blob: cf9a8c0aed808db02e3399befa9c3c40702ac3a9 [file] [log] [blame]
Sean Silvaf722b002012-12-07 10:36:55 +00001==============================
2LLVM Language Reference Manual
3==============================
4
5.. contents::
6 :local:
7 :depth: 3
8
Sean Silvaf722b002012-12-07 10:36:55 +00009Abstract
10========
11
12This document is a reference manual for the LLVM assembly language. LLVM
13is a Static Single Assignment (SSA) based representation that provides
14type safety, low-level operations, flexibility, and the capability of
15representing 'all' high-level languages cleanly. It is the common code
16representation used throughout all phases of the LLVM compilation
17strategy.
18
19Introduction
20============
21
22The LLVM code representation is designed to be used in three different
23forms: 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
25readable assembly language representation. This allows LLVM to provide a
26powerful intermediate representation for efficient compiler
27transformations and analysis, while providing a natural means to debug
28and visualize the transformations. The three different forms of LLVM are
29all equivalent. This document describes the human readable
30representation and notation.
31
32The LLVM representation aims to be light-weight and low-level while
33being expressive, typed, and extensible at the same time. It aims to be
34a "universal IR" of sorts, by being at a low enough level that
35high-level ideas may be cleanly mapped to it (similar to how
36microprocessors are "universal IR's", allowing many source languages to
37be mapped to them). By providing type information, LLVM can be used as
38the target of optimizations: for example, through pointer analysis, it
39can be proven that a C automatic variable is never accessed outside of
40the current function, allowing it to be promoted to a simple SSA value
41instead of a memory location.
42
43.. _wellformed:
44
45Well-Formedness
46---------------
47
48It is important to note that this document describes 'well formed' LLVM
49assembly language. There is a difference between what the parser accepts
50and what is considered 'well formed'. For example, the following
51instruction is syntactically okay, but not well formed:
52
53.. code-block:: llvm
54
55 %x = add i32 1, %x
56
57because the definition of ``%x`` does not dominate all of its uses. The
58LLVM infrastructure provides a verification pass that may be used to
59verify that an LLVM module is well formed. This pass is automatically
60run by the parser after parsing input assembly and by the optimizer
61before it outputs bitcode. The violations pointed out by the verifier
62pass indicate bugs in transformation passes or input to the parser.
63
64.. _identifiers:
65
66Identifiers
67===========
68
69LLVM identifiers come in two basic types: global and local. Global
70identifiers (functions, global variables) begin with the ``'@'``
71character. Local identifiers (register names, types) begin with the
72``'%'`` character. Additionally, there are three different formats for
73identifiers, for different purposes:
74
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.
86
87LLVM requires that values start with a prefix for two reasons: Compilers
88don't need to worry about name clashes with reserved words, and the set
89of reserved words may be expanded in the future without penalty.
90Additionally, unnamed identifiers allow a compiler to quickly come up
91with a temporary variable without having to avoid symbol table
92conflicts.
93
94Reserved words in LLVM are very similar to reserved words in other
95languages. 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
98with variable names, because none of them start with a prefix character
99(``'%'`` or ``'@'``).
100
101Here is an example of LLVM code to multiply the integer variable
102'``%X``' by 8:
103
104The easy way:
105
106.. code-block:: llvm
107
108 %result = mul i32 %X, 8
109
110After strength reduction:
111
112.. code-block:: llvm
113
Dmitri Gribenko126fde52013-01-26 13:30:13 +0000114 %result = shl i32 %X, 3
Sean Silvaf722b002012-12-07 10:36:55 +0000115
116And the hard way:
117
118.. code-block:: llvm
119
120 %0 = add i32 %X, %X ; yields {i32}:%0
121 %1 = add i32 %0, %0 ; yields {i32}:%1
122 %result = add i32 %1, %1
123
124This last way of multiplying ``%X`` by 8 illustrates several important
125lexical features of LLVM:
126
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
131
132It also shows a convention that we follow in this document. When
133demonstrating instructions, we will follow an instruction with a comment
134that defines the type and name of value produced.
135
136High Level Structure
137====================
138
139Module Structure
140----------------
141
142LLVM programs are composed of ``Module``'s, each of which is a
143translation unit of the input programs. Each module consists of
144functions, global variables, and symbol table entries. Modules may be
145combined together with the LLVM linker, which merges function (and
146global variable) definitions, resolves forward declarations, and merges
147symbol table entries. Here is an example of the "hello world" module:
148
149.. code-block:: llvm
150
Daniel Dunbar3389dbc2013-01-17 18:57:32 +0000151 ; Declare the string constant as a global constant.
152 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
Sean Silvaf722b002012-12-07 10:36:55 +0000153
Daniel Dunbar3389dbc2013-01-17 18:57:32 +0000154 ; External declaration of the puts function
155 declare i32 @puts(i8* nocapture) nounwind
Sean Silvaf722b002012-12-07 10:36:55 +0000156
157 ; Definition of main function
Daniel Dunbar3389dbc2013-01-17 18:57:32 +0000158 define i32 @main() { ; i32()*
159 ; Convert [13 x i8]* to i8 *...
Sean Silvaf722b002012-12-07 10:36:55 +0000160 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
161
Daniel Dunbar3389dbc2013-01-17 18:57:32 +0000162 ; Call puts function to write out the string to stdout.
Sean Silvaf722b002012-12-07 10:36:55 +0000163 call i32 @puts(i8* %cast210)
Daniel Dunbar3389dbc2013-01-17 18:57:32 +0000164 ret i32 0
Sean Silvaf722b002012-12-07 10:36:55 +0000165 }
166
167 ; Named metadata
168 !1 = metadata !{i32 42}
169 !foo = !{!1, null}
170
171This example is made up of a :ref:`global variable <globalvars>` named
172"``.str``", an external declaration of the "``puts``" function, a
173:ref:`function definition <functionstructure>` for "``main``" and
174:ref:`named metadata <namedmetadatastructure>` "``foo``".
175
176In general, a module is made up of a list of global values (where both
177functions and global variables are global values). Global values are
178represented by a pointer to a memory location (in this case, a pointer
179to an array of char, and a pointer to a function), and have one of the
180following :ref:`linkage types <linkage>`.
181
182.. _linkage:
183
184Linkage Types
185-------------
186
187All Global Variables and Functions have one of the following types of
188linkage:
189
190``private``
191 Global values with "``private``" linkage are only directly
192 accessible by objects in the current module. In particular, linking
193 code into a module with an private global value may cause the
194 private to be renamed as necessary to avoid collisions. Because the
195 symbol is private to the module, all references can be updated. This
196 doesn't show up in any symbol table in the object file.
197``linker_private``
198 Similar to ``private``, but the symbol is passed through the
199 assembler and evaluated by the linker. Unlike normal strong symbols,
200 they are removed by the linker from the final linked image
201 (executable or dynamic library).
202``linker_private_weak``
203 Similar to "``linker_private``", but the symbol is weak. Note that
204 ``linker_private_weak`` symbols are subject to coalescing by the
205 linker. The symbols are removed by the linker from the final linked
206 image (executable or dynamic library).
207``internal``
208 Similar to private, but the value shows as a local symbol
209 (``STB_LOCAL`` in the case of ELF) in the object file. This
210 corresponds to the notion of the '``static``' keyword in C.
211``available_externally``
212 Globals with "``available_externally``" linkage are never emitted
213 into the object file corresponding to the LLVM module. They exist to
214 allow inlining and other optimizations to take place given knowledge
215 of the definition of the global, which is known to be somewhere
216 outside the module. Globals with ``available_externally`` linkage
217 are allowed to be discarded at will, and are otherwise the same as
218 ``linkonce_odr``. This linkage type is only allowed on definitions,
219 not declarations.
220``linkonce``
221 Globals with "``linkonce``" linkage are merged with other globals of
222 the same name when linkage occurs. This can be used to implement
223 some forms of inline functions, templates, or other code which must
224 be generated in each translation unit that uses it, but where the
225 body may be overridden with a more definitive definition later.
226 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
227 that ``linkonce`` linkage does not actually allow the optimizer to
228 inline the body of this function into callers because it doesn't
229 know if this definition of the function is the definitive definition
230 within the program or whether it will be overridden by a stronger
231 definition. To enable inlining and other optimizations, use
232 "``linkonce_odr``" linkage.
233``weak``
234 "``weak``" linkage has the same merging semantics as ``linkonce``
235 linkage, except that unreferenced globals with ``weak`` linkage may
236 not be discarded. This is used for globals that are declared "weak"
237 in C source code.
238``common``
239 "``common``" linkage is most similar to "``weak``" linkage, but they
240 are used for tentative definitions in C, such as "``int X;``" at
241 global scope. Symbols with "``common``" linkage are merged in the
242 same way as ``weak symbols``, and they may not be deleted if
243 unreferenced. ``common`` symbols may not have an explicit section,
244 must have a zero initializer, and may not be marked
245 ':ref:`constant <globalvars>`'. Functions and aliases may not have
246 common linkage.
247
248.. _linkage_appending:
249
250``appending``
251 "``appending``" linkage may only be applied to global variables of
252 pointer to array type. When two global variables with appending
253 linkage are linked together, the two global arrays are appended
254 together. This is the LLVM, typesafe, equivalent of having the
255 system linker append together "sections" with identical names when
256 .o files are linked.
257``extern_weak``
258 The semantics of this linkage follow the ELF object file model: the
259 symbol is weak until linked, if not linked, the symbol becomes null
260 instead of being an undefined reference.
261``linkonce_odr``, ``weak_odr``
262 Some languages allow differing globals to be merged, such as two
263 functions with different semantics. Other languages, such as
264 ``C++``, ensure that only equivalent globals are ever merged (the
Dmitri Gribenkoae4a9ae2013-01-19 20:34:20 +0000265 "one definition rule" --- "ODR"). Such languages can use the
Sean Silvaf722b002012-12-07 10:36:55 +0000266 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
267 global will only be merged with equivalent globals. These linkage
268 types are otherwise the same as their non-``odr`` versions.
269``linkonce_odr_auto_hide``
270 Similar to "``linkonce_odr``", but nothing in the translation unit
271 takes the address of this definition. For instance, functions that
272 had an inline definition, but the compiler decided not to inline it.
273 ``linkonce_odr_auto_hide`` may have only ``default`` visibility. The
274 symbols are removed by the linker from the final linked image
275 (executable or dynamic library).
276``external``
277 If none of the above identifiers are used, the global is externally
278 visible, meaning that it participates in linkage and can be used to
279 resolve external symbol references.
280
281The next two types of linkage are targeted for Microsoft Windows
282platform only. They are designed to support importing (exporting)
283symbols from (to) DLLs (Dynamic Link Libraries).
284
285``dllimport``
286 "``dllimport``" linkage causes the compiler to reference a function
287 or variable via a global pointer to a pointer that is set up by the
288 DLL exporting the symbol. On Microsoft Windows targets, the pointer
289 name is formed by combining ``__imp_`` and the function or variable
290 name.
291``dllexport``
292 "``dllexport``" linkage causes the compiler to provide a global
293 pointer to a pointer in a DLL, so that it can be referenced with the
294 ``dllimport`` attribute. On Microsoft Windows targets, the pointer
295 name is formed by combining ``__imp_`` and the function or variable
296 name.
297
298For example, since the "``.LC0``" variable is defined to be internal, if
299another module defined a "``.LC0``" variable and was linked with this
300one, one of the two would be renamed, preventing a collision. Since
301"``main``" and "``puts``" are external (i.e., lacking any linkage
302declarations), they are accessible outside of the current module.
303
304It is illegal for a function *declaration* to have any linkage type
305other than ``external``, ``dllimport`` or ``extern_weak``.
306
307Aliases can have only ``external``, ``internal``, ``weak`` or
308``weak_odr`` linkages.
309
310.. _callingconv:
311
312Calling Conventions
313-------------------
314
315LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
316:ref:`invokes <i_invoke>` can all have an optional calling convention
317specified for the call. The calling convention of any pair of dynamic
318caller/callee must match, or the behavior of the program is undefined.
319The following calling conventions are supported by LLVM, and more may be
320added in the future:
321
322"``ccc``" - The C calling convention
323 This calling convention (the default if no other calling convention
324 is specified) matches the target C calling conventions. This calling
325 convention supports varargs function calls and tolerates some
326 mismatch in the declared prototype and implemented declaration of
327 the function (as does normal C).
328"``fastcc``" - The fast calling convention
329 This calling convention attempts to make calls as fast as possible
330 (e.g. by passing things in registers). This calling convention
331 allows the target to use whatever tricks it wants to produce fast
332 code for the target, without having to conform to an externally
333 specified ABI (Application Binary Interface). `Tail calls can only
334 be optimized when this, the GHC or the HiPE convention is
335 used. <CodeGenerator.html#id80>`_ This calling convention does not
336 support varargs and requires the prototype of all callees to exactly
337 match the prototype of the function definition.
338"``coldcc``" - The cold calling convention
339 This calling convention attempts to make code in the caller as
340 efficient as possible under the assumption that the call is not
341 commonly executed. As such, these calls often preserve all registers
342 so that the call does not break any live ranges in the caller side.
343 This calling convention does not support varargs and requires the
344 prototype of all callees to exactly match the prototype of the
345 function definition.
346"``cc 10``" - GHC convention
347 This calling convention has been implemented specifically for use by
348 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
349 It passes everything in registers, going to extremes to achieve this
350 by disabling callee save registers. This calling convention should
351 not be used lightly but only for specific situations such as an
352 alternative to the *register pinning* performance technique often
353 used when implementing functional programming languages. At the
354 moment only X86 supports this convention and it has the following
355 limitations:
356
357 - On *X86-32* only supports up to 4 bit type parameters. No
358 floating point types are supported.
359 - On *X86-64* only supports up to 10 bit type parameters and 6
360 floating point parameters.
361
362 This calling convention supports `tail call
363 optimization <CodeGenerator.html#id80>`_ but requires both the
364 caller and callee are using it.
365"``cc 11``" - The HiPE calling convention
366 This calling convention has been implemented specifically for use by
367 the `High-Performance Erlang
368 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
369 native code compiler of the `Ericsson's Open Source Erlang/OTP
370 system <http://www.erlang.org/download.shtml>`_. It uses more
371 registers for argument passing than the ordinary C calling
372 convention and defines no callee-saved registers. The calling
373 convention properly supports `tail call
374 optimization <CodeGenerator.html#id80>`_ but requires that both the
375 caller and the callee use it. It uses a *register pinning*
376 mechanism, similar to GHC's convention, for keeping frequently
377 accessed runtime components pinned to specific hardware registers.
378 At the moment only X86 supports this convention (both 32 and 64
379 bit).
380"``cc <n>``" - Numbered convention
381 Any calling convention may be specified by number, allowing
382 target-specific calling conventions to be used. Target specific
383 calling conventions start at 64.
384
385More calling conventions can be added/defined on an as-needed basis, to
386support Pascal conventions or any other well-known target-independent
387convention.
388
389Visibility Styles
390-----------------
391
392All Global Variables and Functions have one of the following visibility
393styles:
394
395"``default``" - Default style
396 On targets that use the ELF object file format, default visibility
397 means that the declaration is visible to other modules and, in
398 shared libraries, means that the declared entity may be overridden.
399 On Darwin, default visibility means that the declaration is visible
400 to other modules. Default visibility corresponds to "external
401 linkage" in the language.
402"``hidden``" - Hidden style
403 Two declarations of an object with hidden visibility refer to the
404 same object if they are in the same shared object. Usually, hidden
405 visibility indicates that the symbol will not be placed into the
406 dynamic symbol table, so no other module (executable or shared
407 library) can reference it directly.
408"``protected``" - Protected style
409 On ELF, protected visibility indicates that the symbol will be
410 placed in the dynamic symbol table, but that references within the
411 defining module will bind to the local symbol. That is, the symbol
412 cannot be overridden by another module.
413
414Named Types
415-----------
416
417LLVM IR allows you to specify name aliases for certain types. This can
418make it easier to read the IR and make the IR more condensed
419(particularly when recursive types are involved). An example of a name
420specification is:
421
422.. code-block:: llvm
423
424 %mytype = type { %mytype*, i32 }
425
426You may give a name to any :ref:`type <typesystem>` except
427":ref:`void <t_void>`". Type name aliases may be used anywhere a type is
428expected with the syntax "%mytype".
429
430Note that type names are aliases for the structural type that they
431indicate, and that you can therefore specify multiple names for the same
432type. This often leads to confusing behavior when dumping out a .ll
433file. Since LLVM IR uses structural typing, the name is not part of the
434type. When printing out LLVM IR, the printer will pick *one name* to
435render all types of a particular shape. This means that if you have code
436where two different source types end up having the same LLVM type, that
437the dumper will sometimes print the "wrong" or unexpected type. This is
438an important design point and isn't going to change.
439
440.. _globalvars:
441
442Global Variables
443----------------
444
445Global variables define regions of memory allocated at compilation time
446instead of run-time. Global variables may optionally be initialized, may
447have an explicit section to be placed in, and may have an optional
448explicit alignment specified.
449
450A variable may be defined as ``thread_local``, which means that it will
451not be shared by threads (each thread will have a separated copy of the
452variable). Not all targets support thread-local variables. Optionally, a
453TLS model may be specified:
454
455``localdynamic``
456 For variables that are only used within the current shared library.
457``initialexec``
458 For variables in modules that will not be loaded dynamically.
459``localexec``
460 For variables defined in the executable and only used within it.
461
462The models correspond to the ELF TLS models; see `ELF Handling For
463Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
464more information on under which circumstances the different models may
465be used. The target may choose a different TLS model if the specified
466model is not supported, or if a better choice of model can be made.
467
Michael Gottesmanf5735882013-01-31 05:48:48 +0000468A variable may be defined as a global ``constant``, which indicates that
Sean Silvaf722b002012-12-07 10:36:55 +0000469the contents of the variable will **never** be modified (enabling better
470optimization, allowing the global data to be placed in the read-only
471section of an executable, etc). Note that variables that need runtime
Michael Gottesman34804872013-01-31 05:44:04 +0000472initialization cannot be marked ``constant`` as there is a store to the
Sean Silvaf722b002012-12-07 10:36:55 +0000473variable.
474
475LLVM explicitly allows *declarations* of global variables to be marked
476constant, even if the final definition of the global is not. This
477capability can be used to enable slightly better optimization of the
478program, but requires the language definition to guarantee that
479optimizations based on the 'constantness' are valid for the translation
480units that do not include the definition.
481
482As SSA values, global variables define pointer values that are in scope
483(i.e. they dominate) all basic blocks in the program. Global variables
484always define a pointer to their "content" type because they describe a
485region of memory, and all memory objects in LLVM are accessed through
486pointers.
487
488Global variables can be marked with ``unnamed_addr`` which indicates
489that the address is not significant, only the content. Constants marked
490like this can be merged with other constants if they have the same
491initializer. Note that a constant with significant address *can* be
492merged with a ``unnamed_addr`` constant, the result being a constant
493whose address is significant.
494
495A global variable may be declared to reside in a target-specific
496numbered address space. For targets that support them, address spaces
497may affect how optimizations are performed and/or what target
498instructions are used to access the variable. The default address space
499is zero. The address space qualifier must precede any other attributes.
500
501LLVM allows an explicit section to be specified for globals. If the
502target supports it, it will emit globals to the section specified.
503
Michael Gottesman6c355ee2013-02-04 03:22:00 +0000504By default, global initializers are optimized by assuming that global
Michael Gottesman42834992013-02-03 09:57:15 +0000505variables defined within the module are not modified from their
506initial values before the start of the global initializer. This is
507true even for variables potentially accessible from outside the
508module, including those with external linkage or appearing in
Michael Gottesmanfa987f02013-02-03 09:57:18 +0000509``@llvm.used``. This assumption may be suppressed by marking the
510variable with ``externally_initialized``.
Michael Gottesman42834992013-02-03 09:57:15 +0000511
Sean Silvaf722b002012-12-07 10:36:55 +0000512An explicit alignment may be specified for a global, which must be a
513power of 2. If not present, or if the alignment is set to zero, the
514alignment of the global is set by the target to whatever it feels
515convenient. If an explicit alignment is specified, the global is forced
516to have exactly that alignment. Targets and optimizers are not allowed
517to over-align the global if the global has an assigned section. In this
518case, the extra alignment could be observable: for example, code could
519assume that the globals are densely packed in their section and try to
520iterate over them as an array, alignment padding would break this
521iteration.
522
523For example, the following defines a global in a numbered address space
524with an initializer, section, and alignment:
525
526.. code-block:: llvm
527
528 @G = addrspace(5) constant float 1.0, section "foo", align 4
529
530The following example defines a thread-local global with the
531``initialexec`` TLS model:
532
533.. code-block:: llvm
534
535 @G = thread_local(initialexec) global i32 0, align 4
536
537.. _functionstructure:
538
539Functions
540---------
541
542LLVM function definitions consist of the "``define``" keyword, an
543optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
544style <visibility>`, an optional :ref:`calling convention <callingconv>`,
545an optional ``unnamed_addr`` attribute, a return type, an optional
546:ref:`parameter attribute <paramattrs>` for the return type, a function
547name, a (possibly empty) argument list (each with optional :ref:`parameter
548attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
549an optional section, an optional alignment, an optional :ref:`garbage
550collector name <gc>`, an opening curly brace, a list of basic blocks,
551and a closing curly brace.
552
553LLVM function declarations consist of the "``declare``" keyword, an
554optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
555style <visibility>`, an optional :ref:`calling convention <callingconv>`,
556an optional ``unnamed_addr`` attribute, a return type, an optional
557:ref:`parameter attribute <paramattrs>` for the return type, a function
558name, a possibly empty list of arguments, an optional alignment, and an
559optional :ref:`garbage collector name <gc>`.
560
561A function definition contains a list of basic blocks, forming the CFG
562(Control Flow Graph) for the function. Each basic block may optionally
563start with a label (giving the basic block a symbol table entry),
564contains a list of instructions, and ends with a
565:ref:`terminator <terminators>` instruction (such as a branch or function
566return).
567
568The first basic block in a function is special in two ways: it is
569immediately executed on entrance to the function, and it is not allowed
570to have predecessor basic blocks (i.e. there can not be any branches to
571the entry block of a function). Because the block can have no
572predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
573
574LLVM allows an explicit section to be specified for functions. If the
575target supports it, it will emit functions to the section specified.
576
577An explicit alignment may be specified for a function. If not present,
578or if the alignment is set to zero, the alignment of the function is set
579by the target to whatever it feels convenient. If an explicit alignment
580is specified, the function is forced to have at least that much
581alignment. All alignments must be a power of 2.
582
583If the ``unnamed_addr`` attribute is given, the address is know to not
584be significant and two identical functions can be merged.
585
586Syntax::
587
588 define [linkage] [visibility]
589 [cconv] [ret attrs]
590 <ResultType> @<FunctionName> ([argument list])
591 [fn Attrs] [section "name"] [align N]
592 [gc] { ... }
593
594Aliases
595-------
596
597Aliases act as "second name" for the aliasee value (which can be either
598function, global variable, another alias or bitcast of global value).
599Aliases may have an optional :ref:`linkage type <linkage>`, and an optional
600:ref:`visibility style <visibility>`.
601
602Syntax::
603
604 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
605
606.. _namedmetadatastructure:
607
608Named Metadata
609--------------
610
611Named metadata is a collection of metadata. :ref:`Metadata
612nodes <metadata>` (but not metadata strings) are the only valid
613operands for a named metadata.
614
615Syntax::
616
617 ; Some unnamed metadata nodes, which are referenced by the named metadata.
618 !0 = metadata !{metadata !"zero"}
619 !1 = metadata !{metadata !"one"}
620 !2 = metadata !{metadata !"two"}
621 ; A named metadata.
622 !name = !{!0, !1, !2}
623
624.. _paramattrs:
625
626Parameter Attributes
627--------------------
628
629The return type and each parameter of a function type may have a set of
630*parameter attributes* associated with them. Parameter attributes are
631used to communicate additional information about the result or
632parameters of a function. Parameter attributes are considered to be part
633of the function, not of the function type, so functions with different
634parameter attributes can have the same function type.
635
636Parameter attributes are simple keywords that follow the type specified.
637If multiple parameter attributes are needed, they are space separated.
638For example:
639
640.. code-block:: llvm
641
642 declare i32 @printf(i8* noalias nocapture, ...)
643 declare i32 @atoi(i8 zeroext)
644 declare signext i8 @returns_signed_char()
645
646Note that any attributes for the function result (``nounwind``,
647``readonly``) come immediately after the argument list.
648
649Currently, only the following parameter attributes are defined:
650
651``zeroext``
652 This indicates to the code generator that the parameter or return
653 value should be zero-extended to the extent required by the target's
654 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
655 the caller (for a parameter) or the callee (for a return value).
656``signext``
657 This indicates to the code generator that the parameter or return
658 value should be sign-extended to the extent required by the target's
659 ABI (which is usually 32-bits) by the caller (for a parameter) or
660 the callee (for a return value).
661``inreg``
662 This indicates that this parameter or return value should be treated
663 in a special target-dependent fashion during while emitting code for
664 a function call or return (usually, by putting it in a register as
665 opposed to memory, though some targets use it to distinguish between
666 two different kinds of registers). Use of this attribute is
667 target-specific.
668``byval``
669 This indicates that the pointer parameter should really be passed by
670 value to the function. The attribute implies that a hidden copy of
671 the pointee is made between the caller and the callee, so the callee
672 is unable to modify the value in the caller. This attribute is only
673 valid on LLVM pointer arguments. It is generally used to pass
674 structs and arrays by value, but is also valid on pointers to
675 scalars. The copy is considered to belong to the caller not the
676 callee (for example, ``readonly`` functions should not write to
677 ``byval`` parameters). This is not a valid attribute for return
678 values.
679
680 The byval attribute also supports specifying an alignment with the
681 align attribute. It indicates the alignment of the stack slot to
682 form and the known alignment of the pointer specified to the call
683 site. If the alignment is not specified, then the code generator
684 makes a target-specific assumption.
685
686``sret``
687 This indicates that the pointer parameter specifies the address of a
688 structure that is the return value of the function in the source
689 program. This pointer must be guaranteed by the caller to be valid:
Eli Bendersky98202c02013-01-23 22:05:19 +0000690 loads and stores to the structure may be assumed by the callee
Sean Silvaf722b002012-12-07 10:36:55 +0000691 not to trap and to be properly aligned. This may only be applied to
692 the first parameter. This is not a valid attribute for return
693 values.
694``noalias``
695 This indicates that pointer values `*based* <pointeraliasing>` on
696 the argument or return value do not alias pointer values which are
697 not *based* on it, ignoring certain "irrelevant" dependencies. For a
698 call to the parent function, dependencies between memory references
699 from before or after the call and from those during the call are
700 "irrelevant" to the ``noalias`` keyword for the arguments and return
701 value used in that call. The caller shares the responsibility with
702 the callee for ensuring that these requirements are met. For further
703 details, please see the discussion of the NoAlias response in `alias
704 analysis <AliasAnalysis.html#MustMayNo>`_.
705
706 Note that this definition of ``noalias`` is intentionally similar
707 to the definition of ``restrict`` in C99 for function arguments,
708 though it is slightly weaker.
709
710 For function return values, C99's ``restrict`` is not meaningful,
711 while LLVM's ``noalias`` is.
712``nocapture``
713 This indicates that the callee does not make any copies of the
714 pointer that outlive the callee itself. This is not a valid
715 attribute for return values.
716
717.. _nest:
718
719``nest``
720 This indicates that the pointer parameter can be excised using the
721 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
722 attribute for return values.
723
724.. _gc:
725
726Garbage Collector Names
727-----------------------
728
729Each function may specify a garbage collector name, which is simply a
730string:
731
732.. code-block:: llvm
733
734 define void @f() gc "name" { ... }
735
736The compiler declares the supported values of *name*. Specifying a
737collector which will cause the compiler to alter its output in order to
738support the named garbage collection algorithm.
739
Bill Wendling95ce4c22013-02-06 06:52:58 +0000740.. _attrgrp:
741
742Attribute Groups
743----------------
744
745Attribute groups are groups of attributes that are referenced by objects within
746the IR. They are important for keeping ``.ll`` files readable, because a lot of
747functions will use the same set of attributes. In the degenerative case of a
748``.ll`` file that corresponds to a single ``.c`` file, the single attribute
749group will capture the important command line flags used to build that file.
750
751An attribute group is a module-level object. To use an attribute group, an
752object references the attribute group's ID (e.g. ``#37``). An object may refer
753to more than one attribute group. In that situation, the attributes from the
754different groups are merged.
755
756Here is an example of attribute groups for a function that should always be
757inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
758
759.. code-block:: llvm
760
761 ; Target-independent attributes:
762 #0 = attributes { alwaysinline alignstack=4 }
763
764 ; Target-dependent attributes:
765 #1 = attributes { "no-sse" }
766
767 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
768 define void @f() #0 #1 { ... }
769
Sean Silvaf722b002012-12-07 10:36:55 +0000770.. _fnattrs:
771
772Function Attributes
773-------------------
774
775Function attributes are set to communicate additional information about
776a function. Function attributes are considered to be part of the
777function, not of the function type, so functions with different function
778attributes can have the same function type.
779
780Function attributes are simple keywords that follow the type specified.
781If multiple attributes are needed, they are space separated. For
782example:
783
784.. code-block:: llvm
785
786 define void @f() noinline { ... }
787 define void @f() alwaysinline { ... }
788 define void @f() alwaysinline optsize { ... }
789 define void @f() optsize { ... }
790
791``address_safety``
792 This attribute indicates that the address safety analysis is enabled
793 for this function.
794``alignstack(<n>)``
795 This attribute indicates that, when emitting the prologue and
796 epilogue, the backend should forcibly align the stack pointer.
797 Specify the desired alignment, which must be a power of two, in
798 parentheses.
799``alwaysinline``
800 This attribute indicates that the inliner should attempt to inline
801 this function into callers whenever possible, ignoring any active
802 inlining size threshold for this caller.
803``nonlazybind``
804 This attribute suppresses lazy symbol binding for the function. This
805 may make calls to the function faster, at the cost of extra program
806 startup time if the function is not called during program startup.
807``inlinehint``
808 This attribute indicates that the source code contained a hint that
809 inlining this function is desirable (such as the "inline" keyword in
810 C/C++). It is just a hint; it imposes no requirements on the
811 inliner.
812``naked``
813 This attribute disables prologue / epilogue emission for the
814 function. This can have very system-specific consequences.
Bill Wendlingbe5d7472013-02-06 06:22:58 +0000815``noduplicate``
816 This attribute indicates that calls to the function cannot be
817 duplicated. A call to a ``noduplicate`` function may be moved
818 within its parent function, but may not be duplicated within
819 its parent function.
820
821 A function containing a ``noduplicate`` call may still
822 be an inlining candidate, provided that the call is not
823 duplicated by inlining. That implies that the function has
824 internal linkage and only has one call site, so the original
825 call is dead after inlining.
Sean Silvaf722b002012-12-07 10:36:55 +0000826``noimplicitfloat``
827 This attributes disables implicit floating point instructions.
828``noinline``
829 This attribute indicates that the inliner should never inline this
830 function in any situation. This attribute may not be used together
831 with the ``alwaysinline`` attribute.
832``noredzone``
833 This attribute indicates that the code generator should not use a
834 red zone, even if the target-specific ABI normally permits it.
835``noreturn``
836 This function attribute indicates that the function never returns
837 normally. This produces undefined behavior at runtime if the
838 function ever does dynamically return.
839``nounwind``
840 This function attribute indicates that the function never returns
841 with an unwind or exceptional control flow. If the function does
842 unwind, its runtime behavior is undefined.
843``optsize``
844 This attribute suggests that optimization passes and code generator
845 passes make choices that keep the code size of this function low,
846 and otherwise do optimizations specifically to reduce code size.
847``readnone``
848 This attribute indicates that the function computes its result (or
849 decides to unwind an exception) based strictly on its arguments,
850 without dereferencing any pointer arguments or otherwise accessing
851 any mutable state (e.g. memory, control registers, etc) visible to
852 caller functions. It does not write through any pointer arguments
853 (including ``byval`` arguments) and never changes any state visible
854 to callers. This means that it cannot unwind exceptions by calling
855 the ``C++`` exception throwing methods.
856``readonly``
857 This attribute indicates that the function does not write through
858 any pointer arguments (including ``byval`` arguments) or otherwise
859 modify any state (e.g. memory, control registers, etc) visible to
860 caller functions. It may dereference pointer arguments and read
861 state that may be set in the caller. A readonly function always
862 returns the same value (or unwinds an exception identically) when
863 called with the same set of arguments and global state. It cannot
864 unwind an exception by calling the ``C++`` exception throwing
865 methods.
866``returns_twice``
867 This attribute indicates that this function can return twice. The C
868 ``setjmp`` is an example of such a function. The compiler disables
869 some optimizations (like tail calls) in the caller of these
870 functions.
871``ssp``
872 This attribute indicates that the function should emit a stack
Dmitri Gribenkoae4a9ae2013-01-19 20:34:20 +0000873 smashing protector. It is in the form of a "canary" --- a random value
Sean Silvaf722b002012-12-07 10:36:55 +0000874 placed on the stack before the local variables that's checked upon
875 return from the function to see if it has been overwritten. A
876 heuristic is used to determine if a function needs stack protectors
Bill Wendlinge4957fb2013-01-23 06:43:53 +0000877 or not. The heuristic used will enable protectors for functions with:
Dmitri Gribenkod8acb282013-01-29 23:14:41 +0000878
Bill Wendlinge4957fb2013-01-23 06:43:53 +0000879 - Character arrays larger than ``ssp-buffer-size`` (default 8).
880 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
881 - Calls to alloca() with variable sizes or constant sizes greater than
882 ``ssp-buffer-size``.
Sean Silvaf722b002012-12-07 10:36:55 +0000883
884 If a function that has an ``ssp`` attribute is inlined into a
885 function that doesn't have an ``ssp`` attribute, then the resulting
886 function will have an ``ssp`` attribute.
887``sspreq``
888 This attribute indicates that the function should *always* emit a
889 stack smashing protector. This overrides the ``ssp`` function
890 attribute.
891
892 If a function that has an ``sspreq`` attribute is inlined into a
893 function that doesn't have an ``sspreq`` attribute or which has an
Bill Wendling114baee2013-01-23 06:41:41 +0000894 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
895 an ``sspreq`` attribute.
896``sspstrong``
897 This attribute indicates that the function should emit a stack smashing
Bill Wendlinge4957fb2013-01-23 06:43:53 +0000898 protector. This attribute causes a strong heuristic to be used when
899 determining if a function needs stack protectors. The strong heuristic
900 will enable protectors for functions with:
Dmitri Gribenkod8acb282013-01-29 23:14:41 +0000901
Bill Wendlinge4957fb2013-01-23 06:43:53 +0000902 - Arrays of any size and type
903 - Aggregates containing an array of any size and type.
904 - Calls to alloca().
905 - Local variables that have had their address taken.
906
907 This overrides the ``ssp`` function attribute.
Bill Wendling114baee2013-01-23 06:41:41 +0000908
909 If a function that has an ``sspstrong`` attribute is inlined into a
910 function that doesn't have an ``sspstrong`` attribute, then the
911 resulting function will have an ``sspstrong`` attribute.
Kostya Serebryanyab39afa2013-02-11 08:13:54 +0000912``thread_safety``
913 This attribute indicates that the thread safety analysis is enabled
914 for this function.
915``uninitialized_checks``
916 This attribute indicates that the checks for uses of uninitialized
917 memory are enabled.
Sean Silvaf722b002012-12-07 10:36:55 +0000918``uwtable``
919 This attribute indicates that the ABI being targeted requires that
920 an unwind table entry be produce for this function even if we can
921 show that no exceptions passes by it. This is normally the case for
922 the ELF x86-64 abi, but it can be disabled for some compilation
923 units.
Sean Silvaf722b002012-12-07 10:36:55 +0000924
925.. _moduleasm:
926
927Module-Level Inline Assembly
928----------------------------
929
930Modules may contain "module-level inline asm" blocks, which corresponds
931to the GCC "file scope inline asm" blocks. These blocks are internally
932concatenated by LLVM and treated as a single unit, but may be separated
933in the ``.ll`` file if desired. The syntax is very simple:
934
935.. code-block:: llvm
936
937 module asm "inline asm code goes here"
938 module asm "more can go here"
939
940The strings can contain any character by escaping non-printable
941characters. The escape sequence used is simply "\\xx" where "xx" is the
942two digit hex code for the number.
943
944The inline asm code is simply printed to the machine code .s file when
945assembly code is generated.
946
947Data Layout
948-----------
949
950A module may specify a target specific data layout string that specifies
951how data is to be laid out in memory. The syntax for the data layout is
952simply:
953
954.. code-block:: llvm
955
956 target datalayout = "layout specification"
957
958The *layout specification* consists of a list of specifications
959separated by the minus sign character ('-'). Each specification starts
960with a letter and may include other information after the letter to
961define some aspect of the data layout. The specifications accepted are
962as follows:
963
964``E``
965 Specifies that the target lays out data in big-endian form. That is,
966 the bits with the most significance have the lowest address
967 location.
968``e``
969 Specifies that the target lays out data in little-endian form. That
970 is, the bits with the least significance have the lowest address
971 location.
972``S<size>``
973 Specifies the natural alignment of the stack in bits. Alignment
974 promotion of stack variables is limited to the natural stack
975 alignment to avoid dynamic stack realignment. The stack alignment
976 must be a multiple of 8-bits. If omitted, the natural stack
977 alignment defaults to "unspecified", which does not prevent any
978 alignment promotions.
979``p[n]:<size>:<abi>:<pref>``
980 This specifies the *size* of a pointer and its ``<abi>`` and
981 ``<pref>``\erred alignments for address space ``n``. All sizes are in
982 bits. Specifying the ``<pref>`` alignment is optional. If omitted, the
983 preceding ``:`` should be omitted too. The address space, ``n`` is
984 optional, and if not specified, denotes the default address space 0.
985 The value of ``n`` must be in the range [1,2^23).
986``i<size>:<abi>:<pref>``
987 This specifies the alignment for an integer type of a given bit
988 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
989``v<size>:<abi>:<pref>``
990 This specifies the alignment for a vector type of a given bit
991 ``<size>``.
992``f<size>:<abi>:<pref>``
993 This specifies the alignment for a floating point type of a given bit
994 ``<size>``. Only values of ``<size>`` that are supported by the target
995 will work. 32 (float) and 64 (double) are supported on all targets; 80
996 or 128 (different flavors of long double) are also supported on some
997 targets.
998``a<size>:<abi>:<pref>``
999 This specifies the alignment for an aggregate type of a given bit
1000 ``<size>``.
1001``s<size>:<abi>:<pref>``
1002 This specifies the alignment for a stack object of a given bit
1003 ``<size>``.
1004``n<size1>:<size2>:<size3>...``
1005 This specifies a set of native integer widths for the target CPU in
1006 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1007 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1008 this set are considered to support most general arithmetic operations
1009 efficiently.
1010
1011When constructing the data layout for a given target, LLVM starts with a
1012default set of specifications which are then (possibly) overridden by
1013the specifications in the ``datalayout`` keyword. The default
1014specifications are given in this list:
1015
1016- ``E`` - big endian
1017- ``p:64:64:64`` - 64-bit pointers with 64-bit alignment
Patrik Hagglund3b5f0b02013-01-30 09:02:06 +00001018- ``S0`` - natural stack alignment is unspecified
Sean Silvaf722b002012-12-07 10:36:55 +00001019- ``i1:8:8`` - i1 is 8-bit (byte) aligned
1020- ``i8:8:8`` - i8 is 8-bit (byte) aligned
1021- ``i16:16:16`` - i16 is 16-bit aligned
1022- ``i32:32:32`` - i32 is 32-bit aligned
1023- ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1024 alignment of 64-bits
Patrik Hagglund3b5f0b02013-01-30 09:02:06 +00001025- ``f16:16:16`` - half is 16-bit aligned
Sean Silvaf722b002012-12-07 10:36:55 +00001026- ``f32:32:32`` - float is 32-bit aligned
1027- ``f64:64:64`` - double is 64-bit aligned
Patrik Hagglund3b5f0b02013-01-30 09:02:06 +00001028- ``f128:128:128`` - quad is 128-bit aligned
Sean Silvaf722b002012-12-07 10:36:55 +00001029- ``v64:64:64`` - 64-bit vector is 64-bit aligned
1030- ``v128:128:128`` - 128-bit vector is 128-bit aligned
Patrik Hagglund3b5f0b02013-01-30 09:02:06 +00001031- ``a0:0:64`` - aggregates are 64-bit aligned
Sean Silvaf722b002012-12-07 10:36:55 +00001032
1033When LLVM is determining the alignment for a given type, it uses the
1034following rules:
1035
1036#. If the type sought is an exact match for one of the specifications,
1037 that specification is used.
1038#. If no match is found, and the type sought is an integer type, then
1039 the smallest integer type that is larger than the bitwidth of the
1040 sought type is used. If none of the specifications are larger than
1041 the bitwidth then the largest integer type is used. For example,
1042 given the default specifications above, the i7 type will use the
1043 alignment of i8 (next largest) while both i65 and i256 will use the
1044 alignment of i64 (largest specified).
1045#. If no match is found, and the type sought is a vector type, then the
1046 largest vector type that is smaller than the sought vector type will
1047 be used as a fall back. This happens because <128 x double> can be
1048 implemented in terms of 64 <2 x double>, for example.
1049
1050The function of the data layout string may not be what you expect.
1051Notably, this is not a specification from the frontend of what alignment
1052the code generator should use.
1053
1054Instead, if specified, the target data layout is required to match what
1055the ultimate *code generator* expects. This string is used by the
1056mid-level optimizers to improve code, and this only works if it matches
1057what the ultimate code generator uses. If you would like to generate IR
1058that does not embed this target-specific detail into the IR, then you
1059don't have to specify the string. This will disable some optimizations
1060that require precise layout information, but this also prevents those
1061optimizations from introducing target specificity into the IR.
1062
1063.. _pointeraliasing:
1064
1065Pointer Aliasing Rules
1066----------------------
1067
1068Any memory access must be done through a pointer value associated with
1069an address range of the memory access, otherwise the behavior is
1070undefined. Pointer values are associated with address ranges according
1071to the following rules:
1072
1073- A pointer value is associated with the addresses associated with any
1074 value it is *based* on.
1075- An address of a global variable is associated with the address range
1076 of the variable's storage.
1077- The result value of an allocation instruction is associated with the
1078 address range of the allocated storage.
1079- A null pointer in the default address-space is associated with no
1080 address.
1081- An integer constant other than zero or a pointer value returned from
1082 a function not defined within LLVM may be associated with address
1083 ranges allocated through mechanisms other than those provided by
1084 LLVM. Such ranges shall not overlap with any ranges of addresses
1085 allocated by mechanisms provided by LLVM.
1086
1087A pointer value is *based* on another pointer value according to the
1088following rules:
1089
1090- A pointer value formed from a ``getelementptr`` operation is *based*
1091 on the first operand of the ``getelementptr``.
1092- The result value of a ``bitcast`` is *based* on the operand of the
1093 ``bitcast``.
1094- A pointer value formed by an ``inttoptr`` is *based* on all pointer
1095 values that contribute (directly or indirectly) to the computation of
1096 the pointer's value.
1097- The "*based* on" relationship is transitive.
1098
1099Note that this definition of *"based"* is intentionally similar to the
1100definition of *"based"* in C99, though it is slightly weaker.
1101
1102LLVM IR does not associate types with memory. The result type of a
1103``load`` merely indicates the size and alignment of the memory from
1104which to load, as well as the interpretation of the value. The first
1105operand type of a ``store`` similarly only indicates the size and
1106alignment of the store.
1107
1108Consequently, type-based alias analysis, aka TBAA, aka
1109``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1110:ref:`Metadata <metadata>` may be used to encode additional information
1111which specialized optimization passes may use to implement type-based
1112alias analysis.
1113
1114.. _volatile:
1115
1116Volatile Memory Accesses
1117------------------------
1118
1119Certain memory accesses, such as :ref:`load <i_load>`'s,
1120:ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1121marked ``volatile``. The optimizers must not change the number of
1122volatile operations or change their order of execution relative to other
1123volatile operations. The optimizers *may* change the order of volatile
1124operations relative to non-volatile operations. This is not Java's
1125"volatile" and has no cross-thread synchronization behavior.
1126
Andrew Trick9a6dd022013-01-30 21:19:35 +00001127IR-level volatile loads and stores cannot safely be optimized into
1128llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1129flagged volatile. Likewise, the backend should never split or merge
1130target-legal volatile load/store instructions.
1131
Andrew Trick946317d2013-01-31 00:49:39 +00001132.. admonition:: Rationale
1133
1134 Platforms may rely on volatile loads and stores of natively supported
1135 data width to be executed as single instruction. For example, in C
1136 this holds for an l-value of volatile primitive type with native
1137 hardware support, but not necessarily for aggregate types. The
1138 frontend upholds these expectations, which are intentionally
1139 unspecified in the IR. The rules above ensure that IR transformation
1140 do not violate the frontend's contract with the language.
1141
Sean Silvaf722b002012-12-07 10:36:55 +00001142.. _memmodel:
1143
1144Memory Model for Concurrent Operations
1145--------------------------------------
1146
1147The LLVM IR does not define any way to start parallel threads of
1148execution or to register signal handlers. Nonetheless, there are
1149platform-specific ways to create them, and we define LLVM IR's behavior
1150in their presence. This model is inspired by the C++0x memory model.
1151
1152For a more informal introduction to this model, see the :doc:`Atomics`.
1153
1154We define a *happens-before* partial order as the least partial order
1155that
1156
1157- Is a superset of single-thread program order, and
1158- When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1159 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1160 techniques, like pthread locks, thread creation, thread joining,
1161 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1162 Constraints <ordering>`).
1163
1164Note that program order does not introduce *happens-before* edges
1165between a thread and signals executing inside that thread.
1166
1167Every (defined) read operation (load instructions, memcpy, atomic
1168loads/read-modify-writes, etc.) R reads a series of bytes written by
1169(defined) write operations (store instructions, atomic
1170stores/read-modify-writes, memcpy, etc.). For the purposes of this
1171section, initialized globals are considered to have a write of the
1172initializer which is atomic and happens before any other read or write
1173of the memory in question. For each byte of a read R, R\ :sub:`byte`
1174may see any write to the same byte, except:
1175
1176- If write\ :sub:`1` happens before write\ :sub:`2`, and
1177 write\ :sub:`2` happens before R\ :sub:`byte`, then
1178 R\ :sub:`byte` does not see write\ :sub:`1`.
1179- If R\ :sub:`byte` happens before write\ :sub:`3`, then
1180 R\ :sub:`byte` does not see write\ :sub:`3`.
1181
1182Given that definition, R\ :sub:`byte` is defined as follows:
1183
1184- If R is volatile, the result is target-dependent. (Volatile is
1185 supposed to give guarantees which can support ``sig_atomic_t`` in
1186 C/C++, and may be used for accesses to addresses which do not behave
1187 like normal memory. It does not generally provide cross-thread
1188 synchronization.)
1189- Otherwise, if there is no write to the same byte that happens before
1190 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1191- Otherwise, if R\ :sub:`byte` may see exactly one write,
1192 R\ :sub:`byte` returns the value written by that write.
1193- Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1194 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1195 Memory Ordering Constraints <ordering>` section for additional
1196 constraints on how the choice is made.
1197- Otherwise R\ :sub:`byte` returns ``undef``.
1198
1199R returns the value composed of the series of bytes it read. This
1200implies that some bytes within the value may be ``undef`` **without**
1201the entire value being ``undef``. Note that this only defines the
1202semantics of the operation; it doesn't mean that targets will emit more
1203than one instruction to read the series of bytes.
1204
1205Note that in cases where none of the atomic intrinsics are used, this
1206model places only one restriction on IR transformations on top of what
1207is required for single-threaded execution: introducing a store to a byte
1208which might not otherwise be stored is not allowed in general.
1209(Specifically, in the case where another thread might write to and read
1210from an address, introducing a store can change a load that may see
1211exactly one write into a load that may see multiple writes.)
1212
1213.. _ordering:
1214
1215Atomic Memory Ordering Constraints
1216----------------------------------
1217
1218Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1219:ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1220:ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1221an ordering parameter that determines which other atomic instructions on
1222the same address they *synchronize with*. These semantics are borrowed
1223from Java and C++0x, but are somewhat more colloquial. If these
1224descriptions aren't precise enough, check those specs (see spec
1225references in the :doc:`atomics guide <Atomics>`).
1226:ref:`fence <i_fence>` instructions treat these orderings somewhat
1227differently since they don't take an address. See that instruction's
1228documentation for details.
1229
1230For a simpler introduction to the ordering constraints, see the
1231:doc:`Atomics`.
1232
1233``unordered``
1234 The set of values that can be read is governed by the happens-before
1235 partial order. A value cannot be read unless some operation wrote
1236 it. This is intended to provide a guarantee strong enough to model
1237 Java's non-volatile shared variables. This ordering cannot be
1238 specified for read-modify-write operations; it is not strong enough
1239 to make them atomic in any interesting way.
1240``monotonic``
1241 In addition to the guarantees of ``unordered``, there is a single
1242 total order for modifications by ``monotonic`` operations on each
1243 address. All modification orders must be compatible with the
1244 happens-before order. There is no guarantee that the modification
1245 orders can be combined to a global total order for the whole program
1246 (and this often will not be possible). The read in an atomic
1247 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1248 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1249 order immediately before the value it writes. If one atomic read
1250 happens before another atomic read of the same address, the later
1251 read must see the same value or a later value in the address's
1252 modification order. This disallows reordering of ``monotonic`` (or
1253 stronger) operations on the same address. If an address is written
1254 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1255 read that address repeatedly, the other threads must eventually see
1256 the write. This corresponds to the C++0x/C1x
1257 ``memory_order_relaxed``.
1258``acquire``
1259 In addition to the guarantees of ``monotonic``, a
1260 *synchronizes-with* edge may be formed with a ``release`` operation.
1261 This is intended to model C++'s ``memory_order_acquire``.
1262``release``
1263 In addition to the guarantees of ``monotonic``, if this operation
1264 writes a value which is subsequently read by an ``acquire``
1265 operation, it *synchronizes-with* that operation. (This isn't a
1266 complete description; see the C++0x definition of a release
1267 sequence.) This corresponds to the C++0x/C1x
1268 ``memory_order_release``.
1269``acq_rel`` (acquire+release)
1270 Acts as both an ``acquire`` and ``release`` operation on its
1271 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1272``seq_cst`` (sequentially consistent)
1273 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1274 operation which only reads, ``release`` for an operation which only
1275 writes), there is a global total order on all
1276 sequentially-consistent operations on all addresses, which is
1277 consistent with the *happens-before* partial order and with the
1278 modification orders of all the affected addresses. Each
1279 sequentially-consistent read sees the last preceding write to the
1280 same address in this global order. This corresponds to the C++0x/C1x
1281 ``memory_order_seq_cst`` and Java volatile.
1282
1283.. _singlethread:
1284
1285If an atomic operation is marked ``singlethread``, it only *synchronizes
1286with* or participates in modification and seq\_cst total orderings with
1287other operations running in the same thread (for example, in signal
1288handlers).
1289
1290.. _fastmath:
1291
1292Fast-Math Flags
1293---------------
1294
1295LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1296:ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1297:ref:`frem <i_frem>`) have the following flags that can set to enable
1298otherwise unsafe floating point operations
1299
1300``nnan``
1301 No NaNs - Allow optimizations to assume the arguments and result are not
1302 NaN. Such optimizations are required to retain defined behavior over
1303 NaNs, but the value of the result is undefined.
1304
1305``ninf``
1306 No Infs - Allow optimizations to assume the arguments and result are not
1307 +/-Inf. Such optimizations are required to retain defined behavior over
1308 +/-Inf, but the value of the result is undefined.
1309
1310``nsz``
1311 No Signed Zeros - Allow optimizations to treat the sign of a zero
1312 argument or result as insignificant.
1313
1314``arcp``
1315 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1316 argument rather than perform division.
1317
1318``fast``
1319 Fast - Allow algebraically equivalent transformations that may
1320 dramatically change results in floating point (e.g. reassociate). This
1321 flag implies all the others.
1322
1323.. _typesystem:
1324
1325Type System
1326===========
1327
1328The LLVM type system is one of the most important features of the
1329intermediate representation. Being typed enables a number of
1330optimizations to be performed on the intermediate representation
1331directly, without having to do extra analyses on the side before the
1332transformation. A strong type system makes it easier to read the
1333generated code and enables novel analyses and transformations that are
1334not feasible to perform on normal three address code representations.
1335
1336Type Classifications
1337--------------------
1338
1339The types fall into a few useful classifications:
1340
1341
1342.. list-table::
1343 :header-rows: 1
1344
1345 * - Classification
1346 - Types
1347
1348 * - :ref:`integer <t_integer>`
1349 - ``i1``, ``i2``, ``i3``, ... ``i8``, ... ``i16``, ... ``i32``, ...
1350 ``i64``, ...
1351
1352 * - :ref:`floating point <t_floating>`
1353 - ``half``, ``float``, ``double``, ``x86_fp80``, ``fp128``,
1354 ``ppc_fp128``
1355
1356
1357 * - first class
1358
1359 .. _t_firstclass:
1360
1361 - :ref:`integer <t_integer>`, :ref:`floating point <t_floating>`,
1362 :ref:`pointer <t_pointer>`, :ref:`vector <t_vector>`,
1363 :ref:`structure <t_struct>`, :ref:`array <t_array>`,
1364 :ref:`label <t_label>`, :ref:`metadata <t_metadata>`.
1365
1366 * - :ref:`primitive <t_primitive>`
1367 - :ref:`label <t_label>`,
1368 :ref:`void <t_void>`,
1369 :ref:`integer <t_integer>`,
1370 :ref:`floating point <t_floating>`,
1371 :ref:`x86mmx <t_x86mmx>`,
1372 :ref:`metadata <t_metadata>`.
1373
1374 * - :ref:`derived <t_derived>`
1375 - :ref:`array <t_array>`,
1376 :ref:`function <t_function>`,
1377 :ref:`pointer <t_pointer>`,
1378 :ref:`structure <t_struct>`,
1379 :ref:`vector <t_vector>`,
1380 :ref:`opaque <t_opaque>`.
1381
1382The :ref:`first class <t_firstclass>` types are perhaps the most important.
1383Values of these types are the only ones which can be produced by
1384instructions.
1385
1386.. _t_primitive:
1387
1388Primitive Types
1389---------------
1390
1391The primitive types are the fundamental building blocks of the LLVM
1392system.
1393
1394.. _t_integer:
1395
1396Integer Type
1397^^^^^^^^^^^^
1398
1399Overview:
1400"""""""""
1401
1402The integer type is a very simple type that simply specifies an
1403arbitrary bit width for the integer type desired. Any bit width from 1
1404bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1405
1406Syntax:
1407"""""""
1408
1409::
1410
1411 iN
1412
1413The number of bits the integer will occupy is specified by the ``N``
1414value.
1415
1416Examples:
1417"""""""""
1418
1419+----------------+------------------------------------------------+
1420| ``i1`` | a single-bit integer. |
1421+----------------+------------------------------------------------+
1422| ``i32`` | a 32-bit integer. |
1423+----------------+------------------------------------------------+
1424| ``i1942652`` | a really big integer of over 1 million bits. |
1425+----------------+------------------------------------------------+
1426
1427.. _t_floating:
1428
1429Floating Point Types
1430^^^^^^^^^^^^^^^^^^^^
1431
1432.. list-table::
1433 :header-rows: 1
1434
1435 * - Type
1436 - Description
1437
1438 * - ``half``
1439 - 16-bit floating point value
1440
1441 * - ``float``
1442 - 32-bit floating point value
1443
1444 * - ``double``
1445 - 64-bit floating point value
1446
1447 * - ``fp128``
1448 - 128-bit floating point value (112-bit mantissa)
1449
1450 * - ``x86_fp80``
1451 - 80-bit floating point value (X87)
1452
1453 * - ``ppc_fp128``
1454 - 128-bit floating point value (two 64-bits)
1455
1456.. _t_x86mmx:
1457
1458X86mmx Type
1459^^^^^^^^^^^
1460
1461Overview:
1462"""""""""
1463
1464The x86mmx type represents a value held in an MMX register on an x86
1465machine. The operations allowed on it are quite limited: parameters and
1466return values, load and store, and bitcast. User-specified MMX
1467instructions are represented as intrinsic or asm calls with arguments
1468and/or results of this type. There are no arrays, vectors or constants
1469of this type.
1470
1471Syntax:
1472"""""""
1473
1474::
1475
1476 x86mmx
1477
1478.. _t_void:
1479
1480Void Type
1481^^^^^^^^^
1482
1483Overview:
1484"""""""""
1485
1486The void type does not represent any value and has no size.
1487
1488Syntax:
1489"""""""
1490
1491::
1492
1493 void
1494
1495.. _t_label:
1496
1497Label Type
1498^^^^^^^^^^
1499
1500Overview:
1501"""""""""
1502
1503The label type represents code labels.
1504
1505Syntax:
1506"""""""
1507
1508::
1509
1510 label
1511
1512.. _t_metadata:
1513
1514Metadata Type
1515^^^^^^^^^^^^^
1516
1517Overview:
1518"""""""""
1519
1520The metadata type represents embedded metadata. No derived types may be
1521created from metadata except for :ref:`function <t_function>` arguments.
1522
1523Syntax:
1524"""""""
1525
1526::
1527
1528 metadata
1529
1530.. _t_derived:
1531
1532Derived Types
1533-------------
1534
1535The real power in LLVM comes from the derived types in the system. This
1536is what allows a programmer to represent arrays, functions, pointers,
1537and other useful types. Each of these types contain one or more element
1538types which may be a primitive type, or another derived type. For
1539example, it is possible to have a two dimensional array, using an array
1540as the element type of another array.
1541
1542.. _t_aggregate:
1543
1544Aggregate Types
1545^^^^^^^^^^^^^^^
1546
1547Aggregate Types are a subset of derived types that can contain multiple
1548member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
1549aggregate types. :ref:`Vectors <t_vector>` are not considered to be
1550aggregate types.
1551
1552.. _t_array:
1553
1554Array Type
1555^^^^^^^^^^
1556
1557Overview:
1558"""""""""
1559
1560The array type is a very simple derived type that arranges elements
1561sequentially in memory. The array type requires a size (number of
1562elements) and an underlying data type.
1563
1564Syntax:
1565"""""""
1566
1567::
1568
1569 [<# elements> x <elementtype>]
1570
1571The number of elements is a constant integer value; ``elementtype`` may
1572be any type with a size.
1573
1574Examples:
1575"""""""""
1576
1577+------------------+--------------------------------------+
1578| ``[40 x i32]`` | Array of 40 32-bit integer values. |
1579+------------------+--------------------------------------+
1580| ``[41 x i32]`` | Array of 41 32-bit integer values. |
1581+------------------+--------------------------------------+
1582| ``[4 x i8]`` | Array of 4 8-bit integer values. |
1583+------------------+--------------------------------------+
1584
1585Here are some examples of multidimensional arrays:
1586
1587+-----------------------------+----------------------------------------------------------+
1588| ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
1589+-----------------------------+----------------------------------------------------------+
1590| ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
1591+-----------------------------+----------------------------------------------------------+
1592| ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
1593+-----------------------------+----------------------------------------------------------+
1594
1595There is no restriction on indexing beyond the end of the array implied
1596by a static type (though there are restrictions on indexing beyond the
1597bounds of an allocated object in some cases). This means that
1598single-dimension 'variable sized array' addressing can be implemented in
1599LLVM with a zero length array type. An implementation of 'pascal style
1600arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
1601example.
1602
1603.. _t_function:
1604
1605Function Type
1606^^^^^^^^^^^^^
1607
1608Overview:
1609"""""""""
1610
1611The function type can be thought of as a function signature. It consists
1612of a return type and a list of formal parameter types. The return type
1613of a function type is a first class type or a void type.
1614
1615Syntax:
1616"""""""
1617
1618::
1619
1620 <returntype> (<parameter list>)
1621
1622...where '``<parameter list>``' is a comma-separated list of type
1623specifiers. Optionally, the parameter list may include a type ``...``,
1624which indicates that the function takes a variable number of arguments.
1625Variable argument functions can access their arguments with the
1626:ref:`variable argument handling intrinsic <int_varargs>` functions.
1627'``<returntype>``' is any type except :ref:`label <t_label>`.
1628
1629Examples:
1630"""""""""
1631
1632+---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1633| ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1634+---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
Daniel Dunbar3389dbc2013-01-17 18:57:32 +00001635| ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
Sean Silvaf722b002012-12-07 10:36:55 +00001636+---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1637| ``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. |
1638+---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1639| ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1640+---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1641
1642.. _t_struct:
1643
1644Structure Type
1645^^^^^^^^^^^^^^
1646
1647Overview:
1648"""""""""
1649
1650The structure type is used to represent a collection of data members
1651together in memory. The elements of a structure may be any type that has
1652a size.
1653
1654Structures in memory are accessed using '``load``' and '``store``' by
1655getting a pointer to a field with the '``getelementptr``' instruction.
1656Structures in registers are accessed using the '``extractvalue``' and
1657'``insertvalue``' instructions.
1658
1659Structures may optionally be "packed" structures, which indicate that
1660the alignment of the struct is one byte, and that there is no padding
1661between the elements. In non-packed structs, padding between field types
1662is inserted as defined by the DataLayout string in the module, which is
1663required to match what the underlying code generator expects.
1664
1665Structures can either be "literal" or "identified". A literal structure
1666is defined inline with other types (e.g. ``{i32, i32}*``) whereas
1667identified types are always defined at the top level with a name.
1668Literal types are uniqued by their contents and can never be recursive
1669or opaque since there is no way to write one. Identified types can be
1670recursive, can be opaqued, and are never uniqued.
1671
1672Syntax:
1673"""""""
1674
1675::
1676
1677 %T1 = type { <type list> } ; Identified normal struct type
1678 %T2 = type <{ <type list> }> ; Identified packed struct type
1679
1680Examples:
1681"""""""""
1682
1683+------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1684| ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
1685+------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
Daniel Dunbar3389dbc2013-01-17 18:57:32 +00001686| ``{ 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``. |
Sean Silvaf722b002012-12-07 10:36:55 +00001687+------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1688| ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
1689+------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1690
1691.. _t_opaque:
1692
1693Opaque Structure Types
1694^^^^^^^^^^^^^^^^^^^^^^
1695
1696Overview:
1697"""""""""
1698
1699Opaque structure types are used to represent named structure types that
1700do not have a body specified. This corresponds (for example) to the C
1701notion of a forward declared structure.
1702
1703Syntax:
1704"""""""
1705
1706::
1707
1708 %X = type opaque
1709 %52 = type opaque
1710
1711Examples:
1712"""""""""
1713
1714+--------------+-------------------+
1715| ``opaque`` | An opaque type. |
1716+--------------+-------------------+
1717
1718.. _t_pointer:
1719
1720Pointer Type
1721^^^^^^^^^^^^
1722
1723Overview:
1724"""""""""
1725
1726The pointer type is used to specify memory locations. Pointers are
1727commonly used to reference objects in memory.
1728
1729Pointer types may have an optional address space attribute defining the
1730numbered address space where the pointed-to object resides. The default
1731address space is number zero. The semantics of non-zero address spaces
1732are target-specific.
1733
1734Note that LLVM does not permit pointers to void (``void*``) nor does it
1735permit pointers to labels (``label*``). Use ``i8*`` instead.
1736
1737Syntax:
1738"""""""
1739
1740::
1741
1742 <type> *
1743
1744Examples:
1745"""""""""
1746
1747+-------------------------+--------------------------------------------------------------------------------------------------------------+
1748| ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1749+-------------------------+--------------------------------------------------------------------------------------------------------------+
1750| ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1751+-------------------------+--------------------------------------------------------------------------------------------------------------+
1752| ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1753+-------------------------+--------------------------------------------------------------------------------------------------------------+
1754
1755.. _t_vector:
1756
1757Vector Type
1758^^^^^^^^^^^
1759
1760Overview:
1761"""""""""
1762
1763A vector type is a simple derived type that represents a vector of
1764elements. Vector types are used when multiple primitive data are
1765operated in parallel using a single instruction (SIMD). A vector type
1766requires a size (number of elements) and an underlying primitive data
1767type. Vector types are considered :ref:`first class <t_firstclass>`.
1768
1769Syntax:
1770"""""""
1771
1772::
1773
1774 < <# elements> x <elementtype> >
1775
1776The number of elements is a constant integer value larger than 0;
1777elementtype may be any integer or floating point type, or a pointer to
1778these types. Vectors of size zero are not allowed.
1779
1780Examples:
1781"""""""""
1782
1783+-------------------+--------------------------------------------------+
1784| ``<4 x i32>`` | Vector of 4 32-bit integer values. |
1785+-------------------+--------------------------------------------------+
1786| ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
1787+-------------------+--------------------------------------------------+
1788| ``<2 x i64>`` | Vector of 2 64-bit integer values. |
1789+-------------------+--------------------------------------------------+
1790| ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
1791+-------------------+--------------------------------------------------+
1792
1793Constants
1794=========
1795
1796LLVM has several different basic types of constants. This section
1797describes them all and their syntax.
1798
1799Simple Constants
1800----------------
1801
1802**Boolean constants**
1803 The two strings '``true``' and '``false``' are both valid constants
1804 of the ``i1`` type.
1805**Integer constants**
1806 Standard integers (such as '4') are constants of the
1807 :ref:`integer <t_integer>` type. Negative numbers may be used with
1808 integer types.
1809**Floating point constants**
1810 Floating point constants use standard decimal notation (e.g.
1811 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
1812 hexadecimal notation (see below). The assembler requires the exact
1813 decimal value of a floating-point constant. For example, the
1814 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
1815 decimal in binary. Floating point constants must have a :ref:`floating
1816 point <t_floating>` type.
1817**Null pointer constants**
1818 The identifier '``null``' is recognized as a null pointer constant
1819 and must be of :ref:`pointer type <t_pointer>`.
1820
1821The one non-intuitive notation for constants is the hexadecimal form of
1822floating point constants. For example, the form
1823'``double 0x432ff973cafa8000``' is equivalent to (but harder to read
1824than) '``double 4.5e+15``'. The only time hexadecimal floating point
1825constants are required (and the only time that they are generated by the
1826disassembler) is when a floating point constant must be emitted but it
1827cannot be represented as a decimal floating point number in a reasonable
1828number of digits. For example, NaN's, infinities, and other special
1829values are represented in their IEEE hexadecimal format so that assembly
1830and disassembly do not cause any bits to change in the constants.
1831
1832When using the hexadecimal form, constants of types half, float, and
1833double are represented using the 16-digit form shown above (which
1834matches the IEEE754 representation for double); half and float values
Dmitri Gribenkoc3c8d2a2013-01-16 23:40:37 +00001835must, however, be exactly representable as IEEE 754 half and single
Sean Silvaf722b002012-12-07 10:36:55 +00001836precision, respectively. Hexadecimal format is always used for long
1837double, and there are three forms of long double. The 80-bit format used
1838by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
1839128-bit format used by PowerPC (two adjacent doubles) is represented by
1840``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
1841represented by ``0xL`` followed by 32 hexadecimal digits; no currently
1842supported target uses this format. Long doubles will only work if they
1843match the long double format on your target. The IEEE 16-bit format
1844(half precision) is represented by ``0xH`` followed by 4 hexadecimal
1845digits. All hexadecimal formats are big-endian (sign bit at the left).
1846
1847There are no constants of type x86mmx.
1848
1849Complex Constants
1850-----------------
1851
1852Complex constants are a (potentially recursive) combination of simple
1853constants and smaller complex constants.
1854
1855**Structure constants**
1856 Structure constants are represented with notation similar to
1857 structure type definitions (a comma separated list of elements,
1858 surrounded by braces (``{}``)). For example:
1859 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
1860 "``@G = external global i32``". Structure constants must have
1861 :ref:`structure type <t_struct>`, and the number and types of elements
1862 must match those specified by the type.
1863**Array constants**
1864 Array constants are represented with notation similar to array type
1865 definitions (a comma separated list of elements, surrounded by
1866 square brackets (``[]``)). For example:
1867 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
1868 :ref:`array type <t_array>`, and the number and types of elements must
1869 match those specified by the type.
1870**Vector constants**
1871 Vector constants are represented with notation similar to vector
1872 type definitions (a comma separated list of elements, surrounded by
1873 less-than/greater-than's (``<>``)). For example:
1874 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
1875 must have :ref:`vector type <t_vector>`, and the number and types of
1876 elements must match those specified by the type.
1877**Zero initialization**
1878 The string '``zeroinitializer``' can be used to zero initialize a
1879 value to zero of *any* type, including scalar and
1880 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
1881 having to print large zero initializers (e.g. for large arrays) and
1882 is always exactly equivalent to using explicit zero initializers.
1883**Metadata node**
1884 A metadata node is a structure-like constant with :ref:`metadata
1885 type <t_metadata>`. For example:
1886 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
1887 constants that are meant to be interpreted as part of the
1888 instruction stream, metadata is a place to attach additional
1889 information such as debug info.
1890
1891Global Variable and Function Addresses
1892--------------------------------------
1893
1894The addresses of :ref:`global variables <globalvars>` and
1895:ref:`functions <functionstructure>` are always implicitly valid
1896(link-time) constants. These constants are explicitly referenced when
1897the :ref:`identifier for the global <identifiers>` is used and always have
1898:ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
1899file:
1900
1901.. code-block:: llvm
1902
1903 @X = global i32 17
1904 @Y = global i32 42
1905 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1906
1907.. _undefvalues:
1908
1909Undefined Values
1910----------------
1911
1912The string '``undef``' can be used anywhere a constant is expected, and
1913indicates that the user of the value may receive an unspecified
1914bit-pattern. Undefined values may be of any type (other than '``label``'
1915or '``void``') and be used anywhere a constant is permitted.
1916
1917Undefined values are useful because they indicate to the compiler that
1918the program is well defined no matter what value is used. This gives the
1919compiler more freedom to optimize. Here are some examples of
1920(potentially surprising) transformations that are valid (in pseudo IR):
1921
1922.. code-block:: llvm
1923
1924 %A = add %X, undef
1925 %B = sub %X, undef
1926 %C = xor %X, undef
1927 Safe:
1928 %A = undef
1929 %B = undef
1930 %C = undef
1931
1932This is safe because all of the output bits are affected by the undef
1933bits. Any output bit can have a zero or one depending on the input bits.
1934
1935.. code-block:: llvm
1936
1937 %A = or %X, undef
1938 %B = and %X, undef
1939 Safe:
1940 %A = -1
1941 %B = 0
1942 Unsafe:
1943 %A = undef
1944 %B = undef
1945
1946These logical operations have bits that are not always affected by the
1947input. For example, if ``%X`` has a zero bit, then the output of the
1948'``and``' operation will always be a zero for that bit, no matter what
1949the corresponding bit from the '``undef``' is. As such, it is unsafe to
1950optimize or assume that the result of the '``and``' is '``undef``'.
1951However, it is safe to assume that all bits of the '``undef``' could be
19520, and optimize the '``and``' to 0. Likewise, it is safe to assume that
1953all the bits of the '``undef``' operand to the '``or``' could be set,
1954allowing the '``or``' to be folded to -1.
1955
1956.. code-block:: llvm
1957
1958 %A = select undef, %X, %Y
1959 %B = select undef, 42, %Y
1960 %C = select %X, %Y, undef
1961 Safe:
1962 %A = %X (or %Y)
1963 %B = 42 (or %Y)
1964 %C = %Y
1965 Unsafe:
1966 %A = undef
1967 %B = undef
1968 %C = undef
1969
1970This set of examples shows that undefined '``select``' (and conditional
1971branch) conditions can go *either way*, but they have to come from one
1972of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
1973both known to have a clear low bit, then ``%A`` would have to have a
1974cleared low bit. However, in the ``%C`` example, the optimizer is
1975allowed to assume that the '``undef``' operand could be the same as
1976``%Y``, allowing the whole '``select``' to be eliminated.
1977
1978.. code-block:: llvm
1979
1980 %A = xor undef, undef
1981
1982 %B = undef
1983 %C = xor %B, %B
1984
1985 %D = undef
1986 %E = icmp lt %D, 4
1987 %F = icmp gte %D, 4
1988
1989 Safe:
1990 %A = undef
1991 %B = undef
1992 %C = undef
1993 %D = undef
1994 %E = undef
1995 %F = undef
1996
1997This example points out that two '``undef``' operands are not
1998necessarily the same. This can be surprising to people (and also matches
1999C semantics) where they assume that "``X^X``" is always zero, even if
2000``X`` is undefined. This isn't true for a number of reasons, but the
2001short answer is that an '``undef``' "variable" can arbitrarily change
2002its value over its "live range". This is true because the variable
2003doesn't actually *have a live range*. Instead, the value is logically
2004read from arbitrary registers that happen to be around when needed, so
2005the value is not necessarily consistent over time. In fact, ``%A`` and
2006``%C`` need to have the same semantics or the core LLVM "replace all
2007uses with" concept would not hold.
2008
2009.. code-block:: llvm
2010
2011 %A = fdiv undef, %X
2012 %B = fdiv %X, undef
2013 Safe:
2014 %A = undef
2015 b: unreachable
2016
2017These examples show the crucial difference between an *undefined value*
2018and *undefined behavior*. An undefined value (like '``undef``') is
2019allowed to have an arbitrary bit-pattern. This means that the ``%A``
2020operation can be constant folded to '``undef``', because the '``undef``'
2021could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2022However, in the second example, we can make a more aggressive
2023assumption: because the ``undef`` is allowed to be an arbitrary value,
2024we are allowed to assume that it could be zero. Since a divide by zero
2025has *undefined behavior*, we are allowed to assume that the operation
2026does not execute at all. This allows us to delete the divide and all
2027code after it. Because the undefined operation "can't happen", the
2028optimizer can assume that it occurs in dead code.
2029
2030.. code-block:: llvm
2031
2032 a: store undef -> %X
2033 b: store %X -> undef
2034 Safe:
2035 a: <deleted>
2036 b: unreachable
2037
2038These examples reiterate the ``fdiv`` example: a store *of* an undefined
2039value can be assumed to not have any effect; we can assume that the
2040value is overwritten with bits that happen to match what was already
2041there. However, a store *to* an undefined location could clobber
2042arbitrary memory, therefore, it has undefined behavior.
2043
2044.. _poisonvalues:
2045
2046Poison Values
2047-------------
2048
2049Poison values are similar to :ref:`undef values <undefvalues>`, however
2050they also represent the fact that an instruction or constant expression
2051which cannot evoke side effects has nevertheless detected a condition
2052which results in undefined behavior.
2053
2054There is currently no way of representing a poison value in the IR; they
2055only exist when produced by operations such as :ref:`add <i_add>` with
2056the ``nsw`` flag.
2057
2058Poison value behavior is defined in terms of value *dependence*:
2059
2060- Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2061- :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2062 their dynamic predecessor basic block.
2063- Function arguments depend on the corresponding actual argument values
2064 in the dynamic callers of their functions.
2065- :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2066 instructions that dynamically transfer control back to them.
2067- :ref:`Invoke <i_invoke>` instructions depend on the
2068 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2069 call instructions that dynamically transfer control back to them.
2070- Non-volatile loads and stores depend on the most recent stores to all
2071 of the referenced memory addresses, following the order in the IR
2072 (including loads and stores implied by intrinsics such as
2073 :ref:`@llvm.memcpy <int_memcpy>`.)
2074- An instruction with externally visible side effects depends on the
2075 most recent preceding instruction with externally visible side
2076 effects, following the order in the IR. (This includes :ref:`volatile
2077 operations <volatile>`.)
2078- An instruction *control-depends* on a :ref:`terminator
2079 instruction <terminators>` if the terminator instruction has
2080 multiple successors and the instruction is always executed when
2081 control transfers to one of the successors, and may not be executed
2082 when control is transferred to another.
2083- Additionally, an instruction also *control-depends* on a terminator
2084 instruction if the set of instructions it otherwise depends on would
2085 be different if the terminator had transferred control to a different
2086 successor.
2087- Dependence is transitive.
2088
2089Poison Values have the same behavior as :ref:`undef values <undefvalues>`,
2090with the additional affect that any instruction which has a *dependence*
2091on a poison value has undefined behavior.
2092
2093Here are some examples:
2094
2095.. code-block:: llvm
2096
2097 entry:
2098 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2099 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2100 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2101 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2102
2103 store i32 %poison, i32* @g ; Poison value stored to memory.
2104 %poison2 = load i32* @g ; Poison value loaded back from memory.
2105
2106 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2107
2108 %narrowaddr = bitcast i32* @g to i16*
2109 %wideaddr = bitcast i32* @g to i64*
2110 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2111 %poison4 = load i64* %wideaddr ; Returns a poison value.
2112
2113 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2114 br i1 %cmp, label %true, label %end ; Branch to either destination.
2115
2116 true:
2117 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2118 ; it has undefined behavior.
2119 br label %end
2120
2121 end:
2122 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2123 ; Both edges into this PHI are
2124 ; control-dependent on %cmp, so this
2125 ; always results in a poison value.
2126
2127 store volatile i32 0, i32* @g ; This would depend on the store in %true
2128 ; if %cmp is true, or the store in %entry
2129 ; otherwise, so this is undefined behavior.
2130
2131 br i1 %cmp, label %second_true, label %second_end
2132 ; The same branch again, but this time the
2133 ; true block doesn't have side effects.
2134
2135 second_true:
2136 ; No side effects!
2137 ret void
2138
2139 second_end:
2140 store volatile i32 0, i32* @g ; This time, the instruction always depends
2141 ; on the store in %end. Also, it is
2142 ; control-equivalent to %end, so this is
2143 ; well-defined (ignoring earlier undefined
2144 ; behavior in this example).
2145
2146.. _blockaddress:
2147
2148Addresses of Basic Blocks
2149-------------------------
2150
2151``blockaddress(@function, %block)``
2152
2153The '``blockaddress``' constant computes the address of the specified
2154basic block in the specified function, and always has an ``i8*`` type.
2155Taking the address of the entry block is illegal.
2156
2157This value only has defined behavior when used as an operand to the
2158':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2159against null. Pointer equality tests between labels addresses results in
Dmitri Gribenkoae4a9ae2013-01-19 20:34:20 +00002160undefined behavior --- though, again, comparison against null is ok, and
Sean Silvaf722b002012-12-07 10:36:55 +00002161no label is equal to the null pointer. This may be passed around as an
2162opaque pointer sized value as long as the bits are not inspected. This
2163allows ``ptrtoint`` and arithmetic to be performed on these values so
2164long as the original value is reconstituted before the ``indirectbr``
2165instruction.
2166
2167Finally, some targets may provide defined semantics when using the value
2168as the operand to an inline assembly, but that is target specific.
2169
2170Constant Expressions
2171--------------------
2172
2173Constant expressions are used to allow expressions involving other
2174constants to be used as constants. Constant expressions may be of any
2175:ref:`first class <t_firstclass>` type and may involve any LLVM operation
2176that does not have side effects (e.g. load and call are not supported).
2177The following is the syntax for constant expressions:
2178
2179``trunc (CST to TYPE)``
2180 Truncate a constant to another type. The bit size of CST must be
2181 larger than the bit size of TYPE. Both types must be integers.
2182``zext (CST to TYPE)``
2183 Zero extend a constant to another type. The bit size of CST must be
2184 smaller than the bit size of TYPE. Both types must be integers.
2185``sext (CST to TYPE)``
2186 Sign extend a constant to another type. The bit size of CST must be
2187 smaller than the bit size of TYPE. Both types must be integers.
2188``fptrunc (CST to TYPE)``
2189 Truncate a floating point constant to another floating point type.
2190 The size of CST must be larger than the size of TYPE. Both types
2191 must be floating point.
2192``fpext (CST to TYPE)``
2193 Floating point extend a constant to another type. The size of CST
2194 must be smaller or equal to the size of TYPE. Both types must be
2195 floating point.
2196``fptoui (CST to TYPE)``
2197 Convert a floating point constant to the corresponding unsigned
2198 integer constant. TYPE must be a scalar or vector integer type. CST
2199 must be of scalar or vector floating point type. Both CST and TYPE
2200 must be scalars, or vectors of the same number of elements. If the
2201 value won't fit in the integer type, the results are undefined.
2202``fptosi (CST to TYPE)``
2203 Convert a floating point constant to the corresponding signed
2204 integer constant. TYPE must be a scalar or vector integer type. CST
2205 must be of scalar or vector floating point type. Both CST and TYPE
2206 must be scalars, or vectors of the same number of elements. If the
2207 value won't fit in the integer type, the results are undefined.
2208``uitofp (CST to TYPE)``
2209 Convert an unsigned integer constant to the corresponding floating
2210 point constant. TYPE must be a scalar or vector floating point type.
2211 CST must be of scalar or vector integer type. Both CST and TYPE must
2212 be scalars, or vectors of the same number of elements. If the value
2213 won't fit in the floating point type, the results are undefined.
2214``sitofp (CST to TYPE)``
2215 Convert a signed integer constant to the corresponding floating
2216 point constant. TYPE must be a scalar or vector floating point type.
2217 CST must be of scalar or vector integer type. Both CST and TYPE must
2218 be scalars, or vectors of the same number of elements. If the value
2219 won't fit in the floating point type, the results are undefined.
2220``ptrtoint (CST to TYPE)``
2221 Convert a pointer typed constant to the corresponding integer
2222 constant ``TYPE`` must be an integer type. ``CST`` must be of
2223 pointer type. The ``CST`` value is zero extended, truncated, or
2224 unchanged to make it fit in ``TYPE``.
2225``inttoptr (CST to TYPE)``
2226 Convert an integer constant to a pointer constant. TYPE must be a
2227 pointer type. CST must be of integer type. The CST value is zero
2228 extended, truncated, or unchanged to make it fit in a pointer size.
2229 This one is *really* dangerous!
2230``bitcast (CST to TYPE)``
2231 Convert a constant, CST, to another TYPE. The constraints of the
2232 operands are the same as those for the :ref:`bitcast
2233 instruction <i_bitcast>`.
2234``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2235 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2236 constants. As with the :ref:`getelementptr <i_getelementptr>`
2237 instruction, the index list may have zero or more indexes, which are
2238 required to make sense for the type of "CSTPTR".
2239``select (COND, VAL1, VAL2)``
2240 Perform the :ref:`select operation <i_select>` on constants.
2241``icmp COND (VAL1, VAL2)``
2242 Performs the :ref:`icmp operation <i_icmp>` on constants.
2243``fcmp COND (VAL1, VAL2)``
2244 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2245``extractelement (VAL, IDX)``
2246 Perform the :ref:`extractelement operation <i_extractelement>` on
2247 constants.
2248``insertelement (VAL, ELT, IDX)``
2249 Perform the :ref:`insertelement operation <i_insertelement>` on
2250 constants.
2251``shufflevector (VEC1, VEC2, IDXMASK)``
2252 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2253 constants.
2254``extractvalue (VAL, IDX0, IDX1, ...)``
2255 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2256 constants. The index list is interpreted in a similar manner as
2257 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2258 least one index value must be specified.
2259``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2260 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2261 The index list is interpreted in a similar manner as indices in a
2262 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2263 value must be specified.
2264``OPCODE (LHS, RHS)``
2265 Perform the specified operation of the LHS and RHS constants. OPCODE
2266 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2267 binary <bitwiseops>` operations. The constraints on operands are
2268 the same as those for the corresponding instruction (e.g. no bitwise
2269 operations on floating point values are allowed).
2270
2271Other Values
2272============
2273
2274Inline Assembler Expressions
2275----------------------------
2276
2277LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2278Inline Assembly <moduleasm>`) through the use of a special value. This
2279value represents the inline assembler as a string (containing the
2280instructions to emit), a list of operand constraints (stored as a
2281string), a flag that indicates whether or not the inline asm expression
2282has side effects, and a flag indicating whether the function containing
2283the asm needs to align its stack conservatively. An example inline
2284assembler expression is:
2285
2286.. code-block:: llvm
2287
2288 i32 (i32) asm "bswap $0", "=r,r"
2289
2290Inline assembler expressions may **only** be used as the callee operand
2291of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2292Thus, typically we have:
2293
2294.. code-block:: llvm
2295
2296 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2297
2298Inline asms with side effects not visible in the constraint list must be
2299marked as having side effects. This is done through the use of the
2300'``sideeffect``' keyword, like so:
2301
2302.. code-block:: llvm
2303
2304 call void asm sideeffect "eieio", ""()
2305
2306In some cases inline asms will contain code that will not work unless
2307the stack is aligned in some way, such as calls or SSE instructions on
2308x86, yet will not contain code that does that alignment within the asm.
2309The compiler should make conservative assumptions about what the asm
2310might contain and should generate its usual stack alignment code in the
2311prologue if the '``alignstack``' keyword is present:
2312
2313.. code-block:: llvm
2314
2315 call void asm alignstack "eieio", ""()
2316
2317Inline asms also support using non-standard assembly dialects. The
2318assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2319the inline asm is using the Intel dialect. Currently, ATT and Intel are
2320the only supported dialects. An example is:
2321
2322.. code-block:: llvm
2323
2324 call void asm inteldialect "eieio", ""()
2325
2326If multiple keywords appear the '``sideeffect``' keyword must come
2327first, the '``alignstack``' keyword second and the '``inteldialect``'
2328keyword last.
2329
2330Inline Asm Metadata
2331^^^^^^^^^^^^^^^^^^^
2332
2333The call instructions that wrap inline asm nodes may have a
2334"``!srcloc``" MDNode attached to it that contains a list of constant
2335integers. If present, the code generator will use the integer as the
2336location cookie value when report errors through the ``LLVMContext``
2337error reporting mechanisms. This allows a front-end to correlate backend
2338errors that occur with inline asm back to the source code that produced
2339it. For example:
2340
2341.. code-block:: llvm
2342
2343 call void asm sideeffect "something bad", ""(), !srcloc !42
2344 ...
2345 !42 = !{ i32 1234567 }
2346
2347It is up to the front-end to make sense of the magic numbers it places
2348in the IR. If the MDNode contains multiple constants, the code generator
2349will use the one that corresponds to the line of the asm that the error
2350occurs on.
2351
2352.. _metadata:
2353
2354Metadata Nodes and Metadata Strings
2355-----------------------------------
2356
2357LLVM IR allows metadata to be attached to instructions in the program
2358that can convey extra information about the code to the optimizers and
2359code generator. One example application of metadata is source-level
2360debug information. There are two metadata primitives: strings and nodes.
2361All metadata has the ``metadata`` type and is identified in syntax by a
2362preceding exclamation point ('``!``').
2363
2364A metadata string is a string surrounded by double quotes. It can
2365contain any character by escaping non-printable characters with
2366"``\xx``" where "``xx``" is the two digit hex code. For example:
2367"``!"test\00"``".
2368
2369Metadata nodes are represented with notation similar to structure
2370constants (a comma separated list of elements, surrounded by braces and
2371preceded by an exclamation point). Metadata nodes can have any values as
2372their operand. For example:
2373
2374.. code-block:: llvm
2375
2376 !{ metadata !"test\00", i32 10}
2377
2378A :ref:`named metadata <namedmetadatastructure>` is a collection of
2379metadata nodes, which can be looked up in the module symbol table. For
2380example:
2381
2382.. code-block:: llvm
2383
2384 !foo = metadata !{!4, !3}
2385
2386Metadata can be used as function arguments. Here ``llvm.dbg.value``
2387function is using two metadata arguments:
2388
2389.. code-block:: llvm
2390
2391 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2392
2393Metadata can be attached with an instruction. Here metadata ``!21`` is
2394attached to the ``add`` instruction using the ``!dbg`` identifier:
2395
2396.. code-block:: llvm
2397
2398 %indvar.next = add i64 %indvar, 1, !dbg !21
2399
2400More information about specific metadata nodes recognized by the
2401optimizers and code generator is found below.
2402
2403'``tbaa``' Metadata
2404^^^^^^^^^^^^^^^^^^^
2405
2406In LLVM IR, memory does not have types, so LLVM's own type system is not
2407suitable for doing TBAA. Instead, metadata is added to the IR to
2408describe a type system of a higher level language. This can be used to
2409implement typical C/C++ TBAA, but it can also be used to implement
2410custom alias analysis behavior for other languages.
2411
2412The current metadata format is very simple. TBAA metadata nodes have up
2413to three fields, e.g.:
2414
2415.. code-block:: llvm
2416
2417 !0 = metadata !{ metadata !"an example type tree" }
2418 !1 = metadata !{ metadata !"int", metadata !0 }
2419 !2 = metadata !{ metadata !"float", metadata !0 }
2420 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2421
2422The first field is an identity field. It can be any value, usually a
2423metadata string, which uniquely identifies the type. The most important
2424name in the tree is the name of the root node. Two trees with different
2425root node names are entirely disjoint, even if they have leaves with
2426common names.
2427
2428The second field identifies the type's parent node in the tree, or is
2429null or omitted for a root node. A type is considered to alias all of
2430its descendants and all of its ancestors in the tree. Also, a type is
2431considered to alias all types in other trees, so that bitcode produced
2432from multiple front-ends is handled conservatively.
2433
2434If the third field is present, it's an integer which if equal to 1
2435indicates that the type is "constant" (meaning
2436``pointsToConstantMemory`` should return true; see `other useful
2437AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2438
2439'``tbaa.struct``' Metadata
2440^^^^^^^^^^^^^^^^^^^^^^^^^^
2441
2442The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2443aggregate assignment operations in C and similar languages, however it
2444is defined to copy a contiguous region of memory, which is more than
2445strictly necessary for aggregate types which contain holes due to
2446padding. Also, it doesn't contain any TBAA information about the fields
2447of the aggregate.
2448
2449``!tbaa.struct`` metadata can describe which memory subregions in a
2450memcpy are padding and what the TBAA tags of the struct are.
2451
2452The current metadata format is very simple. ``!tbaa.struct`` metadata
2453nodes are a list of operands which are in conceptual groups of three.
2454For each group of three, the first operand gives the byte offset of a
2455field in bytes, the second gives its size in bytes, and the third gives
2456its tbaa tag. e.g.:
2457
2458.. code-block:: llvm
2459
2460 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2461
2462This describes a struct with two fields. The first is at offset 0 bytes
2463with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2464and has size 4 bytes and has tbaa tag !2.
2465
2466Note that the fields need not be contiguous. In this example, there is a
24674 byte gap between the two fields. This gap represents padding which
2468does not carry useful data and need not be preserved.
2469
2470'``fpmath``' Metadata
2471^^^^^^^^^^^^^^^^^^^^^
2472
2473``fpmath`` metadata may be attached to any instruction of floating point
2474type. It can be used to express the maximum acceptable error in the
2475result of that instruction, in ULPs, thus potentially allowing the
2476compiler to use a more efficient but less accurate method of computing
2477it. ULP is defined as follows:
2478
2479 If ``x`` is a real number that lies between two finite consecutive
2480 floating-point numbers ``a`` and ``b``, without being equal to one
2481 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2482 distance between the two non-equal finite floating-point numbers
2483 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2484
2485The metadata node shall consist of a single positive floating point
2486number representing the maximum relative error, for example:
2487
2488.. code-block:: llvm
2489
2490 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2491
2492'``range``' Metadata
2493^^^^^^^^^^^^^^^^^^^^
2494
2495``range`` metadata may be attached only to loads of integer types. It
2496expresses the possible ranges the loaded value is in. The ranges are
2497represented with a flattened list of integers. The loaded value is known
2498to be in the union of the ranges defined by each consecutive pair. Each
2499pair has the following properties:
2500
2501- The type must match the type loaded by the instruction.
2502- The pair ``a,b`` represents the range ``[a,b)``.
2503- Both ``a`` and ``b`` are constants.
2504- The range is allowed to wrap.
2505- The range should not represent the full or empty set. That is,
2506 ``a!=b``.
2507
2508In addition, the pairs must be in signed order of the lower bound and
2509they must be non-contiguous.
2510
2511Examples:
2512
2513.. code-block:: llvm
2514
2515 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2516 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2517 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
2518 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
2519 ...
2520 !0 = metadata !{ i8 0, i8 2 }
2521 !1 = metadata !{ i8 255, i8 2 }
2522 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2523 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
2524
Pekka Jaaskelainen5d0ce792013-02-13 18:08:57 +00002525'``llvm.loop``'
2526^^^^^^^^^^^^^^^
2527
2528It is sometimes useful to attach information to loop constructs. Currently,
2529loop metadata is implemented as metadata attached to the branch instruction
2530in the loop latch block. This type of metadata refer to a metadata node that is
2531guaranteed to be separate for each loop. The loop-level metadata is prefixed
2532with ``llvm.loop``.
2533
2534The loop identifier metadata is implemented using a metadata that refers to
2535itself as follows:
2536
2537.. code-block:: llvm
Paul Redmond8f0696c2013-02-21 17:20:45 +00002538
Pekka Jaaskelainen5d0ce792013-02-13 18:08:57 +00002539 !0 = metadata !{ metadata !0 }
2540
2541'``llvm.loop.parallel``' Metadata
2542^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2543
2544This loop metadata can be used to communicate that a loop should be considered
2545a parallel loop. The semantics of parallel loops in this case is the one
2546with the strongest cross-iteration instruction ordering freedom: the
2547iterations in the loop can be considered completely independent of each
2548other (also known as embarrassingly parallel loops).
2549
2550This metadata can originate from a programming language with parallel loop
2551constructs. In such a case it is completely the programmer's responsibility
2552to ensure the instructions from the different iterations of the loop can be
2553executed in an arbitrary order, in parallel, or intertwined. No loop-carried
2554dependency checking at all must be expected from the compiler.
2555
2556In order to fulfill the LLVM requirement for metadata to be safely ignored,
2557it is important to ensure that a parallel loop is converted to
2558a sequential loop in case an optimization (agnostic of the parallel loop
2559semantics) converts the loop back to such. This happens when new memory
2560accesses that do not fulfill the requirement of free ordering across iterations
2561are added to the loop. Therefore, this metadata is required, but not
2562sufficient, to consider the loop at hand a parallel loop. For a loop
2563to be parallel, all its memory accessing instructions need to be
2564marked with the ``llvm.mem.parallel_loop_access`` metadata that refer
2565to the same loop identifier metadata that identify the loop at hand.
2566
2567'``llvm.mem``'
2568^^^^^^^^^^^^^^^
2569
2570Metadata types used to annotate memory accesses with information helpful
2571for optimizations are prefixed with ``llvm.mem``.
2572
2573'``llvm.mem.parallel_loop_access``' Metadata
2574^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2575
2576For a loop to be parallel, in addition to using
2577the ``llvm.loop.parallel`` metadata to mark the loop latch branch instruction,
2578also all of the memory accessing instructions in the loop body need to be
2579marked with the ``llvm.mem.parallel_loop_access`` metadata. If there
2580is at least one memory accessing instruction not marked with the metadata,
2581the loop, despite it possibly using the ``llvm.loop.parallel`` metadata,
2582must be considered a sequential loop. This causes parallel loops to be
2583converted to sequential loops due to optimization passes that are unaware of
2584the parallel semantics and that insert new memory instructions to the loop
2585body.
2586
2587Example of a loop that is considered parallel due to its correct use of
2588both ``llvm.loop.parallel`` and ``llvm.mem.parallel_loop_access``
2589metadata types that refer to the same loop identifier metadata.
2590
2591.. code-block:: llvm
2592
2593 for.body:
2594 ...
2595 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2596 ...
2597 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2598 ...
2599 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop.parallel !0
2600
2601 for.end:
2602 ...
2603 !0 = metadata !{ metadata !0 }
2604
2605It is also possible to have nested parallel loops. In that case the
2606memory accesses refer to a list of loop identifier metadata nodes instead of
2607the loop identifier metadata node directly:
2608
2609.. code-block:: llvm
2610
2611 outer.for.body:
2612 ...
2613
2614 inner.for.body:
2615 ...
2616 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2617 ...
2618 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2619 ...
2620 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop.parallel !1
2621
2622 inner.for.end:
2623 ...
2624 %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !0
2625 ...
2626 store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !0
2627 ...
2628 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop.parallel !2
2629
2630 outer.for.end: ; preds = %for.body
2631 ...
2632 !0 = metadata !{ metadata !1, metadata !2 } ; a list of parallel loop identifiers
2633 !1 = metadata !{ metadata !1 } ; an identifier for the inner parallel loop
2634 !2 = metadata !{ metadata !2 } ; an identifier for the outer parallel loop
2635
2636
Sean Silvaf722b002012-12-07 10:36:55 +00002637Module Flags Metadata
2638=====================
2639
2640Information about the module as a whole is difficult to convey to LLVM's
2641subsystems. The LLVM IR isn't sufficient to transmit this information.
2642The ``llvm.module.flags`` named metadata exists in order to facilitate
Dmitri Gribenkoae4a9ae2013-01-19 20:34:20 +00002643this. These flags are in the form of key / value pairs --- much like a
2644dictionary --- making it easy for any subsystem who cares about a flag to
Sean Silvaf722b002012-12-07 10:36:55 +00002645look it up.
2646
2647The ``llvm.module.flags`` metadata contains a list of metadata triplets.
2648Each triplet has the following form:
2649
2650- The first element is a *behavior* flag, which specifies the behavior
2651 when two (or more) modules are merged together, and it encounters two
2652 (or more) metadata with the same ID. The supported behaviors are
2653 described below.
2654- The second element is a metadata string that is a unique ID for the
Daniel Dunbar8dd938e2013-01-15 01:22:53 +00002655 metadata. Each module may only have one flag entry for each unique ID (not
2656 including entries with the **Require** behavior).
Sean Silvaf722b002012-12-07 10:36:55 +00002657- The third element is the value of the flag.
2658
2659When two (or more) modules are merged together, the resulting
Daniel Dunbar8dd938e2013-01-15 01:22:53 +00002660``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
2661each unique metadata ID string, there will be exactly one entry in the merged
2662modules ``llvm.module.flags`` metadata table, and the value for that entry will
2663be determined by the merge behavior flag, as described below. The only exception
2664is that entries with the *Require* behavior are always preserved.
Sean Silvaf722b002012-12-07 10:36:55 +00002665
2666The following behaviors are supported:
2667
2668.. list-table::
2669 :header-rows: 1
2670 :widths: 10 90
2671
2672 * - Value
2673 - Behavior
2674
2675 * - 1
2676 - **Error**
Daniel Dunbar8dd938e2013-01-15 01:22:53 +00002677 Emits an error if two values disagree, otherwise the resulting value
2678 is that of the operands.
Sean Silvaf722b002012-12-07 10:36:55 +00002679
2680 * - 2
2681 - **Warning**
Daniel Dunbar8dd938e2013-01-15 01:22:53 +00002682 Emits a warning if two values disagree. The result value will be the
2683 operand for the flag from the first module being linked.
Sean Silvaf722b002012-12-07 10:36:55 +00002684
2685 * - 3
2686 - **Require**
Daniel Dunbar8dd938e2013-01-15 01:22:53 +00002687 Adds a requirement that another module flag be present and have a
2688 specified value after linking is performed. The value must be a
2689 metadata pair, where the first element of the pair is the ID of the
2690 module flag to be restricted, and the second element of the pair is
2691 the value the module flag should be restricted to. This behavior can
2692 be used to restrict the allowable results (via triggering of an
2693 error) of linking IDs with the **Override** behavior.
Sean Silvaf722b002012-12-07 10:36:55 +00002694
2695 * - 4
2696 - **Override**
Daniel Dunbar8dd938e2013-01-15 01:22:53 +00002697 Uses the specified value, regardless of the behavior or value of the
2698 other module. If both modules specify **Override**, but the values
2699 differ, an error will be emitted.
2700
Daniel Dunbar5db391c2013-01-16 21:38:56 +00002701 * - 5
2702 - **Append**
2703 Appends the two values, which are required to be metadata nodes.
2704
2705 * - 6
2706 - **AppendUnique**
2707 Appends the two values, which are required to be metadata
2708 nodes. However, duplicate entries in the second list are dropped
2709 during the append operation.
2710
Daniel Dunbar8dd938e2013-01-15 01:22:53 +00002711It is an error for a particular unique flag ID to have multiple behaviors,
2712except in the case of **Require** (which adds restrictions on another metadata
2713value) or **Override**.
Sean Silvaf722b002012-12-07 10:36:55 +00002714
2715An example of module flags:
2716
2717.. code-block:: llvm
2718
2719 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
2720 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
2721 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
2722 !3 = metadata !{ i32 3, metadata !"qux",
2723 metadata !{
2724 metadata !"foo", i32 1
2725 }
2726 }
2727 !llvm.module.flags = !{ !0, !1, !2, !3 }
2728
2729- Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
2730 if two or more ``!"foo"`` flags are seen is to emit an error if their
2731 values are not equal.
2732
2733- Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
2734 behavior if two or more ``!"bar"`` flags are seen is to use the value
Daniel Dunbar8dd938e2013-01-15 01:22:53 +00002735 '37'.
Sean Silvaf722b002012-12-07 10:36:55 +00002736
2737- Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
2738 behavior if two or more ``!"qux"`` flags are seen is to emit a
2739 warning if their values are not equal.
2740
2741- Metadata ``!3`` has the ID ``!"qux"`` and the value:
2742
2743 ::
2744
2745 metadata !{ metadata !"foo", i32 1 }
2746
Daniel Dunbar8dd938e2013-01-15 01:22:53 +00002747 The behavior is to emit an error if the ``llvm.module.flags`` does not
2748 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
2749 performed.
Sean Silvaf722b002012-12-07 10:36:55 +00002750
2751Objective-C Garbage Collection Module Flags Metadata
2752----------------------------------------------------
2753
2754On the Mach-O platform, Objective-C stores metadata about garbage
2755collection in a special section called "image info". The metadata
2756consists of a version number and a bitmask specifying what types of
2757garbage collection are supported (if any) by the file. If two or more
2758modules are linked together their garbage collection metadata needs to
2759be merged rather than appended together.
2760
2761The Objective-C garbage collection module flags metadata consists of the
2762following key-value pairs:
2763
2764.. list-table::
2765 :header-rows: 1
2766 :widths: 30 70
2767
2768 * - Key
2769 - Value
2770
Daniel Dunbar3389dbc2013-01-17 18:57:32 +00002771 * - ``Objective-C Version``
Dmitri Gribenkoae4a9ae2013-01-19 20:34:20 +00002772 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
Sean Silvaf722b002012-12-07 10:36:55 +00002773
Daniel Dunbar3389dbc2013-01-17 18:57:32 +00002774 * - ``Objective-C Image Info Version``
Dmitri Gribenkoae4a9ae2013-01-19 20:34:20 +00002775 - **[Required]** --- The version of the image info section. Currently
Sean Silvaf722b002012-12-07 10:36:55 +00002776 always 0.
2777
Daniel Dunbar3389dbc2013-01-17 18:57:32 +00002778 * - ``Objective-C Image Info Section``
Dmitri Gribenkoae4a9ae2013-01-19 20:34:20 +00002779 - **[Required]** --- The section to place the metadata. Valid values are
Sean Silvaf722b002012-12-07 10:36:55 +00002780 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
2781 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
2782 Objective-C ABI version 2.
2783
Daniel Dunbar3389dbc2013-01-17 18:57:32 +00002784 * - ``Objective-C Garbage Collection``
Dmitri Gribenkoae4a9ae2013-01-19 20:34:20 +00002785 - **[Required]** --- Specifies whether garbage collection is supported or
Sean Silvaf722b002012-12-07 10:36:55 +00002786 not. Valid values are 0, for no garbage collection, and 2, for garbage
2787 collection supported.
2788
Daniel Dunbar3389dbc2013-01-17 18:57:32 +00002789 * - ``Objective-C GC Only``
Dmitri Gribenkoae4a9ae2013-01-19 20:34:20 +00002790 - **[Optional]** --- Specifies that only garbage collection is supported.
Sean Silvaf722b002012-12-07 10:36:55 +00002791 If present, its value must be 6. This flag requires that the
2792 ``Objective-C Garbage Collection`` flag have the value 2.
2793
2794Some important flag interactions:
2795
2796- If a module with ``Objective-C Garbage Collection`` set to 0 is
2797 merged with a module with ``Objective-C Garbage Collection`` set to
2798 2, then the resulting module has the
2799 ``Objective-C Garbage Collection`` flag set to 0.
2800- A module with ``Objective-C Garbage Collection`` set to 0 cannot be
2801 merged with a module with ``Objective-C GC Only`` set to 6.
2802
Daniel Dunbare06bfe82013-01-17 00:16:27 +00002803Automatic Linker Flags Module Flags Metadata
2804--------------------------------------------
2805
2806Some targets support embedding flags to the linker inside individual object
2807files. Typically this is used in conjunction with language extensions which
2808allow source files to explicitly declare the libraries they depend on, and have
2809these automatically be transmitted to the linker via object files.
2810
2811These flags are encoded in the IR using metadata in the module flags section,
Daniel Dunbar3389dbc2013-01-17 18:57:32 +00002812using the ``Linker Options`` key. The merge behavior for this flag is required
Daniel Dunbare06bfe82013-01-17 00:16:27 +00002813to be ``AppendUnique``, and the value for the key is expected to be a metadata
2814node which should be a list of other metadata nodes, each of which should be a
2815list of metadata strings defining linker options.
2816
2817For example, the following metadata section specifies two separate sets of
2818linker options, presumably to link against ``libz`` and the ``Cocoa``
2819framework::
2820
Daniel Dunbar6d49b682013-01-18 19:37:00 +00002821 !0 = metadata !{ i32 6, metadata !"Linker Options",
Daniel Dunbare06bfe82013-01-17 00:16:27 +00002822 metadata !{
Daniel Dunbar6d49b682013-01-18 19:37:00 +00002823 metadata !{ metadata !"-lz" },
2824 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
Daniel Dunbare06bfe82013-01-17 00:16:27 +00002825 !llvm.module.flags = !{ !0 }
2826
2827The metadata encoding as lists of lists of options, as opposed to a collapsed
2828list of options, is chosen so that the IR encoding can use multiple option
2829strings to specify e.g., a single library, while still having that specifier be
2830preserved as an atomic element that can be recognized by a target specific
2831assembly writer or object file emitter.
2832
2833Each individual option is required to be either a valid option for the target's
2834linker, or an option that is reserved by the target specific assembly writer or
2835object file emitter. No other aspect of these options is defined by the IR.
2836
Sean Silvaf722b002012-12-07 10:36:55 +00002837Intrinsic Global Variables
2838==========================
2839
2840LLVM has a number of "magic" global variables that contain data that
2841affect code generation or other IR semantics. These are documented here.
2842All globals of this sort should have a section specified as
2843"``llvm.metadata``". This section and all globals that start with
2844"``llvm.``" are reserved for use by LLVM.
2845
2846The '``llvm.used``' Global Variable
2847-----------------------------------
2848
2849The ``@llvm.used`` global is an array with i8\* element type which has
2850:ref:`appending linkage <linkage_appending>`. This array contains a list of
2851pointers to global variables and functions which may optionally have a
2852pointer cast formed of bitcast or getelementptr. For example, a legal
2853use of it is:
2854
2855.. code-block:: llvm
2856
2857 @X = global i8 4
2858 @Y = global i32 123
2859
2860 @llvm.used = appending global [2 x i8*] [
2861 i8* @X,
2862 i8* bitcast (i32* @Y to i8*)
2863 ], section "llvm.metadata"
2864
2865If a global variable appears in the ``@llvm.used`` list, then the
2866compiler, assembler, and linker are required to treat the symbol as if
2867there is a reference to the global that it cannot see. For example, if a
2868variable has internal linkage and no references other than that from the
2869``@llvm.used`` list, it cannot be deleted. This is commonly used to
2870represent references from inline asms and other things the compiler
2871cannot "see", and corresponds to "``attribute((used))``" in GNU C.
2872
2873On some targets, the code generator must emit a directive to the
2874assembler or object file to prevent the assembler and linker from
2875molesting the symbol.
2876
2877The '``llvm.compiler.used``' Global Variable
2878--------------------------------------------
2879
2880The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
2881directive, except that it only prevents the compiler from touching the
2882symbol. On targets that support it, this allows an intelligent linker to
2883optimize references to the symbol without being impeded as it would be
2884by ``@llvm.used``.
2885
2886This is a rare construct that should only be used in rare circumstances,
2887and should not be exposed to source languages.
2888
2889The '``llvm.global_ctors``' Global Variable
2890-------------------------------------------
2891
2892.. code-block:: llvm
2893
2894 %0 = type { i32, void ()* }
2895 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2896
2897The ``@llvm.global_ctors`` array contains a list of constructor
2898functions and associated priorities. The functions referenced by this
2899array will be called in ascending order of priority (i.e. lowest first)
2900when the module is loaded. The order of functions with the same priority
2901is not defined.
2902
2903The '``llvm.global_dtors``' Global Variable
2904-------------------------------------------
2905
2906.. code-block:: llvm
2907
2908 %0 = type { i32, void ()* }
2909 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
2910
2911The ``@llvm.global_dtors`` array contains a list of destructor functions
2912and associated priorities. The functions referenced by this array will
2913be called in descending order of priority (i.e. highest first) when the
2914module is loaded. The order of functions with the same priority is not
2915defined.
2916
2917Instruction Reference
2918=====================
2919
2920The LLVM instruction set consists of several different classifications
2921of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
2922instructions <binaryops>`, :ref:`bitwise binary
2923instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
2924:ref:`other instructions <otherops>`.
2925
2926.. _terminators:
2927
2928Terminator Instructions
2929-----------------------
2930
2931As mentioned :ref:`previously <functionstructure>`, every basic block in a
2932program ends with a "Terminator" instruction, which indicates which
2933block should be executed after the current block is finished. These
2934terminator instructions typically yield a '``void``' value: they produce
2935control flow, not values (the one exception being the
2936':ref:`invoke <i_invoke>`' instruction).
2937
2938The terminator instructions are: ':ref:`ret <i_ret>`',
2939':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
2940':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
2941':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
2942
2943.. _i_ret:
2944
2945'``ret``' Instruction
2946^^^^^^^^^^^^^^^^^^^^^
2947
2948Syntax:
2949"""""""
2950
2951::
2952
2953 ret <type> <value> ; Return a value from a non-void function
2954 ret void ; Return from void function
2955
2956Overview:
2957"""""""""
2958
2959The '``ret``' instruction is used to return control flow (and optionally
2960a value) from a function back to the caller.
2961
2962There are two forms of the '``ret``' instruction: one that returns a
2963value and then causes control flow, and one that just causes control
2964flow to occur.
2965
2966Arguments:
2967""""""""""
2968
2969The '``ret``' instruction optionally accepts a single argument, the
2970return value. The type of the return value must be a ':ref:`first
2971class <t_firstclass>`' type.
2972
2973A function is not :ref:`well formed <wellformed>` if it it has a non-void
2974return type and contains a '``ret``' instruction with no return value or
2975a return value with a type that does not match its type, or if it has a
2976void return type and contains a '``ret``' instruction with a return
2977value.
2978
2979Semantics:
2980""""""""""
2981
2982When the '``ret``' instruction is executed, control flow returns back to
2983the calling function's context. If the caller is a
2984":ref:`call <i_call>`" instruction, execution continues at the
2985instruction after the call. If the caller was an
2986":ref:`invoke <i_invoke>`" instruction, execution continues at the
2987beginning of the "normal" destination block. If the instruction returns
2988a value, that value shall set the call or invoke instruction's return
2989value.
2990
2991Example:
2992""""""""
2993
2994.. code-block:: llvm
2995
2996 ret i32 5 ; Return an integer value of 5
2997 ret void ; Return from a void function
2998 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
2999
3000.. _i_br:
3001
3002'``br``' Instruction
3003^^^^^^^^^^^^^^^^^^^^
3004
3005Syntax:
3006"""""""
3007
3008::
3009
3010 br i1 <cond>, label <iftrue>, label <iffalse>
3011 br label <dest> ; Unconditional branch
3012
3013Overview:
3014"""""""""
3015
3016The '``br``' instruction is used to cause control flow to transfer to a
3017different basic block in the current function. There are two forms of
3018this instruction, corresponding to a conditional branch and an
3019unconditional branch.
3020
3021Arguments:
3022""""""""""
3023
3024The conditional branch form of the '``br``' instruction takes a single
3025'``i1``' value and two '``label``' values. The unconditional form of the
3026'``br``' instruction takes a single '``label``' value as a target.
3027
3028Semantics:
3029""""""""""
3030
3031Upon execution of a conditional '``br``' instruction, the '``i1``'
3032argument is evaluated. If the value is ``true``, control flows to the
3033'``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3034to the '``iffalse``' ``label`` argument.
3035
3036Example:
3037""""""""
3038
3039.. code-block:: llvm
3040
3041 Test:
3042 %cond = icmp eq i32 %a, %b
3043 br i1 %cond, label %IfEqual, label %IfUnequal
3044 IfEqual:
3045 ret i32 1
3046 IfUnequal:
3047 ret i32 0
3048
3049.. _i_switch:
3050
3051'``switch``' Instruction
3052^^^^^^^^^^^^^^^^^^^^^^^^
3053
3054Syntax:
3055"""""""
3056
3057::
3058
3059 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3060
3061Overview:
3062"""""""""
3063
3064The '``switch``' instruction is used to transfer control flow to one of
3065several different places. It is a generalization of the '``br``'
3066instruction, allowing a branch to occur to one of many possible
3067destinations.
3068
3069Arguments:
3070""""""""""
3071
3072The '``switch``' instruction uses three parameters: an integer
3073comparison value '``value``', a default '``label``' destination, and an
3074array of pairs of comparison value constants and '``label``'s. The table
3075is not allowed to contain duplicate constant entries.
3076
3077Semantics:
3078""""""""""
3079
3080The ``switch`` instruction specifies a table of values and destinations.
3081When the '``switch``' instruction is executed, this table is searched
3082for the given value. If the value is found, control flow is transferred
3083to the corresponding destination; otherwise, control flow is transferred
3084to the default destination.
3085
3086Implementation:
3087"""""""""""""""
3088
3089Depending on properties of the target machine and the particular
3090``switch`` instruction, this instruction may be code generated in
3091different ways. For example, it could be generated as a series of
3092chained conditional branches or with a lookup table.
3093
3094Example:
3095""""""""
3096
3097.. code-block:: llvm
3098
3099 ; Emulate a conditional br instruction
3100 %Val = zext i1 %value to i32
3101 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3102
3103 ; Emulate an unconditional br instruction
3104 switch i32 0, label %dest [ ]
3105
3106 ; Implement a jump table:
3107 switch i32 %val, label %otherwise [ i32 0, label %onzero
3108 i32 1, label %onone
3109 i32 2, label %ontwo ]
3110
3111.. _i_indirectbr:
3112
3113'``indirectbr``' Instruction
3114^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3115
3116Syntax:
3117"""""""
3118
3119::
3120
3121 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3122
3123Overview:
3124"""""""""
3125
3126The '``indirectbr``' instruction implements an indirect branch to a
3127label within the current function, whose address is specified by
3128"``address``". Address must be derived from a
3129:ref:`blockaddress <blockaddress>` constant.
3130
3131Arguments:
3132""""""""""
3133
3134The '``address``' argument is the address of the label to jump to. The
3135rest of the arguments indicate the full set of possible destinations
3136that the address may point to. Blocks are allowed to occur multiple
3137times in the destination list, though this isn't particularly useful.
3138
3139This destination list is required so that dataflow analysis has an
3140accurate understanding of the CFG.
3141
3142Semantics:
3143""""""""""
3144
3145Control transfers to the block specified in the address argument. All
3146possible destination blocks must be listed in the label list, otherwise
3147this instruction has undefined behavior. This implies that jumps to
3148labels defined in other functions have undefined behavior as well.
3149
3150Implementation:
3151"""""""""""""""
3152
3153This is typically implemented with a jump through a register.
3154
3155Example:
3156""""""""
3157
3158.. code-block:: llvm
3159
3160 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3161
3162.. _i_invoke:
3163
3164'``invoke``' Instruction
3165^^^^^^^^^^^^^^^^^^^^^^^^
3166
3167Syntax:
3168"""""""
3169
3170::
3171
3172 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3173 to label <normal label> unwind label <exception label>
3174
3175Overview:
3176"""""""""
3177
3178The '``invoke``' instruction causes control to transfer to a specified
3179function, with the possibility of control flow transfer to either the
3180'``normal``' label or the '``exception``' label. If the callee function
3181returns with the "``ret``" instruction, control flow will return to the
3182"normal" label. If the callee (or any indirect callees) returns via the
3183":ref:`resume <i_resume>`" instruction or other exception handling
3184mechanism, control is interrupted and continued at the dynamically
3185nearest "exception" label.
3186
3187The '``exception``' label is a `landing
3188pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3189'``exception``' label is required to have the
3190":ref:`landingpad <i_landingpad>`" instruction, which contains the
3191information about the behavior of the program after unwinding happens,
3192as its first non-PHI instruction. The restrictions on the
3193"``landingpad``" instruction's tightly couples it to the "``invoke``"
3194instruction, so that the important information contained within the
3195"``landingpad``" instruction can't be lost through normal code motion.
3196
3197Arguments:
3198""""""""""
3199
3200This instruction requires several arguments:
3201
3202#. The optional "cconv" marker indicates which :ref:`calling
3203 convention <callingconv>` the call should use. If none is
3204 specified, the call defaults to using C calling conventions.
3205#. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3206 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3207 are valid here.
3208#. '``ptr to function ty``': shall be the signature of the pointer to
3209 function value being invoked. In most cases, this is a direct
3210 function invocation, but indirect ``invoke``'s are just as possible,
3211 branching off an arbitrary pointer to function value.
3212#. '``function ptr val``': An LLVM value containing a pointer to a
3213 function to be invoked.
3214#. '``function args``': argument list whose types match the function
3215 signature argument types and parameter attributes. All arguments must
3216 be of :ref:`first class <t_firstclass>` type. If the function signature
3217 indicates the function accepts a variable number of arguments, the
3218 extra arguments can be specified.
3219#. '``normal label``': the label reached when the called function
3220 executes a '``ret``' instruction.
3221#. '``exception label``': the label reached when a callee returns via
3222 the :ref:`resume <i_resume>` instruction or other exception handling
3223 mechanism.
3224#. The optional :ref:`function attributes <fnattrs>` list. Only
3225 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3226 attributes are valid here.
3227
3228Semantics:
3229""""""""""
3230
3231This instruction is designed to operate as a standard '``call``'
3232instruction in most regards. The primary difference is that it
3233establishes an association with a label, which is used by the runtime
3234library to unwind the stack.
3235
3236This instruction is used in languages with destructors to ensure that
3237proper cleanup is performed in the case of either a ``longjmp`` or a
3238thrown exception. Additionally, this is important for implementation of
3239'``catch``' clauses in high-level languages that support them.
3240
3241For the purposes of the SSA form, the definition of the value returned
3242by the '``invoke``' instruction is deemed to occur on the edge from the
3243current block to the "normal" label. If the callee unwinds then no
3244return value is available.
3245
3246Example:
3247""""""""
3248
3249.. code-block:: llvm
3250
3251 %retval = invoke i32 @Test(i32 15) to label %Continue
3252 unwind label %TestCleanup ; {i32}:retval set
3253 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3254 unwind label %TestCleanup ; {i32}:retval set
3255
3256.. _i_resume:
3257
3258'``resume``' Instruction
3259^^^^^^^^^^^^^^^^^^^^^^^^
3260
3261Syntax:
3262"""""""
3263
3264::
3265
3266 resume <type> <value>
3267
3268Overview:
3269"""""""""
3270
3271The '``resume``' instruction is a terminator instruction that has no
3272successors.
3273
3274Arguments:
3275""""""""""
3276
3277The '``resume``' instruction requires one argument, which must have the
3278same type as the result of any '``landingpad``' instruction in the same
3279function.
3280
3281Semantics:
3282""""""""""
3283
3284The '``resume``' instruction resumes propagation of an existing
3285(in-flight) exception whose unwinding was interrupted with a
3286:ref:`landingpad <i_landingpad>` instruction.
3287
3288Example:
3289""""""""
3290
3291.. code-block:: llvm
3292
3293 resume { i8*, i32 } %exn
3294
3295.. _i_unreachable:
3296
3297'``unreachable``' Instruction
3298^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3299
3300Syntax:
3301"""""""
3302
3303::
3304
3305 unreachable
3306
3307Overview:
3308"""""""""
3309
3310The '``unreachable``' instruction has no defined semantics. This
3311instruction is used to inform the optimizer that a particular portion of
3312the code is not reachable. This can be used to indicate that the code
3313after a no-return function cannot be reached, and other facts.
3314
3315Semantics:
3316""""""""""
3317
3318The '``unreachable``' instruction has no defined semantics.
3319
3320.. _binaryops:
3321
3322Binary Operations
3323-----------------
3324
3325Binary operators are used to do most of the computation in a program.
3326They require two operands of the same type, execute an operation on
3327them, and produce a single value. The operands might represent multiple
3328data, as is the case with the :ref:`vector <t_vector>` data type. The
3329result value has the same type as its operands.
3330
3331There are several different binary operators:
3332
3333.. _i_add:
3334
3335'``add``' Instruction
3336^^^^^^^^^^^^^^^^^^^^^
3337
3338Syntax:
3339"""""""
3340
3341::
3342
3343 <result> = add <ty> <op1>, <op2> ; yields {ty}:result
3344 <result> = add nuw <ty> <op1>, <op2> ; yields {ty}:result
3345 <result> = add nsw <ty> <op1>, <op2> ; yields {ty}:result
3346 <result> = add nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3347
3348Overview:
3349"""""""""
3350
3351The '``add``' instruction returns the sum of its two operands.
3352
3353Arguments:
3354""""""""""
3355
3356The two arguments to the '``add``' instruction must be
3357:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3358arguments must have identical types.
3359
3360Semantics:
3361""""""""""
3362
3363The value produced is the integer sum of the two operands.
3364
3365If the sum has unsigned overflow, the result returned is the
3366mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3367the result.
3368
3369Because LLVM integers use a two's complement representation, this
3370instruction is appropriate for both signed and unsigned integers.
3371
3372``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3373respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3374result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
3375unsigned and/or signed overflow, respectively, occurs.
3376
3377Example:
3378""""""""
3379
3380.. code-block:: llvm
3381
3382 <result> = add i32 4, %var ; yields {i32}:result = 4 + %var
3383
3384.. _i_fadd:
3385
3386'``fadd``' Instruction
3387^^^^^^^^^^^^^^^^^^^^^^
3388
3389Syntax:
3390"""""""
3391
3392::
3393
3394 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3395
3396Overview:
3397"""""""""
3398
3399The '``fadd``' instruction returns the sum of its two operands.
3400
3401Arguments:
3402""""""""""
3403
3404The two arguments to the '``fadd``' instruction must be :ref:`floating
3405point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3406Both arguments must have identical types.
3407
3408Semantics:
3409""""""""""
3410
3411The value produced is the floating point sum of the two operands. This
3412instruction can also take any number of :ref:`fast-math flags <fastmath>`,
3413which are optimization hints to enable otherwise unsafe floating point
3414optimizations:
3415
3416Example:
3417""""""""
3418
3419.. code-block:: llvm
3420
3421 <result> = fadd float 4.0, %var ; yields {float}:result = 4.0 + %var
3422
3423'``sub``' Instruction
3424^^^^^^^^^^^^^^^^^^^^^
3425
3426Syntax:
3427"""""""
3428
3429::
3430
3431 <result> = sub <ty> <op1>, <op2> ; yields {ty}:result
3432 <result> = sub nuw <ty> <op1>, <op2> ; yields {ty}:result
3433 <result> = sub nsw <ty> <op1>, <op2> ; yields {ty}:result
3434 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3435
3436Overview:
3437"""""""""
3438
3439The '``sub``' instruction returns the difference of its two operands.
3440
3441Note that the '``sub``' instruction is used to represent the '``neg``'
3442instruction present in most other intermediate representations.
3443
3444Arguments:
3445""""""""""
3446
3447The two arguments to the '``sub``' instruction must be
3448:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3449arguments must have identical types.
3450
3451Semantics:
3452""""""""""
3453
3454The value produced is the integer difference of the two operands.
3455
3456If the difference has unsigned overflow, the result returned is the
3457mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
3458the result.
3459
3460Because LLVM integers use a two's complement representation, this
3461instruction is appropriate for both signed and unsigned integers.
3462
3463``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3464respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3465result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
3466unsigned and/or signed overflow, respectively, occurs.
3467
3468Example:
3469""""""""
3470
3471.. code-block:: llvm
3472
3473 <result> = sub i32 4, %var ; yields {i32}:result = 4 - %var
3474 <result> = sub i32 0, %val ; yields {i32}:result = -%var
3475
3476.. _i_fsub:
3477
3478'``fsub``' Instruction
3479^^^^^^^^^^^^^^^^^^^^^^
3480
3481Syntax:
3482"""""""
3483
3484::
3485
3486 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3487
3488Overview:
3489"""""""""
3490
3491The '``fsub``' instruction returns the difference of its two operands.
3492
3493Note that the '``fsub``' instruction is used to represent the '``fneg``'
3494instruction present in most other intermediate representations.
3495
3496Arguments:
3497""""""""""
3498
3499The two arguments to the '``fsub``' instruction must be :ref:`floating
3500point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3501Both arguments must have identical types.
3502
3503Semantics:
3504""""""""""
3505
3506The value produced is the floating point difference of the two operands.
3507This instruction can also take any number of :ref:`fast-math
3508flags <fastmath>`, which are optimization hints to enable otherwise
3509unsafe floating point optimizations:
3510
3511Example:
3512""""""""
3513
3514.. code-block:: llvm
3515
3516 <result> = fsub float 4.0, %var ; yields {float}:result = 4.0 - %var
3517 <result> = fsub float -0.0, %val ; yields {float}:result = -%var
3518
3519'``mul``' Instruction
3520^^^^^^^^^^^^^^^^^^^^^
3521
3522Syntax:
3523"""""""
3524
3525::
3526
3527 <result> = mul <ty> <op1>, <op2> ; yields {ty}:result
3528 <result> = mul nuw <ty> <op1>, <op2> ; yields {ty}:result
3529 <result> = mul nsw <ty> <op1>, <op2> ; yields {ty}:result
3530 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3531
3532Overview:
3533"""""""""
3534
3535The '``mul``' instruction returns the product of its two operands.
3536
3537Arguments:
3538""""""""""
3539
3540The two arguments to the '``mul``' instruction must be
3541:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3542arguments must have identical types.
3543
3544Semantics:
3545""""""""""
3546
3547The value produced is the integer product of the two operands.
3548
3549If the result of the multiplication has unsigned overflow, the result
3550returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
3551bit width of the result.
3552
3553Because LLVM integers use a two's complement representation, and the
3554result is the same width as the operands, this instruction returns the
3555correct result for both signed and unsigned integers. If a full product
3556(e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
3557sign-extended or zero-extended as appropriate to the width of the full
3558product.
3559
3560``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
3561respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
3562result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
3563unsigned and/or signed overflow, respectively, occurs.
3564
3565Example:
3566""""""""
3567
3568.. code-block:: llvm
3569
3570 <result> = mul i32 4, %var ; yields {i32}:result = 4 * %var
3571
3572.. _i_fmul:
3573
3574'``fmul``' Instruction
3575^^^^^^^^^^^^^^^^^^^^^^
3576
3577Syntax:
3578"""""""
3579
3580::
3581
3582 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3583
3584Overview:
3585"""""""""
3586
3587The '``fmul``' instruction returns the product of its two operands.
3588
3589Arguments:
3590""""""""""
3591
3592The two arguments to the '``fmul``' instruction must be :ref:`floating
3593point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3594Both arguments must have identical types.
3595
3596Semantics:
3597""""""""""
3598
3599The value produced is the floating point product of the two operands.
3600This instruction can also take any number of :ref:`fast-math
3601flags <fastmath>`, which are optimization hints to enable otherwise
3602unsafe floating point optimizations:
3603
3604Example:
3605""""""""
3606
3607.. code-block:: llvm
3608
3609 <result> = fmul float 4.0, %var ; yields {float}:result = 4.0 * %var
3610
3611'``udiv``' Instruction
3612^^^^^^^^^^^^^^^^^^^^^^
3613
3614Syntax:
3615"""""""
3616
3617::
3618
3619 <result> = udiv <ty> <op1>, <op2> ; yields {ty}:result
3620 <result> = udiv exact <ty> <op1>, <op2> ; yields {ty}:result
3621
3622Overview:
3623"""""""""
3624
3625The '``udiv``' instruction returns the quotient of its two operands.
3626
3627Arguments:
3628""""""""""
3629
3630The two arguments to the '``udiv``' instruction must be
3631:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3632arguments must have identical types.
3633
3634Semantics:
3635""""""""""
3636
3637The value produced is the unsigned integer quotient of the two operands.
3638
3639Note that unsigned integer division and signed integer division are
3640distinct operations; for signed integer division, use '``sdiv``'.
3641
3642Division by zero leads to undefined behavior.
3643
3644If the ``exact`` keyword is present, the result value of the ``udiv`` is
3645a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
3646such, "((a udiv exact b) mul b) == a").
3647
3648Example:
3649""""""""
3650
3651.. code-block:: llvm
3652
3653 <result> = udiv i32 4, %var ; yields {i32}:result = 4 / %var
3654
3655'``sdiv``' Instruction
3656^^^^^^^^^^^^^^^^^^^^^^
3657
3658Syntax:
3659"""""""
3660
3661::
3662
3663 <result> = sdiv <ty> <op1>, <op2> ; yields {ty}:result
3664 <result> = sdiv exact <ty> <op1>, <op2> ; yields {ty}:result
3665
3666Overview:
3667"""""""""
3668
3669The '``sdiv``' instruction returns the quotient of its two operands.
3670
3671Arguments:
3672""""""""""
3673
3674The two arguments to the '``sdiv``' instruction must be
3675:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3676arguments must have identical types.
3677
3678Semantics:
3679""""""""""
3680
3681The value produced is the signed integer quotient of the two operands
3682rounded towards zero.
3683
3684Note that signed integer division and unsigned integer division are
3685distinct operations; for unsigned integer division, use '``udiv``'.
3686
3687Division by zero leads to undefined behavior. Overflow also leads to
3688undefined behavior; this is a rare case, but can occur, for example, by
3689doing a 32-bit division of -2147483648 by -1.
3690
3691If the ``exact`` keyword is present, the result value of the ``sdiv`` is
3692a :ref:`poison value <poisonvalues>` if the result would be rounded.
3693
3694Example:
3695""""""""
3696
3697.. code-block:: llvm
3698
3699 <result> = sdiv i32 4, %var ; yields {i32}:result = 4 / %var
3700
3701.. _i_fdiv:
3702
3703'``fdiv``' Instruction
3704^^^^^^^^^^^^^^^^^^^^^^
3705
3706Syntax:
3707"""""""
3708
3709::
3710
3711 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3712
3713Overview:
3714"""""""""
3715
3716The '``fdiv``' instruction returns the quotient of its two operands.
3717
3718Arguments:
3719""""""""""
3720
3721The two arguments to the '``fdiv``' instruction must be :ref:`floating
3722point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3723Both arguments must have identical types.
3724
3725Semantics:
3726""""""""""
3727
3728The value produced is the floating point quotient of the two operands.
3729This instruction can also take any number of :ref:`fast-math
3730flags <fastmath>`, which are optimization hints to enable otherwise
3731unsafe floating point optimizations:
3732
3733Example:
3734""""""""
3735
3736.. code-block:: llvm
3737
3738 <result> = fdiv float 4.0, %var ; yields {float}:result = 4.0 / %var
3739
3740'``urem``' Instruction
3741^^^^^^^^^^^^^^^^^^^^^^
3742
3743Syntax:
3744"""""""
3745
3746::
3747
3748 <result> = urem <ty> <op1>, <op2> ; yields {ty}:result
3749
3750Overview:
3751"""""""""
3752
3753The '``urem``' instruction returns the remainder from the unsigned
3754division of its two arguments.
3755
3756Arguments:
3757""""""""""
3758
3759The two arguments to the '``urem``' instruction must be
3760:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3761arguments must have identical types.
3762
3763Semantics:
3764""""""""""
3765
3766This instruction returns the unsigned integer *remainder* of a division.
3767This instruction always performs an unsigned division to get the
3768remainder.
3769
3770Note that unsigned integer remainder and signed integer remainder are
3771distinct operations; for signed integer remainder, use '``srem``'.
3772
3773Taking the remainder of a division by zero leads to undefined behavior.
3774
3775Example:
3776""""""""
3777
3778.. code-block:: llvm
3779
3780 <result> = urem i32 4, %var ; yields {i32}:result = 4 % %var
3781
3782'``srem``' Instruction
3783^^^^^^^^^^^^^^^^^^^^^^
3784
3785Syntax:
3786"""""""
3787
3788::
3789
3790 <result> = srem <ty> <op1>, <op2> ; yields {ty}:result
3791
3792Overview:
3793"""""""""
3794
3795The '``srem``' instruction returns the remainder from the signed
3796division of its two operands. This instruction can also take
3797:ref:`vector <t_vector>` versions of the values in which case the elements
3798must be integers.
3799
3800Arguments:
3801""""""""""
3802
3803The two arguments to the '``srem``' instruction must be
3804:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3805arguments must have identical types.
3806
3807Semantics:
3808""""""""""
3809
3810This instruction returns the *remainder* of a division (where the result
3811is either zero or has the same sign as the dividend, ``op1``), not the
3812*modulo* operator (where the result is either zero or has the same sign
3813as the divisor, ``op2``) of a value. For more information about the
3814difference, see `The Math
3815Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
3816table of how this is implemented in various languages, please see
3817`Wikipedia: modulo
3818operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
3819
3820Note that signed integer remainder and unsigned integer remainder are
3821distinct operations; for unsigned integer remainder, use '``urem``'.
3822
3823Taking the remainder of a division by zero leads to undefined behavior.
3824Overflow also leads to undefined behavior; this is a rare case, but can
3825occur, for example, by taking the remainder of a 32-bit division of
3826-2147483648 by -1. (The remainder doesn't actually overflow, but this
3827rule lets srem be implemented using instructions that return both the
3828result of the division and the remainder.)
3829
3830Example:
3831""""""""
3832
3833.. code-block:: llvm
3834
3835 <result> = srem i32 4, %var ; yields {i32}:result = 4 % %var
3836
3837.. _i_frem:
3838
3839'``frem``' Instruction
3840^^^^^^^^^^^^^^^^^^^^^^
3841
3842Syntax:
3843"""""""
3844
3845::
3846
3847 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields {ty}:result
3848
3849Overview:
3850"""""""""
3851
3852The '``frem``' instruction returns the remainder from the division of
3853its two operands.
3854
3855Arguments:
3856""""""""""
3857
3858The two arguments to the '``frem``' instruction must be :ref:`floating
3859point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
3860Both arguments must have identical types.
3861
3862Semantics:
3863""""""""""
3864
3865This instruction returns the *remainder* of a division. The remainder
3866has the same sign as the dividend. This instruction can also take any
3867number of :ref:`fast-math flags <fastmath>`, which are optimization hints
3868to enable otherwise unsafe floating point optimizations:
3869
3870Example:
3871""""""""
3872
3873.. code-block:: llvm
3874
3875 <result> = frem float 4.0, %var ; yields {float}:result = 4.0 % %var
3876
3877.. _bitwiseops:
3878
3879Bitwise Binary Operations
3880-------------------------
3881
3882Bitwise binary operators are used to do various forms of bit-twiddling
3883in a program. They are generally very efficient instructions and can
3884commonly be strength reduced from other instructions. They require two
3885operands of the same type, execute an operation on them, and produce a
3886single value. The resulting value is the same type as its operands.
3887
3888'``shl``' Instruction
3889^^^^^^^^^^^^^^^^^^^^^
3890
3891Syntax:
3892"""""""
3893
3894::
3895
3896 <result> = shl <ty> <op1>, <op2> ; yields {ty}:result
3897 <result> = shl nuw <ty> <op1>, <op2> ; yields {ty}:result
3898 <result> = shl nsw <ty> <op1>, <op2> ; yields {ty}:result
3899 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields {ty}:result
3900
3901Overview:
3902"""""""""
3903
3904The '``shl``' instruction returns the first operand shifted to the left
3905a specified number of bits.
3906
3907Arguments:
3908""""""""""
3909
3910Both arguments to the '``shl``' instruction must be the same
3911:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3912'``op2``' is treated as an unsigned value.
3913
3914Semantics:
3915""""""""""
3916
3917The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
3918where ``n`` is the width of the result. If ``op2`` is (statically or
3919dynamically) negative or equal to or larger than the number of bits in
3920``op1``, the result is undefined. If the arguments are vectors, each
3921vector element of ``op1`` is shifted by the corresponding shift amount
3922in ``op2``.
3923
3924If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
3925value <poisonvalues>` if it shifts out any non-zero bits. If the
3926``nsw`` keyword is present, then the shift produces a :ref:`poison
3927value <poisonvalues>` if it shifts out any bits that disagree with the
3928resultant sign bit. As such, NUW/NSW have the same semantics as they
3929would if the shift were expressed as a mul instruction with the same
3930nsw/nuw bits in (mul %op1, (shl 1, %op2)).
3931
3932Example:
3933""""""""
3934
3935.. code-block:: llvm
3936
3937 <result> = shl i32 4, %var ; yields {i32}: 4 << %var
3938 <result> = shl i32 4, 2 ; yields {i32}: 16
3939 <result> = shl i32 1, 10 ; yields {i32}: 1024
3940 <result> = shl i32 1, 32 ; undefined
3941 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
3942
3943'``lshr``' Instruction
3944^^^^^^^^^^^^^^^^^^^^^^
3945
3946Syntax:
3947"""""""
3948
3949::
3950
3951 <result> = lshr <ty> <op1>, <op2> ; yields {ty}:result
3952 <result> = lshr exact <ty> <op1>, <op2> ; yields {ty}:result
3953
3954Overview:
3955"""""""""
3956
3957The '``lshr``' instruction (logical shift right) returns the first
3958operand shifted to the right a specified number of bits with zero fill.
3959
3960Arguments:
3961""""""""""
3962
3963Both arguments to the '``lshr``' instruction must be the same
3964:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
3965'``op2``' is treated as an unsigned value.
3966
3967Semantics:
3968""""""""""
3969
3970This instruction always performs a logical shift right operation. The
3971most significant bits of the result will be filled with zero bits after
3972the shift. If ``op2`` is (statically or dynamically) equal to or larger
3973than the number of bits in ``op1``, the result is undefined. If the
3974arguments are vectors, each vector element of ``op1`` is shifted by the
3975corresponding shift amount in ``op2``.
3976
3977If the ``exact`` keyword is present, the result value of the ``lshr`` is
3978a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
3979non-zero.
3980
3981Example:
3982""""""""
3983
3984.. code-block:: llvm
3985
3986 <result> = lshr i32 4, 1 ; yields {i32}:result = 2
3987 <result> = lshr i32 4, 2 ; yields {i32}:result = 1
3988 <result> = lshr i8 4, 3 ; yields {i8}:result = 0
3989 <result> = lshr i8 -2, 1 ; yields {i8}:result = 0x7FFFFFFF
3990 <result> = lshr i32 1, 32 ; undefined
3991 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
3992
3993'``ashr``' Instruction
3994^^^^^^^^^^^^^^^^^^^^^^
3995
3996Syntax:
3997"""""""
3998
3999::
4000
4001 <result> = ashr <ty> <op1>, <op2> ; yields {ty}:result
4002 <result> = ashr exact <ty> <op1>, <op2> ; yields {ty}:result
4003
4004Overview:
4005"""""""""
4006
4007The '``ashr``' instruction (arithmetic shift right) returns the first
4008operand shifted to the right a specified number of bits with sign
4009extension.
4010
4011Arguments:
4012""""""""""
4013
4014Both arguments to the '``ashr``' instruction must be the same
4015:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4016'``op2``' is treated as an unsigned value.
4017
4018Semantics:
4019""""""""""
4020
4021This instruction always performs an arithmetic shift right operation,
4022The most significant bits of the result will be filled with the sign bit
4023of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4024than the number of bits in ``op1``, the result is undefined. If the
4025arguments are vectors, each vector element of ``op1`` is shifted by the
4026corresponding shift amount in ``op2``.
4027
4028If the ``exact`` keyword is present, the result value of the ``ashr`` is
4029a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4030non-zero.
4031
4032Example:
4033""""""""
4034
4035.. code-block:: llvm
4036
4037 <result> = ashr i32 4, 1 ; yields {i32}:result = 2
4038 <result> = ashr i32 4, 2 ; yields {i32}:result = 1
4039 <result> = ashr i8 4, 3 ; yields {i8}:result = 0
4040 <result> = ashr i8 -2, 1 ; yields {i8}:result = -1
4041 <result> = ashr i32 1, 32 ; undefined
4042 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4043
4044'``and``' Instruction
4045^^^^^^^^^^^^^^^^^^^^^
4046
4047Syntax:
4048"""""""
4049
4050::
4051
4052 <result> = and <ty> <op1>, <op2> ; yields {ty}:result
4053
4054Overview:
4055"""""""""
4056
4057The '``and``' instruction returns the bitwise logical and of its two
4058operands.
4059
4060Arguments:
4061""""""""""
4062
4063The two arguments to the '``and``' instruction must be
4064:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4065arguments must have identical types.
4066
4067Semantics:
4068""""""""""
4069
4070The truth table used for the '``and``' instruction is:
4071
4072+-----+-----+-----+
4073| In0 | In1 | Out |
4074+-----+-----+-----+
4075| 0 | 0 | 0 |
4076+-----+-----+-----+
4077| 0 | 1 | 0 |
4078+-----+-----+-----+
4079| 1 | 0 | 0 |
4080+-----+-----+-----+
4081| 1 | 1 | 1 |
4082+-----+-----+-----+
4083
4084Example:
4085""""""""
4086
4087.. code-block:: llvm
4088
4089 <result> = and i32 4, %var ; yields {i32}:result = 4 & %var
4090 <result> = and i32 15, 40 ; yields {i32}:result = 8
4091 <result> = and i32 4, 8 ; yields {i32}:result = 0
4092
4093'``or``' Instruction
4094^^^^^^^^^^^^^^^^^^^^
4095
4096Syntax:
4097"""""""
4098
4099::
4100
4101 <result> = or <ty> <op1>, <op2> ; yields {ty}:result
4102
4103Overview:
4104"""""""""
4105
4106The '``or``' instruction returns the bitwise logical inclusive or of its
4107two operands.
4108
4109Arguments:
4110""""""""""
4111
4112The two arguments to the '``or``' instruction must be
4113:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4114arguments must have identical types.
4115
4116Semantics:
4117""""""""""
4118
4119The truth table used for the '``or``' instruction is:
4120
4121+-----+-----+-----+
4122| In0 | In1 | Out |
4123+-----+-----+-----+
4124| 0 | 0 | 0 |
4125+-----+-----+-----+
4126| 0 | 1 | 1 |
4127+-----+-----+-----+
4128| 1 | 0 | 1 |
4129+-----+-----+-----+
4130| 1 | 1 | 1 |
4131+-----+-----+-----+
4132
4133Example:
4134""""""""
4135
4136::
4137
4138 <result> = or i32 4, %var ; yields {i32}:result = 4 | %var
4139 <result> = or i32 15, 40 ; yields {i32}:result = 47
4140 <result> = or i32 4, 8 ; yields {i32}:result = 12
4141
4142'``xor``' Instruction
4143^^^^^^^^^^^^^^^^^^^^^
4144
4145Syntax:
4146"""""""
4147
4148::
4149
4150 <result> = xor <ty> <op1>, <op2> ; yields {ty}:result
4151
4152Overview:
4153"""""""""
4154
4155The '``xor``' instruction returns the bitwise logical exclusive or of
4156its two operands. The ``xor`` is used to implement the "one's
4157complement" operation, which is the "~" operator in C.
4158
4159Arguments:
4160""""""""""
4161
4162The two arguments to the '``xor``' instruction must be
4163:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4164arguments must have identical types.
4165
4166Semantics:
4167""""""""""
4168
4169The truth table used for the '``xor``' instruction is:
4170
4171+-----+-----+-----+
4172| In0 | In1 | Out |
4173+-----+-----+-----+
4174| 0 | 0 | 0 |
4175+-----+-----+-----+
4176| 0 | 1 | 1 |
4177+-----+-----+-----+
4178| 1 | 0 | 1 |
4179+-----+-----+-----+
4180| 1 | 1 | 0 |
4181+-----+-----+-----+
4182
4183Example:
4184""""""""
4185
4186.. code-block:: llvm
4187
4188 <result> = xor i32 4, %var ; yields {i32}:result = 4 ^ %var
4189 <result> = xor i32 15, 40 ; yields {i32}:result = 39
4190 <result> = xor i32 4, 8 ; yields {i32}:result = 12
4191 <result> = xor i32 %V, -1 ; yields {i32}:result = ~%V
4192
4193Vector Operations
4194-----------------
4195
4196LLVM supports several instructions to represent vector operations in a
4197target-independent manner. These instructions cover the element-access
4198and vector-specific operations needed to process vectors effectively.
4199While LLVM does directly support these vector operations, many
4200sophisticated algorithms will want to use target-specific intrinsics to
4201take full advantage of a specific target.
4202
4203.. _i_extractelement:
4204
4205'``extractelement``' Instruction
4206^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4207
4208Syntax:
4209"""""""
4210
4211::
4212
4213 <result> = extractelement <n x <ty>> <val>, i32 <idx> ; yields <ty>
4214
4215Overview:
4216"""""""""
4217
4218The '``extractelement``' instruction extracts a single scalar element
4219from a vector at a specified index.
4220
4221Arguments:
4222""""""""""
4223
4224The first operand of an '``extractelement``' instruction is a value of
4225:ref:`vector <t_vector>` type. The second operand is an index indicating
4226the position from which to extract the element. The index may be a
4227variable.
4228
4229Semantics:
4230""""""""""
4231
4232The result is a scalar of the same type as the element type of ``val``.
4233Its value is the value at position ``idx`` of ``val``. If ``idx``
4234exceeds the length of ``val``, the results are undefined.
4235
4236Example:
4237""""""""
4238
4239.. code-block:: llvm
4240
4241 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4242
4243.. _i_insertelement:
4244
4245'``insertelement``' Instruction
4246^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4247
4248Syntax:
4249"""""""
4250
4251::
4252
4253 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> ; yields <n x <ty>>
4254
4255Overview:
4256"""""""""
4257
4258The '``insertelement``' instruction inserts a scalar element into a
4259vector at a specified index.
4260
4261Arguments:
4262""""""""""
4263
4264The first operand of an '``insertelement``' instruction is a value of
4265:ref:`vector <t_vector>` type. The second operand is a scalar value whose
4266type must equal the element type of the first operand. The third operand
4267is an index indicating the position at which to insert the value. The
4268index may be a variable.
4269
4270Semantics:
4271""""""""""
4272
4273The result is a vector of the same type as ``val``. Its element values
4274are those of ``val`` except at position ``idx``, where it gets the value
4275``elt``. If ``idx`` exceeds the length of ``val``, the results are
4276undefined.
4277
4278Example:
4279""""""""
4280
4281.. code-block:: llvm
4282
4283 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4284
4285.. _i_shufflevector:
4286
4287'``shufflevector``' Instruction
4288^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4289
4290Syntax:
4291"""""""
4292
4293::
4294
4295 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4296
4297Overview:
4298"""""""""
4299
4300The '``shufflevector``' instruction constructs a permutation of elements
4301from two input vectors, returning a vector with the same element type as
4302the input and length that is the same as the shuffle mask.
4303
4304Arguments:
4305""""""""""
4306
4307The first two operands of a '``shufflevector``' instruction are vectors
4308with the same type. The third argument is a shuffle mask whose element
4309type is always 'i32'. The result of the instruction is a vector whose
4310length is the same as the shuffle mask and whose element type is the
4311same as the element type of the first two operands.
4312
4313The shuffle mask operand is required to be a constant vector with either
4314constant integer or undef values.
4315
4316Semantics:
4317""""""""""
4318
4319The elements of the two input vectors are numbered from left to right
4320across both of the vectors. The shuffle mask operand specifies, for each
4321element of the result vector, which element of the two input vectors the
4322result element gets. The element selector may be undef (meaning "don't
4323care") and the second operand may be undef if performing a shuffle from
4324only one vector.
4325
4326Example:
4327""""""""
4328
4329.. code-block:: llvm
4330
4331 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4332 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4333 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4334 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4335 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4336 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4337 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4338 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4339
4340Aggregate Operations
4341--------------------
4342
4343LLVM supports several instructions for working with
4344:ref:`aggregate <t_aggregate>` values.
4345
4346.. _i_extractvalue:
4347
4348'``extractvalue``' Instruction
4349^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4350
4351Syntax:
4352"""""""
4353
4354::
4355
4356 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4357
4358Overview:
4359"""""""""
4360
4361The '``extractvalue``' instruction extracts the value of a member field
4362from an :ref:`aggregate <t_aggregate>` value.
4363
4364Arguments:
4365""""""""""
4366
4367The first operand of an '``extractvalue``' instruction is a value of
4368:ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
4369constant indices to specify which value to extract in a similar manner
4370as indices in a '``getelementptr``' instruction.
4371
4372The major differences to ``getelementptr`` indexing are:
4373
4374- Since the value being indexed is not a pointer, the first index is
4375 omitted and assumed to be zero.
4376- At least one index must be specified.
4377- Not only struct indices but also array indices must be in bounds.
4378
4379Semantics:
4380""""""""""
4381
4382The result is the value at the position in the aggregate specified by
4383the index operands.
4384
4385Example:
4386""""""""
4387
4388.. code-block:: llvm
4389
4390 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
4391
4392.. _i_insertvalue:
4393
4394'``insertvalue``' Instruction
4395^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4396
4397Syntax:
4398"""""""
4399
4400::
4401
4402 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
4403
4404Overview:
4405"""""""""
4406
4407The '``insertvalue``' instruction inserts a value into a member field in
4408an :ref:`aggregate <t_aggregate>` value.
4409
4410Arguments:
4411""""""""""
4412
4413The first operand of an '``insertvalue``' instruction is a value of
4414:ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
4415a first-class value to insert. The following operands are constant
4416indices indicating the position at which to insert the value in a
4417similar manner as indices in a '``extractvalue``' instruction. The value
4418to insert must have the same type as the value identified by the
4419indices.
4420
4421Semantics:
4422""""""""""
4423
4424The result is an aggregate of the same type as ``val``. Its value is
4425that of ``val`` except that the value at the position specified by the
4426indices is that of ``elt``.
4427
4428Example:
4429""""""""
4430
4431.. code-block:: llvm
4432
4433 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
4434 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
4435 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
4436
4437.. _memoryops:
4438
4439Memory Access and Addressing Operations
4440---------------------------------------
4441
4442A key design point of an SSA-based representation is how it represents
4443memory. In LLVM, no memory locations are in SSA form, which makes things
4444very simple. This section describes how to read, write, and allocate
4445memory in LLVM.
4446
4447.. _i_alloca:
4448
4449'``alloca``' Instruction
4450^^^^^^^^^^^^^^^^^^^^^^^^
4451
4452Syntax:
4453"""""""
4454
4455::
4456
4457 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] ; yields {type*}:result
4458
4459Overview:
4460"""""""""
4461
4462The '``alloca``' instruction allocates memory on the stack frame of the
4463currently executing function, to be automatically released when this
4464function returns to its caller. The object is always allocated in the
4465generic address space (address space zero).
4466
4467Arguments:
4468""""""""""
4469
4470The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
4471bytes of memory on the runtime stack, returning a pointer of the
4472appropriate type to the program. If "NumElements" is specified, it is
4473the number of elements allocated, otherwise "NumElements" is defaulted
4474to be one. If a constant alignment is specified, the value result of the
4475allocation is guaranteed to be aligned to at least that boundary. If not
4476specified, or if zero, the target can choose to align the allocation on
4477any convenient boundary compatible with the type.
4478
4479'``type``' may be any sized type.
4480
4481Semantics:
4482""""""""""
4483
4484Memory is allocated; a pointer is returned. The operation is undefined
4485if there is insufficient stack space for the allocation. '``alloca``'d
4486memory is automatically released when the function returns. The
4487'``alloca``' instruction is commonly used to represent automatic
4488variables that must have an address available. When the function returns
4489(either with the ``ret`` or ``resume`` instructions), the memory is
4490reclaimed. Allocating zero bytes is legal, but the result is undefined.
4491The order in which memory is allocated (ie., which way the stack grows)
4492is not specified.
4493
4494Example:
4495""""""""
4496
4497.. code-block:: llvm
4498
4499 %ptr = alloca i32 ; yields {i32*}:ptr
4500 %ptr = alloca i32, i32 4 ; yields {i32*}:ptr
4501 %ptr = alloca i32, i32 4, align 1024 ; yields {i32*}:ptr
4502 %ptr = alloca i32, align 1024 ; yields {i32*}:ptr
4503
4504.. _i_load:
4505
4506'``load``' Instruction
4507^^^^^^^^^^^^^^^^^^^^^^
4508
4509Syntax:
4510"""""""
4511
4512::
4513
4514 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4515 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4516 !<index> = !{ i32 1 }
4517
4518Overview:
4519"""""""""
4520
4521The '``load``' instruction is used to read from memory.
4522
4523Arguments:
4524""""""""""
4525
4526The argument to the '``load``' instruction specifies the memory address
4527from which to load. The pointer must point to a :ref:`first
4528class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
4529then the optimizer is not allowed to modify the number or order of
4530execution of this ``load`` with other :ref:`volatile
4531operations <volatile>`.
4532
4533If the ``load`` is marked as ``atomic``, it takes an extra
4534:ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4535``release`` and ``acq_rel`` orderings are not valid on ``load``
4536instructions. Atomic loads produce :ref:`defined <memmodel>` results
4537when they may see multiple atomic stores. The type of the pointee must
4538be an integer type whose bit width is a power of two greater than or
4539equal to eight and less than or equal to a target-specific size limit.
4540``align`` must be explicitly specified on atomic loads, and the load has
4541undefined behavior if the alignment is not set to a value which is at
4542least the size in bytes of the pointee. ``!nontemporal`` does not have
4543any defined semantics for atomic loads.
4544
4545The optional constant ``align`` argument specifies the alignment of the
4546operation (that is, the alignment of the memory address). A value of 0
4547or an omitted ``align`` argument means that the operation has the abi
4548alignment for the target. It is the responsibility of the code emitter
4549to ensure that the alignment information is correct. Overestimating the
4550alignment results in undefined behavior. Underestimating the alignment
4551may produce less efficient code. An alignment of 1 is always safe.
4552
4553The optional ``!nontemporal`` metadata must reference a single
4554metatadata name <index> corresponding to a metadata node with one
4555``i32`` entry of value 1. The existence of the ``!nontemporal``
4556metatadata on the instruction tells the optimizer and code generator
4557that this load is not expected to be reused in the cache. The code
4558generator may select special instructions to save cache bandwidth, such
4559as the ``MOVNT`` instruction on x86.
4560
4561The optional ``!invariant.load`` metadata must reference a single
4562metatadata name <index> corresponding to a metadata node with no
4563entries. The existence of the ``!invariant.load`` metatadata on the
4564instruction tells the optimizer and code generator that this load
4565address points to memory which does not change value during program
4566execution. The optimizer may then move this load around, for example, by
4567hoisting it out of loops using loop invariant code motion.
4568
4569Semantics:
4570""""""""""
4571
4572The location of memory pointed to is loaded. If the value being loaded
4573is of scalar type then the number of bytes read does not exceed the
4574minimum number of bytes needed to hold all bits of the type. For
4575example, loading an ``i24`` reads at most three bytes. When loading a
4576value of a type like ``i20`` with a size that is not an integral number
4577of bytes, the result is undefined if the value was not originally
4578written using a store of the same type.
4579
4580Examples:
4581"""""""""
4582
4583.. code-block:: llvm
4584
4585 %ptr = alloca i32 ; yields {i32*}:ptr
4586 store i32 3, i32* %ptr ; yields {void}
4587 %val = load i32* %ptr ; yields {i32}:val = i32 3
4588
4589.. _i_store:
4590
4591'``store``' Instruction
4592^^^^^^^^^^^^^^^^^^^^^^^
4593
4594Syntax:
4595"""""""
4596
4597::
4598
4599 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields {void}
4600 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields {void}
4601
4602Overview:
4603"""""""""
4604
4605The '``store``' instruction is used to write to memory.
4606
4607Arguments:
4608""""""""""
4609
4610There are two arguments to the '``store``' instruction: a value to store
4611and an address at which to store it. The type of the '``<pointer>``'
4612operand must be a pointer to the :ref:`first class <t_firstclass>` type of
4613the '``<value>``' operand. If the ``store`` is marked as ``volatile``,
4614then the optimizer is not allowed to modify the number or order of
4615execution of this ``store`` with other :ref:`volatile
4616operations <volatile>`.
4617
4618If the ``store`` is marked as ``atomic``, it takes an extra
4619:ref:`ordering <ordering>` and optional ``singlethread`` argument. The
4620``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
4621instructions. Atomic loads produce :ref:`defined <memmodel>` results
4622when they may see multiple atomic stores. The type of the pointee must
4623be an integer type whose bit width is a power of two greater than or
4624equal to eight and less than or equal to a target-specific size limit.
4625``align`` must be explicitly specified on atomic stores, and the store
4626has undefined behavior if the alignment is not set to a value which is
4627at least the size in bytes of the pointee. ``!nontemporal`` does not
4628have any defined semantics for atomic stores.
4629
4630The optional constant "align" argument specifies the alignment of the
4631operation (that is, the alignment of the memory address). A value of 0
4632or an omitted "align" argument means that the operation has the abi
4633alignment for the target. It is the responsibility of the code emitter
4634to ensure that the alignment information is correct. Overestimating the
4635alignment results in an undefined behavior. Underestimating the
4636alignment may produce less efficient code. An alignment of 1 is always
4637safe.
4638
4639The optional !nontemporal metadata must reference a single metatadata
4640name <index> corresponding to a metadata node with one i32 entry of
4641value 1. The existence of the !nontemporal metatadata on the instruction
4642tells the optimizer and code generator that this load is not expected to
4643be reused in the cache. The code generator may select special
4644instructions to save cache bandwidth, such as the MOVNT instruction on
4645x86.
4646
4647Semantics:
4648""""""""""
4649
4650The contents of memory are updated to contain '``<value>``' at the
4651location specified by the '``<pointer>``' operand. If '``<value>``' is
4652of scalar type then the number of bytes written does not exceed the
4653minimum number of bytes needed to hold all bits of the type. For
4654example, storing an ``i24`` writes at most three bytes. When writing a
4655value of a type like ``i20`` with a size that is not an integral number
4656of bytes, it is unspecified what happens to the extra bits that do not
4657belong to the type, but they will typically be overwritten.
4658
4659Example:
4660""""""""
4661
4662.. code-block:: llvm
4663
4664 %ptr = alloca i32 ; yields {i32*}:ptr
4665 store i32 3, i32* %ptr ; yields {void}
4666 %val = load i32* %ptr ; yields {i32}:val = i32 3
4667
4668.. _i_fence:
4669
4670'``fence``' Instruction
4671^^^^^^^^^^^^^^^^^^^^^^^
4672
4673Syntax:
4674"""""""
4675
4676::
4677
4678 fence [singlethread] <ordering> ; yields {void}
4679
4680Overview:
4681"""""""""
4682
4683The '``fence``' instruction is used to introduce happens-before edges
4684between operations.
4685
4686Arguments:
4687""""""""""
4688
4689'``fence``' instructions take an :ref:`ordering <ordering>` argument which
4690defines what *synchronizes-with* edges they add. They can only be given
4691``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
4692
4693Semantics:
4694""""""""""
4695
4696A fence A which has (at least) ``release`` ordering semantics
4697*synchronizes with* a fence B with (at least) ``acquire`` ordering
4698semantics if and only if there exist atomic operations X and Y, both
4699operating on some atomic object M, such that A is sequenced before X, X
4700modifies M (either directly or through some side effect of a sequence
4701headed by X), Y is sequenced before B, and Y observes M. This provides a
4702*happens-before* dependency between A and B. Rather than an explicit
4703``fence``, one (but not both) of the atomic operations X or Y might
4704provide a ``release`` or ``acquire`` (resp.) ordering constraint and
4705still *synchronize-with* the explicit ``fence`` and establish the
4706*happens-before* edge.
4707
4708A ``fence`` which has ``seq_cst`` ordering, in addition to having both
4709``acquire`` and ``release`` semantics specified above, participates in
4710the global program order of other ``seq_cst`` operations and/or fences.
4711
4712The optional ":ref:`singlethread <singlethread>`" argument specifies
4713that the fence only synchronizes with other fences in the same thread.
4714(This is useful for interacting with signal handlers.)
4715
4716Example:
4717""""""""
4718
4719.. code-block:: llvm
4720
4721 fence acquire ; yields {void}
4722 fence singlethread seq_cst ; yields {void}
4723
4724.. _i_cmpxchg:
4725
4726'``cmpxchg``' Instruction
4727^^^^^^^^^^^^^^^^^^^^^^^^^
4728
4729Syntax:
4730"""""""
4731
4732::
4733
4734 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> ; yields {ty}
4735
4736Overview:
4737"""""""""
4738
4739The '``cmpxchg``' instruction is used to atomically modify memory. It
4740loads a value in memory and compares it to a given value. If they are
4741equal, it stores a new value into the memory.
4742
4743Arguments:
4744""""""""""
4745
4746There are three arguments to the '``cmpxchg``' instruction: an address
4747to operate on, a value to compare to the value currently be at that
4748address, and a new value to place at that address if the compared values
4749are equal. The type of '<cmp>' must be an integer type whose bit width
4750is a power of two greater than or equal to eight and less than or equal
4751to a target-specific size limit. '<cmp>' and '<new>' must have the same
4752type, and the type of '<pointer>' must be a pointer to that type. If the
4753``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
4754to modify the number or order of execution of this ``cmpxchg`` with
4755other :ref:`volatile operations <volatile>`.
4756
4757The :ref:`ordering <ordering>` argument specifies how this ``cmpxchg``
4758synchronizes with other atomic operations.
4759
4760The optional "``singlethread``" argument declares that the ``cmpxchg``
4761is only atomic with respect to code (usually signal handlers) running in
4762the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
4763respect to all other code in the system.
4764
4765The pointer passed into cmpxchg must have alignment greater than or
4766equal to the size in memory of the operand.
4767
4768Semantics:
4769""""""""""
4770
4771The contents of memory at the location specified by the '``<pointer>``'
4772operand is read and compared to '``<cmp>``'; if the read value is the
4773equal, '``<new>``' is written. The original value at the location is
4774returned.
4775
4776A successful ``cmpxchg`` is a read-modify-write instruction for the purpose
4777of identifying release sequences. A failed ``cmpxchg`` is equivalent to an
4778atomic load with an ordering parameter determined by dropping any
4779``release`` part of the ``cmpxchg``'s ordering.
4780
4781Example:
4782""""""""
4783
4784.. code-block:: llvm
4785
4786 entry:
4787 %orig = atomic load i32* %ptr unordered ; yields {i32}
4788 br label %loop
4789
4790 loop:
4791 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
4792 %squared = mul i32 %cmp, %cmp
4793 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared ; yields {i32}
4794 %success = icmp eq i32 %cmp, %old
4795 br i1 %success, label %done, label %loop
4796
4797 done:
4798 ...
4799
4800.. _i_atomicrmw:
4801
4802'``atomicrmw``' Instruction
4803^^^^^^^^^^^^^^^^^^^^^^^^^^^
4804
4805Syntax:
4806"""""""
4807
4808::
4809
4810 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields {ty}
4811
4812Overview:
4813"""""""""
4814
4815The '``atomicrmw``' instruction is used to atomically modify memory.
4816
4817Arguments:
4818""""""""""
4819
4820There are three arguments to the '``atomicrmw``' instruction: an
4821operation to apply, an address whose value to modify, an argument to the
4822operation. The operation must be one of the following keywords:
4823
4824- xchg
4825- add
4826- sub
4827- and
4828- nand
4829- or
4830- xor
4831- max
4832- min
4833- umax
4834- umin
4835
4836The type of '<value>' must be an integer type whose bit width is a power
4837of two greater than or equal to eight and less than or equal to a
4838target-specific size limit. The type of the '``<pointer>``' operand must
4839be a pointer to that type. If the ``atomicrmw`` is marked as
4840``volatile``, then the optimizer is not allowed to modify the number or
4841order of execution of this ``atomicrmw`` with other :ref:`volatile
4842operations <volatile>`.
4843
4844Semantics:
4845""""""""""
4846
4847The contents of memory at the location specified by the '``<pointer>``'
4848operand are atomically read, modified, and written back. The original
4849value at the location is returned. The modification is specified by the
4850operation argument:
4851
4852- xchg: ``*ptr = val``
4853- add: ``*ptr = *ptr + val``
4854- sub: ``*ptr = *ptr - val``
4855- and: ``*ptr = *ptr & val``
4856- nand: ``*ptr = ~(*ptr & val)``
4857- or: ``*ptr = *ptr | val``
4858- xor: ``*ptr = *ptr ^ val``
4859- max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
4860- min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
4861- umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
4862 comparison)
4863- umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
4864 comparison)
4865
4866Example:
4867""""""""
4868
4869.. code-block:: llvm
4870
4871 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields {i32}
4872
4873.. _i_getelementptr:
4874
4875'``getelementptr``' Instruction
4876^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4877
4878Syntax:
4879"""""""
4880
4881::
4882
4883 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4884 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4885 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
4886
4887Overview:
4888"""""""""
4889
4890The '``getelementptr``' instruction is used to get the address of a
4891subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
4892address calculation only and does not access memory.
4893
4894Arguments:
4895""""""""""
4896
4897The first argument is always a pointer or a vector of pointers, and
4898forms the basis of the calculation. The remaining arguments are indices
4899that indicate which of the elements of the aggregate object are indexed.
4900The interpretation of each index is dependent on the type being indexed
4901into. The first index always indexes the pointer value given as the
4902first argument, the second index indexes a value of the type pointed to
4903(not necessarily the value directly pointed to, since the first index
4904can be non-zero), etc. The first type indexed into must be a pointer
4905value, subsequent types can be arrays, vectors, and structs. Note that
4906subsequent types being indexed into can never be pointers, since that
4907would require loading the pointer before continuing calculation.
4908
4909The type of each index argument depends on the type it is indexing into.
4910When indexing into a (optionally packed) structure, only ``i32`` integer
4911**constants** are allowed (when using a vector of indices they must all
4912be the **same** ``i32`` integer constant). When indexing into an array,
4913pointer or vector, integers of any width are allowed, and they are not
4914required to be constant. These integers are treated as signed values
4915where relevant.
4916
4917For example, let's consider a C code fragment and how it gets compiled
4918to LLVM:
4919
4920.. code-block:: c
4921
4922 struct RT {
4923 char A;
4924 int B[10][20];
4925 char C;
4926 };
4927 struct ST {
4928 int X;
4929 double Y;
4930 struct RT Z;
4931 };
4932
4933 int *foo(struct ST *s) {
4934 return &s[1].Z.B[5][13];
4935 }
4936
4937The LLVM code generated by Clang is:
4938
4939.. code-block:: llvm
4940
4941 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
4942 %struct.ST = type { i32, double, %struct.RT }
4943
4944 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
4945 entry:
4946 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
4947 ret i32* %arrayidx
4948 }
4949
4950Semantics:
4951""""""""""
4952
4953In the example above, the first index is indexing into the
4954'``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
4955= '``{ i32, double, %struct.RT }``' type, a structure. The second index
4956indexes into the third element of the structure, yielding a
4957'``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
4958structure. The third index indexes into the second element of the
4959structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
4960dimensions of the array are subscripted into, yielding an '``i32``'
4961type. The '``getelementptr``' instruction returns a pointer to this
4962element, thus computing a value of '``i32*``' type.
4963
4964Note that it is perfectly legal to index partially through a structure,
4965returning a pointer to an inner element. Because of this, the LLVM code
4966for the given testcase is equivalent to:
4967
4968.. code-block:: llvm
4969
4970 define i32* @foo(%struct.ST* %s) {
4971 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
4972 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
4973 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
4974 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
4975 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
4976 ret i32* %t5
4977 }
4978
4979If the ``inbounds`` keyword is present, the result value of the
4980``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
4981pointer is not an *in bounds* address of an allocated object, or if any
4982of the addresses that would be formed by successive addition of the
4983offsets implied by the indices to the base address with infinitely
4984precise signed arithmetic are not an *in bounds* address of that
4985allocated object. The *in bounds* addresses for an allocated object are
4986all the addresses that point into the object, plus the address one byte
4987past the end. In cases where the base is a vector of pointers the
4988``inbounds`` keyword applies to each of the computations element-wise.
4989
4990If the ``inbounds`` keyword is not present, the offsets are added to the
4991base address with silently-wrapping two's complement arithmetic. If the
4992offsets have a different width from the pointer, they are sign-extended
4993or truncated to the width of the pointer. The result value of the
4994``getelementptr`` may be outside the object pointed to by the base
4995pointer. The result value may not necessarily be used to access memory
4996though, even if it happens to point into allocated storage. See the
4997:ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
4998information.
4999
5000The getelementptr instruction is often confusing. For some more insight
5001into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5002
5003Example:
5004""""""""
5005
5006.. code-block:: llvm
5007
5008 ; yields [12 x i8]*:aptr
5009 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5010 ; yields i8*:vptr
5011 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5012 ; yields i8*:eptr
5013 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5014 ; yields i32*:iptr
5015 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5016
5017In cases where the pointer argument is a vector of pointers, each index
5018must be a vector with the same number of elements. For example:
5019
5020.. code-block:: llvm
5021
5022 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5023
5024Conversion Operations
5025---------------------
5026
5027The instructions in this category are the conversion instructions
5028(casting) which all take a single operand and a type. They perform
5029various bit conversions on the operand.
5030
5031'``trunc .. to``' Instruction
5032^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5033
5034Syntax:
5035"""""""
5036
5037::
5038
5039 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5040
5041Overview:
5042"""""""""
5043
5044The '``trunc``' instruction truncates its operand to the type ``ty2``.
5045
5046Arguments:
5047""""""""""
5048
5049The '``trunc``' instruction takes a value to trunc, and a type to trunc
5050it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5051of the same number of integers. The bit size of the ``value`` must be
5052larger than the bit size of the destination type, ``ty2``. Equal sized
5053types are not allowed.
5054
5055Semantics:
5056""""""""""
5057
5058The '``trunc``' instruction truncates the high order bits in ``value``
5059and converts the remaining bits to ``ty2``. Since the source size must
5060be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5061It will always truncate bits.
5062
5063Example:
5064""""""""
5065
5066.. code-block:: llvm
5067
5068 %X = trunc i32 257 to i8 ; yields i8:1
5069 %Y = trunc i32 123 to i1 ; yields i1:true
5070 %Z = trunc i32 122 to i1 ; yields i1:false
5071 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5072
5073'``zext .. to``' Instruction
5074^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5075
5076Syntax:
5077"""""""
5078
5079::
5080
5081 <result> = zext <ty> <value> to <ty2> ; yields ty2
5082
5083Overview:
5084"""""""""
5085
5086The '``zext``' instruction zero extends its operand to type ``ty2``.
5087
5088Arguments:
5089""""""""""
5090
5091The '``zext``' instruction takes a value to cast, and a type to cast it
5092to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5093the same number of integers. The bit size of the ``value`` must be
5094smaller than the bit size of the destination type, ``ty2``.
5095
5096Semantics:
5097""""""""""
5098
5099The ``zext`` fills the high order bits of the ``value`` with zero bits
5100until it reaches the size of the destination type, ``ty2``.
5101
5102When zero extending from i1, the result will always be either 0 or 1.
5103
5104Example:
5105""""""""
5106
5107.. code-block:: llvm
5108
5109 %X = zext i32 257 to i64 ; yields i64:257
5110 %Y = zext i1 true to i32 ; yields i32:1
5111 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5112
5113'``sext .. to``' Instruction
5114^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5115
5116Syntax:
5117"""""""
5118
5119::
5120
5121 <result> = sext <ty> <value> to <ty2> ; yields ty2
5122
5123Overview:
5124"""""""""
5125
5126The '``sext``' sign extends ``value`` to the type ``ty2``.
5127
5128Arguments:
5129""""""""""
5130
5131The '``sext``' instruction takes a value to cast, and a type to cast it
5132to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5133the same number of integers. The bit size of the ``value`` must be
5134smaller than the bit size of the destination type, ``ty2``.
5135
5136Semantics:
5137""""""""""
5138
5139The '``sext``' instruction performs a sign extension by copying the sign
5140bit (highest order bit) of the ``value`` until it reaches the bit size
5141of the type ``ty2``.
5142
5143When sign extending from i1, the extension always results in -1 or 0.
5144
5145Example:
5146""""""""
5147
5148.. code-block:: llvm
5149
5150 %X = sext i8 -1 to i16 ; yields i16 :65535
5151 %Y = sext i1 true to i32 ; yields i32:-1
5152 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5153
5154'``fptrunc .. to``' Instruction
5155^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5156
5157Syntax:
5158"""""""
5159
5160::
5161
5162 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5163
5164Overview:
5165"""""""""
5166
5167The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5168
5169Arguments:
5170""""""""""
5171
5172The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5173value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5174The size of ``value`` must be larger than the size of ``ty2``. This
5175implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5176
5177Semantics:
5178""""""""""
5179
5180The '``fptrunc``' instruction truncates a ``value`` from a larger
5181:ref:`floating point <t_floating>` type to a smaller :ref:`floating
5182point <t_floating>` type. If the value cannot fit within the
5183destination type, ``ty2``, then the results are undefined.
5184
5185Example:
5186""""""""
5187
5188.. code-block:: llvm
5189
5190 %X = fptrunc double 123.0 to float ; yields float:123.0
5191 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5192
5193'``fpext .. to``' Instruction
5194^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5195
5196Syntax:
5197"""""""
5198
5199::
5200
5201 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5202
5203Overview:
5204"""""""""
5205
5206The '``fpext``' extends a floating point ``value`` to a larger floating
5207point value.
5208
5209Arguments:
5210""""""""""
5211
5212The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5213``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5214to. The source type must be smaller than the destination type.
5215
5216Semantics:
5217""""""""""
5218
5219The '``fpext``' instruction extends the ``value`` from a smaller
5220:ref:`floating point <t_floating>` type to a larger :ref:`floating
5221point <t_floating>` type. The ``fpext`` cannot be used to make a
5222*no-op cast* because it always changes bits. Use ``bitcast`` to make a
5223*no-op cast* for a floating point cast.
5224
5225Example:
5226""""""""
5227
5228.. code-block:: llvm
5229
5230 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5231 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5232
5233'``fptoui .. to``' Instruction
5234^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5235
5236Syntax:
5237"""""""
5238
5239::
5240
5241 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5242
5243Overview:
5244"""""""""
5245
5246The '``fptoui``' converts a floating point ``value`` to its unsigned
5247integer equivalent of type ``ty2``.
5248
5249Arguments:
5250""""""""""
5251
5252The '``fptoui``' instruction takes a value to cast, which must be a
5253scalar or vector :ref:`floating point <t_floating>` value, and a type to
5254cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5255``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5256type with the same number of elements as ``ty``
5257
5258Semantics:
5259""""""""""
5260
5261The '``fptoui``' instruction converts its :ref:`floating
5262point <t_floating>` operand into the nearest (rounding towards zero)
5263unsigned integer value. If the value cannot fit in ``ty2``, the results
5264are undefined.
5265
5266Example:
5267""""""""
5268
5269.. code-block:: llvm
5270
5271 %X = fptoui double 123.0 to i32 ; yields i32:123
5272 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5273 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5274
5275'``fptosi .. to``' Instruction
5276^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5277
5278Syntax:
5279"""""""
5280
5281::
5282
5283 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5284
5285Overview:
5286"""""""""
5287
5288The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5289``value`` to type ``ty2``.
5290
5291Arguments:
5292""""""""""
5293
5294The '``fptosi``' instruction takes a value to cast, which must be a
5295scalar or vector :ref:`floating point <t_floating>` value, and a type to
5296cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5297``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5298type with the same number of elements as ``ty``
5299
5300Semantics:
5301""""""""""
5302
5303The '``fptosi``' instruction converts its :ref:`floating
5304point <t_floating>` operand into the nearest (rounding towards zero)
5305signed integer value. If the value cannot fit in ``ty2``, the results
5306are undefined.
5307
5308Example:
5309""""""""
5310
5311.. code-block:: llvm
5312
5313 %X = fptosi double -123.0 to i32 ; yields i32:-123
5314 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5315 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5316
5317'``uitofp .. to``' Instruction
5318^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5319
5320Syntax:
5321"""""""
5322
5323::
5324
5325 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5326
5327Overview:
5328"""""""""
5329
5330The '``uitofp``' instruction regards ``value`` as an unsigned integer
5331and converts that value to the ``ty2`` type.
5332
5333Arguments:
5334""""""""""
5335
5336The '``uitofp``' instruction takes a value to cast, which must be a
5337scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5338``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5339``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5340type with the same number of elements as ``ty``
5341
5342Semantics:
5343""""""""""
5344
5345The '``uitofp``' instruction interprets its operand as an unsigned
5346integer quantity and converts it to the corresponding floating point
5347value. If the value cannot fit in the floating point value, the results
5348are undefined.
5349
5350Example:
5351""""""""
5352
5353.. code-block:: llvm
5354
5355 %X = uitofp i32 257 to float ; yields float:257.0
5356 %Y = uitofp i8 -1 to double ; yields double:255.0
5357
5358'``sitofp .. to``' Instruction
5359^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5360
5361Syntax:
5362"""""""
5363
5364::
5365
5366 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
5367
5368Overview:
5369"""""""""
5370
5371The '``sitofp``' instruction regards ``value`` as a signed integer and
5372converts that value to the ``ty2`` type.
5373
5374Arguments:
5375""""""""""
5376
5377The '``sitofp``' instruction takes a value to cast, which must be a
5378scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5379``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5380``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5381type with the same number of elements as ``ty``
5382
5383Semantics:
5384""""""""""
5385
5386The '``sitofp``' instruction interprets its operand as a signed integer
5387quantity and converts it to the corresponding floating point value. If
5388the value cannot fit in the floating point value, the results are
5389undefined.
5390
5391Example:
5392""""""""
5393
5394.. code-block:: llvm
5395
5396 %X = sitofp i32 257 to float ; yields float:257.0
5397 %Y = sitofp i8 -1 to double ; yields double:-1.0
5398
5399.. _i_ptrtoint:
5400
5401'``ptrtoint .. to``' Instruction
5402^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5403
5404Syntax:
5405"""""""
5406
5407::
5408
5409 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
5410
5411Overview:
5412"""""""""
5413
5414The '``ptrtoint``' instruction converts the pointer or a vector of
5415pointers ``value`` to the integer (or vector of integers) type ``ty2``.
5416
5417Arguments:
5418""""""""""
5419
5420The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
5421a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
5422type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
5423a vector of integers type.
5424
5425Semantics:
5426""""""""""
5427
5428The '``ptrtoint``' instruction converts ``value`` to integer type
5429``ty2`` by interpreting the pointer value as an integer and either
5430truncating or zero extending that value to the size of the integer type.
5431If ``value`` is smaller than ``ty2`` then a zero extension is done. If
5432``value`` is larger than ``ty2`` then a truncation is done. If they are
5433the same size, then nothing is done (*no-op cast*) other than a type
5434change.
5435
5436Example:
5437""""""""
5438
5439.. code-block:: llvm
5440
5441 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
5442 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
5443 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
5444
5445.. _i_inttoptr:
5446
5447'``inttoptr .. to``' Instruction
5448^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5449
5450Syntax:
5451"""""""
5452
5453::
5454
5455 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
5456
5457Overview:
5458"""""""""
5459
5460The '``inttoptr``' instruction converts an integer ``value`` to a
5461pointer type, ``ty2``.
5462
5463Arguments:
5464""""""""""
5465
5466The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
5467cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
5468type.
5469
5470Semantics:
5471""""""""""
5472
5473The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
5474applying either a zero extension or a truncation depending on the size
5475of the integer ``value``. If ``value`` is larger than the size of a
5476pointer then a truncation is done. If ``value`` is smaller than the size
5477of a pointer then a zero extension is done. If they are the same size,
5478nothing is done (*no-op cast*).
5479
5480Example:
5481""""""""
5482
5483.. code-block:: llvm
5484
5485 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
5486 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
5487 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
5488 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
5489
5490.. _i_bitcast:
5491
5492'``bitcast .. to``' Instruction
5493^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5494
5495Syntax:
5496"""""""
5497
5498::
5499
5500 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
5501
5502Overview:
5503"""""""""
5504
5505The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
5506changing any bits.
5507
5508Arguments:
5509""""""""""
5510
5511The '``bitcast``' instruction takes a value to cast, which must be a
5512non-aggregate first class value, and a type to cast it to, which must
5513also be a non-aggregate :ref:`first class <t_firstclass>` type. The bit
5514sizes of ``value`` and the destination type, ``ty2``, must be identical.
5515If the source type is a pointer, the destination type must also be a
5516pointer. This instruction supports bitwise conversion of vectors to
5517integers and to vectors of other types (as long as they have the same
5518size).
5519
5520Semantics:
5521""""""""""
5522
5523The '``bitcast``' instruction converts ``value`` to type ``ty2``. It is
5524always a *no-op cast* because no bits change with this conversion. The
5525conversion is done as if the ``value`` had been stored to memory and
5526read back as type ``ty2``. Pointer (or vector of pointers) types may
5527only be converted to other pointer (or vector of pointers) types with
5528this instruction. To convert pointers to other types, use the
5529:ref:`inttoptr <i_inttoptr>` or :ref:`ptrtoint <i_ptrtoint>` instructions
5530first.
5531
5532Example:
5533""""""""
5534
5535.. code-block:: llvm
5536
5537 %X = bitcast i8 255 to i8 ; yields i8 :-1
5538 %Y = bitcast i32* %x to sint* ; yields sint*:%x
5539 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
5540 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
5541
5542.. _otherops:
5543
5544Other Operations
5545----------------
5546
5547The instructions in this category are the "miscellaneous" instructions,
5548which defy better classification.
5549
5550.. _i_icmp:
5551
5552'``icmp``' Instruction
5553^^^^^^^^^^^^^^^^^^^^^^
5554
5555Syntax:
5556"""""""
5557
5558::
5559
5560 <result> = icmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5561
5562Overview:
5563"""""""""
5564
5565The '``icmp``' instruction returns a boolean value or a vector of
5566boolean values based on comparison of its two integer, integer vector,
5567pointer, or pointer vector operands.
5568
5569Arguments:
5570""""""""""
5571
5572The '``icmp``' instruction takes three operands. The first operand is
5573the condition code indicating the kind of comparison to perform. It is
5574not a value, just a keyword. The possible condition code are:
5575
5576#. ``eq``: equal
5577#. ``ne``: not equal
5578#. ``ugt``: unsigned greater than
5579#. ``uge``: unsigned greater or equal
5580#. ``ult``: unsigned less than
5581#. ``ule``: unsigned less or equal
5582#. ``sgt``: signed greater than
5583#. ``sge``: signed greater or equal
5584#. ``slt``: signed less than
5585#. ``sle``: signed less or equal
5586
5587The remaining two arguments must be :ref:`integer <t_integer>` or
5588:ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
5589must also be identical types.
5590
5591Semantics:
5592""""""""""
5593
5594The '``icmp``' compares ``op1`` and ``op2`` according to the condition
5595code given as ``cond``. The comparison performed always yields either an
5596:ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
5597
5598#. ``eq``: yields ``true`` if the operands are equal, ``false``
5599 otherwise. No sign interpretation is necessary or performed.
5600#. ``ne``: yields ``true`` if the operands are unequal, ``false``
5601 otherwise. No sign interpretation is necessary or performed.
5602#. ``ugt``: interprets the operands as unsigned values and yields
5603 ``true`` if ``op1`` is greater than ``op2``.
5604#. ``uge``: interprets the operands as unsigned values and yields
5605 ``true`` if ``op1`` is greater than or equal to ``op2``.
5606#. ``ult``: interprets the operands as unsigned values and yields
5607 ``true`` if ``op1`` is less than ``op2``.
5608#. ``ule``: interprets the operands as unsigned values and yields
5609 ``true`` if ``op1`` is less than or equal to ``op2``.
5610#. ``sgt``: interprets the operands as signed values and yields ``true``
5611 if ``op1`` is greater than ``op2``.
5612#. ``sge``: interprets the operands as signed values and yields ``true``
5613 if ``op1`` is greater than or equal to ``op2``.
5614#. ``slt``: interprets the operands as signed values and yields ``true``
5615 if ``op1`` is less than ``op2``.
5616#. ``sle``: interprets the operands as signed values and yields ``true``
5617 if ``op1`` is less than or equal to ``op2``.
5618
5619If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
5620are compared as if they were integers.
5621
5622If the operands are integer vectors, then they are compared element by
5623element. The result is an ``i1`` vector with the same number of elements
5624as the values being compared. Otherwise, the result is an ``i1``.
5625
5626Example:
5627""""""""
5628
5629.. code-block:: llvm
5630
5631 <result> = icmp eq i32 4, 5 ; yields: result=false
5632 <result> = icmp ne float* %X, %X ; yields: result=false
5633 <result> = icmp ult i16 4, 5 ; yields: result=true
5634 <result> = icmp sgt i16 4, 5 ; yields: result=false
5635 <result> = icmp ule i16 -4, 5 ; yields: result=false
5636 <result> = icmp sge i16 4, 5 ; yields: result=false
5637
5638Note that the code generator does not yet support vector types with the
5639``icmp`` instruction.
5640
5641.. _i_fcmp:
5642
5643'``fcmp``' Instruction
5644^^^^^^^^^^^^^^^^^^^^^^
5645
5646Syntax:
5647"""""""
5648
5649::
5650
5651 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields {i1} or {<N x i1>}:result
5652
5653Overview:
5654"""""""""
5655
5656The '``fcmp``' instruction returns a boolean value or vector of boolean
5657values based on comparison of its operands.
5658
5659If the operands are floating point scalars, then the result type is a
5660boolean (:ref:`i1 <t_integer>`).
5661
5662If the operands are floating point vectors, then the result type is a
5663vector of boolean with the same number of elements as the operands being
5664compared.
5665
5666Arguments:
5667""""""""""
5668
5669The '``fcmp``' instruction takes three operands. The first operand is
5670the condition code indicating the kind of comparison to perform. It is
5671not a value, just a keyword. The possible condition code are:
5672
5673#. ``false``: no comparison, always returns false
5674#. ``oeq``: ordered and equal
5675#. ``ogt``: ordered and greater than
5676#. ``oge``: ordered and greater than or equal
5677#. ``olt``: ordered and less than
5678#. ``ole``: ordered and less than or equal
5679#. ``one``: ordered and not equal
5680#. ``ord``: ordered (no nans)
5681#. ``ueq``: unordered or equal
5682#. ``ugt``: unordered or greater than
5683#. ``uge``: unordered or greater than or equal
5684#. ``ult``: unordered or less than
5685#. ``ule``: unordered or less than or equal
5686#. ``une``: unordered or not equal
5687#. ``uno``: unordered (either nans)
5688#. ``true``: no comparison, always returns true
5689
5690*Ordered* means that neither operand is a QNAN while *unordered* means
5691that either operand may be a QNAN.
5692
5693Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
5694point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
5695type. They must have identical types.
5696
5697Semantics:
5698""""""""""
5699
5700The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
5701condition code given as ``cond``. If the operands are vectors, then the
5702vectors are compared element by element. Each comparison performed
5703always yields an :ref:`i1 <t_integer>` result, as follows:
5704
5705#. ``false``: always yields ``false``, regardless of operands.
5706#. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
5707 is equal to ``op2``.
5708#. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
5709 is greater than ``op2``.
5710#. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
5711 is greater than or equal to ``op2``.
5712#. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
5713 is less than ``op2``.
5714#. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
5715 is less than or equal to ``op2``.
5716#. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
5717 is not equal to ``op2``.
5718#. ``ord``: yields ``true`` if both operands are not a QNAN.
5719#. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
5720 equal to ``op2``.
5721#. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
5722 greater than ``op2``.
5723#. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
5724 greater than or equal to ``op2``.
5725#. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
5726 less than ``op2``.
5727#. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
5728 less than or equal to ``op2``.
5729#. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
5730 not equal to ``op2``.
5731#. ``uno``: yields ``true`` if either operand is a QNAN.
5732#. ``true``: always yields ``true``, regardless of operands.
5733
5734Example:
5735""""""""
5736
5737.. code-block:: llvm
5738
5739 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
5740 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
5741 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
5742 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
5743
5744Note that the code generator does not yet support vector types with the
5745``fcmp`` instruction.
5746
5747.. _i_phi:
5748
5749'``phi``' Instruction
5750^^^^^^^^^^^^^^^^^^^^^
5751
5752Syntax:
5753"""""""
5754
5755::
5756
5757 <result> = phi <ty> [ <val0>, <label0>], ...
5758
5759Overview:
5760"""""""""
5761
5762The '``phi``' instruction is used to implement the φ node in the SSA
5763graph representing the function.
5764
5765Arguments:
5766""""""""""
5767
5768The type of the incoming values is specified with the first type field.
5769After this, the '``phi``' instruction takes a list of pairs as
5770arguments, with one pair for each predecessor basic block of the current
5771block. Only values of :ref:`first class <t_firstclass>` type may be used as
5772the value arguments to the PHI node. Only labels may be used as the
5773label arguments.
5774
5775There must be no non-phi instructions between the start of a basic block
5776and the PHI instructions: i.e. PHI instructions must be first in a basic
5777block.
5778
5779For the purposes of the SSA form, the use of each incoming value is
5780deemed to occur on the edge from the corresponding predecessor block to
5781the current block (but after any definition of an '``invoke``'
5782instruction's return value on the same edge).
5783
5784Semantics:
5785""""""""""
5786
5787At runtime, the '``phi``' instruction logically takes on the value
5788specified by the pair corresponding to the predecessor basic block that
5789executed just prior to the current block.
5790
5791Example:
5792""""""""
5793
5794.. code-block:: llvm
5795
5796 Loop: ; Infinite loop that counts from 0 on up...
5797 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5798 %nextindvar = add i32 %indvar, 1
5799 br label %Loop
5800
5801.. _i_select:
5802
5803'``select``' Instruction
5804^^^^^^^^^^^^^^^^^^^^^^^^
5805
5806Syntax:
5807"""""""
5808
5809::
5810
5811 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
5812
5813 selty is either i1 or {<N x i1>}
5814
5815Overview:
5816"""""""""
5817
5818The '``select``' instruction is used to choose one value based on a
5819condition, without branching.
5820
5821Arguments:
5822""""""""""
5823
5824The '``select``' instruction requires an 'i1' value or a vector of 'i1'
5825values indicating the condition, and two values of the same :ref:`first
5826class <t_firstclass>` type. If the val1/val2 are vectors and the
5827condition is a scalar, then entire vectors are selected, not individual
5828elements.
5829
5830Semantics:
5831""""""""""
5832
5833If the condition is an i1 and it evaluates to 1, the instruction returns
5834the first value argument; otherwise, it returns the second value
5835argument.
5836
5837If the condition is a vector of i1, then the value arguments must be
5838vectors of the same size, and the selection is done element by element.
5839
5840Example:
5841""""""""
5842
5843.. code-block:: llvm
5844
5845 %X = select i1 true, i8 17, i8 42 ; yields i8:17
5846
5847.. _i_call:
5848
5849'``call``' Instruction
5850^^^^^^^^^^^^^^^^^^^^^^
5851
5852Syntax:
5853"""""""
5854
5855::
5856
5857 <result> = [tail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
5858
5859Overview:
5860"""""""""
5861
5862The '``call``' instruction represents a simple function call.
5863
5864Arguments:
5865""""""""""
5866
5867This instruction requires several arguments:
5868
5869#. The optional "tail" marker indicates that the callee function does
5870 not access any allocas or varargs in the caller. Note that calls may
5871 be marked "tail" even if they do not occur before a
5872 :ref:`ret <i_ret>` instruction. If the "tail" marker is present, the
5873 function call is eligible for tail call optimization, but `might not
5874 in fact be optimized into a jump <CodeGenerator.html#tailcallopt>`_.
5875 The code generator may optimize calls marked "tail" with either 1)
5876 automatic `sibling call
5877 optimization <CodeGenerator.html#sibcallopt>`_ when the caller and
5878 callee have matching signatures, or 2) forced tail call optimization
5879 when the following extra requirements are met:
5880
5881 - Caller and callee both have the calling convention ``fastcc``.
5882 - The call is in tail position (ret immediately follows call and ret
5883 uses value of call or is void).
5884 - Option ``-tailcallopt`` is enabled, or
5885 ``llvm::GuaranteedTailCallOpt`` is ``true``.
5886 - `Platform specific constraints are
5887 met. <CodeGenerator.html#tailcallopt>`_
5888
5889#. The optional "cconv" marker indicates which :ref:`calling
5890 convention <callingconv>` the call should use. If none is
5891 specified, the call defaults to using C calling conventions. The
5892 calling convention of the call must match the calling convention of
5893 the target function, or else the behavior is undefined.
5894#. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5895 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5896 are valid here.
5897#. '``ty``': the type of the call instruction itself which is also the
5898 type of the return value. Functions that return no value are marked
5899 ``void``.
5900#. '``fnty``': shall be the signature of the pointer to function value
5901 being invoked. The argument types must match the types implied by
5902 this signature. This type can be omitted if the function is not
5903 varargs and if the function type does not return a pointer to a
5904 function.
5905#. '``fnptrval``': An LLVM value containing a pointer to a function to
5906 be invoked. In most cases, this is a direct function invocation, but
5907 indirect ``call``'s are just as possible, calling an arbitrary pointer
5908 to function value.
5909#. '``function args``': argument list whose types match the function
5910 signature argument types and parameter attributes. All arguments must
5911 be of :ref:`first class <t_firstclass>` type. If the function signature
5912 indicates the function accepts a variable number of arguments, the
5913 extra arguments can be specified.
5914#. The optional :ref:`function attributes <fnattrs>` list. Only
5915 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5916 attributes are valid here.
5917
5918Semantics:
5919""""""""""
5920
5921The '``call``' instruction is used to cause control flow to transfer to
5922a specified function, with its incoming arguments bound to the specified
5923values. Upon a '``ret``' instruction in the called function, control
5924flow continues with the instruction after the function call, and the
5925return value of the function is bound to the result argument.
5926
5927Example:
5928""""""""
5929
5930.. code-block:: llvm
5931
5932 %retval = call i32 @test(i32 %argc)
5933 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
5934 %X = tail call i32 @foo() ; yields i32
5935 %Y = tail call fastcc i32 @foo() ; yields i32
5936 call void %foo(i8 97 signext)
5937
5938 %struct.A = type { i32, i8 }
5939 %r = call %struct.A @foo() ; yields { 32, i8 }
5940 %gr = extractvalue %struct.A %r, 0 ; yields i32
5941 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
5942 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
5943 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
5944
5945llvm treats calls to some functions with names and arguments that match
5946the standard C99 library as being the C99 library functions, and may
5947perform optimizations or generate code for them under that assumption.
5948This is something we'd like to change in the future to provide better
5949support for freestanding environments and non-C-based languages.
5950
5951.. _i_va_arg:
5952
5953'``va_arg``' Instruction
5954^^^^^^^^^^^^^^^^^^^^^^^^
5955
5956Syntax:
5957"""""""
5958
5959::
5960
5961 <resultval> = va_arg <va_list*> <arglist>, <argty>
5962
5963Overview:
5964"""""""""
5965
5966The '``va_arg``' instruction is used to access arguments passed through
5967the "variable argument" area of a function call. It is used to implement
5968the ``va_arg`` macro in C.
5969
5970Arguments:
5971""""""""""
5972
5973This instruction takes a ``va_list*`` value and the type of the
5974argument. It returns a value of the specified argument type and
5975increments the ``va_list`` to point to the next argument. The actual
5976type of ``va_list`` is target specific.
5977
5978Semantics:
5979""""""""""
5980
5981The '``va_arg``' instruction loads an argument of the specified type
5982from the specified ``va_list`` and causes the ``va_list`` to point to
5983the next argument. For more information, see the variable argument
5984handling :ref:`Intrinsic Functions <int_varargs>`.
5985
5986It is legal for this instruction to be called in a function which does
5987not take a variable number of arguments, for example, the ``vfprintf``
5988function.
5989
5990``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
5991function <intrinsics>` because it takes a type as an argument.
5992
5993Example:
5994""""""""
5995
5996See the :ref:`variable argument processing <int_varargs>` section.
5997
5998Note that the code generator does not yet fully support va\_arg on many
5999targets. Also, it does not currently support va\_arg with aggregate
6000types on any target.
6001
6002.. _i_landingpad:
6003
6004'``landingpad``' Instruction
6005^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6006
6007Syntax:
6008"""""""
6009
6010::
6011
6012 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6013 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6014
6015 <clause> := catch <type> <value>
6016 <clause> := filter <array constant type> <array constant>
6017
6018Overview:
6019"""""""""
6020
6021The '``landingpad``' instruction is used by `LLVM's exception handling
6022system <ExceptionHandling.html#overview>`_ to specify that a basic block
Dmitri Gribenkoae4a9ae2013-01-19 20:34:20 +00006023is a landing pad --- one where the exception lands, and corresponds to the
Sean Silvaf722b002012-12-07 10:36:55 +00006024code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6025defines values supplied by the personality function (``pers_fn``) upon
6026re-entry to the function. The ``resultval`` has the type ``resultty``.
6027
6028Arguments:
6029""""""""""
6030
6031This instruction takes a ``pers_fn`` value. This is the personality
6032function associated with the unwinding mechanism. The optional
6033``cleanup`` flag indicates that the landing pad block is a cleanup.
6034
Dmitri Gribenkoae4a9ae2013-01-19 20:34:20 +00006035A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
Sean Silvaf722b002012-12-07 10:36:55 +00006036contains the global variable representing the "type" that may be caught
6037or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6038clause takes an array constant as its argument. Use
6039"``[0 x i8**] undef``" for a filter which cannot throw. The
6040'``landingpad``' instruction must contain *at least* one ``clause`` or
6041the ``cleanup`` flag.
6042
6043Semantics:
6044""""""""""
6045
6046The '``landingpad``' instruction defines the values which are set by the
6047personality function (``pers_fn``) upon re-entry to the function, and
6048therefore the "result type" of the ``landingpad`` instruction. As with
6049calling conventions, how the personality function results are
6050represented in LLVM IR is target specific.
6051
6052The clauses are applied in order from top to bottom. If two
6053``landingpad`` instructions are merged together through inlining, the
6054clauses from the calling function are appended to the list of clauses.
6055When the call stack is being unwound due to an exception being thrown,
6056the exception is compared against each ``clause`` in turn. If it doesn't
6057match any of the clauses, and the ``cleanup`` flag is not set, then
6058unwinding continues further up the call stack.
6059
6060The ``landingpad`` instruction has several restrictions:
6061
6062- A landing pad block is a basic block which is the unwind destination
6063 of an '``invoke``' instruction.
6064- A landing pad block must have a '``landingpad``' instruction as its
6065 first non-PHI instruction.
6066- There can be only one '``landingpad``' instruction within the landing
6067 pad block.
6068- A basic block that is not a landing pad block may not include a
6069 '``landingpad``' instruction.
6070- All '``landingpad``' instructions in a function must have the same
6071 personality function.
6072
6073Example:
6074""""""""
6075
6076.. code-block:: llvm
6077
6078 ;; A landing pad which can catch an integer.
6079 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6080 catch i8** @_ZTIi
6081 ;; A landing pad that is a cleanup.
6082 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6083 cleanup
6084 ;; A landing pad which can catch an integer and can only throw a double.
6085 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6086 catch i8** @_ZTIi
6087 filter [1 x i8**] [@_ZTId]
6088
6089.. _intrinsics:
6090
6091Intrinsic Functions
6092===================
6093
6094LLVM supports the notion of an "intrinsic function". These functions
6095have well known names and semantics and are required to follow certain
6096restrictions. Overall, these intrinsics represent an extension mechanism
6097for the LLVM language that does not require changing all of the
6098transformations in LLVM when adding to the language (or the bitcode
6099reader/writer, the parser, etc...).
6100
6101Intrinsic function names must all start with an "``llvm.``" prefix. This
6102prefix is reserved in LLVM for intrinsic names; thus, function names may
6103not begin with this prefix. Intrinsic functions must always be external
6104functions: you cannot define the body of intrinsic functions. Intrinsic
6105functions may only be used in call or invoke instructions: it is illegal
6106to take the address of an intrinsic function. Additionally, because
6107intrinsic functions are part of the LLVM language, it is required if any
6108are added that they be documented here.
6109
6110Some intrinsic functions can be overloaded, i.e., the intrinsic
6111represents a family of functions that perform the same operation but on
6112different data types. Because LLVM can represent over 8 million
6113different integer types, overloading is used commonly to allow an
6114intrinsic function to operate on any integer type. One or more of the
6115argument types or the result type can be overloaded to accept any
6116integer type. Argument types may also be defined as exactly matching a
6117previous argument's type or the result type. This allows an intrinsic
6118function which accepts multiple arguments, but needs all of them to be
6119of the same type, to only be overloaded with respect to a single
6120argument or the result.
6121
6122Overloaded intrinsics will have the names of its overloaded argument
6123types encoded into its function name, each preceded by a period. Only
6124those types which are overloaded result in a name suffix. Arguments
6125whose type is matched against another type do not. For example, the
6126``llvm.ctpop`` function can take an integer of any width and returns an
6127integer of exactly the same integer width. This leads to a family of
6128functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6129``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6130overloaded, and only one type suffix is required. Because the argument's
6131type is matched against the return type, it does not require its own
6132name suffix.
6133
6134To learn how to add an intrinsic function, please see the `Extending
6135LLVM Guide <ExtendingLLVM.html>`_.
6136
6137.. _int_varargs:
6138
6139Variable Argument Handling Intrinsics
6140-------------------------------------
6141
6142Variable argument support is defined in LLVM with the
6143:ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6144functions. These functions are related to the similarly named macros
6145defined in the ``<stdarg.h>`` header file.
6146
6147All of these functions operate on arguments that use a target-specific
6148value type "``va_list``". The LLVM assembly language reference manual
6149does not define what this type is, so all transformations should be
6150prepared to handle these functions regardless of the type used.
6151
6152This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6153variable argument handling intrinsic functions are used.
6154
6155.. code-block:: llvm
6156
6157 define i32 @test(i32 %X, ...) {
6158 ; Initialize variable argument processing
6159 %ap = alloca i8*
6160 %ap2 = bitcast i8** %ap to i8*
6161 call void @llvm.va_start(i8* %ap2)
6162
6163 ; Read a single integer argument
6164 %tmp = va_arg i8** %ap, i32
6165
6166 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6167 %aq = alloca i8*
6168 %aq2 = bitcast i8** %aq to i8*
6169 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6170 call void @llvm.va_end(i8* %aq2)
6171
6172 ; Stop processing of arguments.
6173 call void @llvm.va_end(i8* %ap2)
6174 ret i32 %tmp
6175 }
6176
6177 declare void @llvm.va_start(i8*)
6178 declare void @llvm.va_copy(i8*, i8*)
6179 declare void @llvm.va_end(i8*)
6180
6181.. _int_va_start:
6182
6183'``llvm.va_start``' Intrinsic
6184^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6185
6186Syntax:
6187"""""""
6188
6189::
6190
6191 declare void %llvm.va_start(i8* <arglist>)
6192
6193Overview:
6194"""""""""
6195
6196The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6197subsequent use by ``va_arg``.
6198
6199Arguments:
6200""""""""""
6201
6202The argument is a pointer to a ``va_list`` element to initialize.
6203
6204Semantics:
6205""""""""""
6206
6207The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6208available in C. In a target-dependent way, it initializes the
6209``va_list`` element to which the argument points, so that the next call
6210to ``va_arg`` will produce the first variable argument passed to the
6211function. Unlike the C ``va_start`` macro, this intrinsic does not need
6212to know the last argument of the function as the compiler can figure
6213that out.
6214
6215'``llvm.va_end``' Intrinsic
6216^^^^^^^^^^^^^^^^^^^^^^^^^^^
6217
6218Syntax:
6219"""""""
6220
6221::
6222
6223 declare void @llvm.va_end(i8* <arglist>)
6224
6225Overview:
6226"""""""""
6227
6228The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6229initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6230
6231Arguments:
6232""""""""""
6233
6234The argument is a pointer to a ``va_list`` to destroy.
6235
6236Semantics:
6237""""""""""
6238
6239The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6240available in C. In a target-dependent way, it destroys the ``va_list``
6241element to which the argument points. Calls to
6242:ref:`llvm.va_start <int_va_start>` and
6243:ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6244``llvm.va_end``.
6245
6246.. _int_va_copy:
6247
6248'``llvm.va_copy``' Intrinsic
6249^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6250
6251Syntax:
6252"""""""
6253
6254::
6255
6256 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6257
6258Overview:
6259"""""""""
6260
6261The '``llvm.va_copy``' intrinsic copies the current argument position
6262from the source argument list to the destination argument list.
6263
6264Arguments:
6265""""""""""
6266
6267The first argument is a pointer to a ``va_list`` element to initialize.
6268The second argument is a pointer to a ``va_list`` element to copy from.
6269
6270Semantics:
6271""""""""""
6272
6273The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6274available in C. In a target-dependent way, it copies the source
6275``va_list`` element into the destination ``va_list`` element. This
6276intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6277arbitrarily complex and require, for example, memory allocation.
6278
6279Accurate Garbage Collection Intrinsics
6280--------------------------------------
6281
6282LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6283(GC) requires the implementation and generation of these intrinsics.
6284These intrinsics allow identification of :ref:`GC roots on the
6285stack <int_gcroot>`, as well as garbage collector implementations that
6286require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6287Front-ends for type-safe garbage collected languages should generate
6288these intrinsics to make use of the LLVM garbage collectors. For more
6289details, see `Accurate Garbage Collection with
6290LLVM <GarbageCollection.html>`_.
6291
6292The garbage collection intrinsics only operate on objects in the generic
6293address space (address space zero).
6294
6295.. _int_gcroot:
6296
6297'``llvm.gcroot``' Intrinsic
6298^^^^^^^^^^^^^^^^^^^^^^^^^^^
6299
6300Syntax:
6301"""""""
6302
6303::
6304
6305 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6306
6307Overview:
6308"""""""""
6309
6310The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
6311the code generator, and allows some metadata to be associated with it.
6312
6313Arguments:
6314""""""""""
6315
6316The first argument specifies the address of a stack object that contains
6317the root pointer. The second pointer (which must be either a constant or
6318a global value address) contains the meta-data to be associated with the
6319root.
6320
6321Semantics:
6322""""""""""
6323
6324At runtime, a call to this intrinsic stores a null pointer into the
6325"ptrloc" location. At compile-time, the code generator generates
6326information to allow the runtime to find the pointer at GC safe points.
6327The '``llvm.gcroot``' intrinsic may only be used in a function which
6328:ref:`specifies a GC algorithm <gc>`.
6329
6330.. _int_gcread:
6331
6332'``llvm.gcread``' Intrinsic
6333^^^^^^^^^^^^^^^^^^^^^^^^^^^
6334
6335Syntax:
6336"""""""
6337
6338::
6339
6340 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6341
6342Overview:
6343"""""""""
6344
6345The '``llvm.gcread``' intrinsic identifies reads of references from heap
6346locations, allowing garbage collector implementations that require read
6347barriers.
6348
6349Arguments:
6350""""""""""
6351
6352The second argument is the address to read from, which should be an
6353address allocated from the garbage collector. The first object is a
6354pointer to the start of the referenced object, if needed by the language
6355runtime (otherwise null).
6356
6357Semantics:
6358""""""""""
6359
6360The '``llvm.gcread``' intrinsic has the same semantics as a load
6361instruction, but may be replaced with substantially more complex code by
6362the garbage collector runtime, as needed. The '``llvm.gcread``'
6363intrinsic may only be used in a function which :ref:`specifies a GC
6364algorithm <gc>`.
6365
6366.. _int_gcwrite:
6367
6368'``llvm.gcwrite``' Intrinsic
6369^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6370
6371Syntax:
6372"""""""
6373
6374::
6375
6376 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6377
6378Overview:
6379"""""""""
6380
6381The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
6382locations, allowing garbage collector implementations that require write
6383barriers (such as generational or reference counting collectors).
6384
6385Arguments:
6386""""""""""
6387
6388The first argument is the reference to store, the second is the start of
6389the object to store it to, and the third is the address of the field of
6390Obj to store to. If the runtime does not require a pointer to the
6391object, Obj may be null.
6392
6393Semantics:
6394""""""""""
6395
6396The '``llvm.gcwrite``' intrinsic has the same semantics as a store
6397instruction, but may be replaced with substantially more complex code by
6398the garbage collector runtime, as needed. The '``llvm.gcwrite``'
6399intrinsic may only be used in a function which :ref:`specifies a GC
6400algorithm <gc>`.
6401
6402Code Generator Intrinsics
6403-------------------------
6404
6405These intrinsics are provided by LLVM to expose special features that
6406may only be implemented with code generator support.
6407
6408'``llvm.returnaddress``' Intrinsic
6409^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6410
6411Syntax:
6412"""""""
6413
6414::
6415
6416 declare i8 *@llvm.returnaddress(i32 <level>)
6417
6418Overview:
6419"""""""""
6420
6421The '``llvm.returnaddress``' intrinsic attempts to compute a
6422target-specific value indicating the return address of the current
6423function or one of its callers.
6424
6425Arguments:
6426""""""""""
6427
6428The argument to this intrinsic indicates which function to return the
6429address for. Zero indicates the calling function, one indicates its
6430caller, etc. The argument is **required** to be a constant integer
6431value.
6432
6433Semantics:
6434""""""""""
6435
6436The '``llvm.returnaddress``' intrinsic either returns a pointer
6437indicating the return address of the specified call frame, or zero if it
6438cannot be identified. The value returned by this intrinsic is likely to
6439be incorrect or 0 for arguments other than zero, so it should only be
6440used for debugging purposes.
6441
6442Note that calling this intrinsic does not prevent function inlining or
6443other aggressive transformations, so the value returned may not be that
6444of the obvious source-language caller.
6445
6446'``llvm.frameaddress``' Intrinsic
6447^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6448
6449Syntax:
6450"""""""
6451
6452::
6453
6454 declare i8* @llvm.frameaddress(i32 <level>)
6455
6456Overview:
6457"""""""""
6458
6459The '``llvm.frameaddress``' intrinsic attempts to return the
6460target-specific frame pointer value for the specified stack frame.
6461
6462Arguments:
6463""""""""""
6464
6465The argument to this intrinsic indicates which function to return the
6466frame pointer for. Zero indicates the calling function, one indicates
6467its caller, etc. The argument is **required** to be a constant integer
6468value.
6469
6470Semantics:
6471""""""""""
6472
6473The '``llvm.frameaddress``' intrinsic either returns a pointer
6474indicating the frame address of the specified call frame, or zero if it
6475cannot be identified. The value returned by this intrinsic is likely to
6476be incorrect or 0 for arguments other than zero, so it should only be
6477used for debugging purposes.
6478
6479Note that calling this intrinsic does not prevent function inlining or
6480other aggressive transformations, so the value returned may not be that
6481of the obvious source-language caller.
6482
6483.. _int_stacksave:
6484
6485'``llvm.stacksave``' Intrinsic
6486^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6487
6488Syntax:
6489"""""""
6490
6491::
6492
6493 declare i8* @llvm.stacksave()
6494
6495Overview:
6496"""""""""
6497
6498The '``llvm.stacksave``' intrinsic is used to remember the current state
6499of the function stack, for use with
6500:ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
6501implementing language features like scoped automatic variable sized
6502arrays in C99.
6503
6504Semantics:
6505""""""""""
6506
6507This intrinsic returns a opaque pointer value that can be passed to
6508:ref:`llvm.stackrestore <int_stackrestore>`. When an
6509``llvm.stackrestore`` intrinsic is executed with a value saved from
6510``llvm.stacksave``, it effectively restores the state of the stack to
6511the state it was in when the ``llvm.stacksave`` intrinsic executed. In
6512practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
6513were allocated after the ``llvm.stacksave`` was executed.
6514
6515.. _int_stackrestore:
6516
6517'``llvm.stackrestore``' Intrinsic
6518^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6519
6520Syntax:
6521"""""""
6522
6523::
6524
6525 declare void @llvm.stackrestore(i8* %ptr)
6526
6527Overview:
6528"""""""""
6529
6530The '``llvm.stackrestore``' intrinsic is used to restore the state of
6531the function stack to the state it was in when the corresponding
6532:ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
6533useful for implementing language features like scoped automatic variable
6534sized arrays in C99.
6535
6536Semantics:
6537""""""""""
6538
6539See the description for :ref:`llvm.stacksave <int_stacksave>`.
6540
6541'``llvm.prefetch``' Intrinsic
6542^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6543
6544Syntax:
6545"""""""
6546
6547::
6548
6549 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6550
6551Overview:
6552"""""""""
6553
6554The '``llvm.prefetch``' intrinsic is a hint to the code generator to
6555insert a prefetch instruction if supported; otherwise, it is a noop.
6556Prefetches have no effect on the behavior of the program but can change
6557its performance characteristics.
6558
6559Arguments:
6560""""""""""
6561
6562``address`` is the address to be prefetched, ``rw`` is the specifier
6563determining if the fetch should be for a read (0) or write (1), and
6564``locality`` is a temporal locality specifier ranging from (0) - no
6565locality, to (3) - extremely local keep in cache. The ``cache type``
6566specifies whether the prefetch is performed on the data (1) or
6567instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
6568arguments must be constant integers.
6569
6570Semantics:
6571""""""""""
6572
6573This intrinsic does not modify the behavior of the program. In
6574particular, prefetches cannot trap and do not produce a value. On
6575targets that support this intrinsic, the prefetch can provide hints to
6576the processor cache for better performance.
6577
6578'``llvm.pcmarker``' Intrinsic
6579^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6580
6581Syntax:
6582"""""""
6583
6584::
6585
6586 declare void @llvm.pcmarker(i32 <id>)
6587
6588Overview:
6589"""""""""
6590
6591The '``llvm.pcmarker``' intrinsic is a method to export a Program
6592Counter (PC) in a region of code to simulators and other tools. The
6593method is target specific, but it is expected that the marker will use
6594exported symbols to transmit the PC of the marker. The marker makes no
6595guarantees that it will remain with any specific instruction after
6596optimizations. It is possible that the presence of a marker will inhibit
6597optimizations. The intended use is to be inserted after optimizations to
6598allow correlations of simulation runs.
6599
6600Arguments:
6601""""""""""
6602
6603``id`` is a numerical id identifying the marker.
6604
6605Semantics:
6606""""""""""
6607
6608This intrinsic does not modify the behavior of the program. Backends
6609that do not support this intrinsic may ignore it.
6610
6611'``llvm.readcyclecounter``' Intrinsic
6612^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6613
6614Syntax:
6615"""""""
6616
6617::
6618
6619 declare i64 @llvm.readcyclecounter()
6620
6621Overview:
6622"""""""""
6623
6624The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
6625counter register (or similar low latency, high accuracy clocks) on those
6626targets that support it. On X86, it should map to RDTSC. On Alpha, it
6627should map to RPCC. As the backing counters overflow quickly (on the
6628order of 9 seconds on alpha), this should only be used for small
6629timings.
6630
6631Semantics:
6632""""""""""
6633
6634When directly supported, reading the cycle counter should not modify any
6635memory. Implementations are allowed to either return a application
6636specific value or a system wide value. On backends without support, this
6637is lowered to a constant 0.
6638
6639Standard C Library Intrinsics
6640-----------------------------
6641
6642LLVM provides intrinsics for a few important standard C library
6643functions. These intrinsics allow source-language front-ends to pass
6644information about the alignment of the pointer arguments to the code
6645generator, providing opportunity for more efficient code generation.
6646
6647.. _int_memcpy:
6648
6649'``llvm.memcpy``' Intrinsic
6650^^^^^^^^^^^^^^^^^^^^^^^^^^^
6651
6652Syntax:
6653"""""""
6654
6655This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
6656integer bit width and for different address spaces. Not all targets
6657support all bit widths however.
6658
6659::
6660
6661 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6662 i32 <len>, i32 <align>, i1 <isvolatile>)
6663 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6664 i64 <len>, i32 <align>, i1 <isvolatile>)
6665
6666Overview:
6667"""""""""
6668
6669The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6670source location to the destination location.
6671
6672Note that, unlike the standard libc function, the ``llvm.memcpy.*``
6673intrinsics do not return a value, takes extra alignment/isvolatile
6674arguments and the pointers can be in specified address spaces.
6675
6676Arguments:
6677""""""""""
6678
6679The first argument is a pointer to the destination, the second is a
6680pointer to the source. The third argument is an integer argument
6681specifying the number of bytes to copy, the fourth argument is the
6682alignment of the source and destination locations, and the fifth is a
6683boolean indicating a volatile access.
6684
6685If the call to this intrinsic has an alignment value that is not 0 or 1,
6686then the caller guarantees that both the source and destination pointers
6687are aligned to that boundary.
6688
6689If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
6690a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6691very cleanly specified and it is unwise to depend on it.
6692
6693Semantics:
6694""""""""""
6695
6696The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
6697source location to the destination location, which are not allowed to
6698overlap. It copies "len" bytes of memory over. If the argument is known
6699to be aligned to some boundary, this can be specified as the fourth
6700argument, otherwise it should be set to 0 or 1.
6701
6702'``llvm.memmove``' Intrinsic
6703^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6704
6705Syntax:
6706"""""""
6707
6708This is an overloaded intrinsic. You can use llvm.memmove on any integer
6709bit width and for different address space. Not all targets support all
6710bit widths however.
6711
6712::
6713
6714 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6715 i32 <len>, i32 <align>, i1 <isvolatile>)
6716 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6717 i64 <len>, i32 <align>, i1 <isvolatile>)
6718
6719Overview:
6720"""""""""
6721
6722The '``llvm.memmove.*``' intrinsics move a block of memory from the
6723source location to the destination location. It is similar to the
6724'``llvm.memcpy``' intrinsic but allows the two memory locations to
6725overlap.
6726
6727Note that, unlike the standard libc function, the ``llvm.memmove.*``
6728intrinsics do not return a value, takes extra alignment/isvolatile
6729arguments and the pointers can be in specified address spaces.
6730
6731Arguments:
6732""""""""""
6733
6734The first argument is a pointer to the destination, the second is a
6735pointer to the source. The third argument is an integer argument
6736specifying the number of bytes to copy, the fourth argument is the
6737alignment of the source and destination locations, and the fifth is a
6738boolean indicating a volatile access.
6739
6740If the call to this intrinsic has an alignment value that is not 0 or 1,
6741then the caller guarantees that the source and destination pointers are
6742aligned to that boundary.
6743
6744If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
6745is a :ref:`volatile operation <volatile>`. The detailed access behavior is
6746not very cleanly specified and it is unwise to depend on it.
6747
6748Semantics:
6749""""""""""
6750
6751The '``llvm.memmove.*``' intrinsics copy a block of memory from the
6752source location to the destination location, which may overlap. It
6753copies "len" bytes of memory over. If the argument is known to be
6754aligned to some boundary, this can be specified as the fourth argument,
6755otherwise it should be set to 0 or 1.
6756
6757'``llvm.memset.*``' Intrinsics
6758^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6759
6760Syntax:
6761"""""""
6762
6763This is an overloaded intrinsic. You can use llvm.memset on any integer
6764bit width and for different address spaces. However, not all targets
6765support all bit widths.
6766
6767::
6768
6769 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6770 i32 <len>, i32 <align>, i1 <isvolatile>)
6771 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6772 i64 <len>, i32 <align>, i1 <isvolatile>)
6773
6774Overview:
6775"""""""""
6776
6777The '``llvm.memset.*``' intrinsics fill a block of memory with a
6778particular byte value.
6779
6780Note that, unlike the standard libc function, the ``llvm.memset``
6781intrinsic does not return a value and takes extra alignment/volatile
6782arguments. Also, the destination can be in an arbitrary address space.
6783
6784Arguments:
6785""""""""""
6786
6787The first argument is a pointer to the destination to fill, the second
6788is the byte value with which to fill it, the third argument is an
6789integer argument specifying the number of bytes to fill, and the fourth
6790argument is the known alignment of the destination location.
6791
6792If the call to this intrinsic has an alignment value that is not 0 or 1,
6793then the caller guarantees that the destination pointer is aligned to
6794that boundary.
6795
6796If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
6797a :ref:`volatile operation <volatile>`. The detailed access behavior is not
6798very cleanly specified and it is unwise to depend on it.
6799
6800Semantics:
6801""""""""""
6802
6803The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
6804at the destination location. If the argument is known to be aligned to
6805some boundary, this can be specified as the fourth argument, otherwise
6806it should be set to 0 or 1.
6807
6808'``llvm.sqrt.*``' Intrinsic
6809^^^^^^^^^^^^^^^^^^^^^^^^^^^
6810
6811Syntax:
6812"""""""
6813
6814This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
6815floating point or vector of floating point type. Not all targets support
6816all types however.
6817
6818::
6819
6820 declare float @llvm.sqrt.f32(float %Val)
6821 declare double @llvm.sqrt.f64(double %Val)
6822 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6823 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6824 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6825
6826Overview:
6827"""""""""
6828
6829The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
6830returning the same value as the libm '``sqrt``' functions would. Unlike
6831``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
6832negative numbers other than -0.0 (which allows for better optimization,
6833because there is no need to worry about errno being set).
6834``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
6835
6836Arguments:
6837""""""""""
6838
6839The argument and return value are floating point numbers of the same
6840type.
6841
6842Semantics:
6843""""""""""
6844
6845This function returns the sqrt of the specified operand if it is a
6846nonnegative floating point number.
6847
6848'``llvm.powi.*``' Intrinsic
6849^^^^^^^^^^^^^^^^^^^^^^^^^^^
6850
6851Syntax:
6852"""""""
6853
6854This is an overloaded intrinsic. You can use ``llvm.powi`` on any
6855floating point or vector of floating point type. Not all targets support
6856all types however.
6857
6858::
6859
6860 declare float @llvm.powi.f32(float %Val, i32 %power)
6861 declare double @llvm.powi.f64(double %Val, i32 %power)
6862 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6863 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6864 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6865
6866Overview:
6867"""""""""
6868
6869The '``llvm.powi.*``' intrinsics return the first operand raised to the
6870specified (positive or negative) power. The order of evaluation of
6871multiplications is not defined. When a vector of floating point type is
6872used, the second argument remains a scalar integer value.
6873
6874Arguments:
6875""""""""""
6876
6877The second argument is an integer power, and the first is a value to
6878raise to that power.
6879
6880Semantics:
6881""""""""""
6882
6883This function returns the first value raised to the second power with an
6884unspecified sequence of rounding operations.
6885
6886'``llvm.sin.*``' Intrinsic
6887^^^^^^^^^^^^^^^^^^^^^^^^^^
6888
6889Syntax:
6890"""""""
6891
6892This is an overloaded intrinsic. You can use ``llvm.sin`` on any
6893floating point or vector of floating point type. Not all targets support
6894all types however.
6895
6896::
6897
6898 declare float @llvm.sin.f32(float %Val)
6899 declare double @llvm.sin.f64(double %Val)
6900 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6901 declare fp128 @llvm.sin.f128(fp128 %Val)
6902 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6903
6904Overview:
6905"""""""""
6906
6907The '``llvm.sin.*``' intrinsics return the sine of the operand.
6908
6909Arguments:
6910""""""""""
6911
6912The argument and return value are floating point numbers of the same
6913type.
6914
6915Semantics:
6916""""""""""
6917
6918This function returns the sine of the specified operand, returning the
6919same values as the libm ``sin`` functions would, and handles error
6920conditions in the same way.
6921
6922'``llvm.cos.*``' Intrinsic
6923^^^^^^^^^^^^^^^^^^^^^^^^^^
6924
6925Syntax:
6926"""""""
6927
6928This is an overloaded intrinsic. You can use ``llvm.cos`` on any
6929floating point or vector of floating point type. Not all targets support
6930all types however.
6931
6932::
6933
6934 declare float @llvm.cos.f32(float %Val)
6935 declare double @llvm.cos.f64(double %Val)
6936 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6937 declare fp128 @llvm.cos.f128(fp128 %Val)
6938 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6939
6940Overview:
6941"""""""""
6942
6943The '``llvm.cos.*``' intrinsics return the cosine of the operand.
6944
6945Arguments:
6946""""""""""
6947
6948The argument and return value are floating point numbers of the same
6949type.
6950
6951Semantics:
6952""""""""""
6953
6954This function returns the cosine of the specified operand, returning the
6955same values as the libm ``cos`` functions would, and handles error
6956conditions in the same way.
6957
6958'``llvm.pow.*``' Intrinsic
6959^^^^^^^^^^^^^^^^^^^^^^^^^^
6960
6961Syntax:
6962"""""""
6963
6964This is an overloaded intrinsic. You can use ``llvm.pow`` on any
6965floating point or vector of floating point type. Not all targets support
6966all types however.
6967
6968::
6969
6970 declare float @llvm.pow.f32(float %Val, float %Power)
6971 declare double @llvm.pow.f64(double %Val, double %Power)
6972 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6973 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6974 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6975
6976Overview:
6977"""""""""
6978
6979The '``llvm.pow.*``' intrinsics return the first operand raised to the
6980specified (positive or negative) power.
6981
6982Arguments:
6983""""""""""
6984
6985The second argument is a floating point power, and the first is a value
6986to raise to that power.
6987
6988Semantics:
6989""""""""""
6990
6991This function returns the first value raised to the second power,
6992returning the same values as the libm ``pow`` functions would, and
6993handles error conditions in the same way.
6994
6995'``llvm.exp.*``' Intrinsic
6996^^^^^^^^^^^^^^^^^^^^^^^^^^
6997
6998Syntax:
6999"""""""
7000
7001This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7002floating point or vector of floating point type. Not all targets support
7003all types however.
7004
7005::
7006
7007 declare float @llvm.exp.f32(float %Val)
7008 declare double @llvm.exp.f64(double %Val)
7009 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7010 declare fp128 @llvm.exp.f128(fp128 %Val)
7011 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7012
7013Overview:
7014"""""""""
7015
7016The '``llvm.exp.*``' intrinsics perform the exp function.
7017
7018Arguments:
7019""""""""""
7020
7021The argument and return value are floating point numbers of the same
7022type.
7023
7024Semantics:
7025""""""""""
7026
7027This function returns the same values as the libm ``exp`` functions
7028would, and handles error conditions in the same way.
7029
7030'``llvm.exp2.*``' Intrinsic
7031^^^^^^^^^^^^^^^^^^^^^^^^^^^
7032
7033Syntax:
7034"""""""
7035
7036This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7037floating point or vector of floating point type. Not all targets support
7038all types however.
7039
7040::
7041
7042 declare float @llvm.exp2.f32(float %Val)
7043 declare double @llvm.exp2.f64(double %Val)
7044 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7045 declare fp128 @llvm.exp2.f128(fp128 %Val)
7046 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7047
7048Overview:
7049"""""""""
7050
7051The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7052
7053Arguments:
7054""""""""""
7055
7056The argument and return value are floating point numbers of the same
7057type.
7058
7059Semantics:
7060""""""""""
7061
7062This function returns the same values as the libm ``exp2`` functions
7063would, and handles error conditions in the same way.
7064
7065'``llvm.log.*``' Intrinsic
7066^^^^^^^^^^^^^^^^^^^^^^^^^^
7067
7068Syntax:
7069"""""""
7070
7071This is an overloaded intrinsic. You can use ``llvm.log`` on any
7072floating point or vector of floating point type. Not all targets support
7073all types however.
7074
7075::
7076
7077 declare float @llvm.log.f32(float %Val)
7078 declare double @llvm.log.f64(double %Val)
7079 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7080 declare fp128 @llvm.log.f128(fp128 %Val)
7081 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7082
7083Overview:
7084"""""""""
7085
7086The '``llvm.log.*``' intrinsics perform the log function.
7087
7088Arguments:
7089""""""""""
7090
7091The argument and return value are floating point numbers of the same
7092type.
7093
7094Semantics:
7095""""""""""
7096
7097This function returns the same values as the libm ``log`` functions
7098would, and handles error conditions in the same way.
7099
7100'``llvm.log10.*``' Intrinsic
7101^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7102
7103Syntax:
7104"""""""
7105
7106This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7107floating point or vector of floating point type. Not all targets support
7108all types however.
7109
7110::
7111
7112 declare float @llvm.log10.f32(float %Val)
7113 declare double @llvm.log10.f64(double %Val)
7114 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7115 declare fp128 @llvm.log10.f128(fp128 %Val)
7116 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7117
7118Overview:
7119"""""""""
7120
7121The '``llvm.log10.*``' intrinsics perform the log10 function.
7122
7123Arguments:
7124""""""""""
7125
7126The argument and return value are floating point numbers of the same
7127type.
7128
7129Semantics:
7130""""""""""
7131
7132This function returns the same values as the libm ``log10`` functions
7133would, and handles error conditions in the same way.
7134
7135'``llvm.log2.*``' Intrinsic
7136^^^^^^^^^^^^^^^^^^^^^^^^^^^
7137
7138Syntax:
7139"""""""
7140
7141This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7142floating point or vector of floating point type. Not all targets support
7143all types however.
7144
7145::
7146
7147 declare float @llvm.log2.f32(float %Val)
7148 declare double @llvm.log2.f64(double %Val)
7149 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7150 declare fp128 @llvm.log2.f128(fp128 %Val)
7151 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7152
7153Overview:
7154"""""""""
7155
7156The '``llvm.log2.*``' intrinsics perform the log2 function.
7157
7158Arguments:
7159""""""""""
7160
7161The argument and return value are floating point numbers of the same
7162type.
7163
7164Semantics:
7165""""""""""
7166
7167This function returns the same values as the libm ``log2`` functions
7168would, and handles error conditions in the same way.
7169
7170'``llvm.fma.*``' Intrinsic
7171^^^^^^^^^^^^^^^^^^^^^^^^^^
7172
7173Syntax:
7174"""""""
7175
7176This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7177floating point or vector of floating point type. Not all targets support
7178all types however.
7179
7180::
7181
7182 declare float @llvm.fma.f32(float %a, float %b, float %c)
7183 declare double @llvm.fma.f64(double %a, double %b, double %c)
7184 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7185 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7186 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7187
7188Overview:
7189"""""""""
7190
7191The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7192operation.
7193
7194Arguments:
7195""""""""""
7196
7197The argument and return value are floating point numbers of the same
7198type.
7199
7200Semantics:
7201""""""""""
7202
7203This function returns the same values as the libm ``fma`` functions
7204would.
7205
7206'``llvm.fabs.*``' Intrinsic
7207^^^^^^^^^^^^^^^^^^^^^^^^^^^
7208
7209Syntax:
7210"""""""
7211
7212This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
7213floating point or vector of floating point type. Not all targets support
7214all types however.
7215
7216::
7217
7218 declare float @llvm.fabs.f32(float %Val)
7219 declare double @llvm.fabs.f64(double %Val)
7220 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7221 declare fp128 @llvm.fabs.f128(fp128 %Val)
7222 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7223
7224Overview:
7225"""""""""
7226
7227The '``llvm.fabs.*``' intrinsics return the absolute value of the
7228operand.
7229
7230Arguments:
7231""""""""""
7232
7233The argument and return value are floating point numbers of the same
7234type.
7235
7236Semantics:
7237""""""""""
7238
7239This function returns the same values as the libm ``fabs`` functions
7240would, and handles error conditions in the same way.
7241
7242'``llvm.floor.*``' Intrinsic
7243^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7244
7245Syntax:
7246"""""""
7247
7248This is an overloaded intrinsic. You can use ``llvm.floor`` on any
7249floating point or vector of floating point type. Not all targets support
7250all types however.
7251
7252::
7253
7254 declare float @llvm.floor.f32(float %Val)
7255 declare double @llvm.floor.f64(double %Val)
7256 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7257 declare fp128 @llvm.floor.f128(fp128 %Val)
7258 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7259
7260Overview:
7261"""""""""
7262
7263The '``llvm.floor.*``' intrinsics return the floor of the operand.
7264
7265Arguments:
7266""""""""""
7267
7268The argument and return value are floating point numbers of the same
7269type.
7270
7271Semantics:
7272""""""""""
7273
7274This function returns the same values as the libm ``floor`` functions
7275would, and handles error conditions in the same way.
7276
7277'``llvm.ceil.*``' Intrinsic
7278^^^^^^^^^^^^^^^^^^^^^^^^^^^
7279
7280Syntax:
7281"""""""
7282
7283This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
7284floating point or vector of floating point type. Not all targets support
7285all types however.
7286
7287::
7288
7289 declare float @llvm.ceil.f32(float %Val)
7290 declare double @llvm.ceil.f64(double %Val)
7291 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7292 declare fp128 @llvm.ceil.f128(fp128 %Val)
7293 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7294
7295Overview:
7296"""""""""
7297
7298The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
7299
7300Arguments:
7301""""""""""
7302
7303The argument and return value are floating point numbers of the same
7304type.
7305
7306Semantics:
7307""""""""""
7308
7309This function returns the same values as the libm ``ceil`` functions
7310would, and handles error conditions in the same way.
7311
7312'``llvm.trunc.*``' Intrinsic
7313^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7314
7315Syntax:
7316"""""""
7317
7318This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
7319floating point or vector of floating point type. Not all targets support
7320all types however.
7321
7322::
7323
7324 declare float @llvm.trunc.f32(float %Val)
7325 declare double @llvm.trunc.f64(double %Val)
7326 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7327 declare fp128 @llvm.trunc.f128(fp128 %Val)
7328 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7329
7330Overview:
7331"""""""""
7332
7333The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
7334nearest integer not larger in magnitude than the operand.
7335
7336Arguments:
7337""""""""""
7338
7339The argument and return value are floating point numbers of the same
7340type.
7341
7342Semantics:
7343""""""""""
7344
7345This function returns the same values as the libm ``trunc`` functions
7346would, and handles error conditions in the same way.
7347
7348'``llvm.rint.*``' Intrinsic
7349^^^^^^^^^^^^^^^^^^^^^^^^^^^
7350
7351Syntax:
7352"""""""
7353
7354This is an overloaded intrinsic. You can use ``llvm.rint`` on any
7355floating point or vector of floating point type. Not all targets support
7356all types however.
7357
7358::
7359
7360 declare float @llvm.rint.f32(float %Val)
7361 declare double @llvm.rint.f64(double %Val)
7362 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7363 declare fp128 @llvm.rint.f128(fp128 %Val)
7364 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7365
7366Overview:
7367"""""""""
7368
7369The '``llvm.rint.*``' intrinsics returns the operand rounded to the
7370nearest integer. It may raise an inexact floating-point exception if the
7371operand isn't an integer.
7372
7373Arguments:
7374""""""""""
7375
7376The argument and return value are floating point numbers of the same
7377type.
7378
7379Semantics:
7380""""""""""
7381
7382This function returns the same values as the libm ``rint`` functions
7383would, and handles error conditions in the same way.
7384
7385'``llvm.nearbyint.*``' Intrinsic
7386^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7387
7388Syntax:
7389"""""""
7390
7391This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
7392floating point or vector of floating point type. Not all targets support
7393all types however.
7394
7395::
7396
7397 declare float @llvm.nearbyint.f32(float %Val)
7398 declare double @llvm.nearbyint.f64(double %Val)
7399 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7400 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7401 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7402
7403Overview:
7404"""""""""
7405
7406The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
7407nearest integer.
7408
7409Arguments:
7410""""""""""
7411
7412The argument and return value are floating point numbers of the same
7413type.
7414
7415Semantics:
7416""""""""""
7417
7418This function returns the same values as the libm ``nearbyint``
7419functions would, and handles error conditions in the same way.
7420
7421Bit Manipulation Intrinsics
7422---------------------------
7423
7424LLVM provides intrinsics for a few important bit manipulation
7425operations. These allow efficient code generation for some algorithms.
7426
7427'``llvm.bswap.*``' Intrinsics
7428^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7429
7430Syntax:
7431"""""""
7432
7433This is an overloaded intrinsic function. You can use bswap on any
7434integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
7435
7436::
7437
7438 declare i16 @llvm.bswap.i16(i16 <id>)
7439 declare i32 @llvm.bswap.i32(i32 <id>)
7440 declare i64 @llvm.bswap.i64(i64 <id>)
7441
7442Overview:
7443"""""""""
7444
7445The '``llvm.bswap``' family of intrinsics is used to byte swap integer
7446values with an even number of bytes (positive multiple of 16 bits).
7447These are useful for performing operations on data that is not in the
7448target's native byte order.
7449
7450Semantics:
7451""""""""""
7452
7453The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
7454and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
7455intrinsic returns an i32 value that has the four bytes of the input i32
7456swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
7457returned i32 will have its bytes in 3, 2, 1, 0 order. The
7458``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
7459concept to additional even-byte lengths (6 bytes, 8 bytes and more,
7460respectively).
7461
7462'``llvm.ctpop.*``' Intrinsic
7463^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7464
7465Syntax:
7466"""""""
7467
7468This is an overloaded intrinsic. You can use llvm.ctpop on any integer
7469bit width, or on any vector with integer elements. Not all targets
7470support all bit widths or vector types, however.
7471
7472::
7473
7474 declare i8 @llvm.ctpop.i8(i8 <src>)
7475 declare i16 @llvm.ctpop.i16(i16 <src>)
7476 declare i32 @llvm.ctpop.i32(i32 <src>)
7477 declare i64 @llvm.ctpop.i64(i64 <src>)
7478 declare i256 @llvm.ctpop.i256(i256 <src>)
7479 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7480
7481Overview:
7482"""""""""
7483
7484The '``llvm.ctpop``' family of intrinsics counts the number of bits set
7485in a value.
7486
7487Arguments:
7488""""""""""
7489
7490The only argument is the value to be counted. The argument may be of any
7491integer type, or a vector with integer elements. The return type must
7492match the argument type.
7493
7494Semantics:
7495""""""""""
7496
7497The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
7498each element of a vector.
7499
7500'``llvm.ctlz.*``' Intrinsic
7501^^^^^^^^^^^^^^^^^^^^^^^^^^^
7502
7503Syntax:
7504"""""""
7505
7506This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
7507integer bit width, or any vector whose elements are integers. Not all
7508targets support all bit widths or vector types, however.
7509
7510::
7511
7512 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7513 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7514 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7515 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7516 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7517 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7518
7519Overview:
7520"""""""""
7521
7522The '``llvm.ctlz``' family of intrinsic functions counts the number of
7523leading zeros in a variable.
7524
7525Arguments:
7526""""""""""
7527
7528The first argument is the value to be counted. This argument may be of
7529any integer type, or a vectory with integer element type. The return
7530type must match the first argument type.
7531
7532The second argument must be a constant and is a flag to indicate whether
7533the intrinsic should ensure that a zero as the first argument produces a
7534defined result. Historically some architectures did not provide a
7535defined result for zero values as efficiently, and many algorithms are
7536now predicated on avoiding zero-value inputs.
7537
7538Semantics:
7539""""""""""
7540
7541The '``llvm.ctlz``' intrinsic counts the leading (most significant)
7542zeros in a variable, or within each element of the vector. If
7543``src == 0`` then the result is the size in bits of the type of ``src``
7544if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7545``llvm.ctlz(i32 2) = 30``.
7546
7547'``llvm.cttz.*``' Intrinsic
7548^^^^^^^^^^^^^^^^^^^^^^^^^^^
7549
7550Syntax:
7551"""""""
7552
7553This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
7554integer bit width, or any vector of integer elements. Not all targets
7555support all bit widths or vector types, however.
7556
7557::
7558
7559 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7560 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7561 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7562 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7563 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7564 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7565
7566Overview:
7567"""""""""
7568
7569The '``llvm.cttz``' family of intrinsic functions counts the number of
7570trailing zeros.
7571
7572Arguments:
7573""""""""""
7574
7575The first argument is the value to be counted. This argument may be of
7576any integer type, or a vectory with integer element type. The return
7577type must match the first argument type.
7578
7579The second argument must be a constant and is a flag to indicate whether
7580the intrinsic should ensure that a zero as the first argument produces a
7581defined result. Historically some architectures did not provide a
7582defined result for zero values as efficiently, and many algorithms are
7583now predicated on avoiding zero-value inputs.
7584
7585Semantics:
7586""""""""""
7587
7588The '``llvm.cttz``' intrinsic counts the trailing (least significant)
7589zeros in a variable, or within each element of a vector. If ``src == 0``
7590then the result is the size in bits of the type of ``src`` if
7591``is_zero_undef == 0`` and ``undef`` otherwise. For example,
7592``llvm.cttz(2) = 1``.
7593
7594Arithmetic with Overflow Intrinsics
7595-----------------------------------
7596
7597LLVM provides intrinsics for some arithmetic with overflow operations.
7598
7599'``llvm.sadd.with.overflow.*``' Intrinsics
7600^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7601
7602Syntax:
7603"""""""
7604
7605This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
7606on any integer bit width.
7607
7608::
7609
7610 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7611 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7612 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7613
7614Overview:
7615"""""""""
7616
7617The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
7618a signed addition of the two arguments, and indicate whether an overflow
7619occurred during the signed summation.
7620
7621Arguments:
7622""""""""""
7623
7624The arguments (%a and %b) and the first element of the result structure
7625may be of integer types of any bit width, but they must have the same
7626bit width. The second element of the result structure must be of type
7627``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7628addition.
7629
7630Semantics:
7631""""""""""
7632
7633The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
Dmitri Gribenkoae4a9ae2013-01-19 20:34:20 +00007634a signed addition of the two variables. They return a structure --- the
Sean Silvaf722b002012-12-07 10:36:55 +00007635first element of which is the signed summation, and the second element
7636of which is a bit specifying if the signed summation resulted in an
7637overflow.
7638
7639Examples:
7640"""""""""
7641
7642.. code-block:: llvm
7643
7644 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7645 %sum = extractvalue {i32, i1} %res, 0
7646 %obit = extractvalue {i32, i1} %res, 1
7647 br i1 %obit, label %overflow, label %normal
7648
7649'``llvm.uadd.with.overflow.*``' Intrinsics
7650^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7651
7652Syntax:
7653"""""""
7654
7655This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
7656on any integer bit width.
7657
7658::
7659
7660 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7661 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7662 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7663
7664Overview:
7665"""""""""
7666
7667The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
7668an unsigned addition of the two arguments, and indicate whether a carry
7669occurred during the unsigned summation.
7670
7671Arguments:
7672""""""""""
7673
7674The arguments (%a and %b) and the first element of the result structure
7675may be of integer types of any bit width, but they must have the same
7676bit width. The second element of the result structure must be of type
7677``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7678addition.
7679
7680Semantics:
7681""""""""""
7682
7683The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
Dmitri Gribenkoae4a9ae2013-01-19 20:34:20 +00007684an unsigned addition of the two arguments. They return a structure --- the
Sean Silvaf722b002012-12-07 10:36:55 +00007685first element of which is the sum, and the second element of which is a
7686bit specifying if the unsigned summation resulted in a carry.
7687
7688Examples:
7689"""""""""
7690
7691.. code-block:: llvm
7692
7693 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7694 %sum = extractvalue {i32, i1} %res, 0
7695 %obit = extractvalue {i32, i1} %res, 1
7696 br i1 %obit, label %carry, label %normal
7697
7698'``llvm.ssub.with.overflow.*``' Intrinsics
7699^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7700
7701Syntax:
7702"""""""
7703
7704This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
7705on any integer bit width.
7706
7707::
7708
7709 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7710 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7711 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7712
7713Overview:
7714"""""""""
7715
7716The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
7717a signed subtraction of the two arguments, and indicate whether an
7718overflow occurred during the signed subtraction.
7719
7720Arguments:
7721""""""""""
7722
7723The arguments (%a and %b) and the first element of the result structure
7724may be of integer types of any bit width, but they must have the same
7725bit width. The second element of the result structure must be of type
7726``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7727subtraction.
7728
7729Semantics:
7730""""""""""
7731
7732The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
Dmitri Gribenkoae4a9ae2013-01-19 20:34:20 +00007733a signed subtraction of the two arguments. They return a structure --- the
Sean Silvaf722b002012-12-07 10:36:55 +00007734first element of which is the subtraction, and the second element of
7735which is a bit specifying if the signed subtraction resulted in an
7736overflow.
7737
7738Examples:
7739"""""""""
7740
7741.. code-block:: llvm
7742
7743 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7744 %sum = extractvalue {i32, i1} %res, 0
7745 %obit = extractvalue {i32, i1} %res, 1
7746 br i1 %obit, label %overflow, label %normal
7747
7748'``llvm.usub.with.overflow.*``' Intrinsics
7749^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7750
7751Syntax:
7752"""""""
7753
7754This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
7755on any integer bit width.
7756
7757::
7758
7759 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7760 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7761 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7762
7763Overview:
7764"""""""""
7765
7766The '``llvm.usub.with.overflow``' family of intrinsic functions perform
7767an unsigned subtraction of the two arguments, and indicate whether an
7768overflow occurred during the unsigned subtraction.
7769
7770Arguments:
7771""""""""""
7772
7773The arguments (%a and %b) and the first element of the result structure
7774may be of integer types of any bit width, but they must have the same
7775bit width. The second element of the result structure must be of type
7776``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7777subtraction.
7778
7779Semantics:
7780""""""""""
7781
7782The '``llvm.usub.with.overflow``' family of intrinsic functions perform
Dmitri Gribenkoae4a9ae2013-01-19 20:34:20 +00007783an unsigned subtraction of the two arguments. They return a structure ---
Sean Silvaf722b002012-12-07 10:36:55 +00007784the first element of which is the subtraction, and the second element of
7785which is a bit specifying if the unsigned subtraction resulted in an
7786overflow.
7787
7788Examples:
7789"""""""""
7790
7791.. code-block:: llvm
7792
7793 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7794 %sum = extractvalue {i32, i1} %res, 0
7795 %obit = extractvalue {i32, i1} %res, 1
7796 br i1 %obit, label %overflow, label %normal
7797
7798'``llvm.smul.with.overflow.*``' Intrinsics
7799^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7800
7801Syntax:
7802"""""""
7803
7804This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
7805on any integer bit width.
7806
7807::
7808
7809 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7810 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7811 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7812
7813Overview:
7814"""""""""
7815
7816The '``llvm.smul.with.overflow``' family of intrinsic functions perform
7817a signed multiplication of the two arguments, and indicate whether an
7818overflow occurred during the signed multiplication.
7819
7820Arguments:
7821""""""""""
7822
7823The arguments (%a and %b) and the first element of the result structure
7824may be of integer types of any bit width, but they must have the same
7825bit width. The second element of the result structure must be of type
7826``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
7827multiplication.
7828
7829Semantics:
7830""""""""""
7831
7832The '``llvm.smul.with.overflow``' family of intrinsic functions perform
Dmitri Gribenkoae4a9ae2013-01-19 20:34:20 +00007833a signed multiplication of the two arguments. They return a structure ---
Sean Silvaf722b002012-12-07 10:36:55 +00007834the first element of which is the multiplication, and the second element
7835of which is a bit specifying if the signed multiplication resulted in an
7836overflow.
7837
7838Examples:
7839"""""""""
7840
7841.. code-block:: llvm
7842
7843 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7844 %sum = extractvalue {i32, i1} %res, 0
7845 %obit = extractvalue {i32, i1} %res, 1
7846 br i1 %obit, label %overflow, label %normal
7847
7848'``llvm.umul.with.overflow.*``' Intrinsics
7849^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7850
7851Syntax:
7852"""""""
7853
7854This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
7855on any integer bit width.
7856
7857::
7858
7859 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7860 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7861 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7862
7863Overview:
7864"""""""""
7865
7866The '``llvm.umul.with.overflow``' family of intrinsic functions perform
7867a unsigned multiplication of the two arguments, and indicate whether an
7868overflow occurred during the unsigned multiplication.
7869
7870Arguments:
7871""""""""""
7872
7873The arguments (%a and %b) and the first element of the result structure
7874may be of integer types of any bit width, but they must have the same
7875bit width. The second element of the result structure must be of type
7876``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
7877multiplication.
7878
7879Semantics:
7880""""""""""
7881
7882The '``llvm.umul.with.overflow``' family of intrinsic functions perform
Dmitri Gribenkoae4a9ae2013-01-19 20:34:20 +00007883an unsigned multiplication of the two arguments. They return a structure ---
7884the first element of which is the multiplication, and the second
Sean Silvaf722b002012-12-07 10:36:55 +00007885element of which is a bit specifying if the unsigned multiplication
7886resulted in an overflow.
7887
7888Examples:
7889"""""""""
7890
7891.. code-block:: llvm
7892
7893 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7894 %sum = extractvalue {i32, i1} %res, 0
7895 %obit = extractvalue {i32, i1} %res, 1
7896 br i1 %obit, label %overflow, label %normal
7897
7898Specialised Arithmetic Intrinsics
7899---------------------------------
7900
7901'``llvm.fmuladd.*``' Intrinsic
7902^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7903
7904Syntax:
7905"""""""
7906
7907::
7908
7909 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
7910 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
7911
7912Overview:
7913"""""""""
7914
7915The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
Lang Hamesb0ec16b2013-01-17 00:00:49 +00007916expressions that can be fused if the code generator determines that (a) the
7917target instruction set has support for a fused operation, and (b) that the
7918fused operation is more efficient than the equivalent, separate pair of mul
7919and add instructions.
Sean Silvaf722b002012-12-07 10:36:55 +00007920
7921Arguments:
7922""""""""""
7923
7924The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
7925multiplicands, a and b, and an addend c.
7926
7927Semantics:
7928""""""""""
7929
7930The expression:
7931
7932::
7933
7934 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
7935
7936is equivalent to the expression a \* b + c, except that rounding will
7937not be performed between the multiplication and addition steps if the
7938code generator fuses the operations. Fusion is not guaranteed, even if
7939the target platform supports it. If a fused multiply-add is required the
7940corresponding llvm.fma.\* intrinsic function should be used instead.
7941
7942Examples:
7943"""""""""
7944
7945.. code-block:: llvm
7946
7947 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
7948
7949Half Precision Floating Point Intrinsics
7950----------------------------------------
7951
7952For most target platforms, half precision floating point is a
7953storage-only format. This means that it is a dense encoding (in memory)
7954but does not support computation in the format.
7955
7956This means that code must first load the half-precision floating point
7957value as an i16, then convert it to float with
7958:ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
7959then be performed on the float value (including extending to double
7960etc). To store the value back to memory, it is first converted to float
7961if needed, then converted to i16 with
7962:ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
7963i16 value.
7964
7965.. _int_convert_to_fp16:
7966
7967'``llvm.convert.to.fp16``' Intrinsic
7968^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7969
7970Syntax:
7971"""""""
7972
7973::
7974
7975 declare i16 @llvm.convert.to.fp16(f32 %a)
7976
7977Overview:
7978"""""""""
7979
7980The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
7981from single precision floating point format to half precision floating
7982point format.
7983
7984Arguments:
7985""""""""""
7986
7987The intrinsic function contains single argument - the value to be
7988converted.
7989
7990Semantics:
7991""""""""""
7992
7993The '``llvm.convert.to.fp16``' intrinsic function performs a conversion
7994from single precision floating point format to half precision floating
7995point format. The return value is an ``i16`` which contains the
7996converted number.
7997
7998Examples:
7999"""""""""
8000
8001.. code-block:: llvm
8002
8003 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8004 store i16 %res, i16* @x, align 2
8005
8006.. _int_convert_from_fp16:
8007
8008'``llvm.convert.from.fp16``' Intrinsic
8009^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8010
8011Syntax:
8012"""""""
8013
8014::
8015
8016 declare f32 @llvm.convert.from.fp16(i16 %a)
8017
8018Overview:
8019"""""""""
8020
8021The '``llvm.convert.from.fp16``' intrinsic function performs a
8022conversion from half precision floating point format to single precision
8023floating point format.
8024
8025Arguments:
8026""""""""""
8027
8028The intrinsic function contains single argument - the value to be
8029converted.
8030
8031Semantics:
8032""""""""""
8033
8034The '``llvm.convert.from.fp16``' intrinsic function performs a
8035conversion from half single precision floating point format to single
8036precision floating point format. The input half-float value is
8037represented by an ``i16`` value.
8038
8039Examples:
8040"""""""""
8041
8042.. code-block:: llvm
8043
8044 %a = load i16* @x, align 2
8045 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8046
8047Debugger Intrinsics
8048-------------------
8049
8050The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8051prefix), are described in the `LLVM Source Level
8052Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8053document.
8054
8055Exception Handling Intrinsics
8056-----------------------------
8057
8058The LLVM exception handling intrinsics (which all start with
8059``llvm.eh.`` prefix), are described in the `LLVM Exception
8060Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8061
8062.. _int_trampoline:
8063
8064Trampoline Intrinsics
8065---------------------
8066
8067These intrinsics make it possible to excise one parameter, marked with
8068the :ref:`nest <nest>` attribute, from a function. The result is a
8069callable function pointer lacking the nest parameter - the caller does
8070not need to provide a value for it. Instead, the value to use is stored
8071in advance in a "trampoline", a block of memory usually allocated on the
8072stack, which also contains code to splice the nest value into the
8073argument list. This is used to implement the GCC nested function address
8074extension.
8075
8076For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8077then the resulting function pointer has signature ``i32 (i32, i32)*``.
8078It can be created as follows:
8079
8080.. code-block:: llvm
8081
8082 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8083 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8084 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8085 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8086 %fp = bitcast i8* %p to i32 (i32, i32)*
8087
8088The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8089``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8090
8091.. _int_it:
8092
8093'``llvm.init.trampoline``' Intrinsic
8094^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8095
8096Syntax:
8097"""""""
8098
8099::
8100
8101 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8102
8103Overview:
8104"""""""""
8105
8106This fills the memory pointed to by ``tramp`` with executable code,
8107turning it into a trampoline.
8108
8109Arguments:
8110""""""""""
8111
8112The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8113pointers. The ``tramp`` argument must point to a sufficiently large and
8114sufficiently aligned block of memory; this memory is written to by the
8115intrinsic. Note that the size and the alignment are target-specific -
8116LLVM currently provides no portable way of determining them, so a
8117front-end that generates this intrinsic needs to have some
8118target-specific knowledge. The ``func`` argument must hold a function
8119bitcast to an ``i8*``.
8120
8121Semantics:
8122""""""""""
8123
8124The block of memory pointed to by ``tramp`` is filled with target
8125dependent code, turning it into a function. Then ``tramp`` needs to be
8126passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8127be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8128function's signature is the same as that of ``func`` with any arguments
8129marked with the ``nest`` attribute removed. At most one such ``nest``
8130argument is allowed, and it must be of pointer type. Calling the new
8131function is equivalent to calling ``func`` with the same argument list,
8132but with ``nval`` used for the missing ``nest`` argument. If, after
8133calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8134modified, then the effect of any later call to the returned function
8135pointer is undefined.
8136
8137.. _int_at:
8138
8139'``llvm.adjust.trampoline``' Intrinsic
8140^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8141
8142Syntax:
8143"""""""
8144
8145::
8146
8147 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8148
8149Overview:
8150"""""""""
8151
8152This performs any required machine-specific adjustment to the address of
8153a trampoline (passed as ``tramp``).
8154
8155Arguments:
8156""""""""""
8157
8158``tramp`` must point to a block of memory which already has trampoline
8159code filled in by a previous call to
8160:ref:`llvm.init.trampoline <int_it>`.
8161
8162Semantics:
8163""""""""""
8164
8165On some architectures the address of the code to be executed needs to be
8166different to the address where the trampoline is actually stored. This
8167intrinsic returns the executable address corresponding to ``tramp``
8168after performing the required machine specific adjustments. The pointer
8169returned can then be :ref:`bitcast and executed <int_trampoline>`.
8170
8171Memory Use Markers
8172------------------
8173
8174This class of intrinsics exists to information about the lifetime of
8175memory objects and ranges where variables are immutable.
8176
8177'``llvm.lifetime.start``' Intrinsic
8178^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8179
8180Syntax:
8181"""""""
8182
8183::
8184
8185 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8186
8187Overview:
8188"""""""""
8189
8190The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
8191object's lifetime.
8192
8193Arguments:
8194""""""""""
8195
8196The first argument is a constant integer representing the size of the
8197object, or -1 if it is variable sized. The second argument is a pointer
8198to the object.
8199
8200Semantics:
8201""""""""""
8202
8203This intrinsic indicates that before this point in the code, the value
8204of the memory pointed to by ``ptr`` is dead. This means that it is known
8205to never be used and has an undefined value. A load from the pointer
8206that precedes this intrinsic can be replaced with ``'undef'``.
8207
8208'``llvm.lifetime.end``' Intrinsic
8209^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8210
8211Syntax:
8212"""""""
8213
8214::
8215
8216 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8217
8218Overview:
8219"""""""""
8220
8221The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
8222object's lifetime.
8223
8224Arguments:
8225""""""""""
8226
8227The first argument is a constant integer representing the size of the
8228object, or -1 if it is variable sized. The second argument is a pointer
8229to the object.
8230
8231Semantics:
8232""""""""""
8233
8234This intrinsic indicates that after this point in the code, the value of
8235the memory pointed to by ``ptr`` is dead. This means that it is known to
8236never be used and has an undefined value. Any stores into the memory
8237object following this intrinsic may be removed as dead.
8238
8239'``llvm.invariant.start``' Intrinsic
8240^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8241
8242Syntax:
8243"""""""
8244
8245::
8246
8247 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8248
8249Overview:
8250"""""""""
8251
8252The '``llvm.invariant.start``' intrinsic specifies that the contents of
8253a memory object will not change.
8254
8255Arguments:
8256""""""""""
8257
8258The first argument is a constant integer representing the size of the
8259object, or -1 if it is variable sized. The second argument is a pointer
8260to the object.
8261
8262Semantics:
8263""""""""""
8264
8265This intrinsic indicates that until an ``llvm.invariant.end`` that uses
8266the return value, the referenced memory location is constant and
8267unchanging.
8268
8269'``llvm.invariant.end``' Intrinsic
8270^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8271
8272Syntax:
8273"""""""
8274
8275::
8276
8277 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8278
8279Overview:
8280"""""""""
8281
8282The '``llvm.invariant.end``' intrinsic specifies that the contents of a
8283memory object are mutable.
8284
8285Arguments:
8286""""""""""
8287
8288The first argument is the matching ``llvm.invariant.start`` intrinsic.
8289The second argument is a constant integer representing the size of the
8290object, or -1 if it is variable sized and the third argument is a
8291pointer to the object.
8292
8293Semantics:
8294""""""""""
8295
8296This intrinsic indicates that the memory is mutable again.
8297
8298General Intrinsics
8299------------------
8300
8301This class of intrinsics is designed to be generic and has no specific
8302purpose.
8303
8304'``llvm.var.annotation``' Intrinsic
8305^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8306
8307Syntax:
8308"""""""
8309
8310::
8311
8312 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8313
8314Overview:
8315"""""""""
8316
8317The '``llvm.var.annotation``' intrinsic.
8318
8319Arguments:
8320""""""""""
8321
8322The first argument is a pointer to a value, the second is a pointer to a
8323global string, the third is a pointer to a global string which is the
8324source file name, and the last argument is the line number.
8325
8326Semantics:
8327""""""""""
8328
8329This intrinsic allows annotation of local variables with arbitrary
8330strings. This can be useful for special purpose optimizations that want
8331to look for these annotations. These have no other defined use; they are
8332ignored by code generation and optimization.
8333
8334'``llvm.annotation.*``' Intrinsic
8335^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8336
8337Syntax:
8338"""""""
8339
8340This is an overloaded intrinsic. You can use '``llvm.annotation``' on
8341any integer bit width.
8342
8343::
8344
8345 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8346 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8347 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8348 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8349 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8350
8351Overview:
8352"""""""""
8353
8354The '``llvm.annotation``' intrinsic.
8355
8356Arguments:
8357""""""""""
8358
8359The first argument is an integer value (result of some expression), the
8360second is a pointer to a global string, the third is a pointer to a
8361global string which is the source file name, and the last argument is
8362the line number. It returns the value of the first argument.
8363
8364Semantics:
8365""""""""""
8366
8367This intrinsic allows annotations to be put on arbitrary expressions
8368with arbitrary strings. This can be useful for special purpose
8369optimizations that want to look for these annotations. These have no
8370other defined use; they are ignored by code generation and optimization.
8371
8372'``llvm.trap``' Intrinsic
8373^^^^^^^^^^^^^^^^^^^^^^^^^
8374
8375Syntax:
8376"""""""
8377
8378::
8379
8380 declare void @llvm.trap() noreturn nounwind
8381
8382Overview:
8383"""""""""
8384
8385The '``llvm.trap``' intrinsic.
8386
8387Arguments:
8388""""""""""
8389
8390None.
8391
8392Semantics:
8393""""""""""
8394
8395This intrinsic is lowered to the target dependent trap instruction. If
8396the target does not have a trap instruction, this intrinsic will be
8397lowered to a call of the ``abort()`` function.
8398
8399'``llvm.debugtrap``' Intrinsic
8400^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8401
8402Syntax:
8403"""""""
8404
8405::
8406
8407 declare void @llvm.debugtrap() nounwind
8408
8409Overview:
8410"""""""""
8411
8412The '``llvm.debugtrap``' intrinsic.
8413
8414Arguments:
8415""""""""""
8416
8417None.
8418
8419Semantics:
8420""""""""""
8421
8422This intrinsic is lowered to code which is intended to cause an
8423execution trap with the intention of requesting the attention of a
8424debugger.
8425
8426'``llvm.stackprotector``' Intrinsic
8427^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8428
8429Syntax:
8430"""""""
8431
8432::
8433
8434 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8435
8436Overview:
8437"""""""""
8438
8439The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
8440onto the stack at ``slot``. The stack slot is adjusted to ensure that it
8441is placed on the stack before local variables.
8442
8443Arguments:
8444""""""""""
8445
8446The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
8447The first argument is the value loaded from the stack guard
8448``@__stack_chk_guard``. The second variable is an ``alloca`` that has
8449enough space to hold the value of the guard.
8450
8451Semantics:
8452""""""""""
8453
8454This intrinsic causes the prologue/epilogue inserter to force the
8455position of the ``AllocaInst`` stack slot to be before local variables
8456on the stack. This is to ensure that if a local variable on the stack is
8457overwritten, it will destroy the value of the guard. When the function
8458exits, the guard on the stack is checked against the original guard. If
8459they are different, then the program aborts by calling the
8460``__stack_chk_fail()`` function.
8461
8462'``llvm.objectsize``' Intrinsic
8463^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8464
8465Syntax:
8466"""""""
8467
8468::
8469
8470 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8471 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8472
8473Overview:
8474"""""""""
8475
8476The ``llvm.objectsize`` intrinsic is designed to provide information to
8477the optimizers to determine at compile time whether a) an operation
8478(like memcpy) will overflow a buffer that corresponds to an object, or
8479b) that a runtime check for overflow isn't necessary. An object in this
8480context means an allocation of a specific class, structure, array, or
8481other object.
8482
8483Arguments:
8484""""""""""
8485
8486The ``llvm.objectsize`` intrinsic takes two arguments. The first
8487argument is a pointer to or into the ``object``. The second argument is
8488a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
8489or -1 (if false) when the object size is unknown. The second argument
8490only accepts constants.
8491
8492Semantics:
8493""""""""""
8494
8495The ``llvm.objectsize`` intrinsic is lowered to a constant representing
8496the size of the object concerned. If the size cannot be determined at
8497compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
8498on the ``min`` argument).
8499
8500'``llvm.expect``' Intrinsic
8501^^^^^^^^^^^^^^^^^^^^^^^^^^^
8502
8503Syntax:
8504"""""""
8505
8506::
8507
8508 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8509 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8510
8511Overview:
8512"""""""""
8513
8514The ``llvm.expect`` intrinsic provides information about expected (the
8515most probable) value of ``val``, which can be used by optimizers.
8516
8517Arguments:
8518""""""""""
8519
8520The ``llvm.expect`` intrinsic takes two arguments. The first argument is
8521a value. The second argument is an expected value, this needs to be a
8522constant value, variables are not allowed.
8523
8524Semantics:
8525""""""""""
8526
8527This intrinsic is lowered to the ``val``.
8528
8529'``llvm.donothing``' Intrinsic
8530^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8531
8532Syntax:
8533"""""""
8534
8535::
8536
8537 declare void @llvm.donothing() nounwind readnone
8538
8539Overview:
8540"""""""""
8541
8542The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
8543only intrinsic that can be called with an invoke instruction.
8544
8545Arguments:
8546""""""""""
8547
8548None.
8549
8550Semantics:
8551""""""""""
8552
8553This intrinsic does nothing, and it's removed by optimizers and ignored
8554by codegen.