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, i8 3
120 %0 = add i32 %X, %X ; yields {i32}:%0
121 %1 = add i32 %0, %0 ; yields {i32}:%1
122 %result = add i32 %1, %1
124 This last way of multiplying ``%X`` by 8 illustrates several important
125 lexical features of LLVM:
127 #. Comments are delimited with a '``;``' and go until the end of line.
128 #. Unnamed temporaries are created when the result of a computation is
129 not assigned to a named value.
130 #. Unnamed temporaries are numbered sequentially
132 It also shows a convention that we follow in this document. When
133 demonstrating instructions, we will follow an instruction with a comment
134 that defines the type and name of value produced.
142 LLVM programs are composed of ``Module``'s, each of which is a
143 translation unit of the input programs. Each module consists of
144 functions, global variables, and symbol table entries. Modules may be
145 combined together with the LLVM linker, which merges function (and
146 global variable) definitions, resolves forward declarations, and merges
147 symbol table entries. Here is an example of the "hello world" module:
151 ; Declare the string constant as a global constant.
152 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
154 ; External declaration of the puts function
155 declare i32 @puts(i8* nocapture) nounwind
157 ; Definition of main function
158 define i32 @main() { ; i32()*
159 ; Convert [13 x i8]* to i8 *...
160 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
162 ; Call puts function to write out the string to stdout.
163 call i32 @puts(i8* %cast210)
168 !1 = metadata !{i32 42}
171 This example is made up of a :ref:`global variable <globalvars>` named
172 "``.str``", an external declaration of the "``puts``" function, a
173 :ref:`function definition <functionstructure>` for "``main``" and
174 :ref:`named metadata <namedmetadatastructure>` "``foo``".
176 In general, a module is made up of a list of global values (where both
177 functions and global variables are global values). Global values are
178 represented by a pointer to a memory location (in this case, a pointer
179 to an array of char, and a pointer to a function), and have one of the
180 following :ref:`linkage types <linkage>`.
187 All Global Variables and Functions have one of the following types of
191 Global values with "``private``" linkage are only directly
192 accessible by objects in the current module. In particular, linking
193 code into a module with an private global value may cause the
194 private to be renamed as necessary to avoid collisions. Because the
195 symbol is private to the module, all references can be updated. This
196 doesn't show up in any symbol table in the object file.
198 Similar to ``private``, but the symbol is passed through the
199 assembler and evaluated by the linker. Unlike normal strong symbols,
200 they are removed by the linker from the final linked image
201 (executable or dynamic library).
202 ``linker_private_weak``
203 Similar to "``linker_private``", but the symbol is weak. Note that
204 ``linker_private_weak`` symbols are subject to coalescing by the
205 linker. The symbols are removed by the linker from the final linked
206 image (executable or dynamic library).
208 Similar to private, but the value shows as a local symbol
209 (``STB_LOCAL`` in the case of ELF) in the object file. This
210 corresponds to the notion of the '``static``' keyword in C.
211 ``available_externally``
212 Globals with "``available_externally``" linkage are never emitted
213 into the object file corresponding to the LLVM module. They exist to
214 allow inlining and other optimizations to take place given knowledge
215 of the definition of the global, which is known to be somewhere
216 outside the module. Globals with ``available_externally`` linkage
217 are allowed to be discarded at will, and are otherwise the same as
218 ``linkonce_odr``. This linkage type is only allowed on definitions,
221 Globals with "``linkonce``" linkage are merged with other globals of
222 the same name when linkage occurs. This can be used to implement
223 some forms of inline functions, templates, or other code which must
224 be generated in each translation unit that uses it, but where the
225 body may be overridden with a more definitive definition later.
226 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
227 that ``linkonce`` linkage does not actually allow the optimizer to
228 inline the body of this function into callers because it doesn't
229 know if this definition of the function is the definitive definition
230 within the program or whether it will be overridden by a stronger
231 definition. To enable inlining and other optimizations, use
232 "``linkonce_odr``" linkage.
234 "``weak``" linkage has the same merging semantics as ``linkonce``
235 linkage, except that unreferenced globals with ``weak`` linkage may
236 not be discarded. This is used for globals that are declared "weak"
239 "``common``" linkage is most similar to "``weak``" linkage, but they
240 are used for tentative definitions in C, such as "``int X;``" at
241 global scope. Symbols with "``common``" linkage are merged in the
242 same way as ``weak symbols``, and they may not be deleted if
243 unreferenced. ``common`` symbols may not have an explicit section,
244 must have a zero initializer, and may not be marked
245 ':ref:`constant <globalvars>`'. Functions and aliases may not have
248 .. _linkage_appending:
251 "``appending``" linkage may only be applied to global variables of
252 pointer to array type. When two global variables with appending
253 linkage are linked together, the two global arrays are appended
254 together. This is the LLVM, typesafe, equivalent of having the
255 system linker append together "sections" with identical names when
258 The semantics of this linkage follow the ELF object file model: the
259 symbol is weak until linked, if not linked, the symbol becomes null
260 instead of being an undefined reference.
261 ``linkonce_odr``, ``weak_odr``
262 Some languages allow differing globals to be merged, such as two
263 functions with different semantics. Other languages, such as
264 ``C++``, ensure that only equivalent globals are ever merged (the
265 "one definition rule" --- "ODR"). Such languages can use the
266 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
267 global will only be merged with equivalent globals. These linkage
268 types are otherwise the same as their non-``odr`` versions.
269 ``linkonce_odr_auto_hide``
270 Similar to "``linkonce_odr``", but nothing in the translation unit
271 takes the address of this definition. For instance, functions that
272 had an inline definition, but the compiler decided not to inline it.
273 ``linkonce_odr_auto_hide`` may have only ``default`` visibility. The
274 symbols are removed by the linker from the final linked image
275 (executable or dynamic library).
277 If none of the above identifiers are used, the global is externally
278 visible, meaning that it participates in linkage and can be used to
279 resolve external symbol references.
281 The next two types of linkage are targeted for Microsoft Windows
282 platform only. They are designed to support importing (exporting)
283 symbols from (to) DLLs (Dynamic Link Libraries).
286 "``dllimport``" linkage causes the compiler to reference a function
287 or variable via a global pointer to a pointer that is set up by the
288 DLL exporting the symbol. On Microsoft Windows targets, the pointer
289 name is formed by combining ``__imp_`` and the function or variable
292 "``dllexport``" linkage causes the compiler to provide a global
293 pointer to a pointer in a DLL, so that it can be referenced with the
294 ``dllimport`` attribute. On Microsoft Windows targets, the pointer
295 name is formed by combining ``__imp_`` and the function or variable
298 For example, since the "``.LC0``" variable is defined to be internal, if
299 another module defined a "``.LC0``" variable and was linked with this
300 one, one of the two would be renamed, preventing a collision. Since
301 "``main``" and "``puts``" are external (i.e., lacking any linkage
302 declarations), they are accessible outside of the current module.
304 It is illegal for a function *declaration* to have any linkage type
305 other than ``external``, ``dllimport`` or ``extern_weak``.
307 Aliases can have only ``external``, ``internal``, ``weak`` or
308 ``weak_odr`` linkages.
315 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
316 :ref:`invokes <i_invoke>` can all have an optional calling convention
317 specified for the call. The calling convention of any pair of dynamic
318 caller/callee must match, or the behavior of the program is undefined.
319 The following calling conventions are supported by LLVM, and more may be
322 "``ccc``" - The C calling convention
323 This calling convention (the default if no other calling convention
324 is specified) matches the target C calling conventions. This calling
325 convention supports varargs function calls and tolerates some
326 mismatch in the declared prototype and implemented declaration of
327 the function (as does normal C).
328 "``fastcc``" - The fast calling convention
329 This calling convention attempts to make calls as fast as possible
330 (e.g. by passing things in registers). This calling convention
331 allows the target to use whatever tricks it wants to produce fast
332 code for the target, without having to conform to an externally
333 specified ABI (Application Binary Interface). `Tail calls can only
334 be optimized when this, the GHC or the HiPE convention is
335 used. <CodeGenerator.html#id80>`_ This calling convention does not
336 support varargs and requires the prototype of all callees to exactly
337 match the prototype of the function definition.
338 "``coldcc``" - The cold calling convention
339 This calling convention attempts to make code in the caller as
340 efficient as possible under the assumption that the call is not
341 commonly executed. As such, these calls often preserve all registers
342 so that the call does not break any live ranges in the caller side.
343 This calling convention does not support varargs and requires the
344 prototype of all callees to exactly match the prototype of the
346 "``cc 10``" - GHC convention
347 This calling convention has been implemented specifically for use by
348 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
349 It passes everything in registers, going to extremes to achieve this
350 by disabling callee save registers. This calling convention should
351 not be used lightly but only for specific situations such as an
352 alternative to the *register pinning* performance technique often
353 used when implementing functional programming languages. At the
354 moment only X86 supports this convention and it has the following
357 - On *X86-32* only supports up to 4 bit type parameters. No
358 floating point types are supported.
359 - On *X86-64* only supports up to 10 bit type parameters and 6
360 floating point parameters.
362 This calling convention supports `tail call
363 optimization <CodeGenerator.html#id80>`_ but requires both the
364 caller and callee are using it.
365 "``cc 11``" - The HiPE calling convention
366 This calling convention has been implemented specifically for use by
367 the `High-Performance Erlang
368 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
369 native code compiler of the `Ericsson's Open Source Erlang/OTP
370 system <http://www.erlang.org/download.shtml>`_. It uses more
371 registers for argument passing than the ordinary C calling
372 convention and defines no callee-saved registers. The calling
373 convention properly supports `tail call
374 optimization <CodeGenerator.html#id80>`_ but requires that both the
375 caller and the callee use it. It uses a *register pinning*
376 mechanism, similar to GHC's convention, for keeping frequently
377 accessed runtime components pinned to specific hardware registers.
378 At the moment only X86 supports this convention (both 32 and 64
380 "``cc <n>``" - Numbered convention
381 Any calling convention may be specified by number, allowing
382 target-specific calling conventions to be used. Target specific
383 calling conventions start at 64.
385 More calling conventions can be added/defined on an as-needed basis, to
386 support Pascal conventions or any other well-known target-independent
392 All Global Variables and Functions have one of the following visibility
395 "``default``" - Default style
396 On targets that use the ELF object file format, default visibility
397 means that the declaration is visible to other modules and, in
398 shared libraries, means that the declared entity may be overridden.
399 On Darwin, default visibility means that the declaration is visible
400 to other modules. Default visibility corresponds to "external
401 linkage" in the language.
402 "``hidden``" - Hidden style
403 Two declarations of an object with hidden visibility refer to the
404 same object if they are in the same shared object. Usually, hidden
405 visibility indicates that the symbol will not be placed into the
406 dynamic symbol table, so no other module (executable or shared
407 library) can reference it directly.
408 "``protected``" - Protected style
409 On ELF, protected visibility indicates that the symbol will be
410 placed in the dynamic symbol table, but that references within the
411 defining module will bind to the local symbol. That is, the symbol
412 cannot be overridden by another module.
417 LLVM IR allows you to specify name aliases for certain types. This can
418 make it easier to read the IR and make the IR more condensed
419 (particularly when recursive types are involved). An example of a name
424 %mytype = type { %mytype*, i32 }
426 You may give a name to any :ref:`type <typesystem>` except
427 ":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
428 expected with the syntax "%mytype".
430 Note that type names are aliases for the structural type that they
431 indicate, and that you can therefore specify multiple names for the same
432 type. This often leads to confusing behavior when dumping out a .ll
433 file. Since LLVM IR uses structural typing, the name is not part of the
434 type. When printing out LLVM IR, the printer will pick *one name* to
435 render all types of a particular shape. This means that if you have code
436 where two different source types end up having the same LLVM type, that
437 the dumper will sometimes print the "wrong" or unexpected type. This is
438 an important design point and isn't going to change.
445 Global variables define regions of memory allocated at compilation time
446 instead of run-time. Global variables may optionally be initialized, may
447 have an explicit section to be placed in, and may have an optional
448 explicit alignment specified.
450 A variable may be defined as ``thread_local``, which means that it will
451 not be shared by threads (each thread will have a separated copy of the
452 variable). Not all targets support thread-local variables. Optionally, a
453 TLS model may be specified:
456 For variables that are only used within the current shared library.
458 For variables in modules that will not be loaded dynamically.
460 For variables defined in the executable and only used within it.
462 The models correspond to the ELF TLS models; see `ELF Handling For
463 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
464 more information on under which circumstances the different models may
465 be used. The target may choose a different TLS model if the specified
466 model is not supported, or if a better choice of model can be made.
468 A variable may be defined as a global "constant," which indicates that
469 the contents of the variable will **never** be modified (enabling better
470 optimization, allowing the global data to be placed in the read-only
471 section of an executable, etc). Note that variables that need runtime
472 initialization cannot be marked "constant" as there is a store to the
475 LLVM explicitly allows *declarations* of global variables to be marked
476 constant, even if the final definition of the global is not. This
477 capability can be used to enable slightly better optimization of the
478 program, but requires the language definition to guarantee that
479 optimizations based on the 'constantness' are valid for the translation
480 units that do not include the definition.
482 As SSA values, global variables define pointer values that are in scope
483 (i.e. they dominate) all basic blocks in the program. Global variables
484 always define a pointer to their "content" type because they describe a
485 region of memory, and all memory objects in LLVM are accessed through
488 Global variables can be marked with ``unnamed_addr`` which indicates
489 that the address is not significant, only the content. Constants marked
490 like this can be merged with other constants if they have the same
491 initializer. Note that a constant with significant address *can* be
492 merged with a ``unnamed_addr`` constant, the result being a constant
493 whose address is significant.
495 A global variable may be declared to reside in a target-specific
496 numbered address space. For targets that support them, address spaces
497 may affect how optimizations are performed and/or what target
498 instructions are used to access the variable. The default address space
499 is zero. The address space qualifier must precede any other attributes.
501 LLVM allows an explicit section to be specified for globals. If the
502 target supports it, it will emit globals to the section specified.
504 An explicit alignment may be specified for a global, which must be a
505 power of 2. If not present, or if the alignment is set to zero, the
506 alignment of the global is set by the target to whatever it feels
507 convenient. If an explicit alignment is specified, the global is forced
508 to have exactly that alignment. Targets and optimizers are not allowed
509 to over-align the global if the global has an assigned section. In this
510 case, the extra alignment could be observable: for example, code could
511 assume that the globals are densely packed in their section and try to
512 iterate over them as an array, alignment padding would break this
515 For example, the following defines a global in a numbered address space
516 with an initializer, section, and alignment:
520 @G = addrspace(5) constant float 1.0, section "foo", align 4
522 The following example defines a thread-local global with the
523 ``initialexec`` TLS model:
527 @G = thread_local(initialexec) global i32 0, align 4
529 .. _functionstructure:
534 LLVM function definitions consist of the "``define``" keyword, an
535 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
536 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
537 an optional ``unnamed_addr`` attribute, a return type, an optional
538 :ref:`parameter attribute <paramattrs>` for the return type, a function
539 name, a (possibly empty) argument list (each with optional :ref:`parameter
540 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
541 an optional section, an optional alignment, an optional :ref:`garbage
542 collector name <gc>`, an opening curly brace, a list of basic blocks,
543 and a closing curly brace.
545 LLVM function declarations consist of the "``declare``" keyword, an
546 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
547 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
548 an optional ``unnamed_addr`` attribute, a return type, an optional
549 :ref:`parameter attribute <paramattrs>` for the return type, a function
550 name, a possibly empty list of arguments, an optional alignment, and an
551 optional :ref:`garbage collector name <gc>`.
553 A function definition contains a list of basic blocks, forming the CFG
554 (Control Flow Graph) for the function. Each basic block may optionally
555 start with a label (giving the basic block a symbol table entry),
556 contains a list of instructions, and ends with a
557 :ref:`terminator <terminators>` instruction (such as a branch or function
560 The first basic block in a function is special in two ways: it is
561 immediately executed on entrance to the function, and it is not allowed
562 to have predecessor basic blocks (i.e. there can not be any branches to
563 the entry block of a function). Because the block can have no
564 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
566 LLVM allows an explicit section to be specified for functions. If the
567 target supports it, it will emit functions to the section specified.
569 An explicit alignment may be specified for a function. If not present,
570 or if the alignment is set to zero, the alignment of the function is set
571 by the target to whatever it feels convenient. If an explicit alignment
572 is specified, the function is forced to have at least that much
573 alignment. All alignments must be a power of 2.
575 If the ``unnamed_addr`` attribute is given, the address is know to not
576 be significant and two identical functions can be merged.
580 define [linkage] [visibility]
582 <ResultType> @<FunctionName> ([argument list])
583 [fn Attrs] [section "name"] [align N]
589 Aliases act as "second name" for the aliasee value (which can be either
590 function, global variable, another alias or bitcast of global value).
591 Aliases may have an optional :ref:`linkage type <linkage>`, and an optional
592 :ref:`visibility style <visibility>`.
596 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
598 .. _namedmetadatastructure:
603 Named metadata is a collection of metadata. :ref:`Metadata
604 nodes <metadata>` (but not metadata strings) are the only valid
605 operands for a named metadata.
609 ; Some unnamed metadata nodes, which are referenced by the named metadata.
610 !0 = metadata !{metadata !"zero"}
611 !1 = metadata !{metadata !"one"}
612 !2 = metadata !{metadata !"two"}
614 !name = !{!0, !1, !2}
621 The return type and each parameter of a function type may have a set of
622 *parameter attributes* associated with them. Parameter attributes are
623 used to communicate additional information about the result or
624 parameters of a function. Parameter attributes are considered to be part
625 of the function, not of the function type, so functions with different
626 parameter attributes can have the same function type.
628 Parameter attributes are simple keywords that follow the type specified.
629 If multiple parameter attributes are needed, they are space separated.
634 declare i32 @printf(i8* noalias nocapture, ...)
635 declare i32 @atoi(i8 zeroext)
636 declare signext i8 @returns_signed_char()
638 Note that any attributes for the function result (``nounwind``,
639 ``readonly``) come immediately after the argument list.
641 Currently, only the following parameter attributes are defined:
644 This indicates to the code generator that the parameter or return
645 value should be zero-extended to the extent required by the target's
646 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
647 the caller (for a parameter) or the callee (for a return value).
649 This indicates to the code generator that the parameter or return
650 value should be sign-extended to the extent required by the target's
651 ABI (which is usually 32-bits) by the caller (for a parameter) or
652 the callee (for a return value).
654 This indicates that this parameter or return value should be treated
655 in a special target-dependent fashion during while emitting code for
656 a function call or return (usually, by putting it in a register as
657 opposed to memory, though some targets use it to distinguish between
658 two different kinds of registers). Use of this attribute is
661 This indicates that the pointer parameter should really be passed by
662 value to the function. The attribute implies that a hidden copy of
663 the pointee is made between the caller and the callee, so the callee
664 is unable to modify the value in the caller. This attribute is only
665 valid on LLVM pointer arguments. It is generally used to pass
666 structs and arrays by value, but is also valid on pointers to
667 scalars. The copy is considered to belong to the caller not the
668 callee (for example, ``readonly`` functions should not write to
669 ``byval`` parameters). This is not a valid attribute for return
672 The byval attribute also supports specifying an alignment with the
673 align attribute. It indicates the alignment of the stack slot to
674 form and the known alignment of the pointer specified to the call
675 site. If the alignment is not specified, then the code generator
676 makes a target-specific assumption.
679 This indicates that the pointer parameter specifies the address of a
680 structure that is the return value of the function in the source
681 program. This pointer must be guaranteed by the caller to be valid:
682 loads and stores to the structure may be assumed by the callee to
683 not to trap and to be properly aligned. This may only be applied to
684 the first parameter. This is not a valid attribute for return
687 This indicates that pointer values `*based* <pointeraliasing>` on
688 the argument or return value do not alias pointer values which are
689 not *based* on it, ignoring certain "irrelevant" dependencies. For a
690 call to the parent function, dependencies between memory references
691 from before or after the call and from those during the call are
692 "irrelevant" to the ``noalias`` keyword for the arguments and return
693 value used in that call. The caller shares the responsibility with
694 the callee for ensuring that these requirements are met. For further
695 details, please see the discussion of the NoAlias response in `alias
696 analysis <AliasAnalysis.html#MustMayNo>`_.
698 Note that this definition of ``noalias`` is intentionally similar
699 to the definition of ``restrict`` in C99 for function arguments,
700 though it is slightly weaker.
702 For function return values, C99's ``restrict`` is not meaningful,
703 while LLVM's ``noalias`` is.
705 This indicates that the callee does not make any copies of the
706 pointer that outlive the callee itself. This is not a valid
707 attribute for return values.
712 This indicates that the pointer parameter can be excised using the
713 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
714 attribute for return values.
718 Garbage Collector Names
719 -----------------------
721 Each function may specify a garbage collector name, which is simply a
726 define void @f() gc "name" { ... }
728 The compiler declares the supported values of *name*. Specifying a
729 collector which will cause the compiler to alter its output in order to
730 support the named garbage collection algorithm.
737 Function attributes are set to communicate additional information about
738 a function. Function attributes are considered to be part of the
739 function, not of the function type, so functions with different function
740 attributes can have the same function type.
742 Function attributes are simple keywords that follow the type specified.
743 If multiple attributes are needed, they are space separated. For
748 define void @f() noinline { ... }
749 define void @f() alwaysinline { ... }
750 define void @f() alwaysinline optsize { ... }
751 define void @f() optsize { ... }
754 This attribute indicates that the address safety analysis is enabled
757 This attribute indicates that, when emitting the prologue and
758 epilogue, the backend should forcibly align the stack pointer.
759 Specify the desired alignment, which must be a power of two, in
762 This attribute indicates that the inliner should attempt to inline
763 this function into callers whenever possible, ignoring any active
764 inlining size threshold for this caller.
766 This attribute suppresses lazy symbol binding for the function. This
767 may make calls to the function faster, at the cost of extra program
768 startup time if the function is not called during program startup.
770 This attribute indicates that the source code contained a hint that
771 inlining this function is desirable (such as the "inline" keyword in
772 C/C++). It is just a hint; it imposes no requirements on the
775 This attribute disables prologue / epilogue emission for the
776 function. This can have very system-specific consequences.
778 This attributes disables implicit floating point instructions.
780 This attribute indicates that the inliner should never inline this
781 function in any situation. This attribute may not be used together
782 with the ``alwaysinline`` attribute.
784 This attribute indicates that the code generator should not use a
785 red zone, even if the target-specific ABI normally permits it.
787 This function attribute indicates that the function never returns
788 normally. This produces undefined behavior at runtime if the
789 function ever does dynamically return.
791 This function attribute indicates that the function never returns
792 with an unwind or exceptional control flow. If the function does
793 unwind, its runtime behavior is undefined.
795 This attribute suggests that optimization passes and code generator
796 passes make choices that keep the code size of this function low,
797 and otherwise do optimizations specifically to reduce code size.
799 This attribute indicates that the function computes its result (or
800 decides to unwind an exception) based strictly on its arguments,
801 without dereferencing any pointer arguments or otherwise accessing
802 any mutable state (e.g. memory, control registers, etc) visible to
803 caller functions. It does not write through any pointer arguments
804 (including ``byval`` arguments) and never changes any state visible
805 to callers. This means that it cannot unwind exceptions by calling
806 the ``C++`` exception throwing methods.
808 This attribute indicates that the function does not write through
809 any pointer arguments (including ``byval`` arguments) or otherwise
810 modify any state (e.g. memory, control registers, etc) visible to
811 caller functions. It may dereference pointer arguments and read
812 state that may be set in the caller. A readonly function always
813 returns the same value (or unwinds an exception identically) when
814 called with the same set of arguments and global state. It cannot
815 unwind an exception by calling the ``C++`` exception throwing
818 This attribute indicates that this function can return twice. The C
819 ``setjmp`` is an example of such a function. The compiler disables
820 some optimizations (like tail calls) in the caller of these
823 This attribute indicates that the function should emit a stack
824 smashing protector. It is in the form of a "canary" --- a random value
825 placed on the stack before the local variables that's checked upon
826 return from the function to see if it has been overwritten. A
827 heuristic is used to determine if a function needs stack protectors
828 or not. The heuristic used will enable protectors for functions with:
829 - Character arrays larger than ``ssp-buffer-size`` (default 8).
830 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
831 - Calls to alloca() with variable sizes or constant sizes greater than
834 If a function that has an ``ssp`` attribute is inlined into a
835 function that doesn't have an ``ssp`` attribute, then the resulting
836 function will have an ``ssp`` attribute.
838 This attribute indicates that the function should *always* emit a
839 stack smashing protector. This overrides the ``ssp`` function
842 If a function that has an ``sspreq`` attribute is inlined into a
843 function that doesn't have an ``sspreq`` attribute or which has an
844 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
845 an ``sspreq`` attribute.
847 This attribute indicates that the function should emit a stack smashing
848 protector. This attribute causes a strong heuristic to be used when
849 determining if a function needs stack protectors. The strong heuristic
850 will enable protectors for functions with:
851 - Arrays of any size and type
852 - Aggregates containing an array of any size and type.
854 - Local variables that have had their address taken.
856 This overrides the ``ssp`` function attribute.
858 If a function that has an ``sspstrong`` attribute is inlined into a
859 function that doesn't have an ``sspstrong`` attribute, then the
860 resulting function will have an ``sspstrong`` attribute.
862 This attribute indicates that the ABI being targeted requires that
863 an unwind table entry be produce for this function even if we can
864 show that no exceptions passes by it. This is normally the case for
865 the ELF x86-64 abi, but it can be disabled for some compilation
868 This attribute indicates that calls to the function cannot be
869 duplicated. A call to a ``noduplicate`` function may be moved
870 within its parent function, but may not be duplicated within
873 A function containing a ``noduplicate`` call may still
874 be an inlining candidate, provided that the call is not
875 duplicated by inlining. That implies that the function has
876 internal linkage and only has one call site, so the original
877 call is dead after inlining.
881 Module-Level Inline Assembly
882 ----------------------------
884 Modules may contain "module-level inline asm" blocks, which corresponds
885 to the GCC "file scope inline asm" blocks. These blocks are internally
886 concatenated by LLVM and treated as a single unit, but may be separated
887 in the ``.ll`` file if desired. The syntax is very simple:
891 module asm "inline asm code goes here"
892 module asm "more can go here"
894 The strings can contain any character by escaping non-printable
895 characters. The escape sequence used is simply "\\xx" where "xx" is the
896 two digit hex code for the number.
898 The inline asm code is simply printed to the machine code .s file when
899 assembly code is generated.
904 A module may specify a target specific data layout string that specifies
905 how data is to be laid out in memory. The syntax for the data layout is
910 target datalayout = "layout specification"
912 The *layout specification* consists of a list of specifications
913 separated by the minus sign character ('-'). Each specification starts
914 with a letter and may include other information after the letter to
915 define some aspect of the data layout. The specifications accepted are
919 Specifies that the target lays out data in big-endian form. That is,
920 the bits with the most significance have the lowest address
923 Specifies that the target lays out data in little-endian form. That
924 is, the bits with the least significance have the lowest address
927 Specifies the natural alignment of the stack in bits. Alignment
928 promotion of stack variables is limited to the natural stack
929 alignment to avoid dynamic stack realignment. The stack alignment
930 must be a multiple of 8-bits. If omitted, the natural stack
931 alignment defaults to "unspecified", which does not prevent any
932 alignment promotions.
933 ``p[n]:<size>:<abi>:<pref>``
934 This specifies the *size* of a pointer and its ``<abi>`` and
935 ``<pref>``\erred alignments for address space ``n``. All sizes are in
936 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
937 preceding ``:`` should be omitted too. The address space, ``n`` is
938 optional, and if not specified, denotes the default address space 0.
939 The value of ``n`` must be in the range [1,2^23).
940 ``i<size>:<abi>:<pref>``
941 This specifies the alignment for an integer type of a given bit
942 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
943 ``v<size>:<abi>:<pref>``
944 This specifies the alignment for a vector type of a given bit
946 ``f<size>:<abi>:<pref>``
947 This specifies the alignment for a floating point type of a given bit
948 ``<size>``. Only values of ``<size>`` that are supported by the target
949 will work. 32 (float) and 64 (double) are supported on all targets; 80
950 or 128 (different flavors of long double) are also supported on some
952 ``a<size>:<abi>:<pref>``
953 This specifies the alignment for an aggregate type of a given bit
955 ``s<size>:<abi>:<pref>``
956 This specifies the alignment for a stack object of a given bit
958 ``n<size1>:<size2>:<size3>...``
959 This specifies a set of native integer widths for the target CPU in
960 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
961 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
962 this set are considered to support most general arithmetic operations
965 When constructing the data layout for a given target, LLVM starts with a
966 default set of specifications which are then (possibly) overridden by
967 the specifications in the ``datalayout`` keyword. The default
968 specifications are given in this list:
971 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment
972 - ``p1:32:32:32`` - 32-bit pointers with 32-bit alignment for address
974 - ``p2:16:32:32`` - 16-bit pointers with 32-bit alignment for address
976 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
977 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
978 - ``i16:16:16`` - i16 is 16-bit aligned
979 - ``i32:32:32`` - i32 is 32-bit aligned
980 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
982 - ``f32:32:32`` - float is 32-bit aligned
983 - ``f64:64:64`` - double is 64-bit aligned
984 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
985 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
986 - ``a0:0:1`` - aggregates are 8-bit aligned
987 - ``s0:64:64`` - stack objects are 64-bit aligned
989 When LLVM is determining the alignment for a given type, it uses the
992 #. If the type sought is an exact match for one of the specifications,
993 that specification is used.
994 #. If no match is found, and the type sought is an integer type, then
995 the smallest integer type that is larger than the bitwidth of the
996 sought type is used. If none of the specifications are larger than
997 the bitwidth then the largest integer type is used. For example,
998 given the default specifications above, the i7 type will use the
999 alignment of i8 (next largest) while both i65 and i256 will use the
1000 alignment of i64 (largest specified).
1001 #. If no match is found, and the type sought is a vector type, then the
1002 largest vector type that is smaller than the sought vector type will
1003 be used as a fall back. This happens because <128 x double> can be
1004 implemented in terms of 64 <2 x double>, for example.
1006 The function of the data layout string may not be what you expect.
1007 Notably, this is not a specification from the frontend of what alignment
1008 the code generator should use.
1010 Instead, if specified, the target data layout is required to match what
1011 the ultimate *code generator* expects. This string is used by the
1012 mid-level optimizers to improve code, and this only works if it matches
1013 what the ultimate code generator uses. If you would like to generate IR
1014 that does not embed this target-specific detail into the IR, then you
1015 don't have to specify the string. This will disable some optimizations
1016 that require precise layout information, but this also prevents those
1017 optimizations from introducing target specificity into the IR.
1019 .. _pointeraliasing:
1021 Pointer Aliasing Rules
1022 ----------------------
1024 Any memory access must be done through a pointer value associated with
1025 an address range of the memory access, otherwise the behavior is
1026 undefined. Pointer values are associated with address ranges according
1027 to the following rules:
1029 - A pointer value is associated with the addresses associated with any
1030 value it is *based* on.
1031 - An address of a global variable is associated with the address range
1032 of the variable's storage.
1033 - The result value of an allocation instruction is associated with the
1034 address range of the allocated storage.
1035 - A null pointer in the default address-space is associated with no
1037 - An integer constant other than zero or a pointer value returned from
1038 a function not defined within LLVM may be associated with address
1039 ranges allocated through mechanisms other than those provided by
1040 LLVM. Such ranges shall not overlap with any ranges of addresses
1041 allocated by mechanisms provided by LLVM.
1043 A pointer value is *based* on another pointer value according to the
1046 - A pointer value formed from a ``getelementptr`` operation is *based*
1047 on the first operand of the ``getelementptr``.
1048 - The result value of a ``bitcast`` is *based* on the operand of the
1050 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1051 values that contribute (directly or indirectly) to the computation of
1052 the pointer's value.
1053 - The "*based* on" relationship is transitive.
1055 Note that this definition of *"based"* is intentionally similar to the
1056 definition of *"based"* in C99, though it is slightly weaker.
1058 LLVM IR does not associate types with memory. The result type of a
1059 ``load`` merely indicates the size and alignment of the memory from
1060 which to load, as well as the interpretation of the value. The first
1061 operand type of a ``store`` similarly only indicates the size and
1062 alignment of the store.
1064 Consequently, type-based alias analysis, aka TBAA, aka
1065 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1066 :ref:`Metadata <metadata>` may be used to encode additional information
1067 which specialized optimization passes may use to implement type-based
1072 Volatile Memory Accesses
1073 ------------------------
1075 Certain memory accesses, such as :ref:`load <i_load>`'s,
1076 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1077 marked ``volatile``. The optimizers must not change the number of
1078 volatile operations or change their order of execution relative to other
1079 volatile operations. The optimizers *may* change the order of volatile
1080 operations relative to non-volatile operations. This is not Java's
1081 "volatile" and has no cross-thread synchronization behavior.
1085 Memory Model for Concurrent Operations
1086 --------------------------------------
1088 The LLVM IR does not define any way to start parallel threads of
1089 execution or to register signal handlers. Nonetheless, there are
1090 platform-specific ways to create them, and we define LLVM IR's behavior
1091 in their presence. This model is inspired by the C++0x memory model.
1093 For a more informal introduction to this model, see the :doc:`Atomics`.
1095 We define a *happens-before* partial order as the least partial order
1098 - Is a superset of single-thread program order, and
1099 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1100 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1101 techniques, like pthread locks, thread creation, thread joining,
1102 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1103 Constraints <ordering>`).
1105 Note that program order does not introduce *happens-before* edges
1106 between a thread and signals executing inside that thread.
1108 Every (defined) read operation (load instructions, memcpy, atomic
1109 loads/read-modify-writes, etc.) R reads a series of bytes written by
1110 (defined) write operations (store instructions, atomic
1111 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1112 section, initialized globals are considered to have a write of the
1113 initializer which is atomic and happens before any other read or write
1114 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1115 may see any write to the same byte, except:
1117 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1118 write\ :sub:`2` happens before R\ :sub:`byte`, then
1119 R\ :sub:`byte` does not see write\ :sub:`1`.
1120 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1121 R\ :sub:`byte` does not see write\ :sub:`3`.
1123 Given that definition, R\ :sub:`byte` is defined as follows:
1125 - If R is volatile, the result is target-dependent. (Volatile is
1126 supposed to give guarantees which can support ``sig_atomic_t`` in
1127 C/C++, and may be used for accesses to addresses which do not behave
1128 like normal memory. It does not generally provide cross-thread
1130 - Otherwise, if there is no write to the same byte that happens before
1131 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1132 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1133 R\ :sub:`byte` returns the value written by that write.
1134 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1135 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1136 Memory Ordering Constraints <ordering>` section for additional
1137 constraints on how the choice is made.
1138 - Otherwise R\ :sub:`byte` returns ``undef``.
1140 R returns the value composed of the series of bytes it read. This
1141 implies that some bytes within the value may be ``undef`` **without**
1142 the entire value being ``undef``. Note that this only defines the
1143 semantics of the operation; it doesn't mean that targets will emit more
1144 than one instruction to read the series of bytes.
1146 Note that in cases where none of the atomic intrinsics are used, this
1147 model places only one restriction on IR transformations on top of what
1148 is required for single-threaded execution: introducing a store to a byte
1149 which might not otherwise be stored is not allowed in general.
1150 (Specifically, in the case where another thread might write to and read
1151 from an address, introducing a store can change a load that may see
1152 exactly one write into a load that may see multiple writes.)
1156 Atomic Memory Ordering Constraints
1157 ----------------------------------
1159 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1160 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1161 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1162 an ordering parameter that determines which other atomic instructions on
1163 the same address they *synchronize with*. These semantics are borrowed
1164 from Java and C++0x, but are somewhat more colloquial. If these
1165 descriptions aren't precise enough, check those specs (see spec
1166 references in the :doc:`atomics guide <Atomics>`).
1167 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1168 differently since they don't take an address. See that instruction's
1169 documentation for details.
1171 For a simpler introduction to the ordering constraints, see the
1175 The set of values that can be read is governed by the happens-before
1176 partial order. A value cannot be read unless some operation wrote
1177 it. This is intended to provide a guarantee strong enough to model
1178 Java's non-volatile shared variables. This ordering cannot be
1179 specified for read-modify-write operations; it is not strong enough
1180 to make them atomic in any interesting way.
1182 In addition to the guarantees of ``unordered``, there is a single
1183 total order for modifications by ``monotonic`` operations on each
1184 address. All modification orders must be compatible with the
1185 happens-before order. There is no guarantee that the modification
1186 orders can be combined to a global total order for the whole program
1187 (and this often will not be possible). The read in an atomic
1188 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1189 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1190 order immediately before the value it writes. If one atomic read
1191 happens before another atomic read of the same address, the later
1192 read must see the same value or a later value in the address's
1193 modification order. This disallows reordering of ``monotonic`` (or
1194 stronger) operations on the same address. If an address is written
1195 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1196 read that address repeatedly, the other threads must eventually see
1197 the write. This corresponds to the C++0x/C1x
1198 ``memory_order_relaxed``.
1200 In addition to the guarantees of ``monotonic``, a
1201 *synchronizes-with* edge may be formed with a ``release`` operation.
1202 This is intended to model C++'s ``memory_order_acquire``.
1204 In addition to the guarantees of ``monotonic``, if this operation
1205 writes a value which is subsequently read by an ``acquire``
1206 operation, it *synchronizes-with* that operation. (This isn't a
1207 complete description; see the C++0x definition of a release
1208 sequence.) This corresponds to the C++0x/C1x
1209 ``memory_order_release``.
1210 ``acq_rel`` (acquire+release)
1211 Acts as both an ``acquire`` and ``release`` operation on its
1212 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1213 ``seq_cst`` (sequentially consistent)
1214 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1215 operation which only reads, ``release`` for an operation which only
1216 writes), there is a global total order on all
1217 sequentially-consistent operations on all addresses, which is
1218 consistent with the *happens-before* partial order and with the
1219 modification orders of all the affected addresses. Each
1220 sequentially-consistent read sees the last preceding write to the
1221 same address in this global order. This corresponds to the C++0x/C1x
1222 ``memory_order_seq_cst`` and Java volatile.
1226 If an atomic operation is marked ``singlethread``, it only *synchronizes
1227 with* or participates in modification and seq\_cst total orderings with
1228 other operations running in the same thread (for example, in signal
1236 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1237 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1238 :ref:`frem <i_frem>`) have the following flags that can set to enable
1239 otherwise unsafe floating point operations
1242 No NaNs - Allow optimizations to assume the arguments and result are not
1243 NaN. Such optimizations are required to retain defined behavior over
1244 NaNs, but the value of the result is undefined.
1247 No Infs - Allow optimizations to assume the arguments and result are not
1248 +/-Inf. Such optimizations are required to retain defined behavior over
1249 +/-Inf, but the value of the result is undefined.
1252 No Signed Zeros - Allow optimizations to treat the sign of a zero
1253 argument or result as insignificant.
1256 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1257 argument rather than perform division.
1260 Fast - Allow algebraically equivalent transformations that may
1261 dramatically change results in floating point (e.g. reassociate). This
1262 flag implies all the others.
1269 The LLVM type system is one of the most important features of the
1270 intermediate representation. Being typed enables a number of
1271 optimizations to be performed on the intermediate representation
1272 directly, without having to do extra analyses on the side before the
1273 transformation. A strong type system makes it easier to read the
1274 generated code and enables novel analyses and transformations that are
1275 not feasible to perform on normal three address code representations.
1277 Type Classifications
1278 --------------------
1280 The types fall into a few useful classifications:
1289 * - :ref:`integer <t_integer>`
1290 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1293 * - :ref:`floating point <t_floating>`
1294 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1302 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1303 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1304 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1305 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1307 * - :ref:`primitive <t_primitive>`
1308 - :ref:`label <t_label>`,
1309 :ref:`void <t_void>`,
1310 :ref:`integer <t_integer>`,
1311 :ref:`floating point <t_floating>`,
1312 :ref:`x86mmx <t_x86mmx>`,
1313 :ref:`metadata <t_metadata>`.
1315 * - :ref:`derived <t_derived>`
1316 - :ref:`array <t_array>`,
1317 :ref:`function <t_function>`,
1318 :ref:`pointer <t_pointer>`,
1319 :ref:`structure <t_struct>`,
1320 :ref:`vector <t_vector>`,
1321 :ref:`opaque <t_opaque>`.
1323 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1324 Values of these types are the only ones which can be produced by
1332 The primitive types are the fundamental building blocks of the LLVM
1343 The integer type is a very simple type that simply specifies an
1344 arbitrary bit width for the integer type desired. Any bit width from 1
1345 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1354 The number of bits the integer will occupy is specified by the ``N``
1360 +----------------+------------------------------------------------+
1361 | ``i1`` | a single-bit integer. |
1362 +----------------+------------------------------------------------+
1363 | ``i32`` | a 32-bit integer. |
1364 +----------------+------------------------------------------------+
1365 | ``i1942652`` | a really big integer of over 1 million bits. |
1366 +----------------+------------------------------------------------+
1370 Floating Point Types
1371 ^^^^^^^^^^^^^^^^^^^^
1380 - 16-bit floating point value
1383 - 32-bit floating point value
1386 - 64-bit floating point value
1389 - 128-bit floating point value (112-bit mantissa)
1392 - 80-bit floating point value (X87)
1395 - 128-bit floating point value (two 64-bits)
1405 The x86mmx type represents a value held in an MMX register on an x86
1406 machine. The operations allowed on it are quite limited: parameters and
1407 return values, load and store, and bitcast. User-specified MMX
1408 instructions are represented as intrinsic or asm calls with arguments
1409 and/or results of this type. There are no arrays, vectors or constants
1427 The void type does not represent any value and has no size.
1444 The label type represents code labels.
1461 The metadata type represents embedded metadata. No derived types may be
1462 created from metadata except for :ref:`function <t_function>` arguments.
1476 The real power in LLVM comes from the derived types in the system. This
1477 is what allows a programmer to represent arrays, functions, pointers,
1478 and other useful types. Each of these types contain one or more element
1479 types which may be a primitive type, or another derived type. For
1480 example, it is possible to have a two dimensional array, using an array
1481 as the element type of another array.
1488 Aggregate Types are a subset of derived types that can contain multiple
1489 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1490 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1501 The array type is a very simple derived type that arranges elements
1502 sequentially in memory. The array type requires a size (number of
1503 elements) and an underlying data type.
1510 [<# elements> x <elementtype>]
1512 The number of elements is a constant integer value; ``elementtype`` may
1513 be any type with a size.
1518 +------------------+--------------------------------------+
1519 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1520 +------------------+--------------------------------------+
1521 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1522 +------------------+--------------------------------------+
1523 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1524 +------------------+--------------------------------------+
1526 Here are some examples of multidimensional arrays:
1528 +-----------------------------+----------------------------------------------------------+
1529 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1530 +-----------------------------+----------------------------------------------------------+
1531 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1532 +-----------------------------+----------------------------------------------------------+
1533 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1534 +-----------------------------+----------------------------------------------------------+
1536 There is no restriction on indexing beyond the end of the array implied
1537 by a static type (though there are restrictions on indexing beyond the
1538 bounds of an allocated object in some cases). This means that
1539 single-dimension 'variable sized array' addressing can be implemented in
1540 LLVM with a zero length array type. An implementation of 'pascal style
1541 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1552 The function type can be thought of as a function signature. It consists
1553 of a return type and a list of formal parameter types. The return type
1554 of a function type is a first class type or a void type.
1561 <returntype> (<parameter list>)
1563 ...where '``<parameter list>``' is a comma-separated list of type
1564 specifiers. Optionally, the parameter list may include a type ``...``,
1565 which indicates that the function takes a variable number of arguments.
1566 Variable argument functions can access their arguments with the
1567 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1568 '``<returntype>``' is any type except :ref:`label <t_label>`.
1573 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1574 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1575 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1576 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1577 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1578 | ``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. |
1579 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1580 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1581 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1591 The structure type is used to represent a collection of data members
1592 together in memory. The elements of a structure may be any type that has
1595 Structures in memory are accessed using '``load``' and '``store``' by
1596 getting a pointer to a field with the '``getelementptr``' instruction.
1597 Structures in registers are accessed using the '``extractvalue``' and
1598 '``insertvalue``' instructions.
1600 Structures may optionally be "packed" structures, which indicate that
1601 the alignment of the struct is one byte, and that there is no padding
1602 between the elements. In non-packed structs, padding between field types
1603 is inserted as defined by the DataLayout string in the module, which is
1604 required to match what the underlying code generator expects.
1606 Structures can either be "literal" or "identified". A literal structure
1607 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1608 identified types are always defined at the top level with a name.
1609 Literal types are uniqued by their contents and can never be recursive
1610 or opaque since there is no way to write one. Identified types can be
1611 recursive, can be opaqued, and are never uniqued.
1618 %T1 = type { <type list> } ; Identified normal struct type
1619 %T2 = type <{ <type list> }> ; Identified packed struct type
1624 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1625 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1626 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1627 | ``{ 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``. |
1628 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1629 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1630 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1634 Opaque Structure Types
1635 ^^^^^^^^^^^^^^^^^^^^^^
1640 Opaque structure types are used to represent named structure types that
1641 do not have a body specified. This corresponds (for example) to the C
1642 notion of a forward declared structure.
1655 +--------------+-------------------+
1656 | ``opaque`` | An opaque type. |
1657 +--------------+-------------------+
1667 The pointer type is used to specify memory locations. Pointers are
1668 commonly used to reference objects in memory.
1670 Pointer types may have an optional address space attribute defining the
1671 numbered address space where the pointed-to object resides. The default
1672 address space is number zero. The semantics of non-zero address spaces
1673 are target-specific.
1675 Note that LLVM does not permit pointers to void (``void*``) nor does it
1676 permit pointers to labels (``label*``). Use ``i8*`` instead.
1688 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1689 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1690 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1691 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1692 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1693 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1694 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1704 A vector type is a simple derived type that represents a vector of
1705 elements. Vector types are used when multiple primitive data are
1706 operated in parallel using a single instruction (SIMD). A vector type
1707 requires a size (number of elements) and an underlying primitive data
1708 type. Vector types are considered :ref:`first class <t_firstclass>`.
1715 < <# elements> x <elementtype> >
1717 The number of elements is a constant integer value larger than 0;
1718 elementtype may be any integer or floating point type, or a pointer to
1719 these types. Vectors of size zero are not allowed.
1724 +-------------------+--------------------------------------------------+
1725 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1726 +-------------------+--------------------------------------------------+
1727 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1728 +-------------------+--------------------------------------------------+
1729 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1730 +-------------------+--------------------------------------------------+
1731 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1732 +-------------------+--------------------------------------------------+
1737 LLVM has several different basic types of constants. This section
1738 describes them all and their syntax.
1743 **Boolean constants**
1744 The two strings '``true``' and '``false``' are both valid constants
1746 **Integer constants**
1747 Standard integers (such as '4') are constants of the
1748 :ref:`integer <t_integer>` type. Negative numbers may be used with
1750 **Floating point constants**
1751 Floating point constants use standard decimal notation (e.g.
1752 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1753 hexadecimal notation (see below). The assembler requires the exact
1754 decimal value of a floating-point constant. For example, the
1755 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1756 decimal in binary. Floating point constants must have a :ref:`floating
1757 point <t_floating>` type.
1758 **Null pointer constants**
1759 The identifier '``null``' is recognized as a null pointer constant
1760 and must be of :ref:`pointer type <t_pointer>`.
1762 The one non-intuitive notation for constants is the hexadecimal form of
1763 floating point constants. For example, the form
1764 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1765 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1766 constants are required (and the only time that they are generated by the
1767 disassembler) is when a floating point constant must be emitted but it
1768 cannot be represented as a decimal floating point number in a reasonable
1769 number of digits. For example, NaN's, infinities, and other special
1770 values are represented in their IEEE hexadecimal format so that assembly
1771 and disassembly do not cause any bits to change in the constants.
1773 When using the hexadecimal form, constants of types half, float, and
1774 double are represented using the 16-digit form shown above (which
1775 matches the IEEE754 representation for double); half and float values
1776 must, however, be exactly representable as IEEE 754 half and single
1777 precision, respectively. Hexadecimal format is always used for long
1778 double, and there are three forms of long double. The 80-bit format used
1779 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1780 128-bit format used by PowerPC (two adjacent doubles) is represented by
1781 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1782 represented by ``0xL`` followed by 32 hexadecimal digits; no currently
1783 supported target uses this format. Long doubles will only work if they
1784 match the long double format on your target. The IEEE 16-bit format
1785 (half precision) is represented by ``0xH`` followed by 4 hexadecimal
1786 digits. All hexadecimal formats are big-endian (sign bit at the left).
1788 There are no constants of type x86mmx.
1793 Complex constants are a (potentially recursive) combination of simple
1794 constants and smaller complex constants.
1796 **Structure constants**
1797 Structure constants are represented with notation similar to
1798 structure type definitions (a comma separated list of elements,
1799 surrounded by braces (``{}``)). For example:
1800 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1801 "``@G = external global i32``". Structure constants must have
1802 :ref:`structure type <t_struct>`, and the number and types of elements
1803 must match those specified by the type.
1805 Array constants are represented with notation similar to array type
1806 definitions (a comma separated list of elements, surrounded by
1807 square brackets (``[]``)). For example:
1808 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1809 :ref:`array type <t_array>`, and the number and types of elements must
1810 match those specified by the type.
1811 **Vector constants**
1812 Vector constants are represented with notation similar to vector
1813 type definitions (a comma separated list of elements, surrounded by
1814 less-than/greater-than's (``<>``)). For example:
1815 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1816 must have :ref:`vector type <t_vector>`, and the number and types of
1817 elements must match those specified by the type.
1818 **Zero initialization**
1819 The string '``zeroinitializer``' can be used to zero initialize a
1820 value to zero of *any* type, including scalar and
1821 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1822 having to print large zero initializers (e.g. for large arrays) and
1823 is always exactly equivalent to using explicit zero initializers.
1825 A metadata node is a structure-like constant with :ref:`metadata
1826 type <t_metadata>`. For example:
1827 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
1828 constants that are meant to be interpreted as part of the
1829 instruction stream, metadata is a place to attach additional
1830 information such as debug info.
1832 Global Variable and Function Addresses
1833 --------------------------------------
1835 The addresses of :ref:`global variables <globalvars>` and
1836 :ref:`functions <functionstructure>` are always implicitly valid
1837 (link-time) constants. These constants are explicitly referenced when
1838 the :ref:`identifier for the global <identifiers>` is used and always have
1839 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
1842 .. code-block:: llvm
1846 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1853 The string '``undef``' can be used anywhere a constant is expected, and
1854 indicates that the user of the value may receive an unspecified
1855 bit-pattern. Undefined values may be of any type (other than '``label``'
1856 or '``void``') and be used anywhere a constant is permitted.
1858 Undefined values are useful because they indicate to the compiler that
1859 the program is well defined no matter what value is used. This gives the
1860 compiler more freedom to optimize. Here are some examples of
1861 (potentially surprising) transformations that are valid (in pseudo IR):
1863 .. code-block:: llvm
1873 This is safe because all of the output bits are affected by the undef
1874 bits. Any output bit can have a zero or one depending on the input bits.
1876 .. code-block:: llvm
1887 These logical operations have bits that are not always affected by the
1888 input. For example, if ``%X`` has a zero bit, then the output of the
1889 '``and``' operation will always be a zero for that bit, no matter what
1890 the corresponding bit from the '``undef``' is. As such, it is unsafe to
1891 optimize or assume that the result of the '``and``' is '``undef``'.
1892 However, it is safe to assume that all bits of the '``undef``' could be
1893 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
1894 all the bits of the '``undef``' operand to the '``or``' could be set,
1895 allowing the '``or``' to be folded to -1.
1897 .. code-block:: llvm
1899 %A = select undef, %X, %Y
1900 %B = select undef, 42, %Y
1901 %C = select %X, %Y, undef
1911 This set of examples shows that undefined '``select``' (and conditional
1912 branch) conditions can go *either way*, but they have to come from one
1913 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
1914 both known to have a clear low bit, then ``%A`` would have to have a
1915 cleared low bit. However, in the ``%C`` example, the optimizer is
1916 allowed to assume that the '``undef``' operand could be the same as
1917 ``%Y``, allowing the whole '``select``' to be eliminated.
1919 .. code-block:: llvm
1921 %A = xor undef, undef
1938 This example points out that two '``undef``' operands are not
1939 necessarily the same. This can be surprising to people (and also matches
1940 C semantics) where they assume that "``X^X``" is always zero, even if
1941 ``X`` is undefined. This isn't true for a number of reasons, but the
1942 short answer is that an '``undef``' "variable" can arbitrarily change
1943 its value over its "live range". This is true because the variable
1944 doesn't actually *have a live range*. Instead, the value is logically
1945 read from arbitrary registers that happen to be around when needed, so
1946 the value is not necessarily consistent over time. In fact, ``%A`` and
1947 ``%C`` need to have the same semantics or the core LLVM "replace all
1948 uses with" concept would not hold.
1950 .. code-block:: llvm
1958 These examples show the crucial difference between an *undefined value*
1959 and *undefined behavior*. An undefined value (like '``undef``') is
1960 allowed to have an arbitrary bit-pattern. This means that the ``%A``
1961 operation can be constant folded to '``undef``', because the '``undef``'
1962 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
1963 However, in the second example, we can make a more aggressive
1964 assumption: because the ``undef`` is allowed to be an arbitrary value,
1965 we are allowed to assume that it could be zero. Since a divide by zero
1966 has *undefined behavior*, we are allowed to assume that the operation
1967 does not execute at all. This allows us to delete the divide and all
1968 code after it. Because the undefined operation "can't happen", the
1969 optimizer can assume that it occurs in dead code.
1971 .. code-block:: llvm
1973 a: store undef -> %X
1974 b: store %X -> undef
1979 These examples reiterate the ``fdiv`` example: a store *of* an undefined
1980 value can be assumed to not have any effect; we can assume that the
1981 value is overwritten with bits that happen to match what was already
1982 there. However, a store *to* an undefined location could clobber
1983 arbitrary memory, therefore, it has undefined behavior.
1990 Poison values are similar to :ref:`undef values <undefvalues>`, however
1991 they also represent the fact that an instruction or constant expression
1992 which cannot evoke side effects has nevertheless detected a condition
1993 which results in undefined behavior.
1995 There is currently no way of representing a poison value in the IR; they
1996 only exist when produced by operations such as :ref:`add <i_add>` with
1999 Poison value behavior is defined in terms of value *dependence*:
2001 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2002 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2003 their dynamic predecessor basic block.
2004 - Function arguments depend on the corresponding actual argument values
2005 in the dynamic callers of their functions.
2006 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2007 instructions that dynamically transfer control back to them.
2008 - :ref:`Invoke <i_invoke>` instructions depend on the
2009 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2010 call instructions that dynamically transfer control back to them.
2011 - Non-volatile loads and stores depend on the most recent stores to all
2012 of the referenced memory addresses, following the order in the IR
2013 (including loads and stores implied by intrinsics such as
2014 :ref:`@llvm.memcpy <int_memcpy>`.)
2015 - An instruction with externally visible side effects depends on the
2016 most recent preceding instruction with externally visible side
2017 effects, following the order in the IR. (This includes :ref:`volatile
2018 operations <volatile>`.)
2019 - An instruction *control-depends* on a :ref:`terminator
2020 instruction <terminators>` if the terminator instruction has
2021 multiple successors and the instruction is always executed when
2022 control transfers to one of the successors, and may not be executed
2023 when control is transferred to another.
2024 - Additionally, an instruction also *control-depends* on a terminator
2025 instruction if the set of instructions it otherwise depends on would
2026 be different if the terminator had transferred control to a different
2028 - Dependence is transitive.
2030 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2031 with the additional affect that any instruction which has a *dependence*
2032 on a poison value has undefined behavior.
2034 Here are some examples:
2036 .. code-block:: llvm
2039 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2040 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2041 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2042 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2044 store i32 %poison, i32* @g ; Poison value stored to memory.
2045 %poison2 = load i32* @g ; Poison value loaded back from memory.
2047 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2049 %narrowaddr = bitcast i32* @g to i16*
2050 %wideaddr = bitcast i32* @g to i64*
2051 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2052 %poison4 = load i64* %wideaddr ; Returns a poison value.
2054 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2055 br i1 %cmp, label %true, label %end ; Branch to either destination.
2058 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2059 ; it has undefined behavior.
2063 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2064 ; Both edges into this PHI are
2065 ; control-dependent on %cmp, so this
2066 ; always results in a poison value.
2068 store volatile i32 0, i32* @g ; This would depend on the store in %true
2069 ; if %cmp is true, or the store in %entry
2070 ; otherwise, so this is undefined behavior.
2072 br i1 %cmp, label %second_true, label %second_end
2073 ; The same branch again, but this time the
2074 ; true block doesn't have side effects.
2081 store volatile i32 0, i32* @g ; This time, the instruction always depends
2082 ; on the store in %end. Also, it is
2083 ; control-equivalent to %end, so this is
2084 ; well-defined (ignoring earlier undefined
2085 ; behavior in this example).
2089 Addresses of Basic Blocks
2090 -------------------------
2092 ``blockaddress(@function, %block)``
2094 The '``blockaddress``' constant computes the address of the specified
2095 basic block in the specified function, and always has an ``i8*`` type.
2096 Taking the address of the entry block is illegal.
2098 This value only has defined behavior when used as an operand to the
2099 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2100 against null. Pointer equality tests between labels addresses results in
2101 undefined behavior --- though, again, comparison against null is ok, and
2102 no label is equal to the null pointer. This may be passed around as an
2103 opaque pointer sized value as long as the bits are not inspected. This
2104 allows ``ptrtoint`` and arithmetic to be performed on these values so
2105 long as the original value is reconstituted before the ``indirectbr``
2108 Finally, some targets may provide defined semantics when using the value
2109 as the operand to an inline assembly, but that is target specific.
2111 Constant Expressions
2112 --------------------
2114 Constant expressions are used to allow expressions involving other
2115 constants to be used as constants. Constant expressions may be of any
2116 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2117 that does not have side effects (e.g. load and call are not supported).
2118 The following is the syntax for constant expressions:
2120 ``trunc (CST to TYPE)``
2121 Truncate a constant to another type. The bit size of CST must be
2122 larger than the bit size of TYPE. Both types must be integers.
2123 ``zext (CST to TYPE)``
2124 Zero extend a constant to another type. The bit size of CST must be
2125 smaller than the bit size of TYPE. Both types must be integers.
2126 ``sext (CST to TYPE)``
2127 Sign extend a constant to another type. The bit size of CST must be
2128 smaller than the bit size of TYPE. Both types must be integers.
2129 ``fptrunc (CST to TYPE)``
2130 Truncate a floating point constant to another floating point type.
2131 The size of CST must be larger than the size of TYPE. Both types
2132 must be floating point.
2133 ``fpext (CST to TYPE)``
2134 Floating point extend a constant to another type. The size of CST
2135 must be smaller or equal to the size of TYPE. Both types must be
2137 ``fptoui (CST to TYPE)``
2138 Convert a floating point constant to the corresponding unsigned
2139 integer constant. TYPE must be a scalar or vector integer type. CST
2140 must be of scalar or vector floating point type. Both CST and TYPE
2141 must be scalars, or vectors of the same number of elements. If the
2142 value won't fit in the integer type, the results are undefined.
2143 ``fptosi (CST to TYPE)``
2144 Convert a floating point constant to the corresponding signed
2145 integer constant. TYPE must be a scalar or vector integer type. CST
2146 must be of scalar or vector floating point type. Both CST and TYPE
2147 must be scalars, or vectors of the same number of elements. If the
2148 value won't fit in the integer type, the results are undefined.
2149 ``uitofp (CST to TYPE)``
2150 Convert an unsigned integer constant to the corresponding floating
2151 point constant. TYPE must be a scalar or vector floating point type.
2152 CST must be of scalar or vector integer type. Both CST and TYPE must
2153 be scalars, or vectors of the same number of elements. If the value
2154 won't fit in the floating point type, the results are undefined.
2155 ``sitofp (CST to TYPE)``
2156 Convert a signed integer constant to the corresponding floating
2157 point constant. TYPE must be a scalar or vector floating point type.
2158 CST must be of scalar or vector integer type. Both CST and TYPE must
2159 be scalars, or vectors of the same number of elements. If the value
2160 won't fit in the floating point type, the results are undefined.
2161 ``ptrtoint (CST to TYPE)``
2162 Convert a pointer typed constant to the corresponding integer
2163 constant ``TYPE`` must be an integer type. ``CST`` must be of
2164 pointer type. The ``CST`` value is zero extended, truncated, or
2165 unchanged to make it fit in ``TYPE``.
2166 ``inttoptr (CST to TYPE)``
2167 Convert an integer constant to a pointer constant. TYPE must be a
2168 pointer type. CST must be of integer type. The CST value is zero
2169 extended, truncated, or unchanged to make it fit in a pointer size.
2170 This one is *really* dangerous!
2171 ``bitcast (CST to TYPE)``
2172 Convert a constant, CST, to another TYPE. The constraints of the
2173 operands are the same as those for the :ref:`bitcast
2174 instruction <i_bitcast>`.
2175 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2176 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2177 constants. As with the :ref:`getelementptr <i_getelementptr>`
2178 instruction, the index list may have zero or more indexes, which are
2179 required to make sense for the type of "CSTPTR".
2180 ``select (COND, VAL1, VAL2)``
2181 Perform the :ref:`select operation <i_select>` on constants.
2182 ``icmp COND (VAL1, VAL2)``
2183 Performs the :ref:`icmp operation <i_icmp>` on constants.
2184 ``fcmp COND (VAL1, VAL2)``
2185 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2186 ``extractelement (VAL, IDX)``
2187 Perform the :ref:`extractelement operation <i_extractelement>` on
2189 ``insertelement (VAL, ELT, IDX)``
2190 Perform the :ref:`insertelement operation <i_insertelement>` on
2192 ``shufflevector (VEC1, VEC2, IDXMASK)``
2193 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2195 ``extractvalue (VAL, IDX0, IDX1, ...)``
2196 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2197 constants. The index list is interpreted in a similar manner as
2198 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2199 least one index value must be specified.
2200 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2201 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2202 The index list is interpreted in a similar manner as indices in a
2203 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2204 value must be specified.
2205 ``OPCODE (LHS, RHS)``
2206 Perform the specified operation of the LHS and RHS constants. OPCODE
2207 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2208 binary <bitwiseops>` operations. The constraints on operands are
2209 the same as those for the corresponding instruction (e.g. no bitwise
2210 operations on floating point values are allowed).
2215 Inline Assembler Expressions
2216 ----------------------------
2218 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2219 Inline Assembly <moduleasm>`) through the use of a special value. This
2220 value represents the inline assembler as a string (containing the
2221 instructions to emit), a list of operand constraints (stored as a
2222 string), a flag that indicates whether or not the inline asm expression
2223 has side effects, and a flag indicating whether the function containing
2224 the asm needs to align its stack conservatively. An example inline
2225 assembler expression is:
2227 .. code-block:: llvm
2229 i32 (i32) asm "bswap $0", "=r,r"
2231 Inline assembler expressions may **only** be used as the callee operand
2232 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2233 Thus, typically we have:
2235 .. code-block:: llvm
2237 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2239 Inline asms with side effects not visible in the constraint list must be
2240 marked as having side effects. This is done through the use of the
2241 '``sideeffect``' keyword, like so:
2243 .. code-block:: llvm
2245 call void asm sideeffect "eieio", ""()
2247 In some cases inline asms will contain code that will not work unless
2248 the stack is aligned in some way, such as calls or SSE instructions on
2249 x86, yet will not contain code that does that alignment within the asm.
2250 The compiler should make conservative assumptions about what the asm
2251 might contain and should generate its usual stack alignment code in the
2252 prologue if the '``alignstack``' keyword is present:
2254 .. code-block:: llvm
2256 call void asm alignstack "eieio", ""()
2258 Inline asms also support using non-standard assembly dialects. The
2259 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2260 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2261 the only supported dialects. An example is:
2263 .. code-block:: llvm
2265 call void asm inteldialect "eieio", ""()
2267 If multiple keywords appear the '``sideeffect``' keyword must come
2268 first, the '``alignstack``' keyword second and the '``inteldialect``'
2274 The call instructions that wrap inline asm nodes may have a
2275 "``!srcloc``" MDNode attached to it that contains a list of constant
2276 integers. If present, the code generator will use the integer as the
2277 location cookie value when report errors through the ``LLVMContext``
2278 error reporting mechanisms. This allows a front-end to correlate backend
2279 errors that occur with inline asm back to the source code that produced
2282 .. code-block:: llvm
2284 call void asm sideeffect "something bad", ""(), !srcloc !42
2286 !42 = !{ i32 1234567 }
2288 It is up to the front-end to make sense of the magic numbers it places
2289 in the IR. If the MDNode contains multiple constants, the code generator
2290 will use the one that corresponds to the line of the asm that the error
2295 Metadata Nodes and Metadata Strings
2296 -----------------------------------
2298 LLVM IR allows metadata to be attached to instructions in the program
2299 that can convey extra information about the code to the optimizers and
2300 code generator. One example application of metadata is source-level
2301 debug information. There are two metadata primitives: strings and nodes.
2302 All metadata has the ``metadata`` type and is identified in syntax by a
2303 preceding exclamation point ('``!``').
2305 A metadata string is a string surrounded by double quotes. It can
2306 contain any character by escaping non-printable characters with
2307 "``\xx``" where "``xx``" is the two digit hex code. For example:
2310 Metadata nodes are represented with notation similar to structure
2311 constants (a comma separated list of elements, surrounded by braces and
2312 preceded by an exclamation point). Metadata nodes can have any values as
2313 their operand. For example:
2315 .. code-block:: llvm
2317 !{ metadata !"test\00", i32 10}
2319 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2320 metadata nodes, which can be looked up in the module symbol table. For
2323 .. code-block:: llvm
2325 !foo = metadata !{!4, !3}
2327 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2328 function is using two metadata arguments:
2330 .. code-block:: llvm
2332 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2334 Metadata can be attached with an instruction. Here metadata ``!21`` is
2335 attached to the ``add`` instruction using the ``!dbg`` identifier:
2337 .. code-block:: llvm
2339 %indvar.next = add i64 %indvar, 1, !dbg !21
2341 More information about specific metadata nodes recognized by the
2342 optimizers and code generator is found below.
2347 In LLVM IR, memory does not have types, so LLVM's own type system is not
2348 suitable for doing TBAA. Instead, metadata is added to the IR to
2349 describe a type system of a higher level language. This can be used to
2350 implement typical C/C++ TBAA, but it can also be used to implement
2351 custom alias analysis behavior for other languages.
2353 The current metadata format is very simple. TBAA metadata nodes have up
2354 to three fields, e.g.:
2356 .. code-block:: llvm
2358 !0 = metadata !{ metadata !"an example type tree" }
2359 !1 = metadata !{ metadata !"int", metadata !0 }
2360 !2 = metadata !{ metadata !"float", metadata !0 }
2361 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2363 The first field is an identity field. It can be any value, usually a
2364 metadata string, which uniquely identifies the type. The most important
2365 name in the tree is the name of the root node. Two trees with different
2366 root node names are entirely disjoint, even if they have leaves with
2369 The second field identifies the type's parent node in the tree, or is
2370 null or omitted for a root node. A type is considered to alias all of
2371 its descendants and all of its ancestors in the tree. Also, a type is
2372 considered to alias all types in other trees, so that bitcode produced
2373 from multiple front-ends is handled conservatively.
2375 If the third field is present, it's an integer which if equal to 1
2376 indicates that the type is "constant" (meaning
2377 ``pointsToConstantMemory`` should return true; see `other useful
2378 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2380 '``tbaa.struct``' Metadata
2381 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2383 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2384 aggregate assignment operations in C and similar languages, however it
2385 is defined to copy a contiguous region of memory, which is more than
2386 strictly necessary for aggregate types which contain holes due to
2387 padding. Also, it doesn't contain any TBAA information about the fields
2390 ``!tbaa.struct`` metadata can describe which memory subregions in a
2391 memcpy are padding and what the TBAA tags of the struct are.
2393 The current metadata format is very simple. ``!tbaa.struct`` metadata
2394 nodes are a list of operands which are in conceptual groups of three.
2395 For each group of three, the first operand gives the byte offset of a
2396 field in bytes, the second gives its size in bytes, and the third gives
2399 .. code-block:: llvm
2401 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2403 This describes a struct with two fields. The first is at offset 0 bytes
2404 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2405 and has size 4 bytes and has tbaa tag !2.
2407 Note that the fields need not be contiguous. In this example, there is a
2408 4 byte gap between the two fields. This gap represents padding which
2409 does not carry useful data and need not be preserved.
2411 '``fpmath``' Metadata
2412 ^^^^^^^^^^^^^^^^^^^^^
2414 ``fpmath`` metadata may be attached to any instruction of floating point
2415 type. It can be used to express the maximum acceptable error in the
2416 result of that instruction, in ULPs, thus potentially allowing the
2417 compiler to use a more efficient but less accurate method of computing
2418 it. ULP is defined as follows:
2420 If ``x`` is a real number that lies between two finite consecutive
2421 floating-point numbers ``a`` and ``b``, without being equal to one
2422 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2423 distance between the two non-equal finite floating-point numbers
2424 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2426 The metadata node shall consist of a single positive floating point
2427 number representing the maximum relative error, for example:
2429 .. code-block:: llvm
2431 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2433 '``range``' Metadata
2434 ^^^^^^^^^^^^^^^^^^^^
2436 ``range`` metadata may be attached only to loads of integer types. It
2437 expresses the possible ranges the loaded value is in. The ranges are
2438 represented with a flattened list of integers. The loaded value is known
2439 to be in the union of the ranges defined by each consecutive pair. Each
2440 pair has the following properties:
2442 - The type must match the type loaded by the instruction.
2443 - The pair ``a,b`` represents the range ``[a,b)``.
2444 - Both ``a`` and ``b`` are constants.
2445 - The range is allowed to wrap.
2446 - The range should not represent the full or empty set. That is,
2449 In addition, the pairs must be in signed order of the lower bound and
2450 they must be non-contiguous.
2454 .. code-block:: llvm
2456 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2457 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2458 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2459 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2461 !0 = metadata !{ i8 0, i8 2 }
2462 !1 = metadata !{ i8 255, i8 2 }
2463 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2464 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2466 Module Flags Metadata
2467 =====================
2469 Information about the module as a whole is difficult to convey to LLVM's
2470 subsystems. The LLVM IR isn't sufficient to transmit this information.
2471 The ``llvm.module.flags`` named metadata exists in order to facilitate
2472 this. These flags are in the form of key / value pairs --- much like a
2473 dictionary --- making it easy for any subsystem who cares about a flag to
2476 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2477 Each triplet has the following form:
2479 - The first element is a *behavior* flag, which specifies the behavior
2480 when two (or more) modules are merged together, and it encounters two
2481 (or more) metadata with the same ID. The supported behaviors are
2483 - The second element is a metadata string that is a unique ID for the
2484 metadata. Each module may only have one flag entry for each unique ID (not
2485 including entries with the **Require** behavior).
2486 - The third element is the value of the flag.
2488 When two (or more) modules are merged together, the resulting
2489 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2490 each unique metadata ID string, there will be exactly one entry in the merged
2491 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2492 be determined by the merge behavior flag, as described below. The only exception
2493 is that entries with the *Require* behavior are always preserved.
2495 The following behaviors are supported:
2506 Emits an error if two values disagree, otherwise the resulting value
2507 is that of the operands.
2511 Emits a warning if two values disagree. The result value will be the
2512 operand for the flag from the first module being linked.
2516 Adds a requirement that another module flag be present and have a
2517 specified value after linking is performed. The value must be a
2518 metadata pair, where the first element of the pair is the ID of the
2519 module flag to be restricted, and the second element of the pair is
2520 the value the module flag should be restricted to. This behavior can
2521 be used to restrict the allowable results (via triggering of an
2522 error) of linking IDs with the **Override** behavior.
2526 Uses the specified value, regardless of the behavior or value of the
2527 other module. If both modules specify **Override**, but the values
2528 differ, an error will be emitted.
2532 Appends the two values, which are required to be metadata nodes.
2536 Appends the two values, which are required to be metadata
2537 nodes. However, duplicate entries in the second list are dropped
2538 during the append operation.
2540 It is an error for a particular unique flag ID to have multiple behaviors,
2541 except in the case of **Require** (which adds restrictions on another metadata
2542 value) or **Override**.
2544 An example of module flags:
2546 .. code-block:: llvm
2548 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2549 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2550 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2551 !3 = metadata !{ i32 3, metadata !"qux",
2553 metadata !"foo", i32 1
2556 !llvm.module.flags = !{ !0, !1, !2, !3 }
2558 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2559 if two or more ``!"foo"`` flags are seen is to emit an error if their
2560 values are not equal.
2562 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2563 behavior if two or more ``!"bar"`` flags are seen is to use the value
2566 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2567 behavior if two or more ``!"qux"`` flags are seen is to emit a
2568 warning if their values are not equal.
2570 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2574 metadata !{ metadata !"foo", i32 1 }
2576 The behavior is to emit an error if the ``llvm.module.flags`` does not
2577 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2580 Objective-C Garbage Collection Module Flags Metadata
2581 ----------------------------------------------------
2583 On the Mach-O platform, Objective-C stores metadata about garbage
2584 collection in a special section called "image info". The metadata
2585 consists of a version number and a bitmask specifying what types of
2586 garbage collection are supported (if any) by the file. If two or more
2587 modules are linked together their garbage collection metadata needs to
2588 be merged rather than appended together.
2590 The Objective-C garbage collection module flags metadata consists of the
2591 following key-value pairs:
2600 * - ``Objective-C Version``
2601 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2603 * - ``Objective-C Image Info Version``
2604 - **[Required]** --- The version of the image info section. Currently
2607 * - ``Objective-C Image Info Section``
2608 - **[Required]** --- The section to place the metadata. Valid values are
2609 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2610 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2611 Objective-C ABI version 2.
2613 * - ``Objective-C Garbage Collection``
2614 - **[Required]** --- Specifies whether garbage collection is supported or
2615 not. Valid values are 0, for no garbage collection, and 2, for garbage
2616 collection supported.
2618 * - ``Objective-C GC Only``
2619 - **[Optional]** --- Specifies that only garbage collection is supported.
2620 If present, its value must be 6. This flag requires that the
2621 ``Objective-C Garbage Collection`` flag have the value 2.
2623 Some important flag interactions:
2625 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2626 merged with a module with ``Objective-C Garbage Collection`` set to
2627 2, then the resulting module has the
2628 ``Objective-C Garbage Collection`` flag set to 0.
2629 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2630 merged with a module with ``Objective-C GC Only`` set to 6.
2632 Automatic Linker Flags Module Flags Metadata
2633 --------------------------------------------
2635 Some targets support embedding flags to the linker inside individual object
2636 files. Typically this is used in conjunction with language extensions which
2637 allow source files to explicitly declare the libraries they depend on, and have
2638 these automatically be transmitted to the linker via object files.
2640 These flags are encoded in the IR using metadata in the module flags section,
2641 using the ``Linker Options`` key. The merge behavior for this flag is required
2642 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2643 node which should be a list of other metadata nodes, each of which should be a
2644 list of metadata strings defining linker options.
2646 For example, the following metadata section specifies two separate sets of
2647 linker options, presumably to link against ``libz`` and the ``Cocoa``
2650 !0 = metadata !{ i32 6, metadata !"Linker Options",
2652 metadata !{ metadata !"-lz" },
2653 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2654 !llvm.module.flags = !{ !0 }
2656 The metadata encoding as lists of lists of options, as opposed to a collapsed
2657 list of options, is chosen so that the IR encoding can use multiple option
2658 strings to specify e.g., a single library, while still having that specifier be
2659 preserved as an atomic element that can be recognized by a target specific
2660 assembly writer or object file emitter.
2662 Each individual option is required to be either a valid option for the target's
2663 linker, or an option that is reserved by the target specific assembly writer or
2664 object file emitter. No other aspect of these options is defined by the IR.
2666 Intrinsic Global Variables
2667 ==========================
2669 LLVM has a number of "magic" global variables that contain data that
2670 affect code generation or other IR semantics. These are documented here.
2671 All globals of this sort should have a section specified as
2672 "``llvm.metadata``". This section and all globals that start with
2673 "``llvm.``" are reserved for use by LLVM.
2675 The '``llvm.used``' Global Variable
2676 -----------------------------------
2678 The ``@llvm.used`` global is an array with i8\* element type which has
2679 :ref:`appending linkage <linkage_appending>`. This array contains a list of
2680 pointers to global variables and functions which may optionally have a
2681 pointer cast formed of bitcast or getelementptr. For example, a legal
2684 .. code-block:: llvm
2689 @llvm.used = appending global [2 x i8*] [
2691 i8* bitcast (i32* @Y to i8*)
2692 ], section "llvm.metadata"
2694 If a global variable appears in the ``@llvm.used`` list, then the
2695 compiler, assembler, and linker are required to treat the symbol as if
2696 there is a reference to the global that it cannot see. For example, if a
2697 variable has internal linkage and no references other than that from the
2698 ``@llvm.used`` list, it cannot be deleted. This is commonly used to
2699 represent references from inline asms and other things the compiler
2700 cannot "see", and corresponds to "``attribute((used))``" in GNU C.
2702 On some targets, the code generator must emit a directive to the
2703 assembler or object file to prevent the assembler and linker from
2704 molesting the symbol.
2706 The '``llvm.compiler.used``' Global Variable
2707 --------------------------------------------
2709 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
2710 directive, except that it only prevents the compiler from touching the
2711 symbol. On targets that support it, this allows an intelligent linker to
2712 optimize references to the symbol without being impeded as it would be
2715 This is a rare construct that should only be used in rare circumstances,
2716 and should not be exposed to source languages.
2718 The '``llvm.global_ctors``' Global Variable
2719 -------------------------------------------
2721 .. code-block:: llvm
2723 %0 = type { i32, void ()* }
2724 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2726 The ``@llvm.global_ctors`` array contains a list of constructor
2727 functions and associated priorities. The functions referenced by this
2728 array will be called in ascending order of priority (i.e. lowest first)
2729 when the module is loaded. The order of functions with the same priority
2732 The '``llvm.global_dtors``' Global Variable
2733 -------------------------------------------
2735 .. code-block:: llvm
2737 %0 = type { i32, void ()* }
2738 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
2740 The ``@llvm.global_dtors`` array contains a list of destructor functions
2741 and associated priorities. The functions referenced by this array will
2742 be called in descending order of priority (i.e. highest first) when the
2743 module is loaded. The order of functions with the same priority is not
2746 Instruction Reference
2747 =====================
2749 The LLVM instruction set consists of several different classifications
2750 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
2751 instructions <binaryops>`, :ref:`bitwise binary
2752 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
2753 :ref:`other instructions <otherops>`.
2757 Terminator Instructions
2758 -----------------------
2760 As mentioned :ref:`previously <functionstructure>`, every basic block in a
2761 program ends with a "Terminator" instruction, which indicates which
2762 block should be executed after the current block is finished. These
2763 terminator instructions typically yield a '``void``' value: they produce
2764 control flow, not values (the one exception being the
2765 ':ref:`invoke <i_invoke>`' instruction).
2767 The terminator instructions are: ':ref:`ret <i_ret>`',
2768 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
2769 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
2770 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
2774 '``ret``' Instruction
2775 ^^^^^^^^^^^^^^^^^^^^^
2782 ret <type> <value> ; Return a value from a non-void function
2783 ret void ; Return from void function
2788 The '``ret``' instruction is used to return control flow (and optionally
2789 a value) from a function back to the caller.
2791 There are two forms of the '``ret``' instruction: one that returns a
2792 value and then causes control flow, and one that just causes control
2798 The '``ret``' instruction optionally accepts a single argument, the
2799 return value. The type of the return value must be a ':ref:`first
2800 class <t_firstclass>`' type.
2802 A function is not :ref:`well formed <wellformed>` if it it has a non-void
2803 return type and contains a '``ret``' instruction with no return value or
2804 a return value with a type that does not match its type, or if it has a
2805 void return type and contains a '``ret``' instruction with a return
2811 When the '``ret``' instruction is executed, control flow returns back to
2812 the calling function's context. If the caller is a
2813 ":ref:`call <i_call>`" instruction, execution continues at the
2814 instruction after the call. If the caller was an
2815 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
2816 beginning of the "normal" destination block. If the instruction returns
2817 a value, that value shall set the call or invoke instruction's return
2823 .. code-block:: llvm
2825 ret i32 5 ; Return an integer value of 5
2826 ret void ; Return from a void function
2827 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
2831 '``br``' Instruction
2832 ^^^^^^^^^^^^^^^^^^^^
2839 br i1 <cond>, label <iftrue>, label <iffalse>
2840 br label <dest> ; Unconditional branch
2845 The '``br``' instruction is used to cause control flow to transfer to a
2846 different basic block in the current function. There are two forms of
2847 this instruction, corresponding to a conditional branch and an
2848 unconditional branch.
2853 The conditional branch form of the '``br``' instruction takes a single
2854 '``i1``' value and two '``label``' values. The unconditional form of the
2855 '``br``' instruction takes a single '``label``' value as a target.
2860 Upon execution of a conditional '``br``' instruction, the '``i1``'
2861 argument is evaluated. If the value is ``true``, control flows to the
2862 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
2863 to the '``iffalse``' ``label`` argument.
2868 .. code-block:: llvm
2871 %cond = icmp eq i32 %a, %b
2872 br i1 %cond, label %IfEqual, label %IfUnequal
2880 '``switch``' Instruction
2881 ^^^^^^^^^^^^^^^^^^^^^^^^
2888 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2893 The '``switch``' instruction is used to transfer control flow to one of
2894 several different places. It is a generalization of the '``br``'
2895 instruction, allowing a branch to occur to one of many possible
2901 The '``switch``' instruction uses three parameters: an integer
2902 comparison value '``value``', a default '``label``' destination, and an
2903 array of pairs of comparison value constants and '``label``'s. The table
2904 is not allowed to contain duplicate constant entries.
2909 The ``switch`` instruction specifies a table of values and destinations.
2910 When the '``switch``' instruction is executed, this table is searched
2911 for the given value. If the value is found, control flow is transferred
2912 to the corresponding destination; otherwise, control flow is transferred
2913 to the default destination.
2918 Depending on properties of the target machine and the particular
2919 ``switch`` instruction, this instruction may be code generated in
2920 different ways. For example, it could be generated as a series of
2921 chained conditional branches or with a lookup table.
2926 .. code-block:: llvm
2928 ; Emulate a conditional br instruction
2929 %Val = zext i1 %value to i32
2930 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2932 ; Emulate an unconditional br instruction
2933 switch i32 0, label %dest [ ]
2935 ; Implement a jump table:
2936 switch i32 %val, label %otherwise [ i32 0, label %onzero
2938 i32 2, label %ontwo ]
2942 '``indirectbr``' Instruction
2943 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2950 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
2955 The '``indirectbr``' instruction implements an indirect branch to a
2956 label within the current function, whose address is specified by
2957 "``address``". Address must be derived from a
2958 :ref:`blockaddress <blockaddress>` constant.
2963 The '``address``' argument is the address of the label to jump to. The
2964 rest of the arguments indicate the full set of possible destinations
2965 that the address may point to. Blocks are allowed to occur multiple
2966 times in the destination list, though this isn't particularly useful.
2968 This destination list is required so that dataflow analysis has an
2969 accurate understanding of the CFG.
2974 Control transfers to the block specified in the address argument. All
2975 possible destination blocks must be listed in the label list, otherwise
2976 this instruction has undefined behavior. This implies that jumps to
2977 labels defined in other functions have undefined behavior as well.
2982 This is typically implemented with a jump through a register.
2987 .. code-block:: llvm
2989 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
2993 '``invoke``' Instruction
2994 ^^^^^^^^^^^^^^^^^^^^^^^^
3001 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3002 to label <normal label> unwind label <exception label>
3007 The '``invoke``' instruction causes control to transfer to a specified
3008 function, with the possibility of control flow transfer to either the
3009 '``normal``' label or the '``exception``' label. If the callee function
3010 returns with the "``ret``" instruction, control flow will return to the
3011 "normal" label. If the callee (or any indirect callees) returns via the
3012 ":ref:`resume <i_resume>`" instruction or other exception handling
3013 mechanism, control is interrupted and continued at the dynamically
3014 nearest "exception" label.
3016 The '``exception``' label is a `landing
3017 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3018 '``exception``' label is required to have the
3019 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3020 information about the behavior of the program after unwinding happens,
3021 as its first non-PHI instruction. The restrictions on the
3022 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3023 instruction, so that the important information contained within the
3024 "``landingpad``" instruction can't be lost through normal code motion.
3029 This instruction requires several arguments:
3031 #. The optional "cconv" marker indicates which :ref:`calling
3032 convention <callingconv>` the call should use. If none is
3033 specified, the call defaults to using C calling conventions.
3034 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3035 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3037 #. '``ptr to function ty``': shall be the signature of the pointer to
3038 function value being invoked. In most cases, this is a direct
3039 function invocation, but indirect ``invoke``'s are just as possible,
3040 branching off an arbitrary pointer to function value.
3041 #. '``function ptr val``': An LLVM value containing a pointer to a
3042 function to be invoked.
3043 #. '``function args``': argument list whose types match the function
3044 signature argument types and parameter attributes. All arguments must
3045 be of :ref:`first class <t_firstclass>` type. If the function signature
3046 indicates the function accepts a variable number of arguments, the
3047 extra arguments can be specified.
3048 #. '``normal label``': the label reached when the called function
3049 executes a '``ret``' instruction.
3050 #. '``exception label``': the label reached when a callee returns via
3051 the :ref:`resume <i_resume>` instruction or other exception handling
3053 #. The optional :ref:`function attributes <fnattrs>` list. Only
3054 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3055 attributes are valid here.
3060 This instruction is designed to operate as a standard '``call``'
3061 instruction in most regards. The primary difference is that it
3062 establishes an association with a label, which is used by the runtime
3063 library to unwind the stack.
3065 This instruction is used in languages with destructors to ensure that
3066 proper cleanup is performed in the case of either a ``longjmp`` or a
3067 thrown exception. Additionally, this is important for implementation of
3068 '``catch``' clauses in high-level languages that support them.
3070 For the purposes of the SSA form, the definition of the value returned
3071 by the '``invoke``' instruction is deemed to occur on the edge from the
3072 current block to the "normal" label. If the callee unwinds then no
3073 return value is available.
3078 .. code-block:: llvm
3080 %retval = invoke i32 @Test(i32 15) to label %Continue
3081 unwind label %TestCleanup ; {i32}:retval set
3082 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3083 unwind label %TestCleanup ; {i32}:retval set
3087 '``resume``' Instruction
3088 ^^^^^^^^^^^^^^^^^^^^^^^^
3095 resume <type> <value>
3100 The '``resume``' instruction is a terminator instruction that has no
3106 The '``resume``' instruction requires one argument, which must have the
3107 same type as the result of any '``landingpad``' instruction in the same
3113 The '``resume``' instruction resumes propagation of an existing
3114 (in-flight) exception whose unwinding was interrupted with a
3115 :ref:`landingpad <i_landingpad>` instruction.
3120 .. code-block:: llvm
3122 resume { i8*, i32 } %exn
3126 '``unreachable``' Instruction
3127 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3139 The '``unreachable``' instruction has no defined semantics. This
3140 instruction is used to inform the optimizer that a particular portion of
3141 the code is not reachable. This can be used to indicate that the code
3142 after a no-return function cannot be reached, and other facts.
3147 The '``unreachable``' instruction has no defined semantics.
3154 Binary operators are used to do most of the computation in a program.
3155 They require two operands of the same type, execute an operation on
3156 them, and produce a single value. The operands might represent multiple
3157 data, as is the case with the :ref:`vector <t_vector>` data type. The
3158 result value has the same type as its operands.
3160 There are several different binary operators:
3164 '``add``' Instruction
3165 ^^^^^^^^^^^^^^^^^^^^^
3172 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3173 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3174 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3175 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3180 The '``add``' instruction returns the sum of its two operands.
3185 The two arguments to the '``add``' instruction must be
3186 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3187 arguments must have identical types.
3192 The value produced is the integer sum of the two operands.
3194 If the sum has unsigned overflow, the result returned is the
3195 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3198 Because LLVM integers use a two's complement representation, this
3199 instruction is appropriate for both signed and unsigned integers.
3201 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3202 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3203 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3204 unsigned and/or signed overflow, respectively, occurs.
3209 .. code-block:: llvm
3211 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3215 '``fadd``' Instruction
3216 ^^^^^^^^^^^^^^^^^^^^^^
3223 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3228 The '``fadd``' instruction returns the sum of its two operands.
3233 The two arguments to the '``fadd``' instruction must be :ref:`floating
3234 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3235 Both arguments must have identical types.
3240 The value produced is the floating point sum of the two operands. This
3241 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3242 which are optimization hints to enable otherwise unsafe floating point
3248 .. code-block:: llvm
3250 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3252 '``sub``' Instruction
3253 ^^^^^^^^^^^^^^^^^^^^^
3260 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3261 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3262 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3263 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3268 The '``sub``' instruction returns the difference of its two operands.
3270 Note that the '``sub``' instruction is used to represent the '``neg``'
3271 instruction present in most other intermediate representations.
3276 The two arguments to the '``sub``' instruction must be
3277 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3278 arguments must have identical types.
3283 The value produced is the integer difference of the two operands.
3285 If the difference has unsigned overflow, the result returned is the
3286 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3289 Because LLVM integers use a two's complement representation, this
3290 instruction is appropriate for both signed and unsigned integers.
3292 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3293 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3294 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3295 unsigned and/or signed overflow, respectively, occurs.
3300 .. code-block:: llvm
3302 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3303 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3307 '``fsub``' Instruction
3308 ^^^^^^^^^^^^^^^^^^^^^^
3315 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3320 The '``fsub``' instruction returns the difference of its two operands.
3322 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3323 instruction present in most other intermediate representations.
3328 The two arguments to the '``fsub``' instruction must be :ref:`floating
3329 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3330 Both arguments must have identical types.
3335 The value produced is the floating point difference of the two operands.
3336 This instruction can also take any number of :ref:`fast-math
3337 flags <fastmath>`, which are optimization hints to enable otherwise
3338 unsafe floating point optimizations:
3343 .. code-block:: llvm
3345 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3346 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3348 '``mul``' Instruction
3349 ^^^^^^^^^^^^^^^^^^^^^
3356 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3357 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3358 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3359 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3364 The '``mul``' instruction returns the product of its two operands.
3369 The two arguments to the '``mul``' instruction must be
3370 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3371 arguments must have identical types.
3376 The value produced is the integer product of the two operands.
3378 If the result of the multiplication has unsigned overflow, the result
3379 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3380 bit width of the result.
3382 Because LLVM integers use a two's complement representation, and the
3383 result is the same width as the operands, this instruction returns the
3384 correct result for both signed and unsigned integers. If a full product
3385 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3386 sign-extended or zero-extended as appropriate to the width of the full
3389 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3390 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3391 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3392 unsigned and/or signed overflow, respectively, occurs.
3397 .. code-block:: llvm
3399 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3403 '``fmul``' Instruction
3404 ^^^^^^^^^^^^^^^^^^^^^^
3411 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3416 The '``fmul``' instruction returns the product of its two operands.
3421 The two arguments to the '``fmul``' instruction must be :ref:`floating
3422 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3423 Both arguments must have identical types.
3428 The value produced is the floating point product of the two operands.
3429 This instruction can also take any number of :ref:`fast-math
3430 flags <fastmath>`, which are optimization hints to enable otherwise
3431 unsafe floating point optimizations:
3436 .. code-block:: llvm
3438 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3440 '``udiv``' Instruction
3441 ^^^^^^^^^^^^^^^^^^^^^^
3448 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3449 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3454 The '``udiv``' instruction returns the quotient of its two operands.
3459 The two arguments to the '``udiv``' instruction must be
3460 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3461 arguments must have identical types.
3466 The value produced is the unsigned integer quotient of the two operands.
3468 Note that unsigned integer division and signed integer division are
3469 distinct operations; for signed integer division, use '``sdiv``'.
3471 Division by zero leads to undefined behavior.
3473 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3474 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3475 such, "((a udiv exact b) mul b) == a").
3480 .. code-block:: llvm
3482 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3484 '``sdiv``' Instruction
3485 ^^^^^^^^^^^^^^^^^^^^^^
3492 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3493 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3498 The '``sdiv``' instruction returns the quotient of its two operands.
3503 The two arguments to the '``sdiv``' instruction must be
3504 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3505 arguments must have identical types.
3510 The value produced is the signed integer quotient of the two operands
3511 rounded towards zero.
3513 Note that signed integer division and unsigned integer division are
3514 distinct operations; for unsigned integer division, use '``udiv``'.
3516 Division by zero leads to undefined behavior. Overflow also leads to
3517 undefined behavior; this is a rare case, but can occur, for example, by
3518 doing a 32-bit division of -2147483648 by -1.
3520 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3521 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3526 .. code-block:: llvm
3528 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3532 '``fdiv``' Instruction
3533 ^^^^^^^^^^^^^^^^^^^^^^
3540 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3545 The '``fdiv``' instruction returns the quotient of its two operands.
3550 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3551 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3552 Both arguments must have identical types.
3557 The value produced is the floating point quotient of the two operands.
3558 This instruction can also take any number of :ref:`fast-math
3559 flags <fastmath>`, which are optimization hints to enable otherwise
3560 unsafe floating point optimizations:
3565 .. code-block:: llvm
3567 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3569 '``urem``' Instruction
3570 ^^^^^^^^^^^^^^^^^^^^^^
3577 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3582 The '``urem``' instruction returns the remainder from the unsigned
3583 division of its two arguments.
3588 The two arguments to the '``urem``' instruction must be
3589 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3590 arguments must have identical types.
3595 This instruction returns the unsigned integer *remainder* of a division.
3596 This instruction always performs an unsigned division to get the
3599 Note that unsigned integer remainder and signed integer remainder are
3600 distinct operations; for signed integer remainder, use '``srem``'.
3602 Taking the remainder of a division by zero leads to undefined behavior.
3607 .. code-block:: llvm
3609 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3611 '``srem``' Instruction
3612 ^^^^^^^^^^^^^^^^^^^^^^
3619 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3624 The '``srem``' instruction returns the remainder from the signed
3625 division of its two operands. This instruction can also take
3626 :ref:`vector <t_vector>` versions of the values in which case the elements
3632 The two arguments to the '``srem``' instruction must be
3633 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3634 arguments must have identical types.
3639 This instruction returns the *remainder* of a division (where the result
3640 is either zero or has the same sign as the dividend, ``op1``), not the
3641 *modulo* operator (where the result is either zero or has the same sign
3642 as the divisor, ``op2``) of a value. For more information about the
3643 difference, see `The Math
3644 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3645 table of how this is implemented in various languages, please see
3647 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3649 Note that signed integer remainder and unsigned integer remainder are
3650 distinct operations; for unsigned integer remainder, use '``urem``'.
3652 Taking the remainder of a division by zero leads to undefined behavior.
3653 Overflow also leads to undefined behavior; this is a rare case, but can
3654 occur, for example, by taking the remainder of a 32-bit division of
3655 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3656 rule lets srem be implemented using instructions that return both the
3657 result of the division and the remainder.)
3662 .. code-block:: llvm
3664 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
3668 '``frem``' Instruction
3669 ^^^^^^^^^^^^^^^^^^^^^^
3676 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3681 The '``frem``' instruction returns the remainder from the division of
3687 The two arguments to the '``frem``' instruction must be :ref:`floating
3688 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3689 Both arguments must have identical types.
3694 This instruction returns the *remainder* of a division. The remainder
3695 has the same sign as the dividend. This instruction can also take any
3696 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
3697 to enable otherwise unsafe floating point optimizations:
3702 .. code-block:: llvm
3704 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
3708 Bitwise Binary Operations
3709 -------------------------
3711 Bitwise binary operators are used to do various forms of bit-twiddling
3712 in a program. They are generally very efficient instructions and can
3713 commonly be strength reduced from other instructions. They require two
3714 operands of the same type, execute an operation on them, and produce a
3715 single value. The resulting value is the same type as its operands.
3717 '``shl``' Instruction
3718 ^^^^^^^^^^^^^^^^^^^^^
3725 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
3726 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
3727 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
3728 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3733 The '``shl``' instruction returns the first operand shifted to the left
3734 a specified number of bits.
3739 Both arguments to the '``shl``' instruction must be the same
3740 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3741 '``op2``' is treated as an unsigned value.
3746 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
3747 where ``n`` is the width of the result. If ``op2`` is (statically or
3748 dynamically) negative or equal to or larger than the number of bits in
3749 ``op1``, the result is undefined. If the arguments are vectors, each
3750 vector element of ``op1`` is shifted by the corresponding shift amount
3753 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
3754 value <poisonvalues>` if it shifts out any non-zero bits. If the
3755 ``nsw`` keyword is present, then the shift produces a :ref:`poison
3756 value <poisonvalues>` if it shifts out any bits that disagree with the
3757 resultant sign bit. As such, NUW/NSW have the same semantics as they
3758 would if the shift were expressed as a mul instruction with the same
3759 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
3764 .. code-block:: llvm
3766 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
3767 <result> = shl i32 4, 2 ; yields {i32}: 16
3768 <result> = shl i32 1, 10 ; yields {i32}: 1024
3769 <result> = shl i32 1, 32 ; undefined
3770 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
3772 '``lshr``' Instruction
3773 ^^^^^^^^^^^^^^^^^^^^^^
3780 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
3781 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
3786 The '``lshr``' instruction (logical shift right) returns the first
3787 operand shifted to the right a specified number of bits with zero fill.
3792 Both arguments to the '``lshr``' instruction must be the same
3793 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3794 '``op2``' is treated as an unsigned value.
3799 This instruction always performs a logical shift right operation. The
3800 most significant bits of the result will be filled with zero bits after
3801 the shift. If ``op2`` is (statically or dynamically) equal to or larger
3802 than the number of bits in ``op1``, the result is undefined. If the
3803 arguments are vectors, each vector element of ``op1`` is shifted by the
3804 corresponding shift amount in ``op2``.
3806 If the ``exact`` keyword is present, the result value of the ``lshr`` is
3807 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
3813 .. code-block:: llvm
3815 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
3816 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
3817 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
3818 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7FFFFFFF
3819 <result> = lshr i32 1, 32 ; undefined
3820 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
3822 '``ashr``' Instruction
3823 ^^^^^^^^^^^^^^^^^^^^^^
3830 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
3831 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
3836 The '``ashr``' instruction (arithmetic shift right) returns the first
3837 operand shifted to the right a specified number of bits with sign
3843 Both arguments to the '``ashr``' instruction must be the same
3844 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3845 '``op2``' is treated as an unsigned value.
3850 This instruction always performs an arithmetic shift right operation,
3851 The most significant bits of the result will be filled with the sign bit
3852 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
3853 than the number of bits in ``op1``, the result is undefined. If the
3854 arguments are vectors, each vector element of ``op1`` is shifted by the
3855 corresponding shift amount in ``op2``.
3857 If the ``exact`` keyword is present, the result value of the ``ashr`` is
3858 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
3864 .. code-block:: llvm
3866 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
3867 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
3868 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
3869 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
3870 <result> = ashr i32 1, 32 ; undefined
3871 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
3873 '``and``' Instruction
3874 ^^^^^^^^^^^^^^^^^^^^^
3881 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
3886 The '``and``' instruction returns the bitwise logical and of its two
3892 The two arguments to the '``and``' instruction must be
3893 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3894 arguments must have identical types.
3899 The truth table used for the '``and``' instruction is:
3916 .. code-block:: llvm
3918 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
3919 <result> = and i32 15, 40 ; yields {i32}:result = 8
3920 <result> = and i32 4, 8 ; yields {i32}:result = 0
3922 '``or``' Instruction
3923 ^^^^^^^^^^^^^^^^^^^^
3930 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
3935 The '``or``' instruction returns the bitwise logical inclusive or of its
3941 The two arguments to the '``or``' instruction must be
3942 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3943 arguments must have identical types.
3948 The truth table used for the '``or``' instruction is:
3967 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
3968 <result> = or i32 15, 40 ; yields {i32}:result = 47
3969 <result> = or i32 4, 8 ; yields {i32}:result = 12
3971 '``xor``' Instruction
3972 ^^^^^^^^^^^^^^^^^^^^^
3979 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
3984 The '``xor``' instruction returns the bitwise logical exclusive or of
3985 its two operands. The ``xor`` is used to implement the "one's
3986 complement" operation, which is the "~" operator in C.
3991 The two arguments to the '``xor``' instruction must be
3992 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3993 arguments must have identical types.
3998 The truth table used for the '``xor``' instruction is:
4015 .. code-block:: llvm
4017 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4018 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4019 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4020 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4025 LLVM supports several instructions to represent vector operations in a
4026 target-independent manner. These instructions cover the element-access
4027 and vector-specific operations needed to process vectors effectively.
4028 While LLVM does directly support these vector operations, many
4029 sophisticated algorithms will want to use target-specific intrinsics to
4030 take full advantage of a specific target.
4032 .. _i_extractelement:
4034 '``extractelement``' Instruction
4035 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4042 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4047 The '``extractelement``' instruction extracts a single scalar element
4048 from a vector at a specified index.
4053 The first operand of an '``extractelement``' instruction is a value of
4054 :ref:`vector <t_vector>` type. The second operand is an index indicating
4055 the position from which to extract the element. The index may be a
4061 The result is a scalar of the same type as the element type of ``val``.
4062 Its value is the value at position ``idx`` of ``val``. If ``idx``
4063 exceeds the length of ``val``, the results are undefined.
4068 .. code-block:: llvm
4070 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4072 .. _i_insertelement:
4074 '``insertelement``' Instruction
4075 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4082 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4087 The '``insertelement``' instruction inserts a scalar element into a
4088 vector at a specified index.
4093 The first operand of an '``insertelement``' instruction is a value of
4094 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4095 type must equal the element type of the first operand. The third operand
4096 is an index indicating the position at which to insert the value. The
4097 index may be a variable.
4102 The result is a vector of the same type as ``val``. Its element values
4103 are those of ``val`` except at position ``idx``, where it gets the value
4104 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4110 .. code-block:: llvm
4112 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4114 .. _i_shufflevector:
4116 '``shufflevector``' Instruction
4117 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4124 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4129 The '``shufflevector``' instruction constructs a permutation of elements
4130 from two input vectors, returning a vector with the same element type as
4131 the input and length that is the same as the shuffle mask.
4136 The first two operands of a '``shufflevector``' instruction are vectors
4137 with the same type. The third argument is a shuffle mask whose element
4138 type is always 'i32'. The result of the instruction is a vector whose
4139 length is the same as the shuffle mask and whose element type is the
4140 same as the element type of the first two operands.
4142 The shuffle mask operand is required to be a constant vector with either
4143 constant integer or undef values.
4148 The elements of the two input vectors are numbered from left to right
4149 across both of the vectors. The shuffle mask operand specifies, for each
4150 element of the result vector, which element of the two input vectors the
4151 result element gets. The element selector may be undef (meaning "don't
4152 care") and the second operand may be undef if performing a shuffle from
4158 .. code-block:: llvm
4160 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4161 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4162 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4163 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4164 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4165 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4166 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4167 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4169 Aggregate Operations
4170 --------------------
4172 LLVM supports several instructions for working with
4173 :ref:`aggregate <t_aggregate>` values.
4177 '``extractvalue``' Instruction
4178 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4185 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4190 The '``extractvalue``' instruction extracts the value of a member field
4191 from an :ref:`aggregate <t_aggregate>` value.
4196 The first operand of an '``extractvalue``' instruction is a value of
4197 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4198 constant indices to specify which value to extract in a similar manner
4199 as indices in a '``getelementptr``' instruction.
4201 The major differences to ``getelementptr`` indexing are:
4203 - Since the value being indexed is not a pointer, the first index is
4204 omitted and assumed to be zero.
4205 - At least one index must be specified.
4206 - Not only struct indices but also array indices must be in bounds.
4211 The result is the value at the position in the aggregate specified by
4217 .. code-block:: llvm
4219 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4223 '``insertvalue``' Instruction
4224 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4231 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4236 The '``insertvalue``' instruction inserts a value into a member field in
4237 an :ref:`aggregate <t_aggregate>` value.
4242 The first operand of an '``insertvalue``' instruction is a value of
4243 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4244 a first-class value to insert. The following operands are constant
4245 indices indicating the position at which to insert the value in a
4246 similar manner as indices in a '``extractvalue``' instruction. The value
4247 to insert must have the same type as the value identified by the
4253 The result is an aggregate of the same type as ``val``. Its value is
4254 that of ``val`` except that the value at the position specified by the
4255 indices is that of ``elt``.
4260 .. code-block:: llvm
4262 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4263 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4264 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4268 Memory Access and Addressing Operations
4269 ---------------------------------------
4271 A key design point of an SSA-based representation is how it represents
4272 memory. In LLVM, no memory locations are in SSA form, which makes things
4273 very simple. This section describes how to read, write, and allocate
4278 '``alloca``' Instruction
4279 ^^^^^^^^^^^^^^^^^^^^^^^^
4286 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4291 The '``alloca``' instruction allocates memory on the stack frame of the
4292 currently executing function, to be automatically released when this
4293 function returns to its caller. The object is always allocated in the
4294 generic address space (address space zero).
4299 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4300 bytes of memory on the runtime stack, returning a pointer of the
4301 appropriate type to the program. If "NumElements" is specified, it is
4302 the number of elements allocated, otherwise "NumElements" is defaulted
4303 to be one. If a constant alignment is specified, the value result of the
4304 allocation is guaranteed to be aligned to at least that boundary. If not
4305 specified, or if zero, the target can choose to align the allocation on
4306 any convenient boundary compatible with the type.
4308 '``type``' may be any sized type.
4313 Memory is allocated; a pointer is returned. The operation is undefined
4314 if there is insufficient stack space for the allocation. '``alloca``'d
4315 memory is automatically released when the function returns. The
4316 '``alloca``' instruction is commonly used to represent automatic
4317 variables that must have an address available. When the function returns
4318 (either with the ``ret`` or ``resume`` instructions), the memory is
4319 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4320 The order in which memory is allocated (ie., which way the stack grows)
4326 .. code-block:: llvm
4328 %ptr = alloca i32 ; yields {i32*}:ptr
4329 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4330 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4331 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4335 '``load``' Instruction
4336 ^^^^^^^^^^^^^^^^^^^^^^
4343 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4344 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4345 !<index> = !{ i32 1 }
4350 The '``load``' instruction is used to read from memory.
4355 The argument to the '``load``' instruction specifies the memory address
4356 from which to load. The pointer must point to a :ref:`first
4357 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4358 then the optimizer is not allowed to modify the number or order of
4359 execution of this ``load`` with other :ref:`volatile
4360 operations <volatile>`.
4362 If the ``load`` is marked as ``atomic``, it takes an extra
4363 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4364 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4365 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4366 when they may see multiple atomic stores. The type of the pointee must
4367 be an integer type whose bit width is a power of two greater than or
4368 equal to eight and less than or equal to a target-specific size limit.
4369 ``align`` must be explicitly specified on atomic loads, and the load has
4370 undefined behavior if the alignment is not set to a value which is at
4371 least the size in bytes of the pointee. ``!nontemporal`` does not have
4372 any defined semantics for atomic loads.
4374 The optional constant ``align`` argument specifies the alignment of the
4375 operation (that is, the alignment of the memory address). A value of 0
4376 or an omitted ``align`` argument means that the operation has the abi
4377 alignment for the target. It is the responsibility of the code emitter
4378 to ensure that the alignment information is correct. Overestimating the
4379 alignment results in undefined behavior. Underestimating the alignment
4380 may produce less efficient code. An alignment of 1 is always safe.
4382 The optional ``!nontemporal`` metadata must reference a single
4383 metatadata name <index> corresponding to a metadata node with one
4384 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4385 metatadata on the instruction tells the optimizer and code generator
4386 that this load is not expected to be reused in the cache. The code
4387 generator may select special instructions to save cache bandwidth, such
4388 as the ``MOVNT`` instruction on x86.
4390 The optional ``!invariant.load`` metadata must reference a single
4391 metatadata name <index> corresponding to a metadata node with no
4392 entries. The existence of the ``!invariant.load`` metatadata on the
4393 instruction tells the optimizer and code generator that this load
4394 address points to memory which does not change value during program
4395 execution. The optimizer may then move this load around, for example, by
4396 hoisting it out of loops using loop invariant code motion.
4401 The location of memory pointed to is loaded. If the value being loaded
4402 is of scalar type then the number of bytes read does not exceed the
4403 minimum number of bytes needed to hold all bits of the type. For
4404 example, loading an ``i24`` reads at most three bytes. When loading a
4405 value of a type like ``i20`` with a size that is not an integral number
4406 of bytes, the result is undefined if the value was not originally
4407 written using a store of the same type.
4412 .. code-block:: llvm
4414 %ptr = alloca i32 ; yields {i32*}:ptr
4415 store i32 3, i32* %ptr ; yields {void}
4416 %val = load i32* %ptr ; yields {i32}:val = i32 3
4420 '``store``' Instruction
4421 ^^^^^^^^^^^^^^^^^^^^^^^
4428 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4429 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4434 The '``store``' instruction is used to write to memory.
4439 There are two arguments to the '``store``' instruction: a value to store
4440 and an address at which to store it. The type of the '``<pointer>``'
4441 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4442 the '``<value>``' operand. If the ``store`` is marked as ``volatile``,
4443 then the optimizer is not allowed to modify the number or order of
4444 execution of this ``store`` with other :ref:`volatile
4445 operations <volatile>`.
4447 If the ``store`` is marked as ``atomic``, it takes an extra
4448 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4449 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4450 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4451 when they may see multiple atomic stores. The type of the pointee must
4452 be an integer type whose bit width is a power of two greater than or
4453 equal to eight and less than or equal to a target-specific size limit.
4454 ``align`` must be explicitly specified on atomic stores, and the store
4455 has undefined behavior if the alignment is not set to a value which is
4456 at least the size in bytes of the pointee. ``!nontemporal`` does not
4457 have any defined semantics for atomic stores.
4459 The optional constant "align" argument specifies the alignment of the
4460 operation (that is, the alignment of the memory address). A value of 0
4461 or an omitted "align" argument means that the operation has the abi
4462 alignment for the target. It is the responsibility of the code emitter
4463 to ensure that the alignment information is correct. Overestimating the
4464 alignment results in an undefined behavior. Underestimating the
4465 alignment may produce less efficient code. An alignment of 1 is always
4468 The optional !nontemporal metadata must reference a single metatadata
4469 name <index> corresponding to a metadata node with one i32 entry of
4470 value 1. The existence of the !nontemporal metatadata on the instruction
4471 tells the optimizer and code generator that this load is not expected to
4472 be reused in the cache. The code generator may select special
4473 instructions to save cache bandwidth, such as the MOVNT instruction on
4479 The contents of memory are updated to contain '``<value>``' at the
4480 location specified by the '``<pointer>``' operand. If '``<value>``' is
4481 of scalar type then the number of bytes written does not exceed the
4482 minimum number of bytes needed to hold all bits of the type. For
4483 example, storing an ``i24`` writes at most three bytes. When writing a
4484 value of a type like ``i20`` with a size that is not an integral number
4485 of bytes, it is unspecified what happens to the extra bits that do not
4486 belong to the type, but they will typically be overwritten.
4491 .. code-block:: llvm
4493 %ptr = alloca i32 ; yields {i32*}:ptr
4494 store i32 3, i32* %ptr ; yields {void}
4495 %val = load i32* %ptr ; yields {i32}:val = i32 3
4499 '``fence``' Instruction
4500 ^^^^^^^^^^^^^^^^^^^^^^^
4507 fence [singlethread] <ordering> ; yields {void}
4512 The '``fence``' instruction is used to introduce happens-before edges
4518 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4519 defines what *synchronizes-with* edges they add. They can only be given
4520 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4525 A fence A which has (at least) ``release`` ordering semantics
4526 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4527 semantics if and only if there exist atomic operations X and Y, both
4528 operating on some atomic object M, such that A is sequenced before X, X
4529 modifies M (either directly or through some side effect of a sequence
4530 headed by X), Y is sequenced before B, and Y observes M. This provides a
4531 *happens-before* dependency between A and B. Rather than an explicit
4532 ``fence``, one (but not both) of the atomic operations X or Y might
4533 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4534 still *synchronize-with* the explicit ``fence`` and establish the
4535 *happens-before* edge.
4537 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4538 ``acquire`` and ``release`` semantics specified above, participates in
4539 the global program order of other ``seq_cst`` operations and/or fences.
4541 The optional ":ref:`singlethread <singlethread>`" argument specifies
4542 that the fence only synchronizes with other fences in the same thread.
4543 (This is useful for interacting with signal handlers.)
4548 .. code-block:: llvm
4550 fence acquire ; yields {void}
4551 fence singlethread seq_cst ; yields {void}
4555 '``cmpxchg``' Instruction
4556 ^^^^^^^^^^^^^^^^^^^^^^^^^
4563 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4568 The '``cmpxchg``' instruction is used to atomically modify memory. It
4569 loads a value in memory and compares it to a given value. If they are
4570 equal, it stores a new value into the memory.
4575 There are three arguments to the '``cmpxchg``' instruction: an address
4576 to operate on, a value to compare to the value currently be at that
4577 address, and a new value to place at that address if the compared values
4578 are equal. The type of '<cmp>' must be an integer type whose bit width
4579 is a power of two greater than or equal to eight and less than or equal
4580 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4581 type, and the type of '<pointer>' must be a pointer to that type. If the
4582 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4583 to modify the number or order of execution of this ``cmpxchg`` with
4584 other :ref:`volatile operations <volatile>`.
4586 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4587 synchronizes with other atomic operations.
4589 The optional "``singlethread``" argument declares that the ``cmpxchg``
4590 is only atomic with respect to code (usually signal handlers) running in
4591 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4592 respect to all other code in the system.
4594 The pointer passed into cmpxchg must have alignment greater than or
4595 equal to the size in memory of the operand.
4600 The contents of memory at the location specified by the '``<pointer>``'
4601 operand is read and compared to '``<cmp>``'; if the read value is the
4602 equal, '``<new>``' is written. The original value at the location is
4605 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4606 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4607 atomic load with an ordering parameter determined by dropping any
4608 ``release`` part of the ``cmpxchg``'s ordering.
4613 .. code-block:: llvm
4616 %orig = atomic load i32* %ptr unordered ; yields {i32}
4620 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4621 %squared = mul i32 %cmp, %cmp
4622 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4623 %success = icmp eq i32 %cmp, %old
4624 br i1 %success, label %done, label %loop
4631 '``atomicrmw``' Instruction
4632 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4639 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4644 The '``atomicrmw``' instruction is used to atomically modify memory.
4649 There are three arguments to the '``atomicrmw``' instruction: an
4650 operation to apply, an address whose value to modify, an argument to the
4651 operation. The operation must be one of the following keywords:
4665 The type of '<value>' must be an integer type whose bit width is a power
4666 of two greater than or equal to eight and less than or equal to a
4667 target-specific size limit. The type of the '``<pointer>``' operand must
4668 be a pointer to that type. If the ``atomicrmw`` is marked as
4669 ``volatile``, then the optimizer is not allowed to modify the number or
4670 order of execution of this ``atomicrmw`` with other :ref:`volatile
4671 operations <volatile>`.
4676 The contents of memory at the location specified by the '``<pointer>``'
4677 operand are atomically read, modified, and written back. The original
4678 value at the location is returned. The modification is specified by the
4681 - xchg: ``*ptr = val``
4682 - add: ``*ptr = *ptr + val``
4683 - sub: ``*ptr = *ptr - val``
4684 - and: ``*ptr = *ptr & val``
4685 - nand: ``*ptr = ~(*ptr & val)``
4686 - or: ``*ptr = *ptr | val``
4687 - xor: ``*ptr = *ptr ^ val``
4688 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
4689 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
4690 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
4692 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
4698 .. code-block:: llvm
4700 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
4702 .. _i_getelementptr:
4704 '``getelementptr``' Instruction
4705 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4712 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4713 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4714 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
4719 The '``getelementptr``' instruction is used to get the address of a
4720 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
4721 address calculation only and does not access memory.
4726 The first argument is always a pointer or a vector of pointers, and
4727 forms the basis of the calculation. The remaining arguments are indices
4728 that indicate which of the elements of the aggregate object are indexed.
4729 The interpretation of each index is dependent on the type being indexed
4730 into. The first index always indexes the pointer value given as the
4731 first argument, the second index indexes a value of the type pointed to
4732 (not necessarily the value directly pointed to, since the first index
4733 can be non-zero), etc. The first type indexed into must be a pointer
4734 value, subsequent types can be arrays, vectors, and structs. Note that
4735 subsequent types being indexed into can never be pointers, since that
4736 would require loading the pointer before continuing calculation.
4738 The type of each index argument depends on the type it is indexing into.
4739 When indexing into a (optionally packed) structure, only ``i32`` integer
4740 **constants** are allowed (when using a vector of indices they must all
4741 be the **same** ``i32`` integer constant). When indexing into an array,
4742 pointer or vector, integers of any width are allowed, and they are not
4743 required to be constant. These integers are treated as signed values
4746 For example, let's consider a C code fragment and how it gets compiled
4762 int *foo(struct ST *s) {
4763 return &s[1].Z.B[5][13];
4766 The LLVM code generated by Clang is:
4768 .. code-block:: llvm
4770 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
4771 %struct.ST = type { i32, double, %struct.RT }
4773 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
4775 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
4782 In the example above, the first index is indexing into the
4783 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
4784 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
4785 indexes into the third element of the structure, yielding a
4786 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
4787 structure. The third index indexes into the second element of the
4788 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
4789 dimensions of the array are subscripted into, yielding an '``i32``'
4790 type. The '``getelementptr``' instruction returns a pointer to this
4791 element, thus computing a value of '``i32*``' type.
4793 Note that it is perfectly legal to index partially through a structure,
4794 returning a pointer to an inner element. Because of this, the LLVM code
4795 for the given testcase is equivalent to:
4797 .. code-block:: llvm
4799 define i32* @foo(%struct.ST* %s) {
4800 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
4801 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
4802 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
4803 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
4804 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
4808 If the ``inbounds`` keyword is present, the result value of the
4809 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
4810 pointer is not an *in bounds* address of an allocated object, or if any
4811 of the addresses that would be formed by successive addition of the
4812 offsets implied by the indices to the base address with infinitely
4813 precise signed arithmetic are not an *in bounds* address of that
4814 allocated object. The *in bounds* addresses for an allocated object are
4815 all the addresses that point into the object, plus the address one byte
4816 past the end. In cases where the base is a vector of pointers the
4817 ``inbounds`` keyword applies to each of the computations element-wise.
4819 If the ``inbounds`` keyword is not present, the offsets are added to the
4820 base address with silently-wrapping two's complement arithmetic. If the
4821 offsets have a different width from the pointer, they are sign-extended
4822 or truncated to the width of the pointer. The result value of the
4823 ``getelementptr`` may be outside the object pointed to by the base
4824 pointer. The result value may not necessarily be used to access memory
4825 though, even if it happens to point into allocated storage. See the
4826 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
4829 The getelementptr instruction is often confusing. For some more insight
4830 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
4835 .. code-block:: llvm
4837 ; yields [12 x i8]*:aptr
4838 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
4840 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
4842 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
4844 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
4846 In cases where the pointer argument is a vector of pointers, each index
4847 must be a vector with the same number of elements. For example:
4849 .. code-block:: llvm
4851 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
4853 Conversion Operations
4854 ---------------------
4856 The instructions in this category are the conversion instructions
4857 (casting) which all take a single operand and a type. They perform
4858 various bit conversions on the operand.
4860 '``trunc .. to``' Instruction
4861 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4868 <result> = trunc <ty> <value> to <ty2> ; yields ty2
4873 The '``trunc``' instruction truncates its operand to the type ``ty2``.
4878 The '``trunc``' instruction takes a value to trunc, and a type to trunc
4879 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
4880 of the same number of integers. The bit size of the ``value`` must be
4881 larger than the bit size of the destination type, ``ty2``. Equal sized
4882 types are not allowed.
4887 The '``trunc``' instruction truncates the high order bits in ``value``
4888 and converts the remaining bits to ``ty2``. Since the source size must
4889 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
4890 It will always truncate bits.
4895 .. code-block:: llvm
4897 %X = trunc i32 257 to i8 ; yields i8:1
4898 %Y = trunc i32 123 to i1 ; yields i1:true
4899 %Z = trunc i32 122 to i1 ; yields i1:false
4900 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
4902 '``zext .. to``' Instruction
4903 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4910 <result> = zext <ty> <value> to <ty2> ; yields ty2
4915 The '``zext``' instruction zero extends its operand to type ``ty2``.
4920 The '``zext``' instruction takes a value to cast, and a type to cast it
4921 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
4922 the same number of integers. The bit size of the ``value`` must be
4923 smaller than the bit size of the destination type, ``ty2``.
4928 The ``zext`` fills the high order bits of the ``value`` with zero bits
4929 until it reaches the size of the destination type, ``ty2``.
4931 When zero extending from i1, the result will always be either 0 or 1.
4936 .. code-block:: llvm
4938 %X = zext i32 257 to i64 ; yields i64:257
4939 %Y = zext i1 true to i32 ; yields i32:1
4940 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
4942 '``sext .. to``' Instruction
4943 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4950 <result> = sext <ty> <value> to <ty2> ; yields ty2
4955 The '``sext``' sign extends ``value`` to the type ``ty2``.
4960 The '``sext``' instruction takes a value to cast, and a type to cast it
4961 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
4962 the same number of integers. The bit size of the ``value`` must be
4963 smaller than the bit size of the destination type, ``ty2``.
4968 The '``sext``' instruction performs a sign extension by copying the sign
4969 bit (highest order bit) of the ``value`` until it reaches the bit size
4970 of the type ``ty2``.
4972 When sign extending from i1, the extension always results in -1 or 0.
4977 .. code-block:: llvm
4979 %X = sext i8 -1 to i16 ; yields i16 :65535
4980 %Y = sext i1 true to i32 ; yields i32:-1
4981 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
4983 '``fptrunc .. to``' Instruction
4984 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4991 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
4996 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5001 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5002 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5003 The size of ``value`` must be larger than the size of ``ty2``. This
5004 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5009 The '``fptrunc``' instruction truncates a ``value`` from a larger
5010 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5011 point <t_floating>` type. If the value cannot fit within the
5012 destination type, ``ty2``, then the results are undefined.
5017 .. code-block:: llvm
5019 %X = fptrunc double 123.0 to float ; yields float:123.0
5020 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5022 '``fpext .. to``' Instruction
5023 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5030 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5035 The '``fpext``' extends a floating point ``value`` to a larger floating
5041 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5042 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5043 to. The source type must be smaller than the destination type.
5048 The '``fpext``' instruction extends the ``value`` from a smaller
5049 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5050 point <t_floating>` type. The ``fpext`` cannot be used to make a
5051 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5052 *no-op cast* for a floating point cast.
5057 .. code-block:: llvm
5059 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5060 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5062 '``fptoui .. to``' Instruction
5063 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5070 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5075 The '``fptoui``' converts a floating point ``value`` to its unsigned
5076 integer equivalent of type ``ty2``.
5081 The '``fptoui``' instruction takes a value to cast, which must be a
5082 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5083 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5084 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5085 type with the same number of elements as ``ty``
5090 The '``fptoui``' instruction converts its :ref:`floating
5091 point <t_floating>` operand into the nearest (rounding towards zero)
5092 unsigned integer value. If the value cannot fit in ``ty2``, the results
5098 .. code-block:: llvm
5100 %X = fptoui double 123.0 to i32 ; yields i32:123
5101 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5102 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5104 '``fptosi .. to``' Instruction
5105 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5112 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5117 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5118 ``value`` to type ``ty2``.
5123 The '``fptosi``' instruction takes a value to cast, which must be a
5124 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5125 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5126 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5127 type with the same number of elements as ``ty``
5132 The '``fptosi``' instruction converts its :ref:`floating
5133 point <t_floating>` operand into the nearest (rounding towards zero)
5134 signed integer value. If the value cannot fit in ``ty2``, the results
5140 .. code-block:: llvm
5142 %X = fptosi double -123.0 to i32 ; yields i32:-123
5143 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5144 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5146 '``uitofp .. to``' Instruction
5147 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5154 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5159 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5160 and converts that value to the ``ty2`` type.
5165 The '``uitofp``' instruction takes a value to cast, which must be a
5166 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5167 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5168 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5169 type with the same number of elements as ``ty``
5174 The '``uitofp``' instruction interprets its operand as an unsigned
5175 integer quantity and converts it to the corresponding floating point
5176 value. If the value cannot fit in the floating point value, the results
5182 .. code-block:: llvm
5184 %X = uitofp i32 257 to float ; yields float:257.0
5185 %Y = uitofp i8 -1 to double ; yields double:255.0
5187 '``sitofp .. to``' Instruction
5188 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5195 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5200 The '``sitofp``' instruction regards ``value`` as a signed integer and
5201 converts that value to the ``ty2`` type.
5206 The '``sitofp``' instruction takes a value to cast, which must be a
5207 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5208 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5209 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5210 type with the same number of elements as ``ty``
5215 The '``sitofp``' instruction interprets its operand as a signed integer
5216 quantity and converts it to the corresponding floating point value. If
5217 the value cannot fit in the floating point value, the results are
5223 .. code-block:: llvm
5225 %X = sitofp i32 257 to float ; yields float:257.0
5226 %Y = sitofp i8 -1 to double ; yields double:-1.0
5230 '``ptrtoint .. to``' Instruction
5231 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5238 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5243 The '``ptrtoint``' instruction converts the pointer or a vector of
5244 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5249 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5250 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5251 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5252 a vector of integers type.
5257 The '``ptrtoint``' instruction converts ``value`` to integer type
5258 ``ty2`` by interpreting the pointer value as an integer and either
5259 truncating or zero extending that value to the size of the integer type.
5260 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5261 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5262 the same size, then nothing is done (*no-op cast*) other than a type
5268 .. code-block:: llvm
5270 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5271 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5272 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5276 '``inttoptr .. to``' Instruction
5277 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5284 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5289 The '``inttoptr``' instruction converts an integer ``value`` to a
5290 pointer type, ``ty2``.
5295 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5296 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5302 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5303 applying either a zero extension or a truncation depending on the size
5304 of the integer ``value``. If ``value`` is larger than the size of a
5305 pointer then a truncation is done. If ``value`` is smaller than the size
5306 of a pointer then a zero extension is done. If they are the same size,
5307 nothing is done (*no-op cast*).
5312 .. code-block:: llvm
5314 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5315 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5316 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5317 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5321 '``bitcast .. to``' Instruction
5322 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5329 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5334 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5340 The '``bitcast``' instruction takes a value to cast, which must be a
5341 non-aggregate first class value, and a type to cast it to, which must
5342 also be a non-aggregate :ref:`first class <t_firstclass>` type. The bit
5343 sizes of ``value`` and the destination type, ``ty2``, must be identical.
5344 If the source type is a pointer, the destination type must also be a
5345 pointer. This instruction supports bitwise conversion of vectors to
5346 integers and to vectors of other types (as long as they have the same
5352 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It is
5353 always a *no-op cast* because no bits change with this conversion. The
5354 conversion is done as if the ``value`` had been stored to memory and
5355 read back as type ``ty2``. Pointer (or vector of pointers) types may
5356 only be converted to other pointer (or vector of pointers) types with
5357 this instruction. To convert pointers to other types, use the
5358 :ref:`inttoptr <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions
5364 .. code-block:: llvm
5366 %X = bitcast i8 255 to i8 ; yields i8 :-1
5367 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5368 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5369 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5376 The instructions in this category are the "miscellaneous" instructions,
5377 which defy better classification.
5381 '``icmp``' Instruction
5382 ^^^^^^^^^^^^^^^^^^^^^^
5389 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5394 The '``icmp``' instruction returns a boolean value or a vector of
5395 boolean values based on comparison of its two integer, integer vector,
5396 pointer, or pointer vector operands.
5401 The '``icmp``' instruction takes three operands. The first operand is
5402 the condition code indicating the kind of comparison to perform. It is
5403 not a value, just a keyword. The possible condition code are:
5406 #. ``ne``: not equal
5407 #. ``ugt``: unsigned greater than
5408 #. ``uge``: unsigned greater or equal
5409 #. ``ult``: unsigned less than
5410 #. ``ule``: unsigned less or equal
5411 #. ``sgt``: signed greater than
5412 #. ``sge``: signed greater or equal
5413 #. ``slt``: signed less than
5414 #. ``sle``: signed less or equal
5416 The remaining two arguments must be :ref:`integer <t_integer>` or
5417 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5418 must also be identical types.
5423 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5424 code given as ``cond``. The comparison performed always yields either an
5425 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5427 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5428 otherwise. No sign interpretation is necessary or performed.
5429 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5430 otherwise. No sign interpretation is necessary or performed.
5431 #. ``ugt``: interprets the operands as unsigned values and yields
5432 ``true`` if ``op1`` is greater than ``op2``.
5433 #. ``uge``: interprets the operands as unsigned values and yields
5434 ``true`` if ``op1`` is greater than or equal to ``op2``.
5435 #. ``ult``: interprets the operands as unsigned values and yields
5436 ``true`` if ``op1`` is less than ``op2``.
5437 #. ``ule``: interprets the operands as unsigned values and yields
5438 ``true`` if ``op1`` is less than or equal to ``op2``.
5439 #. ``sgt``: interprets the operands as signed values and yields ``true``
5440 if ``op1`` is greater than ``op2``.
5441 #. ``sge``: interprets the operands as signed values and yields ``true``
5442 if ``op1`` is greater than or equal to ``op2``.
5443 #. ``slt``: interprets the operands as signed values and yields ``true``
5444 if ``op1`` is less than ``op2``.
5445 #. ``sle``: interprets the operands as signed values and yields ``true``
5446 if ``op1`` is less than or equal to ``op2``.
5448 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5449 are compared as if they were integers.
5451 If the operands are integer vectors, then they are compared element by
5452 element. The result is an ``i1`` vector with the same number of elements
5453 as the values being compared. Otherwise, the result is an ``i1``.
5458 .. code-block:: llvm
5460 <result> = icmp eq i32 4, 5 ; yields: result=false
5461 <result> = icmp ne float* %X, %X ; yields: result=false
5462 <result> = icmp ult i16 4, 5 ; yields: result=true
5463 <result> = icmp sgt i16 4, 5 ; yields: result=false
5464 <result> = icmp ule i16 -4, 5 ; yields: result=false
5465 <result> = icmp sge i16 4, 5 ; yields: result=false
5467 Note that the code generator does not yet support vector types with the
5468 ``icmp`` instruction.
5472 '``fcmp``' Instruction
5473 ^^^^^^^^^^^^^^^^^^^^^^
5480 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5485 The '``fcmp``' instruction returns a boolean value or vector of boolean
5486 values based on comparison of its operands.
5488 If the operands are floating point scalars, then the result type is a
5489 boolean (:ref:`i1 <t_integer>`).
5491 If the operands are floating point vectors, then the result type is a
5492 vector of boolean with the same number of elements as the operands being
5498 The '``fcmp``' instruction takes three operands. The first operand is
5499 the condition code indicating the kind of comparison to perform. It is
5500 not a value, just a keyword. The possible condition code are:
5502 #. ``false``: no comparison, always returns false
5503 #. ``oeq``: ordered and equal
5504 #. ``ogt``: ordered and greater than
5505 #. ``oge``: ordered and greater than or equal
5506 #. ``olt``: ordered and less than
5507 #. ``ole``: ordered and less than or equal
5508 #. ``one``: ordered and not equal
5509 #. ``ord``: ordered (no nans)
5510 #. ``ueq``: unordered or equal
5511 #. ``ugt``: unordered or greater than
5512 #. ``uge``: unordered or greater than or equal
5513 #. ``ult``: unordered or less than
5514 #. ``ule``: unordered or less than or equal
5515 #. ``une``: unordered or not equal
5516 #. ``uno``: unordered (either nans)
5517 #. ``true``: no comparison, always returns true
5519 *Ordered* means that neither operand is a QNAN while *unordered* means
5520 that either operand may be a QNAN.
5522 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5523 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5524 type. They must have identical types.
5529 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5530 condition code given as ``cond``. If the operands are vectors, then the
5531 vectors are compared element by element. Each comparison performed
5532 always yields an :ref:`i1 <t_integer>` result, as follows:
5534 #. ``false``: always yields ``false``, regardless of operands.
5535 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5536 is equal to ``op2``.
5537 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5538 is greater than ``op2``.
5539 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5540 is greater than or equal to ``op2``.
5541 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5542 is less than ``op2``.
5543 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5544 is less than or equal to ``op2``.
5545 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5546 is not equal to ``op2``.
5547 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5548 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5550 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5551 greater than ``op2``.
5552 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5553 greater than or equal to ``op2``.
5554 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5556 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5557 less than or equal to ``op2``.
5558 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5559 not equal to ``op2``.
5560 #. ``uno``: yields ``true`` if either operand is a QNAN.
5561 #. ``true``: always yields ``true``, regardless of operands.
5566 .. code-block:: llvm
5568 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5569 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5570 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5571 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5573 Note that the code generator does not yet support vector types with the
5574 ``fcmp`` instruction.
5578 '``phi``' Instruction
5579 ^^^^^^^^^^^^^^^^^^^^^
5586 <result> = phi <ty> [ <val0>, <label0>], ...
5591 The '``phi``' instruction is used to implement the φ node in the SSA
5592 graph representing the function.
5597 The type of the incoming values is specified with the first type field.
5598 After this, the '``phi``' instruction takes a list of pairs as
5599 arguments, with one pair for each predecessor basic block of the current
5600 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5601 the value arguments to the PHI node. Only labels may be used as the
5604 There must be no non-phi instructions between the start of a basic block
5605 and the PHI instructions: i.e. PHI instructions must be first in a basic
5608 For the purposes of the SSA form, the use of each incoming value is
5609 deemed to occur on the edge from the corresponding predecessor block to
5610 the current block (but after any definition of an '``invoke``'
5611 instruction's return value on the same edge).
5616 At runtime, the '``phi``' instruction logically takes on the value
5617 specified by the pair corresponding to the predecessor basic block that
5618 executed just prior to the current block.
5623 .. code-block:: llvm
5625 Loop: ; Infinite loop that counts from 0 on up...
5626 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5627 %nextindvar = add i32 %indvar, 1
5632 '``select``' Instruction
5633 ^^^^^^^^^^^^^^^^^^^^^^^^
5640 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5642 selty is either i1 or {<N x i1>}
5647 The '``select``' instruction is used to choose one value based on a
5648 condition, without branching.
5653 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
5654 values indicating the condition, and two values of the same :ref:`first
5655 class <t_firstclass>` type. If the val1/val2 are vectors and the
5656 condition is a scalar, then entire vectors are selected, not individual
5662 If the condition is an i1 and it evaluates to 1, the instruction returns
5663 the first value argument; otherwise, it returns the second value
5666 If the condition is a vector of i1, then the value arguments must be
5667 vectors of the same size, and the selection is done element by element.
5672 .. code-block:: llvm
5674 %X = select i1 true, i8 17, i8 42 ; yields i8:17
5678 '``call``' Instruction
5679 ^^^^^^^^^^^^^^^^^^^^^^
5686 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
5691 The '``call``' instruction represents a simple function call.
5696 This instruction requires several arguments:
5698 #. The optional "tail" marker indicates that the callee function does
5699 not access any allocas or varargs in the caller. Note that calls may
5700 be marked "tail" even if they do not occur before a
5701 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
5702 function call is eligible for tail call optimization, but `might not
5703 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
5704 The code generator may optimize calls marked "tail" with either 1)
5705 automatic `sibling call
5706 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
5707 callee have matching signatures, or 2) forced tail call optimization
5708 when the following extra requirements are met:
5710 - Caller and callee both have the calling convention ``fastcc``.
5711 - The call is in tail position (ret immediately follows call and ret
5712 uses value of call or is void).
5713 - Option ``-tailcallopt`` is enabled, or
5714 ``llvm::GuaranteedTailCallOpt`` is ``true``.
5715 - `Platform specific constraints are
5716 met. <CodeGenerator.html#tailcallopt>`_
5718 #. The optional "cconv" marker indicates which :ref:`calling
5719 convention <callingconv>` the call should use. If none is
5720 specified, the call defaults to using C calling conventions. The
5721 calling convention of the call must match the calling convention of
5722 the target function, or else the behavior is undefined.
5723 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5724 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5726 #. '``ty``': the type of the call instruction itself which is also the
5727 type of the return value. Functions that return no value are marked
5729 #. '``fnty``': shall be the signature of the pointer to function value
5730 being invoked. The argument types must match the types implied by
5731 this signature. This type can be omitted if the function is not
5732 varargs and if the function type does not return a pointer to a
5734 #. '``fnptrval``': An LLVM value containing a pointer to a function to
5735 be invoked. In most cases, this is a direct function invocation, but
5736 indirect ``call``'s are just as possible, calling an arbitrary pointer
5738 #. '``function args``': argument list whose types match the function
5739 signature argument types and parameter attributes. All arguments must
5740 be of :ref:`first class <t_firstclass>` type. If the function signature
5741 indicates the function accepts a variable number of arguments, the
5742 extra arguments can be specified.
5743 #. The optional :ref:`function attributes <fnattrs>` list. Only
5744 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5745 attributes are valid here.
5750 The '``call``' instruction is used to cause control flow to transfer to
5751 a specified function, with its incoming arguments bound to the specified
5752 values. Upon a '``ret``' instruction in the called function, control
5753 flow continues with the instruction after the function call, and the
5754 return value of the function is bound to the result argument.
5759 .. code-block:: llvm
5761 %retval = call i32 @test(i32 %argc)
5762 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
5763 %X = tail call i32 @foo() ; yields i32
5764 %Y = tail call fastcc i32 @foo() ; yields i32
5765 call void %foo(i8 97 signext)
5767 %struct.A = type { i32, i8 }
5768 %r = call %struct.A @foo() ; yields { 32, i8 }
5769 %gr = extractvalue %struct.A %r, 0 ; yields i32
5770 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
5771 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
5772 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
5774 llvm treats calls to some functions with names and arguments that match
5775 the standard C99 library as being the C99 library functions, and may
5776 perform optimizations or generate code for them under that assumption.
5777 This is something we'd like to change in the future to provide better
5778 support for freestanding environments and non-C-based languages.
5782 '``va_arg``' Instruction
5783 ^^^^^^^^^^^^^^^^^^^^^^^^
5790 <resultval> = va_arg <va_list*> <arglist>, <argty>
5795 The '``va_arg``' instruction is used to access arguments passed through
5796 the "variable argument" area of a function call. It is used to implement
5797 the ``va_arg`` macro in C.
5802 This instruction takes a ``va_list*`` value and the type of the
5803 argument. It returns a value of the specified argument type and
5804 increments the ``va_list`` to point to the next argument. The actual
5805 type of ``va_list`` is target specific.
5810 The '``va_arg``' instruction loads an argument of the specified type
5811 from the specified ``va_list`` and causes the ``va_list`` to point to
5812 the next argument. For more information, see the variable argument
5813 handling :ref:`Intrinsic Functions <int_varargs>`.
5815 It is legal for this instruction to be called in a function which does
5816 not take a variable number of arguments, for example, the ``vfprintf``
5819 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
5820 function <intrinsics>` because it takes a type as an argument.
5825 See the :ref:`variable argument processing <int_varargs>` section.
5827 Note that the code generator does not yet fully support va\_arg on many
5828 targets. Also, it does not currently support va\_arg with aggregate
5829 types on any target.
5833 '``landingpad``' Instruction
5834 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5841 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
5842 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
5844 <clause> := catch <type> <value>
5845 <clause> := filter <array constant type> <array constant>
5850 The '``landingpad``' instruction is used by `LLVM's exception handling
5851 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5852 is a landing pad --- one where the exception lands, and corresponds to the
5853 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
5854 defines values supplied by the personality function (``pers_fn``) upon
5855 re-entry to the function. The ``resultval`` has the type ``resultty``.
5860 This instruction takes a ``pers_fn`` value. This is the personality
5861 function associated with the unwinding mechanism. The optional
5862 ``cleanup`` flag indicates that the landing pad block is a cleanup.
5864 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
5865 contains the global variable representing the "type" that may be caught
5866 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
5867 clause takes an array constant as its argument. Use
5868 "``[0 x i8**] undef``" for a filter which cannot throw. The
5869 '``landingpad``' instruction must contain *at least* one ``clause`` or
5870 the ``cleanup`` flag.
5875 The '``landingpad``' instruction defines the values which are set by the
5876 personality function (``pers_fn``) upon re-entry to the function, and
5877 therefore the "result type" of the ``landingpad`` instruction. As with
5878 calling conventions, how the personality function results are
5879 represented in LLVM IR is target specific.
5881 The clauses are applied in order from top to bottom. If two
5882 ``landingpad`` instructions are merged together through inlining, the
5883 clauses from the calling function are appended to the list of clauses.
5884 When the call stack is being unwound due to an exception being thrown,
5885 the exception is compared against each ``clause`` in turn. If it doesn't
5886 match any of the clauses, and the ``cleanup`` flag is not set, then
5887 unwinding continues further up the call stack.
5889 The ``landingpad`` instruction has several restrictions:
5891 - A landing pad block is a basic block which is the unwind destination
5892 of an '``invoke``' instruction.
5893 - A landing pad block must have a '``landingpad``' instruction as its
5894 first non-PHI instruction.
5895 - There can be only one '``landingpad``' instruction within the landing
5897 - A basic block that is not a landing pad block may not include a
5898 '``landingpad``' instruction.
5899 - All '``landingpad``' instructions in a function must have the same
5900 personality function.
5905 .. code-block:: llvm
5907 ;; A landing pad which can catch an integer.
5908 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
5910 ;; A landing pad that is a cleanup.
5911 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
5913 ;; A landing pad which can catch an integer and can only throw a double.
5914 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
5916 filter [1 x i8**] [@_ZTId]
5923 LLVM supports the notion of an "intrinsic function". These functions
5924 have well known names and semantics and are required to follow certain
5925 restrictions. Overall, these intrinsics represent an extension mechanism
5926 for the LLVM language that does not require changing all of the
5927 transformations in LLVM when adding to the language (or the bitcode
5928 reader/writer, the parser, etc...).
5930 Intrinsic function names must all start with an "``llvm.``" prefix. This
5931 prefix is reserved in LLVM for intrinsic names; thus, function names may
5932 not begin with this prefix. Intrinsic functions must always be external
5933 functions: you cannot define the body of intrinsic functions. Intrinsic
5934 functions may only be used in call or invoke instructions: it is illegal
5935 to take the address of an intrinsic function. Additionally, because
5936 intrinsic functions are part of the LLVM language, it is required if any
5937 are added that they be documented here.
5939 Some intrinsic functions can be overloaded, i.e., the intrinsic
5940 represents a family of functions that perform the same operation but on
5941 different data types. Because LLVM can represent over 8 million
5942 different integer types, overloading is used commonly to allow an
5943 intrinsic function to operate on any integer type. One or more of the
5944 argument types or the result type can be overloaded to accept any
5945 integer type. Argument types may also be defined as exactly matching a
5946 previous argument's type or the result type. This allows an intrinsic
5947 function which accepts multiple arguments, but needs all of them to be
5948 of the same type, to only be overloaded with respect to a single
5949 argument or the result.
5951 Overloaded intrinsics will have the names of its overloaded argument
5952 types encoded into its function name, each preceded by a period. Only
5953 those types which are overloaded result in a name suffix. Arguments
5954 whose type is matched against another type do not. For example, the
5955 ``llvm.ctpop`` function can take an integer of any width and returns an
5956 integer of exactly the same integer width. This leads to a family of
5957 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
5958 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
5959 overloaded, and only one type suffix is required. Because the argument's
5960 type is matched against the return type, it does not require its own
5963 To learn how to add an intrinsic function, please see the `Extending
5964 LLVM Guide <ExtendingLLVM.html>`_.
5968 Variable Argument Handling Intrinsics
5969 -------------------------------------
5971 Variable argument support is defined in LLVM with the
5972 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
5973 functions. These functions are related to the similarly named macros
5974 defined in the ``<stdarg.h>`` header file.
5976 All of these functions operate on arguments that use a target-specific
5977 value type "``va_list``". The LLVM assembly language reference manual
5978 does not define what this type is, so all transformations should be
5979 prepared to handle these functions regardless of the type used.
5981 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
5982 variable argument handling intrinsic functions are used.
5984 .. code-block:: llvm
5986 define i32 @test(i32 %X, ...) {
5987 ; Initialize variable argument processing
5989 %ap2 = bitcast i8** %ap to i8*
5990 call void @llvm.va_start(i8* %ap2)
5992 ; Read a single integer argument
5993 %tmp = va_arg i8** %ap, i32
5995 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5997 %aq2 = bitcast i8** %aq to i8*
5998 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5999 call void @llvm.va_end(i8* %aq2)
6001 ; Stop processing of arguments.
6002 call void @llvm.va_end(i8* %ap2)
6006 declare void @llvm.va_start(i8*)
6007 declare void @llvm.va_copy(i8*, i8*)
6008 declare void @llvm.va_end(i8*)
6012 '``llvm.va_start``' Intrinsic
6013 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6020 declare void %llvm.va_start(i8* <arglist>)
6025 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6026 subsequent use by ``va_arg``.
6031 The argument is a pointer to a ``va_list`` element to initialize.
6036 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6037 available in C. In a target-dependent way, it initializes the
6038 ``va_list`` element to which the argument points, so that the next call
6039 to ``va_arg`` will produce the first variable argument passed to the
6040 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6041 to know the last argument of the function as the compiler can figure
6044 '``llvm.va_end``' Intrinsic
6045 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6052 declare void @llvm.va_end(i8* <arglist>)
6057 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6058 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6063 The argument is a pointer to a ``va_list`` to destroy.
6068 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6069 available in C. In a target-dependent way, it destroys the ``va_list``
6070 element to which the argument points. Calls to
6071 :ref:`llvm.va_start <int_va_start>` and
6072 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6077 '``llvm.va_copy``' Intrinsic
6078 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6085 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6090 The '``llvm.va_copy``' intrinsic copies the current argument position
6091 from the source argument list to the destination argument list.
6096 The first argument is a pointer to a ``va_list`` element to initialize.
6097 The second argument is a pointer to a ``va_list`` element to copy from.
6102 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6103 available in C. In a target-dependent way, it copies the source
6104 ``va_list`` element into the destination ``va_list`` element. This
6105 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6106 arbitrarily complex and require, for example, memory allocation.
6108 Accurate Garbage Collection Intrinsics
6109 --------------------------------------
6111 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6112 (GC) requires the implementation and generation of these intrinsics.
6113 These intrinsics allow identification of :ref:`GC roots on the
6114 stack <int_gcroot>`, as well as garbage collector implementations that
6115 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6116 Front-ends for type-safe garbage collected languages should generate
6117 these intrinsics to make use of the LLVM garbage collectors. For more
6118 details, see `Accurate Garbage Collection with
6119 LLVM <GarbageCollection.html>`_.
6121 The garbage collection intrinsics only operate on objects in the generic
6122 address space (address space zero).
6126 '``llvm.gcroot``' Intrinsic
6127 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6134 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6139 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6140 the code generator, and allows some metadata to be associated with it.
6145 The first argument specifies the address of a stack object that contains
6146 the root pointer. The second pointer (which must be either a constant or
6147 a global value address) contains the meta-data to be associated with the
6153 At runtime, a call to this intrinsic stores a null pointer into the
6154 "ptrloc" location. At compile-time, the code generator generates
6155 information to allow the runtime to find the pointer at GC safe points.
6156 The '``llvm.gcroot``' intrinsic may only be used in a function which
6157 :ref:`specifies a GC algorithm <gc>`.
6161 '``llvm.gcread``' Intrinsic
6162 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6169 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6174 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6175 locations, allowing garbage collector implementations that require read
6181 The second argument is the address to read from, which should be an
6182 address allocated from the garbage collector. The first object is a
6183 pointer to the start of the referenced object, if needed by the language
6184 runtime (otherwise null).
6189 The '``llvm.gcread``' intrinsic has the same semantics as a load
6190 instruction, but may be replaced with substantially more complex code by
6191 the garbage collector runtime, as needed. The '``llvm.gcread``'
6192 intrinsic may only be used in a function which :ref:`specifies a GC
6197 '``llvm.gcwrite``' Intrinsic
6198 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6205 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6210 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6211 locations, allowing garbage collector implementations that require write
6212 barriers (such as generational or reference counting collectors).
6217 The first argument is the reference to store, the second is the start of
6218 the object to store it to, and the third is the address of the field of
6219 Obj to store to. If the runtime does not require a pointer to the
6220 object, Obj may be null.
6225 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6226 instruction, but may be replaced with substantially more complex code by
6227 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6228 intrinsic may only be used in a function which :ref:`specifies a GC
6231 Code Generator Intrinsics
6232 -------------------------
6234 These intrinsics are provided by LLVM to expose special features that
6235 may only be implemented with code generator support.
6237 '``llvm.returnaddress``' Intrinsic
6238 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6245 declare i8 *@llvm.returnaddress(i32 <level>)
6250 The '``llvm.returnaddress``' intrinsic attempts to compute a
6251 target-specific value indicating the return address of the current
6252 function or one of its callers.
6257 The argument to this intrinsic indicates which function to return the
6258 address for. Zero indicates the calling function, one indicates its
6259 caller, etc. The argument is **required** to be a constant integer
6265 The '``llvm.returnaddress``' intrinsic either returns a pointer
6266 indicating the return address of the specified call frame, or zero if it
6267 cannot be identified. The value returned by this intrinsic is likely to
6268 be incorrect or 0 for arguments other than zero, so it should only be
6269 used for debugging purposes.
6271 Note that calling this intrinsic does not prevent function inlining or
6272 other aggressive transformations, so the value returned may not be that
6273 of the obvious source-language caller.
6275 '``llvm.frameaddress``' Intrinsic
6276 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6283 declare i8* @llvm.frameaddress(i32 <level>)
6288 The '``llvm.frameaddress``' intrinsic attempts to return the
6289 target-specific frame pointer value for the specified stack frame.
6294 The argument to this intrinsic indicates which function to return the
6295 frame pointer for. Zero indicates the calling function, one indicates
6296 its caller, etc. The argument is **required** to be a constant integer
6302 The '``llvm.frameaddress``' intrinsic either returns a pointer
6303 indicating the frame address of the specified call frame, or zero if it
6304 cannot be identified. The value returned by this intrinsic is likely to
6305 be incorrect or 0 for arguments other than zero, so it should only be
6306 used for debugging purposes.
6308 Note that calling this intrinsic does not prevent function inlining or
6309 other aggressive transformations, so the value returned may not be that
6310 of the obvious source-language caller.
6314 '``llvm.stacksave``' Intrinsic
6315 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6322 declare i8* @llvm.stacksave()
6327 The '``llvm.stacksave``' intrinsic is used to remember the current state
6328 of the function stack, for use with
6329 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6330 implementing language features like scoped automatic variable sized
6336 This intrinsic returns a opaque pointer value that can be passed to
6337 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6338 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6339 ``llvm.stacksave``, it effectively restores the state of the stack to
6340 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6341 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6342 were allocated after the ``llvm.stacksave`` was executed.
6344 .. _int_stackrestore:
6346 '``llvm.stackrestore``' Intrinsic
6347 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6354 declare void @llvm.stackrestore(i8* %ptr)
6359 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6360 the function stack to the state it was in when the corresponding
6361 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6362 useful for implementing language features like scoped automatic variable
6363 sized arrays in C99.
6368 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6370 '``llvm.prefetch``' Intrinsic
6371 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6378 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6383 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6384 insert a prefetch instruction if supported; otherwise, it is a noop.
6385 Prefetches have no effect on the behavior of the program but can change
6386 its performance characteristics.
6391 ``address`` is the address to be prefetched, ``rw`` is the specifier
6392 determining if the fetch should be for a read (0) or write (1), and
6393 ``locality`` is a temporal locality specifier ranging from (0) - no
6394 locality, to (3) - extremely local keep in cache. The ``cache type``
6395 specifies whether the prefetch is performed on the data (1) or
6396 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6397 arguments must be constant integers.
6402 This intrinsic does not modify the behavior of the program. In
6403 particular, prefetches cannot trap and do not produce a value. On
6404 targets that support this intrinsic, the prefetch can provide hints to
6405 the processor cache for better performance.
6407 '``llvm.pcmarker``' Intrinsic
6408 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6415 declare void @llvm.pcmarker(i32 <id>)
6420 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6421 Counter (PC) in a region of code to simulators and other tools. The
6422 method is target specific, but it is expected that the marker will use
6423 exported symbols to transmit the PC of the marker. The marker makes no
6424 guarantees that it will remain with any specific instruction after
6425 optimizations. It is possible that the presence of a marker will inhibit
6426 optimizations. The intended use is to be inserted after optimizations to
6427 allow correlations of simulation runs.
6432 ``id`` is a numerical id identifying the marker.
6437 This intrinsic does not modify the behavior of the program. Backends
6438 that do not support this intrinsic may ignore it.
6440 '``llvm.readcyclecounter``' Intrinsic
6441 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6448 declare i64 @llvm.readcyclecounter()
6453 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6454 counter register (or similar low latency, high accuracy clocks) on those
6455 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6456 should map to RPCC. As the backing counters overflow quickly (on the
6457 order of 9 seconds on alpha), this should only be used for small
6463 When directly supported, reading the cycle counter should not modify any
6464 memory. Implementations are allowed to either return a application
6465 specific value or a system wide value. On backends without support, this
6466 is lowered to a constant 0.
6468 Standard C Library Intrinsics
6469 -----------------------------
6471 LLVM provides intrinsics for a few important standard C library
6472 functions. These intrinsics allow source-language front-ends to pass
6473 information about the alignment of the pointer arguments to the code
6474 generator, providing opportunity for more efficient code generation.
6478 '``llvm.memcpy``' Intrinsic
6479 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6484 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6485 integer bit width and for different address spaces. Not all targets
6486 support all bit widths however.
6490 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6491 i32 <len>, i32 <align>, i1 <isvolatile>)
6492 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6493 i64 <len>, i32 <align>, i1 <isvolatile>)
6498 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6499 source location to the destination location.
6501 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6502 intrinsics do not return a value, takes extra alignment/isvolatile
6503 arguments and the pointers can be in specified address spaces.
6508 The first argument is a pointer to the destination, the second is a
6509 pointer to the source. The third argument is an integer argument
6510 specifying the number of bytes to copy, the fourth argument is the
6511 alignment of the source and destination locations, and the fifth is a
6512 boolean indicating a volatile access.
6514 If the call to this intrinsic has an alignment value that is not 0 or 1,
6515 then the caller guarantees that both the source and destination pointers
6516 are aligned to that boundary.
6518 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6519 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6520 very cleanly specified and it is unwise to depend on it.
6525 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6526 source location to the destination location, which are not allowed to
6527 overlap. It copies "len" bytes of memory over. If the argument is known
6528 to be aligned to some boundary, this can be specified as the fourth
6529 argument, otherwise it should be set to 0 or 1.
6531 '``llvm.memmove``' Intrinsic
6532 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6537 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6538 bit width and for different address space. Not all targets support all
6543 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6544 i32 <len>, i32 <align>, i1 <isvolatile>)
6545 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6546 i64 <len>, i32 <align>, i1 <isvolatile>)
6551 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6552 source location to the destination location. It is similar to the
6553 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6556 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6557 intrinsics do not return a value, takes extra alignment/isvolatile
6558 arguments and the pointers can be in specified address spaces.
6563 The first argument is a pointer to the destination, the second is a
6564 pointer to the source. The third argument is an integer argument
6565 specifying the number of bytes to copy, the fourth argument is the
6566 alignment of the source and destination locations, and the fifth is a
6567 boolean indicating a volatile access.
6569 If the call to this intrinsic has an alignment value that is not 0 or 1,
6570 then the caller guarantees that the source and destination pointers are
6571 aligned to that boundary.
6573 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6574 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6575 not very cleanly specified and it is unwise to depend on it.
6580 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6581 source location to the destination location, which may overlap. It
6582 copies "len" bytes of memory over. If the argument is known to be
6583 aligned to some boundary, this can be specified as the fourth argument,
6584 otherwise it should be set to 0 or 1.
6586 '``llvm.memset.*``' Intrinsics
6587 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6592 This is an overloaded intrinsic. You can use llvm.memset on any integer
6593 bit width and for different address spaces. However, not all targets
6594 support all bit widths.
6598 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6599 i32 <len>, i32 <align>, i1 <isvolatile>)
6600 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6601 i64 <len>, i32 <align>, i1 <isvolatile>)
6606 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6607 particular byte value.
6609 Note that, unlike the standard libc function, the ``llvm.memset``
6610 intrinsic does not return a value and takes extra alignment/volatile
6611 arguments. Also, the destination can be in an arbitrary address space.
6616 The first argument is a pointer to the destination to fill, the second
6617 is the byte value with which to fill it, the third argument is an
6618 integer argument specifying the number of bytes to fill, and the fourth
6619 argument is the known alignment of the destination location.
6621 If the call to this intrinsic has an alignment value that is not 0 or 1,
6622 then the caller guarantees that the destination pointer is aligned to
6625 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6626 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6627 very cleanly specified and it is unwise to depend on it.
6632 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6633 at the destination location. If the argument is known to be aligned to
6634 some boundary, this can be specified as the fourth argument, otherwise
6635 it should be set to 0 or 1.
6637 '``llvm.sqrt.*``' Intrinsic
6638 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6643 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6644 floating point or vector of floating point type. Not all targets support
6649 declare float @llvm.sqrt.f32(float %Val)
6650 declare double @llvm.sqrt.f64(double %Val)
6651 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6652 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6653 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6658 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
6659 returning the same value as the libm '``sqrt``' functions would. Unlike
6660 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
6661 negative numbers other than -0.0 (which allows for better optimization,
6662 because there is no need to worry about errno being set).
6663 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
6668 The argument and return value are floating point numbers of the same
6674 This function returns the sqrt of the specified operand if it is a
6675 nonnegative floating point number.
6677 '``llvm.powi.*``' Intrinsic
6678 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6683 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
6684 floating point or vector of floating point type. Not all targets support
6689 declare float @llvm.powi.f32(float %Val, i32 %power)
6690 declare double @llvm.powi.f64(double %Val, i32 %power)
6691 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6692 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6693 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6698 The '``llvm.powi.*``' intrinsics return the first operand raised to the
6699 specified (positive or negative) power. The order of evaluation of
6700 multiplications is not defined. When a vector of floating point type is
6701 used, the second argument remains a scalar integer value.
6706 The second argument is an integer power, and the first is a value to
6707 raise to that power.
6712 This function returns the first value raised to the second power with an
6713 unspecified sequence of rounding operations.
6715 '``llvm.sin.*``' Intrinsic
6716 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6721 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
6722 floating point or vector of floating point type. Not all targets support
6727 declare float @llvm.sin.f32(float %Val)
6728 declare double @llvm.sin.f64(double %Val)
6729 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6730 declare fp128 @llvm.sin.f128(fp128 %Val)
6731 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6736 The '``llvm.sin.*``' intrinsics return the sine of the operand.
6741 The argument and return value are floating point numbers of the same
6747 This function returns the sine of the specified operand, returning the
6748 same values as the libm ``sin`` functions would, and handles error
6749 conditions in the same way.
6751 '``llvm.cos.*``' Intrinsic
6752 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6757 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
6758 floating point or vector of floating point type. Not all targets support
6763 declare float @llvm.cos.f32(float %Val)
6764 declare double @llvm.cos.f64(double %Val)
6765 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6766 declare fp128 @llvm.cos.f128(fp128 %Val)
6767 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6772 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
6777 The argument and return value are floating point numbers of the same
6783 This function returns the cosine of the specified operand, returning the
6784 same values as the libm ``cos`` functions would, and handles error
6785 conditions in the same way.
6787 '``llvm.pow.*``' Intrinsic
6788 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6793 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
6794 floating point or vector of floating point type. Not all targets support
6799 declare float @llvm.pow.f32(float %Val, float %Power)
6800 declare double @llvm.pow.f64(double %Val, double %Power)
6801 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6802 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6803 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6808 The '``llvm.pow.*``' intrinsics return the first operand raised to the
6809 specified (positive or negative) power.
6814 The second argument is a floating point power, and the first is a value
6815 to raise to that power.
6820 This function returns the first value raised to the second power,
6821 returning the same values as the libm ``pow`` functions would, and
6822 handles error conditions in the same way.
6824 '``llvm.exp.*``' Intrinsic
6825 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6830 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
6831 floating point or vector of floating point type. Not all targets support
6836 declare float @llvm.exp.f32(float %Val)
6837 declare double @llvm.exp.f64(double %Val)
6838 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
6839 declare fp128 @llvm.exp.f128(fp128 %Val)
6840 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
6845 The '``llvm.exp.*``' intrinsics perform the exp function.
6850 The argument and return value are floating point numbers of the same
6856 This function returns the same values as the libm ``exp`` functions
6857 would, and handles error conditions in the same way.
6859 '``llvm.exp2.*``' Intrinsic
6860 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6865 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
6866 floating point or vector of floating point type. Not all targets support
6871 declare float @llvm.exp2.f32(float %Val)
6872 declare double @llvm.exp2.f64(double %Val)
6873 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
6874 declare fp128 @llvm.exp2.f128(fp128 %Val)
6875 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
6880 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
6885 The argument and return value are floating point numbers of the same
6891 This function returns the same values as the libm ``exp2`` functions
6892 would, and handles error conditions in the same way.
6894 '``llvm.log.*``' Intrinsic
6895 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6900 This is an overloaded intrinsic. You can use ``llvm.log`` on any
6901 floating point or vector of floating point type. Not all targets support
6906 declare float @llvm.log.f32(float %Val)
6907 declare double @llvm.log.f64(double %Val)
6908 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
6909 declare fp128 @llvm.log.f128(fp128 %Val)
6910 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
6915 The '``llvm.log.*``' intrinsics perform the log function.
6920 The argument and return value are floating point numbers of the same
6926 This function returns the same values as the libm ``log`` functions
6927 would, and handles error conditions in the same way.
6929 '``llvm.log10.*``' Intrinsic
6930 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6935 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
6936 floating point or vector of floating point type. Not all targets support
6941 declare float @llvm.log10.f32(float %Val)
6942 declare double @llvm.log10.f64(double %Val)
6943 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
6944 declare fp128 @llvm.log10.f128(fp128 %Val)
6945 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
6950 The '``llvm.log10.*``' intrinsics perform the log10 function.
6955 The argument and return value are floating point numbers of the same
6961 This function returns the same values as the libm ``log10`` functions
6962 would, and handles error conditions in the same way.
6964 '``llvm.log2.*``' Intrinsic
6965 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6970 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
6971 floating point or vector of floating point type. Not all targets support
6976 declare float @llvm.log2.f32(float %Val)
6977 declare double @llvm.log2.f64(double %Val)
6978 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
6979 declare fp128 @llvm.log2.f128(fp128 %Val)
6980 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
6985 The '``llvm.log2.*``' intrinsics perform the log2 function.
6990 The argument and return value are floating point numbers of the same
6996 This function returns the same values as the libm ``log2`` functions
6997 would, and handles error conditions in the same way.
6999 '``llvm.fma.*``' Intrinsic
7000 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7005 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7006 floating point or vector of floating point type. Not all targets support
7011 declare float @llvm.fma.f32(float %a, float %b, float %c)
7012 declare double @llvm.fma.f64(double %a, double %b, double %c)
7013 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7014 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7015 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7020 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7026 The argument and return value are floating point numbers of the same
7032 This function returns the same values as the libm ``fma`` functions
7035 '``llvm.fabs.*``' Intrinsic
7036 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7041 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7042 floating point or vector of floating point type. Not all targets support
7047 declare float @llvm.fabs.f32(float %Val)
7048 declare double @llvm.fabs.f64(double %Val)
7049 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7050 declare fp128 @llvm.fabs.f128(fp128 %Val)
7051 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7056 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7062 The argument and return value are floating point numbers of the same
7068 This function returns the same values as the libm ``fabs`` functions
7069 would, and handles error conditions in the same way.
7071 '``llvm.floor.*``' Intrinsic
7072 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7077 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7078 floating point or vector of floating point type. Not all targets support
7083 declare float @llvm.floor.f32(float %Val)
7084 declare double @llvm.floor.f64(double %Val)
7085 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7086 declare fp128 @llvm.floor.f128(fp128 %Val)
7087 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7092 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7097 The argument and return value are floating point numbers of the same
7103 This function returns the same values as the libm ``floor`` functions
7104 would, and handles error conditions in the same way.
7106 '``llvm.ceil.*``' Intrinsic
7107 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7112 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7113 floating point or vector of floating point type. Not all targets support
7118 declare float @llvm.ceil.f32(float %Val)
7119 declare double @llvm.ceil.f64(double %Val)
7120 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7121 declare fp128 @llvm.ceil.f128(fp128 %Val)
7122 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7127 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7132 The argument and return value are floating point numbers of the same
7138 This function returns the same values as the libm ``ceil`` functions
7139 would, and handles error conditions in the same way.
7141 '``llvm.trunc.*``' Intrinsic
7142 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7147 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7148 floating point or vector of floating point type. Not all targets support
7153 declare float @llvm.trunc.f32(float %Val)
7154 declare double @llvm.trunc.f64(double %Val)
7155 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7156 declare fp128 @llvm.trunc.f128(fp128 %Val)
7157 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7162 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7163 nearest integer not larger in magnitude than the operand.
7168 The argument and return value are floating point numbers of the same
7174 This function returns the same values as the libm ``trunc`` functions
7175 would, and handles error conditions in the same way.
7177 '``llvm.rint.*``' Intrinsic
7178 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7183 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7184 floating point or vector of floating point type. Not all targets support
7189 declare float @llvm.rint.f32(float %Val)
7190 declare double @llvm.rint.f64(double %Val)
7191 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7192 declare fp128 @llvm.rint.f128(fp128 %Val)
7193 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7198 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7199 nearest integer. It may raise an inexact floating-point exception if the
7200 operand isn't an integer.
7205 The argument and return value are floating point numbers of the same
7211 This function returns the same values as the libm ``rint`` functions
7212 would, and handles error conditions in the same way.
7214 '``llvm.nearbyint.*``' Intrinsic
7215 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7220 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7221 floating point or vector of floating point type. Not all targets support
7226 declare float @llvm.nearbyint.f32(float %Val)
7227 declare double @llvm.nearbyint.f64(double %Val)
7228 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7229 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7230 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7235 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7241 The argument and return value are floating point numbers of the same
7247 This function returns the same values as the libm ``nearbyint``
7248 functions would, and handles error conditions in the same way.
7250 Bit Manipulation Intrinsics
7251 ---------------------------
7253 LLVM provides intrinsics for a few important bit manipulation
7254 operations. These allow efficient code generation for some algorithms.
7256 '``llvm.bswap.*``' Intrinsics
7257 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7262 This is an overloaded intrinsic function. You can use bswap on any
7263 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7267 declare i16 @llvm.bswap.i16(i16 <id>)
7268 declare i32 @llvm.bswap.i32(i32 <id>)
7269 declare i64 @llvm.bswap.i64(i64 <id>)
7274 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7275 values with an even number of bytes (positive multiple of 16 bits).
7276 These are useful for performing operations on data that is not in the
7277 target's native byte order.
7282 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7283 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7284 intrinsic returns an i32 value that has the four bytes of the input i32
7285 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7286 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7287 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7288 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7291 '``llvm.ctpop.*``' Intrinsic
7292 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7297 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7298 bit width, or on any vector with integer elements. Not all targets
7299 support all bit widths or vector types, however.
7303 declare i8 @llvm.ctpop.i8(i8 <src>)
7304 declare i16 @llvm.ctpop.i16(i16 <src>)
7305 declare i32 @llvm.ctpop.i32(i32 <src>)
7306 declare i64 @llvm.ctpop.i64(i64 <src>)
7307 declare i256 @llvm.ctpop.i256(i256 <src>)
7308 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7313 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7319 The only argument is the value to be counted. The argument may be of any
7320 integer type, or a vector with integer elements. The return type must
7321 match the argument type.
7326 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7327 each element of a vector.
7329 '``llvm.ctlz.*``' Intrinsic
7330 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7335 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7336 integer bit width, or any vector whose elements are integers. Not all
7337 targets support all bit widths or vector types, however.
7341 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7342 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7343 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7344 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7345 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7346 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7351 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7352 leading zeros in a variable.
7357 The first argument is the value to be counted. This argument may be of
7358 any integer type, or a vectory with integer element type. The return
7359 type must match the first argument type.
7361 The second argument must be a constant and is a flag to indicate whether
7362 the intrinsic should ensure that a zero as the first argument produces a
7363 defined result. Historically some architectures did not provide a
7364 defined result for zero values as efficiently, and many algorithms are
7365 now predicated on avoiding zero-value inputs.
7370 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7371 zeros in a variable, or within each element of the vector. If
7372 ``src == 0`` then the result is the size in bits of the type of ``src``
7373 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7374 ``llvm.ctlz(i32 2) = 30``.
7376 '``llvm.cttz.*``' Intrinsic
7377 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7382 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7383 integer bit width, or any vector of integer elements. Not all targets
7384 support all bit widths or vector types, however.
7388 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7389 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7390 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7391 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7392 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7393 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7398 The '``llvm.cttz``' family of intrinsic functions counts the number of
7404 The first argument is the value to be counted. This argument may be of
7405 any integer type, or a vectory with integer element type. The return
7406 type must match the first argument type.
7408 The second argument must be a constant and is a flag to indicate whether
7409 the intrinsic should ensure that a zero as the first argument produces a
7410 defined result. Historically some architectures did not provide a
7411 defined result for zero values as efficiently, and many algorithms are
7412 now predicated on avoiding zero-value inputs.
7417 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7418 zeros in a variable, or within each element of a vector. If ``src == 0``
7419 then the result is the size in bits of the type of ``src`` if
7420 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7421 ``llvm.cttz(2) = 1``.
7423 Arithmetic with Overflow Intrinsics
7424 -----------------------------------
7426 LLVM provides intrinsics for some arithmetic with overflow operations.
7428 '``llvm.sadd.with.overflow.*``' Intrinsics
7429 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7434 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7435 on any integer bit width.
7439 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7440 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7441 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7446 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7447 a signed addition of the two arguments, and indicate whether an overflow
7448 occurred during the signed summation.
7453 The arguments (%a and %b) and the first element of the result structure
7454 may be of integer types of any bit width, but they must have the same
7455 bit width. The second element of the result structure must be of type
7456 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7462 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7463 a signed addition of the two variables. They return a structure --- the
7464 first element of which is the signed summation, and the second element
7465 of which is a bit specifying if the signed summation resulted in an
7471 .. code-block:: llvm
7473 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7474 %sum = extractvalue {i32, i1} %res, 0
7475 %obit = extractvalue {i32, i1} %res, 1
7476 br i1 %obit, label %overflow, label %normal
7478 '``llvm.uadd.with.overflow.*``' Intrinsics
7479 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7484 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7485 on any integer bit width.
7489 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7490 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7491 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7496 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7497 an unsigned addition of the two arguments, and indicate whether a carry
7498 occurred during the unsigned summation.
7503 The arguments (%a and %b) and the first element of the result structure
7504 may be of integer types of any bit width, but they must have the same
7505 bit width. The second element of the result structure must be of type
7506 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7512 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7513 an unsigned addition of the two arguments. They return a structure --- the
7514 first element of which is the sum, and the second element of which is a
7515 bit specifying if the unsigned summation resulted in a carry.
7520 .. code-block:: llvm
7522 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7523 %sum = extractvalue {i32, i1} %res, 0
7524 %obit = extractvalue {i32, i1} %res, 1
7525 br i1 %obit, label %carry, label %normal
7527 '``llvm.ssub.with.overflow.*``' Intrinsics
7528 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7533 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7534 on any integer bit width.
7538 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7539 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7540 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7545 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7546 a signed subtraction of the two arguments, and indicate whether an
7547 overflow occurred during the signed subtraction.
7552 The arguments (%a and %b) and the first element of the result structure
7553 may be of integer types of any bit width, but they must have the same
7554 bit width. The second element of the result structure must be of type
7555 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7561 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7562 a signed subtraction of the two arguments. They return a structure --- the
7563 first element of which is the subtraction, and the second element of
7564 which is a bit specifying if the signed subtraction resulted in an
7570 .. code-block:: llvm
7572 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7573 %sum = extractvalue {i32, i1} %res, 0
7574 %obit = extractvalue {i32, i1} %res, 1
7575 br i1 %obit, label %overflow, label %normal
7577 '``llvm.usub.with.overflow.*``' Intrinsics
7578 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7583 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
7584 on any integer bit width.
7588 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7589 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7590 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7595 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7596 an unsigned subtraction of the two arguments, and indicate whether an
7597 overflow occurred during the unsigned subtraction.
7602 The arguments (%a and %b) and the first element of the result structure
7603 may be of integer types of any bit width, but they must have the same
7604 bit width. The second element of the result structure must be of type
7605 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7611 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7612 an unsigned subtraction of the two arguments. They return a structure ---
7613 the first element of which is the subtraction, and the second element of
7614 which is a bit specifying if the unsigned subtraction resulted in an
7620 .. code-block:: llvm
7622 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7623 %sum = extractvalue {i32, i1} %res, 0
7624 %obit = extractvalue {i32, i1} %res, 1
7625 br i1 %obit, label %overflow, label %normal
7627 '``llvm.smul.with.overflow.*``' Intrinsics
7628 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7633 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
7634 on any integer bit width.
7638 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7639 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7640 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7645 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7646 a signed multiplication of the two arguments, and indicate whether an
7647 overflow occurred during the signed multiplication.
7652 The arguments (%a and %b) and the first element of the result structure
7653 may be of integer types of any bit width, but they must have the same
7654 bit width. The second element of the result structure must be of type
7655 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7661 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7662 a signed multiplication of the two arguments. They return a structure ---
7663 the first element of which is the multiplication, and the second element
7664 of which is a bit specifying if the signed multiplication resulted in an
7670 .. code-block:: llvm
7672 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7673 %sum = extractvalue {i32, i1} %res, 0
7674 %obit = extractvalue {i32, i1} %res, 1
7675 br i1 %obit, label %overflow, label %normal
7677 '``llvm.umul.with.overflow.*``' Intrinsics
7678 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7683 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
7684 on any integer bit width.
7688 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7689 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7690 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7695 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7696 a unsigned multiplication of the two arguments, and indicate whether an
7697 overflow occurred during the unsigned multiplication.
7702 The arguments (%a and %b) and the first element of the result structure
7703 may be of integer types of any bit width, but they must have the same
7704 bit width. The second element of the result structure must be of type
7705 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7711 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7712 an unsigned multiplication of the two arguments. They return a structure ---
7713 the first element of which is the multiplication, and the second
7714 element of which is a bit specifying if the unsigned multiplication
7715 resulted in an overflow.
7720 .. code-block:: llvm
7722 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7723 %sum = extractvalue {i32, i1} %res, 0
7724 %obit = extractvalue {i32, i1} %res, 1
7725 br i1 %obit, label %overflow, label %normal
7727 Specialised Arithmetic Intrinsics
7728 ---------------------------------
7730 '``llvm.fmuladd.*``' Intrinsic
7731 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7738 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
7739 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
7744 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
7745 expressions that can be fused if the code generator determines that (a) the
7746 target instruction set has support for a fused operation, and (b) that the
7747 fused operation is more efficient than the equivalent, separate pair of mul
7748 and add instructions.
7753 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
7754 multiplicands, a and b, and an addend c.
7763 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
7765 is equivalent to the expression a \* b + c, except that rounding will
7766 not be performed between the multiplication and addition steps if the
7767 code generator fuses the operations. Fusion is not guaranteed, even if
7768 the target platform supports it. If a fused multiply-add is required the
7769 corresponding llvm.fma.\* intrinsic function should be used instead.
7774 .. code-block:: llvm
7776 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
7778 Half Precision Floating Point Intrinsics
7779 ----------------------------------------
7781 For most target platforms, half precision floating point is a
7782 storage-only format. This means that it is a dense encoding (in memory)
7783 but does not support computation in the format.
7785 This means that code must first load the half-precision floating point
7786 value as an i16, then convert it to float with
7787 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
7788 then be performed on the float value (including extending to double
7789 etc). To store the value back to memory, it is first converted to float
7790 if needed, then converted to i16 with
7791 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
7794 .. _int_convert_to_fp16:
7796 '``llvm.convert.to.fp16``' Intrinsic
7797 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7804 declare i16 @llvm.convert.to.fp16(f32 %a)
7809 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
7810 from single precision floating point format to half precision floating
7816 The intrinsic function contains single argument - the value to be
7822 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
7823 from single precision floating point format to half precision floating
7824 point format. The return value is an ``i16`` which contains the
7830 .. code-block:: llvm
7832 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7833 store i16 %res, i16* @x, align 2
7835 .. _int_convert_from_fp16:
7837 '``llvm.convert.from.fp16``' Intrinsic
7838 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7845 declare f32 @llvm.convert.from.fp16(i16 %a)
7850 The '``llvm.convert.from.fp16``' intrinsic function performs a
7851 conversion from half precision floating point format to single precision
7852 floating point format.
7857 The intrinsic function contains single argument - the value to be
7863 The '``llvm.convert.from.fp16``' intrinsic function performs a
7864 conversion from half single precision floating point format to single
7865 precision floating point format. The input half-float value is
7866 represented by an ``i16`` value.
7871 .. code-block:: llvm
7873 %a = load i16* @x, align 2
7874 %res = call f32 @llvm.convert.from.fp16(i16 %a)
7879 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
7880 prefix), are described in the `LLVM Source Level
7881 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
7884 Exception Handling Intrinsics
7885 -----------------------------
7887 The LLVM exception handling intrinsics (which all start with
7888 ``llvm.eh.`` prefix), are described in the `LLVM Exception
7889 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
7893 Trampoline Intrinsics
7894 ---------------------
7896 These intrinsics make it possible to excise one parameter, marked with
7897 the :ref:`nest <nest>` attribute, from a function. The result is a
7898 callable function pointer lacking the nest parameter - the caller does
7899 not need to provide a value for it. Instead, the value to use is stored
7900 in advance in a "trampoline", a block of memory usually allocated on the
7901 stack, which also contains code to splice the nest value into the
7902 argument list. This is used to implement the GCC nested function address
7905 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
7906 then the resulting function pointer has signature ``i32 (i32, i32)*``.
7907 It can be created as follows:
7909 .. code-block:: llvm
7911 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
7912 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
7913 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
7914 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
7915 %fp = bitcast i8* %p to i32 (i32, i32)*
7917 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
7918 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
7922 '``llvm.init.trampoline``' Intrinsic
7923 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7930 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
7935 This fills the memory pointed to by ``tramp`` with executable code,
7936 turning it into a trampoline.
7941 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
7942 pointers. The ``tramp`` argument must point to a sufficiently large and
7943 sufficiently aligned block of memory; this memory is written to by the
7944 intrinsic. Note that the size and the alignment are target-specific -
7945 LLVM currently provides no portable way of determining them, so a
7946 front-end that generates this intrinsic needs to have some
7947 target-specific knowledge. The ``func`` argument must hold a function
7948 bitcast to an ``i8*``.
7953 The block of memory pointed to by ``tramp`` is filled with target
7954 dependent code, turning it into a function. Then ``tramp`` needs to be
7955 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
7956 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
7957 function's signature is the same as that of ``func`` with any arguments
7958 marked with the ``nest`` attribute removed. At most one such ``nest``
7959 argument is allowed, and it must be of pointer type. Calling the new
7960 function is equivalent to calling ``func`` with the same argument list,
7961 but with ``nval`` used for the missing ``nest`` argument. If, after
7962 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
7963 modified, then the effect of any later call to the returned function
7964 pointer is undefined.
7968 '``llvm.adjust.trampoline``' Intrinsic
7969 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7976 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
7981 This performs any required machine-specific adjustment to the address of
7982 a trampoline (passed as ``tramp``).
7987 ``tramp`` must point to a block of memory which already has trampoline
7988 code filled in by a previous call to
7989 :ref:`llvm.init.trampoline <int_it>`.
7994 On some architectures the address of the code to be executed needs to be
7995 different to the address where the trampoline is actually stored. This
7996 intrinsic returns the executable address corresponding to ``tramp``
7997 after performing the required machine specific adjustments. The pointer
7998 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8003 This class of intrinsics exists to information about the lifetime of
8004 memory objects and ranges where variables are immutable.
8006 '``llvm.lifetime.start``' Intrinsic
8007 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8014 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8019 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8025 The first argument is a constant integer representing the size of the
8026 object, or -1 if it is variable sized. The second argument is a pointer
8032 This intrinsic indicates that before this point in the code, the value
8033 of the memory pointed to by ``ptr`` is dead. This means that it is known
8034 to never be used and has an undefined value. A load from the pointer
8035 that precedes this intrinsic can be replaced with ``'undef'``.
8037 '``llvm.lifetime.end``' Intrinsic
8038 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8045 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8050 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8056 The first argument is a constant integer representing the size of the
8057 object, or -1 if it is variable sized. The second argument is a pointer
8063 This intrinsic indicates that after this point in the code, the value of
8064 the memory pointed to by ``ptr`` is dead. This means that it is known to
8065 never be used and has an undefined value. Any stores into the memory
8066 object following this intrinsic may be removed as dead.
8068 '``llvm.invariant.start``' Intrinsic
8069 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8076 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8081 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8082 a memory object will not change.
8087 The first argument is a constant integer representing the size of the
8088 object, or -1 if it is variable sized. The second argument is a pointer
8094 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8095 the return value, the referenced memory location is constant and
8098 '``llvm.invariant.end``' Intrinsic
8099 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8106 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8111 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8112 memory object are mutable.
8117 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8118 The second argument is a constant integer representing the size of the
8119 object, or -1 if it is variable sized and the third argument is a
8120 pointer to the object.
8125 This intrinsic indicates that the memory is mutable again.
8130 This class of intrinsics is designed to be generic and has no specific
8133 '``llvm.var.annotation``' Intrinsic
8134 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8141 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8146 The '``llvm.var.annotation``' intrinsic.
8151 The first argument is a pointer to a value, the second is a pointer to a
8152 global string, the third is a pointer to a global string which is the
8153 source file name, and the last argument is the line number.
8158 This intrinsic allows annotation of local variables with arbitrary
8159 strings. This can be useful for special purpose optimizations that want
8160 to look for these annotations. These have no other defined use; they are
8161 ignored by code generation and optimization.
8163 '``llvm.annotation.*``' Intrinsic
8164 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8169 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8170 any integer bit width.
8174 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8175 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8176 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8177 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8178 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8183 The '``llvm.annotation``' intrinsic.
8188 The first argument is an integer value (result of some expression), the
8189 second is a pointer to a global string, the third is a pointer to a
8190 global string which is the source file name, and the last argument is
8191 the line number. It returns the value of the first argument.
8196 This intrinsic allows annotations to be put on arbitrary expressions
8197 with arbitrary strings. This can be useful for special purpose
8198 optimizations that want to look for these annotations. These have no
8199 other defined use; they are ignored by code generation and optimization.
8201 '``llvm.trap``' Intrinsic
8202 ^^^^^^^^^^^^^^^^^^^^^^^^^
8209 declare void @llvm.trap() noreturn nounwind
8214 The '``llvm.trap``' intrinsic.
8224 This intrinsic is lowered to the target dependent trap instruction. If
8225 the target does not have a trap instruction, this intrinsic will be
8226 lowered to a call of the ``abort()`` function.
8228 '``llvm.debugtrap``' Intrinsic
8229 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8236 declare void @llvm.debugtrap() nounwind
8241 The '``llvm.debugtrap``' intrinsic.
8251 This intrinsic is lowered to code which is intended to cause an
8252 execution trap with the intention of requesting the attention of a
8255 '``llvm.stackprotector``' Intrinsic
8256 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8263 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8268 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8269 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8270 is placed on the stack before local variables.
8275 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8276 The first argument is the value loaded from the stack guard
8277 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8278 enough space to hold the value of the guard.
8283 This intrinsic causes the prologue/epilogue inserter to force the
8284 position of the ``AllocaInst`` stack slot to be before local variables
8285 on the stack. This is to ensure that if a local variable on the stack is
8286 overwritten, it will destroy the value of the guard. When the function
8287 exits, the guard on the stack is checked against the original guard. If
8288 they are different, then the program aborts by calling the
8289 ``__stack_chk_fail()`` function.
8291 '``llvm.objectsize``' Intrinsic
8292 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8299 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8300 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8305 The ``llvm.objectsize`` intrinsic is designed to provide information to
8306 the optimizers to determine at compile time whether a) an operation
8307 (like memcpy) will overflow a buffer that corresponds to an object, or
8308 b) that a runtime check for overflow isn't necessary. An object in this
8309 context means an allocation of a specific class, structure, array, or
8315 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8316 argument is a pointer to or into the ``object``. The second argument is
8317 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8318 or -1 (if false) when the object size is unknown. The second argument
8319 only accepts constants.
8324 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8325 the size of the object concerned. If the size cannot be determined at
8326 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8327 on the ``min`` argument).
8329 '``llvm.expect``' Intrinsic
8330 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8337 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8338 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8343 The ``llvm.expect`` intrinsic provides information about expected (the
8344 most probable) value of ``val``, which can be used by optimizers.
8349 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8350 a value. The second argument is an expected value, this needs to be a
8351 constant value, variables are not allowed.
8356 This intrinsic is lowered to the ``val``.
8358 '``llvm.donothing``' Intrinsic
8359 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8366 declare void @llvm.donothing() nounwind readnone
8371 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8372 only intrinsic that can be called with an invoke instruction.
8382 This intrinsic does nothing, and it's removed by optimizers and ignored