1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][a-zA-Z$._][a-zA-Z$._0-9]*``'. Identifiers which require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves.
83 #. Unnamed values are represented as an unsigned numeric value with
84 their prefix. For example, ``%12``, ``@2``, ``%44``.
85 #. Constants, which are described in the section Constants_ below.
87 LLVM requires that values start with a prefix for two reasons: Compilers
88 don't need to worry about name clashes with reserved words, and the set
89 of reserved words may be expanded in the future without penalty.
90 Additionally, unnamed identifiers allow a compiler to quickly come up
91 with a temporary variable without having to avoid symbol table
94 Reserved words in LLVM are very similar to reserved words in other
95 languages. There are keywords for different opcodes ('``add``',
96 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
97 '``i32``', etc...), and others. These reserved words cannot conflict
98 with variable names, because none of them start with a prefix character
101 Here is an example of LLVM code to multiply the integer variable
108 %result = mul i32 %X, 8
110 After strength reduction:
114 %result = shl i32 %X, 3
120 %0 = add i32 %X, %X ; yields {i32}:%0
121 %1 = add i32 %0, %0 ; yields {i32}:%1
122 %result = add i32 %1, %1
124 This last way of multiplying ``%X`` by 8 illustrates several important
125 lexical features of LLVM:
127 #. Comments are delimited with a '``;``' and go until the end of line.
128 #. Unnamed temporaries are created when the result of a computation is
129 not assigned to a named value.
130 #. Unnamed temporaries are numbered sequentially (using a per-function
131 incrementing counter, starting with 0).
133 It also shows a convention that we follow in this document. When
134 demonstrating instructions, we will follow an instruction with a comment
135 that defines the type and name of value produced.
143 LLVM programs are composed of ``Module``'s, each of which is a
144 translation unit of the input programs. Each module consists of
145 functions, global variables, and symbol table entries. Modules may be
146 combined together with the LLVM linker, which merges function (and
147 global variable) definitions, resolves forward declarations, and merges
148 symbol table entries. Here is an example of the "hello world" module:
152 ; Declare the string constant as a global constant.
153 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
155 ; External declaration of the puts function
156 declare i32 @puts(i8* nocapture) nounwind
158 ; Definition of main function
159 define i32 @main() { ; i32()*
160 ; Convert [13 x i8]* to i8 *...
161 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
163 ; Call puts function to write out the string to stdout.
164 call i32 @puts(i8* %cast210)
169 !1 = metadata !{i32 42}
172 This example is made up of a :ref:`global variable <globalvars>` named
173 "``.str``", an external declaration of the "``puts``" function, a
174 :ref:`function definition <functionstructure>` for "``main``" and
175 :ref:`named metadata <namedmetadatastructure>` "``foo``".
177 In general, a module is made up of a list of global values (where both
178 functions and global variables are global values). Global values are
179 represented by a pointer to a memory location (in this case, a pointer
180 to an array of char, and a pointer to a function), and have one of the
181 following :ref:`linkage types <linkage>`.
188 All Global Variables and Functions have one of the following types of
192 Global values with "``private``" linkage are only directly
193 accessible by objects in the current module. In particular, linking
194 code into a module with an private global value may cause the
195 private to be renamed as necessary to avoid collisions. Because the
196 symbol is private to the module, all references can be updated. This
197 doesn't show up in any symbol table in the object file.
199 Similar to ``private``, but the symbol is passed through the
200 assembler and evaluated by the linker. Unlike normal strong symbols,
201 they are removed by the linker from the final linked image
202 (executable or dynamic library).
203 ``linker_private_weak``
204 Similar to "``linker_private``", but the symbol is weak. Note that
205 ``linker_private_weak`` symbols are subject to coalescing by the
206 linker. The symbols are removed by the linker from the final linked
207 image (executable or dynamic library).
209 Similar to private, but the value shows as a local symbol
210 (``STB_LOCAL`` in the case of ELF) in the object file. This
211 corresponds to the notion of the '``static``' keyword in C.
212 ``available_externally``
213 Globals with "``available_externally``" linkage are never emitted
214 into the object file corresponding to the LLVM module. They exist to
215 allow inlining and other optimizations to take place given knowledge
216 of the definition of the global, which is known to be somewhere
217 outside the module. Globals with ``available_externally`` linkage
218 are allowed to be discarded at will, and are otherwise the same as
219 ``linkonce_odr``. This linkage type is only allowed on definitions,
222 Globals with "``linkonce``" linkage are merged with other globals of
223 the same name when linkage occurs. This can be used to implement
224 some forms of inline functions, templates, or other code which must
225 be generated in each translation unit that uses it, but where the
226 body may be overridden with a more definitive definition later.
227 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
228 that ``linkonce`` linkage does not actually allow the optimizer to
229 inline the body of this function into callers because it doesn't
230 know if this definition of the function is the definitive definition
231 within the program or whether it will be overridden by a stronger
232 definition. To enable inlining and other optimizations, use
233 "``linkonce_odr``" linkage.
235 "``weak``" linkage has the same merging semantics as ``linkonce``
236 linkage, except that unreferenced globals with ``weak`` linkage may
237 not be discarded. This is used for globals that are declared "weak"
240 "``common``" linkage is most similar to "``weak``" linkage, but they
241 are used for tentative definitions in C, such as "``int X;``" at
242 global scope. Symbols with "``common``" linkage are merged in the
243 same way as ``weak symbols``, and they may not be deleted if
244 unreferenced. ``common`` symbols may not have an explicit section,
245 must have a zero initializer, and may not be marked
246 ':ref:`constant <globalvars>`'. Functions and aliases may not have
249 .. _linkage_appending:
252 "``appending``" linkage may only be applied to global variables of
253 pointer to array type. When two global variables with appending
254 linkage are linked together, the two global arrays are appended
255 together. This is the LLVM, typesafe, equivalent of having the
256 system linker append together "sections" with identical names when
259 The semantics of this linkage follow the ELF object file model: the
260 symbol is weak until linked, if not linked, the symbol becomes null
261 instead of being an undefined reference.
262 ``linkonce_odr``, ``weak_odr``
263 Some languages allow differing globals to be merged, such as two
264 functions with different semantics. Other languages, such as
265 ``C++``, ensure that only equivalent globals are ever merged (the
266 "one definition rule" --- "ODR"). Such languages can use the
267 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
268 global will only be merged with equivalent globals. These linkage
269 types are otherwise the same as their non-``odr`` versions.
270 ``linkonce_odr_auto_hide``
271 Similar to "``linkonce_odr``", but nothing in the translation unit
272 takes the address of this definition. For instance, functions that
273 had an inline definition, but the compiler decided not to inline it.
274 ``linkonce_odr_auto_hide`` may have only ``default`` visibility. The
275 symbols are removed by the linker from the final linked image
276 (executable or dynamic library).
278 If none of the above identifiers are used, the global is externally
279 visible, meaning that it participates in linkage and can be used to
280 resolve external symbol references.
282 The next two types of linkage are targeted for Microsoft Windows
283 platform only. They are designed to support importing (exporting)
284 symbols from (to) DLLs (Dynamic Link Libraries).
287 "``dllimport``" linkage causes the compiler to reference a function
288 or variable via a global pointer to a pointer that is set up by the
289 DLL exporting the symbol. On Microsoft Windows targets, the pointer
290 name is formed by combining ``__imp_`` and the function or variable
293 "``dllexport``" linkage causes the compiler to provide a global
294 pointer to a pointer in a DLL, so that it can be referenced with the
295 ``dllimport`` attribute. On Microsoft Windows targets, the pointer
296 name is formed by combining ``__imp_`` and the function or variable
299 For example, since the "``.LC0``" variable is defined to be internal, if
300 another module defined a "``.LC0``" variable and was linked with this
301 one, one of the two would be renamed, preventing a collision. Since
302 "``main``" and "``puts``" are external (i.e., lacking any linkage
303 declarations), they are accessible outside of the current module.
305 It is illegal for a function *declaration* to have any linkage type
306 other than ``external``, ``dllimport`` or ``extern_weak``.
308 Aliases can have only ``external``, ``internal``, ``weak`` or
309 ``weak_odr`` linkages.
316 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
317 :ref:`invokes <i_invoke>` can all have an optional calling convention
318 specified for the call. The calling convention of any pair of dynamic
319 caller/callee must match, or the behavior of the program is undefined.
320 The following calling conventions are supported by LLVM, and more may be
323 "``ccc``" - The C calling convention
324 This calling convention (the default if no other calling convention
325 is specified) matches the target C calling conventions. This calling
326 convention supports varargs function calls and tolerates some
327 mismatch in the declared prototype and implemented declaration of
328 the function (as does normal C).
329 "``fastcc``" - The fast calling convention
330 This calling convention attempts to make calls as fast as possible
331 (e.g. by passing things in registers). This calling convention
332 allows the target to use whatever tricks it wants to produce fast
333 code for the target, without having to conform to an externally
334 specified ABI (Application Binary Interface). `Tail calls can only
335 be optimized when this, the GHC or the HiPE convention is
336 used. <CodeGenerator.html#id80>`_ This calling convention does not
337 support varargs and requires the prototype of all callees to exactly
338 match the prototype of the function definition.
339 "``coldcc``" - The cold calling convention
340 This calling convention attempts to make code in the caller as
341 efficient as possible under the assumption that the call is not
342 commonly executed. As such, these calls often preserve all registers
343 so that the call does not break any live ranges in the caller side.
344 This calling convention does not support varargs and requires the
345 prototype of all callees to exactly match the prototype of the
347 "``cc 10``" - GHC convention
348 This calling convention has been implemented specifically for use by
349 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
350 It passes everything in registers, going to extremes to achieve this
351 by disabling callee save registers. This calling convention should
352 not be used lightly but only for specific situations such as an
353 alternative to the *register pinning* performance technique often
354 used when implementing functional programming languages. At the
355 moment only X86 supports this convention and it has the following
358 - On *X86-32* only supports up to 4 bit type parameters. No
359 floating point types are supported.
360 - On *X86-64* only supports up to 10 bit type parameters and 6
361 floating point parameters.
363 This calling convention supports `tail call
364 optimization <CodeGenerator.html#id80>`_ but requires both the
365 caller and callee are using it.
366 "``cc 11``" - The HiPE calling convention
367 This calling convention has been implemented specifically for use by
368 the `High-Performance Erlang
369 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
370 native code compiler of the `Ericsson's Open Source Erlang/OTP
371 system <http://www.erlang.org/download.shtml>`_. It uses more
372 registers for argument passing than the ordinary C calling
373 convention and defines no callee-saved registers. The calling
374 convention properly supports `tail call
375 optimization <CodeGenerator.html#id80>`_ but requires that both the
376 caller and the callee use it. It uses a *register pinning*
377 mechanism, similar to GHC's convention, for keeping frequently
378 accessed runtime components pinned to specific hardware registers.
379 At the moment only X86 supports this convention (both 32 and 64
381 "``cc <n>``" - Numbered convention
382 Any calling convention may be specified by number, allowing
383 target-specific calling conventions to be used. Target specific
384 calling conventions start at 64.
386 More calling conventions can be added/defined on an as-needed basis, to
387 support Pascal conventions or any other well-known target-independent
390 .. _visibilitystyles:
395 All Global Variables and Functions have one of the following visibility
398 "``default``" - Default style
399 On targets that use the ELF object file format, default visibility
400 means that the declaration is visible to other modules and, in
401 shared libraries, means that the declared entity may be overridden.
402 On Darwin, default visibility means that the declaration is visible
403 to other modules. Default visibility corresponds to "external
404 linkage" in the language.
405 "``hidden``" - Hidden style
406 Two declarations of an object with hidden visibility refer to the
407 same object if they are in the same shared object. Usually, hidden
408 visibility indicates that the symbol will not be placed into the
409 dynamic symbol table, so no other module (executable or shared
410 library) can reference it directly.
411 "``protected``" - Protected style
412 On ELF, protected visibility indicates that the symbol will be
413 placed in the dynamic symbol table, but that references within the
414 defining module will bind to the local symbol. That is, the symbol
415 cannot be overridden by another module.
422 LLVM IR allows you to specify name aliases for certain types. This can
423 make it easier to read the IR and make the IR more condensed
424 (particularly when recursive types are involved). An example of a name
429 %mytype = type { %mytype*, i32 }
431 You may give a name to any :ref:`type <typesystem>` except
432 ":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
433 expected with the syntax "%mytype".
435 Note that type names are aliases for the structural type that they
436 indicate, and that you can therefore specify multiple names for the same
437 type. This often leads to confusing behavior when dumping out a .ll
438 file. Since LLVM IR uses structural typing, the name is not part of the
439 type. When printing out LLVM IR, the printer will pick *one name* to
440 render all types of a particular shape. This means that if you have code
441 where two different source types end up having the same LLVM type, that
442 the dumper will sometimes print the "wrong" or unexpected type. This is
443 an important design point and isn't going to change.
450 Global variables define regions of memory allocated at compilation time
451 instead of run-time. Global variables may optionally be initialized, may
452 have an explicit section to be placed in, and may have an optional
453 explicit alignment specified.
455 A variable may be defined as ``thread_local``, which means that it will
456 not be shared by threads (each thread will have a separated copy of the
457 variable). Not all targets support thread-local variables. Optionally, a
458 TLS model may be specified:
461 For variables that are only used within the current shared library.
463 For variables in modules that will not be loaded dynamically.
465 For variables defined in the executable and only used within it.
467 The models correspond to the ELF TLS models; see `ELF Handling For
468 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
469 more information on under which circumstances the different models may
470 be used. The target may choose a different TLS model if the specified
471 model is not supported, or if a better choice of model can be made.
473 A variable may be defined as a global ``constant``, which indicates that
474 the contents of the variable will **never** be modified (enabling better
475 optimization, allowing the global data to be placed in the read-only
476 section of an executable, etc). Note that variables that need runtime
477 initialization cannot be marked ``constant`` as there is a store to the
480 LLVM explicitly allows *declarations* of global variables to be marked
481 constant, even if the final definition of the global is not. This
482 capability can be used to enable slightly better optimization of the
483 program, but requires the language definition to guarantee that
484 optimizations based on the 'constantness' are valid for the translation
485 units that do not include the definition.
487 As SSA values, global variables define pointer values that are in scope
488 (i.e. they dominate) all basic blocks in the program. Global variables
489 always define a pointer to their "content" type because they describe a
490 region of memory, and all memory objects in LLVM are accessed through
493 Global variables can be marked with ``unnamed_addr`` which indicates
494 that the address is not significant, only the content. Constants marked
495 like this can be merged with other constants if they have the same
496 initializer. Note that a constant with significant address *can* be
497 merged with a ``unnamed_addr`` constant, the result being a constant
498 whose address is significant.
500 A global variable may be declared to reside in a target-specific
501 numbered address space. For targets that support them, address spaces
502 may affect how optimizations are performed and/or what target
503 instructions are used to access the variable. The default address space
504 is zero. The address space qualifier must precede any other attributes.
506 LLVM allows an explicit section to be specified for globals. If the
507 target supports it, it will emit globals to the section specified.
509 By default, global initializers are optimized by assuming that global
510 variables defined within the module are not modified from their
511 initial values before the start of the global initializer. This is
512 true even for variables potentially accessible from outside the
513 module, including those with external linkage or appearing in
514 ``@llvm.used``. This assumption may be suppressed by marking the
515 variable with ``externally_initialized``.
517 An explicit alignment may be specified for a global, which must be a
518 power of 2. If not present, or if the alignment is set to zero, the
519 alignment of the global is set by the target to whatever it feels
520 convenient. If an explicit alignment is specified, the global is forced
521 to have exactly that alignment. Targets and optimizers are not allowed
522 to over-align the global if the global has an assigned section. In this
523 case, the extra alignment could be observable: for example, code could
524 assume that the globals are densely packed in their section and try to
525 iterate over them as an array, alignment padding would break this
528 For example, the following defines a global in a numbered address space
529 with an initializer, section, and alignment:
533 @G = addrspace(5) constant float 1.0, section "foo", align 4
535 The following example defines a thread-local global with the
536 ``initialexec`` TLS model:
540 @G = thread_local(initialexec) global i32 0, align 4
542 .. _functionstructure:
547 LLVM function definitions consist of the "``define``" keyword, an
548 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
549 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
550 an optional ``unnamed_addr`` attribute, a return type, an optional
551 :ref:`parameter attribute <paramattrs>` for the return type, a function
552 name, a (possibly empty) argument list (each with optional :ref:`parameter
553 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
554 an optional section, an optional alignment, an optional :ref:`garbage
555 collector name <gc>`, an opening curly brace, a list of basic blocks,
556 and a closing curly brace.
558 LLVM function declarations consist of the "``declare``" keyword, an
559 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
560 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
561 an optional ``unnamed_addr`` attribute, a return type, an optional
562 :ref:`parameter attribute <paramattrs>` for the return type, a function
563 name, a possibly empty list of arguments, an optional alignment, and an
564 optional :ref:`garbage collector name <gc>`.
566 A function definition contains a list of basic blocks, forming the CFG
567 (Control Flow Graph) for the function. Each basic block may optionally
568 start with a label (giving the basic block a symbol table entry),
569 contains a list of instructions, and ends with a
570 :ref:`terminator <terminators>` instruction (such as a branch or function
571 return). If explicit label is not provided, a block is assigned an
572 implicit numbered label, using a next value from the same counter as used
573 for unnamed temporaries (:ref:`see above<identifiers>`). For example, if a
574 function entry block does not have explicit label, it will be assigned
575 label "%0", then first unnamed temporary in that block will be "%1", etc.
577 The first basic block in a function is special in two ways: it is
578 immediately executed on entrance to the function, and it is not allowed
579 to have predecessor basic blocks (i.e. there can not be any branches to
580 the entry block of a function). Because the block can have no
581 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
583 LLVM allows an explicit section to be specified for functions. If the
584 target supports it, it will emit functions to the section specified.
586 An explicit alignment may be specified for a function. If not present,
587 or if the alignment is set to zero, the alignment of the function is set
588 by the target to whatever it feels convenient. If an explicit alignment
589 is specified, the function is forced to have at least that much
590 alignment. All alignments must be a power of 2.
592 If the ``unnamed_addr`` attribute is given, the address is know to not
593 be significant and two identical functions can be merged.
597 define [linkage] [visibility]
599 <ResultType> @<FunctionName> ([argument list])
600 [fn Attrs] [section "name"] [align N]
608 Aliases act as "second name" for the aliasee value (which can be either
609 function, global variable, another alias or bitcast of global value).
610 Aliases may have an optional :ref:`linkage type <linkage>`, and an optional
611 :ref:`visibility style <visibility>`.
615 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
617 .. _namedmetadatastructure:
622 Named metadata is a collection of metadata. :ref:`Metadata
623 nodes <metadata>` (but not metadata strings) are the only valid
624 operands for a named metadata.
628 ; Some unnamed metadata nodes, which are referenced by the named metadata.
629 !0 = metadata !{metadata !"zero"}
630 !1 = metadata !{metadata !"one"}
631 !2 = metadata !{metadata !"two"}
633 !name = !{!0, !1, !2}
640 The return type and each parameter of a function type may have a set of
641 *parameter attributes* associated with them. Parameter attributes are
642 used to communicate additional information about the result or
643 parameters of a function. Parameter attributes are considered to be part
644 of the function, not of the function type, so functions with different
645 parameter attributes can have the same function type.
647 Parameter attributes are simple keywords that follow the type specified.
648 If multiple parameter attributes are needed, they are space separated.
653 declare i32 @printf(i8* noalias nocapture, ...)
654 declare i32 @atoi(i8 zeroext)
655 declare signext i8 @returns_signed_char()
657 Note that any attributes for the function result (``nounwind``,
658 ``readonly``) come immediately after the argument list.
660 Currently, only the following parameter attributes are defined:
663 This indicates to the code generator that the parameter or return
664 value should be zero-extended to the extent required by the target's
665 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
666 the caller (for a parameter) or the callee (for a return value).
668 This indicates to the code generator that the parameter or return
669 value should be sign-extended to the extent required by the target's
670 ABI (which is usually 32-bits) by the caller (for a parameter) or
671 the callee (for a return value).
673 This indicates that this parameter or return value should be treated
674 in a special target-dependent fashion during while emitting code for
675 a function call or return (usually, by putting it in a register as
676 opposed to memory, though some targets use it to distinguish between
677 two different kinds of registers). Use of this attribute is
680 This indicates that the pointer parameter should really be passed by
681 value to the function. The attribute implies that a hidden copy of
682 the pointee is made between the caller and the callee, so the callee
683 is unable to modify the value in the caller. This attribute is only
684 valid on LLVM pointer arguments. It is generally used to pass
685 structs and arrays by value, but is also valid on pointers to
686 scalars. The copy is considered to belong to the caller not the
687 callee (for example, ``readonly`` functions should not write to
688 ``byval`` parameters). This is not a valid attribute for return
691 The byval attribute also supports specifying an alignment with the
692 align attribute. It indicates the alignment of the stack slot to
693 form and the known alignment of the pointer specified to the call
694 site. If the alignment is not specified, then the code generator
695 makes a target-specific assumption.
698 This indicates that the pointer parameter specifies the address of a
699 structure that is the return value of the function in the source
700 program. This pointer must be guaranteed by the caller to be valid:
701 loads and stores to the structure may be assumed by the callee
702 not to trap and to be properly aligned. This may only be applied to
703 the first parameter. This is not a valid attribute for return
706 This indicates that pointer values :ref:`based <pointeraliasing>` on
707 the argument or return value do not alias pointer values which are
708 not *based* on it, ignoring certain "irrelevant" dependencies. For a
709 call to the parent function, dependencies between memory references
710 from before or after the call and from those during the call are
711 "irrelevant" to the ``noalias`` keyword for the arguments and return
712 value used in that call. The caller shares the responsibility with
713 the callee for ensuring that these requirements are met. For further
714 details, please see the discussion of the NoAlias response in `alias
715 analysis <AliasAnalysis.html#MustMayNo>`_.
717 Note that this definition of ``noalias`` is intentionally similar
718 to the definition of ``restrict`` in C99 for function arguments,
719 though it is slightly weaker.
721 For function return values, C99's ``restrict`` is not meaningful,
722 while LLVM's ``noalias`` is.
724 This indicates that the callee does not make any copies of the
725 pointer that outlive the callee itself. This is not a valid
726 attribute for return values.
731 This indicates that the pointer parameter can be excised using the
732 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
733 attribute for return values and can only be applied to one parameter.
736 This indicates that the function always returns the argument as its return
737 value. This is an optimization hint to the code generator when generating
738 the caller, allowing tail call optimization and omission of register saves
739 and restores in some cases; it is not checked or enforced when generating
740 the callee. The parameter and the function return type must be valid
741 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
742 valid attribute for return values and can only be applied to one parameter.
746 Garbage Collector Names
747 -----------------------
749 Each function may specify a garbage collector name, which is simply a
754 define void @f() gc "name" { ... }
756 The compiler declares the supported values of *name*. Specifying a
757 collector which will cause the compiler to alter its output in order to
758 support the named garbage collection algorithm.
765 Attribute groups are groups of attributes that are referenced by objects within
766 the IR. They are important for keeping ``.ll`` files readable, because a lot of
767 functions will use the same set of attributes. In the degenerative case of a
768 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
769 group will capture the important command line flags used to build that file.
771 An attribute group is a module-level object. To use an attribute group, an
772 object references the attribute group's ID (e.g. ``#37``). An object may refer
773 to more than one attribute group. In that situation, the attributes from the
774 different groups are merged.
776 Here is an example of attribute groups for a function that should always be
777 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
781 ; Target-independent attributes:
782 attributes #0 = { alwaysinline alignstack=4 }
784 ; Target-dependent attributes:
785 attributes #1 = { "no-sse" }
787 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
788 define void @f() #0 #1 { ... }
795 Function attributes are set to communicate additional information about
796 a function. Function attributes are considered to be part of the
797 function, not of the function type, so functions with different function
798 attributes can have the same function type.
800 Function attributes are simple keywords that follow the type specified.
801 If multiple attributes are needed, they are space separated. For
806 define void @f() noinline { ... }
807 define void @f() alwaysinline { ... }
808 define void @f() alwaysinline optsize { ... }
809 define void @f() optsize { ... }
812 This attribute indicates that, when emitting the prologue and
813 epilogue, the backend should forcibly align the stack pointer.
814 Specify the desired alignment, which must be a power of two, in
817 This attribute indicates that the inliner should attempt to inline
818 this function into callers whenever possible, ignoring any active
819 inlining size threshold for this caller.
821 This indicates that the callee function at a call site should be
822 recognized as a built-in function, even though the function's declaration
823 uses the ``nobuiltin`` attribute. This is only valid at call sites for
824 direct calls to functions which are declared with the ``nobuiltin``
827 This attribute indicates that this function is rarely called. When
828 computing edge weights, basic blocks post-dominated by a cold
829 function call are also considered to be cold; and, thus, given low
832 This attribute indicates that the source code contained a hint that
833 inlining this function is desirable (such as the "inline" keyword in
834 C/C++). It is just a hint; it imposes no requirements on the
837 This attribute disables prologue / epilogue emission for the
838 function. This can have very system-specific consequences.
840 This indicates that the callee function at a call site is not recognized as
841 a built-in function. LLVM will retain the original call and not replace it
842 with equivalent code based on the semantics of the built-in function, unless
843 the call site uses the ``builtin`` attribute. This is valid at call sites
844 and on function declarations and definitions.
846 This attribute indicates that calls to the function cannot be
847 duplicated. A call to a ``noduplicate`` function may be moved
848 within its parent function, but may not be duplicated within
851 A function containing a ``noduplicate`` call may still
852 be an inlining candidate, provided that the call is not
853 duplicated by inlining. That implies that the function has
854 internal linkage and only has one call site, so the original
855 call is dead after inlining.
857 This attributes disables implicit floating point instructions.
859 This attribute indicates that the inliner should never inline this
860 function in any situation. This attribute may not be used together
861 with the ``alwaysinline`` attribute.
863 This attribute suppresses lazy symbol binding for the function. This
864 may make calls to the function faster, at the cost of extra program
865 startup time if the function is not called during program startup.
867 This attribute indicates that the code generator should not use a
868 red zone, even if the target-specific ABI normally permits it.
870 This function attribute indicates that the function never returns
871 normally. This produces undefined behavior at runtime if the
872 function ever does dynamically return.
874 This function attribute indicates that the function never returns
875 with an unwind or exceptional control flow. If the function does
876 unwind, its runtime behavior is undefined.
878 This attribute suggests that optimization passes and code generator
879 passes make choices that keep the code size of this function low,
880 and otherwise do optimizations specifically to reduce code size.
882 On a function, this attribute indicates that the function computes its
883 result (or decides to unwind an exception) based strictly on its arguments,
884 without dereferencing any pointer arguments or otherwise accessing
885 any mutable state (e.g. memory, control registers, etc) visible to
886 caller functions. It does not write through any pointer arguments
887 (including ``byval`` arguments) and never changes any state visible
888 to callers. This means that it cannot unwind exceptions by calling
889 the ``C++`` exception throwing methods.
891 On an argument, this attribute indicates that the function does not
892 dereference that pointer argument, even though it may read or write the
893 memory that the pointer points to if accessed through other pointers.
895 On a function, this attribute indicates that the function does not write
896 through any pointer arguments (including ``byval`` arguments) or otherwise
897 modify any state (e.g. memory, control registers, etc) visible to
898 caller functions. It may dereference pointer arguments and read
899 state that may be set in the caller. A readonly function always
900 returns the same value (or unwinds an exception identically) when
901 called with the same set of arguments and global state. It cannot
902 unwind an exception by calling the ``C++`` exception throwing
905 On an argument, this attribute indicates that the function does not write
906 through this pointer argument, even though it may write to the memory that
907 the pointer points to.
909 This attribute indicates that this function can return twice. The C
910 ``setjmp`` is an example of such a function. The compiler disables
911 some optimizations (like tail calls) in the caller of these
914 This attribute indicates that AddressSanitizer checks
915 (dynamic address safety analysis) are enabled for this function.
917 This attribute indicates that MemorySanitizer checks (dynamic detection
918 of accesses to uninitialized memory) are enabled for this function.
920 This attribute indicates that ThreadSanitizer checks
921 (dynamic thread safety analysis) are enabled for this function.
923 This attribute indicates that the function should emit a stack
924 smashing protector. It is in the form of a "canary" --- a random value
925 placed on the stack before the local variables that's checked upon
926 return from the function to see if it has been overwritten. A
927 heuristic is used to determine if a function needs stack protectors
928 or not. The heuristic used will enable protectors for functions with:
930 - Character arrays larger than ``ssp-buffer-size`` (default 8).
931 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
932 - Calls to alloca() with variable sizes or constant sizes greater than
935 If a function that has an ``ssp`` attribute is inlined into a
936 function that doesn't have an ``ssp`` attribute, then the resulting
937 function will have an ``ssp`` attribute.
939 This attribute indicates that the function should *always* emit a
940 stack smashing protector. This overrides the ``ssp`` function
943 If a function that has an ``sspreq`` attribute is inlined into a
944 function that doesn't have an ``sspreq`` attribute or which has an
945 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
946 an ``sspreq`` attribute.
948 This attribute indicates that the function should emit a stack smashing
949 protector. This attribute causes a strong heuristic to be used when
950 determining if a function needs stack protectors. The strong heuristic
951 will enable protectors for functions with:
953 - Arrays of any size and type
954 - Aggregates containing an array of any size and type.
956 - Local variables that have had their address taken.
958 This overrides the ``ssp`` function attribute.
960 If a function that has an ``sspstrong`` attribute is inlined into a
961 function that doesn't have an ``sspstrong`` attribute, then the
962 resulting function will have an ``sspstrong`` attribute.
964 This attribute indicates that the ABI being targeted requires that
965 an unwind table entry be produce for this function even if we can
966 show that no exceptions passes by it. This is normally the case for
967 the ELF x86-64 abi, but it can be disabled for some compilation
972 Module-Level Inline Assembly
973 ----------------------------
975 Modules may contain "module-level inline asm" blocks, which corresponds
976 to the GCC "file scope inline asm" blocks. These blocks are internally
977 concatenated by LLVM and treated as a single unit, but may be separated
978 in the ``.ll`` file if desired. The syntax is very simple:
982 module asm "inline asm code goes here"
983 module asm "more can go here"
985 The strings can contain any character by escaping non-printable
986 characters. The escape sequence used is simply "\\xx" where "xx" is the
987 two digit hex code for the number.
989 The inline asm code is simply printed to the machine code .s file when
990 assembly code is generated.
992 .. _langref_datalayout:
997 A module may specify a target specific data layout string that specifies
998 how data is to be laid out in memory. The syntax for the data layout is
1001 .. code-block:: llvm
1003 target datalayout = "layout specification"
1005 The *layout specification* consists of a list of specifications
1006 separated by the minus sign character ('-'). Each specification starts
1007 with a letter and may include other information after the letter to
1008 define some aspect of the data layout. The specifications accepted are
1012 Specifies that the target lays out data in big-endian form. That is,
1013 the bits with the most significance have the lowest address
1016 Specifies that the target lays out data in little-endian form. That
1017 is, the bits with the least significance have the lowest address
1020 Specifies the natural alignment of the stack in bits. Alignment
1021 promotion of stack variables is limited to the natural stack
1022 alignment to avoid dynamic stack realignment. The stack alignment
1023 must be a multiple of 8-bits. If omitted, the natural stack
1024 alignment defaults to "unspecified", which does not prevent any
1025 alignment promotions.
1026 ``p[n]:<size>:<abi>:<pref>``
1027 This specifies the *size* of a pointer and its ``<abi>`` and
1028 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1029 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
1030 preceding ``:`` should be omitted too. The address space, ``n`` is
1031 optional, and if not specified, denotes the default address space 0.
1032 The value of ``n`` must be in the range [1,2^23).
1033 ``i<size>:<abi>:<pref>``
1034 This specifies the alignment for an integer type of a given bit
1035 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1036 ``v<size>:<abi>:<pref>``
1037 This specifies the alignment for a vector type of a given bit
1039 ``f<size>:<abi>:<pref>``
1040 This specifies the alignment for a floating point type of a given bit
1041 ``<size>``. Only values of ``<size>`` that are supported by the target
1042 will work. 32 (float) and 64 (double) are supported on all targets; 80
1043 or 128 (different flavors of long double) are also supported on some
1045 ``a<size>:<abi>:<pref>``
1046 This specifies the alignment for an aggregate type of a given bit
1048 ``s<size>:<abi>:<pref>``
1049 This specifies the alignment for a stack object of a given bit
1051 ``n<size1>:<size2>:<size3>...``
1052 This specifies a set of native integer widths for the target CPU in
1053 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1054 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1055 this set are considered to support most general arithmetic operations
1058 When constructing the data layout for a given target, LLVM starts with a
1059 default set of specifications which are then (possibly) overridden by
1060 the specifications in the ``datalayout`` keyword. The default
1061 specifications are given in this list:
1063 - ``E`` - big endian
1064 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1065 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1066 same as the default address space.
1067 - ``S0`` - natural stack alignment is unspecified
1068 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1069 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1070 - ``i16:16:16`` - i16 is 16-bit aligned
1071 - ``i32:32:32`` - i32 is 32-bit aligned
1072 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1073 alignment of 64-bits
1074 - ``f16:16:16`` - half is 16-bit aligned
1075 - ``f32:32:32`` - float is 32-bit aligned
1076 - ``f64:64:64`` - double is 64-bit aligned
1077 - ``f128:128:128`` - quad is 128-bit aligned
1078 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1079 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1080 - ``a0:0:64`` - aggregates are 64-bit aligned
1082 When LLVM is determining the alignment for a given type, it uses the
1085 #. If the type sought is an exact match for one of the specifications,
1086 that specification is used.
1087 #. If no match is found, and the type sought is an integer type, then
1088 the smallest integer type that is larger than the bitwidth of the
1089 sought type is used. If none of the specifications are larger than
1090 the bitwidth then the largest integer type is used. For example,
1091 given the default specifications above, the i7 type will use the
1092 alignment of i8 (next largest) while both i65 and i256 will use the
1093 alignment of i64 (largest specified).
1094 #. If no match is found, and the type sought is a vector type, then the
1095 largest vector type that is smaller than the sought vector type will
1096 be used as a fall back. This happens because <128 x double> can be
1097 implemented in terms of 64 <2 x double>, for example.
1099 The function of the data layout string may not be what you expect.
1100 Notably, this is not a specification from the frontend of what alignment
1101 the code generator should use.
1103 Instead, if specified, the target data layout is required to match what
1104 the ultimate *code generator* expects. This string is used by the
1105 mid-level optimizers to improve code, and this only works if it matches
1106 what the ultimate code generator uses. If you would like to generate IR
1107 that does not embed this target-specific detail into the IR, then you
1108 don't have to specify the string. This will disable some optimizations
1109 that require precise layout information, but this also prevents those
1110 optimizations from introducing target specificity into the IR.
1112 .. _pointeraliasing:
1114 Pointer Aliasing Rules
1115 ----------------------
1117 Any memory access must be done through a pointer value associated with
1118 an address range of the memory access, otherwise the behavior is
1119 undefined. Pointer values are associated with address ranges according
1120 to the following rules:
1122 - A pointer value is associated with the addresses associated with any
1123 value it is *based* on.
1124 - An address of a global variable is associated with the address range
1125 of the variable's storage.
1126 - The result value of an allocation instruction is associated with the
1127 address range of the allocated storage.
1128 - A null pointer in the default address-space is associated with no
1130 - An integer constant other than zero or a pointer value returned from
1131 a function not defined within LLVM may be associated with address
1132 ranges allocated through mechanisms other than those provided by
1133 LLVM. Such ranges shall not overlap with any ranges of addresses
1134 allocated by mechanisms provided by LLVM.
1136 A pointer value is *based* on another pointer value according to the
1139 - A pointer value formed from a ``getelementptr`` operation is *based*
1140 on the first operand of the ``getelementptr``.
1141 - The result value of a ``bitcast`` is *based* on the operand of the
1143 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1144 values that contribute (directly or indirectly) to the computation of
1145 the pointer's value.
1146 - The "*based* on" relationship is transitive.
1148 Note that this definition of *"based"* is intentionally similar to the
1149 definition of *"based"* in C99, though it is slightly weaker.
1151 LLVM IR does not associate types with memory. The result type of a
1152 ``load`` merely indicates the size and alignment of the memory from
1153 which to load, as well as the interpretation of the value. The first
1154 operand type of a ``store`` similarly only indicates the size and
1155 alignment of the store.
1157 Consequently, type-based alias analysis, aka TBAA, aka
1158 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1159 :ref:`Metadata <metadata>` may be used to encode additional information
1160 which specialized optimization passes may use to implement type-based
1165 Volatile Memory Accesses
1166 ------------------------
1168 Certain memory accesses, such as :ref:`load <i_load>`'s,
1169 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1170 marked ``volatile``. The optimizers must not change the number of
1171 volatile operations or change their order of execution relative to other
1172 volatile operations. The optimizers *may* change the order of volatile
1173 operations relative to non-volatile operations. This is not Java's
1174 "volatile" and has no cross-thread synchronization behavior.
1176 IR-level volatile loads and stores cannot safely be optimized into
1177 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1178 flagged volatile. Likewise, the backend should never split or merge
1179 target-legal volatile load/store instructions.
1181 .. admonition:: Rationale
1183 Platforms may rely on volatile loads and stores of natively supported
1184 data width to be executed as single instruction. For example, in C
1185 this holds for an l-value of volatile primitive type with native
1186 hardware support, but not necessarily for aggregate types. The
1187 frontend upholds these expectations, which are intentionally
1188 unspecified in the IR. The rules above ensure that IR transformation
1189 do not violate the frontend's contract with the language.
1193 Memory Model for Concurrent Operations
1194 --------------------------------------
1196 The LLVM IR does not define any way to start parallel threads of
1197 execution or to register signal handlers. Nonetheless, there are
1198 platform-specific ways to create them, and we define LLVM IR's behavior
1199 in their presence. This model is inspired by the C++0x memory model.
1201 For a more informal introduction to this model, see the :doc:`Atomics`.
1203 We define a *happens-before* partial order as the least partial order
1206 - Is a superset of single-thread program order, and
1207 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1208 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1209 techniques, like pthread locks, thread creation, thread joining,
1210 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1211 Constraints <ordering>`).
1213 Note that program order does not introduce *happens-before* edges
1214 between a thread and signals executing inside that thread.
1216 Every (defined) read operation (load instructions, memcpy, atomic
1217 loads/read-modify-writes, etc.) R reads a series of bytes written by
1218 (defined) write operations (store instructions, atomic
1219 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1220 section, initialized globals are considered to have a write of the
1221 initializer which is atomic and happens before any other read or write
1222 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1223 may see any write to the same byte, except:
1225 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1226 write\ :sub:`2` happens before R\ :sub:`byte`, then
1227 R\ :sub:`byte` does not see write\ :sub:`1`.
1228 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1229 R\ :sub:`byte` does not see write\ :sub:`3`.
1231 Given that definition, R\ :sub:`byte` is defined as follows:
1233 - If R is volatile, the result is target-dependent. (Volatile is
1234 supposed to give guarantees which can support ``sig_atomic_t`` in
1235 C/C++, and may be used for accesses to addresses which do not behave
1236 like normal memory. It does not generally provide cross-thread
1238 - Otherwise, if there is no write to the same byte that happens before
1239 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1240 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1241 R\ :sub:`byte` returns the value written by that write.
1242 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1243 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1244 Memory Ordering Constraints <ordering>` section for additional
1245 constraints on how the choice is made.
1246 - Otherwise R\ :sub:`byte` returns ``undef``.
1248 R returns the value composed of the series of bytes it read. This
1249 implies that some bytes within the value may be ``undef`` **without**
1250 the entire value being ``undef``. Note that this only defines the
1251 semantics of the operation; it doesn't mean that targets will emit more
1252 than one instruction to read the series of bytes.
1254 Note that in cases where none of the atomic intrinsics are used, this
1255 model places only one restriction on IR transformations on top of what
1256 is required for single-threaded execution: introducing a store to a byte
1257 which might not otherwise be stored is not allowed in general.
1258 (Specifically, in the case where another thread might write to and read
1259 from an address, introducing a store can change a load that may see
1260 exactly one write into a load that may see multiple writes.)
1264 Atomic Memory Ordering Constraints
1265 ----------------------------------
1267 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1268 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1269 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1270 an ordering parameter that determines which other atomic instructions on
1271 the same address they *synchronize with*. These semantics are borrowed
1272 from Java and C++0x, but are somewhat more colloquial. If these
1273 descriptions aren't precise enough, check those specs (see spec
1274 references in the :doc:`atomics guide <Atomics>`).
1275 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1276 differently since they don't take an address. See that instruction's
1277 documentation for details.
1279 For a simpler introduction to the ordering constraints, see the
1283 The set of values that can be read is governed by the happens-before
1284 partial order. A value cannot be read unless some operation wrote
1285 it. This is intended to provide a guarantee strong enough to model
1286 Java's non-volatile shared variables. This ordering cannot be
1287 specified for read-modify-write operations; it is not strong enough
1288 to make them atomic in any interesting way.
1290 In addition to the guarantees of ``unordered``, there is a single
1291 total order for modifications by ``monotonic`` operations on each
1292 address. All modification orders must be compatible with the
1293 happens-before order. There is no guarantee that the modification
1294 orders can be combined to a global total order for the whole program
1295 (and this often will not be possible). The read in an atomic
1296 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1297 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1298 order immediately before the value it writes. If one atomic read
1299 happens before another atomic read of the same address, the later
1300 read must see the same value or a later value in the address's
1301 modification order. This disallows reordering of ``monotonic`` (or
1302 stronger) operations on the same address. If an address is written
1303 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1304 read that address repeatedly, the other threads must eventually see
1305 the write. This corresponds to the C++0x/C1x
1306 ``memory_order_relaxed``.
1308 In addition to the guarantees of ``monotonic``, a
1309 *synchronizes-with* edge may be formed with a ``release`` operation.
1310 This is intended to model C++'s ``memory_order_acquire``.
1312 In addition to the guarantees of ``monotonic``, if this operation
1313 writes a value which is subsequently read by an ``acquire``
1314 operation, it *synchronizes-with* that operation. (This isn't a
1315 complete description; see the C++0x definition of a release
1316 sequence.) This corresponds to the C++0x/C1x
1317 ``memory_order_release``.
1318 ``acq_rel`` (acquire+release)
1319 Acts as both an ``acquire`` and ``release`` operation on its
1320 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1321 ``seq_cst`` (sequentially consistent)
1322 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1323 operation which only reads, ``release`` for an operation which only
1324 writes), there is a global total order on all
1325 sequentially-consistent operations on all addresses, which is
1326 consistent with the *happens-before* partial order and with the
1327 modification orders of all the affected addresses. Each
1328 sequentially-consistent read sees the last preceding write to the
1329 same address in this global order. This corresponds to the C++0x/C1x
1330 ``memory_order_seq_cst`` and Java volatile.
1334 If an atomic operation is marked ``singlethread``, it only *synchronizes
1335 with* or participates in modification and seq\_cst total orderings with
1336 other operations running in the same thread (for example, in signal
1344 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1345 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1346 :ref:`frem <i_frem>`) have the following flags that can set to enable
1347 otherwise unsafe floating point operations
1350 No NaNs - Allow optimizations to assume the arguments and result are not
1351 NaN. Such optimizations are required to retain defined behavior over
1352 NaNs, but the value of the result is undefined.
1355 No Infs - Allow optimizations to assume the arguments and result are not
1356 +/-Inf. Such optimizations are required to retain defined behavior over
1357 +/-Inf, but the value of the result is undefined.
1360 No Signed Zeros - Allow optimizations to treat the sign of a zero
1361 argument or result as insignificant.
1364 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1365 argument rather than perform division.
1368 Fast - Allow algebraically equivalent transformations that may
1369 dramatically change results in floating point (e.g. reassociate). This
1370 flag implies all the others.
1377 The LLVM type system is one of the most important features of the
1378 intermediate representation. Being typed enables a number of
1379 optimizations to be performed on the intermediate representation
1380 directly, without having to do extra analyses on the side before the
1381 transformation. A strong type system makes it easier to read the
1382 generated code and enables novel analyses and transformations that are
1383 not feasible to perform on normal three address code representations.
1385 .. _typeclassifications:
1387 Type Classifications
1388 --------------------
1390 The types fall into a few useful classifications:
1399 * - :ref:`integer <t_integer>`
1400 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1403 * - :ref:`floating point <t_floating>`
1404 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1412 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1413 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1414 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1415 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1417 * - :ref:`primitive <t_primitive>`
1418 - :ref:`label <t_label>`,
1419 :ref:`void <t_void>`,
1420 :ref:`integer <t_integer>`,
1421 :ref:`floating point <t_floating>`,
1422 :ref:`x86mmx <t_x86mmx>`,
1423 :ref:`metadata <t_metadata>`.
1425 * - :ref:`derived <t_derived>`
1426 - :ref:`array <t_array>`,
1427 :ref:`function <t_function>`,
1428 :ref:`pointer <t_pointer>`,
1429 :ref:`structure <t_struct>`,
1430 :ref:`vector <t_vector>`,
1431 :ref:`opaque <t_opaque>`.
1433 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1434 Values of these types are the only ones which can be produced by
1442 The primitive types are the fundamental building blocks of the LLVM
1453 The integer type is a very simple type that simply specifies an
1454 arbitrary bit width for the integer type desired. Any bit width from 1
1455 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1464 The number of bits the integer will occupy is specified by the ``N``
1470 +----------------+------------------------------------------------+
1471 | ``i1`` | a single-bit integer. |
1472 +----------------+------------------------------------------------+
1473 | ``i32`` | a 32-bit integer. |
1474 +----------------+------------------------------------------------+
1475 | ``i1942652`` | a really big integer of over 1 million bits. |
1476 +----------------+------------------------------------------------+
1480 Floating Point Types
1481 ^^^^^^^^^^^^^^^^^^^^
1490 - 16-bit floating point value
1493 - 32-bit floating point value
1496 - 64-bit floating point value
1499 - 128-bit floating point value (112-bit mantissa)
1502 - 80-bit floating point value (X87)
1505 - 128-bit floating point value (two 64-bits)
1515 The x86mmx type represents a value held in an MMX register on an x86
1516 machine. The operations allowed on it are quite limited: parameters and
1517 return values, load and store, and bitcast. User-specified MMX
1518 instructions are represented as intrinsic or asm calls with arguments
1519 and/or results of this type. There are no arrays, vectors or constants
1537 The void type does not represent any value and has no size.
1554 The label type represents code labels.
1571 The metadata type represents embedded metadata. No derived types may be
1572 created from metadata except for :ref:`function <t_function>` arguments.
1586 The real power in LLVM comes from the derived types in the system. This
1587 is what allows a programmer to represent arrays, functions, pointers,
1588 and other useful types. Each of these types contain one or more element
1589 types which may be a primitive type, or another derived type. For
1590 example, it is possible to have a two dimensional array, using an array
1591 as the element type of another array.
1598 Aggregate Types are a subset of derived types that can contain multiple
1599 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1600 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1611 The array type is a very simple derived type that arranges elements
1612 sequentially in memory. The array type requires a size (number of
1613 elements) and an underlying data type.
1620 [<# elements> x <elementtype>]
1622 The number of elements is a constant integer value; ``elementtype`` may
1623 be any type with a size.
1628 +------------------+--------------------------------------+
1629 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1630 +------------------+--------------------------------------+
1631 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1632 +------------------+--------------------------------------+
1633 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1634 +------------------+--------------------------------------+
1636 Here are some examples of multidimensional arrays:
1638 +-----------------------------+----------------------------------------------------------+
1639 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1640 +-----------------------------+----------------------------------------------------------+
1641 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1642 +-----------------------------+----------------------------------------------------------+
1643 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1644 +-----------------------------+----------------------------------------------------------+
1646 There is no restriction on indexing beyond the end of the array implied
1647 by a static type (though there are restrictions on indexing beyond the
1648 bounds of an allocated object in some cases). This means that
1649 single-dimension 'variable sized array' addressing can be implemented in
1650 LLVM with a zero length array type. An implementation of 'pascal style
1651 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1662 The function type can be thought of as a function signature. It consists
1663 of a return type and a list of formal parameter types. The return type
1664 of a function type is a first class type or a void type.
1671 <returntype> (<parameter list>)
1673 ...where '``<parameter list>``' is a comma-separated list of type
1674 specifiers. Optionally, the parameter list may include a type ``...``,
1675 which indicates that the function takes a variable number of arguments.
1676 Variable argument functions can access their arguments with the
1677 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1678 '``<returntype>``' is any type except :ref:`label <t_label>`.
1683 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1684 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1685 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1686 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1687 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1688 | ``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. |
1689 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1690 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1691 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1701 The structure type is used to represent a collection of data members
1702 together in memory. The elements of a structure may be any type that has
1705 Structures in memory are accessed using '``load``' and '``store``' by
1706 getting a pointer to a field with the '``getelementptr``' instruction.
1707 Structures in registers are accessed using the '``extractvalue``' and
1708 '``insertvalue``' instructions.
1710 Structures may optionally be "packed" structures, which indicate that
1711 the alignment of the struct is one byte, and that there is no padding
1712 between the elements. In non-packed structs, padding between field types
1713 is inserted as defined by the DataLayout string in the module, which is
1714 required to match what the underlying code generator expects.
1716 Structures can either be "literal" or "identified". A literal structure
1717 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1718 identified types are always defined at the top level with a name.
1719 Literal types are uniqued by their contents and can never be recursive
1720 or opaque since there is no way to write one. Identified types can be
1721 recursive, can be opaqued, and are never uniqued.
1728 %T1 = type { <type list> } ; Identified normal struct type
1729 %T2 = type <{ <type list> }> ; Identified packed struct type
1734 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1735 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1736 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1737 | ``{ 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``. |
1738 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1739 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1740 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1744 Opaque Structure Types
1745 ^^^^^^^^^^^^^^^^^^^^^^
1750 Opaque structure types are used to represent named structure types that
1751 do not have a body specified. This corresponds (for example) to the C
1752 notion of a forward declared structure.
1765 +--------------+-------------------+
1766 | ``opaque`` | An opaque type. |
1767 +--------------+-------------------+
1777 The pointer type is used to specify memory locations. Pointers are
1778 commonly used to reference objects in memory.
1780 Pointer types may have an optional address space attribute defining the
1781 numbered address space where the pointed-to object resides. The default
1782 address space is number zero. The semantics of non-zero address spaces
1783 are target-specific.
1785 Note that LLVM does not permit pointers to void (``void*``) nor does it
1786 permit pointers to labels (``label*``). Use ``i8*`` instead.
1798 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1799 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1800 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1801 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1802 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1803 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1804 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1814 A vector type is a simple derived type that represents a vector of
1815 elements. Vector types are used when multiple primitive data are
1816 operated in parallel using a single instruction (SIMD). A vector type
1817 requires a size (number of elements) and an underlying primitive data
1818 type. Vector types are considered :ref:`first class <t_firstclass>`.
1825 < <# elements> x <elementtype> >
1827 The number of elements is a constant integer value larger than 0;
1828 elementtype may be any integer or floating point type, or a pointer to
1829 these types. Vectors of size zero are not allowed.
1834 +-------------------+--------------------------------------------------+
1835 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1836 +-------------------+--------------------------------------------------+
1837 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1838 +-------------------+--------------------------------------------------+
1839 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1840 +-------------------+--------------------------------------------------+
1841 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1842 +-------------------+--------------------------------------------------+
1847 LLVM has several different basic types of constants. This section
1848 describes them all and their syntax.
1853 **Boolean constants**
1854 The two strings '``true``' and '``false``' are both valid constants
1856 **Integer constants**
1857 Standard integers (such as '4') are constants of the
1858 :ref:`integer <t_integer>` type. Negative numbers may be used with
1860 **Floating point constants**
1861 Floating point constants use standard decimal notation (e.g.
1862 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1863 hexadecimal notation (see below). The assembler requires the exact
1864 decimal value of a floating-point constant. For example, the
1865 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1866 decimal in binary. Floating point constants must have a :ref:`floating
1867 point <t_floating>` type.
1868 **Null pointer constants**
1869 The identifier '``null``' is recognized as a null pointer constant
1870 and must be of :ref:`pointer type <t_pointer>`.
1872 The one non-intuitive notation for constants is the hexadecimal form of
1873 floating point constants. For example, the form
1874 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1875 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1876 constants are required (and the only time that they are generated by the
1877 disassembler) is when a floating point constant must be emitted but it
1878 cannot be represented as a decimal floating point number in a reasonable
1879 number of digits. For example, NaN's, infinities, and other special
1880 values are represented in their IEEE hexadecimal format so that assembly
1881 and disassembly do not cause any bits to change in the constants.
1883 When using the hexadecimal form, constants of types half, float, and
1884 double are represented using the 16-digit form shown above (which
1885 matches the IEEE754 representation for double); half and float values
1886 must, however, be exactly representable as IEEE 754 half and single
1887 precision, respectively. Hexadecimal format is always used for long
1888 double, and there are three forms of long double. The 80-bit format used
1889 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1890 128-bit format used by PowerPC (two adjacent doubles) is represented by
1891 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1892 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
1893 will only work if they match the long double format on your target.
1894 The IEEE 16-bit format (half precision) is represented by ``0xH``
1895 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
1896 (sign bit at the left).
1898 There are no constants of type x86mmx.
1900 .. _complexconstants:
1905 Complex constants are a (potentially recursive) combination of simple
1906 constants and smaller complex constants.
1908 **Structure constants**
1909 Structure constants are represented with notation similar to
1910 structure type definitions (a comma separated list of elements,
1911 surrounded by braces (``{}``)). For example:
1912 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1913 "``@G = external global i32``". Structure constants must have
1914 :ref:`structure type <t_struct>`, and the number and types of elements
1915 must match those specified by the type.
1917 Array constants are represented with notation similar to array type
1918 definitions (a comma separated list of elements, surrounded by
1919 square brackets (``[]``)). For example:
1920 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1921 :ref:`array type <t_array>`, and the number and types of elements must
1922 match those specified by the type.
1923 **Vector constants**
1924 Vector constants are represented with notation similar to vector
1925 type definitions (a comma separated list of elements, surrounded by
1926 less-than/greater-than's (``<>``)). For example:
1927 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1928 must have :ref:`vector type <t_vector>`, and the number and types of
1929 elements must match those specified by the type.
1930 **Zero initialization**
1931 The string '``zeroinitializer``' can be used to zero initialize a
1932 value to zero of *any* type, including scalar and
1933 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1934 having to print large zero initializers (e.g. for large arrays) and
1935 is always exactly equivalent to using explicit zero initializers.
1937 A metadata node is a structure-like constant with :ref:`metadata
1938 type <t_metadata>`. For example:
1939 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
1940 constants that are meant to be interpreted as part of the
1941 instruction stream, metadata is a place to attach additional
1942 information such as debug info.
1944 Global Variable and Function Addresses
1945 --------------------------------------
1947 The addresses of :ref:`global variables <globalvars>` and
1948 :ref:`functions <functionstructure>` are always implicitly valid
1949 (link-time) constants. These constants are explicitly referenced when
1950 the :ref:`identifier for the global <identifiers>` is used and always have
1951 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
1954 .. code-block:: llvm
1958 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1965 The string '``undef``' can be used anywhere a constant is expected, and
1966 indicates that the user of the value may receive an unspecified
1967 bit-pattern. Undefined values may be of any type (other than '``label``'
1968 or '``void``') and be used anywhere a constant is permitted.
1970 Undefined values are useful because they indicate to the compiler that
1971 the program is well defined no matter what value is used. This gives the
1972 compiler more freedom to optimize. Here are some examples of
1973 (potentially surprising) transformations that are valid (in pseudo IR):
1975 .. code-block:: llvm
1985 This is safe because all of the output bits are affected by the undef
1986 bits. Any output bit can have a zero or one depending on the input bits.
1988 .. code-block:: llvm
1999 These logical operations have bits that are not always affected by the
2000 input. For example, if ``%X`` has a zero bit, then the output of the
2001 '``and``' operation will always be a zero for that bit, no matter what
2002 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2003 optimize or assume that the result of the '``and``' is '``undef``'.
2004 However, it is safe to assume that all bits of the '``undef``' could be
2005 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2006 all the bits of the '``undef``' operand to the '``or``' could be set,
2007 allowing the '``or``' to be folded to -1.
2009 .. code-block:: llvm
2011 %A = select undef, %X, %Y
2012 %B = select undef, 42, %Y
2013 %C = select %X, %Y, undef
2023 This set of examples shows that undefined '``select``' (and conditional
2024 branch) conditions can go *either way*, but they have to come from one
2025 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2026 both known to have a clear low bit, then ``%A`` would have to have a
2027 cleared low bit. However, in the ``%C`` example, the optimizer is
2028 allowed to assume that the '``undef``' operand could be the same as
2029 ``%Y``, allowing the whole '``select``' to be eliminated.
2031 .. code-block:: llvm
2033 %A = xor undef, undef
2050 This example points out that two '``undef``' operands are not
2051 necessarily the same. This can be surprising to people (and also matches
2052 C semantics) where they assume that "``X^X``" is always zero, even if
2053 ``X`` is undefined. This isn't true for a number of reasons, but the
2054 short answer is that an '``undef``' "variable" can arbitrarily change
2055 its value over its "live range". This is true because the variable
2056 doesn't actually *have a live range*. Instead, the value is logically
2057 read from arbitrary registers that happen to be around when needed, so
2058 the value is not necessarily consistent over time. In fact, ``%A`` and
2059 ``%C`` need to have the same semantics or the core LLVM "replace all
2060 uses with" concept would not hold.
2062 .. code-block:: llvm
2070 These examples show the crucial difference between an *undefined value*
2071 and *undefined behavior*. An undefined value (like '``undef``') is
2072 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2073 operation can be constant folded to '``undef``', because the '``undef``'
2074 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2075 However, in the second example, we can make a more aggressive
2076 assumption: because the ``undef`` is allowed to be an arbitrary value,
2077 we are allowed to assume that it could be zero. Since a divide by zero
2078 has *undefined behavior*, we are allowed to assume that the operation
2079 does not execute at all. This allows us to delete the divide and all
2080 code after it. Because the undefined operation "can't happen", the
2081 optimizer can assume that it occurs in dead code.
2083 .. code-block:: llvm
2085 a: store undef -> %X
2086 b: store %X -> undef
2091 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2092 value can be assumed to not have any effect; we can assume that the
2093 value is overwritten with bits that happen to match what was already
2094 there. However, a store *to* an undefined location could clobber
2095 arbitrary memory, therefore, it has undefined behavior.
2102 Poison values are similar to :ref:`undef values <undefvalues>`, however
2103 they also represent the fact that an instruction or constant expression
2104 which cannot evoke side effects has nevertheless detected a condition
2105 which results in undefined behavior.
2107 There is currently no way of representing a poison value in the IR; they
2108 only exist when produced by operations such as :ref:`add <i_add>` with
2111 Poison value behavior is defined in terms of value *dependence*:
2113 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2114 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2115 their dynamic predecessor basic block.
2116 - Function arguments depend on the corresponding actual argument values
2117 in the dynamic callers of their functions.
2118 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2119 instructions that dynamically transfer control back to them.
2120 - :ref:`Invoke <i_invoke>` instructions depend on the
2121 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2122 call instructions that dynamically transfer control back to them.
2123 - Non-volatile loads and stores depend on the most recent stores to all
2124 of the referenced memory addresses, following the order in the IR
2125 (including loads and stores implied by intrinsics such as
2126 :ref:`@llvm.memcpy <int_memcpy>`.)
2127 - An instruction with externally visible side effects depends on the
2128 most recent preceding instruction with externally visible side
2129 effects, following the order in the IR. (This includes :ref:`volatile
2130 operations <volatile>`.)
2131 - An instruction *control-depends* on a :ref:`terminator
2132 instruction <terminators>` if the terminator instruction has
2133 multiple successors and the instruction is always executed when
2134 control transfers to one of the successors, and may not be executed
2135 when control is transferred to another.
2136 - Additionally, an instruction also *control-depends* on a terminator
2137 instruction if the set of instructions it otherwise depends on would
2138 be different if the terminator had transferred control to a different
2140 - Dependence is transitive.
2142 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2143 with the additional affect that any instruction which has a *dependence*
2144 on a poison value has undefined behavior.
2146 Here are some examples:
2148 .. code-block:: llvm
2151 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2152 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2153 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2154 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2156 store i32 %poison, i32* @g ; Poison value stored to memory.
2157 %poison2 = load i32* @g ; Poison value loaded back from memory.
2159 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2161 %narrowaddr = bitcast i32* @g to i16*
2162 %wideaddr = bitcast i32* @g to i64*
2163 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2164 %poison4 = load i64* %wideaddr ; Returns a poison value.
2166 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2167 br i1 %cmp, label %true, label %end ; Branch to either destination.
2170 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2171 ; it has undefined behavior.
2175 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2176 ; Both edges into this PHI are
2177 ; control-dependent on %cmp, so this
2178 ; always results in a poison value.
2180 store volatile i32 0, i32* @g ; This would depend on the store in %true
2181 ; if %cmp is true, or the store in %entry
2182 ; otherwise, so this is undefined behavior.
2184 br i1 %cmp, label %second_true, label %second_end
2185 ; The same branch again, but this time the
2186 ; true block doesn't have side effects.
2193 store volatile i32 0, i32* @g ; This time, the instruction always depends
2194 ; on the store in %end. Also, it is
2195 ; control-equivalent to %end, so this is
2196 ; well-defined (ignoring earlier undefined
2197 ; behavior in this example).
2201 Addresses of Basic Blocks
2202 -------------------------
2204 ``blockaddress(@function, %block)``
2206 The '``blockaddress``' constant computes the address of the specified
2207 basic block in the specified function, and always has an ``i8*`` type.
2208 Taking the address of the entry block is illegal.
2210 This value only has defined behavior when used as an operand to the
2211 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2212 against null. Pointer equality tests between labels addresses results in
2213 undefined behavior --- though, again, comparison against null is ok, and
2214 no label is equal to the null pointer. This may be passed around as an
2215 opaque pointer sized value as long as the bits are not inspected. This
2216 allows ``ptrtoint`` and arithmetic to be performed on these values so
2217 long as the original value is reconstituted before the ``indirectbr``
2220 Finally, some targets may provide defined semantics when using the value
2221 as the operand to an inline assembly, but that is target specific.
2225 Constant Expressions
2226 --------------------
2228 Constant expressions are used to allow expressions involving other
2229 constants to be used as constants. Constant expressions may be of any
2230 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2231 that does not have side effects (e.g. load and call are not supported).
2232 The following is the syntax for constant expressions:
2234 ``trunc (CST to TYPE)``
2235 Truncate a constant to another type. The bit size of CST must be
2236 larger than the bit size of TYPE. Both types must be integers.
2237 ``zext (CST to TYPE)``
2238 Zero extend a constant to another type. The bit size of CST must be
2239 smaller than the bit size of TYPE. Both types must be integers.
2240 ``sext (CST to TYPE)``
2241 Sign extend a constant to another type. The bit size of CST must be
2242 smaller than the bit size of TYPE. Both types must be integers.
2243 ``fptrunc (CST to TYPE)``
2244 Truncate a floating point constant to another floating point type.
2245 The size of CST must be larger than the size of TYPE. Both types
2246 must be floating point.
2247 ``fpext (CST to TYPE)``
2248 Floating point extend a constant to another type. The size of CST
2249 must be smaller or equal to the size of TYPE. Both types must be
2251 ``fptoui (CST to TYPE)``
2252 Convert a floating point constant to the corresponding unsigned
2253 integer constant. TYPE must be a scalar or vector integer type. CST
2254 must be of scalar or vector floating point type. Both CST and TYPE
2255 must be scalars, or vectors of the same number of elements. If the
2256 value won't fit in the integer type, the results are undefined.
2257 ``fptosi (CST to TYPE)``
2258 Convert a floating point constant to the corresponding signed
2259 integer constant. TYPE must be a scalar or vector integer type. CST
2260 must be of scalar or vector floating point type. Both CST and TYPE
2261 must be scalars, or vectors of the same number of elements. If the
2262 value won't fit in the integer type, the results are undefined.
2263 ``uitofp (CST to TYPE)``
2264 Convert an unsigned integer constant to the corresponding floating
2265 point constant. TYPE must be a scalar or vector floating point type.
2266 CST must be of scalar or vector integer type. Both CST and TYPE must
2267 be scalars, or vectors of the same number of elements. If the value
2268 won't fit in the floating point type, the results are undefined.
2269 ``sitofp (CST to TYPE)``
2270 Convert a signed integer constant to the corresponding floating
2271 point constant. TYPE must be a scalar or vector floating point type.
2272 CST must be of scalar or vector integer type. Both CST and TYPE must
2273 be scalars, or vectors of the same number of elements. If the value
2274 won't fit in the floating point type, the results are undefined.
2275 ``ptrtoint (CST to TYPE)``
2276 Convert a pointer typed constant to the corresponding integer
2277 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2278 pointer type. The ``CST`` value is zero extended, truncated, or
2279 unchanged to make it fit in ``TYPE``.
2280 ``inttoptr (CST to TYPE)``
2281 Convert an integer constant to a pointer constant. TYPE must be a
2282 pointer type. CST must be of integer type. The CST value is zero
2283 extended, truncated, or unchanged to make it fit in a pointer size.
2284 This one is *really* dangerous!
2285 ``bitcast (CST to TYPE)``
2286 Convert a constant, CST, to another TYPE. The constraints of the
2287 operands are the same as those for the :ref:`bitcast
2288 instruction <i_bitcast>`.
2289 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2290 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2291 constants. As with the :ref:`getelementptr <i_getelementptr>`
2292 instruction, the index list may have zero or more indexes, which are
2293 required to make sense for the type of "CSTPTR".
2294 ``select (COND, VAL1, VAL2)``
2295 Perform the :ref:`select operation <i_select>` on constants.
2296 ``icmp COND (VAL1, VAL2)``
2297 Performs the :ref:`icmp operation <i_icmp>` on constants.
2298 ``fcmp COND (VAL1, VAL2)``
2299 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2300 ``extractelement (VAL, IDX)``
2301 Perform the :ref:`extractelement operation <i_extractelement>` on
2303 ``insertelement (VAL, ELT, IDX)``
2304 Perform the :ref:`insertelement operation <i_insertelement>` on
2306 ``shufflevector (VEC1, VEC2, IDXMASK)``
2307 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2309 ``extractvalue (VAL, IDX0, IDX1, ...)``
2310 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2311 constants. The index list is interpreted in a similar manner as
2312 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2313 least one index value must be specified.
2314 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2315 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2316 The index list is interpreted in a similar manner as indices in a
2317 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2318 value must be specified.
2319 ``OPCODE (LHS, RHS)``
2320 Perform the specified operation of the LHS and RHS constants. OPCODE
2321 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2322 binary <bitwiseops>` operations. The constraints on operands are
2323 the same as those for the corresponding instruction (e.g. no bitwise
2324 operations on floating point values are allowed).
2331 Inline Assembler Expressions
2332 ----------------------------
2334 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2335 Inline Assembly <moduleasm>`) through the use of a special value. This
2336 value represents the inline assembler as a string (containing the
2337 instructions to emit), a list of operand constraints (stored as a
2338 string), a flag that indicates whether or not the inline asm expression
2339 has side effects, and a flag indicating whether the function containing
2340 the asm needs to align its stack conservatively. An example inline
2341 assembler expression is:
2343 .. code-block:: llvm
2345 i32 (i32) asm "bswap $0", "=r,r"
2347 Inline assembler expressions may **only** be used as the callee operand
2348 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2349 Thus, typically we have:
2351 .. code-block:: llvm
2353 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2355 Inline asms with side effects not visible in the constraint list must be
2356 marked as having side effects. This is done through the use of the
2357 '``sideeffect``' keyword, like so:
2359 .. code-block:: llvm
2361 call void asm sideeffect "eieio", ""()
2363 In some cases inline asms will contain code that will not work unless
2364 the stack is aligned in some way, such as calls or SSE instructions on
2365 x86, yet will not contain code that does that alignment within the asm.
2366 The compiler should make conservative assumptions about what the asm
2367 might contain and should generate its usual stack alignment code in the
2368 prologue if the '``alignstack``' keyword is present:
2370 .. code-block:: llvm
2372 call void asm alignstack "eieio", ""()
2374 Inline asms also support using non-standard assembly dialects. The
2375 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2376 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2377 the only supported dialects. An example is:
2379 .. code-block:: llvm
2381 call void asm inteldialect "eieio", ""()
2383 If multiple keywords appear the '``sideeffect``' keyword must come
2384 first, the '``alignstack``' keyword second and the '``inteldialect``'
2390 The call instructions that wrap inline asm nodes may have a
2391 "``!srcloc``" MDNode attached to it that contains a list of constant
2392 integers. If present, the code generator will use the integer as the
2393 location cookie value when report errors through the ``LLVMContext``
2394 error reporting mechanisms. This allows a front-end to correlate backend
2395 errors that occur with inline asm back to the source code that produced
2398 .. code-block:: llvm
2400 call void asm sideeffect "something bad", ""(), !srcloc !42
2402 !42 = !{ i32 1234567 }
2404 It is up to the front-end to make sense of the magic numbers it places
2405 in the IR. If the MDNode contains multiple constants, the code generator
2406 will use the one that corresponds to the line of the asm that the error
2411 Metadata Nodes and Metadata Strings
2412 -----------------------------------
2414 LLVM IR allows metadata to be attached to instructions in the program
2415 that can convey extra information about the code to the optimizers and
2416 code generator. One example application of metadata is source-level
2417 debug information. There are two metadata primitives: strings and nodes.
2418 All metadata has the ``metadata`` type and is identified in syntax by a
2419 preceding exclamation point ('``!``').
2421 A metadata string is a string surrounded by double quotes. It can
2422 contain any character by escaping non-printable characters with
2423 "``\xx``" where "``xx``" is the two digit hex code. For example:
2426 Metadata nodes are represented with notation similar to structure
2427 constants (a comma separated list of elements, surrounded by braces and
2428 preceded by an exclamation point). Metadata nodes can have any values as
2429 their operand. For example:
2431 .. code-block:: llvm
2433 !{ metadata !"test\00", i32 10}
2435 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2436 metadata nodes, which can be looked up in the module symbol table. For
2439 .. code-block:: llvm
2441 !foo = metadata !{!4, !3}
2443 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2444 function is using two metadata arguments:
2446 .. code-block:: llvm
2448 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2450 Metadata can be attached with an instruction. Here metadata ``!21`` is
2451 attached to the ``add`` instruction using the ``!dbg`` identifier:
2453 .. code-block:: llvm
2455 %indvar.next = add i64 %indvar, 1, !dbg !21
2457 More information about specific metadata nodes recognized by the
2458 optimizers and code generator is found below.
2463 In LLVM IR, memory does not have types, so LLVM's own type system is not
2464 suitable for doing TBAA. Instead, metadata is added to the IR to
2465 describe a type system of a higher level language. This can be used to
2466 implement typical C/C++ TBAA, but it can also be used to implement
2467 custom alias analysis behavior for other languages.
2469 The current metadata format is very simple. TBAA metadata nodes have up
2470 to three fields, e.g.:
2472 .. code-block:: llvm
2474 !0 = metadata !{ metadata !"an example type tree" }
2475 !1 = metadata !{ metadata !"int", metadata !0 }
2476 !2 = metadata !{ metadata !"float", metadata !0 }
2477 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2479 The first field is an identity field. It can be any value, usually a
2480 metadata string, which uniquely identifies the type. The most important
2481 name in the tree is the name of the root node. Two trees with different
2482 root node names are entirely disjoint, even if they have leaves with
2485 The second field identifies the type's parent node in the tree, or is
2486 null or omitted for a root node. A type is considered to alias all of
2487 its descendants and all of its ancestors in the tree. Also, a type is
2488 considered to alias all types in other trees, so that bitcode produced
2489 from multiple front-ends is handled conservatively.
2491 If the third field is present, it's an integer which if equal to 1
2492 indicates that the type is "constant" (meaning
2493 ``pointsToConstantMemory`` should return true; see `other useful
2494 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2496 '``tbaa.struct``' Metadata
2497 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2499 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2500 aggregate assignment operations in C and similar languages, however it
2501 is defined to copy a contiguous region of memory, which is more than
2502 strictly necessary for aggregate types which contain holes due to
2503 padding. Also, it doesn't contain any TBAA information about the fields
2506 ``!tbaa.struct`` metadata can describe which memory subregions in a
2507 memcpy are padding and what the TBAA tags of the struct are.
2509 The current metadata format is very simple. ``!tbaa.struct`` metadata
2510 nodes are a list of operands which are in conceptual groups of three.
2511 For each group of three, the first operand gives the byte offset of a
2512 field in bytes, the second gives its size in bytes, and the third gives
2515 .. code-block:: llvm
2517 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2519 This describes a struct with two fields. The first is at offset 0 bytes
2520 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2521 and has size 4 bytes and has tbaa tag !2.
2523 Note that the fields need not be contiguous. In this example, there is a
2524 4 byte gap between the two fields. This gap represents padding which
2525 does not carry useful data and need not be preserved.
2527 '``fpmath``' Metadata
2528 ^^^^^^^^^^^^^^^^^^^^^
2530 ``fpmath`` metadata may be attached to any instruction of floating point
2531 type. It can be used to express the maximum acceptable error in the
2532 result of that instruction, in ULPs, thus potentially allowing the
2533 compiler to use a more efficient but less accurate method of computing
2534 it. ULP is defined as follows:
2536 If ``x`` is a real number that lies between two finite consecutive
2537 floating-point numbers ``a`` and ``b``, without being equal to one
2538 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2539 distance between the two non-equal finite floating-point numbers
2540 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2542 The metadata node shall consist of a single positive floating point
2543 number representing the maximum relative error, for example:
2545 .. code-block:: llvm
2547 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2549 '``range``' Metadata
2550 ^^^^^^^^^^^^^^^^^^^^
2552 ``range`` metadata may be attached only to loads of integer types. It
2553 expresses the possible ranges the loaded value is in. The ranges are
2554 represented with a flattened list of integers. The loaded value is known
2555 to be in the union of the ranges defined by each consecutive pair. Each
2556 pair has the following properties:
2558 - The type must match the type loaded by the instruction.
2559 - The pair ``a,b`` represents the range ``[a,b)``.
2560 - Both ``a`` and ``b`` are constants.
2561 - The range is allowed to wrap.
2562 - The range should not represent the full or empty set. That is,
2565 In addition, the pairs must be in signed order of the lower bound and
2566 they must be non-contiguous.
2570 .. code-block:: llvm
2572 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2573 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2574 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2575 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2577 !0 = metadata !{ i8 0, i8 2 }
2578 !1 = metadata !{ i8 255, i8 2 }
2579 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2580 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2585 It is sometimes useful to attach information to loop constructs. Currently,
2586 loop metadata is implemented as metadata attached to the branch instruction
2587 in the loop latch block. This type of metadata refer to a metadata node that is
2588 guaranteed to be separate for each loop. The loop identifier metadata is
2589 specified with the name ``llvm.loop``.
2591 The loop identifier metadata is implemented using a metadata that refers to
2592 itself to avoid merging it with any other identifier metadata, e.g.,
2593 during module linkage or function inlining. That is, each loop should refer
2594 to their own identification metadata even if they reside in separate functions.
2595 The following example contains loop identifier metadata for two separate loop
2598 .. code-block:: llvm
2600 !0 = metadata !{ metadata !0 }
2601 !1 = metadata !{ metadata !1 }
2603 The loop identifier metadata can be used to specify additional per-loop
2604 metadata. Any operands after the first operand can be treated as user-defined
2605 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2606 by the loop vectorizer to indicate how many times to unroll the loop:
2608 .. code-block:: llvm
2610 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2612 !0 = metadata !{ metadata !0, metadata !1 }
2613 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2618 Metadata types used to annotate memory accesses with information helpful
2619 for optimizations are prefixed with ``llvm.mem``.
2621 '``llvm.mem.parallel_loop_access``' Metadata
2622 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2624 For a loop to be parallel, in addition to using
2625 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2626 also all of the memory accessing instructions in the loop body need to be
2627 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2628 is at least one memory accessing instruction not marked with the metadata,
2629 the loop must be considered a sequential loop. This causes parallel loops to be
2630 converted to sequential loops due to optimization passes that are unaware of
2631 the parallel semantics and that insert new memory instructions to the loop
2634 Example of a loop that is considered parallel due to its correct use of
2635 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2636 metadata types that refer to the same loop identifier metadata.
2638 .. code-block:: llvm
2642 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2644 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2646 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2650 !0 = metadata !{ metadata !0 }
2652 It is also possible to have nested parallel loops. In that case the
2653 memory accesses refer to a list of loop identifier metadata nodes instead of
2654 the loop identifier metadata node directly:
2656 .. code-block:: llvm
2663 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2665 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2667 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2671 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2673 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2675 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2677 outer.for.end: ; preds = %for.body
2679 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2680 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2681 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2683 '``llvm.vectorizer``'
2684 ^^^^^^^^^^^^^^^^^^^^^
2686 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2687 vectorization parameters such as vectorization factor and unroll factor.
2689 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2690 loop identification metadata.
2692 '``llvm.vectorizer.unroll``' Metadata
2693 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2695 This metadata instructs the loop vectorizer to unroll the specified
2696 loop exactly ``N`` times.
2698 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2699 operand is an integer specifying the unroll factor. For example:
2701 .. code-block:: llvm
2703 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2705 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2708 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2709 determined automatically.
2711 '``llvm.vectorizer.width``' Metadata
2712 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2714 This metadata sets the target width of the vectorizer to ``N``. Without
2715 this metadata, the vectorizer will choose a width automatically.
2716 Regardless of this metadata, the vectorizer will only vectorize loops if
2717 it believes it is valid to do so.
2719 The first operand is the string ``llvm.vectorizer.width`` and the second
2720 operand is an integer specifying the width. For example:
2722 .. code-block:: llvm
2724 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2726 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2729 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2732 Module Flags Metadata
2733 =====================
2735 Information about the module as a whole is difficult to convey to LLVM's
2736 subsystems. The LLVM IR isn't sufficient to transmit this information.
2737 The ``llvm.module.flags`` named metadata exists in order to facilitate
2738 this. These flags are in the form of key / value pairs --- much like a
2739 dictionary --- making it easy for any subsystem who cares about a flag to
2742 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2743 Each triplet has the following form:
2745 - The first element is a *behavior* flag, which specifies the behavior
2746 when two (or more) modules are merged together, and it encounters two
2747 (or more) metadata with the same ID. The supported behaviors are
2749 - The second element is a metadata string that is a unique ID for the
2750 metadata. Each module may only have one flag entry for each unique ID (not
2751 including entries with the **Require** behavior).
2752 - The third element is the value of the flag.
2754 When two (or more) modules are merged together, the resulting
2755 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2756 each unique metadata ID string, there will be exactly one entry in the merged
2757 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2758 be determined by the merge behavior flag, as described below. The only exception
2759 is that entries with the *Require* behavior are always preserved.
2761 The following behaviors are supported:
2772 Emits an error if two values disagree, otherwise the resulting value
2773 is that of the operands.
2777 Emits a warning if two values disagree. The result value will be the
2778 operand for the flag from the first module being linked.
2782 Adds a requirement that another module flag be present and have a
2783 specified value after linking is performed. The value must be a
2784 metadata pair, where the first element of the pair is the ID of the
2785 module flag to be restricted, and the second element of the pair is
2786 the value the module flag should be restricted to. This behavior can
2787 be used to restrict the allowable results (via triggering of an
2788 error) of linking IDs with the **Override** behavior.
2792 Uses the specified value, regardless of the behavior or value of the
2793 other module. If both modules specify **Override**, but the values
2794 differ, an error will be emitted.
2798 Appends the two values, which are required to be metadata nodes.
2802 Appends the two values, which are required to be metadata
2803 nodes. However, duplicate entries in the second list are dropped
2804 during the append operation.
2806 It is an error for a particular unique flag ID to have multiple behaviors,
2807 except in the case of **Require** (which adds restrictions on another metadata
2808 value) or **Override**.
2810 An example of module flags:
2812 .. code-block:: llvm
2814 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2815 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2816 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2817 !3 = metadata !{ i32 3, metadata !"qux",
2819 metadata !"foo", i32 1
2822 !llvm.module.flags = !{ !0, !1, !2, !3 }
2824 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2825 if two or more ``!"foo"`` flags are seen is to emit an error if their
2826 values are not equal.
2828 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2829 behavior if two or more ``!"bar"`` flags are seen is to use the value
2832 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2833 behavior if two or more ``!"qux"`` flags are seen is to emit a
2834 warning if their values are not equal.
2836 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2840 metadata !{ metadata !"foo", i32 1 }
2842 The behavior is to emit an error if the ``llvm.module.flags`` does not
2843 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2846 Objective-C Garbage Collection Module Flags Metadata
2847 ----------------------------------------------------
2849 On the Mach-O platform, Objective-C stores metadata about garbage
2850 collection in a special section called "image info". The metadata
2851 consists of a version number and a bitmask specifying what types of
2852 garbage collection are supported (if any) by the file. If two or more
2853 modules are linked together their garbage collection metadata needs to
2854 be merged rather than appended together.
2856 The Objective-C garbage collection module flags metadata consists of the
2857 following key-value pairs:
2866 * - ``Objective-C Version``
2867 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2869 * - ``Objective-C Image Info Version``
2870 - **[Required]** --- The version of the image info section. Currently
2873 * - ``Objective-C Image Info Section``
2874 - **[Required]** --- The section to place the metadata. Valid values are
2875 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2876 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2877 Objective-C ABI version 2.
2879 * - ``Objective-C Garbage Collection``
2880 - **[Required]** --- Specifies whether garbage collection is supported or
2881 not. Valid values are 0, for no garbage collection, and 2, for garbage
2882 collection supported.
2884 * - ``Objective-C GC Only``
2885 - **[Optional]** --- Specifies that only garbage collection is supported.
2886 If present, its value must be 6. This flag requires that the
2887 ``Objective-C Garbage Collection`` flag have the value 2.
2889 Some important flag interactions:
2891 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2892 merged with a module with ``Objective-C Garbage Collection`` set to
2893 2, then the resulting module has the
2894 ``Objective-C Garbage Collection`` flag set to 0.
2895 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2896 merged with a module with ``Objective-C GC Only`` set to 6.
2898 Automatic Linker Flags Module Flags Metadata
2899 --------------------------------------------
2901 Some targets support embedding flags to the linker inside individual object
2902 files. Typically this is used in conjunction with language extensions which
2903 allow source files to explicitly declare the libraries they depend on, and have
2904 these automatically be transmitted to the linker via object files.
2906 These flags are encoded in the IR using metadata in the module flags section,
2907 using the ``Linker Options`` key. The merge behavior for this flag is required
2908 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2909 node which should be a list of other metadata nodes, each of which should be a
2910 list of metadata strings defining linker options.
2912 For example, the following metadata section specifies two separate sets of
2913 linker options, presumably to link against ``libz`` and the ``Cocoa``
2916 !0 = metadata !{ i32 6, metadata !"Linker Options",
2918 metadata !{ metadata !"-lz" },
2919 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2920 !llvm.module.flags = !{ !0 }
2922 The metadata encoding as lists of lists of options, as opposed to a collapsed
2923 list of options, is chosen so that the IR encoding can use multiple option
2924 strings to specify e.g., a single library, while still having that specifier be
2925 preserved as an atomic element that can be recognized by a target specific
2926 assembly writer or object file emitter.
2928 Each individual option is required to be either a valid option for the target's
2929 linker, or an option that is reserved by the target specific assembly writer or
2930 object file emitter. No other aspect of these options is defined by the IR.
2932 .. _intrinsicglobalvariables:
2934 Intrinsic Global Variables
2935 ==========================
2937 LLVM has a number of "magic" global variables that contain data that
2938 affect code generation or other IR semantics. These are documented here.
2939 All globals of this sort should have a section specified as
2940 "``llvm.metadata``". This section and all globals that start with
2941 "``llvm.``" are reserved for use by LLVM.
2945 The '``llvm.used``' Global Variable
2946 -----------------------------------
2948 The ``@llvm.used`` global is an array which has
2949 :ref:`appending linkage <linkage_appending>`. This array contains a list of
2950 pointers to named global variables, functions and aliases which may optionally
2951 have a pointer cast formed of bitcast or getelementptr. For example, a legal
2954 .. code-block:: llvm
2959 @llvm.used = appending global [2 x i8*] [
2961 i8* bitcast (i32* @Y to i8*)
2962 ], section "llvm.metadata"
2964 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
2965 and linker are required to treat the symbol as if there is a reference to the
2966 symbol that it cannot see (which is why they have to be named). For example, if
2967 a variable has internal linkage and no references other than that from the
2968 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
2969 references from inline asms and other things the compiler cannot "see", and
2970 corresponds to "``attribute((used))``" in GNU C.
2972 On some targets, the code generator must emit a directive to the
2973 assembler or object file to prevent the assembler and linker from
2974 molesting the symbol.
2976 .. _gv_llvmcompilerused:
2978 The '``llvm.compiler.used``' Global Variable
2979 --------------------------------------------
2981 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
2982 directive, except that it only prevents the compiler from touching the
2983 symbol. On targets that support it, this allows an intelligent linker to
2984 optimize references to the symbol without being impeded as it would be
2987 This is a rare construct that should only be used in rare circumstances,
2988 and should not be exposed to source languages.
2990 .. _gv_llvmglobalctors:
2992 The '``llvm.global_ctors``' Global Variable
2993 -------------------------------------------
2995 .. code-block:: llvm
2997 %0 = type { i32, void ()* }
2998 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3000 The ``@llvm.global_ctors`` array contains a list of constructor
3001 functions and associated priorities. The functions referenced by this
3002 array will be called in ascending order of priority (i.e. lowest first)
3003 when the module is loaded. The order of functions with the same priority
3006 .. _llvmglobaldtors:
3008 The '``llvm.global_dtors``' Global Variable
3009 -------------------------------------------
3011 .. code-block:: llvm
3013 %0 = type { i32, void ()* }
3014 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3016 The ``@llvm.global_dtors`` array contains a list of destructor functions
3017 and associated priorities. The functions referenced by this array will
3018 be called in descending order of priority (i.e. highest first) when the
3019 module is loaded. The order of functions with the same priority is not
3022 Instruction Reference
3023 =====================
3025 The LLVM instruction set consists of several different classifications
3026 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3027 instructions <binaryops>`, :ref:`bitwise binary
3028 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3029 :ref:`other instructions <otherops>`.
3033 Terminator Instructions
3034 -----------------------
3036 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3037 program ends with a "Terminator" instruction, which indicates which
3038 block should be executed after the current block is finished. These
3039 terminator instructions typically yield a '``void``' value: they produce
3040 control flow, not values (the one exception being the
3041 ':ref:`invoke <i_invoke>`' instruction).
3043 The terminator instructions are: ':ref:`ret <i_ret>`',
3044 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3045 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3046 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3050 '``ret``' Instruction
3051 ^^^^^^^^^^^^^^^^^^^^^
3058 ret <type> <value> ; Return a value from a non-void function
3059 ret void ; Return from void function
3064 The '``ret``' instruction is used to return control flow (and optionally
3065 a value) from a function back to the caller.
3067 There are two forms of the '``ret``' instruction: one that returns a
3068 value and then causes control flow, and one that just causes control
3074 The '``ret``' instruction optionally accepts a single argument, the
3075 return value. The type of the return value must be a ':ref:`first
3076 class <t_firstclass>`' type.
3078 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3079 return type and contains a '``ret``' instruction with no return value or
3080 a return value with a type that does not match its type, or if it has a
3081 void return type and contains a '``ret``' instruction with a return
3087 When the '``ret``' instruction is executed, control flow returns back to
3088 the calling function's context. If the caller is a
3089 ":ref:`call <i_call>`" instruction, execution continues at the
3090 instruction after the call. If the caller was an
3091 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3092 beginning of the "normal" destination block. If the instruction returns
3093 a value, that value shall set the call or invoke instruction's return
3099 .. code-block:: llvm
3101 ret i32 5 ; Return an integer value of 5
3102 ret void ; Return from a void function
3103 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3107 '``br``' Instruction
3108 ^^^^^^^^^^^^^^^^^^^^
3115 br i1 <cond>, label <iftrue>, label <iffalse>
3116 br label <dest> ; Unconditional branch
3121 The '``br``' instruction is used to cause control flow to transfer to a
3122 different basic block in the current function. There are two forms of
3123 this instruction, corresponding to a conditional branch and an
3124 unconditional branch.
3129 The conditional branch form of the '``br``' instruction takes a single
3130 '``i1``' value and two '``label``' values. The unconditional form of the
3131 '``br``' instruction takes a single '``label``' value as a target.
3136 Upon execution of a conditional '``br``' instruction, the '``i1``'
3137 argument is evaluated. If the value is ``true``, control flows to the
3138 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3139 to the '``iffalse``' ``label`` argument.
3144 .. code-block:: llvm
3147 %cond = icmp eq i32 %a, %b
3148 br i1 %cond, label %IfEqual, label %IfUnequal
3156 '``switch``' Instruction
3157 ^^^^^^^^^^^^^^^^^^^^^^^^
3164 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3169 The '``switch``' instruction is used to transfer control flow to one of
3170 several different places. It is a generalization of the '``br``'
3171 instruction, allowing a branch to occur to one of many possible
3177 The '``switch``' instruction uses three parameters: an integer
3178 comparison value '``value``', a default '``label``' destination, and an
3179 array of pairs of comparison value constants and '``label``'s. The table
3180 is not allowed to contain duplicate constant entries.
3185 The ``switch`` instruction specifies a table of values and destinations.
3186 When the '``switch``' instruction is executed, this table is searched
3187 for the given value. If the value is found, control flow is transferred
3188 to the corresponding destination; otherwise, control flow is transferred
3189 to the default destination.
3194 Depending on properties of the target machine and the particular
3195 ``switch`` instruction, this instruction may be code generated in
3196 different ways. For example, it could be generated as a series of
3197 chained conditional branches or with a lookup table.
3202 .. code-block:: llvm
3204 ; Emulate a conditional br instruction
3205 %Val = zext i1 %value to i32
3206 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3208 ; Emulate an unconditional br instruction
3209 switch i32 0, label %dest [ ]
3211 ; Implement a jump table:
3212 switch i32 %val, label %otherwise [ i32 0, label %onzero
3214 i32 2, label %ontwo ]
3218 '``indirectbr``' Instruction
3219 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3226 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3231 The '``indirectbr``' instruction implements an indirect branch to a
3232 label within the current function, whose address is specified by
3233 "``address``". Address must be derived from a
3234 :ref:`blockaddress <blockaddress>` constant.
3239 The '``address``' argument is the address of the label to jump to. The
3240 rest of the arguments indicate the full set of possible destinations
3241 that the address may point to. Blocks are allowed to occur multiple
3242 times in the destination list, though this isn't particularly useful.
3244 This destination list is required so that dataflow analysis has an
3245 accurate understanding of the CFG.
3250 Control transfers to the block specified in the address argument. All
3251 possible destination blocks must be listed in the label list, otherwise
3252 this instruction has undefined behavior. This implies that jumps to
3253 labels defined in other functions have undefined behavior as well.
3258 This is typically implemented with a jump through a register.
3263 .. code-block:: llvm
3265 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3269 '``invoke``' Instruction
3270 ^^^^^^^^^^^^^^^^^^^^^^^^
3277 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3278 to label <normal label> unwind label <exception label>
3283 The '``invoke``' instruction causes control to transfer to a specified
3284 function, with the possibility of control flow transfer to either the
3285 '``normal``' label or the '``exception``' label. If the callee function
3286 returns with the "``ret``" instruction, control flow will return to the
3287 "normal" label. If the callee (or any indirect callees) returns via the
3288 ":ref:`resume <i_resume>`" instruction or other exception handling
3289 mechanism, control is interrupted and continued at the dynamically
3290 nearest "exception" label.
3292 The '``exception``' label is a `landing
3293 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3294 '``exception``' label is required to have the
3295 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3296 information about the behavior of the program after unwinding happens,
3297 as its first non-PHI instruction. The restrictions on the
3298 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3299 instruction, so that the important information contained within the
3300 "``landingpad``" instruction can't be lost through normal code motion.
3305 This instruction requires several arguments:
3307 #. The optional "cconv" marker indicates which :ref:`calling
3308 convention <callingconv>` the call should use. If none is
3309 specified, the call defaults to using C calling conventions.
3310 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3311 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3313 #. '``ptr to function ty``': shall be the signature of the pointer to
3314 function value being invoked. In most cases, this is a direct
3315 function invocation, but indirect ``invoke``'s are just as possible,
3316 branching off an arbitrary pointer to function value.
3317 #. '``function ptr val``': An LLVM value containing a pointer to a
3318 function to be invoked.
3319 #. '``function args``': argument list whose types match the function
3320 signature argument types and parameter attributes. All arguments must
3321 be of :ref:`first class <t_firstclass>` type. If the function signature
3322 indicates the function accepts a variable number of arguments, the
3323 extra arguments can be specified.
3324 #. '``normal label``': the label reached when the called function
3325 executes a '``ret``' instruction.
3326 #. '``exception label``': the label reached when a callee returns via
3327 the :ref:`resume <i_resume>` instruction or other exception handling
3329 #. The optional :ref:`function attributes <fnattrs>` list. Only
3330 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3331 attributes are valid here.
3336 This instruction is designed to operate as a standard '``call``'
3337 instruction in most regards. The primary difference is that it
3338 establishes an association with a label, which is used by the runtime
3339 library to unwind the stack.
3341 This instruction is used in languages with destructors to ensure that
3342 proper cleanup is performed in the case of either a ``longjmp`` or a
3343 thrown exception. Additionally, this is important for implementation of
3344 '``catch``' clauses in high-level languages that support them.
3346 For the purposes of the SSA form, the definition of the value returned
3347 by the '``invoke``' instruction is deemed to occur on the edge from the
3348 current block to the "normal" label. If the callee unwinds then no
3349 return value is available.
3354 .. code-block:: llvm
3356 %retval = invoke i32 @Test(i32 15) to label %Continue
3357 unwind label %TestCleanup ; {i32}:retval set
3358 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3359 unwind label %TestCleanup ; {i32}:retval set
3363 '``resume``' Instruction
3364 ^^^^^^^^^^^^^^^^^^^^^^^^
3371 resume <type> <value>
3376 The '``resume``' instruction is a terminator instruction that has no
3382 The '``resume``' instruction requires one argument, which must have the
3383 same type as the result of any '``landingpad``' instruction in the same
3389 The '``resume``' instruction resumes propagation of an existing
3390 (in-flight) exception whose unwinding was interrupted with a
3391 :ref:`landingpad <i_landingpad>` instruction.
3396 .. code-block:: llvm
3398 resume { i8*, i32 } %exn
3402 '``unreachable``' Instruction
3403 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3415 The '``unreachable``' instruction has no defined semantics. This
3416 instruction is used to inform the optimizer that a particular portion of
3417 the code is not reachable. This can be used to indicate that the code
3418 after a no-return function cannot be reached, and other facts.
3423 The '``unreachable``' instruction has no defined semantics.
3430 Binary operators are used to do most of the computation in a program.
3431 They require two operands of the same type, execute an operation on
3432 them, and produce a single value. The operands might represent multiple
3433 data, as is the case with the :ref:`vector <t_vector>` data type. The
3434 result value has the same type as its operands.
3436 There are several different binary operators:
3440 '``add``' Instruction
3441 ^^^^^^^^^^^^^^^^^^^^^
3448 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3449 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3450 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3451 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3456 The '``add``' instruction returns the sum of its two operands.
3461 The two arguments to the '``add``' instruction must be
3462 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3463 arguments must have identical types.
3468 The value produced is the integer sum of the two operands.
3470 If the sum has unsigned overflow, the result returned is the
3471 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3474 Because LLVM integers use a two's complement representation, this
3475 instruction is appropriate for both signed and unsigned integers.
3477 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3478 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3479 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3480 unsigned and/or signed overflow, respectively, occurs.
3485 .. code-block:: llvm
3487 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3491 '``fadd``' Instruction
3492 ^^^^^^^^^^^^^^^^^^^^^^
3499 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3504 The '``fadd``' instruction returns the sum of its two operands.
3509 The two arguments to the '``fadd``' instruction must be :ref:`floating
3510 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3511 Both arguments must have identical types.
3516 The value produced is the floating point sum of the two operands. This
3517 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3518 which are optimization hints to enable otherwise unsafe floating point
3524 .. code-block:: llvm
3526 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3528 '``sub``' Instruction
3529 ^^^^^^^^^^^^^^^^^^^^^
3536 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3537 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3538 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3539 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3544 The '``sub``' instruction returns the difference of its two operands.
3546 Note that the '``sub``' instruction is used to represent the '``neg``'
3547 instruction present in most other intermediate representations.
3552 The two arguments to the '``sub``' instruction must be
3553 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3554 arguments must have identical types.
3559 The value produced is the integer difference of the two operands.
3561 If the difference has unsigned overflow, the result returned is the
3562 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3565 Because LLVM integers use a two's complement representation, this
3566 instruction is appropriate for both signed and unsigned integers.
3568 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3569 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3570 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3571 unsigned and/or signed overflow, respectively, occurs.
3576 .. code-block:: llvm
3578 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3579 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3583 '``fsub``' Instruction
3584 ^^^^^^^^^^^^^^^^^^^^^^
3591 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3596 The '``fsub``' instruction returns the difference of its two operands.
3598 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3599 instruction present in most other intermediate representations.
3604 The two arguments to the '``fsub``' instruction must be :ref:`floating
3605 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3606 Both arguments must have identical types.
3611 The value produced is the floating point difference of the two operands.
3612 This instruction can also take any number of :ref:`fast-math
3613 flags <fastmath>`, which are optimization hints to enable otherwise
3614 unsafe floating point optimizations:
3619 .. code-block:: llvm
3621 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3622 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3624 '``mul``' Instruction
3625 ^^^^^^^^^^^^^^^^^^^^^
3632 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3633 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3634 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3635 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3640 The '``mul``' instruction returns the product of its two operands.
3645 The two arguments to the '``mul``' instruction must be
3646 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3647 arguments must have identical types.
3652 The value produced is the integer product of the two operands.
3654 If the result of the multiplication has unsigned overflow, the result
3655 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3656 bit width of the result.
3658 Because LLVM integers use a two's complement representation, and the
3659 result is the same width as the operands, this instruction returns the
3660 correct result for both signed and unsigned integers. If a full product
3661 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3662 sign-extended or zero-extended as appropriate to the width of the full
3665 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3666 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3667 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3668 unsigned and/or signed overflow, respectively, occurs.
3673 .. code-block:: llvm
3675 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3679 '``fmul``' Instruction
3680 ^^^^^^^^^^^^^^^^^^^^^^
3687 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3692 The '``fmul``' instruction returns the product of its two operands.
3697 The two arguments to the '``fmul``' instruction must be :ref:`floating
3698 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3699 Both arguments must have identical types.
3704 The value produced is the floating point product of the two operands.
3705 This instruction can also take any number of :ref:`fast-math
3706 flags <fastmath>`, which are optimization hints to enable otherwise
3707 unsafe floating point optimizations:
3712 .. code-block:: llvm
3714 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3716 '``udiv``' Instruction
3717 ^^^^^^^^^^^^^^^^^^^^^^
3724 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3725 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3730 The '``udiv``' instruction returns the quotient of its two operands.
3735 The two arguments to the '``udiv``' instruction must be
3736 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3737 arguments must have identical types.
3742 The value produced is the unsigned integer quotient of the two operands.
3744 Note that unsigned integer division and signed integer division are
3745 distinct operations; for signed integer division, use '``sdiv``'.
3747 Division by zero leads to undefined behavior.
3749 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3750 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3751 such, "((a udiv exact b) mul b) == a").
3756 .. code-block:: llvm
3758 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3760 '``sdiv``' Instruction
3761 ^^^^^^^^^^^^^^^^^^^^^^
3768 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3769 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3774 The '``sdiv``' instruction returns the quotient of its two operands.
3779 The two arguments to the '``sdiv``' instruction must be
3780 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3781 arguments must have identical types.
3786 The value produced is the signed integer quotient of the two operands
3787 rounded towards zero.
3789 Note that signed integer division and unsigned integer division are
3790 distinct operations; for unsigned integer division, use '``udiv``'.
3792 Division by zero leads to undefined behavior. Overflow also leads to
3793 undefined behavior; this is a rare case, but can occur, for example, by
3794 doing a 32-bit division of -2147483648 by -1.
3796 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3797 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3802 .. code-block:: llvm
3804 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3808 '``fdiv``' Instruction
3809 ^^^^^^^^^^^^^^^^^^^^^^
3816 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3821 The '``fdiv``' instruction returns the quotient of its two operands.
3826 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3827 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3828 Both arguments must have identical types.
3833 The value produced is the floating point quotient of the two operands.
3834 This instruction can also take any number of :ref:`fast-math
3835 flags <fastmath>`, which are optimization hints to enable otherwise
3836 unsafe floating point optimizations:
3841 .. code-block:: llvm
3843 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3845 '``urem``' Instruction
3846 ^^^^^^^^^^^^^^^^^^^^^^
3853 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3858 The '``urem``' instruction returns the remainder from the unsigned
3859 division of its two arguments.
3864 The two arguments to the '``urem``' instruction must be
3865 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3866 arguments must have identical types.
3871 This instruction returns the unsigned integer *remainder* of a division.
3872 This instruction always performs an unsigned division to get the
3875 Note that unsigned integer remainder and signed integer remainder are
3876 distinct operations; for signed integer remainder, use '``srem``'.
3878 Taking the remainder of a division by zero leads to undefined behavior.
3883 .. code-block:: llvm
3885 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3887 '``srem``' Instruction
3888 ^^^^^^^^^^^^^^^^^^^^^^
3895 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3900 The '``srem``' instruction returns the remainder from the signed
3901 division of its two operands. This instruction can also take
3902 :ref:`vector <t_vector>` versions of the values in which case the elements
3908 The two arguments to the '``srem``' instruction must be
3909 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3910 arguments must have identical types.
3915 This instruction returns the *remainder* of a division (where the result
3916 is either zero or has the same sign as the dividend, ``op1``), not the
3917 *modulo* operator (where the result is either zero or has the same sign
3918 as the divisor, ``op2``) of a value. For more information about the
3919 difference, see `The Math
3920 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3921 table of how this is implemented in various languages, please see
3923 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3925 Note that signed integer remainder and unsigned integer remainder are
3926 distinct operations; for unsigned integer remainder, use '``urem``'.
3928 Taking the remainder of a division by zero leads to undefined behavior.
3929 Overflow also leads to undefined behavior; this is a rare case, but can
3930 occur, for example, by taking the remainder of a 32-bit division of
3931 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3932 rule lets srem be implemented using instructions that return both the
3933 result of the division and the remainder.)
3938 .. code-block:: llvm
3940 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
3944 '``frem``' Instruction
3945 ^^^^^^^^^^^^^^^^^^^^^^
3952 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3957 The '``frem``' instruction returns the remainder from the division of
3963 The two arguments to the '``frem``' instruction must be :ref:`floating
3964 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3965 Both arguments must have identical types.
3970 This instruction returns the *remainder* of a division. The remainder
3971 has the same sign as the dividend. This instruction can also take any
3972 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
3973 to enable otherwise unsafe floating point optimizations:
3978 .. code-block:: llvm
3980 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
3984 Bitwise Binary Operations
3985 -------------------------
3987 Bitwise binary operators are used to do various forms of bit-twiddling
3988 in a program. They are generally very efficient instructions and can
3989 commonly be strength reduced from other instructions. They require two
3990 operands of the same type, execute an operation on them, and produce a
3991 single value. The resulting value is the same type as its operands.
3993 '``shl``' Instruction
3994 ^^^^^^^^^^^^^^^^^^^^^
4001 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4002 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4003 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4004 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4009 The '``shl``' instruction returns the first operand shifted to the left
4010 a specified number of bits.
4015 Both arguments to the '``shl``' instruction must be the same
4016 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4017 '``op2``' is treated as an unsigned value.
4022 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4023 where ``n`` is the width of the result. If ``op2`` is (statically or
4024 dynamically) negative or equal to or larger than the number of bits in
4025 ``op1``, the result is undefined. If the arguments are vectors, each
4026 vector element of ``op1`` is shifted by the corresponding shift amount
4029 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4030 value <poisonvalues>` if it shifts out any non-zero bits. If the
4031 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4032 value <poisonvalues>` if it shifts out any bits that disagree with the
4033 resultant sign bit. As such, NUW/NSW have the same semantics as they
4034 would if the shift were expressed as a mul instruction with the same
4035 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4040 .. code-block:: llvm
4042 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4043 <result> = shl i32 4, 2 ; yields {i32}: 16
4044 <result> = shl i32 1, 10 ; yields {i32}: 1024
4045 <result> = shl i32 1, 32 ; undefined
4046 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4048 '``lshr``' Instruction
4049 ^^^^^^^^^^^^^^^^^^^^^^
4056 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4057 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4062 The '``lshr``' instruction (logical shift right) returns the first
4063 operand shifted to the right a specified number of bits with zero fill.
4068 Both arguments to the '``lshr``' instruction must be the same
4069 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4070 '``op2``' is treated as an unsigned value.
4075 This instruction always performs a logical shift right operation. The
4076 most significant bits of the result will be filled with zero bits after
4077 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4078 than the number of bits in ``op1``, the result is undefined. If the
4079 arguments are vectors, each vector element of ``op1`` is shifted by the
4080 corresponding shift amount in ``op2``.
4082 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4083 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4089 .. code-block:: llvm
4091 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4092 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4093 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4094 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4095 <result> = lshr i32 1, 32 ; undefined
4096 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4098 '``ashr``' Instruction
4099 ^^^^^^^^^^^^^^^^^^^^^^
4106 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4107 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4112 The '``ashr``' instruction (arithmetic shift right) returns the first
4113 operand shifted to the right a specified number of bits with sign
4119 Both arguments to the '``ashr``' instruction must be the same
4120 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4121 '``op2``' is treated as an unsigned value.
4126 This instruction always performs an arithmetic shift right operation,
4127 The most significant bits of the result will be filled with the sign bit
4128 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4129 than the number of bits in ``op1``, the result is undefined. If the
4130 arguments are vectors, each vector element of ``op1`` is shifted by the
4131 corresponding shift amount in ``op2``.
4133 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4134 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4140 .. code-block:: llvm
4142 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4143 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4144 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4145 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4146 <result> = ashr i32 1, 32 ; undefined
4147 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4149 '``and``' Instruction
4150 ^^^^^^^^^^^^^^^^^^^^^
4157 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4162 The '``and``' instruction returns the bitwise logical and of its two
4168 The two arguments to the '``and``' instruction must be
4169 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4170 arguments must have identical types.
4175 The truth table used for the '``and``' instruction is:
4192 .. code-block:: llvm
4194 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4195 <result> = and i32 15, 40 ; yields {i32}:result = 8
4196 <result> = and i32 4, 8 ; yields {i32}:result = 0
4198 '``or``' Instruction
4199 ^^^^^^^^^^^^^^^^^^^^
4206 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4211 The '``or``' instruction returns the bitwise logical inclusive or of its
4217 The two arguments to the '``or``' instruction must be
4218 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4219 arguments must have identical types.
4224 The truth table used for the '``or``' instruction is:
4243 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4244 <result> = or i32 15, 40 ; yields {i32}:result = 47
4245 <result> = or i32 4, 8 ; yields {i32}:result = 12
4247 '``xor``' Instruction
4248 ^^^^^^^^^^^^^^^^^^^^^
4255 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4260 The '``xor``' instruction returns the bitwise logical exclusive or of
4261 its two operands. The ``xor`` is used to implement the "one's
4262 complement" operation, which is the "~" operator in C.
4267 The two arguments to the '``xor``' instruction must be
4268 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4269 arguments must have identical types.
4274 The truth table used for the '``xor``' instruction is:
4291 .. code-block:: llvm
4293 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4294 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4295 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4296 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4301 LLVM supports several instructions to represent vector operations in a
4302 target-independent manner. These instructions cover the element-access
4303 and vector-specific operations needed to process vectors effectively.
4304 While LLVM does directly support these vector operations, many
4305 sophisticated algorithms will want to use target-specific intrinsics to
4306 take full advantage of a specific target.
4308 .. _i_extractelement:
4310 '``extractelement``' Instruction
4311 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4318 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4323 The '``extractelement``' instruction extracts a single scalar element
4324 from a vector at a specified index.
4329 The first operand of an '``extractelement``' instruction is a value of
4330 :ref:`vector <t_vector>` type. The second operand is an index indicating
4331 the position from which to extract the element. The index may be a
4337 The result is a scalar of the same type as the element type of ``val``.
4338 Its value is the value at position ``idx`` of ``val``. If ``idx``
4339 exceeds the length of ``val``, the results are undefined.
4344 .. code-block:: llvm
4346 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4348 .. _i_insertelement:
4350 '``insertelement``' Instruction
4351 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4358 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4363 The '``insertelement``' instruction inserts a scalar element into a
4364 vector at a specified index.
4369 The first operand of an '``insertelement``' instruction is a value of
4370 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4371 type must equal the element type of the first operand. The third operand
4372 is an index indicating the position at which to insert the value. The
4373 index may be a variable.
4378 The result is a vector of the same type as ``val``. Its element values
4379 are those of ``val`` except at position ``idx``, where it gets the value
4380 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4386 .. code-block:: llvm
4388 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4390 .. _i_shufflevector:
4392 '``shufflevector``' Instruction
4393 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4400 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4405 The '``shufflevector``' instruction constructs a permutation of elements
4406 from two input vectors, returning a vector with the same element type as
4407 the input and length that is the same as the shuffle mask.
4412 The first two operands of a '``shufflevector``' instruction are vectors
4413 with the same type. The third argument is a shuffle mask whose element
4414 type is always 'i32'. The result of the instruction is a vector whose
4415 length is the same as the shuffle mask and whose element type is the
4416 same as the element type of the first two operands.
4418 The shuffle mask operand is required to be a constant vector with either
4419 constant integer or undef values.
4424 The elements of the two input vectors are numbered from left to right
4425 across both of the vectors. The shuffle mask operand specifies, for each
4426 element of the result vector, which element of the two input vectors the
4427 result element gets. The element selector may be undef (meaning "don't
4428 care") and the second operand may be undef if performing a shuffle from
4434 .. code-block:: llvm
4436 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4437 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4438 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4439 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4440 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4441 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4442 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4443 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4445 Aggregate Operations
4446 --------------------
4448 LLVM supports several instructions for working with
4449 :ref:`aggregate <t_aggregate>` values.
4453 '``extractvalue``' Instruction
4454 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4461 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4466 The '``extractvalue``' instruction extracts the value of a member field
4467 from an :ref:`aggregate <t_aggregate>` value.
4472 The first operand of an '``extractvalue``' instruction is a value of
4473 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4474 constant indices to specify which value to extract in a similar manner
4475 as indices in a '``getelementptr``' instruction.
4477 The major differences to ``getelementptr`` indexing are:
4479 - Since the value being indexed is not a pointer, the first index is
4480 omitted and assumed to be zero.
4481 - At least one index must be specified.
4482 - Not only struct indices but also array indices must be in bounds.
4487 The result is the value at the position in the aggregate specified by
4493 .. code-block:: llvm
4495 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4499 '``insertvalue``' Instruction
4500 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4507 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4512 The '``insertvalue``' instruction inserts a value into a member field in
4513 an :ref:`aggregate <t_aggregate>` value.
4518 The first operand of an '``insertvalue``' instruction is a value of
4519 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4520 a first-class value to insert. The following operands are constant
4521 indices indicating the position at which to insert the value in a
4522 similar manner as indices in a '``extractvalue``' instruction. The value
4523 to insert must have the same type as the value identified by the
4529 The result is an aggregate of the same type as ``val``. Its value is
4530 that of ``val`` except that the value at the position specified by the
4531 indices is that of ``elt``.
4536 .. code-block:: llvm
4538 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4539 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4540 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4544 Memory Access and Addressing Operations
4545 ---------------------------------------
4547 A key design point of an SSA-based representation is how it represents
4548 memory. In LLVM, no memory locations are in SSA form, which makes things
4549 very simple. This section describes how to read, write, and allocate
4554 '``alloca``' Instruction
4555 ^^^^^^^^^^^^^^^^^^^^^^^^
4562 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4567 The '``alloca``' instruction allocates memory on the stack frame of the
4568 currently executing function, to be automatically released when this
4569 function returns to its caller. The object is always allocated in the
4570 generic address space (address space zero).
4575 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4576 bytes of memory on the runtime stack, returning a pointer of the
4577 appropriate type to the program. If "NumElements" is specified, it is
4578 the number of elements allocated, otherwise "NumElements" is defaulted
4579 to be one. If a constant alignment is specified, the value result of the
4580 allocation is guaranteed to be aligned to at least that boundary. If not
4581 specified, or if zero, the target can choose to align the allocation on
4582 any convenient boundary compatible with the type.
4584 '``type``' may be any sized type.
4589 Memory is allocated; a pointer is returned. The operation is undefined
4590 if there is insufficient stack space for the allocation. '``alloca``'d
4591 memory is automatically released when the function returns. The
4592 '``alloca``' instruction is commonly used to represent automatic
4593 variables that must have an address available. When the function returns
4594 (either with the ``ret`` or ``resume`` instructions), the memory is
4595 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4596 The order in which memory is allocated (ie., which way the stack grows)
4602 .. code-block:: llvm
4604 %ptr = alloca i32 ; yields {i32*}:ptr
4605 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4606 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4607 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4611 '``load``' Instruction
4612 ^^^^^^^^^^^^^^^^^^^^^^
4619 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4620 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4621 !<index> = !{ i32 1 }
4626 The '``load``' instruction is used to read from memory.
4631 The argument to the ``load`` instruction specifies the memory address
4632 from which to load. The pointer must point to a :ref:`first
4633 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4634 then the optimizer is not allowed to modify the number or order of
4635 execution of this ``load`` with other :ref:`volatile
4636 operations <volatile>`.
4638 If the ``load`` is marked as ``atomic``, it takes an extra
4639 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4640 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4641 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4642 when they may see multiple atomic stores. The type of the pointee must
4643 be an integer type whose bit width is a power of two greater than or
4644 equal to eight and less than or equal to a target-specific size limit.
4645 ``align`` must be explicitly specified on atomic loads, and the load has
4646 undefined behavior if the alignment is not set to a value which is at
4647 least the size in bytes of the pointee. ``!nontemporal`` does not have
4648 any defined semantics for atomic loads.
4650 The optional constant ``align`` argument specifies the alignment of the
4651 operation (that is, the alignment of the memory address). A value of 0
4652 or an omitted ``align`` argument means that the operation has the ABI
4653 alignment for the target. It is the responsibility of the code emitter
4654 to ensure that the alignment information is correct. Overestimating the
4655 alignment results in undefined behavior. Underestimating the alignment
4656 may produce less efficient code. An alignment of 1 is always safe.
4658 The optional ``!nontemporal`` metadata must reference a single
4659 metadata name ``<index>`` corresponding to a metadata node with one
4660 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4661 metadata on the instruction tells the optimizer and code generator
4662 that this load is not expected to be reused in the cache. The code
4663 generator may select special instructions to save cache bandwidth, such
4664 as the ``MOVNT`` instruction on x86.
4666 The optional ``!invariant.load`` metadata must reference a single
4667 metadata name ``<index>`` corresponding to a metadata node with no
4668 entries. The existence of the ``!invariant.load`` metadata on the
4669 instruction tells the optimizer and code generator that this load
4670 address points to memory which does not change value during program
4671 execution. The optimizer may then move this load around, for example, by
4672 hoisting it out of loops using loop invariant code motion.
4677 The location of memory pointed to is loaded. If the value being loaded
4678 is of scalar type then the number of bytes read does not exceed the
4679 minimum number of bytes needed to hold all bits of the type. For
4680 example, loading an ``i24`` reads at most three bytes. When loading a
4681 value of a type like ``i20`` with a size that is not an integral number
4682 of bytes, the result is undefined if the value was not originally
4683 written using a store of the same type.
4688 .. code-block:: llvm
4690 %ptr = alloca i32 ; yields {i32*}:ptr
4691 store i32 3, i32* %ptr ; yields {void}
4692 %val = load i32* %ptr ; yields {i32}:val = i32 3
4696 '``store``' Instruction
4697 ^^^^^^^^^^^^^^^^^^^^^^^
4704 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4705 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4710 The '``store``' instruction is used to write to memory.
4715 There are two arguments to the ``store`` instruction: a value to store
4716 and an address at which to store it. The type of the ``<pointer>``
4717 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4718 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4719 then the optimizer is not allowed to modify the number or order of
4720 execution of this ``store`` with other :ref:`volatile
4721 operations <volatile>`.
4723 If the ``store`` is marked as ``atomic``, it takes an extra
4724 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4725 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4726 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4727 when they may see multiple atomic stores. The type of the pointee must
4728 be an integer type whose bit width is a power of two greater than or
4729 equal to eight and less than or equal to a target-specific size limit.
4730 ``align`` must be explicitly specified on atomic stores, and the store
4731 has undefined behavior if the alignment is not set to a value which is
4732 at least the size in bytes of the pointee. ``!nontemporal`` does not
4733 have any defined semantics for atomic stores.
4735 The optional constant ``align`` argument specifies the alignment of the
4736 operation (that is, the alignment of the memory address). A value of 0
4737 or an omitted ``align`` argument means that the operation has the ABI
4738 alignment for the target. It is the responsibility of the code emitter
4739 to ensure that the alignment information is correct. Overestimating the
4740 alignment results in undefined behavior. Underestimating the
4741 alignment may produce less efficient code. An alignment of 1 is always
4744 The optional ``!nontemporal`` metadata must reference a single metadata
4745 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4746 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4747 tells the optimizer and code generator that this load is not expected to
4748 be reused in the cache. The code generator may select special
4749 instructions to save cache bandwidth, such as the MOVNT instruction on
4755 The contents of memory are updated to contain ``<value>`` at the
4756 location specified by the ``<pointer>`` operand. If ``<value>`` is
4757 of scalar type then the number of bytes written does not exceed the
4758 minimum number of bytes needed to hold all bits of the type. For
4759 example, storing an ``i24`` writes at most three bytes. When writing a
4760 value of a type like ``i20`` with a size that is not an integral number
4761 of bytes, it is unspecified what happens to the extra bits that do not
4762 belong to the type, but they will typically be overwritten.
4767 .. code-block:: llvm
4769 %ptr = alloca i32 ; yields {i32*}:ptr
4770 store i32 3, i32* %ptr ; yields {void}
4771 %val = load i32* %ptr ; yields {i32}:val = i32 3
4775 '``fence``' Instruction
4776 ^^^^^^^^^^^^^^^^^^^^^^^
4783 fence [singlethread] <ordering> ; yields {void}
4788 The '``fence``' instruction is used to introduce happens-before edges
4794 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4795 defines what *synchronizes-with* edges they add. They can only be given
4796 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4801 A fence A which has (at least) ``release`` ordering semantics
4802 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4803 semantics if and only if there exist atomic operations X and Y, both
4804 operating on some atomic object M, such that A is sequenced before X, X
4805 modifies M (either directly or through some side effect of a sequence
4806 headed by X), Y is sequenced before B, and Y observes M. This provides a
4807 *happens-before* dependency between A and B. Rather than an explicit
4808 ``fence``, one (but not both) of the atomic operations X or Y might
4809 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4810 still *synchronize-with* the explicit ``fence`` and establish the
4811 *happens-before* edge.
4813 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4814 ``acquire`` and ``release`` semantics specified above, participates in
4815 the global program order of other ``seq_cst`` operations and/or fences.
4817 The optional ":ref:`singlethread <singlethread>`" argument specifies
4818 that the fence only synchronizes with other fences in the same thread.
4819 (This is useful for interacting with signal handlers.)
4824 .. code-block:: llvm
4826 fence acquire ; yields {void}
4827 fence singlethread seq_cst ; yields {void}
4831 '``cmpxchg``' Instruction
4832 ^^^^^^^^^^^^^^^^^^^^^^^^^
4839 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4844 The '``cmpxchg``' instruction is used to atomically modify memory. It
4845 loads a value in memory and compares it to a given value. If they are
4846 equal, it stores a new value into the memory.
4851 There are three arguments to the '``cmpxchg``' instruction: an address
4852 to operate on, a value to compare to the value currently be at that
4853 address, and a new value to place at that address if the compared values
4854 are equal. The type of '<cmp>' must be an integer type whose bit width
4855 is a power of two greater than or equal to eight and less than or equal
4856 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4857 type, and the type of '<pointer>' must be a pointer to that type. If the
4858 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4859 to modify the number or order of execution of this ``cmpxchg`` with
4860 other :ref:`volatile operations <volatile>`.
4862 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4863 synchronizes with other atomic operations.
4865 The optional "``singlethread``" argument declares that the ``cmpxchg``
4866 is only atomic with respect to code (usually signal handlers) running in
4867 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4868 respect to all other code in the system.
4870 The pointer passed into cmpxchg must have alignment greater than or
4871 equal to the size in memory of the operand.
4876 The contents of memory at the location specified by the '``<pointer>``'
4877 operand is read and compared to '``<cmp>``'; if the read value is the
4878 equal, '``<new>``' is written. The original value at the location is
4881 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4882 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4883 atomic load with an ordering parameter determined by dropping any
4884 ``release`` part of the ``cmpxchg``'s ordering.
4889 .. code-block:: llvm
4892 %orig = atomic load i32* %ptr unordered ; yields {i32}
4896 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4897 %squared = mul i32 %cmp, %cmp
4898 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4899 %success = icmp eq i32 %cmp, %old
4900 br i1 %success, label %done, label %loop
4907 '``atomicrmw``' Instruction
4908 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4915 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4920 The '``atomicrmw``' instruction is used to atomically modify memory.
4925 There are three arguments to the '``atomicrmw``' instruction: an
4926 operation to apply, an address whose value to modify, an argument to the
4927 operation. The operation must be one of the following keywords:
4941 The type of '<value>' must be an integer type whose bit width is a power
4942 of two greater than or equal to eight and less than or equal to a
4943 target-specific size limit. The type of the '``<pointer>``' operand must
4944 be a pointer to that type. If the ``atomicrmw`` is marked as
4945 ``volatile``, then the optimizer is not allowed to modify the number or
4946 order of execution of this ``atomicrmw`` with other :ref:`volatile
4947 operations <volatile>`.
4952 The contents of memory at the location specified by the '``<pointer>``'
4953 operand are atomically read, modified, and written back. The original
4954 value at the location is returned. The modification is specified by the
4957 - xchg: ``*ptr = val``
4958 - add: ``*ptr = *ptr + val``
4959 - sub: ``*ptr = *ptr - val``
4960 - and: ``*ptr = *ptr & val``
4961 - nand: ``*ptr = ~(*ptr & val)``
4962 - or: ``*ptr = *ptr | val``
4963 - xor: ``*ptr = *ptr ^ val``
4964 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
4965 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
4966 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
4968 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
4974 .. code-block:: llvm
4976 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
4978 .. _i_getelementptr:
4980 '``getelementptr``' Instruction
4981 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4988 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4989 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4990 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
4995 The '``getelementptr``' instruction is used to get the address of a
4996 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
4997 address calculation only and does not access memory.
5002 The first argument is always a pointer or a vector of pointers, and
5003 forms the basis of the calculation. The remaining arguments are indices
5004 that indicate which of the elements of the aggregate object are indexed.
5005 The interpretation of each index is dependent on the type being indexed
5006 into. The first index always indexes the pointer value given as the
5007 first argument, the second index indexes a value of the type pointed to
5008 (not necessarily the value directly pointed to, since the first index
5009 can be non-zero), etc. The first type indexed into must be a pointer
5010 value, subsequent types can be arrays, vectors, and structs. Note that
5011 subsequent types being indexed into can never be pointers, since that
5012 would require loading the pointer before continuing calculation.
5014 The type of each index argument depends on the type it is indexing into.
5015 When indexing into a (optionally packed) structure, only ``i32`` integer
5016 **constants** are allowed (when using a vector of indices they must all
5017 be the **same** ``i32`` integer constant). When indexing into an array,
5018 pointer or vector, integers of any width are allowed, and they are not
5019 required to be constant. These integers are treated as signed values
5022 For example, let's consider a C code fragment and how it gets compiled
5038 int *foo(struct ST *s) {
5039 return &s[1].Z.B[5][13];
5042 The LLVM code generated by Clang is:
5044 .. code-block:: llvm
5046 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5047 %struct.ST = type { i32, double, %struct.RT }
5049 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5051 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5058 In the example above, the first index is indexing into the
5059 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5060 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5061 indexes into the third element of the structure, yielding a
5062 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5063 structure. The third index indexes into the second element of the
5064 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5065 dimensions of the array are subscripted into, yielding an '``i32``'
5066 type. The '``getelementptr``' instruction returns a pointer to this
5067 element, thus computing a value of '``i32*``' type.
5069 Note that it is perfectly legal to index partially through a structure,
5070 returning a pointer to an inner element. Because of this, the LLVM code
5071 for the given testcase is equivalent to:
5073 .. code-block:: llvm
5075 define i32* @foo(%struct.ST* %s) {
5076 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5077 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5078 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5079 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5080 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5084 If the ``inbounds`` keyword is present, the result value of the
5085 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5086 pointer is not an *in bounds* address of an allocated object, or if any
5087 of the addresses that would be formed by successive addition of the
5088 offsets implied by the indices to the base address with infinitely
5089 precise signed arithmetic are not an *in bounds* address of that
5090 allocated object. The *in bounds* addresses for an allocated object are
5091 all the addresses that point into the object, plus the address one byte
5092 past the end. In cases where the base is a vector of pointers the
5093 ``inbounds`` keyword applies to each of the computations element-wise.
5095 If the ``inbounds`` keyword is not present, the offsets are added to the
5096 base address with silently-wrapping two's complement arithmetic. If the
5097 offsets have a different width from the pointer, they are sign-extended
5098 or truncated to the width of the pointer. The result value of the
5099 ``getelementptr`` may be outside the object pointed to by the base
5100 pointer. The result value may not necessarily be used to access memory
5101 though, even if it happens to point into allocated storage. See the
5102 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5105 The getelementptr instruction is often confusing. For some more insight
5106 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5111 .. code-block:: llvm
5113 ; yields [12 x i8]*:aptr
5114 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5116 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5118 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5120 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5122 In cases where the pointer argument is a vector of pointers, each index
5123 must be a vector with the same number of elements. For example:
5125 .. code-block:: llvm
5127 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5129 Conversion Operations
5130 ---------------------
5132 The instructions in this category are the conversion instructions
5133 (casting) which all take a single operand and a type. They perform
5134 various bit conversions on the operand.
5136 '``trunc .. to``' Instruction
5137 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5144 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5149 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5154 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5155 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5156 of the same number of integers. The bit size of the ``value`` must be
5157 larger than the bit size of the destination type, ``ty2``. Equal sized
5158 types are not allowed.
5163 The '``trunc``' instruction truncates the high order bits in ``value``
5164 and converts the remaining bits to ``ty2``. Since the source size must
5165 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5166 It will always truncate bits.
5171 .. code-block:: llvm
5173 %X = trunc i32 257 to i8 ; yields i8:1
5174 %Y = trunc i32 123 to i1 ; yields i1:true
5175 %Z = trunc i32 122 to i1 ; yields i1:false
5176 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5178 '``zext .. to``' Instruction
5179 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5186 <result> = zext <ty> <value> to <ty2> ; yields ty2
5191 The '``zext``' instruction zero extends its operand to type ``ty2``.
5196 The '``zext``' instruction takes a value to cast, and a type to cast it
5197 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5198 the same number of integers. The bit size of the ``value`` must be
5199 smaller than the bit size of the destination type, ``ty2``.
5204 The ``zext`` fills the high order bits of the ``value`` with zero bits
5205 until it reaches the size of the destination type, ``ty2``.
5207 When zero extending from i1, the result will always be either 0 or 1.
5212 .. code-block:: llvm
5214 %X = zext i32 257 to i64 ; yields i64:257
5215 %Y = zext i1 true to i32 ; yields i32:1
5216 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5218 '``sext .. to``' Instruction
5219 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5226 <result> = sext <ty> <value> to <ty2> ; yields ty2
5231 The '``sext``' sign extends ``value`` to the type ``ty2``.
5236 The '``sext``' instruction takes a value to cast, and a type to cast it
5237 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5238 the same number of integers. The bit size of the ``value`` must be
5239 smaller than the bit size of the destination type, ``ty2``.
5244 The '``sext``' instruction performs a sign extension by copying the sign
5245 bit (highest order bit) of the ``value`` until it reaches the bit size
5246 of the type ``ty2``.
5248 When sign extending from i1, the extension always results in -1 or 0.
5253 .. code-block:: llvm
5255 %X = sext i8 -1 to i16 ; yields i16 :65535
5256 %Y = sext i1 true to i32 ; yields i32:-1
5257 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5259 '``fptrunc .. to``' Instruction
5260 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5267 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5272 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5277 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5278 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5279 The size of ``value`` must be larger than the size of ``ty2``. This
5280 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5285 The '``fptrunc``' instruction truncates a ``value`` from a larger
5286 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5287 point <t_floating>` type. If the value cannot fit within the
5288 destination type, ``ty2``, then the results are undefined.
5293 .. code-block:: llvm
5295 %X = fptrunc double 123.0 to float ; yields float:123.0
5296 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5298 '``fpext .. to``' Instruction
5299 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5306 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5311 The '``fpext``' extends a floating point ``value`` to a larger floating
5317 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5318 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5319 to. The source type must be smaller than the destination type.
5324 The '``fpext``' instruction extends the ``value`` from a smaller
5325 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5326 point <t_floating>` type. The ``fpext`` cannot be used to make a
5327 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5328 *no-op cast* for a floating point cast.
5333 .. code-block:: llvm
5335 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5336 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5338 '``fptoui .. to``' Instruction
5339 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5346 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5351 The '``fptoui``' converts a floating point ``value`` to its unsigned
5352 integer equivalent of type ``ty2``.
5357 The '``fptoui``' instruction takes a value to cast, which must be a
5358 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5359 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5360 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5361 type with the same number of elements as ``ty``
5366 The '``fptoui``' instruction converts its :ref:`floating
5367 point <t_floating>` operand into the nearest (rounding towards zero)
5368 unsigned integer value. If the value cannot fit in ``ty2``, the results
5374 .. code-block:: llvm
5376 %X = fptoui double 123.0 to i32 ; yields i32:123
5377 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5378 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5380 '``fptosi .. to``' Instruction
5381 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5388 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5393 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5394 ``value`` to type ``ty2``.
5399 The '``fptosi``' instruction takes a value to cast, which must be a
5400 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5401 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5402 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5403 type with the same number of elements as ``ty``
5408 The '``fptosi``' instruction converts its :ref:`floating
5409 point <t_floating>` operand into the nearest (rounding towards zero)
5410 signed integer value. If the value cannot fit in ``ty2``, the results
5416 .. code-block:: llvm
5418 %X = fptosi double -123.0 to i32 ; yields i32:-123
5419 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5420 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5422 '``uitofp .. to``' Instruction
5423 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5430 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5435 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5436 and converts that value to the ``ty2`` type.
5441 The '``uitofp``' instruction takes a value to cast, which must be a
5442 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5443 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5444 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5445 type with the same number of elements as ``ty``
5450 The '``uitofp``' instruction interprets its operand as an unsigned
5451 integer quantity and converts it to the corresponding floating point
5452 value. If the value cannot fit in the floating point value, the results
5458 .. code-block:: llvm
5460 %X = uitofp i32 257 to float ; yields float:257.0
5461 %Y = uitofp i8 -1 to double ; yields double:255.0
5463 '``sitofp .. to``' Instruction
5464 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5471 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5476 The '``sitofp``' instruction regards ``value`` as a signed integer and
5477 converts that value to the ``ty2`` type.
5482 The '``sitofp``' instruction takes a value to cast, which must be a
5483 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5484 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5485 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5486 type with the same number of elements as ``ty``
5491 The '``sitofp``' instruction interprets its operand as a signed integer
5492 quantity and converts it to the corresponding floating point value. If
5493 the value cannot fit in the floating point value, the results are
5499 .. code-block:: llvm
5501 %X = sitofp i32 257 to float ; yields float:257.0
5502 %Y = sitofp i8 -1 to double ; yields double:-1.0
5506 '``ptrtoint .. to``' Instruction
5507 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5514 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5519 The '``ptrtoint``' instruction converts the pointer or a vector of
5520 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5525 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5526 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5527 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5528 a vector of integers type.
5533 The '``ptrtoint``' instruction converts ``value`` to integer type
5534 ``ty2`` by interpreting the pointer value as an integer and either
5535 truncating or zero extending that value to the size of the integer type.
5536 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5537 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5538 the same size, then nothing is done (*no-op cast*) other than a type
5544 .. code-block:: llvm
5546 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5547 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5548 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5552 '``inttoptr .. to``' Instruction
5553 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5560 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5565 The '``inttoptr``' instruction converts an integer ``value`` to a
5566 pointer type, ``ty2``.
5571 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5572 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5578 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5579 applying either a zero extension or a truncation depending on the size
5580 of the integer ``value``. If ``value`` is larger than the size of a
5581 pointer then a truncation is done. If ``value`` is smaller than the size
5582 of a pointer then a zero extension is done. If they are the same size,
5583 nothing is done (*no-op cast*).
5588 .. code-block:: llvm
5590 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5591 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5592 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5593 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5597 '``bitcast .. to``' Instruction
5598 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5605 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5610 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5616 The '``bitcast``' instruction takes a value to cast, which must be a
5617 non-aggregate first class value, and a type to cast it to, which must
5618 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5619 bit sizes of ``value`` and the destination type, ``ty2``, must be
5620 identical. If the source type is a pointer, the destination type must
5621 also be a pointer of the same size. This instruction supports bitwise
5622 conversion of vectors to integers and to vectors of other types (as
5623 long as they have the same size).
5628 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5629 is always a *no-op cast* because no bits change with this
5630 conversion. The conversion is done as if the ``value`` had been stored
5631 to memory and read back as type ``ty2``. Pointer (or vector of
5632 pointers) types may only be converted to other pointer (or vector of
5633 pointers) types with this instruction if the pointer sizes are
5634 equal. To convert pointers to other types, use the :ref:`inttoptr
5635 <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5640 .. code-block:: llvm
5642 %X = bitcast i8 255 to i8 ; yields i8 :-1
5643 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5644 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5645 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5652 The instructions in this category are the "miscellaneous" instructions,
5653 which defy better classification.
5657 '``icmp``' Instruction
5658 ^^^^^^^^^^^^^^^^^^^^^^
5665 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5670 The '``icmp``' instruction returns a boolean value or a vector of
5671 boolean values based on comparison of its two integer, integer vector,
5672 pointer, or pointer vector operands.
5677 The '``icmp``' instruction takes three operands. The first operand is
5678 the condition code indicating the kind of comparison to perform. It is
5679 not a value, just a keyword. The possible condition code are:
5682 #. ``ne``: not equal
5683 #. ``ugt``: unsigned greater than
5684 #. ``uge``: unsigned greater or equal
5685 #. ``ult``: unsigned less than
5686 #. ``ule``: unsigned less or equal
5687 #. ``sgt``: signed greater than
5688 #. ``sge``: signed greater or equal
5689 #. ``slt``: signed less than
5690 #. ``sle``: signed less or equal
5692 The remaining two arguments must be :ref:`integer <t_integer>` or
5693 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5694 must also be identical types.
5699 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5700 code given as ``cond``. The comparison performed always yields either an
5701 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5703 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5704 otherwise. No sign interpretation is necessary or performed.
5705 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5706 otherwise. No sign interpretation is necessary or performed.
5707 #. ``ugt``: interprets the operands as unsigned values and yields
5708 ``true`` if ``op1`` is greater than ``op2``.
5709 #. ``uge``: interprets the operands as unsigned values and yields
5710 ``true`` if ``op1`` is greater than or equal to ``op2``.
5711 #. ``ult``: interprets the operands as unsigned values and yields
5712 ``true`` if ``op1`` is less than ``op2``.
5713 #. ``ule``: interprets the operands as unsigned values and yields
5714 ``true`` if ``op1`` is less than or equal to ``op2``.
5715 #. ``sgt``: interprets the operands as signed values and yields ``true``
5716 if ``op1`` is greater than ``op2``.
5717 #. ``sge``: interprets the operands as signed values and yields ``true``
5718 if ``op1`` is greater than or equal to ``op2``.
5719 #. ``slt``: interprets the operands as signed values and yields ``true``
5720 if ``op1`` is less than ``op2``.
5721 #. ``sle``: interprets the operands as signed values and yields ``true``
5722 if ``op1`` is less than or equal to ``op2``.
5724 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5725 are compared as if they were integers.
5727 If the operands are integer vectors, then they are compared element by
5728 element. The result is an ``i1`` vector with the same number of elements
5729 as the values being compared. Otherwise, the result is an ``i1``.
5734 .. code-block:: llvm
5736 <result> = icmp eq i32 4, 5 ; yields: result=false
5737 <result> = icmp ne float* %X, %X ; yields: result=false
5738 <result> = icmp ult i16 4, 5 ; yields: result=true
5739 <result> = icmp sgt i16 4, 5 ; yields: result=false
5740 <result> = icmp ule i16 -4, 5 ; yields: result=false
5741 <result> = icmp sge i16 4, 5 ; yields: result=false
5743 Note that the code generator does not yet support vector types with the
5744 ``icmp`` instruction.
5748 '``fcmp``' Instruction
5749 ^^^^^^^^^^^^^^^^^^^^^^
5756 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5761 The '``fcmp``' instruction returns a boolean value or vector of boolean
5762 values based on comparison of its operands.
5764 If the operands are floating point scalars, then the result type is a
5765 boolean (:ref:`i1 <t_integer>`).
5767 If the operands are floating point vectors, then the result type is a
5768 vector of boolean with the same number of elements as the operands being
5774 The '``fcmp``' instruction takes three operands. The first operand is
5775 the condition code indicating the kind of comparison to perform. It is
5776 not a value, just a keyword. The possible condition code are:
5778 #. ``false``: no comparison, always returns false
5779 #. ``oeq``: ordered and equal
5780 #. ``ogt``: ordered and greater than
5781 #. ``oge``: ordered and greater than or equal
5782 #. ``olt``: ordered and less than
5783 #. ``ole``: ordered and less than or equal
5784 #. ``one``: ordered and not equal
5785 #. ``ord``: ordered (no nans)
5786 #. ``ueq``: unordered or equal
5787 #. ``ugt``: unordered or greater than
5788 #. ``uge``: unordered or greater than or equal
5789 #. ``ult``: unordered or less than
5790 #. ``ule``: unordered or less than or equal
5791 #. ``une``: unordered or not equal
5792 #. ``uno``: unordered (either nans)
5793 #. ``true``: no comparison, always returns true
5795 *Ordered* means that neither operand is a QNAN while *unordered* means
5796 that either operand may be a QNAN.
5798 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5799 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5800 type. They must have identical types.
5805 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5806 condition code given as ``cond``. If the operands are vectors, then the
5807 vectors are compared element by element. Each comparison performed
5808 always yields an :ref:`i1 <t_integer>` result, as follows:
5810 #. ``false``: always yields ``false``, regardless of operands.
5811 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5812 is equal to ``op2``.
5813 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5814 is greater than ``op2``.
5815 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5816 is greater than or equal to ``op2``.
5817 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5818 is less than ``op2``.
5819 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5820 is less than or equal to ``op2``.
5821 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5822 is not equal to ``op2``.
5823 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5824 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5826 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5827 greater than ``op2``.
5828 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5829 greater than or equal to ``op2``.
5830 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5832 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5833 less than or equal to ``op2``.
5834 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5835 not equal to ``op2``.
5836 #. ``uno``: yields ``true`` if either operand is a QNAN.
5837 #. ``true``: always yields ``true``, regardless of operands.
5842 .. code-block:: llvm
5844 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5845 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5846 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5847 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5849 Note that the code generator does not yet support vector types with the
5850 ``fcmp`` instruction.
5854 '``phi``' Instruction
5855 ^^^^^^^^^^^^^^^^^^^^^
5862 <result> = phi <ty> [ <val0>, <label0>], ...
5867 The '``phi``' instruction is used to implement the φ node in the SSA
5868 graph representing the function.
5873 The type of the incoming values is specified with the first type field.
5874 After this, the '``phi``' instruction takes a list of pairs as
5875 arguments, with one pair for each predecessor basic block of the current
5876 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5877 the value arguments to the PHI node. Only labels may be used as the
5880 There must be no non-phi instructions between the start of a basic block
5881 and the PHI instructions: i.e. PHI instructions must be first in a basic
5884 For the purposes of the SSA form, the use of each incoming value is
5885 deemed to occur on the edge from the corresponding predecessor block to
5886 the current block (but after any definition of an '``invoke``'
5887 instruction's return value on the same edge).
5892 At runtime, the '``phi``' instruction logically takes on the value
5893 specified by the pair corresponding to the predecessor basic block that
5894 executed just prior to the current block.
5899 .. code-block:: llvm
5901 Loop: ; Infinite loop that counts from 0 on up...
5902 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5903 %nextindvar = add i32 %indvar, 1
5908 '``select``' Instruction
5909 ^^^^^^^^^^^^^^^^^^^^^^^^
5916 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5918 selty is either i1 or {<N x i1>}
5923 The '``select``' instruction is used to choose one value based on a
5924 condition, without branching.
5929 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
5930 values indicating the condition, and two values of the same :ref:`first
5931 class <t_firstclass>` type. If the val1/val2 are vectors and the
5932 condition is a scalar, then entire vectors are selected, not individual
5938 If the condition is an i1 and it evaluates to 1, the instruction returns
5939 the first value argument; otherwise, it returns the second value
5942 If the condition is a vector of i1, then the value arguments must be
5943 vectors of the same size, and the selection is done element by element.
5948 .. code-block:: llvm
5950 %X = select i1 true, i8 17, i8 42 ; yields i8:17
5954 '``call``' Instruction
5955 ^^^^^^^^^^^^^^^^^^^^^^
5962 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
5967 The '``call``' instruction represents a simple function call.
5972 This instruction requires several arguments:
5974 #. The optional "tail" marker indicates that the callee function does
5975 not access any allocas or varargs in the caller. Note that calls may
5976 be marked "tail" even if they do not occur before a
5977 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
5978 function call is eligible for tail call optimization, but `might not
5979 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
5980 The code generator may optimize calls marked "tail" with either 1)
5981 automatic `sibling call
5982 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
5983 callee have matching signatures, or 2) forced tail call optimization
5984 when the following extra requirements are met:
5986 - Caller and callee both have the calling convention ``fastcc``.
5987 - The call is in tail position (ret immediately follows call and ret
5988 uses value of call or is void).
5989 - Option ``-tailcallopt`` is enabled, or
5990 ``llvm::GuaranteedTailCallOpt`` is ``true``.
5991 - `Platform specific constraints are
5992 met. <CodeGenerator.html#tailcallopt>`_
5994 #. The optional "cconv" marker indicates which :ref:`calling
5995 convention <callingconv>` the call should use. If none is
5996 specified, the call defaults to using C calling conventions. The
5997 calling convention of the call must match the calling convention of
5998 the target function, or else the behavior is undefined.
5999 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6000 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6002 #. '``ty``': the type of the call instruction itself which is also the
6003 type of the return value. Functions that return no value are marked
6005 #. '``fnty``': shall be the signature of the pointer to function value
6006 being invoked. The argument types must match the types implied by
6007 this signature. This type can be omitted if the function is not
6008 varargs and if the function type does not return a pointer to a
6010 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6011 be invoked. In most cases, this is a direct function invocation, but
6012 indirect ``call``'s are just as possible, calling an arbitrary pointer
6014 #. '``function args``': argument list whose types match the function
6015 signature argument types and parameter attributes. All arguments must
6016 be of :ref:`first class <t_firstclass>` type. If the function signature
6017 indicates the function accepts a variable number of arguments, the
6018 extra arguments can be specified.
6019 #. The optional :ref:`function attributes <fnattrs>` list. Only
6020 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6021 attributes are valid here.
6026 The '``call``' instruction is used to cause control flow to transfer to
6027 a specified function, with its incoming arguments bound to the specified
6028 values. Upon a '``ret``' instruction in the called function, control
6029 flow continues with the instruction after the function call, and the
6030 return value of the function is bound to the result argument.
6035 .. code-block:: llvm
6037 %retval = call i32 @test(i32 %argc)
6038 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6039 %X = tail call i32 @foo() ; yields i32
6040 %Y = tail call fastcc i32 @foo() ; yields i32
6041 call void %foo(i8 97 signext)
6043 %struct.A = type { i32, i8 }
6044 %r = call %struct.A @foo() ; yields { 32, i8 }
6045 %gr = extractvalue %struct.A %r, 0 ; yields i32
6046 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6047 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6048 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6050 llvm treats calls to some functions with names and arguments that match
6051 the standard C99 library as being the C99 library functions, and may
6052 perform optimizations or generate code for them under that assumption.
6053 This is something we'd like to change in the future to provide better
6054 support for freestanding environments and non-C-based languages.
6058 '``va_arg``' Instruction
6059 ^^^^^^^^^^^^^^^^^^^^^^^^
6066 <resultval> = va_arg <va_list*> <arglist>, <argty>
6071 The '``va_arg``' instruction is used to access arguments passed through
6072 the "variable argument" area of a function call. It is used to implement
6073 the ``va_arg`` macro in C.
6078 This instruction takes a ``va_list*`` value and the type of the
6079 argument. It returns a value of the specified argument type and
6080 increments the ``va_list`` to point to the next argument. The actual
6081 type of ``va_list`` is target specific.
6086 The '``va_arg``' instruction loads an argument of the specified type
6087 from the specified ``va_list`` and causes the ``va_list`` to point to
6088 the next argument. For more information, see the variable argument
6089 handling :ref:`Intrinsic Functions <int_varargs>`.
6091 It is legal for this instruction to be called in a function which does
6092 not take a variable number of arguments, for example, the ``vfprintf``
6095 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6096 function <intrinsics>` because it takes a type as an argument.
6101 See the :ref:`variable argument processing <int_varargs>` section.
6103 Note that the code generator does not yet fully support va\_arg on many
6104 targets. Also, it does not currently support va\_arg with aggregate
6105 types on any target.
6109 '``landingpad``' Instruction
6110 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6117 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6118 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6120 <clause> := catch <type> <value>
6121 <clause> := filter <array constant type> <array constant>
6126 The '``landingpad``' instruction is used by `LLVM's exception handling
6127 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6128 is a landing pad --- one where the exception lands, and corresponds to the
6129 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6130 defines values supplied by the personality function (``pers_fn``) upon
6131 re-entry to the function. The ``resultval`` has the type ``resultty``.
6136 This instruction takes a ``pers_fn`` value. This is the personality
6137 function associated with the unwinding mechanism. The optional
6138 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6140 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6141 contains the global variable representing the "type" that may be caught
6142 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6143 clause takes an array constant as its argument. Use
6144 "``[0 x i8**] undef``" for a filter which cannot throw. The
6145 '``landingpad``' instruction must contain *at least* one ``clause`` or
6146 the ``cleanup`` flag.
6151 The '``landingpad``' instruction defines the values which are set by the
6152 personality function (``pers_fn``) upon re-entry to the function, and
6153 therefore the "result type" of the ``landingpad`` instruction. As with
6154 calling conventions, how the personality function results are
6155 represented in LLVM IR is target specific.
6157 The clauses are applied in order from top to bottom. If two
6158 ``landingpad`` instructions are merged together through inlining, the
6159 clauses from the calling function are appended to the list of clauses.
6160 When the call stack is being unwound due to an exception being thrown,
6161 the exception is compared against each ``clause`` in turn. If it doesn't
6162 match any of the clauses, and the ``cleanup`` flag is not set, then
6163 unwinding continues further up the call stack.
6165 The ``landingpad`` instruction has several restrictions:
6167 - A landing pad block is a basic block which is the unwind destination
6168 of an '``invoke``' instruction.
6169 - A landing pad block must have a '``landingpad``' instruction as its
6170 first non-PHI instruction.
6171 - There can be only one '``landingpad``' instruction within the landing
6173 - A basic block that is not a landing pad block may not include a
6174 '``landingpad``' instruction.
6175 - All '``landingpad``' instructions in a function must have the same
6176 personality function.
6181 .. code-block:: llvm
6183 ;; A landing pad which can catch an integer.
6184 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6186 ;; A landing pad that is a cleanup.
6187 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6189 ;; A landing pad which can catch an integer and can only throw a double.
6190 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6192 filter [1 x i8**] [@_ZTId]
6199 LLVM supports the notion of an "intrinsic function". These functions
6200 have well known names and semantics and are required to follow certain
6201 restrictions. Overall, these intrinsics represent an extension mechanism
6202 for the LLVM language that does not require changing all of the
6203 transformations in LLVM when adding to the language (or the bitcode
6204 reader/writer, the parser, etc...).
6206 Intrinsic function names must all start with an "``llvm.``" prefix. This
6207 prefix is reserved in LLVM for intrinsic names; thus, function names may
6208 not begin with this prefix. Intrinsic functions must always be external
6209 functions: you cannot define the body of intrinsic functions. Intrinsic
6210 functions may only be used in call or invoke instructions: it is illegal
6211 to take the address of an intrinsic function. Additionally, because
6212 intrinsic functions are part of the LLVM language, it is required if any
6213 are added that they be documented here.
6215 Some intrinsic functions can be overloaded, i.e., the intrinsic
6216 represents a family of functions that perform the same operation but on
6217 different data types. Because LLVM can represent over 8 million
6218 different integer types, overloading is used commonly to allow an
6219 intrinsic function to operate on any integer type. One or more of the
6220 argument types or the result type can be overloaded to accept any
6221 integer type. Argument types may also be defined as exactly matching a
6222 previous argument's type or the result type. This allows an intrinsic
6223 function which accepts multiple arguments, but needs all of them to be
6224 of the same type, to only be overloaded with respect to a single
6225 argument or the result.
6227 Overloaded intrinsics will have the names of its overloaded argument
6228 types encoded into its function name, each preceded by a period. Only
6229 those types which are overloaded result in a name suffix. Arguments
6230 whose type is matched against another type do not. For example, the
6231 ``llvm.ctpop`` function can take an integer of any width and returns an
6232 integer of exactly the same integer width. This leads to a family of
6233 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6234 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6235 overloaded, and only one type suffix is required. Because the argument's
6236 type is matched against the return type, it does not require its own
6239 To learn how to add an intrinsic function, please see the `Extending
6240 LLVM Guide <ExtendingLLVM.html>`_.
6244 Variable Argument Handling Intrinsics
6245 -------------------------------------
6247 Variable argument support is defined in LLVM with the
6248 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6249 functions. These functions are related to the similarly named macros
6250 defined in the ``<stdarg.h>`` header file.
6252 All of these functions operate on arguments that use a target-specific
6253 value type "``va_list``". The LLVM assembly language reference manual
6254 does not define what this type is, so all transformations should be
6255 prepared to handle these functions regardless of the type used.
6257 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6258 variable argument handling intrinsic functions are used.
6260 .. code-block:: llvm
6262 define i32 @test(i32 %X, ...) {
6263 ; Initialize variable argument processing
6265 %ap2 = bitcast i8** %ap to i8*
6266 call void @llvm.va_start(i8* %ap2)
6268 ; Read a single integer argument
6269 %tmp = va_arg i8** %ap, i32
6271 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6273 %aq2 = bitcast i8** %aq to i8*
6274 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6275 call void @llvm.va_end(i8* %aq2)
6277 ; Stop processing of arguments.
6278 call void @llvm.va_end(i8* %ap2)
6282 declare void @llvm.va_start(i8*)
6283 declare void @llvm.va_copy(i8*, i8*)
6284 declare void @llvm.va_end(i8*)
6288 '``llvm.va_start``' Intrinsic
6289 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6296 declare void %llvm.va_start(i8* <arglist>)
6301 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6302 subsequent use by ``va_arg``.
6307 The argument is a pointer to a ``va_list`` element to initialize.
6312 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6313 available in C. In a target-dependent way, it initializes the
6314 ``va_list`` element to which the argument points, so that the next call
6315 to ``va_arg`` will produce the first variable argument passed to the
6316 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6317 to know the last argument of the function as the compiler can figure
6320 '``llvm.va_end``' Intrinsic
6321 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6328 declare void @llvm.va_end(i8* <arglist>)
6333 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6334 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6339 The argument is a pointer to a ``va_list`` to destroy.
6344 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6345 available in C. In a target-dependent way, it destroys the ``va_list``
6346 element to which the argument points. Calls to
6347 :ref:`llvm.va_start <int_va_start>` and
6348 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6353 '``llvm.va_copy``' Intrinsic
6354 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6361 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6366 The '``llvm.va_copy``' intrinsic copies the current argument position
6367 from the source argument list to the destination argument list.
6372 The first argument is a pointer to a ``va_list`` element to initialize.
6373 The second argument is a pointer to a ``va_list`` element to copy from.
6378 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6379 available in C. In a target-dependent way, it copies the source
6380 ``va_list`` element into the destination ``va_list`` element. This
6381 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6382 arbitrarily complex and require, for example, memory allocation.
6384 Accurate Garbage Collection Intrinsics
6385 --------------------------------------
6387 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6388 (GC) requires the implementation and generation of these intrinsics.
6389 These intrinsics allow identification of :ref:`GC roots on the
6390 stack <int_gcroot>`, as well as garbage collector implementations that
6391 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6392 Front-ends for type-safe garbage collected languages should generate
6393 these intrinsics to make use of the LLVM garbage collectors. For more
6394 details, see `Accurate Garbage Collection with
6395 LLVM <GarbageCollection.html>`_.
6397 The garbage collection intrinsics only operate on objects in the generic
6398 address space (address space zero).
6402 '``llvm.gcroot``' Intrinsic
6403 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6410 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6415 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6416 the code generator, and allows some metadata to be associated with it.
6421 The first argument specifies the address of a stack object that contains
6422 the root pointer. The second pointer (which must be either a constant or
6423 a global value address) contains the meta-data to be associated with the
6429 At runtime, a call to this intrinsic stores a null pointer into the
6430 "ptrloc" location. At compile-time, the code generator generates
6431 information to allow the runtime to find the pointer at GC safe points.
6432 The '``llvm.gcroot``' intrinsic may only be used in a function which
6433 :ref:`specifies a GC algorithm <gc>`.
6437 '``llvm.gcread``' Intrinsic
6438 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6445 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6450 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6451 locations, allowing garbage collector implementations that require read
6457 The second argument is the address to read from, which should be an
6458 address allocated from the garbage collector. The first object is a
6459 pointer to the start of the referenced object, if needed by the language
6460 runtime (otherwise null).
6465 The '``llvm.gcread``' intrinsic has the same semantics as a load
6466 instruction, but may be replaced with substantially more complex code by
6467 the garbage collector runtime, as needed. The '``llvm.gcread``'
6468 intrinsic may only be used in a function which :ref:`specifies a GC
6473 '``llvm.gcwrite``' Intrinsic
6474 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6481 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6486 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6487 locations, allowing garbage collector implementations that require write
6488 barriers (such as generational or reference counting collectors).
6493 The first argument is the reference to store, the second is the start of
6494 the object to store it to, and the third is the address of the field of
6495 Obj to store to. If the runtime does not require a pointer to the
6496 object, Obj may be null.
6501 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6502 instruction, but may be replaced with substantially more complex code by
6503 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6504 intrinsic may only be used in a function which :ref:`specifies a GC
6507 Code Generator Intrinsics
6508 -------------------------
6510 These intrinsics are provided by LLVM to expose special features that
6511 may only be implemented with code generator support.
6513 '``llvm.returnaddress``' Intrinsic
6514 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6521 declare i8 *@llvm.returnaddress(i32 <level>)
6526 The '``llvm.returnaddress``' intrinsic attempts to compute a
6527 target-specific value indicating the return address of the current
6528 function or one of its callers.
6533 The argument to this intrinsic indicates which function to return the
6534 address for. Zero indicates the calling function, one indicates its
6535 caller, etc. The argument is **required** to be a constant integer
6541 The '``llvm.returnaddress``' intrinsic either returns a pointer
6542 indicating the return address of the specified call frame, or zero if it
6543 cannot be identified. The value returned by this intrinsic is likely to
6544 be incorrect or 0 for arguments other than zero, so it should only be
6545 used for debugging purposes.
6547 Note that calling this intrinsic does not prevent function inlining or
6548 other aggressive transformations, so the value returned may not be that
6549 of the obvious source-language caller.
6551 '``llvm.frameaddress``' Intrinsic
6552 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6559 declare i8* @llvm.frameaddress(i32 <level>)
6564 The '``llvm.frameaddress``' intrinsic attempts to return the
6565 target-specific frame pointer value for the specified stack frame.
6570 The argument to this intrinsic indicates which function to return the
6571 frame pointer for. Zero indicates the calling function, one indicates
6572 its caller, etc. The argument is **required** to be a constant integer
6578 The '``llvm.frameaddress``' intrinsic either returns a pointer
6579 indicating the frame address of the specified call frame, or zero if it
6580 cannot be identified. The value returned by this intrinsic is likely to
6581 be incorrect or 0 for arguments other than zero, so it should only be
6582 used for debugging purposes.
6584 Note that calling this intrinsic does not prevent function inlining or
6585 other aggressive transformations, so the value returned may not be that
6586 of the obvious source-language caller.
6590 '``llvm.stacksave``' Intrinsic
6591 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6598 declare i8* @llvm.stacksave()
6603 The '``llvm.stacksave``' intrinsic is used to remember the current state
6604 of the function stack, for use with
6605 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6606 implementing language features like scoped automatic variable sized
6612 This intrinsic returns a opaque pointer value that can be passed to
6613 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6614 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6615 ``llvm.stacksave``, it effectively restores the state of the stack to
6616 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6617 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6618 were allocated after the ``llvm.stacksave`` was executed.
6620 .. _int_stackrestore:
6622 '``llvm.stackrestore``' Intrinsic
6623 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6630 declare void @llvm.stackrestore(i8* %ptr)
6635 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6636 the function stack to the state it was in when the corresponding
6637 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6638 useful for implementing language features like scoped automatic variable
6639 sized arrays in C99.
6644 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6646 '``llvm.prefetch``' Intrinsic
6647 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6654 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6659 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6660 insert a prefetch instruction if supported; otherwise, it is a noop.
6661 Prefetches have no effect on the behavior of the program but can change
6662 its performance characteristics.
6667 ``address`` is the address to be prefetched, ``rw`` is the specifier
6668 determining if the fetch should be for a read (0) or write (1), and
6669 ``locality`` is a temporal locality specifier ranging from (0) - no
6670 locality, to (3) - extremely local keep in cache. The ``cache type``
6671 specifies whether the prefetch is performed on the data (1) or
6672 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6673 arguments must be constant integers.
6678 This intrinsic does not modify the behavior of the program. In
6679 particular, prefetches cannot trap and do not produce a value. On
6680 targets that support this intrinsic, the prefetch can provide hints to
6681 the processor cache for better performance.
6683 '``llvm.pcmarker``' Intrinsic
6684 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6691 declare void @llvm.pcmarker(i32 <id>)
6696 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6697 Counter (PC) in a region of code to simulators and other tools. The
6698 method is target specific, but it is expected that the marker will use
6699 exported symbols to transmit the PC of the marker. The marker makes no
6700 guarantees that it will remain with any specific instruction after
6701 optimizations. It is possible that the presence of a marker will inhibit
6702 optimizations. The intended use is to be inserted after optimizations to
6703 allow correlations of simulation runs.
6708 ``id`` is a numerical id identifying the marker.
6713 This intrinsic does not modify the behavior of the program. Backends
6714 that do not support this intrinsic may ignore it.
6716 '``llvm.readcyclecounter``' Intrinsic
6717 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6724 declare i64 @llvm.readcyclecounter()
6729 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6730 counter register (or similar low latency, high accuracy clocks) on those
6731 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6732 should map to RPCC. As the backing counters overflow quickly (on the
6733 order of 9 seconds on alpha), this should only be used for small
6739 When directly supported, reading the cycle counter should not modify any
6740 memory. Implementations are allowed to either return a application
6741 specific value or a system wide value. On backends without support, this
6742 is lowered to a constant 0.
6744 Note that runtime support may be conditional on the privilege-level code is
6745 running at and the host platform.
6747 Standard C Library Intrinsics
6748 -----------------------------
6750 LLVM provides intrinsics for a few important standard C library
6751 functions. These intrinsics allow source-language front-ends to pass
6752 information about the alignment of the pointer arguments to the code
6753 generator, providing opportunity for more efficient code generation.
6757 '``llvm.memcpy``' Intrinsic
6758 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6763 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6764 integer bit width and for different address spaces. Not all targets
6765 support all bit widths however.
6769 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6770 i32 <len>, i32 <align>, i1 <isvolatile>)
6771 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6772 i64 <len>, i32 <align>, i1 <isvolatile>)
6777 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6778 source location to the destination location.
6780 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6781 intrinsics do not return a value, takes extra alignment/isvolatile
6782 arguments and the pointers can be in specified address spaces.
6787 The first argument is a pointer to the destination, the second is a
6788 pointer to the source. The third argument is an integer argument
6789 specifying the number of bytes to copy, the fourth argument is the
6790 alignment of the source and destination locations, and the fifth is a
6791 boolean indicating a volatile access.
6793 If the call to this intrinsic has an alignment value that is not 0 or 1,
6794 then the caller guarantees that both the source and destination pointers
6795 are aligned to that boundary.
6797 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6798 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6799 very cleanly specified and it is unwise to depend on it.
6804 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6805 source location to the destination location, which are not allowed to
6806 overlap. It copies "len" bytes of memory over. If the argument is known
6807 to be aligned to some boundary, this can be specified as the fourth
6808 argument, otherwise it should be set to 0 or 1.
6810 '``llvm.memmove``' Intrinsic
6811 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6816 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6817 bit width and for different address space. Not all targets support all
6822 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6823 i32 <len>, i32 <align>, i1 <isvolatile>)
6824 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6825 i64 <len>, i32 <align>, i1 <isvolatile>)
6830 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6831 source location to the destination location. It is similar to the
6832 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6835 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6836 intrinsics do not return a value, takes extra alignment/isvolatile
6837 arguments and the pointers can be in specified address spaces.
6842 The first argument is a pointer to the destination, the second is a
6843 pointer to the source. The third argument is an integer argument
6844 specifying the number of bytes to copy, the fourth argument is the
6845 alignment of the source and destination locations, and the fifth is a
6846 boolean indicating a volatile access.
6848 If the call to this intrinsic has an alignment value that is not 0 or 1,
6849 then the caller guarantees that the source and destination pointers are
6850 aligned to that boundary.
6852 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6853 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6854 not very cleanly specified and it is unwise to depend on it.
6859 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6860 source location to the destination location, which may overlap. It
6861 copies "len" bytes of memory over. If the argument is known to be
6862 aligned to some boundary, this can be specified as the fourth argument,
6863 otherwise it should be set to 0 or 1.
6865 '``llvm.memset.*``' Intrinsics
6866 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6871 This is an overloaded intrinsic. You can use llvm.memset on any integer
6872 bit width and for different address spaces. However, not all targets
6873 support all bit widths.
6877 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6878 i32 <len>, i32 <align>, i1 <isvolatile>)
6879 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6880 i64 <len>, i32 <align>, i1 <isvolatile>)
6885 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6886 particular byte value.
6888 Note that, unlike the standard libc function, the ``llvm.memset``
6889 intrinsic does not return a value and takes extra alignment/volatile
6890 arguments. Also, the destination can be in an arbitrary address space.
6895 The first argument is a pointer to the destination to fill, the second
6896 is the byte value with which to fill it, the third argument is an
6897 integer argument specifying the number of bytes to fill, and the fourth
6898 argument is the known alignment of the destination location.
6900 If the call to this intrinsic has an alignment value that is not 0 or 1,
6901 then the caller guarantees that the destination pointer is aligned to
6904 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6905 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6906 very cleanly specified and it is unwise to depend on it.
6911 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6912 at the destination location. If the argument is known to be aligned to
6913 some boundary, this can be specified as the fourth argument, otherwise
6914 it should be set to 0 or 1.
6916 '``llvm.sqrt.*``' Intrinsic
6917 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6922 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6923 floating point or vector of floating point type. Not all targets support
6928 declare float @llvm.sqrt.f32(float %Val)
6929 declare double @llvm.sqrt.f64(double %Val)
6930 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6931 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6932 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6937 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
6938 returning the same value as the libm '``sqrt``' functions would. Unlike
6939 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
6940 negative numbers other than -0.0 (which allows for better optimization,
6941 because there is no need to worry about errno being set).
6942 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
6947 The argument and return value are floating point numbers of the same
6953 This function returns the sqrt of the specified operand if it is a
6954 nonnegative floating point number.
6956 '``llvm.powi.*``' Intrinsic
6957 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6962 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
6963 floating point or vector of floating point type. Not all targets support
6968 declare float @llvm.powi.f32(float %Val, i32 %power)
6969 declare double @llvm.powi.f64(double %Val, i32 %power)
6970 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6971 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6972 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6977 The '``llvm.powi.*``' intrinsics return the first operand raised to the
6978 specified (positive or negative) power. The order of evaluation of
6979 multiplications is not defined. When a vector of floating point type is
6980 used, the second argument remains a scalar integer value.
6985 The second argument is an integer power, and the first is a value to
6986 raise to that power.
6991 This function returns the first value raised to the second power with an
6992 unspecified sequence of rounding operations.
6994 '``llvm.sin.*``' Intrinsic
6995 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7000 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7001 floating point or vector of floating point type. Not all targets support
7006 declare float @llvm.sin.f32(float %Val)
7007 declare double @llvm.sin.f64(double %Val)
7008 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7009 declare fp128 @llvm.sin.f128(fp128 %Val)
7010 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7015 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7020 The argument and return value are floating point numbers of the same
7026 This function returns the sine of the specified operand, returning the
7027 same values as the libm ``sin`` functions would, and handles error
7028 conditions in the same way.
7030 '``llvm.cos.*``' Intrinsic
7031 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7036 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7037 floating point or vector of floating point type. Not all targets support
7042 declare float @llvm.cos.f32(float %Val)
7043 declare double @llvm.cos.f64(double %Val)
7044 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7045 declare fp128 @llvm.cos.f128(fp128 %Val)
7046 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7051 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7056 The argument and return value are floating point numbers of the same
7062 This function returns the cosine of the specified operand, returning the
7063 same values as the libm ``cos`` functions would, and handles error
7064 conditions in the same way.
7066 '``llvm.pow.*``' Intrinsic
7067 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7072 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7073 floating point or vector of floating point type. Not all targets support
7078 declare float @llvm.pow.f32(float %Val, float %Power)
7079 declare double @llvm.pow.f64(double %Val, double %Power)
7080 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7081 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7082 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7087 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7088 specified (positive or negative) power.
7093 The second argument is a floating point power, and the first is a value
7094 to raise to that power.
7099 This function returns the first value raised to the second power,
7100 returning the same values as the libm ``pow`` functions would, and
7101 handles error conditions in the same way.
7103 '``llvm.exp.*``' Intrinsic
7104 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7109 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7110 floating point or vector of floating point type. Not all targets support
7115 declare float @llvm.exp.f32(float %Val)
7116 declare double @llvm.exp.f64(double %Val)
7117 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7118 declare fp128 @llvm.exp.f128(fp128 %Val)
7119 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7124 The '``llvm.exp.*``' intrinsics perform the exp function.
7129 The argument and return value are floating point numbers of the same
7135 This function returns the same values as the libm ``exp`` functions
7136 would, and handles error conditions in the same way.
7138 '``llvm.exp2.*``' Intrinsic
7139 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7144 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7145 floating point or vector of floating point type. Not all targets support
7150 declare float @llvm.exp2.f32(float %Val)
7151 declare double @llvm.exp2.f64(double %Val)
7152 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7153 declare fp128 @llvm.exp2.f128(fp128 %Val)
7154 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7159 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7164 The argument and return value are floating point numbers of the same
7170 This function returns the same values as the libm ``exp2`` functions
7171 would, and handles error conditions in the same way.
7173 '``llvm.log.*``' Intrinsic
7174 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7179 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7180 floating point or vector of floating point type. Not all targets support
7185 declare float @llvm.log.f32(float %Val)
7186 declare double @llvm.log.f64(double %Val)
7187 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7188 declare fp128 @llvm.log.f128(fp128 %Val)
7189 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7194 The '``llvm.log.*``' intrinsics perform the log function.
7199 The argument and return value are floating point numbers of the same
7205 This function returns the same values as the libm ``log`` functions
7206 would, and handles error conditions in the same way.
7208 '``llvm.log10.*``' Intrinsic
7209 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7214 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7215 floating point or vector of floating point type. Not all targets support
7220 declare float @llvm.log10.f32(float %Val)
7221 declare double @llvm.log10.f64(double %Val)
7222 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7223 declare fp128 @llvm.log10.f128(fp128 %Val)
7224 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7229 The '``llvm.log10.*``' intrinsics perform the log10 function.
7234 The argument and return value are floating point numbers of the same
7240 This function returns the same values as the libm ``log10`` functions
7241 would, and handles error conditions in the same way.
7243 '``llvm.log2.*``' Intrinsic
7244 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7249 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7250 floating point or vector of floating point type. Not all targets support
7255 declare float @llvm.log2.f32(float %Val)
7256 declare double @llvm.log2.f64(double %Val)
7257 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7258 declare fp128 @llvm.log2.f128(fp128 %Val)
7259 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7264 The '``llvm.log2.*``' intrinsics perform the log2 function.
7269 The argument and return value are floating point numbers of the same
7275 This function returns the same values as the libm ``log2`` functions
7276 would, and handles error conditions in the same way.
7278 '``llvm.fma.*``' Intrinsic
7279 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7284 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7285 floating point or vector of floating point type. Not all targets support
7290 declare float @llvm.fma.f32(float %a, float %b, float %c)
7291 declare double @llvm.fma.f64(double %a, double %b, double %c)
7292 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7293 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7294 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7299 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7305 The argument and return value are floating point numbers of the same
7311 This function returns the same values as the libm ``fma`` functions
7314 '``llvm.fabs.*``' Intrinsic
7315 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7320 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7321 floating point or vector of floating point type. Not all targets support
7326 declare float @llvm.fabs.f32(float %Val)
7327 declare double @llvm.fabs.f64(double %Val)
7328 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7329 declare fp128 @llvm.fabs.f128(fp128 %Val)
7330 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7335 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7341 The argument and return value are floating point numbers of the same
7347 This function returns the same values as the libm ``fabs`` functions
7348 would, and handles error conditions in the same way.
7350 '``llvm.floor.*``' Intrinsic
7351 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7356 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7357 floating point or vector of floating point type. Not all targets support
7362 declare float @llvm.floor.f32(float %Val)
7363 declare double @llvm.floor.f64(double %Val)
7364 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7365 declare fp128 @llvm.floor.f128(fp128 %Val)
7366 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7371 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7376 The argument and return value are floating point numbers of the same
7382 This function returns the same values as the libm ``floor`` functions
7383 would, and handles error conditions in the same way.
7385 '``llvm.ceil.*``' Intrinsic
7386 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7391 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7392 floating point or vector of floating point type. Not all targets support
7397 declare float @llvm.ceil.f32(float %Val)
7398 declare double @llvm.ceil.f64(double %Val)
7399 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7400 declare fp128 @llvm.ceil.f128(fp128 %Val)
7401 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7406 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7411 The argument and return value are floating point numbers of the same
7417 This function returns the same values as the libm ``ceil`` functions
7418 would, and handles error conditions in the same way.
7420 '``llvm.trunc.*``' Intrinsic
7421 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7426 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7427 floating point or vector of floating point type. Not all targets support
7432 declare float @llvm.trunc.f32(float %Val)
7433 declare double @llvm.trunc.f64(double %Val)
7434 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7435 declare fp128 @llvm.trunc.f128(fp128 %Val)
7436 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7441 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7442 nearest integer not larger in magnitude than the operand.
7447 The argument and return value are floating point numbers of the same
7453 This function returns the same values as the libm ``trunc`` functions
7454 would, and handles error conditions in the same way.
7456 '``llvm.rint.*``' Intrinsic
7457 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7462 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7463 floating point or vector of floating point type. Not all targets support
7468 declare float @llvm.rint.f32(float %Val)
7469 declare double @llvm.rint.f64(double %Val)
7470 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7471 declare fp128 @llvm.rint.f128(fp128 %Val)
7472 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7477 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7478 nearest integer. It may raise an inexact floating-point exception if the
7479 operand isn't an integer.
7484 The argument and return value are floating point numbers of the same
7490 This function returns the same values as the libm ``rint`` functions
7491 would, and handles error conditions in the same way.
7493 '``llvm.nearbyint.*``' Intrinsic
7494 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7499 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7500 floating point or vector of floating point type. Not all targets support
7505 declare float @llvm.nearbyint.f32(float %Val)
7506 declare double @llvm.nearbyint.f64(double %Val)
7507 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7508 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7509 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7514 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7520 The argument and return value are floating point numbers of the same
7526 This function returns the same values as the libm ``nearbyint``
7527 functions would, and handles error conditions in the same way.
7529 Bit Manipulation Intrinsics
7530 ---------------------------
7532 LLVM provides intrinsics for a few important bit manipulation
7533 operations. These allow efficient code generation for some algorithms.
7535 '``llvm.bswap.*``' Intrinsics
7536 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7541 This is an overloaded intrinsic function. You can use bswap on any
7542 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7546 declare i16 @llvm.bswap.i16(i16 <id>)
7547 declare i32 @llvm.bswap.i32(i32 <id>)
7548 declare i64 @llvm.bswap.i64(i64 <id>)
7553 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7554 values with an even number of bytes (positive multiple of 16 bits).
7555 These are useful for performing operations on data that is not in the
7556 target's native byte order.
7561 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7562 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7563 intrinsic returns an i32 value that has the four bytes of the input i32
7564 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7565 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7566 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7567 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7570 '``llvm.ctpop.*``' Intrinsic
7571 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7576 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7577 bit width, or on any vector with integer elements. Not all targets
7578 support all bit widths or vector types, however.
7582 declare i8 @llvm.ctpop.i8(i8 <src>)
7583 declare i16 @llvm.ctpop.i16(i16 <src>)
7584 declare i32 @llvm.ctpop.i32(i32 <src>)
7585 declare i64 @llvm.ctpop.i64(i64 <src>)
7586 declare i256 @llvm.ctpop.i256(i256 <src>)
7587 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7592 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7598 The only argument is the value to be counted. The argument may be of any
7599 integer type, or a vector with integer elements. The return type must
7600 match the argument type.
7605 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7606 each element of a vector.
7608 '``llvm.ctlz.*``' Intrinsic
7609 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7614 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7615 integer bit width, or any vector whose elements are integers. Not all
7616 targets support all bit widths or vector types, however.
7620 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7621 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7622 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7623 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7624 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7625 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7630 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7631 leading zeros in a variable.
7636 The first argument is the value to be counted. This argument may be of
7637 any integer type, or a vectory with integer element type. The return
7638 type must match the first argument type.
7640 The second argument must be a constant and is a flag to indicate whether
7641 the intrinsic should ensure that a zero as the first argument produces a
7642 defined result. Historically some architectures did not provide a
7643 defined result for zero values as efficiently, and many algorithms are
7644 now predicated on avoiding zero-value inputs.
7649 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7650 zeros in a variable, or within each element of the vector. If
7651 ``src == 0`` then the result is the size in bits of the type of ``src``
7652 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7653 ``llvm.ctlz(i32 2) = 30``.
7655 '``llvm.cttz.*``' Intrinsic
7656 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7661 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7662 integer bit width, or any vector of integer elements. Not all targets
7663 support all bit widths or vector types, however.
7667 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7668 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7669 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7670 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7671 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7672 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7677 The '``llvm.cttz``' family of intrinsic functions counts the number of
7683 The first argument is the value to be counted. This argument may be of
7684 any integer type, or a vectory with integer element type. The return
7685 type must match the first argument type.
7687 The second argument must be a constant and is a flag to indicate whether
7688 the intrinsic should ensure that a zero as the first argument produces a
7689 defined result. Historically some architectures did not provide a
7690 defined result for zero values as efficiently, and many algorithms are
7691 now predicated on avoiding zero-value inputs.
7696 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7697 zeros in a variable, or within each element of a vector. If ``src == 0``
7698 then the result is the size in bits of the type of ``src`` if
7699 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7700 ``llvm.cttz(2) = 1``.
7702 Arithmetic with Overflow Intrinsics
7703 -----------------------------------
7705 LLVM provides intrinsics for some arithmetic with overflow operations.
7707 '``llvm.sadd.with.overflow.*``' Intrinsics
7708 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7713 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7714 on any integer bit width.
7718 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7719 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7720 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7725 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7726 a signed addition of the two arguments, and indicate whether an overflow
7727 occurred during the signed summation.
7732 The arguments (%a and %b) and the first element of the result structure
7733 may be of integer types of any bit width, but they must have the same
7734 bit width. The second element of the result structure must be of type
7735 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7741 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7742 a signed addition of the two variables. They return a structure --- the
7743 first element of which is the signed summation, and the second element
7744 of which is a bit specifying if the signed summation resulted in an
7750 .. code-block:: llvm
7752 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7753 %sum = extractvalue {i32, i1} %res, 0
7754 %obit = extractvalue {i32, i1} %res, 1
7755 br i1 %obit, label %overflow, label %normal
7757 '``llvm.uadd.with.overflow.*``' Intrinsics
7758 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7763 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7764 on any integer bit width.
7768 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7769 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7770 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7775 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7776 an unsigned addition of the two arguments, and indicate whether a carry
7777 occurred during the unsigned summation.
7782 The arguments (%a and %b) and the first element of the result structure
7783 may be of integer types of any bit width, but they must have the same
7784 bit width. The second element of the result structure must be of type
7785 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7791 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7792 an unsigned addition of the two arguments. They return a structure --- the
7793 first element of which is the sum, and the second element of which is a
7794 bit specifying if the unsigned summation resulted in a carry.
7799 .. code-block:: llvm
7801 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7802 %sum = extractvalue {i32, i1} %res, 0
7803 %obit = extractvalue {i32, i1} %res, 1
7804 br i1 %obit, label %carry, label %normal
7806 '``llvm.ssub.with.overflow.*``' Intrinsics
7807 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7812 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7813 on any integer bit width.
7817 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7818 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7819 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7824 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7825 a signed subtraction of the two arguments, and indicate whether an
7826 overflow occurred during the signed subtraction.
7831 The arguments (%a and %b) and the first element of the result structure
7832 may be of integer types of any bit width, but they must have the same
7833 bit width. The second element of the result structure must be of type
7834 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7840 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7841 a signed subtraction of the two arguments. They return a structure --- the
7842 first element of which is the subtraction, and the second element of
7843 which is a bit specifying if the signed subtraction resulted in an
7849 .. code-block:: llvm
7851 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7852 %sum = extractvalue {i32, i1} %res, 0
7853 %obit = extractvalue {i32, i1} %res, 1
7854 br i1 %obit, label %overflow, label %normal
7856 '``llvm.usub.with.overflow.*``' Intrinsics
7857 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7862 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
7863 on any integer bit width.
7867 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7868 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7869 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7874 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7875 an unsigned subtraction of the two arguments, and indicate whether an
7876 overflow occurred during the unsigned subtraction.
7881 The arguments (%a and %b) and the first element of the result structure
7882 may be of integer types of any bit width, but they must have the same
7883 bit width. The second element of the result structure must be of type
7884 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7890 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7891 an unsigned subtraction of the two arguments. They return a structure ---
7892 the first element of which is the subtraction, and the second element of
7893 which is a bit specifying if the unsigned subtraction resulted in an
7899 .. code-block:: llvm
7901 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7902 %sum = extractvalue {i32, i1} %res, 0
7903 %obit = extractvalue {i32, i1} %res, 1
7904 br i1 %obit, label %overflow, label %normal
7906 '``llvm.smul.with.overflow.*``' Intrinsics
7907 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7912 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
7913 on any integer bit width.
7917 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7918 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7919 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7924 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7925 a signed multiplication of the two arguments, and indicate whether an
7926 overflow occurred during the signed multiplication.
7931 The arguments (%a and %b) and the first element of the result structure
7932 may be of integer types of any bit width, but they must have the same
7933 bit width. The second element of the result structure must be of type
7934 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7940 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7941 a signed multiplication of the two arguments. They return a structure ---
7942 the first element of which is the multiplication, and the second element
7943 of which is a bit specifying if the signed multiplication resulted in an
7949 .. code-block:: llvm
7951 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7952 %sum = extractvalue {i32, i1} %res, 0
7953 %obit = extractvalue {i32, i1} %res, 1
7954 br i1 %obit, label %overflow, label %normal
7956 '``llvm.umul.with.overflow.*``' Intrinsics
7957 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7962 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
7963 on any integer bit width.
7967 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7968 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7969 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7974 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7975 a unsigned multiplication of the two arguments, and indicate whether an
7976 overflow occurred during the unsigned multiplication.
7981 The arguments (%a and %b) and the first element of the result structure
7982 may be of integer types of any bit width, but they must have the same
7983 bit width. The second element of the result structure must be of type
7984 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7990 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7991 an unsigned multiplication of the two arguments. They return a structure ---
7992 the first element of which is the multiplication, and the second
7993 element of which is a bit specifying if the unsigned multiplication
7994 resulted in an overflow.
7999 .. code-block:: llvm
8001 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8002 %sum = extractvalue {i32, i1} %res, 0
8003 %obit = extractvalue {i32, i1} %res, 1
8004 br i1 %obit, label %overflow, label %normal
8006 Specialised Arithmetic Intrinsics
8007 ---------------------------------
8009 '``llvm.fmuladd.*``' Intrinsic
8010 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8017 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8018 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8023 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8024 expressions that can be fused if the code generator determines that (a) the
8025 target instruction set has support for a fused operation, and (b) that the
8026 fused operation is more efficient than the equivalent, separate pair of mul
8027 and add instructions.
8032 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8033 multiplicands, a and b, and an addend c.
8042 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8044 is equivalent to the expression a \* b + c, except that rounding will
8045 not be performed between the multiplication and addition steps if the
8046 code generator fuses the operations. Fusion is not guaranteed, even if
8047 the target platform supports it. If a fused multiply-add is required the
8048 corresponding llvm.fma.\* intrinsic function should be used instead.
8053 .. code-block:: llvm
8055 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8057 Half Precision Floating Point Intrinsics
8058 ----------------------------------------
8060 For most target platforms, half precision floating point is a
8061 storage-only format. This means that it is a dense encoding (in memory)
8062 but does not support computation in the format.
8064 This means that code must first load the half-precision floating point
8065 value as an i16, then convert it to float with
8066 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8067 then be performed on the float value (including extending to double
8068 etc). To store the value back to memory, it is first converted to float
8069 if needed, then converted to i16 with
8070 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8073 .. _int_convert_to_fp16:
8075 '``llvm.convert.to.fp16``' Intrinsic
8076 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8083 declare i16 @llvm.convert.to.fp16(f32 %a)
8088 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8089 from single precision floating point format to half precision floating
8095 The intrinsic function contains single argument - the value to be
8101 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8102 from single precision floating point format to half precision floating
8103 point format. The return value is an ``i16`` which contains the
8109 .. code-block:: llvm
8111 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8112 store i16 %res, i16* @x, align 2
8114 .. _int_convert_from_fp16:
8116 '``llvm.convert.from.fp16``' Intrinsic
8117 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8124 declare f32 @llvm.convert.from.fp16(i16 %a)
8129 The '``llvm.convert.from.fp16``' intrinsic function performs a
8130 conversion from half precision floating point format to single precision
8131 floating point format.
8136 The intrinsic function contains single argument - the value to be
8142 The '``llvm.convert.from.fp16``' intrinsic function performs a
8143 conversion from half single precision floating point format to single
8144 precision floating point format. The input half-float value is
8145 represented by an ``i16`` value.
8150 .. code-block:: llvm
8152 %a = load i16* @x, align 2
8153 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8158 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8159 prefix), are described in the `LLVM Source Level
8160 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8163 Exception Handling Intrinsics
8164 -----------------------------
8166 The LLVM exception handling intrinsics (which all start with
8167 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8168 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8172 Trampoline Intrinsics
8173 ---------------------
8175 These intrinsics make it possible to excise one parameter, marked with
8176 the :ref:`nest <nest>` attribute, from a function. The result is a
8177 callable function pointer lacking the nest parameter - the caller does
8178 not need to provide a value for it. Instead, the value to use is stored
8179 in advance in a "trampoline", a block of memory usually allocated on the
8180 stack, which also contains code to splice the nest value into the
8181 argument list. This is used to implement the GCC nested function address
8184 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8185 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8186 It can be created as follows:
8188 .. code-block:: llvm
8190 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8191 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8192 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8193 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8194 %fp = bitcast i8* %p to i32 (i32, i32)*
8196 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8197 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8201 '``llvm.init.trampoline``' Intrinsic
8202 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8209 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8214 This fills the memory pointed to by ``tramp`` with executable code,
8215 turning it into a trampoline.
8220 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8221 pointers. The ``tramp`` argument must point to a sufficiently large and
8222 sufficiently aligned block of memory; this memory is written to by the
8223 intrinsic. Note that the size and the alignment are target-specific -
8224 LLVM currently provides no portable way of determining them, so a
8225 front-end that generates this intrinsic needs to have some
8226 target-specific knowledge. The ``func`` argument must hold a function
8227 bitcast to an ``i8*``.
8232 The block of memory pointed to by ``tramp`` is filled with target
8233 dependent code, turning it into a function. Then ``tramp`` needs to be
8234 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8235 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8236 function's signature is the same as that of ``func`` with any arguments
8237 marked with the ``nest`` attribute removed. At most one such ``nest``
8238 argument is allowed, and it must be of pointer type. Calling the new
8239 function is equivalent to calling ``func`` with the same argument list,
8240 but with ``nval`` used for the missing ``nest`` argument. If, after
8241 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8242 modified, then the effect of any later call to the returned function
8243 pointer is undefined.
8247 '``llvm.adjust.trampoline``' Intrinsic
8248 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8255 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8260 This performs any required machine-specific adjustment to the address of
8261 a trampoline (passed as ``tramp``).
8266 ``tramp`` must point to a block of memory which already has trampoline
8267 code filled in by a previous call to
8268 :ref:`llvm.init.trampoline <int_it>`.
8273 On some architectures the address of the code to be executed needs to be
8274 different to the address where the trampoline is actually stored. This
8275 intrinsic returns the executable address corresponding to ``tramp``
8276 after performing the required machine specific adjustments. The pointer
8277 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8282 This class of intrinsics exists to information about the lifetime of
8283 memory objects and ranges where variables are immutable.
8285 '``llvm.lifetime.start``' Intrinsic
8286 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8293 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8298 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8304 The first argument is a constant integer representing the size of the
8305 object, or -1 if it is variable sized. The second argument is a pointer
8311 This intrinsic indicates that before this point in the code, the value
8312 of the memory pointed to by ``ptr`` is dead. This means that it is known
8313 to never be used and has an undefined value. A load from the pointer
8314 that precedes this intrinsic can be replaced with ``'undef'``.
8316 '``llvm.lifetime.end``' Intrinsic
8317 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8324 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8329 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8335 The first argument is a constant integer representing the size of the
8336 object, or -1 if it is variable sized. The second argument is a pointer
8342 This intrinsic indicates that after this point in the code, the value of
8343 the memory pointed to by ``ptr`` is dead. This means that it is known to
8344 never be used and has an undefined value. Any stores into the memory
8345 object following this intrinsic may be removed as dead.
8347 '``llvm.invariant.start``' Intrinsic
8348 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8355 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8360 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8361 a memory object will not change.
8366 The first argument is a constant integer representing the size of the
8367 object, or -1 if it is variable sized. The second argument is a pointer
8373 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8374 the return value, the referenced memory location is constant and
8377 '``llvm.invariant.end``' Intrinsic
8378 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8385 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8390 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8391 memory object are mutable.
8396 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8397 The second argument is a constant integer representing the size of the
8398 object, or -1 if it is variable sized and the third argument is a
8399 pointer to the object.
8404 This intrinsic indicates that the memory is mutable again.
8409 This class of intrinsics is designed to be generic and has no specific
8412 '``llvm.var.annotation``' Intrinsic
8413 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8420 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8425 The '``llvm.var.annotation``' intrinsic.
8430 The first argument is a pointer to a value, the second is a pointer to a
8431 global string, the third is a pointer to a global string which is the
8432 source file name, and the last argument is the line number.
8437 This intrinsic allows annotation of local variables with arbitrary
8438 strings. This can be useful for special purpose optimizations that want
8439 to look for these annotations. These have no other defined use; they are
8440 ignored by code generation and optimization.
8442 '``llvm.ptr.annotation.*``' Intrinsic
8443 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8448 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8449 pointer to an integer of any width. *NOTE* you must specify an address space for
8450 the pointer. The identifier for the default address space is the integer
8455 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8456 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8457 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8458 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8459 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8464 The '``llvm.ptr.annotation``' intrinsic.
8469 The first argument is a pointer to an integer value of arbitrary bitwidth
8470 (result of some expression), the second is a pointer to a global string, the
8471 third is a pointer to a global string which is the source file name, and the
8472 last argument is the line number. It returns the value of the first argument.
8477 This intrinsic allows annotation of a pointer to an integer with arbitrary
8478 strings. This can be useful for special purpose optimizations that want to look
8479 for these annotations. These have no other defined use; they are ignored by code
8480 generation and optimization.
8482 '``llvm.annotation.*``' Intrinsic
8483 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8488 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8489 any integer bit width.
8493 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8494 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8495 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8496 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8497 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8502 The '``llvm.annotation``' intrinsic.
8507 The first argument is an integer value (result of some expression), the
8508 second is a pointer to a global string, the third is a pointer to a
8509 global string which is the source file name, and the last argument is
8510 the line number. It returns the value of the first argument.
8515 This intrinsic allows annotations to be put on arbitrary expressions
8516 with arbitrary strings. This can be useful for special purpose
8517 optimizations that want to look for these annotations. These have no
8518 other defined use; they are ignored by code generation and optimization.
8520 '``llvm.trap``' Intrinsic
8521 ^^^^^^^^^^^^^^^^^^^^^^^^^
8528 declare void @llvm.trap() noreturn nounwind
8533 The '``llvm.trap``' intrinsic.
8543 This intrinsic is lowered to the target dependent trap instruction. If
8544 the target does not have a trap instruction, this intrinsic will be
8545 lowered to a call of the ``abort()`` function.
8547 '``llvm.debugtrap``' Intrinsic
8548 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8555 declare void @llvm.debugtrap() nounwind
8560 The '``llvm.debugtrap``' intrinsic.
8570 This intrinsic is lowered to code which is intended to cause an
8571 execution trap with the intention of requesting the attention of a
8574 '``llvm.stackprotector``' Intrinsic
8575 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8582 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8587 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8588 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8589 is placed on the stack before local variables.
8594 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8595 The first argument is the value loaded from the stack guard
8596 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8597 enough space to hold the value of the guard.
8602 This intrinsic causes the prologue/epilogue inserter to force the
8603 position of the ``AllocaInst`` stack slot to be before local variables
8604 on the stack. This is to ensure that if a local variable on the stack is
8605 overwritten, it will destroy the value of the guard. When the function
8606 exits, the guard on the stack is checked against the original guard. If
8607 they are different, then the program aborts by calling the
8608 ``__stack_chk_fail()`` function.
8610 '``llvm.objectsize``' Intrinsic
8611 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8618 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8619 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8624 The ``llvm.objectsize`` intrinsic is designed to provide information to
8625 the optimizers to determine at compile time whether a) an operation
8626 (like memcpy) will overflow a buffer that corresponds to an object, or
8627 b) that a runtime check for overflow isn't necessary. An object in this
8628 context means an allocation of a specific class, structure, array, or
8634 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8635 argument is a pointer to or into the ``object``. The second argument is
8636 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8637 or -1 (if false) when the object size is unknown. The second argument
8638 only accepts constants.
8643 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8644 the size of the object concerned. If the size cannot be determined at
8645 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8646 on the ``min`` argument).
8648 '``llvm.expect``' Intrinsic
8649 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8656 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8657 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8662 The ``llvm.expect`` intrinsic provides information about expected (the
8663 most probable) value of ``val``, which can be used by optimizers.
8668 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8669 a value. The second argument is an expected value, this needs to be a
8670 constant value, variables are not allowed.
8675 This intrinsic is lowered to the ``val``.
8677 '``llvm.donothing``' Intrinsic
8678 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8685 declare void @llvm.donothing() nounwind readnone
8690 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8691 only intrinsic that can be called with an invoke instruction.
8701 This intrinsic does nothing, and it's removed by optimizers and ignored