1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][a-zA-Z$._][a-zA-Z$._0-9]*``'. Identifiers which require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves.
83 #. Unnamed values are represented as an unsigned numeric value with
84 their prefix. For example, ``%12``, ``@2``, ``%44``.
85 #. Constants, which are described in the section Constants_ below.
87 LLVM requires that values start with a prefix for two reasons: Compilers
88 don't need to worry about name clashes with reserved words, and the set
89 of reserved words may be expanded in the future without penalty.
90 Additionally, unnamed identifiers allow a compiler to quickly come up
91 with a temporary variable without having to avoid symbol table
94 Reserved words in LLVM are very similar to reserved words in other
95 languages. There are keywords for different opcodes ('``add``',
96 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
97 '``i32``', etc...), and others. These reserved words cannot conflict
98 with variable names, because none of them start with a prefix character
101 Here is an example of LLVM code to multiply the integer variable
108 %result = mul i32 %X, 8
110 After strength reduction:
114 %result = shl i32 %X, 3
120 %0 = add i32 %X, %X ; yields {i32}:%0
121 %1 = add i32 %0, %0 ; yields {i32}:%1
122 %result = add i32 %1, %1
124 This last way of multiplying ``%X`` by 8 illustrates several important
125 lexical features of LLVM:
127 #. Comments are delimited with a '``;``' and go until the end of line.
128 #. Unnamed temporaries are created when the result of a computation is
129 not assigned to a named value.
130 #. Unnamed temporaries are numbered sequentially (using a per-function
131 incrementing counter, starting with 0).
133 It also shows a convention that we follow in this document. When
134 demonstrating instructions, we will follow an instruction with a comment
135 that defines the type and name of value produced.
143 LLVM programs are composed of ``Module``'s, each of which is a
144 translation unit of the input programs. Each module consists of
145 functions, global variables, and symbol table entries. Modules may be
146 combined together with the LLVM linker, which merges function (and
147 global variable) definitions, resolves forward declarations, and merges
148 symbol table entries. Here is an example of the "hello world" module:
152 ; Declare the string constant as a global constant.
153 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
155 ; External declaration of the puts function
156 declare i32 @puts(i8* nocapture) nounwind
158 ; Definition of main function
159 define i32 @main() { ; i32()*
160 ; Convert [13 x i8]* to i8 *...
161 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
163 ; Call puts function to write out the string to stdout.
164 call i32 @puts(i8* %cast210)
169 !1 = metadata !{i32 42}
172 This example is made up of a :ref:`global variable <globalvars>` named
173 "``.str``", an external declaration of the "``puts``" function, a
174 :ref:`function definition <functionstructure>` for "``main``" and
175 :ref:`named metadata <namedmetadatastructure>` "``foo``".
177 In general, a module is made up of a list of global values (where both
178 functions and global variables are global values). Global values are
179 represented by a pointer to a memory location (in this case, a pointer
180 to an array of char, and a pointer to a function), and have one of the
181 following :ref:`linkage types <linkage>`.
188 All Global Variables and Functions have one of the following types of
192 Global values with "``private``" linkage are only directly
193 accessible by objects in the current module. In particular, linking
194 code into a module with an private global value may cause the
195 private to be renamed as necessary to avoid collisions. Because the
196 symbol is private to the module, all references can be updated. This
197 doesn't show up in any symbol table in the object file.
199 Similar to ``private``, but the symbol is passed through the
200 assembler and evaluated by the linker. Unlike normal strong symbols,
201 they are removed by the linker from the final linked image
202 (executable or dynamic library).
203 ``linker_private_weak``
204 Similar to "``linker_private``", but the symbol is weak. Note that
205 ``linker_private_weak`` symbols are subject to coalescing by the
206 linker. The symbols are removed by the linker from the final linked
207 image (executable or dynamic library).
209 Similar to private, but the value shows as a local symbol
210 (``STB_LOCAL`` in the case of ELF) in the object file. This
211 corresponds to the notion of the '``static``' keyword in C.
212 ``available_externally``
213 Globals with "``available_externally``" linkage are never emitted
214 into the object file corresponding to the LLVM module. They exist to
215 allow inlining and other optimizations to take place given knowledge
216 of the definition of the global, which is known to be somewhere
217 outside the module. Globals with ``available_externally`` linkage
218 are allowed to be discarded at will, and are otherwise the same as
219 ``linkonce_odr``. This linkage type is only allowed on definitions,
222 Globals with "``linkonce``" linkage are merged with other globals of
223 the same name when linkage occurs. This can be used to implement
224 some forms of inline functions, templates, or other code which must
225 be generated in each translation unit that uses it, but where the
226 body may be overridden with a more definitive definition later.
227 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
228 that ``linkonce`` linkage does not actually allow the optimizer to
229 inline the body of this function into callers because it doesn't
230 know if this definition of the function is the definitive definition
231 within the program or whether it will be overridden by a stronger
232 definition. To enable inlining and other optimizations, use
233 "``linkonce_odr``" linkage.
235 "``weak``" linkage has the same merging semantics as ``linkonce``
236 linkage, except that unreferenced globals with ``weak`` linkage may
237 not be discarded. This is used for globals that are declared "weak"
240 "``common``" linkage is most similar to "``weak``" linkage, but they
241 are used for tentative definitions in C, such as "``int X;``" at
242 global scope. Symbols with "``common``" linkage are merged in the
243 same way as ``weak symbols``, and they may not be deleted if
244 unreferenced. ``common`` symbols may not have an explicit section,
245 must have a zero initializer, and may not be marked
246 ':ref:`constant <globalvars>`'. Functions and aliases may not have
249 .. _linkage_appending:
252 "``appending``" linkage may only be applied to global variables of
253 pointer to array type. When two global variables with appending
254 linkage are linked together, the two global arrays are appended
255 together. This is the LLVM, typesafe, equivalent of having the
256 system linker append together "sections" with identical names when
259 The semantics of this linkage follow the ELF object file model: the
260 symbol is weak until linked, if not linked, the symbol becomes null
261 instead of being an undefined reference.
262 ``linkonce_odr``, ``weak_odr``
263 Some languages allow differing globals to be merged, such as two
264 functions with different semantics. Other languages, such as
265 ``C++``, ensure that only equivalent globals are ever merged (the
266 "one definition rule" --- "ODR"). Such languages can use the
267 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
268 global will only be merged with equivalent globals. These linkage
269 types are otherwise the same as their non-``odr`` versions.
270 ``linkonce_odr_auto_hide``
271 Similar to "``linkonce_odr``", but nothing in the translation unit
272 takes the address of this definition. For instance, functions that
273 had an inline definition, but the compiler decided not to inline it.
274 ``linkonce_odr_auto_hide`` may have only ``default`` visibility. The
275 symbols are removed by the linker from the final linked image
276 (executable or dynamic library).
278 If none of the above identifiers are used, the global is externally
279 visible, meaning that it participates in linkage and can be used to
280 resolve external symbol references.
282 The next two types of linkage are targeted for Microsoft Windows
283 platform only. They are designed to support importing (exporting)
284 symbols from (to) DLLs (Dynamic Link Libraries).
287 "``dllimport``" linkage causes the compiler to reference a function
288 or variable via a global pointer to a pointer that is set up by the
289 DLL exporting the symbol. On Microsoft Windows targets, the pointer
290 name is formed by combining ``__imp_`` and the function or variable
293 "``dllexport``" linkage causes the compiler to provide a global
294 pointer to a pointer in a DLL, so that it can be referenced with the
295 ``dllimport`` attribute. On Microsoft Windows targets, the pointer
296 name is formed by combining ``__imp_`` and the function or variable
299 For example, since the "``.LC0``" variable is defined to be internal, if
300 another module defined a "``.LC0``" variable and was linked with this
301 one, one of the two would be renamed, preventing a collision. Since
302 "``main``" and "``puts``" are external (i.e., lacking any linkage
303 declarations), they are accessible outside of the current module.
305 It is illegal for a function *declaration* to have any linkage type
306 other than ``external``, ``dllimport`` or ``extern_weak``.
308 Aliases can have only ``external``, ``internal``, ``weak`` or
309 ``weak_odr`` linkages.
316 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
317 :ref:`invokes <i_invoke>` can all have an optional calling convention
318 specified for the call. The calling convention of any pair of dynamic
319 caller/callee must match, or the behavior of the program is undefined.
320 The following calling conventions are supported by LLVM, and more may be
323 "``ccc``" - The C calling convention
324 This calling convention (the default if no other calling convention
325 is specified) matches the target C calling conventions. This calling
326 convention supports varargs function calls and tolerates some
327 mismatch in the declared prototype and implemented declaration of
328 the function (as does normal C).
329 "``fastcc``" - The fast calling convention
330 This calling convention attempts to make calls as fast as possible
331 (e.g. by passing things in registers). This calling convention
332 allows the target to use whatever tricks it wants to produce fast
333 code for the target, without having to conform to an externally
334 specified ABI (Application Binary Interface). `Tail calls can only
335 be optimized when this, the GHC or the HiPE convention is
336 used. <CodeGenerator.html#id80>`_ This calling convention does not
337 support varargs and requires the prototype of all callees to exactly
338 match the prototype of the function definition.
339 "``coldcc``" - The cold calling convention
340 This calling convention attempts to make code in the caller as
341 efficient as possible under the assumption that the call is not
342 commonly executed. As such, these calls often preserve all registers
343 so that the call does not break any live ranges in the caller side.
344 This calling convention does not support varargs and requires the
345 prototype of all callees to exactly match the prototype of the
347 "``cc 10``" - GHC convention
348 This calling convention has been implemented specifically for use by
349 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
350 It passes everything in registers, going to extremes to achieve this
351 by disabling callee save registers. This calling convention should
352 not be used lightly but only for specific situations such as an
353 alternative to the *register pinning* performance technique often
354 used when implementing functional programming languages. At the
355 moment only X86 supports this convention and it has the following
358 - On *X86-32* only supports up to 4 bit type parameters. No
359 floating point types are supported.
360 - On *X86-64* only supports up to 10 bit type parameters and 6
361 floating point parameters.
363 This calling convention supports `tail call
364 optimization <CodeGenerator.html#id80>`_ but requires both the
365 caller and callee are using it.
366 "``cc 11``" - The HiPE calling convention
367 This calling convention has been implemented specifically for use by
368 the `High-Performance Erlang
369 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
370 native code compiler of the `Ericsson's Open Source Erlang/OTP
371 system <http://www.erlang.org/download.shtml>`_. It uses more
372 registers for argument passing than the ordinary C calling
373 convention and defines no callee-saved registers. The calling
374 convention properly supports `tail call
375 optimization <CodeGenerator.html#id80>`_ but requires that both the
376 caller and the callee use it. It uses a *register pinning*
377 mechanism, similar to GHC's convention, for keeping frequently
378 accessed runtime components pinned to specific hardware registers.
379 At the moment only X86 supports this convention (both 32 and 64
381 "``cc <n>``" - Numbered convention
382 Any calling convention may be specified by number, allowing
383 target-specific calling conventions to be used. Target specific
384 calling conventions start at 64.
386 More calling conventions can be added/defined on an as-needed basis, to
387 support Pascal conventions or any other well-known target-independent
390 .. _visibilitystyles:
395 All Global Variables and Functions have one of the following visibility
398 "``default``" - Default style
399 On targets that use the ELF object file format, default visibility
400 means that the declaration is visible to other modules and, in
401 shared libraries, means that the declared entity may be overridden.
402 On Darwin, default visibility means that the declaration is visible
403 to other modules. Default visibility corresponds to "external
404 linkage" in the language.
405 "``hidden``" - Hidden style
406 Two declarations of an object with hidden visibility refer to the
407 same object if they are in the same shared object. Usually, hidden
408 visibility indicates that the symbol will not be placed into the
409 dynamic symbol table, so no other module (executable or shared
410 library) can reference it directly.
411 "``protected``" - Protected style
412 On ELF, protected visibility indicates that the symbol will be
413 placed in the dynamic symbol table, but that references within the
414 defining module will bind to the local symbol. That is, the symbol
415 cannot be overridden by another module.
422 LLVM IR allows you to specify name aliases for certain types. This can
423 make it easier to read the IR and make the IR more condensed
424 (particularly when recursive types are involved). An example of a name
429 %mytype = type { %mytype*, i32 }
431 You may give a name to any :ref:`type <typesystem>` except
432 ":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
433 expected with the syntax "%mytype".
435 Note that type names are aliases for the structural type that they
436 indicate, and that you can therefore specify multiple names for the same
437 type. This often leads to confusing behavior when dumping out a .ll
438 file. Since LLVM IR uses structural typing, the name is not part of the
439 type. When printing out LLVM IR, the printer will pick *one name* to
440 render all types of a particular shape. This means that if you have code
441 where two different source types end up having the same LLVM type, that
442 the dumper will sometimes print the "wrong" or unexpected type. This is
443 an important design point and isn't going to change.
450 Global variables define regions of memory allocated at compilation time
451 instead of run-time. Global variables may optionally be initialized, may
452 have an explicit section to be placed in, and may have an optional
453 explicit alignment specified.
455 A variable may be defined as ``thread_local``, which means that it will
456 not be shared by threads (each thread will have a separated copy of the
457 variable). Not all targets support thread-local variables. Optionally, a
458 TLS model may be specified:
461 For variables that are only used within the current shared library.
463 For variables in modules that will not be loaded dynamically.
465 For variables defined in the executable and only used within it.
467 The models correspond to the ELF TLS models; see `ELF Handling For
468 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
469 more information on under which circumstances the different models may
470 be used. The target may choose a different TLS model if the specified
471 model is not supported, or if a better choice of model can be made.
473 A variable may be defined as a global ``constant``, which indicates that
474 the contents of the variable will **never** be modified (enabling better
475 optimization, allowing the global data to be placed in the read-only
476 section of an executable, etc). Note that variables that need runtime
477 initialization cannot be marked ``constant`` as there is a store to the
480 LLVM explicitly allows *declarations* of global variables to be marked
481 constant, even if the final definition of the global is not. This
482 capability can be used to enable slightly better optimization of the
483 program, but requires the language definition to guarantee that
484 optimizations based on the 'constantness' are valid for the translation
485 units that do not include the definition.
487 As SSA values, global variables define pointer values that are in scope
488 (i.e. they dominate) all basic blocks in the program. Global variables
489 always define a pointer to their "content" type because they describe a
490 region of memory, and all memory objects in LLVM are accessed through
493 Global variables can be marked with ``unnamed_addr`` which indicates
494 that the address is not significant, only the content. Constants marked
495 like this can be merged with other constants if they have the same
496 initializer. Note that a constant with significant address *can* be
497 merged with a ``unnamed_addr`` constant, the result being a constant
498 whose address is significant.
500 A global variable may be declared to reside in a target-specific
501 numbered address space. For targets that support them, address spaces
502 may affect how optimizations are performed and/or what target
503 instructions are used to access the variable. The default address space
504 is zero. The address space qualifier must precede any other attributes.
506 LLVM allows an explicit section to be specified for globals. If the
507 target supports it, it will emit globals to the section specified.
509 By default, global initializers are optimized by assuming that global
510 variables defined within the module are not modified from their
511 initial values before the start of the global initializer. This is
512 true even for variables potentially accessible from outside the
513 module, including those with external linkage or appearing in
514 ``@llvm.used``. This assumption may be suppressed by marking the
515 variable with ``externally_initialized``.
517 An explicit alignment may be specified for a global, which must be a
518 power of 2. If not present, or if the alignment is set to zero, the
519 alignment of the global is set by the target to whatever it feels
520 convenient. If an explicit alignment is specified, the global is forced
521 to have exactly that alignment. Targets and optimizers are not allowed
522 to over-align the global if the global has an assigned section. In this
523 case, the extra alignment could be observable: for example, code could
524 assume that the globals are densely packed in their section and try to
525 iterate over them as an array, alignment padding would break this
528 For example, the following defines a global in a numbered address space
529 with an initializer, section, and alignment:
533 @G = addrspace(5) constant float 1.0, section "foo", align 4
535 The following example defines a thread-local global with the
536 ``initialexec`` TLS model:
540 @G = thread_local(initialexec) global i32 0, align 4
542 .. _functionstructure:
547 LLVM function definitions consist of the "``define``" keyword, an
548 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
549 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
550 an optional ``unnamed_addr`` attribute, a return type, an optional
551 :ref:`parameter attribute <paramattrs>` for the return type, a function
552 name, a (possibly empty) argument list (each with optional :ref:`parameter
553 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
554 an optional section, an optional alignment, an optional :ref:`garbage
555 collector name <gc>`, an opening curly brace, a list of basic blocks,
556 and a closing curly brace.
558 LLVM function declarations consist of the "``declare``" keyword, an
559 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
560 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
561 an optional ``unnamed_addr`` attribute, a return type, an optional
562 :ref:`parameter attribute <paramattrs>` for the return type, a function
563 name, a possibly empty list of arguments, an optional alignment, and an
564 optional :ref:`garbage collector name <gc>`.
566 A function definition contains a list of basic blocks, forming the CFG
567 (Control Flow Graph) for the function. Each basic block may optionally
568 start with a label (giving the basic block a symbol table entry),
569 contains a list of instructions, and ends with a
570 :ref:`terminator <terminators>` instruction (such as a branch or function
571 return). If explicit label is not provided, a block is assigned an
572 implicit numbered label, using a next value from the same counter as used
573 for unnamed temporaries (:ref:`see above<identifiers>`). For example, if a
574 function entry block does not have explicit label, it will be assigned
575 label "%0", then first unnamed temporary in that block will be "%1", etc.
577 The first basic block in a function is special in two ways: it is
578 immediately executed on entrance to the function, and it is not allowed
579 to have predecessor basic blocks (i.e. there can not be any branches to
580 the entry block of a function). Because the block can have no
581 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
583 LLVM allows an explicit section to be specified for functions. If the
584 target supports it, it will emit functions to the section specified.
586 An explicit alignment may be specified for a function. If not present,
587 or if the alignment is set to zero, the alignment of the function is set
588 by the target to whatever it feels convenient. If an explicit alignment
589 is specified, the function is forced to have at least that much
590 alignment. All alignments must be a power of 2.
592 If the ``unnamed_addr`` attribute is given, the address is know to not
593 be significant and two identical functions can be merged.
597 define [linkage] [visibility]
599 <ResultType> @<FunctionName> ([argument list])
600 [fn Attrs] [section "name"] [align N]
608 Aliases act as "second name" for the aliasee value (which can be either
609 function, global variable, another alias or bitcast of global value).
610 Aliases may have an optional :ref:`linkage type <linkage>`, and an optional
611 :ref:`visibility style <visibility>`.
615 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
617 .. _namedmetadatastructure:
622 Named metadata is a collection of metadata. :ref:`Metadata
623 nodes <metadata>` (but not metadata strings) are the only valid
624 operands for a named metadata.
628 ; Some unnamed metadata nodes, which are referenced by the named metadata.
629 !0 = metadata !{metadata !"zero"}
630 !1 = metadata !{metadata !"one"}
631 !2 = metadata !{metadata !"two"}
633 !name = !{!0, !1, !2}
640 The return type and each parameter of a function type may have a set of
641 *parameter attributes* associated with them. Parameter attributes are
642 used to communicate additional information about the result or
643 parameters of a function. Parameter attributes are considered to be part
644 of the function, not of the function type, so functions with different
645 parameter attributes can have the same function type.
647 Parameter attributes are simple keywords that follow the type specified.
648 If multiple parameter attributes are needed, they are space separated.
653 declare i32 @printf(i8* noalias nocapture, ...)
654 declare i32 @atoi(i8 zeroext)
655 declare signext i8 @returns_signed_char()
657 Note that any attributes for the function result (``nounwind``,
658 ``readonly``) come immediately after the argument list.
660 Currently, only the following parameter attributes are defined:
663 This indicates to the code generator that the parameter or return
664 value should be zero-extended to the extent required by the target's
665 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
666 the caller (for a parameter) or the callee (for a return value).
668 This indicates to the code generator that the parameter or return
669 value should be sign-extended to the extent required by the target's
670 ABI (which is usually 32-bits) by the caller (for a parameter) or
671 the callee (for a return value).
673 This indicates that this parameter or return value should be treated
674 in a special target-dependent fashion during while emitting code for
675 a function call or return (usually, by putting it in a register as
676 opposed to memory, though some targets use it to distinguish between
677 two different kinds of registers). Use of this attribute is
680 This indicates that the pointer parameter should really be passed by
681 value to the function. The attribute implies that a hidden copy of
682 the pointee is made between the caller and the callee, so the callee
683 is unable to modify the value in the caller. This attribute is only
684 valid on LLVM pointer arguments. It is generally used to pass
685 structs and arrays by value, but is also valid on pointers to
686 scalars. The copy is considered to belong to the caller not the
687 callee (for example, ``readonly`` functions should not write to
688 ``byval`` parameters). This is not a valid attribute for return
691 The byval attribute also supports specifying an alignment with the
692 align attribute. It indicates the alignment of the stack slot to
693 form and the known alignment of the pointer specified to the call
694 site. If the alignment is not specified, then the code generator
695 makes a target-specific assumption.
698 This indicates that the pointer parameter specifies the address of a
699 structure that is the return value of the function in the source
700 program. This pointer must be guaranteed by the caller to be valid:
701 loads and stores to the structure may be assumed by the callee
702 not to trap and to be properly aligned. This may only be applied to
703 the first parameter. This is not a valid attribute for return
706 This indicates that pointer values :ref:`based <pointeraliasing>` on
707 the argument or return value do not alias pointer values which are
708 not *based* on it, ignoring certain "irrelevant" dependencies. For a
709 call to the parent function, dependencies between memory references
710 from before or after the call and from those during the call are
711 "irrelevant" to the ``noalias`` keyword for the arguments and return
712 value used in that call. The caller shares the responsibility with
713 the callee for ensuring that these requirements are met. For further
714 details, please see the discussion of the NoAlias response in `alias
715 analysis <AliasAnalysis.html#MustMayNo>`_.
717 Note that this definition of ``noalias`` is intentionally similar
718 to the definition of ``restrict`` in C99 for function arguments,
719 though it is slightly weaker.
721 For function return values, C99's ``restrict`` is not meaningful,
722 while LLVM's ``noalias`` is.
724 This indicates that the callee does not make any copies of the
725 pointer that outlive the callee itself. This is not a valid
726 attribute for return values.
731 This indicates that the pointer parameter can be excised using the
732 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
733 attribute for return values and can only be applied to one parameter.
736 This indicates that the function always returns the argument as its return
737 value. This is an optimization hint to the code generator when generating
738 the caller, allowing tail call optimization and omission of register saves
739 and restores in some cases; it is not checked or enforced when generating
740 the callee. The parameter and the function return type must be valid
741 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
742 valid attribute for return values and can only be applied to one parameter.
746 Garbage Collector Names
747 -----------------------
749 Each function may specify a garbage collector name, which is simply a
754 define void @f() gc "name" { ... }
756 The compiler declares the supported values of *name*. Specifying a
757 collector which will cause the compiler to alter its output in order to
758 support the named garbage collection algorithm.
765 Attribute groups are groups of attributes that are referenced by objects within
766 the IR. They are important for keeping ``.ll`` files readable, because a lot of
767 functions will use the same set of attributes. In the degenerative case of a
768 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
769 group will capture the important command line flags used to build that file.
771 An attribute group is a module-level object. To use an attribute group, an
772 object references the attribute group's ID (e.g. ``#37``). An object may refer
773 to more than one attribute group. In that situation, the attributes from the
774 different groups are merged.
776 Here is an example of attribute groups for a function that should always be
777 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
781 ; Target-independent attributes:
782 attributes #0 = { alwaysinline alignstack=4 }
784 ; Target-dependent attributes:
785 attributes #1 = { "no-sse" }
787 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
788 define void @f() #0 #1 { ... }
795 Function attributes are set to communicate additional information about
796 a function. Function attributes are considered to be part of the
797 function, not of the function type, so functions with different function
798 attributes can have the same function type.
800 Function attributes are simple keywords that follow the type specified.
801 If multiple attributes are needed, they are space separated. For
806 define void @f() noinline { ... }
807 define void @f() alwaysinline { ... }
808 define void @f() alwaysinline optsize { ... }
809 define void @f() optsize { ... }
812 This attribute indicates that, when emitting the prologue and
813 epilogue, the backend should forcibly align the stack pointer.
814 Specify the desired alignment, which must be a power of two, in
817 This attribute indicates that the inliner should attempt to inline
818 this function into callers whenever possible, ignoring any active
819 inlining size threshold for this caller.
821 This indicates that the callee function at a call site should be
822 recognized as a built-in function, even though the function's declaration
823 uses the ``nobuiltin`` attribute. This is only valid at call sites for
824 direct calls to functions which are declared with the ``nobuiltin``
827 This attribute indicates that this function is rarely called. When
828 computing edge weights, basic blocks post-dominated by a cold
829 function call are also considered to be cold; and, thus, given low
832 This attribute indicates that the source code contained a hint that
833 inlining this function is desirable (such as the "inline" keyword in
834 C/C++). It is just a hint; it imposes no requirements on the
837 This attribute suggests that optimization passes and code generator
838 passes make choices that keep the code size of this function as small
839 as possible and perform optimizations that may sacrifice runtime
840 performance in order to minimize the size of the generated code.
842 This attribute disables prologue / epilogue emission for the
843 function. This can have very system-specific consequences.
845 This indicates that the callee function at a call site is not recognized as
846 a built-in function. LLVM will retain the original call and not replace it
847 with equivalent code based on the semantics of the built-in function, unless
848 the call site uses the ``builtin`` attribute. This is valid at call sites
849 and on function declarations and definitions.
851 This attribute indicates that calls to the function cannot be
852 duplicated. A call to a ``noduplicate`` function may be moved
853 within its parent function, but may not be duplicated within
856 A function containing a ``noduplicate`` call may still
857 be an inlining candidate, provided that the call is not
858 duplicated by inlining. That implies that the function has
859 internal linkage and only has one call site, so the original
860 call is dead after inlining.
862 This attributes disables implicit floating point instructions.
864 This attribute indicates that the inliner should never inline this
865 function in any situation. This attribute may not be used together
866 with the ``alwaysinline`` attribute.
868 This attribute suppresses lazy symbol binding for the function. This
869 may make calls to the function faster, at the cost of extra program
870 startup time if the function is not called during program startup.
872 This attribute indicates that the code generator should not use a
873 red zone, even if the target-specific ABI normally permits it.
875 This function attribute indicates that the function never returns
876 normally. This produces undefined behavior at runtime if the
877 function ever does dynamically return.
879 This function attribute indicates that the function never returns
880 with an unwind or exceptional control flow. If the function does
881 unwind, its runtime behavior is undefined.
883 This function attribute indicates that the function is not optimized
884 by any optimization or code generator passes with the
885 exception of interprocedural optimization passes.
886 This attribute cannot be used together with the ``alwaysinline``
887 attribute; this attribute is also incompatible
888 with the ``minsize`` attribute and the ``optsize`` attribute.
890 The inliner should never inline this function in any situation.
891 Only functions with the ``alwaysinline`` attribute are valid
892 candidates for inlining inside the body of this function.
894 This attribute suggests that optimization passes and code generator
895 passes make choices that keep the code size of this function low,
896 and otherwise do optimizations specifically to reduce code size as
897 long as they do not significantly impact runtime performance.
899 On a function, this attribute indicates that the function computes its
900 result (or decides to unwind an exception) based strictly on its arguments,
901 without dereferencing any pointer arguments or otherwise accessing
902 any mutable state (e.g. memory, control registers, etc) visible to
903 caller functions. It does not write through any pointer arguments
904 (including ``byval`` arguments) and never changes any state visible
905 to callers. This means that it cannot unwind exceptions by calling
906 the ``C++`` exception throwing methods.
908 On an argument, this attribute indicates that the function does not
909 dereference that pointer argument, even though it may read or write the
910 memory that the pointer points to if accessed through other pointers.
912 On a function, this attribute indicates that the function does not write
913 through any pointer arguments (including ``byval`` arguments) or otherwise
914 modify any state (e.g. memory, control registers, etc) visible to
915 caller functions. It may dereference pointer arguments and read
916 state that may be set in the caller. A readonly function always
917 returns the same value (or unwinds an exception identically) when
918 called with the same set of arguments and global state. It cannot
919 unwind an exception by calling the ``C++`` exception throwing
922 On an argument, this attribute indicates that the function does not write
923 through this pointer argument, even though it may write to the memory that
924 the pointer points to.
926 This attribute indicates that this function can return twice. The C
927 ``setjmp`` is an example of such a function. The compiler disables
928 some optimizations (like tail calls) in the caller of these
931 This attribute indicates that AddressSanitizer checks
932 (dynamic address safety analysis) are enabled for this function.
934 This attribute indicates that MemorySanitizer checks (dynamic detection
935 of accesses to uninitialized memory) are enabled for this function.
937 This attribute indicates that ThreadSanitizer checks
938 (dynamic thread safety analysis) are enabled for this function.
940 This attribute indicates that the function should emit a stack
941 smashing protector. It is in the form of a "canary" --- a random value
942 placed on the stack before the local variables that's checked upon
943 return from the function to see if it has been overwritten. A
944 heuristic is used to determine if a function needs stack protectors
945 or not. The heuristic used will enable protectors for functions with:
947 - Character arrays larger than ``ssp-buffer-size`` (default 8).
948 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
949 - Calls to alloca() with variable sizes or constant sizes greater than
952 If a function that has an ``ssp`` attribute is inlined into a
953 function that doesn't have an ``ssp`` attribute, then the resulting
954 function will have an ``ssp`` attribute.
956 This attribute indicates that the function should *always* emit a
957 stack smashing protector. This overrides the ``ssp`` function
960 If a function that has an ``sspreq`` attribute is inlined into a
961 function that doesn't have an ``sspreq`` attribute or which has an
962 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
963 an ``sspreq`` attribute.
965 This attribute indicates that the function should emit a stack smashing
966 protector. This attribute causes a strong heuristic to be used when
967 determining if a function needs stack protectors. The strong heuristic
968 will enable protectors for functions with:
970 - Arrays of any size and type
971 - Aggregates containing an array of any size and type.
973 - Local variables that have had their address taken.
975 This overrides the ``ssp`` function attribute.
977 If a function that has an ``sspstrong`` attribute is inlined into a
978 function that doesn't have an ``sspstrong`` attribute, then the
979 resulting function will have an ``sspstrong`` attribute.
981 This attribute indicates that the ABI being targeted requires that
982 an unwind table entry be produce for this function even if we can
983 show that no exceptions passes by it. This is normally the case for
984 the ELF x86-64 abi, but it can be disabled for some compilation
989 Module-Level Inline Assembly
990 ----------------------------
992 Modules may contain "module-level inline asm" blocks, which corresponds
993 to the GCC "file scope inline asm" blocks. These blocks are internally
994 concatenated by LLVM and treated as a single unit, but may be separated
995 in the ``.ll`` file if desired. The syntax is very simple:
999 module asm "inline asm code goes here"
1000 module asm "more can go here"
1002 The strings can contain any character by escaping non-printable
1003 characters. The escape sequence used is simply "\\xx" where "xx" is the
1004 two digit hex code for the number.
1006 The inline asm code is simply printed to the machine code .s file when
1007 assembly code is generated.
1009 .. _langref_datalayout:
1014 A module may specify a target specific data layout string that specifies
1015 how data is to be laid out in memory. The syntax for the data layout is
1018 .. code-block:: llvm
1020 target datalayout = "layout specification"
1022 The *layout specification* consists of a list of specifications
1023 separated by the minus sign character ('-'). Each specification starts
1024 with a letter and may include other information after the letter to
1025 define some aspect of the data layout. The specifications accepted are
1029 Specifies that the target lays out data in big-endian form. That is,
1030 the bits with the most significance have the lowest address
1033 Specifies that the target lays out data in little-endian form. That
1034 is, the bits with the least significance have the lowest address
1037 Specifies the natural alignment of the stack in bits. Alignment
1038 promotion of stack variables is limited to the natural stack
1039 alignment to avoid dynamic stack realignment. The stack alignment
1040 must be a multiple of 8-bits. If omitted, the natural stack
1041 alignment defaults to "unspecified", which does not prevent any
1042 alignment promotions.
1043 ``p[n]:<size>:<abi>:<pref>``
1044 This specifies the *size* of a pointer and its ``<abi>`` and
1045 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1046 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
1047 preceding ``:`` should be omitted too. The address space, ``n`` is
1048 optional, and if not specified, denotes the default address space 0.
1049 The value of ``n`` must be in the range [1,2^23).
1050 ``i<size>:<abi>:<pref>``
1051 This specifies the alignment for an integer type of a given bit
1052 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1053 ``v<size>:<abi>:<pref>``
1054 This specifies the alignment for a vector type of a given bit
1056 ``f<size>:<abi>:<pref>``
1057 This specifies the alignment for a floating point type of a given bit
1058 ``<size>``. Only values of ``<size>`` that are supported by the target
1059 will work. 32 (float) and 64 (double) are supported on all targets; 80
1060 or 128 (different flavors of long double) are also supported on some
1062 ``a<size>:<abi>:<pref>``
1063 This specifies the alignment for an aggregate type of a given bit
1065 ``s<size>:<abi>:<pref>``
1066 This specifies the alignment for a stack object of a given bit
1068 ``n<size1>:<size2>:<size3>...``
1069 This specifies a set of native integer widths for the target CPU in
1070 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1071 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1072 this set are considered to support most general arithmetic operations
1075 When constructing the data layout for a given target, LLVM starts with a
1076 default set of specifications which are then (possibly) overridden by
1077 the specifications in the ``datalayout`` keyword. The default
1078 specifications are given in this list:
1080 - ``E`` - big endian
1081 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1082 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1083 same as the default address space.
1084 - ``S0`` - natural stack alignment is unspecified
1085 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1086 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1087 - ``i16:16:16`` - i16 is 16-bit aligned
1088 - ``i32:32:32`` - i32 is 32-bit aligned
1089 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1090 alignment of 64-bits
1091 - ``f16:16:16`` - half is 16-bit aligned
1092 - ``f32:32:32`` - float is 32-bit aligned
1093 - ``f64:64:64`` - double is 64-bit aligned
1094 - ``f128:128:128`` - quad is 128-bit aligned
1095 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1096 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1097 - ``a0:0:64`` - aggregates are 64-bit aligned
1099 When LLVM is determining the alignment for a given type, it uses the
1102 #. If the type sought is an exact match for one of the specifications,
1103 that specification is used.
1104 #. If no match is found, and the type sought is an integer type, then
1105 the smallest integer type that is larger than the bitwidth of the
1106 sought type is used. If none of the specifications are larger than
1107 the bitwidth then the largest integer type is used. For example,
1108 given the default specifications above, the i7 type will use the
1109 alignment of i8 (next largest) while both i65 and i256 will use the
1110 alignment of i64 (largest specified).
1111 #. If no match is found, and the type sought is a vector type, then the
1112 largest vector type that is smaller than the sought vector type will
1113 be used as a fall back. This happens because <128 x double> can be
1114 implemented in terms of 64 <2 x double>, for example.
1116 The function of the data layout string may not be what you expect.
1117 Notably, this is not a specification from the frontend of what alignment
1118 the code generator should use.
1120 Instead, if specified, the target data layout is required to match what
1121 the ultimate *code generator* expects. This string is used by the
1122 mid-level optimizers to improve code, and this only works if it matches
1123 what the ultimate code generator uses. If you would like to generate IR
1124 that does not embed this target-specific detail into the IR, then you
1125 don't have to specify the string. This will disable some optimizations
1126 that require precise layout information, but this also prevents those
1127 optimizations from introducing target specificity into the IR.
1129 .. _pointeraliasing:
1131 Pointer Aliasing Rules
1132 ----------------------
1134 Any memory access must be done through a pointer value associated with
1135 an address range of the memory access, otherwise the behavior is
1136 undefined. Pointer values are associated with address ranges according
1137 to the following rules:
1139 - A pointer value is associated with the addresses associated with any
1140 value it is *based* on.
1141 - An address of a global variable is associated with the address range
1142 of the variable's storage.
1143 - The result value of an allocation instruction is associated with the
1144 address range of the allocated storage.
1145 - A null pointer in the default address-space is associated with no
1147 - An integer constant other than zero or a pointer value returned from
1148 a function not defined within LLVM may be associated with address
1149 ranges allocated through mechanisms other than those provided by
1150 LLVM. Such ranges shall not overlap with any ranges of addresses
1151 allocated by mechanisms provided by LLVM.
1153 A pointer value is *based* on another pointer value according to the
1156 - A pointer value formed from a ``getelementptr`` operation is *based*
1157 on the first operand of the ``getelementptr``.
1158 - The result value of a ``bitcast`` is *based* on the operand of the
1160 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1161 values that contribute (directly or indirectly) to the computation of
1162 the pointer's value.
1163 - The "*based* on" relationship is transitive.
1165 Note that this definition of *"based"* is intentionally similar to the
1166 definition of *"based"* in C99, though it is slightly weaker.
1168 LLVM IR does not associate types with memory. The result type of a
1169 ``load`` merely indicates the size and alignment of the memory from
1170 which to load, as well as the interpretation of the value. The first
1171 operand type of a ``store`` similarly only indicates the size and
1172 alignment of the store.
1174 Consequently, type-based alias analysis, aka TBAA, aka
1175 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1176 :ref:`Metadata <metadata>` may be used to encode additional information
1177 which specialized optimization passes may use to implement type-based
1182 Volatile Memory Accesses
1183 ------------------------
1185 Certain memory accesses, such as :ref:`load <i_load>`'s,
1186 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1187 marked ``volatile``. The optimizers must not change the number of
1188 volatile operations or change their order of execution relative to other
1189 volatile operations. The optimizers *may* change the order of volatile
1190 operations relative to non-volatile operations. This is not Java's
1191 "volatile" and has no cross-thread synchronization behavior.
1193 IR-level volatile loads and stores cannot safely be optimized into
1194 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1195 flagged volatile. Likewise, the backend should never split or merge
1196 target-legal volatile load/store instructions.
1198 .. admonition:: Rationale
1200 Platforms may rely on volatile loads and stores of natively supported
1201 data width to be executed as single instruction. For example, in C
1202 this holds for an l-value of volatile primitive type with native
1203 hardware support, but not necessarily for aggregate types. The
1204 frontend upholds these expectations, which are intentionally
1205 unspecified in the IR. The rules above ensure that IR transformation
1206 do not violate the frontend's contract with the language.
1210 Memory Model for Concurrent Operations
1211 --------------------------------------
1213 The LLVM IR does not define any way to start parallel threads of
1214 execution or to register signal handlers. Nonetheless, there are
1215 platform-specific ways to create them, and we define LLVM IR's behavior
1216 in their presence. This model is inspired by the C++0x memory model.
1218 For a more informal introduction to this model, see the :doc:`Atomics`.
1220 We define a *happens-before* partial order as the least partial order
1223 - Is a superset of single-thread program order, and
1224 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1225 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1226 techniques, like pthread locks, thread creation, thread joining,
1227 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1228 Constraints <ordering>`).
1230 Note that program order does not introduce *happens-before* edges
1231 between a thread and signals executing inside that thread.
1233 Every (defined) read operation (load instructions, memcpy, atomic
1234 loads/read-modify-writes, etc.) R reads a series of bytes written by
1235 (defined) write operations (store instructions, atomic
1236 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1237 section, initialized globals are considered to have a write of the
1238 initializer which is atomic and happens before any other read or write
1239 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1240 may see any write to the same byte, except:
1242 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1243 write\ :sub:`2` happens before R\ :sub:`byte`, then
1244 R\ :sub:`byte` does not see write\ :sub:`1`.
1245 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1246 R\ :sub:`byte` does not see write\ :sub:`3`.
1248 Given that definition, R\ :sub:`byte` is defined as follows:
1250 - If R is volatile, the result is target-dependent. (Volatile is
1251 supposed to give guarantees which can support ``sig_atomic_t`` in
1252 C/C++, and may be used for accesses to addresses which do not behave
1253 like normal memory. It does not generally provide cross-thread
1255 - Otherwise, if there is no write to the same byte that happens before
1256 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1257 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1258 R\ :sub:`byte` returns the value written by that write.
1259 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1260 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1261 Memory Ordering Constraints <ordering>` section for additional
1262 constraints on how the choice is made.
1263 - Otherwise R\ :sub:`byte` returns ``undef``.
1265 R returns the value composed of the series of bytes it read. This
1266 implies that some bytes within the value may be ``undef`` **without**
1267 the entire value being ``undef``. Note that this only defines the
1268 semantics of the operation; it doesn't mean that targets will emit more
1269 than one instruction to read the series of bytes.
1271 Note that in cases where none of the atomic intrinsics are used, this
1272 model places only one restriction on IR transformations on top of what
1273 is required for single-threaded execution: introducing a store to a byte
1274 which might not otherwise be stored is not allowed in general.
1275 (Specifically, in the case where another thread might write to and read
1276 from an address, introducing a store can change a load that may see
1277 exactly one write into a load that may see multiple writes.)
1281 Atomic Memory Ordering Constraints
1282 ----------------------------------
1284 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1285 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1286 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1287 an ordering parameter that determines which other atomic instructions on
1288 the same address they *synchronize with*. These semantics are borrowed
1289 from Java and C++0x, but are somewhat more colloquial. If these
1290 descriptions aren't precise enough, check those specs (see spec
1291 references in the :doc:`atomics guide <Atomics>`).
1292 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1293 differently since they don't take an address. See that instruction's
1294 documentation for details.
1296 For a simpler introduction to the ordering constraints, see the
1300 The set of values that can be read is governed by the happens-before
1301 partial order. A value cannot be read unless some operation wrote
1302 it. This is intended to provide a guarantee strong enough to model
1303 Java's non-volatile shared variables. This ordering cannot be
1304 specified for read-modify-write operations; it is not strong enough
1305 to make them atomic in any interesting way.
1307 In addition to the guarantees of ``unordered``, there is a single
1308 total order for modifications by ``monotonic`` operations on each
1309 address. All modification orders must be compatible with the
1310 happens-before order. There is no guarantee that the modification
1311 orders can be combined to a global total order for the whole program
1312 (and this often will not be possible). The read in an atomic
1313 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1314 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1315 order immediately before the value it writes. If one atomic read
1316 happens before another atomic read of the same address, the later
1317 read must see the same value or a later value in the address's
1318 modification order. This disallows reordering of ``monotonic`` (or
1319 stronger) operations on the same address. If an address is written
1320 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1321 read that address repeatedly, the other threads must eventually see
1322 the write. This corresponds to the C++0x/C1x
1323 ``memory_order_relaxed``.
1325 In addition to the guarantees of ``monotonic``, a
1326 *synchronizes-with* edge may be formed with a ``release`` operation.
1327 This is intended to model C++'s ``memory_order_acquire``.
1329 In addition to the guarantees of ``monotonic``, if this operation
1330 writes a value which is subsequently read by an ``acquire``
1331 operation, it *synchronizes-with* that operation. (This isn't a
1332 complete description; see the C++0x definition of a release
1333 sequence.) This corresponds to the C++0x/C1x
1334 ``memory_order_release``.
1335 ``acq_rel`` (acquire+release)
1336 Acts as both an ``acquire`` and ``release`` operation on its
1337 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1338 ``seq_cst`` (sequentially consistent)
1339 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1340 operation which only reads, ``release`` for an operation which only
1341 writes), there is a global total order on all
1342 sequentially-consistent operations on all addresses, which is
1343 consistent with the *happens-before* partial order and with the
1344 modification orders of all the affected addresses. Each
1345 sequentially-consistent read sees the last preceding write to the
1346 same address in this global order. This corresponds to the C++0x/C1x
1347 ``memory_order_seq_cst`` and Java volatile.
1351 If an atomic operation is marked ``singlethread``, it only *synchronizes
1352 with* or participates in modification and seq\_cst total orderings with
1353 other operations running in the same thread (for example, in signal
1361 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1362 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1363 :ref:`frem <i_frem>`) have the following flags that can set to enable
1364 otherwise unsafe floating point operations
1367 No NaNs - Allow optimizations to assume the arguments and result are not
1368 NaN. Such optimizations are required to retain defined behavior over
1369 NaNs, but the value of the result is undefined.
1372 No Infs - Allow optimizations to assume the arguments and result are not
1373 +/-Inf. Such optimizations are required to retain defined behavior over
1374 +/-Inf, but the value of the result is undefined.
1377 No Signed Zeros - Allow optimizations to treat the sign of a zero
1378 argument or result as insignificant.
1381 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1382 argument rather than perform division.
1385 Fast - Allow algebraically equivalent transformations that may
1386 dramatically change results in floating point (e.g. reassociate). This
1387 flag implies all the others.
1394 The LLVM type system is one of the most important features of the
1395 intermediate representation. Being typed enables a number of
1396 optimizations to be performed on the intermediate representation
1397 directly, without having to do extra analyses on the side before the
1398 transformation. A strong type system makes it easier to read the
1399 generated code and enables novel analyses and transformations that are
1400 not feasible to perform on normal three address code representations.
1402 .. _typeclassifications:
1404 Type Classifications
1405 --------------------
1407 The types fall into a few useful classifications:
1416 * - :ref:`integer <t_integer>`
1417 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1420 * - :ref:`floating point <t_floating>`
1421 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1429 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1430 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1431 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1432 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1434 * - :ref:`primitive <t_primitive>`
1435 - :ref:`label <t_label>`,
1436 :ref:`void <t_void>`,
1437 :ref:`integer <t_integer>`,
1438 :ref:`floating point <t_floating>`,
1439 :ref:`x86mmx <t_x86mmx>`,
1440 :ref:`metadata <t_metadata>`.
1442 * - :ref:`derived <t_derived>`
1443 - :ref:`array <t_array>`,
1444 :ref:`function <t_function>`,
1445 :ref:`pointer <t_pointer>`,
1446 :ref:`structure <t_struct>`,
1447 :ref:`vector <t_vector>`,
1448 :ref:`opaque <t_opaque>`.
1450 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1451 Values of these types are the only ones which can be produced by
1459 The primitive types are the fundamental building blocks of the LLVM
1470 The integer type is a very simple type that simply specifies an
1471 arbitrary bit width for the integer type desired. Any bit width from 1
1472 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1481 The number of bits the integer will occupy is specified by the ``N``
1487 +----------------+------------------------------------------------+
1488 | ``i1`` | a single-bit integer. |
1489 +----------------+------------------------------------------------+
1490 | ``i32`` | a 32-bit integer. |
1491 +----------------+------------------------------------------------+
1492 | ``i1942652`` | a really big integer of over 1 million bits. |
1493 +----------------+------------------------------------------------+
1497 Floating Point Types
1498 ^^^^^^^^^^^^^^^^^^^^
1507 - 16-bit floating point value
1510 - 32-bit floating point value
1513 - 64-bit floating point value
1516 - 128-bit floating point value (112-bit mantissa)
1519 - 80-bit floating point value (X87)
1522 - 128-bit floating point value (two 64-bits)
1532 The x86mmx type represents a value held in an MMX register on an x86
1533 machine. The operations allowed on it are quite limited: parameters and
1534 return values, load and store, and bitcast. User-specified MMX
1535 instructions are represented as intrinsic or asm calls with arguments
1536 and/or results of this type. There are no arrays, vectors or constants
1554 The void type does not represent any value and has no size.
1571 The label type represents code labels.
1588 The metadata type represents embedded metadata. No derived types may be
1589 created from metadata except for :ref:`function <t_function>` arguments.
1603 The real power in LLVM comes from the derived types in the system. This
1604 is what allows a programmer to represent arrays, functions, pointers,
1605 and other useful types. Each of these types contain one or more element
1606 types which may be a primitive type, or another derived type. For
1607 example, it is possible to have a two dimensional array, using an array
1608 as the element type of another array.
1615 Aggregate Types are a subset of derived types that can contain multiple
1616 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1617 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1628 The array type is a very simple derived type that arranges elements
1629 sequentially in memory. The array type requires a size (number of
1630 elements) and an underlying data type.
1637 [<# elements> x <elementtype>]
1639 The number of elements is a constant integer value; ``elementtype`` may
1640 be any type with a size.
1645 +------------------+--------------------------------------+
1646 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
1647 +------------------+--------------------------------------+
1648 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
1649 +------------------+--------------------------------------+
1650 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
1651 +------------------+--------------------------------------+
1653 Here are some examples of multidimensional arrays:
1655 +-----------------------------+----------------------------------------------------------+
1656 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1657 +-----------------------------+----------------------------------------------------------+
1658 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1659 +-----------------------------+----------------------------------------------------------+
1660 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1661 +-----------------------------+----------------------------------------------------------+
1663 There is no restriction on indexing beyond the end of the array implied
1664 by a static type (though there are restrictions on indexing beyond the
1665 bounds of an allocated object in some cases). This means that
1666 single-dimension 'variable sized array' addressing can be implemented in
1667 LLVM with a zero length array type. An implementation of 'pascal style
1668 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1679 The function type can be thought of as a function signature. It consists
1680 of a return type and a list of formal parameter types. The return type
1681 of a function type is a first class type or a void type.
1688 <returntype> (<parameter list>)
1690 ...where '``<parameter list>``' is a comma-separated list of type
1691 specifiers. Optionally, the parameter list may include a type ``...``,
1692 which indicates that the function takes a variable number of arguments.
1693 Variable argument functions can access their arguments with the
1694 :ref:`variable argument handling intrinsic <int_varargs>` functions.
1695 '``<returntype>``' is any type except :ref:`label <t_label>`.
1700 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1701 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1702 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1703 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1704 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1705 | ``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. |
1706 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1707 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1708 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1718 The structure type is used to represent a collection of data members
1719 together in memory. The elements of a structure may be any type that has
1722 Structures in memory are accessed using '``load``' and '``store``' by
1723 getting a pointer to a field with the '``getelementptr``' instruction.
1724 Structures in registers are accessed using the '``extractvalue``' and
1725 '``insertvalue``' instructions.
1727 Structures may optionally be "packed" structures, which indicate that
1728 the alignment of the struct is one byte, and that there is no padding
1729 between the elements. In non-packed structs, padding between field types
1730 is inserted as defined by the DataLayout string in the module, which is
1731 required to match what the underlying code generator expects.
1733 Structures can either be "literal" or "identified". A literal structure
1734 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1735 identified types are always defined at the top level with a name.
1736 Literal types are uniqued by their contents and can never be recursive
1737 or opaque since there is no way to write one. Identified types can be
1738 recursive, can be opaqued, and are never uniqued.
1745 %T1 = type { <type list> } ; Identified normal struct type
1746 %T2 = type <{ <type list> }> ; Identified packed struct type
1751 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1752 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1753 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1754 | ``{ 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``. |
1755 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1756 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1757 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1761 Opaque Structure Types
1762 ^^^^^^^^^^^^^^^^^^^^^^
1767 Opaque structure types are used to represent named structure types that
1768 do not have a body specified. This corresponds (for example) to the C
1769 notion of a forward declared structure.
1782 +--------------+-------------------+
1783 | ``opaque`` | An opaque type. |
1784 +--------------+-------------------+
1794 The pointer type is used to specify memory locations. Pointers are
1795 commonly used to reference objects in memory.
1797 Pointer types may have an optional address space attribute defining the
1798 numbered address space where the pointed-to object resides. The default
1799 address space is number zero. The semantics of non-zero address spaces
1800 are target-specific.
1802 Note that LLVM does not permit pointers to void (``void*``) nor does it
1803 permit pointers to labels (``label*``). Use ``i8*`` instead.
1815 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1816 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1817 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1818 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1819 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1820 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1821 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1831 A vector type is a simple derived type that represents a vector of
1832 elements. Vector types are used when multiple primitive data are
1833 operated in parallel using a single instruction (SIMD). A vector type
1834 requires a size (number of elements) and an underlying primitive data
1835 type. Vector types are considered :ref:`first class <t_firstclass>`.
1842 < <# elements> x <elementtype> >
1844 The number of elements is a constant integer value larger than 0;
1845 elementtype may be any integer or floating point type, or a pointer to
1846 these types. Vectors of size zero are not allowed.
1851 +-------------------+--------------------------------------------------+
1852 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1853 +-------------------+--------------------------------------------------+
1854 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1855 +-------------------+--------------------------------------------------+
1856 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1857 +-------------------+--------------------------------------------------+
1858 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1859 +-------------------+--------------------------------------------------+
1864 LLVM has several different basic types of constants. This section
1865 describes them all and their syntax.
1870 **Boolean constants**
1871 The two strings '``true``' and '``false``' are both valid constants
1873 **Integer constants**
1874 Standard integers (such as '4') are constants of the
1875 :ref:`integer <t_integer>` type. Negative numbers may be used with
1877 **Floating point constants**
1878 Floating point constants use standard decimal notation (e.g.
1879 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1880 hexadecimal notation (see below). The assembler requires the exact
1881 decimal value of a floating-point constant. For example, the
1882 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1883 decimal in binary. Floating point constants must have a :ref:`floating
1884 point <t_floating>` type.
1885 **Null pointer constants**
1886 The identifier '``null``' is recognized as a null pointer constant
1887 and must be of :ref:`pointer type <t_pointer>`.
1889 The one non-intuitive notation for constants is the hexadecimal form of
1890 floating point constants. For example, the form
1891 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1892 than) '``double 4.5e+15``'. The only time hexadecimal floating point
1893 constants are required (and the only time that they are generated by the
1894 disassembler) is when a floating point constant must be emitted but it
1895 cannot be represented as a decimal floating point number in a reasonable
1896 number of digits. For example, NaN's, infinities, and other special
1897 values are represented in their IEEE hexadecimal format so that assembly
1898 and disassembly do not cause any bits to change in the constants.
1900 When using the hexadecimal form, constants of types half, float, and
1901 double are represented using the 16-digit form shown above (which
1902 matches the IEEE754 representation for double); half and float values
1903 must, however, be exactly representable as IEEE 754 half and single
1904 precision, respectively. Hexadecimal format is always used for long
1905 double, and there are three forms of long double. The 80-bit format used
1906 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1907 128-bit format used by PowerPC (two adjacent doubles) is represented by
1908 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1909 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
1910 will only work if they match the long double format on your target.
1911 The IEEE 16-bit format (half precision) is represented by ``0xH``
1912 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
1913 (sign bit at the left).
1915 There are no constants of type x86mmx.
1917 .. _complexconstants:
1922 Complex constants are a (potentially recursive) combination of simple
1923 constants and smaller complex constants.
1925 **Structure constants**
1926 Structure constants are represented with notation similar to
1927 structure type definitions (a comma separated list of elements,
1928 surrounded by braces (``{}``)). For example:
1929 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1930 "``@G = external global i32``". Structure constants must have
1931 :ref:`structure type <t_struct>`, and the number and types of elements
1932 must match those specified by the type.
1934 Array constants are represented with notation similar to array type
1935 definitions (a comma separated list of elements, surrounded by
1936 square brackets (``[]``)). For example:
1937 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1938 :ref:`array type <t_array>`, and the number and types of elements must
1939 match those specified by the type.
1940 **Vector constants**
1941 Vector constants are represented with notation similar to vector
1942 type definitions (a comma separated list of elements, surrounded by
1943 less-than/greater-than's (``<>``)). For example:
1944 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1945 must have :ref:`vector type <t_vector>`, and the number and types of
1946 elements must match those specified by the type.
1947 **Zero initialization**
1948 The string '``zeroinitializer``' can be used to zero initialize a
1949 value to zero of *any* type, including scalar and
1950 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1951 having to print large zero initializers (e.g. for large arrays) and
1952 is always exactly equivalent to using explicit zero initializers.
1954 A metadata node is a structure-like constant with :ref:`metadata
1955 type <t_metadata>`. For example:
1956 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
1957 constants that are meant to be interpreted as part of the
1958 instruction stream, metadata is a place to attach additional
1959 information such as debug info.
1961 Global Variable and Function Addresses
1962 --------------------------------------
1964 The addresses of :ref:`global variables <globalvars>` and
1965 :ref:`functions <functionstructure>` are always implicitly valid
1966 (link-time) constants. These constants are explicitly referenced when
1967 the :ref:`identifier for the global <identifiers>` is used and always have
1968 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
1971 .. code-block:: llvm
1975 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1982 The string '``undef``' can be used anywhere a constant is expected, and
1983 indicates that the user of the value may receive an unspecified
1984 bit-pattern. Undefined values may be of any type (other than '``label``'
1985 or '``void``') and be used anywhere a constant is permitted.
1987 Undefined values are useful because they indicate to the compiler that
1988 the program is well defined no matter what value is used. This gives the
1989 compiler more freedom to optimize. Here are some examples of
1990 (potentially surprising) transformations that are valid (in pseudo IR):
1992 .. code-block:: llvm
2002 This is safe because all of the output bits are affected by the undef
2003 bits. Any output bit can have a zero or one depending on the input bits.
2005 .. code-block:: llvm
2016 These logical operations have bits that are not always affected by the
2017 input. For example, if ``%X`` has a zero bit, then the output of the
2018 '``and``' operation will always be a zero for that bit, no matter what
2019 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2020 optimize or assume that the result of the '``and``' is '``undef``'.
2021 However, it is safe to assume that all bits of the '``undef``' could be
2022 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2023 all the bits of the '``undef``' operand to the '``or``' could be set,
2024 allowing the '``or``' to be folded to -1.
2026 .. code-block:: llvm
2028 %A = select undef, %X, %Y
2029 %B = select undef, 42, %Y
2030 %C = select %X, %Y, undef
2040 This set of examples shows that undefined '``select``' (and conditional
2041 branch) conditions can go *either way*, but they have to come from one
2042 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2043 both known to have a clear low bit, then ``%A`` would have to have a
2044 cleared low bit. However, in the ``%C`` example, the optimizer is
2045 allowed to assume that the '``undef``' operand could be the same as
2046 ``%Y``, allowing the whole '``select``' to be eliminated.
2048 .. code-block:: llvm
2050 %A = xor undef, undef
2067 This example points out that two '``undef``' operands are not
2068 necessarily the same. This can be surprising to people (and also matches
2069 C semantics) where they assume that "``X^X``" is always zero, even if
2070 ``X`` is undefined. This isn't true for a number of reasons, but the
2071 short answer is that an '``undef``' "variable" can arbitrarily change
2072 its value over its "live range". This is true because the variable
2073 doesn't actually *have a live range*. Instead, the value is logically
2074 read from arbitrary registers that happen to be around when needed, so
2075 the value is not necessarily consistent over time. In fact, ``%A`` and
2076 ``%C`` need to have the same semantics or the core LLVM "replace all
2077 uses with" concept would not hold.
2079 .. code-block:: llvm
2087 These examples show the crucial difference between an *undefined value*
2088 and *undefined behavior*. An undefined value (like '``undef``') is
2089 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2090 operation can be constant folded to '``undef``', because the '``undef``'
2091 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2092 However, in the second example, we can make a more aggressive
2093 assumption: because the ``undef`` is allowed to be an arbitrary value,
2094 we are allowed to assume that it could be zero. Since a divide by zero
2095 has *undefined behavior*, we are allowed to assume that the operation
2096 does not execute at all. This allows us to delete the divide and all
2097 code after it. Because the undefined operation "can't happen", the
2098 optimizer can assume that it occurs in dead code.
2100 .. code-block:: llvm
2102 a: store undef -> %X
2103 b: store %X -> undef
2108 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2109 value can be assumed to not have any effect; we can assume that the
2110 value is overwritten with bits that happen to match what was already
2111 there. However, a store *to* an undefined location could clobber
2112 arbitrary memory, therefore, it has undefined behavior.
2119 Poison values are similar to :ref:`undef values <undefvalues>`, however
2120 they also represent the fact that an instruction or constant expression
2121 which cannot evoke side effects has nevertheless detected a condition
2122 which results in undefined behavior.
2124 There is currently no way of representing a poison value in the IR; they
2125 only exist when produced by operations such as :ref:`add <i_add>` with
2128 Poison value behavior is defined in terms of value *dependence*:
2130 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2131 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2132 their dynamic predecessor basic block.
2133 - Function arguments depend on the corresponding actual argument values
2134 in the dynamic callers of their functions.
2135 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2136 instructions that dynamically transfer control back to them.
2137 - :ref:`Invoke <i_invoke>` instructions depend on the
2138 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2139 call instructions that dynamically transfer control back to them.
2140 - Non-volatile loads and stores depend on the most recent stores to all
2141 of the referenced memory addresses, following the order in the IR
2142 (including loads and stores implied by intrinsics such as
2143 :ref:`@llvm.memcpy <int_memcpy>`.)
2144 - An instruction with externally visible side effects depends on the
2145 most recent preceding instruction with externally visible side
2146 effects, following the order in the IR. (This includes :ref:`volatile
2147 operations <volatile>`.)
2148 - An instruction *control-depends* on a :ref:`terminator
2149 instruction <terminators>` if the terminator instruction has
2150 multiple successors and the instruction is always executed when
2151 control transfers to one of the successors, and may not be executed
2152 when control is transferred to another.
2153 - Additionally, an instruction also *control-depends* on a terminator
2154 instruction if the set of instructions it otherwise depends on would
2155 be different if the terminator had transferred control to a different
2157 - Dependence is transitive.
2159 Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2160 with the additional affect that any instruction which has a *dependence*
2161 on a poison value has undefined behavior.
2163 Here are some examples:
2165 .. code-block:: llvm
2168 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2169 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2170 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2171 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2173 store i32 %poison, i32* @g ; Poison value stored to memory.
2174 %poison2 = load i32* @g ; Poison value loaded back from memory.
2176 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2178 %narrowaddr = bitcast i32* @g to i16*
2179 %wideaddr = bitcast i32* @g to i64*
2180 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2181 %poison4 = load i64* %wideaddr ; Returns a poison value.
2183 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2184 br i1 %cmp, label %true, label %end ; Branch to either destination.
2187 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2188 ; it has undefined behavior.
2192 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2193 ; Both edges into this PHI are
2194 ; control-dependent on %cmp, so this
2195 ; always results in a poison value.
2197 store volatile i32 0, i32* @g ; This would depend on the store in %true
2198 ; if %cmp is true, or the store in %entry
2199 ; otherwise, so this is undefined behavior.
2201 br i1 %cmp, label %second_true, label %second_end
2202 ; The same branch again, but this time the
2203 ; true block doesn't have side effects.
2210 store volatile i32 0, i32* @g ; This time, the instruction always depends
2211 ; on the store in %end. Also, it is
2212 ; control-equivalent to %end, so this is
2213 ; well-defined (ignoring earlier undefined
2214 ; behavior in this example).
2218 Addresses of Basic Blocks
2219 -------------------------
2221 ``blockaddress(@function, %block)``
2223 The '``blockaddress``' constant computes the address of the specified
2224 basic block in the specified function, and always has an ``i8*`` type.
2225 Taking the address of the entry block is illegal.
2227 This value only has defined behavior when used as an operand to the
2228 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2229 against null. Pointer equality tests between labels addresses results in
2230 undefined behavior --- though, again, comparison against null is ok, and
2231 no label is equal to the null pointer. This may be passed around as an
2232 opaque pointer sized value as long as the bits are not inspected. This
2233 allows ``ptrtoint`` and arithmetic to be performed on these values so
2234 long as the original value is reconstituted before the ``indirectbr``
2237 Finally, some targets may provide defined semantics when using the value
2238 as the operand to an inline assembly, but that is target specific.
2242 Constant Expressions
2243 --------------------
2245 Constant expressions are used to allow expressions involving other
2246 constants to be used as constants. Constant expressions may be of any
2247 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2248 that does not have side effects (e.g. load and call are not supported).
2249 The following is the syntax for constant expressions:
2251 ``trunc (CST to TYPE)``
2252 Truncate a constant to another type. The bit size of CST must be
2253 larger than the bit size of TYPE. Both types must be integers.
2254 ``zext (CST to TYPE)``
2255 Zero extend a constant to another type. The bit size of CST must be
2256 smaller than the bit size of TYPE. Both types must be integers.
2257 ``sext (CST to TYPE)``
2258 Sign extend a constant to another type. The bit size of CST must be
2259 smaller than the bit size of TYPE. Both types must be integers.
2260 ``fptrunc (CST to TYPE)``
2261 Truncate a floating point constant to another floating point type.
2262 The size of CST must be larger than the size of TYPE. Both types
2263 must be floating point.
2264 ``fpext (CST to TYPE)``
2265 Floating point extend a constant to another type. The size of CST
2266 must be smaller or equal to the size of TYPE. Both types must be
2268 ``fptoui (CST to TYPE)``
2269 Convert a floating point constant to the corresponding unsigned
2270 integer constant. TYPE must be a scalar or vector integer type. CST
2271 must be of scalar or vector floating point type. Both CST and TYPE
2272 must be scalars, or vectors of the same number of elements. If the
2273 value won't fit in the integer type, the results are undefined.
2274 ``fptosi (CST to TYPE)``
2275 Convert a floating point constant to the corresponding signed
2276 integer constant. TYPE must be a scalar or vector integer type. CST
2277 must be of scalar or vector floating point type. Both CST and TYPE
2278 must be scalars, or vectors of the same number of elements. If the
2279 value won't fit in the integer type, the results are undefined.
2280 ``uitofp (CST to TYPE)``
2281 Convert an unsigned integer constant to the corresponding floating
2282 point constant. TYPE must be a scalar or vector floating point type.
2283 CST must be of scalar or vector integer type. Both CST and TYPE must
2284 be scalars, or vectors of the same number of elements. If the value
2285 won't fit in the floating point type, the results are undefined.
2286 ``sitofp (CST to TYPE)``
2287 Convert a signed integer constant to the corresponding floating
2288 point constant. TYPE must be a scalar or vector floating point type.
2289 CST must be of scalar or vector integer type. Both CST and TYPE must
2290 be scalars, or vectors of the same number of elements. If the value
2291 won't fit in the floating point type, the results are undefined.
2292 ``ptrtoint (CST to TYPE)``
2293 Convert a pointer typed constant to the corresponding integer
2294 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2295 pointer type. The ``CST`` value is zero extended, truncated, or
2296 unchanged to make it fit in ``TYPE``.
2297 ``inttoptr (CST to TYPE)``
2298 Convert an integer constant to a pointer constant. TYPE must be a
2299 pointer type. CST must be of integer type. The CST value is zero
2300 extended, truncated, or unchanged to make it fit in a pointer size.
2301 This one is *really* dangerous!
2302 ``bitcast (CST to TYPE)``
2303 Convert a constant, CST, to another TYPE. The constraints of the
2304 operands are the same as those for the :ref:`bitcast
2305 instruction <i_bitcast>`.
2306 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2307 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2308 constants. As with the :ref:`getelementptr <i_getelementptr>`
2309 instruction, the index list may have zero or more indexes, which are
2310 required to make sense for the type of "CSTPTR".
2311 ``select (COND, VAL1, VAL2)``
2312 Perform the :ref:`select operation <i_select>` on constants.
2313 ``icmp COND (VAL1, VAL2)``
2314 Performs the :ref:`icmp operation <i_icmp>` on constants.
2315 ``fcmp COND (VAL1, VAL2)``
2316 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2317 ``extractelement (VAL, IDX)``
2318 Perform the :ref:`extractelement operation <i_extractelement>` on
2320 ``insertelement (VAL, ELT, IDX)``
2321 Perform the :ref:`insertelement operation <i_insertelement>` on
2323 ``shufflevector (VEC1, VEC2, IDXMASK)``
2324 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2326 ``extractvalue (VAL, IDX0, IDX1, ...)``
2327 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2328 constants. The index list is interpreted in a similar manner as
2329 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2330 least one index value must be specified.
2331 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2332 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2333 The index list is interpreted in a similar manner as indices in a
2334 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2335 value must be specified.
2336 ``OPCODE (LHS, RHS)``
2337 Perform the specified operation of the LHS and RHS constants. OPCODE
2338 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2339 binary <bitwiseops>` operations. The constraints on operands are
2340 the same as those for the corresponding instruction (e.g. no bitwise
2341 operations on floating point values are allowed).
2348 Inline Assembler Expressions
2349 ----------------------------
2351 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2352 Inline Assembly <moduleasm>`) through the use of a special value. This
2353 value represents the inline assembler as a string (containing the
2354 instructions to emit), a list of operand constraints (stored as a
2355 string), a flag that indicates whether or not the inline asm expression
2356 has side effects, and a flag indicating whether the function containing
2357 the asm needs to align its stack conservatively. An example inline
2358 assembler expression is:
2360 .. code-block:: llvm
2362 i32 (i32) asm "bswap $0", "=r,r"
2364 Inline assembler expressions may **only** be used as the callee operand
2365 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2366 Thus, typically we have:
2368 .. code-block:: llvm
2370 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2372 Inline asms with side effects not visible in the constraint list must be
2373 marked as having side effects. This is done through the use of the
2374 '``sideeffect``' keyword, like so:
2376 .. code-block:: llvm
2378 call void asm sideeffect "eieio", ""()
2380 In some cases inline asms will contain code that will not work unless
2381 the stack is aligned in some way, such as calls or SSE instructions on
2382 x86, yet will not contain code that does that alignment within the asm.
2383 The compiler should make conservative assumptions about what the asm
2384 might contain and should generate its usual stack alignment code in the
2385 prologue if the '``alignstack``' keyword is present:
2387 .. code-block:: llvm
2389 call void asm alignstack "eieio", ""()
2391 Inline asms also support using non-standard assembly dialects. The
2392 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2393 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2394 the only supported dialects. An example is:
2396 .. code-block:: llvm
2398 call void asm inteldialect "eieio", ""()
2400 If multiple keywords appear the '``sideeffect``' keyword must come
2401 first, the '``alignstack``' keyword second and the '``inteldialect``'
2407 The call instructions that wrap inline asm nodes may have a
2408 "``!srcloc``" MDNode attached to it that contains a list of constant
2409 integers. If present, the code generator will use the integer as the
2410 location cookie value when report errors through the ``LLVMContext``
2411 error reporting mechanisms. This allows a front-end to correlate backend
2412 errors that occur with inline asm back to the source code that produced
2415 .. code-block:: llvm
2417 call void asm sideeffect "something bad", ""(), !srcloc !42
2419 !42 = !{ i32 1234567 }
2421 It is up to the front-end to make sense of the magic numbers it places
2422 in the IR. If the MDNode contains multiple constants, the code generator
2423 will use the one that corresponds to the line of the asm that the error
2428 Metadata Nodes and Metadata Strings
2429 -----------------------------------
2431 LLVM IR allows metadata to be attached to instructions in the program
2432 that can convey extra information about the code to the optimizers and
2433 code generator. One example application of metadata is source-level
2434 debug information. There are two metadata primitives: strings and nodes.
2435 All metadata has the ``metadata`` type and is identified in syntax by a
2436 preceding exclamation point ('``!``').
2438 A metadata string is a string surrounded by double quotes. It can
2439 contain any character by escaping non-printable characters with
2440 "``\xx``" where "``xx``" is the two digit hex code. For example:
2443 Metadata nodes are represented with notation similar to structure
2444 constants (a comma separated list of elements, surrounded by braces and
2445 preceded by an exclamation point). Metadata nodes can have any values as
2446 their operand. For example:
2448 .. code-block:: llvm
2450 !{ metadata !"test\00", i32 10}
2452 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2453 metadata nodes, which can be looked up in the module symbol table. For
2456 .. code-block:: llvm
2458 !foo = metadata !{!4, !3}
2460 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2461 function is using two metadata arguments:
2463 .. code-block:: llvm
2465 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2467 Metadata can be attached with an instruction. Here metadata ``!21`` is
2468 attached to the ``add`` instruction using the ``!dbg`` identifier:
2470 .. code-block:: llvm
2472 %indvar.next = add i64 %indvar, 1, !dbg !21
2474 More information about specific metadata nodes recognized by the
2475 optimizers and code generator is found below.
2480 In LLVM IR, memory does not have types, so LLVM's own type system is not
2481 suitable for doing TBAA. Instead, metadata is added to the IR to
2482 describe a type system of a higher level language. This can be used to
2483 implement typical C/C++ TBAA, but it can also be used to implement
2484 custom alias analysis behavior for other languages.
2486 The current metadata format is very simple. TBAA metadata nodes have up
2487 to three fields, e.g.:
2489 .. code-block:: llvm
2491 !0 = metadata !{ metadata !"an example type tree" }
2492 !1 = metadata !{ metadata !"int", metadata !0 }
2493 !2 = metadata !{ metadata !"float", metadata !0 }
2494 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2496 The first field is an identity field. It can be any value, usually a
2497 metadata string, which uniquely identifies the type. The most important
2498 name in the tree is the name of the root node. Two trees with different
2499 root node names are entirely disjoint, even if they have leaves with
2502 The second field identifies the type's parent node in the tree, or is
2503 null or omitted for a root node. A type is considered to alias all of
2504 its descendants and all of its ancestors in the tree. Also, a type is
2505 considered to alias all types in other trees, so that bitcode produced
2506 from multiple front-ends is handled conservatively.
2508 If the third field is present, it's an integer which if equal to 1
2509 indicates that the type is "constant" (meaning
2510 ``pointsToConstantMemory`` should return true; see `other useful
2511 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2513 '``tbaa.struct``' Metadata
2514 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2516 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2517 aggregate assignment operations in C and similar languages, however it
2518 is defined to copy a contiguous region of memory, which is more than
2519 strictly necessary for aggregate types which contain holes due to
2520 padding. Also, it doesn't contain any TBAA information about the fields
2523 ``!tbaa.struct`` metadata can describe which memory subregions in a
2524 memcpy are padding and what the TBAA tags of the struct are.
2526 The current metadata format is very simple. ``!tbaa.struct`` metadata
2527 nodes are a list of operands which are in conceptual groups of three.
2528 For each group of three, the first operand gives the byte offset of a
2529 field in bytes, the second gives its size in bytes, and the third gives
2532 .. code-block:: llvm
2534 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2536 This describes a struct with two fields. The first is at offset 0 bytes
2537 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2538 and has size 4 bytes and has tbaa tag !2.
2540 Note that the fields need not be contiguous. In this example, there is a
2541 4 byte gap between the two fields. This gap represents padding which
2542 does not carry useful data and need not be preserved.
2544 '``fpmath``' Metadata
2545 ^^^^^^^^^^^^^^^^^^^^^
2547 ``fpmath`` metadata may be attached to any instruction of floating point
2548 type. It can be used to express the maximum acceptable error in the
2549 result of that instruction, in ULPs, thus potentially allowing the
2550 compiler to use a more efficient but less accurate method of computing
2551 it. ULP is defined as follows:
2553 If ``x`` is a real number that lies between two finite consecutive
2554 floating-point numbers ``a`` and ``b``, without being equal to one
2555 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2556 distance between the two non-equal finite floating-point numbers
2557 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2559 The metadata node shall consist of a single positive floating point
2560 number representing the maximum relative error, for example:
2562 .. code-block:: llvm
2564 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2566 '``range``' Metadata
2567 ^^^^^^^^^^^^^^^^^^^^
2569 ``range`` metadata may be attached only to loads of integer types. It
2570 expresses the possible ranges the loaded value is in. The ranges are
2571 represented with a flattened list of integers. The loaded value is known
2572 to be in the union of the ranges defined by each consecutive pair. Each
2573 pair has the following properties:
2575 - The type must match the type loaded by the instruction.
2576 - The pair ``a,b`` represents the range ``[a,b)``.
2577 - Both ``a`` and ``b`` are constants.
2578 - The range is allowed to wrap.
2579 - The range should not represent the full or empty set. That is,
2582 In addition, the pairs must be in signed order of the lower bound and
2583 they must be non-contiguous.
2587 .. code-block:: llvm
2589 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2590 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2591 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2592 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2594 !0 = metadata !{ i8 0, i8 2 }
2595 !1 = metadata !{ i8 255, i8 2 }
2596 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2597 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2602 It is sometimes useful to attach information to loop constructs. Currently,
2603 loop metadata is implemented as metadata attached to the branch instruction
2604 in the loop latch block. This type of metadata refer to a metadata node that is
2605 guaranteed to be separate for each loop. The loop identifier metadata is
2606 specified with the name ``llvm.loop``.
2608 The loop identifier metadata is implemented using a metadata that refers to
2609 itself to avoid merging it with any other identifier metadata, e.g.,
2610 during module linkage or function inlining. That is, each loop should refer
2611 to their own identification metadata even if they reside in separate functions.
2612 The following example contains loop identifier metadata for two separate loop
2615 .. code-block:: llvm
2617 !0 = metadata !{ metadata !0 }
2618 !1 = metadata !{ metadata !1 }
2620 The loop identifier metadata can be used to specify additional per-loop
2621 metadata. Any operands after the first operand can be treated as user-defined
2622 metadata. For example the ``llvm.vectorizer.unroll`` metadata is understood
2623 by the loop vectorizer to indicate how many times to unroll the loop:
2625 .. code-block:: llvm
2627 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
2629 !0 = metadata !{ metadata !0, metadata !1 }
2630 !1 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 2 }
2635 Metadata types used to annotate memory accesses with information helpful
2636 for optimizations are prefixed with ``llvm.mem``.
2638 '``llvm.mem.parallel_loop_access``' Metadata
2639 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2641 For a loop to be parallel, in addition to using
2642 the ``llvm.loop`` metadata to mark the loop latch branch instruction,
2643 also all of the memory accessing instructions in the loop body need to be
2644 marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2645 is at least one memory accessing instruction not marked with the metadata,
2646 the loop must be considered a sequential loop. This causes parallel loops to be
2647 converted to sequential loops due to optimization passes that are unaware of
2648 the parallel semantics and that insert new memory instructions to the loop
2651 Example of a loop that is considered parallel due to its correct use of
2652 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
2653 metadata types that refer to the same loop identifier metadata.
2655 .. code-block:: llvm
2659 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2661 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2663 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
2667 !0 = metadata !{ metadata !0 }
2669 It is also possible to have nested parallel loops. In that case the
2670 memory accesses refer to a list of loop identifier metadata nodes instead of
2671 the loop identifier metadata node directly:
2673 .. code-block:: llvm
2680 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2682 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2684 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
2688 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2690 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2692 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
2694 outer.for.end: ; preds = %for.body
2696 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
2697 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
2698 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
2700 '``llvm.vectorizer``'
2701 ^^^^^^^^^^^^^^^^^^^^^
2703 Metadata prefixed with ``llvm.vectorizer`` is used to control per-loop
2704 vectorization parameters such as vectorization factor and unroll factor.
2706 ``llvm.vectorizer`` metadata should be used in conjunction with ``llvm.loop``
2707 loop identification metadata.
2709 '``llvm.vectorizer.unroll``' Metadata
2710 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2712 This metadata instructs the loop vectorizer to unroll the specified
2713 loop exactly ``N`` times.
2715 The first operand is the string ``llvm.vectorizer.unroll`` and the second
2716 operand is an integer specifying the unroll factor. For example:
2718 .. code-block:: llvm
2720 !0 = metadata !{ metadata !"llvm.vectorizer.unroll", i32 4 }
2722 Note that setting ``llvm.vectorizer.unroll`` to 1 disables unrolling of the
2725 If ``llvm.vectorizer.unroll`` is set to 0 then the amount of unrolling will be
2726 determined automatically.
2728 '``llvm.vectorizer.width``' Metadata
2729 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2731 This metadata sets the target width of the vectorizer to ``N``. Without
2732 this metadata, the vectorizer will choose a width automatically.
2733 Regardless of this metadata, the vectorizer will only vectorize loops if
2734 it believes it is valid to do so.
2736 The first operand is the string ``llvm.vectorizer.width`` and the second
2737 operand is an integer specifying the width. For example:
2739 .. code-block:: llvm
2741 !0 = metadata !{ metadata !"llvm.vectorizer.width", i32 4 }
2743 Note that setting ``llvm.vectorizer.width`` to 1 disables vectorization of the
2746 If ``llvm.vectorizer.width`` is set to 0 then the width will be determined
2749 Module Flags Metadata
2750 =====================
2752 Information about the module as a whole is difficult to convey to LLVM's
2753 subsystems. The LLVM IR isn't sufficient to transmit this information.
2754 The ``llvm.module.flags`` named metadata exists in order to facilitate
2755 this. These flags are in the form of key / value pairs --- much like a
2756 dictionary --- making it easy for any subsystem who cares about a flag to
2759 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2760 Each triplet has the following form:
2762 - The first element is a *behavior* flag, which specifies the behavior
2763 when two (or more) modules are merged together, and it encounters two
2764 (or more) metadata with the same ID. The supported behaviors are
2766 - The second element is a metadata string that is a unique ID for the
2767 metadata. Each module may only have one flag entry for each unique ID (not
2768 including entries with the **Require** behavior).
2769 - The third element is the value of the flag.
2771 When two (or more) modules are merged together, the resulting
2772 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2773 each unique metadata ID string, there will be exactly one entry in the merged
2774 modules ``llvm.module.flags`` metadata table, and the value for that entry will
2775 be determined by the merge behavior flag, as described below. The only exception
2776 is that entries with the *Require* behavior are always preserved.
2778 The following behaviors are supported:
2789 Emits an error if two values disagree, otherwise the resulting value
2790 is that of the operands.
2794 Emits a warning if two values disagree. The result value will be the
2795 operand for the flag from the first module being linked.
2799 Adds a requirement that another module flag be present and have a
2800 specified value after linking is performed. The value must be a
2801 metadata pair, where the first element of the pair is the ID of the
2802 module flag to be restricted, and the second element of the pair is
2803 the value the module flag should be restricted to. This behavior can
2804 be used to restrict the allowable results (via triggering of an
2805 error) of linking IDs with the **Override** behavior.
2809 Uses the specified value, regardless of the behavior or value of the
2810 other module. If both modules specify **Override**, but the values
2811 differ, an error will be emitted.
2815 Appends the two values, which are required to be metadata nodes.
2819 Appends the two values, which are required to be metadata
2820 nodes. However, duplicate entries in the second list are dropped
2821 during the append operation.
2823 It is an error for a particular unique flag ID to have multiple behaviors,
2824 except in the case of **Require** (which adds restrictions on another metadata
2825 value) or **Override**.
2827 An example of module flags:
2829 .. code-block:: llvm
2831 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2832 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2833 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2834 !3 = metadata !{ i32 3, metadata !"qux",
2836 metadata !"foo", i32 1
2839 !llvm.module.flags = !{ !0, !1, !2, !3 }
2841 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2842 if two or more ``!"foo"`` flags are seen is to emit an error if their
2843 values are not equal.
2845 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2846 behavior if two or more ``!"bar"`` flags are seen is to use the value
2849 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2850 behavior if two or more ``!"qux"`` flags are seen is to emit a
2851 warning if their values are not equal.
2853 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
2857 metadata !{ metadata !"foo", i32 1 }
2859 The behavior is to emit an error if the ``llvm.module.flags`` does not
2860 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2863 Objective-C Garbage Collection Module Flags Metadata
2864 ----------------------------------------------------
2866 On the Mach-O platform, Objective-C stores metadata about garbage
2867 collection in a special section called "image info". The metadata
2868 consists of a version number and a bitmask specifying what types of
2869 garbage collection are supported (if any) by the file. If two or more
2870 modules are linked together their garbage collection metadata needs to
2871 be merged rather than appended together.
2873 The Objective-C garbage collection module flags metadata consists of the
2874 following key-value pairs:
2883 * - ``Objective-C Version``
2884 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
2886 * - ``Objective-C Image Info Version``
2887 - **[Required]** --- The version of the image info section. Currently
2890 * - ``Objective-C Image Info Section``
2891 - **[Required]** --- The section to place the metadata. Valid values are
2892 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2893 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2894 Objective-C ABI version 2.
2896 * - ``Objective-C Garbage Collection``
2897 - **[Required]** --- Specifies whether garbage collection is supported or
2898 not. Valid values are 0, for no garbage collection, and 2, for garbage
2899 collection supported.
2901 * - ``Objective-C GC Only``
2902 - **[Optional]** --- Specifies that only garbage collection is supported.
2903 If present, its value must be 6. This flag requires that the
2904 ``Objective-C Garbage Collection`` flag have the value 2.
2906 Some important flag interactions:
2908 - If a module with ``Objective-C Garbage Collection`` set to 0 is
2909 merged with a module with ``Objective-C Garbage Collection`` set to
2910 2, then the resulting module has the
2911 ``Objective-C Garbage Collection`` flag set to 0.
2912 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2913 merged with a module with ``Objective-C GC Only`` set to 6.
2915 Automatic Linker Flags Module Flags Metadata
2916 --------------------------------------------
2918 Some targets support embedding flags to the linker inside individual object
2919 files. Typically this is used in conjunction with language extensions which
2920 allow source files to explicitly declare the libraries they depend on, and have
2921 these automatically be transmitted to the linker via object files.
2923 These flags are encoded in the IR using metadata in the module flags section,
2924 using the ``Linker Options`` key. The merge behavior for this flag is required
2925 to be ``AppendUnique``, and the value for the key is expected to be a metadata
2926 node which should be a list of other metadata nodes, each of which should be a
2927 list of metadata strings defining linker options.
2929 For example, the following metadata section specifies two separate sets of
2930 linker options, presumably to link against ``libz`` and the ``Cocoa``
2933 !0 = metadata !{ i32 6, metadata !"Linker Options",
2935 metadata !{ metadata !"-lz" },
2936 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
2937 !llvm.module.flags = !{ !0 }
2939 The metadata encoding as lists of lists of options, as opposed to a collapsed
2940 list of options, is chosen so that the IR encoding can use multiple option
2941 strings to specify e.g., a single library, while still having that specifier be
2942 preserved as an atomic element that can be recognized by a target specific
2943 assembly writer or object file emitter.
2945 Each individual option is required to be either a valid option for the target's
2946 linker, or an option that is reserved by the target specific assembly writer or
2947 object file emitter. No other aspect of these options is defined by the IR.
2949 .. _intrinsicglobalvariables:
2951 Intrinsic Global Variables
2952 ==========================
2954 LLVM has a number of "magic" global variables that contain data that
2955 affect code generation or other IR semantics. These are documented here.
2956 All globals of this sort should have a section specified as
2957 "``llvm.metadata``". This section and all globals that start with
2958 "``llvm.``" are reserved for use by LLVM.
2962 The '``llvm.used``' Global Variable
2963 -----------------------------------
2965 The ``@llvm.used`` global is an array which has
2966 :ref:`appending linkage <linkage_appending>`. This array contains a list of
2967 pointers to named global variables, functions and aliases which may optionally
2968 have a pointer cast formed of bitcast or getelementptr. For example, a legal
2971 .. code-block:: llvm
2976 @llvm.used = appending global [2 x i8*] [
2978 i8* bitcast (i32* @Y to i8*)
2979 ], section "llvm.metadata"
2981 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
2982 and linker are required to treat the symbol as if there is a reference to the
2983 symbol that it cannot see (which is why they have to be named). For example, if
2984 a variable has internal linkage and no references other than that from the
2985 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
2986 references from inline asms and other things the compiler cannot "see", and
2987 corresponds to "``attribute((used))``" in GNU C.
2989 On some targets, the code generator must emit a directive to the
2990 assembler or object file to prevent the assembler and linker from
2991 molesting the symbol.
2993 .. _gv_llvmcompilerused:
2995 The '``llvm.compiler.used``' Global Variable
2996 --------------------------------------------
2998 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
2999 directive, except that it only prevents the compiler from touching the
3000 symbol. On targets that support it, this allows an intelligent linker to
3001 optimize references to the symbol without being impeded as it would be
3004 This is a rare construct that should only be used in rare circumstances,
3005 and should not be exposed to source languages.
3007 .. _gv_llvmglobalctors:
3009 The '``llvm.global_ctors``' Global Variable
3010 -------------------------------------------
3012 .. code-block:: llvm
3014 %0 = type { i32, void ()* }
3015 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3017 The ``@llvm.global_ctors`` array contains a list of constructor
3018 functions and associated priorities. The functions referenced by this
3019 array will be called in ascending order of priority (i.e. lowest first)
3020 when the module is loaded. The order of functions with the same priority
3023 .. _llvmglobaldtors:
3025 The '``llvm.global_dtors``' Global Variable
3026 -------------------------------------------
3028 .. code-block:: llvm
3030 %0 = type { i32, void ()* }
3031 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3033 The ``@llvm.global_dtors`` array contains a list of destructor functions
3034 and associated priorities. The functions referenced by this array will
3035 be called in descending order of priority (i.e. highest first) when the
3036 module is loaded. The order of functions with the same priority is not
3039 Instruction Reference
3040 =====================
3042 The LLVM instruction set consists of several different classifications
3043 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3044 instructions <binaryops>`, :ref:`bitwise binary
3045 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3046 :ref:`other instructions <otherops>`.
3050 Terminator Instructions
3051 -----------------------
3053 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3054 program ends with a "Terminator" instruction, which indicates which
3055 block should be executed after the current block is finished. These
3056 terminator instructions typically yield a '``void``' value: they produce
3057 control flow, not values (the one exception being the
3058 ':ref:`invoke <i_invoke>`' instruction).
3060 The terminator instructions are: ':ref:`ret <i_ret>`',
3061 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3062 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3063 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3067 '``ret``' Instruction
3068 ^^^^^^^^^^^^^^^^^^^^^
3075 ret <type> <value> ; Return a value from a non-void function
3076 ret void ; Return from void function
3081 The '``ret``' instruction is used to return control flow (and optionally
3082 a value) from a function back to the caller.
3084 There are two forms of the '``ret``' instruction: one that returns a
3085 value and then causes control flow, and one that just causes control
3091 The '``ret``' instruction optionally accepts a single argument, the
3092 return value. The type of the return value must be a ':ref:`first
3093 class <t_firstclass>`' type.
3095 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3096 return type and contains a '``ret``' instruction with no return value or
3097 a return value with a type that does not match its type, or if it has a
3098 void return type and contains a '``ret``' instruction with a return
3104 When the '``ret``' instruction is executed, control flow returns back to
3105 the calling function's context. If the caller is a
3106 ":ref:`call <i_call>`" instruction, execution continues at the
3107 instruction after the call. If the caller was an
3108 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3109 beginning of the "normal" destination block. If the instruction returns
3110 a value, that value shall set the call or invoke instruction's return
3116 .. code-block:: llvm
3118 ret i32 5 ; Return an integer value of 5
3119 ret void ; Return from a void function
3120 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3124 '``br``' Instruction
3125 ^^^^^^^^^^^^^^^^^^^^
3132 br i1 <cond>, label <iftrue>, label <iffalse>
3133 br label <dest> ; Unconditional branch
3138 The '``br``' instruction is used to cause control flow to transfer to a
3139 different basic block in the current function. There are two forms of
3140 this instruction, corresponding to a conditional branch and an
3141 unconditional branch.
3146 The conditional branch form of the '``br``' instruction takes a single
3147 '``i1``' value and two '``label``' values. The unconditional form of the
3148 '``br``' instruction takes a single '``label``' value as a target.
3153 Upon execution of a conditional '``br``' instruction, the '``i1``'
3154 argument is evaluated. If the value is ``true``, control flows to the
3155 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3156 to the '``iffalse``' ``label`` argument.
3161 .. code-block:: llvm
3164 %cond = icmp eq i32 %a, %b
3165 br i1 %cond, label %IfEqual, label %IfUnequal
3173 '``switch``' Instruction
3174 ^^^^^^^^^^^^^^^^^^^^^^^^
3181 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3186 The '``switch``' instruction is used to transfer control flow to one of
3187 several different places. It is a generalization of the '``br``'
3188 instruction, allowing a branch to occur to one of many possible
3194 The '``switch``' instruction uses three parameters: an integer
3195 comparison value '``value``', a default '``label``' destination, and an
3196 array of pairs of comparison value constants and '``label``'s. The table
3197 is not allowed to contain duplicate constant entries.
3202 The ``switch`` instruction specifies a table of values and destinations.
3203 When the '``switch``' instruction is executed, this table is searched
3204 for the given value. If the value is found, control flow is transferred
3205 to the corresponding destination; otherwise, control flow is transferred
3206 to the default destination.
3211 Depending on properties of the target machine and the particular
3212 ``switch`` instruction, this instruction may be code generated in
3213 different ways. For example, it could be generated as a series of
3214 chained conditional branches or with a lookup table.
3219 .. code-block:: llvm
3221 ; Emulate a conditional br instruction
3222 %Val = zext i1 %value to i32
3223 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3225 ; Emulate an unconditional br instruction
3226 switch i32 0, label %dest [ ]
3228 ; Implement a jump table:
3229 switch i32 %val, label %otherwise [ i32 0, label %onzero
3231 i32 2, label %ontwo ]
3235 '``indirectbr``' Instruction
3236 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3243 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3248 The '``indirectbr``' instruction implements an indirect branch to a
3249 label within the current function, whose address is specified by
3250 "``address``". Address must be derived from a
3251 :ref:`blockaddress <blockaddress>` constant.
3256 The '``address``' argument is the address of the label to jump to. The
3257 rest of the arguments indicate the full set of possible destinations
3258 that the address may point to. Blocks are allowed to occur multiple
3259 times in the destination list, though this isn't particularly useful.
3261 This destination list is required so that dataflow analysis has an
3262 accurate understanding of the CFG.
3267 Control transfers to the block specified in the address argument. All
3268 possible destination blocks must be listed in the label list, otherwise
3269 this instruction has undefined behavior. This implies that jumps to
3270 labels defined in other functions have undefined behavior as well.
3275 This is typically implemented with a jump through a register.
3280 .. code-block:: llvm
3282 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3286 '``invoke``' Instruction
3287 ^^^^^^^^^^^^^^^^^^^^^^^^
3294 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3295 to label <normal label> unwind label <exception label>
3300 The '``invoke``' instruction causes control to transfer to a specified
3301 function, with the possibility of control flow transfer to either the
3302 '``normal``' label or the '``exception``' label. If the callee function
3303 returns with the "``ret``" instruction, control flow will return to the
3304 "normal" label. If the callee (or any indirect callees) returns via the
3305 ":ref:`resume <i_resume>`" instruction or other exception handling
3306 mechanism, control is interrupted and continued at the dynamically
3307 nearest "exception" label.
3309 The '``exception``' label is a `landing
3310 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3311 '``exception``' label is required to have the
3312 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3313 information about the behavior of the program after unwinding happens,
3314 as its first non-PHI instruction. The restrictions on the
3315 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3316 instruction, so that the important information contained within the
3317 "``landingpad``" instruction can't be lost through normal code motion.
3322 This instruction requires several arguments:
3324 #. The optional "cconv" marker indicates which :ref:`calling
3325 convention <callingconv>` the call should use. If none is
3326 specified, the call defaults to using C calling conventions.
3327 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3328 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3330 #. '``ptr to function ty``': shall be the signature of the pointer to
3331 function value being invoked. In most cases, this is a direct
3332 function invocation, but indirect ``invoke``'s are just as possible,
3333 branching off an arbitrary pointer to function value.
3334 #. '``function ptr val``': An LLVM value containing a pointer to a
3335 function to be invoked.
3336 #. '``function args``': argument list whose types match the function
3337 signature argument types and parameter attributes. All arguments must
3338 be of :ref:`first class <t_firstclass>` type. If the function signature
3339 indicates the function accepts a variable number of arguments, the
3340 extra arguments can be specified.
3341 #. '``normal label``': the label reached when the called function
3342 executes a '``ret``' instruction.
3343 #. '``exception label``': the label reached when a callee returns via
3344 the :ref:`resume <i_resume>` instruction or other exception handling
3346 #. The optional :ref:`function attributes <fnattrs>` list. Only
3347 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3348 attributes are valid here.
3353 This instruction is designed to operate as a standard '``call``'
3354 instruction in most regards. The primary difference is that it
3355 establishes an association with a label, which is used by the runtime
3356 library to unwind the stack.
3358 This instruction is used in languages with destructors to ensure that
3359 proper cleanup is performed in the case of either a ``longjmp`` or a
3360 thrown exception. Additionally, this is important for implementation of
3361 '``catch``' clauses in high-level languages that support them.
3363 For the purposes of the SSA form, the definition of the value returned
3364 by the '``invoke``' instruction is deemed to occur on the edge from the
3365 current block to the "normal" label. If the callee unwinds then no
3366 return value is available.
3371 .. code-block:: llvm
3373 %retval = invoke i32 @Test(i32 15) to label %Continue
3374 unwind label %TestCleanup ; {i32}:retval set
3375 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3376 unwind label %TestCleanup ; {i32}:retval set
3380 '``resume``' Instruction
3381 ^^^^^^^^^^^^^^^^^^^^^^^^
3388 resume <type> <value>
3393 The '``resume``' instruction is a terminator instruction that has no
3399 The '``resume``' instruction requires one argument, which must have the
3400 same type as the result of any '``landingpad``' instruction in the same
3406 The '``resume``' instruction resumes propagation of an existing
3407 (in-flight) exception whose unwinding was interrupted with a
3408 :ref:`landingpad <i_landingpad>` instruction.
3413 .. code-block:: llvm
3415 resume { i8*, i32 } %exn
3419 '``unreachable``' Instruction
3420 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3432 The '``unreachable``' instruction has no defined semantics. This
3433 instruction is used to inform the optimizer that a particular portion of
3434 the code is not reachable. This can be used to indicate that the code
3435 after a no-return function cannot be reached, and other facts.
3440 The '``unreachable``' instruction has no defined semantics.
3447 Binary operators are used to do most of the computation in a program.
3448 They require two operands of the same type, execute an operation on
3449 them, and produce a single value. The operands might represent multiple
3450 data, as is the case with the :ref:`vector <t_vector>` data type. The
3451 result value has the same type as its operands.
3453 There are several different binary operators:
3457 '``add``' Instruction
3458 ^^^^^^^^^^^^^^^^^^^^^
3465 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3466 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3467 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3468 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3473 The '``add``' instruction returns the sum of its two operands.
3478 The two arguments to the '``add``' instruction must be
3479 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3480 arguments must have identical types.
3485 The value produced is the integer sum of the two operands.
3487 If the sum has unsigned overflow, the result returned is the
3488 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3491 Because LLVM integers use a two's complement representation, this
3492 instruction is appropriate for both signed and unsigned integers.
3494 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3495 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3496 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3497 unsigned and/or signed overflow, respectively, occurs.
3502 .. code-block:: llvm
3504 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3508 '``fadd``' Instruction
3509 ^^^^^^^^^^^^^^^^^^^^^^
3516 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3521 The '``fadd``' instruction returns the sum of its two operands.
3526 The two arguments to the '``fadd``' instruction must be :ref:`floating
3527 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3528 Both arguments must have identical types.
3533 The value produced is the floating point sum of the two operands. This
3534 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3535 which are optimization hints to enable otherwise unsafe floating point
3541 .. code-block:: llvm
3543 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3545 '``sub``' Instruction
3546 ^^^^^^^^^^^^^^^^^^^^^
3553 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3554 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3555 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3556 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3561 The '``sub``' instruction returns the difference of its two operands.
3563 Note that the '``sub``' instruction is used to represent the '``neg``'
3564 instruction present in most other intermediate representations.
3569 The two arguments to the '``sub``' instruction must be
3570 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3571 arguments must have identical types.
3576 The value produced is the integer difference of the two operands.
3578 If the difference has unsigned overflow, the result returned is the
3579 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3582 Because LLVM integers use a two's complement representation, this
3583 instruction is appropriate for both signed and unsigned integers.
3585 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3586 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3587 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3588 unsigned and/or signed overflow, respectively, occurs.
3593 .. code-block:: llvm
3595 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3596 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3600 '``fsub``' Instruction
3601 ^^^^^^^^^^^^^^^^^^^^^^
3608 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3613 The '``fsub``' instruction returns the difference of its two operands.
3615 Note that the '``fsub``' instruction is used to represent the '``fneg``'
3616 instruction present in most other intermediate representations.
3621 The two arguments to the '``fsub``' instruction must be :ref:`floating
3622 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3623 Both arguments must have identical types.
3628 The value produced is the floating point difference of the two operands.
3629 This instruction can also take any number of :ref:`fast-math
3630 flags <fastmath>`, which are optimization hints to enable otherwise
3631 unsafe floating point optimizations:
3636 .. code-block:: llvm
3638 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3639 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3641 '``mul``' Instruction
3642 ^^^^^^^^^^^^^^^^^^^^^
3649 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3650 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3651 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3652 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3657 The '``mul``' instruction returns the product of its two operands.
3662 The two arguments to the '``mul``' instruction must be
3663 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3664 arguments must have identical types.
3669 The value produced is the integer product of the two operands.
3671 If the result of the multiplication has unsigned overflow, the result
3672 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3673 bit width of the result.
3675 Because LLVM integers use a two's complement representation, and the
3676 result is the same width as the operands, this instruction returns the
3677 correct result for both signed and unsigned integers. If a full product
3678 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3679 sign-extended or zero-extended as appropriate to the width of the full
3682 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3683 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3684 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3685 unsigned and/or signed overflow, respectively, occurs.
3690 .. code-block:: llvm
3692 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3696 '``fmul``' Instruction
3697 ^^^^^^^^^^^^^^^^^^^^^^
3704 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3709 The '``fmul``' instruction returns the product of its two operands.
3714 The two arguments to the '``fmul``' instruction must be :ref:`floating
3715 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3716 Both arguments must have identical types.
3721 The value produced is the floating point product of the two operands.
3722 This instruction can also take any number of :ref:`fast-math
3723 flags <fastmath>`, which are optimization hints to enable otherwise
3724 unsafe floating point optimizations:
3729 .. code-block:: llvm
3731 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3733 '``udiv``' Instruction
3734 ^^^^^^^^^^^^^^^^^^^^^^
3741 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3742 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3747 The '``udiv``' instruction returns the quotient of its two operands.
3752 The two arguments to the '``udiv``' instruction must be
3753 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3754 arguments must have identical types.
3759 The value produced is the unsigned integer quotient of the two operands.
3761 Note that unsigned integer division and signed integer division are
3762 distinct operations; for signed integer division, use '``sdiv``'.
3764 Division by zero leads to undefined behavior.
3766 If the ``exact`` keyword is present, the result value of the ``udiv`` is
3767 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3768 such, "((a udiv exact b) mul b) == a").
3773 .. code-block:: llvm
3775 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3777 '``sdiv``' Instruction
3778 ^^^^^^^^^^^^^^^^^^^^^^
3785 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3786 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3791 The '``sdiv``' instruction returns the quotient of its two operands.
3796 The two arguments to the '``sdiv``' instruction must be
3797 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3798 arguments must have identical types.
3803 The value produced is the signed integer quotient of the two operands
3804 rounded towards zero.
3806 Note that signed integer division and unsigned integer division are
3807 distinct operations; for unsigned integer division, use '``udiv``'.
3809 Division by zero leads to undefined behavior. Overflow also leads to
3810 undefined behavior; this is a rare case, but can occur, for example, by
3811 doing a 32-bit division of -2147483648 by -1.
3813 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3814 a :ref:`poison value <poisonvalues>` if the result would be rounded.
3819 .. code-block:: llvm
3821 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3825 '``fdiv``' Instruction
3826 ^^^^^^^^^^^^^^^^^^^^^^
3833 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3838 The '``fdiv``' instruction returns the quotient of its two operands.
3843 The two arguments to the '``fdiv``' instruction must be :ref:`floating
3844 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3845 Both arguments must have identical types.
3850 The value produced is the floating point quotient of the two operands.
3851 This instruction can also take any number of :ref:`fast-math
3852 flags <fastmath>`, which are optimization hints to enable otherwise
3853 unsafe floating point optimizations:
3858 .. code-block:: llvm
3860 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3862 '``urem``' Instruction
3863 ^^^^^^^^^^^^^^^^^^^^^^
3870 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3875 The '``urem``' instruction returns the remainder from the unsigned
3876 division of its two arguments.
3881 The two arguments to the '``urem``' instruction must be
3882 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3883 arguments must have identical types.
3888 This instruction returns the unsigned integer *remainder* of a division.
3889 This instruction always performs an unsigned division to get the
3892 Note that unsigned integer remainder and signed integer remainder are
3893 distinct operations; for signed integer remainder, use '``srem``'.
3895 Taking the remainder of a division by zero leads to undefined behavior.
3900 .. code-block:: llvm
3902 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3904 '``srem``' Instruction
3905 ^^^^^^^^^^^^^^^^^^^^^^
3912 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3917 The '``srem``' instruction returns the remainder from the signed
3918 division of its two operands. This instruction can also take
3919 :ref:`vector <t_vector>` versions of the values in which case the elements
3925 The two arguments to the '``srem``' instruction must be
3926 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3927 arguments must have identical types.
3932 This instruction returns the *remainder* of a division (where the result
3933 is either zero or has the same sign as the dividend, ``op1``), not the
3934 *modulo* operator (where the result is either zero or has the same sign
3935 as the divisor, ``op2``) of a value. For more information about the
3936 difference, see `The Math
3937 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3938 table of how this is implemented in various languages, please see
3940 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3942 Note that signed integer remainder and unsigned integer remainder are
3943 distinct operations; for unsigned integer remainder, use '``urem``'.
3945 Taking the remainder of a division by zero leads to undefined behavior.
3946 Overflow also leads to undefined behavior; this is a rare case, but can
3947 occur, for example, by taking the remainder of a 32-bit division of
3948 -2147483648 by -1. (The remainder doesn't actually overflow, but this
3949 rule lets srem be implemented using instructions that return both the
3950 result of the division and the remainder.)
3955 .. code-block:: llvm
3957 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
3961 '``frem``' Instruction
3962 ^^^^^^^^^^^^^^^^^^^^^^
3969 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3974 The '``frem``' instruction returns the remainder from the division of
3980 The two arguments to the '``frem``' instruction must be :ref:`floating
3981 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3982 Both arguments must have identical types.
3987 This instruction returns the *remainder* of a division. The remainder
3988 has the same sign as the dividend. This instruction can also take any
3989 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
3990 to enable otherwise unsafe floating point optimizations:
3995 .. code-block:: llvm
3997 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
4001 Bitwise Binary Operations
4002 -------------------------
4004 Bitwise binary operators are used to do various forms of bit-twiddling
4005 in a program. They are generally very efficient instructions and can
4006 commonly be strength reduced from other instructions. They require two
4007 operands of the same type, execute an operation on them, and produce a
4008 single value. The resulting value is the same type as its operands.
4010 '``shl``' Instruction
4011 ^^^^^^^^^^^^^^^^^^^^^
4018 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
4019 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
4020 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
4021 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
4026 The '``shl``' instruction returns the first operand shifted to the left
4027 a specified number of bits.
4032 Both arguments to the '``shl``' instruction must be the same
4033 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4034 '``op2``' is treated as an unsigned value.
4039 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4040 where ``n`` is the width of the result. If ``op2`` is (statically or
4041 dynamically) negative or equal to or larger than the number of bits in
4042 ``op1``, the result is undefined. If the arguments are vectors, each
4043 vector element of ``op1`` is shifted by the corresponding shift amount
4046 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4047 value <poisonvalues>` if it shifts out any non-zero bits. If the
4048 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4049 value <poisonvalues>` if it shifts out any bits that disagree with the
4050 resultant sign bit. As such, NUW/NSW have the same semantics as they
4051 would if the shift were expressed as a mul instruction with the same
4052 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4057 .. code-block:: llvm
4059 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
4060 <result> = shl i32 4, 2 ; yields {i32}: 16
4061 <result> = shl i32 1, 10 ; yields {i32}: 1024
4062 <result> = shl i32 1, 32 ; undefined
4063 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4065 '``lshr``' Instruction
4066 ^^^^^^^^^^^^^^^^^^^^^^
4073 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
4074 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
4079 The '``lshr``' instruction (logical shift right) returns the first
4080 operand shifted to the right a specified number of bits with zero fill.
4085 Both arguments to the '``lshr``' instruction must be the same
4086 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4087 '``op2``' is treated as an unsigned value.
4092 This instruction always performs a logical shift right operation. The
4093 most significant bits of the result will be filled with zero bits after
4094 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4095 than the number of bits in ``op1``, the result is undefined. If the
4096 arguments are vectors, each vector element of ``op1`` is shifted by the
4097 corresponding shift amount in ``op2``.
4099 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4100 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4106 .. code-block:: llvm
4108 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
4109 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
4110 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
4111 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7F
4112 <result> = lshr i32 1, 32 ; undefined
4113 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4115 '``ashr``' Instruction
4116 ^^^^^^^^^^^^^^^^^^^^^^
4123 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4124 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4129 The '``ashr``' instruction (arithmetic shift right) returns the first
4130 operand shifted to the right a specified number of bits with sign
4136 Both arguments to the '``ashr``' instruction must be the same
4137 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4138 '``op2``' is treated as an unsigned value.
4143 This instruction always performs an arithmetic shift right operation,
4144 The most significant bits of the result will be filled with the sign bit
4145 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4146 than the number of bits in ``op1``, the result is undefined. If the
4147 arguments are vectors, each vector element of ``op1`` is shifted by the
4148 corresponding shift amount in ``op2``.
4150 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4151 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4157 .. code-block:: llvm
4159 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4160 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4161 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4162 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4163 <result> = ashr i32 1, 32 ; undefined
4164 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4166 '``and``' Instruction
4167 ^^^^^^^^^^^^^^^^^^^^^
4174 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4179 The '``and``' instruction returns the bitwise logical and of its two
4185 The two arguments to the '``and``' instruction must be
4186 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4187 arguments must have identical types.
4192 The truth table used for the '``and``' instruction is:
4209 .. code-block:: llvm
4211 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4212 <result> = and i32 15, 40 ; yields {i32}:result = 8
4213 <result> = and i32 4, 8 ; yields {i32}:result = 0
4215 '``or``' Instruction
4216 ^^^^^^^^^^^^^^^^^^^^
4223 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4228 The '``or``' instruction returns the bitwise logical inclusive or of its
4234 The two arguments to the '``or``' instruction must be
4235 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4236 arguments must have identical types.
4241 The truth table used for the '``or``' instruction is:
4260 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4261 <result> = or i32 15, 40 ; yields {i32}:result = 47
4262 <result> = or i32 4, 8 ; yields {i32}:result = 12
4264 '``xor``' Instruction
4265 ^^^^^^^^^^^^^^^^^^^^^
4272 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4277 The '``xor``' instruction returns the bitwise logical exclusive or of
4278 its two operands. The ``xor`` is used to implement the "one's
4279 complement" operation, which is the "~" operator in C.
4284 The two arguments to the '``xor``' instruction must be
4285 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4286 arguments must have identical types.
4291 The truth table used for the '``xor``' instruction is:
4308 .. code-block:: llvm
4310 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4311 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4312 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4313 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4318 LLVM supports several instructions to represent vector operations in a
4319 target-independent manner. These instructions cover the element-access
4320 and vector-specific operations needed to process vectors effectively.
4321 While LLVM does directly support these vector operations, many
4322 sophisticated algorithms will want to use target-specific intrinsics to
4323 take full advantage of a specific target.
4325 .. _i_extractelement:
4327 '``extractelement``' Instruction
4328 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4335 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4340 The '``extractelement``' instruction extracts a single scalar element
4341 from a vector at a specified index.
4346 The first operand of an '``extractelement``' instruction is a value of
4347 :ref:`vector <t_vector>` type. The second operand is an index indicating
4348 the position from which to extract the element. The index may be a
4354 The result is a scalar of the same type as the element type of ``val``.
4355 Its value is the value at position ``idx`` of ``val``. If ``idx``
4356 exceeds the length of ``val``, the results are undefined.
4361 .. code-block:: llvm
4363 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4365 .. _i_insertelement:
4367 '``insertelement``' Instruction
4368 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4375 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4380 The '``insertelement``' instruction inserts a scalar element into a
4381 vector at a specified index.
4386 The first operand of an '``insertelement``' instruction is a value of
4387 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4388 type must equal the element type of the first operand. The third operand
4389 is an index indicating the position at which to insert the value. The
4390 index may be a variable.
4395 The result is a vector of the same type as ``val``. Its element values
4396 are those of ``val`` except at position ``idx``, where it gets the value
4397 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4403 .. code-block:: llvm
4405 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4407 .. _i_shufflevector:
4409 '``shufflevector``' Instruction
4410 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4417 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4422 The '``shufflevector``' instruction constructs a permutation of elements
4423 from two input vectors, returning a vector with the same element type as
4424 the input and length that is the same as the shuffle mask.
4429 The first two operands of a '``shufflevector``' instruction are vectors
4430 with the same type. The third argument is a shuffle mask whose element
4431 type is always 'i32'. The result of the instruction is a vector whose
4432 length is the same as the shuffle mask and whose element type is the
4433 same as the element type of the first two operands.
4435 The shuffle mask operand is required to be a constant vector with either
4436 constant integer or undef values.
4441 The elements of the two input vectors are numbered from left to right
4442 across both of the vectors. The shuffle mask operand specifies, for each
4443 element of the result vector, which element of the two input vectors the
4444 result element gets. The element selector may be undef (meaning "don't
4445 care") and the second operand may be undef if performing a shuffle from
4451 .. code-block:: llvm
4453 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4454 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4455 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4456 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4457 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4458 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4459 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4460 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4462 Aggregate Operations
4463 --------------------
4465 LLVM supports several instructions for working with
4466 :ref:`aggregate <t_aggregate>` values.
4470 '``extractvalue``' Instruction
4471 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4478 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4483 The '``extractvalue``' instruction extracts the value of a member field
4484 from an :ref:`aggregate <t_aggregate>` value.
4489 The first operand of an '``extractvalue``' instruction is a value of
4490 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4491 constant indices to specify which value to extract in a similar manner
4492 as indices in a '``getelementptr``' instruction.
4494 The major differences to ``getelementptr`` indexing are:
4496 - Since the value being indexed is not a pointer, the first index is
4497 omitted and assumed to be zero.
4498 - At least one index must be specified.
4499 - Not only struct indices but also array indices must be in bounds.
4504 The result is the value at the position in the aggregate specified by
4510 .. code-block:: llvm
4512 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4516 '``insertvalue``' Instruction
4517 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4524 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4529 The '``insertvalue``' instruction inserts a value into a member field in
4530 an :ref:`aggregate <t_aggregate>` value.
4535 The first operand of an '``insertvalue``' instruction is a value of
4536 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4537 a first-class value to insert. The following operands are constant
4538 indices indicating the position at which to insert the value in a
4539 similar manner as indices in a '``extractvalue``' instruction. The value
4540 to insert must have the same type as the value identified by the
4546 The result is an aggregate of the same type as ``val``. Its value is
4547 that of ``val`` except that the value at the position specified by the
4548 indices is that of ``elt``.
4553 .. code-block:: llvm
4555 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4556 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4557 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4561 Memory Access and Addressing Operations
4562 ---------------------------------------
4564 A key design point of an SSA-based representation is how it represents
4565 memory. In LLVM, no memory locations are in SSA form, which makes things
4566 very simple. This section describes how to read, write, and allocate
4571 '``alloca``' Instruction
4572 ^^^^^^^^^^^^^^^^^^^^^^^^
4579 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4584 The '``alloca``' instruction allocates memory on the stack frame of the
4585 currently executing function, to be automatically released when this
4586 function returns to its caller. The object is always allocated in the
4587 generic address space (address space zero).
4592 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4593 bytes of memory on the runtime stack, returning a pointer of the
4594 appropriate type to the program. If "NumElements" is specified, it is
4595 the number of elements allocated, otherwise "NumElements" is defaulted
4596 to be one. If a constant alignment is specified, the value result of the
4597 allocation is guaranteed to be aligned to at least that boundary. If not
4598 specified, or if zero, the target can choose to align the allocation on
4599 any convenient boundary compatible with the type.
4601 '``type``' may be any sized type.
4606 Memory is allocated; a pointer is returned. The operation is undefined
4607 if there is insufficient stack space for the allocation. '``alloca``'d
4608 memory is automatically released when the function returns. The
4609 '``alloca``' instruction is commonly used to represent automatic
4610 variables that must have an address available. When the function returns
4611 (either with the ``ret`` or ``resume`` instructions), the memory is
4612 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4613 The order in which memory is allocated (ie., which way the stack grows)
4619 .. code-block:: llvm
4621 %ptr = alloca i32 ; yields {i32*}:ptr
4622 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4623 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4624 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4628 '``load``' Instruction
4629 ^^^^^^^^^^^^^^^^^^^^^^
4636 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4637 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4638 !<index> = !{ i32 1 }
4643 The '``load``' instruction is used to read from memory.
4648 The argument to the ``load`` instruction specifies the memory address
4649 from which to load. The pointer must point to a :ref:`first
4650 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4651 then the optimizer is not allowed to modify the number or order of
4652 execution of this ``load`` with other :ref:`volatile
4653 operations <volatile>`.
4655 If the ``load`` is marked as ``atomic``, it takes an extra
4656 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4657 ``release`` and ``acq_rel`` orderings are not valid on ``load``
4658 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4659 when they may see multiple atomic stores. The type of the pointee must
4660 be an integer type whose bit width is a power of two greater than or
4661 equal to eight and less than or equal to a target-specific size limit.
4662 ``align`` must be explicitly specified on atomic loads, and the load has
4663 undefined behavior if the alignment is not set to a value which is at
4664 least the size in bytes of the pointee. ``!nontemporal`` does not have
4665 any defined semantics for atomic loads.
4667 The optional constant ``align`` argument specifies the alignment of the
4668 operation (that is, the alignment of the memory address). A value of 0
4669 or an omitted ``align`` argument means that the operation has the ABI
4670 alignment for the target. It is the responsibility of the code emitter
4671 to ensure that the alignment information is correct. Overestimating the
4672 alignment results in undefined behavior. Underestimating the alignment
4673 may produce less efficient code. An alignment of 1 is always safe.
4675 The optional ``!nontemporal`` metadata must reference a single
4676 metadata name ``<index>`` corresponding to a metadata node with one
4677 ``i32`` entry of value 1. The existence of the ``!nontemporal``
4678 metadata on the instruction tells the optimizer and code generator
4679 that this load is not expected to be reused in the cache. The code
4680 generator may select special instructions to save cache bandwidth, such
4681 as the ``MOVNT`` instruction on x86.
4683 The optional ``!invariant.load`` metadata must reference a single
4684 metadata name ``<index>`` corresponding to a metadata node with no
4685 entries. The existence of the ``!invariant.load`` metadata on the
4686 instruction tells the optimizer and code generator that this load
4687 address points to memory which does not change value during program
4688 execution. The optimizer may then move this load around, for example, by
4689 hoisting it out of loops using loop invariant code motion.
4694 The location of memory pointed to is loaded. If the value being loaded
4695 is of scalar type then the number of bytes read does not exceed the
4696 minimum number of bytes needed to hold all bits of the type. For
4697 example, loading an ``i24`` reads at most three bytes. When loading a
4698 value of a type like ``i20`` with a size that is not an integral number
4699 of bytes, the result is undefined if the value was not originally
4700 written using a store of the same type.
4705 .. code-block:: llvm
4707 %ptr = alloca i32 ; yields {i32*}:ptr
4708 store i32 3, i32* %ptr ; yields {void}
4709 %val = load i32* %ptr ; yields {i32}:val = i32 3
4713 '``store``' Instruction
4714 ^^^^^^^^^^^^^^^^^^^^^^^
4721 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4722 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4727 The '``store``' instruction is used to write to memory.
4732 There are two arguments to the ``store`` instruction: a value to store
4733 and an address at which to store it. The type of the ``<pointer>``
4734 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4735 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
4736 then the optimizer is not allowed to modify the number or order of
4737 execution of this ``store`` with other :ref:`volatile
4738 operations <volatile>`.
4740 If the ``store`` is marked as ``atomic``, it takes an extra
4741 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4742 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4743 instructions. Atomic loads produce :ref:`defined <memmodel>` results
4744 when they may see multiple atomic stores. The type of the pointee must
4745 be an integer type whose bit width is a power of two greater than or
4746 equal to eight and less than or equal to a target-specific size limit.
4747 ``align`` must be explicitly specified on atomic stores, and the store
4748 has undefined behavior if the alignment is not set to a value which is
4749 at least the size in bytes of the pointee. ``!nontemporal`` does not
4750 have any defined semantics for atomic stores.
4752 The optional constant ``align`` argument specifies the alignment of the
4753 operation (that is, the alignment of the memory address). A value of 0
4754 or an omitted ``align`` argument means that the operation has the ABI
4755 alignment for the target. It is the responsibility of the code emitter
4756 to ensure that the alignment information is correct. Overestimating the
4757 alignment results in undefined behavior. Underestimating the
4758 alignment may produce less efficient code. An alignment of 1 is always
4761 The optional ``!nontemporal`` metadata must reference a single metadata
4762 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
4763 value 1. The existence of the ``!nontemporal`` metadata on the instruction
4764 tells the optimizer and code generator that this load is not expected to
4765 be reused in the cache. The code generator may select special
4766 instructions to save cache bandwidth, such as the MOVNT instruction on
4772 The contents of memory are updated to contain ``<value>`` at the
4773 location specified by the ``<pointer>`` operand. If ``<value>`` is
4774 of scalar type then the number of bytes written does not exceed the
4775 minimum number of bytes needed to hold all bits of the type. For
4776 example, storing an ``i24`` writes at most three bytes. When writing a
4777 value of a type like ``i20`` with a size that is not an integral number
4778 of bytes, it is unspecified what happens to the extra bits that do not
4779 belong to the type, but they will typically be overwritten.
4784 .. code-block:: llvm
4786 %ptr = alloca i32 ; yields {i32*}:ptr
4787 store i32 3, i32* %ptr ; yields {void}
4788 %val = load i32* %ptr ; yields {i32}:val = i32 3
4792 '``fence``' Instruction
4793 ^^^^^^^^^^^^^^^^^^^^^^^
4800 fence [singlethread] <ordering> ; yields {void}
4805 The '``fence``' instruction is used to introduce happens-before edges
4811 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
4812 defines what *synchronizes-with* edges they add. They can only be given
4813 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4818 A fence A which has (at least) ``release`` ordering semantics
4819 *synchronizes with* a fence B with (at least) ``acquire`` ordering
4820 semantics if and only if there exist atomic operations X and Y, both
4821 operating on some atomic object M, such that A is sequenced before X, X
4822 modifies M (either directly or through some side effect of a sequence
4823 headed by X), Y is sequenced before B, and Y observes M. This provides a
4824 *happens-before* dependency between A and B. Rather than an explicit
4825 ``fence``, one (but not both) of the atomic operations X or Y might
4826 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4827 still *synchronize-with* the explicit ``fence`` and establish the
4828 *happens-before* edge.
4830 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4831 ``acquire`` and ``release`` semantics specified above, participates in
4832 the global program order of other ``seq_cst`` operations and/or fences.
4834 The optional ":ref:`singlethread <singlethread>`" argument specifies
4835 that the fence only synchronizes with other fences in the same thread.
4836 (This is useful for interacting with signal handlers.)
4841 .. code-block:: llvm
4843 fence acquire ; yields {void}
4844 fence singlethread seq_cst ; yields {void}
4848 '``cmpxchg``' Instruction
4849 ^^^^^^^^^^^^^^^^^^^^^^^^^
4856 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4861 The '``cmpxchg``' instruction is used to atomically modify memory. It
4862 loads a value in memory and compares it to a given value. If they are
4863 equal, it stores a new value into the memory.
4868 There are three arguments to the '``cmpxchg``' instruction: an address
4869 to operate on, a value to compare to the value currently be at that
4870 address, and a new value to place at that address if the compared values
4871 are equal. The type of '<cmp>' must be an integer type whose bit width
4872 is a power of two greater than or equal to eight and less than or equal
4873 to a target-specific size limit. '<cmp>' and '<new>' must have the same
4874 type, and the type of '<pointer>' must be a pointer to that type. If the
4875 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4876 to modify the number or order of execution of this ``cmpxchg`` with
4877 other :ref:`volatile operations <volatile>`.
4879 The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4880 synchronizes with other atomic operations.
4882 The optional "``singlethread``" argument declares that the ``cmpxchg``
4883 is only atomic with respect to code (usually signal handlers) running in
4884 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4885 respect to all other code in the system.
4887 The pointer passed into cmpxchg must have alignment greater than or
4888 equal to the size in memory of the operand.
4893 The contents of memory at the location specified by the '``<pointer>``'
4894 operand is read and compared to '``<cmp>``'; if the read value is the
4895 equal, '``<new>``' is written. The original value at the location is
4898 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4899 of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4900 atomic load with an ordering parameter determined by dropping any
4901 ``release`` part of the ``cmpxchg``'s ordering.
4906 .. code-block:: llvm
4909 %orig = atomic load i32* %ptr unordered ; yields {i32}
4913 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4914 %squared = mul i32 %cmp, %cmp
4915 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4916 %success = icmp eq i32 %cmp, %old
4917 br i1 %success, label %done, label %loop
4924 '``atomicrmw``' Instruction
4925 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
4932 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4937 The '``atomicrmw``' instruction is used to atomically modify memory.
4942 There are three arguments to the '``atomicrmw``' instruction: an
4943 operation to apply, an address whose value to modify, an argument to the
4944 operation. The operation must be one of the following keywords:
4958 The type of '<value>' must be an integer type whose bit width is a power
4959 of two greater than or equal to eight and less than or equal to a
4960 target-specific size limit. The type of the '``<pointer>``' operand must
4961 be a pointer to that type. If the ``atomicrmw`` is marked as
4962 ``volatile``, then the optimizer is not allowed to modify the number or
4963 order of execution of this ``atomicrmw`` with other :ref:`volatile
4964 operations <volatile>`.
4969 The contents of memory at the location specified by the '``<pointer>``'
4970 operand are atomically read, modified, and written back. The original
4971 value at the location is returned. The modification is specified by the
4974 - xchg: ``*ptr = val``
4975 - add: ``*ptr = *ptr + val``
4976 - sub: ``*ptr = *ptr - val``
4977 - and: ``*ptr = *ptr & val``
4978 - nand: ``*ptr = ~(*ptr & val)``
4979 - or: ``*ptr = *ptr | val``
4980 - xor: ``*ptr = *ptr ^ val``
4981 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
4982 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
4983 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
4985 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
4991 .. code-block:: llvm
4993 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
4995 .. _i_getelementptr:
4997 '``getelementptr``' Instruction
4998 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5005 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5006 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5007 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5012 The '``getelementptr``' instruction is used to get the address of a
5013 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5014 address calculation only and does not access memory.
5019 The first argument is always a pointer or a vector of pointers, and
5020 forms the basis of the calculation. The remaining arguments are indices
5021 that indicate which of the elements of the aggregate object are indexed.
5022 The interpretation of each index is dependent on the type being indexed
5023 into. The first index always indexes the pointer value given as the
5024 first argument, the second index indexes a value of the type pointed to
5025 (not necessarily the value directly pointed to, since the first index
5026 can be non-zero), etc. The first type indexed into must be a pointer
5027 value, subsequent types can be arrays, vectors, and structs. Note that
5028 subsequent types being indexed into can never be pointers, since that
5029 would require loading the pointer before continuing calculation.
5031 The type of each index argument depends on the type it is indexing into.
5032 When indexing into a (optionally packed) structure, only ``i32`` integer
5033 **constants** are allowed (when using a vector of indices they must all
5034 be the **same** ``i32`` integer constant). When indexing into an array,
5035 pointer or vector, integers of any width are allowed, and they are not
5036 required to be constant. These integers are treated as signed values
5039 For example, let's consider a C code fragment and how it gets compiled
5055 int *foo(struct ST *s) {
5056 return &s[1].Z.B[5][13];
5059 The LLVM code generated by Clang is:
5061 .. code-block:: llvm
5063 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5064 %struct.ST = type { i32, double, %struct.RT }
5066 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5068 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5075 In the example above, the first index is indexing into the
5076 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5077 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5078 indexes into the third element of the structure, yielding a
5079 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5080 structure. The third index indexes into the second element of the
5081 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5082 dimensions of the array are subscripted into, yielding an '``i32``'
5083 type. The '``getelementptr``' instruction returns a pointer to this
5084 element, thus computing a value of '``i32*``' type.
5086 Note that it is perfectly legal to index partially through a structure,
5087 returning a pointer to an inner element. Because of this, the LLVM code
5088 for the given testcase is equivalent to:
5090 .. code-block:: llvm
5092 define i32* @foo(%struct.ST* %s) {
5093 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5094 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5095 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5096 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5097 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5101 If the ``inbounds`` keyword is present, the result value of the
5102 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5103 pointer is not an *in bounds* address of an allocated object, or if any
5104 of the addresses that would be formed by successive addition of the
5105 offsets implied by the indices to the base address with infinitely
5106 precise signed arithmetic are not an *in bounds* address of that
5107 allocated object. The *in bounds* addresses for an allocated object are
5108 all the addresses that point into the object, plus the address one byte
5109 past the end. In cases where the base is a vector of pointers the
5110 ``inbounds`` keyword applies to each of the computations element-wise.
5112 If the ``inbounds`` keyword is not present, the offsets are added to the
5113 base address with silently-wrapping two's complement arithmetic. If the
5114 offsets have a different width from the pointer, they are sign-extended
5115 or truncated to the width of the pointer. The result value of the
5116 ``getelementptr`` may be outside the object pointed to by the base
5117 pointer. The result value may not necessarily be used to access memory
5118 though, even if it happens to point into allocated storage. See the
5119 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5122 The getelementptr instruction is often confusing. For some more insight
5123 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5128 .. code-block:: llvm
5130 ; yields [12 x i8]*:aptr
5131 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5133 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5135 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5137 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5139 In cases where the pointer argument is a vector of pointers, each index
5140 must be a vector with the same number of elements. For example:
5142 .. code-block:: llvm
5144 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5146 Conversion Operations
5147 ---------------------
5149 The instructions in this category are the conversion instructions
5150 (casting) which all take a single operand and a type. They perform
5151 various bit conversions on the operand.
5153 '``trunc .. to``' Instruction
5154 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5161 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5166 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5171 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5172 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5173 of the same number of integers. The bit size of the ``value`` must be
5174 larger than the bit size of the destination type, ``ty2``. Equal sized
5175 types are not allowed.
5180 The '``trunc``' instruction truncates the high order bits in ``value``
5181 and converts the remaining bits to ``ty2``. Since the source size must
5182 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5183 It will always truncate bits.
5188 .. code-block:: llvm
5190 %X = trunc i32 257 to i8 ; yields i8:1
5191 %Y = trunc i32 123 to i1 ; yields i1:true
5192 %Z = trunc i32 122 to i1 ; yields i1:false
5193 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5195 '``zext .. to``' Instruction
5196 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5203 <result> = zext <ty> <value> to <ty2> ; yields ty2
5208 The '``zext``' instruction zero extends its operand to type ``ty2``.
5213 The '``zext``' instruction takes a value to cast, and a type to cast it
5214 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5215 the same number of integers. The bit size of the ``value`` must be
5216 smaller than the bit size of the destination type, ``ty2``.
5221 The ``zext`` fills the high order bits of the ``value`` with zero bits
5222 until it reaches the size of the destination type, ``ty2``.
5224 When zero extending from i1, the result will always be either 0 or 1.
5229 .. code-block:: llvm
5231 %X = zext i32 257 to i64 ; yields i64:257
5232 %Y = zext i1 true to i32 ; yields i32:1
5233 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5235 '``sext .. to``' Instruction
5236 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5243 <result> = sext <ty> <value> to <ty2> ; yields ty2
5248 The '``sext``' sign extends ``value`` to the type ``ty2``.
5253 The '``sext``' instruction takes a value to cast, and a type to cast it
5254 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5255 the same number of integers. The bit size of the ``value`` must be
5256 smaller than the bit size of the destination type, ``ty2``.
5261 The '``sext``' instruction performs a sign extension by copying the sign
5262 bit (highest order bit) of the ``value`` until it reaches the bit size
5263 of the type ``ty2``.
5265 When sign extending from i1, the extension always results in -1 or 0.
5270 .. code-block:: llvm
5272 %X = sext i8 -1 to i16 ; yields i16 :65535
5273 %Y = sext i1 true to i32 ; yields i32:-1
5274 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5276 '``fptrunc .. to``' Instruction
5277 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5284 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5289 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5294 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5295 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5296 The size of ``value`` must be larger than the size of ``ty2``. This
5297 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5302 The '``fptrunc``' instruction truncates a ``value`` from a larger
5303 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5304 point <t_floating>` type. If the value cannot fit within the
5305 destination type, ``ty2``, then the results are undefined.
5310 .. code-block:: llvm
5312 %X = fptrunc double 123.0 to float ; yields float:123.0
5313 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5315 '``fpext .. to``' Instruction
5316 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5323 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5328 The '``fpext``' extends a floating point ``value`` to a larger floating
5334 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5335 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5336 to. The source type must be smaller than the destination type.
5341 The '``fpext``' instruction extends the ``value`` from a smaller
5342 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5343 point <t_floating>` type. The ``fpext`` cannot be used to make a
5344 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5345 *no-op cast* for a floating point cast.
5350 .. code-block:: llvm
5352 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5353 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5355 '``fptoui .. to``' Instruction
5356 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5363 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5368 The '``fptoui``' converts a floating point ``value`` to its unsigned
5369 integer equivalent of type ``ty2``.
5374 The '``fptoui``' instruction takes a value to cast, which must be a
5375 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5376 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5377 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5378 type with the same number of elements as ``ty``
5383 The '``fptoui``' instruction converts its :ref:`floating
5384 point <t_floating>` operand into the nearest (rounding towards zero)
5385 unsigned integer value. If the value cannot fit in ``ty2``, the results
5391 .. code-block:: llvm
5393 %X = fptoui double 123.0 to i32 ; yields i32:123
5394 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5395 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5397 '``fptosi .. to``' Instruction
5398 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5405 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5410 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5411 ``value`` to type ``ty2``.
5416 The '``fptosi``' instruction takes a value to cast, which must be a
5417 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5418 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5419 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5420 type with the same number of elements as ``ty``
5425 The '``fptosi``' instruction converts its :ref:`floating
5426 point <t_floating>` operand into the nearest (rounding towards zero)
5427 signed integer value. If the value cannot fit in ``ty2``, the results
5433 .. code-block:: llvm
5435 %X = fptosi double -123.0 to i32 ; yields i32:-123
5436 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5437 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5439 '``uitofp .. to``' Instruction
5440 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5447 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5452 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5453 and converts that value to the ``ty2`` type.
5458 The '``uitofp``' instruction takes a value to cast, which must be a
5459 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5460 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5461 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5462 type with the same number of elements as ``ty``
5467 The '``uitofp``' instruction interprets its operand as an unsigned
5468 integer quantity and converts it to the corresponding floating point
5469 value. If the value cannot fit in the floating point value, the results
5475 .. code-block:: llvm
5477 %X = uitofp i32 257 to float ; yields float:257.0
5478 %Y = uitofp i8 -1 to double ; yields double:255.0
5480 '``sitofp .. to``' Instruction
5481 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5488 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5493 The '``sitofp``' instruction regards ``value`` as a signed integer and
5494 converts that value to the ``ty2`` type.
5499 The '``sitofp``' instruction takes a value to cast, which must be a
5500 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5501 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5502 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5503 type with the same number of elements as ``ty``
5508 The '``sitofp``' instruction interprets its operand as a signed integer
5509 quantity and converts it to the corresponding floating point value. If
5510 the value cannot fit in the floating point value, the results are
5516 .. code-block:: llvm
5518 %X = sitofp i32 257 to float ; yields float:257.0
5519 %Y = sitofp i8 -1 to double ; yields double:-1.0
5523 '``ptrtoint .. to``' Instruction
5524 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5531 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5536 The '``ptrtoint``' instruction converts the pointer or a vector of
5537 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5542 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5543 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5544 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5545 a vector of integers type.
5550 The '``ptrtoint``' instruction converts ``value`` to integer type
5551 ``ty2`` by interpreting the pointer value as an integer and either
5552 truncating or zero extending that value to the size of the integer type.
5553 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5554 ``value`` is larger than ``ty2`` then a truncation is done. If they are
5555 the same size, then nothing is done (*no-op cast*) other than a type
5561 .. code-block:: llvm
5563 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5564 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5565 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5569 '``inttoptr .. to``' Instruction
5570 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5577 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5582 The '``inttoptr``' instruction converts an integer ``value`` to a
5583 pointer type, ``ty2``.
5588 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5589 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5595 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5596 applying either a zero extension or a truncation depending on the size
5597 of the integer ``value``. If ``value`` is larger than the size of a
5598 pointer then a truncation is done. If ``value`` is smaller than the size
5599 of a pointer then a zero extension is done. If they are the same size,
5600 nothing is done (*no-op cast*).
5605 .. code-block:: llvm
5607 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5608 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5609 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5610 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5614 '``bitcast .. to``' Instruction
5615 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5622 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5627 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5633 The '``bitcast``' instruction takes a value to cast, which must be a
5634 non-aggregate first class value, and a type to cast it to, which must
5635 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
5636 bit sizes of ``value`` and the destination type, ``ty2``, must be
5637 identical. If the source type is a pointer, the destination type must
5638 also be a pointer of the same size. This instruction supports bitwise
5639 conversion of vectors to integers and to vectors of other types (as
5640 long as they have the same size).
5645 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
5646 is always a *no-op cast* because no bits change with this
5647 conversion. The conversion is done as if the ``value`` had been stored
5648 to memory and read back as type ``ty2``. Pointer (or vector of
5649 pointers) types may only be converted to other pointer (or vector of
5650 pointers) types with this instruction if the pointer sizes are
5651 equal. To convert pointers to other types, use the :ref:`inttoptr
5652 <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions first.
5657 .. code-block:: llvm
5659 %X = bitcast i8 255 to i8 ; yields i8 :-1
5660 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5661 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5662 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5669 The instructions in this category are the "miscellaneous" instructions,
5670 which defy better classification.
5674 '``icmp``' Instruction
5675 ^^^^^^^^^^^^^^^^^^^^^^
5682 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5687 The '``icmp``' instruction returns a boolean value or a vector of
5688 boolean values based on comparison of its two integer, integer vector,
5689 pointer, or pointer vector operands.
5694 The '``icmp``' instruction takes three operands. The first operand is
5695 the condition code indicating the kind of comparison to perform. It is
5696 not a value, just a keyword. The possible condition code are:
5699 #. ``ne``: not equal
5700 #. ``ugt``: unsigned greater than
5701 #. ``uge``: unsigned greater or equal
5702 #. ``ult``: unsigned less than
5703 #. ``ule``: unsigned less or equal
5704 #. ``sgt``: signed greater than
5705 #. ``sge``: signed greater or equal
5706 #. ``slt``: signed less than
5707 #. ``sle``: signed less or equal
5709 The remaining two arguments must be :ref:`integer <t_integer>` or
5710 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5711 must also be identical types.
5716 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5717 code given as ``cond``. The comparison performed always yields either an
5718 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5720 #. ``eq``: yields ``true`` if the operands are equal, ``false``
5721 otherwise. No sign interpretation is necessary or performed.
5722 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
5723 otherwise. No sign interpretation is necessary or performed.
5724 #. ``ugt``: interprets the operands as unsigned values and yields
5725 ``true`` if ``op1`` is greater than ``op2``.
5726 #. ``uge``: interprets the operands as unsigned values and yields
5727 ``true`` if ``op1`` is greater than or equal to ``op2``.
5728 #. ``ult``: interprets the operands as unsigned values and yields
5729 ``true`` if ``op1`` is less than ``op2``.
5730 #. ``ule``: interprets the operands as unsigned values and yields
5731 ``true`` if ``op1`` is less than or equal to ``op2``.
5732 #. ``sgt``: interprets the operands as signed values and yields ``true``
5733 if ``op1`` is greater than ``op2``.
5734 #. ``sge``: interprets the operands as signed values and yields ``true``
5735 if ``op1`` is greater than or equal to ``op2``.
5736 #. ``slt``: interprets the operands as signed values and yields ``true``
5737 if ``op1`` is less than ``op2``.
5738 #. ``sle``: interprets the operands as signed values and yields ``true``
5739 if ``op1`` is less than or equal to ``op2``.
5741 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5742 are compared as if they were integers.
5744 If the operands are integer vectors, then they are compared element by
5745 element. The result is an ``i1`` vector with the same number of elements
5746 as the values being compared. Otherwise, the result is an ``i1``.
5751 .. code-block:: llvm
5753 <result> = icmp eq i32 4, 5 ; yields: result=false
5754 <result> = icmp ne float* %X, %X ; yields: result=false
5755 <result> = icmp ult i16 4, 5 ; yields: result=true
5756 <result> = icmp sgt i16 4, 5 ; yields: result=false
5757 <result> = icmp ule i16 -4, 5 ; yields: result=false
5758 <result> = icmp sge i16 4, 5 ; yields: result=false
5760 Note that the code generator does not yet support vector types with the
5761 ``icmp`` instruction.
5765 '``fcmp``' Instruction
5766 ^^^^^^^^^^^^^^^^^^^^^^
5773 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5778 The '``fcmp``' instruction returns a boolean value or vector of boolean
5779 values based on comparison of its operands.
5781 If the operands are floating point scalars, then the result type is a
5782 boolean (:ref:`i1 <t_integer>`).
5784 If the operands are floating point vectors, then the result type is a
5785 vector of boolean with the same number of elements as the operands being
5791 The '``fcmp``' instruction takes three operands. The first operand is
5792 the condition code indicating the kind of comparison to perform. It is
5793 not a value, just a keyword. The possible condition code are:
5795 #. ``false``: no comparison, always returns false
5796 #. ``oeq``: ordered and equal
5797 #. ``ogt``: ordered and greater than
5798 #. ``oge``: ordered and greater than or equal
5799 #. ``olt``: ordered and less than
5800 #. ``ole``: ordered and less than or equal
5801 #. ``one``: ordered and not equal
5802 #. ``ord``: ordered (no nans)
5803 #. ``ueq``: unordered or equal
5804 #. ``ugt``: unordered or greater than
5805 #. ``uge``: unordered or greater than or equal
5806 #. ``ult``: unordered or less than
5807 #. ``ule``: unordered or less than or equal
5808 #. ``une``: unordered or not equal
5809 #. ``uno``: unordered (either nans)
5810 #. ``true``: no comparison, always returns true
5812 *Ordered* means that neither operand is a QNAN while *unordered* means
5813 that either operand may be a QNAN.
5815 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5816 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5817 type. They must have identical types.
5822 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5823 condition code given as ``cond``. If the operands are vectors, then the
5824 vectors are compared element by element. Each comparison performed
5825 always yields an :ref:`i1 <t_integer>` result, as follows:
5827 #. ``false``: always yields ``false``, regardless of operands.
5828 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5829 is equal to ``op2``.
5830 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5831 is greater than ``op2``.
5832 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5833 is greater than or equal to ``op2``.
5834 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5835 is less than ``op2``.
5836 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5837 is less than or equal to ``op2``.
5838 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5839 is not equal to ``op2``.
5840 #. ``ord``: yields ``true`` if both operands are not a QNAN.
5841 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5843 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5844 greater than ``op2``.
5845 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5846 greater than or equal to ``op2``.
5847 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5849 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5850 less than or equal to ``op2``.
5851 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5852 not equal to ``op2``.
5853 #. ``uno``: yields ``true`` if either operand is a QNAN.
5854 #. ``true``: always yields ``true``, regardless of operands.
5859 .. code-block:: llvm
5861 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5862 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5863 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5864 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5866 Note that the code generator does not yet support vector types with the
5867 ``fcmp`` instruction.
5871 '``phi``' Instruction
5872 ^^^^^^^^^^^^^^^^^^^^^
5879 <result> = phi <ty> [ <val0>, <label0>], ...
5884 The '``phi``' instruction is used to implement the φ node in the SSA
5885 graph representing the function.
5890 The type of the incoming values is specified with the first type field.
5891 After this, the '``phi``' instruction takes a list of pairs as
5892 arguments, with one pair for each predecessor basic block of the current
5893 block. Only values of :ref:`first class <t_firstclass>` type may be used as
5894 the value arguments to the PHI node. Only labels may be used as the
5897 There must be no non-phi instructions between the start of a basic block
5898 and the PHI instructions: i.e. PHI instructions must be first in a basic
5901 For the purposes of the SSA form, the use of each incoming value is
5902 deemed to occur on the edge from the corresponding predecessor block to
5903 the current block (but after any definition of an '``invoke``'
5904 instruction's return value on the same edge).
5909 At runtime, the '``phi``' instruction logically takes on the value
5910 specified by the pair corresponding to the predecessor basic block that
5911 executed just prior to the current block.
5916 .. code-block:: llvm
5918 Loop: ; Infinite loop that counts from 0 on up...
5919 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5920 %nextindvar = add i32 %indvar, 1
5925 '``select``' Instruction
5926 ^^^^^^^^^^^^^^^^^^^^^^^^
5933 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5935 selty is either i1 or {<N x i1>}
5940 The '``select``' instruction is used to choose one value based on a
5941 condition, without branching.
5946 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
5947 values indicating the condition, and two values of the same :ref:`first
5948 class <t_firstclass>` type. If the val1/val2 are vectors and the
5949 condition is a scalar, then entire vectors are selected, not individual
5955 If the condition is an i1 and it evaluates to 1, the instruction returns
5956 the first value argument; otherwise, it returns the second value
5959 If the condition is a vector of i1, then the value arguments must be
5960 vectors of the same size, and the selection is done element by element.
5965 .. code-block:: llvm
5967 %X = select i1 true, i8 17, i8 42 ; yields i8:17
5971 '``call``' Instruction
5972 ^^^^^^^^^^^^^^^^^^^^^^
5979 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
5984 The '``call``' instruction represents a simple function call.
5989 This instruction requires several arguments:
5991 #. The optional "tail" marker indicates that the callee function does
5992 not access any allocas or varargs in the caller. Note that calls may
5993 be marked "tail" even if they do not occur before a
5994 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
5995 function call is eligible for tail call optimization, but `might not
5996 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
5997 The code generator may optimize calls marked "tail" with either 1)
5998 automatic `sibling call
5999 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
6000 callee have matching signatures, or 2) forced tail call optimization
6001 when the following extra requirements are met:
6003 - Caller and callee both have the calling convention ``fastcc``.
6004 - The call is in tail position (ret immediately follows call and ret
6005 uses value of call or is void).
6006 - Option ``-tailcallopt`` is enabled, or
6007 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6008 - `Platform specific constraints are
6009 met. <CodeGenerator.html#tailcallopt>`_
6011 #. The optional "cconv" marker indicates which :ref:`calling
6012 convention <callingconv>` the call should use. If none is
6013 specified, the call defaults to using C calling conventions. The
6014 calling convention of the call must match the calling convention of
6015 the target function, or else the behavior is undefined.
6016 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6017 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6019 #. '``ty``': the type of the call instruction itself which is also the
6020 type of the return value. Functions that return no value are marked
6022 #. '``fnty``': shall be the signature of the pointer to function value
6023 being invoked. The argument types must match the types implied by
6024 this signature. This type can be omitted if the function is not
6025 varargs and if the function type does not return a pointer to a
6027 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6028 be invoked. In most cases, this is a direct function invocation, but
6029 indirect ``call``'s are just as possible, calling an arbitrary pointer
6031 #. '``function args``': argument list whose types match the function
6032 signature argument types and parameter attributes. All arguments must
6033 be of :ref:`first class <t_firstclass>` type. If the function signature
6034 indicates the function accepts a variable number of arguments, the
6035 extra arguments can be specified.
6036 #. The optional :ref:`function attributes <fnattrs>` list. Only
6037 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6038 attributes are valid here.
6043 The '``call``' instruction is used to cause control flow to transfer to
6044 a specified function, with its incoming arguments bound to the specified
6045 values. Upon a '``ret``' instruction in the called function, control
6046 flow continues with the instruction after the function call, and the
6047 return value of the function is bound to the result argument.
6052 .. code-block:: llvm
6054 %retval = call i32 @test(i32 %argc)
6055 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6056 %X = tail call i32 @foo() ; yields i32
6057 %Y = tail call fastcc i32 @foo() ; yields i32
6058 call void %foo(i8 97 signext)
6060 %struct.A = type { i32, i8 }
6061 %r = call %struct.A @foo() ; yields { 32, i8 }
6062 %gr = extractvalue %struct.A %r, 0 ; yields i32
6063 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6064 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6065 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6067 llvm treats calls to some functions with names and arguments that match
6068 the standard C99 library as being the C99 library functions, and may
6069 perform optimizations or generate code for them under that assumption.
6070 This is something we'd like to change in the future to provide better
6071 support for freestanding environments and non-C-based languages.
6075 '``va_arg``' Instruction
6076 ^^^^^^^^^^^^^^^^^^^^^^^^
6083 <resultval> = va_arg <va_list*> <arglist>, <argty>
6088 The '``va_arg``' instruction is used to access arguments passed through
6089 the "variable argument" area of a function call. It is used to implement
6090 the ``va_arg`` macro in C.
6095 This instruction takes a ``va_list*`` value and the type of the
6096 argument. It returns a value of the specified argument type and
6097 increments the ``va_list`` to point to the next argument. The actual
6098 type of ``va_list`` is target specific.
6103 The '``va_arg``' instruction loads an argument of the specified type
6104 from the specified ``va_list`` and causes the ``va_list`` to point to
6105 the next argument. For more information, see the variable argument
6106 handling :ref:`Intrinsic Functions <int_varargs>`.
6108 It is legal for this instruction to be called in a function which does
6109 not take a variable number of arguments, for example, the ``vfprintf``
6112 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6113 function <intrinsics>` because it takes a type as an argument.
6118 See the :ref:`variable argument processing <int_varargs>` section.
6120 Note that the code generator does not yet fully support va\_arg on many
6121 targets. Also, it does not currently support va\_arg with aggregate
6122 types on any target.
6126 '``landingpad``' Instruction
6127 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6134 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6135 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6137 <clause> := catch <type> <value>
6138 <clause> := filter <array constant type> <array constant>
6143 The '``landingpad``' instruction is used by `LLVM's exception handling
6144 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6145 is a landing pad --- one where the exception lands, and corresponds to the
6146 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6147 defines values supplied by the personality function (``pers_fn``) upon
6148 re-entry to the function. The ``resultval`` has the type ``resultty``.
6153 This instruction takes a ``pers_fn`` value. This is the personality
6154 function associated with the unwinding mechanism. The optional
6155 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6157 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6158 contains the global variable representing the "type" that may be caught
6159 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6160 clause takes an array constant as its argument. Use
6161 "``[0 x i8**] undef``" for a filter which cannot throw. The
6162 '``landingpad``' instruction must contain *at least* one ``clause`` or
6163 the ``cleanup`` flag.
6168 The '``landingpad``' instruction defines the values which are set by the
6169 personality function (``pers_fn``) upon re-entry to the function, and
6170 therefore the "result type" of the ``landingpad`` instruction. As with
6171 calling conventions, how the personality function results are
6172 represented in LLVM IR is target specific.
6174 The clauses are applied in order from top to bottom. If two
6175 ``landingpad`` instructions are merged together through inlining, the
6176 clauses from the calling function are appended to the list of clauses.
6177 When the call stack is being unwound due to an exception being thrown,
6178 the exception is compared against each ``clause`` in turn. If it doesn't
6179 match any of the clauses, and the ``cleanup`` flag is not set, then
6180 unwinding continues further up the call stack.
6182 The ``landingpad`` instruction has several restrictions:
6184 - A landing pad block is a basic block which is the unwind destination
6185 of an '``invoke``' instruction.
6186 - A landing pad block must have a '``landingpad``' instruction as its
6187 first non-PHI instruction.
6188 - There can be only one '``landingpad``' instruction within the landing
6190 - A basic block that is not a landing pad block may not include a
6191 '``landingpad``' instruction.
6192 - All '``landingpad``' instructions in a function must have the same
6193 personality function.
6198 .. code-block:: llvm
6200 ;; A landing pad which can catch an integer.
6201 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6203 ;; A landing pad that is a cleanup.
6204 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6206 ;; A landing pad which can catch an integer and can only throw a double.
6207 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6209 filter [1 x i8**] [@_ZTId]
6216 LLVM supports the notion of an "intrinsic function". These functions
6217 have well known names and semantics and are required to follow certain
6218 restrictions. Overall, these intrinsics represent an extension mechanism
6219 for the LLVM language that does not require changing all of the
6220 transformations in LLVM when adding to the language (or the bitcode
6221 reader/writer, the parser, etc...).
6223 Intrinsic function names must all start with an "``llvm.``" prefix. This
6224 prefix is reserved in LLVM for intrinsic names; thus, function names may
6225 not begin with this prefix. Intrinsic functions must always be external
6226 functions: you cannot define the body of intrinsic functions. Intrinsic
6227 functions may only be used in call or invoke instructions: it is illegal
6228 to take the address of an intrinsic function. Additionally, because
6229 intrinsic functions are part of the LLVM language, it is required if any
6230 are added that they be documented here.
6232 Some intrinsic functions can be overloaded, i.e., the intrinsic
6233 represents a family of functions that perform the same operation but on
6234 different data types. Because LLVM can represent over 8 million
6235 different integer types, overloading is used commonly to allow an
6236 intrinsic function to operate on any integer type. One or more of the
6237 argument types or the result type can be overloaded to accept any
6238 integer type. Argument types may also be defined as exactly matching a
6239 previous argument's type or the result type. This allows an intrinsic
6240 function which accepts multiple arguments, but needs all of them to be
6241 of the same type, to only be overloaded with respect to a single
6242 argument or the result.
6244 Overloaded intrinsics will have the names of its overloaded argument
6245 types encoded into its function name, each preceded by a period. Only
6246 those types which are overloaded result in a name suffix. Arguments
6247 whose type is matched against another type do not. For example, the
6248 ``llvm.ctpop`` function can take an integer of any width and returns an
6249 integer of exactly the same integer width. This leads to a family of
6250 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6251 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6252 overloaded, and only one type suffix is required. Because the argument's
6253 type is matched against the return type, it does not require its own
6256 To learn how to add an intrinsic function, please see the `Extending
6257 LLVM Guide <ExtendingLLVM.html>`_.
6261 Variable Argument Handling Intrinsics
6262 -------------------------------------
6264 Variable argument support is defined in LLVM with the
6265 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6266 functions. These functions are related to the similarly named macros
6267 defined in the ``<stdarg.h>`` header file.
6269 All of these functions operate on arguments that use a target-specific
6270 value type "``va_list``". The LLVM assembly language reference manual
6271 does not define what this type is, so all transformations should be
6272 prepared to handle these functions regardless of the type used.
6274 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6275 variable argument handling intrinsic functions are used.
6277 .. code-block:: llvm
6279 define i32 @test(i32 %X, ...) {
6280 ; Initialize variable argument processing
6282 %ap2 = bitcast i8** %ap to i8*
6283 call void @llvm.va_start(i8* %ap2)
6285 ; Read a single integer argument
6286 %tmp = va_arg i8** %ap, i32
6288 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6290 %aq2 = bitcast i8** %aq to i8*
6291 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6292 call void @llvm.va_end(i8* %aq2)
6294 ; Stop processing of arguments.
6295 call void @llvm.va_end(i8* %ap2)
6299 declare void @llvm.va_start(i8*)
6300 declare void @llvm.va_copy(i8*, i8*)
6301 declare void @llvm.va_end(i8*)
6305 '``llvm.va_start``' Intrinsic
6306 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6313 declare void %llvm.va_start(i8* <arglist>)
6318 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6319 subsequent use by ``va_arg``.
6324 The argument is a pointer to a ``va_list`` element to initialize.
6329 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6330 available in C. In a target-dependent way, it initializes the
6331 ``va_list`` element to which the argument points, so that the next call
6332 to ``va_arg`` will produce the first variable argument passed to the
6333 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6334 to know the last argument of the function as the compiler can figure
6337 '``llvm.va_end``' Intrinsic
6338 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6345 declare void @llvm.va_end(i8* <arglist>)
6350 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6351 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6356 The argument is a pointer to a ``va_list`` to destroy.
6361 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6362 available in C. In a target-dependent way, it destroys the ``va_list``
6363 element to which the argument points. Calls to
6364 :ref:`llvm.va_start <int_va_start>` and
6365 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6370 '``llvm.va_copy``' Intrinsic
6371 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6378 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6383 The '``llvm.va_copy``' intrinsic copies the current argument position
6384 from the source argument list to the destination argument list.
6389 The first argument is a pointer to a ``va_list`` element to initialize.
6390 The second argument is a pointer to a ``va_list`` element to copy from.
6395 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6396 available in C. In a target-dependent way, it copies the source
6397 ``va_list`` element into the destination ``va_list`` element. This
6398 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6399 arbitrarily complex and require, for example, memory allocation.
6401 Accurate Garbage Collection Intrinsics
6402 --------------------------------------
6404 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6405 (GC) requires the implementation and generation of these intrinsics.
6406 These intrinsics allow identification of :ref:`GC roots on the
6407 stack <int_gcroot>`, as well as garbage collector implementations that
6408 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6409 Front-ends for type-safe garbage collected languages should generate
6410 these intrinsics to make use of the LLVM garbage collectors. For more
6411 details, see `Accurate Garbage Collection with
6412 LLVM <GarbageCollection.html>`_.
6414 The garbage collection intrinsics only operate on objects in the generic
6415 address space (address space zero).
6419 '``llvm.gcroot``' Intrinsic
6420 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6427 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6432 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6433 the code generator, and allows some metadata to be associated with it.
6438 The first argument specifies the address of a stack object that contains
6439 the root pointer. The second pointer (which must be either a constant or
6440 a global value address) contains the meta-data to be associated with the
6446 At runtime, a call to this intrinsic stores a null pointer into the
6447 "ptrloc" location. At compile-time, the code generator generates
6448 information to allow the runtime to find the pointer at GC safe points.
6449 The '``llvm.gcroot``' intrinsic may only be used in a function which
6450 :ref:`specifies a GC algorithm <gc>`.
6454 '``llvm.gcread``' Intrinsic
6455 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6462 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6467 The '``llvm.gcread``' intrinsic identifies reads of references from heap
6468 locations, allowing garbage collector implementations that require read
6474 The second argument is the address to read from, which should be an
6475 address allocated from the garbage collector. The first object is a
6476 pointer to the start of the referenced object, if needed by the language
6477 runtime (otherwise null).
6482 The '``llvm.gcread``' intrinsic has the same semantics as a load
6483 instruction, but may be replaced with substantially more complex code by
6484 the garbage collector runtime, as needed. The '``llvm.gcread``'
6485 intrinsic may only be used in a function which :ref:`specifies a GC
6490 '``llvm.gcwrite``' Intrinsic
6491 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6498 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6503 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6504 locations, allowing garbage collector implementations that require write
6505 barriers (such as generational or reference counting collectors).
6510 The first argument is the reference to store, the second is the start of
6511 the object to store it to, and the third is the address of the field of
6512 Obj to store to. If the runtime does not require a pointer to the
6513 object, Obj may be null.
6518 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6519 instruction, but may be replaced with substantially more complex code by
6520 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6521 intrinsic may only be used in a function which :ref:`specifies a GC
6524 Code Generator Intrinsics
6525 -------------------------
6527 These intrinsics are provided by LLVM to expose special features that
6528 may only be implemented with code generator support.
6530 '``llvm.returnaddress``' Intrinsic
6531 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6538 declare i8 *@llvm.returnaddress(i32 <level>)
6543 The '``llvm.returnaddress``' intrinsic attempts to compute a
6544 target-specific value indicating the return address of the current
6545 function or one of its callers.
6550 The argument to this intrinsic indicates which function to return the
6551 address for. Zero indicates the calling function, one indicates its
6552 caller, etc. The argument is **required** to be a constant integer
6558 The '``llvm.returnaddress``' intrinsic either returns a pointer
6559 indicating the return address of the specified call frame, or zero if it
6560 cannot be identified. The value returned by this intrinsic is likely to
6561 be incorrect or 0 for arguments other than zero, so it should only be
6562 used for debugging purposes.
6564 Note that calling this intrinsic does not prevent function inlining or
6565 other aggressive transformations, so the value returned may not be that
6566 of the obvious source-language caller.
6568 '``llvm.frameaddress``' Intrinsic
6569 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6576 declare i8* @llvm.frameaddress(i32 <level>)
6581 The '``llvm.frameaddress``' intrinsic attempts to return the
6582 target-specific frame pointer value for the specified stack frame.
6587 The argument to this intrinsic indicates which function to return the
6588 frame pointer for. Zero indicates the calling function, one indicates
6589 its caller, etc. The argument is **required** to be a constant integer
6595 The '``llvm.frameaddress``' intrinsic either returns a pointer
6596 indicating the frame address of the specified call frame, or zero if it
6597 cannot be identified. The value returned by this intrinsic is likely to
6598 be incorrect or 0 for arguments other than zero, so it should only be
6599 used for debugging purposes.
6601 Note that calling this intrinsic does not prevent function inlining or
6602 other aggressive transformations, so the value returned may not be that
6603 of the obvious source-language caller.
6607 '``llvm.stacksave``' Intrinsic
6608 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6615 declare i8* @llvm.stacksave()
6620 The '``llvm.stacksave``' intrinsic is used to remember the current state
6621 of the function stack, for use with
6622 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6623 implementing language features like scoped automatic variable sized
6629 This intrinsic returns a opaque pointer value that can be passed to
6630 :ref:`llvm.stackrestore <int_stackrestore>`. When an
6631 ``llvm.stackrestore`` intrinsic is executed with a value saved from
6632 ``llvm.stacksave``, it effectively restores the state of the stack to
6633 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6634 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6635 were allocated after the ``llvm.stacksave`` was executed.
6637 .. _int_stackrestore:
6639 '``llvm.stackrestore``' Intrinsic
6640 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6647 declare void @llvm.stackrestore(i8* %ptr)
6652 The '``llvm.stackrestore``' intrinsic is used to restore the state of
6653 the function stack to the state it was in when the corresponding
6654 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6655 useful for implementing language features like scoped automatic variable
6656 sized arrays in C99.
6661 See the description for :ref:`llvm.stacksave <int_stacksave>`.
6663 '``llvm.prefetch``' Intrinsic
6664 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6671 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6676 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6677 insert a prefetch instruction if supported; otherwise, it is a noop.
6678 Prefetches have no effect on the behavior of the program but can change
6679 its performance characteristics.
6684 ``address`` is the address to be prefetched, ``rw`` is the specifier
6685 determining if the fetch should be for a read (0) or write (1), and
6686 ``locality`` is a temporal locality specifier ranging from (0) - no
6687 locality, to (3) - extremely local keep in cache. The ``cache type``
6688 specifies whether the prefetch is performed on the data (1) or
6689 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6690 arguments must be constant integers.
6695 This intrinsic does not modify the behavior of the program. In
6696 particular, prefetches cannot trap and do not produce a value. On
6697 targets that support this intrinsic, the prefetch can provide hints to
6698 the processor cache for better performance.
6700 '``llvm.pcmarker``' Intrinsic
6701 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6708 declare void @llvm.pcmarker(i32 <id>)
6713 The '``llvm.pcmarker``' intrinsic is a method to export a Program
6714 Counter (PC) in a region of code to simulators and other tools. The
6715 method is target specific, but it is expected that the marker will use
6716 exported symbols to transmit the PC of the marker. The marker makes no
6717 guarantees that it will remain with any specific instruction after
6718 optimizations. It is possible that the presence of a marker will inhibit
6719 optimizations. The intended use is to be inserted after optimizations to
6720 allow correlations of simulation runs.
6725 ``id`` is a numerical id identifying the marker.
6730 This intrinsic does not modify the behavior of the program. Backends
6731 that do not support this intrinsic may ignore it.
6733 '``llvm.readcyclecounter``' Intrinsic
6734 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6741 declare i64 @llvm.readcyclecounter()
6746 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6747 counter register (or similar low latency, high accuracy clocks) on those
6748 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6749 should map to RPCC. As the backing counters overflow quickly (on the
6750 order of 9 seconds on alpha), this should only be used for small
6756 When directly supported, reading the cycle counter should not modify any
6757 memory. Implementations are allowed to either return a application
6758 specific value or a system wide value. On backends without support, this
6759 is lowered to a constant 0.
6761 Note that runtime support may be conditional on the privilege-level code is
6762 running at and the host platform.
6764 Standard C Library Intrinsics
6765 -----------------------------
6767 LLVM provides intrinsics for a few important standard C library
6768 functions. These intrinsics allow source-language front-ends to pass
6769 information about the alignment of the pointer arguments to the code
6770 generator, providing opportunity for more efficient code generation.
6774 '``llvm.memcpy``' Intrinsic
6775 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6780 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6781 integer bit width and for different address spaces. Not all targets
6782 support all bit widths however.
6786 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6787 i32 <len>, i32 <align>, i1 <isvolatile>)
6788 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6789 i64 <len>, i32 <align>, i1 <isvolatile>)
6794 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6795 source location to the destination location.
6797 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6798 intrinsics do not return a value, takes extra alignment/isvolatile
6799 arguments and the pointers can be in specified address spaces.
6804 The first argument is a pointer to the destination, the second is a
6805 pointer to the source. The third argument is an integer argument
6806 specifying the number of bytes to copy, the fourth argument is the
6807 alignment of the source and destination locations, and the fifth is a
6808 boolean indicating a volatile access.
6810 If the call to this intrinsic has an alignment value that is not 0 or 1,
6811 then the caller guarantees that both the source and destination pointers
6812 are aligned to that boundary.
6814 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6815 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6816 very cleanly specified and it is unwise to depend on it.
6821 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6822 source location to the destination location, which are not allowed to
6823 overlap. It copies "len" bytes of memory over. If the argument is known
6824 to be aligned to some boundary, this can be specified as the fourth
6825 argument, otherwise it should be set to 0 or 1.
6827 '``llvm.memmove``' Intrinsic
6828 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6833 This is an overloaded intrinsic. You can use llvm.memmove on any integer
6834 bit width and for different address space. Not all targets support all
6839 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6840 i32 <len>, i32 <align>, i1 <isvolatile>)
6841 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6842 i64 <len>, i32 <align>, i1 <isvolatile>)
6847 The '``llvm.memmove.*``' intrinsics move a block of memory from the
6848 source location to the destination location. It is similar to the
6849 '``llvm.memcpy``' intrinsic but allows the two memory locations to
6852 Note that, unlike the standard libc function, the ``llvm.memmove.*``
6853 intrinsics do not return a value, takes extra alignment/isvolatile
6854 arguments and the pointers can be in specified address spaces.
6859 The first argument is a pointer to the destination, the second is a
6860 pointer to the source. The third argument is an integer argument
6861 specifying the number of bytes to copy, the fourth argument is the
6862 alignment of the source and destination locations, and the fifth is a
6863 boolean indicating a volatile access.
6865 If the call to this intrinsic has an alignment value that is not 0 or 1,
6866 then the caller guarantees that the source and destination pointers are
6867 aligned to that boundary.
6869 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6870 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6871 not very cleanly specified and it is unwise to depend on it.
6876 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6877 source location to the destination location, which may overlap. It
6878 copies "len" bytes of memory over. If the argument is known to be
6879 aligned to some boundary, this can be specified as the fourth argument,
6880 otherwise it should be set to 0 or 1.
6882 '``llvm.memset.*``' Intrinsics
6883 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6888 This is an overloaded intrinsic. You can use llvm.memset on any integer
6889 bit width and for different address spaces. However, not all targets
6890 support all bit widths.
6894 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6895 i32 <len>, i32 <align>, i1 <isvolatile>)
6896 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6897 i64 <len>, i32 <align>, i1 <isvolatile>)
6902 The '``llvm.memset.*``' intrinsics fill a block of memory with a
6903 particular byte value.
6905 Note that, unlike the standard libc function, the ``llvm.memset``
6906 intrinsic does not return a value and takes extra alignment/volatile
6907 arguments. Also, the destination can be in an arbitrary address space.
6912 The first argument is a pointer to the destination to fill, the second
6913 is the byte value with which to fill it, the third argument is an
6914 integer argument specifying the number of bytes to fill, and the fourth
6915 argument is the known alignment of the destination location.
6917 If the call to this intrinsic has an alignment value that is not 0 or 1,
6918 then the caller guarantees that the destination pointer is aligned to
6921 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6922 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6923 very cleanly specified and it is unwise to depend on it.
6928 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6929 at the destination location. If the argument is known to be aligned to
6930 some boundary, this can be specified as the fourth argument, otherwise
6931 it should be set to 0 or 1.
6933 '``llvm.sqrt.*``' Intrinsic
6934 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6939 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6940 floating point or vector of floating point type. Not all targets support
6945 declare float @llvm.sqrt.f32(float %Val)
6946 declare double @llvm.sqrt.f64(double %Val)
6947 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6948 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6949 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6954 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
6955 returning the same value as the libm '``sqrt``' functions would. Unlike
6956 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
6957 negative numbers other than -0.0 (which allows for better optimization,
6958 because there is no need to worry about errno being set).
6959 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
6964 The argument and return value are floating point numbers of the same
6970 This function returns the sqrt of the specified operand if it is a
6971 nonnegative floating point number.
6973 '``llvm.powi.*``' Intrinsic
6974 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6979 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
6980 floating point or vector of floating point type. Not all targets support
6985 declare float @llvm.powi.f32(float %Val, i32 %power)
6986 declare double @llvm.powi.f64(double %Val, i32 %power)
6987 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6988 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6989 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6994 The '``llvm.powi.*``' intrinsics return the first operand raised to the
6995 specified (positive or negative) power. The order of evaluation of
6996 multiplications is not defined. When a vector of floating point type is
6997 used, the second argument remains a scalar integer value.
7002 The second argument is an integer power, and the first is a value to
7003 raise to that power.
7008 This function returns the first value raised to the second power with an
7009 unspecified sequence of rounding operations.
7011 '``llvm.sin.*``' Intrinsic
7012 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7017 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7018 floating point or vector of floating point type. Not all targets support
7023 declare float @llvm.sin.f32(float %Val)
7024 declare double @llvm.sin.f64(double %Val)
7025 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7026 declare fp128 @llvm.sin.f128(fp128 %Val)
7027 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7032 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7037 The argument and return value are floating point numbers of the same
7043 This function returns the sine of the specified operand, returning the
7044 same values as the libm ``sin`` functions would, and handles error
7045 conditions in the same way.
7047 '``llvm.cos.*``' Intrinsic
7048 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7053 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7054 floating point or vector of floating point type. Not all targets support
7059 declare float @llvm.cos.f32(float %Val)
7060 declare double @llvm.cos.f64(double %Val)
7061 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7062 declare fp128 @llvm.cos.f128(fp128 %Val)
7063 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7068 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7073 The argument and return value are floating point numbers of the same
7079 This function returns the cosine of the specified operand, returning the
7080 same values as the libm ``cos`` functions would, and handles error
7081 conditions in the same way.
7083 '``llvm.pow.*``' Intrinsic
7084 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7089 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7090 floating point or vector of floating point type. Not all targets support
7095 declare float @llvm.pow.f32(float %Val, float %Power)
7096 declare double @llvm.pow.f64(double %Val, double %Power)
7097 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7098 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7099 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7104 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7105 specified (positive or negative) power.
7110 The second argument is a floating point power, and the first is a value
7111 to raise to that power.
7116 This function returns the first value raised to the second power,
7117 returning the same values as the libm ``pow`` functions would, and
7118 handles error conditions in the same way.
7120 '``llvm.exp.*``' Intrinsic
7121 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7126 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7127 floating point or vector of floating point type. Not all targets support
7132 declare float @llvm.exp.f32(float %Val)
7133 declare double @llvm.exp.f64(double %Val)
7134 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7135 declare fp128 @llvm.exp.f128(fp128 %Val)
7136 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7141 The '``llvm.exp.*``' intrinsics perform the exp function.
7146 The argument and return value are floating point numbers of the same
7152 This function returns the same values as the libm ``exp`` functions
7153 would, and handles error conditions in the same way.
7155 '``llvm.exp2.*``' Intrinsic
7156 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7161 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7162 floating point or vector of floating point type. Not all targets support
7167 declare float @llvm.exp2.f32(float %Val)
7168 declare double @llvm.exp2.f64(double %Val)
7169 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7170 declare fp128 @llvm.exp2.f128(fp128 %Val)
7171 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7176 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7181 The argument and return value are floating point numbers of the same
7187 This function returns the same values as the libm ``exp2`` functions
7188 would, and handles error conditions in the same way.
7190 '``llvm.log.*``' Intrinsic
7191 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7196 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7197 floating point or vector of floating point type. Not all targets support
7202 declare float @llvm.log.f32(float %Val)
7203 declare double @llvm.log.f64(double %Val)
7204 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7205 declare fp128 @llvm.log.f128(fp128 %Val)
7206 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7211 The '``llvm.log.*``' intrinsics perform the log function.
7216 The argument and return value are floating point numbers of the same
7222 This function returns the same values as the libm ``log`` functions
7223 would, and handles error conditions in the same way.
7225 '``llvm.log10.*``' Intrinsic
7226 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7231 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7232 floating point or vector of floating point type. Not all targets support
7237 declare float @llvm.log10.f32(float %Val)
7238 declare double @llvm.log10.f64(double %Val)
7239 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7240 declare fp128 @llvm.log10.f128(fp128 %Val)
7241 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7246 The '``llvm.log10.*``' intrinsics perform the log10 function.
7251 The argument and return value are floating point numbers of the same
7257 This function returns the same values as the libm ``log10`` functions
7258 would, and handles error conditions in the same way.
7260 '``llvm.log2.*``' Intrinsic
7261 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7266 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7267 floating point or vector of floating point type. Not all targets support
7272 declare float @llvm.log2.f32(float %Val)
7273 declare double @llvm.log2.f64(double %Val)
7274 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7275 declare fp128 @llvm.log2.f128(fp128 %Val)
7276 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7281 The '``llvm.log2.*``' intrinsics perform the log2 function.
7286 The argument and return value are floating point numbers of the same
7292 This function returns the same values as the libm ``log2`` functions
7293 would, and handles error conditions in the same way.
7295 '``llvm.fma.*``' Intrinsic
7296 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7301 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7302 floating point or vector of floating point type. Not all targets support
7307 declare float @llvm.fma.f32(float %a, float %b, float %c)
7308 declare double @llvm.fma.f64(double %a, double %b, double %c)
7309 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7310 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7311 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7316 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7322 The argument and return value are floating point numbers of the same
7328 This function returns the same values as the libm ``fma`` functions
7331 '``llvm.fabs.*``' Intrinsic
7332 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7337 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7338 floating point or vector of floating point type. Not all targets support
7343 declare float @llvm.fabs.f32(float %Val)
7344 declare double @llvm.fabs.f64(double %Val)
7345 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7346 declare fp128 @llvm.fabs.f128(fp128 %Val)
7347 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7352 The '``llvm.fabs.*``' intrinsics return the absolute value of the
7358 The argument and return value are floating point numbers of the same
7364 This function returns the same values as the libm ``fabs`` functions
7365 would, and handles error conditions in the same way.
7367 '``llvm.copysign.*``' Intrinsic
7368 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7373 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
7374 floating point or vector of floating point type. Not all targets support
7379 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
7380 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
7381 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
7382 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
7383 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
7388 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
7389 first operand and the sign of the second operand.
7394 The arguments and return value are floating point numbers of the same
7400 This function returns the same values as the libm ``copysign``
7401 functions would, and handles error conditions in the same way.
7403 '``llvm.floor.*``' Intrinsic
7404 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7409 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7410 floating point or vector of floating point type. Not all targets support
7415 declare float @llvm.floor.f32(float %Val)
7416 declare double @llvm.floor.f64(double %Val)
7417 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7418 declare fp128 @llvm.floor.f128(fp128 %Val)
7419 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7424 The '``llvm.floor.*``' intrinsics return the floor of the operand.
7429 The argument and return value are floating point numbers of the same
7435 This function returns the same values as the libm ``floor`` functions
7436 would, and handles error conditions in the same way.
7438 '``llvm.ceil.*``' Intrinsic
7439 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7444 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7445 floating point or vector of floating point type. Not all targets support
7450 declare float @llvm.ceil.f32(float %Val)
7451 declare double @llvm.ceil.f64(double %Val)
7452 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7453 declare fp128 @llvm.ceil.f128(fp128 %Val)
7454 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7459 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7464 The argument and return value are floating point numbers of the same
7470 This function returns the same values as the libm ``ceil`` functions
7471 would, and handles error conditions in the same way.
7473 '``llvm.trunc.*``' Intrinsic
7474 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7479 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7480 floating point or vector of floating point type. Not all targets support
7485 declare float @llvm.trunc.f32(float %Val)
7486 declare double @llvm.trunc.f64(double %Val)
7487 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7488 declare fp128 @llvm.trunc.f128(fp128 %Val)
7489 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7494 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7495 nearest integer not larger in magnitude than the operand.
7500 The argument and return value are floating point numbers of the same
7506 This function returns the same values as the libm ``trunc`` functions
7507 would, and handles error conditions in the same way.
7509 '``llvm.rint.*``' Intrinsic
7510 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7515 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7516 floating point or vector of floating point type. Not all targets support
7521 declare float @llvm.rint.f32(float %Val)
7522 declare double @llvm.rint.f64(double %Val)
7523 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7524 declare fp128 @llvm.rint.f128(fp128 %Val)
7525 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7530 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7531 nearest integer. It may raise an inexact floating-point exception if the
7532 operand isn't an integer.
7537 The argument and return value are floating point numbers of the same
7543 This function returns the same values as the libm ``rint`` functions
7544 would, and handles error conditions in the same way.
7546 '``llvm.nearbyint.*``' Intrinsic
7547 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7552 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7553 floating point or vector of floating point type. Not all targets support
7558 declare float @llvm.nearbyint.f32(float %Val)
7559 declare double @llvm.nearbyint.f64(double %Val)
7560 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7561 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7562 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7567 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7573 The argument and return value are floating point numbers of the same
7579 This function returns the same values as the libm ``nearbyint``
7580 functions would, and handles error conditions in the same way.
7582 '``llvm.round.*``' Intrinsic
7583 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7588 This is an overloaded intrinsic. You can use ``llvm.round`` on any
7589 floating point or vector of floating point type. Not all targets support
7594 declare float @llvm.round.f32(float %Val)
7595 declare double @llvm.round.f64(double %Val)
7596 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
7597 declare fp128 @llvm.round.f128(fp128 %Val)
7598 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
7603 The '``llvm.round.*``' intrinsics returns the operand rounded to the
7609 The argument and return value are floating point numbers of the same
7615 This function returns the same values as the libm ``round``
7616 functions would, and handles error conditions in the same way.
7618 Bit Manipulation Intrinsics
7619 ---------------------------
7621 LLVM provides intrinsics for a few important bit manipulation
7622 operations. These allow efficient code generation for some algorithms.
7624 '``llvm.bswap.*``' Intrinsics
7625 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7630 This is an overloaded intrinsic function. You can use bswap on any
7631 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7635 declare i16 @llvm.bswap.i16(i16 <id>)
7636 declare i32 @llvm.bswap.i32(i32 <id>)
7637 declare i64 @llvm.bswap.i64(i64 <id>)
7642 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7643 values with an even number of bytes (positive multiple of 16 bits).
7644 These are useful for performing operations on data that is not in the
7645 target's native byte order.
7650 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7651 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7652 intrinsic returns an i32 value that has the four bytes of the input i32
7653 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7654 returned i32 will have its bytes in 3, 2, 1, 0 order. The
7655 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7656 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7659 '``llvm.ctpop.*``' Intrinsic
7660 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7665 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7666 bit width, or on any vector with integer elements. Not all targets
7667 support all bit widths or vector types, however.
7671 declare i8 @llvm.ctpop.i8(i8 <src>)
7672 declare i16 @llvm.ctpop.i16(i16 <src>)
7673 declare i32 @llvm.ctpop.i32(i32 <src>)
7674 declare i64 @llvm.ctpop.i64(i64 <src>)
7675 declare i256 @llvm.ctpop.i256(i256 <src>)
7676 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7681 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7687 The only argument is the value to be counted. The argument may be of any
7688 integer type, or a vector with integer elements. The return type must
7689 match the argument type.
7694 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7695 each element of a vector.
7697 '``llvm.ctlz.*``' Intrinsic
7698 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7703 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7704 integer bit width, or any vector whose elements are integers. Not all
7705 targets support all bit widths or vector types, however.
7709 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7710 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7711 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7712 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7713 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7714 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7719 The '``llvm.ctlz``' family of intrinsic functions counts the number of
7720 leading zeros in a variable.
7725 The first argument is the value to be counted. This argument may be of
7726 any integer type, or a vectory with integer element type. The return
7727 type must match the first argument type.
7729 The second argument must be a constant and is a flag to indicate whether
7730 the intrinsic should ensure that a zero as the first argument produces a
7731 defined result. Historically some architectures did not provide a
7732 defined result for zero values as efficiently, and many algorithms are
7733 now predicated on avoiding zero-value inputs.
7738 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7739 zeros in a variable, or within each element of the vector. If
7740 ``src == 0`` then the result is the size in bits of the type of ``src``
7741 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7742 ``llvm.ctlz(i32 2) = 30``.
7744 '``llvm.cttz.*``' Intrinsic
7745 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7750 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7751 integer bit width, or any vector of integer elements. Not all targets
7752 support all bit widths or vector types, however.
7756 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7757 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7758 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7759 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7760 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7761 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7766 The '``llvm.cttz``' family of intrinsic functions counts the number of
7772 The first argument is the value to be counted. This argument may be of
7773 any integer type, or a vectory with integer element type. The return
7774 type must match the first argument type.
7776 The second argument must be a constant and is a flag to indicate whether
7777 the intrinsic should ensure that a zero as the first argument produces a
7778 defined result. Historically some architectures did not provide a
7779 defined result for zero values as efficiently, and many algorithms are
7780 now predicated on avoiding zero-value inputs.
7785 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7786 zeros in a variable, or within each element of a vector. If ``src == 0``
7787 then the result is the size in bits of the type of ``src`` if
7788 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7789 ``llvm.cttz(2) = 1``.
7791 Arithmetic with Overflow Intrinsics
7792 -----------------------------------
7794 LLVM provides intrinsics for some arithmetic with overflow operations.
7796 '``llvm.sadd.with.overflow.*``' Intrinsics
7797 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7802 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7803 on any integer bit width.
7807 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7808 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7809 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7814 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7815 a signed addition of the two arguments, and indicate whether an overflow
7816 occurred during the signed summation.
7821 The arguments (%a and %b) and the first element of the result structure
7822 may be of integer types of any bit width, but they must have the same
7823 bit width. The second element of the result structure must be of type
7824 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7830 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7831 a signed addition of the two variables. They return a structure --- the
7832 first element of which is the signed summation, and the second element
7833 of which is a bit specifying if the signed summation resulted in an
7839 .. code-block:: llvm
7841 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7842 %sum = extractvalue {i32, i1} %res, 0
7843 %obit = extractvalue {i32, i1} %res, 1
7844 br i1 %obit, label %overflow, label %normal
7846 '``llvm.uadd.with.overflow.*``' Intrinsics
7847 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7852 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7853 on any integer bit width.
7857 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7858 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7859 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7864 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7865 an unsigned addition of the two arguments, and indicate whether a carry
7866 occurred during the unsigned summation.
7871 The arguments (%a and %b) and the first element of the result structure
7872 may be of integer types of any bit width, but they must have the same
7873 bit width. The second element of the result structure must be of type
7874 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7880 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7881 an unsigned addition of the two arguments. They return a structure --- the
7882 first element of which is the sum, and the second element of which is a
7883 bit specifying if the unsigned summation resulted in a carry.
7888 .. code-block:: llvm
7890 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7891 %sum = extractvalue {i32, i1} %res, 0
7892 %obit = extractvalue {i32, i1} %res, 1
7893 br i1 %obit, label %carry, label %normal
7895 '``llvm.ssub.with.overflow.*``' Intrinsics
7896 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7901 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7902 on any integer bit width.
7906 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7907 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7908 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7913 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7914 a signed subtraction of the two arguments, and indicate whether an
7915 overflow occurred during the signed subtraction.
7920 The arguments (%a and %b) and the first element of the result structure
7921 may be of integer types of any bit width, but they must have the same
7922 bit width. The second element of the result structure must be of type
7923 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7929 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7930 a signed subtraction of the two arguments. They return a structure --- the
7931 first element of which is the subtraction, and the second element of
7932 which is a bit specifying if the signed subtraction resulted in an
7938 .. code-block:: llvm
7940 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7941 %sum = extractvalue {i32, i1} %res, 0
7942 %obit = extractvalue {i32, i1} %res, 1
7943 br i1 %obit, label %overflow, label %normal
7945 '``llvm.usub.with.overflow.*``' Intrinsics
7946 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7951 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
7952 on any integer bit width.
7956 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7957 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7958 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7963 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7964 an unsigned subtraction of the two arguments, and indicate whether an
7965 overflow occurred during the unsigned subtraction.
7970 The arguments (%a and %b) and the first element of the result structure
7971 may be of integer types of any bit width, but they must have the same
7972 bit width. The second element of the result structure must be of type
7973 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7979 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7980 an unsigned subtraction of the two arguments. They return a structure ---
7981 the first element of which is the subtraction, and the second element of
7982 which is a bit specifying if the unsigned subtraction resulted in an
7988 .. code-block:: llvm
7990 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7991 %sum = extractvalue {i32, i1} %res, 0
7992 %obit = extractvalue {i32, i1} %res, 1
7993 br i1 %obit, label %overflow, label %normal
7995 '``llvm.smul.with.overflow.*``' Intrinsics
7996 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8001 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8002 on any integer bit width.
8006 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8007 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8008 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8013 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8014 a signed multiplication of the two arguments, and indicate whether an
8015 overflow occurred during the signed multiplication.
8020 The arguments (%a and %b) and the first element of the result structure
8021 may be of integer types of any bit width, but they must have the same
8022 bit width. The second element of the result structure must be of type
8023 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8029 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8030 a signed multiplication of the two arguments. They return a structure ---
8031 the first element of which is the multiplication, and the second element
8032 of which is a bit specifying if the signed multiplication resulted in an
8038 .. code-block:: llvm
8040 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8041 %sum = extractvalue {i32, i1} %res, 0
8042 %obit = extractvalue {i32, i1} %res, 1
8043 br i1 %obit, label %overflow, label %normal
8045 '``llvm.umul.with.overflow.*``' Intrinsics
8046 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8051 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8052 on any integer bit width.
8056 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8057 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8058 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8063 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8064 a unsigned multiplication of the two arguments, and indicate whether an
8065 overflow occurred during the unsigned multiplication.
8070 The arguments (%a and %b) and the first element of the result structure
8071 may be of integer types of any bit width, but they must have the same
8072 bit width. The second element of the result structure must be of type
8073 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8079 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8080 an unsigned multiplication of the two arguments. They return a structure ---
8081 the first element of which is the multiplication, and the second
8082 element of which is a bit specifying if the unsigned multiplication
8083 resulted in an overflow.
8088 .. code-block:: llvm
8090 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8091 %sum = extractvalue {i32, i1} %res, 0
8092 %obit = extractvalue {i32, i1} %res, 1
8093 br i1 %obit, label %overflow, label %normal
8095 Specialised Arithmetic Intrinsics
8096 ---------------------------------
8098 '``llvm.fmuladd.*``' Intrinsic
8099 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8106 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8107 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8112 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8113 expressions that can be fused if the code generator determines that (a) the
8114 target instruction set has support for a fused operation, and (b) that the
8115 fused operation is more efficient than the equivalent, separate pair of mul
8116 and add instructions.
8121 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8122 multiplicands, a and b, and an addend c.
8131 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8133 is equivalent to the expression a \* b + c, except that rounding will
8134 not be performed between the multiplication and addition steps if the
8135 code generator fuses the operations. Fusion is not guaranteed, even if
8136 the target platform supports it. If a fused multiply-add is required the
8137 corresponding llvm.fma.\* intrinsic function should be used instead.
8142 .. code-block:: llvm
8144 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8146 Half Precision Floating Point Intrinsics
8147 ----------------------------------------
8149 For most target platforms, half precision floating point is a
8150 storage-only format. This means that it is a dense encoding (in memory)
8151 but does not support computation in the format.
8153 This means that code must first load the half-precision floating point
8154 value as an i16, then convert it to float with
8155 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8156 then be performed on the float value (including extending to double
8157 etc). To store the value back to memory, it is first converted to float
8158 if needed, then converted to i16 with
8159 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8162 .. _int_convert_to_fp16:
8164 '``llvm.convert.to.fp16``' Intrinsic
8165 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8172 declare i16 @llvm.convert.to.fp16(f32 %a)
8177 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8178 from single precision floating point format to half precision floating
8184 The intrinsic function contains single argument - the value to be
8190 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
8191 from single precision floating point format to half precision floating
8192 point format. The return value is an ``i16`` which contains the
8198 .. code-block:: llvm
8200 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8201 store i16 %res, i16* @x, align 2
8203 .. _int_convert_from_fp16:
8205 '``llvm.convert.from.fp16``' Intrinsic
8206 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8213 declare f32 @llvm.convert.from.fp16(i16 %a)
8218 The '``llvm.convert.from.fp16``' intrinsic function performs a
8219 conversion from half precision floating point format to single precision
8220 floating point format.
8225 The intrinsic function contains single argument - the value to be
8231 The '``llvm.convert.from.fp16``' intrinsic function performs a
8232 conversion from half single precision floating point format to single
8233 precision floating point format. The input half-float value is
8234 represented by an ``i16`` value.
8239 .. code-block:: llvm
8241 %a = load i16* @x, align 2
8242 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8247 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8248 prefix), are described in the `LLVM Source Level
8249 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8252 Exception Handling Intrinsics
8253 -----------------------------
8255 The LLVM exception handling intrinsics (which all start with
8256 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8257 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8261 Trampoline Intrinsics
8262 ---------------------
8264 These intrinsics make it possible to excise one parameter, marked with
8265 the :ref:`nest <nest>` attribute, from a function. The result is a
8266 callable function pointer lacking the nest parameter - the caller does
8267 not need to provide a value for it. Instead, the value to use is stored
8268 in advance in a "trampoline", a block of memory usually allocated on the
8269 stack, which also contains code to splice the nest value into the
8270 argument list. This is used to implement the GCC nested function address
8273 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8274 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8275 It can be created as follows:
8277 .. code-block:: llvm
8279 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8280 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8281 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8282 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8283 %fp = bitcast i8* %p to i32 (i32, i32)*
8285 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8286 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8290 '``llvm.init.trampoline``' Intrinsic
8291 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8298 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8303 This fills the memory pointed to by ``tramp`` with executable code,
8304 turning it into a trampoline.
8309 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8310 pointers. The ``tramp`` argument must point to a sufficiently large and
8311 sufficiently aligned block of memory; this memory is written to by the
8312 intrinsic. Note that the size and the alignment are target-specific -
8313 LLVM currently provides no portable way of determining them, so a
8314 front-end that generates this intrinsic needs to have some
8315 target-specific knowledge. The ``func`` argument must hold a function
8316 bitcast to an ``i8*``.
8321 The block of memory pointed to by ``tramp`` is filled with target
8322 dependent code, turning it into a function. Then ``tramp`` needs to be
8323 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8324 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8325 function's signature is the same as that of ``func`` with any arguments
8326 marked with the ``nest`` attribute removed. At most one such ``nest``
8327 argument is allowed, and it must be of pointer type. Calling the new
8328 function is equivalent to calling ``func`` with the same argument list,
8329 but with ``nval`` used for the missing ``nest`` argument. If, after
8330 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8331 modified, then the effect of any later call to the returned function
8332 pointer is undefined.
8336 '``llvm.adjust.trampoline``' Intrinsic
8337 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8344 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8349 This performs any required machine-specific adjustment to the address of
8350 a trampoline (passed as ``tramp``).
8355 ``tramp`` must point to a block of memory which already has trampoline
8356 code filled in by a previous call to
8357 :ref:`llvm.init.trampoline <int_it>`.
8362 On some architectures the address of the code to be executed needs to be
8363 different to the address where the trampoline is actually stored. This
8364 intrinsic returns the executable address corresponding to ``tramp``
8365 after performing the required machine specific adjustments. The pointer
8366 returned can then be :ref:`bitcast and executed <int_trampoline>`.
8371 This class of intrinsics exists to information about the lifetime of
8372 memory objects and ranges where variables are immutable.
8374 '``llvm.lifetime.start``' Intrinsic
8375 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8382 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8387 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8393 The first argument is a constant integer representing the size of the
8394 object, or -1 if it is variable sized. The second argument is a pointer
8400 This intrinsic indicates that before this point in the code, the value
8401 of the memory pointed to by ``ptr`` is dead. This means that it is known
8402 to never be used and has an undefined value. A load from the pointer
8403 that precedes this intrinsic can be replaced with ``'undef'``.
8405 '``llvm.lifetime.end``' Intrinsic
8406 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8413 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8418 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8424 The first argument is a constant integer representing the size of the
8425 object, or -1 if it is variable sized. The second argument is a pointer
8431 This intrinsic indicates that after this point in the code, the value of
8432 the memory pointed to by ``ptr`` is dead. This means that it is known to
8433 never be used and has an undefined value. Any stores into the memory
8434 object following this intrinsic may be removed as dead.
8436 '``llvm.invariant.start``' Intrinsic
8437 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8444 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8449 The '``llvm.invariant.start``' intrinsic specifies that the contents of
8450 a memory object will not change.
8455 The first argument is a constant integer representing the size of the
8456 object, or -1 if it is variable sized. The second argument is a pointer
8462 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8463 the return value, the referenced memory location is constant and
8466 '``llvm.invariant.end``' Intrinsic
8467 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8474 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8479 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8480 memory object are mutable.
8485 The first argument is the matching ``llvm.invariant.start`` intrinsic.
8486 The second argument is a constant integer representing the size of the
8487 object, or -1 if it is variable sized and the third argument is a
8488 pointer to the object.
8493 This intrinsic indicates that the memory is mutable again.
8498 This class of intrinsics is designed to be generic and has no specific
8501 '``llvm.var.annotation``' Intrinsic
8502 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8509 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8514 The '``llvm.var.annotation``' intrinsic.
8519 The first argument is a pointer to a value, the second is a pointer to a
8520 global string, the third is a pointer to a global string which is the
8521 source file name, and the last argument is the line number.
8526 This intrinsic allows annotation of local variables with arbitrary
8527 strings. This can be useful for special purpose optimizations that want
8528 to look for these annotations. These have no other defined use; they are
8529 ignored by code generation and optimization.
8531 '``llvm.ptr.annotation.*``' Intrinsic
8532 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8537 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
8538 pointer to an integer of any width. *NOTE* you must specify an address space for
8539 the pointer. The identifier for the default address space is the integer
8544 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8545 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
8546 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
8547 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
8548 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
8553 The '``llvm.ptr.annotation``' intrinsic.
8558 The first argument is a pointer to an integer value of arbitrary bitwidth
8559 (result of some expression), the second is a pointer to a global string, the
8560 third is a pointer to a global string which is the source file name, and the
8561 last argument is the line number. It returns the value of the first argument.
8566 This intrinsic allows annotation of a pointer to an integer with arbitrary
8567 strings. This can be useful for special purpose optimizations that want to look
8568 for these annotations. These have no other defined use; they are ignored by code
8569 generation and optimization.
8571 '``llvm.annotation.*``' Intrinsic
8572 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8577 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8578 any integer bit width.
8582 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8583 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8584 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8585 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8586 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8591 The '``llvm.annotation``' intrinsic.
8596 The first argument is an integer value (result of some expression), the
8597 second is a pointer to a global string, the third is a pointer to a
8598 global string which is the source file name, and the last argument is
8599 the line number. It returns the value of the first argument.
8604 This intrinsic allows annotations to be put on arbitrary expressions
8605 with arbitrary strings. This can be useful for special purpose
8606 optimizations that want to look for these annotations. These have no
8607 other defined use; they are ignored by code generation and optimization.
8609 '``llvm.trap``' Intrinsic
8610 ^^^^^^^^^^^^^^^^^^^^^^^^^
8617 declare void @llvm.trap() noreturn nounwind
8622 The '``llvm.trap``' intrinsic.
8632 This intrinsic is lowered to the target dependent trap instruction. If
8633 the target does not have a trap instruction, this intrinsic will be
8634 lowered to a call of the ``abort()`` function.
8636 '``llvm.debugtrap``' Intrinsic
8637 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8644 declare void @llvm.debugtrap() nounwind
8649 The '``llvm.debugtrap``' intrinsic.
8659 This intrinsic is lowered to code which is intended to cause an
8660 execution trap with the intention of requesting the attention of a
8663 '``llvm.stackprotector``' Intrinsic
8664 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8671 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8676 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8677 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8678 is placed on the stack before local variables.
8683 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8684 The first argument is the value loaded from the stack guard
8685 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8686 enough space to hold the value of the guard.
8691 This intrinsic causes the prologue/epilogue inserter to force the position of
8692 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
8693 to ensure that if a local variable on the stack is overwritten, it will destroy
8694 the value of the guard. When the function exits, the guard on the stack is
8695 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
8696 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
8697 calling the ``__stack_chk_fail()`` function.
8699 '``llvm.stackprotectorcheck``' Intrinsic
8700 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8707 declare void @llvm.stackprotectorcheck(i8** <guard>)
8712 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
8713 created stack protector and if they are not equal calls the
8714 ``__stack_chk_fail()`` function.
8719 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
8720 the variable ``@__stack_chk_guard``.
8725 This intrinsic is provided to perform the stack protector check by comparing
8726 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
8727 values do not match call the ``__stack_chk_fail()`` function.
8729 The reason to provide this as an IR level intrinsic instead of implementing it
8730 via other IR operations is that in order to perform this operation at the IR
8731 level without an intrinsic, one would need to create additional basic blocks to
8732 handle the success/failure cases. This makes it difficult to stop the stack
8733 protector check from disrupting sibling tail calls in Codegen. With this
8734 intrinsic, we are able to generate the stack protector basic blocks late in
8735 codegen after the tail call decision has occured.
8737 '``llvm.objectsize``' Intrinsic
8738 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8745 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8746 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8751 The ``llvm.objectsize`` intrinsic is designed to provide information to
8752 the optimizers to determine at compile time whether a) an operation
8753 (like memcpy) will overflow a buffer that corresponds to an object, or
8754 b) that a runtime check for overflow isn't necessary. An object in this
8755 context means an allocation of a specific class, structure, array, or
8761 The ``llvm.objectsize`` intrinsic takes two arguments. The first
8762 argument is a pointer to or into the ``object``. The second argument is
8763 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8764 or -1 (if false) when the object size is unknown. The second argument
8765 only accepts constants.
8770 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8771 the size of the object concerned. If the size cannot be determined at
8772 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8773 on the ``min`` argument).
8775 '``llvm.expect``' Intrinsic
8776 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8783 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8784 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8789 The ``llvm.expect`` intrinsic provides information about expected (the
8790 most probable) value of ``val``, which can be used by optimizers.
8795 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8796 a value. The second argument is an expected value, this needs to be a
8797 constant value, variables are not allowed.
8802 This intrinsic is lowered to the ``val``.
8804 '``llvm.donothing``' Intrinsic
8805 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8812 declare void @llvm.donothing() nounwind readnone
8817 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8818 only intrinsic that can be called with an invoke instruction.
8828 This intrinsic does nothing, and it's removed by optimizers and ignored