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 indicates that the source code contained a hint that
833 inlining this function is desirable (such as the "inline" keyword in
834 C/C++). It is just a hint; it imposes no requirements on the
837 This attribute suggests that optimization passes and code generator
838 passes make choices that keep the code size of this function as small
839 as possible and perform optimizations that may sacrifice runtime
840 performance in order to minimize the size of the generated code.
842 This attribute disables prologue / epilogue emission for the
843 function. This can have very system-specific consequences.
845 This indicates that the callee function at a call site is not recognized as
846 a built-in function. LLVM will retain the original call and not replace it
847 with equivalent code based on the semantics of the built-in function, unless
848 the call site uses the ``builtin`` attribute. This is valid at call sites
849 and on function declarations and definitions.
851 This attribute indicates that calls to the function cannot be
852 duplicated. A call to a ``noduplicate`` function may be moved
853 within its parent function, but may not be duplicated within
856 A function containing a ``noduplicate`` call may still
857 be an inlining candidate, provided that the call is not
858 duplicated by inlining. That implies that the function has
859 internal linkage and only has one call site, so the original
860 call is dead after inlining.
862 This attributes disables implicit floating point instructions.
864 This attribute indicates that the inliner should never inline this
865 function in any situation. This attribute may not be used together
866 with the ``alwaysinline`` attribute.
868 This attribute suppresses lazy symbol binding for the function. This
869 may make calls to the function faster, at the cost of extra program
870 startup time if the function is not called during program startup.
872 This attribute indicates that the code generator should not use a
873 red zone, even if the target-specific ABI normally permits it.
875 This function attribute indicates that the function never returns
876 normally. This produces undefined behavior at runtime if the
877 function ever does dynamically return.
879 This function attribute indicates that the function never returns
880 with an unwind or exceptional control flow. If the function does
881 unwind, its runtime behavior is undefined.
883 This attribute suggests that optimization passes and code generator
884 passes make choices that keep the code size of this function low,
885 and otherwise do optimizations specifically to reduce code size as
886 long as they do not significantly impact runtime performance.
888 On a function, this attribute indicates that the function computes its
889 result (or decides to unwind an exception) based strictly on its arguments,
890 without dereferencing any pointer arguments or otherwise accessing
891 any mutable state (e.g. memory, control registers, etc) visible to
892 caller functions. It does not write through any pointer arguments
893 (including ``byval`` arguments) and never changes any state visible
894 to callers. This means that it cannot unwind exceptions by calling
895 the ``C++`` exception throwing methods.
897 On an argument, this attribute indicates that the function does not
898 dereference that pointer argument, even though it may read or write the
899 memory that the pointer points to if accessed through other pointers.
901 On a function, this attribute indicates that the function does not write
902 through any pointer arguments (including ``byval`` arguments) or otherwise
903 modify any state (e.g. memory, control registers, etc) visible to
904 caller functions. It may dereference pointer arguments and read
905 state that may be set in the caller. A readonly function always
906 returns the same value (or unwinds an exception identically) when
907 called with the same set of arguments and global state. It cannot
908 unwind an exception by calling the ``C++`` exception throwing
911 On an argument, this attribute indicates that the function does not write
912 through this pointer argument, even though it may write to the memory that
913 the pointer points to.
915 This attribute indicates that this function can return twice. The C
916 ``setjmp`` is an example of such a function. The compiler disables
917 some optimizations (like tail calls) in the caller of these
920 This attribute indicates that AddressSanitizer checks
921 (dynamic address safety analysis) are enabled for this function.
923 This attribute indicates that MemorySanitizer checks (dynamic detection
924 of accesses to uninitialized memory) are enabled for this function.
926 This attribute indicates that ThreadSanitizer checks
927 (dynamic thread safety analysis) are enabled for this function.
929 This attribute indicates that the function should emit a stack
930 smashing protector. It is in the form of a "canary" --- a random value
931 placed on the stack before the local variables that's checked upon
932 return from the function to see if it has been overwritten. A
933 heuristic is used to determine if a function needs stack protectors
934 or not. The heuristic used will enable protectors for functions with:
936 - Character arrays larger than ``ssp-buffer-size`` (default 8).
937 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
938 - Calls to alloca() with variable sizes or constant sizes greater than
941 If a function that has an ``ssp`` attribute is inlined into a
942 function that doesn't have an ``ssp`` attribute, then the resulting
943 function will have an ``ssp`` attribute.
945 This attribute indicates that the function should *always* emit a
946 stack smashing protector. This overrides the ``ssp`` function
949 If a function that has an ``sspreq`` attribute is inlined into a
950 function that doesn't have an ``sspreq`` attribute or which has an
951 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
952 an ``sspreq`` attribute.
954 This attribute indicates that the function should emit a stack smashing
955 protector. This attribute causes a strong heuristic to be used when
956 determining if a function needs stack protectors. The strong heuristic
957 will enable protectors for functions with:
959 - Arrays of any size and type
960 - Aggregates containing an array of any size and type.
962 - Local variables that have had their address taken.
964 This overrides the ``ssp`` function attribute.
966 If a function that has an ``sspstrong`` attribute is inlined into a
967 function that doesn't have an ``sspstrong`` attribute, then the
968 resulting function will have an ``sspstrong`` attribute.
970 This attribute indicates that the ABI being targeted requires that
971 an unwind table entry be produce for this function even if we can
972 show that no exceptions passes by it. This is normally the case for
973 the ELF x86-64 abi, but it can be disabled for some compilation
978 Module-Level Inline Assembly
979 ----------------------------
981 Modules may contain "module-level inline asm" blocks, which corresponds
982 to the GCC "file scope inline asm" blocks. These blocks are internally
983 concatenated by LLVM and treated as a single unit, but may be separated
984 in the ``.ll`` file if desired. The syntax is very simple:
988 module asm "inline asm code goes here"
989 module asm "more can go here"
991 The strings can contain any character by escaping non-printable
992 characters. The escape sequence used is simply "\\xx" where "xx" is the
993 two digit hex code for the number.
995 The inline asm code is simply printed to the machine code .s file when
996 assembly code is generated.
998 .. _langref_datalayout:
1003 A module may specify a target specific data layout string that specifies
1004 how data is to be laid out in memory. The syntax for the data layout is
1007 .. code-block:: llvm
1009 target datalayout = "layout specification"
1011 The *layout specification* consists of a list of specifications
1012 separated by the minus sign character ('-'). Each specification starts
1013 with a letter and may include other information after the letter to
1014 define some aspect of the data layout. The specifications accepted are
1018 Specifies that the target lays out data in big-endian form. That is,
1019 the bits with the most significance have the lowest address
1022 Specifies that the target lays out data in little-endian form. That
1023 is, the bits with the least significance have the lowest address
1026 Specifies the natural alignment of the stack in bits. Alignment
1027 promotion of stack variables is limited to the natural stack
1028 alignment to avoid dynamic stack realignment. The stack alignment
1029 must be a multiple of 8-bits. If omitted, the natural stack
1030 alignment defaults to "unspecified", which does not prevent any
1031 alignment promotions.
1032 ``p[n]:<size>:<abi>:<pref>``
1033 This specifies the *size* of a pointer and its ``<abi>`` and
1034 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1035 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
1036 preceding ``:`` should be omitted too. The address space, ``n`` is
1037 optional, and if not specified, denotes the default address space 0.
1038 The value of ``n`` must be in the range [1,2^23).
1039 ``i<size>:<abi>:<pref>``
1040 This specifies the alignment for an integer type of a given bit
1041 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1042 ``v<size>:<abi>:<pref>``
1043 This specifies the alignment for a vector type of a given bit
1045 ``f<size>:<abi>:<pref>``
1046 This specifies the alignment for a floating point type of a given bit
1047 ``<size>``. Only values of ``<size>`` that are supported by the target
1048 will work. 32 (float) and 64 (double) are supported on all targets; 80
1049 or 128 (different flavors of long double) are also supported on some
1051 ``a<size>:<abi>:<pref>``
1052 This specifies the alignment for an aggregate type of a given bit
1054 ``s<size>:<abi>:<pref>``
1055 This specifies the alignment for a stack object of a given bit
1057 ``n<size1>:<size2>:<size3>...``
1058 This specifies a set of native integer widths for the target CPU in
1059 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1060 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1061 this set are considered to support most general arithmetic operations
1064 When constructing the data layout for a given target, LLVM starts with a
1065 default set of specifications which are then (possibly) overridden by
1066 the specifications in the ``datalayout`` keyword. The default
1067 specifications are given in this list:
1069 - ``E`` - big endian
1070 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1071 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1072 same as the default address space.
1073 - ``S0`` - natural stack alignment is unspecified
1074 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1075 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1076 - ``i16:16:16`` - i16 is 16-bit aligned
1077 - ``i32:32:32`` - i32 is 32-bit aligned
1078 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1079 alignment of 64-bits
1080 - ``f16:16:16`` - half is 16-bit aligned
1081 - ``f32:32:32`` - float is 32-bit aligned
1082 - ``f64:64:64`` - double is 64-bit aligned
1083 - ``f128:128:128`` - quad is 128-bit aligned
1084 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1085 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1086 - ``a0:0:64`` - aggregates are 64-bit aligned
1088 When LLVM is determining the alignment for a given type, it uses the
1091 #. If the type sought is an exact match for one of the specifications,
1092 that specification is used.
1093 #. If no match is found, and the type sought is an integer type, then
1094 the smallest integer type that is larger than the bitwidth of the
1095 sought type is used. If none of the specifications are larger than
1096 the bitwidth then the largest integer type is used. For example,
1097 given the default specifications above, the i7 type will use the
1098 alignment of i8 (next largest) while both i65 and i256 will use the
1099 alignment of i64 (largest specified).
1100 #. If no match is found, and the type sought is a vector type, then the
1101 largest vector type that is smaller than the sought vector type will
1102 be used as a fall back. This happens because <128 x double> can be
1103 implemented in terms of 64 <2 x double>, for example.
1105 The function of the data layout string may not be what you expect.
1106 Notably, this is not a specification from the frontend of what alignment
1107 the code generator should use.
1109 Instead, if specified, the target data layout is required to match what
1110 the ultimate *code generator* expects. This string is used by the
1111 mid-level optimizers to improve code, and this only works if it matches
1112 what the ultimate code generator uses. If you would like to generate IR
1113 that does not embed this target-specific detail into the IR, then you
1114 don't have to specify the string. This will disable some optimizations
1115 that require precise layout information, but this also prevents those
1116 optimizations from introducing target specificity into the IR.
1118 .. _pointeraliasing:
1120 Pointer Aliasing Rules
1121 ----------------------
1123 Any memory access must be done through a pointer value associated with
1124 an address range of the memory access, otherwise the behavior is
1125 undefined. Pointer values are associated with address ranges according
1126 to the following rules:
1128 - A pointer value is associated with the addresses associated with any
1129 value it is *based* on.
1130 - An address of a global variable is associated with the address range
1131 of the variable's storage.
1132 - The result value of an allocation instruction is associated with the
1133 address range of the allocated storage.
1134 - A null pointer in the default address-space is associated with no
1136 - An integer constant other than zero or a pointer value returned from
1137 a function not defined within LLVM may be associated with address
1138 ranges allocated through mechanisms other than those provided by
1139 LLVM. Such ranges shall not overlap with any ranges of addresses
1140 allocated by mechanisms provided by LLVM.
1142 A pointer value is *based* on another pointer value according to the
1145 - A pointer value formed from a ``getelementptr`` operation is *based*
1146 on the first operand of the ``getelementptr``.
1147 - The result value of a ``bitcast`` is *based* on the operand of the
1149 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1150 values that contribute (directly or indirectly) to the computation of
1151 the pointer's value.
1152 - The "*based* on" relationship is transitive.
1154 Note that this definition of *"based"* is intentionally similar to the
1155 definition of *"based"* in C99, though it is slightly weaker.
1157 LLVM IR does not associate types with memory. The result type of a
1158 ``load`` merely indicates the size and alignment of the memory from
1159 which to load, as well as the interpretation of the value. The first
1160 operand type of a ``store`` similarly only indicates the size and
1161 alignment of the store.
1163 Consequently, type-based alias analysis, aka TBAA, aka
1164 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1165 :ref:`Metadata <metadata>` may be used to encode additional information
1166 which specialized optimization passes may use to implement type-based
1171 Volatile Memory Accesses
1172 ------------------------
1174 Certain memory accesses, such as :ref:`load <i_load>`'s,
1175 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1176 marked ``volatile``. The optimizers must not change the number of
1177 volatile operations or change their order of execution relative to other
1178 volatile operations. The optimizers *may* change the order of volatile
1179 operations relative to non-volatile operations. This is not Java's
1180 "volatile" and has no cross-thread synchronization behavior.
1182 IR-level volatile loads and stores cannot safely be optimized into
1183 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1184 flagged volatile. Likewise, the backend should never split or merge
1185 target-legal volatile load/store instructions.
1187 .. admonition:: Rationale
1189 Platforms may rely on volatile loads and stores of natively supported
1190 data width to be executed as single instruction. For example, in C
1191 this holds for an l-value of volatile primitive type with native
1192 hardware support, but not necessarily for aggregate types. The
1193 frontend upholds these expectations, which are intentionally
1194 unspecified in the IR. The rules above ensure that IR transformation
1195 do not violate the frontend's contract with the language.
1199 Memory Model for Concurrent Operations
1200 --------------------------------------
1202 The LLVM IR does not define any way to start parallel threads of
1203 execution or to register signal handlers. Nonetheless, there are
1204 platform-specific ways to create them, and we define LLVM IR's behavior
1205 in their presence. This model is inspired by the C++0x memory model.
1207 For a more informal introduction to this model, see the :doc:`Atomics`.
1209 We define a *happens-before* partial order as the least partial order
1212 - Is a superset of single-thread program order, and
1213 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1214 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1215 techniques, like pthread locks, thread creation, thread joining,
1216 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1217 Constraints <ordering>`).
1219 Note that program order does not introduce *happens-before* edges
1220 between a thread and signals executing inside that thread.
1222 Every (defined) read operation (load instructions, memcpy, atomic
1223 loads/read-modify-writes, etc.) R reads a series of bytes written by
1224 (defined) write operations (store instructions, atomic
1225 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1226 section, initialized globals are considered to have a write of the
1227 initializer which is atomic and happens before any other read or write
1228 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1229 may see any write to the same byte, except:
1231 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1232 write\ :sub:`2` happens before R\ :sub:`byte`, then
1233 R\ :sub:`byte` does not see write\ :sub:`1`.
1234 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1235 R\ :sub:`byte` does not see write\ :sub:`3`.
1237 Given that definition, R\ :sub:`byte` is defined as follows:
1239 - If R is volatile, the result is target-dependent. (Volatile is
1240 supposed to give guarantees which can support ``sig_atomic_t`` in
1241 C/C++, and may be used for accesses to addresses which do not behave
1242 like normal memory. It does not generally provide cross-thread
1244 - Otherwise, if there is no write to the same byte that happens before
1245 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1246 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1247 R\ :sub:`byte` returns the value written by that write.
1248 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1249 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1250 Memory Ordering Constraints <ordering>` section for additional
1251 constraints on how the choice is made.
1252 - Otherwise R\ :sub:`byte` returns ``undef``.
1254 R returns the value composed of the series of bytes it read. This
1255 implies that some bytes within the value may be ``undef`` **without**
1256 the entire value being ``undef``. Note that this only defines the
1257 semantics of the operation; it doesn't mean that targets will emit more
1258 than one instruction to read the series of bytes.
1260 Note that in cases where none of the atomic intrinsics are used, this
1261 model places only one restriction on IR transformations on top of what
1262 is required for single-threaded execution: introducing a store to a byte
1263 which might not otherwise be stored is not allowed in general.
1264 (Specifically, in the case where another thread might write to and read
1265 from an address, introducing a store can change a load that may see
1266 exactly one write into a load that may see multiple writes.)
1270 Atomic Memory Ordering Constraints
1271 ----------------------------------
1273 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1274 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1275 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1276 an ordering parameter that determines which other atomic instructions on
1277 the same address they *synchronize with*. These semantics are borrowed
1278 from Java and C++0x, but are somewhat more colloquial. If these
1279 descriptions aren't precise enough, check those specs (see spec
1280 references in the :doc:`atomics guide <Atomics>`).
1281 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1282 differently since they don't take an address. See that instruction's
1283 documentation for details.
1285 For a simpler introduction to the ordering constraints, see the
1289 The set of values that can be read is governed by the happens-before
1290 partial order. A value cannot be read unless some operation wrote
1291 it. This is intended to provide a guarantee strong enough to model
1292 Java's non-volatile shared variables. This ordering cannot be
1293 specified for read-modify-write operations; it is not strong enough
1294 to make them atomic in any interesting way.
1296 In addition to the guarantees of ``unordered``, there is a single
1297 total order for modifications by ``monotonic`` operations on each
1298 address. All modification orders must be compatible with the
1299 happens-before order. There is no guarantee that the modification
1300 orders can be combined to a global total order for the whole program
1301 (and this often will not be possible). The read in an atomic
1302 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1303 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1304 order immediately before the value it writes. If one atomic read
1305 happens before another atomic read of the same address, the later
1306 read must see the same value or a later value in the address's
1307 modification order. This disallows reordering of ``monotonic`` (or
1308 stronger) operations on the same address. If an address is written
1309 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1310 read that address repeatedly, the other threads must eventually see
1311 the write. This corresponds to the C++0x/C1x
1312 ``memory_order_relaxed``.
1314 In addition to the guarantees of ``monotonic``, a
1315 *synchronizes-with* edge may be formed with a ``release`` operation.
1316 This is intended to model C++'s ``memory_order_acquire``.
1318 In addition to the guarantees of ``monotonic``, if this operation
1319 writes a value which is subsequently read by an ``acquire``
1320 operation, it *synchronizes-with* that operation. (This isn't a
1321 complete description; see the C++0x definition of a release
1322 sequence.) This corresponds to the C++0x/C1x
1323 ``memory_order_release``.
1324 ``acq_rel`` (acquire+release)
1325 Acts as both an ``acquire`` and ``release`` operation on its
1326 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1327 ``seq_cst`` (sequentially consistent)
1328 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1329 operation which only reads, ``release`` for an operation which only
1330 writes), there is a global total order on all
1331 sequentially-consistent operations on all addresses, which is
1332 consistent with the *happens-before* partial order and with the
1333 modification orders of all the affected addresses. Each
1334 sequentially-consistent read sees the last preceding write to the
1335 same address in this global order. This corresponds to the C++0x/C1x
1336 ``memory_order_seq_cst`` and Java volatile.
1340 If an atomic operation is marked ``singlethread``, it only *synchronizes
1341 with* or participates in modification and seq\_cst total orderings with
1342 other operations running in the same thread (for example, in signal
1350 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1351 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1352 :ref:`frem <i_frem>`) have the following flags that can set to enable
1353 otherwise unsafe floating point operations
1356 No NaNs - Allow optimizations to assume the arguments and result are not
1357 NaN. Such optimizations are required to retain defined behavior over
1358 NaNs, but the value of the result is undefined.
1361 No Infs - Allow optimizations to assume the arguments and result are not
1362 +/-Inf. Such optimizations are required to retain defined behavior over
1363 +/-Inf, but the value of the result is undefined.
1366 No Signed Zeros - Allow optimizations to treat the sign of a zero
1367 argument or result as insignificant.
1370 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1371 argument rather than perform division.
1374 Fast - Allow algebraically equivalent transformations that may
1375 dramatically change results in floating point (e.g. reassociate). This
1376 flag implies all the others.
1383 The LLVM type system is one of the most important features of the
1384 intermediate representation. Being typed enables a number of
1385 optimizations to be performed on the intermediate representation
1386 directly, without having to do extra analyses on the side before the
1387 transformation. A strong type system makes it easier to read the
1388 generated code and enables novel analyses and transformations that are
1389 not feasible to perform on normal three address code representations.
1391 .. _typeclassifications:
1393 Type Classifications
1394 --------------------
1396 The types fall into a few useful classifications:
1405 * - :ref:`integer <t_integer>`
1406 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1409 * - :ref:`floating point <t_floating>`
1410 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1418 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1419 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1420 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1421 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1423 * - :ref:`primitive <t_primitive>`
1424 - :ref:`label <t_label>`,
1425 :ref:`void <t_void>`,
1426 :ref:`integer <t_integer>`,
1427 :ref:`floating point <t_floating>`,
1428 :ref:`x86mmx <t_x86mmx>`,
1429 :ref:`metadata <t_metadata>`.
1431 * - :ref:`derived <t_derived>`
1432 - :ref:`array <t_array>`,
1433 :ref:`function <t_function>`,
1434 :ref:`pointer <t_pointer>`,
1435 :ref:`structure <t_struct>`,
1436 :ref:`vector <t_vector>`,
1437 :ref:`opaque <t_opaque>`.
1439 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1440 Values of these types are the only ones which can be produced by
1448 The primitive types are the fundamental building blocks of the LLVM
1459 The integer type is a very simple type that simply specifies an
1460 arbitrary bit width for the integer type desired. Any bit width from 1
1461 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1470 The number of bits the integer will occupy is specified by the ``N``
1476 +----------------+------------------------------------------------+
1477 | ``i1`` | a single-bit integer. |
1478 +----------------+------------------------------------------------+
1479 | ``i32`` | a 32-bit integer. |
1480 +----------------+------------------------------------------------+
1481 | ``i1942652`` | a really big integer of over 1 million bits. |
1482 +----------------+------------------------------------------------+
1486 Floating Point Types
1487 ^^^^^^^^^^^^^^^^^^^^
1496 - 16-bit floating point value
1499 - 32-bit floating point value
1502 - 64-bit floating point value
1505 - 128-bit floating point value (112-bit mantissa)
1508 - 80-bit floating point value (X87)
1511 - 128-bit floating point value (two 64-bits)
1521 The x86mmx type represents a value held in an MMX register on an x86
1522 machine. The operations allowed on it are quite limited: parameters and
1523 return values, load and store, and bitcast. User-specified MMX
1524 instructions are represented as intrinsic or asm calls with arguments
1525 and/or results of this type. There are no arrays, vectors or constants
1543 The void type does not represent any value and has no size.
1560 The label type represents code labels.
1577 The metadata type represents embedded metadata. No derived types may be
1578 created from metadata except for :ref:`function <t_function>` arguments.
1592 The real power in LLVM comes from the derived types in the system. This
1593 is what allows a programmer to represent arrays, functions, pointers,
1594 and other useful types. Each of these types contain one or more element
1595 types which may be a primitive type, or another derived type. For
1596 example, it is possible to have a two dimensional array, using an array
1597 as the element type of another array.
1604 Aggregate Types are a subset of derived types that can contain multiple
1605 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1606 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1617 The array type is a very simple derived type that arranges elements
1618 sequentially in memory. The array type requires a size (number of
1619 elements) and an underlying data type.
1626 [<# elements> x <elementtype>]
1628 The number of elements is a constant integer value; ``elementtype`` may
1629 be any type with a size.
1634 +------------------+--------------------------------------+
1635 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1636 +------------------+--------------------------------------+
1637 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1638 +------------------+--------------------------------------+
1639 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1640 +------------------+--------------------------------------+
1642 Here are some examples of multidimensional arrays:
1644 +-----------------------------+----------------------------------------------------------+
1645 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1646 +-----------------------------+----------------------------------------------------------+
1647 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1648 +-----------------------------+----------------------------------------------------------+
1649 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1650 +-----------------------------+----------------------------------------------------------+
1652 There is no restriction on indexing beyond the end of the array implied
1653 by a static type (though there are restrictions on indexing beyond the
1654 bounds of an allocated object in some cases). This means that
1655 single-dimension 'variable sized array' addressing can be implemented in
1656 LLVM with a zero length array type. An implementation of 'pascal style
1657 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1668 The function type can be thought of as a function signature. It consists
1669 of a return type and a list of formal parameter types. The return type
1670 of a function type is a first class type or a void type.
1677 <returntype> (<parameter list>)
1679 ...where '``<parameter list>``' is a comma-separated list of type
1680 specifiers. Optionally, the parameter list may include a type ``...``,
1681 which indicates that the function takes a variable number of arguments.
1682 Variable argument functions can access their arguments with the
1683 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1684 '``<returntype>``' is any type except :ref:`label <t_label>`.
1689 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1690 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1691 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1692 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1693 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1694 | ``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. |
1695 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1696 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1697 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1707 The structure type is used to represent a collection of data members
1708 together in memory. The elements of a structure may be any type that has
1711 Structures in memory are accessed using '``load``' and '``store``' by
1712 getting a pointer to a field with the '``getelementptr``' instruction.
1713 Structures in registers are accessed using the '``extractvalue``' and
1714 '``insertvalue``' instructions.
1716 Structures may optionally be "packed" structures, which indicate that
1717 the alignment of the struct is one byte, and that there is no padding
1718 between the elements. In non-packed structs, padding between field types
1719 is inserted as defined by the DataLayout string in the module, which is
1720 required to match what the underlying code generator expects.
1722 Structures can either be "literal" or "identified". A literal structure
1723 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1724 identified types are always defined at the top level with a name.
1725 Literal types are uniqued by their contents and can never be recursive
1726 or opaque since there is no way to write one. Identified types can be
1727 recursive, can be opaqued, and are never uniqued.
1734 %T1 = type { <type list> } ; Identified normal struct type
1735 %T2 = type <{ <type list> }> ; Identified packed struct type
1740 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1741 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1742 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1743 | ``{ 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``. |
1744 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1745 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1746 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1750 Opaque Structure Types
1751 ^^^^^^^^^^^^^^^^^^^^^^
1756 Opaque structure types are used to represent named structure types that
1757 do not have a body specified. This corresponds (for example) to the C
1758 notion of a forward declared structure.
1771 +--------------+-------------------+
1772 | ``opaque`` | An opaque type. |
1773 +--------------+-------------------+
1783 The pointer type is used to specify memory locations. Pointers are
1784 commonly used to reference objects in memory.
1786 Pointer types may have an optional address space attribute defining the
1787 numbered address space where the pointed-to object resides. The default
1788 address space is number zero. The semantics of non-zero address spaces
1789 are target-specific.
1791 Note that LLVM does not permit pointers to void (``void*``) nor does it
1792 permit pointers to labels (``label*``). Use ``i8*`` instead.
1804 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1805 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1806 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1807 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1808 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1809 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1810 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1820 A vector type is a simple derived type that represents a vector of
1821 elements. Vector types are used when multiple primitive data are
1822 operated in parallel using a single instruction (SIMD). A vector type
1823 requires a size (number of elements) and an underlying primitive data
1824 type. Vector types are considered :ref:`first class <t_firstclass>`.
1831 < <# elements> x <elementtype> >
1833 The number of elements is a constant integer value larger than 0;
1834 elementtype may be any integer or floating point type, or a pointer to
1835 these types. Vectors of size zero are not allowed.
1840 +-------------------+--------------------------------------------------+
1841 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1842 +-------------------+--------------------------------------------------+
1843 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1844 +-------------------+--------------------------------------------------+
1845 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1846 +-------------------+--------------------------------------------------+
1847 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1848 +-------------------+--------------------------------------------------+
1853 LLVM has several different basic types of constants. This section
1854 describes them all and their syntax.
1859 **Boolean constants**
1860 The two strings '``true``' and '``false``' are both valid constants
1862 **Integer constants**
1863 Standard integers (such as '4') are constants of the
1864 :ref:`integer <t_integer>` type. Negative numbers may be used with
1866 **Floating point constants**
1867 Floating point constants use standard decimal notation (e.g.
1868 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1869 hexadecimal notation (see below). The assembler requires the exact
1870 decimal value of a floating-point constant. For example, the
1871 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1872 decimal in binary. Floating point constants must have a :ref:`floating
1873 point <t_floating>` type.
1874 **Null pointer constants**
1875 The identifier '``null``' is recognized as a null pointer constant
1876 and must be of :ref:`pointer type <t_pointer>`.
1878 The one non-intuitive notation for constants is the hexadecimal form of
1879 floating point constants. For example, the form
1880 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1881 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1882 constants are required (and the only time that they are generated by the
1883 disassembler) is when a floating point constant must be emitted but it
1884 cannot be represented as a decimal floating point number in a reasonable
1885 number of digits. For example, NaN's, infinities, and other special
1886 values are represented in their IEEE hexadecimal format so that assembly
1887 and disassembly do not cause any bits to change in the constants.
1889 When using the hexadecimal form, constants of types half, float, and
1890 double are represented using the 16-digit form shown above (which
1891 matches the IEEE754 representation for double); half and float values
1892 must, however, be exactly representable as IEEE 754 half and single
1893 precision, respectively. Hexadecimal format is always used for long
1894 double, and there are three forms of long double. The 80-bit format used
1895 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1896 128-bit format used by PowerPC (two adjacent doubles) is represented by
1897 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1898 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
1899 will only work if they match the long double format on your target.
1900 The IEEE 16-bit format (half precision) is represented by ``0xH``
1901 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
1902 (sign bit at the left).
1904 There are no constants of type x86mmx.
1906 .. _complexconstants:
1911 Complex constants are a (potentially recursive) combination of simple
1912 constants and smaller complex constants.
1914 **Structure constants**
1915 Structure constants are represented with notation similar to
1916 structure type definitions (a comma separated list of elements,
1917 surrounded by braces (``{}``)). For example:
1918 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1919 "``@G = external global i32``". Structure constants must have
1920 :ref:`structure type <t_struct>`, and the number and types of elements
1921 must match those specified by the type.
1923 Array constants are represented with notation similar to array type
1924 definitions (a comma separated list of elements, surrounded by
1925 square brackets (``[]``)). For example:
1926 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1927 :ref:`array type <t_array>`, and the number and types of elements must
1928 match those specified by the type.
1929 **Vector constants**
1930 Vector constants are represented with notation similar to vector
1931 type definitions (a comma separated list of elements, surrounded by
1932 less-than/greater-than's (``<>``)). For example:
1933 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1934 must have :ref:`vector type <t_vector>`, and the number and types of
1935 elements must match those specified by the type.
1936 **Zero initialization**
1937 The string '``zeroinitializer``' can be used to zero initialize a
1938 value to zero of *any* type, including scalar and
1939 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1940 having to print large zero initializers (e.g. for large arrays) and
1941 is always exactly equivalent to using explicit zero initializers.
1943 A metadata node is a structure-like constant with :ref:`metadata
1944 type <t_metadata>`. For example:
1945 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
1946 constants that are meant to be interpreted as part of the
1947 instruction stream, metadata is a place to attach additional
1948 information such as debug info.
1950 Global Variable and Function Addresses
1951 --------------------------------------
1953 The addresses of :ref:`global variables <globalvars>` and
1954 :ref:`functions <functionstructure>` are always implicitly valid
1955 (link-time) constants. These constants are explicitly referenced when
1956 the :ref:`identifier for the global <identifiers>` is used and always have
1957 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
1960 .. code-block:: llvm
1964 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1971 The string '``undef``' can be used anywhere a constant is expected, and
1972 indicates that the user of the value may receive an unspecified
1973 bit-pattern. Undefined values may be of any type (other than '``label``'
1974 or '``void``') and be used anywhere a constant is permitted.
1976 Undefined values are useful because they indicate to the compiler that
1977 the program is well defined no matter what value is used. This gives the
1978 compiler more freedom to optimize. Here are some examples of
1979 (potentially surprising) transformations that are valid (in pseudo IR):
1981 .. code-block:: llvm
1991 This is safe because all of the output bits are affected by the undef
1992 bits. Any output bit can have a zero or one depending on the input bits.
1994 .. code-block:: llvm
2005 These logical operations have bits that are not always affected by the
2006 input. For example, if ``%X`` has a zero bit, then the output of the
2007 '``and``' operation will always be a zero for that bit, no matter what
2008 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2009 optimize or assume that the result of the '``and``' is '``undef``'.
2010 However, it is safe to assume that all bits of the '``undef``' could be
2011 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2012 all the bits of the '``undef``' operand to the '``or``' could be set,
2013 allowing the '``or``' to be folded to -1.
2015 .. code-block:: llvm
2017 %A = select undef, %X, %Y
2018 %B = select undef, 42, %Y
2019 %C = select %X, %Y, undef
2029 This set of examples shows that undefined '``select``' (and conditional
2030 branch) conditions can go *either way*, but they have to come from one
2031 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2032 both known to have a clear low bit, then ``%A`` would have to have a
2033 cleared low bit. However, in the ``%C`` example, the optimizer is
2034 allowed to assume that the '``undef``' operand could be the same as
2035 ``%Y``, allowing the whole '``select``' to be eliminated.
2037 .. code-block:: llvm
2039 %A = xor undef, undef
2056 This example points out that two '``undef``' operands are not
2057 necessarily the same. This can be surprising to people (and also matches
2058 C semantics) where they assume that "``X^X``" is always zero, even if
2059 ``X`` is undefined. This isn't true for a number of reasons, but the
2060 short answer is that an '``undef``' "variable" can arbitrarily change
2061 its value over its "live range". This is true because the variable
2062 doesn't actually *have a live range*. Instead, the value is logically
2063 read from arbitrary registers that happen to be around when needed, so
2064 the value is not necessarily consistent over time. In fact, ``%A`` and
2065 ``%C`` need to have the same semantics or the core LLVM "replace all
2066 uses with" concept would not hold.
2068 .. code-block:: llvm
2076 These examples show the crucial difference between an *undefined value*
2077 and *undefined behavior*. An undefined value (like '``undef``') is
2078 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2079 operation can be constant folded to '``undef``', because the '``undef``'
2080 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2081 However, in the second example, we can make a more aggressive
2082 assumption: because the ``undef`` is allowed to be an arbitrary value,
2083 we are allowed to assume that it could be zero. Since a divide by zero
2084 has *undefined behavior*, we are allowed to assume that the operation
2085 does not execute at all. This allows us to delete the divide and all
2086 code after it. Because the undefined operation "can't happen", the
2087 optimizer can assume that it occurs in dead code.
2089 .. code-block:: llvm
2091 a: store undef -> %X
2092 b: store %X -> undef
2097 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2098 value can be assumed to not have any effect; we can assume that the
2099 value is overwritten with bits that happen to match what was already
2100 there. However, a store *to* an undefined location could clobber
2101 arbitrary memory, therefore, it has undefined behavior.
2108 Poison values are similar to :ref:`undef values <undefvalues>`, however
2109 they also represent the fact that an instruction or constant expression
2110 which cannot evoke side effects has nevertheless detected a condition
2111 which results in undefined behavior.
2113 There is currently no way of representing a poison value in the IR; they
2114 only exist when produced by operations such as :ref:`add <i_add>` with
2117 Poison value behavior is defined in terms of value *dependence*:
2119 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2120 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2121 their dynamic predecessor basic block.
2122 - Function arguments depend on the corresponding actual argument values
2123 in the dynamic callers of their functions.
2124 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2125 instructions that dynamically transfer control back to them.
2126 - :ref:`Invoke <i_invoke>` instructions depend on the
2127 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2128 call instructions that dynamically transfer control back to them.
2129 - Non-volatile loads and stores depend on the most recent stores to all
2130 of the referenced memory addresses, following the order in the IR
2131 (including loads and stores implied by intrinsics such as
2132 :ref:`@llvm.memcpy <int_memcpy>`.)
2133 - An instruction with externally visible side effects depends on the
2134 most recent preceding instruction with externally visible side
2135 effects, following the order in the IR. (This includes :ref:`volatile
2136 operations <volatile>`.)
2137 - An instruction *control-depends* on a :ref:`terminator
2138 instruction <terminators>` if the terminator instruction has
2139 multiple successors and the instruction is always executed when
2140 control transfers to one of the successors, and may not be executed
2141 when control is transferred to another.
2142 - Additionally, an instruction also *control-depends* on a terminator
2143 instruction if the set of instructions it otherwise depends on would
2144 be different if the terminator had transferred control to a different
2146 - Dependence is transitive.
2148 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2149 with the additional affect that any instruction which has a *dependence*
2150 on a poison value has undefined behavior.
2152 Here are some examples:
2154 .. code-block:: llvm
2157 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2158 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2159 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2160 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2162 store i32 %poison, i32* @g ; Poison value stored to memory.
2163 %poison2 = load i32* @g ; Poison value loaded back from memory.
2165 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2167 %narrowaddr = bitcast i32* @g to i16*
2168 %wideaddr = bitcast i32* @g to i64*
2169 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2170 %poison4 = load i64* %wideaddr ; Returns a poison value.
2172 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2173 br i1 %cmp, label %true, label %end ; Branch to either destination.
2176 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2177 ; it has undefined behavior.
2181 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2182 ; Both edges into this PHI are
2183 ; control-dependent on %cmp, so this
2184 ; always results in a poison value.
2186 store volatile i32 0, i32* @g ; This would depend on the store in %true
2187 ; if %cmp is true, or the store in %entry
2188 ; otherwise, so this is undefined behavior.
2190 br i1 %cmp, label %second_true, label %second_end
2191 ; The same branch again, but this time the
2192 ; true block doesn't have side effects.
2199 store volatile i32 0, i32* @g ; This time, the instruction always depends
2200 ; on the store in %end. Also, it is
2201 ; control-equivalent to %end, so this is
2202 ; well-defined (ignoring earlier undefined
2203 ; behavior in this example).
2207 Addresses of Basic Blocks
2208 -------------------------
2210 ``blockaddress(@function, %block)``
2212 The '``blockaddress``' constant computes the address of the specified
2213 basic block in the specified function, and always has an ``i8*`` type.
2214 Taking the address of the entry block is illegal.
2216 This value only has defined behavior when used as an operand to the
2217 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2218 against null. Pointer equality tests between labels addresses results in
2219 undefined behavior --- though, again, comparison against null is ok, and
2220 no label is equal to the null pointer. This may be passed around as an
2221 opaque pointer sized value as long as the bits are not inspected. This
2222 allows ``ptrtoint`` and arithmetic to be performed on these values so
2223 long as the original value is reconstituted before the ``indirectbr``
2226 Finally, some targets may provide defined semantics when using the value
2227 as the operand to an inline assembly, but that is target specific.
2231 Constant Expressions
2232 --------------------
2234 Constant expressions are used to allow expressions involving other
2235 constants to be used as constants. Constant expressions may be of any
2236 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2237 that does not have side effects (e.g. load and call are not supported).
2238 The following is the syntax for constant expressions:
2240 ``trunc (CST to TYPE)``
2241 Truncate a constant to another type. The bit size of CST must be
2242 larger than the bit size of TYPE. Both types must be integers.
2243 ``zext (CST to TYPE)``
2244 Zero extend a constant to another type. The bit size of CST must be
2245 smaller than the bit size of TYPE. Both types must be integers.
2246 ``sext (CST to TYPE)``
2247 Sign extend a constant to another type. The bit size of CST must be
2248 smaller than the bit size of TYPE. Both types must be integers.
2249 ``fptrunc (CST to TYPE)``
2250 Truncate a floating point constant to another floating point type.
2251 The size of CST must be larger than the size of TYPE. Both types
2252 must be floating point.
2253 ``fpext (CST to TYPE)``
2254 Floating point extend a constant to another type. The size of CST
2255 must be smaller or equal to the size of TYPE. Both types must be
2257 ``fptoui (CST to TYPE)``
2258 Convert a floating point constant to the corresponding unsigned
2259 integer constant. TYPE must be a scalar or vector integer type. CST
2260 must be of scalar or vector floating point type. Both CST and TYPE
2261 must be scalars, or vectors of the same number of elements. If the
2262 value won't fit in the integer type, the results are undefined.
2263 ``fptosi (CST to TYPE)``
2264 Convert a floating point constant to the corresponding signed
2265 integer constant. TYPE must be a scalar or vector integer type. CST
2266 must be of scalar or vector floating point type. Both CST and TYPE
2267 must be scalars, or vectors of the same number of elements. If the
2268 value won't fit in the integer type, the results are undefined.
2269 ``uitofp (CST to TYPE)``
2270 Convert an unsigned integer constant to the corresponding floating
2271 point constant. TYPE must be a scalar or vector floating point type.
2272 CST must be of scalar or vector integer type. Both CST and TYPE must
2273 be scalars, or vectors of the same number of elements. If the value
2274 won't fit in the floating point type, the results are undefined.
2275 ``sitofp (CST to TYPE)``
2276 Convert a signed integer constant to the corresponding floating
2277 point constant. TYPE must be a scalar or vector floating point type.
2278 CST must be of scalar or vector integer type. Both CST and TYPE must
2279 be scalars, or vectors of the same number of elements. If the value
2280 won't fit in the floating point type, the results are undefined.
2281 ``ptrtoint (CST to TYPE)``
2282 Convert a pointer typed constant to the corresponding integer
2283 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2284 pointer type. The ``CST`` value is zero extended, truncated, or
2285 unchanged to make it fit in ``TYPE``.
2286 ``inttoptr (CST to TYPE)``
2287 Convert an integer constant to a pointer constant. TYPE must be a
2288 pointer type. CST must be of integer type. The CST value is zero
2289 extended, truncated, or unchanged to make it fit in a pointer size.
2290 This one is *really* dangerous!
2291 ``bitcast (CST to TYPE)``
2292 Convert a constant, CST, to another TYPE. The constraints of the
2293 operands are the same as those for the :ref:`bitcast
2294 instruction <i_bitcast>`.
2295 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2296 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2297 constants. As with the :ref:`getelementptr <i_getelementptr>`
2298 instruction, the index list may have zero or more indexes, which are
2299 required to make sense for the type of "CSTPTR".
2300 ``select (COND, VAL1, VAL2)``
2301 Perform the :ref:`select operation <i_select>` on constants.
2302 ``icmp COND (VAL1, VAL2)``
2303 Performs the :ref:`icmp operation <i_icmp>` on constants.
2304 ``fcmp COND (VAL1, VAL2)``
2305 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2306 ``extractelement (VAL, IDX)``
2307 Perform the :ref:`extractelement operation <i_extractelement>` on
2309 ``insertelement (VAL, ELT, IDX)``
2310 Perform the :ref:`insertelement operation <i_insertelement>` on
2312 ``shufflevector (VEC1, VEC2, IDXMASK)``
2313 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2315 ``extractvalue (VAL, IDX0, IDX1, ...)``
2316 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2317 constants. The index list is interpreted in a similar manner as
2318 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2319 least one index value must be specified.
2320 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2321 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2322 The index list is interpreted in a similar manner as indices in a
2323 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2324 value must be specified.
2325 ``OPCODE (LHS, RHS)``
2326 Perform the specified operation of the LHS and RHS constants. OPCODE
2327 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2328 binary <bitwiseops>` operations. The constraints on operands are
2329 the same as those for the corresponding instruction (e.g. no bitwise
2330 operations on floating point values are allowed).
2337 Inline Assembler Expressions
2338 ----------------------------
2340 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2341 Inline Assembly <moduleasm>`) through the use of a special value. This
2342 value represents the inline assembler as a string (containing the
2343 instructions to emit), a list of operand constraints (stored as a
2344 string), a flag that indicates whether or not the inline asm expression
2345 has side effects, and a flag indicating whether the function containing
2346 the asm needs to align its stack conservatively. An example inline
2347 assembler expression is:
2349 .. code-block:: llvm
2351 i32 (i32) asm "bswap $0", "=r,r"
2353 Inline assembler expressions may **only** be used as the callee operand
2354 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2355 Thus, typically we have:
2357 .. code-block:: llvm
2359 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2361 Inline asms with side effects not visible in the constraint list must be
2362 marked as having side effects. This is done through the use of the
2363 '``sideeffect``' keyword, like so:
2365 .. code-block:: llvm
2367 call void asm sideeffect "eieio", ""()
2369 In some cases inline asms will contain code that will not work unless
2370 the stack is aligned in some way, such as calls or SSE instructions on
2371 x86, yet will not contain code that does that alignment within the asm.
2372 The compiler should make conservative assumptions about what the asm
2373 might contain and should generate its usual stack alignment code in the
2374 prologue if the '``alignstack``' keyword is present:
2376 .. code-block:: llvm
2378 call void asm alignstack "eieio", ""()
2380 Inline asms also support using non-standard assembly dialects. The
2381 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2382 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2383 the only supported dialects. An example is:
2385 .. code-block:: llvm
2387 call void asm inteldialect "eieio", ""()
2389 If multiple keywords appear the '``sideeffect``' keyword must come
2390 first, the '``alignstack``' keyword second and the '``inteldialect``'
2396 The call instructions that wrap inline asm nodes may have a
2397 "``!srcloc``" MDNode attached to it that contains a list of constant
2398 integers. If present, the code generator will use the integer as the
2399 location cookie value when report errors through the ``LLVMContext``
2400 error reporting mechanisms. This allows a front-end to correlate backend
2401 errors that occur with inline asm back to the source code that produced
2404 .. code-block:: llvm
2406 call void asm sideeffect "something bad", ""(), !srcloc !42
2408 !42 = !{ i32 1234567 }
2410 It is up to the front-end to make sense of the magic numbers it places
2411 in the IR. If the MDNode contains multiple constants, the code generator
2412 will use the one that corresponds to the line of the asm that the error
2417 Metadata Nodes and Metadata Strings
2418 -----------------------------------
2420 LLVM IR allows metadata to be attached to instructions in the program
2421 that can convey extra information about the code to the optimizers and
2422 code generator. One example application of metadata is source-level
2423 debug information. There are two metadata primitives: strings and nodes.
2424 All metadata has the ``metadata`` type and is identified in syntax by a
2425 preceding exclamation point ('``!``').
2427 A metadata string is a string surrounded by double quotes. It can
2428 contain any character by escaping non-printable characters with
2429 "``\xx``" where "``xx``" is the two digit hex code. For example:
2432 Metadata nodes are represented with notation similar to structure
2433 constants (a comma separated list of elements, surrounded by braces and
2434 preceded by an exclamation point). Metadata nodes can have any values as
2435 their operand. For example:
2437 .. code-block:: llvm
2439 !{ metadata !"test\00", i32 10}
2441 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2442 metadata nodes, which can be looked up in the module symbol table. For
2445 .. code-block:: llvm
2447 !foo = metadata !{!4, !3}
2449 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2450 function is using two metadata arguments:
2452 .. code-block:: llvm
2454 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2456 Metadata can be attached with an instruction. Here metadata ``!21`` is
2457 attached to the ``add`` instruction using the ``!dbg`` identifier:
2459 .. code-block:: llvm
2461 %indvar.next = add i64 %indvar, 1, !dbg !21
2463 More information about specific metadata nodes recognized by the
2464 optimizers and code generator is found below.
2469 In LLVM IR, memory does not have types, so LLVM's own type system is not
2470 suitable for doing TBAA. Instead, metadata is added to the IR to
2471 describe a type system of a higher level language. This can be used to
2472 implement typical C/C++ TBAA, but it can also be used to implement
2473 custom alias analysis behavior for other languages.
2475 The current metadata format is very simple. TBAA metadata nodes have up
2476 to three fields, e.g.:
2478 .. code-block:: llvm
2480 !0 = metadata !{ metadata !"an example type tree" }
2481 !1 = metadata !{ metadata !"int", metadata !0 }
2482 !2 = metadata !{ metadata !"float", metadata !0 }
2483 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2485 The first field is an identity field. It can be any value, usually a
2486 metadata string, which uniquely identifies the type. The most important
2487 name in the tree is the name of the root node. Two trees with different
2488 root node names are entirely disjoint, even if they have leaves with
2491 The second field identifies the type's parent node in the tree, or is
2492 null or omitted for a root node. A type is considered to alias all of
2493 its descendants and all of its ancestors in the tree. Also, a type is
2494 considered to alias all types in other trees, so that bitcode produced
2495 from multiple front-ends is handled conservatively.
2497 If the third field is present, it's an integer which if equal to 1
2498 indicates that the type is "constant" (meaning
2499 ``pointsToConstantMemory`` should return true; see `other useful
2500 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2502 '``tbaa.struct``' Metadata
2503 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2505 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2506 aggregate assignment operations in C and similar languages, however it
2507 is defined to copy a contiguous region of memory, which is more than
2508 strictly necessary for aggregate types which contain holes due to
2509 padding. Also, it doesn't contain any TBAA information about the fields
2512 ``!tbaa.struct`` metadata can describe which memory subregions in a
2513 memcpy are padding and what the TBAA tags of the struct are.
2515 The current metadata format is very simple. ``!tbaa.struct`` metadata
2516 nodes are a list of operands which are in conceptual groups of three.
2517 For each group of three, the first operand gives the byte offset of a
2518 field in bytes, the second gives its size in bytes, and the third gives
2521 .. code-block:: llvm
2523 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2525 This describes a struct with two fields. The first is at offset 0 bytes
2526 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2527 and has size 4 bytes and has tbaa tag !2.
2529 Note that the fields need not be contiguous. In this example, there is a
2530 4 byte gap between the two fields. This gap represents padding which
2531 does not carry useful data and need not be preserved.
2533 '``fpmath``' Metadata
2534 ^^^^^^^^^^^^^^^^^^^^^
2536 ``fpmath`` metadata may be attached to any instruction of floating point
2537 type. It can be used to express the maximum acceptable error in the
2538 result of that instruction, in ULPs, thus potentially allowing the
2539 compiler to use a more efficient but less accurate method of computing
2540 it. ULP is defined as follows:
2542 If ``x`` is a real number that lies between two finite consecutive
2543 floating-point numbers ``a`` and ``b``, without being equal to one
2544 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2545 distance between the two non-equal finite floating-point numbers
2546 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2548 The metadata node shall consist of a single positive floating point
2549 number representing the maximum relative error, for example:
2551 .. code-block:: llvm
2553 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2555 '``range``' Metadata
2556 ^^^^^^^^^^^^^^^^^^^^
2558 ``range`` metadata may be attached only to loads of integer types. It
2559 expresses the possible ranges the loaded value is in. The ranges are
2560 represented with a flattened list of integers. The loaded value is known
2561 to be in the union of the ranges defined by each consecutive pair. Each
2562 pair has the following properties:
2564 - The type must match the type loaded by the instruction.
2565 - The pair ``a,b`` represents the range ``[a,b)``.
2566 - Both ``a`` and ``b`` are constants.
2567 - The range is allowed to wrap.
2568 - The range should not represent the full or empty set. That is,
2571 In addition, the pairs must be in signed order of the lower bound and
2572 they must be non-contiguous.
2576 .. code-block:: llvm
2578 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2579 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2580 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2581 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2583 !0 = metadata !{ i8 0, i8 2 }
2584 !1 = metadata !{ i8 255, i8 2 }
2585 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2586 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2591 It is sometimes useful to attach information to loop constructs. Currently,
2592 loop metadata is implemented as metadata attached to the branch instruction
2593 in the loop latch block. This type of metadata refer to a metadata node that is
2594 guaranteed to be separate for each loop. The loop identifier metadata is
2595 specified with the name ``llvm.loop``.
2597 The loop identifier metadata is implemented using a metadata that refers to
2598 itself to avoid merging it with any other identifier metadata, e.g.,
2599 during module linkage or function inlining. That is, each loop should refer
2600 to their own identification metadata even if they reside in separate functions.
2601 The following example contains loop identifier metadata for two separate loop
2604 .. code-block:: llvm
2606 !0 = metadata !{ metadata !0 }
2607 !1 = metadata !{ metadata !1 }
2609 The loop identifier metadata can be used to specify additional per-loop
2610 metadata. Any operands after the first operand can be treated as user-defined
2611 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2612 by the loop vectorizer to indicate how many times to unroll the loop:
2614 .. code-block:: llvm
2616 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2618 !0 = metadata !{ metadata !0, metadata !1 }
2619 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2624 Metadata types used to annotate memory accesses with information helpful
2625 for optimizations are prefixed with ``llvm.mem``.
2627 '``llvm.mem.parallel_loop_access``' Metadata
2628 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2630 For a loop to be parallel, in addition to using
2631 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2632 also all of the memory accessing instructions in the loop body need to be
2633 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2634 is at least one memory accessing instruction not marked with the metadata,
2635 the loop must be considered a sequential loop. This causes parallel loops to be
2636 converted to sequential loops due to optimization passes that are unaware of
2637 the parallel semantics and that insert new memory instructions to the loop
2640 Example of a loop that is considered parallel due to its correct use of
2641 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2642 metadata types that refer to the same loop identifier metadata.
2644 .. code-block:: llvm
2648 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2650 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2652 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2656 !0 = metadata !{ metadata !0 }
2658 It is also possible to have nested parallel loops. In that case the
2659 memory accesses refer to a list of loop identifier metadata nodes instead of
2660 the loop identifier metadata node directly:
2662 .. code-block:: llvm
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 %inner.for.end, label %inner.for.body, !llvm.loop !1
2677 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2679 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2681 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2683 outer.for.end: ; preds = %for.body
2685 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2686 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2687 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2689 '``llvm.vectorizer``'
2690 ^^^^^^^^^^^^^^^^^^^^^
2692 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2693 vectorization parameters such as vectorization factor and unroll factor.
2695 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2696 loop identification metadata.
2698 '``llvm.vectorizer.unroll``' Metadata
2699 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2701 This metadata instructs the loop vectorizer to unroll the specified
2702 loop exactly ``N`` times.
2704 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2705 operand is an integer specifying the unroll factor. For example:
2707 .. code-block:: llvm
2709 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2711 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2714 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2715 determined automatically.
2717 '``llvm.vectorizer.width``' Metadata
2718 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2720 This metadata sets the target width of the vectorizer to ``N``. Without
2721 this metadata, the vectorizer will choose a width automatically.
2722 Regardless of this metadata, the vectorizer will only vectorize loops if
2723 it believes it is valid to do so.
2725 The first operand is the string ``llvm.vectorizer.width`` and the second
2726 operand is an integer specifying the width. For example:
2728 .. code-block:: llvm
2730 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2732 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2735 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2738 Module Flags Metadata
2739 =====================
2741 Information about the module as a whole is difficult to convey to LLVM's
2742 subsystems. The LLVM IR isn't sufficient to transmit this information.
2743 The ``llvm.module.flags`` named metadata exists in order to facilitate
2744 this. These flags are in the form of key / value pairs --- much like a
2745 dictionary --- making it easy for any subsystem who cares about a flag to
2748 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2749 Each triplet has the following form:
2751 - The first element is a *behavior* flag, which specifies the behavior
2752 when two (or more) modules are merged together, and it encounters two
2753 (or more) metadata with the same ID. The supported behaviors are
2755 - The second element is a metadata string that is a unique ID for the
2756 metadata. Each module may only have one flag entry for each unique ID (not
2757 including entries with the **Require** behavior).
2758 - The third element is the value of the flag.
2760 When two (or more) modules are merged together, the resulting
2761 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2762 each unique metadata ID string, there will be exactly one entry in the merged
2763 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2764 be determined by the merge behavior flag, as described below. The only exception
2765 is that entries with the *Require* behavior are always preserved.
2767 The following behaviors are supported:
2778 Emits an error if two values disagree, otherwise the resulting value
2779 is that of the operands.
2783 Emits a warning if two values disagree. The result value will be the
2784 operand for the flag from the first module being linked.
2788 Adds a requirement that another module flag be present and have a
2789 specified value after linking is performed. The value must be a
2790 metadata pair, where the first element of the pair is the ID of the
2791 module flag to be restricted, and the second element of the pair is
2792 the value the module flag should be restricted to. This behavior can
2793 be used to restrict the allowable results (via triggering of an
2794 error) of linking IDs with the **Override** behavior.
2798 Uses the specified value, regardless of the behavior or value of the
2799 other module. If both modules specify **Override**, but the values
2800 differ, an error will be emitted.
2804 Appends the two values, which are required to be metadata nodes.
2808 Appends the two values, which are required to be metadata
2809 nodes. However, duplicate entries in the second list are dropped
2810 during the append operation.
2812 It is an error for a particular unique flag ID to have multiple behaviors,
2813 except in the case of **Require** (which adds restrictions on another metadata
2814 value) or **Override**.
2816 An example of module flags:
2818 .. code-block:: llvm
2820 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2821 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2822 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2823 !3 = metadata !{ i32 3, metadata !"qux",
2825 metadata !"foo", i32 1
2828 !llvm.module.flags = !{ !0, !1, !2, !3 }
2830 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2831 if two or more ``!"foo"`` flags are seen is to emit an error if their
2832 values are not equal.
2834 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2835 behavior if two or more ``!"bar"`` flags are seen is to use the value
2838 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2839 behavior if two or more ``!"qux"`` flags are seen is to emit a
2840 warning if their values are not equal.
2842 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2846 metadata !{ metadata !"foo", i32 1 }
2848 The behavior is to emit an error if the ``llvm.module.flags`` does not
2849 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2852 Objective-C Garbage Collection Module Flags Metadata
2853 ----------------------------------------------------
2855 On the Mach-O platform, Objective-C stores metadata about garbage
2856 collection in a special section called "image info". The metadata
2857 consists of a version number and a bitmask specifying what types of
2858 garbage collection are supported (if any) by the file. If two or more
2859 modules are linked together their garbage collection metadata needs to
2860 be merged rather than appended together.
2862 The Objective-C garbage collection module flags metadata consists of the
2863 following key-value pairs:
2872 * - ``Objective-C Version``
2873 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2875 * - ``Objective-C Image Info Version``
2876 - **[Required]** --- The version of the image info section. Currently
2879 * - ``Objective-C Image Info Section``
2880 - **[Required]** --- The section to place the metadata. Valid values are
2881 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2882 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2883 Objective-C ABI version 2.
2885 * - ``Objective-C Garbage Collection``
2886 - **[Required]** --- Specifies whether garbage collection is supported or
2887 not. Valid values are 0, for no garbage collection, and 2, for garbage
2888 collection supported.
2890 * - ``Objective-C GC Only``
2891 - **[Optional]** --- Specifies that only garbage collection is supported.
2892 If present, its value must be 6. This flag requires that the
2893 ``Objective-C Garbage Collection`` flag have the value 2.
2895 Some important flag interactions:
2897 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2898 merged with a module with ``Objective-C Garbage Collection`` set to
2899 2, then the resulting module has the
2900 ``Objective-C Garbage Collection`` flag set to 0.
2901 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2902 merged with a module with ``Objective-C GC Only`` set to 6.
2904 Automatic Linker Flags Module Flags Metadata
2905 --------------------------------------------
2907 Some targets support embedding flags to the linker inside individual object
2908 files. Typically this is used in conjunction with language extensions which
2909 allow source files to explicitly declare the libraries they depend on, and have
2910 these automatically be transmitted to the linker via object files.
2912 These flags are encoded in the IR using metadata in the module flags section,
2913 using the ``Linker Options`` key. The merge behavior for this flag is required
2914 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2915 node which should be a list of other metadata nodes, each of which should be a
2916 list of metadata strings defining linker options.
2918 For example, the following metadata section specifies two separate sets of
2919 linker options, presumably to link against ``libz`` and the ``Cocoa``
2922 !0 = metadata !{ i32 6, metadata !"Linker Options",
2924 metadata !{ metadata !"-lz" },
2925 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2926 !llvm.module.flags = !{ !0 }
2928 The metadata encoding as lists of lists of options, as opposed to a collapsed
2929 list of options, is chosen so that the IR encoding can use multiple option
2930 strings to specify e.g., a single library, while still having that specifier be
2931 preserved as an atomic element that can be recognized by a target specific
2932 assembly writer or object file emitter.
2934 Each individual option is required to be either a valid option for the target's
2935 linker, or an option that is reserved by the target specific assembly writer or
2936 object file emitter. No other aspect of these options is defined by the IR.
2938 .. _intrinsicglobalvariables:
2940 Intrinsic Global Variables
2941 ==========================
2943 LLVM has a number of "magic" global variables that contain data that
2944 affect code generation or other IR semantics. These are documented here.
2945 All globals of this sort should have a section specified as
2946 "``llvm.metadata``". This section and all globals that start with
2947 "``llvm.``" are reserved for use by LLVM.
2951 The '``llvm.used``' Global Variable
2952 -----------------------------------
2954 The ``@llvm.used`` global is an array which has
2955 :ref:`appending linkage <linkage_appending>`. This array contains a list of
2956 pointers to named global variables, functions and aliases which may optionally
2957 have a pointer cast formed of bitcast or getelementptr. For example, a legal
2960 .. code-block:: llvm
2965 @llvm.used = appending global [2 x i8*] [
2967 i8* bitcast (i32* @Y to i8*)
2968 ], section "llvm.metadata"
2970 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
2971 and linker are required to treat the symbol as if there is a reference to the
2972 symbol that it cannot see (which is why they have to be named). For example, if
2973 a variable has internal linkage and no references other than that from the
2974 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
2975 references from inline asms and other things the compiler cannot "see", and
2976 corresponds to "``attribute((used))``" in GNU C.
2978 On some targets, the code generator must emit a directive to the
2979 assembler or object file to prevent the assembler and linker from
2980 molesting the symbol.
2982 .. _gv_llvmcompilerused:
2984 The '``llvm.compiler.used``' Global Variable
2985 --------------------------------------------
2987 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
2988 directive, except that it only prevents the compiler from touching the
2989 symbol. On targets that support it, this allows an intelligent linker to
2990 optimize references to the symbol without being impeded as it would be
2993 This is a rare construct that should only be used in rare circumstances,
2994 and should not be exposed to source languages.
2996 .. _gv_llvmglobalctors:
2998 The '``llvm.global_ctors``' Global Variable
2999 -------------------------------------------
3001 .. code-block:: llvm
3003 %0 = type { i32, void ()* }
3004 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3006 The ``@llvm.global_ctors`` array contains a list of constructor
3007 functions and associated priorities. The functions referenced by this
3008 array will be called in ascending order of priority (i.e. lowest first)
3009 when the module is loaded. The order of functions with the same priority
3012 .. _llvmglobaldtors:
3014 The '``llvm.global_dtors``' Global Variable
3015 -------------------------------------------
3017 .. code-block:: llvm
3019 %0 = type { i32, void ()* }
3020 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3022 The ``@llvm.global_dtors`` array contains a list of destructor functions
3023 and associated priorities. The functions referenced by this array will
3024 be called in descending order of priority (i.e. highest first) when the
3025 module is loaded. The order of functions with the same priority is not
3028 Instruction Reference
3029 =====================
3031 The LLVM instruction set consists of several different classifications
3032 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3033 instructions <binaryops>`, :ref:`bitwise binary
3034 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3035 :ref:`other instructions <otherops>`.
3039 Terminator Instructions
3040 -----------------------
3042 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3043 program ends with a "Terminator" instruction, which indicates which
3044 block should be executed after the current block is finished. These
3045 terminator instructions typically yield a '``void``' value: they produce
3046 control flow, not values (the one exception being the
3047 ':ref:`invoke <i_invoke>`' instruction).
3049 The terminator instructions are: ':ref:`ret <i_ret>`',
3050 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3051 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3052 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3056 '``ret``' Instruction
3057 ^^^^^^^^^^^^^^^^^^^^^
3064 ret <type> <value> ; Return a value from a non-void function
3065 ret void ; Return from void function
3070 The '``ret``' instruction is used to return control flow (and optionally
3071 a value) from a function back to the caller.
3073 There are two forms of the '``ret``' instruction: one that returns a
3074 value and then causes control flow, and one that just causes control
3080 The '``ret``' instruction optionally accepts a single argument, the
3081 return value. The type of the return value must be a ':ref:`first
3082 class <t_firstclass>`' type.
3084 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3085 return type and contains a '``ret``' instruction with no return value or
3086 a return value with a type that does not match its type, or if it has a
3087 void return type and contains a '``ret``' instruction with a return
3093 When the '``ret``' instruction is executed, control flow returns back to
3094 the calling function's context. If the caller is a
3095 ":ref:`call <i_call>`" instruction, execution continues at the
3096 instruction after the call. If the caller was an
3097 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3098 beginning of the "normal" destination block. If the instruction returns
3099 a value, that value shall set the call or invoke instruction's return
3105 .. code-block:: llvm
3107 ret i32 5 ; Return an integer value of 5
3108 ret void ; Return from a void function
3109 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3113 '``br``' Instruction
3114 ^^^^^^^^^^^^^^^^^^^^
3121 br i1 <cond>, label <iftrue>, label <iffalse>
3122 br label <dest> ; Unconditional branch
3127 The '``br``' instruction is used to cause control flow to transfer to a
3128 different basic block in the current function. There are two forms of
3129 this instruction, corresponding to a conditional branch and an
3130 unconditional branch.
3135 The conditional branch form of the '``br``' instruction takes a single
3136 '``i1``' value and two '``label``' values. The unconditional form of the
3137 '``br``' instruction takes a single '``label``' value as a target.
3142 Upon execution of a conditional '``br``' instruction, the '``i1``'
3143 argument is evaluated. If the value is ``true``, control flows to the
3144 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3145 to the '``iffalse``' ``label`` argument.
3150 .. code-block:: llvm
3153 %cond = icmp eq i32 %a, %b
3154 br i1 %cond, label %IfEqual, label %IfUnequal
3162 '``switch``' Instruction
3163 ^^^^^^^^^^^^^^^^^^^^^^^^
3170 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3175 The '``switch``' instruction is used to transfer control flow to one of
3176 several different places. It is a generalization of the '``br``'
3177 instruction, allowing a branch to occur to one of many possible
3183 The '``switch``' instruction uses three parameters: an integer
3184 comparison value '``value``', a default '``label``' destination, and an
3185 array of pairs of comparison value constants and '``label``'s. The table
3186 is not allowed to contain duplicate constant entries.
3191 The ``switch`` instruction specifies a table of values and destinations.
3192 When the '``switch``' instruction is executed, this table is searched
3193 for the given value. If the value is found, control flow is transferred
3194 to the corresponding destination; otherwise, control flow is transferred
3195 to the default destination.
3200 Depending on properties of the target machine and the particular
3201 ``switch`` instruction, this instruction may be code generated in
3202 different ways. For example, it could be generated as a series of
3203 chained conditional branches or with a lookup table.
3208 .. code-block:: llvm
3210 ; Emulate a conditional br instruction
3211 %Val = zext i1 %value to i32
3212 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3214 ; Emulate an unconditional br instruction
3215 switch i32 0, label %dest [ ]
3217 ; Implement a jump table:
3218 switch i32 %val, label %otherwise [ i32 0, label %onzero
3220 i32 2, label %ontwo ]
3224 '``indirectbr``' Instruction
3225 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3232 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3237 The '``indirectbr``' instruction implements an indirect branch to a
3238 label within the current function, whose address is specified by
3239 "``address``". Address must be derived from a
3240 :ref:`blockaddress <blockaddress>` constant.
3245 The '``address``' argument is the address of the label to jump to. The
3246 rest of the arguments indicate the full set of possible destinations
3247 that the address may point to. Blocks are allowed to occur multiple
3248 times in the destination list, though this isn't particularly useful.
3250 This destination list is required so that dataflow analysis has an
3251 accurate understanding of the CFG.
3256 Control transfers to the block specified in the address argument. All
3257 possible destination blocks must be listed in the label list, otherwise
3258 this instruction has undefined behavior. This implies that jumps to
3259 labels defined in other functions have undefined behavior as well.
3264 This is typically implemented with a jump through a register.
3269 .. code-block:: llvm
3271 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3275 '``invoke``' Instruction
3276 ^^^^^^^^^^^^^^^^^^^^^^^^
3283 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3284 to label <normal label> unwind label <exception label>
3289 The '``invoke``' instruction causes control to transfer to a specified
3290 function, with the possibility of control flow transfer to either the
3291 '``normal``' label or the '``exception``' label. If the callee function
3292 returns with the "``ret``" instruction, control flow will return to the
3293 "normal" label. If the callee (or any indirect callees) returns via the
3294 ":ref:`resume <i_resume>`" instruction or other exception handling
3295 mechanism, control is interrupted and continued at the dynamically
3296 nearest "exception" label.
3298 The '``exception``' label is a `landing
3299 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3300 '``exception``' label is required to have the
3301 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3302 information about the behavior of the program after unwinding happens,
3303 as its first non-PHI instruction. The restrictions on the
3304 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3305 instruction, so that the important information contained within the
3306 "``landingpad``" instruction can't be lost through normal code motion.
3311 This instruction requires several arguments:
3313 #. The optional "cconv" marker indicates which :ref:`calling
3314 convention <callingconv>` the call should use. If none is
3315 specified, the call defaults to using C calling conventions.
3316 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3317 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3319 #. '``ptr to function ty``': shall be the signature of the pointer to
3320 function value being invoked. In most cases, this is a direct
3321 function invocation, but indirect ``invoke``'s are just as possible,
3322 branching off an arbitrary pointer to function value.
3323 #. '``function ptr val``': An LLVM value containing a pointer to a
3324 function to be invoked.
3325 #. '``function args``': argument list whose types match the function
3326 signature argument types and parameter attributes. All arguments must
3327 be of :ref:`first class <t_firstclass>` type. If the function signature
3328 indicates the function accepts a variable number of arguments, the
3329 extra arguments can be specified.
3330 #. '``normal label``': the label reached when the called function
3331 executes a '``ret``' instruction.
3332 #. '``exception label``': the label reached when a callee returns via
3333 the :ref:`resume <i_resume>` instruction or other exception handling
3335 #. The optional :ref:`function attributes <fnattrs>` list. Only
3336 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3337 attributes are valid here.
3342 This instruction is designed to operate as a standard '``call``'
3343 instruction in most regards. The primary difference is that it
3344 establishes an association with a label, which is used by the runtime
3345 library to unwind the stack.
3347 This instruction is used in languages with destructors to ensure that
3348 proper cleanup is performed in the case of either a ``longjmp`` or a
3349 thrown exception. Additionally, this is important for implementation of
3350 '``catch``' clauses in high-level languages that support them.
3352 For the purposes of the SSA form, the definition of the value returned
3353 by the '``invoke``' instruction is deemed to occur on the edge from the
3354 current block to the "normal" label. If the callee unwinds then no
3355 return value is available.
3360 .. code-block:: llvm
3362 %retval = invoke i32 @Test(i32 15) to label %Continue
3363 unwind label %TestCleanup ; {i32}:retval set
3364 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3365 unwind label %TestCleanup ; {i32}:retval set
3369 '``resume``' Instruction
3370 ^^^^^^^^^^^^^^^^^^^^^^^^
3377 resume <type> <value>
3382 The '``resume``' instruction is a terminator instruction that has no
3388 The '``resume``' instruction requires one argument, which must have the
3389 same type as the result of any '``landingpad``' instruction in the same
3395 The '``resume``' instruction resumes propagation of an existing
3396 (in-flight) exception whose unwinding was interrupted with a
3397 :ref:`landingpad <i_landingpad>` instruction.
3402 .. code-block:: llvm
3404 resume { i8*, i32 } %exn
3408 '``unreachable``' Instruction
3409 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3421 The '``unreachable``' instruction has no defined semantics. This
3422 instruction is used to inform the optimizer that a particular portion of
3423 the code is not reachable. This can be used to indicate that the code
3424 after a no-return function cannot be reached, and other facts.
3429 The '``unreachable``' instruction has no defined semantics.
3436 Binary operators are used to do most of the computation in a program.
3437 They require two operands of the same type, execute an operation on
3438 them, and produce a single value. The operands might represent multiple
3439 data, as is the case with the :ref:`vector <t_vector>` data type. The
3440 result value has the same type as its operands.
3442 There are several different binary operators:
3446 '``add``' Instruction
3447 ^^^^^^^^^^^^^^^^^^^^^
3454 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3455 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3456 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3457 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3462 The '``add``' instruction returns the sum of its two operands.
3467 The two arguments to the '``add``' instruction must be
3468 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3469 arguments must have identical types.
3474 The value produced is the integer sum of the two operands.
3476 If the sum has unsigned overflow, the result returned is the
3477 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3480 Because LLVM integers use a two's complement representation, this
3481 instruction is appropriate for both signed and unsigned integers.
3483 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3484 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3485 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3486 unsigned and/or signed overflow, respectively, occurs.
3491 .. code-block:: llvm
3493 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3497 '``fadd``' Instruction
3498 ^^^^^^^^^^^^^^^^^^^^^^
3505 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3510 The '``fadd``' instruction returns the sum of its two operands.
3515 The two arguments to the '``fadd``' instruction must be :ref:`floating
3516 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3517 Both arguments must have identical types.
3522 The value produced is the floating point sum of the two operands. This
3523 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3524 which are optimization hints to enable otherwise unsafe floating point
3530 .. code-block:: llvm
3532 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3534 '``sub``' Instruction
3535 ^^^^^^^^^^^^^^^^^^^^^
3542 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3543 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3544 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3545 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3550 The '``sub``' instruction returns the difference of its two operands.
3552 Note that the '``sub``' instruction is used to represent the '``neg``'
3553 instruction present in most other intermediate representations.
3558 The two arguments to the '``sub``' instruction must be
3559 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3560 arguments must have identical types.
3565 The value produced is the integer difference of the two operands.
3567 If the difference has unsigned overflow, the result returned is the
3568 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3571 Because LLVM integers use a two's complement representation, this
3572 instruction is appropriate for both signed and unsigned integers.
3574 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3575 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3576 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3577 unsigned and/or signed overflow, respectively, occurs.
3582 .. code-block:: llvm
3584 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3585 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3589 '``fsub``' Instruction
3590 ^^^^^^^^^^^^^^^^^^^^^^
3597 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3602 The '``fsub``' instruction returns the difference of its two operands.
3604 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3605 instruction present in most other intermediate representations.
3610 The two arguments to the '``fsub``' instruction must be :ref:`floating
3611 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3612 Both arguments must have identical types.
3617 The value produced is the floating point difference of the two operands.
3618 This instruction can also take any number of :ref:`fast-math
3619 flags <fastmath>`, which are optimization hints to enable otherwise
3620 unsafe floating point optimizations:
3625 .. code-block:: llvm
3627 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3628 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3630 '``mul``' Instruction
3631 ^^^^^^^^^^^^^^^^^^^^^
3638 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3639 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3640 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3641 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3646 The '``mul``' instruction returns the product of its two operands.
3651 The two arguments to the '``mul``' instruction must be
3652 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3653 arguments must have identical types.
3658 The value produced is the integer product of the two operands.
3660 If the result of the multiplication has unsigned overflow, the result
3661 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3662 bit width of the result.
3664 Because LLVM integers use a two's complement representation, and the
3665 result is the same width as the operands, this instruction returns the
3666 correct result for both signed and unsigned integers. If a full product
3667 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3668 sign-extended or zero-extended as appropriate to the width of the full
3671 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3672 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3673 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3674 unsigned and/or signed overflow, respectively, occurs.
3679 .. code-block:: llvm
3681 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3685 '``fmul``' Instruction
3686 ^^^^^^^^^^^^^^^^^^^^^^
3693 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3698 The '``fmul``' instruction returns the product of its two operands.
3703 The two arguments to the '``fmul``' instruction must be :ref:`floating
3704 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3705 Both arguments must have identical types.
3710 The value produced is the floating point product of the two operands.
3711 This instruction can also take any number of :ref:`fast-math
3712 flags <fastmath>`, which are optimization hints to enable otherwise
3713 unsafe floating point optimizations:
3718 .. code-block:: llvm
3720 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3722 '``udiv``' Instruction
3723 ^^^^^^^^^^^^^^^^^^^^^^
3730 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3731 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3736 The '``udiv``' instruction returns the quotient of its two operands.
3741 The two arguments to the '``udiv``' instruction must be
3742 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3743 arguments must have identical types.
3748 The value produced is the unsigned integer quotient of the two operands.
3750 Note that unsigned integer division and signed integer division are
3751 distinct operations; for signed integer division, use '``sdiv``'.
3753 Division by zero leads to undefined behavior.
3755 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3756 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3757 such, "((a udiv exact b) mul b) == a").
3762 .. code-block:: llvm
3764 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3766 '``sdiv``' Instruction
3767 ^^^^^^^^^^^^^^^^^^^^^^
3774 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3775 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3780 The '``sdiv``' instruction returns the quotient of its two operands.
3785 The two arguments to the '``sdiv``' instruction must be
3786 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3787 arguments must have identical types.
3792 The value produced is the signed integer quotient of the two operands
3793 rounded towards zero.
3795 Note that signed integer division and unsigned integer division are
3796 distinct operations; for unsigned integer division, use '``udiv``'.
3798 Division by zero leads to undefined behavior. Overflow also leads to
3799 undefined behavior; this is a rare case, but can occur, for example, by
3800 doing a 32-bit division of -2147483648 by -1.
3802 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3803 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3808 .. code-block:: llvm
3810 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3814 '``fdiv``' Instruction
3815 ^^^^^^^^^^^^^^^^^^^^^^
3822 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3827 The '``fdiv``' instruction returns the quotient of its two operands.
3832 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3833 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3834 Both arguments must have identical types.
3839 The value produced is the floating point quotient of the two operands.
3840 This instruction can also take any number of :ref:`fast-math
3841 flags <fastmath>`, which are optimization hints to enable otherwise
3842 unsafe floating point optimizations:
3847 .. code-block:: llvm
3849 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3851 '``urem``' Instruction
3852 ^^^^^^^^^^^^^^^^^^^^^^
3859 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3864 The '``urem``' instruction returns the remainder from the unsigned
3865 division of its two arguments.
3870 The two arguments to the '``urem``' instruction must be
3871 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3872 arguments must have identical types.
3877 This instruction returns the unsigned integer *remainder* of a division.
3878 This instruction always performs an unsigned division to get the
3881 Note that unsigned integer remainder and signed integer remainder are
3882 distinct operations; for signed integer remainder, use '``srem``'.
3884 Taking the remainder of a division by zero leads to undefined behavior.
3889 .. code-block:: llvm
3891 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3893 '``srem``' Instruction
3894 ^^^^^^^^^^^^^^^^^^^^^^
3901 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3906 The '``srem``' instruction returns the remainder from the signed
3907 division of its two operands. This instruction can also take
3908 :ref:`vector <t_vector>` versions of the values in which case the elements
3914 The two arguments to the '``srem``' instruction must be
3915 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3916 arguments must have identical types.
3921 This instruction returns the *remainder* of a division (where the result
3922 is either zero or has the same sign as the dividend, ``op1``), not the
3923 *modulo* operator (where the result is either zero or has the same sign
3924 as the divisor, ``op2``) of a value. For more information about the
3925 difference, see `The Math
3926 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3927 table of how this is implemented in various languages, please see
3929 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3931 Note that signed integer remainder and unsigned integer remainder are
3932 distinct operations; for unsigned integer remainder, use '``urem``'.
3934 Taking the remainder of a division by zero leads to undefined behavior.
3935 Overflow also leads to undefined behavior; this is a rare case, but can
3936 occur, for example, by taking the remainder of a 32-bit division of
3937 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3938 rule lets srem be implemented using instructions that return both the
3939 result of the division and the remainder.)
3944 .. code-block:: llvm
3946 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
3950 '``frem``' Instruction
3951 ^^^^^^^^^^^^^^^^^^^^^^
3958 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3963 The '``frem``' instruction returns the remainder from the division of
3969 The two arguments to the '``frem``' instruction must be :ref:`floating
3970 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3971 Both arguments must have identical types.
3976 This instruction returns the *remainder* of a division. The remainder
3977 has the same sign as the dividend. This instruction can also take any
3978 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
3979 to enable otherwise unsafe floating point optimizations:
3984 .. code-block:: llvm
3986 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
3990 Bitwise Binary Operations
3991 -------------------------
3993 Bitwise binary operators are used to do various forms of bit-twiddling
3994 in a program. They are generally very efficient instructions and can
3995 commonly be strength reduced from other instructions. They require two
3996 operands of the same type, execute an operation on them, and produce a
3997 single value. The resulting value is the same type as its operands.
3999 '``shl``' Instruction
4000 ^^^^^^^^^^^^^^^^^^^^^
4007 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4008 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4009 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4010 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4015 The '``shl``' instruction returns the first operand shifted to the left
4016 a specified number of bits.
4021 Both arguments to the '``shl``' instruction must be the same
4022 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4023 '``op2``' is treated as an unsigned value.
4028 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4029 where ``n`` is the width of the result. If ``op2`` is (statically or
4030 dynamically) negative or equal to or larger than the number of bits in
4031 ``op1``, the result is undefined. If the arguments are vectors, each
4032 vector element of ``op1`` is shifted by the corresponding shift amount
4035 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4036 value <poisonvalues>` if it shifts out any non-zero bits. If the
4037 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4038 value <poisonvalues>` if it shifts out any bits that disagree with the
4039 resultant sign bit. As such, NUW/NSW have the same semantics as they
4040 would if the shift were expressed as a mul instruction with the same
4041 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4046 .. code-block:: llvm
4048 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4049 <result> = shl i32 4, 2 ; yields {i32}: 16
4050 <result> = shl i32 1, 10 ; yields {i32}: 1024
4051 <result> = shl i32 1, 32 ; undefined
4052 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4054 '``lshr``' Instruction
4055 ^^^^^^^^^^^^^^^^^^^^^^
4062 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4063 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4068 The '``lshr``' instruction (logical shift right) returns the first
4069 operand shifted to the right a specified number of bits with zero fill.
4074 Both arguments to the '``lshr``' instruction must be the same
4075 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4076 '``op2``' is treated as an unsigned value.
4081 This instruction always performs a logical shift right operation. The
4082 most significant bits of the result will be filled with zero bits after
4083 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4084 than the number of bits in ``op1``, the result is undefined. If the
4085 arguments are vectors, each vector element of ``op1`` is shifted by the
4086 corresponding shift amount in ``op2``.
4088 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4089 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4095 .. code-block:: llvm
4097 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4098 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4099 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4100 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4101 <result> = lshr i32 1, 32 ; undefined
4102 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4104 '``ashr``' Instruction
4105 ^^^^^^^^^^^^^^^^^^^^^^
4112 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4113 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4118 The '``ashr``' instruction (arithmetic shift right) returns the first
4119 operand shifted to the right a specified number of bits with sign
4125 Both arguments to the '``ashr``' instruction must be the same
4126 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4127 '``op2``' is treated as an unsigned value.
4132 This instruction always performs an arithmetic shift right operation,
4133 The most significant bits of the result will be filled with the sign bit
4134 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4135 than the number of bits in ``op1``, the result is undefined. If the
4136 arguments are vectors, each vector element of ``op1`` is shifted by the
4137 corresponding shift amount in ``op2``.
4139 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4140 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4146 .. code-block:: llvm
4148 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4149 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4150 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4151 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4152 <result> = ashr i32 1, 32 ; undefined
4153 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4155 '``and``' Instruction
4156 ^^^^^^^^^^^^^^^^^^^^^
4163 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4168 The '``and``' instruction returns the bitwise logical and of its two
4174 The two arguments to the '``and``' instruction must be
4175 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4176 arguments must have identical types.
4181 The truth table used for the '``and``' instruction is:
4198 .. code-block:: llvm
4200 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4201 <result> = and i32 15, 40 ; yields {i32}:result = 8
4202 <result> = and i32 4, 8 ; yields {i32}:result = 0
4204 '``or``' Instruction
4205 ^^^^^^^^^^^^^^^^^^^^
4212 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4217 The '``or``' instruction returns the bitwise logical inclusive or of its
4223 The two arguments to the '``or``' instruction must be
4224 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4225 arguments must have identical types.
4230 The truth table used for the '``or``' instruction is:
4249 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4250 <result> = or i32 15, 40 ; yields {i32}:result = 47
4251 <result> = or i32 4, 8 ; yields {i32}:result = 12
4253 '``xor``' Instruction
4254 ^^^^^^^^^^^^^^^^^^^^^
4261 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4266 The '``xor``' instruction returns the bitwise logical exclusive or of
4267 its two operands. The ``xor`` is used to implement the "one's
4268 complement" operation, which is the "~" operator in C.
4273 The two arguments to the '``xor``' instruction must be
4274 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4275 arguments must have identical types.
4280 The truth table used for the '``xor``' instruction is:
4297 .. code-block:: llvm
4299 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4300 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4301 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4302 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4307 LLVM supports several instructions to represent vector operations in a
4308 target-independent manner. These instructions cover the element-access
4309 and vector-specific operations needed to process vectors effectively.
4310 While LLVM does directly support these vector operations, many
4311 sophisticated algorithms will want to use target-specific intrinsics to
4312 take full advantage of a specific target.
4314 .. _i_extractelement:
4316 '``extractelement``' Instruction
4317 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4324 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4329 The '``extractelement``' instruction extracts a single scalar element
4330 from a vector at a specified index.
4335 The first operand of an '``extractelement``' instruction is a value of
4336 :ref:`vector <t_vector>` type. The second operand is an index indicating
4337 the position from which to extract the element. The index may be a
4343 The result is a scalar of the same type as the element type of ``val``.
4344 Its value is the value at position ``idx`` of ``val``. If ``idx``
4345 exceeds the length of ``val``, the results are undefined.
4350 .. code-block:: llvm
4352 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4354 .. _i_insertelement:
4356 '``insertelement``' Instruction
4357 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4364 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4369 The '``insertelement``' instruction inserts a scalar element into a
4370 vector at a specified index.
4375 The first operand of an '``insertelement``' instruction is a value of
4376 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4377 type must equal the element type of the first operand. The third operand
4378 is an index indicating the position at which to insert the value. The
4379 index may be a variable.
4384 The result is a vector of the same type as ``val``. Its element values
4385 are those of ``val`` except at position ``idx``, where it gets the value
4386 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4392 .. code-block:: llvm
4394 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4396 .. _i_shufflevector:
4398 '``shufflevector``' Instruction
4399 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4406 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4411 The '``shufflevector``' instruction constructs a permutation of elements
4412 from two input vectors, returning a vector with the same element type as
4413 the input and length that is the same as the shuffle mask.
4418 The first two operands of a '``shufflevector``' instruction are vectors
4419 with the same type. The third argument is a shuffle mask whose element
4420 type is always 'i32'. The result of the instruction is a vector whose
4421 length is the same as the shuffle mask and whose element type is the
4422 same as the element type of the first two operands.
4424 The shuffle mask operand is required to be a constant vector with either
4425 constant integer or undef values.
4430 The elements of the two input vectors are numbered from left to right
4431 across both of the vectors. The shuffle mask operand specifies, for each
4432 element of the result vector, which element of the two input vectors the
4433 result element gets. The element selector may be undef (meaning "don't
4434 care") and the second operand may be undef if performing a shuffle from
4440 .. code-block:: llvm
4442 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4443 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4444 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4445 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4446 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4447 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4448 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4449 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4451 Aggregate Operations
4452 --------------------
4454 LLVM supports several instructions for working with
4455 :ref:`aggregate <t_aggregate>` values.
4459 '``extractvalue``' Instruction
4460 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4467 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4472 The '``extractvalue``' instruction extracts the value of a member field
4473 from an :ref:`aggregate <t_aggregate>` value.
4478 The first operand of an '``extractvalue``' instruction is a value of
4479 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4480 constant indices to specify which value to extract in a similar manner
4481 as indices in a '``getelementptr``' instruction.
4483 The major differences to ``getelementptr`` indexing are:
4485 - Since the value being indexed is not a pointer, the first index is
4486 omitted and assumed to be zero.
4487 - At least one index must be specified.
4488 - Not only struct indices but also array indices must be in bounds.
4493 The result is the value at the position in the aggregate specified by
4499 .. code-block:: llvm
4501 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4505 '``insertvalue``' Instruction
4506 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4513 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4518 The '``insertvalue``' instruction inserts a value into a member field in
4519 an :ref:`aggregate <t_aggregate>` value.
4524 The first operand of an '``insertvalue``' instruction is a value of
4525 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4526 a first-class value to insert. The following operands are constant
4527 indices indicating the position at which to insert the value in a
4528 similar manner as indices in a '``extractvalue``' instruction. The value
4529 to insert must have the same type as the value identified by the
4535 The result is an aggregate of the same type as ``val``. Its value is
4536 that of ``val`` except that the value at the position specified by the
4537 indices is that of ``elt``.
4542 .. code-block:: llvm
4544 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4545 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4546 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4550 Memory Access and Addressing Operations
4551 ---------------------------------------
4553 A key design point of an SSA-based representation is how it represents
4554 memory. In LLVM, no memory locations are in SSA form, which makes things
4555 very simple. This section describes how to read, write, and allocate
4560 '``alloca``' Instruction
4561 ^^^^^^^^^^^^^^^^^^^^^^^^
4568 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4573 The '``alloca``' instruction allocates memory on the stack frame of the
4574 currently executing function, to be automatically released when this
4575 function returns to its caller. The object is always allocated in the
4576 generic address space (address space zero).
4581 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4582 bytes of memory on the runtime stack, returning a pointer of the
4583 appropriate type to the program. If "NumElements" is specified, it is
4584 the number of elements allocated, otherwise "NumElements" is defaulted
4585 to be one. If a constant alignment is specified, the value result of the
4586 allocation is guaranteed to be aligned to at least that boundary. If not
4587 specified, or if zero, the target can choose to align the allocation on
4588 any convenient boundary compatible with the type.
4590 '``type``' may be any sized type.
4595 Memory is allocated; a pointer is returned. The operation is undefined
4596 if there is insufficient stack space for the allocation. '``alloca``'d
4597 memory is automatically released when the function returns. The
4598 '``alloca``' instruction is commonly used to represent automatic
4599 variables that must have an address available. When the function returns
4600 (either with the ``ret`` or ``resume`` instructions), the memory is
4601 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4602 The order in which memory is allocated (ie., which way the stack grows)
4608 .. code-block:: llvm
4610 %ptr = alloca i32 ; yields {i32*}:ptr
4611 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4612 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4613 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4617 '``load``' Instruction
4618 ^^^^^^^^^^^^^^^^^^^^^^
4625 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4626 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4627 !<index> = !{ i32 1 }
4632 The '``load``' instruction is used to read from memory.
4637 The argument to the ``load`` instruction specifies the memory address
4638 from which to load. The pointer must point to a :ref:`first
4639 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4640 then the optimizer is not allowed to modify the number or order of
4641 execution of this ``load`` with other :ref:`volatile
4642 operations <volatile>`.
4644 If the ``load`` is marked as ``atomic``, it takes an extra
4645 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4646 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4647 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4648 when they may see multiple atomic stores. The type of the pointee must
4649 be an integer type whose bit width is a power of two greater than or
4650 equal to eight and less than or equal to a target-specific size limit.
4651 ``align`` must be explicitly specified on atomic loads, and the load has
4652 undefined behavior if the alignment is not set to a value which is at
4653 least the size in bytes of the pointee. ``!nontemporal`` does not have
4654 any defined semantics for atomic loads.
4656 The optional constant ``align`` argument specifies the alignment of the
4657 operation (that is, the alignment of the memory address). A value of 0
4658 or an omitted ``align`` argument means that the operation has the ABI
4659 alignment for the target. It is the responsibility of the code emitter
4660 to ensure that the alignment information is correct. Overestimating the
4661 alignment results in undefined behavior. Underestimating the alignment
4662 may produce less efficient code. An alignment of 1 is always safe.
4664 The optional ``!nontemporal`` metadata must reference a single
4665 metadata name ``<index>`` corresponding to a metadata node with one
4666 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4667 metadata on the instruction tells the optimizer and code generator
4668 that this load is not expected to be reused in the cache. The code
4669 generator may select special instructions to save cache bandwidth, such
4670 as the ``MOVNT`` instruction on x86.
4672 The optional ``!invariant.load`` metadata must reference a single
4673 metadata name ``<index>`` corresponding to a metadata node with no
4674 entries. The existence of the ``!invariant.load`` metadata on the
4675 instruction tells the optimizer and code generator that this load
4676 address points to memory which does not change value during program
4677 execution. The optimizer may then move this load around, for example, by
4678 hoisting it out of loops using loop invariant code motion.
4683 The location of memory pointed to is loaded. If the value being loaded
4684 is of scalar type then the number of bytes read does not exceed the
4685 minimum number of bytes needed to hold all bits of the type. For
4686 example, loading an ``i24`` reads at most three bytes. When loading a
4687 value of a type like ``i20`` with a size that is not an integral number
4688 of bytes, the result is undefined if the value was not originally
4689 written using a store of the same type.
4694 .. code-block:: llvm
4696 %ptr = alloca i32 ; yields {i32*}:ptr
4697 store i32 3, i32* %ptr ; yields {void}
4698 %val = load i32* %ptr ; yields {i32}:val = i32 3
4702 '``store``' Instruction
4703 ^^^^^^^^^^^^^^^^^^^^^^^
4710 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4711 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4716 The '``store``' instruction is used to write to memory.
4721 There are two arguments to the ``store`` instruction: a value to store
4722 and an address at which to store it. The type of the ``<pointer>``
4723 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4724 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4725 then the optimizer is not allowed to modify the number or order of
4726 execution of this ``store`` with other :ref:`volatile
4727 operations <volatile>`.
4729 If the ``store`` is marked as ``atomic``, it takes an extra
4730 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4731 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4732 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4733 when they may see multiple atomic stores. The type of the pointee must
4734 be an integer type whose bit width is a power of two greater than or
4735 equal to eight and less than or equal to a target-specific size limit.
4736 ``align`` must be explicitly specified on atomic stores, and the store
4737 has undefined behavior if the alignment is not set to a value which is
4738 at least the size in bytes of the pointee. ``!nontemporal`` does not
4739 have any defined semantics for atomic stores.
4741 The optional constant ``align`` argument specifies the alignment of the
4742 operation (that is, the alignment of the memory address). A value of 0
4743 or an omitted ``align`` argument means that the operation has the ABI
4744 alignment for the target. It is the responsibility of the code emitter
4745 to ensure that the alignment information is correct. Overestimating the
4746 alignment results in undefined behavior. Underestimating the
4747 alignment may produce less efficient code. An alignment of 1 is always
4750 The optional ``!nontemporal`` metadata must reference a single metadata
4751 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4752 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4753 tells the optimizer and code generator that this load is not expected to
4754 be reused in the cache. The code generator may select special
4755 instructions to save cache bandwidth, such as the MOVNT instruction on
4761 The contents of memory are updated to contain ``<value>`` at the
4762 location specified by the ``<pointer>`` operand. If ``<value>`` is
4763 of scalar type then the number of bytes written does not exceed the
4764 minimum number of bytes needed to hold all bits of the type. For
4765 example, storing an ``i24`` writes at most three bytes. When writing a
4766 value of a type like ``i20`` with a size that is not an integral number
4767 of bytes, it is unspecified what happens to the extra bits that do not
4768 belong to the type, but they will typically be overwritten.
4773 .. code-block:: llvm
4775 %ptr = alloca i32 ; yields {i32*}:ptr
4776 store i32 3, i32* %ptr ; yields {void}
4777 %val = load i32* %ptr ; yields {i32}:val = i32 3
4781 '``fence``' Instruction
4782 ^^^^^^^^^^^^^^^^^^^^^^^
4789 fence [singlethread] <ordering> ; yields {void}
4794 The '``fence``' instruction is used to introduce happens-before edges
4800 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4801 defines what *synchronizes-with* edges they add. They can only be given
4802 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4807 A fence A which has (at least) ``release`` ordering semantics
4808 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4809 semantics if and only if there exist atomic operations X and Y, both
4810 operating on some atomic object M, such that A is sequenced before X, X
4811 modifies M (either directly or through some side effect of a sequence
4812 headed by X), Y is sequenced before B, and Y observes M. This provides a
4813 *happens-before* dependency between A and B. Rather than an explicit
4814 ``fence``, one (but not both) of the atomic operations X or Y might
4815 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4816 still *synchronize-with* the explicit ``fence`` and establish the
4817 *happens-before* edge.
4819 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4820 ``acquire`` and ``release`` semantics specified above, participates in
4821 the global program order of other ``seq_cst`` operations and/or fences.
4823 The optional ":ref:`singlethread <singlethread>`" argument specifies
4824 that the fence only synchronizes with other fences in the same thread.
4825 (This is useful for interacting with signal handlers.)
4830 .. code-block:: llvm
4832 fence acquire ; yields {void}
4833 fence singlethread seq_cst ; yields {void}
4837 '``cmpxchg``' Instruction
4838 ^^^^^^^^^^^^^^^^^^^^^^^^^
4845 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4850 The '``cmpxchg``' instruction is used to atomically modify memory. It
4851 loads a value in memory and compares it to a given value. If they are
4852 equal, it stores a new value into the memory.
4857 There are three arguments to the '``cmpxchg``' instruction: an address
4858 to operate on, a value to compare to the value currently be at that
4859 address, and a new value to place at that address if the compared values
4860 are equal. The type of '<cmp>' must be an integer type whose bit width
4861 is a power of two greater than or equal to eight and less than or equal
4862 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4863 type, and the type of '<pointer>' must be a pointer to that type. If the
4864 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4865 to modify the number or order of execution of this ``cmpxchg`` with
4866 other :ref:`volatile operations <volatile>`.
4868 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4869 synchronizes with other atomic operations.
4871 The optional "``singlethread``" argument declares that the ``cmpxchg``
4872 is only atomic with respect to code (usually signal handlers) running in
4873 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4874 respect to all other code in the system.
4876 The pointer passed into cmpxchg must have alignment greater than or
4877 equal to the size in memory of the operand.
4882 The contents of memory at the location specified by the '``<pointer>``'
4883 operand is read and compared to '``<cmp>``'; if the read value is the
4884 equal, '``<new>``' is written. The original value at the location is
4887 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4888 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4889 atomic load with an ordering parameter determined by dropping any
4890 ``release`` part of the ``cmpxchg``'s ordering.
4895 .. code-block:: llvm
4898 %orig = atomic load i32* %ptr unordered ; yields {i32}
4902 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4903 %squared = mul i32 %cmp, %cmp
4904 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4905 %success = icmp eq i32 %cmp, %old
4906 br i1 %success, label %done, label %loop
4913 '``atomicrmw``' Instruction
4914 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4921 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4926 The '``atomicrmw``' instruction is used to atomically modify memory.
4931 There are three arguments to the '``atomicrmw``' instruction: an
4932 operation to apply, an address whose value to modify, an argument to the
4933 operation. The operation must be one of the following keywords:
4947 The type of '<value>' must be an integer type whose bit width is a power
4948 of two greater than or equal to eight and less than or equal to a
4949 target-specific size limit. The type of the '``<pointer>``' operand must
4950 be a pointer to that type. If the ``atomicrmw`` is marked as
4951 ``volatile``, then the optimizer is not allowed to modify the number or
4952 order of execution of this ``atomicrmw`` with other :ref:`volatile
4953 operations <volatile>`.
4958 The contents of memory at the location specified by the '``<pointer>``'
4959 operand are atomically read, modified, and written back. The original
4960 value at the location is returned. The modification is specified by the
4963 - xchg: ``*ptr = val``
4964 - add: ``*ptr = *ptr + val``
4965 - sub: ``*ptr = *ptr - val``
4966 - and: ``*ptr = *ptr & val``
4967 - nand: ``*ptr = ~(*ptr & val)``
4968 - or: ``*ptr = *ptr | val``
4969 - xor: ``*ptr = *ptr ^ val``
4970 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
4971 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
4972 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
4974 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
4980 .. code-block:: llvm
4982 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
4984 .. _i_getelementptr:
4986 '``getelementptr``' Instruction
4987 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4994 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4995 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4996 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5001 The '``getelementptr``' instruction is used to get the address of a
5002 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5003 address calculation only and does not access memory.
5008 The first argument is always a pointer or a vector of pointers, and
5009 forms the basis of the calculation. The remaining arguments are indices
5010 that indicate which of the elements of the aggregate object are indexed.
5011 The interpretation of each index is dependent on the type being indexed
5012 into. The first index always indexes the pointer value given as the
5013 first argument, the second index indexes a value of the type pointed to
5014 (not necessarily the value directly pointed to, since the first index
5015 can be non-zero), etc. The first type indexed into must be a pointer
5016 value, subsequent types can be arrays, vectors, and structs. Note that
5017 subsequent types being indexed into can never be pointers, since that
5018 would require loading the pointer before continuing calculation.
5020 The type of each index argument depends on the type it is indexing into.
5021 When indexing into a (optionally packed) structure, only ``i32`` integer
5022 **constants** are allowed (when using a vector of indices they must all
5023 be the **same** ``i32`` integer constant). When indexing into an array,
5024 pointer or vector, integers of any width are allowed, and they are not
5025 required to be constant. These integers are treated as signed values
5028 For example, let's consider a C code fragment and how it gets compiled
5044 int *foo(struct ST *s) {
5045 return &s[1].Z.B[5][13];
5048 The LLVM code generated by Clang is:
5050 .. code-block:: llvm
5052 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5053 %struct.ST = type { i32, double, %struct.RT }
5055 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5057 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5064 In the example above, the first index is indexing into the
5065 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5066 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5067 indexes into the third element of the structure, yielding a
5068 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5069 structure. The third index indexes into the second element of the
5070 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5071 dimensions of the array are subscripted into, yielding an '``i32``'
5072 type. The '``getelementptr``' instruction returns a pointer to this
5073 element, thus computing a value of '``i32*``' type.
5075 Note that it is perfectly legal to index partially through a structure,
5076 returning a pointer to an inner element. Because of this, the LLVM code
5077 for the given testcase is equivalent to:
5079 .. code-block:: llvm
5081 define i32* @foo(%struct.ST* %s) {
5082 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5083 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5084 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5085 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5086 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5090 If the ``inbounds`` keyword is present, the result value of the
5091 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5092 pointer is not an *in bounds* address of an allocated object, or if any
5093 of the addresses that would be formed by successive addition of the
5094 offsets implied by the indices to the base address with infinitely
5095 precise signed arithmetic are not an *in bounds* address of that
5096 allocated object. The *in bounds* addresses for an allocated object are
5097 all the addresses that point into the object, plus the address one byte
5098 past the end. In cases where the base is a vector of pointers the
5099 ``inbounds`` keyword applies to each of the computations element-wise.
5101 If the ``inbounds`` keyword is not present, the offsets are added to the
5102 base address with silently-wrapping two's complement arithmetic. If the
5103 offsets have a different width from the pointer, they are sign-extended
5104 or truncated to the width of the pointer. The result value of the
5105 ``getelementptr`` may be outside the object pointed to by the base
5106 pointer. The result value may not necessarily be used to access memory
5107 though, even if it happens to point into allocated storage. See the
5108 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5111 The getelementptr instruction is often confusing. For some more insight
5112 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5117 .. code-block:: llvm
5119 ; yields [12 x i8]*:aptr
5120 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5122 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5124 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5126 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5128 In cases where the pointer argument is a vector of pointers, each index
5129 must be a vector with the same number of elements. For example:
5131 .. code-block:: llvm
5133 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5135 Conversion Operations
5136 ---------------------
5138 The instructions in this category are the conversion instructions
5139 (casting) which all take a single operand and a type. They perform
5140 various bit conversions on the operand.
5142 '``trunc .. to``' Instruction
5143 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5150 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5155 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5160 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5161 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5162 of the same number of integers. The bit size of the ``value`` must be
5163 larger than the bit size of the destination type, ``ty2``. Equal sized
5164 types are not allowed.
5169 The '``trunc``' instruction truncates the high order bits in ``value``
5170 and converts the remaining bits to ``ty2``. Since the source size must
5171 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5172 It will always truncate bits.
5177 .. code-block:: llvm
5179 %X = trunc i32 257 to i8 ; yields i8:1
5180 %Y = trunc i32 123 to i1 ; yields i1:true
5181 %Z = trunc i32 122 to i1 ; yields i1:false
5182 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5184 '``zext .. to``' Instruction
5185 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5192 <result> = zext <ty> <value> to <ty2> ; yields ty2
5197 The '``zext``' instruction zero extends its operand to type ``ty2``.
5202 The '``zext``' instruction takes a value to cast, and a type to cast it
5203 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5204 the same number of integers. The bit size of the ``value`` must be
5205 smaller than the bit size of the destination type, ``ty2``.
5210 The ``zext`` fills the high order bits of the ``value`` with zero bits
5211 until it reaches the size of the destination type, ``ty2``.
5213 When zero extending from i1, the result will always be either 0 or 1.
5218 .. code-block:: llvm
5220 %X = zext i32 257 to i64 ; yields i64:257
5221 %Y = zext i1 true to i32 ; yields i32:1
5222 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5224 '``sext .. to``' Instruction
5225 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5232 <result> = sext <ty> <value> to <ty2> ; yields ty2
5237 The '``sext``' sign extends ``value`` to the type ``ty2``.
5242 The '``sext``' instruction takes a value to cast, and a type to cast it
5243 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5244 the same number of integers. The bit size of the ``value`` must be
5245 smaller than the bit size of the destination type, ``ty2``.
5250 The '``sext``' instruction performs a sign extension by copying the sign
5251 bit (highest order bit) of the ``value`` until it reaches the bit size
5252 of the type ``ty2``.
5254 When sign extending from i1, the extension always results in -1 or 0.
5259 .. code-block:: llvm
5261 %X = sext i8 -1 to i16 ; yields i16 :65535
5262 %Y = sext i1 true to i32 ; yields i32:-1
5263 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5265 '``fptrunc .. to``' Instruction
5266 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5273 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5278 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5283 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5284 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5285 The size of ``value`` must be larger than the size of ``ty2``. This
5286 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5291 The '``fptrunc``' instruction truncates a ``value`` from a larger
5292 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5293 point <t_floating>` type. If the value cannot fit within the
5294 destination type, ``ty2``, then the results are undefined.
5299 .. code-block:: llvm
5301 %X = fptrunc double 123.0 to float ; yields float:123.0
5302 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5304 '``fpext .. to``' Instruction
5305 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5312 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5317 The '``fpext``' extends a floating point ``value`` to a larger floating
5323 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5324 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5325 to. The source type must be smaller than the destination type.
5330 The '``fpext``' instruction extends the ``value`` from a smaller
5331 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5332 point <t_floating>` type. The ``fpext`` cannot be used to make a
5333 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5334 *no-op cast* for a floating point cast.
5339 .. code-block:: llvm
5341 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5342 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5344 '``fptoui .. to``' Instruction
5345 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5352 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5357 The '``fptoui``' converts a floating point ``value`` to its unsigned
5358 integer equivalent of type ``ty2``.
5363 The '``fptoui``' instruction takes a value to cast, which must be a
5364 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5365 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5366 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5367 type with the same number of elements as ``ty``
5372 The '``fptoui``' instruction converts its :ref:`floating
5373 point <t_floating>` operand into the nearest (rounding towards zero)
5374 unsigned integer value. If the value cannot fit in ``ty2``, the results
5380 .. code-block:: llvm
5382 %X = fptoui double 123.0 to i32 ; yields i32:123
5383 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5384 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5386 '``fptosi .. to``' Instruction
5387 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5394 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5399 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5400 ``value`` to type ``ty2``.
5405 The '``fptosi``' instruction takes a value to cast, which must be a
5406 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5407 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5408 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5409 type with the same number of elements as ``ty``
5414 The '``fptosi``' instruction converts its :ref:`floating
5415 point <t_floating>` operand into the nearest (rounding towards zero)
5416 signed integer value. If the value cannot fit in ``ty2``, the results
5422 .. code-block:: llvm
5424 %X = fptosi double -123.0 to i32 ; yields i32:-123
5425 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5426 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5428 '``uitofp .. to``' Instruction
5429 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5436 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5441 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5442 and converts that value to the ``ty2`` type.
5447 The '``uitofp``' instruction takes a value to cast, which must be a
5448 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5449 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5450 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5451 type with the same number of elements as ``ty``
5456 The '``uitofp``' instruction interprets its operand as an unsigned
5457 integer quantity and converts it to the corresponding floating point
5458 value. If the value cannot fit in the floating point value, the results
5464 .. code-block:: llvm
5466 %X = uitofp i32 257 to float ; yields float:257.0
5467 %Y = uitofp i8 -1 to double ; yields double:255.0
5469 '``sitofp .. to``' Instruction
5470 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5477 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5482 The '``sitofp``' instruction regards ``value`` as a signed integer and
5483 converts that value to the ``ty2`` type.
5488 The '``sitofp``' instruction takes a value to cast, which must be a
5489 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5490 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5491 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5492 type with the same number of elements as ``ty``
5497 The '``sitofp``' instruction interprets its operand as a signed integer
5498 quantity and converts it to the corresponding floating point value. If
5499 the value cannot fit in the floating point value, the results are
5505 .. code-block:: llvm
5507 %X = sitofp i32 257 to float ; yields float:257.0
5508 %Y = sitofp i8 -1 to double ; yields double:-1.0
5512 '``ptrtoint .. to``' Instruction
5513 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5520 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5525 The '``ptrtoint``' instruction converts the pointer or a vector of
5526 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5531 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5532 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5533 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5534 a vector of integers type.
5539 The '``ptrtoint``' instruction converts ``value`` to integer type
5540 ``ty2`` by interpreting the pointer value as an integer and either
5541 truncating or zero extending that value to the size of the integer type.
5542 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5543 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5544 the same size, then nothing is done (*no-op cast*) other than a type
5550 .. code-block:: llvm
5552 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5553 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5554 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5558 '``inttoptr .. to``' Instruction
5559 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5566 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5571 The '``inttoptr``' instruction converts an integer ``value`` to a
5572 pointer type, ``ty2``.
5577 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5578 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5584 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5585 applying either a zero extension or a truncation depending on the size
5586 of the integer ``value``. If ``value`` is larger than the size of a
5587 pointer then a truncation is done. If ``value`` is smaller than the size
5588 of a pointer then a zero extension is done. If they are the same size,
5589 nothing is done (*no-op cast*).
5594 .. code-block:: llvm
5596 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5597 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5598 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5599 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5603 '``bitcast .. to``' Instruction
5604 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5611 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5616 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5622 The '``bitcast``' instruction takes a value to cast, which must be a
5623 non-aggregate first class value, and a type to cast it to, which must
5624 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5625 bit sizes of ``value`` and the destination type, ``ty2``, must be
5626 identical. If the source type is a pointer, the destination type must
5627 also be a pointer of the same size. This instruction supports bitwise
5628 conversion of vectors to integers and to vectors of other types (as
5629 long as they have the same size).
5634 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5635 is always a *no-op cast* because no bits change with this
5636 conversion. The conversion is done as if the ``value`` had been stored
5637 to memory and read back as type ``ty2``. Pointer (or vector of
5638 pointers) types may only be converted to other pointer (or vector of
5639 pointers) types with this instruction if the pointer sizes are
5640 equal. To convert pointers to other types, use the :ref:`inttoptr
5641 <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5646 .. code-block:: llvm
5648 %X = bitcast i8 255 to i8 ; yields i8 :-1
5649 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5650 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5651 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5658 The instructions in this category are the "miscellaneous" instructions,
5659 which defy better classification.
5663 '``icmp``' Instruction
5664 ^^^^^^^^^^^^^^^^^^^^^^
5671 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5676 The '``icmp``' instruction returns a boolean value or a vector of
5677 boolean values based on comparison of its two integer, integer vector,
5678 pointer, or pointer vector operands.
5683 The '``icmp``' instruction takes three operands. The first operand is
5684 the condition code indicating the kind of comparison to perform. It is
5685 not a value, just a keyword. The possible condition code are:
5688 #. ``ne``: not equal
5689 #. ``ugt``: unsigned greater than
5690 #. ``uge``: unsigned greater or equal
5691 #. ``ult``: unsigned less than
5692 #. ``ule``: unsigned less or equal
5693 #. ``sgt``: signed greater than
5694 #. ``sge``: signed greater or equal
5695 #. ``slt``: signed less than
5696 #. ``sle``: signed less or equal
5698 The remaining two arguments must be :ref:`integer <t_integer>` or
5699 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5700 must also be identical types.
5705 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5706 code given as ``cond``. The comparison performed always yields either an
5707 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5709 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5710 otherwise. No sign interpretation is necessary or performed.
5711 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5712 otherwise. No sign interpretation is necessary or performed.
5713 #. ``ugt``: interprets the operands as unsigned values and yields
5714 ``true`` if ``op1`` is greater than ``op2``.
5715 #. ``uge``: interprets the operands as unsigned values and yields
5716 ``true`` if ``op1`` is greater than or equal to ``op2``.
5717 #. ``ult``: interprets the operands as unsigned values and yields
5718 ``true`` if ``op1`` is less than ``op2``.
5719 #. ``ule``: interprets the operands as unsigned values and yields
5720 ``true`` if ``op1`` is less than or equal to ``op2``.
5721 #. ``sgt``: interprets the operands as signed values and yields ``true``
5722 if ``op1`` is greater than ``op2``.
5723 #. ``sge``: interprets the operands as signed values and yields ``true``
5724 if ``op1`` is greater than or equal to ``op2``.
5725 #. ``slt``: interprets the operands as signed values and yields ``true``
5726 if ``op1`` is less than ``op2``.
5727 #. ``sle``: interprets the operands as signed values and yields ``true``
5728 if ``op1`` is less than or equal to ``op2``.
5730 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5731 are compared as if they were integers.
5733 If the operands are integer vectors, then they are compared element by
5734 element. The result is an ``i1`` vector with the same number of elements
5735 as the values being compared. Otherwise, the result is an ``i1``.
5740 .. code-block:: llvm
5742 <result> = icmp eq i32 4, 5 ; yields: result=false
5743 <result> = icmp ne float* %X, %X ; yields: result=false
5744 <result> = icmp ult i16 4, 5 ; yields: result=true
5745 <result> = icmp sgt i16 4, 5 ; yields: result=false
5746 <result> = icmp ule i16 -4, 5 ; yields: result=false
5747 <result> = icmp sge i16 4, 5 ; yields: result=false
5749 Note that the code generator does not yet support vector types with the
5750 ``icmp`` instruction.
5754 '``fcmp``' Instruction
5755 ^^^^^^^^^^^^^^^^^^^^^^
5762 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5767 The '``fcmp``' instruction returns a boolean value or vector of boolean
5768 values based on comparison of its operands.
5770 If the operands are floating point scalars, then the result type is a
5771 boolean (:ref:`i1 <t_integer>`).
5773 If the operands are floating point vectors, then the result type is a
5774 vector of boolean with the same number of elements as the operands being
5780 The '``fcmp``' instruction takes three operands. The first operand is
5781 the condition code indicating the kind of comparison to perform. It is
5782 not a value, just a keyword. The possible condition code are:
5784 #. ``false``: no comparison, always returns false
5785 #. ``oeq``: ordered and equal
5786 #. ``ogt``: ordered and greater than
5787 #. ``oge``: ordered and greater than or equal
5788 #. ``olt``: ordered and less than
5789 #. ``ole``: ordered and less than or equal
5790 #. ``one``: ordered and not equal
5791 #. ``ord``: ordered (no nans)
5792 #. ``ueq``: unordered or equal
5793 #. ``ugt``: unordered or greater than
5794 #. ``uge``: unordered or greater than or equal
5795 #. ``ult``: unordered or less than
5796 #. ``ule``: unordered or less than or equal
5797 #. ``une``: unordered or not equal
5798 #. ``uno``: unordered (either nans)
5799 #. ``true``: no comparison, always returns true
5801 *Ordered* means that neither operand is a QNAN while *unordered* means
5802 that either operand may be a QNAN.
5804 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5805 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5806 type. They must have identical types.
5811 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5812 condition code given as ``cond``. If the operands are vectors, then the
5813 vectors are compared element by element. Each comparison performed
5814 always yields an :ref:`i1 <t_integer>` result, as follows:
5816 #. ``false``: always yields ``false``, regardless of operands.
5817 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5818 is equal to ``op2``.
5819 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5820 is greater than ``op2``.
5821 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5822 is greater than or equal to ``op2``.
5823 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5824 is less than ``op2``.
5825 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5826 is less than or equal to ``op2``.
5827 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5828 is not equal to ``op2``.
5829 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5830 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5832 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5833 greater than ``op2``.
5834 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5835 greater than or equal to ``op2``.
5836 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5838 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5839 less than or equal to ``op2``.
5840 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5841 not equal to ``op2``.
5842 #. ``uno``: yields ``true`` if either operand is a QNAN.
5843 #. ``true``: always yields ``true``, regardless of operands.
5848 .. code-block:: llvm
5850 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5851 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5852 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5853 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5855 Note that the code generator does not yet support vector types with the
5856 ``fcmp`` instruction.
5860 '``phi``' Instruction
5861 ^^^^^^^^^^^^^^^^^^^^^
5868 <result> = phi <ty> [ <val0>, <label0>], ...
5873 The '``phi``' instruction is used to implement the φ node in the SSA
5874 graph representing the function.
5879 The type of the incoming values is specified with the first type field.
5880 After this, the '``phi``' instruction takes a list of pairs as
5881 arguments, with one pair for each predecessor basic block of the current
5882 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5883 the value arguments to the PHI node. Only labels may be used as the
5886 There must be no non-phi instructions between the start of a basic block
5887 and the PHI instructions: i.e. PHI instructions must be first in a basic
5890 For the purposes of the SSA form, the use of each incoming value is
5891 deemed to occur on the edge from the corresponding predecessor block to
5892 the current block (but after any definition of an '``invoke``'
5893 instruction's return value on the same edge).
5898 At runtime, the '``phi``' instruction logically takes on the value
5899 specified by the pair corresponding to the predecessor basic block that
5900 executed just prior to the current block.
5905 .. code-block:: llvm
5907 Loop: ; Infinite loop that counts from 0 on up...
5908 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5909 %nextindvar = add i32 %indvar, 1
5914 '``select``' Instruction
5915 ^^^^^^^^^^^^^^^^^^^^^^^^
5922 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5924 selty is either i1 or {<N x i1>}
5929 The '``select``' instruction is used to choose one value based on a
5930 condition, without branching.
5935 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
5936 values indicating the condition, and two values of the same :ref:`first
5937 class <t_firstclass>` type. If the val1/val2 are vectors and the
5938 condition is a scalar, then entire vectors are selected, not individual
5944 If the condition is an i1 and it evaluates to 1, the instruction returns
5945 the first value argument; otherwise, it returns the second value
5948 If the condition is a vector of i1, then the value arguments must be
5949 vectors of the same size, and the selection is done element by element.
5954 .. code-block:: llvm
5956 %X = select i1 true, i8 17, i8 42 ; yields i8:17
5960 '``call``' Instruction
5961 ^^^^^^^^^^^^^^^^^^^^^^
5968 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
5973 The '``call``' instruction represents a simple function call.
5978 This instruction requires several arguments:
5980 #. The optional "tail" marker indicates that the callee function does
5981 not access any allocas or varargs in the caller. Note that calls may
5982 be marked "tail" even if they do not occur before a
5983 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
5984 function call is eligible for tail call optimization, but `might not
5985 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
5986 The code generator may optimize calls marked "tail" with either 1)
5987 automatic `sibling call
5988 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
5989 callee have matching signatures, or 2) forced tail call optimization
5990 when the following extra requirements are met:
5992 - Caller and callee both have the calling convention ``fastcc``.
5993 - The call is in tail position (ret immediately follows call and ret
5994 uses value of call or is void).
5995 - Option ``-tailcallopt`` is enabled, or
5996 ``llvm::GuaranteedTailCallOpt`` is ``true``.
5997 - `Platform specific constraints are
5998 met. <CodeGenerator.html#tailcallopt>`_
6000 #. The optional "cconv" marker indicates which :ref:`calling
6001 convention <callingconv>` the call should use. If none is
6002 specified, the call defaults to using C calling conventions. The
6003 calling convention of the call must match the calling convention of
6004 the target function, or else the behavior is undefined.
6005 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6006 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6008 #. '``ty``': the type of the call instruction itself which is also the
6009 type of the return value. Functions that return no value are marked
6011 #. '``fnty``': shall be the signature of the pointer to function value
6012 being invoked. The argument types must match the types implied by
6013 this signature. This type can be omitted if the function is not
6014 varargs and if the function type does not return a pointer to a
6016 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6017 be invoked. In most cases, this is a direct function invocation, but
6018 indirect ``call``'s are just as possible, calling an arbitrary pointer
6020 #. '``function args``': argument list whose types match the function
6021 signature argument types and parameter attributes. All arguments must
6022 be of :ref:`first class <t_firstclass>` type. If the function signature
6023 indicates the function accepts a variable number of arguments, the
6024 extra arguments can be specified.
6025 #. The optional :ref:`function attributes <fnattrs>` list. Only
6026 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6027 attributes are valid here.
6032 The '``call``' instruction is used to cause control flow to transfer to
6033 a specified function, with its incoming arguments bound to the specified
6034 values. Upon a '``ret``' instruction in the called function, control
6035 flow continues with the instruction after the function call, and the
6036 return value of the function is bound to the result argument.
6041 .. code-block:: llvm
6043 %retval = call i32 @test(i32 %argc)
6044 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6045 %X = tail call i32 @foo() ; yields i32
6046 %Y = tail call fastcc i32 @foo() ; yields i32
6047 call void %foo(i8 97 signext)
6049 %struct.A = type { i32, i8 }
6050 %r = call %struct.A @foo() ; yields { 32, i8 }
6051 %gr = extractvalue %struct.A %r, 0 ; yields i32
6052 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6053 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6054 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6056 llvm treats calls to some functions with names and arguments that match
6057 the standard C99 library as being the C99 library functions, and may
6058 perform optimizations or generate code for them under that assumption.
6059 This is something we'd like to change in the future to provide better
6060 support for freestanding environments and non-C-based languages.
6064 '``va_arg``' Instruction
6065 ^^^^^^^^^^^^^^^^^^^^^^^^
6072 <resultval> = va_arg <va_list*> <arglist>, <argty>
6077 The '``va_arg``' instruction is used to access arguments passed through
6078 the "variable argument" area of a function call. It is used to implement
6079 the ``va_arg`` macro in C.
6084 This instruction takes a ``va_list*`` value and the type of the
6085 argument. It returns a value of the specified argument type and
6086 increments the ``va_list`` to point to the next argument. The actual
6087 type of ``va_list`` is target specific.
6092 The '``va_arg``' instruction loads an argument of the specified type
6093 from the specified ``va_list`` and causes the ``va_list`` to point to
6094 the next argument. For more information, see the variable argument
6095 handling :ref:`Intrinsic Functions <int_varargs>`.
6097 It is legal for this instruction to be called in a function which does
6098 not take a variable number of arguments, for example, the ``vfprintf``
6101 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6102 function <intrinsics>` because it takes a type as an argument.
6107 See the :ref:`variable argument processing <int_varargs>` section.
6109 Note that the code generator does not yet fully support va\_arg on many
6110 targets. Also, it does not currently support va\_arg with aggregate
6111 types on any target.
6115 '``landingpad``' Instruction
6116 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6123 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6124 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6126 <clause> := catch <type> <value>
6127 <clause> := filter <array constant type> <array constant>
6132 The '``landingpad``' instruction is used by `LLVM's exception handling
6133 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6134 is a landing pad --- one where the exception lands, and corresponds to the
6135 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6136 defines values supplied by the personality function (``pers_fn``) upon
6137 re-entry to the function. The ``resultval`` has the type ``resultty``.
6142 This instruction takes a ``pers_fn`` value. This is the personality
6143 function associated with the unwinding mechanism. The optional
6144 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6146 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6147 contains the global variable representing the "type" that may be caught
6148 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6149 clause takes an array constant as its argument. Use
6150 "``[0 x i8**] undef``" for a filter which cannot throw. The
6151 '``landingpad``' instruction must contain *at least* one ``clause`` or
6152 the ``cleanup`` flag.
6157 The '``landingpad``' instruction defines the values which are set by the
6158 personality function (``pers_fn``) upon re-entry to the function, and
6159 therefore the "result type" of the ``landingpad`` instruction. As with
6160 calling conventions, how the personality function results are
6161 represented in LLVM IR is target specific.
6163 The clauses are applied in order from top to bottom. If two
6164 ``landingpad`` instructions are merged together through inlining, the
6165 clauses from the calling function are appended to the list of clauses.
6166 When the call stack is being unwound due to an exception being thrown,
6167 the exception is compared against each ``clause`` in turn. If it doesn't
6168 match any of the clauses, and the ``cleanup`` flag is not set, then
6169 unwinding continues further up the call stack.
6171 The ``landingpad`` instruction has several restrictions:
6173 - A landing pad block is a basic block which is the unwind destination
6174 of an '``invoke``' instruction.
6175 - A landing pad block must have a '``landingpad``' instruction as its
6176 first non-PHI instruction.
6177 - There can be only one '``landingpad``' instruction within the landing
6179 - A basic block that is not a landing pad block may not include a
6180 '``landingpad``' instruction.
6181 - All '``landingpad``' instructions in a function must have the same
6182 personality function.
6187 .. code-block:: llvm
6189 ;; A landing pad which can catch an integer.
6190 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6192 ;; A landing pad that is a cleanup.
6193 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6195 ;; A landing pad which can catch an integer and can only throw a double.
6196 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6198 filter [1 x i8**] [@_ZTId]
6205 LLVM supports the notion of an "intrinsic function". These functions
6206 have well known names and semantics and are required to follow certain
6207 restrictions. Overall, these intrinsics represent an extension mechanism
6208 for the LLVM language that does not require changing all of the
6209 transformations in LLVM when adding to the language (or the bitcode
6210 reader/writer, the parser, etc...).
6212 Intrinsic function names must all start with an "``llvm.``" prefix. This
6213 prefix is reserved in LLVM for intrinsic names; thus, function names may
6214 not begin with this prefix. Intrinsic functions must always be external
6215 functions: you cannot define the body of intrinsic functions. Intrinsic
6216 functions may only be used in call or invoke instructions: it is illegal
6217 to take the address of an intrinsic function. Additionally, because
6218 intrinsic functions are part of the LLVM language, it is required if any
6219 are added that they be documented here.
6221 Some intrinsic functions can be overloaded, i.e., the intrinsic
6222 represents a family of functions that perform the same operation but on
6223 different data types. Because LLVM can represent over 8 million
6224 different integer types, overloading is used commonly to allow an
6225 intrinsic function to operate on any integer type. One or more of the
6226 argument types or the result type can be overloaded to accept any
6227 integer type. Argument types may also be defined as exactly matching a
6228 previous argument's type or the result type. This allows an intrinsic
6229 function which accepts multiple arguments, but needs all of them to be
6230 of the same type, to only be overloaded with respect to a single
6231 argument or the result.
6233 Overloaded intrinsics will have the names of its overloaded argument
6234 types encoded into its function name, each preceded by a period. Only
6235 those types which are overloaded result in a name suffix. Arguments
6236 whose type is matched against another type do not. For example, the
6237 ``llvm.ctpop`` function can take an integer of any width and returns an
6238 integer of exactly the same integer width. This leads to a family of
6239 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6240 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6241 overloaded, and only one type suffix is required. Because the argument's
6242 type is matched against the return type, it does not require its own
6245 To learn how to add an intrinsic function, please see the `Extending
6246 LLVM Guide <ExtendingLLVM.html>`_.
6250 Variable Argument Handling Intrinsics
6251 -------------------------------------
6253 Variable argument support is defined in LLVM with the
6254 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6255 functions. These functions are related to the similarly named macros
6256 defined in the ``<stdarg.h>`` header file.
6258 All of these functions operate on arguments that use a target-specific
6259 value type "``va_list``". The LLVM assembly language reference manual
6260 does not define what this type is, so all transformations should be
6261 prepared to handle these functions regardless of the type used.
6263 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6264 variable argument handling intrinsic functions are used.
6266 .. code-block:: llvm
6268 define i32 @test(i32 %X, ...) {
6269 ; Initialize variable argument processing
6271 %ap2 = bitcast i8** %ap to i8*
6272 call void @llvm.va_start(i8* %ap2)
6274 ; Read a single integer argument
6275 %tmp = va_arg i8** %ap, i32
6277 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6279 %aq2 = bitcast i8** %aq to i8*
6280 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6281 call void @llvm.va_end(i8* %aq2)
6283 ; Stop processing of arguments.
6284 call void @llvm.va_end(i8* %ap2)
6288 declare void @llvm.va_start(i8*)
6289 declare void @llvm.va_copy(i8*, i8*)
6290 declare void @llvm.va_end(i8*)
6294 '``llvm.va_start``' Intrinsic
6295 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6302 declare void %llvm.va_start(i8* <arglist>)
6307 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6308 subsequent use by ``va_arg``.
6313 The argument is a pointer to a ``va_list`` element to initialize.
6318 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6319 available in C. In a target-dependent way, it initializes the
6320 ``va_list`` element to which the argument points, so that the next call
6321 to ``va_arg`` will produce the first variable argument passed to the
6322 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6323 to know the last argument of the function as the compiler can figure
6326 '``llvm.va_end``' Intrinsic
6327 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6334 declare void @llvm.va_end(i8* <arglist>)
6339 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6340 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6345 The argument is a pointer to a ``va_list`` to destroy.
6350 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6351 available in C. In a target-dependent way, it destroys the ``va_list``
6352 element to which the argument points. Calls to
6353 :ref:`llvm.va_start <int_va_start>` and
6354 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6359 '``llvm.va_copy``' Intrinsic
6360 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6367 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6372 The '``llvm.va_copy``' intrinsic copies the current argument position
6373 from the source argument list to the destination argument list.
6378 The first argument is a pointer to a ``va_list`` element to initialize.
6379 The second argument is a pointer to a ``va_list`` element to copy from.
6384 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6385 available in C. In a target-dependent way, it copies the source
6386 ``va_list`` element into the destination ``va_list`` element. This
6387 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6388 arbitrarily complex and require, for example, memory allocation.
6390 Accurate Garbage Collection Intrinsics
6391 --------------------------------------
6393 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6394 (GC) requires the implementation and generation of these intrinsics.
6395 These intrinsics allow identification of :ref:`GC roots on the
6396 stack <int_gcroot>`, as well as garbage collector implementations that
6397 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6398 Front-ends for type-safe garbage collected languages should generate
6399 these intrinsics to make use of the LLVM garbage collectors. For more
6400 details, see `Accurate Garbage Collection with
6401 LLVM <GarbageCollection.html>`_.
6403 The garbage collection intrinsics only operate on objects in the generic
6404 address space (address space zero).
6408 '``llvm.gcroot``' Intrinsic
6409 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6416 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6421 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6422 the code generator, and allows some metadata to be associated with it.
6427 The first argument specifies the address of a stack object that contains
6428 the root pointer. The second pointer (which must be either a constant or
6429 a global value address) contains the meta-data to be associated with the
6435 At runtime, a call to this intrinsic stores a null pointer into the
6436 "ptrloc" location. At compile-time, the code generator generates
6437 information to allow the runtime to find the pointer at GC safe points.
6438 The '``llvm.gcroot``' intrinsic may only be used in a function which
6439 :ref:`specifies a GC algorithm <gc>`.
6443 '``llvm.gcread``' Intrinsic
6444 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6451 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6456 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6457 locations, allowing garbage collector implementations that require read
6463 The second argument is the address to read from, which should be an
6464 address allocated from the garbage collector. The first object is a
6465 pointer to the start of the referenced object, if needed by the language
6466 runtime (otherwise null).
6471 The '``llvm.gcread``' intrinsic has the same semantics as a load
6472 instruction, but may be replaced with substantially more complex code by
6473 the garbage collector runtime, as needed. The '``llvm.gcread``'
6474 intrinsic may only be used in a function which :ref:`specifies a GC
6479 '``llvm.gcwrite``' Intrinsic
6480 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6487 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6492 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6493 locations, allowing garbage collector implementations that require write
6494 barriers (such as generational or reference counting collectors).
6499 The first argument is the reference to store, the second is the start of
6500 the object to store it to, and the third is the address of the field of
6501 Obj to store to. If the runtime does not require a pointer to the
6502 object, Obj may be null.
6507 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6508 instruction, but may be replaced with substantially more complex code by
6509 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6510 intrinsic may only be used in a function which :ref:`specifies a GC
6513 Code Generator Intrinsics
6514 -------------------------
6516 These intrinsics are provided by LLVM to expose special features that
6517 may only be implemented with code generator support.
6519 '``llvm.returnaddress``' Intrinsic
6520 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6527 declare i8 *@llvm.returnaddress(i32 <level>)
6532 The '``llvm.returnaddress``' intrinsic attempts to compute a
6533 target-specific value indicating the return address of the current
6534 function or one of its callers.
6539 The argument to this intrinsic indicates which function to return the
6540 address for. Zero indicates the calling function, one indicates its
6541 caller, etc. The argument is **required** to be a constant integer
6547 The '``llvm.returnaddress``' intrinsic either returns a pointer
6548 indicating the return address of the specified call frame, or zero if it
6549 cannot be identified. The value returned by this intrinsic is likely to
6550 be incorrect or 0 for arguments other than zero, so it should only be
6551 used for debugging purposes.
6553 Note that calling this intrinsic does not prevent function inlining or
6554 other aggressive transformations, so the value returned may not be that
6555 of the obvious source-language caller.
6557 '``llvm.frameaddress``' Intrinsic
6558 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6565 declare i8* @llvm.frameaddress(i32 <level>)
6570 The '``llvm.frameaddress``' intrinsic attempts to return the
6571 target-specific frame pointer value for the specified stack frame.
6576 The argument to this intrinsic indicates which function to return the
6577 frame pointer for. Zero indicates the calling function, one indicates
6578 its caller, etc. The argument is **required** to be a constant integer
6584 The '``llvm.frameaddress``' intrinsic either returns a pointer
6585 indicating the frame address of the specified call frame, or zero if it
6586 cannot be identified. The value returned by this intrinsic is likely to
6587 be incorrect or 0 for arguments other than zero, so it should only be
6588 used for debugging purposes.
6590 Note that calling this intrinsic does not prevent function inlining or
6591 other aggressive transformations, so the value returned may not be that
6592 of the obvious source-language caller.
6596 '``llvm.stacksave``' Intrinsic
6597 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6604 declare i8* @llvm.stacksave()
6609 The '``llvm.stacksave``' intrinsic is used to remember the current state
6610 of the function stack, for use with
6611 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6612 implementing language features like scoped automatic variable sized
6618 This intrinsic returns a opaque pointer value that can be passed to
6619 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6620 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6621 ``llvm.stacksave``, it effectively restores the state of the stack to
6622 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6623 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6624 were allocated after the ``llvm.stacksave`` was executed.
6626 .. _int_stackrestore:
6628 '``llvm.stackrestore``' Intrinsic
6629 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6636 declare void @llvm.stackrestore(i8* %ptr)
6641 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6642 the function stack to the state it was in when the corresponding
6643 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6644 useful for implementing language features like scoped automatic variable
6645 sized arrays in C99.
6650 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6652 '``llvm.prefetch``' Intrinsic
6653 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6660 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6665 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6666 insert a prefetch instruction if supported; otherwise, it is a noop.
6667 Prefetches have no effect on the behavior of the program but can change
6668 its performance characteristics.
6673 ``address`` is the address to be prefetched, ``rw`` is the specifier
6674 determining if the fetch should be for a read (0) or write (1), and
6675 ``locality`` is a temporal locality specifier ranging from (0) - no
6676 locality, to (3) - extremely local keep in cache. The ``cache type``
6677 specifies whether the prefetch is performed on the data (1) or
6678 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6679 arguments must be constant integers.
6684 This intrinsic does not modify the behavior of the program. In
6685 particular, prefetches cannot trap and do not produce a value. On
6686 targets that support this intrinsic, the prefetch can provide hints to
6687 the processor cache for better performance.
6689 '``llvm.pcmarker``' Intrinsic
6690 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6697 declare void @llvm.pcmarker(i32 <id>)
6702 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6703 Counter (PC) in a region of code to simulators and other tools. The
6704 method is target specific, but it is expected that the marker will use
6705 exported symbols to transmit the PC of the marker. The marker makes no
6706 guarantees that it will remain with any specific instruction after
6707 optimizations. It is possible that the presence of a marker will inhibit
6708 optimizations. The intended use is to be inserted after optimizations to
6709 allow correlations of simulation runs.
6714 ``id`` is a numerical id identifying the marker.
6719 This intrinsic does not modify the behavior of the program. Backends
6720 that do not support this intrinsic may ignore it.
6722 '``llvm.readcyclecounter``' Intrinsic
6723 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6730 declare i64 @llvm.readcyclecounter()
6735 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6736 counter register (or similar low latency, high accuracy clocks) on those
6737 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6738 should map to RPCC. As the backing counters overflow quickly (on the
6739 order of 9 seconds on alpha), this should only be used for small
6745 When directly supported, reading the cycle counter should not modify any
6746 memory. Implementations are allowed to either return a application
6747 specific value or a system wide value. On backends without support, this
6748 is lowered to a constant 0.
6750 Note that runtime support may be conditional on the privilege-level code is
6751 running at and the host platform.
6753 Standard C Library Intrinsics
6754 -----------------------------
6756 LLVM provides intrinsics for a few important standard C library
6757 functions. These intrinsics allow source-language front-ends to pass
6758 information about the alignment of the pointer arguments to the code
6759 generator, providing opportunity for more efficient code generation.
6763 '``llvm.memcpy``' Intrinsic
6764 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6769 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6770 integer bit width and for different address spaces. Not all targets
6771 support all bit widths however.
6775 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6776 i32 <len>, i32 <align>, i1 <isvolatile>)
6777 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6778 i64 <len>, i32 <align>, i1 <isvolatile>)
6783 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6784 source location to the destination location.
6786 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6787 intrinsics do not return a value, takes extra alignment/isvolatile
6788 arguments and the pointers can be in specified address spaces.
6793 The first argument is a pointer to the destination, the second is a
6794 pointer to the source. The third argument is an integer argument
6795 specifying the number of bytes to copy, the fourth argument is the
6796 alignment of the source and destination locations, and the fifth is a
6797 boolean indicating a volatile access.
6799 If the call to this intrinsic has an alignment value that is not 0 or 1,
6800 then the caller guarantees that both the source and destination pointers
6801 are aligned to that boundary.
6803 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6804 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6805 very cleanly specified and it is unwise to depend on it.
6810 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6811 source location to the destination location, which are not allowed to
6812 overlap. It copies "len" bytes of memory over. If the argument is known
6813 to be aligned to some boundary, this can be specified as the fourth
6814 argument, otherwise it should be set to 0 or 1.
6816 '``llvm.memmove``' Intrinsic
6817 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6822 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6823 bit width and for different address space. Not all targets support all
6828 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6829 i32 <len>, i32 <align>, i1 <isvolatile>)
6830 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6831 i64 <len>, i32 <align>, i1 <isvolatile>)
6836 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6837 source location to the destination location. It is similar to the
6838 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6841 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6842 intrinsics do not return a value, takes extra alignment/isvolatile
6843 arguments and the pointers can be in specified address spaces.
6848 The first argument is a pointer to the destination, the second is a
6849 pointer to the source. The third argument is an integer argument
6850 specifying the number of bytes to copy, the fourth argument is the
6851 alignment of the source and destination locations, and the fifth is a
6852 boolean indicating a volatile access.
6854 If the call to this intrinsic has an alignment value that is not 0 or 1,
6855 then the caller guarantees that the source and destination pointers are
6856 aligned to that boundary.
6858 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6859 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6860 not very cleanly specified and it is unwise to depend on it.
6865 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6866 source location to the destination location, which may overlap. It
6867 copies "len" bytes of memory over. If the argument is known to be
6868 aligned to some boundary, this can be specified as the fourth argument,
6869 otherwise it should be set to 0 or 1.
6871 '``llvm.memset.*``' Intrinsics
6872 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6877 This is an overloaded intrinsic. You can use llvm.memset on any integer
6878 bit width and for different address spaces. However, not all targets
6879 support all bit widths.
6883 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6884 i32 <len>, i32 <align>, i1 <isvolatile>)
6885 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6886 i64 <len>, i32 <align>, i1 <isvolatile>)
6891 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6892 particular byte value.
6894 Note that, unlike the standard libc function, the ``llvm.memset``
6895 intrinsic does not return a value and takes extra alignment/volatile
6896 arguments. Also, the destination can be in an arbitrary address space.
6901 The first argument is a pointer to the destination to fill, the second
6902 is the byte value with which to fill it, the third argument is an
6903 integer argument specifying the number of bytes to fill, and the fourth
6904 argument is the known alignment of the destination location.
6906 If the call to this intrinsic has an alignment value that is not 0 or 1,
6907 then the caller guarantees that the destination pointer is aligned to
6910 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6911 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6912 very cleanly specified and it is unwise to depend on it.
6917 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6918 at the destination location. If the argument is known to be aligned to
6919 some boundary, this can be specified as the fourth argument, otherwise
6920 it should be set to 0 or 1.
6922 '``llvm.sqrt.*``' Intrinsic
6923 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6928 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6929 floating point or vector of floating point type. Not all targets support
6934 declare float @llvm.sqrt.f32(float %Val)
6935 declare double @llvm.sqrt.f64(double %Val)
6936 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6937 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6938 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6943 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
6944 returning the same value as the libm '``sqrt``' functions would. Unlike
6945 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
6946 negative numbers other than -0.0 (which allows for better optimization,
6947 because there is no need to worry about errno being set).
6948 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
6953 The argument and return value are floating point numbers of the same
6959 This function returns the sqrt of the specified operand if it is a
6960 nonnegative floating point number.
6962 '``llvm.powi.*``' Intrinsic
6963 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6968 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
6969 floating point or vector of floating point type. Not all targets support
6974 declare float @llvm.powi.f32(float %Val, i32 %power)
6975 declare double @llvm.powi.f64(double %Val, i32 %power)
6976 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6977 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6978 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6983 The '``llvm.powi.*``' intrinsics return the first operand raised to the
6984 specified (positive or negative) power. The order of evaluation of
6985 multiplications is not defined. When a vector of floating point type is
6986 used, the second argument remains a scalar integer value.
6991 The second argument is an integer power, and the first is a value to
6992 raise to that power.
6997 This function returns the first value raised to the second power with an
6998 unspecified sequence of rounding operations.
7000 '``llvm.sin.*``' Intrinsic
7001 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7006 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7007 floating point or vector of floating point type. Not all targets support
7012 declare float @llvm.sin.f32(float %Val)
7013 declare double @llvm.sin.f64(double %Val)
7014 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7015 declare fp128 @llvm.sin.f128(fp128 %Val)
7016 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7021 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7026 The argument and return value are floating point numbers of the same
7032 This function returns the sine of the specified operand, returning the
7033 same values as the libm ``sin`` functions would, and handles error
7034 conditions in the same way.
7036 '``llvm.cos.*``' Intrinsic
7037 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7042 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7043 floating point or vector of floating point type. Not all targets support
7048 declare float @llvm.cos.f32(float %Val)
7049 declare double @llvm.cos.f64(double %Val)
7050 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7051 declare fp128 @llvm.cos.f128(fp128 %Val)
7052 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7057 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7062 The argument and return value are floating point numbers of the same
7068 This function returns the cosine of the specified operand, returning the
7069 same values as the libm ``cos`` functions would, and handles error
7070 conditions in the same way.
7072 '``llvm.pow.*``' Intrinsic
7073 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7078 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7079 floating point or vector of floating point type. Not all targets support
7084 declare float @llvm.pow.f32(float %Val, float %Power)
7085 declare double @llvm.pow.f64(double %Val, double %Power)
7086 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7087 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7088 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7093 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7094 specified (positive or negative) power.
7099 The second argument is a floating point power, and the first is a value
7100 to raise to that power.
7105 This function returns the first value raised to the second power,
7106 returning the same values as the libm ``pow`` functions would, and
7107 handles error conditions in the same way.
7109 '``llvm.exp.*``' Intrinsic
7110 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7115 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7116 floating point or vector of floating point type. Not all targets support
7121 declare float @llvm.exp.f32(float %Val)
7122 declare double @llvm.exp.f64(double %Val)
7123 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7124 declare fp128 @llvm.exp.f128(fp128 %Val)
7125 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7130 The '``llvm.exp.*``' intrinsics perform the exp function.
7135 The argument and return value are floating point numbers of the same
7141 This function returns the same values as the libm ``exp`` functions
7142 would, and handles error conditions in the same way.
7144 '``llvm.exp2.*``' Intrinsic
7145 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7150 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7151 floating point or vector of floating point type. Not all targets support
7156 declare float @llvm.exp2.f32(float %Val)
7157 declare double @llvm.exp2.f64(double %Val)
7158 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7159 declare fp128 @llvm.exp2.f128(fp128 %Val)
7160 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7165 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7170 The argument and return value are floating point numbers of the same
7176 This function returns the same values as the libm ``exp2`` functions
7177 would, and handles error conditions in the same way.
7179 '``llvm.log.*``' Intrinsic
7180 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7185 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7186 floating point or vector of floating point type. Not all targets support
7191 declare float @llvm.log.f32(float %Val)
7192 declare double @llvm.log.f64(double %Val)
7193 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7194 declare fp128 @llvm.log.f128(fp128 %Val)
7195 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7200 The '``llvm.log.*``' intrinsics perform the log function.
7205 The argument and return value are floating point numbers of the same
7211 This function returns the same values as the libm ``log`` functions
7212 would, and handles error conditions in the same way.
7214 '``llvm.log10.*``' Intrinsic
7215 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7220 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7221 floating point or vector of floating point type. Not all targets support
7226 declare float @llvm.log10.f32(float %Val)
7227 declare double @llvm.log10.f64(double %Val)
7228 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7229 declare fp128 @llvm.log10.f128(fp128 %Val)
7230 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7235 The '``llvm.log10.*``' intrinsics perform the log10 function.
7240 The argument and return value are floating point numbers of the same
7246 This function returns the same values as the libm ``log10`` functions
7247 would, and handles error conditions in the same way.
7249 '``llvm.log2.*``' Intrinsic
7250 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7255 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7256 floating point or vector of floating point type. Not all targets support
7261 declare float @llvm.log2.f32(float %Val)
7262 declare double @llvm.log2.f64(double %Val)
7263 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7264 declare fp128 @llvm.log2.f128(fp128 %Val)
7265 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7270 The '``llvm.log2.*``' intrinsics perform the log2 function.
7275 The argument and return value are floating point numbers of the same
7281 This function returns the same values as the libm ``log2`` functions
7282 would, and handles error conditions in the same way.
7284 '``llvm.fma.*``' Intrinsic
7285 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7290 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7291 floating point or vector of floating point type. Not all targets support
7296 declare float @llvm.fma.f32(float %a, float %b, float %c)
7297 declare double @llvm.fma.f64(double %a, double %b, double %c)
7298 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7299 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7300 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7305 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7311 The argument and return value are floating point numbers of the same
7317 This function returns the same values as the libm ``fma`` functions
7320 '``llvm.fabs.*``' Intrinsic
7321 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7326 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7327 floating point or vector of floating point type. Not all targets support
7332 declare float @llvm.fabs.f32(float %Val)
7333 declare double @llvm.fabs.f64(double %Val)
7334 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7335 declare fp128 @llvm.fabs.f128(fp128 %Val)
7336 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7341 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7347 The argument and return value are floating point numbers of the same
7353 This function returns the same values as the libm ``fabs`` functions
7354 would, and handles error conditions in the same way.
7356 '``llvm.floor.*``' Intrinsic
7357 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7362 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7363 floating point or vector of floating point type. Not all targets support
7368 declare float @llvm.floor.f32(float %Val)
7369 declare double @llvm.floor.f64(double %Val)
7370 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7371 declare fp128 @llvm.floor.f128(fp128 %Val)
7372 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7377 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7382 The argument and return value are floating point numbers of the same
7388 This function returns the same values as the libm ``floor`` functions
7389 would, and handles error conditions in the same way.
7391 '``llvm.ceil.*``' Intrinsic
7392 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7397 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7398 floating point or vector of floating point type. Not all targets support
7403 declare float @llvm.ceil.f32(float %Val)
7404 declare double @llvm.ceil.f64(double %Val)
7405 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7406 declare fp128 @llvm.ceil.f128(fp128 %Val)
7407 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7412 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7417 The argument and return value are floating point numbers of the same
7423 This function returns the same values as the libm ``ceil`` functions
7424 would, and handles error conditions in the same way.
7426 '``llvm.trunc.*``' Intrinsic
7427 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7432 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7433 floating point or vector of floating point type. Not all targets support
7438 declare float @llvm.trunc.f32(float %Val)
7439 declare double @llvm.trunc.f64(double %Val)
7440 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7441 declare fp128 @llvm.trunc.f128(fp128 %Val)
7442 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7447 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7448 nearest integer not larger in magnitude than the operand.
7453 The argument and return value are floating point numbers of the same
7459 This function returns the same values as the libm ``trunc`` functions
7460 would, and handles error conditions in the same way.
7462 '``llvm.rint.*``' Intrinsic
7463 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7468 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7469 floating point or vector of floating point type. Not all targets support
7474 declare float @llvm.rint.f32(float %Val)
7475 declare double @llvm.rint.f64(double %Val)
7476 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7477 declare fp128 @llvm.rint.f128(fp128 %Val)
7478 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7483 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7484 nearest integer. It may raise an inexact floating-point exception if the
7485 operand isn't an integer.
7490 The argument and return value are floating point numbers of the same
7496 This function returns the same values as the libm ``rint`` functions
7497 would, and handles error conditions in the same way.
7499 '``llvm.nearbyint.*``' Intrinsic
7500 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7505 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7506 floating point or vector of floating point type. Not all targets support
7511 declare float @llvm.nearbyint.f32(float %Val)
7512 declare double @llvm.nearbyint.f64(double %Val)
7513 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7514 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7515 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7520 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7526 The argument and return value are floating point numbers of the same
7532 This function returns the same values as the libm ``nearbyint``
7533 functions would, and handles error conditions in the same way.
7535 '``llvm.round.*``' Intrinsic
7536 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7541 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7542 floating point or vector of floating point type. Not all targets support
7547 declare float @llvm.round.f32(float %Val)
7548 declare double @llvm.round.f64(double %Val)
7549 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7550 declare fp128 @llvm.round.f128(fp128 %Val)
7551 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7556 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7562 The argument and return value are floating point numbers of the same
7568 This function returns the same values as the libm ``round``
7569 functions would, and handles error conditions in the same way.
7571 Bit Manipulation Intrinsics
7572 ---------------------------
7574 LLVM provides intrinsics for a few important bit manipulation
7575 operations. These allow efficient code generation for some algorithms.
7577 '``llvm.bswap.*``' Intrinsics
7578 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7583 This is an overloaded intrinsic function. You can use bswap on any
7584 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7588 declare i16 @llvm.bswap.i16(i16 <id>)
7589 declare i32 @llvm.bswap.i32(i32 <id>)
7590 declare i64 @llvm.bswap.i64(i64 <id>)
7595 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7596 values with an even number of bytes (positive multiple of 16 bits).
7597 These are useful for performing operations on data that is not in the
7598 target's native byte order.
7603 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7604 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7605 intrinsic returns an i32 value that has the four bytes of the input i32
7606 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7607 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7608 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7609 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7612 '``llvm.ctpop.*``' Intrinsic
7613 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7618 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7619 bit width, or on any vector with integer elements. Not all targets
7620 support all bit widths or vector types, however.
7624 declare i8 @llvm.ctpop.i8(i8 <src>)
7625 declare i16 @llvm.ctpop.i16(i16 <src>)
7626 declare i32 @llvm.ctpop.i32(i32 <src>)
7627 declare i64 @llvm.ctpop.i64(i64 <src>)
7628 declare i256 @llvm.ctpop.i256(i256 <src>)
7629 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7634 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7640 The only argument is the value to be counted. The argument may be of any
7641 integer type, or a vector with integer elements. The return type must
7642 match the argument type.
7647 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7648 each element of a vector.
7650 '``llvm.ctlz.*``' Intrinsic
7651 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7656 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7657 integer bit width, or any vector whose elements are integers. Not all
7658 targets support all bit widths or vector types, however.
7662 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7663 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7664 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7665 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7666 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7667 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7672 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7673 leading zeros in a variable.
7678 The first argument is the value to be counted. This argument may be of
7679 any integer type, or a vectory with integer element type. The return
7680 type must match the first argument type.
7682 The second argument must be a constant and is a flag to indicate whether
7683 the intrinsic should ensure that a zero as the first argument produces a
7684 defined result. Historically some architectures did not provide a
7685 defined result for zero values as efficiently, and many algorithms are
7686 now predicated on avoiding zero-value inputs.
7691 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7692 zeros in a variable, or within each element of the vector. If
7693 ``src == 0`` then the result is the size in bits of the type of ``src``
7694 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7695 ``llvm.ctlz(i32 2) = 30``.
7697 '``llvm.cttz.*``' Intrinsic
7698 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7703 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7704 integer bit width, or any vector of integer elements. Not all targets
7705 support all bit widths or vector types, however.
7709 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7710 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7711 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7712 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7713 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7714 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7719 The '``llvm.cttz``' family of intrinsic functions counts the number of
7725 The first argument is the value to be counted. This argument may be of
7726 any integer type, or a vectory with integer element type. The return
7727 type must match the first argument type.
7729 The second argument must be a constant and is a flag to indicate whether
7730 the intrinsic should ensure that a zero as the first argument produces a
7731 defined result. Historically some architectures did not provide a
7732 defined result for zero values as efficiently, and many algorithms are
7733 now predicated on avoiding zero-value inputs.
7738 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7739 zeros in a variable, or within each element of a vector. If ``src == 0``
7740 then the result is the size in bits of the type of ``src`` if
7741 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7742 ``llvm.cttz(2) = 1``.
7744 Arithmetic with Overflow Intrinsics
7745 -----------------------------------
7747 LLVM provides intrinsics for some arithmetic with overflow operations.
7749 '``llvm.sadd.with.overflow.*``' Intrinsics
7750 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7755 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7756 on any integer bit width.
7760 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7761 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7762 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7767 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7768 a signed addition of the two arguments, and indicate whether an overflow
7769 occurred during the signed summation.
7774 The arguments (%a and %b) and the first element of the result structure
7775 may be of integer types of any bit width, but they must have the same
7776 bit width. The second element of the result structure must be of type
7777 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7783 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7784 a signed addition of the two variables. They return a structure --- the
7785 first element of which is the signed summation, and the second element
7786 of which is a bit specifying if the signed summation resulted in an
7792 .. code-block:: llvm
7794 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7795 %sum = extractvalue {i32, i1} %res, 0
7796 %obit = extractvalue {i32, i1} %res, 1
7797 br i1 %obit, label %overflow, label %normal
7799 '``llvm.uadd.with.overflow.*``' Intrinsics
7800 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7805 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7806 on any integer bit width.
7810 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7811 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7812 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7817 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7818 an unsigned addition of the two arguments, and indicate whether a carry
7819 occurred during the unsigned summation.
7824 The arguments (%a and %b) and the first element of the result structure
7825 may be of integer types of any bit width, but they must have the same
7826 bit width. The second element of the result structure must be of type
7827 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7833 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7834 an unsigned addition of the two arguments. They return a structure --- the
7835 first element of which is the sum, and the second element of which is a
7836 bit specifying if the unsigned summation resulted in a carry.
7841 .. code-block:: llvm
7843 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7844 %sum = extractvalue {i32, i1} %res, 0
7845 %obit = extractvalue {i32, i1} %res, 1
7846 br i1 %obit, label %carry, label %normal
7848 '``llvm.ssub.with.overflow.*``' Intrinsics
7849 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7854 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7855 on any integer bit width.
7859 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7860 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7861 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7866 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7867 a signed subtraction of the two arguments, and indicate whether an
7868 overflow occurred during the signed subtraction.
7873 The arguments (%a and %b) and the first element of the result structure
7874 may be of integer types of any bit width, but they must have the same
7875 bit width. The second element of the result structure must be of type
7876 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7882 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7883 a signed subtraction of the two arguments. They return a structure --- the
7884 first element of which is the subtraction, and the second element of
7885 which is a bit specifying if the signed subtraction resulted in an
7891 .. code-block:: llvm
7893 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7894 %sum = extractvalue {i32, i1} %res, 0
7895 %obit = extractvalue {i32, i1} %res, 1
7896 br i1 %obit, label %overflow, label %normal
7898 '``llvm.usub.with.overflow.*``' Intrinsics
7899 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7904 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
7905 on any integer bit width.
7909 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7910 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7911 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7916 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7917 an unsigned subtraction of the two arguments, and indicate whether an
7918 overflow occurred during the unsigned subtraction.
7923 The arguments (%a and %b) and the first element of the result structure
7924 may be of integer types of any bit width, but they must have the same
7925 bit width. The second element of the result structure must be of type
7926 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7932 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7933 an unsigned subtraction of the two arguments. They return a structure ---
7934 the first element of which is the subtraction, and the second element of
7935 which is a bit specifying if the unsigned subtraction resulted in an
7941 .. code-block:: llvm
7943 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7944 %sum = extractvalue {i32, i1} %res, 0
7945 %obit = extractvalue {i32, i1} %res, 1
7946 br i1 %obit, label %overflow, label %normal
7948 '``llvm.smul.with.overflow.*``' Intrinsics
7949 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7954 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
7955 on any integer bit width.
7959 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7960 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7961 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7966 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7967 a signed multiplication of the two arguments, and indicate whether an
7968 overflow occurred during the signed multiplication.
7973 The arguments (%a and %b) and the first element of the result structure
7974 may be of integer types of any bit width, but they must have the same
7975 bit width. The second element of the result structure must be of type
7976 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7982 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7983 a signed multiplication of the two arguments. They return a structure ---
7984 the first element of which is the multiplication, and the second element
7985 of which is a bit specifying if the signed multiplication resulted in an
7991 .. code-block:: llvm
7993 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7994 %sum = extractvalue {i32, i1} %res, 0
7995 %obit = extractvalue {i32, i1} %res, 1
7996 br i1 %obit, label %overflow, label %normal
7998 '``llvm.umul.with.overflow.*``' Intrinsics
7999 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8004 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8005 on any integer bit width.
8009 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8010 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8011 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8016 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8017 a unsigned multiplication of the two arguments, and indicate whether an
8018 overflow occurred during the unsigned multiplication.
8023 The arguments (%a and %b) and the first element of the result structure
8024 may be of integer types of any bit width, but they must have the same
8025 bit width. The second element of the result structure must be of type
8026 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8032 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8033 an unsigned multiplication of the two arguments. They return a structure ---
8034 the first element of which is the multiplication, and the second
8035 element of which is a bit specifying if the unsigned multiplication
8036 resulted in an overflow.
8041 .. code-block:: llvm
8043 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8044 %sum = extractvalue {i32, i1} %res, 0
8045 %obit = extractvalue {i32, i1} %res, 1
8046 br i1 %obit, label %overflow, label %normal
8048 Specialised Arithmetic Intrinsics
8049 ---------------------------------
8051 '``llvm.fmuladd.*``' Intrinsic
8052 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8059 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8060 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8065 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8066 expressions that can be fused if the code generator determines that (a) the
8067 target instruction set has support for a fused operation, and (b) that the
8068 fused operation is more efficient than the equivalent, separate pair of mul
8069 and add instructions.
8074 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8075 multiplicands, a and b, and an addend c.
8084 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8086 is equivalent to the expression a \* b + c, except that rounding will
8087 not be performed between the multiplication and addition steps if the
8088 code generator fuses the operations. Fusion is not guaranteed, even if
8089 the target platform supports it. If a fused multiply-add is required the
8090 corresponding llvm.fma.\* intrinsic function should be used instead.
8095 .. code-block:: llvm
8097 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8099 Half Precision Floating Point Intrinsics
8100 ----------------------------------------
8102 For most target platforms, half precision floating point is a
8103 storage-only format. This means that it is a dense encoding (in memory)
8104 but does not support computation in the format.
8106 This means that code must first load the half-precision floating point
8107 value as an i16, then convert it to float with
8108 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8109 then be performed on the float value (including extending to double
8110 etc). To store the value back to memory, it is first converted to float
8111 if needed, then converted to i16 with
8112 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8115 .. _int_convert_to_fp16:
8117 '``llvm.convert.to.fp16``' Intrinsic
8118 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8125 declare i16 @llvm.convert.to.fp16(f32 %a)
8130 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8131 from single precision floating point format to half precision floating
8137 The intrinsic function contains single argument - the value to be
8143 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8144 from single precision floating point format to half precision floating
8145 point format. The return value is an ``i16`` which contains the
8151 .. code-block:: llvm
8153 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8154 store i16 %res, i16* @x, align 2
8156 .. _int_convert_from_fp16:
8158 '``llvm.convert.from.fp16``' Intrinsic
8159 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8166 declare f32 @llvm.convert.from.fp16(i16 %a)
8171 The '``llvm.convert.from.fp16``' intrinsic function performs a
8172 conversion from half precision floating point format to single precision
8173 floating point format.
8178 The intrinsic function contains single argument - the value to be
8184 The '``llvm.convert.from.fp16``' intrinsic function performs a
8185 conversion from half single precision floating point format to single
8186 precision floating point format. The input half-float value is
8187 represented by an ``i16`` value.
8192 .. code-block:: llvm
8194 %a = load i16* @x, align 2
8195 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8200 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8201 prefix), are described in the `LLVM Source Level
8202 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8205 Exception Handling Intrinsics
8206 -----------------------------
8208 The LLVM exception handling intrinsics (which all start with
8209 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8210 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8214 Trampoline Intrinsics
8215 ---------------------
8217 These intrinsics make it possible to excise one parameter, marked with
8218 the :ref:`nest <nest>` attribute, from a function. The result is a
8219 callable function pointer lacking the nest parameter - the caller does
8220 not need to provide a value for it. Instead, the value to use is stored
8221 in advance in a "trampoline", a block of memory usually allocated on the
8222 stack, which also contains code to splice the nest value into the
8223 argument list. This is used to implement the GCC nested function address
8226 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8227 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8228 It can be created as follows:
8230 .. code-block:: llvm
8232 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8233 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8234 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8235 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8236 %fp = bitcast i8* %p to i32 (i32, i32)*
8238 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8239 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8243 '``llvm.init.trampoline``' Intrinsic
8244 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8251 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8256 This fills the memory pointed to by ``tramp`` with executable code,
8257 turning it into a trampoline.
8262 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8263 pointers. The ``tramp`` argument must point to a sufficiently large and
8264 sufficiently aligned block of memory; this memory is written to by the
8265 intrinsic. Note that the size and the alignment are target-specific -
8266 LLVM currently provides no portable way of determining them, so a
8267 front-end that generates this intrinsic needs to have some
8268 target-specific knowledge. The ``func`` argument must hold a function
8269 bitcast to an ``i8*``.
8274 The block of memory pointed to by ``tramp`` is filled with target
8275 dependent code, turning it into a function. Then ``tramp`` needs to be
8276 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8277 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8278 function's signature is the same as that of ``func`` with any arguments
8279 marked with the ``nest`` attribute removed. At most one such ``nest``
8280 argument is allowed, and it must be of pointer type. Calling the new
8281 function is equivalent to calling ``func`` with the same argument list,
8282 but with ``nval`` used for the missing ``nest`` argument. If, after
8283 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8284 modified, then the effect of any later call to the returned function
8285 pointer is undefined.
8289 '``llvm.adjust.trampoline``' Intrinsic
8290 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8297 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8302 This performs any required machine-specific adjustment to the address of
8303 a trampoline (passed as ``tramp``).
8308 ``tramp`` must point to a block of memory which already has trampoline
8309 code filled in by a previous call to
8310 :ref:`llvm.init.trampoline <int_it>`.
8315 On some architectures the address of the code to be executed needs to be
8316 different to the address where the trampoline is actually stored. This
8317 intrinsic returns the executable address corresponding to ``tramp``
8318 after performing the required machine specific adjustments. The pointer
8319 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8324 This class of intrinsics exists to information about the lifetime of
8325 memory objects and ranges where variables are immutable.
8327 '``llvm.lifetime.start``' Intrinsic
8328 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8335 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8340 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8346 The first argument is a constant integer representing the size of the
8347 object, or -1 if it is variable sized. The second argument is a pointer
8353 This intrinsic indicates that before this point in the code, the value
8354 of the memory pointed to by ``ptr`` is dead. This means that it is known
8355 to never be used and has an undefined value. A load from the pointer
8356 that precedes this intrinsic can be replaced with ``'undef'``.
8358 '``llvm.lifetime.end``' Intrinsic
8359 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8366 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8371 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8377 The first argument is a constant integer representing the size of the
8378 object, or -1 if it is variable sized. The second argument is a pointer
8384 This intrinsic indicates that after this point in the code, the value of
8385 the memory pointed to by ``ptr`` is dead. This means that it is known to
8386 never be used and has an undefined value. Any stores into the memory
8387 object following this intrinsic may be removed as dead.
8389 '``llvm.invariant.start``' Intrinsic
8390 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8397 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8402 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8403 a memory object will not change.
8408 The first argument is a constant integer representing the size of the
8409 object, or -1 if it is variable sized. The second argument is a pointer
8415 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8416 the return value, the referenced memory location is constant and
8419 '``llvm.invariant.end``' Intrinsic
8420 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8427 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8432 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8433 memory object are mutable.
8438 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8439 The second argument is a constant integer representing the size of the
8440 object, or -1 if it is variable sized and the third argument is a
8441 pointer to the object.
8446 This intrinsic indicates that the memory is mutable again.
8451 This class of intrinsics is designed to be generic and has no specific
8454 '``llvm.var.annotation``' Intrinsic
8455 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8462 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8467 The '``llvm.var.annotation``' intrinsic.
8472 The first argument is a pointer to a value, the second is a pointer to a
8473 global string, the third is a pointer to a global string which is the
8474 source file name, and the last argument is the line number.
8479 This intrinsic allows annotation of local variables with arbitrary
8480 strings. This can be useful for special purpose optimizations that want
8481 to look for these annotations. These have no other defined use; they are
8482 ignored by code generation and optimization.
8484 '``llvm.ptr.annotation.*``' Intrinsic
8485 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8490 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8491 pointer to an integer of any width. *NOTE* you must specify an address space for
8492 the pointer. The identifier for the default address space is the integer
8497 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8498 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8499 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8500 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8501 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8506 The '``llvm.ptr.annotation``' intrinsic.
8511 The first argument is a pointer to an integer value of arbitrary bitwidth
8512 (result of some expression), the second is a pointer to a global string, the
8513 third is a pointer to a global string which is the source file name, and the
8514 last argument is the line number. It returns the value of the first argument.
8519 This intrinsic allows annotation of a pointer to an integer with arbitrary
8520 strings. This can be useful for special purpose optimizations that want to look
8521 for these annotations. These have no other defined use; they are ignored by code
8522 generation and optimization.
8524 '``llvm.annotation.*``' Intrinsic
8525 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8530 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8531 any integer bit width.
8535 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8536 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8537 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8538 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8539 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8544 The '``llvm.annotation``' intrinsic.
8549 The first argument is an integer value (result of some expression), the
8550 second is a pointer to a global string, the third is a pointer to a
8551 global string which is the source file name, and the last argument is
8552 the line number. It returns the value of the first argument.
8557 This intrinsic allows annotations to be put on arbitrary expressions
8558 with arbitrary strings. This can be useful for special purpose
8559 optimizations that want to look for these annotations. These have no
8560 other defined use; they are ignored by code generation and optimization.
8562 '``llvm.trap``' Intrinsic
8563 ^^^^^^^^^^^^^^^^^^^^^^^^^
8570 declare void @llvm.trap() noreturn nounwind
8575 The '``llvm.trap``' intrinsic.
8585 This intrinsic is lowered to the target dependent trap instruction. If
8586 the target does not have a trap instruction, this intrinsic will be
8587 lowered to a call of the ``abort()`` function.
8589 '``llvm.debugtrap``' Intrinsic
8590 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8597 declare void @llvm.debugtrap() nounwind
8602 The '``llvm.debugtrap``' intrinsic.
8612 This intrinsic is lowered to code which is intended to cause an
8613 execution trap with the intention of requesting the attention of a
8616 '``llvm.stackprotector``' Intrinsic
8617 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8624 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8629 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8630 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8631 is placed on the stack before local variables.
8636 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8637 The first argument is the value loaded from the stack guard
8638 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8639 enough space to hold the value of the guard.
8644 This intrinsic causes the prologue/epilogue inserter to force the
8645 position of the ``AllocaInst`` stack slot to be before local variables
8646 on the stack. This is to ensure that if a local variable on the stack is
8647 overwritten, it will destroy the value of the guard. When the function
8648 exits, the guard on the stack is checked against the original guard. If
8649 they are different, then the program aborts by calling the
8650 ``__stack_chk_fail()`` function.
8652 '``llvm.objectsize``' Intrinsic
8653 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8660 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8661 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8666 The ``llvm.objectsize`` intrinsic is designed to provide information to
8667 the optimizers to determine at compile time whether a) an operation
8668 (like memcpy) will overflow a buffer that corresponds to an object, or
8669 b) that a runtime check for overflow isn't necessary. An object in this
8670 context means an allocation of a specific class, structure, array, or
8676 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8677 argument is a pointer to or into the ``object``. The second argument is
8678 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8679 or -1 (if false) when the object size is unknown. The second argument
8680 only accepts constants.
8685 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8686 the size of the object concerned. If the size cannot be determined at
8687 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8688 on the ``min`` argument).
8690 '``llvm.expect``' Intrinsic
8691 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8698 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8699 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8704 The ``llvm.expect`` intrinsic provides information about expected (the
8705 most probable) value of ``val``, which can be used by optimizers.
8710 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8711 a value. The second argument is an expected value, this needs to be a
8712 constant value, variables are not allowed.
8717 This intrinsic is lowered to the ``val``.
8719 '``llvm.donothing``' Intrinsic
8720 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8727 declare void @llvm.donothing() nounwind readnone
8732 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8733 only intrinsic that can be called with an invoke instruction.
8743 This intrinsic does nothing, and it's removed by optimizers and ignored