument is a pointer to a ``va_list`` element to copy from.
6376 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6377 available in C. In a target-dependent way, it copies the source
6378 ``va_list`` element into the destination ``va_list`` element. This
6379 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6380 arbitrarily complex and require, for example, memory allocation.
6382 Accurate Garbage Collection Intrinsics
6383 --------------------------------------
6385 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6386 (GC) requires the implementation and generation of these intrinsics.
6387 These intrinsics allow identification of :ref:`GC roots on the
6388 stack <int_gcroot>`, as well as garbage collector implementations that
6389 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6390 Front-ends for type-safe garbage collected languages should generate
6391 these intrinsics to make use of the LLVM garbage collectors. For more
6392 details, see `Accurate Garbage Collection with
6393 LLVM <GarbageCollection.html>`_.
6395 The garbage collection intrinsics only operate on objects in the generic
6396 address space (address space zero).
6400 '``llvm.gcroot``' Intrinsic
6401 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6408 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6413 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6414 the code generator, and allows some metadata to be associated with it.
6419 The first argument specifies the address of a stack object that contains
6420 the root pointer. The second pointer (which must be either a constant or
6421 a global value address) contains the meta-data to be associated with the
6427 At runtime, a call to this intrinsic stores a null pointer into the
6428 "ptrloc" location. At compile-time, the code generator generates
6429 information to allow the runtime to find the pointer at GC safe points.
6430 The '``llvm.gcroot``' intrinsic may only be used in a function which
6431 :ref:`specifies a GC algorithm <gc>`.
6435 '``llvm.gcread``' Intrinsic
6436 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6443 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6448 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6449 locations, allowing garbage collector implementations that require read
6455 The second argument is the address to read from, which should be an
6456 address allocated from the garbage collector. The first object is a
6457 pointer to the start of the referenced object, if needed by the language
6458 runtime (otherwise null).
6463 The '``llvm.gcread``' intrinsic has the same semantics as a load
6464 instruction, but may be replaced with substantially more complex code by
6465 the garbage collector runtime, as needed. The '``llvm.gcread``'
6466 intrinsic may only be used in a function which :ref:`specifies a GC
6471 '``llvm.gcwrite``' Intrinsic
6472 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6479 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6484 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6485 locations, allowing garbage collector implementations that require write
6486 barriers (such as generational or reference counting collectors).
6491 The first argument is the reference to store, the second is the start of
6492 the object to store it to, and the third is the address of the field of
6493 Obj to store to. If the runtime does not require a pointer to the
6494 object, Obj may be null.
6499 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6500 instruction, but may be replaced with substantially more complex code by
6501 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6502 intrinsic may only be used in a function which :ref:`specifies a GC
6505 Code Generator Intrinsics
6506 -------------------------
6508 These intrinsics are provided by LLVM to expose special features that
6509 may only be implemented with code generator support.
6511 '``llvm.returnaddress``' Intrinsic
6512 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6519 declare i8 *@llvm.returnaddress(i32 <level>)
6524 The '``llvm.returnaddress``' intrinsic attempts to compute a
6525 target-specific value indicating the return address of the current
6526 function or one of its callers.
6531 The argument to this intrinsic indicates which function to return the
6532 address for. Zero indicates the calling function, one indicates its
6533 caller, etc. The argument is **required** to be a constant integer
6539 The '``llvm.returnaddress``' intrinsic either returns a pointer
6540 indicating the return address of the specified call frame, or zero if it
6541 cannot be identified. The value returned by this intrinsic is likely to
6542 be incorrect or 0 for arguments other than zero, so it should only be
6543 used for debugging purposes.
6545 Note that calling this intrinsic does not prevent function inlining or
6546 other aggressive transformations, so the value returned may not be that
6547 of the obvious source-language caller.
6549 '``llvm.frameaddress``' Intrinsic
6550 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6557 declare i8* @llvm.frameaddress(i32 <level>)
6562 The '``llvm.frameaddress``' intrinsic attempts to return the
6563 target-specific frame pointer value for the specified stack frame.
6568 The argument to this intrinsic indicates which function to return the
6569 frame pointer for. Zero indicates the calling function, one indicates
6570 its caller, etc. The argument is **required** to be a constant integer
6576 The '``llvm.frameaddress``' intrinsic either returns a pointer
6577 indicating the frame address of the specified call frame, or zero if it
6578 cannot be identified. The value returned by this intrinsic is likely to
6579 be incorrect or 0 for arguments other than zero, so it should only be
6580 used for debugging purposes.
6582 Note that calling this intrinsic does not prevent function inlining or
6583 other aggressive transformations, so the value returned may not be that
6584 of the obvious source-language caller.
6588 '``llvm.stacksave``' Intrinsic
6589 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6596 declare i8* @llvm.stacksave()
6601 The '``llvm.stacksave``' intrinsic is used to remember the current state
6602 of the function stack, for use with
6603 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6604 implementing language features like scoped automatic variable sized
6610 This intrinsic returns a opaque pointer value that can be passed to
6611 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6612 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6613 ``llvm.stacksave``, it effectively restores the state of the stack to
6614 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6615 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6616 were allocated after the ``llvm.stacksave`` was executed.
6618 .. _int_stackrestore:
6620 '``llvm.stackrestore``' Intrinsic
6621 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6628 declare void @llvm.stackrestore(i8* %ptr)
6633 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6634 the function stack to the state it was in when the corresponding
6635 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6636 useful for implementing language features like scoped automatic variable
6637 sized arrays in C99.
6642 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6644 '``llvm.prefetch``' Intrinsic
6645 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6652 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6657 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6658 insert a prefetch instruction if supported; otherwise, it is a noop.
6659 Prefetches have no effect on the behavior of the program but can change
6660 its performance characteristics.
6665 ``address`` is the address to be prefetched, ``rw`` is the specifier
6666 determining if the fetch should be for a read (0) or write (1), and
6667 ``locality`` is a temporal locality specifier ranging from (0) - no
6668 locality, to (3) - extremely local keep in cache. The ``cache type``
6669 specifies whether the prefetch is performed on the data (1) or
6670 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6671 arguments must be constant integers.
6676 This intrinsic does not modify the behavior of the program. In
6677 particular, prefetches cannot trap and do not produce a value. On
6678 targets that support this intrinsic, the prefetch can provide hints to
6679 the processor cache for better performance.
6681 '``llvm.pcmarker``' Intrinsic
6682 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6689 declare void @llvm.pcmarker(i32 <id>)
6694 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6695 Counter (PC) in a region of code to simulators and other tools. The
6696 method is target specific, but it is expected that the marker will use
6697 exported symbols to transmit the PC of the marker. The marker makes no
6698 guarantees that it will remain with any specific instruction after
6699 optimizations. It is possible that the presence of a marker will inhibit
6700 optimizations. The intended use is to be inserted after optimizations to
6701 allow correlations of simulation runs.
6706 ``id`` is a numerical id identifying the marker.
6711 This intrinsic does not modify the behavior of the program. Backends
6712 that do not support this intrinsic may ignore it.
6714 '``llvm.readcyclecounter``' Intrinsic
6715 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6722 declare i64 @llvm.readcyclecounter()
6727 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6728 counter register (or similar low latency, high accuracy clocks) on those
6729 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6730 should map to RPCC. As the backing counters overflow quickly (on the
6731 order of 9 seconds on alpha), this should only be used for small
6737 When directly supported, reading the cycle counter should not modify any
6738 memory. Implementations are allowed to either return a application
6739 specific value or a system wide value. On backends without support, this
6740 is lowered to a constant 0.
6742 Note that runtime support may be conditional on the privilege-level code is
6743 running at and the host platform.
6745 Standard C Library Intrinsics
6746 -----------------------------
6748 LLVM provides intrinsics for a few important standard C library
6749 functions. These intrinsics allow source-language front-ends to pass
6750 information about the alignment of the pointer arguments to the code
6751 generator, providing opportunity for more efficient code generation.
6755 '``llvm.memcpy``' Intrinsic
6756 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6761 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6762 integer bit width and for different address spaces. Not all targets
6763 support all bit widths however.
6767 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6768 i32 <len>, i32 <align>, i1 <isvolatile>)
6769 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6770 i64 <len>, i32 <align>, i1 <isvolatile>)
6775 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6776 source location to the destination location.
6778 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6779 intrinsics do not return a value, takes extra alignment/isvolatile
6780 arguments and the pointers can be in specified address spaces.
6785 The first argument is a pointer to the destination, the second is a
6786 pointer to the source. The third argument is an integer argument
6787 specifying the number of bytes to copy, the fourth argument is the
6788 alignment of the source and destination locations, and the fifth is a
6789 boolean indicating a volatile access.
6791 If the call to this intrinsic has an alignment value that is not 0 or 1,
6792 then the caller guarantees that both the source and destination pointers
6793 are aligned to that boundary.
6795 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6796 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6797 very cleanly specified and it is unwise to depend on it.
6802 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6803 source location to the destination location, which are not allowed to
6804 overlap. It copies "len" bytes of memory over. If the argument is known
6805 to be aligned to some boundary, this can be specified as the fourth
6806 argument, otherwise it should be set to 0 or 1.
6808 '``llvm.memmove``' Intrinsic
6809 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6814 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6815 bit width and for different address space. Not all targets support all
6820 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6821 i32 <len>, i32 <align>, i1 <isvolatile>)
6822 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6823 i64 <len>, i32 <align>, i1 <isvolatile>)
6828 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6829 source location to the destination location. It is similar to the
6830 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6833 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6834 intrinsics do not return a value, takes extra alignment/isvolatile
6835 arguments and the pointers can be in specified address spaces.
6840 The first argument is a pointer to the destination, the second is a
6841 pointer to the source. The third argument is an integer argument
6842 specifying the number of bytes to copy, the fourth argument is the
6843 alignment of the source and destination locations, and the fifth is a
6844 boolean indicating a volatile access.
6846 If the call to this intrinsic has an alignment value that is not 0 or 1,
6847 then the caller guarantees that the source and destination pointers are
6848 aligned to that boundary.
6850 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6851 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6852 not very cleanly specified and it is unwise to depend on it.
6857 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6858 source location to the destination location, which may overlap. It
6859 copies "len" bytes of memory over. If the argument is known to be
6860 aligned to some boundary, this can be specified as the fourth argument,
6861 otherwise it should be set to 0 or 1.
6863 '``llvm.memset.*``' Intrinsics
6864 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6869 This is an overloaded intrinsic. You can use llvm.memset on any integer
6870 bit width and for different address spaces. However, not all targets
6871 support all bit widths.
6875 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6876 i32 <len>, i32 <align>, i1 <isvolatile>)
6877 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6878 i64 <len>, i32 <align>, i1 <isvolatile>)
6883 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6884 particular byte value.
6886 Note that, unlike the standard libc function, the ``llvm.memset``
6887 intrinsic does not return a value and takes extra alignment/volatile
6888 arguments. Also, the destination can be in an arbitrary address space.
6893 The first argument is a pointer to the destination to fill, the second
6894 is the byte value with which to fill it, the third argument is an
6895 integer argument specifying the number of bytes to fill, and the fourth
6896 argument is the known alignment of the destination location.
6898 If the call to this intrinsic has an alignment value that is not 0 or 1,
6899 then the caller guarantees that the destination pointer is aligned to
6902 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6903 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6904 very cleanly specified and it is unwise to depend on it.
6909 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6910 at the destination location. If the argument is known to be aligned to
6911 some boundary, this can be specified as the fourth argument, otherwise
6912 it should be set to 0 or 1.
6914 '``llvm.sqrt.*``' Intrinsic
6915 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6920 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6921 floating point or vector of floating point type. Not all targets support
6926 declare float @llvm.sqrt.f32(float %Val)
6927 declare double @llvm.sqrt.f64(double %Val)
6928 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6929 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6930 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6935 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
6936 returning the same value as the libm '``sqrt``' functions would. Unlike
6937 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
6938 negative numbers other than -0.0 (which allows for better optimization,
6939 because there is no need to worry about errno being set).
6940 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
6945 The argument and return value are floating point numbers of the same
6951 This function returns the sqrt of the specified operand if it is a
6952 nonnegative floating point number.
6954 '``llvm.powi.*``' Intrinsic
6955 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6960 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
6961 floating point or vector of floating point type. Not all targets support
6966 declare float @llvm.powi.f32(float %Val, i32 %power)
6967 declare double @llvm.powi.f64(double %Val, i32 %power)
6968 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6969 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6970 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6975 The '``llvm.powi.*``' intrinsics return the first operand raised to the
6976 specified (positive or negative) power. The order of evaluation of
6977 multiplications is not defined. When a vector of floating point type is
6978 used, the second argument remains a scalar integer value.
6983 The second argument is an integer power, and the first is a value to
6984 raise to that power.
6989 This function returns the first value raised to the second power with an
6990 unspecified sequence of rounding operations.
6992 '``llvm.sin.*``' Intrinsic
6993 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6998 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
6999 floating point or vector of floating point type. Not all targets support
7004 declare float @llvm.sin.f32(float %Val)
7005 declare double @llvm.sin.f64(double %Val)
7006 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7007 declare fp128 @llvm.sin.f128(fp128 %Val)
7008 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7013 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7018 The argument and return value are floating point numbers of the same
7024 This function returns the sine of the specified operand, returning the
7025 same values as the libm ``sin`` functions would, and handles error
7026 conditions in the same way.
7028 '``llvm.cos.*``' Intrinsic
7029 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7034 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7035 floating point or vector of floating point type. Not all targets support
7040 declare float @llvm.cos.f32(float %Val)
7041 declare double @llvm.cos.f64(double %Val)
7042 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7043 declare fp128 @llvm.cos.f128(fp128 %Val)
7044 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7049 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7054 The argument and return value are floating point numbers of the same
7060 This function returns the cosine of the specified operand, returning the
7061 same values as the libm ``cos`` functions would, and handles error
7062 conditions in the same way.
7064 '``llvm.pow.*``' Intrinsic
7065 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7070 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7071 floating point or vector of floating point type. Not all targets support
7076 declare float @llvm.pow.f32(float %Val, float %Power)
7077 declare double @llvm.pow.f64(double %Val, double %Power)
7078 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7079 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7080 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7085 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7086 specified (positive or negative) power.
7091 The second argument is a floating point power, and the first is a value
7092 to raise to that power.
7097 This function returns the first value raised to the second power,
7098 returning the same values as the libm ``pow`` functions would, and
7099 handles error conditions in the same way.
7101 '``llvm.exp.*``' Intrinsic
7102 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7107 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7108 floating point or vector of floating point type. Not all targets support
7113 declare float @llvm.exp.f32(float %Val)
7114 declare double @llvm.exp.f64(double %Val)
7115 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7116 declare fp128 @llvm.exp.f128(fp128 %Val)
7117 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7122 The '``llvm.exp.*``' intrinsics perform the exp function.
7127 The argument and return value are floating point numbers of the same
7133 This function returns the same values as the libm ``exp`` functions
7134 would, and handles error conditions in the same way.
7136 '``llvm.exp2.*``' Intrinsic
7137 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7142 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7143 floating point or vector of floating point type. Not all targets support
7148 declare float @llvm.exp2.f32(float %Val)
7149 declare double @llvm.exp2.f64(double %Val)
7150 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7151 declare fp128 @llvm.exp2.f128(fp128 %Val)
7152 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7157 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7162 The argument and return value are floating point numbers of the same
7168 This function returns the same values as the libm ``exp2`` functions
7169 would, and handles error conditions in the same way.
7171 '``llvm.log.*``' Intrinsic
7172 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7177 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7178 floating point or vector of floating point type. Not all targets support
7183 declare float @llvm.log.f32(float %Val)
7184 declare double @llvm.log.f64(double %Val)
7185 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7186 declare fp128 @llvm.log.f128(fp128 %Val)
7187 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7192 The '``llvm.log.*``' intrinsics perform the log function.
7197 The argument and return value are floating point numbers of the same
7203 This function returns the same values as the libm ``log`` functions
7204 would, and handles error conditions in the same way.
7206 '``llvm.log10.*``' Intrinsic
7207 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7212 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7213 floating point or vector of floating point type. Not all targets support
7218 declare float @llvm.log10.f32(float %Val)
7219 declare double @llvm.log10.f64(double %Val)
7220 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7221 declare fp128 @llvm.log10.f128(fp128 %Val)
7222 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7227 The '``llvm.log10.*``' intrinsics perform the log10 function.
7232 The argument and return value are floating point numbers of the same
7238 This function returns the same values as the libm ``log10`` functions
7239 would, and handles error conditions in the same way.
7241 '``llvm.log2.*``' Intrinsic
7242 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7247 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7248 floating point or vector of floating point type. Not all targets support
7253 declare float @llvm.log2.f32(float %Val)
7254 declare double @llvm.log2.f64(double %Val)
7255 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7256 declare fp128 @llvm.log2.f128(fp128 %Val)
7257 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7262 The '``llvm.log2.*``' intrinsics perform the log2 function.
7267 The argument and return value are floating point numbers of the same
7273 This function returns the same values as the libm ``log2`` functions
7274 would, and handles error conditions in the same way.
7276 '``llvm.fma.*``' Intrinsic
7277 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7282 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7283 floating point or vector of floating point type. Not all targets support
7288 declare float @llvm.fma.f32(float %a, float %b, float %c)
7289 declare double @llvm.fma.f64(double %a, double %b, double %c)
7290 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7291 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7292 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7297 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7303 The argument and return value are floating point numbers of the same
7309 This function returns the same values as the libm ``fma`` functions
7312 '``llvm.fabs.*``' Intrinsic
7313 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7318 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7319 floating point or vector of floating point type. Not all targets support
7324 declare float @llvm.fabs.f32(float %Val)
7325 declare double @llvm.fabs.f64(double %Val)
7326 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7327 declare fp128 @llvm.fabs.f128(fp128 %Val)
7328 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7333 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7339 The argument and return value are floating point numbers of the same
7345 This function returns the same values as the libm ``fabs`` functions
7346 would, and handles error conditions in the same way.
7348 '``llvm.floor.*``' Intrinsic
7349 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7354 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7355 floating point or vector of floating point type. Not all targets support
7360 declare float @llvm.floor.f32(float %Val)
7361 declare double @llvm.floor.f64(double %Val)
7362 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7363 declare fp128 @llvm.floor.f128(fp128 %Val)
7364 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7369 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7374 The argument and return value are floating point numbers of the same
7380 This function returns the same values as the libm ``floor`` functions
7381 would, and handles error conditions in the same way.
7383 '``llvm.ceil.*``' Intrinsic
7384 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7389 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7390 floating point or vector of floating point type. Not all targets support
7395 declare float @llvm.ceil.f32(float %Val)
7396 declare double @llvm.ceil.f64(double %Val)
7397 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7398 declare fp128 @llvm.ceil.f128(fp128 %Val)
7399 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7404 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7409 The argument and return value are floating point numbers of the same
7415 This function returns the same values as the libm ``ceil`` functions
7416 would, and handles error conditions in the same way.
7418 '``llvm.trunc.*``' Intrinsic
7419 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7424 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7425 floating point or vector of floating point type. Not all targets support
7430 declare float @llvm.trunc.f32(float %Val)
7431 declare double @llvm.trunc.f64(double %Val)
7432 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7433 declare fp128 @llvm.trunc.f128(fp128 %Val)
7434 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7439 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7440 nearest integer not larger in magnitude than the operand.
7445 The argument and return value are floating point numbers of the same
7451 This function returns the same values as the libm ``trunc`` functions
7452 would, and handles error conditions in the same way.
7454 '``llvm.rint.*``' Intrinsic
7455 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7460 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7461 floating point or vector of floating point type. Not all targets support
7466 declare float @llvm.rint.f32(float %Val)
7467 declare double @llvm.rint.f64(double %Val)
7468 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7469 declare fp128 @llvm.rint.f128(fp128 %Val)
7470 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7475 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7476 nearest integer. It may raise an inexact floating-point exception if the
7477 operand isn't an integer.
7482 The argument and return value are floating point numbers of the same
7488 This function returns the same values as the libm ``rint`` functions
7489 would, and handles error conditions in the same way.
7491 '``llvm.nearbyint.*``' Intrinsic
7492 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7497 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7498 floating point or vector of floating point type. Not all targets support
7503 declare float @llvm.nearbyint.f32(float %Val)
7504 declare double @llvm.nearbyint.f64(double %Val)
7505 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7506 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7507 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7512 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7518 The argument and return value are floating point numbers of the same
7524 This function returns the same values as the libm ``nearbyint``
7525 functions would, and handles error conditions in the same way.
7527 Bit Manipulation Intrinsics
7528 ---------------------------
7530 LLVM provides intrinsics for a few important bit manipulation
7531 operations. These allow efficient code generation for some algorithms.
7533 '``llvm.bswap.*``' Intrinsics
7534 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7539 This is an overloaded intrinsic function. You can use bswap on any
7540 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7544 declare i16 @llvm.bswap.i16(i16 <id>)
7545 declare i32 @llvm.bswap.i32(i32 <id>)
7546 declare i64 @llvm.bswap.i64(i64 <id>)
7551 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7552 values with an even number of bytes (positive multiple of 16 bits).
7553 These are useful for performing operations on data that is not in the
7554 target's native byte order.
7559 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7560 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7561 intrinsic returns an i32 value that has the four bytes of the input i32
7562 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7563 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7564 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7565 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7568 '``llvm.ctpop.*``' Intrinsic
7569 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7574 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7575 bit width, or on any vector with integer elements. Not all targets
7576 support all bit widths or vector types, however.
7580 declare i8 @llvm.ctpop.i8(i8 <src>)
7581 declare i16 @llvm.ctpop.i16(i16 <src>)
7582 declare i32 @llvm.ctpop.i32(i32 <src>)
7583 declare i64 @llvm.ctpop.i64(i64 <src>)
7584 declare i256 @llvm.ctpop.i256(i256 <src>)
7585 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7590 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7596 The only argument is the value to be counted. The argument may be of any
7597 integer type, or a vector with integer elements. The return type must
7598 match the argument type.
7603 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7604 each element of a vector.
7606 '``llvm.ctlz.*``' Intrinsic
7607 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7612 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7613 integer bit width, or any vector whose elements are integers. Not all
7614 targets support all bit widths or vector types, however.
7618 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7619 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7620 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7621 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7622 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7623 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7628 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7629 leading zeros in a variable.
7634 The first argument is the value to be counted. This argument may be of
7635 any integer type, or a vectory with integer element type. The return
7636 type must match the first argument type.
7638 The second argument must be a constant and is a flag to indicate whether
7639 the intrinsic should ensure that a zero as the first argument produces a
7640 defined result. Historically some architectures did not provide a
7641 defined result for zero values as efficiently, and many algorithms are
7642 now predicated on avoiding zero-value inputs.
7647 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7648 zeros in a variable, or within each element of the vector. If
7649 ``src == 0`` then the result is the size in bits of the type of ``src``
7650 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7651 ``llvm.ctlz(i32 2) = 30``.
7653 '``llvm.cttz.*``' Intrinsic
7654 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7659 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7660 integer bit width, or any vector of integer elements. Not all targets
7661 support all bit widths or vector types, however.
7665 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7666 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7667 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7668 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7669 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7670 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7675 The '``llvm.cttz``' family of intrinsic functions counts the number of
7681 The first argument is the value to be counted. This argument may be of
7682 any integer type, or a vectory with integer element type. The return
7683 type must match the first argument type.
7685 The second argument must be a constant and is a flag to indicate whether
7686 the intrinsic should ensure that a zero as the first argument produces a
7687 defined result. Historically some architectures did not provide a
7688 defined result for zero values as efficiently, and many algorithms are
7689 now predicated on avoiding zero-value inputs.
7694 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7695 zeros in a variable, or within each element of a vector. If ``src == 0``
7696 then the result is the size in bits of the type of ``src`` if
7697 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7698 ``llvm.cttz(2) = 1``.
7700 Arithmetic with Overflow Intrinsics
7701 -----------------------------------
7703 LLVM provides intrinsics for some arithmetic with overflow operations.
7705 '``llvm.sadd.with.overflow.*``' Intrinsics
7706 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7711 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7712 on any integer bit width.
7716 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7717 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7718 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7723 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7724 a signed addition of the two arguments, and indicate whether an overflow
7725 occurred during the signed summation.
7730 The arguments (%a and %b) and the first element of the result structure
7731 may be of integer types of any bit width, but they must have the same
7732 bit width. The second element of the result structure must be of type
7733 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7739 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7740 a signed addition of the two variables. They return a structure --- the
7741 first element of which is the signed summation, and the second element
7742 of which is a bit specifying if the signed summation resulted in an
7748 .. code-block:: llvm
7750 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7751 %sum = extractvalue {i32, i1} %res, 0
7752 %obit = extractvalue {i32, i1} %res, 1
7753 br i1 %obit, label %overflow, label %normal
7755 '``llvm.uadd.with.overflow.*``' Intrinsics
7756 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7761 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7762 on any integer bit width.
7766 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7767 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7768 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7773 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7774 an unsigned addition of the two arguments, and indicate whether a carry
7775 occurred during the unsigned summation.
7780 The arguments (%a and %b) and the first element of the result structure
7781 may be of integer types of any bit width, but they must have the same
7782 bit width. The second element of the result structure must be of type
7783 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7789 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7790 an unsigned addition of the two arguments. They return a structure --- the
7791 first element of which is the sum, and the second element of which is a
7792 bit specifying if the unsigned summation resulted in a carry.
7797 .. code-block:: llvm
7799 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7800 %sum = extractvalue {i32, i1} %res, 0
7801 %obit = extractvalue {i32, i1} %res, 1
7802 br i1 %obit, label %carry, label %normal
7804 '``llvm.ssub.with.overflow.*``' Intrinsics
7805 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7810 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7811 on any integer bit width.
7815 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7816 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7817 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7822 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7823 a signed subtraction of the two arguments, and indicate whether an
7824 overflow occurred during the signed subtraction.
7829 The arguments (%a and %b) and the first element of the result structure
7830 may be of integer types of any bit width, but they must have the same
7831 bit width. The second element of the result structure must be of type
7832 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7838 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7839 a signed subtraction of the two arguments. They return a structure --- the
7840 first element of which is the subtraction, and the second element of
7841 which is a bit specifying if the signed subtraction resulted in an
7847 .. code-block:: llvm
7849 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7850 %sum = extractvalue {i32, i1} %res, 0
7851 %obit = extractvalue {i32, i1} %res, 1
7852 br i1 %obit, label %overflow, label %normal
7854 '``llvm.usub.with.overflow.*``' Intrinsics
7855 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7860 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
7861 on any integer bit width.
7865 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7866 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7867 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7872 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7873 an unsigned subtraction of the two arguments, and indicate whether an
7874 overflow occurred during the unsigned subtraction.
7879 The arguments (%a and %b) and the first element of the result structure
7880 may be of integer types of any bit width, but they must have the same
7881 bit width. The second element of the result structure must be of type
7882 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7888 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7889 an unsigned subtraction of the two arguments. They return a structure ---
7890 the first element of which is the subtraction, and the second element of
7891 which is a bit specifying if the unsigned subtraction resulted in an
7897 .. code-block:: llvm
7899 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7900 %sum = extractvalue {i32, i1} %res, 0
7901 %obit = extractvalue {i32, i1} %res, 1
7902 br i1 %obit, label %overflow, label %normal
7904 '``llvm.smul.with.overflow.*``' Intrinsics
7905 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7910 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
7911 on any integer bit width.
7915 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7916 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7917 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7922 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7923 a signed multiplication of the two arguments, and indicate whether an
7924 overflow occurred during the signed multiplication.
7929 The arguments (%a and %b) and the first element of the result structure
7930 may be of integer types of any bit width, but they must have the same
7931 bit width. The second element of the result structure must be of type
7932 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7938 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7939 a signed multiplication of the two arguments. They return a structure ---
7940 the first element of which is the multiplication, and the second element
7941 of which is a bit specifying if the signed multiplication resulted in an
7947 .. code-block:: llvm
7949 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7950 %sum = extractvalue {i32, i1} %res, 0
7951 %obit = extractvalue {i32, i1} %res, 1
7952 br i1 %obit, label %overflow, label %normal
7954 '``llvm.umul.with.overflow.*``' Intrinsics
7955 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7960 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
7961 on any integer bit width.
7965 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7966 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7967 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7972 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7973 a unsigned multiplication of the two arguments, and indicate whether an
7974 overflow occurred during the unsigned multiplication.
7979 The arguments (%a and %b) and the first element of the result structure
7980 may be of integer types of any bit width, but they must have the same
7981 bit width. The second element of the result structure must be of type
7982 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7988 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7989 an unsigned multiplication of the two arguments. They return a structure ---
7990 the first element of which is the multiplication, and the second
7991 element of which is a bit specifying if the unsigned multiplication
7992 resulted in an overflow.
7997 .. code-block:: llvm
7999 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8000 %sum = extractvalue {i32, i1} %res, 0
8001 %obit = extractvalue {i32, i1} %res, 1
8002 br i1 %obit, label %overflow, label %normal
8004 Specialised Arithmetic Intrinsics
8005 ---------------------------------
8007 '``llvm.fmuladd.*``' Intrinsic
8008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8015 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8016 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8021 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8022 expressions that can be fused if the code generator determines that (a) the
8023 target instruction set has support for a fused operation, and (b) that the
8024 fused operation is more efficient than the equivalent, separate pair of mul
8025 and add instructions.
8030 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8031 multiplicands, a and b, and an addend c.
8040 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8042 is equivalent to the expression a \* b + c, except that rounding will
8043 not be performed between the multiplication and addition steps if the
8044 code generator fuses the operations. Fusion is not guaranteed, even if
8045 the target platform supports it. If a fused multiply-add is required the
8046 corresponding llvm.fma.\* intrinsic function should be used instead.
8051 .. code-block:: llvm
8053 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8055 Half Precision Floating Point Intrinsics
8056 ----------------------------------------
8058 For most target platforms, half precision floating point is a
8059 storage-only format. This means that it is a dense encoding (in memory)
8060 but does not support computation in the format.
8062 This means that code must first load the half-precision floating point
8063 value as an i16, then convert it to float with
8064 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8065 then be performed on the float value (including extending to double
8066 etc). To store the value back to memory, it is first converted to float
8067 if needed, then converted to i16 with
8068 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8071 .. _int_convert_to_fp16:
8073 '``llvm.convert.to.fp16``' Intrinsic
8074 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8081 declare i16 @llvm.convert.to.fp16(f32 %a)
8086 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8087 from single precision floating point format to half precision floating
8093 The intrinsic function contains single argument - the value to be
8099 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8100 from single precision floating point format to half precision floating
8101 point format. The return value is an ``i16`` which contains the
8107 .. code-block:: llvm
8109 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8110 store i16 %res, i16* @x, align 2
8112 .. _int_convert_from_fp16:
8114 '``llvm.convert.from.fp16``' Intrinsic
8115 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8122 declare f32 @llvm.convert.from.fp16(i16 %a)
8127 The '``llvm.convert.from.fp16``' intrinsic function performs a
8128 conversion from half precision floating point format to single precision
8129 floating point format.
8134 The intrinsic function contains single argument - the value to be
8140 The '``llvm.convert.from.fp16``' intrinsic function performs a
8141 conversion from half single precision floating point format to single
8142 precision floating point format. The input half-float value is
8143 represented by an ``i16`` value.
8148 .. code-block:: llvm
8150 %a = load i16* @x, align 2
8151 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8156 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8157 prefix), are described in the `LLVM Source Level
8158 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8161 Exception Handling Intrinsics
8162 -----------------------------
8164 The LLVM exception handling intrinsics (which all start with
8165 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8166 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8170 Trampoline Intrinsics
8171 ---------------------
8173 These intrinsics make it possible to excise one parameter, marked with
8174 the :ref:`nest <nest>` attribute, from a function. The result is a
8175 callable function pointer lacking the nest parameter - the caller does
8176 not need to provide a value for it. Instead, the value to use is stored
8177 in advance in a "trampoline", a block of memory usually allocated on the
8178 stack, which also contains code to splice the nest value into the
8179 argument list. This is used to implement the GCC nested function address
8182 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8183 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8184 It can be created as follows:
8186 .. code-block:: llvm
8188 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8189 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8190 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8191 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8192 %fp = bitcast i8* %p to i32 (i32, i32)*
8194 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8195 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8199 '``llvm.init.trampoline``' Intrinsic
8200 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8207 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8212 This fills the memory pointed to by ``tramp`` with executable code,
8213 turning it into a trampoline.
8218 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8219 pointers. The ``tramp`` argument must point to a sufficiently large and
8220 sufficiently aligned block of memory; this memory is written to by the
8221 intrinsic. Note that the size and the alignment are target-specific -
8222 LLVM currently provides no portable way of determining them, so a
8223 front-end that generates this intrinsic needs to have some
8224 target-specific knowledge. The ``func`` argument must hold a function
8225 bitcast to an ``i8*``.
8230 The block of memory pointed to by ``tramp`` is filled with target
8231 dependent code, turning it into a function. Then ``tramp`` needs to be
8232 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8233 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8234 function's signature is the same as that of ``func`` with any arguments
8235 marked with the ``nest`` attribute removed. At most one such ``nest``
8236 argument is allowed, and it must be of pointer type. Calling the new
8237 function is equivalent to calling ``func`` with the same argument list,
8238 but with ``nval`` used for the missing ``nest`` argument. If, after
8239 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8240 modified, then the effect of any later call to the returned function
8241 pointer is undefined.
8245 '``llvm.adjust.trampoline``' Intrinsic
8246 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8253 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8258 This performs any required machine-specific adjustment to the address of
8259 a trampoline (passed as ``tramp``).
8264 ``tramp`` must point to a block of memory which already has trampoline
8265 code filled in by a previous call to
8266 :ref:`llvm.init.trampoline <int_it>`.
8271 On some architectures the address of the code to be executed needs to be
8272 different to the address where the trampoline is actually stored. This
8273 intrinsic returns the executable address corresponding to ``tramp``
8274 after performing the required machine specific adjustments. The pointer
8275 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8280 This class of intrinsics exists to information about the lifetime of
8281 memory objects and ranges where variables are immutable.
8283 '``llvm.lifetime.start``' Intrinsic
8284 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8291 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8296 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8302 The first argument is a constant integer representing the size of the
8303 object, or -1 if it is variable sized. The second argument is a pointer
8309 This intrinsic indicates that before this point in the code, the value
8310 of the memory pointed to by ``ptr`` is dead. This means that it is known
8311 to never be used and has an undefined value. A load from the pointer
8312 that precedes this intrinsic can be replaced with ``'undef'``.
8314 '``llvm.lifetime.end``' Intrinsic
8315 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8322 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8327 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8333 The first argument is a constant integer representing the size of the
8334 object, or -1 if it is variable sized. The second argument is a pointer
8340 This intrinsic indicates that after this point in the code, the value of
8341 the memory pointed to by ``ptr`` is dead. This means that it is known to
8342 never be used and has an undefined value. Any stores into the memory
8343 object following this intrinsic may be removed as dead.
8345 '``llvm.invariant.start``' Intrinsic
8346 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8353 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8358 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8359 a memory object will not change.
8364 The first argument is a constant integer representing the size of the
8365 object, or -1 if it is variable sized. The second argument is a pointer
8371 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8372 the return value, the referenced memory location is constant and
8375 '``llvm.invariant.end``' Intrinsic
8376 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8383 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8388 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8389 memory object are mutable.
8394 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8395 The second argument is a constant integer representing the size of the
8396 object, or -1 if it is variable sized and the third argument is a
8397 pointer to the object.
8402 This intrinsic indicates that the memory is mutable again.
8407 This class of intrinsics is designed to be generic and has no specific
8410 '``llvm.var.annotation``' Intrinsic
8411 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8418 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8423 The '``llvm.var.annotation``' intrinsic.
8428 The first argument is a pointer to a value, the second is a pointer to a
8429 global string, the third is a pointer to a global string which is the
8430 source file name, and the last argument is the line number.
8435 This intrinsic allows annotation of local variables with arbitrary
8436 strings. This can be useful for special purpose optimizations that want
8437 to look for these annotations. These have no other defined use; they are
8438 ignored by code generation and optimization.
8440 '``llvm.ptr.annotation.*``' Intrinsic
8441 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8446 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8447 pointer to an integer of any width. *NOTE* you must specify an address space for
8448 the pointer. The identifier for the default address space is the integer
8453 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8454 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8455 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8456 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8457 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8462 The '``llvm.ptr.annotation``' intrinsic.
8467 The first argument is a pointer to an integer value of arbitrary bitwidth
8468 (result of some expression), the second is a pointer to a global string, the
8469 third is a pointer to a global string which is the source file name, and the
8470 last argument is the line number. It returns the value of the first argument.
8475 This intrinsic allows annotation of a pointer to an integer with arbitrary
8476 strings. This can be useful for special purpose optimizations that want to look
8477 for these annotations. These have no other defined use; they are ignored by code
8478 generation and optimization.
8480 '``llvm.annotation.*``' Intrinsic
8481 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8486 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8487 any integer bit width.
8491 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8492 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8493 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8494 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8495 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8500 The '``llvm.annotation``' intrinsic.
8505 The first argument is an integer value (result of some expression), the
8506 second is a pointer to a global string, the third is a pointer to a
8507 global string which is the source file name, and the last argument is
8508 the line number. It returns the value of the first argument.
8513 This intrinsic allows annotations to be put on arbitrary expressions
8514 with arbitrary strings. This can be useful for special purpose
8515 optimizations that want to look for these annotations. These have no
8516 other defined use; they are ignored by code generation and optimization.
8518 '``llvm.trap``' Intrinsic
8519 ^^^^^^^^^^^^^^^^^^^^^^^^^
8526 declare void @llvm.trap() noreturn nounwind
8531 The '``llvm.trap``' intrinsic.
8541 This intrinsic is lowered to the target dependent trap instruction. If
8542 the target does not have a trap instruction, this intrinsic will be
8543 lowered to a call of the ``abort()`` function.
8545 '``llvm.debugtrap``' Intrinsic
8546 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8553 declare void @llvm.debugtrap() nounwind
8558 The '``llvm.debugtrap``' intrinsic.
8568 This intrinsic is lowered to code which is intended to cause an
8569 execution trap with the intention of requesting the attention of a
8572 '``llvm.stackprotector``' Intrinsic
8573 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8580 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8585 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8586 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8587 is placed on the stack before local variables.
8592 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8593 The first argument is the value loaded from the stack guard
8594 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8595 enough space to hold the value of the guard.
8600 This intrinsic causes the prologue/epilogue inserter to force the
8601 position of the ``AllocaInst`` stack slot to be before local variables
8602 on the stack. This is to ensure that if a local variable on the stack is
8603 overwritten, it will destroy the value of the guard. When the function
8604 exits, the guard on the stack is checked against the original guard. If
8605 they are different, then the program aborts by calling the
8606 ``__stack_chk_fail()`` function.
8608 '``llvm.objectsize``' Intrinsic
8609 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8616 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8617 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8622 The ``llvm.objectsize`` intrinsic is designed to provide information to
8623 the optimizers to determine at compile time whether a) an operation
8624 (like memcpy) will overflow a buffer that corresponds to an object, or
8625 b) that a runtime check for overflow isn't necessary. An object in this
8626 context means an allocation of a specific class, structure, array, or
8632 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8633 argument is a pointer to or into the ``object``. The second argument is
8634 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8635 or -1 (if false) when the object size is unknown. The second argument
8636 only accepts constants.
8641 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8642 the size of the object concerned. If the size cannot be determined at
8643 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8644 on the ``min`` argument).
8646 '``llvm.expect``' Intrinsic
8647 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8654 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8655 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8660 The ``llvm.expect`` intrinsic provides information about expected (the
8661 most probable) value of ``val``, which can be used by optimizers.
8666 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8667 a value. The second argument is an expected value, this needs to be a
8668 constant value, variables are not allowed.
8673 This intrinsic is lowered to the ``val``.
8675 '``llvm.donothing``' Intrinsic
8676 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8683 declare void @llvm.donothing() nounwind readnone
8688 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8689 only intrinsic that can be called with an invoke instruction.
8699 This intrinsic does nothing, and it's removed by optimizers and ignored