1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][a-zA-Z$._][a-zA-Z$._0-9]*``'. Identifiers which require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves.
83 #. Unnamed values are represented as an unsigned numeric value with
84 their prefix. For example, ``%12``, ``@2``, ``%44``.
85 #. Constants, which are described in the section Constants_ below.
87 LLVM requires that values start with a prefix for two reasons: Compilers
88 don't need to worry about name clashes with reserved words, and the set
89 of reserved words may be expanded in the future without penalty.
90 Additionally, unnamed identifiers allow a compiler to quickly come up
91 with a temporary variable without having to avoid symbol table
94 Reserved words in LLVM are very similar to reserved words in other
95 languages. There are keywords for different opcodes ('``add``',
96 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
97 '``i32``', etc...), and others. These reserved words cannot conflict
98 with variable names, because none of them start with a prefix character
101 Here is an example of LLVM code to multiply the integer variable
108 %result = mul i32 %X, 8
110 After strength reduction:
114 %result = shl i32 %X, 3
120 %0 = add i32 %X, %X ; yields {i32}:%0
121 %1 = add i32 %0, %0 ; yields {i32}:%1
122 %result = add i32 %1, %1
124 This last way of multiplying ``%X`` by 8 illustrates several important
125 lexical features of LLVM:
127 #. Comments are delimited with a '``;``' and go until the end of line.
128 #. Unnamed temporaries are created when the result of a computation is
129 not assigned to a named value.
130 #. Unnamed temporaries are numbered sequentially (using a per-function
131 incrementing counter, starting with 0).
133 It also shows a convention that we follow in this document. When
134 demonstrating instructions, we will follow an instruction with a comment
135 that defines the type and name of value produced.
143 LLVM programs are composed of ``Module``'s, each of which is a
144 translation unit of the input programs. Each module consists of
145 functions, global variables, and symbol table entries. Modules may be
146 combined together with the LLVM linker, which merges function (and
147 global variable) definitions, resolves forward declarations, and merges
148 symbol table entries. Here is an example of the "hello world" module:
152 ; Declare the string constant as a global constant.
153 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
155 ; External declaration of the puts function
156 declare i32 @puts(i8* nocapture) nounwind
158 ; Definition of main function
159 define i32 @main() { ; i32()*
160 ; Convert [13 x i8]* to i8 *...
161 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
163 ; Call puts function to write out the string to stdout.
164 call i32 @puts(i8* %cast210)
169 !1 = metadata !{i32 42}
172 This example is made up of a :ref:`global variable <globalvars>` named
173 "``.str``", an external declaration of the "``puts``" function, a
174 :ref:`function definition <functionstructure>` for "``main``" and
175 :ref:`named metadata <namedmetadatastructure>` "``foo``".
177 In general, a module is made up of a list of global values (where both
178 functions and global variables are global values). Global values are
179 represented by a pointer to a memory location (in this case, a pointer
180 to an array of char, and a pointer to a function), and have one of the
181 following :ref:`linkage types <linkage>`.
188 All Global Variables and Functions have one of the following types of
192 Global values with "``private``" linkage are only directly
193 accessible by objects in the current module. In particular, linking
194 code into a module with an private global value may cause the
195 private to be renamed as necessary to avoid collisions. Because the
196 symbol is private to the module, all references can be updated. This
197 doesn't show up in any symbol table in the object file.
199 Similar to ``private``, but the symbol is passed through the
200 assembler and evaluated by the linker. Unlike normal strong symbols,
201 they are removed by the linker from the final linked image
202 (executable or dynamic library).
203 ``linker_private_weak``
204 Similar to "``linker_private``", but the symbol is weak. Note that
205 ``linker_private_weak`` symbols are subject to coalescing by the
206 linker. The symbols are removed by the linker from the final linked
207 image (executable or dynamic library).
209 Similar to private, but the value shows as a local symbol
210 (``STB_LOCAL`` in the case of ELF) in the object file. This
211 corresponds to the notion of the '``static``' keyword in C.
212 ``available_externally``
213 Globals with "``available_externally``" linkage are never emitted
214 into the object file corresponding to the LLVM module. They exist to
215 allow inlining and other optimizations to take place given knowledge
216 of the definition of the global, which is known to be somewhere
217 outside the module. Globals with ``available_externally`` linkage
218 are allowed to be discarded at will, and are otherwise the same as
219 ``linkonce_odr``. This linkage type is only allowed on definitions,
222 Globals with "``linkonce``" linkage are merged with other globals of
223 the same name when linkage occurs. This can be used to implement
224 some forms of inline functions, templates, or other code which must
225 be generated in each translation unit that uses it, but where the
226 body may be overridden with a more definitive definition later.
227 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
228 that ``linkonce`` linkage does not actually allow the optimizer to
229 inline the body of this function into callers because it doesn't
230 know if this definition of the function is the definitive definition
231 within the program or whether it will be overridden by a stronger
232 definition. To enable inlining and other optimizations, use
233 "``linkonce_odr``" linkage.
235 "``weak``" linkage has the same merging semantics as ``linkonce``
236 linkage, except that unreferenced globals with ``weak`` linkage may
237 not be discarded. This is used for globals that are declared "weak"
240 "``common``" linkage is most similar to "``weak``" linkage, but they
241 are used for tentative definitions in C, such as "``int X;``" at
242 global scope. Symbols with "``common``" linkage are merged in the
243 same way as ``weak symbols``, and they may not be deleted if
244 unreferenced. ``common`` symbols may not have an explicit section,
245 must have a zero initializer, and may not be marked
246 ':ref:`constant <globalvars>`'. Functions and aliases may not have
249 .. _linkage_appending:
252 "``appending``" linkage may only be applied to global variables of
253 pointer to array type. When two global variables with appending
254 linkage are linked together, the two global arrays are appended
255 together. This is the LLVM, typesafe, equivalent of having the
256 system linker append together "sections" with identical names when
259 The semantics of this linkage follow the ELF object file model: the
260 symbol is weak until linked, if not linked, the symbol becomes null
261 instead of being an undefined reference.
262 ``linkonce_odr``, ``weak_odr``
263 Some languages allow differing globals to be merged, such as two
264 functions with different semantics. Other languages, such as
265 ``C++``, ensure that only equivalent globals are ever merged (the
266 "one definition rule" --- "ODR"). Such languages can use the
267 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
268 global will only be merged with equivalent globals. These linkage
269 types are otherwise the same as their non-``odr`` versions.
270 ``linkonce_odr_auto_hide``
271 Similar to "``linkonce_odr``", but nothing in the translation unit
272 takes the address of this definition. For instance, functions that
273 had an inline definition, but the compiler decided not to inline it.
274 ``linkonce_odr_auto_hide`` may have only ``default`` visibility. The
275 symbols are removed by the linker from the final linked image
276 (executable or dynamic library).
278 If none of the above identifiers are used, the global is externally
279 visible, meaning that it participates in linkage and can be used to
280 resolve external symbol references.
282 The next two types of linkage are targeted for Microsoft Windows
283 platform only. They are designed to support importing (exporting)
284 symbols from (to) DLLs (Dynamic Link Libraries).
287 "``dllimport``" linkage causes the compiler to reference a function
288 or variable via a global pointer to a pointer that is set up by the
289 DLL exporting the symbol. On Microsoft Windows targets, the pointer
290 name is formed by combining ``__imp_`` and the function or variable
293 "``dllexport``" linkage causes the compiler to provide a global
294 pointer to a pointer in a DLL, so that it can be referenced with the
295 ``dllimport`` attribute. On Microsoft Windows targets, the pointer
296 name is formed by combining ``__imp_`` and the function or variable
299 For example, since the "``.LC0``" variable is defined to be internal, if
300 another module defined a "``.LC0``" variable and was linked with this
301 one, one of the two would be renamed, preventing a collision. Since
302 "``main``" and "``puts``" are external (i.e., lacking any linkage
303 declarations), they are accessible outside of the current module.
305 It is illegal for a function *declaration* to have any linkage type
306 other than ``external``, ``dllimport`` or ``extern_weak``.
308 Aliases can have only ``external``, ``internal``, ``weak`` or
309 ``weak_odr`` linkages.
316 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
317 :ref:`invokes <i_invoke>` can all have an optional calling convention
318 specified for the call. The calling convention of any pair of dynamic
319 caller/callee must match, or the behavior of the program is undefined.
320 The following calling conventions are supported by LLVM, and more may be
323 "``ccc``" - The C calling convention
324 This calling convention (the default if no other calling convention
325 is specified) matches the target C calling conventions. This calling
326 convention supports varargs function calls and tolerates some
327 mismatch in the declared prototype and implemented declaration of
328 the function (as does normal C).
329 "``fastcc``" - The fast calling convention
330 This calling convention attempts to make calls as fast as possible
331 (e.g. by passing things in registers). This calling convention
332 allows the target to use whatever tricks it wants to produce fast
333 code for the target, without having to conform to an externally
334 specified ABI (Application Binary Interface). `Tail calls can only
335 be optimized when this, the GHC or the HiPE convention is
336 used. <CodeGenerator.html#id80>`_ This calling convention does not
337 support varargs and requires the prototype of all callees to exactly
338 match the prototype of the function definition.
339 "``coldcc``" - The cold calling convention
340 This calling convention attempts to make code in the caller as
341 efficient as possible under the assumption that the call is not
342 commonly executed. As such, these calls often preserve all registers
343 so that the call does not break any live ranges in the caller side.
344 This calling convention does not support varargs and requires the
345 prototype of all callees to exactly match the prototype of the
347 "``cc 10``" - GHC convention
348 This calling convention has been implemented specifically for use by
349 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
350 It passes everything in registers, going to extremes to achieve this
351 by disabling callee save registers. This calling convention should
352 not be used lightly but only for specific situations such as an
353 alternative to the *register pinning* performance technique often
354 used when implementing functional programming languages. At the
355 moment only X86 supports this convention and it has the following
358 - On *X86-32* only supports up to 4 bit type parameters. No
359 floating point types are supported.
360 - On *X86-64* only supports up to 10 bit type parameters and 6
361 floating point parameters.
363 This calling convention supports `tail call
364 optimization <CodeGenerator.html#id80>`_ but requires both the
365 caller and callee are using it.
366 "``cc 11``" - The HiPE calling convention
367 This calling convention has been implemented specifically for use by
368 the `High-Performance Erlang
369 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
370 native code compiler of the `Ericsson's Open Source Erlang/OTP
371 system <http://www.erlang.org/download.shtml>`_. It uses more
372 registers for argument passing than the ordinary C calling
373 convention and defines no callee-saved registers. The calling
374 convention properly supports `tail call
375 optimization <CodeGenerator.html#id80>`_ but requires that both the
376 caller and the callee use it. It uses a *register pinning*
377 mechanism, similar to GHC's convention, for keeping frequently
378 accessed runtime components pinned to specific hardware registers.
379 At the moment only X86 supports this convention (both 32 and 64
381 "``cc <n>``" - Numbered convention
382 Any calling convention may be specified by number, allowing
383 target-specific calling conventions to be used. Target specific
384 calling conventions start at 64.
386 More calling conventions can be added/defined on an as-needed basis, to
387 support Pascal conventions or any other well-known target-independent
390 .. _visibilitystyles:
395 All Global Variables and Functions have one of the following visibility
398 "``default``" - Default style
399 On targets that use the ELF object file format, default visibility
400 means that the declaration is visible to other modules and, in
401 shared libraries, means that the declared entity may be overridden.
402 On Darwin, default visibility means that the declaration is visible
403 to other modules. Default visibility corresponds to "external
404 linkage" in the language.
405 "``hidden``" - Hidden style
406 Two declarations of an object with hidden visibility refer to the
407 same object if they are in the same shared object. Usually, hidden
408 visibility indicates that the symbol will not be placed into the
409 dynamic symbol table, so no other module (executable or shared
410 library) can reference it directly.
411 "``protected``" - Protected style
412 On ELF, protected visibility indicates that the symbol will be
413 placed in the dynamic symbol table, but that references within the
414 defining module will bind to the local symbol. That is, the symbol
415 cannot be overridden by another module.
422 LLVM IR allows you to specify name aliases for certain types. This can
423 make it easier to read the IR and make the IR more condensed
424 (particularly when recursive types are involved). An example of a name
429 %mytype = type { %mytype*, i32 }
431 You may give a name to any :ref:`type <typesystem>` except
432 ":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
433 expected with the syntax "%mytype".
435 Note that type names are aliases for the structural type that they
436 indicate, and that you can therefore specify multiple names for the same
437 type. This often leads to confusing behavior when dumping out a .ll
438 file. Since LLVM IR uses structural typing, the name is not part of the
439 type. When printing out LLVM IR, the printer will pick *one name* to
440 render all types of a particular shape. This means that if you have code
441 where two different source types end up having the same LLVM type, that
442 the dumper will sometimes print the "wrong" or unexpected type. This is
443 an important design point and isn't going to change.
450 Global variables define regions of memory allocated at compilation time
451 instead of run-time. Global variables may optionally be initialized, may
452 have an explicit section to be placed in, and may have an optional
453 explicit alignment specified.
455 A variable may be defined as ``thread_local``, which means that it will
456 not be shared by threads (each thread will have a separated copy of the
457 variable). Not all targets support thread-local variables. Optionally, a
458 TLS model may be specified:
461 For variables that are only used within the current shared library.
463 For variables in modules that will not be loaded dynamically.
465 For variables defined in the executable and only used within it.
467 The models correspond to the ELF TLS models; see `ELF Handling For
468 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
469 more information on under which circumstances the different models may
470 be used. The target may choose a different TLS model if the specified
471 model is not supported, or if a better choice of model can be made.
473 A variable may be defined as a global ``constant``, which indicates that
474 the contents of the variable will **never** be modified (enabling better
475 optimization, allowing the global data to be placed in the read-only
476 section of an executable, etc). Note that variables that need runtime
477 initialization cannot be marked ``constant`` as there is a store to the
480 LLVM explicitly allows *declarations* of global variables to be marked
481 constant, even if the final definition of the global is not. This
482 capability can be used to enable slightly better optimization of the
483 program, but requires the language definition to guarantee that
484 optimizations based on the 'constantness' are valid for the translation
485 units that do not include the definition.
487 As SSA values, global variables define pointer values that are in scope
488 (i.e. they dominate) all basic blocks in the program. Global variables
489 always define a pointer to their "content" type because they describe a
490 region of memory, and all memory objects in LLVM are accessed through
493 Global variables can be marked with ``unnamed_addr`` which indicates
494 that the address is not significant, only the content. Constants marked
495 like this can be merged with other constants if they have the same
496 initializer. Note that a constant with significant address *can* be
497 merged with a ``unnamed_addr`` constant, the result being a constant
498 whose address is significant.
500 A global variable may be declared to reside in a target-specific
501 numbered address space. For targets that support them, address spaces
502 may affect how optimizations are performed and/or what target
503 instructions are used to access the variable. The default address space
504 is zero. The address space qualifier must precede any other attributes.
506 LLVM allows an explicit section to be specified for globals. If the
507 target supports it, it will emit globals to the section specified.
509 By default, global initializers are optimized by assuming that global
510 variables defined within the module are not modified from their
511 initial values before the start of the global initializer. This is
512 true even for variables potentially accessible from outside the
513 module, including those with external linkage or appearing in
514 ``@llvm.used``. This assumption may be suppressed by marking the
515 variable with ``externally_initialized``.
517 An explicit alignment may be specified for a global, which must be a
518 power of 2. If not present, or if the alignment is set to zero, the
519 alignment of the global is set by the target to whatever it feels
520 convenient. If an explicit alignment is specified, the global is forced
521 to have exactly that alignment. Targets and optimizers are not allowed
522 to over-align the global if the global has an assigned section. In this
523 case, the extra alignment could be observable: for example, code could
524 assume that the globals are densely packed in their section and try to
525 iterate over them as an array, alignment padding would break this
528 For example, the following defines a global in a numbered address space
529 with an initializer, section, and alignment:
533 @G = addrspace(5) constant float 1.0, section "foo", align 4
535 The following example defines a thread-local global with the
536 ``initialexec`` TLS model:
540 @G = thread_local(initialexec) global i32 0, align 4
542 .. _functionstructure:
547 LLVM function definitions consist of the "``define``" keyword, an
548 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
549 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
550 an optional ``unnamed_addr`` attribute, a return type, an optional
551 :ref:`parameter attribute <paramattrs>` for the return type, a function
552 name, a (possibly empty) argument list (each with optional :ref:`parameter
553 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
554 an optional section, an optional alignment, an optional :ref:`garbage
555 collector name <gc>`, an opening curly brace, a list of basic blocks,
556 and a closing curly brace.
558 LLVM function declarations consist of the "``declare``" keyword, an
559 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
560 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
561 an optional ``unnamed_addr`` attribute, a return type, an optional
562 :ref:`parameter attribute <paramattrs>` for the return type, a function
563 name, a possibly empty list of arguments, an optional alignment, and an
564 optional :ref:`garbage collector name <gc>`.
566 A function definition contains a list of basic blocks, forming the CFG
567 (Control Flow Graph) for the function. Each basic block may optionally
568 start with a label (giving the basic block a symbol table entry),
569 contains a list of instructions, and ends with a
570 :ref:`terminator <terminators>` instruction (such as a branch or function
571 return). If explicit label is not provided, a block is assigned an
572 implicit numbered label, using a next value from the same counter as used
573 for unnamed temporaries (:ref:`see above<identifiers>`). For example, if a
574 function entry block does not have explicit label, it will be assigned
575 label "%0", then first unnamed temporary in that block will be "%1", etc.
577 The first basic block in a function is special in two ways: it is
578 immediately executed on entrance to the function, and it is not allowed
579 to have predecessor basic blocks (i.e. there can not be any branches to
580 the entry block of a function). Because the block can have no
581 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
583 LLVM allows an explicit section to be specified for functions. If the
584 target supports it, it will emit functions to the section specified.
586 An explicit alignment may be specified for a function. If not present,
587 or if the alignment is set to zero, the alignment of the function is set
588 by the target to whatever it feels convenient. If an explicit alignment
589 is specified, the function is forced to have at least that much
590 alignment. All alignments must be a power of 2.
592 If the ``unnamed_addr`` attribute is given, the address is know to not
593 be significant and two identical functions can be merged.
597 define [linkage] [visibility]
599 <ResultType> @<FunctionName> ([argument list])
600 [fn Attrs] [section "name"] [align N]
608 Aliases act as "second name" for the aliasee value (which can be either
609 function, global variable, another alias or bitcast of global value).
610 Aliases may have an optional :ref:`linkage type <linkage>`, and an optional
611 :ref:`visibility style <visibility>`.
615 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
617 .. _namedmetadatastructure:
622 Named metadata is a collection of metadata. :ref:`Metadata
623 nodes <metadata>` (but not metadata strings) are the only valid
624 operands for a named metadata.
628 ; Some unnamed metadata nodes, which are referenced by the named metadata.
629 !0 = metadata !{metadata !"zero"}
630 !1 = metadata !{metadata !"one"}
631 !2 = metadata !{metadata !"two"}
633 !name = !{!0, !1, !2}
640 The return type and each parameter of a function type may have a set of
641 *parameter attributes* associated with them. Parameter attributes are
642 used to communicate additional information about the result or
643 parameters of a function. Parameter attributes are considered to be part
644 of the function, not of the function type, so functions with different
645 parameter attributes can have the same function type.
647 Parameter attributes are simple keywords that follow the type specified.
648 If multiple parameter attributes are needed, they are space separated.
653 declare i32 @printf(i8* noalias nocapture, ...)
654 declare i32 @atoi(i8 zeroext)
655 declare signext i8 @returns_signed_char()
657 Note that any attributes for the function result (``nounwind``,
658 ``readonly``) come immediately after the argument list.
660 Currently, only the following parameter attributes are defined:
663 This indicates to the code generator that the parameter or return
664 value should be zero-extended to the extent required by the target's
665 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
666 the caller (for a parameter) or the callee (for a return value).
668 This indicates to the code generator that the parameter or return
669 value should be sign-extended to the extent required by the target's
670 ABI (which is usually 32-bits) by the caller (for a parameter) or
671 the callee (for a return value).
673 This indicates that this parameter or return value should be treated
674 in a special target-dependent fashion during while emitting code for
675 a function call or return (usually, by putting it in a register as
676 opposed to memory, though some targets use it to distinguish between
677 two different kinds of registers). Use of this attribute is
680 This indicates that the pointer parameter should really be passed by
681 value to the function. The attribute implies that a hidden copy of
682 the pointee is made between the caller and the callee, so the callee
683 is unable to modify the value in the caller. This attribute is only
684 valid on LLVM pointer arguments. It is generally used to pass
685 structs and arrays by value, but is also valid on pointers to
686 scalars. The copy is considered to belong to the caller not the
687 callee (for example, ``readonly`` functions should not write to
688 ``byval`` parameters). This is not a valid attribute for return
691 The byval attribute also supports specifying an alignment with the
692 align attribute. It indicates the alignment of the stack slot to
693 form and the known alignment of the pointer specified to the call
694 site. If the alignment is not specified, then the code generator
695 makes a target-specific assumption.
698 This indicates that the pointer parameter specifies the address of a
699 structure that is the return value of the function in the source
700 program. This pointer must be guaranteed by the caller to be valid:
701 loads and stores to the structure may be assumed by the callee
702 not to trap and to be properly aligned. This may only be applied to
703 the first parameter. This is not a valid attribute for return
706 This indicates that pointer values :ref:`based <pointeraliasing>` on
707 the argument or return value do not alias pointer values which are
708 not *based* on it, ignoring certain "irrelevant" dependencies. For a
709 call to the parent function, dependencies between memory references
710 from before or after the call and from those during the call are
711 "irrelevant" to the ``noalias`` keyword for the arguments and return
712 value used in that call. The caller shares the responsibility with
713 the callee for ensuring that these requirements are met. For further
714 details, please see the discussion of the NoAlias response in `alias
715 analysis <AliasAnalysis.html#MustMayNo>`_.
717 Note that this definition of ``noalias`` is intentionally similar
718 to the definition of ``restrict`` in C99 for function arguments,
719 though it is slightly weaker.
721 For function return values, C99's ``restrict`` is not meaningful,
722 while LLVM's ``noalias`` is.
724 This indicates that the callee does not make any copies of the
725 pointer that outlive the callee itself. This is not a valid
726 attribute for return values.
731 This indicates that the pointer parameter can be excised using the
732 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
733 attribute for return values and can only be applied to one parameter.
736 This indicates that the function always returns the argument as its return
737 value. This is an optimization hint to the code generator when generating
738 the caller, allowing tail call optimization and omission of register saves
739 and restores in some cases; it is not checked or enforced when generating
740 the callee. The parameter and the function return type must be valid
741 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
742 valid attribute for return values and can only be applied to one parameter.
746 Garbage Collector Names
747 -----------------------
749 Each function may specify a garbage collector name, which is simply a
754 define void @f() gc "name" { ... }
756 The compiler declares the supported values of *name*. Specifying a
757 collector which will cause the compiler to alter its output in order to
758 support the named garbage collection algorithm.
765 Attribute groups are groups of attributes that are referenced by objects within
766 the IR. They are important for keeping ``.ll`` files readable, because a lot of
767 functions will use the same set of attributes. In the degenerative case of a
768 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
769 group will capture the important command line flags used to build that file.
771 An attribute group is a module-level object. To use an attribute group, an
772 object references the attribute group's ID (e.g. ``#37``). An object may refer
773 to more than one attribute group. In that situation, the attributes from the
774 different groups are merged.
776 Here is an example of attribute groups for a function that should always be
777 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
781 ; Target-independent attributes:
782 attributes #0 = { alwaysinline alignstack=4 }
784 ; Target-dependent attributes:
785 attributes #1 = { "no-sse" }
787 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
788 define void @f() #0 #1 { ... }
795 Function attributes are set to communicate additional information about
796 a function. Function attributes are considered to be part of the
797 function, not of the function type, so functions with different function
798 attributes can have the same function type.
800 Function attributes are simple keywords that follow the type specified.
801 If multiple attributes are needed, they are space separated. For
806 define void @f() noinline { ... }
807 define void @f() alwaysinline { ... }
808 define void @f() alwaysinline optsize { ... }
809 define void @f() optsize { ... }
812 This attribute indicates that, when emitting the prologue and
813 epilogue, the backend should forcibly align the stack pointer.
814 Specify the desired alignment, which must be a power of two, in
817 This attribute indicates that the inliner should attempt to inline
818 this function into callers whenever possible, ignoring any active
819 inlining size threshold for this caller.
821 This indicates that the callee function at a call site should be
822 recognized as a built-in function, even though the function's declaration
823 uses the ``nobuiltin'' attribute. This is only valid at call sites for
824 direct calls to functions which are declared with the ``nobuiltin``
827 This attribute indicates that this function is rarely called. When
828 computing edge weights, basic blocks post-dominated by a cold
829 function call are also considered to be cold; and, thus, given low
832 This attribute suppresses lazy symbol binding for the function. This
833 may make calls to the function faster, at the cost of extra program
834 startup time if the function is not called during program startup.
836 This attribute indicates that the source code contained a hint that
837 inlining this function is desirable (such as the "inline" keyword in
838 C/C++). It is just a hint; it imposes no requirements on the
841 This attribute disables prologue / epilogue emission for the
842 function. This can have very system-specific consequences.
844 This indicates that the callee function at a call site is not recognized as
845 a built-in function. LLVM will retain the original call and not replace it
846 with equivalent code based on the semantics of the built-in function, unless
847 the call site uses the ``builtin`` attribute. This is valid at call sites
848 and on function declarations and definitions.
850 This attribute indicates that calls to the function cannot be
851 duplicated. A call to a ``noduplicate`` function may be moved
852 within its parent function, but may not be duplicated within
855 A function containing a ``noduplicate`` call may still
856 be an inlining candidate, provided that the call is not
857 duplicated by inlining. That implies that the function has
858 internal linkage and only has one call site, so the original
859 call is dead after inlining.
861 This attributes disables implicit floating point instructions.
863 This attribute indicates that the inliner should never inline this
864 function in any situation. This attribute may not be used together
865 with the ``alwaysinline`` attribute.
867 This attribute indicates that the code generator should not use a
868 red zone, even if the target-specific ABI normally permits it.
870 This function attribute indicates that the function never returns
871 normally. This produces undefined behavior at runtime if the
872 function ever does dynamically return.
874 This function attribute indicates that the function never returns
875 with an unwind or exceptional control flow. If the function does
876 unwind, its runtime behavior is undefined.
878 This attribute suggests that optimization passes and code generator
879 passes make choices that keep the code size of this function low,
880 and otherwise do optimizations specifically to reduce code size.
882 This attribute indicates that the function computes its result (or
883 decides to unwind an exception) based strictly on its arguments,
884 without dereferencing any pointer arguments or otherwise accessing
885 any mutable state (e.g. memory, control registers, etc) visible to
886 caller functions. It does not write through any pointer arguments
887 (including ``byval`` arguments) and never changes any state visible
888 to callers. This means that it cannot unwind exceptions by calling
889 the ``C++`` exception throwing methods.
891 This attribute indicates that the function does not write through
892 any pointer arguments (including ``byval`` arguments) or otherwise
893 modify any state (e.g. memory, control registers, etc) visible to
894 caller functions. It may dereference pointer arguments and read
895 state that may be set in the caller. A readonly function always
896 returns the same value (or unwinds an exception identically) when
897 called with the same set of arguments and global state. It cannot
898 unwind an exception by calling the ``C++`` exception throwing
901 This attribute indicates that this function can return twice. The C
902 ``setjmp`` is an example of such a function. The compiler disables
903 some optimizations (like tail calls) in the caller of these
906 This attribute indicates that AddressSanitizer checks
907 (dynamic address safety analysis) are enabled for this function.
909 This attribute indicates that MemorySanitizer checks (dynamic detection
910 of accesses to uninitialized memory) are enabled for this function.
912 This attribute indicates that ThreadSanitizer checks
913 (dynamic thread safety analysis) are enabled for this function.
915 This attribute indicates that the function should emit a stack
916 smashing protector. It is in the form of a "canary" --- a random value
917 placed on the stack before the local variables that's checked upon
918 return from the function to see if it has been overwritten. A
919 heuristic is used to determine if a function needs stack protectors
920 or not. The heuristic used will enable protectors for functions with:
922 - Character arrays larger than ``ssp-buffer-size`` (default 8).
923 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
924 - Calls to alloca() with variable sizes or constant sizes greater than
927 If a function that has an ``ssp`` attribute is inlined into a
928 function that doesn't have an ``ssp`` attribute, then the resulting
929 function will have an ``ssp`` attribute.
931 This attribute indicates that the function should *always* emit a
932 stack smashing protector. This overrides the ``ssp`` function
935 If a function that has an ``sspreq`` attribute is inlined into a
936 function that doesn't have an ``sspreq`` attribute or which has an
937 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
938 an ``sspreq`` attribute.
940 This attribute indicates that the function should emit a stack smashing
941 protector. This attribute causes a strong heuristic to be used when
942 determining if a function needs stack protectors. The strong heuristic
943 will enable protectors for functions with:
945 - Arrays of any size and type
946 - Aggregates containing an array of any size and type.
948 - Local variables that have had their address taken.
950 This overrides the ``ssp`` function attribute.
952 If a function that has an ``sspstrong`` attribute is inlined into a
953 function that doesn't have an ``sspstrong`` attribute, then the
954 resulting function will have an ``sspstrong`` attribute.
956 This attribute indicates that the ABI being targeted requires that
957 an unwind table entry be produce for this function even if we can
958 show that no exceptions passes by it. This is normally the case for
959 the ELF x86-64 abi, but it can be disabled for some compilation
964 Module-Level Inline Assembly
965 ----------------------------
967 Modules may contain "module-level inline asm" blocks, which corresponds
968 to the GCC "file scope inline asm" blocks. These blocks are internally
969 concatenated by LLVM and treated as a single unit, but may be separated
970 in the ``.ll`` file if desired. The syntax is very simple:
974 module asm "inline asm code goes here"
975 module asm "more can go here"
977 The strings can contain any character by escaping non-printable
978 characters. The escape sequence used is simply "\\xx" where "xx" is the
979 two digit hex code for the number.
981 The inline asm code is simply printed to the machine code .s file when
982 assembly code is generated.
984 .. _langref_datalayout:
989 A module may specify a target specific data layout string that specifies
990 how data is to be laid out in memory. The syntax for the data layout is
995 target datalayout = "layout specification"
997 The *layout specification* consists of a list of specifications
998 separated by the minus sign character ('-'). Each specification starts
999 with a letter and may include other information after the letter to
1000 define some aspect of the data layout. The specifications accepted are
1004 Specifies that the target lays out data in big-endian form. That is,
1005 the bits with the most significance have the lowest address
1008 Specifies that the target lays out data in little-endian form. That
1009 is, the bits with the least significance have the lowest address
1012 Specifies the natural alignment of the stack in bits. Alignment
1013 promotion of stack variables is limited to the natural stack
1014 alignment to avoid dynamic stack realignment. The stack alignment
1015 must be a multiple of 8-bits. If omitted, the natural stack
1016 alignment defaults to "unspecified", which does not prevent any
1017 alignment promotions.
1018 ``p[n]:<size>:<abi>:<pref>``
1019 This specifies the *size* of a pointer and its ``<abi>`` and
1020 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1021 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
1022 preceding ``:`` should be omitted too. The address space, ``n`` is
1023 optional, and if not specified, denotes the default address space 0.
1024 The value of ``n`` must be in the range [1,2^23).
1025 ``i<size>:<abi>:<pref>``
1026 This specifies the alignment for an integer type of a given bit
1027 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1028 ``v<size>:<abi>:<pref>``
1029 This specifies the alignment for a vector type of a given bit
1031 ``f<size>:<abi>:<pref>``
1032 This specifies the alignment for a floating point type of a given bit
1033 ``<size>``. Only values of ``<size>`` that are supported by the target
1034 will work. 32 (float) and 64 (double) are supported on all targets; 80
1035 or 128 (different flavors of long double) are also supported on some
1037 ``a<size>:<abi>:<pref>``
1038 This specifies the alignment for an aggregate type of a given bit
1040 ``s<size>:<abi>:<pref>``
1041 This specifies the alignment for a stack object of a given bit
1043 ``n<size1>:<size2>:<size3>...``
1044 This specifies a set of native integer widths for the target CPU in
1045 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1046 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1047 this set are considered to support most general arithmetic operations
1050 When constructing the data layout for a given target, LLVM starts with a
1051 default set of specifications which are then (possibly) overridden by
1052 the specifications in the ``datalayout`` keyword. The default
1053 specifications are given in this list:
1055 - ``E`` - big endian
1056 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment
1057 - ``S0`` - natural stack alignment is unspecified
1058 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1059 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1060 - ``i16:16:16`` - i16 is 16-bit aligned
1061 - ``i32:32:32`` - i32 is 32-bit aligned
1062 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1063 alignment of 64-bits
1064 - ``f16:16:16`` - half is 16-bit aligned
1065 - ``f32:32:32`` - float is 32-bit aligned
1066 - ``f64:64:64`` - double is 64-bit aligned
1067 - ``f128:128:128`` - quad is 128-bit aligned
1068 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1069 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1070 - ``a0:0:64`` - aggregates are 64-bit aligned
1072 When LLVM is determining the alignment for a given type, it uses the
1075 #. If the type sought is an exact match for one of the specifications,
1076 that specification is used.
1077 #. If no match is found, and the type sought is an integer type, then
1078 the smallest integer type that is larger than the bitwidth of the
1079 sought type is used. If none of the specifications are larger than
1080 the bitwidth then the largest integer type is used. For example,
1081 given the default specifications above, the i7 type will use the
1082 alignment of i8 (next largest) while both i65 and i256 will use the
1083 alignment of i64 (largest specified).
1084 #. If no match is found, and the type sought is a vector type, then the
1085 largest vector type that is smaller than the sought vector type will
1086 be used as a fall back. This happens because <128 x double> can be
1087 implemented in terms of 64 <2 x double>, for example.
1089 The function of the data layout string may not be what you expect.
1090 Notably, this is not a specification from the frontend of what alignment
1091 the code generator should use.
1093 Instead, if specified, the target data layout is required to match what
1094 the ultimate *code generator* expects. This string is used by the
1095 mid-level optimizers to improve code, and this only works if it matches
1096 what the ultimate code generator uses. If you would like to generate IR
1097 that does not embed this target-specific detail into the IR, then you
1098 don't have to specify the string. This will disable some optimizations
1099 that require precise layout information, but this also prevents those
1100 optimizations from introducing target specificity into the IR.
1102 .. _pointeraliasing:
1104 Pointer Aliasing Rules
1105 ----------------------
1107 Any memory access must be done through a pointer value associated with
1108 an address range of the memory access, otherwise the behavior is
1109 undefined. Pointer values are associated with address ranges according
1110 to the following rules:
1112 - A pointer value is associated with the addresses associated with any
1113 value it is *based* on.
1114 - An address of a global variable is associated with the address range
1115 of the variable's storage.
1116 - The result value of an allocation instruction is associated with the
1117 address range of the allocated storage.
1118 - A null pointer in the default address-space is associated with no
1120 - An integer constant other than zero or a pointer value returned from
1121 a function not defined within LLVM may be associated with address
1122 ranges allocated through mechanisms other than those provided by
1123 LLVM. Such ranges shall not overlap with any ranges of addresses
1124 allocated by mechanisms provided by LLVM.
1126 A pointer value is *based* on another pointer value according to the
1129 - A pointer value formed from a ``getelementptr`` operation is *based*
1130 on the first operand of the ``getelementptr``.
1131 - The result value of a ``bitcast`` is *based* on the operand of the
1133 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1134 values that contribute (directly or indirectly) to the computation of
1135 the pointer's value.
1136 - The "*based* on" relationship is transitive.
1138 Note that this definition of *"based"* is intentionally similar to the
1139 definition of *"based"* in C99, though it is slightly weaker.
1141 LLVM IR does not associate types with memory. The result type of a
1142 ``load`` merely indicates the size and alignment of the memory from
1143 which to load, as well as the interpretation of the value. The first
1144 operand type of a ``store`` similarly only indicates the size and
1145 alignment of the store.
1147 Consequently, type-based alias analysis, aka TBAA, aka
1148 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1149 :ref:`Metadata <metadata>` may be used to encode additional information
1150 which specialized optimization passes may use to implement type-based
1155 Volatile Memory Accesses
1156 ------------------------
1158 Certain memory accesses, such as :ref:`load <i_load>`'s,
1159 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1160 marked ``volatile``. The optimizers must not change the number of
1161 volatile operations or change their order of execution relative to other
1162 volatile operations. The optimizers *may* change the order of volatile
1163 operations relative to non-volatile operations. This is not Java's
1164 "volatile" and has no cross-thread synchronization behavior.
1166 IR-level volatile loads and stores cannot safely be optimized into
1167 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1168 flagged volatile. Likewise, the backend should never split or merge
1169 target-legal volatile load/store instructions.
1171 .. admonition:: Rationale
1173 Platforms may rely on volatile loads and stores of natively supported
1174 data width to be executed as single instruction. For example, in C
1175 this holds for an l-value of volatile primitive type with native
1176 hardware support, but not necessarily for aggregate types. The
1177 frontend upholds these expectations, which are intentionally
1178 unspecified in the IR. The rules above ensure that IR transformation
1179 do not violate the frontend's contract with the language.
1183 Memory Model for Concurrent Operations
1184 --------------------------------------
1186 The LLVM IR does not define any way to start parallel threads of
1187 execution or to register signal handlers. Nonetheless, there are
1188 platform-specific ways to create them, and we define LLVM IR's behavior
1189 in their presence. This model is inspired by the C++0x memory model.
1191 For a more informal introduction to this model, see the :doc:`Atomics`.
1193 We define a *happens-before* partial order as the least partial order
1196 - Is a superset of single-thread program order, and
1197 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1198 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1199 techniques, like pthread locks, thread creation, thread joining,
1200 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1201 Constraints <ordering>`).
1203 Note that program order does not introduce *happens-before* edges
1204 between a thread and signals executing inside that thread.
1206 Every (defined) read operation (load instructions, memcpy, atomic
1207 loads/read-modify-writes, etc.) R reads a series of bytes written by
1208 (defined) write operations (store instructions, atomic
1209 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1210 section, initialized globals are considered to have a write of the
1211 initializer which is atomic and happens before any other read or write
1212 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1213 may see any write to the same byte, except:
1215 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1216 write\ :sub:`2` happens before R\ :sub:`byte`, then
1217 R\ :sub:`byte` does not see write\ :sub:`1`.
1218 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1219 R\ :sub:`byte` does not see write\ :sub:`3`.
1221 Given that definition, R\ :sub:`byte` is defined as follows:
1223 - If R is volatile, the result is target-dependent. (Volatile is
1224 supposed to give guarantees which can support ``sig_atomic_t`` in
1225 C/C++, and may be used for accesses to addresses which do not behave
1226 like normal memory. It does not generally provide cross-thread
1228 - Otherwise, if there is no write to the same byte that happens before
1229 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1230 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1231 R\ :sub:`byte` returns the value written by that write.
1232 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1233 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1234 Memory Ordering Constraints <ordering>` section for additional
1235 constraints on how the choice is made.
1236 - Otherwise R\ :sub:`byte` returns ``undef``.
1238 R returns the value composed of the series of bytes it read. This
1239 implies that some bytes within the value may be ``undef`` **without**
1240 the entire value being ``undef``. Note that this only defines the
1241 semantics of the operation; it doesn't mean that targets will emit more
1242 than one instruction to read the series of bytes.
1244 Note that in cases where none of the atomic intrinsics are used, this
1245 model places only one restriction on IR transformations on top of what
1246 is required for single-threaded execution: introducing a store to a byte
1247 which might not otherwise be stored is not allowed in general.
1248 (Specifically, in the case where another thread might write to and read
1249 from an address, introducing a store can change a load that may see
1250 exactly one write into a load that may see multiple writes.)
1254 Atomic Memory Ordering Constraints
1255 ----------------------------------
1257 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1258 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1259 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1260 an ordering parameter that determines which other atomic instructions on
1261 the same address they *synchronize with*. These semantics are borrowed
1262 from Java and C++0x, but are somewhat more colloquial. If these
1263 descriptions aren't precise enough, check those specs (see spec
1264 references in the :doc:`atomics guide <Atomics>`).
1265 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1266 differently since they don't take an address. See that instruction's
1267 documentation for details.
1269 For a simpler introduction to the ordering constraints, see the
1273 The set of values that can be read is governed by the happens-before
1274 partial order. A value cannot be read unless some operation wrote
1275 it. This is intended to provide a guarantee strong enough to model
1276 Java's non-volatile shared variables. This ordering cannot be
1277 specified for read-modify-write operations; it is not strong enough
1278 to make them atomic in any interesting way.
1280 In addition to the guarantees of ``unordered``, there is a single
1281 total order for modifications by ``monotonic`` operations on each
1282 address. All modification orders must be compatible with the
1283 happens-before order. There is no guarantee that the modification
1284 orders can be combined to a global total order for the whole program
1285 (and this often will not be possible). The read in an atomic
1286 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1287 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1288 order immediately before the value it writes. If one atomic read
1289 happens before another atomic read of the same address, the later
1290 read must see the same value or a later value in the address's
1291 modification order. This disallows reordering of ``monotonic`` (or
1292 stronger) operations on the same address. If an address is written
1293 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1294 read that address repeatedly, the other threads must eventually see
1295 the write. This corresponds to the C++0x/C1x
1296 ``memory_order_relaxed``.
1298 In addition to the guarantees of ``monotonic``, a
1299 *synchronizes-with* edge may be formed with a ``release`` operation.
1300 This is intended to model C++'s ``memory_order_acquire``.
1302 In addition to the guarantees of ``monotonic``, if this operation
1303 writes a value which is subsequently read by an ``acquire``
1304 operation, it *synchronizes-with* that operation. (This isn't a
1305 complete description; see the C++0x definition of a release
1306 sequence.) This corresponds to the C++0x/C1x
1307 ``memory_order_release``.
1308 ``acq_rel`` (acquire+release)
1309 Acts as both an ``acquire`` and ``release`` operation on its
1310 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1311 ``seq_cst`` (sequentially consistent)
1312 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1313 operation which only reads, ``release`` for an operation which only
1314 writes), there is a global total order on all
1315 sequentially-consistent operations on all addresses, which is
1316 consistent with the *happens-before* partial order and with the
1317 modification orders of all the affected addresses. Each
1318 sequentially-consistent read sees the last preceding write to the
1319 same address in this global order. This corresponds to the C++0x/C1x
1320 ``memory_order_seq_cst`` and Java volatile.
1324 If an atomic operation is marked ``singlethread``, it only *synchronizes
1325 with* or participates in modification and seq\_cst total orderings with
1326 other operations running in the same thread (for example, in signal
1334 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1335 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1336 :ref:`frem <i_frem>`) have the following flags that can set to enable
1337 otherwise unsafe floating point operations
1340 No NaNs - Allow optimizations to assume the arguments and result are not
1341 NaN. Such optimizations are required to retain defined behavior over
1342 NaNs, but the value of the result is undefined.
1345 No Infs - Allow optimizations to assume the arguments and result are not
1346 +/-Inf. Such optimizations are required to retain defined behavior over
1347 +/-Inf, but the value of the result is undefined.
1350 No Signed Zeros - Allow optimizations to treat the sign of a zero
1351 argument or result as insignificant.
1354 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1355 argument rather than perform division.
1358 Fast - Allow algebraically equivalent transformations that may
1359 dramatically change results in floating point (e.g. reassociate). This
1360 flag implies all the others.
1367 The LLVM type system is one of the most important features of the
1368 intermediate representation. Being typed enables a number of
1369 optimizations to be performed on the intermediate representation
1370 directly, without having to do extra analyses on the side before the
1371 transformation. A strong type system makes it easier to read the
1372 generated code and enables novel analyses and transformations that are
1373 not feasible to perform on normal three address code representations.
1375 .. _typeclassifications:
1377 Type Classifications
1378 --------------------
1380 The types fall into a few useful classifications:
1389 * - :ref:`integer <t_integer>`
1390 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1393 * - :ref:`floating point <t_floating>`
1394 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1402 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1403 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1404 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1405 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1407 * - :ref:`primitive <t_primitive>`
1408 - :ref:`label <t_label>`,
1409 :ref:`void <t_void>`,
1410 :ref:`integer <t_integer>`,
1411 :ref:`floating point <t_floating>`,
1412 :ref:`x86mmx <t_x86mmx>`,
1413 :ref:`metadata <t_metadata>`.
1415 * - :ref:`derived <t_derived>`
1416 - :ref:`array <t_array>`,
1417 :ref:`function <t_function>`,
1418 :ref:`pointer <t_pointer>`,
1419 :ref:`structure <t_struct>`,
1420 :ref:`vector <t_vector>`,
1421 :ref:`opaque <t_opaque>`.
1423 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1424 Values of these types are the only ones which can be produced by
1432 The primitive types are the fundamental building blocks of the LLVM
1443 The integer type is a very simple type that simply specifies an
1444 arbitrary bit width for the integer type desired. Any bit width from 1
1445 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1454 The number of bits the integer will occupy is specified by the ``N``
1460 +----------------+------------------------------------------------+
1461 | ``i1`` | a single-bit integer. |
1462 +----------------+------------------------------------------------+
1463 | ``i32`` | a 32-bit integer. |
1464 +----------------+------------------------------------------------+
1465 | ``i1942652`` | a really big integer of over 1 million bits. |
1466 +----------------+------------------------------------------------+
1470 Floating Point Types
1471 ^^^^^^^^^^^^^^^^^^^^
1480 - 16-bit floating point value
1483 - 32-bit floating point value
1486 - 64-bit floating point value
1489 - 128-bit floating point value (112-bit mantissa)
1492 - 80-bit floating point value (X87)
1495 - 128-bit floating point value (two 64-bits)
1505 The x86mmx type represents a value held in an MMX register on an x86
1506 machine. The operations allowed on it are quite limited: parameters and
1507 return values, load and store, and bitcast. User-specified MMX
1508 instructions are represented as intrinsic or asm calls with arguments
1509 and/or results of this type. There are no arrays, vectors or constants
1527 The void type does not represent any value and has no size.
1544 The label type represents code labels.
1561 The metadata type represents embedded metadata. No derived types may be
1562 created from metadata except for :ref:`function <t_function>` arguments.
1576 The real power in LLVM comes from the derived types in the system. This
1577 is what allows a programmer to represent arrays, functions, pointers,
1578 and other useful types. Each of these types contain one or more element
1579 types which may be a primitive type, or another derived type. For
1580 example, it is possible to have a two dimensional array, using an array
1581 as the element type of another array.
1588 Aggregate Types are a subset of derived types that can contain multiple
1589 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1590 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1601 The array type is a very simple derived type that arranges elements
1602 sequentially in memory. The array type requires a size (number of
1603 elements) and an underlying data type.
1610 [<# elements> x <elementtype>]
1612 The number of elements is a constant integer value; ``elementtype`` may
1613 be any type with a size.
1618 +------------------+--------------------------------------+
1619 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1620 +------------------+--------------------------------------+
1621 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1622 +------------------+--------------------------------------+
1623 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1624 +------------------+--------------------------------------+
1626 Here are some examples of multidimensional arrays:
1628 +-----------------------------+----------------------------------------------------------+
1629 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1630 +-----------------------------+----------------------------------------------------------+
1631 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1632 +-----------------------------+----------------------------------------------------------+
1633 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1634 +-----------------------------+----------------------------------------------------------+
1636 There is no restriction on indexing beyond the end of the array implied
1637 by a static type (though there are restrictions on indexing beyond the
1638 bounds of an allocated object in some cases). This means that
1639 single-dimension 'variable sized array' addressing can be implemented in
1640 LLVM with a zero length array type. An implementation of 'pascal style
1641 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1652 The function type can be thought of as a function signature. It consists
1653 of a return type and a list of formal parameter types. The return type
1654 of a function type is a first class type or a void type.
1661 <returntype> (<parameter list>)
1663 ...where '``<parameter list>``' is a comma-separated list of type
1664 specifiers. Optionally, the parameter list may include a type ``...``,
1665 which indicates that the function takes a variable number of arguments.
1666 Variable argument functions can access their arguments with the
1667 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1668 '``<returntype>``' is any type except :ref:`label <t_label>`.
1673 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1674 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1675 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1676 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1677 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1678 | ``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. |
1679 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1680 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1681 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1691 The structure type is used to represent a collection of data members
1692 together in memory. The elements of a structure may be any type that has
1695 Structures in memory are accessed using '``load``' and '``store``' by
1696 getting a pointer to a field with the '``getelementptr``' instruction.
1697 Structures in registers are accessed using the '``extractvalue``' and
1698 '``insertvalue``' instructions.
1700 Structures may optionally be "packed" structures, which indicate that
1701 the alignment of the struct is one byte, and that there is no padding
1702 between the elements. In non-packed structs, padding between field types
1703 is inserted as defined by the DataLayout string in the module, which is
1704 required to match what the underlying code generator expects.
1706 Structures can either be "literal" or "identified". A literal structure
1707 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1708 identified types are always defined at the top level with a name.
1709 Literal types are uniqued by their contents and can never be recursive
1710 or opaque since there is no way to write one. Identified types can be
1711 recursive, can be opaqued, and are never uniqued.
1718 %T1 = type { <type list> } ; Identified normal struct type
1719 %T2 = type <{ <type list> }> ; Identified packed struct type
1724 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1725 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1726 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1727 | ``{ 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``. |
1728 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1729 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1730 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1734 Opaque Structure Types
1735 ^^^^^^^^^^^^^^^^^^^^^^
1740 Opaque structure types are used to represent named structure types that
1741 do not have a body specified. This corresponds (for example) to the C
1742 notion of a forward declared structure.
1755 +--------------+-------------------+
1756 | ``opaque`` | An opaque type. |
1757 +--------------+-------------------+
1767 The pointer type is used to specify memory locations. Pointers are
1768 commonly used to reference objects in memory.
1770 Pointer types may have an optional address space attribute defining the
1771 numbered address space where the pointed-to object resides. The default
1772 address space is number zero. The semantics of non-zero address spaces
1773 are target-specific.
1775 Note that LLVM does not permit pointers to void (``void*``) nor does it
1776 permit pointers to labels (``label*``). Use ``i8*`` instead.
1788 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1789 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1790 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1791 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1792 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1793 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1794 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1804 A vector type is a simple derived type that represents a vector of
1805 elements. Vector types are used when multiple primitive data are
1806 operated in parallel using a single instruction (SIMD). A vector type
1807 requires a size (number of elements) and an underlying primitive data
1808 type. Vector types are considered :ref:`first class <t_firstclass>`.
1815 < <# elements> x <elementtype> >
1817 The number of elements is a constant integer value larger than 0;
1818 elementtype may be any integer or floating point type, or a pointer to
1819 these types. Vectors of size zero are not allowed.
1824 +-------------------+--------------------------------------------------+
1825 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1826 +-------------------+--------------------------------------------------+
1827 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1828 +-------------------+--------------------------------------------------+
1829 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1830 +-------------------+--------------------------------------------------+
1831 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1832 +-------------------+--------------------------------------------------+
1837 LLVM has several different basic types of constants. This section
1838 describes them all and their syntax.
1843 **Boolean constants**
1844 The two strings '``true``' and '``false``' are both valid constants
1846 **Integer constants**
1847 Standard integers (such as '4') are constants of the
1848 :ref:`integer <t_integer>` type. Negative numbers may be used with
1850 **Floating point constants**
1851 Floating point constants use standard decimal notation (e.g.
1852 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1853 hexadecimal notation (see below). The assembler requires the exact
1854 decimal value of a floating-point constant. For example, the
1855 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1856 decimal in binary. Floating point constants must have a :ref:`floating
1857 point <t_floating>` type.
1858 **Null pointer constants**
1859 The identifier '``null``' is recognized as a null pointer constant
1860 and must be of :ref:`pointer type <t_pointer>`.
1862 The one non-intuitive notation for constants is the hexadecimal form of
1863 floating point constants. For example, the form
1864 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1865 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1866 constants are required (and the only time that they are generated by the
1867 disassembler) is when a floating point constant must be emitted but it
1868 cannot be represented as a decimal floating point number in a reasonable
1869 number of digits. For example, NaN's, infinities, and other special
1870 values are represented in their IEEE hexadecimal format so that assembly
1871 and disassembly do not cause any bits to change in the constants.
1873 When using the hexadecimal form, constants of types half, float, and
1874 double are represented using the 16-digit form shown above (which
1875 matches the IEEE754 representation for double); half and float values
1876 must, however, be exactly representable as IEEE 754 half and single
1877 precision, respectively. Hexadecimal format is always used for long
1878 double, and there are three forms of long double. The 80-bit format used
1879 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1880 128-bit format used by PowerPC (two adjacent doubles) is represented by
1881 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1882 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
1883 will only work if they match the long double format on your target.
1884 The IEEE 16-bit format (half precision) is represented by ``0xH``
1885 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
1886 (sign bit at the left).
1888 There are no constants of type x86mmx.
1890 .. _complexconstants:
1895 Complex constants are a (potentially recursive) combination of simple
1896 constants and smaller complex constants.
1898 **Structure constants**
1899 Structure constants are represented with notation similar to
1900 structure type definitions (a comma separated list of elements,
1901 surrounded by braces (``{}``)). For example:
1902 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1903 "``@G = external global i32``". Structure constants must have
1904 :ref:`structure type <t_struct>`, and the number and types of elements
1905 must match those specified by the type.
1907 Array constants are represented with notation similar to array type
1908 definitions (a comma separated list of elements, surrounded by
1909 square brackets (``[]``)). For example:
1910 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1911 :ref:`array type <t_array>`, and the number and types of elements must
1912 match those specified by the type.
1913 **Vector constants**
1914 Vector constants are represented with notation similar to vector
1915 type definitions (a comma separated list of elements, surrounded by
1916 less-than/greater-than's (``<>``)). For example:
1917 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1918 must have :ref:`vector type <t_vector>`, and the number and types of
1919 elements must match those specified by the type.
1920 **Zero initialization**
1921 The string '``zeroinitializer``' can be used to zero initialize a
1922 value to zero of *any* type, including scalar and
1923 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1924 having to print large zero initializers (e.g. for large arrays) and
1925 is always exactly equivalent to using explicit zero initializers.
1927 A metadata node is a structure-like constant with :ref:`metadata
1928 type <t_metadata>`. For example:
1929 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
1930 constants that are meant to be interpreted as part of the
1931 instruction stream, metadata is a place to attach additional
1932 information such as debug info.
1934 Global Variable and Function Addresses
1935 --------------------------------------
1937 The addresses of :ref:`global variables <globalvars>` and
1938 :ref:`functions <functionstructure>` are always implicitly valid
1939 (link-time) constants. These constants are explicitly referenced when
1940 the :ref:`identifier for the global <identifiers>` is used and always have
1941 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
1944 .. code-block:: llvm
1948 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1955 The string '``undef``' can be used anywhere a constant is expected, and
1956 indicates that the user of the value may receive an unspecified
1957 bit-pattern. Undefined values may be of any type (other than '``label``'
1958 or '``void``') and be used anywhere a constant is permitted.
1960 Undefined values are useful because they indicate to the compiler that
1961 the program is well defined no matter what value is used. This gives the
1962 compiler more freedom to optimize. Here are some examples of
1963 (potentially surprising) transformations that are valid (in pseudo IR):
1965 .. code-block:: llvm
1975 This is safe because all of the output bits are affected by the undef
1976 bits. Any output bit can have a zero or one depending on the input bits.
1978 .. code-block:: llvm
1989 These logical operations have bits that are not always affected by the
1990 input. For example, if ``%X`` has a zero bit, then the output of the
1991 '``and``' operation will always be a zero for that bit, no matter what
1992 the corresponding bit from the '``undef``' is. As such, it is unsafe to
1993 optimize or assume that the result of the '``and``' is '``undef``'.
1994 However, it is safe to assume that all bits of the '``undef``' could be
1995 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
1996 all the bits of the '``undef``' operand to the '``or``' could be set,
1997 allowing the '``or``' to be folded to -1.
1999 .. code-block:: llvm
2001 %A = select undef, %X, %Y
2002 %B = select undef, 42, %Y
2003 %C = select %X, %Y, undef
2013 This set of examples shows that undefined '``select``' (and conditional
2014 branch) conditions can go *either way*, but they have to come from one
2015 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2016 both known to have a clear low bit, then ``%A`` would have to have a
2017 cleared low bit. However, in the ``%C`` example, the optimizer is
2018 allowed to assume that the '``undef``' operand could be the same as
2019 ``%Y``, allowing the whole '``select``' to be eliminated.
2021 .. code-block:: llvm
2023 %A = xor undef, undef
2040 This example points out that two '``undef``' operands are not
2041 necessarily the same. This can be surprising to people (and also matches
2042 C semantics) where they assume that "``X^X``" is always zero, even if
2043 ``X`` is undefined. This isn't true for a number of reasons, but the
2044 short answer is that an '``undef``' "variable" can arbitrarily change
2045 its value over its "live range". This is true because the variable
2046 doesn't actually *have a live range*. Instead, the value is logically
2047 read from arbitrary registers that happen to be around when needed, so
2048 the value is not necessarily consistent over time. In fact, ``%A`` and
2049 ``%C`` need to have the same semantics or the core LLVM "replace all
2050 uses with" concept would not hold.
2052 .. code-block:: llvm
2060 These examples show the crucial difference between an *undefined value*
2061 and *undefined behavior*. An undefined value (like '``undef``') is
2062 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2063 operation can be constant folded to '``undef``', because the '``undef``'
2064 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2065 However, in the second example, we can make a more aggressive
2066 assumption: because the ``undef`` is allowed to be an arbitrary value,
2067 we are allowed to assume that it could be zero. Since a divide by zero
2068 has *undefined behavior*, we are allowed to assume that the operation
2069 does not execute at all. This allows us to delete the divide and all
2070 code after it. Because the undefined operation "can't happen", the
2071 optimizer can assume that it occurs in dead code.
2073 .. code-block:: llvm
2075 a: store undef -> %X
2076 b: store %X -> undef
2081 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2082 value can be assumed to not have any effect; we can assume that the
2083 value is overwritten with bits that happen to match what was already
2084 there. However, a store *to* an undefined location could clobber
2085 arbitrary memory, therefore, it has undefined behavior.
2092 Poison values are similar to :ref:`undef values <undefvalues>`, however
2093 they also represent the fact that an instruction or constant expression
2094 which cannot evoke side effects has nevertheless detected a condition
2095 which results in undefined behavior.
2097 There is currently no way of representing a poison value in the IR; they
2098 only exist when produced by operations such as :ref:`add <i_add>` with
2101 Poison value behavior is defined in terms of value *dependence*:
2103 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2104 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2105 their dynamic predecessor basic block.
2106 - Function arguments depend on the corresponding actual argument values
2107 in the dynamic callers of their functions.
2108 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2109 instructions that dynamically transfer control back to them.
2110 - :ref:`Invoke <i_invoke>` instructions depend on the
2111 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2112 call instructions that dynamically transfer control back to them.
2113 - Non-volatile loads and stores depend on the most recent stores to all
2114 of the referenced memory addresses, following the order in the IR
2115 (including loads and stores implied by intrinsics such as
2116 :ref:`@llvm.memcpy <int_memcpy>`.)
2117 - An instruction with externally visible side effects depends on the
2118 most recent preceding instruction with externally visible side
2119 effects, following the order in the IR. (This includes :ref:`volatile
2120 operations <volatile>`.)
2121 - An instruction *control-depends* on a :ref:`terminator
2122 instruction <terminators>` if the terminator instruction has
2123 multiple successors and the instruction is always executed when
2124 control transfers to one of the successors, and may not be executed
2125 when control is transferred to another.
2126 - Additionally, an instruction also *control-depends* on a terminator
2127 instruction if the set of instructions it otherwise depends on would
2128 be different if the terminator had transferred control to a different
2130 - Dependence is transitive.
2132 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2133 with the additional affect that any instruction which has a *dependence*
2134 on a poison value has undefined behavior.
2136 Here are some examples:
2138 .. code-block:: llvm
2141 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2142 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2143 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2144 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2146 store i32 %poison, i32* @g ; Poison value stored to memory.
2147 %poison2 = load i32* @g ; Poison value loaded back from memory.
2149 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2151 %narrowaddr = bitcast i32* @g to i16*
2152 %wideaddr = bitcast i32* @g to i64*
2153 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2154 %poison4 = load i64* %wideaddr ; Returns a poison value.
2156 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2157 br i1 %cmp, label %true, label %end ; Branch to either destination.
2160 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2161 ; it has undefined behavior.
2165 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2166 ; Both edges into this PHI are
2167 ; control-dependent on %cmp, so this
2168 ; always results in a poison value.
2170 store volatile i32 0, i32* @g ; This would depend on the store in %true
2171 ; if %cmp is true, or the store in %entry
2172 ; otherwise, so this is undefined behavior.
2174 br i1 %cmp, label %second_true, label %second_end
2175 ; The same branch again, but this time the
2176 ; true block doesn't have side effects.
2183 store volatile i32 0, i32* @g ; This time, the instruction always depends
2184 ; on the store in %end. Also, it is
2185 ; control-equivalent to %end, so this is
2186 ; well-defined (ignoring earlier undefined
2187 ; behavior in this example).
2191 Addresses of Basic Blocks
2192 -------------------------
2194 ``blockaddress(@function, %block)``
2196 The '``blockaddress``' constant computes the address of the specified
2197 basic block in the specified function, and always has an ``i8*`` type.
2198 Taking the address of the entry block is illegal.
2200 This value only has defined behavior when used as an operand to the
2201 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2202 against null. Pointer equality tests between labels addresses results in
2203 undefined behavior --- though, again, comparison against null is ok, and
2204 no label is equal to the null pointer. This may be passed around as an
2205 opaque pointer sized value as long as the bits are not inspected. This
2206 allows ``ptrtoint`` and arithmetic to be performed on these values so
2207 long as the original value is reconstituted before the ``indirectbr``
2210 Finally, some targets may provide defined semantics when using the value
2211 as the operand to an inline assembly, but that is target specific.
2215 Constant Expressions
2216 --------------------
2218 Constant expressions are used to allow expressions involving other
2219 constants to be used as constants. Constant expressions may be of any
2220 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2221 that does not have side effects (e.g. load and call are not supported).
2222 The following is the syntax for constant expressions:
2224 ``trunc (CST to TYPE)``
2225 Truncate a constant to another type. The bit size of CST must be
2226 larger than the bit size of TYPE. Both types must be integers.
2227 ``zext (CST to TYPE)``
2228 Zero extend a constant to another type. The bit size of CST must be
2229 smaller than the bit size of TYPE. Both types must be integers.
2230 ``sext (CST to TYPE)``
2231 Sign extend a constant to another type. The bit size of CST must be
2232 smaller than the bit size of TYPE. Both types must be integers.
2233 ``fptrunc (CST to TYPE)``
2234 Truncate a floating point constant to another floating point type.
2235 The size of CST must be larger than the size of TYPE. Both types
2236 must be floating point.
2237 ``fpext (CST to TYPE)``
2238 Floating point extend a constant to another type. The size of CST
2239 must be smaller or equal to the size of TYPE. Both types must be
2241 ``fptoui (CST to TYPE)``
2242 Convert a floating point constant to the corresponding unsigned
2243 integer constant. TYPE must be a scalar or vector integer type. CST
2244 must be of scalar or vector floating point type. Both CST and TYPE
2245 must be scalars, or vectors of the same number of elements. If the
2246 value won't fit in the integer type, the results are undefined.
2247 ``fptosi (CST to TYPE)``
2248 Convert a floating point constant to the corresponding signed
2249 integer constant. TYPE must be a scalar or vector integer type. CST
2250 must be of scalar or vector floating point type. Both CST and TYPE
2251 must be scalars, or vectors of the same number of elements. If the
2252 value won't fit in the integer type, the results are undefined.
2253 ``uitofp (CST to TYPE)``
2254 Convert an unsigned integer constant to the corresponding floating
2255 point constant. TYPE must be a scalar or vector floating point type.
2256 CST must be of scalar or vector integer type. Both CST and TYPE must
2257 be scalars, or vectors of the same number of elements. If the value
2258 won't fit in the floating point type, the results are undefined.
2259 ``sitofp (CST to TYPE)``
2260 Convert a signed integer constant to the corresponding floating
2261 point constant. TYPE must be a scalar or vector floating point type.
2262 CST must be of scalar or vector integer type. Both CST and TYPE must
2263 be scalars, or vectors of the same number of elements. If the value
2264 won't fit in the floating point type, the results are undefined.
2265 ``ptrtoint (CST to TYPE)``
2266 Convert a pointer typed constant to the corresponding integer
2267 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2268 pointer type. The ``CST`` value is zero extended, truncated, or
2269 unchanged to make it fit in ``TYPE``.
2270 ``inttoptr (CST to TYPE)``
2271 Convert an integer constant to a pointer constant. TYPE must be a
2272 pointer type. CST must be of integer type. The CST value is zero
2273 extended, truncated, or unchanged to make it fit in a pointer size.
2274 This one is *really* dangerous!
2275 ``bitcast (CST to TYPE)``
2276 Convert a constant, CST, to another TYPE. The constraints of the
2277 operands are the same as those for the :ref:`bitcast
2278 instruction <i_bitcast>`.
2279 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2280 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2281 constants. As with the :ref:`getelementptr <i_getelementptr>`
2282 instruction, the index list may have zero or more indexes, which are
2283 required to make sense for the type of "CSTPTR".
2284 ``select (COND, VAL1, VAL2)``
2285 Perform the :ref:`select operation <i_select>` on constants.
2286 ``icmp COND (VAL1, VAL2)``
2287 Performs the :ref:`icmp operation <i_icmp>` on constants.
2288 ``fcmp COND (VAL1, VAL2)``
2289 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2290 ``extractelement (VAL, IDX)``
2291 Perform the :ref:`extractelement operation <i_extractelement>` on
2293 ``insertelement (VAL, ELT, IDX)``
2294 Perform the :ref:`insertelement operation <i_insertelement>` on
2296 ``shufflevector (VEC1, VEC2, IDXMASK)``
2297 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2299 ``extractvalue (VAL, IDX0, IDX1, ...)``
2300 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2301 constants. The index list is interpreted in a similar manner as
2302 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2303 least one index value must be specified.
2304 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2305 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2306 The index list is interpreted in a similar manner as indices in a
2307 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2308 value must be specified.
2309 ``OPCODE (LHS, RHS)``
2310 Perform the specified operation of the LHS and RHS constants. OPCODE
2311 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2312 binary <bitwiseops>` operations. The constraints on operands are
2313 the same as those for the corresponding instruction (e.g. no bitwise
2314 operations on floating point values are allowed).
2321 Inline Assembler Expressions
2322 ----------------------------
2324 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2325 Inline Assembly <moduleasm>`) through the use of a special value. This
2326 value represents the inline assembler as a string (containing the
2327 instructions to emit), a list of operand constraints (stored as a
2328 string), a flag that indicates whether or not the inline asm expression
2329 has side effects, and a flag indicating whether the function containing
2330 the asm needs to align its stack conservatively. An example inline
2331 assembler expression is:
2333 .. code-block:: llvm
2335 i32 (i32) asm "bswap $0", "=r,r"
2337 Inline assembler expressions may **only** be used as the callee operand
2338 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2339 Thus, typically we have:
2341 .. code-block:: llvm
2343 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2345 Inline asms with side effects not visible in the constraint list must be
2346 marked as having side effects. This is done through the use of the
2347 '``sideeffect``' keyword, like so:
2349 .. code-block:: llvm
2351 call void asm sideeffect "eieio", ""()
2353 In some cases inline asms will contain code that will not work unless
2354 the stack is aligned in some way, such as calls or SSE instructions on
2355 x86, yet will not contain code that does that alignment within the asm.
2356 The compiler should make conservative assumptions about what the asm
2357 might contain and should generate its usual stack alignment code in the
2358 prologue if the '``alignstack``' keyword is present:
2360 .. code-block:: llvm
2362 call void asm alignstack "eieio", ""()
2364 Inline asms also support using non-standard assembly dialects. The
2365 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2366 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2367 the only supported dialects. An example is:
2369 .. code-block:: llvm
2371 call void asm inteldialect "eieio", ""()
2373 If multiple keywords appear the '``sideeffect``' keyword must come
2374 first, the '``alignstack``' keyword second and the '``inteldialect``'
2380 The call instructions that wrap inline asm nodes may have a
2381 "``!srcloc``" MDNode attached to it that contains a list of constant
2382 integers. If present, the code generator will use the integer as the
2383 location cookie value when report errors through the ``LLVMContext``
2384 error reporting mechanisms. This allows a front-end to correlate backend
2385 errors that occur with inline asm back to the source code that produced
2388 .. code-block:: llvm
2390 call void asm sideeffect "something bad", ""(), !srcloc !42
2392 !42 = !{ i32 1234567 }
2394 It is up to the front-end to make sense of the magic numbers it places
2395 in the IR. If the MDNode contains multiple constants, the code generator
2396 will use the one that corresponds to the line of the asm that the error
2401 Metadata Nodes and Metadata Strings
2402 -----------------------------------
2404 LLVM IR allows metadata to be attached to instructions in the program
2405 that can convey extra information about the code to the optimizers and
2406 code generator. One example application of metadata is source-level
2407 debug information. There are two metadata primitives: strings and nodes.
2408 All metadata has the ``metadata`` type and is identified in syntax by a
2409 preceding exclamation point ('``!``').
2411 A metadata string is a string surrounded by double quotes. It can
2412 contain any character by escaping non-printable characters with
2413 "``\xx``" where "``xx``" is the two digit hex code. For example:
2416 Metadata nodes are represented with notation similar to structure
2417 constants (a comma separated list of elements, surrounded by braces and
2418 preceded by an exclamation point). Metadata nodes can have any values as
2419 their operand. For example:
2421 .. code-block:: llvm
2423 !{ metadata !"test\00", i32 10}
2425 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2426 metadata nodes, which can be looked up in the module symbol table. For
2429 .. code-block:: llvm
2431 !foo = metadata !{!4, !3}
2433 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2434 function is using two metadata arguments:
2436 .. code-block:: llvm
2438 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2440 Metadata can be attached with an instruction. Here metadata ``!21`` is
2441 attached to the ``add`` instruction using the ``!dbg`` identifier:
2443 .. code-block:: llvm
2445 %indvar.next = add i64 %indvar, 1, !dbg !21
2447 More information about specific metadata nodes recognized by the
2448 optimizers and code generator is found below.
2453 In LLVM IR, memory does not have types, so LLVM's own type system is not
2454 suitable for doing TBAA. Instead, metadata is added to the IR to
2455 describe a type system of a higher level language. This can be used to
2456 implement typical C/C++ TBAA, but it can also be used to implement
2457 custom alias analysis behavior for other languages.
2459 The current metadata format is very simple. TBAA metadata nodes have up
2460 to three fields, e.g.:
2462 .. code-block:: llvm
2464 !0 = metadata !{ metadata !"an example type tree" }
2465 !1 = metadata !{ metadata !"int", metadata !0 }
2466 !2 = metadata !{ metadata !"float", metadata !0 }
2467 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2469 The first field is an identity field. It can be any value, usually a
2470 metadata string, which uniquely identifies the type. The most important
2471 name in the tree is the name of the root node. Two trees with different
2472 root node names are entirely disjoint, even if they have leaves with
2475 The second field identifies the type's parent node in the tree, or is
2476 null or omitted for a root node. A type is considered to alias all of
2477 its descendants and all of its ancestors in the tree. Also, a type is
2478 considered to alias all types in other trees, so that bitcode produced
2479 from multiple front-ends is handled conservatively.
2481 If the third field is present, it's an integer which if equal to 1
2482 indicates that the type is "constant" (meaning
2483 ``pointsToConstantMemory`` should return true; see `other useful
2484 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2486 '``tbaa.struct``' Metadata
2487 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2489 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2490 aggregate assignment operations in C and similar languages, however it
2491 is defined to copy a contiguous region of memory, which is more than
2492 strictly necessary for aggregate types which contain holes due to
2493 padding. Also, it doesn't contain any TBAA information about the fields
2496 ``!tbaa.struct`` metadata can describe which memory subregions in a
2497 memcpy are padding and what the TBAA tags of the struct are.
2499 The current metadata format is very simple. ``!tbaa.struct`` metadata
2500 nodes are a list of operands which are in conceptual groups of three.
2501 For each group of three, the first operand gives the byte offset of a
2502 field in bytes, the second gives its size in bytes, and the third gives
2505 .. code-block:: llvm
2507 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2509 This describes a struct with two fields. The first is at offset 0 bytes
2510 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2511 and has size 4 bytes and has tbaa tag !2.
2513 Note that the fields need not be contiguous. In this example, there is a
2514 4 byte gap between the two fields. This gap represents padding which
2515 does not carry useful data and need not be preserved.
2517 '``fpmath``' Metadata
2518 ^^^^^^^^^^^^^^^^^^^^^
2520 ``fpmath`` metadata may be attached to any instruction of floating point
2521 type. It can be used to express the maximum acceptable error in the
2522 result of that instruction, in ULPs, thus potentially allowing the
2523 compiler to use a more efficient but less accurate method of computing
2524 it. ULP is defined as follows:
2526 If ``x`` is a real number that lies between two finite consecutive
2527 floating-point numbers ``a`` and ``b``, without being equal to one
2528 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2529 distance between the two non-equal finite floating-point numbers
2530 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2532 The metadata node shall consist of a single positive floating point
2533 number representing the maximum relative error, for example:
2535 .. code-block:: llvm
2537 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2539 '``range``' Metadata
2540 ^^^^^^^^^^^^^^^^^^^^
2542 ``range`` metadata may be attached only to loads of integer types. It
2543 expresses the possible ranges the loaded value is in. The ranges are
2544 represented with a flattened list of integers. The loaded value is known
2545 to be in the union of the ranges defined by each consecutive pair. Each
2546 pair has the following properties:
2548 - The type must match the type loaded by the instruction.
2549 - The pair ``a,b`` represents the range ``[a,b)``.
2550 - Both ``a`` and ``b`` are constants.
2551 - The range is allowed to wrap.
2552 - The range should not represent the full or empty set. That is,
2555 In addition, the pairs must be in signed order of the lower bound and
2556 they must be non-contiguous.
2560 .. code-block:: llvm
2562 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2563 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2564 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2565 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2567 !0 = metadata !{ i8 0, i8 2 }
2568 !1 = metadata !{ i8 255, i8 2 }
2569 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2570 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2575 It is sometimes useful to attach information to loop constructs. Currently,
2576 loop metadata is implemented as metadata attached to the branch instruction
2577 in the loop latch block. This type of metadata refer to a metadata node that is
2578 guaranteed to be separate for each loop. The loop identifier metadata is
2579 specified with the name ``llvm.loop``.
2581 The loop identifier metadata is implemented using a metadata that refers to
2582 itself to avoid merging it with any other identifier metadata, e.g.,
2583 during module linkage or function inlining. That is, each loop should refer
2584 to their own identification metadata even if they reside in separate functions.
2585 The following example contains loop identifier metadata for two separate loop
2588 .. code-block:: llvm
2590 !0 = metadata !{ metadata !0 }
2591 !1 = metadata !{ metadata !1 }
2593 The loop identifier metadata can be used to specify additional per-loop
2594 metadata. Any operands after the first operand can be treated as user-defined
2595 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2596 by the loop vectorizer to indicate how many times to unroll the loop:
2598 .. code-block:: llvm
2600 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2602 !0 = metadata !{ metadata !0, metadata !1 }
2603 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2608 Metadata types used to annotate memory accesses with information helpful
2609 for optimizations are prefixed with ``llvm.mem``.
2611 '``llvm.mem.parallel_loop_access``' Metadata
2612 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2614 For a loop to be parallel, in addition to using
2615 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2616 also all of the memory accessing instructions in the loop body need to be
2617 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2618 is at least one memory accessing instruction not marked with the metadata,
2619 the loop must be considered a sequential loop. This causes parallel loops to be
2620 converted to sequential loops due to optimization passes that are unaware of
2621 the parallel semantics and that insert new memory instructions to the loop
2624 Example of a loop that is considered parallel due to its correct use of
2625 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2626 metadata types that refer to the same loop identifier metadata.
2628 .. code-block:: llvm
2632 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2634 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2636 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2640 !0 = metadata !{ metadata !0 }
2642 It is also possible to have nested parallel loops. In that case the
2643 memory accesses refer to a list of loop identifier metadata nodes instead of
2644 the loop identifier metadata node directly:
2646 .. code-block:: llvm
2653 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2655 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2657 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2661 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2663 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2665 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2667 outer.for.end: ; preds = %for.body
2669 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2670 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2671 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2673 '``llvm.vectorizer``'
2674 ^^^^^^^^^^^^^^^^^^^^^
2676 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2677 vectorization parameters such as vectorization factor and unroll factor.
2679 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2680 loop identification metadata.
2682 '``llvm.vectorizer.unroll``' Metadata
2683 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2685 This metadata instructs the loop vectorizer to unroll the specified
2686 loop exactly ``N`` times.
2688 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2689 operand is an integer specifying the unroll factor. For example:
2691 .. code-block:: llvm
2693 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2695 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2698 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2699 determined automatically.
2701 '``llvm.vectorizer.width``' Metadata
2702 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2704 This metadata sets the target width of the vectorizer to ``N``. Without
2705 this metadata, the vectorizer will choose a width automatically.
2706 Regardless of this metadata, the vectorizer will only vectorize loops if
2707 it believes it is valid to do so.
2709 The first operand is the string ``llvm.vectorizer.width`` and the second
2710 operand is an integer specifying the width. For example:
2712 .. code-block:: llvm
2714 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2716 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2719 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2722 Module Flags Metadata
2723 =====================
2725 Information about the module as a whole is difficult to convey to LLVM's
2726 subsystems. The LLVM IR isn't sufficient to transmit this information.
2727 The ``llvm.module.flags`` named metadata exists in order to facilitate
2728 this. These flags are in the form of key / value pairs --- much like a
2729 dictionary --- making it easy for any subsystem who cares about a flag to
2732 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2733 Each triplet has the following form:
2735 - The first element is a *behavior* flag, which specifies the behavior
2736 when two (or more) modules are merged together, and it encounters two
2737 (or more) metadata with the same ID. The supported behaviors are
2739 - The second element is a metadata string that is a unique ID for the
2740 metadata. Each module may only have one flag entry for each unique ID (not
2741 including entries with the **Require** behavior).
2742 - The third element is the value of the flag.
2744 When two (or more) modules are merged together, the resulting
2745 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2746 each unique metadata ID string, there will be exactly one entry in the merged
2747 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2748 be determined by the merge behavior flag, as described below. The only exception
2749 is that entries with the *Require* behavior are always preserved.
2751 The following behaviors are supported:
2762 Emits an error if two values disagree, otherwise the resulting value
2763 is that of the operands.
2767 Emits a warning if two values disagree. The result value will be the
2768 operand for the flag from the first module being linked.
2772 Adds a requirement that another module flag be present and have a
2773 specified value after linking is performed. The value must be a
2774 metadata pair, where the first element of the pair is the ID of the
2775 module flag to be restricted, and the second element of the pair is
2776 the value the module flag should be restricted to. This behavior can
2777 be used to restrict the allowable results (via triggering of an
2778 error) of linking IDs with the **Override** behavior.
2782 Uses the specified value, regardless of the behavior or value of the
2783 other module. If both modules specify **Override**, but the values
2784 differ, an error will be emitted.
2788 Appends the two values, which are required to be metadata nodes.
2792 Appends the two values, which are required to be metadata
2793 nodes. However, duplicate entries in the second list are dropped
2794 during the append operation.
2796 It is an error for a particular unique flag ID to have multiple behaviors,
2797 except in the case of **Require** (which adds restrictions on another metadata
2798 value) or **Override**.
2800 An example of module flags:
2802 .. code-block:: llvm
2804 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2805 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2806 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2807 !3 = metadata !{ i32 3, metadata !"qux",
2809 metadata !"foo", i32 1
2812 !llvm.module.flags = !{ !0, !1, !2, !3 }
2814 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2815 if two or more ``!"foo"`` flags are seen is to emit an error if their
2816 values are not equal.
2818 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2819 behavior if two or more ``!"bar"`` flags are seen is to use the value
2822 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2823 behavior if two or more ``!"qux"`` flags are seen is to emit a
2824 warning if their values are not equal.
2826 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2830 metadata !{ metadata !"foo", i32 1 }
2832 The behavior is to emit an error if the ``llvm.module.flags`` does not
2833 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2836 Objective-C Garbage Collection Module Flags Metadata
2837 ----------------------------------------------------
2839 On the Mach-O platform, Objective-C stores metadata about garbage
2840 collection in a special section called "image info". The metadata
2841 consists of a version number and a bitmask specifying what types of
2842 garbage collection are supported (if any) by the file. If two or more
2843 modules are linked together their garbage collection metadata needs to
2844 be merged rather than appended together.
2846 The Objective-C garbage collection module flags metadata consists of the
2847 following key-value pairs:
2856 * - ``Objective-C Version``
2857 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2859 * - ``Objective-C Image Info Version``
2860 - **[Required]** --- The version of the image info section. Currently
2863 * - ``Objective-C Image Info Section``
2864 - **[Required]** --- The section to place the metadata. Valid values are
2865 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2866 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2867 Objective-C ABI version 2.
2869 * - ``Objective-C Garbage Collection``
2870 - **[Required]** --- Specifies whether garbage collection is supported or
2871 not. Valid values are 0, for no garbage collection, and 2, for garbage
2872 collection supported.
2874 * - ``Objective-C GC Only``
2875 - **[Optional]** --- Specifies that only garbage collection is supported.
2876 If present, its value must be 6. This flag requires that the
2877 ``Objective-C Garbage Collection`` flag have the value 2.
2879 Some important flag interactions:
2881 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2882 merged with a module with ``Objective-C Garbage Collection`` set to
2883 2, then the resulting module has the
2884 ``Objective-C Garbage Collection`` flag set to 0.
2885 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2886 merged with a module with ``Objective-C GC Only`` set to 6.
2888 Automatic Linker Flags Module Flags Metadata
2889 --------------------------------------------
2891 Some targets support embedding flags to the linker inside individual object
2892 files. Typically this is used in conjunction with language extensions which
2893 allow source files to explicitly declare the libraries they depend on, and have
2894 these automatically be transmitted to the linker via object files.
2896 These flags are encoded in the IR using metadata in the module flags section,
2897 using the ``Linker Options`` key. The merge behavior for this flag is required
2898 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2899 node which should be a list of other metadata nodes, each of which should be a
2900 list of metadata strings defining linker options.
2902 For example, the following metadata section specifies two separate sets of
2903 linker options, presumably to link against ``libz`` and the ``Cocoa``
2906 !0 = metadata !{ i32 6, metadata !"Linker Options",
2908 metadata !{ metadata !"-lz" },
2909 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2910 !llvm.module.flags = !{ !0 }
2912 The metadata encoding as lists of lists of options, as opposed to a collapsed
2913 list of options, is chosen so that the IR encoding can use multiple option
2914 strings to specify e.g., a single library, while still having that specifier be
2915 preserved as an atomic element that can be recognized by a target specific
2916 assembly writer or object file emitter.
2918 Each individual option is required to be either a valid option for the target's
2919 linker, or an option that is reserved by the target specific assembly writer or
2920 object file emitter. No other aspect of these options is defined by the IR.
2922 .. _intrinsicglobalvariables:
2924 Intrinsic Global Variables
2925 ==========================
2927 LLVM has a number of "magic" global variables that contain data that
2928 affect code generation or other IR semantics. These are documented here.
2929 All globals of this sort should have a section specified as
2930 "``llvm.metadata``". This section and all globals that start with
2931 "``llvm.``" are reserved for use by LLVM.
2935 The '``llvm.used``' Global Variable
2936 -----------------------------------
2938 The ``@llvm.used`` global is an array which has
2939 :ref:`appending linkage <linkage_appending>`. This array contains a list of
2940 pointers to named global variables, functions and aliases which may optionally
2941 have a pointer cast formed of bitcast or getelementptr. For example, a legal
2944 .. code-block:: llvm
2949 @llvm.used = appending global [2 x i8*] [
2951 i8* bitcast (i32* @Y to i8*)
2952 ], section "llvm.metadata"
2954 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
2955 and linker are required to treat the symbol as if there is a reference to the
2956 symbol that it cannot see (which is why they have to be named). For example, if
2957 a variable has internal linkage and no references other than that from the
2958 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
2959 references from inline asms and other things the compiler cannot "see", and
2960 corresponds to "``attribute((used))``" in GNU C.
2962 On some targets, the code generator must emit a directive to the
2963 assembler or object file to prevent the assembler and linker from
2964 molesting the symbol.
2966 .. _gv_llvmcompilerused:
2968 The '``llvm.compiler.used``' Global Variable
2969 --------------------------------------------
2971 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
2972 directive, except that it only prevents the compiler from touching the
2973 symbol. On targets that support it, this allows an intelligent linker to
2974 optimize references to the symbol without being impeded as it would be
2977 This is a rare construct that should only be used in rare circumstances,
2978 and should not be exposed to source languages.
2980 .. _gv_llvmglobalctors:
2982 The '``llvm.global_ctors``' Global Variable
2983 -------------------------------------------
2985 .. code-block:: llvm
2987 %0 = type { i32, void ()* }
2988 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2990 The ``@llvm.global_ctors`` array contains a list of constructor
2991 functions and associated priorities. The functions referenced by this
2992 array will be called in ascending order of priority (i.e. lowest first)
2993 when the module is loaded. The order of functions with the same priority
2996 .. _llvmglobaldtors:
2998 The '``llvm.global_dtors``' Global Variable
2999 -------------------------------------------
3001 .. code-block:: llvm
3003 %0 = type { i32, void ()* }
3004 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3006 The ``@llvm.global_dtors`` array contains a list of destructor functions
3007 and associated priorities. The functions referenced by this array will
3008 be called in descending order of priority (i.e. highest first) when the
3009 module is loaded. The order of functions with the same priority is not
3012 Instruction Reference
3013 =====================
3015 The LLVM instruction set consists of several different classifications
3016 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3017 instructions <binaryops>`, :ref:`bitwise binary
3018 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3019 :ref:`other instructions <otherops>`.
3023 Terminator Instructions
3024 -----------------------
3026 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3027 program ends with a "Terminator" instruction, which indicates which
3028 block should be executed after the current block is finished. These
3029 terminator instructions typically yield a '``void``' value: they produce
3030 control flow, not values (the one exception being the
3031 ':ref:`invoke <i_invoke>`' instruction).
3033 The terminator instructions are: ':ref:`ret <i_ret>`',
3034 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3035 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3036 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3040 '``ret``' Instruction
3041 ^^^^^^^^^^^^^^^^^^^^^
3048 ret <type> <value> ; Return a value from a non-void function
3049 ret void ; Return from void function
3054 The '``ret``' instruction is used to return control flow (and optionally
3055 a value) from a function back to the caller.
3057 There are two forms of the '``ret``' instruction: one that returns a
3058 value and then causes control flow, and one that just causes control
3064 The '``ret``' instruction optionally accepts a single argument, the
3065 return value. The type of the return value must be a ':ref:`first
3066 class <t_firstclass>`' type.
3068 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3069 return type and contains a '``ret``' instruction with no return value or
3070 a return value with a type that does not match its type, or if it has a
3071 void return type and contains a '``ret``' instruction with a return
3077 When the '``ret``' instruction is executed, control flow returns back to
3078 the calling function's context. If the caller is a
3079 ":ref:`call <i_call>`" instruction, execution continues at the
3080 instruction after the call. If the caller was an
3081 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3082 beginning of the "normal" destination block. If the instruction returns
3083 a value, that value shall set the call or invoke instruction's return
3089 .. code-block:: llvm
3091 ret i32 5 ; Return an integer value of 5
3092 ret void ; Return from a void function
3093 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3097 '``br``' Instruction
3098 ^^^^^^^^^^^^^^^^^^^^
3105 br i1 <cond>, label <iftrue>, label <iffalse>
3106 br label <dest> ; Unconditional branch
3111 The '``br``' instruction is used to cause control flow to transfer to a
3112 different basic block in the current function. There are two forms of
3113 this instruction, corresponding to a conditional branch and an
3114 unconditional branch.
3119 The conditional branch form of the '``br``' instruction takes a single
3120 '``i1``' value and two '``label``' values. The unconditional form of the
3121 '``br``' instruction takes a single '``label``' value as a target.
3126 Upon execution of a conditional '``br``' instruction, the '``i1``'
3127 argument is evaluated. If the value is ``true``, control flows to the
3128 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3129 to the '``iffalse``' ``label`` argument.
3134 .. code-block:: llvm
3137 %cond = icmp eq i32 %a, %b
3138 br i1 %cond, label %IfEqual, label %IfUnequal
3146 '``switch``' Instruction
3147 ^^^^^^^^^^^^^^^^^^^^^^^^
3154 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3159 The '``switch``' instruction is used to transfer control flow to one of
3160 several different places. It is a generalization of the '``br``'
3161 instruction, allowing a branch to occur to one of many possible
3167 The '``switch``' instruction uses three parameters: an integer
3168 comparison value '``value``', a default '``label``' destination, and an
3169 array of pairs of comparison value constants and '``label``'s. The table
3170 is not allowed to contain duplicate constant entries.
3175 The ``switch`` instruction specifies a table of values and destinations.
3176 When the '``switch``' instruction is executed, this table is searched
3177 for the given value. If the value is found, control flow is transferred
3178 to the corresponding destination; otherwise, control flow is transferred
3179 to the default destination.
3184 Depending on properties of the target machine and the particular
3185 ``switch`` instruction, this instruction may be code generated in
3186 different ways. For example, it could be generated as a series of
3187 chained conditional branches or with a lookup table.
3192 .. code-block:: llvm
3194 ; Emulate a conditional br instruction
3195 %Val = zext i1 %value to i32
3196 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3198 ; Emulate an unconditional br instruction
3199 switch i32 0, label %dest [ ]
3201 ; Implement a jump table:
3202 switch i32 %val, label %otherwise [ i32 0, label %onzero
3204 i32 2, label %ontwo ]
3208 '``indirectbr``' Instruction
3209 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3216 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3221 The '``indirectbr``' instruction implements an indirect branch to a
3222 label within the current function, whose address is specified by
3223 "``address``". Address must be derived from a
3224 :ref:`blockaddress <blockaddress>` constant.
3229 The '``address``' argument is the address of the label to jump to. The
3230 rest of the arguments indicate the full set of possible destinations
3231 that the address may point to. Blocks are allowed to occur multiple
3232 times in the destination list, though this isn't particularly useful.
3234 This destination list is required so that dataflow analysis has an
3235 accurate understanding of the CFG.
3240 Control transfers to the block specified in the address argument. All
3241 possible destination blocks must be listed in the label list, otherwise
3242 this instruction has undefined behavior. This implies that jumps to
3243 labels defined in other functions have undefined behavior as well.
3248 This is typically implemented with a jump through a register.
3253 .. code-block:: llvm
3255 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3259 '``invoke``' Instruction
3260 ^^^^^^^^^^^^^^^^^^^^^^^^
3267 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3268 to label <normal label> unwind label <exception label>
3273 The '``invoke``' instruction causes control to transfer to a specified
3274 function, with the possibility of control flow transfer to either the
3275 '``normal``' label or the '``exception``' label. If the callee function
3276 returns with the "``ret``" instruction, control flow will return to the
3277 "normal" label. If the callee (or any indirect callees) returns via the
3278 ":ref:`resume <i_resume>`" instruction or other exception handling
3279 mechanism, control is interrupted and continued at the dynamically
3280 nearest "exception" label.
3282 The '``exception``' label is a `landing
3283 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3284 '``exception``' label is required to have the
3285 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3286 information about the behavior of the program after unwinding happens,
3287 as its first non-PHI instruction. The restrictions on the
3288 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3289 instruction, so that the important information contained within the
3290 "``landingpad``" instruction can't be lost through normal code motion.
3295 This instruction requires several arguments:
3297 #. The optional "cconv" marker indicates which :ref:`calling
3298 convention <callingconv>` the call should use. If none is
3299 specified, the call defaults to using C calling conventions.
3300 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3301 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3303 #. '``ptr to function ty``': shall be the signature of the pointer to
3304 function value being invoked. In most cases, this is a direct
3305 function invocation, but indirect ``invoke``'s are just as possible,
3306 branching off an arbitrary pointer to function value.
3307 #. '``function ptr val``': An LLVM value containing a pointer to a
3308 function to be invoked.
3309 #. '``function args``': argument list whose types match the function
3310 signature argument types and parameter attributes. All arguments must
3311 be of :ref:`first class <t_firstclass>` type. If the function signature
3312 indicates the function accepts a variable number of arguments, the
3313 extra arguments can be specified.
3314 #. '``normal label``': the label reached when the called function
3315 executes a '``ret``' instruction.
3316 #. '``exception label``': the label reached when a callee returns via
3317 the :ref:`resume <i_resume>` instruction or other exception handling
3319 #. The optional :ref:`function attributes <fnattrs>` list. Only
3320 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3321 attributes are valid here.
3326 This instruction is designed to operate as a standard '``call``'
3327 instruction in most regards. The primary difference is that it
3328 establishes an association with a label, which is used by the runtime
3329 library to unwind the stack.
3331 This instruction is used in languages with destructors to ensure that
3332 proper cleanup is performed in the case of either a ``longjmp`` or a
3333 thrown exception. Additionally, this is important for implementation of
3334 '``catch``' clauses in high-level languages that support them.
3336 For the purposes of the SSA form, the definition of the value returned
3337 by the '``invoke``' instruction is deemed to occur on the edge from the
3338 current block to the "normal" label. If the callee unwinds then no
3339 return value is available.
3344 .. code-block:: llvm
3346 %retval = invoke i32 @Test(i32 15) to label %Continue
3347 unwind label %TestCleanup ; {i32}:retval set
3348 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3349 unwind label %TestCleanup ; {i32}:retval set
3353 '``resume``' Instruction
3354 ^^^^^^^^^^^^^^^^^^^^^^^^
3361 resume <type> <value>
3366 The '``resume``' instruction is a terminator instruction that has no
3372 The '``resume``' instruction requires one argument, which must have the
3373 same type as the result of any '``landingpad``' instruction in the same
3379 The '``resume``' instruction resumes propagation of an existing
3380 (in-flight) exception whose unwinding was interrupted with a
3381 :ref:`landingpad <i_landingpad>` instruction.
3386 .. code-block:: llvm
3388 resume { i8*, i32 } %exn
3392 '``unreachable``' Instruction
3393 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3405 The '``unreachable``' instruction has no defined semantics. This
3406 instruction is used to inform the optimizer that a particular portion of
3407 the code is not reachable. This can be used to indicate that the code
3408 after a no-return function cannot be reached, and other facts.
3413 The '``unreachable``' instruction has no defined semantics.
3420 Binary operators are used to do most of the computation in a program.
3421 They require two operands of the same type, execute an operation on
3422 them, and produce a single value. The operands might represent multiple
3423 data, as is the case with the :ref:`vector <t_vector>` data type. The
3424 result value has the same type as its operands.
3426 There are several different binary operators:
3430 '``add``' Instruction
3431 ^^^^^^^^^^^^^^^^^^^^^
3438 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3439 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3440 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3441 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3446 The '``add``' instruction returns the sum of its two operands.
3451 The two arguments to the '``add``' instruction must be
3452 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3453 arguments must have identical types.
3458 The value produced is the integer sum of the two operands.
3460 If the sum has unsigned overflow, the result returned is the
3461 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3464 Because LLVM integers use a two's complement representation, this
3465 instruction is appropriate for both signed and unsigned integers.
3467 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3468 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3469 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3470 unsigned and/or signed overflow, respectively, occurs.
3475 .. code-block:: llvm
3477 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3481 '``fadd``' Instruction
3482 ^^^^^^^^^^^^^^^^^^^^^^
3489 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3494 The '``fadd``' instruction returns the sum of its two operands.
3499 The two arguments to the '``fadd``' instruction must be :ref:`floating
3500 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3501 Both arguments must have identical types.
3506 The value produced is the floating point sum of the two operands. This
3507 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3508 which are optimization hints to enable otherwise unsafe floating point
3514 .. code-block:: llvm
3516 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3518 '``sub``' Instruction
3519 ^^^^^^^^^^^^^^^^^^^^^
3526 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3527 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3528 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3529 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3534 The '``sub``' instruction returns the difference of its two operands.
3536 Note that the '``sub``' instruction is used to represent the '``neg``'
3537 instruction present in most other intermediate representations.
3542 The two arguments to the '``sub``' instruction must be
3543 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3544 arguments must have identical types.
3549 The value produced is the integer difference of the two operands.
3551 If the difference has unsigned overflow, the result returned is the
3552 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3555 Because LLVM integers use a two's complement representation, this
3556 instruction is appropriate for both signed and unsigned integers.
3558 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3559 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3560 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3561 unsigned and/or signed overflow, respectively, occurs.
3566 .. code-block:: llvm
3568 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3569 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3573 '``fsub``' Instruction
3574 ^^^^^^^^^^^^^^^^^^^^^^
3581 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3586 The '``fsub``' instruction returns the difference of its two operands.
3588 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3589 instruction present in most other intermediate representations.
3594 The two arguments to the '``fsub``' instruction must be :ref:`floating
3595 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3596 Both arguments must have identical types.
3601 The value produced is the floating point difference of the two operands.
3602 This instruction can also take any number of :ref:`fast-math
3603 flags <fastmath>`, which are optimization hints to enable otherwise
3604 unsafe floating point optimizations:
3609 .. code-block:: llvm
3611 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3612 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3614 '``mul``' Instruction
3615 ^^^^^^^^^^^^^^^^^^^^^
3622 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3623 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3624 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3625 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3630 The '``mul``' instruction returns the product of its two operands.
3635 The two arguments to the '``mul``' instruction must be
3636 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3637 arguments must have identical types.
3642 The value produced is the integer product of the two operands.
3644 If the result of the multiplication has unsigned overflow, the result
3645 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3646 bit width of the result.
3648 Because LLVM integers use a two's complement representation, and the
3649 result is the same width as the operands, this instruction returns the
3650 correct result for both signed and unsigned integers. If a full product
3651 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3652 sign-extended or zero-extended as appropriate to the width of the full
3655 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3656 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3657 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3658 unsigned and/or signed overflow, respectively, occurs.
3663 .. code-block:: llvm
3665 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3669 '``fmul``' Instruction
3670 ^^^^^^^^^^^^^^^^^^^^^^
3677 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3682 The '``fmul``' instruction returns the product of its two operands.
3687 The two arguments to the '``fmul``' 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 The value produced is the floating point product of the two operands.
3695 This instruction can also take any number of :ref:`fast-math
3696 flags <fastmath>`, which are optimization hints to enable otherwise
3697 unsafe floating point optimizations:
3702 .. code-block:: llvm
3704 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3706 '``udiv``' Instruction
3707 ^^^^^^^^^^^^^^^^^^^^^^
3714 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3715 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3720 The '``udiv``' instruction returns the quotient of its two operands.
3725 The two arguments to the '``udiv``' instruction must be
3726 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3727 arguments must have identical types.
3732 The value produced is the unsigned integer quotient of the two operands.
3734 Note that unsigned integer division and signed integer division are
3735 distinct operations; for signed integer division, use '``sdiv``'.
3737 Division by zero leads to undefined behavior.
3739 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3740 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3741 such, "((a udiv exact b) mul b) == a").
3746 .. code-block:: llvm
3748 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3750 '``sdiv``' Instruction
3751 ^^^^^^^^^^^^^^^^^^^^^^
3758 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3759 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3764 The '``sdiv``' instruction returns the quotient of its two operands.
3769 The two arguments to the '``sdiv``' instruction must be
3770 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3771 arguments must have identical types.
3776 The value produced is the signed integer quotient of the two operands
3777 rounded towards zero.
3779 Note that signed integer division and unsigned integer division are
3780 distinct operations; for unsigned integer division, use '``udiv``'.
3782 Division by zero leads to undefined behavior. Overflow also leads to
3783 undefined behavior; this is a rare case, but can occur, for example, by
3784 doing a 32-bit division of -2147483648 by -1.
3786 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3787 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3792 .. code-block:: llvm
3794 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3798 '``fdiv``' Instruction
3799 ^^^^^^^^^^^^^^^^^^^^^^
3806 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3811 The '``fdiv``' instruction returns the quotient of its two operands.
3816 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3817 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3818 Both arguments must have identical types.
3823 The value produced is the floating point quotient of the two operands.
3824 This instruction can also take any number of :ref:`fast-math
3825 flags <fastmath>`, which are optimization hints to enable otherwise
3826 unsafe floating point optimizations:
3831 .. code-block:: llvm
3833 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3835 '``urem``' Instruction
3836 ^^^^^^^^^^^^^^^^^^^^^^
3843 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3848 The '``urem``' instruction returns the remainder from the unsigned
3849 division of its two arguments.
3854 The two arguments to the '``urem``' instruction must be
3855 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3856 arguments must have identical types.
3861 This instruction returns the unsigned integer *remainder* of a division.
3862 This instruction always performs an unsigned division to get the
3865 Note that unsigned integer remainder and signed integer remainder are
3866 distinct operations; for signed integer remainder, use '``srem``'.
3868 Taking the remainder of a division by zero leads to undefined behavior.
3873 .. code-block:: llvm
3875 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3877 '``srem``' Instruction
3878 ^^^^^^^^^^^^^^^^^^^^^^
3885 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3890 The '``srem``' instruction returns the remainder from the signed
3891 division of its two operands. This instruction can also take
3892 :ref:`vector <t_vector>` versions of the values in which case the elements
3898 The two arguments to the '``srem``' instruction must be
3899 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3900 arguments must have identical types.
3905 This instruction returns the *remainder* of a division (where the result
3906 is either zero or has the same sign as the dividend, ``op1``), not the
3907 *modulo* operator (where the result is either zero or has the same sign
3908 as the divisor, ``op2``) of a value. For more information about the
3909 difference, see `The Math
3910 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3911 table of how this is implemented in various languages, please see
3913 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3915 Note that signed integer remainder and unsigned integer remainder are
3916 distinct operations; for unsigned integer remainder, use '``urem``'.
3918 Taking the remainder of a division by zero leads to undefined behavior.
3919 Overflow also leads to undefined behavior; this is a rare case, but can
3920 occur, for example, by taking the remainder of a 32-bit division of
3921 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3922 rule lets srem be implemented using instructions that return both the
3923 result of the division and the remainder.)
3928 .. code-block:: llvm
3930 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
3934 '``frem``' Instruction
3935 ^^^^^^^^^^^^^^^^^^^^^^
3942 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3947 The '``frem``' instruction returns the remainder from the division of
3953 The two arguments to the '``frem``' instruction must be :ref:`floating
3954 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3955 Both arguments must have identical types.
3960 This instruction returns the *remainder* of a division. The remainder
3961 has the same sign as the dividend. This instruction can also take any
3962 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
3963 to enable otherwise unsafe floating point optimizations:
3968 .. code-block:: llvm
3970 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
3974 Bitwise Binary Operations
3975 -------------------------
3977 Bitwise binary operators are used to do various forms of bit-twiddling
3978 in a program. They are generally very efficient instructions and can
3979 commonly be strength reduced from other instructions. They require two
3980 operands of the same type, execute an operation on them, and produce a
3981 single value. The resulting value is the same type as its operands.
3983 '``shl``' Instruction
3984 ^^^^^^^^^^^^^^^^^^^^^
3991 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
3992 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
3993 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
3994 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3999 The '``shl``' instruction returns the first operand shifted to the left
4000 a specified number of bits.
4005 Both arguments to the '``shl``' instruction must be the same
4006 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4007 '``op2``' is treated as an unsigned value.
4012 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4013 where ``n`` is the width of the result. If ``op2`` is (statically or
4014 dynamically) negative or equal to or larger than the number of bits in
4015 ``op1``, the result is undefined. If the arguments are vectors, each
4016 vector element of ``op1`` is shifted by the corresponding shift amount
4019 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4020 value <poisonvalues>` if it shifts out any non-zero bits. If the
4021 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4022 value <poisonvalues>` if it shifts out any bits that disagree with the
4023 resultant sign bit. As such, NUW/NSW have the same semantics as they
4024 would if the shift were expressed as a mul instruction with the same
4025 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4030 .. code-block:: llvm
4032 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4033 <result> = shl i32 4, 2 ; yields {i32}: 16
4034 <result> = shl i32 1, 10 ; yields {i32}: 1024
4035 <result> = shl i32 1, 32 ; undefined
4036 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4038 '``lshr``' Instruction
4039 ^^^^^^^^^^^^^^^^^^^^^^
4046 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4047 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4052 The '``lshr``' instruction (logical shift right) returns the first
4053 operand shifted to the right a specified number of bits with zero fill.
4058 Both arguments to the '``lshr``' instruction must be the same
4059 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4060 '``op2``' is treated as an unsigned value.
4065 This instruction always performs a logical shift right operation. The
4066 most significant bits of the result will be filled with zero bits after
4067 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4068 than the number of bits in ``op1``, the result is undefined. If the
4069 arguments are vectors, each vector element of ``op1`` is shifted by the
4070 corresponding shift amount in ``op2``.
4072 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4073 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4079 .. code-block:: llvm
4081 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4082 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4083 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4084 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4085 <result> = lshr i32 1, 32 ; undefined
4086 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4088 '``ashr``' Instruction
4089 ^^^^^^^^^^^^^^^^^^^^^^
4096 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4097 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4102 The '``ashr``' instruction (arithmetic shift right) returns the first
4103 operand shifted to the right a specified number of bits with sign
4109 Both arguments to the '``ashr``' instruction must be the same
4110 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4111 '``op2``' is treated as an unsigned value.
4116 This instruction always performs an arithmetic shift right operation,
4117 The most significant bits of the result will be filled with the sign bit
4118 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4119 than the number of bits in ``op1``, the result is undefined. If the
4120 arguments are vectors, each vector element of ``op1`` is shifted by the
4121 corresponding shift amount in ``op2``.
4123 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4124 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4130 .. code-block:: llvm
4132 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4133 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4134 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4135 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4136 <result> = ashr i32 1, 32 ; undefined
4137 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4139 '``and``' Instruction
4140 ^^^^^^^^^^^^^^^^^^^^^
4147 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4152 The '``and``' instruction returns the bitwise logical and of its two
4158 The two arguments to the '``and``' instruction must be
4159 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4160 arguments must have identical types.
4165 The truth table used for the '``and``' instruction is:
4182 .. code-block:: llvm
4184 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4185 <result> = and i32 15, 40 ; yields {i32}:result = 8
4186 <result> = and i32 4, 8 ; yields {i32}:result = 0
4188 '``or``' Instruction
4189 ^^^^^^^^^^^^^^^^^^^^
4196 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4201 The '``or``' instruction returns the bitwise logical inclusive or of its
4207 The two arguments to the '``or``' instruction must be
4208 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4209 arguments must have identical types.
4214 The truth table used for the '``or``' instruction is:
4233 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4234 <result> = or i32 15, 40 ; yields {i32}:result = 47
4235 <result> = or i32 4, 8 ; yields {i32}:result = 12
4237 '``xor``' Instruction
4238 ^^^^^^^^^^^^^^^^^^^^^
4245 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4250 The '``xor``' instruction returns the bitwise logical exclusive or of
4251 its two operands. The ``xor`` is used to implement the "one's
4252 complement" operation, which is the "~" operator in C.
4257 The two arguments to the '``xor``' instruction must be
4258 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4259 arguments must have identical types.
4264 The truth table used for the '``xor``' instruction is:
4281 .. code-block:: llvm
4283 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4284 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4285 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4286 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4291 LLVM supports several instructions to represent vector operations in a
4292 target-independent manner. These instructions cover the element-access
4293 and vector-specific operations needed to process vectors effectively.
4294 While LLVM does directly support these vector operations, many
4295 sophisticated algorithms will want to use target-specific intrinsics to
4296 take full advantage of a specific target.
4298 .. _i_extractelement:
4300 '``extractelement``' Instruction
4301 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4308 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4313 The '``extractelement``' instruction extracts a single scalar element
4314 from a vector at a specified index.
4319 The first operand of an '``extractelement``' instruction is a value of
4320 :ref:`vector <t_vector>` type. The second operand is an index indicating
4321 the position from which to extract the element. The index may be a
4327 The result is a scalar of the same type as the element type of ``val``.
4328 Its value is the value at position ``idx`` of ``val``. If ``idx``
4329 exceeds the length of ``val``, the results are undefined.
4334 .. code-block:: llvm
4336 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4338 .. _i_insertelement:
4340 '``insertelement``' Instruction
4341 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4348 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4353 The '``insertelement``' instruction inserts a scalar element into a
4354 vector at a specified index.
4359 The first operand of an '``insertelement``' instruction is a value of
4360 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4361 type must equal the element type of the first operand. The third operand
4362 is an index indicating the position at which to insert the value. The
4363 index may be a variable.
4368 The result is a vector of the same type as ``val``. Its element values
4369 are those of ``val`` except at position ``idx``, where it gets the value
4370 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4376 .. code-block:: llvm
4378 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4380 .. _i_shufflevector:
4382 '``shufflevector``' Instruction
4383 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4390 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4395 The '``shufflevector``' instruction constructs a permutation of elements
4396 from two input vectors, returning a vector with the same element type as
4397 the input and length that is the same as the shuffle mask.
4402 The first two operands of a '``shufflevector``' instruction are vectors
4403 with the same type. The third argument is a shuffle mask whose element
4404 type is always 'i32'. The result of the instruction is a vector whose
4405 length is the same as the shuffle mask and whose element type is the
4406 same as the element type of the first two operands.
4408 The shuffle mask operand is required to be a constant vector with either
4409 constant integer or undef values.
4414 The elements of the two input vectors are numbered from left to right
4415 across both of the vectors. The shuffle mask operand specifies, for each
4416 element of the result vector, which element of the two input vectors the
4417 result element gets. The element selector may be undef (meaning "don't
4418 care") and the second operand may be undef if performing a shuffle from
4424 .. code-block:: llvm
4426 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4427 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4428 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4429 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4430 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4431 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4432 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4433 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4435 Aggregate Operations
4436 --------------------
4438 LLVM supports several instructions for working with
4439 :ref:`aggregate <t_aggregate>` values.
4443 '``extractvalue``' Instruction
4444 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4451 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4456 The '``extractvalue``' instruction extracts the value of a member field
4457 from an :ref:`aggregate <t_aggregate>` value.
4462 The first operand of an '``extractvalue``' instruction is a value of
4463 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4464 constant indices to specify which value to extract in a similar manner
4465 as indices in a '``getelementptr``' instruction.
4467 The major differences to ``getelementptr`` indexing are:
4469 - Since the value being indexed is not a pointer, the first index is
4470 omitted and assumed to be zero.
4471 - At least one index must be specified.
4472 - Not only struct indices but also array indices must be in bounds.
4477 The result is the value at the position in the aggregate specified by
4483 .. code-block:: llvm
4485 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4489 '``insertvalue``' Instruction
4490 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4497 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4502 The '``insertvalue``' instruction inserts a value into a member field in
4503 an :ref:`aggregate <t_aggregate>` value.
4508 The first operand of an '``insertvalue``' instruction is a value of
4509 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4510 a first-class value to insert. The following operands are constant
4511 indices indicating the position at which to insert the value in a
4512 similar manner as indices in a '``extractvalue``' instruction. The value
4513 to insert must have the same type as the value identified by the
4519 The result is an aggregate of the same type as ``val``. Its value is
4520 that of ``val`` except that the value at the position specified by the
4521 indices is that of ``elt``.
4526 .. code-block:: llvm
4528 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4529 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4530 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4534 Memory Access and Addressing Operations
4535 ---------------------------------------
4537 A key design point of an SSA-based representation is how it represents
4538 memory. In LLVM, no memory locations are in SSA form, which makes things
4539 very simple. This section describes how to read, write, and allocate
4544 '``alloca``' Instruction
4545 ^^^^^^^^^^^^^^^^^^^^^^^^
4552 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4557 The '``alloca``' instruction allocates memory on the stack frame of the
4558 currently executing function, to be automatically released when this
4559 function returns to its caller. The object is always allocated in the
4560 generic address space (address space zero).
4565 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4566 bytes of memory on the runtime stack, returning a pointer of the
4567 appropriate type to the program. If "NumElements" is specified, it is
4568 the number of elements allocated, otherwise "NumElements" is defaulted
4569 to be one. If a constant alignment is specified, the value result of the
4570 allocation is guaranteed to be aligned to at least that boundary. If not
4571 specified, or if zero, the target can choose to align the allocation on
4572 any convenient boundary compatible with the type.
4574 '``type``' may be any sized type.
4579 Memory is allocated; a pointer is returned. The operation is undefined
4580 if there is insufficient stack space for the allocation. '``alloca``'d
4581 memory is automatically released when the function returns. The
4582 '``alloca``' instruction is commonly used to represent automatic
4583 variables that must have an address available. When the function returns
4584 (either with the ``ret`` or ``resume`` instructions), the memory is
4585 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4586 The order in which memory is allocated (ie., which way the stack grows)
4592 .. code-block:: llvm
4594 %ptr = alloca i32 ; yields {i32*}:ptr
4595 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4596 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4597 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4601 '``load``' Instruction
4602 ^^^^^^^^^^^^^^^^^^^^^^
4609 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4610 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4611 !<index> = !{ i32 1 }
4616 The '``load``' instruction is used to read from memory.
4621 The argument to the ``load`` instruction specifies the memory address
4622 from which to load. The pointer must point to a :ref:`first
4623 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4624 then the optimizer is not allowed to modify the number or order of
4625 execution of this ``load`` with other :ref:`volatile
4626 operations <volatile>`.
4628 If the ``load`` is marked as ``atomic``, it takes an extra
4629 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4630 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4631 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4632 when they may see multiple atomic stores. The type of the pointee must
4633 be an integer type whose bit width is a power of two greater than or
4634 equal to eight and less than or equal to a target-specific size limit.
4635 ``align`` must be explicitly specified on atomic loads, and the load has
4636 undefined behavior if the alignment is not set to a value which is at
4637 least the size in bytes of the pointee. ``!nontemporal`` does not have
4638 any defined semantics for atomic loads.
4640 The optional constant ``align`` argument specifies the alignment of the
4641 operation (that is, the alignment of the memory address). A value of 0
4642 or an omitted ``align`` argument means that the operation has the ABI
4643 alignment for the target. It is the responsibility of the code emitter
4644 to ensure that the alignment information is correct. Overestimating the
4645 alignment results in undefined behavior. Underestimating the alignment
4646 may produce less efficient code. An alignment of 1 is always safe.
4648 The optional ``!nontemporal`` metadata must reference a single
4649 metadata name ``<index>`` corresponding to a metadata node with one
4650 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4651 metadata on the instruction tells the optimizer and code generator
4652 that this load is not expected to be reused in the cache. The code
4653 generator may select special instructions to save cache bandwidth, such
4654 as the ``MOVNT`` instruction on x86.
4656 The optional ``!invariant.load`` metadata must reference a single
4657 metadata name ``<index>`` corresponding to a metadata node with no
4658 entries. The existence of the ``!invariant.load`` metadata on the
4659 instruction tells the optimizer and code generator that this load
4660 address points to memory which does not change value during program
4661 execution. The optimizer may then move this load around, for example, by
4662 hoisting it out of loops using loop invariant code motion.
4667 The location of memory pointed to is loaded. If the value being loaded
4668 is of scalar type then the number of bytes read does not exceed the
4669 minimum number of bytes needed to hold all bits of the type. For
4670 example, loading an ``i24`` reads at most three bytes. When loading a
4671 value of a type like ``i20`` with a size that is not an integral number
4672 of bytes, the result is undefined if the value was not originally
4673 written using a store of the same type.
4678 .. code-block:: llvm
4680 %ptr = alloca i32 ; yields {i32*}:ptr
4681 store i32 3, i32* %ptr ; yields {void}
4682 %val = load i32* %ptr ; yields {i32}:val = i32 3
4686 '``store``' Instruction
4687 ^^^^^^^^^^^^^^^^^^^^^^^
4694 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4695 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4700 The '``store``' instruction is used to write to memory.
4705 There are two arguments to the ``store`` instruction: a value to store
4706 and an address at which to store it. The type of the ``<pointer>``
4707 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4708 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4709 then the optimizer is not allowed to modify the number or order of
4710 execution of this ``store`` with other :ref:`volatile
4711 operations <volatile>`.
4713 If the ``store`` is marked as ``atomic``, it takes an extra
4714 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4715 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4716 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4717 when they may see multiple atomic stores. The type of the pointee must
4718 be an integer type whose bit width is a power of two greater than or
4719 equal to eight and less than or equal to a target-specific size limit.
4720 ``align`` must be explicitly specified on atomic stores, and the store
4721 has undefined behavior if the alignment is not set to a value which is
4722 at least the size in bytes of the pointee. ``!nontemporal`` does not
4723 have any defined semantics for atomic stores.
4725 The optional constant ``align`` argument specifies the alignment of the
4726 operation (that is, the alignment of the memory address). A value of 0
4727 or an omitted ``align`` argument means that the operation has the ABI
4728 alignment for the target. It is the responsibility of the code emitter
4729 to ensure that the alignment information is correct. Overestimating the
4730 alignment results in undefined behavior. Underestimating the
4731 alignment may produce less efficient code. An alignment of 1 is always
4734 The optional ``!nontemporal`` metadata must reference a single metadata
4735 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4736 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4737 tells the optimizer and code generator that this load is not expected to
4738 be reused in the cache. The code generator may select special
4739 instructions to save cache bandwidth, such as the MOVNT instruction on
4745 The contents of memory are updated to contain ``<value>`` at the
4746 location specified by the ``<pointer>`` operand. If ``<value>`` is
4747 of scalar type then the number of bytes written does not exceed the
4748 minimum number of bytes needed to hold all bits of the type. For
4749 example, storing an ``i24`` writes at most three bytes. When writing a
4750 value of a type like ``i20`` with a size that is not an integral number
4751 of bytes, it is unspecified what happens to the extra bits that do not
4752 belong to the type, but they will typically be overwritten.
4757 .. code-block:: llvm
4759 %ptr = alloca i32 ; yields {i32*}:ptr
4760 store i32 3, i32* %ptr ; yields {void}
4761 %val = load i32* %ptr ; yields {i32}:val = i32 3
4765 '``fence``' Instruction
4766 ^^^^^^^^^^^^^^^^^^^^^^^
4773 fence [singlethread] <ordering> ; yields {void}
4778 The '``fence``' instruction is used to introduce happens-before edges
4784 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4785 defines what *synchronizes-with* edges they add. They can only be given
4786 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4791 A fence A which has (at least) ``release`` ordering semantics
4792 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4793 semantics if and only if there exist atomic operations X and Y, both
4794 operating on some atomic object M, such that A is sequenced before X, X
4795 modifies M (either directly or through some side effect of a sequence
4796 headed by X), Y is sequenced before B, and Y observes M. This provides a
4797 *happens-before* dependency between A and B. Rather than an explicit
4798 ``fence``, one (but not both) of the atomic operations X or Y might
4799 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4800 still *synchronize-with* the explicit ``fence`` and establish the
4801 *happens-before* edge.
4803 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4804 ``acquire`` and ``release`` semantics specified above, participates in
4805 the global program order of other ``seq_cst`` operations and/or fences.
4807 The optional ":ref:`singlethread <singlethread>`" argument specifies
4808 that the fence only synchronizes with other fences in the same thread.
4809 (This is useful for interacting with signal handlers.)
4814 .. code-block:: llvm
4816 fence acquire ; yields {void}
4817 fence singlethread seq_cst ; yields {void}
4821 '``cmpxchg``' Instruction
4822 ^^^^^^^^^^^^^^^^^^^^^^^^^
4829 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4834 The '``cmpxchg``' instruction is used to atomically modify memory. It
4835 loads a value in memory and compares it to a given value. If they are
4836 equal, it stores a new value into the memory.
4841 There are three arguments to the '``cmpxchg``' instruction: an address
4842 to operate on, a value to compare to the value currently be at that
4843 address, and a new value to place at that address if the compared values
4844 are equal. The type of '<cmp>' must be an integer type whose bit width
4845 is a power of two greater than or equal to eight and less than or equal
4846 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4847 type, and the type of '<pointer>' must be a pointer to that type. If the
4848 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4849 to modify the number or order of execution of this ``cmpxchg`` with
4850 other :ref:`volatile operations <volatile>`.
4852 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4853 synchronizes with other atomic operations.
4855 The optional "``singlethread``" argument declares that the ``cmpxchg``
4856 is only atomic with respect to code (usually signal handlers) running in
4857 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4858 respect to all other code in the system.
4860 The pointer passed into cmpxchg must have alignment greater than or
4861 equal to the size in memory of the operand.
4866 The contents of memory at the location specified by the '``<pointer>``'
4867 operand is read and compared to '``<cmp>``'; if the read value is the
4868 equal, '``<new>``' is written. The original value at the location is
4871 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4872 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4873 atomic load with an ordering parameter determined by dropping any
4874 ``release`` part of the ``cmpxchg``'s ordering.
4879 .. code-block:: llvm
4882 %orig = atomic load i32* %ptr unordered ; yields {i32}
4886 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4887 %squared = mul i32 %cmp, %cmp
4888 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4889 %success = icmp eq i32 %cmp, %old
4890 br i1 %success, label %done, label %loop
4897 '``atomicrmw``' Instruction
4898 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4905 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4910 The '``atomicrmw``' instruction is used to atomically modify memory.
4915 There are three arguments to the '``atomicrmw``' instruction: an
4916 operation to apply, an address whose value to modify, an argument to the
4917 operation. The operation must be one of the following keywords:
4931 The type of '<value>' must be an integer type whose bit width is a power
4932 of two greater than or equal to eight and less than or equal to a
4933 target-specific size limit. The type of the '``<pointer>``' operand must
4934 be a pointer to that type. If the ``atomicrmw`` is marked as
4935 ``volatile``, then the optimizer is not allowed to modify the number or
4936 order of execution of this ``atomicrmw`` with other :ref:`volatile
4937 operations <volatile>`.
4942 The contents of memory at the location specified by the '``<pointer>``'
4943 operand are atomically read, modified, and written back. The original
4944 value at the location is returned. The modification is specified by the
4947 - xchg: ``*ptr = val``
4948 - add: ``*ptr = *ptr + val``
4949 - sub: ``*ptr = *ptr - val``
4950 - and: ``*ptr = *ptr & val``
4951 - nand: ``*ptr = ~(*ptr & val)``
4952 - or: ``*ptr = *ptr | val``
4953 - xor: ``*ptr = *ptr ^ val``
4954 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
4955 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
4956 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
4958 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
4964 .. code-block:: llvm
4966 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
4968 .. _i_getelementptr:
4970 '``getelementptr``' Instruction
4971 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4978 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4979 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4980 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
4985 The '``getelementptr``' instruction is used to get the address of a
4986 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
4987 address calculation only and does not access memory.
4992 The first argument is always a pointer or a vector of pointers, and
4993 forms the basis of the calculation. The remaining arguments are indices
4994 that indicate which of the elements of the aggregate object are indexed.
4995 The interpretation of each index is dependent on the type being indexed
4996 into. The first index always indexes the pointer value given as the
4997 first argument, the second index indexes a value of the type pointed to
4998 (not necessarily the value directly pointed to, since the first index
4999 can be non-zero), etc. The first type indexed into must be a pointer
5000 value, subsequent types can be arrays, vectors, and structs. Note that
5001 subsequent types being indexed into can never be pointers, since that
5002 would require loading the pointer before continuing calculation.
5004 The type of each index argument depends on the type it is indexing into.
5005 When indexing into a (optionally packed) structure, only ``i32`` integer
5006 **constants** are allowed (when using a vector of indices they must all
5007 be the **same** ``i32`` integer constant). When indexing into an array,
5008 pointer or vector, integers of any width are allowed, and they are not
5009 required to be constant. These integers are treated as signed values
5012 For example, let's consider a C code fragment and how it gets compiled
5028 int *foo(struct ST *s) {
5029 return &s[1].Z.B[5][13];
5032 The LLVM code generated by Clang is:
5034 .. code-block:: llvm
5036 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5037 %struct.ST = type { i32, double, %struct.RT }
5039 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5041 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5048 In the example above, the first index is indexing into the
5049 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5050 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5051 indexes into the third element of the structure, yielding a
5052 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5053 structure. The third index indexes into the second element of the
5054 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5055 dimensions of the array are subscripted into, yielding an '``i32``'
5056 type. The '``getelementptr``' instruction returns a pointer to this
5057 element, thus computing a value of '``i32*``' type.
5059 Note that it is perfectly legal to index partially through a structure,
5060 returning a pointer to an inner element. Because of this, the LLVM code
5061 for the given testcase is equivalent to:
5063 .. code-block:: llvm
5065 define i32* @foo(%struct.ST* %s) {
5066 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5067 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5068 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5069 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5070 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5074 If the ``inbounds`` keyword is present, the result value of the
5075 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5076 pointer is not an *in bounds* address of an allocated object, or if any
5077 of the addresses that would be formed by successive addition of the
5078 offsets implied by the indices to the base address with infinitely
5079 precise signed arithmetic are not an *in bounds* address of that
5080 allocated object. The *in bounds* addresses for an allocated object are
5081 all the addresses that point into the object, plus the address one byte
5082 past the end. In cases where the base is a vector of pointers the
5083 ``inbounds`` keyword applies to each of the computations element-wise.
5085 If the ``inbounds`` keyword is not present, the offsets are added to the
5086 base address with silently-wrapping two's complement arithmetic. If the
5087 offsets have a different width from the pointer, they are sign-extended
5088 or truncated to the width of the pointer. The result value of the
5089 ``getelementptr`` may be outside the object pointed to by the base
5090 pointer. The result value may not necessarily be used to access memory
5091 though, even if it happens to point into allocated storage. See the
5092 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5095 The getelementptr instruction is often confusing. For some more insight
5096 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5101 .. code-block:: llvm
5103 ; yields [12 x i8]*:aptr
5104 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5106 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5108 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5110 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5112 In cases where the pointer argument is a vector of pointers, each index
5113 must be a vector with the same number of elements. For example:
5115 .. code-block:: llvm
5117 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5119 Conversion Operations
5120 ---------------------
5122 The instructions in this category are the conversion instructions
5123 (casting) which all take a single operand and a type. They perform
5124 various bit conversions on the operand.
5126 '``trunc .. to``' Instruction
5127 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5134 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5139 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5144 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5145 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5146 of the same number of integers. The bit size of the ``value`` must be
5147 larger than the bit size of the destination type, ``ty2``. Equal sized
5148 types are not allowed.
5153 The '``trunc``' instruction truncates the high order bits in ``value``
5154 and converts the remaining bits to ``ty2``. Since the source size must
5155 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5156 It will always truncate bits.
5161 .. code-block:: llvm
5163 %X = trunc i32 257 to i8 ; yields i8:1
5164 %Y = trunc i32 123 to i1 ; yields i1:true
5165 %Z = trunc i32 122 to i1 ; yields i1:false
5166 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5168 '``zext .. to``' Instruction
5169 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5176 <result> = zext <ty> <value> to <ty2> ; yields ty2
5181 The '``zext``' instruction zero extends its operand to type ``ty2``.
5186 The '``zext``' instruction takes a value to cast, and a type to cast it
5187 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5188 the same number of integers. The bit size of the ``value`` must be
5189 smaller than the bit size of the destination type, ``ty2``.
5194 The ``zext`` fills the high order bits of the ``value`` with zero bits
5195 until it reaches the size of the destination type, ``ty2``.
5197 When zero extending from i1, the result will always be either 0 or 1.
5202 .. code-block:: llvm
5204 %X = zext i32 257 to i64 ; yields i64:257
5205 %Y = zext i1 true to i32 ; yields i32:1
5206 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5208 '``sext .. to``' Instruction
5209 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5216 <result> = sext <ty> <value> to <ty2> ; yields ty2
5221 The '``sext``' sign extends ``value`` to the type ``ty2``.
5226 The '``sext``' instruction takes a value to cast, and a type to cast it
5227 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5228 the same number of integers. The bit size of the ``value`` must be
5229 smaller than the bit size of the destination type, ``ty2``.
5234 The '``sext``' instruction performs a sign extension by copying the sign
5235 bit (highest order bit) of the ``value`` until it reaches the bit size
5236 of the type ``ty2``.
5238 When sign extending from i1, the extension always results in -1 or 0.
5243 .. code-block:: llvm
5245 %X = sext i8 -1 to i16 ; yields i16 :65535
5246 %Y = sext i1 true to i32 ; yields i32:-1
5247 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5249 '``fptrunc .. to``' Instruction
5250 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5257 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5262 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5267 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5268 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5269 The size of ``value`` must be larger than the size of ``ty2``. This
5270 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5275 The '``fptrunc``' instruction truncates a ``value`` from a larger
5276 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5277 point <t_floating>` type. If the value cannot fit within the
5278 destination type, ``ty2``, then the results are undefined.
5283 .. code-block:: llvm
5285 %X = fptrunc double 123.0 to float ; yields float:123.0
5286 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5288 '``fpext .. to``' Instruction
5289 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5296 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5301 The '``fpext``' extends a floating point ``value`` to a larger floating
5307 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5308 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5309 to. The source type must be smaller than the destination type.
5314 The '``fpext``' instruction extends the ``value`` from a smaller
5315 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5316 point <t_floating>` type. The ``fpext`` cannot be used to make a
5317 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5318 *no-op cast* for a floating point cast.
5323 .. code-block:: llvm
5325 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5326 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5328 '``fptoui .. to``' Instruction
5329 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5336 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5341 The '``fptoui``' converts a floating point ``value`` to its unsigned
5342 integer equivalent of type ``ty2``.
5347 The '``fptoui``' instruction takes a value to cast, which must be a
5348 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5349 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5350 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5351 type with the same number of elements as ``ty``
5356 The '``fptoui``' instruction converts its :ref:`floating
5357 point <t_floating>` operand into the nearest (rounding towards zero)
5358 unsigned integer value. If the value cannot fit in ``ty2``, the results
5364 .. code-block:: llvm
5366 %X = fptoui double 123.0 to i32 ; yields i32:123
5367 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5368 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5370 '``fptosi .. to``' Instruction
5371 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5378 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5383 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5384 ``value`` to type ``ty2``.
5389 The '``fptosi``' instruction takes a value to cast, which must be a
5390 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5391 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5392 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5393 type with the same number of elements as ``ty``
5398 The '``fptosi``' instruction converts its :ref:`floating
5399 point <t_floating>` operand into the nearest (rounding towards zero)
5400 signed integer value. If the value cannot fit in ``ty2``, the results
5406 .. code-block:: llvm
5408 %X = fptosi double -123.0 to i32 ; yields i32:-123
5409 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5410 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5412 '``uitofp .. to``' Instruction
5413 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5420 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5425 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5426 and converts that value to the ``ty2`` type.
5431 The '``uitofp``' instruction takes a value to cast, which must be a
5432 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5433 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5434 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5435 type with the same number of elements as ``ty``
5440 The '``uitofp``' instruction interprets its operand as an unsigned
5441 integer quantity and converts it to the corresponding floating point
5442 value. If the value cannot fit in the floating point value, the results
5448 .. code-block:: llvm
5450 %X = uitofp i32 257 to float ; yields float:257.0
5451 %Y = uitofp i8 -1 to double ; yields double:255.0
5453 '``sitofp .. to``' Instruction
5454 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5461 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5466 The '``sitofp``' instruction regards ``value`` as a signed integer and
5467 converts that value to the ``ty2`` type.
5472 The '``sitofp``' instruction takes a value to cast, which must be a
5473 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5474 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5475 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5476 type with the same number of elements as ``ty``
5481 The '``sitofp``' instruction interprets its operand as a signed integer
5482 quantity and converts it to the corresponding floating point value. If
5483 the value cannot fit in the floating point value, the results are
5489 .. code-block:: llvm
5491 %X = sitofp i32 257 to float ; yields float:257.0
5492 %Y = sitofp i8 -1 to double ; yields double:-1.0
5496 '``ptrtoint .. to``' Instruction
5497 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5504 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5509 The '``ptrtoint``' instruction converts the pointer or a vector of
5510 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5515 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5516 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5517 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5518 a vector of integers type.
5523 The '``ptrtoint``' instruction converts ``value`` to integer type
5524 ``ty2`` by interpreting the pointer value as an integer and either
5525 truncating or zero extending that value to the size of the integer type.
5526 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5527 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5528 the same size, then nothing is done (*no-op cast*) other than a type
5534 .. code-block:: llvm
5536 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5537 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5538 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5542 '``inttoptr .. to``' Instruction
5543 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5550 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5555 The '``inttoptr``' instruction converts an integer ``value`` to a
5556 pointer type, ``ty2``.
5561 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5562 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5568 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5569 applying either a zero extension or a truncation depending on the size
5570 of the integer ``value``. If ``value`` is larger than the size of a
5571 pointer then a truncation is done. If ``value`` is smaller than the size
5572 of a pointer then a zero extension is done. If they are the same size,
5573 nothing is done (*no-op cast*).
5578 .. code-block:: llvm
5580 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5581 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5582 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5583 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5587 '``bitcast .. to``' Instruction
5588 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5595 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5600 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5606 The '``bitcast``' instruction takes a value to cast, which must be a
5607 non-aggregate first class value, and a type to cast it to, which must
5608 also be a non-aggregate :ref:`first class <t_firstclass>` type. The bit
5609 sizes of ``value`` and the destination type, ``ty2``, must be identical.
5610 If the source type is a pointer, the destination type must also be a
5611 pointer. This instruction supports bitwise conversion of vectors to
5612 integers and to vectors of other types (as long as they have the same
5618 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It is
5619 always a *no-op cast* because no bits change with this conversion. The
5620 conversion is done as if the ``value`` had been stored to memory and
5621 read back as type ``ty2``. Pointer (or vector of pointers) types may
5622 only be converted to other pointer (or vector of pointers) types with
5623 this instruction. To convert pointers to other types, use the
5624 :ref:`inttoptr <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions
5630 .. code-block:: llvm
5632 %X = bitcast i8 255 to i8 ; yields i8 :-1
5633 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5634 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5635 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5642 The instructions in this category are the "miscellaneous" instructions,
5643 which defy better classification.
5647 '``icmp``' Instruction
5648 ^^^^^^^^^^^^^^^^^^^^^^
5655 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5660 The '``icmp``' instruction returns a boolean value or a vector of
5661 boolean values based on comparison of its two integer, integer vector,
5662 pointer, or pointer vector operands.
5667 The '``icmp``' instruction takes three operands. The first operand is
5668 the condition code indicating the kind of comparison to perform. It is
5669 not a value, just a keyword. The possible condition code are:
5672 #. ``ne``: not equal
5673 #. ``ugt``: unsigned greater than
5674 #. ``uge``: unsigned greater or equal
5675 #. ``ult``: unsigned less than
5676 #. ``ule``: unsigned less or equal
5677 #. ``sgt``: signed greater than
5678 #. ``sge``: signed greater or equal
5679 #. ``slt``: signed less than
5680 #. ``sle``: signed less or equal
5682 The remaining two arguments must be :ref:`integer <t_integer>` or
5683 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5684 must also be identical types.
5689 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5690 code given as ``cond``. The comparison performed always yields either an
5691 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5693 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5694 otherwise. No sign interpretation is necessary or performed.
5695 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5696 otherwise. No sign interpretation is necessary or performed.
5697 #. ``ugt``: interprets the operands as unsigned values and yields
5698 ``true`` if ``op1`` is greater than ``op2``.
5699 #. ``uge``: interprets the operands as unsigned values and yields
5700 ``true`` if ``op1`` is greater than or equal to ``op2``.
5701 #. ``ult``: interprets the operands as unsigned values and yields
5702 ``true`` if ``op1`` is less than ``op2``.
5703 #. ``ule``: interprets the operands as unsigned values and yields
5704 ``true`` if ``op1`` is less than or equal to ``op2``.
5705 #. ``sgt``: interprets the operands as signed values and yields ``true``
5706 if ``op1`` is greater than ``op2``.
5707 #. ``sge``: interprets the operands as signed values and yields ``true``
5708 if ``op1`` is greater than or equal to ``op2``.
5709 #. ``slt``: interprets the operands as signed values and yields ``true``
5710 if ``op1`` is less than ``op2``.
5711 #. ``sle``: interprets the operands as signed values and yields ``true``
5712 if ``op1`` is less than or equal to ``op2``.
5714 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5715 are compared as if they were integers.
5717 If the operands are integer vectors, then they are compared element by
5718 element. The result is an ``i1`` vector with the same number of elements
5719 as the values being compared. Otherwise, the result is an ``i1``.
5724 .. code-block:: llvm
5726 <result> = icmp eq i32 4, 5 ; yields: result=false
5727 <result> = icmp ne float* %X, %X ; yields: result=false
5728 <result> = icmp ult i16 4, 5 ; yields: result=true
5729 <result> = icmp sgt i16 4, 5 ; yields: result=false
5730 <result> = icmp ule i16 -4, 5 ; yields: result=false
5731 <result> = icmp sge i16 4, 5 ; yields: result=false
5733 Note that the code generator does not yet support vector types with the
5734 ``icmp`` instruction.
5738 '``fcmp``' Instruction
5739 ^^^^^^^^^^^^^^^^^^^^^^
5746 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5751 The '``fcmp``' instruction returns a boolean value or vector of boolean
5752 values based on comparison of its operands.
5754 If the operands are floating point scalars, then the result type is a
5755 boolean (:ref:`i1 <t_integer>`).
5757 If the operands are floating point vectors, then the result type is a
5758 vector of boolean with the same number of elements as the operands being
5764 The '``fcmp``' instruction takes three operands. The first operand is
5765 the condition code indicating the kind of comparison to perform. It is
5766 not a value, just a keyword. The possible condition code are:
5768 #. ``false``: no comparison, always returns false
5769 #. ``oeq``: ordered and equal
5770 #. ``ogt``: ordered and greater than
5771 #. ``oge``: ordered and greater than or equal
5772 #. ``olt``: ordered and less than
5773 #. ``ole``: ordered and less than or equal
5774 #. ``one``: ordered and not equal
5775 #. ``ord``: ordered (no nans)
5776 #. ``ueq``: unordered or equal
5777 #. ``ugt``: unordered or greater than
5778 #. ``uge``: unordered or greater than or equal
5779 #. ``ult``: unordered or less than
5780 #. ``ule``: unordered or less than or equal
5781 #. ``une``: unordered or not equal
5782 #. ``uno``: unordered (either nans)
5783 #. ``true``: no comparison, always returns true
5785 *Ordered* means that neither operand is a QNAN while *unordered* means
5786 that either operand may be a QNAN.
5788 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5789 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5790 type. They must have identical types.
5795 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5796 condition code given as ``cond``. If the operands are vectors, then the
5797 vectors are compared element by element. Each comparison performed
5798 always yields an :ref:`i1 <t_integer>` result, as follows:
5800 #. ``false``: always yields ``false``, regardless of operands.
5801 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5802 is equal to ``op2``.
5803 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5804 is greater than ``op2``.
5805 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5806 is greater than or equal to ``op2``.
5807 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5808 is less than ``op2``.
5809 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5810 is less than or equal to ``op2``.
5811 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5812 is not equal to ``op2``.
5813 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5814 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5816 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5817 greater than ``op2``.
5818 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5819 greater than or equal to ``op2``.
5820 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5822 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5823 less than or equal to ``op2``.
5824 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5825 not equal to ``op2``.
5826 #. ``uno``: yields ``true`` if either operand is a QNAN.
5827 #. ``true``: always yields ``true``, regardless of operands.
5832 .. code-block:: llvm
5834 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5835 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5836 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5837 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5839 Note that the code generator does not yet support vector types with the
5840 ``fcmp`` instruction.
5844 '``phi``' Instruction
5845 ^^^^^^^^^^^^^^^^^^^^^
5852 <result> = phi <ty> [ <val0>, <label0>], ...
5857 The '``phi``' instruction is used to implement the φ node in the SSA
5858 graph representing the function.
5863 The type of the incoming values is specified with the first type field.
5864 After this, the '``phi``' instruction takes a list of pairs as
5865 arguments, with one pair for each predecessor basic block of the current
5866 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5867 the value arguments to the PHI node. Only labels may be used as the
5870 There must be no non-phi instructions between the start of a basic block
5871 and the PHI instructions: i.e. PHI instructions must be first in a basic
5874 For the purposes of the SSA form, the use of each incoming value is
5875 deemed to occur on the edge from the corresponding predecessor block to
5876 the current block (but after any definition of an '``invoke``'
5877 instruction's return value on the same edge).
5882 At runtime, the '``phi``' instruction logically takes on the value
5883 specified by the pair corresponding to the predecessor basic block that
5884 executed just prior to the current block.
5889 .. code-block:: llvm
5891 Loop: ; Infinite loop that counts from 0 on up...
5892 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5893 %nextindvar = add i32 %indvar, 1
5898 '``select``' Instruction
5899 ^^^^^^^^^^^^^^^^^^^^^^^^
5906 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5908 selty is either i1 or {<N x i1>}
5913 The '``select``' instruction is used to choose one value based on a
5914 condition, without branching.
5919 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
5920 values indicating the condition, and two values of the same :ref:`first
5921 class <t_firstclass>` type. If the val1/val2 are vectors and the
5922 condition is a scalar, then entire vectors are selected, not individual
5928 If the condition is an i1 and it evaluates to 1, the instruction returns
5929 the first value argument; otherwise, it returns the second value
5932 If the condition is a vector of i1, then the value arguments must be
5933 vectors of the same size, and the selection is done element by element.
5938 .. code-block:: llvm
5940 %X = select i1 true, i8 17, i8 42 ; yields i8:17
5944 '``call``' Instruction
5945 ^^^^^^^^^^^^^^^^^^^^^^
5952 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
5957 The '``call``' instruction represents a simple function call.
5962 This instruction requires several arguments:
5964 #. The optional "tail" marker indicates that the callee function does
5965 not access any allocas or varargs in the caller. Note that calls may
5966 be marked "tail" even if they do not occur before a
5967 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
5968 function call is eligible for tail call optimization, but `might not
5969 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
5970 The code generator may optimize calls marked "tail" with either 1)
5971 automatic `sibling call
5972 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
5973 callee have matching signatures, or 2) forced tail call optimization
5974 when the following extra requirements are met:
5976 - Caller and callee both have the calling convention ``fastcc``.
5977 - The call is in tail position (ret immediately follows call and ret
5978 uses value of call or is void).
5979 - Option ``-tailcallopt`` is enabled, or
5980 ``llvm::GuaranteedTailCallOpt`` is ``true``.
5981 - `Platform specific constraints are
5982 met. <CodeGenerator.html#tailcallopt>`_
5984 #. The optional "cconv" marker indicates which :ref:`calling
5985 convention <callingconv>` the call should use. If none is
5986 specified, the call defaults to using C calling conventions. The
5987 calling convention of the call must match the calling convention of
5988 the target function, or else the behavior is undefined.
5989 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5990 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5992 #. '``ty``': the type of the call instruction itself which is also the
5993 type of the return value. Functions that return no value are marked
5995 #. '``fnty``': shall be the signature of the pointer to function value
5996 being invoked. The argument types must match the types implied by
5997 this signature. This type can be omitted if the function is not
5998 varargs and if the function type does not return a pointer to a
6000 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6001 be invoked. In most cases, this is a direct function invocation, but
6002 indirect ``call``'s are just as possible, calling an arbitrary pointer
6004 #. '``function args``': argument list whose types match the function
6005 signature argument types and parameter attributes. All arguments must
6006 be of :ref:`first class <t_firstclass>` type. If the function signature
6007 indicates the function accepts a variable number of arguments, the
6008 extra arguments can be specified.
6009 #. The optional :ref:`function attributes <fnattrs>` list. Only
6010 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6011 attributes are valid here.
6016 The '``call``' instruction is used to cause control flow to transfer to
6017 a specified function, with its incoming arguments bound to the specified
6018 values. Upon a '``ret``' instruction in the called function, control
6019 flow continues with the instruction after the function call, and the
6020 return value of the function is bound to the result argument.
6025 .. code-block:: llvm
6027 %retval = call i32 @test(i32 %argc)
6028 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6029 %X = tail call i32 @foo() ; yields i32
6030 %Y = tail call fastcc i32 @foo() ; yields i32
6031 call void %foo(i8 97 signext)
6033 %struct.A = type { i32, i8 }
6034 %r = call %struct.A @foo() ; yields { 32, i8 }
6035 %gr = extractvalue %struct.A %r, 0 ; yields i32
6036 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6037 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6038 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6040 llvm treats calls to some functions with names and arguments that match
6041 the standard C99 library as being the C99 library functions, and may
6042 perform optimizations or generate code for them under that assumption.
6043 This is something we'd like to change in the future to provide better
6044 support for freestanding environments and non-C-based languages.
6048 '``va_arg``' Instruction
6049 ^^^^^^^^^^^^^^^^^^^^^^^^
6056 <resultval> = va_arg <va_list*> <arglist>, <argty>
6061 The '``va_arg``' instruction is used to access arguments passed through
6062 the "variable argument" area of a function call. It is used to implement
6063 the ``va_arg`` macro in C.
6068 This instruction takes a ``va_list*`` value and the type of the
6069 argument. It returns a value of the specified argument type and
6070 increments the ``va_list`` to point to the next argument. The actual
6071 type of ``va_list`` is target specific.
6076 The '``va_arg``' instruction loads an argument of the specified type
6077 from the specified ``va_list`` and causes the ``va_list`` to point to
6078 the next argument. For more information, see the variable argument
6079 handling :ref:`Intrinsic Functions <int_varargs>`.
6081 It is legal for this instruction to be called in a function which does
6082 not take a variable number of arguments, for example, the ``vfprintf``
6085 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6086 function <intrinsics>` because it takes a type as an argument.
6091 See the :ref:`variable argument processing <int_varargs>` section.
6093 Note that the code generator does not yet fully support va\_arg on many
6094 targets. Also, it does not currently support va\_arg with aggregate
6095 types on any target.
6099 '``landingpad``' Instruction
6100 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6107 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6108 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6110 <clause> := catch <type> <value>
6111 <clause> := filter <array constant type> <array constant>
6116 The '``landingpad``' instruction is used by `LLVM's exception handling
6117 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6118 is a landing pad --- one where the exception lands, and corresponds to the
6119 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6120 defines values supplied by the personality function (``pers_fn``) upon
6121 re-entry to the function. The ``resultval`` has the type ``resultty``.
6126 This instruction takes a ``pers_fn`` value. This is the personality
6127 function associated with the unwinding mechanism. The optional
6128 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6130 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6131 contains the global variable representing the "type" that may be caught
6132 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6133 clause takes an array constant as its argument. Use
6134 "``[0 x i8**] undef``" for a filter which cannot throw. The
6135 '``landingpad``' instruction must contain *at least* one ``clause`` or
6136 the ``cleanup`` flag.
6141 The '``landingpad``' instruction defines the values which are set by the
6142 personality function (``pers_fn``) upon re-entry to the function, and
6143 therefore the "result type" of the ``landingpad`` instruction. As with
6144 calling conventions, how the personality function results are
6145 represented in LLVM IR is target specific.
6147 The clauses are applied in order from top to bottom. If two
6148 ``landingpad`` instructions are merged together through inlining, the
6149 clauses from the calling function are appended to the list of clauses.
6150 When the call stack is being unwound due to an exception being thrown,
6151 the exception is compared against each ``clause`` in turn. If it doesn't
6152 match any of the clauses, and the ``cleanup`` flag is not set, then
6153 unwinding continues further up the call stack.
6155 The ``landingpad`` instruction has several restrictions:
6157 - A landing pad block is a basic block which is the unwind destination
6158 of an '``invoke``' instruction.
6159 - A landing pad block must have a '``landingpad``' instruction as its
6160 first non-PHI instruction.
6161 - There can be only one '``landingpad``' instruction within the landing
6163 - A basic block that is not a landing pad block may not include a
6164 '``landingpad``' instruction.
6165 - All '``landingpad``' instructions in a function must have the same
6166 personality function.
6171 .. code-block:: llvm
6173 ;; A landing pad which can catch an integer.
6174 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6176 ;; A landing pad that is a cleanup.
6177 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6179 ;; A landing pad which can catch an integer and can only throw a double.
6180 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6182 filter [1 x i8**] [@_ZTId]
6189 LLVM supports the notion of an "intrinsic function". These functions
6190 have well known names and semantics and are required to follow certain
6191 restrictions. Overall, these intrinsics represent an extension mechanism
6192 for the LLVM language that does not require changing all of the
6193 transformations in LLVM when adding to the language (or the bitcode
6194 reader/writer, the parser, etc...).
6196 Intrinsic function names must all start with an "``llvm.``" prefix. This
6197 prefix is reserved in LLVM for intrinsic names; thus, function names may
6198 not begin with this prefix. Intrinsic functions must always be external
6199 functions: you cannot define the body of intrinsic functions. Intrinsic
6200 functions may only be used in call or invoke instructions: it is illegal
6201 to take the address of an intrinsic function. Additionally, because
6202 intrinsic functions are part of the LLVM language, it is required if any
6203 are added that they be documented here.
6205 Some intrinsic functions can be overloaded, i.e., the intrinsic
6206 represents a family of functions that perform the same operation but on
6207 different data types. Because LLVM can represent over 8 million
6208 different integer types, overloading is used commonly to allow an
6209 intrinsic function to operate on any integer type. One or more of the
6210 argument types or the result type can be overloaded to accept any
6211 integer type. Argument types may also be defined as exactly matching a
6212 previous argument's type or the result type. This allows an intrinsic
6213 function which accepts multiple arguments, but needs all of them to be
6214 of the same type, to only be overloaded with respect to a single
6215 argument or the result.
6217 Overloaded intrinsics will have the names of its overloaded argument
6218 types encoded into its function name, each preceded by a period. Only
6219 those types which are overloaded result in a name suffix. Arguments
6220 whose type is matched against another type do not. For example, the
6221 ``llvm.ctpop`` function can take an integer of any width and returns an
6222 integer of exactly the same integer width. This leads to a family of
6223 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6224 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6225 overloaded, and only one type suffix is required. Because the argument's
6226 type is matched against the return type, it does not require its own
6229 To learn how to add an intrinsic function, please see the `Extending
6230 LLVM Guide <ExtendingLLVM.html>`_.
6234 Variable Argument Handling Intrinsics
6235 -------------------------------------
6237 Variable argument support is defined in LLVM with the
6238 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6239 functions. These functions are related to the similarly named macros
6240 defined in the ``<stdarg.h>`` header file.
6242 All of these functions operate on arguments that use a target-specific
6243 value type "``va_list``". The LLVM assembly language reference manual
6244 does not define what this type is, so all transformations should be
6245 prepared to handle these functions regardless of the type used.
6247 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6248 variable argument handling intrinsic functions are used.
6250 .. code-block:: llvm
6252 define i32 @test(i32 %X, ...) {
6253 ; Initialize variable argument processing
6255 %ap2 = bitcast i8** %ap to i8*
6256 call void @llvm.va_start(i8* %ap2)
6258 ; Read a single integer argument
6259 %tmp = va_arg i8** %ap, i32
6261 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6263 %aq2 = bitcast i8** %aq to i8*
6264 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6265 call void @llvm.va_end(i8* %aq2)
6267 ; Stop processing of arguments.
6268 call void @llvm.va_end(i8* %ap2)
6272 declare void @llvm.va_start(i8*)
6273 declare void @llvm.va_copy(i8*, i8*)
6274 declare void @llvm.va_end(i8*)
6278 '``llvm.va_start``' Intrinsic
6279 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6286 declare void %llvm.va_start(i8* <arglist>)
6291 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6292 subsequent use by ``va_arg``.
6297 The argument is a pointer to a ``va_list`` element to initialize.
6302 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6303 available in C. In a target-dependent way, it initializes the
6304 ``va_list`` element to which the argument points, so that the next call
6305 to ``va_arg`` will produce the first variable argument passed to the
6306 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6307 to know the last argument of the function as the compiler can figure
6310 '``llvm.va_end``' Intrinsic
6311 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6318 declare void @llvm.va_end(i8* <arglist>)
6323 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6324 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6329 The argument is a pointer to a ``va_list`` to destroy.
6334 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6335 available in C. In a target-dependent way, it destroys the ``va_list``
6336 element to which the argument points. Calls to
6337 :ref:`llvm.va_start <int_va_start>` and
6338 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6343 '``llvm.va_copy``' Intrinsic
6344 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6351 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6356 The '``llvm.va_copy``' intrinsic copies the current argument position
6357 from the source argument list to the destination argument list.
6362 The first argument is a pointer to a ``va_list`` element to initialize.
6363 The second argument is a pointer to a ``va_list`` element to copy from.
6368 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6369 available in C. In a target-dependent way, it copies the source
6370 ``va_list`` element into the destination ``va_list`` element. This
6371 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6372 arbitrarily complex and require, for example, memory allocation.
6374 Accurate Garbage Collection Intrinsics
6375 --------------------------------------
6377 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6378 (GC) requires the implementation and generation of these intrinsics.
6379 These intrinsics allow identification of :ref:`GC roots on the
6380 stack <int_gcroot>`, as well as garbage collector implementations that
6381 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6382 Front-ends for type-safe garbage collected languages should generate
6383 these intrinsics to make use of the LLVM garbage collectors. For more
6384 details, see `Accurate Garbage Collection with
6385 LLVM <GarbageCollection.html>`_.
6387 The garbage collection intrinsics only operate on objects in the generic
6388 address space (address space zero).
6392 '``llvm.gcroot``' Intrinsic
6393 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6400 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6405 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6406 the code generator, and allows some metadata to be associated with it.
6411 The first argument specifies the address of a stack object that contains
6412 the root pointer. The second pointer (which must be either a constant or
6413 a global value address) contains the meta-data to be associated with the
6419 At runtime, a call to this intrinsic stores a null pointer into the
6420 "ptrloc" location. At compile-time, the code generator generates
6421 information to allow the runtime to find the pointer at GC safe points.
6422 The '``llvm.gcroot``' intrinsic may only be used in a function which
6423 :ref:`specifies a GC algorithm <gc>`.
6427 '``llvm.gcread``' Intrinsic
6428 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6435 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6440 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6441 locations, allowing garbage collector implementations that require read
6447 The second argument is the address to read from, which should be an
6448 address allocated from the garbage collector. The first object is a
6449 pointer to the start of the referenced object, if needed by the language
6450 runtime (otherwise null).
6455 The '``llvm.gcread``' intrinsic has the same semantics as a load
6456 instruction, but may be replaced with substantially more complex code by
6457 the garbage collector runtime, as needed. The '``llvm.gcread``'
6458 intrinsic may only be used in a function which :ref:`specifies a GC
6463 '``llvm.gcwrite``' Intrinsic
6464 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6471 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6476 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6477 locations, allowing garbage collector implementations that require write
6478 barriers (such as generational or reference counting collectors).
6483 The first argument is the reference to store, the second is the start of
6484 the object to store it to, and the third is the address of the field of
6485 Obj to store to. If the runtime does not require a pointer to the
6486 object, Obj may be null.
6491 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6492 instruction, but may be replaced with substantially more complex code by
6493 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6494 intrinsic may only be used in a function which :ref:`specifies a GC
6497 Code Generator Intrinsics
6498 -------------------------
6500 These intrinsics are provided by LLVM to expose special features that
6501 may only be implemented with code generator support.
6503 '``llvm.returnaddress``' Intrinsic
6504 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6511 declare i8 *@llvm.returnaddress(i32 <level>)
6516 The '``llvm.returnaddress``' intrinsic attempts to compute a
6517 target-specific value indicating the return address of the current
6518 function or one of its callers.
6523 The argument to this intrinsic indicates which function to return the
6524 address for. Zero indicates the calling function, one indicates its
6525 caller, etc. The argument is **required** to be a constant integer
6531 The '``llvm.returnaddress``' intrinsic either returns a pointer
6532 indicating the return address of the specified call frame, or zero if it
6533 cannot be identified. The value returned by this intrinsic is likely to
6534 be incorrect or 0 for arguments other than zero, so it should only be
6535 used for debugging purposes.
6537 Note that calling this intrinsic does not prevent function inlining or
6538 other aggressive transformations, so the value returned may not be that
6539 of the obvious source-language caller.
6541 '``llvm.frameaddress``' Intrinsic
6542 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6549 declare i8* @llvm.frameaddress(i32 <level>)
6554 The '``llvm.frameaddress``' intrinsic attempts to return the
6555 target-specific frame pointer value for the specified stack frame.
6560 The argument to this intrinsic indicates which function to return the
6561 frame pointer for. Zero indicates the calling function, one indicates
6562 its caller, etc. The argument is **required** to be a constant integer
6568 The '``llvm.frameaddress``' intrinsic either returns a pointer
6569 indicating the frame address of the specified call frame, or zero if it
6570 cannot be identified. The value returned by this intrinsic is likely to
6571 be incorrect or 0 for arguments other than zero, so it should only be
6572 used for debugging purposes.
6574 Note that calling this intrinsic does not prevent function inlining or
6575 other aggressive transformations, so the value returned may not be that
6576 of the obvious source-language caller.
6580 '``llvm.stacksave``' Intrinsic
6581 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6588 declare i8* @llvm.stacksave()
6593 The '``llvm.stacksave``' intrinsic is used to remember the current state
6594 of the function stack, for use with
6595 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6596 implementing language features like scoped automatic variable sized
6602 This intrinsic returns a opaque pointer value that can be passed to
6603 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6604 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6605 ``llvm.stacksave``, it effectively restores the state of the stack to
6606 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6607 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6608 were allocated after the ``llvm.stacksave`` was executed.
6610 .. _int_stackrestore:
6612 '``llvm.stackrestore``' Intrinsic
6613 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6620 declare void @llvm.stackrestore(i8* %ptr)
6625 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6626 the function stack to the state it was in when the corresponding
6627 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6628 useful for implementing language features like scoped automatic variable
6629 sized arrays in C99.
6634 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6636 '``llvm.prefetch``' Intrinsic
6637 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6644 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6649 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6650 insert a prefetch instruction if supported; otherwise, it is a noop.
6651 Prefetches have no effect on the behavior of the program but can change
6652 its performance characteristics.
6657 ``address`` is the address to be prefetched, ``rw`` is the specifier
6658 determining if the fetch should be for a read (0) or write (1), and
6659 ``locality`` is a temporal locality specifier ranging from (0) - no
6660 locality, to (3) - extremely local keep in cache. The ``cache type``
6661 specifies whether the prefetch is performed on the data (1) or
6662 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6663 arguments must be constant integers.
6668 This intrinsic does not modify the behavior of the program. In
6669 particular, prefetches cannot trap and do not produce a value. On
6670 targets that support this intrinsic, the prefetch can provide hints to
6671 the processor cache for better performance.
6673 '``llvm.pcmarker``' Intrinsic
6674 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6681 declare void @llvm.pcmarker(i32 <id>)
6686 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6687 Counter (PC) in a region of code to simulators and other tools. The
6688 method is target specific, but it is expected that the marker will use
6689 exported symbols to transmit the PC of the marker. The marker makes no
6690 guarantees that it will remain with any specific instruction after
6691 optimizations. It is possible that the presence of a marker will inhibit
6692 optimizations. The intended use is to be inserted after optimizations to
6693 allow correlations of simulation runs.
6698 ``id`` is a numerical id identifying the marker.
6703 This intrinsic does not modify the behavior of the program. Backends
6704 that do not support this intrinsic may ignore it.
6706 '``llvm.readcyclecounter``' Intrinsic
6707 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6714 declare i64 @llvm.readcyclecounter()
6719 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6720 counter register (or similar low latency, high accuracy clocks) on those
6721 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6722 should map to RPCC. As the backing counters overflow quickly (on the
6723 order of 9 seconds on alpha), this should only be used for small
6729 When directly supported, reading the cycle counter should not modify any
6730 memory. Implementations are allowed to either return a application
6731 specific value or a system wide value. On backends without support, this
6732 is lowered to a constant 0.
6734 Note that runtime support may be conditional on the privilege-level code is
6735 running at and the host platform.
6737 Standard C Library Intrinsics
6738 -----------------------------
6740 LLVM provides intrinsics for a few important standard C library
6741 functions. These intrinsics allow source-language front-ends to pass
6742 information about the alignment of the pointer arguments to the code
6743 generator, providing opportunity for more efficient code generation.
6747 '``llvm.memcpy``' Intrinsic
6748 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6753 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6754 integer bit width and for different address spaces. Not all targets
6755 support all bit widths however.
6759 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6760 i32 <len>, i32 <align>, i1 <isvolatile>)
6761 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6762 i64 <len>, i32 <align>, i1 <isvolatile>)
6767 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6768 source location to the destination location.
6770 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6771 intrinsics do not return a value, takes extra alignment/isvolatile
6772 arguments and the pointers can be in specified address spaces.
6777 The first argument is a pointer to the destination, the second is a
6778 pointer to the source. The third argument is an integer argument
6779 specifying the number of bytes to copy, the fourth argument is the
6780 alignment of the source and destination locations, and the fifth is a
6781 boolean indicating a volatile access.
6783 If the call to this intrinsic has an alignment value that is not 0 or 1,
6784 then the caller guarantees that both the source and destination pointers
6785 are aligned to that boundary.
6787 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6788 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6789 very cleanly specified and it is unwise to depend on it.
6794 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6795 source location to the destination location, which are not allowed to
6796 overlap. It copies "len" bytes of memory over. If the argument is known
6797 to be aligned to some boundary, this can be specified as the fourth
6798 argument, otherwise it should be set to 0 or 1.
6800 '``llvm.memmove``' Intrinsic
6801 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6806 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6807 bit width and for different address space. Not all targets support all
6812 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6813 i32 <len>, i32 <align>, i1 <isvolatile>)
6814 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6815 i64 <len>, i32 <align>, i1 <isvolatile>)
6820 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6821 source location to the destination location. It is similar to the
6822 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6825 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6826 intrinsics do not return a value, takes extra alignment/isvolatile
6827 arguments and the pointers can be in specified address spaces.
6832 The first argument is a pointer to the destination, the second is a
6833 pointer to the source. The third argument is an integer argument
6834 specifying the number of bytes to copy, the fourth argument is the
6835 alignment of the source and destination locations, and the fifth is a
6836 boolean indicating a volatile access.
6838 If the call to this intrinsic has an alignment value that is not 0 or 1,
6839 then the caller guarantees that the source and destination pointers are
6840 aligned to that boundary.
6842 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6843 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6844 not very cleanly specified and it is unwise to depend on it.
6849 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6850 source location to the destination location, which may overlap. It
6851 copies "len" bytes of memory over. If the argument is known to be
6852 aligned to some boundary, this can be specified as the fourth argument,
6853 otherwise it should be set to 0 or 1.
6855 '``llvm.memset.*``' Intrinsics
6856 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6861 This is an overloaded intrinsic. You can use llvm.memset on any integer
6862 bit width and for different address spaces. However, not all targets
6863 support all bit widths.
6867 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6868 i32 <len>, i32 <align>, i1 <isvolatile>)
6869 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6870 i64 <len>, i32 <align>, i1 <isvolatile>)
6875 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6876 particular byte value.
6878 Note that, unlike the standard libc function, the ``llvm.memset``
6879 intrinsic does not return a value and takes extra alignment/volatile
6880 arguments. Also, the destination can be in an arbitrary address space.
6885 The first argument is a pointer to the destination to fill, the second
6886 is the byte value with which to fill it, the third argument is an
6887 integer argument specifying the number of bytes to fill, and the fourth
6888 argument is the known alignment of the destination location.
6890 If the call to this intrinsic has an alignment value that is not 0 or 1,
6891 then the caller guarantees that the destination pointer is aligned to
6894 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6895 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6896 very cleanly specified and it is unwise to depend on it.
6901 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6902 at the destination location. If the argument is known to be aligned to
6903 some boundary, this can be specified as the fourth argument, otherwise
6904 it should be set to 0 or 1.
6906 '``llvm.sqrt.*``' Intrinsic
6907 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6912 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6913 floating point or vector of floating point type. Not all targets support
6918 declare float @llvm.sqrt.f32(float %Val)
6919 declare double @llvm.sqrt.f64(double %Val)
6920 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6921 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6922 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6927 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
6928 returning the same value as the libm '``sqrt``' functions would. Unlike
6929 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
6930 negative numbers other than -0.0 (which allows for better optimization,
6931 because there is no need to worry about errno being set).
6932 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
6937 The argument and return value are floating point numbers of the same
6943 This function returns the sqrt of the specified operand if it is a
6944 nonnegative floating point number.
6946 '``llvm.powi.*``' Intrinsic
6947 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6952 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
6953 floating point or vector of floating point type. Not all targets support
6958 declare float @llvm.powi.f32(float %Val, i32 %power)
6959 declare double @llvm.powi.f64(double %Val, i32 %power)
6960 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6961 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6962 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6967 The '``llvm.powi.*``' intrinsics return the first operand raised to the
6968 specified (positive or negative) power. The order of evaluation of
6969 multiplications is not defined. When a vector of floating point type is
6970 used, the second argument remains a scalar integer value.
6975 The second argument is an integer power, and the first is a value to
6976 raise to that power.
6981 This function returns the first value raised to the second power with an
6982 unspecified sequence of rounding operations.
6984 '``llvm.sin.*``' Intrinsic
6985 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6990 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
6991 floating point or vector of floating point type. Not all targets support
6996 declare float @llvm.sin.f32(float %Val)
6997 declare double @llvm.sin.f64(double %Val)
6998 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6999 declare fp128 @llvm.sin.f128(fp128 %Val)
7000 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7005 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7010 The argument and return value are floating point numbers of the same
7016 This function returns the sine of the specified operand, returning the
7017 same values as the libm ``sin`` functions would, and handles error
7018 conditions in the same way.
7020 '``llvm.cos.*``' Intrinsic
7021 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7026 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7027 floating point or vector of floating point type. Not all targets support
7032 declare float @llvm.cos.f32(float %Val)
7033 declare double @llvm.cos.f64(double %Val)
7034 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7035 declare fp128 @llvm.cos.f128(fp128 %Val)
7036 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7041 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7046 The argument and return value are floating point numbers of the same
7052 This function returns the cosine of the specified operand, returning the
7053 same values as the libm ``cos`` functions would, and handles error
7054 conditions in the same way.
7056 '``llvm.pow.*``' Intrinsic
7057 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7062 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7063 floating point or vector of floating point type. Not all targets support
7068 declare float @llvm.pow.f32(float %Val, float %Power)
7069 declare double @llvm.pow.f64(double %Val, double %Power)
7070 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7071 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7072 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7077 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7078 specified (positive or negative) power.
7083 The second argument is a floating point power, and the first is a value
7084 to raise to that power.
7089 This function returns the first value raised to the second power,
7090 returning the same values as the libm ``pow`` functions would, and
7091 handles error conditions in the same way.
7093 '``llvm.exp.*``' Intrinsic
7094 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7099 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7100 floating point or vector of floating point type. Not all targets support
7105 declare float @llvm.exp.f32(float %Val)
7106 declare double @llvm.exp.f64(double %Val)
7107 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7108 declare fp128 @llvm.exp.f128(fp128 %Val)
7109 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7114 The '``llvm.exp.*``' intrinsics perform the exp function.
7119 The argument and return value are floating point numbers of the same
7125 This function returns the same values as the libm ``exp`` functions
7126 would, and handles error conditions in the same way.
7128 '``llvm.exp2.*``' Intrinsic
7129 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7134 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7135 floating point or vector of floating point type. Not all targets support
7140 declare float @llvm.exp2.f32(float %Val)
7141 declare double @llvm.exp2.f64(double %Val)
7142 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7143 declare fp128 @llvm.exp2.f128(fp128 %Val)
7144 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7149 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7154 The argument and return value are floating point numbers of the same
7160 This function returns the same values as the libm ``exp2`` functions
7161 would, and handles error conditions in the same way.
7163 '``llvm.log.*``' Intrinsic
7164 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7169 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7170 floating point or vector of floating point type. Not all targets support
7175 declare float @llvm.log.f32(float %Val)
7176 declare double @llvm.log.f64(double %Val)
7177 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7178 declare fp128 @llvm.log.f128(fp128 %Val)
7179 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7184 The '``llvm.log.*``' intrinsics perform the log function.
7189 The argument and return value are floating point numbers of the same
7195 This function returns the same values as the libm ``log`` functions
7196 would, and handles error conditions in the same way.
7198 '``llvm.log10.*``' Intrinsic
7199 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7204 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7205 floating point or vector of floating point type. Not all targets support
7210 declare float @llvm.log10.f32(float %Val)
7211 declare double @llvm.log10.f64(double %Val)
7212 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7213 declare fp128 @llvm.log10.f128(fp128 %Val)
7214 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7219 The '``llvm.log10.*``' intrinsics perform the log10 function.
7224 The argument and return value are floating point numbers of the same
7230 This function returns the same values as the libm ``log10`` functions
7231 would, and handles error conditions in the same way.
7233 '``llvm.log2.*``' Intrinsic
7234 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7239 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7240 floating point or vector of floating point type. Not all targets support
7245 declare float @llvm.log2.f32(float %Val)
7246 declare double @llvm.log2.f64(double %Val)
7247 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7248 declare fp128 @llvm.log2.f128(fp128 %Val)
7249 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7254 The '``llvm.log2.*``' intrinsics perform the log2 function.
7259 The argument and return value are floating point numbers of the same
7265 This function returns the same values as the libm ``log2`` functions
7266 would, and handles error conditions in the same way.
7268 '``llvm.fma.*``' Intrinsic
7269 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7274 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7275 floating point or vector of floating point type. Not all targets support
7280 declare float @llvm.fma.f32(float %a, float %b, float %c)
7281 declare double @llvm.fma.f64(double %a, double %b, double %c)
7282 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7283 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7284 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7289 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7295 The argument and return value are floating point numbers of the same
7301 This function returns the same values as the libm ``fma`` functions
7304 '``llvm.fabs.*``' Intrinsic
7305 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7310 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7311 floating point or vector of floating point type. Not all targets support
7316 declare float @llvm.fabs.f32(float %Val)
7317 declare double @llvm.fabs.f64(double %Val)
7318 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7319 declare fp128 @llvm.fabs.f128(fp128 %Val)
7320 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7325 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7331 The argument and return value are floating point numbers of the same
7337 This function returns the same values as the libm ``fabs`` functions
7338 would, and handles error conditions in the same way.
7340 '``llvm.floor.*``' Intrinsic
7341 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7346 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7347 floating point or vector of floating point type. Not all targets support
7352 declare float @llvm.floor.f32(float %Val)
7353 declare double @llvm.floor.f64(double %Val)
7354 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7355 declare fp128 @llvm.floor.f128(fp128 %Val)
7356 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7361 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7366 The argument and return value are floating point numbers of the same
7372 This function returns the same values as the libm ``floor`` functions
7373 would, and handles error conditions in the same way.
7375 '``llvm.ceil.*``' Intrinsic
7376 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7381 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7382 floating point or vector of floating point type. Not all targets support
7387 declare float @llvm.ceil.f32(float %Val)
7388 declare double @llvm.ceil.f64(double %Val)
7389 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7390 declare fp128 @llvm.ceil.f128(fp128 %Val)
7391 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7396 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7401 The argument and return value are floating point numbers of the same
7407 This function returns the same values as the libm ``ceil`` functions
7408 would, and handles error conditions in the same way.
7410 '``llvm.trunc.*``' Intrinsic
7411 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7416 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7417 floating point or vector of floating point type. Not all targets support
7422 declare float @llvm.trunc.f32(float %Val)
7423 declare double @llvm.trunc.f64(double %Val)
7424 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7425 declare fp128 @llvm.trunc.f128(fp128 %Val)
7426 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7431 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7432 nearest integer not larger in magnitude than the operand.
7437 The argument and return value are floating point numbers of the same
7443 This function returns the same values as the libm ``trunc`` functions
7444 would, and handles error conditions in the same way.
7446 '``llvm.rint.*``' Intrinsic
7447 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7452 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7453 floating point or vector of floating point type. Not all targets support
7458 declare float @llvm.rint.f32(float %Val)
7459 declare double @llvm.rint.f64(double %Val)
7460 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7461 declare fp128 @llvm.rint.f128(fp128 %Val)
7462 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7467 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7468 nearest integer. It may raise an inexact floating-point exception if the
7469 operand isn't an integer.
7474 The argument and return value are floating point numbers of the same
7480 This function returns the same values as the libm ``rint`` functions
7481 would, and handles error conditions in the same way.
7483 '``llvm.nearbyint.*``' Intrinsic
7484 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7489 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7490 floating point or vector of floating point type. Not all targets support
7495 declare float @llvm.nearbyint.f32(float %Val)
7496 declare double @llvm.nearbyint.f64(double %Val)
7497 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7498 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7499 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7504 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7510 The argument and return value are floating point numbers of the same
7516 This function returns the same values as the libm ``nearbyint``
7517 functions would, and handles error conditions in the same way.
7519 Bit Manipulation Intrinsics
7520 ---------------------------
7522 LLVM provides intrinsics for a few important bit manipulation
7523 operations. These allow efficient code generation for some algorithms.
7525 '``llvm.bswap.*``' Intrinsics
7526 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7531 This is an overloaded intrinsic function. You can use bswap on any
7532 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7536 declare i16 @llvm.bswap.i16(i16 <id>)
7537 declare i32 @llvm.bswap.i32(i32 <id>)
7538 declare i64 @llvm.bswap.i64(i64 <id>)
7543 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7544 values with an even number of bytes (positive multiple of 16 bits).
7545 These are useful for performing operations on data that is not in the
7546 target's native byte order.
7551 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7552 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7553 intrinsic returns an i32 value that has the four bytes of the input i32
7554 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7555 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7556 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7557 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7560 '``llvm.ctpop.*``' Intrinsic
7561 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7566 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7567 bit width, or on any vector with integer elements. Not all targets
7568 support all bit widths or vector types, however.
7572 declare i8 @llvm.ctpop.i8(i8 <src>)
7573 declare i16 @llvm.ctpop.i16(i16 <src>)
7574 declare i32 @llvm.ctpop.i32(i32 <src>)
7575 declare i64 @llvm.ctpop.i64(i64 <src>)
7576 declare i256 @llvm.ctpop.i256(i256 <src>)
7577 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7582 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7588 The only argument is the value to be counted. The argument may be of any
7589 integer type, or a vector with integer elements. The return type must
7590 match the argument type.
7595 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7596 each element of a vector.
7598 '``llvm.ctlz.*``' Intrinsic
7599 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7604 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7605 integer bit width, or any vector whose elements are integers. Not all
7606 targets support all bit widths or vector types, however.
7610 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7611 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7612 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7613 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7614 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7615 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7620 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7621 leading zeros in a variable.
7626 The first argument is the value to be counted. This argument may be of
7627 any integer type, or a vectory with integer element type. The return
7628 type must match the first argument type.
7630 The second argument must be a constant and is a flag to indicate whether
7631 the intrinsic should ensure that a zero as the first argument produces a
7632 defined result. Historically some architectures did not provide a
7633 defined result for zero values as efficiently, and many algorithms are
7634 now predicated on avoiding zero-value inputs.
7639 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7640 zeros in a variable, or within each element of the vector. If
7641 ``src == 0`` then the result is the size in bits of the type of ``src``
7642 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7643 ``llvm.ctlz(i32 2) = 30``.
7645 '``llvm.cttz.*``' Intrinsic
7646 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7651 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7652 integer bit width, or any vector of integer elements. Not all targets
7653 support all bit widths or vector types, however.
7657 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7658 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7659 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7660 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7661 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7662 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7667 The '``llvm.cttz``' family of intrinsic functions counts the number of
7673 The first argument is the value to be counted. This argument may be of
7674 any integer type, or a vectory with integer element type. The return
7675 type must match the first argument type.
7677 The second argument must be a constant and is a flag to indicate whether
7678 the intrinsic should ensure that a zero as the first argument produces a
7679 defined result. Historically some architectures did not provide a
7680 defined result for zero values as efficiently, and many algorithms are
7681 now predicated on avoiding zero-value inputs.
7686 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7687 zeros in a variable, or within each element of a vector. If ``src == 0``
7688 then the result is the size in bits of the type of ``src`` if
7689 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7690 ``llvm.cttz(2) = 1``.
7692 Arithmetic with Overflow Intrinsics
7693 -----------------------------------
7695 LLVM provides intrinsics for some arithmetic with overflow operations.
7697 '``llvm.sadd.with.overflow.*``' Intrinsics
7698 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7703 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7704 on any integer bit width.
7708 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7709 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7710 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7715 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7716 a signed addition of the two arguments, and indicate whether an overflow
7717 occurred during the signed summation.
7722 The arguments (%a and %b) and the first element of the result structure
7723 may be of integer types of any bit width, but they must have the same
7724 bit width. The second element of the result structure must be of type
7725 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7731 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7732 a signed addition of the two variables. They return a structure --- the
7733 first element of which is the signed summation, and the second element
7734 of which is a bit specifying if the signed summation resulted in an
7740 .. code-block:: llvm
7742 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7743 %sum = extractvalue {i32, i1} %res, 0
7744 %obit = extractvalue {i32, i1} %res, 1
7745 br i1 %obit, label %overflow, label %normal
7747 '``llvm.uadd.with.overflow.*``' Intrinsics
7748 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7753 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7754 on any integer bit width.
7758 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7759 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7760 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7765 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7766 an unsigned addition of the two arguments, and indicate whether a carry
7767 occurred during the unsigned summation.
7772 The arguments (%a and %b) and the first element of the result structure
7773 may be of integer types of any bit width, but they must have the same
7774 bit width. The second element of the result structure must be of type
7775 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7781 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7782 an unsigned addition of the two arguments. They return a structure --- the
7783 first element of which is the sum, and the second element of which is a
7784 bit specifying if the unsigned summation resulted in a carry.
7789 .. code-block:: llvm
7791 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7792 %sum = extractvalue {i32, i1} %res, 0
7793 %obit = extractvalue {i32, i1} %res, 1
7794 br i1 %obit, label %carry, label %normal
7796 '``llvm.ssub.with.overflow.*``' Intrinsics
7797 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7802 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7803 on any integer bit width.
7807 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7808 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7809 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7814 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7815 a signed subtraction of the two arguments, and indicate whether an
7816 overflow occurred during the signed subtraction.
7821 The arguments (%a and %b) and the first element of the result structure
7822 may be of integer types of any bit width, but they must have the same
7823 bit width. The second element of the result structure must be of type
7824 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7830 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7831 a signed subtraction of the two arguments. They return a structure --- the
7832 first element of which is the subtraction, and the second element of
7833 which is a bit specifying if the signed subtraction resulted in an
7839 .. code-block:: llvm
7841 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7842 %sum = extractvalue {i32, i1} %res, 0
7843 %obit = extractvalue {i32, i1} %res, 1
7844 br i1 %obit, label %overflow, label %normal
7846 '``llvm.usub.with.overflow.*``' Intrinsics
7847 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7852 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
7853 on any integer bit width.
7857 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7858 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7859 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7864 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7865 an unsigned subtraction of the two arguments, and indicate whether an
7866 overflow occurred during the unsigned subtraction.
7871 The arguments (%a and %b) and the first element of the result structure
7872 may be of integer types of any bit width, but they must have the same
7873 bit width. The second element of the result structure must be of type
7874 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7880 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7881 an unsigned subtraction of the two arguments. They return a structure ---
7882 the first element of which is the subtraction, and the second element of
7883 which is a bit specifying if the unsigned subtraction resulted in an
7889 .. code-block:: llvm
7891 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7892 %sum = extractvalue {i32, i1} %res, 0
7893 %obit = extractvalue {i32, i1} %res, 1
7894 br i1 %obit, label %overflow, label %normal
7896 '``llvm.smul.with.overflow.*``' Intrinsics
7897 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7902 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
7903 on any integer bit width.
7907 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7908 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7909 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7914 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7915 a signed multiplication of the two arguments, and indicate whether an
7916 overflow occurred during the signed multiplication.
7921 The arguments (%a and %b) and the first element of the result structure
7922 may be of integer types of any bit width, but they must have the same
7923 bit width. The second element of the result structure must be of type
7924 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7930 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7931 a signed multiplication of the two arguments. They return a structure ---
7932 the first element of which is the multiplication, and the second element
7933 of which is a bit specifying if the signed multiplication resulted in an
7939 .. code-block:: llvm
7941 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7942 %sum = extractvalue {i32, i1} %res, 0
7943 %obit = extractvalue {i32, i1} %res, 1
7944 br i1 %obit, label %overflow, label %normal
7946 '``llvm.umul.with.overflow.*``' Intrinsics
7947 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7952 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
7953 on any integer bit width.
7957 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7958 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7959 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7964 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7965 a unsigned multiplication of the two arguments, and indicate whether an
7966 overflow occurred during the unsigned multiplication.
7971 The arguments (%a and %b) and the first element of the result structure
7972 may be of integer types of any bit width, but they must have the same
7973 bit width. The second element of the result structure must be of type
7974 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7980 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7981 an unsigned multiplication of the two arguments. They return a structure ---
7982 the first element of which is the multiplication, and the second
7983 element of which is a bit specifying if the unsigned multiplication
7984 resulted in an overflow.
7989 .. code-block:: llvm
7991 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7992 %sum = extractvalue {i32, i1} %res, 0
7993 %obit = extractvalue {i32, i1} %res, 1
7994 br i1 %obit, label %overflow, label %normal
7996 Specialised Arithmetic Intrinsics
7997 ---------------------------------
7999 '``llvm.fmuladd.*``' Intrinsic
8000 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8007 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8008 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8013 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8014 expressions that can be fused if the code generator determines that (a) the
8015 target instruction set has support for a fused operation, and (b) that the
8016 fused operation is more efficient than the equivalent, separate pair of mul
8017 and add instructions.
8022 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8023 multiplicands, a and b, and an addend c.
8032 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8034 is equivalent to the expression a \* b + c, except that rounding will
8035 not be performed between the multiplication and addition steps if the
8036 code generator fuses the operations. Fusion is not guaranteed, even if
8037 the target platform supports it. If a fused multiply-add is required the
8038 corresponding llvm.fma.\* intrinsic function should be used instead.
8043 .. code-block:: llvm
8045 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8047 Half Precision Floating Point Intrinsics
8048 ----------------------------------------
8050 For most target platforms, half precision floating point is a
8051 storage-only format. This means that it is a dense encoding (in memory)
8052 but does not support computation in the format.
8054 This means that code must first load the half-precision floating point
8055 value as an i16, then convert it to float with
8056 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8057 then be performed on the float value (including extending to double
8058 etc). To store the value back to memory, it is first converted to float
8059 if needed, then converted to i16 with
8060 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8063 .. _int_convert_to_fp16:
8065 '``llvm.convert.to.fp16``' Intrinsic
8066 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8073 declare i16 @llvm.convert.to.fp16(f32 %a)
8078 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8079 from single precision floating point format to half precision floating
8085 The intrinsic function contains single argument - the value to be
8091 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8092 from single precision floating point format to half precision floating
8093 point format. The return value is an ``i16`` which contains the
8099 .. code-block:: llvm
8101 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8102 store i16 %res, i16* @x, align 2
8104 .. _int_convert_from_fp16:
8106 '``llvm.convert.from.fp16``' Intrinsic
8107 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8114 declare f32 @llvm.convert.from.fp16(i16 %a)
8119 The '``llvm.convert.from.fp16``' intrinsic function performs a
8120 conversion from half precision floating point format to single precision
8121 floating point format.
8126 The intrinsic function contains single argument - the value to be
8132 The '``llvm.convert.from.fp16``' intrinsic function performs a
8133 conversion from half single precision floating point format to single
8134 precision floating point format. The input half-float value is
8135 represented by an ``i16`` value.
8140 .. code-block:: llvm
8142 %a = load i16* @x, align 2
8143 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8148 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8149 prefix), are described in the `LLVM Source Level
8150 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8153 Exception Handling Intrinsics
8154 -----------------------------
8156 The LLVM exception handling intrinsics (which all start with
8157 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8158 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8162 Trampoline Intrinsics
8163 ---------------------
8165 These intrinsics make it possible to excise one parameter, marked with
8166 the :ref:`nest <nest>` attribute, from a function. The result is a
8167 callable function pointer lacking the nest parameter - the caller does
8168 not need to provide a value for it. Instead, the value to use is stored
8169 in advance in a "trampoline", a block of memory usually allocated on the
8170 stack, which also contains code to splice the nest value into the
8171 argument list. This is used to implement the GCC nested function address
8174 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8175 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8176 It can be created as follows:
8178 .. code-block:: llvm
8180 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8181 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8182 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8183 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8184 %fp = bitcast i8* %p to i32 (i32, i32)*
8186 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8187 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8191 '``llvm.init.trampoline``' Intrinsic
8192 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8199 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8204 This fills the memory pointed to by ``tramp`` with executable code,
8205 turning it into a trampoline.
8210 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8211 pointers. The ``tramp`` argument must point to a sufficiently large and
8212 sufficiently aligned block of memory; this memory is written to by the
8213 intrinsic. Note that the size and the alignment are target-specific -
8214 LLVM currently provides no portable way of determining them, so a
8215 front-end that generates this intrinsic needs to have some
8216 target-specific knowledge. The ``func`` argument must hold a function
8217 bitcast to an ``i8*``.
8222 The block of memory pointed to by ``tramp`` is filled with target
8223 dependent code, turning it into a function. Then ``tramp`` needs to be
8224 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8225 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8226 function's signature is the same as that of ``func`` with any arguments
8227 marked with the ``nest`` attribute removed. At most one such ``nest``
8228 argument is allowed, and it must be of pointer type. Calling the new
8229 function is equivalent to calling ``func`` with the same argument list,
8230 but with ``nval`` used for the missing ``nest`` argument. If, after
8231 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8232 modified, then the effect of any later call to the returned function
8233 pointer is undefined.
8237 '``llvm.adjust.trampoline``' Intrinsic
8238 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8245 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8250 This performs any required machine-specific adjustment to the address of
8251 a trampoline (passed as ``tramp``).
8256 ``tramp`` must point to a block of memory which already has trampoline
8257 code filled in by a previous call to
8258 :ref:`llvm.init.trampoline <int_it>`.
8263 On some architectures the address of the code to be executed needs to be
8264 different to the address where the trampoline is actually stored. This
8265 intrinsic returns the executable address corresponding to ``tramp``
8266 after performing the required machine specific adjustments. The pointer
8267 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8272 This class of intrinsics exists to information about the lifetime of
8273 memory objects and ranges where variables are immutable.
8275 '``llvm.lifetime.start``' Intrinsic
8276 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8283 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8288 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8294 The first argument is a constant integer representing the size of the
8295 object, or -1 if it is variable sized. The second argument is a pointer
8301 This intrinsic indicates that before this point in the code, the value
8302 of the memory pointed to by ``ptr`` is dead. This means that it is known
8303 to never be used and has an undefined value. A load from the pointer
8304 that precedes this intrinsic can be replaced with ``'undef'``.
8306 '``llvm.lifetime.end``' Intrinsic
8307 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8314 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8319 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8325 The first argument is a constant integer representing the size of the
8326 object, or -1 if it is variable sized. The second argument is a pointer
8332 This intrinsic indicates that after this point in the code, the value of
8333 the memory pointed to by ``ptr`` is dead. This means that it is known to
8334 never be used and has an undefined value. Any stores into the memory
8335 object following this intrinsic may be removed as dead.
8337 '``llvm.invariant.start``' Intrinsic
8338 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8345 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8350 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8351 a memory object will not change.
8356 The first argument is a constant integer representing the size of the
8357 object, or -1 if it is variable sized. The second argument is a pointer
8363 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8364 the return value, the referenced memory location is constant and
8367 '``llvm.invariant.end``' Intrinsic
8368 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8375 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8380 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8381 memory object are mutable.
8386 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8387 The second argument is a constant integer representing the size of the
8388 object, or -1 if it is variable sized and the third argument is a
8389 pointer to the object.
8394 This intrinsic indicates that the memory is mutable again.
8399 This class of intrinsics is designed to be generic and has no specific
8402 '``llvm.var.annotation``' Intrinsic
8403 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8410 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8415 The '``llvm.var.annotation``' intrinsic.
8420 The first argument is a pointer to a value, the second is a pointer to a
8421 global string, the third is a pointer to a global string which is the
8422 source file name, and the last argument is the line number.
8427 This intrinsic allows annotation of local variables with arbitrary
8428 strings. This can be useful for special purpose optimizations that want
8429 to look for these annotations. These have no other defined use; they are
8430 ignored by code generation and optimization.
8432 '``llvm.ptr.annotation.*``' Intrinsic
8433 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8438 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8439 pointer to an integer of any width. *NOTE* you must specify an address space for
8440 the pointer. The identifier for the default address space is the integer
8445 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8446 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8447 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8448 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8449 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8454 The '``llvm.ptr.annotation``' intrinsic.
8459 The first argument is a pointer to an integer value of arbitrary bitwidth
8460 (result of some expression), the second is a pointer to a global string, the
8461 third is a pointer to a global string which is the source file name, and the
8462 last argument is the line number. It returns the value of the first argument.
8467 This intrinsic allows annotation of a pointer to an integer with arbitrary
8468 strings. This can be useful for special purpose optimizations that want to look
8469 for these annotations. These have no other defined use; they are ignored by code
8470 generation and optimization.
8472 '``llvm.annotation.*``' Intrinsic
8473 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8478 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8479 any integer bit width.
8483 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8484 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8485 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8486 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8487 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8492 The '``llvm.annotation``' intrinsic.
8497 The first argument is an integer value (result of some expression), the
8498 second is a pointer to a global string, the third is a pointer to a
8499 global string which is the source file name, and the last argument is
8500 the line number. It returns the value of the first argument.
8505 This intrinsic allows annotations to be put on arbitrary expressions
8506 with arbitrary strings. This can be useful for special purpose
8507 optimizations that want to look for these annotations. These have no
8508 other defined use; they are ignored by code generation and optimization.
8510 '``llvm.trap``' Intrinsic
8511 ^^^^^^^^^^^^^^^^^^^^^^^^^
8518 declare void @llvm.trap() noreturn nounwind
8523 The '``llvm.trap``' intrinsic.
8533 This intrinsic is lowered to the target dependent trap instruction. If
8534 the target does not have a trap instruction, this intrinsic will be
8535 lowered to a call of the ``abort()`` function.
8537 '``llvm.debugtrap``' Intrinsic
8538 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8545 declare void @llvm.debugtrap() nounwind
8550 The '``llvm.debugtrap``' intrinsic.
8560 This intrinsic is lowered to code which is intended to cause an
8561 execution trap with the intention of requesting the attention of a
8564 '``llvm.stackprotector``' Intrinsic
8565 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8572 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8577 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8578 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8579 is placed on the stack before local variables.
8584 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8585 The first argument is the value loaded from the stack guard
8586 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8587 enough space to hold the value of the guard.
8592 This intrinsic causes the prologue/epilogue inserter to force the
8593 position of the ``AllocaInst`` stack slot to be before local variables
8594 on the stack. This is to ensure that if a local variable on the stack is
8595 overwritten, it will destroy the value of the guard. When the function
8596 exits, the guard on the stack is checked against the original guard. If
8597 they are different, then the program aborts by calling the
8598 ``__stack_chk_fail()`` function.
8600 '``llvm.objectsize``' Intrinsic
8601 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8608 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8609 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8614 The ``llvm.objectsize`` intrinsic is designed to provide information to
8615 the optimizers to determine at compile time whether a) an operation
8616 (like memcpy) will overflow a buffer that corresponds to an object, or
8617 b) that a runtime check for overflow isn't necessary. An object in this
8618 context means an allocation of a specific class, structure, array, or
8624 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8625 argument is a pointer to or into the ``object``. The second argument is
8626 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8627 or -1 (if false) when the object size is unknown. The second argument
8628 only accepts constants.
8633 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8634 the size of the object concerned. If the size cannot be determined at
8635 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8636 on the ``min`` argument).
8638 '``llvm.expect``' Intrinsic
8639 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8646 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8647 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8652 The ``llvm.expect`` intrinsic provides information about expected (the
8653 most probable) value of ``val``, which can be used by optimizers.
8658 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8659 a value. The second argument is an expected value, this needs to be a
8660 constant value, variables are not allowed.
8665 This intrinsic is lowered to the ``val``.
8667 '``llvm.donothing``' Intrinsic
8668 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8675 declare void @llvm.donothing() nounwind readnone
8680 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8681 only intrinsic that can be called with an invoke instruction.
8691 This intrinsic does nothing, and it's removed by optimizers and ignored