1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][-a-zA-Z$._][-a-zA-Z$._0-9]*``'. Identifiers that 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. The ``"\01"`` prefix
83 can be used on global variables to suppress mangling.
84 #. Unnamed values are represented as an unsigned numeric value with
85 their prefix. For example, ``%12``, ``@2``, ``%44``.
86 #. Constants, which are described in the section Constants_ below.
88 LLVM requires that values start with a prefix for two reasons: Compilers
89 don't need to worry about name clashes with reserved words, and the set
90 of reserved words may be expanded in the future without penalty.
91 Additionally, unnamed identifiers allow a compiler to quickly come up
92 with a temporary variable without having to avoid symbol table
95 Reserved words in LLVM are very similar to reserved words in other
96 languages. There are keywords for different opcodes ('``add``',
97 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
98 '``i32``', etc...), and others. These reserved words cannot conflict
99 with variable names, because none of them start with a prefix character
100 (``'%'`` or ``'@'``).
102 Here is an example of LLVM code to multiply the integer variable
109 %result = mul i32 %X, 8
111 After strength reduction:
115 %result = shl i32 %X, 3
121 %0 = add i32 %X, %X ; yields i32:%0
122 %1 = add i32 %0, %0 ; yields i32:%1
123 %result = add i32 %1, %1
125 This last way of multiplying ``%X`` by 8 illustrates several important
126 lexical features of LLVM:
128 #. Comments are delimited with a '``;``' and go until the end of line.
129 #. Unnamed temporaries are created when the result of a computation is
130 not assigned to a named value.
131 #. Unnamed temporaries are numbered sequentially (using a per-function
132 incrementing counter, starting with 0). Note that basic blocks and unnamed
133 function parameters are included in this numbering. For example, if the
134 entry basic block is not given a label name and all function parameters are
135 named, then it will get number 0.
137 It also shows a convention that we follow in this document. When
138 demonstrating instructions, we will follow an instruction with a comment
139 that defines the type and name of value produced.
147 LLVM programs are composed of ``Module``'s, each of which is a
148 translation unit of the input programs. Each module consists of
149 functions, global variables, and symbol table entries. Modules may be
150 combined together with the LLVM linker, which merges function (and
151 global variable) definitions, resolves forward declarations, and merges
152 symbol table entries. Here is an example of the "hello world" module:
156 ; Declare the string constant as a global constant.
157 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
159 ; External declaration of the puts function
160 declare i32 @puts(i8* nocapture) nounwind
162 ; Definition of main function
163 define i32 @main() { ; i32()*
164 ; Convert [13 x i8]* to i8 *...
165 %cast210 = getelementptr [13 x i8], [13 x i8]* @.str, i64 0, i64 0
167 ; Call puts function to write out the string to stdout.
168 call i32 @puts(i8* %cast210)
173 !0 = !{i32 42, null, !"string"}
176 This example is made up of a :ref:`global variable <globalvars>` named
177 "``.str``", an external declaration of the "``puts``" function, a
178 :ref:`function definition <functionstructure>` for "``main``" and
179 :ref:`named metadata <namedmetadatastructure>` "``foo``".
181 In general, a module is made up of a list of global values (where both
182 functions and global variables are global values). Global values are
183 represented by a pointer to a memory location (in this case, a pointer
184 to an array of char, and a pointer to a function), and have one of the
185 following :ref:`linkage types <linkage>`.
192 All Global Variables and Functions have one of the following types of
196 Global values with "``private``" linkage are only directly
197 accessible by objects in the current module. In particular, linking
198 code into a module with an private global value may cause the
199 private to be renamed as necessary to avoid collisions. Because the
200 symbol is private to the module, all references can be updated. This
201 doesn't show up in any symbol table in the object file.
203 Similar to private, but the value shows as a local symbol
204 (``STB_LOCAL`` in the case of ELF) in the object file. This
205 corresponds to the notion of the '``static``' keyword in C.
206 ``available_externally``
207 Globals with "``available_externally``" linkage are never emitted
208 into the object file corresponding to the LLVM module. They exist to
209 allow inlining and other optimizations to take place given knowledge
210 of the definition of the global, which is known to be somewhere
211 outside the module. Globals with ``available_externally`` linkage
212 are allowed to be discarded at will, and are otherwise the same as
213 ``linkonce_odr``. This linkage type is only allowed on definitions,
216 Globals with "``linkonce``" linkage are merged with other globals of
217 the same name when linkage occurs. This can be used to implement
218 some forms of inline functions, templates, or other code which must
219 be generated in each translation unit that uses it, but where the
220 body may be overridden with a more definitive definition later.
221 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
222 that ``linkonce`` linkage does not actually allow the optimizer to
223 inline the body of this function into callers because it doesn't
224 know if this definition of the function is the definitive definition
225 within the program or whether it will be overridden by a stronger
226 definition. To enable inlining and other optimizations, use
227 "``linkonce_odr``" linkage.
229 "``weak``" linkage has the same merging semantics as ``linkonce``
230 linkage, except that unreferenced globals with ``weak`` linkage may
231 not be discarded. This is used for globals that are declared "weak"
234 "``common``" linkage is most similar to "``weak``" linkage, but they
235 are used for tentative definitions in C, such as "``int X;``" at
236 global scope. Symbols with "``common``" linkage are merged in the
237 same way as ``weak symbols``, and they may not be deleted if
238 unreferenced. ``common`` symbols may not have an explicit section,
239 must have a zero initializer, and may not be marked
240 ':ref:`constant <globalvars>`'. Functions and aliases may not have
243 .. _linkage_appending:
246 "``appending``" linkage may only be applied to global variables of
247 pointer to array type. When two global variables with appending
248 linkage are linked together, the two global arrays are appended
249 together. This is the LLVM, typesafe, equivalent of having the
250 system linker append together "sections" with identical names when
253 The semantics of this linkage follow the ELF object file model: the
254 symbol is weak until linked, if not linked, the symbol becomes null
255 instead of being an undefined reference.
256 ``linkonce_odr``, ``weak_odr``
257 Some languages allow differing globals to be merged, such as two
258 functions with different semantics. Other languages, such as
259 ``C++``, ensure that only equivalent globals are ever merged (the
260 "one definition rule" --- "ODR"). Such languages can use the
261 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
262 global will only be merged with equivalent globals. These linkage
263 types are otherwise the same as their non-``odr`` versions.
265 If none of the above identifiers are used, the global is externally
266 visible, meaning that it participates in linkage and can be used to
267 resolve external symbol references.
269 It is illegal for a function *declaration* to have any linkage type
270 other than ``external`` or ``extern_weak``.
277 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
278 :ref:`invokes <i_invoke>` can all have an optional calling convention
279 specified for the call. The calling convention of any pair of dynamic
280 caller/callee must match, or the behavior of the program is undefined.
281 The following calling conventions are supported by LLVM, and more may be
284 "``ccc``" - The C calling convention
285 This calling convention (the default if no other calling convention
286 is specified) matches the target C calling conventions. This calling
287 convention supports varargs function calls and tolerates some
288 mismatch in the declared prototype and implemented declaration of
289 the function (as does normal C).
290 "``fastcc``" - The fast calling convention
291 This calling convention attempts to make calls as fast as possible
292 (e.g. by passing things in registers). This calling convention
293 allows the target to use whatever tricks it wants to produce fast
294 code for the target, without having to conform to an externally
295 specified ABI (Application Binary Interface). `Tail calls can only
296 be optimized when this, the GHC or the HiPE convention is
297 used. <CodeGenerator.html#id80>`_ This calling convention does not
298 support varargs and requires the prototype of all callees to exactly
299 match the prototype of the function definition.
300 "``coldcc``" - The cold calling convention
301 This calling convention attempts to make code in the caller as
302 efficient as possible under the assumption that the call is not
303 commonly executed. As such, these calls often preserve all registers
304 so that the call does not break any live ranges in the caller side.
305 This calling convention does not support varargs and requires the
306 prototype of all callees to exactly match the prototype of the
307 function definition. Furthermore the inliner doesn't consider such function
309 "``cc 10``" - GHC convention
310 This calling convention has been implemented specifically for use by
311 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
312 It passes everything in registers, going to extremes to achieve this
313 by disabling callee save registers. This calling convention should
314 not be used lightly but only for specific situations such as an
315 alternative to the *register pinning* performance technique often
316 used when implementing functional programming languages. At the
317 moment only X86 supports this convention and it has the following
320 - On *X86-32* only supports up to 4 bit type parameters. No
321 floating point types are supported.
322 - On *X86-64* only supports up to 10 bit type parameters and 6
323 floating point parameters.
325 This calling convention supports `tail call
326 optimization <CodeGenerator.html#id80>`_ but requires both the
327 caller and callee are using it.
328 "``cc 11``" - The HiPE calling convention
329 This calling convention has been implemented specifically for use by
330 the `High-Performance Erlang
331 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
332 native code compiler of the `Ericsson's Open Source Erlang/OTP
333 system <http://www.erlang.org/download.shtml>`_. It uses more
334 registers for argument passing than the ordinary C calling
335 convention and defines no callee-saved registers. The calling
336 convention properly supports `tail call
337 optimization <CodeGenerator.html#id80>`_ but requires that both the
338 caller and the callee use it. It uses a *register pinning*
339 mechanism, similar to GHC's convention, for keeping frequently
340 accessed runtime components pinned to specific hardware registers.
341 At the moment only X86 supports this convention (both 32 and 64
343 "``webkit_jscc``" - WebKit's JavaScript calling convention
344 This calling convention has been implemented for `WebKit FTL JIT
345 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
346 stack right to left (as cdecl does), and returns a value in the
347 platform's customary return register.
348 "``anyregcc``" - Dynamic calling convention for code patching
349 This is a special convention that supports patching an arbitrary code
350 sequence in place of a call site. This convention forces the call
351 arguments into registers but allows them to be dynamically
352 allocated. This can currently only be used with calls to
353 llvm.experimental.patchpoint because only this intrinsic records
354 the location of its arguments in a side table. See :doc:`StackMaps`.
355 "``preserve_mostcc``" - The `PreserveMost` calling convention
356 This calling convention attempts to make the code in the caller as
357 unintrusive as possible. This convention behaves identically to the `C`
358 calling convention on how arguments and return values are passed, but it
359 uses a different set of caller/callee-saved registers. This alleviates the
360 burden of saving and recovering a large register set before and after the
361 call in the caller. If the arguments are passed in callee-saved registers,
362 then they will be preserved by the callee across the call. This doesn't
363 apply for values returned in callee-saved registers.
365 - On X86-64 the callee preserves all general purpose registers, except for
366 R11. R11 can be used as a scratch register. Floating-point registers
367 (XMMs/YMMs) are not preserved and need to be saved by the caller.
369 The idea behind this convention is to support calls to runtime functions
370 that have a hot path and a cold path. The hot path is usually a small piece
371 of code that doesn't use many registers. The cold path might need to call out to
372 another function and therefore only needs to preserve the caller-saved
373 registers, which haven't already been saved by the caller. The
374 `PreserveMost` calling convention is very similar to the `cold` calling
375 convention in terms of caller/callee-saved registers, but they are used for
376 different types of function calls. `coldcc` is for function calls that are
377 rarely executed, whereas `preserve_mostcc` function calls are intended to be
378 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
379 doesn't prevent the inliner from inlining the function call.
381 This calling convention will be used by a future version of the ObjectiveC
382 runtime and should therefore still be considered experimental at this time.
383 Although this convention was created to optimize certain runtime calls to
384 the ObjectiveC runtime, it is not limited to this runtime and might be used
385 by other runtimes in the future too. The current implementation only
386 supports X86-64, but the intention is to support more architectures in the
388 "``preserve_allcc``" - The `PreserveAll` calling convention
389 This calling convention attempts to make the code in the caller even less
390 intrusive than the `PreserveMost` calling convention. This calling
391 convention also behaves identical to the `C` calling convention on how
392 arguments and return values are passed, but it uses a different set of
393 caller/callee-saved registers. This removes the burden of saving and
394 recovering a large register set before and after the call in the caller. If
395 the arguments are passed in callee-saved registers, then they will be
396 preserved by the callee across the call. This doesn't apply for values
397 returned in callee-saved registers.
399 - On X86-64 the callee preserves all general purpose registers, except for
400 R11. R11 can be used as a scratch register. Furthermore it also preserves
401 all floating-point registers (XMMs/YMMs).
403 The idea behind this convention is to support calls to runtime functions
404 that don't need to call out to any other functions.
406 This calling convention, like the `PreserveMost` calling convention, will be
407 used by a future version of the ObjectiveC runtime and should be considered
408 experimental at this time.
409 "``cc <n>``" - Numbered convention
410 Any calling convention may be specified by number, allowing
411 target-specific calling conventions to be used. Target specific
412 calling conventions start at 64.
414 More calling conventions can be added/defined on an as-needed basis, to
415 support Pascal conventions or any other well-known target-independent
418 .. _visibilitystyles:
423 All Global Variables and Functions have one of the following visibility
426 "``default``" - Default style
427 On targets that use the ELF object file format, default visibility
428 means that the declaration is visible to other modules and, in
429 shared libraries, means that the declared entity may be overridden.
430 On Darwin, default visibility means that the declaration is visible
431 to other modules. Default visibility corresponds to "external
432 linkage" in the language.
433 "``hidden``" - Hidden style
434 Two declarations of an object with hidden visibility refer to the
435 same object if they are in the same shared object. Usually, hidden
436 visibility indicates that the symbol will not be placed into the
437 dynamic symbol table, so no other module (executable or shared
438 library) can reference it directly.
439 "``protected``" - Protected style
440 On ELF, protected visibility indicates that the symbol will be
441 placed in the dynamic symbol table, but that references within the
442 defining module will bind to the local symbol. That is, the symbol
443 cannot be overridden by another module.
445 A symbol with ``internal`` or ``private`` linkage must have ``default``
453 All Global Variables, Functions and Aliases can have one of the following
457 "``dllimport``" causes the compiler to reference a function or variable via
458 a global pointer to a pointer that is set up by the DLL exporting the
459 symbol. On Microsoft Windows targets, the pointer name is formed by
460 combining ``__imp_`` and the function or variable name.
462 "``dllexport``" causes the compiler to provide a global pointer to a pointer
463 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
464 Microsoft Windows targets, the pointer name is formed by combining
465 ``__imp_`` and the function or variable name. Since this storage class
466 exists for defining a dll interface, the compiler, assembler and linker know
467 it is externally referenced and must refrain from deleting the symbol.
471 Thread Local Storage Models
472 ---------------------------
474 A variable may be defined as ``thread_local``, which means that it will
475 not be shared by threads (each thread will have a separated copy of the
476 variable). Not all targets support thread-local variables. Optionally, a
477 TLS model may be specified:
480 For variables that are only used within the current shared library.
482 For variables in modules that will not be loaded dynamically.
484 For variables defined in the executable and only used within it.
486 If no explicit model is given, the "general dynamic" model is used.
488 The models correspond to the ELF TLS models; see `ELF Handling For
489 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
490 more information on under which circumstances the different models may
491 be used. The target may choose a different TLS model if the specified
492 model is not supported, or if a better choice of model can be made.
494 A model can also be specified in a alias, but then it only governs how
495 the alias is accessed. It will not have any effect in the aliasee.
502 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
503 types <t_struct>`. Literal types are uniqued structurally, but identified types
504 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
505 to forward declare a type that is not yet available.
507 An example of a identified structure specification is:
511 %mytype = type { %mytype*, i32 }
513 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
514 literal types are uniqued in recent versions of LLVM.
521 Global variables define regions of memory allocated at compilation time
524 Global variable definitions must be initialized.
526 Global variables in other translation units can also be declared, in which
527 case they don't have an initializer.
529 Either global variable definitions or declarations may have an explicit section
530 to be placed in and may have an optional explicit alignment specified.
532 A variable may be defined as a global ``constant``, which indicates that
533 the contents of the variable will **never** be modified (enabling better
534 optimization, allowing the global data to be placed in the read-only
535 section of an executable, etc). Note that variables that need runtime
536 initialization cannot be marked ``constant`` as there is a store to the
539 LLVM explicitly allows *declarations* of global variables to be marked
540 constant, even if the final definition of the global is not. This
541 capability can be used to enable slightly better optimization of the
542 program, but requires the language definition to guarantee that
543 optimizations based on the 'constantness' are valid for the translation
544 units that do not include the definition.
546 As SSA values, global variables define pointer values that are in scope
547 (i.e. they dominate) all basic blocks in the program. Global variables
548 always define a pointer to their "content" type because they describe a
549 region of memory, and all memory objects in LLVM are accessed through
552 Global variables can be marked with ``unnamed_addr`` which indicates
553 that the address is not significant, only the content. Constants marked
554 like this can be merged with other constants if they have the same
555 initializer. Note that a constant with significant address *can* be
556 merged with a ``unnamed_addr`` constant, the result being a constant
557 whose address is significant.
559 A global variable may be declared to reside in a target-specific
560 numbered address space. For targets that support them, address spaces
561 may affect how optimizations are performed and/or what target
562 instructions are used to access the variable. The default address space
563 is zero. The address space qualifier must precede any other attributes.
565 LLVM allows an explicit section to be specified for globals. If the
566 target supports it, it will emit globals to the section specified.
567 Additionally, the global can placed in a comdat if the target has the necessary
570 By default, global initializers are optimized by assuming that global
571 variables defined within the module are not modified from their
572 initial values before the start of the global initializer. This is
573 true even for variables potentially accessible from outside the
574 module, including those with external linkage or appearing in
575 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
576 by marking the variable with ``externally_initialized``.
578 An explicit alignment may be specified for a global, which must be a
579 power of 2. If not present, or if the alignment is set to zero, the
580 alignment of the global is set by the target to whatever it feels
581 convenient. If an explicit alignment is specified, the global is forced
582 to have exactly that alignment. Targets and optimizers are not allowed
583 to over-align the global if the global has an assigned section. In this
584 case, the extra alignment could be observable: for example, code could
585 assume that the globals are densely packed in their section and try to
586 iterate over them as an array, alignment padding would break this
587 iteration. The maximum alignment is ``1 << 29``.
589 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
591 Variables and aliases can have a
592 :ref:`Thread Local Storage Model <tls_model>`.
596 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
597 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
598 <global | constant> <Type> [<InitializerConstant>]
599 [, section "name"] [, comdat [($name)]]
600 [, align <Alignment>]
602 For example, the following defines a global in a numbered address space
603 with an initializer, section, and alignment:
607 @G = addrspace(5) constant float 1.0, section "foo", align 4
609 The following example just declares a global variable
613 @G = external global i32
615 The following example defines a thread-local global with the
616 ``initialexec`` TLS model:
620 @G = thread_local(initialexec) global i32 0, align 4
622 .. _functionstructure:
627 LLVM function definitions consist of the "``define``" keyword, an
628 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
629 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
630 an optional :ref:`calling convention <callingconv>`,
631 an optional ``unnamed_addr`` attribute, a return type, an optional
632 :ref:`parameter attribute <paramattrs>` for the return type, a function
633 name, a (possibly empty) argument list (each with optional :ref:`parameter
634 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
635 an optional section, an optional alignment,
636 an optional :ref:`comdat <langref_comdats>`,
637 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
638 an optional :ref:`prologue <prologuedata>`,
639 an optional :ref:`personality <personalityfn>`,
640 an opening curly brace, a list of basic blocks, and a closing curly brace.
642 LLVM function declarations consist of the "``declare``" keyword, an
643 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
644 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
645 an optional :ref:`calling convention <callingconv>`,
646 an optional ``unnamed_addr`` attribute, a return type, an optional
647 :ref:`parameter attribute <paramattrs>` for the return type, a function
648 name, a possibly empty list of arguments, an optional alignment, an optional
649 :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
650 and an optional :ref:`prologue <prologuedata>`.
652 A function definition contains a list of basic blocks, forming the CFG (Control
653 Flow Graph) for the function. Each basic block may optionally start with a label
654 (giving the basic block a symbol table entry), contains a list of instructions,
655 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
656 function return). If an explicit label is not provided, a block is assigned an
657 implicit numbered label, using the next value from the same counter as used for
658 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
659 entry block does not have an explicit label, it will be assigned label "%0",
660 then the first unnamed temporary in that block will be "%1", etc.
662 The first basic block in a function is special in two ways: it is
663 immediately executed on entrance to the function, and it is not allowed
664 to have predecessor basic blocks (i.e. there can not be any branches to
665 the entry block of a function). Because the block can have no
666 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
668 LLVM allows an explicit section to be specified for functions. If the
669 target supports it, it will emit functions to the section specified.
670 Additionally, the function can be placed in a COMDAT.
672 An explicit alignment may be specified for a function. If not present,
673 or if the alignment is set to zero, the alignment of the function is set
674 by the target to whatever it feels convenient. If an explicit alignment
675 is specified, the function is forced to have at least that much
676 alignment. All alignments must be a power of 2.
678 If the ``unnamed_addr`` attribute is given, the address is known to not
679 be significant and two identical functions can be merged.
683 define [linkage] [visibility] [DLLStorageClass]
685 <ResultType> @<FunctionName> ([argument list])
686 [unnamed_addr] [fn Attrs] [section "name"] [comdat [($name)]]
687 [align N] [gc] [prefix Constant] [prologue Constant]
688 [personality Constant] { ... }
690 The argument list is a comma seperated sequence of arguments where each
691 argument is of the following form
695 <type> [parameter Attrs] [name]
703 Aliases, unlike function or variables, don't create any new data. They
704 are just a new symbol and metadata for an existing position.
706 Aliases have a name and an aliasee that is either a global value or a
709 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
710 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
711 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
715 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy> @<Aliasee>
717 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
718 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
719 might not correctly handle dropping a weak symbol that is aliased.
721 Aliases that are not ``unnamed_addr`` are guaranteed to have the same address as
722 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
725 Since aliases are only a second name, some restrictions apply, of which
726 some can only be checked when producing an object file:
728 * The expression defining the aliasee must be computable at assembly
729 time. Since it is just a name, no relocations can be used.
731 * No alias in the expression can be weak as the possibility of the
732 intermediate alias being overridden cannot be represented in an
735 * No global value in the expression can be a declaration, since that
736 would require a relocation, which is not possible.
743 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
745 Comdats have a name which represents the COMDAT key. All global objects that
746 specify this key will only end up in the final object file if the linker chooses
747 that key over some other key. Aliases are placed in the same COMDAT that their
748 aliasee computes to, if any.
750 Comdats have a selection kind to provide input on how the linker should
751 choose between keys in two different object files.
755 $<Name> = comdat SelectionKind
757 The selection kind must be one of the following:
760 The linker may choose any COMDAT key, the choice is arbitrary.
762 The linker may choose any COMDAT key but the sections must contain the
765 The linker will choose the section containing the largest COMDAT key.
767 The linker requires that only section with this COMDAT key exist.
769 The linker may choose any COMDAT key but the sections must contain the
772 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
773 ``any`` as a selection kind.
775 Here is an example of a COMDAT group where a function will only be selected if
776 the COMDAT key's section is the largest:
780 $foo = comdat largest
781 @foo = global i32 2, comdat($foo)
783 define void @bar() comdat($foo) {
787 As a syntactic sugar the ``$name`` can be omitted if the name is the same as
793 @foo = global i32 2, comdat
796 In a COFF object file, this will create a COMDAT section with selection kind
797 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
798 and another COMDAT section with selection kind
799 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
800 section and contains the contents of the ``@bar`` symbol.
802 There are some restrictions on the properties of the global object.
803 It, or an alias to it, must have the same name as the COMDAT group when
805 The contents and size of this object may be used during link-time to determine
806 which COMDAT groups get selected depending on the selection kind.
807 Because the name of the object must match the name of the COMDAT group, the
808 linkage of the global object must not be local; local symbols can get renamed
809 if a collision occurs in the symbol table.
811 The combined use of COMDATS and section attributes may yield surprising results.
818 @g1 = global i32 42, section "sec", comdat($foo)
819 @g2 = global i32 42, section "sec", comdat($bar)
821 From the object file perspective, this requires the creation of two sections
822 with the same name. This is necessary because both globals belong to different
823 COMDAT groups and COMDATs, at the object file level, are represented by
826 Note that certain IR constructs like global variables and functions may create
827 COMDATs in the object file in addition to any which are specified using COMDAT
828 IR. This arises, for example, when a global variable has linkonce_odr linkage.
830 .. _namedmetadatastructure:
835 Named metadata is a collection of metadata. :ref:`Metadata
836 nodes <metadata>` (but not metadata strings) are the only valid
837 operands for a named metadata.
839 #. Named metadata are represented as a string of characters with the
840 metadata prefix. The rules for metadata names are the same as for
841 identifiers, but quoted names are not allowed. ``"\xx"`` type escapes
842 are still valid, which allows any character to be part of a name.
846 ; Some unnamed metadata nodes, which are referenced by the named metadata.
851 !name = !{!0, !1, !2}
858 The return type and each parameter of a function type may have a set of
859 *parameter attributes* associated with them. Parameter attributes are
860 used to communicate additional information about the result or
861 parameters of a function. Parameter attributes are considered to be part
862 of the function, not of the function type, so functions with different
863 parameter attributes can have the same function type.
865 Parameter attributes are simple keywords that follow the type specified.
866 If multiple parameter attributes are needed, they are space separated.
871 declare i32 @printf(i8* noalias nocapture, ...)
872 declare i32 @atoi(i8 zeroext)
873 declare signext i8 @returns_signed_char()
875 Note that any attributes for the function result (``nounwind``,
876 ``readonly``) come immediately after the argument list.
878 Currently, only the following parameter attributes are defined:
881 This indicates to the code generator that the parameter or return
882 value should be zero-extended to the extent required by the target's
883 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
884 the caller (for a parameter) or the callee (for a return value).
886 This indicates to the code generator that the parameter or return
887 value should be sign-extended to the extent required by the target's
888 ABI (which is usually 32-bits) by the caller (for a parameter) or
889 the callee (for a return value).
891 This indicates that this parameter or return value should be treated
892 in a special target-dependent fashion during while emitting code for
893 a function call or return (usually, by putting it in a register as
894 opposed to memory, though some targets use it to distinguish between
895 two different kinds of registers). Use of this attribute is
898 This indicates that the pointer parameter should really be passed by
899 value to the function. The attribute implies that a hidden copy of
900 the pointee is made between the caller and the callee, so the callee
901 is unable to modify the value in the caller. This attribute is only
902 valid on LLVM pointer arguments. It is generally used to pass
903 structs and arrays by value, but is also valid on pointers to
904 scalars. The copy is considered to belong to the caller not the
905 callee (for example, ``readonly`` functions should not write to
906 ``byval`` parameters). This is not a valid attribute for return
909 The byval attribute also supports specifying an alignment with the
910 align attribute. It indicates the alignment of the stack slot to
911 form and the known alignment of the pointer specified to the call
912 site. If the alignment is not specified, then the code generator
913 makes a target-specific assumption.
919 The ``inalloca`` argument attribute allows the caller to take the
920 address of outgoing stack arguments. An ``inalloca`` argument must
921 be a pointer to stack memory produced by an ``alloca`` instruction.
922 The alloca, or argument allocation, must also be tagged with the
923 inalloca keyword. Only the last argument may have the ``inalloca``
924 attribute, and that argument is guaranteed to be passed in memory.
926 An argument allocation may be used by a call at most once because
927 the call may deallocate it. The ``inalloca`` attribute cannot be
928 used in conjunction with other attributes that affect argument
929 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
930 ``inalloca`` attribute also disables LLVM's implicit lowering of
931 large aggregate return values, which means that frontend authors
932 must lower them with ``sret`` pointers.
934 When the call site is reached, the argument allocation must have
935 been the most recent stack allocation that is still live, or the
936 results are undefined. It is possible to allocate additional stack
937 space after an argument allocation and before its call site, but it
938 must be cleared off with :ref:`llvm.stackrestore
941 See :doc:`InAlloca` for more information on how to use this
945 This indicates that the pointer parameter specifies the address of a
946 structure that is the return value of the function in the source
947 program. This pointer must be guaranteed by the caller to be valid:
948 loads and stores to the structure may be assumed by the callee
949 not to trap and to be properly aligned. This may only be applied to
950 the first parameter. This is not a valid attribute for return
954 This indicates that the pointer value may be assumed by the optimizer to
955 have the specified alignment.
957 Note that this attribute has additional semantics when combined with the
963 This indicates that objects accessed via pointer values
964 :ref:`based <pointeraliasing>` on the argument or return value are not also
965 accessed, during the execution of the function, via pointer values not
966 *based* on the argument or return value. The attribute on a return value
967 also has additional semantics described below. The caller shares the
968 responsibility with the callee for ensuring that these requirements are met.
969 For further details, please see the discussion of the NoAlias response in
970 :ref:`alias analysis <Must, May, or No>`.
972 Note that this definition of ``noalias`` is intentionally similar
973 to the definition of ``restrict`` in C99 for function arguments.
975 For function return values, C99's ``restrict`` is not meaningful,
976 while LLVM's ``noalias`` is. Furthermore, the semantics of the ``noalias``
977 attribute on return values are stronger than the semantics of the attribute
978 when used on function arguments. On function return values, the ``noalias``
979 attribute indicates that the function acts like a system memory allocation
980 function, returning a pointer to allocated storage disjoint from the
981 storage for any other object accessible to the caller.
984 This indicates that the callee does not make any copies of the
985 pointer that outlive the callee itself. This is not a valid
986 attribute for return values.
991 This indicates that the pointer parameter can be excised using the
992 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
993 attribute for return values and can only be applied to one parameter.
996 This indicates that the function always returns the argument as its return
997 value. This is an optimization hint to the code generator when generating
998 the caller, allowing tail call optimization and omission of register saves
999 and restores in some cases; it is not checked or enforced when generating
1000 the callee. The parameter and the function return type must be valid
1001 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
1002 valid attribute for return values and can only be applied to one parameter.
1005 This indicates that the parameter or return pointer is not null. This
1006 attribute may only be applied to pointer typed parameters. This is not
1007 checked or enforced by LLVM, the caller must ensure that the pointer
1008 passed in is non-null, or the callee must ensure that the returned pointer
1011 ``dereferenceable(<n>)``
1012 This indicates that the parameter or return pointer is dereferenceable. This
1013 attribute may only be applied to pointer typed parameters. A pointer that
1014 is dereferenceable can be loaded from speculatively without a risk of
1015 trapping. The number of bytes known to be dereferenceable must be provided
1016 in parentheses. It is legal for the number of bytes to be less than the
1017 size of the pointee type. The ``nonnull`` attribute does not imply
1018 dereferenceability (consider a pointer to one element past the end of an
1019 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
1020 ``addrspace(0)`` (which is the default address space).
1022 ``dereferenceable_or_null(<n>)``
1023 This indicates that the parameter or return value isn't both
1024 non-null and non-dereferenceable (up to ``<n>`` bytes) at the same
1025 time. All non-null pointers tagged with
1026 ``dereferenceable_or_null(<n>)`` are ``dereferenceable(<n>)``.
1027 For address space 0 ``dereferenceable_or_null(<n>)`` implies that
1028 a pointer is exactly one of ``dereferenceable(<n>)`` or ``null``,
1029 and in other address spaces ``dereferenceable_or_null(<n>)``
1030 implies that a pointer is at least one of ``dereferenceable(<n>)``
1031 or ``null`` (i.e. it may be both ``null`` and
1032 ``dereferenceable(<n>)``). This attribute may only be applied to
1033 pointer typed parameters.
1037 Garbage Collector Strategy Names
1038 --------------------------------
1040 Each function may specify a garbage collector strategy name, which is simply a
1043 .. code-block:: llvm
1045 define void @f() gc "name" { ... }
1047 The supported values of *name* includes those :ref:`built in to LLVM
1048 <builtin-gc-strategies>` and any provided by loaded plugins. Specifying a GC
1049 strategy will cause the compiler to alter its output in order to support the
1050 named garbage collection algorithm. Note that LLVM itself does not contain a
1051 garbage collector, this functionality is restricted to generating machine code
1052 which can interoperate with a collector provided externally.
1059 Prefix data is data associated with a function which the code
1060 generator will emit immediately before the function's entrypoint.
1061 The purpose of this feature is to allow frontends to associate
1062 language-specific runtime metadata with specific functions and make it
1063 available through the function pointer while still allowing the
1064 function pointer to be called.
1066 To access the data for a given function, a program may bitcast the
1067 function pointer to a pointer to the constant's type and dereference
1068 index -1. This implies that the IR symbol points just past the end of
1069 the prefix data. For instance, take the example of a function annotated
1070 with a single ``i32``,
1072 .. code-block:: llvm
1074 define void @f() prefix i32 123 { ... }
1076 The prefix data can be referenced as,
1078 .. code-block:: llvm
1080 %0 = bitcast void* () @f to i32*
1081 %a = getelementptr inbounds i32, i32* %0, i32 -1
1082 %b = load i32, i32* %a
1084 Prefix data is laid out as if it were an initializer for a global variable
1085 of the prefix data's type. The function will be placed such that the
1086 beginning of the prefix data is aligned. This means that if the size
1087 of the prefix data is not a multiple of the alignment size, the
1088 function's entrypoint will not be aligned. If alignment of the
1089 function's entrypoint is desired, padding must be added to the prefix
1092 A function may have prefix data but no body. This has similar semantics
1093 to the ``available_externally`` linkage in that the data may be used by the
1094 optimizers but will not be emitted in the object file.
1101 The ``prologue`` attribute allows arbitrary code (encoded as bytes) to
1102 be inserted prior to the function body. This can be used for enabling
1103 function hot-patching and instrumentation.
1105 To maintain the semantics of ordinary function calls, the prologue data must
1106 have a particular format. Specifically, it must begin with a sequence of
1107 bytes which decode to a sequence of machine instructions, valid for the
1108 module's target, which transfer control to the point immediately succeeding
1109 the prologue data, without performing any other visible action. This allows
1110 the inliner and other passes to reason about the semantics of the function
1111 definition without needing to reason about the prologue data. Obviously this
1112 makes the format of the prologue data highly target dependent.
1114 A trivial example of valid prologue data for the x86 architecture is ``i8 144``,
1115 which encodes the ``nop`` instruction:
1117 .. code-block:: llvm
1119 define void @f() prologue i8 144 { ... }
1121 Generally prologue data can be formed by encoding a relative branch instruction
1122 which skips the metadata, as in this example of valid prologue data for the
1123 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1125 .. code-block:: llvm
1127 %0 = type <{ i8, i8, i8* }>
1129 define void @f() prologue %0 <{ i8 235, i8 8, i8* @md}> { ... }
1131 A function may have prologue data but no body. This has similar semantics
1132 to the ``available_externally`` linkage in that the data may be used by the
1133 optimizers but will not be emitted in the object file.
1137 Personality Function
1138 --------------------
1140 The ``personality`` attribute permits functions to specify what function
1141 to use for exception handling.
1148 Attribute groups are groups of attributes that are referenced by objects within
1149 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1150 functions will use the same set of attributes. In the degenerative case of a
1151 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1152 group will capture the important command line flags used to build that file.
1154 An attribute group is a module-level object. To use an attribute group, an
1155 object references the attribute group's ID (e.g. ``#37``). An object may refer
1156 to more than one attribute group. In that situation, the attributes from the
1157 different groups are merged.
1159 Here is an example of attribute groups for a function that should always be
1160 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1162 .. code-block:: llvm
1164 ; Target-independent attributes:
1165 attributes #0 = { alwaysinline alignstack=4 }
1167 ; Target-dependent attributes:
1168 attributes #1 = { "no-sse" }
1170 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1171 define void @f() #0 #1 { ... }
1178 Function attributes are set to communicate additional information about
1179 a function. Function attributes are considered to be part of the
1180 function, not of the function type, so functions with different function
1181 attributes can have the same function type.
1183 Function attributes are simple keywords that follow the type specified.
1184 If multiple attributes are needed, they are space separated. For
1187 .. code-block:: llvm
1189 define void @f() noinline { ... }
1190 define void @f() alwaysinline { ... }
1191 define void @f() alwaysinline optsize { ... }
1192 define void @f() optsize { ... }
1195 This attribute indicates that, when emitting the prologue and
1196 epilogue, the backend should forcibly align the stack pointer.
1197 Specify the desired alignment, which must be a power of two, in
1200 This attribute indicates that the inliner should attempt to inline
1201 this function into callers whenever possible, ignoring any active
1202 inlining size threshold for this caller.
1204 This indicates that the callee function at a call site should be
1205 recognized as a built-in function, even though the function's declaration
1206 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1207 direct calls to functions that are declared with the ``nobuiltin``
1210 This attribute indicates that this function is rarely called. When
1211 computing edge weights, basic blocks post-dominated by a cold
1212 function call are also considered to be cold; and, thus, given low
1215 This attribute indicates that the callee is dependent on a convergent
1216 thread execution pattern under certain parallel execution models.
1217 Transformations that are execution model agnostic may only move or
1218 tranform this call if the final location is control equivalent to its
1219 original position in the program, where control equivalence is defined as
1220 A dominates B and B post-dominates A, or vice versa.
1222 This attribute indicates that the source code contained a hint that
1223 inlining this function is desirable (such as the "inline" keyword in
1224 C/C++). It is just a hint; it imposes no requirements on the
1227 This attribute indicates that the function should be added to a
1228 jump-instruction table at code-generation time, and that all address-taken
1229 references to this function should be replaced with a reference to the
1230 appropriate jump-instruction-table function pointer. Note that this creates
1231 a new pointer for the original function, which means that code that depends
1232 on function-pointer identity can break. So, any function annotated with
1233 ``jumptable`` must also be ``unnamed_addr``.
1235 This attribute suggests that optimization passes and code generator
1236 passes make choices that keep the code size of this function as small
1237 as possible and perform optimizations that may sacrifice runtime
1238 performance in order to minimize the size of the generated code.
1240 This attribute disables prologue / epilogue emission for the
1241 function. This can have very system-specific consequences.
1243 This indicates that the callee function at a call site is not recognized as
1244 a built-in function. LLVM will retain the original call and not replace it
1245 with equivalent code based on the semantics of the built-in function, unless
1246 the call site uses the ``builtin`` attribute. This is valid at call sites
1247 and on function declarations and definitions.
1249 This attribute indicates that calls to the function cannot be
1250 duplicated. A call to a ``noduplicate`` function may be moved
1251 within its parent function, but may not be duplicated within
1252 its parent function.
1254 A function containing a ``noduplicate`` call may still
1255 be an inlining candidate, provided that the call is not
1256 duplicated by inlining. That implies that the function has
1257 internal linkage and only has one call site, so the original
1258 call is dead after inlining.
1260 This attributes disables implicit floating point instructions.
1262 This attribute indicates that the inliner should never inline this
1263 function in any situation. This attribute may not be used together
1264 with the ``alwaysinline`` attribute.
1266 This attribute suppresses lazy symbol binding for the function. This
1267 may make calls to the function faster, at the cost of extra program
1268 startup time if the function is not called during program startup.
1270 This attribute indicates that the code generator should not use a
1271 red zone, even if the target-specific ABI normally permits it.
1273 This function attribute indicates that the function never returns
1274 normally. This produces undefined behavior at runtime if the
1275 function ever does dynamically return.
1277 This function attribute indicates that the function never raises an
1278 exception. If the function does raise an exception, its runtime
1279 behavior is undefined. However, functions marked nounwind may still
1280 trap or generate asynchronous exceptions. Exception handling schemes
1281 that are recognized by LLVM to handle asynchronous exceptions, such
1282 as SEH, will still provide their implementation defined semantics.
1284 This function attribute indicates that the function is not optimized
1285 by any optimization or code generator passes with the
1286 exception of interprocedural optimization passes.
1287 This attribute cannot be used together with the ``alwaysinline``
1288 attribute; this attribute is also incompatible
1289 with the ``minsize`` attribute and the ``optsize`` attribute.
1291 This attribute requires the ``noinline`` attribute to be specified on
1292 the function as well, so the function is never inlined into any caller.
1293 Only functions with the ``alwaysinline`` attribute are valid
1294 candidates for inlining into the body of this function.
1296 This attribute suggests that optimization passes and code generator
1297 passes make choices that keep the code size of this function low,
1298 and otherwise do optimizations specifically to reduce code size as
1299 long as they do not significantly impact runtime performance.
1301 On a function, this attribute indicates that the function computes its
1302 result (or decides to unwind an exception) based strictly on its arguments,
1303 without dereferencing any pointer arguments or otherwise accessing
1304 any mutable state (e.g. memory, control registers, etc) visible to
1305 caller functions. It does not write through any pointer arguments
1306 (including ``byval`` arguments) and never changes any state visible
1307 to callers. This means that it cannot unwind exceptions by calling
1308 the ``C++`` exception throwing methods.
1310 On an argument, this attribute indicates that the function does not
1311 dereference that pointer argument, even though it may read or write the
1312 memory that the pointer points to if accessed through other pointers.
1314 On a function, this attribute indicates that the function does not write
1315 through any pointer arguments (including ``byval`` arguments) or otherwise
1316 modify any state (e.g. memory, control registers, etc) visible to
1317 caller functions. It may dereference pointer arguments and read
1318 state that may be set in the caller. A readonly function always
1319 returns the same value (or unwinds an exception identically) when
1320 called with the same set of arguments and global state. It cannot
1321 unwind an exception by calling the ``C++`` exception throwing
1324 On an argument, this attribute indicates that the function does not write
1325 through this pointer argument, even though it may write to the memory that
1326 the pointer points to.
1328 This attribute indicates that this function can return twice. The C
1329 ``setjmp`` is an example of such a function. The compiler disables
1330 some optimizations (like tail calls) in the caller of these
1333 This attribute indicates that
1334 `SafeStack <http://clang.llvm.org/docs/SafeStack.html>`_
1335 protection is enabled for this function.
1337 If a function that has a ``safestack`` attribute is inlined into a
1338 function that doesn't have a ``safestack`` attribute or which has an
1339 ``ssp``, ``sspstrong`` or ``sspreq`` attribute, then the resulting
1340 function will have a ``safestack`` attribute.
1341 ``sanitize_address``
1342 This attribute indicates that AddressSanitizer checks
1343 (dynamic address safety analysis) are enabled for this function.
1345 This attribute indicates that MemorySanitizer checks (dynamic detection
1346 of accesses to uninitialized memory) are enabled for this function.
1348 This attribute indicates that ThreadSanitizer checks
1349 (dynamic thread safety analysis) are enabled for this function.
1351 This attribute indicates that the function should emit a stack
1352 smashing protector. It is in the form of a "canary" --- a random value
1353 placed on the stack before the local variables that's checked upon
1354 return from the function to see if it has been overwritten. A
1355 heuristic is used to determine if a function needs stack protectors
1356 or not. The heuristic used will enable protectors for functions with:
1358 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1359 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1360 - Calls to alloca() with variable sizes or constant sizes greater than
1361 ``ssp-buffer-size``.
1363 Variables that are identified as requiring a protector will be arranged
1364 on the stack such that they are adjacent to the stack protector guard.
1366 If a function that has an ``ssp`` attribute is inlined into a
1367 function that doesn't have an ``ssp`` attribute, then the resulting
1368 function will have an ``ssp`` attribute.
1370 This attribute indicates that the function should *always* emit a
1371 stack smashing protector. This overrides the ``ssp`` function
1374 Variables that are identified as requiring a protector will be arranged
1375 on the stack such that they are adjacent to the stack protector guard.
1376 The specific layout rules are:
1378 #. Large arrays and structures containing large arrays
1379 (``>= ssp-buffer-size``) are closest to the stack protector.
1380 #. Small arrays and structures containing small arrays
1381 (``< ssp-buffer-size``) are 2nd closest to the protector.
1382 #. Variables that have had their address taken are 3rd closest to the
1385 If a function that has an ``sspreq`` attribute is inlined into a
1386 function that doesn't have an ``sspreq`` attribute or which has an
1387 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1388 an ``sspreq`` attribute.
1390 This attribute indicates that the function should emit a stack smashing
1391 protector. This attribute causes a strong heuristic to be used when
1392 determining if a function needs stack protectors. The strong heuristic
1393 will enable protectors for functions with:
1395 - Arrays of any size and type
1396 - Aggregates containing an array of any size and type.
1397 - Calls to alloca().
1398 - Local variables that have had their address taken.
1400 Variables that are identified as requiring a protector will be arranged
1401 on the stack such that they are adjacent to the stack protector guard.
1402 The specific layout rules are:
1404 #. Large arrays and structures containing large arrays
1405 (``>= ssp-buffer-size``) are closest to the stack protector.
1406 #. Small arrays and structures containing small arrays
1407 (``< ssp-buffer-size``) are 2nd closest to the protector.
1408 #. Variables that have had their address taken are 3rd closest to the
1411 This overrides the ``ssp`` function attribute.
1413 If a function that has an ``sspstrong`` attribute is inlined into a
1414 function that doesn't have an ``sspstrong`` attribute, then the
1415 resulting function will have an ``sspstrong`` attribute.
1417 This attribute indicates that the function will delegate to some other
1418 function with a tail call. The prototype of a thunk should not be used for
1419 optimization purposes. The caller is expected to cast the thunk prototype to
1420 match the thunk target prototype.
1422 This attribute indicates that the ABI being targeted requires that
1423 an unwind table entry be produce for this function even if we can
1424 show that no exceptions passes by it. This is normally the case for
1425 the ELF x86-64 abi, but it can be disabled for some compilation
1430 Module-Level Inline Assembly
1431 ----------------------------
1433 Modules may contain "module-level inline asm" blocks, which corresponds
1434 to the GCC "file scope inline asm" blocks. These blocks are internally
1435 concatenated by LLVM and treated as a single unit, but may be separated
1436 in the ``.ll`` file if desired. The syntax is very simple:
1438 .. code-block:: llvm
1440 module asm "inline asm code goes here"
1441 module asm "more can go here"
1443 The strings can contain any character by escaping non-printable
1444 characters. The escape sequence used is simply "\\xx" where "xx" is the
1445 two digit hex code for the number.
1447 The inline asm code is simply printed to the machine code .s file when
1448 assembly code is generated.
1450 .. _langref_datalayout:
1455 A module may specify a target specific data layout string that specifies
1456 how data is to be laid out in memory. The syntax for the data layout is
1459 .. code-block:: llvm
1461 target datalayout = "layout specification"
1463 The *layout specification* consists of a list of specifications
1464 separated by the minus sign character ('-'). Each specification starts
1465 with a letter and may include other information after the letter to
1466 define some aspect of the data layout. The specifications accepted are
1470 Specifies that the target lays out data in big-endian form. That is,
1471 the bits with the most significance have the lowest address
1474 Specifies that the target lays out data in little-endian form. That
1475 is, the bits with the least significance have the lowest address
1478 Specifies the natural alignment of the stack in bits. Alignment
1479 promotion of stack variables is limited to the natural stack
1480 alignment to avoid dynamic stack realignment. The stack alignment
1481 must be a multiple of 8-bits. If omitted, the natural stack
1482 alignment defaults to "unspecified", which does not prevent any
1483 alignment promotions.
1484 ``p[n]:<size>:<abi>:<pref>``
1485 This specifies the *size* of a pointer and its ``<abi>`` and
1486 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1487 bits. The address space, ``n`` is optional, and if not specified,
1488 denotes the default address space 0. The value of ``n`` must be
1489 in the range [1,2^23).
1490 ``i<size>:<abi>:<pref>``
1491 This specifies the alignment for an integer type of a given bit
1492 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1493 ``v<size>:<abi>:<pref>``
1494 This specifies the alignment for a vector type of a given bit
1496 ``f<size>:<abi>:<pref>``
1497 This specifies the alignment for a floating point type of a given bit
1498 ``<size>``. Only values of ``<size>`` that are supported by the target
1499 will work. 32 (float) and 64 (double) are supported on all targets; 80
1500 or 128 (different flavors of long double) are also supported on some
1503 This specifies the alignment for an object of aggregate type.
1505 If present, specifies that llvm names are mangled in the output. The
1508 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1509 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1510 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1511 symbols get a ``_`` prefix.
1512 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1513 functions also get a suffix based on the frame size.
1514 ``n<size1>:<size2>:<size3>...``
1515 This specifies a set of native integer widths for the target CPU in
1516 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1517 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1518 this set are considered to support most general arithmetic operations
1521 On every specification that takes a ``<abi>:<pref>``, specifying the
1522 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1523 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1525 When constructing the data layout for a given target, LLVM starts with a
1526 default set of specifications which are then (possibly) overridden by
1527 the specifications in the ``datalayout`` keyword. The default
1528 specifications are given in this list:
1530 - ``E`` - big endian
1531 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1532 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1533 same as the default address space.
1534 - ``S0`` - natural stack alignment is unspecified
1535 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1536 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1537 - ``i16:16:16`` - i16 is 16-bit aligned
1538 - ``i32:32:32`` - i32 is 32-bit aligned
1539 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1540 alignment of 64-bits
1541 - ``f16:16:16`` - half is 16-bit aligned
1542 - ``f32:32:32`` - float is 32-bit aligned
1543 - ``f64:64:64`` - double is 64-bit aligned
1544 - ``f128:128:128`` - quad is 128-bit aligned
1545 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1546 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1547 - ``a:0:64`` - aggregates are 64-bit aligned
1549 When LLVM is determining the alignment for a given type, it uses the
1552 #. If the type sought is an exact match for one of the specifications,
1553 that specification is used.
1554 #. If no match is found, and the type sought is an integer type, then
1555 the smallest integer type that is larger than the bitwidth of the
1556 sought type is used. If none of the specifications are larger than
1557 the bitwidth then the largest integer type is used. For example,
1558 given the default specifications above, the i7 type will use the
1559 alignment of i8 (next largest) while both i65 and i256 will use the
1560 alignment of i64 (largest specified).
1561 #. If no match is found, and the type sought is a vector type, then the
1562 largest vector type that is smaller than the sought vector type will
1563 be used as a fall back. This happens because <128 x double> can be
1564 implemented in terms of 64 <2 x double>, for example.
1566 The function of the data layout string may not be what you expect.
1567 Notably, this is not a specification from the frontend of what alignment
1568 the code generator should use.
1570 Instead, if specified, the target data layout is required to match what
1571 the ultimate *code generator* expects. This string is used by the
1572 mid-level optimizers to improve code, and this only works if it matches
1573 what the ultimate code generator uses. There is no way to generate IR
1574 that does not embed this target-specific detail into the IR. If you
1575 don't specify the string, the default specifications will be used to
1576 generate a Data Layout and the optimization phases will operate
1577 accordingly and introduce target specificity into the IR with respect to
1578 these default specifications.
1585 A module may specify a target triple string that describes the target
1586 host. The syntax for the target triple is simply:
1588 .. code-block:: llvm
1590 target triple = "x86_64-apple-macosx10.7.0"
1592 The *target triple* string consists of a series of identifiers delimited
1593 by the minus sign character ('-'). The canonical forms are:
1597 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1598 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1600 This information is passed along to the backend so that it generates
1601 code for the proper architecture. It's possible to override this on the
1602 command line with the ``-mtriple`` command line option.
1604 .. _pointeraliasing:
1606 Pointer Aliasing Rules
1607 ----------------------
1609 Any memory access must be done through a pointer value associated with
1610 an address range of the memory access, otherwise the behavior is
1611 undefined. Pointer values are associated with address ranges according
1612 to the following rules:
1614 - A pointer value is associated with the addresses associated with any
1615 value it is *based* on.
1616 - An address of a global variable is associated with the address range
1617 of the variable's storage.
1618 - The result value of an allocation instruction is associated with the
1619 address range of the allocated storage.
1620 - A null pointer in the default address-space is associated with no
1622 - An integer constant other than zero or a pointer value returned from
1623 a function not defined within LLVM may be associated with address
1624 ranges allocated through mechanisms other than those provided by
1625 LLVM. Such ranges shall not overlap with any ranges of addresses
1626 allocated by mechanisms provided by LLVM.
1628 A pointer value is *based* on another pointer value according to the
1631 - A pointer value formed from a ``getelementptr`` operation is *based*
1632 on the first value operand of the ``getelementptr``.
1633 - The result value of a ``bitcast`` is *based* on the operand of the
1635 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1636 values that contribute (directly or indirectly) to the computation of
1637 the pointer's value.
1638 - The "*based* on" relationship is transitive.
1640 Note that this definition of *"based"* is intentionally similar to the
1641 definition of *"based"* in C99, though it is slightly weaker.
1643 LLVM IR does not associate types with memory. The result type of a
1644 ``load`` merely indicates the size and alignment of the memory from
1645 which to load, as well as the interpretation of the value. The first
1646 operand type of a ``store`` similarly only indicates the size and
1647 alignment of the store.
1649 Consequently, type-based alias analysis, aka TBAA, aka
1650 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1651 :ref:`Metadata <metadata>` may be used to encode additional information
1652 which specialized optimization passes may use to implement type-based
1657 Volatile Memory Accesses
1658 ------------------------
1660 Certain memory accesses, such as :ref:`load <i_load>`'s,
1661 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1662 marked ``volatile``. The optimizers must not change the number of
1663 volatile operations or change their order of execution relative to other
1664 volatile operations. The optimizers *may* change the order of volatile
1665 operations relative to non-volatile operations. This is not Java's
1666 "volatile" and has no cross-thread synchronization behavior.
1668 IR-level volatile loads and stores cannot safely be optimized into
1669 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1670 flagged volatile. Likewise, the backend should never split or merge
1671 target-legal volatile load/store instructions.
1673 .. admonition:: Rationale
1675 Platforms may rely on volatile loads and stores of natively supported
1676 data width to be executed as single instruction. For example, in C
1677 this holds for an l-value of volatile primitive type with native
1678 hardware support, but not necessarily for aggregate types. The
1679 frontend upholds these expectations, which are intentionally
1680 unspecified in the IR. The rules above ensure that IR transformation
1681 do not violate the frontend's contract with the language.
1685 Memory Model for Concurrent Operations
1686 --------------------------------------
1688 The LLVM IR does not define any way to start parallel threads of
1689 execution or to register signal handlers. Nonetheless, there are
1690 platform-specific ways to create them, and we define LLVM IR's behavior
1691 in their presence. This model is inspired by the C++0x memory model.
1693 For a more informal introduction to this model, see the :doc:`Atomics`.
1695 We define a *happens-before* partial order as the least partial order
1698 - Is a superset of single-thread program order, and
1699 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1700 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1701 techniques, like pthread locks, thread creation, thread joining,
1702 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1703 Constraints <ordering>`).
1705 Note that program order does not introduce *happens-before* edges
1706 between a thread and signals executing inside that thread.
1708 Every (defined) read operation (load instructions, memcpy, atomic
1709 loads/read-modify-writes, etc.) R reads a series of bytes written by
1710 (defined) write operations (store instructions, atomic
1711 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1712 section, initialized globals are considered to have a write of the
1713 initializer which is atomic and happens before any other read or write
1714 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1715 may see any write to the same byte, except:
1717 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1718 write\ :sub:`2` happens before R\ :sub:`byte`, then
1719 R\ :sub:`byte` does not see write\ :sub:`1`.
1720 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1721 R\ :sub:`byte` does not see write\ :sub:`3`.
1723 Given that definition, R\ :sub:`byte` is defined as follows:
1725 - If R is volatile, the result is target-dependent. (Volatile is
1726 supposed to give guarantees which can support ``sig_atomic_t`` in
1727 C/C++, and may be used for accesses to addresses that do not behave
1728 like normal memory. It does not generally provide cross-thread
1730 - Otherwise, if there is no write to the same byte that happens before
1731 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1732 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1733 R\ :sub:`byte` returns the value written by that write.
1734 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1735 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1736 Memory Ordering Constraints <ordering>` section for additional
1737 constraints on how the choice is made.
1738 - Otherwise R\ :sub:`byte` returns ``undef``.
1740 R returns the value composed of the series of bytes it read. This
1741 implies that some bytes within the value may be ``undef`` **without**
1742 the entire value being ``undef``. Note that this only defines the
1743 semantics of the operation; it doesn't mean that targets will emit more
1744 than one instruction to read the series of bytes.
1746 Note that in cases where none of the atomic intrinsics are used, this
1747 model places only one restriction on IR transformations on top of what
1748 is required for single-threaded execution: introducing a store to a byte
1749 which might not otherwise be stored is not allowed in general.
1750 (Specifically, in the case where another thread might write to and read
1751 from an address, introducing a store can change a load that may see
1752 exactly one write into a load that may see multiple writes.)
1756 Atomic Memory Ordering Constraints
1757 ----------------------------------
1759 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1760 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1761 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1762 ordering parameters that determine which other atomic instructions on
1763 the same address they *synchronize with*. These semantics are borrowed
1764 from Java and C++0x, but are somewhat more colloquial. If these
1765 descriptions aren't precise enough, check those specs (see spec
1766 references in the :doc:`atomics guide <Atomics>`).
1767 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1768 differently since they don't take an address. See that instruction's
1769 documentation for details.
1771 For a simpler introduction to the ordering constraints, see the
1775 The set of values that can be read is governed by the happens-before
1776 partial order. A value cannot be read unless some operation wrote
1777 it. This is intended to provide a guarantee strong enough to model
1778 Java's non-volatile shared variables. This ordering cannot be
1779 specified for read-modify-write operations; it is not strong enough
1780 to make them atomic in any interesting way.
1782 In addition to the guarantees of ``unordered``, there is a single
1783 total order for modifications by ``monotonic`` operations on each
1784 address. All modification orders must be compatible with the
1785 happens-before order. There is no guarantee that the modification
1786 orders can be combined to a global total order for the whole program
1787 (and this often will not be possible). The read in an atomic
1788 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1789 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1790 order immediately before the value it writes. If one atomic read
1791 happens before another atomic read of the same address, the later
1792 read must see the same value or a later value in the address's
1793 modification order. This disallows reordering of ``monotonic`` (or
1794 stronger) operations on the same address. If an address is written
1795 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1796 read that address repeatedly, the other threads must eventually see
1797 the write. This corresponds to the C++0x/C1x
1798 ``memory_order_relaxed``.
1800 In addition to the guarantees of ``monotonic``, a
1801 *synchronizes-with* edge may be formed with a ``release`` operation.
1802 This is intended to model C++'s ``memory_order_acquire``.
1804 In addition to the guarantees of ``monotonic``, if this operation
1805 writes a value which is subsequently read by an ``acquire``
1806 operation, it *synchronizes-with* that operation. (This isn't a
1807 complete description; see the C++0x definition of a release
1808 sequence.) This corresponds to the C++0x/C1x
1809 ``memory_order_release``.
1810 ``acq_rel`` (acquire+release)
1811 Acts as both an ``acquire`` and ``release`` operation on its
1812 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1813 ``seq_cst`` (sequentially consistent)
1814 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1815 operation that only reads, ``release`` for an operation that only
1816 writes), there is a global total order on all
1817 sequentially-consistent operations on all addresses, which is
1818 consistent with the *happens-before* partial order and with the
1819 modification orders of all the affected addresses. Each
1820 sequentially-consistent read sees the last preceding write to the
1821 same address in this global order. This corresponds to the C++0x/C1x
1822 ``memory_order_seq_cst`` and Java volatile.
1826 If an atomic operation is marked ``singlethread``, it only *synchronizes
1827 with* or participates in modification and seq\_cst total orderings with
1828 other operations running in the same thread (for example, in signal
1836 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1837 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1838 :ref:`frem <i_frem>`) have the following flags that can be set to enable
1839 otherwise unsafe floating point operations
1842 No NaNs - Allow optimizations to assume the arguments and result are not
1843 NaN. Such optimizations are required to retain defined behavior over
1844 NaNs, but the value of the result is undefined.
1847 No Infs - Allow optimizations to assume the arguments and result are not
1848 +/-Inf. Such optimizations are required to retain defined behavior over
1849 +/-Inf, but the value of the result is undefined.
1852 No Signed Zeros - Allow optimizations to treat the sign of a zero
1853 argument or result as insignificant.
1856 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1857 argument rather than perform division.
1860 Fast - Allow algebraically equivalent transformations that may
1861 dramatically change results in floating point (e.g. reassociate). This
1862 flag implies all the others.
1866 Use-list Order Directives
1867 -------------------------
1869 Use-list directives encode the in-memory order of each use-list, allowing the
1870 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1871 indexes that are assigned to the referenced value's uses. The referenced
1872 value's use-list is immediately sorted by these indexes.
1874 Use-list directives may appear at function scope or global scope. They are not
1875 instructions, and have no effect on the semantics of the IR. When they're at
1876 function scope, they must appear after the terminator of the final basic block.
1878 If basic blocks have their address taken via ``blockaddress()`` expressions,
1879 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1886 uselistorder <ty> <value>, { <order-indexes> }
1887 uselistorder_bb @function, %block { <order-indexes> }
1893 define void @foo(i32 %arg1, i32 %arg2) {
1895 ; ... instructions ...
1897 ; ... instructions ...
1899 ; At function scope.
1900 uselistorder i32 %arg1, { 1, 0, 2 }
1901 uselistorder label %bb, { 1, 0 }
1905 uselistorder i32* @global, { 1, 2, 0 }
1906 uselistorder i32 7, { 1, 0 }
1907 uselistorder i32 (i32) @bar, { 1, 0 }
1908 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1915 The LLVM type system is one of the most important features of the
1916 intermediate representation. Being typed enables a number of
1917 optimizations to be performed on the intermediate representation
1918 directly, without having to do extra analyses on the side before the
1919 transformation. A strong type system makes it easier to read the
1920 generated code and enables novel analyses and transformations that are
1921 not feasible to perform on normal three address code representations.
1931 The void type does not represent any value and has no size.
1949 The function type can be thought of as a function signature. It consists of a
1950 return type and a list of formal parameter types. The return type of a function
1951 type is a void type or first class type --- except for :ref:`label <t_label>`
1952 and :ref:`metadata <t_metadata>` types.
1958 <returntype> (<parameter list>)
1960 ...where '``<parameter list>``' is a comma-separated list of type
1961 specifiers. Optionally, the parameter list may include a type ``...``, which
1962 indicates that the function takes a variable number of arguments. Variable
1963 argument functions can access their arguments with the :ref:`variable argument
1964 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1965 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1969 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1970 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1971 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1972 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1973 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1974 | ``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. |
1975 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1976 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1977 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1984 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1985 Values of these types are the only ones which can be produced by
1993 These are the types that are valid in registers from CodeGen's perspective.
2002 The integer type is a very simple type that simply specifies an
2003 arbitrary bit width for the integer type desired. Any bit width from 1
2004 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
2012 The number of bits the integer will occupy is specified by the ``N``
2018 +----------------+------------------------------------------------+
2019 | ``i1`` | a single-bit integer. |
2020 +----------------+------------------------------------------------+
2021 | ``i32`` | a 32-bit integer. |
2022 +----------------+------------------------------------------------+
2023 | ``i1942652`` | a really big integer of over 1 million bits. |
2024 +----------------+------------------------------------------------+
2028 Floating Point Types
2029 """"""""""""""""""""
2038 - 16-bit floating point value
2041 - 32-bit floating point value
2044 - 64-bit floating point value
2047 - 128-bit floating point value (112-bit mantissa)
2050 - 80-bit floating point value (X87)
2053 - 128-bit floating point value (two 64-bits)
2060 The x86_mmx type represents a value held in an MMX register on an x86
2061 machine. The operations allowed on it are quite limited: parameters and
2062 return values, load and store, and bitcast. User-specified MMX
2063 instructions are represented as intrinsic or asm calls with arguments
2064 and/or results of this type. There are no arrays, vectors or constants
2081 The pointer type is used to specify memory locations. Pointers are
2082 commonly used to reference objects in memory.
2084 Pointer types may have an optional address space attribute defining the
2085 numbered address space where the pointed-to object resides. The default
2086 address space is number zero. The semantics of non-zero address spaces
2087 are target-specific.
2089 Note that LLVM does not permit pointers to void (``void*``) nor does it
2090 permit pointers to labels (``label*``). Use ``i8*`` instead.
2100 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2101 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2102 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2103 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2104 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2105 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2106 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2115 A vector type is a simple derived type that represents a vector of
2116 elements. Vector types are used when multiple primitive data are
2117 operated in parallel using a single instruction (SIMD). A vector type
2118 requires a size (number of elements) and an underlying primitive data
2119 type. Vector types are considered :ref:`first class <t_firstclass>`.
2125 < <# elements> x <elementtype> >
2127 The number of elements is a constant integer value larger than 0;
2128 elementtype may be any integer, floating point or pointer type. Vectors
2129 of size zero are not allowed.
2133 +-------------------+--------------------------------------------------+
2134 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2135 +-------------------+--------------------------------------------------+
2136 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2137 +-------------------+--------------------------------------------------+
2138 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2139 +-------------------+--------------------------------------------------+
2140 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2141 +-------------------+--------------------------------------------------+
2150 The label type represents code labels.
2165 The metadata type represents embedded metadata. No derived types may be
2166 created from metadata except for :ref:`function <t_function>` arguments.
2179 Aggregate Types are a subset of derived types that can contain multiple
2180 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2181 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2191 The array type is a very simple derived type that arranges elements
2192 sequentially in memory. The array type requires a size (number of
2193 elements) and an underlying data type.
2199 [<# elements> x <elementtype>]
2201 The number of elements is a constant integer value; ``elementtype`` may
2202 be any type with a size.
2206 +------------------+--------------------------------------+
2207 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2208 +------------------+--------------------------------------+
2209 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2210 +------------------+--------------------------------------+
2211 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2212 +------------------+--------------------------------------+
2214 Here are some examples of multidimensional arrays:
2216 +-----------------------------+----------------------------------------------------------+
2217 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2218 +-----------------------------+----------------------------------------------------------+
2219 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2220 +-----------------------------+----------------------------------------------------------+
2221 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2222 +-----------------------------+----------------------------------------------------------+
2224 There is no restriction on indexing beyond the end of the array implied
2225 by a static type (though there are restrictions on indexing beyond the
2226 bounds of an allocated object in some cases). This means that
2227 single-dimension 'variable sized array' addressing can be implemented in
2228 LLVM with a zero length array type. An implementation of 'pascal style
2229 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2239 The structure type is used to represent a collection of data members
2240 together in memory. The elements of a structure may be any type that has
2243 Structures in memory are accessed using '``load``' and '``store``' by
2244 getting a pointer to a field with the '``getelementptr``' instruction.
2245 Structures in registers are accessed using the '``extractvalue``' and
2246 '``insertvalue``' instructions.
2248 Structures may optionally be "packed" structures, which indicate that
2249 the alignment of the struct is one byte, and that there is no padding
2250 between the elements. In non-packed structs, padding between field types
2251 is inserted as defined by the DataLayout string in the module, which is
2252 required to match what the underlying code generator expects.
2254 Structures can either be "literal" or "identified". A literal structure
2255 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2256 identified types are always defined at the top level with a name.
2257 Literal types are uniqued by their contents and can never be recursive
2258 or opaque since there is no way to write one. Identified types can be
2259 recursive, can be opaqued, and are never uniqued.
2265 %T1 = type { <type list> } ; Identified normal struct type
2266 %T2 = type <{ <type list> }> ; Identified packed struct type
2270 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2271 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2272 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2273 | ``{ 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``. |
2274 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2275 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2276 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2280 Opaque Structure Types
2281 """"""""""""""""""""""
2285 Opaque structure types are used to represent named structure types that
2286 do not have a body specified. This corresponds (for example) to the C
2287 notion of a forward declared structure.
2298 +--------------+-------------------+
2299 | ``opaque`` | An opaque type. |
2300 +--------------+-------------------+
2307 LLVM has several different basic types of constants. This section
2308 describes them all and their syntax.
2313 **Boolean constants**
2314 The two strings '``true``' and '``false``' are both valid constants
2316 **Integer constants**
2317 Standard integers (such as '4') are constants of the
2318 :ref:`integer <t_integer>` type. Negative numbers may be used with
2320 **Floating point constants**
2321 Floating point constants use standard decimal notation (e.g.
2322 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2323 hexadecimal notation (see below). The assembler requires the exact
2324 decimal value of a floating-point constant. For example, the
2325 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2326 decimal in binary. Floating point constants must have a :ref:`floating
2327 point <t_floating>` type.
2328 **Null pointer constants**
2329 The identifier '``null``' is recognized as a null pointer constant
2330 and must be of :ref:`pointer type <t_pointer>`.
2332 The one non-intuitive notation for constants is the hexadecimal form of
2333 floating point constants. For example, the form
2334 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2335 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2336 constants are required (and the only time that they are generated by the
2337 disassembler) is when a floating point constant must be emitted but it
2338 cannot be represented as a decimal floating point number in a reasonable
2339 number of digits. For example, NaN's, infinities, and other special
2340 values are represented in their IEEE hexadecimal format so that assembly
2341 and disassembly do not cause any bits to change in the constants.
2343 When using the hexadecimal form, constants of types half, float, and
2344 double are represented using the 16-digit form shown above (which
2345 matches the IEEE754 representation for double); half and float values
2346 must, however, be exactly representable as IEEE 754 half and single
2347 precision, respectively. Hexadecimal format is always used for long
2348 double, and there are three forms of long double. The 80-bit format used
2349 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2350 128-bit format used by PowerPC (two adjacent doubles) is represented by
2351 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2352 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2353 will only work if they match the long double format on your target.
2354 The IEEE 16-bit format (half precision) is represented by ``0xH``
2355 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2356 (sign bit at the left).
2358 There are no constants of type x86_mmx.
2360 .. _complexconstants:
2365 Complex constants are a (potentially recursive) combination of simple
2366 constants and smaller complex constants.
2368 **Structure constants**
2369 Structure constants are represented with notation similar to
2370 structure type definitions (a comma separated list of elements,
2371 surrounded by braces (``{}``)). For example:
2372 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2373 "``@G = external global i32``". Structure constants must have
2374 :ref:`structure type <t_struct>`, and the number and types of elements
2375 must match those specified by the type.
2377 Array constants are represented with notation similar to array type
2378 definitions (a comma separated list of elements, surrounded by
2379 square brackets (``[]``)). For example:
2380 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2381 :ref:`array type <t_array>`, and the number and types of elements must
2382 match those specified by the type. As a special case, character array
2383 constants may also be represented as a double-quoted string using the ``c``
2384 prefix. For example: "``c"Hello World\0A\00"``".
2385 **Vector constants**
2386 Vector constants are represented with notation similar to vector
2387 type definitions (a comma separated list of elements, surrounded by
2388 less-than/greater-than's (``<>``)). For example:
2389 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2390 must have :ref:`vector type <t_vector>`, and the number and types of
2391 elements must match those specified by the type.
2392 **Zero initialization**
2393 The string '``zeroinitializer``' can be used to zero initialize a
2394 value to zero of *any* type, including scalar and
2395 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2396 having to print large zero initializers (e.g. for large arrays) and
2397 is always exactly equivalent to using explicit zero initializers.
2399 A metadata node is a constant tuple without types. For example:
2400 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2401 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2402 Unlike other typed constants that are meant to be interpreted as part of
2403 the instruction stream, metadata is a place to attach additional
2404 information such as debug info.
2406 Global Variable and Function Addresses
2407 --------------------------------------
2409 The addresses of :ref:`global variables <globalvars>` and
2410 :ref:`functions <functionstructure>` are always implicitly valid
2411 (link-time) constants. These constants are explicitly referenced when
2412 the :ref:`identifier for the global <identifiers>` is used and always have
2413 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2416 .. code-block:: llvm
2420 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2427 The string '``undef``' can be used anywhere a constant is expected, and
2428 indicates that the user of the value may receive an unspecified
2429 bit-pattern. Undefined values may be of any type (other than '``label``'
2430 or '``void``') and be used anywhere a constant is permitted.
2432 Undefined values are useful because they indicate to the compiler that
2433 the program is well defined no matter what value is used. This gives the
2434 compiler more freedom to optimize. Here are some examples of
2435 (potentially surprising) transformations that are valid (in pseudo IR):
2437 .. code-block:: llvm
2447 This is safe because all of the output bits are affected by the undef
2448 bits. Any output bit can have a zero or one depending on the input bits.
2450 .. code-block:: llvm
2461 These logical operations have bits that are not always affected by the
2462 input. For example, if ``%X`` has a zero bit, then the output of the
2463 '``and``' operation will always be a zero for that bit, no matter what
2464 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2465 optimize or assume that the result of the '``and``' is '``undef``'.
2466 However, it is safe to assume that all bits of the '``undef``' could be
2467 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2468 all the bits of the '``undef``' operand to the '``or``' could be set,
2469 allowing the '``or``' to be folded to -1.
2471 .. code-block:: llvm
2473 %A = select undef, %X, %Y
2474 %B = select undef, 42, %Y
2475 %C = select %X, %Y, undef
2485 This set of examples shows that undefined '``select``' (and conditional
2486 branch) conditions can go *either way*, but they have to come from one
2487 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2488 both known to have a clear low bit, then ``%A`` would have to have a
2489 cleared low bit. However, in the ``%C`` example, the optimizer is
2490 allowed to assume that the '``undef``' operand could be the same as
2491 ``%Y``, allowing the whole '``select``' to be eliminated.
2493 .. code-block:: llvm
2495 %A = xor undef, undef
2512 This example points out that two '``undef``' operands are not
2513 necessarily the same. This can be surprising to people (and also matches
2514 C semantics) where they assume that "``X^X``" is always zero, even if
2515 ``X`` is undefined. This isn't true for a number of reasons, but the
2516 short answer is that an '``undef``' "variable" can arbitrarily change
2517 its value over its "live range". This is true because the variable
2518 doesn't actually *have a live range*. Instead, the value is logically
2519 read from arbitrary registers that happen to be around when needed, so
2520 the value is not necessarily consistent over time. In fact, ``%A`` and
2521 ``%C`` need to have the same semantics or the core LLVM "replace all
2522 uses with" concept would not hold.
2524 .. code-block:: llvm
2532 These examples show the crucial difference between an *undefined value*
2533 and *undefined behavior*. An undefined value (like '``undef``') is
2534 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2535 operation can be constant folded to '``undef``', because the '``undef``'
2536 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2537 However, in the second example, we can make a more aggressive
2538 assumption: because the ``undef`` is allowed to be an arbitrary value,
2539 we are allowed to assume that it could be zero. Since a divide by zero
2540 has *undefined behavior*, we are allowed to assume that the operation
2541 does not execute at all. This allows us to delete the divide and all
2542 code after it. Because the undefined operation "can't happen", the
2543 optimizer can assume that it occurs in dead code.
2545 .. code-block:: llvm
2547 a: store undef -> %X
2548 b: store %X -> undef
2553 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2554 value can be assumed to not have any effect; we can assume that the
2555 value is overwritten with bits that happen to match what was already
2556 there. However, a store *to* an undefined location could clobber
2557 arbitrary memory, therefore, it has undefined behavior.
2564 Poison values are similar to :ref:`undef values <undefvalues>`, however
2565 they also represent the fact that an instruction or constant expression
2566 that cannot evoke side effects has nevertheless detected a condition
2567 that results in undefined behavior.
2569 There is currently no way of representing a poison value in the IR; they
2570 only exist when produced by operations such as :ref:`add <i_add>` with
2573 Poison value behavior is defined in terms of value *dependence*:
2575 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2576 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2577 their dynamic predecessor basic block.
2578 - Function arguments depend on the corresponding actual argument values
2579 in the dynamic callers of their functions.
2580 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2581 instructions that dynamically transfer control back to them.
2582 - :ref:`Invoke <i_invoke>` instructions depend on the
2583 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2584 call instructions that dynamically transfer control back to them.
2585 - Non-volatile loads and stores depend on the most recent stores to all
2586 of the referenced memory addresses, following the order in the IR
2587 (including loads and stores implied by intrinsics such as
2588 :ref:`@llvm.memcpy <int_memcpy>`.)
2589 - An instruction with externally visible side effects depends on the
2590 most recent preceding instruction with externally visible side
2591 effects, following the order in the IR. (This includes :ref:`volatile
2592 operations <volatile>`.)
2593 - An instruction *control-depends* on a :ref:`terminator
2594 instruction <terminators>` if the terminator instruction has
2595 multiple successors and the instruction is always executed when
2596 control transfers to one of the successors, and may not be executed
2597 when control is transferred to another.
2598 - Additionally, an instruction also *control-depends* on a terminator
2599 instruction if the set of instructions it otherwise depends on would
2600 be different if the terminator had transferred control to a different
2602 - Dependence is transitive.
2604 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2605 with the additional effect that any instruction that has a *dependence*
2606 on a poison value has undefined behavior.
2608 Here are some examples:
2610 .. code-block:: llvm
2613 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2614 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2615 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
2616 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2618 store i32 %poison, i32* @g ; Poison value stored to memory.
2619 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
2621 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2623 %narrowaddr = bitcast i32* @g to i16*
2624 %wideaddr = bitcast i32* @g to i64*
2625 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2626 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
2628 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2629 br i1 %cmp, label %true, label %end ; Branch to either destination.
2632 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2633 ; it has undefined behavior.
2637 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2638 ; Both edges into this PHI are
2639 ; control-dependent on %cmp, so this
2640 ; always results in a poison value.
2642 store volatile i32 0, i32* @g ; This would depend on the store in %true
2643 ; if %cmp is true, or the store in %entry
2644 ; otherwise, so this is undefined behavior.
2646 br i1 %cmp, label %second_true, label %second_end
2647 ; The same branch again, but this time the
2648 ; true block doesn't have side effects.
2655 store volatile i32 0, i32* @g ; This time, the instruction always depends
2656 ; on the store in %end. Also, it is
2657 ; control-equivalent to %end, so this is
2658 ; well-defined (ignoring earlier undefined
2659 ; behavior in this example).
2663 Addresses of Basic Blocks
2664 -------------------------
2666 ``blockaddress(@function, %block)``
2668 The '``blockaddress``' constant computes the address of the specified
2669 basic block in the specified function, and always has an ``i8*`` type.
2670 Taking the address of the entry block is illegal.
2672 This value only has defined behavior when used as an operand to the
2673 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2674 against null. Pointer equality tests between labels addresses results in
2675 undefined behavior --- though, again, comparison against null is ok, and
2676 no label is equal to the null pointer. This may be passed around as an
2677 opaque pointer sized value as long as the bits are not inspected. This
2678 allows ``ptrtoint`` and arithmetic to be performed on these values so
2679 long as the original value is reconstituted before the ``indirectbr``
2682 Finally, some targets may provide defined semantics when using the value
2683 as the operand to an inline assembly, but that is target specific.
2687 Constant Expressions
2688 --------------------
2690 Constant expressions are used to allow expressions involving other
2691 constants to be used as constants. Constant expressions may be of any
2692 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2693 that does not have side effects (e.g. load and call are not supported).
2694 The following is the syntax for constant expressions:
2696 ``trunc (CST to TYPE)``
2697 Truncate a constant to another type. The bit size of CST must be
2698 larger than the bit size of TYPE. Both types must be integers.
2699 ``zext (CST to TYPE)``
2700 Zero extend a constant to another type. The bit size of CST must be
2701 smaller than the bit size of TYPE. Both types must be integers.
2702 ``sext (CST to TYPE)``
2703 Sign extend a constant to another type. The bit size of CST must be
2704 smaller than the bit size of TYPE. Both types must be integers.
2705 ``fptrunc (CST to TYPE)``
2706 Truncate a floating point constant to another floating point type.
2707 The size of CST must be larger than the size of TYPE. Both types
2708 must be floating point.
2709 ``fpext (CST to TYPE)``
2710 Floating point extend a constant to another type. The size of CST
2711 must be smaller or equal to the size of TYPE. Both types must be
2713 ``fptoui (CST to TYPE)``
2714 Convert a floating point constant to the corresponding unsigned
2715 integer constant. TYPE must be a scalar or vector integer type. CST
2716 must be of scalar or vector floating point type. Both CST and TYPE
2717 must be scalars, or vectors of the same number of elements. If the
2718 value won't fit in the integer type, the results are undefined.
2719 ``fptosi (CST to TYPE)``
2720 Convert a floating point constant to the corresponding signed
2721 integer constant. TYPE must be a scalar or vector integer type. CST
2722 must be of scalar or vector floating point type. Both CST and TYPE
2723 must be scalars, or vectors of the same number of elements. If the
2724 value won't fit in the integer type, the results are undefined.
2725 ``uitofp (CST to TYPE)``
2726 Convert an unsigned integer constant to the corresponding floating
2727 point constant. TYPE must be a scalar or vector floating point type.
2728 CST must be of scalar or vector integer type. Both CST and TYPE must
2729 be scalars, or vectors of the same number of elements. If the value
2730 won't fit in the floating point type, the results are undefined.
2731 ``sitofp (CST to TYPE)``
2732 Convert a signed integer constant to the corresponding floating
2733 point constant. TYPE must be a scalar or vector floating point type.
2734 CST must be of scalar or vector integer type. Both CST and TYPE must
2735 be scalars, or vectors of the same number of elements. If the value
2736 won't fit in the floating point type, the results are undefined.
2737 ``ptrtoint (CST to TYPE)``
2738 Convert a pointer typed constant to the corresponding integer
2739 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2740 pointer type. The ``CST`` value is zero extended, truncated, or
2741 unchanged to make it fit in ``TYPE``.
2742 ``inttoptr (CST to TYPE)``
2743 Convert an integer constant to a pointer constant. TYPE must be a
2744 pointer type. CST must be of integer type. The CST value is zero
2745 extended, truncated, or unchanged to make it fit in a pointer size.
2746 This one is *really* dangerous!
2747 ``bitcast (CST to TYPE)``
2748 Convert a constant, CST, to another TYPE. The constraints of the
2749 operands are the same as those for the :ref:`bitcast
2750 instruction <i_bitcast>`.
2751 ``addrspacecast (CST to TYPE)``
2752 Convert a constant pointer or constant vector of pointer, CST, to another
2753 TYPE in a different address space. The constraints of the operands are the
2754 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2755 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
2756 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2757 constants. As with the :ref:`getelementptr <i_getelementptr>`
2758 instruction, the index list may have zero or more indexes, which are
2759 required to make sense for the type of "pointer to TY".
2760 ``select (COND, VAL1, VAL2)``
2761 Perform the :ref:`select operation <i_select>` on constants.
2762 ``icmp COND (VAL1, VAL2)``
2763 Performs the :ref:`icmp operation <i_icmp>` on constants.
2764 ``fcmp COND (VAL1, VAL2)``
2765 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2766 ``extractelement (VAL, IDX)``
2767 Perform the :ref:`extractelement operation <i_extractelement>` on
2769 ``insertelement (VAL, ELT, IDX)``
2770 Perform the :ref:`insertelement operation <i_insertelement>` on
2772 ``shufflevector (VEC1, VEC2, IDXMASK)``
2773 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2775 ``extractvalue (VAL, IDX0, IDX1, ...)``
2776 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2777 constants. The index list is interpreted in a similar manner as
2778 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2779 least one index value must be specified.
2780 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2781 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2782 The index list is interpreted in a similar manner as indices in a
2783 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2784 value must be specified.
2785 ``OPCODE (LHS, RHS)``
2786 Perform the specified operation of the LHS and RHS constants. OPCODE
2787 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2788 binary <bitwiseops>` operations. The constraints on operands are
2789 the same as those for the corresponding instruction (e.g. no bitwise
2790 operations on floating point values are allowed).
2797 Inline Assembler Expressions
2798 ----------------------------
2800 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2801 Inline Assembly <moduleasm>`) through the use of a special value. This
2802 value represents the inline assembler as a string (containing the
2803 instructions to emit), a list of operand constraints (stored as a
2804 string), a flag that indicates whether or not the inline asm expression
2805 has side effects, and a flag indicating whether the function containing
2806 the asm needs to align its stack conservatively. An example inline
2807 assembler expression is:
2809 .. code-block:: llvm
2811 i32 (i32) asm "bswap $0", "=r,r"
2813 Inline assembler expressions may **only** be used as the callee operand
2814 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2815 Thus, typically we have:
2817 .. code-block:: llvm
2819 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2821 Inline asms with side effects not visible in the constraint list must be
2822 marked as having side effects. This is done through the use of the
2823 '``sideeffect``' keyword, like so:
2825 .. code-block:: llvm
2827 call void asm sideeffect "eieio", ""()
2829 In some cases inline asms will contain code that will not work unless
2830 the stack is aligned in some way, such as calls or SSE instructions on
2831 x86, yet will not contain code that does that alignment within the asm.
2832 The compiler should make conservative assumptions about what the asm
2833 might contain and should generate its usual stack alignment code in the
2834 prologue if the '``alignstack``' keyword is present:
2836 .. code-block:: llvm
2838 call void asm alignstack "eieio", ""()
2840 Inline asms also support using non-standard assembly dialects. The
2841 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2842 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2843 the only supported dialects. An example is:
2845 .. code-block:: llvm
2847 call void asm inteldialect "eieio", ""()
2849 If multiple keywords appear the '``sideeffect``' keyword must come
2850 first, the '``alignstack``' keyword second and the '``inteldialect``'
2856 The call instructions that wrap inline asm nodes may have a
2857 "``!srcloc``" MDNode attached to it that contains a list of constant
2858 integers. If present, the code generator will use the integer as the
2859 location cookie value when report errors through the ``LLVMContext``
2860 error reporting mechanisms. This allows a front-end to correlate backend
2861 errors that occur with inline asm back to the source code that produced
2864 .. code-block:: llvm
2866 call void asm sideeffect "something bad", ""(), !srcloc !42
2868 !42 = !{ i32 1234567 }
2870 It is up to the front-end to make sense of the magic numbers it places
2871 in the IR. If the MDNode contains multiple constants, the code generator
2872 will use the one that corresponds to the line of the asm that the error
2880 LLVM IR allows metadata to be attached to instructions in the program
2881 that can convey extra information about the code to the optimizers and
2882 code generator. One example application of metadata is source-level
2883 debug information. There are two metadata primitives: strings and nodes.
2885 Metadata does not have a type, and is not a value. If referenced from a
2886 ``call`` instruction, it uses the ``metadata`` type.
2888 All metadata are identified in syntax by a exclamation point ('``!``').
2890 .. _metadata-string:
2892 Metadata Nodes and Metadata Strings
2893 -----------------------------------
2895 A metadata string is a string surrounded by double quotes. It can
2896 contain any character by escaping non-printable characters with
2897 "``\xx``" where "``xx``" is the two digit hex code. For example:
2900 Metadata nodes are represented with notation similar to structure
2901 constants (a comma separated list of elements, surrounded by braces and
2902 preceded by an exclamation point). Metadata nodes can have any values as
2903 their operand. For example:
2905 .. code-block:: llvm
2907 !{ !"test\00", i32 10}
2909 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
2911 .. code-block:: llvm
2913 !0 = distinct !{!"test\00", i32 10}
2915 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
2916 content. They can also occur when transformations cause uniquing collisions
2917 when metadata operands change.
2919 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2920 metadata nodes, which can be looked up in the module symbol table. For
2923 .. code-block:: llvm
2927 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2928 function is using two metadata arguments:
2930 .. code-block:: llvm
2932 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2934 Metadata can be attached with an instruction. Here metadata ``!21`` is
2935 attached to the ``add`` instruction using the ``!dbg`` identifier:
2937 .. code-block:: llvm
2939 %indvar.next = add i64 %indvar, 1, !dbg !21
2941 More information about specific metadata nodes recognized by the
2942 optimizers and code generator is found below.
2944 .. _specialized-metadata:
2946 Specialized Metadata Nodes
2947 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2949 Specialized metadata nodes are custom data structures in metadata (as opposed
2950 to generic tuples). Their fields are labelled, and can be specified in any
2953 These aren't inherently debug info centric, but currently all the specialized
2954 metadata nodes are related to debug info.
2961 ``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
2962 ``retainedTypes:``, ``subprograms:``, ``globals:`` and ``imports:`` fields are
2963 tuples containing the debug info to be emitted along with the compile unit,
2964 regardless of code optimizations (some nodes are only emitted if there are
2965 references to them from instructions).
2967 .. code-block:: llvm
2969 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
2970 isOptimized: true, flags: "-O2", runtimeVersion: 2,
2971 splitDebugFilename: "abc.debug", emissionKind: 1,
2972 enums: !2, retainedTypes: !3, subprograms: !4,
2973 globals: !5, imports: !6)
2975 Compile unit descriptors provide the root scope for objects declared in a
2976 specific compilation unit. File descriptors are defined using this scope.
2977 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
2978 keep track of subprograms, global variables, type information, and imported
2979 entities (declarations and namespaces).
2986 ``DIFile`` nodes represent files. The ``filename:`` can include slashes.
2988 .. code-block:: llvm
2990 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
2992 Files are sometimes used in ``scope:`` fields, and are the only valid target
2993 for ``file:`` fields.
3000 ``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
3001 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
3003 .. code-block:: llvm
3005 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3006 encoding: DW_ATE_unsigned_char)
3007 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
3009 The ``encoding:`` describes the details of the type. Usually it's one of the
3012 .. code-block:: llvm
3018 DW_ATE_signed_char = 6
3020 DW_ATE_unsigned_char = 8
3022 .. _DISubroutineType:
3027 ``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
3028 refers to a tuple; the first operand is the return type, while the rest are the
3029 types of the formal arguments in order. If the first operand is ``null``, that
3030 represents a function with no return value (such as ``void foo() {}`` in C++).
3032 .. code-block:: llvm
3034 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
3035 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
3036 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
3043 ``DIDerivedType`` nodes represent types derived from other types, such as
3046 .. code-block:: llvm
3048 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3049 encoding: DW_ATE_unsigned_char)
3050 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
3053 The following ``tag:`` values are valid:
3055 .. code-block:: llvm
3057 DW_TAG_formal_parameter = 5
3059 DW_TAG_pointer_type = 15
3060 DW_TAG_reference_type = 16
3062 DW_TAG_ptr_to_member_type = 31
3063 DW_TAG_const_type = 38
3064 DW_TAG_volatile_type = 53
3065 DW_TAG_restrict_type = 55
3067 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
3068 <DICompositeType>` or :ref:`subprogram <DISubprogram>`. The type of the member
3069 is the ``baseType:``. The ``offset:`` is the member's bit offset.
3070 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
3071 argument of a subprogram.
3073 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3075 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3076 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3079 Note that the ``void *`` type is expressed as a type derived from NULL.
3081 .. _DICompositeType:
3086 ``DICompositeType`` nodes represent types composed of other types, like
3087 structures and unions. ``elements:`` points to a tuple of the composed types.
3089 If the source language supports ODR, the ``identifier:`` field gives the unique
3090 identifier used for type merging between modules. When specified, other types
3091 can refer to composite types indirectly via a :ref:`metadata string
3092 <metadata-string>` that matches their identifier.
3094 .. code-block:: llvm
3096 !0 = !DIEnumerator(name: "SixKind", value: 7)
3097 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3098 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3099 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
3100 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3101 elements: !{!0, !1, !2})
3103 The following ``tag:`` values are valid:
3105 .. code-block:: llvm
3107 DW_TAG_array_type = 1
3108 DW_TAG_class_type = 2
3109 DW_TAG_enumeration_type = 4
3110 DW_TAG_structure_type = 19
3111 DW_TAG_union_type = 23
3112 DW_TAG_subroutine_type = 21
3113 DW_TAG_inheritance = 28
3116 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
3117 descriptors <DISubrange>`, each representing the range of subscripts at that
3118 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3119 array type is a native packed vector.
3121 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
3122 descriptors <DIEnumerator>`, each representing the definition of an enumeration
3123 value for the set. All enumeration type descriptors are collected in the
3124 ``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
3126 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3127 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
3128 <DIDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
3135 ``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3136 :ref:`DICompositeType`. ``count: -1`` indicates an empty array.
3138 .. code-block:: llvm
3140 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
3141 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
3142 !2 = !DISubrange(count: -1) ; empty array.
3149 ``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3150 variants of :ref:`DICompositeType`.
3152 .. code-block:: llvm
3154 !0 = !DIEnumerator(name: "SixKind", value: 7)
3155 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3156 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3158 DITemplateTypeParameter
3159 """""""""""""""""""""""
3161 ``DITemplateTypeParameter`` nodes represent type parameters to generic source
3162 language constructs. They are used (optionally) in :ref:`DICompositeType` and
3163 :ref:`DISubprogram` ``templateParams:`` fields.
3165 .. code-block:: llvm
3167 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
3169 DITemplateValueParameter
3170 """"""""""""""""""""""""
3172 ``DITemplateValueParameter`` nodes represent value parameters to generic source
3173 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3174 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3175 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
3176 :ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
3178 .. code-block:: llvm
3180 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
3185 ``DINamespace`` nodes represent namespaces in the source language.
3187 .. code-block:: llvm
3189 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
3194 ``DIGlobalVariable`` nodes represent global variables in the source language.
3196 .. code-block:: llvm
3198 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
3199 file: !2, line: 7, type: !3, isLocal: true,
3200 isDefinition: false, variable: i32* @foo,
3203 All global variables should be referenced by the `globals:` field of a
3204 :ref:`compile unit <DICompileUnit>`.
3211 ``DISubprogram`` nodes represent functions from the source language. The
3212 ``variables:`` field points at :ref:`variables <DILocalVariable>` that must be
3213 retained, even if their IR counterparts are optimized out of the IR. The
3214 ``type:`` field must point at an :ref:`DISubroutineType`.
3216 .. code-block:: llvm
3218 !0 = !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
3219 file: !2, line: 7, type: !3, isLocal: true,
3220 isDefinition: false, scopeLine: 8, containingType: !4,
3221 virtuality: DW_VIRTUALITY_pure_virtual, virtualIndex: 10,
3222 flags: DIFlagPrototyped, isOptimized: true,
3223 function: void ()* @_Z3foov,
3224 templateParams: !5, declaration: !6, variables: !7)
3231 ``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
3232 <DISubprogram>`. The line number and column numbers are used to dinstinguish
3233 two lexical blocks at same depth. They are valid targets for ``scope:``
3236 .. code-block:: llvm
3238 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
3240 Usually lexical blocks are ``distinct`` to prevent node merging based on
3243 .. _DILexicalBlockFile:
3248 ``DILexicalBlockFile`` nodes are used to discriminate between sections of a
3249 :ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
3250 indicate textual inclusion, or the ``discriminator:`` field can be used to
3251 discriminate between control flow within a single block in the source language.
3253 .. code-block:: llvm
3255 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
3256 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
3257 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
3264 ``DILocation`` nodes represent source debug locations. The ``scope:`` field is
3265 mandatory, and points at an :ref:`DILexicalBlockFile`, an
3266 :ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
3268 .. code-block:: llvm
3270 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
3272 .. _DILocalVariable:
3277 ``DILocalVariable`` nodes represent local variables in the source language.
3278 Instead of ``DW_TAG_variable``, they use LLVM-specific fake tags to
3279 discriminate between local variables (``DW_TAG_auto_variable``) and subprogram
3280 arguments (``DW_TAG_arg_variable``). In the latter case, the ``arg:`` field
3281 specifies the argument position, and this variable will be included in the
3282 ``variables:`` field of its :ref:`DISubprogram`.
3284 .. code-block:: llvm
3286 !0 = !DILocalVariable(tag: DW_TAG_arg_variable, name: "this", arg: 0,
3287 scope: !3, file: !2, line: 7, type: !3,
3288 flags: DIFlagArtificial)
3289 !1 = !DILocalVariable(tag: DW_TAG_arg_variable, name: "x", arg: 1,
3290 scope: !4, file: !2, line: 7, type: !3)
3291 !1 = !DILocalVariable(tag: DW_TAG_auto_variable, name: "y",
3292 scope: !5, file: !2, line: 7, type: !3)
3297 ``DIExpression`` nodes represent DWARF expression sequences. They are used in
3298 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
3299 describe how the referenced LLVM variable relates to the source language
3302 The current supported vocabulary is limited:
3304 - ``DW_OP_deref`` dereferences the working expression.
3305 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
3306 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
3307 here, respectively) of the variable piece from the working expression.
3309 .. code-block:: llvm
3311 !0 = !DIExpression(DW_OP_deref)
3312 !1 = !DIExpression(DW_OP_plus, 3)
3313 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
3314 !3 = !DIExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
3319 ``DIObjCProperty`` nodes represent Objective-C property nodes.
3321 .. code-block:: llvm
3323 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
3324 getter: "getFoo", attributes: 7, type: !2)
3329 ``DIImportedEntity`` nodes represent entities (such as modules) imported into a
3332 .. code-block:: llvm
3334 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
3335 entity: !1, line: 7)
3340 In LLVM IR, memory does not have types, so LLVM's own type system is not
3341 suitable for doing TBAA. Instead, metadata is added to the IR to
3342 describe a type system of a higher level language. This can be used to
3343 implement typical C/C++ TBAA, but it can also be used to implement
3344 custom alias analysis behavior for other languages.
3346 The current metadata format is very simple. TBAA metadata nodes have up
3347 to three fields, e.g.:
3349 .. code-block:: llvm
3351 !0 = !{ !"an example type tree" }
3352 !1 = !{ !"int", !0 }
3353 !2 = !{ !"float", !0 }
3354 !3 = !{ !"const float", !2, i64 1 }
3356 The first field is an identity field. It can be any value, usually a
3357 metadata string, which uniquely identifies the type. The most important
3358 name in the tree is the name of the root node. Two trees with different
3359 root node names are entirely disjoint, even if they have leaves with
3362 The second field identifies the type's parent node in the tree, or is
3363 null or omitted for a root node. A type is considered to alias all of
3364 its descendants and all of its ancestors in the tree. Also, a type is
3365 considered to alias all types in other trees, so that bitcode produced
3366 from multiple front-ends is handled conservatively.
3368 If the third field is present, it's an integer which if equal to 1
3369 indicates that the type is "constant" (meaning
3370 ``pointsToConstantMemory`` should return true; see `other useful
3371 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
3373 '``tbaa.struct``' Metadata
3374 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3376 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
3377 aggregate assignment operations in C and similar languages, however it
3378 is defined to copy a contiguous region of memory, which is more than
3379 strictly necessary for aggregate types which contain holes due to
3380 padding. Also, it doesn't contain any TBAA information about the fields
3383 ``!tbaa.struct`` metadata can describe which memory subregions in a
3384 memcpy are padding and what the TBAA tags of the struct are.
3386 The current metadata format is very simple. ``!tbaa.struct`` metadata
3387 nodes are a list of operands which are in conceptual groups of three.
3388 For each group of three, the first operand gives the byte offset of a
3389 field in bytes, the second gives its size in bytes, and the third gives
3392 .. code-block:: llvm
3394 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
3396 This describes a struct with two fields. The first is at offset 0 bytes
3397 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
3398 and has size 4 bytes and has tbaa tag !2.
3400 Note that the fields need not be contiguous. In this example, there is a
3401 4 byte gap between the two fields. This gap represents padding which
3402 does not carry useful data and need not be preserved.
3404 '``noalias``' and '``alias.scope``' Metadata
3405 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3407 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
3408 noalias memory-access sets. This means that some collection of memory access
3409 instructions (loads, stores, memory-accessing calls, etc.) that carry
3410 ``noalias`` metadata can specifically be specified not to alias with some other
3411 collection of memory access instructions that carry ``alias.scope`` metadata.
3412 Each type of metadata specifies a list of scopes where each scope has an id and
3413 a domain. When evaluating an aliasing query, if for some domain, the set
3414 of scopes with that domain in one instruction's ``alias.scope`` list is a
3415 subset of (or equal to) the set of scopes for that domain in another
3416 instruction's ``noalias`` list, then the two memory accesses are assumed not to
3419 The metadata identifying each domain is itself a list containing one or two
3420 entries. The first entry is the name of the domain. Note that if the name is a
3421 string then it can be combined accross functions and translation units. A
3422 self-reference can be used to create globally unique domain names. A
3423 descriptive string may optionally be provided as a second list entry.
3425 The metadata identifying each scope is also itself a list containing two or
3426 three entries. The first entry is the name of the scope. Note that if the name
3427 is a string then it can be combined accross functions and translation units. A
3428 self-reference can be used to create globally unique scope names. A metadata
3429 reference to the scope's domain is the second entry. A descriptive string may
3430 optionally be provided as a third list entry.
3434 .. code-block:: llvm
3436 ; Two scope domains:
3440 ; Some scopes in these domains:
3446 !5 = !{!4} ; A list containing only scope !4
3450 ; These two instructions don't alias:
3451 %0 = load float, float* %c, align 4, !alias.scope !5
3452 store float %0, float* %arrayidx.i, align 4, !noalias !5
3454 ; These two instructions also don't alias (for domain !1, the set of scopes
3455 ; in the !alias.scope equals that in the !noalias list):
3456 %2 = load float, float* %c, align 4, !alias.scope !5
3457 store float %2, float* %arrayidx.i2, align 4, !noalias !6
3459 ; These two instructions may alias (for domain !0, the set of scopes in
3460 ; the !noalias list is not a superset of, or equal to, the scopes in the
3461 ; !alias.scope list):
3462 %2 = load float, float* %c, align 4, !alias.scope !6
3463 store float %0, float* %arrayidx.i, align 4, !noalias !7
3465 '``fpmath``' Metadata
3466 ^^^^^^^^^^^^^^^^^^^^^
3468 ``fpmath`` metadata may be attached to any instruction of floating point
3469 type. It can be used to express the maximum acceptable error in the
3470 result of that instruction, in ULPs, thus potentially allowing the
3471 compiler to use a more efficient but less accurate method of computing
3472 it. ULP is defined as follows:
3474 If ``x`` is a real number that lies between two finite consecutive
3475 floating-point numbers ``a`` and ``b``, without being equal to one
3476 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
3477 distance between the two non-equal finite floating-point numbers
3478 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
3480 The metadata node shall consist of a single positive floating point
3481 number representing the maximum relative error, for example:
3483 .. code-block:: llvm
3485 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
3489 '``range``' Metadata
3490 ^^^^^^^^^^^^^^^^^^^^
3492 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
3493 integer types. It expresses the possible ranges the loaded value or the value
3494 returned by the called function at this call site is in. The ranges are
3495 represented with a flattened list of integers. The loaded value or the value
3496 returned is known to be in the union of the ranges defined by each consecutive
3497 pair. Each pair has the following properties:
3499 - The type must match the type loaded by the instruction.
3500 - The pair ``a,b`` represents the range ``[a,b)``.
3501 - Both ``a`` and ``b`` are constants.
3502 - The range is allowed to wrap.
3503 - The range should not represent the full or empty set. That is,
3506 In addition, the pairs must be in signed order of the lower bound and
3507 they must be non-contiguous.
3511 .. code-block:: llvm
3513 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
3514 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3515 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
3516 %d = invoke i8 @bar() to label %cont
3517 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
3519 !0 = !{ i8 0, i8 2 }
3520 !1 = !{ i8 255, i8 2 }
3521 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
3522 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
3527 It is sometimes useful to attach information to loop constructs. Currently,
3528 loop metadata is implemented as metadata attached to the branch instruction
3529 in the loop latch block. This type of metadata refer to a metadata node that is
3530 guaranteed to be separate for each loop. The loop identifier metadata is
3531 specified with the name ``llvm.loop``.
3533 The loop identifier metadata is implemented using a metadata that refers to
3534 itself to avoid merging it with any other identifier metadata, e.g.,
3535 during module linkage or function inlining. That is, each loop should refer
3536 to their own identification metadata even if they reside in separate functions.
3537 The following example contains loop identifier metadata for two separate loop
3540 .. code-block:: llvm
3545 The loop identifier metadata can be used to specify additional
3546 per-loop metadata. Any operands after the first operand can be treated
3547 as user-defined metadata. For example the ``llvm.loop.unroll.count``
3548 suggests an unroll factor to the loop unroller:
3550 .. code-block:: llvm
3552 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
3555 !1 = !{!"llvm.loop.unroll.count", i32 4}
3557 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
3558 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3560 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
3561 used to control per-loop vectorization and interleaving parameters such as
3562 vectorization width and interleave count. These metadata should be used in
3563 conjunction with ``llvm.loop`` loop identification metadata. The
3564 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
3565 optimization hints and the optimizer will only interleave and vectorize loops if
3566 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
3567 which contains information about loop-carried memory dependencies can be helpful
3568 in determining the safety of these transformations.
3570 '``llvm.loop.interleave.count``' Metadata
3571 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3573 This metadata suggests an interleave count to the loop interleaver.
3574 The first operand is the string ``llvm.loop.interleave.count`` and the
3575 second operand is an integer specifying the interleave count. For
3578 .. code-block:: llvm
3580 !0 = !{!"llvm.loop.interleave.count", i32 4}
3582 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
3583 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
3584 then the interleave count will be determined automatically.
3586 '``llvm.loop.vectorize.enable``' Metadata
3587 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3589 This metadata selectively enables or disables vectorization for the loop. The
3590 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
3591 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
3592 0 disables vectorization:
3594 .. code-block:: llvm
3596 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
3597 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
3599 '``llvm.loop.vectorize.width``' Metadata
3600 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3602 This metadata sets the target width of the vectorizer. The first
3603 operand is the string ``llvm.loop.vectorize.width`` and the second
3604 operand is an integer specifying the width. For example:
3606 .. code-block:: llvm
3608 !0 = !{!"llvm.loop.vectorize.width", i32 4}
3610 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
3611 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
3612 0 or if the loop does not have this metadata the width will be
3613 determined automatically.
3615 '``llvm.loop.unroll``'
3616 ^^^^^^^^^^^^^^^^^^^^^^
3618 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
3619 optimization hints such as the unroll factor. ``llvm.loop.unroll``
3620 metadata should be used in conjunction with ``llvm.loop`` loop
3621 identification metadata. The ``llvm.loop.unroll`` metadata are only
3622 optimization hints and the unrolling will only be performed if the
3623 optimizer believes it is safe to do so.
3625 '``llvm.loop.unroll.count``' Metadata
3626 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3628 This metadata suggests an unroll factor to the loop unroller. The
3629 first operand is the string ``llvm.loop.unroll.count`` and the second
3630 operand is a positive integer specifying the unroll factor. For
3633 .. code-block:: llvm
3635 !0 = !{!"llvm.loop.unroll.count", i32 4}
3637 If the trip count of the loop is less than the unroll count the loop
3638 will be partially unrolled.
3640 '``llvm.loop.unroll.disable``' Metadata
3641 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3643 This metadata either disables loop unrolling. The metadata has a single operand
3644 which is the string ``llvm.loop.unroll.disable``. For example:
3646 .. code-block:: llvm
3648 !0 = !{!"llvm.loop.unroll.disable"}
3650 '``llvm.loop.unroll.runtime.disable``' Metadata
3651 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3653 This metadata either disables runtime loop unrolling. The metadata has a single
3654 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
3656 .. code-block:: llvm
3658 !0 = !{!"llvm.loop.unroll.runtime.disable"}
3660 '``llvm.loop.unroll.full``' Metadata
3661 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3663 This metadata either suggests that the loop should be unrolled fully. The
3664 metadata has a single operand which is the string ``llvm.loop.unroll.disable``.
3667 .. code-block:: llvm
3669 !0 = !{!"llvm.loop.unroll.full"}
3674 Metadata types used to annotate memory accesses with information helpful
3675 for optimizations are prefixed with ``llvm.mem``.
3677 '``llvm.mem.parallel_loop_access``' Metadata
3678 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3680 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
3681 or metadata containing a list of loop identifiers for nested loops.
3682 The metadata is attached to memory accessing instructions and denotes that
3683 no loop carried memory dependence exist between it and other instructions denoted
3684 with the same loop identifier.
3686 Precisely, given two instructions ``m1`` and ``m2`` that both have the
3687 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
3688 set of loops associated with that metadata, respectively, then there is no loop
3689 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
3692 As a special case, if all memory accessing instructions in a loop have
3693 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
3694 loop has no loop carried memory dependences and is considered to be a parallel
3697 Note that if not all memory access instructions have such metadata referring to
3698 the loop, then the loop is considered not being trivially parallel. Additional
3699 memory dependence analysis is required to make that determination. As a fail
3700 safe mechanism, this causes loops that were originally parallel to be considered
3701 sequential (if optimization passes that are unaware of the parallel semantics
3702 insert new memory instructions into the loop body).
3704 Example of a loop that is considered parallel due to its correct use of
3705 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
3706 metadata types that refer to the same loop identifier metadata.
3708 .. code-block:: llvm
3712 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
3714 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3716 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
3722 It is also possible to have nested parallel loops. In that case the
3723 memory accesses refer to a list of loop identifier metadata nodes instead of
3724 the loop identifier metadata node directly:
3726 .. code-block:: llvm
3730 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
3732 br label %inner.for.body
3736 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3738 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
3740 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
3744 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
3746 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
3748 outer.for.end: ; preds = %for.body
3750 !0 = !{!1, !2} ; a list of loop identifiers
3751 !1 = !{!1} ; an identifier for the inner loop
3752 !2 = !{!2} ; an identifier for the outer loop
3757 The ``llvm.bitsets`` global metadata is used to implement
3758 :doc:`bitsets <BitSets>`.
3760 Module Flags Metadata
3761 =====================
3763 Information about the module as a whole is difficult to convey to LLVM's
3764 subsystems. The LLVM IR isn't sufficient to transmit this information.
3765 The ``llvm.module.flags`` named metadata exists in order to facilitate
3766 this. These flags are in the form of key / value pairs --- much like a
3767 dictionary --- making it easy for any subsystem who cares about a flag to
3770 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
3771 Each triplet has the following form:
3773 - The first element is a *behavior* flag, which specifies the behavior
3774 when two (or more) modules are merged together, and it encounters two
3775 (or more) metadata with the same ID. The supported behaviors are
3777 - The second element is a metadata string that is a unique ID for the
3778 metadata. Each module may only have one flag entry for each unique ID (not
3779 including entries with the **Require** behavior).
3780 - The third element is the value of the flag.
3782 When two (or more) modules are merged together, the resulting
3783 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
3784 each unique metadata ID string, there will be exactly one entry in the merged
3785 modules ``llvm.module.flags`` metadata table, and the value for that entry will
3786 be determined by the merge behavior flag, as described below. The only exception
3787 is that entries with the *Require* behavior are always preserved.
3789 The following behaviors are supported:
3800 Emits an error if two values disagree, otherwise the resulting value
3801 is that of the operands.
3805 Emits a warning if two values disagree. The result value will be the
3806 operand for the flag from the first module being linked.
3810 Adds a requirement that another module flag be present and have a
3811 specified value after linking is performed. The value must be a
3812 metadata pair, where the first element of the pair is the ID of the
3813 module flag to be restricted, and the second element of the pair is
3814 the value the module flag should be restricted to. This behavior can
3815 be used to restrict the allowable results (via triggering of an
3816 error) of linking IDs with the **Override** behavior.
3820 Uses the specified value, regardless of the behavior or value of the
3821 other module. If both modules specify **Override**, but the values
3822 differ, an error will be emitted.
3826 Appends the two values, which are required to be metadata nodes.
3830 Appends the two values, which are required to be metadata
3831 nodes. However, duplicate entries in the second list are dropped
3832 during the append operation.
3834 It is an error for a particular unique flag ID to have multiple behaviors,
3835 except in the case of **Require** (which adds restrictions on another metadata
3836 value) or **Override**.
3838 An example of module flags:
3840 .. code-block:: llvm
3842 !0 = !{ i32 1, !"foo", i32 1 }
3843 !1 = !{ i32 4, !"bar", i32 37 }
3844 !2 = !{ i32 2, !"qux", i32 42 }
3845 !3 = !{ i32 3, !"qux",
3850 !llvm.module.flags = !{ !0, !1, !2, !3 }
3852 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
3853 if two or more ``!"foo"`` flags are seen is to emit an error if their
3854 values are not equal.
3856 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
3857 behavior if two or more ``!"bar"`` flags are seen is to use the value
3860 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
3861 behavior if two or more ``!"qux"`` flags are seen is to emit a
3862 warning if their values are not equal.
3864 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
3870 The behavior is to emit an error if the ``llvm.module.flags`` does not
3871 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
3874 Objective-C Garbage Collection Module Flags Metadata
3875 ----------------------------------------------------
3877 On the Mach-O platform, Objective-C stores metadata about garbage
3878 collection in a special section called "image info". The metadata
3879 consists of a version number and a bitmask specifying what types of
3880 garbage collection are supported (if any) by the file. If two or more
3881 modules are linked together their garbage collection metadata needs to
3882 be merged rather than appended together.
3884 The Objective-C garbage collection module flags metadata consists of the
3885 following key-value pairs:
3894 * - ``Objective-C Version``
3895 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3897 * - ``Objective-C Image Info Version``
3898 - **[Required]** --- The version of the image info section. Currently
3901 * - ``Objective-C Image Info Section``
3902 - **[Required]** --- The section to place the metadata. Valid values are
3903 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3904 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3905 Objective-C ABI version 2.
3907 * - ``Objective-C Garbage Collection``
3908 - **[Required]** --- Specifies whether garbage collection is supported or
3909 not. Valid values are 0, for no garbage collection, and 2, for garbage
3910 collection supported.
3912 * - ``Objective-C GC Only``
3913 - **[Optional]** --- Specifies that only garbage collection is supported.
3914 If present, its value must be 6. This flag requires that the
3915 ``Objective-C Garbage Collection`` flag have the value 2.
3917 Some important flag interactions:
3919 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3920 merged with a module with ``Objective-C Garbage Collection`` set to
3921 2, then the resulting module has the
3922 ``Objective-C Garbage Collection`` flag set to 0.
3923 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3924 merged with a module with ``Objective-C GC Only`` set to 6.
3926 Automatic Linker Flags Module Flags Metadata
3927 --------------------------------------------
3929 Some targets support embedding flags to the linker inside individual object
3930 files. Typically this is used in conjunction with language extensions which
3931 allow source files to explicitly declare the libraries they depend on, and have
3932 these automatically be transmitted to the linker via object files.
3934 These flags are encoded in the IR using metadata in the module flags section,
3935 using the ``Linker Options`` key. The merge behavior for this flag is required
3936 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3937 node which should be a list of other metadata nodes, each of which should be a
3938 list of metadata strings defining linker options.
3940 For example, the following metadata section specifies two separate sets of
3941 linker options, presumably to link against ``libz`` and the ``Cocoa``
3944 !0 = !{ i32 6, !"Linker Options",
3947 !{ !"-framework", !"Cocoa" } } }
3948 !llvm.module.flags = !{ !0 }
3950 The metadata encoding as lists of lists of options, as opposed to a collapsed
3951 list of options, is chosen so that the IR encoding can use multiple option
3952 strings to specify e.g., a single library, while still having that specifier be
3953 preserved as an atomic element that can be recognized by a target specific
3954 assembly writer or object file emitter.
3956 Each individual option is required to be either a valid option for the target's
3957 linker, or an option that is reserved by the target specific assembly writer or
3958 object file emitter. No other aspect of these options is defined by the IR.
3960 C type width Module Flags Metadata
3961 ----------------------------------
3963 The ARM backend emits a section into each generated object file describing the
3964 options that it was compiled with (in a compiler-independent way) to prevent
3965 linking incompatible objects, and to allow automatic library selection. Some
3966 of these options are not visible at the IR level, namely wchar_t width and enum
3969 To pass this information to the backend, these options are encoded in module
3970 flags metadata, using the following key-value pairs:
3980 - * 0 --- sizeof(wchar_t) == 4
3981 * 1 --- sizeof(wchar_t) == 2
3984 - * 0 --- Enums are at least as large as an ``int``.
3985 * 1 --- Enums are stored in the smallest integer type which can
3986 represent all of its values.
3988 For example, the following metadata section specifies that the module was
3989 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
3990 enum is the smallest type which can represent all of its values::
3992 !llvm.module.flags = !{!0, !1}
3993 !0 = !{i32 1, !"short_wchar", i32 1}
3994 !1 = !{i32 1, !"short_enum", i32 0}
3996 .. _intrinsicglobalvariables:
3998 Intrinsic Global Variables
3999 ==========================
4001 LLVM has a number of "magic" global variables that contain data that
4002 affect code generation or other IR semantics. These are documented here.
4003 All globals of this sort should have a section specified as
4004 "``llvm.metadata``". This section and all globals that start with
4005 "``llvm.``" are reserved for use by LLVM.
4009 The '``llvm.used``' Global Variable
4010 -----------------------------------
4012 The ``@llvm.used`` global is an array which has
4013 :ref:`appending linkage <linkage_appending>`. This array contains a list of
4014 pointers to named global variables, functions and aliases which may optionally
4015 have a pointer cast formed of bitcast or getelementptr. For example, a legal
4018 .. code-block:: llvm
4023 @llvm.used = appending global [2 x i8*] [
4025 i8* bitcast (i32* @Y to i8*)
4026 ], section "llvm.metadata"
4028 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
4029 and linker are required to treat the symbol as if there is a reference to the
4030 symbol that it cannot see (which is why they have to be named). For example, if
4031 a variable has internal linkage and no references other than that from the
4032 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
4033 references from inline asms and other things the compiler cannot "see", and
4034 corresponds to "``attribute((used))``" in GNU C.
4036 On some targets, the code generator must emit a directive to the
4037 assembler or object file to prevent the assembler and linker from
4038 molesting the symbol.
4040 .. _gv_llvmcompilerused:
4042 The '``llvm.compiler.used``' Global Variable
4043 --------------------------------------------
4045 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4046 directive, except that it only prevents the compiler from touching the
4047 symbol. On targets that support it, this allows an intelligent linker to
4048 optimize references to the symbol without being impeded as it would be
4051 This is a rare construct that should only be used in rare circumstances,
4052 and should not be exposed to source languages.
4054 .. _gv_llvmglobalctors:
4056 The '``llvm.global_ctors``' Global Variable
4057 -------------------------------------------
4059 .. code-block:: llvm
4061 %0 = type { i32, void ()*, i8* }
4062 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
4064 The ``@llvm.global_ctors`` array contains a list of constructor
4065 functions, priorities, and an optional associated global or function.
4066 The functions referenced by this array will be called in ascending order
4067 of priority (i.e. lowest first) when the module is loaded. The order of
4068 functions with the same priority is not defined.
4070 If the third field is present, non-null, and points to a global variable
4071 or function, the initializer function will only run if the associated
4072 data from the current module is not discarded.
4074 .. _llvmglobaldtors:
4076 The '``llvm.global_dtors``' Global Variable
4077 -------------------------------------------
4079 .. code-block:: llvm
4081 %0 = type { i32, void ()*, i8* }
4082 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
4084 The ``@llvm.global_dtors`` array contains a list of destructor
4085 functions, priorities, and an optional associated global or function.
4086 The functions referenced by this array will be called in descending
4087 order of priority (i.e. highest first) when the module is unloaded. The
4088 order of functions with the same priority is not defined.
4090 If the third field is present, non-null, and points to a global variable
4091 or function, the destructor function will only run if the associated
4092 data from the current module is not discarded.
4094 Instruction Reference
4095 =====================
4097 The LLVM instruction set consists of several different classifications
4098 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
4099 instructions <binaryops>`, :ref:`bitwise binary
4100 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
4101 :ref:`other instructions <otherops>`.
4105 Terminator Instructions
4106 -----------------------
4108 As mentioned :ref:`previously <functionstructure>`, every basic block in a
4109 program ends with a "Terminator" instruction, which indicates which
4110 block should be executed after the current block is finished. These
4111 terminator instructions typically yield a '``void``' value: they produce
4112 control flow, not values (the one exception being the
4113 ':ref:`invoke <i_invoke>`' instruction).
4115 The terminator instructions are: ':ref:`ret <i_ret>`',
4116 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
4117 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
4118 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
4122 '``ret``' Instruction
4123 ^^^^^^^^^^^^^^^^^^^^^
4130 ret <type> <value> ; Return a value from a non-void function
4131 ret void ; Return from void function
4136 The '``ret``' instruction is used to return control flow (and optionally
4137 a value) from a function back to the caller.
4139 There are two forms of the '``ret``' instruction: one that returns a
4140 value and then causes control flow, and one that just causes control
4146 The '``ret``' instruction optionally accepts a single argument, the
4147 return value. The type of the return value must be a ':ref:`first
4148 class <t_firstclass>`' type.
4150 A function is not :ref:`well formed <wellformed>` if it it has a non-void
4151 return type and contains a '``ret``' instruction with no return value or
4152 a return value with a type that does not match its type, or if it has a
4153 void return type and contains a '``ret``' instruction with a return
4159 When the '``ret``' instruction is executed, control flow returns back to
4160 the calling function's context. If the caller is a
4161 ":ref:`call <i_call>`" instruction, execution continues at the
4162 instruction after the call. If the caller was an
4163 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
4164 beginning of the "normal" destination block. If the instruction returns
4165 a value, that value shall set the call or invoke instruction's return
4171 .. code-block:: llvm
4173 ret i32 5 ; Return an integer value of 5
4174 ret void ; Return from a void function
4175 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
4179 '``br``' Instruction
4180 ^^^^^^^^^^^^^^^^^^^^
4187 br i1 <cond>, label <iftrue>, label <iffalse>
4188 br label <dest> ; Unconditional branch
4193 The '``br``' instruction is used to cause control flow to transfer to a
4194 different basic block in the current function. There are two forms of
4195 this instruction, corresponding to a conditional branch and an
4196 unconditional branch.
4201 The conditional branch form of the '``br``' instruction takes a single
4202 '``i1``' value and two '``label``' values. The unconditional form of the
4203 '``br``' instruction takes a single '``label``' value as a target.
4208 Upon execution of a conditional '``br``' instruction, the '``i1``'
4209 argument is evaluated. If the value is ``true``, control flows to the
4210 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
4211 to the '``iffalse``' ``label`` argument.
4216 .. code-block:: llvm
4219 %cond = icmp eq i32 %a, %b
4220 br i1 %cond, label %IfEqual, label %IfUnequal
4228 '``switch``' Instruction
4229 ^^^^^^^^^^^^^^^^^^^^^^^^
4236 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
4241 The '``switch``' instruction is used to transfer control flow to one of
4242 several different places. It is a generalization of the '``br``'
4243 instruction, allowing a branch to occur to one of many possible
4249 The '``switch``' instruction uses three parameters: an integer
4250 comparison value '``value``', a default '``label``' destination, and an
4251 array of pairs of comparison value constants and '``label``'s. The table
4252 is not allowed to contain duplicate constant entries.
4257 The ``switch`` instruction specifies a table of values and destinations.
4258 When the '``switch``' instruction is executed, this table is searched
4259 for the given value. If the value is found, control flow is transferred
4260 to the corresponding destination; otherwise, control flow is transferred
4261 to the default destination.
4266 Depending on properties of the target machine and the particular
4267 ``switch`` instruction, this instruction may be code generated in
4268 different ways. For example, it could be generated as a series of
4269 chained conditional branches or with a lookup table.
4274 .. code-block:: llvm
4276 ; Emulate a conditional br instruction
4277 %Val = zext i1 %value to i32
4278 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
4280 ; Emulate an unconditional br instruction
4281 switch i32 0, label %dest [ ]
4283 ; Implement a jump table:
4284 switch i32 %val, label %otherwise [ i32 0, label %onzero
4286 i32 2, label %ontwo ]
4290 '``indirectbr``' Instruction
4291 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4298 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
4303 The '``indirectbr``' instruction implements an indirect branch to a
4304 label within the current function, whose address is specified by
4305 "``address``". Address must be derived from a
4306 :ref:`blockaddress <blockaddress>` constant.
4311 The '``address``' argument is the address of the label to jump to. The
4312 rest of the arguments indicate the full set of possible destinations
4313 that the address may point to. Blocks are allowed to occur multiple
4314 times in the destination list, though this isn't particularly useful.
4316 This destination list is required so that dataflow analysis has an
4317 accurate understanding of the CFG.
4322 Control transfers to the block specified in the address argument. All
4323 possible destination blocks must be listed in the label list, otherwise
4324 this instruction has undefined behavior. This implies that jumps to
4325 labels defined in other functions have undefined behavior as well.
4330 This is typically implemented with a jump through a register.
4335 .. code-block:: llvm
4337 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
4341 '``invoke``' Instruction
4342 ^^^^^^^^^^^^^^^^^^^^^^^^
4349 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
4350 to label <normal label> unwind label <exception label>
4355 The '``invoke``' instruction causes control to transfer to a specified
4356 function, with the possibility of control flow transfer to either the
4357 '``normal``' label or the '``exception``' label. If the callee function
4358 returns with the "``ret``" instruction, control flow will return to the
4359 "normal" label. If the callee (or any indirect callees) returns via the
4360 ":ref:`resume <i_resume>`" instruction or other exception handling
4361 mechanism, control is interrupted and continued at the dynamically
4362 nearest "exception" label.
4364 The '``exception``' label is a `landing
4365 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
4366 '``exception``' label is required to have the
4367 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
4368 information about the behavior of the program after unwinding happens,
4369 as its first non-PHI instruction. The restrictions on the
4370 "``landingpad``" instruction's tightly couples it to the "``invoke``"
4371 instruction, so that the important information contained within the
4372 "``landingpad``" instruction can't be lost through normal code motion.
4377 This instruction requires several arguments:
4379 #. The optional "cconv" marker indicates which :ref:`calling
4380 convention <callingconv>` the call should use. If none is
4381 specified, the call defaults to using C calling conventions.
4382 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
4383 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
4385 #. '``ptr to function ty``': shall be the signature of the pointer to
4386 function value being invoked. In most cases, this is a direct
4387 function invocation, but indirect ``invoke``'s are just as possible,
4388 branching off an arbitrary pointer to function value.
4389 #. '``function ptr val``': An LLVM value containing a pointer to a
4390 function to be invoked.
4391 #. '``function args``': argument list whose types match the function
4392 signature argument types and parameter attributes. All arguments must
4393 be of :ref:`first class <t_firstclass>` type. If the function signature
4394 indicates the function accepts a variable number of arguments, the
4395 extra arguments can be specified.
4396 #. '``normal label``': the label reached when the called function
4397 executes a '``ret``' instruction.
4398 #. '``exception label``': the label reached when a callee returns via
4399 the :ref:`resume <i_resume>` instruction or other exception handling
4401 #. The optional :ref:`function attributes <fnattrs>` list. Only
4402 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
4403 attributes are valid here.
4408 This instruction is designed to operate as a standard '``call``'
4409 instruction in most regards. The primary difference is that it
4410 establishes an association with a label, which is used by the runtime
4411 library to unwind the stack.
4413 This instruction is used in languages with destructors to ensure that
4414 proper cleanup is performed in the case of either a ``longjmp`` or a
4415 thrown exception. Additionally, this is important for implementation of
4416 '``catch``' clauses in high-level languages that support them.
4418 For the purposes of the SSA form, the definition of the value returned
4419 by the '``invoke``' instruction is deemed to occur on the edge from the
4420 current block to the "normal" label. If the callee unwinds then no
4421 return value is available.
4426 .. code-block:: llvm
4428 %retval = invoke i32 @Test(i32 15) to label %Continue
4429 unwind label %TestCleanup ; i32:retval set
4430 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
4431 unwind label %TestCleanup ; i32:retval set
4435 '``resume``' Instruction
4436 ^^^^^^^^^^^^^^^^^^^^^^^^
4443 resume <type> <value>
4448 The '``resume``' instruction is a terminator instruction that has no
4454 The '``resume``' instruction requires one argument, which must have the
4455 same type as the result of any '``landingpad``' instruction in the same
4461 The '``resume``' instruction resumes propagation of an existing
4462 (in-flight) exception whose unwinding was interrupted with a
4463 :ref:`landingpad <i_landingpad>` instruction.
4468 .. code-block:: llvm
4470 resume { i8*, i32 } %exn
4474 '``unreachable``' Instruction
4475 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4487 The '``unreachable``' instruction has no defined semantics. This
4488 instruction is used to inform the optimizer that a particular portion of
4489 the code is not reachable. This can be used to indicate that the code
4490 after a no-return function cannot be reached, and other facts.
4495 The '``unreachable``' instruction has no defined semantics.
4502 Binary operators are used to do most of the computation in a program.
4503 They require two operands of the same type, execute an operation on
4504 them, and produce a single value. The operands might represent multiple
4505 data, as is the case with the :ref:`vector <t_vector>` data type. The
4506 result value has the same type as its operands.
4508 There are several different binary operators:
4512 '``add``' Instruction
4513 ^^^^^^^^^^^^^^^^^^^^^
4520 <result> = add <ty> <op1>, <op2> ; yields ty:result
4521 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
4522 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
4523 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
4528 The '``add``' instruction returns the sum of its two operands.
4533 The two arguments to the '``add``' instruction must be
4534 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4535 arguments must have identical types.
4540 The value produced is the integer sum of the two operands.
4542 If the sum has unsigned overflow, the result returned is the
4543 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4546 Because LLVM integers use a two's complement representation, this
4547 instruction is appropriate for both signed and unsigned integers.
4549 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4550 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4551 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
4552 unsigned and/or signed overflow, respectively, occurs.
4557 .. code-block:: llvm
4559 <result> = add i32 4, %var ; yields i32:result = 4 + %var
4563 '``fadd``' Instruction
4564 ^^^^^^^^^^^^^^^^^^^^^^
4571 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4576 The '``fadd``' instruction returns the sum of its two operands.
4581 The two arguments to the '``fadd``' instruction must be :ref:`floating
4582 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4583 Both arguments must have identical types.
4588 The value produced is the floating point sum of the two operands. This
4589 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
4590 which are optimization hints to enable otherwise unsafe floating point
4596 .. code-block:: llvm
4598 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
4600 '``sub``' Instruction
4601 ^^^^^^^^^^^^^^^^^^^^^
4608 <result> = sub <ty> <op1>, <op2> ; yields ty:result
4609 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
4610 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
4611 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
4616 The '``sub``' instruction returns the difference of its two operands.
4618 Note that the '``sub``' instruction is used to represent the '``neg``'
4619 instruction present in most other intermediate representations.
4624 The two arguments to the '``sub``' instruction must be
4625 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4626 arguments must have identical types.
4631 The value produced is the integer difference of the two operands.
4633 If the difference has unsigned overflow, the result returned is the
4634 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4637 Because LLVM integers use a two's complement representation, this
4638 instruction is appropriate for both signed and unsigned integers.
4640 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4641 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4642 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
4643 unsigned and/or signed overflow, respectively, occurs.
4648 .. code-block:: llvm
4650 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
4651 <result> = sub i32 0, %val ; yields i32:result = -%var
4655 '``fsub``' Instruction
4656 ^^^^^^^^^^^^^^^^^^^^^^
4663 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4668 The '``fsub``' instruction returns the difference of its two operands.
4670 Note that the '``fsub``' instruction is used to represent the '``fneg``'
4671 instruction present in most other intermediate representations.
4676 The two arguments to the '``fsub``' instruction must be :ref:`floating
4677 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4678 Both arguments must have identical types.
4683 The value produced is the floating point difference of the two operands.
4684 This instruction can also take any number of :ref:`fast-math
4685 flags <fastmath>`, which are optimization hints to enable otherwise
4686 unsafe floating point optimizations:
4691 .. code-block:: llvm
4693 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
4694 <result> = fsub float -0.0, %val ; yields float:result = -%var
4696 '``mul``' Instruction
4697 ^^^^^^^^^^^^^^^^^^^^^
4704 <result> = mul <ty> <op1>, <op2> ; yields ty:result
4705 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
4706 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
4707 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
4712 The '``mul``' instruction returns the product of its two operands.
4717 The two arguments to the '``mul``' instruction must be
4718 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4719 arguments must have identical types.
4724 The value produced is the integer product of the two operands.
4726 If the result of the multiplication has unsigned overflow, the result
4727 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
4728 bit width of the result.
4730 Because LLVM integers use a two's complement representation, and the
4731 result is the same width as the operands, this instruction returns the
4732 correct result for both signed and unsigned integers. If a full product
4733 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
4734 sign-extended or zero-extended as appropriate to the width of the full
4737 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4738 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4739 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
4740 unsigned and/or signed overflow, respectively, occurs.
4745 .. code-block:: llvm
4747 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
4751 '``fmul``' Instruction
4752 ^^^^^^^^^^^^^^^^^^^^^^
4759 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4764 The '``fmul``' instruction returns the product of its two operands.
4769 The two arguments to the '``fmul``' instruction must be :ref:`floating
4770 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4771 Both arguments must have identical types.
4776 The value produced is the floating point product of the two operands.
4777 This instruction can also take any number of :ref:`fast-math
4778 flags <fastmath>`, which are optimization hints to enable otherwise
4779 unsafe floating point optimizations:
4784 .. code-block:: llvm
4786 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
4788 '``udiv``' Instruction
4789 ^^^^^^^^^^^^^^^^^^^^^^
4796 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
4797 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
4802 The '``udiv``' instruction returns the quotient of its two operands.
4807 The two arguments to the '``udiv``' instruction must be
4808 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4809 arguments must have identical types.
4814 The value produced is the unsigned integer quotient of the two operands.
4816 Note that unsigned integer division and signed integer division are
4817 distinct operations; for signed integer division, use '``sdiv``'.
4819 Division by zero leads to undefined behavior.
4821 If the ``exact`` keyword is present, the result value of the ``udiv`` is
4822 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
4823 such, "((a udiv exact b) mul b) == a").
4828 .. code-block:: llvm
4830 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
4832 '``sdiv``' Instruction
4833 ^^^^^^^^^^^^^^^^^^^^^^
4840 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
4841 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
4846 The '``sdiv``' instruction returns the quotient of its two operands.
4851 The two arguments to the '``sdiv``' instruction must be
4852 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4853 arguments must have identical types.
4858 The value produced is the signed integer quotient of the two operands
4859 rounded towards zero.
4861 Note that signed integer division and unsigned integer division are
4862 distinct operations; for unsigned integer division, use '``udiv``'.
4864 Division by zero leads to undefined behavior. Overflow also leads to
4865 undefined behavior; this is a rare case, but can occur, for example, by
4866 doing a 32-bit division of -2147483648 by -1.
4868 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
4869 a :ref:`poison value <poisonvalues>` if the result would be rounded.
4874 .. code-block:: llvm
4876 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
4880 '``fdiv``' Instruction
4881 ^^^^^^^^^^^^^^^^^^^^^^
4888 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4893 The '``fdiv``' instruction returns the quotient of its two operands.
4898 The two arguments to the '``fdiv``' instruction must be :ref:`floating
4899 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4900 Both arguments must have identical types.
4905 The value produced is the floating point quotient of the two operands.
4906 This instruction can also take any number of :ref:`fast-math
4907 flags <fastmath>`, which are optimization hints to enable otherwise
4908 unsafe floating point optimizations:
4913 .. code-block:: llvm
4915 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
4917 '``urem``' Instruction
4918 ^^^^^^^^^^^^^^^^^^^^^^
4925 <result> = urem <ty> <op1>, <op2> ; yields ty:result
4930 The '``urem``' instruction returns the remainder from the unsigned
4931 division of its two arguments.
4936 The two arguments to the '``urem``' instruction must be
4937 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4938 arguments must have identical types.
4943 This instruction returns the unsigned integer *remainder* of a division.
4944 This instruction always performs an unsigned division to get the
4947 Note that unsigned integer remainder and signed integer remainder are
4948 distinct operations; for signed integer remainder, use '``srem``'.
4950 Taking the remainder of a division by zero leads to undefined behavior.
4955 .. code-block:: llvm
4957 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
4959 '``srem``' Instruction
4960 ^^^^^^^^^^^^^^^^^^^^^^
4967 <result> = srem <ty> <op1>, <op2> ; yields ty:result
4972 The '``srem``' instruction returns the remainder from the signed
4973 division of its two operands. This instruction can also take
4974 :ref:`vector <t_vector>` versions of the values in which case the elements
4980 The two arguments to the '``srem``' instruction must be
4981 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4982 arguments must have identical types.
4987 This instruction returns the *remainder* of a division (where the result
4988 is either zero or has the same sign as the dividend, ``op1``), not the
4989 *modulo* operator (where the result is either zero or has the same sign
4990 as the divisor, ``op2``) of a value. For more information about the
4991 difference, see `The Math
4992 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4993 table of how this is implemented in various languages, please see
4995 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4997 Note that signed integer remainder and unsigned integer remainder are
4998 distinct operations; for unsigned integer remainder, use '``urem``'.
5000 Taking the remainder of a division by zero leads to undefined behavior.
5001 Overflow also leads to undefined behavior; this is a rare case, but can
5002 occur, for example, by taking the remainder of a 32-bit division of
5003 -2147483648 by -1. (The remainder doesn't actually overflow, but this
5004 rule lets srem be implemented using instructions that return both the
5005 result of the division and the remainder.)
5010 .. code-block:: llvm
5012 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
5016 '``frem``' Instruction
5017 ^^^^^^^^^^^^^^^^^^^^^^
5024 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5029 The '``frem``' instruction returns the remainder from the division of
5035 The two arguments to the '``frem``' instruction must be :ref:`floating
5036 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5037 Both arguments must have identical types.
5042 This instruction returns the *remainder* of a division. The remainder
5043 has the same sign as the dividend. This instruction can also take any
5044 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
5045 to enable otherwise unsafe floating point optimizations:
5050 .. code-block:: llvm
5052 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
5056 Bitwise Binary Operations
5057 -------------------------
5059 Bitwise binary operators are used to do various forms of bit-twiddling
5060 in a program. They are generally very efficient instructions and can
5061 commonly be strength reduced from other instructions. They require two
5062 operands of the same type, execute an operation on them, and produce a
5063 single value. The resulting value is the same type as its operands.
5065 '``shl``' Instruction
5066 ^^^^^^^^^^^^^^^^^^^^^
5073 <result> = shl <ty> <op1>, <op2> ; yields ty:result
5074 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
5075 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
5076 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
5081 The '``shl``' instruction returns the first operand shifted to the left
5082 a specified number of bits.
5087 Both arguments to the '``shl``' instruction must be the same
5088 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5089 '``op2``' is treated as an unsigned value.
5094 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
5095 where ``n`` is the width of the result. If ``op2`` is (statically or
5096 dynamically) equal to or larger than the number of bits in
5097 ``op1``, the result is undefined. If the arguments are vectors, each
5098 vector element of ``op1`` is shifted by the corresponding shift amount
5101 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
5102 value <poisonvalues>` if it shifts out any non-zero bits. If the
5103 ``nsw`` keyword is present, then the shift produces a :ref:`poison
5104 value <poisonvalues>` if it shifts out any bits that disagree with the
5105 resultant sign bit. As such, NUW/NSW have the same semantics as they
5106 would if the shift were expressed as a mul instruction with the same
5107 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
5112 .. code-block:: llvm
5114 <result> = shl i32 4, %var ; yields i32: 4 << %var
5115 <result> = shl i32 4, 2 ; yields i32: 16
5116 <result> = shl i32 1, 10 ; yields i32: 1024
5117 <result> = shl i32 1, 32 ; undefined
5118 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
5120 '``lshr``' Instruction
5121 ^^^^^^^^^^^^^^^^^^^^^^
5128 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
5129 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
5134 The '``lshr``' instruction (logical shift right) returns the first
5135 operand shifted to the right a specified number of bits with zero fill.
5140 Both arguments to the '``lshr``' instruction must be the same
5141 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5142 '``op2``' is treated as an unsigned value.
5147 This instruction always performs a logical shift right operation. The
5148 most significant bits of the result will be filled with zero bits after
5149 the shift. If ``op2`` is (statically or dynamically) equal to or larger
5150 than the number of bits in ``op1``, the result is undefined. If the
5151 arguments are vectors, each vector element of ``op1`` is shifted by the
5152 corresponding shift amount in ``op2``.
5154 If the ``exact`` keyword is present, the result value of the ``lshr`` is
5155 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
5161 .. code-block:: llvm
5163 <result> = lshr i32 4, 1 ; yields i32:result = 2
5164 <result> = lshr i32 4, 2 ; yields i32:result = 1
5165 <result> = lshr i8 4, 3 ; yields i8:result = 0
5166 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
5167 <result> = lshr i32 1, 32 ; undefined
5168 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
5170 '``ashr``' Instruction
5171 ^^^^^^^^^^^^^^^^^^^^^^
5178 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
5179 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
5184 The '``ashr``' instruction (arithmetic shift right) returns the first
5185 operand shifted to the right a specified number of bits with sign
5191 Both arguments to the '``ashr``' instruction must be the same
5192 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5193 '``op2``' is treated as an unsigned value.
5198 This instruction always performs an arithmetic shift right operation,
5199 The most significant bits of the result will be filled with the sign bit
5200 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
5201 than the number of bits in ``op1``, the result is undefined. If the
5202 arguments are vectors, each vector element of ``op1`` is shifted by the
5203 corresponding shift amount in ``op2``.
5205 If the ``exact`` keyword is present, the result value of the ``ashr`` is
5206 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
5212 .. code-block:: llvm
5214 <result> = ashr i32 4, 1 ; yields i32:result = 2
5215 <result> = ashr i32 4, 2 ; yields i32:result = 1
5216 <result> = ashr i8 4, 3 ; yields i8:result = 0
5217 <result> = ashr i8 -2, 1 ; yields i8:result = -1
5218 <result> = ashr i32 1, 32 ; undefined
5219 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
5221 '``and``' Instruction
5222 ^^^^^^^^^^^^^^^^^^^^^
5229 <result> = and <ty> <op1>, <op2> ; yields ty:result
5234 The '``and``' instruction returns the bitwise logical and of its two
5240 The two arguments to the '``and``' instruction must be
5241 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5242 arguments must have identical types.
5247 The truth table used for the '``and``' instruction is:
5264 .. code-block:: llvm
5266 <result> = and i32 4, %var ; yields i32:result = 4 & %var
5267 <result> = and i32 15, 40 ; yields i32:result = 8
5268 <result> = and i32 4, 8 ; yields i32:result = 0
5270 '``or``' Instruction
5271 ^^^^^^^^^^^^^^^^^^^^
5278 <result> = or <ty> <op1>, <op2> ; yields ty:result
5283 The '``or``' instruction returns the bitwise logical inclusive or of its
5289 The two arguments to the '``or``' instruction must be
5290 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5291 arguments must have identical types.
5296 The truth table used for the '``or``' instruction is:
5315 <result> = or i32 4, %var ; yields i32:result = 4 | %var
5316 <result> = or i32 15, 40 ; yields i32:result = 47
5317 <result> = or i32 4, 8 ; yields i32:result = 12
5319 '``xor``' Instruction
5320 ^^^^^^^^^^^^^^^^^^^^^
5327 <result> = xor <ty> <op1>, <op2> ; yields ty:result
5332 The '``xor``' instruction returns the bitwise logical exclusive or of
5333 its two operands. The ``xor`` is used to implement the "one's
5334 complement" operation, which is the "~" operator in C.
5339 The two arguments to the '``xor``' instruction must be
5340 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5341 arguments must have identical types.
5346 The truth table used for the '``xor``' instruction is:
5363 .. code-block:: llvm
5365 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
5366 <result> = xor i32 15, 40 ; yields i32:result = 39
5367 <result> = xor i32 4, 8 ; yields i32:result = 12
5368 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
5373 LLVM supports several instructions to represent vector operations in a
5374 target-independent manner. These instructions cover the element-access
5375 and vector-specific operations needed to process vectors effectively.
5376 While LLVM does directly support these vector operations, many
5377 sophisticated algorithms will want to use target-specific intrinsics to
5378 take full advantage of a specific target.
5380 .. _i_extractelement:
5382 '``extractelement``' Instruction
5383 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5390 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
5395 The '``extractelement``' instruction extracts a single scalar element
5396 from a vector at a specified index.
5401 The first operand of an '``extractelement``' instruction is a value of
5402 :ref:`vector <t_vector>` type. The second operand is an index indicating
5403 the position from which to extract the element. The index may be a
5404 variable of any integer type.
5409 The result is a scalar of the same type as the element type of ``val``.
5410 Its value is the value at position ``idx`` of ``val``. If ``idx``
5411 exceeds the length of ``val``, the results are undefined.
5416 .. code-block:: llvm
5418 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
5420 .. _i_insertelement:
5422 '``insertelement``' Instruction
5423 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5430 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
5435 The '``insertelement``' instruction inserts a scalar element into a
5436 vector at a specified index.
5441 The first operand of an '``insertelement``' instruction is a value of
5442 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
5443 type must equal the element type of the first operand. The third operand
5444 is an index indicating the position at which to insert the value. The
5445 index may be a variable of any integer type.
5450 The result is a vector of the same type as ``val``. Its element values
5451 are those of ``val`` except at position ``idx``, where it gets the value
5452 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
5458 .. code-block:: llvm
5460 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
5462 .. _i_shufflevector:
5464 '``shufflevector``' Instruction
5465 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5472 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
5477 The '``shufflevector``' instruction constructs a permutation of elements
5478 from two input vectors, returning a vector with the same element type as
5479 the input and length that is the same as the shuffle mask.
5484 The first two operands of a '``shufflevector``' instruction are vectors
5485 with the same type. The third argument is a shuffle mask whose element
5486 type is always 'i32'. The result of the instruction is a vector whose
5487 length is the same as the shuffle mask and whose element type is the
5488 same as the element type of the first two operands.
5490 The shuffle mask operand is required to be a constant vector with either
5491 constant integer or undef values.
5496 The elements of the two input vectors are numbered from left to right
5497 across both of the vectors. The shuffle mask operand specifies, for each
5498 element of the result vector, which element of the two input vectors the
5499 result element gets. The element selector may be undef (meaning "don't
5500 care") and the second operand may be undef if performing a shuffle from
5506 .. code-block:: llvm
5508 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
5509 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
5510 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
5511 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
5512 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
5513 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
5514 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
5515 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
5517 Aggregate Operations
5518 --------------------
5520 LLVM supports several instructions for working with
5521 :ref:`aggregate <t_aggregate>` values.
5525 '``extractvalue``' Instruction
5526 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5533 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
5538 The '``extractvalue``' instruction extracts the value of a member field
5539 from an :ref:`aggregate <t_aggregate>` value.
5544 The first operand of an '``extractvalue``' instruction is a value of
5545 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
5546 constant indices to specify which value to extract in a similar manner
5547 as indices in a '``getelementptr``' instruction.
5549 The major differences to ``getelementptr`` indexing are:
5551 - Since the value being indexed is not a pointer, the first index is
5552 omitted and assumed to be zero.
5553 - At least one index must be specified.
5554 - Not only struct indices but also array indices must be in bounds.
5559 The result is the value at the position in the aggregate specified by
5565 .. code-block:: llvm
5567 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
5571 '``insertvalue``' Instruction
5572 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5579 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
5584 The '``insertvalue``' instruction inserts a value into a member field in
5585 an :ref:`aggregate <t_aggregate>` value.
5590 The first operand of an '``insertvalue``' instruction is a value of
5591 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
5592 a first-class value to insert. The following operands are constant
5593 indices indicating the position at which to insert the value in a
5594 similar manner as indices in a '``extractvalue``' instruction. The value
5595 to insert must have the same type as the value identified by the
5601 The result is an aggregate of the same type as ``val``. Its value is
5602 that of ``val`` except that the value at the position specified by the
5603 indices is that of ``elt``.
5608 .. code-block:: llvm
5610 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
5611 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
5612 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
5616 Memory Access and Addressing Operations
5617 ---------------------------------------
5619 A key design point of an SSA-based representation is how it represents
5620 memory. In LLVM, no memory locations are in SSA form, which makes things
5621 very simple. This section describes how to read, write, and allocate
5626 '``alloca``' Instruction
5627 ^^^^^^^^^^^^^^^^^^^^^^^^
5634 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
5639 The '``alloca``' instruction allocates memory on the stack frame of the
5640 currently executing function, to be automatically released when this
5641 function returns to its caller. The object is always allocated in the
5642 generic address space (address space zero).
5647 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
5648 bytes of memory on the runtime stack, returning a pointer of the
5649 appropriate type to the program. If "NumElements" is specified, it is
5650 the number of elements allocated, otherwise "NumElements" is defaulted
5651 to be one. If a constant alignment is specified, the value result of the
5652 allocation is guaranteed to be aligned to at least that boundary. The
5653 alignment may not be greater than ``1 << 29``. If not specified, or if
5654 zero, the target can choose to align the allocation on any convenient
5655 boundary compatible with the type.
5657 '``type``' may be any sized type.
5662 Memory is allocated; a pointer is returned. The operation is undefined
5663 if there is insufficient stack space for the allocation. '``alloca``'d
5664 memory is automatically released when the function returns. The
5665 '``alloca``' instruction is commonly used to represent automatic
5666 variables that must have an address available. When the function returns
5667 (either with the ``ret`` or ``resume`` instructions), the memory is
5668 reclaimed. Allocating zero bytes is legal, but the result is undefined.
5669 The order in which memory is allocated (ie., which way the stack grows)
5675 .. code-block:: llvm
5677 %ptr = alloca i32 ; yields i32*:ptr
5678 %ptr = alloca i32, i32 4 ; yields i32*:ptr
5679 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
5680 %ptr = alloca i32, align 1024 ; yields i32*:ptr
5684 '``load``' Instruction
5685 ^^^^^^^^^^^^^^^^^^^^^^
5692 <result> = load [volatile] <ty>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>][, !nonnull !<index>][, !dereferenceable !<index>][, !dereferenceable_or_null !<index>]
5693 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
5694 !<index> = !{ i32 1 }
5699 The '``load``' instruction is used to read from memory.
5704 The argument to the ``load`` instruction specifies the memory address
5705 from which to load. The type specified must be a :ref:`first
5706 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
5707 then the optimizer is not allowed to modify the number or order of
5708 execution of this ``load`` with other :ref:`volatile
5709 operations <volatile>`.
5711 If the ``load`` is marked as ``atomic``, it takes an extra
5712 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5713 ``release`` and ``acq_rel`` orderings are not valid on ``load``
5714 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5715 when they may see multiple atomic stores. The type of the pointee must
5716 be an integer type whose bit width is a power of two greater than or
5717 equal to eight and less than or equal to a target-specific size limit.
5718 ``align`` must be explicitly specified on atomic loads, and the load has
5719 undefined behavior if the alignment is not set to a value which is at
5720 least the size in bytes of the pointee. ``!nontemporal`` does not have
5721 any defined semantics for atomic loads.
5723 The optional constant ``align`` argument specifies the alignment of the
5724 operation (that is, the alignment of the memory address). A value of 0
5725 or an omitted ``align`` argument means that the operation has the ABI
5726 alignment for the target. It is the responsibility of the code emitter
5727 to ensure that the alignment information is correct. Overestimating the
5728 alignment results in undefined behavior. Underestimating the alignment
5729 may produce less efficient code. An alignment of 1 is always safe. The
5730 maximum possible alignment is ``1 << 29``.
5732 The optional ``!nontemporal`` metadata must reference a single
5733 metadata name ``<index>`` corresponding to a metadata node with one
5734 ``i32`` entry of value 1. The existence of the ``!nontemporal``
5735 metadata on the instruction tells the optimizer and code generator
5736 that this load is not expected to be reused in the cache. The code
5737 generator may select special instructions to save cache bandwidth, such
5738 as the ``MOVNT`` instruction on x86.
5740 The optional ``!invariant.load`` metadata must reference a single
5741 metadata name ``<index>`` corresponding to a metadata node with no
5742 entries. The existence of the ``!invariant.load`` metadata on the
5743 instruction tells the optimizer and code generator that the address
5744 operand to this load points to memory which can be assumed unchanged.
5745 Being invariant does not imply that a location is dereferenceable,
5746 but it does imply that once the location is known dereferenceable
5747 its value is henceforth unchanging.
5749 The optional ``!nonnull`` metadata must reference a single
5750 metadata name ``<index>`` corresponding to a metadata node with no
5751 entries. The existence of the ``!nonnull`` metadata on the
5752 instruction tells the optimizer that the value loaded is known to
5753 never be null. This is analogous to the ''nonnull'' attribute
5754 on parameters and return values. This metadata can only be applied
5755 to loads of a pointer type.
5757 The optional ``!dereferenceable`` metadata must reference a single
5758 metadata name ``<index>`` corresponding to a metadata node with one ``i64``
5759 entry. The existence of the ``!dereferenceable`` metadata on the instruction
5760 tells the optimizer that the value loaded is known to be dereferenceable.
5761 The number of bytes known to be dereferenceable is specified by the integer
5762 value in the metadata node. This is analogous to the ''dereferenceable''
5763 attribute on parameters and return values. This metadata can only be applied
5764 to loads of a pointer type.
5766 The optional ``!dereferenceable_or_null`` metadata must reference a single
5767 metadata name ``<index>`` corresponding to a metadata node with one ``i64``
5768 entry. The existence of the ``!dereferenceable_or_null`` metadata on the
5769 instruction tells the optimizer that the value loaded is known to be either
5770 dereferenceable or null.
5771 The number of bytes known to be dereferenceable is specified by the integer
5772 value in the metadata node. This is analogous to the ''dereferenceable_or_null''
5773 attribute on parameters and return values. This metadata can only be applied
5774 to loads of a pointer type.
5779 The location of memory pointed to is loaded. If the value being loaded
5780 is of scalar type then the number of bytes read does not exceed the
5781 minimum number of bytes needed to hold all bits of the type. For
5782 example, loading an ``i24`` reads at most three bytes. When loading a
5783 value of a type like ``i20`` with a size that is not an integral number
5784 of bytes, the result is undefined if the value was not originally
5785 written using a store of the same type.
5790 .. code-block:: llvm
5792 %ptr = alloca i32 ; yields i32*:ptr
5793 store i32 3, i32* %ptr ; yields void
5794 %val = load i32, i32* %ptr ; yields i32:val = i32 3
5798 '``store``' Instruction
5799 ^^^^^^^^^^^^^^^^^^^^^^^
5806 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields void
5807 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields void
5812 The '``store``' instruction is used to write to memory.
5817 There are two arguments to the ``store`` instruction: a value to store
5818 and an address at which to store it. The type of the ``<pointer>``
5819 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
5820 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
5821 then the optimizer is not allowed to modify the number or order of
5822 execution of this ``store`` with other :ref:`volatile
5823 operations <volatile>`.
5825 If the ``store`` is marked as ``atomic``, it takes an extra
5826 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5827 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
5828 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5829 when they may see multiple atomic stores. The type of the pointee must
5830 be an integer type whose bit width is a power of two greater than or
5831 equal to eight and less than or equal to a target-specific size limit.
5832 ``align`` must be explicitly specified on atomic stores, and the store
5833 has undefined behavior if the alignment is not set to a value which is
5834 at least the size in bytes of the pointee. ``!nontemporal`` does not
5835 have any defined semantics for atomic stores.
5837 The optional constant ``align`` argument specifies the alignment of the
5838 operation (that is, the alignment of the memory address). A value of 0
5839 or an omitted ``align`` argument means that the operation has the ABI
5840 alignment for the target. It is the responsibility of the code emitter
5841 to ensure that the alignment information is correct. Overestimating the
5842 alignment results in undefined behavior. Underestimating the
5843 alignment may produce less efficient code. An alignment of 1 is always
5844 safe. The maximum possible alignment is ``1 << 29``.
5846 The optional ``!nontemporal`` metadata must reference a single metadata
5847 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
5848 value 1. The existence of the ``!nontemporal`` metadata on the instruction
5849 tells the optimizer and code generator that this load is not expected to
5850 be reused in the cache. The code generator may select special
5851 instructions to save cache bandwidth, such as the MOVNT instruction on
5857 The contents of memory are updated to contain ``<value>`` at the
5858 location specified by the ``<pointer>`` operand. If ``<value>`` is
5859 of scalar type then the number of bytes written does not exceed the
5860 minimum number of bytes needed to hold all bits of the type. For
5861 example, storing an ``i24`` writes at most three bytes. When writing a
5862 value of a type like ``i20`` with a size that is not an integral number
5863 of bytes, it is unspecified what happens to the extra bits that do not
5864 belong to the type, but they will typically be overwritten.
5869 .. code-block:: llvm
5871 %ptr = alloca i32 ; yields i32*:ptr
5872 store i32 3, i32* %ptr ; yields void
5873 %val = load i32* %ptr ; yields i32:val = i32 3
5877 '``fence``' Instruction
5878 ^^^^^^^^^^^^^^^^^^^^^^^
5885 fence [singlethread] <ordering> ; yields void
5890 The '``fence``' instruction is used to introduce happens-before edges
5896 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
5897 defines what *synchronizes-with* edges they add. They can only be given
5898 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
5903 A fence A which has (at least) ``release`` ordering semantics
5904 *synchronizes with* a fence B with (at least) ``acquire`` ordering
5905 semantics if and only if there exist atomic operations X and Y, both
5906 operating on some atomic object M, such that A is sequenced before X, X
5907 modifies M (either directly or through some side effect of a sequence
5908 headed by X), Y is sequenced before B, and Y observes M. This provides a
5909 *happens-before* dependency between A and B. Rather than an explicit
5910 ``fence``, one (but not both) of the atomic operations X or Y might
5911 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
5912 still *synchronize-with* the explicit ``fence`` and establish the
5913 *happens-before* edge.
5915 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
5916 ``acquire`` and ``release`` semantics specified above, participates in
5917 the global program order of other ``seq_cst`` operations and/or fences.
5919 The optional ":ref:`singlethread <singlethread>`" argument specifies
5920 that the fence only synchronizes with other fences in the same thread.
5921 (This is useful for interacting with signal handlers.)
5926 .. code-block:: llvm
5928 fence acquire ; yields void
5929 fence singlethread seq_cst ; yields void
5933 '``cmpxchg``' Instruction
5934 ^^^^^^^^^^^^^^^^^^^^^^^^^
5941 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
5946 The '``cmpxchg``' instruction is used to atomically modify memory. It
5947 loads a value in memory and compares it to a given value. If they are
5948 equal, it tries to store a new value into the memory.
5953 There are three arguments to the '``cmpxchg``' instruction: an address
5954 to operate on, a value to compare to the value currently be at that
5955 address, and a new value to place at that address if the compared values
5956 are equal. The type of '<cmp>' must be an integer type whose bit width
5957 is a power of two greater than or equal to eight and less than or equal
5958 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5959 type, and the type of '<pointer>' must be a pointer to that type. If the
5960 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5961 to modify the number or order of execution of this ``cmpxchg`` with
5962 other :ref:`volatile operations <volatile>`.
5964 The success and failure :ref:`ordering <ordering>` arguments specify how this
5965 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
5966 must be at least ``monotonic``, the ordering constraint on failure must be no
5967 stronger than that on success, and the failure ordering cannot be either
5968 ``release`` or ``acq_rel``.
5970 The optional "``singlethread``" argument declares that the ``cmpxchg``
5971 is only atomic with respect to code (usually signal handlers) running in
5972 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5973 respect to all other code in the system.
5975 The pointer passed into cmpxchg must have alignment greater than or
5976 equal to the size in memory of the operand.
5981 The contents of memory at the location specified by the '``<pointer>``' operand
5982 is read and compared to '``<cmp>``'; if the read value is the equal, the
5983 '``<new>``' is written. The original value at the location is returned, together
5984 with a flag indicating success (true) or failure (false).
5986 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
5987 permitted: the operation may not write ``<new>`` even if the comparison
5990 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
5991 if the value loaded equals ``cmp``.
5993 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5994 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5995 load with an ordering parameter determined the second ordering parameter.
6000 .. code-block:: llvm
6003 %orig = atomic load i32, i32* %ptr unordered ; yields i32
6007 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
6008 %squared = mul i32 %cmp, %cmp
6009 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
6010 %value_loaded = extractvalue { i32, i1 } %val_success, 0
6011 %success = extractvalue { i32, i1 } %val_success, 1
6012 br i1 %success, label %done, label %loop
6019 '``atomicrmw``' Instruction
6020 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6027 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
6032 The '``atomicrmw``' instruction is used to atomically modify memory.
6037 There are three arguments to the '``atomicrmw``' instruction: an
6038 operation to apply, an address whose value to modify, an argument to the
6039 operation. The operation must be one of the following keywords:
6053 The type of '<value>' must be an integer type whose bit width is a power
6054 of two greater than or equal to eight and less than or equal to a
6055 target-specific size limit. The type of the '``<pointer>``' operand must
6056 be a pointer to that type. If the ``atomicrmw`` is marked as
6057 ``volatile``, then the optimizer is not allowed to modify the number or
6058 order of execution of this ``atomicrmw`` with other :ref:`volatile
6059 operations <volatile>`.
6064 The contents of memory at the location specified by the '``<pointer>``'
6065 operand are atomically read, modified, and written back. The original
6066 value at the location is returned. The modification is specified by the
6069 - xchg: ``*ptr = val``
6070 - add: ``*ptr = *ptr + val``
6071 - sub: ``*ptr = *ptr - val``
6072 - and: ``*ptr = *ptr & val``
6073 - nand: ``*ptr = ~(*ptr & val)``
6074 - or: ``*ptr = *ptr | val``
6075 - xor: ``*ptr = *ptr ^ val``
6076 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
6077 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
6078 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
6080 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
6086 .. code-block:: llvm
6088 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
6090 .. _i_getelementptr:
6092 '``getelementptr``' Instruction
6093 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6100 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
6101 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
6102 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
6107 The '``getelementptr``' instruction is used to get the address of a
6108 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
6109 address calculation only and does not access memory.
6114 The first argument is always a type used as the basis for the calculations.
6115 The second argument is always a pointer or a vector of pointers, and is the
6116 base address to start from. The remaining arguments are indices
6117 that indicate which of the elements of the aggregate object are indexed.
6118 The interpretation of each index is dependent on the type being indexed
6119 into. The first index always indexes the pointer value given as the
6120 first argument, the second index indexes a value of the type pointed to
6121 (not necessarily the value directly pointed to, since the first index
6122 can be non-zero), etc. The first type indexed into must be a pointer
6123 value, subsequent types can be arrays, vectors, and structs. Note that
6124 subsequent types being indexed into can never be pointers, since that
6125 would require loading the pointer before continuing calculation.
6127 The type of each index argument depends on the type it is indexing into.
6128 When indexing into a (optionally packed) structure, only ``i32`` integer
6129 **constants** are allowed (when using a vector of indices they must all
6130 be the **same** ``i32`` integer constant). When indexing into an array,
6131 pointer or vector, integers of any width are allowed, and they are not
6132 required to be constant. These integers are treated as signed values
6135 For example, let's consider a C code fragment and how it gets compiled
6151 int *foo(struct ST *s) {
6152 return &s[1].Z.B[5][13];
6155 The LLVM code generated by Clang is:
6157 .. code-block:: llvm
6159 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
6160 %struct.ST = type { i32, double, %struct.RT }
6162 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
6164 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
6171 In the example above, the first index is indexing into the
6172 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
6173 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
6174 indexes into the third element of the structure, yielding a
6175 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
6176 structure. The third index indexes into the second element of the
6177 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
6178 dimensions of the array are subscripted into, yielding an '``i32``'
6179 type. The '``getelementptr``' instruction returns a pointer to this
6180 element, thus computing a value of '``i32*``' type.
6182 Note that it is perfectly legal to index partially through a structure,
6183 returning a pointer to an inner element. Because of this, the LLVM code
6184 for the given testcase is equivalent to:
6186 .. code-block:: llvm
6188 define i32* @foo(%struct.ST* %s) {
6189 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
6190 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
6191 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
6192 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
6193 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
6197 If the ``inbounds`` keyword is present, the result value of the
6198 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
6199 pointer is not an *in bounds* address of an allocated object, or if any
6200 of the addresses that would be formed by successive addition of the
6201 offsets implied by the indices to the base address with infinitely
6202 precise signed arithmetic are not an *in bounds* address of that
6203 allocated object. The *in bounds* addresses for an allocated object are
6204 all the addresses that point into the object, plus the address one byte
6205 past the end. In cases where the base is a vector of pointers the
6206 ``inbounds`` keyword applies to each of the computations element-wise.
6208 If the ``inbounds`` keyword is not present, the offsets are added to the
6209 base address with silently-wrapping two's complement arithmetic. If the
6210 offsets have a different width from the pointer, they are sign-extended
6211 or truncated to the width of the pointer. The result value of the
6212 ``getelementptr`` may be outside the object pointed to by the base
6213 pointer. The result value may not necessarily be used to access memory
6214 though, even if it happens to point into allocated storage. See the
6215 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
6218 The getelementptr instruction is often confusing. For some more insight
6219 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
6224 .. code-block:: llvm
6226 ; yields [12 x i8]*:aptr
6227 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
6229 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
6231 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
6233 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
6235 In cases where the pointer argument is a vector of pointers, each index
6236 must be a vector with the same number of elements. For example:
6238 .. code-block:: llvm
6240 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets,
6242 Conversion Operations
6243 ---------------------
6245 The instructions in this category are the conversion instructions
6246 (casting) which all take a single operand and a type. They perform
6247 various bit conversions on the operand.
6249 '``trunc .. to``' Instruction
6250 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6257 <result> = trunc <ty> <value> to <ty2> ; yields ty2
6262 The '``trunc``' instruction truncates its operand to the type ``ty2``.
6267 The '``trunc``' instruction takes a value to trunc, and a type to trunc
6268 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
6269 of the same number of integers. The bit size of the ``value`` must be
6270 larger than the bit size of the destination type, ``ty2``. Equal sized
6271 types are not allowed.
6276 The '``trunc``' instruction truncates the high order bits in ``value``
6277 and converts the remaining bits to ``ty2``. Since the source size must
6278 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
6279 It will always truncate bits.
6284 .. code-block:: llvm
6286 %X = trunc i32 257 to i8 ; yields i8:1
6287 %Y = trunc i32 123 to i1 ; yields i1:true
6288 %Z = trunc i32 122 to i1 ; yields i1:false
6289 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
6291 '``zext .. to``' Instruction
6292 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6299 <result> = zext <ty> <value> to <ty2> ; yields ty2
6304 The '``zext``' instruction zero extends its operand to type ``ty2``.
6309 The '``zext``' instruction takes a value to cast, and a type to cast it
6310 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
6311 the same number of integers. The bit size of the ``value`` must be
6312 smaller than the bit size of the destination type, ``ty2``.
6317 The ``zext`` fills the high order bits of the ``value`` with zero bits
6318 until it reaches the size of the destination type, ``ty2``.
6320 When zero extending from i1, the result will always be either 0 or 1.
6325 .. code-block:: llvm
6327 %X = zext i32 257 to i64 ; yields i64:257
6328 %Y = zext i1 true to i32 ; yields i32:1
6329 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
6331 '``sext .. to``' Instruction
6332 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6339 <result> = sext <ty> <value> to <ty2> ; yields ty2
6344 The '``sext``' sign extends ``value`` to the type ``ty2``.
6349 The '``sext``' instruction takes a value to cast, and a type to cast it
6350 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
6351 the same number of integers. The bit size of the ``value`` must be
6352 smaller than the bit size of the destination type, ``ty2``.
6357 The '``sext``' instruction performs a sign extension by copying the sign
6358 bit (highest order bit) of the ``value`` until it reaches the bit size
6359 of the type ``ty2``.
6361 When sign extending from i1, the extension always results in -1 or 0.
6366 .. code-block:: llvm
6368 %X = sext i8 -1 to i16 ; yields i16 :65535
6369 %Y = sext i1 true to i32 ; yields i32:-1
6370 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
6372 '``fptrunc .. to``' Instruction
6373 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6380 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
6385 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
6390 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
6391 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
6392 The size of ``value`` must be larger than the size of ``ty2``. This
6393 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
6398 The '``fptrunc``' instruction truncates a ``value`` from a larger
6399 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
6400 point <t_floating>` type. If the value cannot fit within the
6401 destination type, ``ty2``, then the results are undefined.
6406 .. code-block:: llvm
6408 %X = fptrunc double 123.0 to float ; yields float:123.0
6409 %Y = fptrunc double 1.0E+300 to float ; yields undefined
6411 '``fpext .. to``' Instruction
6412 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6419 <result> = fpext <ty> <value> to <ty2> ; yields ty2
6424 The '``fpext``' extends a floating point ``value`` to a larger floating
6430 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
6431 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
6432 to. The source type must be smaller than the destination type.
6437 The '``fpext``' instruction extends the ``value`` from a smaller
6438 :ref:`floating point <t_floating>` type to a larger :ref:`floating
6439 point <t_floating>` type. The ``fpext`` cannot be used to make a
6440 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
6441 *no-op cast* for a floating point cast.
6446 .. code-block:: llvm
6448 %X = fpext float 3.125 to double ; yields double:3.125000e+00
6449 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
6451 '``fptoui .. to``' Instruction
6452 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6459 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
6464 The '``fptoui``' converts a floating point ``value`` to its unsigned
6465 integer equivalent of type ``ty2``.
6470 The '``fptoui``' instruction takes a value to cast, which must be a
6471 scalar or vector :ref:`floating point <t_floating>` value, and a type to
6472 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
6473 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
6474 type with the same number of elements as ``ty``
6479 The '``fptoui``' instruction converts its :ref:`floating
6480 point <t_floating>` operand into the nearest (rounding towards zero)
6481 unsigned integer value. If the value cannot fit in ``ty2``, the results
6487 .. code-block:: llvm
6489 %X = fptoui double 123.0 to i32 ; yields i32:123
6490 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
6491 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
6493 '``fptosi .. to``' Instruction
6494 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6501 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
6506 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
6507 ``value`` to type ``ty2``.
6512 The '``fptosi``' instruction takes a value to cast, which must be a
6513 scalar or vector :ref:`floating point <t_floating>` value, and a type to
6514 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
6515 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
6516 type with the same number of elements as ``ty``
6521 The '``fptosi``' instruction converts its :ref:`floating
6522 point <t_floating>` operand into the nearest (rounding towards zero)
6523 signed integer value. If the value cannot fit in ``ty2``, the results
6529 .. code-block:: llvm
6531 %X = fptosi double -123.0 to i32 ; yields i32:-123
6532 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
6533 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
6535 '``uitofp .. to``' Instruction
6536 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6543 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
6548 The '``uitofp``' instruction regards ``value`` as an unsigned integer
6549 and converts that value to the ``ty2`` type.
6554 The '``uitofp``' instruction takes a value to cast, which must be a
6555 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6556 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6557 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6558 type with the same number of elements as ``ty``
6563 The '``uitofp``' instruction interprets its operand as an unsigned
6564 integer quantity and converts it to the corresponding floating point
6565 value. If the value cannot fit in the floating point value, the results
6571 .. code-block:: llvm
6573 %X = uitofp i32 257 to float ; yields float:257.0
6574 %Y = uitofp i8 -1 to double ; yields double:255.0
6576 '``sitofp .. to``' Instruction
6577 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6584 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
6589 The '``sitofp``' instruction regards ``value`` as a signed integer and
6590 converts that value to the ``ty2`` type.
6595 The '``sitofp``' instruction takes a value to cast, which must be a
6596 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6597 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6598 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6599 type with the same number of elements as ``ty``
6604 The '``sitofp``' instruction interprets its operand as a signed integer
6605 quantity and converts it to the corresponding floating point value. If
6606 the value cannot fit in the floating point value, the results are
6612 .. code-block:: llvm
6614 %X = sitofp i32 257 to float ; yields float:257.0
6615 %Y = sitofp i8 -1 to double ; yields double:-1.0
6619 '``ptrtoint .. to``' Instruction
6620 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6627 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
6632 The '``ptrtoint``' instruction converts the pointer or a vector of
6633 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
6638 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
6639 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
6640 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
6641 a vector of integers type.
6646 The '``ptrtoint``' instruction converts ``value`` to integer type
6647 ``ty2`` by interpreting the pointer value as an integer and either
6648 truncating or zero extending that value to the size of the integer type.
6649 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
6650 ``value`` is larger than ``ty2`` then a truncation is done. If they are
6651 the same size, then nothing is done (*no-op cast*) other than a type
6657 .. code-block:: llvm
6659 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
6660 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
6661 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
6665 '``inttoptr .. to``' Instruction
6666 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6673 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
6678 The '``inttoptr``' instruction converts an integer ``value`` to a
6679 pointer type, ``ty2``.
6684 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
6685 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
6691 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
6692 applying either a zero extension or a truncation depending on the size
6693 of the integer ``value``. If ``value`` is larger than the size of a
6694 pointer then a truncation is done. If ``value`` is smaller than the size
6695 of a pointer then a zero extension is done. If they are the same size,
6696 nothing is done (*no-op cast*).
6701 .. code-block:: llvm
6703 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
6704 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
6705 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
6706 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
6710 '``bitcast .. to``' Instruction
6711 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6718 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
6723 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
6729 The '``bitcast``' instruction takes a value to cast, which must be a
6730 non-aggregate first class value, and a type to cast it to, which must
6731 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
6732 bit sizes of ``value`` and the destination type, ``ty2``, must be
6733 identical. If the source type is a pointer, the destination type must
6734 also be a pointer of the same size. This instruction supports bitwise
6735 conversion of vectors to integers and to vectors of other types (as
6736 long as they have the same size).
6741 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
6742 is always a *no-op cast* because no bits change with this
6743 conversion. The conversion is done as if the ``value`` had been stored
6744 to memory and read back as type ``ty2``. Pointer (or vector of
6745 pointers) types may only be converted to other pointer (or vector of
6746 pointers) types with the same address space through this instruction.
6747 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
6748 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
6753 .. code-block:: llvm
6755 %X = bitcast i8 255 to i8 ; yields i8 :-1
6756 %Y = bitcast i32* %x to sint* ; yields sint*:%x
6757 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
6758 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
6760 .. _i_addrspacecast:
6762 '``addrspacecast .. to``' Instruction
6763 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6770 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
6775 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
6776 address space ``n`` to type ``pty2`` in address space ``m``.
6781 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
6782 to cast and a pointer type to cast it to, which must have a different
6788 The '``addrspacecast``' instruction converts the pointer value
6789 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
6790 value modification, depending on the target and the address space
6791 pair. Pointer conversions within the same address space must be
6792 performed with the ``bitcast`` instruction. Note that if the address space
6793 conversion is legal then both result and operand refer to the same memory
6799 .. code-block:: llvm
6801 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
6802 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
6803 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
6810 The instructions in this category are the "miscellaneous" instructions,
6811 which defy better classification.
6815 '``icmp``' Instruction
6816 ^^^^^^^^^^^^^^^^^^^^^^
6823 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6828 The '``icmp``' instruction returns a boolean value or a vector of
6829 boolean values based on comparison of its two integer, integer vector,
6830 pointer, or pointer vector operands.
6835 The '``icmp``' instruction takes three operands. The first operand is
6836 the condition code indicating the kind of comparison to perform. It is
6837 not a value, just a keyword. The possible condition code are:
6840 #. ``ne``: not equal
6841 #. ``ugt``: unsigned greater than
6842 #. ``uge``: unsigned greater or equal
6843 #. ``ult``: unsigned less than
6844 #. ``ule``: unsigned less or equal
6845 #. ``sgt``: signed greater than
6846 #. ``sge``: signed greater or equal
6847 #. ``slt``: signed less than
6848 #. ``sle``: signed less or equal
6850 The remaining two arguments must be :ref:`integer <t_integer>` or
6851 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
6852 must also be identical types.
6857 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
6858 code given as ``cond``. The comparison performed always yields either an
6859 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
6861 #. ``eq``: yields ``true`` if the operands are equal, ``false``
6862 otherwise. No sign interpretation is necessary or performed.
6863 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
6864 otherwise. No sign interpretation is necessary or performed.
6865 #. ``ugt``: interprets the operands as unsigned values and yields
6866 ``true`` if ``op1`` is greater than ``op2``.
6867 #. ``uge``: interprets the operands as unsigned values and yields
6868 ``true`` if ``op1`` is greater than or equal to ``op2``.
6869 #. ``ult``: interprets the operands as unsigned values and yields
6870 ``true`` if ``op1`` is less than ``op2``.
6871 #. ``ule``: interprets the operands as unsigned values and yields
6872 ``true`` if ``op1`` is less than or equal to ``op2``.
6873 #. ``sgt``: interprets the operands as signed values and yields ``true``
6874 if ``op1`` is greater than ``op2``.
6875 #. ``sge``: interprets the operands as signed values and yields ``true``
6876 if ``op1`` is greater than or equal to ``op2``.
6877 #. ``slt``: interprets the operands as signed values and yields ``true``
6878 if ``op1`` is less than ``op2``.
6879 #. ``sle``: interprets the operands as signed values and yields ``true``
6880 if ``op1`` is less than or equal to ``op2``.
6882 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
6883 are compared as if they were integers.
6885 If the operands are integer vectors, then they are compared element by
6886 element. The result is an ``i1`` vector with the same number of elements
6887 as the values being compared. Otherwise, the result is an ``i1``.
6892 .. code-block:: llvm
6894 <result> = icmp eq i32 4, 5 ; yields: result=false
6895 <result> = icmp ne float* %X, %X ; yields: result=false
6896 <result> = icmp ult i16 4, 5 ; yields: result=true
6897 <result> = icmp sgt i16 4, 5 ; yields: result=false
6898 <result> = icmp ule i16 -4, 5 ; yields: result=false
6899 <result> = icmp sge i16 4, 5 ; yields: result=false
6901 Note that the code generator does not yet support vector types with the
6902 ``icmp`` instruction.
6906 '``fcmp``' Instruction
6907 ^^^^^^^^^^^^^^^^^^^^^^
6914 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6919 The '``fcmp``' instruction returns a boolean value or vector of boolean
6920 values based on comparison of its operands.
6922 If the operands are floating point scalars, then the result type is a
6923 boolean (:ref:`i1 <t_integer>`).
6925 If the operands are floating point vectors, then the result type is a
6926 vector of boolean with the same number of elements as the operands being
6932 The '``fcmp``' instruction takes three operands. The first operand is
6933 the condition code indicating the kind of comparison to perform. It is
6934 not a value, just a keyword. The possible condition code are:
6936 #. ``false``: no comparison, always returns false
6937 #. ``oeq``: ordered and equal
6938 #. ``ogt``: ordered and greater than
6939 #. ``oge``: ordered and greater than or equal
6940 #. ``olt``: ordered and less than
6941 #. ``ole``: ordered and less than or equal
6942 #. ``one``: ordered and not equal
6943 #. ``ord``: ordered (no nans)
6944 #. ``ueq``: unordered or equal
6945 #. ``ugt``: unordered or greater than
6946 #. ``uge``: unordered or greater than or equal
6947 #. ``ult``: unordered or less than
6948 #. ``ule``: unordered or less than or equal
6949 #. ``une``: unordered or not equal
6950 #. ``uno``: unordered (either nans)
6951 #. ``true``: no comparison, always returns true
6953 *Ordered* means that neither operand is a QNAN while *unordered* means
6954 that either operand may be a QNAN.
6956 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
6957 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
6958 type. They must have identical types.
6963 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6964 condition code given as ``cond``. If the operands are vectors, then the
6965 vectors are compared element by element. Each comparison performed
6966 always yields an :ref:`i1 <t_integer>` result, as follows:
6968 #. ``false``: always yields ``false``, regardless of operands.
6969 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6970 is equal to ``op2``.
6971 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6972 is greater than ``op2``.
6973 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6974 is greater than or equal to ``op2``.
6975 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6976 is less than ``op2``.
6977 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6978 is less than or equal to ``op2``.
6979 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6980 is not equal to ``op2``.
6981 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6982 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6984 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6985 greater than ``op2``.
6986 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6987 greater than or equal to ``op2``.
6988 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6990 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6991 less than or equal to ``op2``.
6992 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6993 not equal to ``op2``.
6994 #. ``uno``: yields ``true`` if either operand is a QNAN.
6995 #. ``true``: always yields ``true``, regardless of operands.
7000 .. code-block:: llvm
7002 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
7003 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
7004 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
7005 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
7007 Note that the code generator does not yet support vector types with the
7008 ``fcmp`` instruction.
7012 '``phi``' Instruction
7013 ^^^^^^^^^^^^^^^^^^^^^
7020 <result> = phi <ty> [ <val0>, <label0>], ...
7025 The '``phi``' instruction is used to implement the φ node in the SSA
7026 graph representing the function.
7031 The type of the incoming values is specified with the first type field.
7032 After this, the '``phi``' instruction takes a list of pairs as
7033 arguments, with one pair for each predecessor basic block of the current
7034 block. Only values of :ref:`first class <t_firstclass>` type may be used as
7035 the value arguments to the PHI node. Only labels may be used as the
7038 There must be no non-phi instructions between the start of a basic block
7039 and the PHI instructions: i.e. PHI instructions must be first in a basic
7042 For the purposes of the SSA form, the use of each incoming value is
7043 deemed to occur on the edge from the corresponding predecessor block to
7044 the current block (but after any definition of an '``invoke``'
7045 instruction's return value on the same edge).
7050 At runtime, the '``phi``' instruction logically takes on the value
7051 specified by the pair corresponding to the predecessor basic block that
7052 executed just prior to the current block.
7057 .. code-block:: llvm
7059 Loop: ; Infinite loop that counts from 0 on up...
7060 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
7061 %nextindvar = add i32 %indvar, 1
7066 '``select``' Instruction
7067 ^^^^^^^^^^^^^^^^^^^^^^^^
7074 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
7076 selty is either i1 or {<N x i1>}
7081 The '``select``' instruction is used to choose one value based on a
7082 condition, without IR-level branching.
7087 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
7088 values indicating the condition, and two values of the same :ref:`first
7089 class <t_firstclass>` type.
7094 If the condition is an i1 and it evaluates to 1, the instruction returns
7095 the first value argument; otherwise, it returns the second value
7098 If the condition is a vector of i1, then the value arguments must be
7099 vectors of the same size, and the selection is done element by element.
7101 If the condition is an i1 and the value arguments are vectors of the
7102 same size, then an entire vector is selected.
7107 .. code-block:: llvm
7109 %X = select i1 true, i8 17, i8 42 ; yields i8:17
7113 '``call``' Instruction
7114 ^^^^^^^^^^^^^^^^^^^^^^
7121 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
7126 The '``call``' instruction represents a simple function call.
7131 This instruction requires several arguments:
7133 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
7134 should perform tail call optimization. The ``tail`` marker is a hint that
7135 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
7136 means that the call must be tail call optimized in order for the program to
7137 be correct. The ``musttail`` marker provides these guarantees:
7139 #. The call will not cause unbounded stack growth if it is part of a
7140 recursive cycle in the call graph.
7141 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
7144 Both markers imply that the callee does not access allocas or varargs from
7145 the caller. Calls marked ``musttail`` must obey the following additional
7148 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
7149 or a pointer bitcast followed by a ret instruction.
7150 - The ret instruction must return the (possibly bitcasted) value
7151 produced by the call or void.
7152 - The caller and callee prototypes must match. Pointer types of
7153 parameters or return types may differ in pointee type, but not
7155 - The calling conventions of the caller and callee must match.
7156 - All ABI-impacting function attributes, such as sret, byval, inreg,
7157 returned, and inalloca, must match.
7158 - The callee must be varargs iff the caller is varargs. Bitcasting a
7159 non-varargs function to the appropriate varargs type is legal so
7160 long as the non-varargs prefixes obey the other rules.
7162 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
7163 the following conditions are met:
7165 - Caller and callee both have the calling convention ``fastcc``.
7166 - The call is in tail position (ret immediately follows call and ret
7167 uses value of call or is void).
7168 - Option ``-tailcallopt`` is enabled, or
7169 ``llvm::GuaranteedTailCallOpt`` is ``true``.
7170 - `Platform-specific constraints are
7171 met. <CodeGenerator.html#tailcallopt>`_
7173 #. The optional "cconv" marker indicates which :ref:`calling
7174 convention <callingconv>` the call should use. If none is
7175 specified, the call defaults to using C calling conventions. The
7176 calling convention of the call must match the calling convention of
7177 the target function, or else the behavior is undefined.
7178 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
7179 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
7181 #. '``ty``': the type of the call instruction itself which is also the
7182 type of the return value. Functions that return no value are marked
7184 #. '``fnty``': shall be the signature of the pointer to function value
7185 being invoked. The argument types must match the types implied by
7186 this signature. This type can be omitted if the function is not
7187 varargs and if the function type does not return a pointer to a
7189 #. '``fnptrval``': An LLVM value containing a pointer to a function to
7190 be invoked. In most cases, this is a direct function invocation, but
7191 indirect ``call``'s are just as possible, calling an arbitrary pointer
7193 #. '``function args``': argument list whose types match the function
7194 signature argument types and parameter attributes. All arguments must
7195 be of :ref:`first class <t_firstclass>` type. If the function signature
7196 indicates the function accepts a variable number of arguments, the
7197 extra arguments can be specified.
7198 #. The optional :ref:`function attributes <fnattrs>` list. Only
7199 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
7200 attributes are valid here.
7205 The '``call``' instruction is used to cause control flow to transfer to
7206 a specified function, with its incoming arguments bound to the specified
7207 values. Upon a '``ret``' instruction in the called function, control
7208 flow continues with the instruction after the function call, and the
7209 return value of the function is bound to the result argument.
7214 .. code-block:: llvm
7216 %retval = call i32 @test(i32 %argc)
7217 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
7218 %X = tail call i32 @foo() ; yields i32
7219 %Y = tail call fastcc i32 @foo() ; yields i32
7220 call void %foo(i8 97 signext)
7222 %struct.A = type { i32, i8 }
7223 %r = call %struct.A @foo() ; yields { i32, i8 }
7224 %gr = extractvalue %struct.A %r, 0 ; yields i32
7225 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
7226 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
7227 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
7229 llvm treats calls to some functions with names and arguments that match
7230 the standard C99 library as being the C99 library functions, and may
7231 perform optimizations or generate code for them under that assumption.
7232 This is something we'd like to change in the future to provide better
7233 support for freestanding environments and non-C-based languages.
7237 '``va_arg``' Instruction
7238 ^^^^^^^^^^^^^^^^^^^^^^^^
7245 <resultval> = va_arg <va_list*> <arglist>, <argty>
7250 The '``va_arg``' instruction is used to access arguments passed through
7251 the "variable argument" area of a function call. It is used to implement
7252 the ``va_arg`` macro in C.
7257 This instruction takes a ``va_list*`` value and the type of the
7258 argument. It returns a value of the specified argument type and
7259 increments the ``va_list`` to point to the next argument. The actual
7260 type of ``va_list`` is target specific.
7265 The '``va_arg``' instruction loads an argument of the specified type
7266 from the specified ``va_list`` and causes the ``va_list`` to point to
7267 the next argument. For more information, see the variable argument
7268 handling :ref:`Intrinsic Functions <int_varargs>`.
7270 It is legal for this instruction to be called in a function which does
7271 not take a variable number of arguments, for example, the ``vfprintf``
7274 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
7275 function <intrinsics>` because it takes a type as an argument.
7280 See the :ref:`variable argument processing <int_varargs>` section.
7282 Note that the code generator does not yet fully support va\_arg on many
7283 targets. Also, it does not currently support va\_arg with aggregate
7284 types on any target.
7288 '``landingpad``' Instruction
7289 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7296 <resultval> = landingpad <resultty> <clause>+
7297 <resultval> = landingpad <resultty> cleanup <clause>*
7299 <clause> := catch <type> <value>
7300 <clause> := filter <array constant type> <array constant>
7305 The '``landingpad``' instruction is used by `LLVM's exception handling
7306 system <ExceptionHandling.html#overview>`_ to specify that a basic block
7307 is a landing pad --- one where the exception lands, and corresponds to the
7308 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
7309 defines values supplied by the :ref:`personality function <personalityfn>` upon
7310 re-entry to the function. The ``resultval`` has the type ``resultty``.
7316 ``cleanup`` flag indicates that the landing pad block is a cleanup.
7318 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
7319 contains the global variable representing the "type" that may be caught
7320 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
7321 clause takes an array constant as its argument. Use
7322 "``[0 x i8**] undef``" for a filter which cannot throw. The
7323 '``landingpad``' instruction must contain *at least* one ``clause`` or
7324 the ``cleanup`` flag.
7329 The '``landingpad``' instruction defines the values which are set by the
7330 :ref:`personality function <personalityfn>` upon re-entry to the function, and
7331 therefore the "result type" of the ``landingpad`` instruction. As with
7332 calling conventions, how the personality function results are
7333 represented in LLVM IR is target specific.
7335 The clauses are applied in order from top to bottom. If two
7336 ``landingpad`` instructions are merged together through inlining, the
7337 clauses from the calling function are appended to the list of clauses.
7338 When the call stack is being unwound due to an exception being thrown,
7339 the exception is compared against each ``clause`` in turn. If it doesn't
7340 match any of the clauses, and the ``cleanup`` flag is not set, then
7341 unwinding continues further up the call stack.
7343 The ``landingpad`` instruction has several restrictions:
7345 - A landing pad block is a basic block which is the unwind destination
7346 of an '``invoke``' instruction.
7347 - A landing pad block must have a '``landingpad``' instruction as its
7348 first non-PHI instruction.
7349 - There can be only one '``landingpad``' instruction within the landing
7351 - A basic block that is not a landing pad block may not include a
7352 '``landingpad``' instruction.
7357 .. code-block:: llvm
7359 ;; A landing pad which can catch an integer.
7360 %res = landingpad { i8*, i32 }
7362 ;; A landing pad that is a cleanup.
7363 %res = landingpad { i8*, i32 }
7365 ;; A landing pad which can catch an integer and can only throw a double.
7366 %res = landingpad { i8*, i32 }
7368 filter [1 x i8**] [@_ZTId]
7375 LLVM supports the notion of an "intrinsic function". These functions
7376 have well known names and semantics and are required to follow certain
7377 restrictions. Overall, these intrinsics represent an extension mechanism
7378 for the LLVM language that does not require changing all of the
7379 transformations in LLVM when adding to the language (or the bitcode
7380 reader/writer, the parser, etc...).
7382 Intrinsic function names must all start with an "``llvm.``" prefix. This
7383 prefix is reserved in LLVM for intrinsic names; thus, function names may
7384 not begin with this prefix. Intrinsic functions must always be external
7385 functions: you cannot define the body of intrinsic functions. Intrinsic
7386 functions may only be used in call or invoke instructions: it is illegal
7387 to take the address of an intrinsic function. Additionally, because
7388 intrinsic functions are part of the LLVM language, it is required if any
7389 are added that they be documented here.
7391 Some intrinsic functions can be overloaded, i.e., the intrinsic
7392 represents a family of functions that perform the same operation but on
7393 different data types. Because LLVM can represent over 8 million
7394 different integer types, overloading is used commonly to allow an
7395 intrinsic function to operate on any integer type. One or more of the
7396 argument types or the result type can be overloaded to accept any
7397 integer type. Argument types may also be defined as exactly matching a
7398 previous argument's type or the result type. This allows an intrinsic
7399 function which accepts multiple arguments, but needs all of them to be
7400 of the same type, to only be overloaded with respect to a single
7401 argument or the result.
7403 Overloaded intrinsics will have the names of its overloaded argument
7404 types encoded into its function name, each preceded by a period. Only
7405 those types which are overloaded result in a name suffix. Arguments
7406 whose type is matched against another type do not. For example, the
7407 ``llvm.ctpop`` function can take an integer of any width and returns an
7408 integer of exactly the same integer width. This leads to a family of
7409 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
7410 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
7411 overloaded, and only one type suffix is required. Because the argument's
7412 type is matched against the return type, it does not require its own
7415 To learn how to add an intrinsic function, please see the `Extending
7416 LLVM Guide <ExtendingLLVM.html>`_.
7420 Variable Argument Handling Intrinsics
7421 -------------------------------------
7423 Variable argument support is defined in LLVM with the
7424 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
7425 functions. These functions are related to the similarly named macros
7426 defined in the ``<stdarg.h>`` header file.
7428 All of these functions operate on arguments that use a target-specific
7429 value type "``va_list``". The LLVM assembly language reference manual
7430 does not define what this type is, so all transformations should be
7431 prepared to handle these functions regardless of the type used.
7433 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
7434 variable argument handling intrinsic functions are used.
7436 .. code-block:: llvm
7438 ; This struct is different for every platform. For most platforms,
7439 ; it is merely an i8*.
7440 %struct.va_list = type { i8* }
7442 ; For Unix x86_64 platforms, va_list is the following struct:
7443 ; %struct.va_list = type { i32, i32, i8*, i8* }
7445 define i32 @test(i32 %X, ...) {
7446 ; Initialize variable argument processing
7447 %ap = alloca %struct.va_list
7448 %ap2 = bitcast %struct.va_list* %ap to i8*
7449 call void @llvm.va_start(i8* %ap2)
7451 ; Read a single integer argument
7452 %tmp = va_arg i8* %ap2, i32
7454 ; Demonstrate usage of llvm.va_copy and llvm.va_end
7456 %aq2 = bitcast i8** %aq to i8*
7457 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
7458 call void @llvm.va_end(i8* %aq2)
7460 ; Stop processing of arguments.
7461 call void @llvm.va_end(i8* %ap2)
7465 declare void @llvm.va_start(i8*)
7466 declare void @llvm.va_copy(i8*, i8*)
7467 declare void @llvm.va_end(i8*)
7471 '``llvm.va_start``' Intrinsic
7472 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7479 declare void @llvm.va_start(i8* <arglist>)
7484 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
7485 subsequent use by ``va_arg``.
7490 The argument is a pointer to a ``va_list`` element to initialize.
7495 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
7496 available in C. In a target-dependent way, it initializes the
7497 ``va_list`` element to which the argument points, so that the next call
7498 to ``va_arg`` will produce the first variable argument passed to the
7499 function. Unlike the C ``va_start`` macro, this intrinsic does not need
7500 to know the last argument of the function as the compiler can figure
7503 '``llvm.va_end``' Intrinsic
7504 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7511 declare void @llvm.va_end(i8* <arglist>)
7516 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
7517 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
7522 The argument is a pointer to a ``va_list`` to destroy.
7527 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
7528 available in C. In a target-dependent way, it destroys the ``va_list``
7529 element to which the argument points. Calls to
7530 :ref:`llvm.va_start <int_va_start>` and
7531 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
7536 '``llvm.va_copy``' Intrinsic
7537 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7544 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
7549 The '``llvm.va_copy``' intrinsic copies the current argument position
7550 from the source argument list to the destination argument list.
7555 The first argument is a pointer to a ``va_list`` element to initialize.
7556 The second argument is a pointer to a ``va_list`` element to copy from.
7561 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
7562 available in C. In a target-dependent way, it copies the source
7563 ``va_list`` element into the destination ``va_list`` element. This
7564 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
7565 arbitrarily complex and require, for example, memory allocation.
7567 Accurate Garbage Collection Intrinsics
7568 --------------------------------------
7570 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
7571 (GC) requires the frontend to generate code containing appropriate intrinsic
7572 calls and select an appropriate GC strategy which knows how to lower these
7573 intrinsics in a manner which is appropriate for the target collector.
7575 These intrinsics allow identification of :ref:`GC roots on the
7576 stack <int_gcroot>`, as well as garbage collector implementations that
7577 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
7578 Frontends for type-safe garbage collected languages should generate
7579 these intrinsics to make use of the LLVM garbage collectors. For more
7580 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
7582 Experimental Statepoint Intrinsics
7583 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7585 LLVM provides an second experimental set of intrinsics for describing garbage
7586 collection safepoints in compiled code. These intrinsics are an alternative
7587 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
7588 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
7589 differences in approach are covered in the `Garbage Collection with LLVM
7590 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
7591 described in :doc:`Statepoints`.
7595 '``llvm.gcroot``' Intrinsic
7596 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7603 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
7608 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
7609 the code generator, and allows some metadata to be associated with it.
7614 The first argument specifies the address of a stack object that contains
7615 the root pointer. The second pointer (which must be either a constant or
7616 a global value address) contains the meta-data to be associated with the
7622 At runtime, a call to this intrinsic stores a null pointer into the
7623 "ptrloc" location. At compile-time, the code generator generates
7624 information to allow the runtime to find the pointer at GC safe points.
7625 The '``llvm.gcroot``' intrinsic may only be used in a function which
7626 :ref:`specifies a GC algorithm <gc>`.
7630 '``llvm.gcread``' Intrinsic
7631 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7638 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
7643 The '``llvm.gcread``' intrinsic identifies reads of references from heap
7644 locations, allowing garbage collector implementations that require read
7650 The second argument is the address to read from, which should be an
7651 address allocated from the garbage collector. The first object is a
7652 pointer to the start of the referenced object, if needed by the language
7653 runtime (otherwise null).
7658 The '``llvm.gcread``' intrinsic has the same semantics as a load
7659 instruction, but may be replaced with substantially more complex code by
7660 the garbage collector runtime, as needed. The '``llvm.gcread``'
7661 intrinsic may only be used in a function which :ref:`specifies a GC
7666 '``llvm.gcwrite``' Intrinsic
7667 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7674 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
7679 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
7680 locations, allowing garbage collector implementations that require write
7681 barriers (such as generational or reference counting collectors).
7686 The first argument is the reference to store, the second is the start of
7687 the object to store it to, and the third is the address of the field of
7688 Obj to store to. If the runtime does not require a pointer to the
7689 object, Obj may be null.
7694 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
7695 instruction, but may be replaced with substantially more complex code by
7696 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
7697 intrinsic may only be used in a function which :ref:`specifies a GC
7700 Code Generator Intrinsics
7701 -------------------------
7703 These intrinsics are provided by LLVM to expose special features that
7704 may only be implemented with code generator support.
7706 '``llvm.returnaddress``' Intrinsic
7707 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7714 declare i8 *@llvm.returnaddress(i32 <level>)
7719 The '``llvm.returnaddress``' intrinsic attempts to compute a
7720 target-specific value indicating the return address of the current
7721 function or one of its callers.
7726 The argument to this intrinsic indicates which function to return the
7727 address for. Zero indicates the calling function, one indicates its
7728 caller, etc. The argument is **required** to be a constant integer
7734 The '``llvm.returnaddress``' intrinsic either returns a pointer
7735 indicating the return address of the specified call frame, or zero if it
7736 cannot be identified. The value returned by this intrinsic is likely to
7737 be incorrect or 0 for arguments other than zero, so it should only be
7738 used for debugging purposes.
7740 Note that calling this intrinsic does not prevent function inlining or
7741 other aggressive transformations, so the value returned may not be that
7742 of the obvious source-language caller.
7744 '``llvm.frameaddress``' Intrinsic
7745 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7752 declare i8* @llvm.frameaddress(i32 <level>)
7757 The '``llvm.frameaddress``' intrinsic attempts to return the
7758 target-specific frame pointer value for the specified stack frame.
7763 The argument to this intrinsic indicates which function to return the
7764 frame pointer for. Zero indicates the calling function, one indicates
7765 its caller, etc. The argument is **required** to be a constant integer
7771 The '``llvm.frameaddress``' intrinsic either returns a pointer
7772 indicating the frame address of the specified call frame, or zero if it
7773 cannot be identified. The value returned by this intrinsic is likely to
7774 be incorrect or 0 for arguments other than zero, so it should only be
7775 used for debugging purposes.
7777 Note that calling this intrinsic does not prevent function inlining or
7778 other aggressive transformations, so the value returned may not be that
7779 of the obvious source-language caller.
7781 '``llvm.frameescape``' and '``llvm.framerecover``' Intrinsics
7782 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7789 declare void @llvm.frameescape(...)
7790 declare i8* @llvm.framerecover(i8* %func, i8* %fp, i32 %idx)
7795 The '``llvm.frameescape``' intrinsic escapes offsets of a collection of static
7796 allocas, and the '``llvm.framerecover``' intrinsic applies those offsets to a
7797 live frame pointer to recover the address of the allocation. The offset is
7798 computed during frame layout of the caller of ``llvm.frameescape``.
7803 All arguments to '``llvm.frameescape``' must be pointers to static allocas or
7804 casts of static allocas. Each function can only call '``llvm.frameescape``'
7805 once, and it can only do so from the entry block.
7807 The ``func`` argument to '``llvm.framerecover``' must be a constant
7808 bitcasted pointer to a function defined in the current module. The code
7809 generator cannot determine the frame allocation offset of functions defined in
7812 The ``fp`` argument to '``llvm.framerecover``' must be a frame
7813 pointer of a call frame that is currently live. The return value of
7814 '``llvm.frameaddress``' is one way to produce such a value, but most platforms
7815 also expose the frame pointer through stack unwinding mechanisms.
7817 The ``idx`` argument to '``llvm.framerecover``' indicates which alloca passed to
7818 '``llvm.frameescape``' to recover. It is zero-indexed.
7823 These intrinsics allow a group of functions to access one stack memory
7824 allocation in an ancestor stack frame. The memory returned from
7825 '``llvm.frameallocate``' may be allocated prior to stack realignment, so the
7826 memory is only aligned to the ABI-required stack alignment. Each function may
7827 only call '``llvm.frameallocate``' one or zero times from the function entry
7828 block. The frame allocation intrinsic inhibits inlining, as any frame
7829 allocations in the inlined function frame are likely to be at a different
7830 offset from the one used by '``llvm.framerecover``' called with the
7833 .. _int_read_register:
7834 .. _int_write_register:
7836 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
7837 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7844 declare i32 @llvm.read_register.i32(metadata)
7845 declare i64 @llvm.read_register.i64(metadata)
7846 declare void @llvm.write_register.i32(metadata, i32 @value)
7847 declare void @llvm.write_register.i64(metadata, i64 @value)
7853 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
7854 provides access to the named register. The register must be valid on
7855 the architecture being compiled to. The type needs to be compatible
7856 with the register being read.
7861 The '``llvm.read_register``' intrinsic returns the current value of the
7862 register, where possible. The '``llvm.write_register``' intrinsic sets
7863 the current value of the register, where possible.
7865 This is useful to implement named register global variables that need
7866 to always be mapped to a specific register, as is common practice on
7867 bare-metal programs including OS kernels.
7869 The compiler doesn't check for register availability or use of the used
7870 register in surrounding code, including inline assembly. Because of that,
7871 allocatable registers are not supported.
7873 Warning: So far it only works with the stack pointer on selected
7874 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
7875 work is needed to support other registers and even more so, allocatable
7880 '``llvm.stacksave``' Intrinsic
7881 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7888 declare i8* @llvm.stacksave()
7893 The '``llvm.stacksave``' intrinsic is used to remember the current state
7894 of the function stack, for use with
7895 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
7896 implementing language features like scoped automatic variable sized
7902 This intrinsic returns a opaque pointer value that can be passed to
7903 :ref:`llvm.stackrestore <int_stackrestore>`. When an
7904 ``llvm.stackrestore`` intrinsic is executed with a value saved from
7905 ``llvm.stacksave``, it effectively restores the state of the stack to
7906 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
7907 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
7908 were allocated after the ``llvm.stacksave`` was executed.
7910 .. _int_stackrestore:
7912 '``llvm.stackrestore``' Intrinsic
7913 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7920 declare void @llvm.stackrestore(i8* %ptr)
7925 The '``llvm.stackrestore``' intrinsic is used to restore the state of
7926 the function stack to the state it was in when the corresponding
7927 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
7928 useful for implementing language features like scoped automatic variable
7929 sized arrays in C99.
7934 See the description for :ref:`llvm.stacksave <int_stacksave>`.
7936 '``llvm.prefetch``' Intrinsic
7937 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7944 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
7949 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
7950 insert a prefetch instruction if supported; otherwise, it is a noop.
7951 Prefetches have no effect on the behavior of the program but can change
7952 its performance characteristics.
7957 ``address`` is the address to be prefetched, ``rw`` is the specifier
7958 determining if the fetch should be for a read (0) or write (1), and
7959 ``locality`` is a temporal locality specifier ranging from (0) - no
7960 locality, to (3) - extremely local keep in cache. The ``cache type``
7961 specifies whether the prefetch is performed on the data (1) or
7962 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
7963 arguments must be constant integers.
7968 This intrinsic does not modify the behavior of the program. In
7969 particular, prefetches cannot trap and do not produce a value. On
7970 targets that support this intrinsic, the prefetch can provide hints to
7971 the processor cache for better performance.
7973 '``llvm.pcmarker``' Intrinsic
7974 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7981 declare void @llvm.pcmarker(i32 <id>)
7986 The '``llvm.pcmarker``' intrinsic is a method to export a Program
7987 Counter (PC) in a region of code to simulators and other tools. The
7988 method is target specific, but it is expected that the marker will use
7989 exported symbols to transmit the PC of the marker. The marker makes no
7990 guarantees that it will remain with any specific instruction after
7991 optimizations. It is possible that the presence of a marker will inhibit
7992 optimizations. The intended use is to be inserted after optimizations to
7993 allow correlations of simulation runs.
7998 ``id`` is a numerical id identifying the marker.
8003 This intrinsic does not modify the behavior of the program. Backends
8004 that do not support this intrinsic may ignore it.
8006 '``llvm.readcyclecounter``' Intrinsic
8007 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8014 declare i64 @llvm.readcyclecounter()
8019 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
8020 counter register (or similar low latency, high accuracy clocks) on those
8021 targets that support it. On X86, it should map to RDTSC. On Alpha, it
8022 should map to RPCC. As the backing counters overflow quickly (on the
8023 order of 9 seconds on alpha), this should only be used for small
8029 When directly supported, reading the cycle counter should not modify any
8030 memory. Implementations are allowed to either return a application
8031 specific value or a system wide value. On backends without support, this
8032 is lowered to a constant 0.
8034 Note that runtime support may be conditional on the privilege-level code is
8035 running at and the host platform.
8037 '``llvm.clear_cache``' Intrinsic
8038 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8045 declare void @llvm.clear_cache(i8*, i8*)
8050 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
8051 in the specified range to the execution unit of the processor. On
8052 targets with non-unified instruction and data cache, the implementation
8053 flushes the instruction cache.
8058 On platforms with coherent instruction and data caches (e.g. x86), this
8059 intrinsic is a nop. On platforms with non-coherent instruction and data
8060 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
8061 instructions or a system call, if cache flushing requires special
8064 The default behavior is to emit a call to ``__clear_cache`` from the run
8067 This instrinsic does *not* empty the instruction pipeline. Modifications
8068 of the current function are outside the scope of the intrinsic.
8070 '``llvm.instrprof_increment``' Intrinsic
8071 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8078 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
8079 i32 <num-counters>, i32 <index>)
8084 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
8085 frontend for use with instrumentation based profiling. These will be
8086 lowered by the ``-instrprof`` pass to generate execution counts of a
8092 The first argument is a pointer to a global variable containing the
8093 name of the entity being instrumented. This should generally be the
8094 (mangled) function name for a set of counters.
8096 The second argument is a hash value that can be used by the consumer
8097 of the profile data to detect changes to the instrumented source, and
8098 the third is the number of counters associated with ``name``. It is an
8099 error if ``hash`` or ``num-counters`` differ between two instances of
8100 ``instrprof_increment`` that refer to the same name.
8102 The last argument refers to which of the counters for ``name`` should
8103 be incremented. It should be a value between 0 and ``num-counters``.
8108 This intrinsic represents an increment of a profiling counter. It will
8109 cause the ``-instrprof`` pass to generate the appropriate data
8110 structures and the code to increment the appropriate value, in a
8111 format that can be written out by a compiler runtime and consumed via
8112 the ``llvm-profdata`` tool.
8114 Standard C Library Intrinsics
8115 -----------------------------
8117 LLVM provides intrinsics for a few important standard C library
8118 functions. These intrinsics allow source-language front-ends to pass
8119 information about the alignment of the pointer arguments to the code
8120 generator, providing opportunity for more efficient code generation.
8124 '``llvm.memcpy``' Intrinsic
8125 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8130 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
8131 integer bit width and for different address spaces. Not all targets
8132 support all bit widths however.
8136 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
8137 i32 <len>, i32 <align>, i1 <isvolatile>)
8138 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
8139 i64 <len>, i32 <align>, i1 <isvolatile>)
8144 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
8145 source location to the destination location.
8147 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
8148 intrinsics do not return a value, takes extra alignment/isvolatile
8149 arguments and the pointers can be in specified address spaces.
8154 The first argument is a pointer to the destination, the second is a
8155 pointer to the source. The third argument is an integer argument
8156 specifying the number of bytes to copy, the fourth argument is the
8157 alignment of the source and destination locations, and the fifth is a
8158 boolean indicating a volatile access.
8160 If the call to this intrinsic has an alignment value that is not 0 or 1,
8161 then the caller guarantees that both the source and destination pointers
8162 are aligned to that boundary.
8164 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
8165 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
8166 very cleanly specified and it is unwise to depend on it.
8171 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
8172 source location to the destination location, which are not allowed to
8173 overlap. It copies "len" bytes of memory over. If the argument is known
8174 to be aligned to some boundary, this can be specified as the fourth
8175 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
8177 '``llvm.memmove``' Intrinsic
8178 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8183 This is an overloaded intrinsic. You can use llvm.memmove on any integer
8184 bit width and for different address space. Not all targets support all
8189 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
8190 i32 <len>, i32 <align>, i1 <isvolatile>)
8191 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
8192 i64 <len>, i32 <align>, i1 <isvolatile>)
8197 The '``llvm.memmove.*``' intrinsics move a block of memory from the
8198 source location to the destination location. It is similar to the
8199 '``llvm.memcpy``' intrinsic but allows the two memory locations to
8202 Note that, unlike the standard libc function, the ``llvm.memmove.*``
8203 intrinsics do not return a value, takes extra alignment/isvolatile
8204 arguments and the pointers can be in specified address spaces.
8209 The first argument is a pointer to the destination, the second is a
8210 pointer to the source. The third argument is an integer argument
8211 specifying the number of bytes to copy, the fourth argument is the
8212 alignment of the source and destination locations, and the fifth is a
8213 boolean indicating a volatile access.
8215 If the call to this intrinsic has an alignment value that is not 0 or 1,
8216 then the caller guarantees that the source and destination pointers are
8217 aligned to that boundary.
8219 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
8220 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
8221 not very cleanly specified and it is unwise to depend on it.
8226 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
8227 source location to the destination location, which may overlap. It
8228 copies "len" bytes of memory over. If the argument is known to be
8229 aligned to some boundary, this can be specified as the fourth argument,
8230 otherwise it should be set to 0 or 1 (both meaning no alignment).
8232 '``llvm.memset.*``' Intrinsics
8233 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8238 This is an overloaded intrinsic. You can use llvm.memset on any integer
8239 bit width and for different address spaces. However, not all targets
8240 support all bit widths.
8244 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
8245 i32 <len>, i32 <align>, i1 <isvolatile>)
8246 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
8247 i64 <len>, i32 <align>, i1 <isvolatile>)
8252 The '``llvm.memset.*``' intrinsics fill a block of memory with a
8253 particular byte value.
8255 Note that, unlike the standard libc function, the ``llvm.memset``
8256 intrinsic does not return a value and takes extra alignment/volatile
8257 arguments. Also, the destination can be in an arbitrary address space.
8262 The first argument is a pointer to the destination to fill, the second
8263 is the byte value with which to fill it, the third argument is an
8264 integer argument specifying the number of bytes to fill, and the fourth
8265 argument is the known alignment of the destination location.
8267 If the call to this intrinsic has an alignment value that is not 0 or 1,
8268 then the caller guarantees that the destination pointer is aligned to
8271 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
8272 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
8273 very cleanly specified and it is unwise to depend on it.
8278 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
8279 at the destination location. If the argument is known to be aligned to
8280 some boundary, this can be specified as the fourth argument, otherwise
8281 it should be set to 0 or 1 (both meaning no alignment).
8283 '``llvm.sqrt.*``' Intrinsic
8284 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8289 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
8290 floating point or vector of floating point type. Not all targets support
8295 declare float @llvm.sqrt.f32(float %Val)
8296 declare double @llvm.sqrt.f64(double %Val)
8297 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
8298 declare fp128 @llvm.sqrt.f128(fp128 %Val)
8299 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
8304 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
8305 returning the same value as the libm '``sqrt``' functions would. Unlike
8306 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
8307 negative numbers other than -0.0 (which allows for better optimization,
8308 because there is no need to worry about errno being set).
8309 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
8314 The argument and return value are floating point numbers of the same
8320 This function returns the sqrt of the specified operand if it is a
8321 nonnegative floating point number.
8323 '``llvm.powi.*``' Intrinsic
8324 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8329 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
8330 floating point or vector of floating point type. Not all targets support
8335 declare float @llvm.powi.f32(float %Val, i32 %power)
8336 declare double @llvm.powi.f64(double %Val, i32 %power)
8337 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
8338 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
8339 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
8344 The '``llvm.powi.*``' intrinsics return the first operand raised to the
8345 specified (positive or negative) power. The order of evaluation of
8346 multiplications is not defined. When a vector of floating point type is
8347 used, the second argument remains a scalar integer value.
8352 The second argument is an integer power, and the first is a value to
8353 raise to that power.
8358 This function returns the first value raised to the second power with an
8359 unspecified sequence of rounding operations.
8361 '``llvm.sin.*``' Intrinsic
8362 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8367 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
8368 floating point or vector of floating point type. Not all targets support
8373 declare float @llvm.sin.f32(float %Val)
8374 declare double @llvm.sin.f64(double %Val)
8375 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
8376 declare fp128 @llvm.sin.f128(fp128 %Val)
8377 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
8382 The '``llvm.sin.*``' intrinsics return the sine of the operand.
8387 The argument and return value are floating point numbers of the same
8393 This function returns the sine of the specified operand, returning the
8394 same values as the libm ``sin`` functions would, and handles error
8395 conditions in the same way.
8397 '``llvm.cos.*``' Intrinsic
8398 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8403 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
8404 floating point or vector of floating point type. Not all targets support
8409 declare float @llvm.cos.f32(float %Val)
8410 declare double @llvm.cos.f64(double %Val)
8411 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
8412 declare fp128 @llvm.cos.f128(fp128 %Val)
8413 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
8418 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
8423 The argument and return value are floating point numbers of the same
8429 This function returns the cosine of the specified operand, returning the
8430 same values as the libm ``cos`` functions would, and handles error
8431 conditions in the same way.
8433 '``llvm.pow.*``' Intrinsic
8434 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8439 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
8440 floating point or vector of floating point type. Not all targets support
8445 declare float @llvm.pow.f32(float %Val, float %Power)
8446 declare double @llvm.pow.f64(double %Val, double %Power)
8447 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
8448 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
8449 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
8454 The '``llvm.pow.*``' intrinsics return the first operand raised to the
8455 specified (positive or negative) power.
8460 The second argument is a floating point power, and the first is a value
8461 to raise to that power.
8466 This function returns the first value raised to the second power,
8467 returning the same values as the libm ``pow`` functions would, and
8468 handles error conditions in the same way.
8470 '``llvm.exp.*``' Intrinsic
8471 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8476 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
8477 floating point or vector of floating point type. Not all targets support
8482 declare float @llvm.exp.f32(float %Val)
8483 declare double @llvm.exp.f64(double %Val)
8484 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
8485 declare fp128 @llvm.exp.f128(fp128 %Val)
8486 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
8491 The '``llvm.exp.*``' intrinsics perform the exp function.
8496 The argument and return value are floating point numbers of the same
8502 This function returns the same values as the libm ``exp`` functions
8503 would, and handles error conditions in the same way.
8505 '``llvm.exp2.*``' Intrinsic
8506 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8511 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
8512 floating point or vector of floating point type. Not all targets support
8517 declare float @llvm.exp2.f32(float %Val)
8518 declare double @llvm.exp2.f64(double %Val)
8519 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
8520 declare fp128 @llvm.exp2.f128(fp128 %Val)
8521 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
8526 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
8531 The argument and return value are floating point numbers of the same
8537 This function returns the same values as the libm ``exp2`` functions
8538 would, and handles error conditions in the same way.
8540 '``llvm.log.*``' Intrinsic
8541 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8546 This is an overloaded intrinsic. You can use ``llvm.log`` on any
8547 floating point or vector of floating point type. Not all targets support
8552 declare float @llvm.log.f32(float %Val)
8553 declare double @llvm.log.f64(double %Val)
8554 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
8555 declare fp128 @llvm.log.f128(fp128 %Val)
8556 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
8561 The '``llvm.log.*``' intrinsics perform the log function.
8566 The argument and return value are floating point numbers of the same
8572 This function returns the same values as the libm ``log`` functions
8573 would, and handles error conditions in the same way.
8575 '``llvm.log10.*``' Intrinsic
8576 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8581 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
8582 floating point or vector of floating point type. Not all targets support
8587 declare float @llvm.log10.f32(float %Val)
8588 declare double @llvm.log10.f64(double %Val)
8589 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
8590 declare fp128 @llvm.log10.f128(fp128 %Val)
8591 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
8596 The '``llvm.log10.*``' intrinsics perform the log10 function.
8601 The argument and return value are floating point numbers of the same
8607 This function returns the same values as the libm ``log10`` functions
8608 would, and handles error conditions in the same way.
8610 '``llvm.log2.*``' Intrinsic
8611 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8616 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
8617 floating point or vector of floating point type. Not all targets support
8622 declare float @llvm.log2.f32(float %Val)
8623 declare double @llvm.log2.f64(double %Val)
8624 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
8625 declare fp128 @llvm.log2.f128(fp128 %Val)
8626 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
8631 The '``llvm.log2.*``' intrinsics perform the log2 function.
8636 The argument and return value are floating point numbers of the same
8642 This function returns the same values as the libm ``log2`` functions
8643 would, and handles error conditions in the same way.
8645 '``llvm.fma.*``' Intrinsic
8646 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8651 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
8652 floating point or vector of floating point type. Not all targets support
8657 declare float @llvm.fma.f32(float %a, float %b, float %c)
8658 declare double @llvm.fma.f64(double %a, double %b, double %c)
8659 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
8660 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
8661 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
8666 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
8672 The argument and return value are floating point numbers of the same
8678 This function returns the same values as the libm ``fma`` functions
8679 would, and does not set errno.
8681 '``llvm.fabs.*``' Intrinsic
8682 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8687 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
8688 floating point or vector of floating point type. Not all targets support
8693 declare float @llvm.fabs.f32(float %Val)
8694 declare double @llvm.fabs.f64(double %Val)
8695 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
8696 declare fp128 @llvm.fabs.f128(fp128 %Val)
8697 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
8702 The '``llvm.fabs.*``' intrinsics return the absolute value of the
8708 The argument and return value are floating point numbers of the same
8714 This function returns the same values as the libm ``fabs`` functions
8715 would, and handles error conditions in the same way.
8717 '``llvm.minnum.*``' Intrinsic
8718 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8723 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
8724 floating point or vector of floating point type. Not all targets support
8729 declare float @llvm.minnum.f32(float %Val0, float %Val1)
8730 declare double @llvm.minnum.f64(double %Val0, double %Val1)
8731 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8732 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
8733 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8738 The '``llvm.minnum.*``' intrinsics return the minimum of the two
8745 The arguments and return value are floating point numbers of the same
8751 Follows the IEEE-754 semantics for minNum, which also match for libm's
8754 If either operand is a NaN, returns the other non-NaN operand. Returns
8755 NaN only if both operands are NaN. If the operands compare equal,
8756 returns a value that compares equal to both operands. This means that
8757 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8759 '``llvm.maxnum.*``' Intrinsic
8760 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8765 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
8766 floating point or vector of floating point type. Not all targets support
8771 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
8772 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
8773 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8774 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
8775 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8780 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
8787 The arguments and return value are floating point numbers of the same
8792 Follows the IEEE-754 semantics for maxNum, which also match for libm's
8795 If either operand is a NaN, returns the other non-NaN operand. Returns
8796 NaN only if both operands are NaN. If the operands compare equal,
8797 returns a value that compares equal to both operands. This means that
8798 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8800 '``llvm.copysign.*``' Intrinsic
8801 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8806 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
8807 floating point or vector of floating point type. Not all targets support
8812 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
8813 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
8814 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
8815 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
8816 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
8821 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
8822 first operand and the sign of the second operand.
8827 The arguments and return value are floating point numbers of the same
8833 This function returns the same values as the libm ``copysign``
8834 functions would, and handles error conditions in the same way.
8836 '``llvm.floor.*``' Intrinsic
8837 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8842 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
8843 floating point or vector of floating point type. Not all targets support
8848 declare float @llvm.floor.f32(float %Val)
8849 declare double @llvm.floor.f64(double %Val)
8850 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
8851 declare fp128 @llvm.floor.f128(fp128 %Val)
8852 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
8857 The '``llvm.floor.*``' intrinsics return the floor of the operand.
8862 The argument and return value are floating point numbers of the same
8868 This function returns the same values as the libm ``floor`` functions
8869 would, and handles error conditions in the same way.
8871 '``llvm.ceil.*``' Intrinsic
8872 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8877 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
8878 floating point or vector of floating point type. Not all targets support
8883 declare float @llvm.ceil.f32(float %Val)
8884 declare double @llvm.ceil.f64(double %Val)
8885 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
8886 declare fp128 @llvm.ceil.f128(fp128 %Val)
8887 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
8892 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
8897 The argument and return value are floating point numbers of the same
8903 This function returns the same values as the libm ``ceil`` functions
8904 would, and handles error conditions in the same way.
8906 '``llvm.trunc.*``' Intrinsic
8907 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8912 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
8913 floating point or vector of floating point type. Not all targets support
8918 declare float @llvm.trunc.f32(float %Val)
8919 declare double @llvm.trunc.f64(double %Val)
8920 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
8921 declare fp128 @llvm.trunc.f128(fp128 %Val)
8922 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
8927 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
8928 nearest integer not larger in magnitude than the operand.
8933 The argument and return value are floating point numbers of the same
8939 This function returns the same values as the libm ``trunc`` functions
8940 would, and handles error conditions in the same way.
8942 '``llvm.rint.*``' Intrinsic
8943 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8948 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
8949 floating point or vector of floating point type. Not all targets support
8954 declare float @llvm.rint.f32(float %Val)
8955 declare double @llvm.rint.f64(double %Val)
8956 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
8957 declare fp128 @llvm.rint.f128(fp128 %Val)
8958 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
8963 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
8964 nearest integer. It may raise an inexact floating-point exception if the
8965 operand isn't an integer.
8970 The argument and return value are floating point numbers of the same
8976 This function returns the same values as the libm ``rint`` functions
8977 would, and handles error conditions in the same way.
8979 '``llvm.nearbyint.*``' Intrinsic
8980 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8985 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
8986 floating point or vector of floating point type. Not all targets support
8991 declare float @llvm.nearbyint.f32(float %Val)
8992 declare double @llvm.nearbyint.f64(double %Val)
8993 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
8994 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
8995 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
9000 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
9006 The argument and return value are floating point numbers of the same
9012 This function returns the same values as the libm ``nearbyint``
9013 functions would, and handles error conditions in the same way.
9015 '``llvm.round.*``' Intrinsic
9016 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9021 This is an overloaded intrinsic. You can use ``llvm.round`` on any
9022 floating point or vector of floating point type. Not all targets support
9027 declare float @llvm.round.f32(float %Val)
9028 declare double @llvm.round.f64(double %Val)
9029 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
9030 declare fp128 @llvm.round.f128(fp128 %Val)
9031 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
9036 The '``llvm.round.*``' intrinsics returns the operand rounded to the
9042 The argument and return value are floating point numbers of the same
9048 This function returns the same values as the libm ``round``
9049 functions would, and handles error conditions in the same way.
9051 Bit Manipulation Intrinsics
9052 ---------------------------
9054 LLVM provides intrinsics for a few important bit manipulation
9055 operations. These allow efficient code generation for some algorithms.
9057 '``llvm.bswap.*``' Intrinsics
9058 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9063 This is an overloaded intrinsic function. You can use bswap on any
9064 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
9068 declare i16 @llvm.bswap.i16(i16 <id>)
9069 declare i32 @llvm.bswap.i32(i32 <id>)
9070 declare i64 @llvm.bswap.i64(i64 <id>)
9075 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
9076 values with an even number of bytes (positive multiple of 16 bits).
9077 These are useful for performing operations on data that is not in the
9078 target's native byte order.
9083 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
9084 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
9085 intrinsic returns an i32 value that has the four bytes of the input i32
9086 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
9087 returned i32 will have its bytes in 3, 2, 1, 0 order. The
9088 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
9089 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
9092 '``llvm.ctpop.*``' Intrinsic
9093 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9098 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
9099 bit width, or on any vector with integer elements. Not all targets
9100 support all bit widths or vector types, however.
9104 declare i8 @llvm.ctpop.i8(i8 <src>)
9105 declare i16 @llvm.ctpop.i16(i16 <src>)
9106 declare i32 @llvm.ctpop.i32(i32 <src>)
9107 declare i64 @llvm.ctpop.i64(i64 <src>)
9108 declare i256 @llvm.ctpop.i256(i256 <src>)
9109 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
9114 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
9120 The only argument is the value to be counted. The argument may be of any
9121 integer type, or a vector with integer elements. The return type must
9122 match the argument type.
9127 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
9128 each element of a vector.
9130 '``llvm.ctlz.*``' Intrinsic
9131 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9136 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
9137 integer bit width, or any vector whose elements are integers. Not all
9138 targets support all bit widths or vector types, however.
9142 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
9143 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
9144 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
9145 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
9146 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
9147 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
9152 The '``llvm.ctlz``' family of intrinsic functions counts the number of
9153 leading zeros in a variable.
9158 The first argument is the value to be counted. This argument may be of
9159 any integer type, or a vector with integer element type. The return
9160 type must match the first argument type.
9162 The second argument must be a constant and is a flag to indicate whether
9163 the intrinsic should ensure that a zero as the first argument produces a
9164 defined result. Historically some architectures did not provide a
9165 defined result for zero values as efficiently, and many algorithms are
9166 now predicated on avoiding zero-value inputs.
9171 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
9172 zeros in a variable, or within each element of the vector. If
9173 ``src == 0`` then the result is the size in bits of the type of ``src``
9174 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
9175 ``llvm.ctlz(i32 2) = 30``.
9177 '``llvm.cttz.*``' Intrinsic
9178 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9183 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
9184 integer bit width, or any vector of integer elements. Not all targets
9185 support all bit widths or vector types, however.
9189 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
9190 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
9191 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
9192 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
9193 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
9194 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
9199 The '``llvm.cttz``' family of intrinsic functions counts the number of
9205 The first argument is the value to be counted. This argument may be of
9206 any integer type, or a vector with integer element type. The return
9207 type must match the first argument type.
9209 The second argument must be a constant and is a flag to indicate whether
9210 the intrinsic should ensure that a zero as the first argument produces a
9211 defined result. Historically some architectures did not provide a
9212 defined result for zero values as efficiently, and many algorithms are
9213 now predicated on avoiding zero-value inputs.
9218 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
9219 zeros in a variable, or within each element of a vector. If ``src == 0``
9220 then the result is the size in bits of the type of ``src`` if
9221 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
9222 ``llvm.cttz(2) = 1``.
9226 Arithmetic with Overflow Intrinsics
9227 -----------------------------------
9229 LLVM provides intrinsics for some arithmetic with overflow operations.
9231 '``llvm.sadd.with.overflow.*``' Intrinsics
9232 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9237 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
9238 on any integer bit width.
9242 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
9243 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
9244 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
9249 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
9250 a signed addition of the two arguments, and indicate whether an overflow
9251 occurred during the signed summation.
9256 The arguments (%a and %b) and the first element of the result structure
9257 may be of integer types of any bit width, but they must have the same
9258 bit width. The second element of the result structure must be of type
9259 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9265 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
9266 a signed addition of the two variables. They return a structure --- the
9267 first element of which is the signed summation, and the second element
9268 of which is a bit specifying if the signed summation resulted in an
9274 .. code-block:: llvm
9276 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
9277 %sum = extractvalue {i32, i1} %res, 0
9278 %obit = extractvalue {i32, i1} %res, 1
9279 br i1 %obit, label %overflow, label %normal
9281 '``llvm.uadd.with.overflow.*``' Intrinsics
9282 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9287 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
9288 on any integer bit width.
9292 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
9293 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
9294 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
9299 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
9300 an unsigned addition of the two arguments, and indicate whether a carry
9301 occurred during the unsigned summation.
9306 The arguments (%a and %b) and the first element of the result structure
9307 may be of integer types of any bit width, but they must have the same
9308 bit width. The second element of the result structure must be of type
9309 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9315 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
9316 an unsigned addition of the two arguments. They return a structure --- the
9317 first element of which is the sum, and the second element of which is a
9318 bit specifying if the unsigned summation resulted in a carry.
9323 .. code-block:: llvm
9325 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
9326 %sum = extractvalue {i32, i1} %res, 0
9327 %obit = extractvalue {i32, i1} %res, 1
9328 br i1 %obit, label %carry, label %normal
9330 '``llvm.ssub.with.overflow.*``' Intrinsics
9331 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9336 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
9337 on any integer bit width.
9341 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
9342 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
9343 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
9348 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
9349 a signed subtraction of the two arguments, and indicate whether an
9350 overflow occurred during the signed subtraction.
9355 The arguments (%a and %b) and the first element of the result structure
9356 may be of integer types of any bit width, but they must have the same
9357 bit width. The second element of the result structure must be of type
9358 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9364 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
9365 a signed subtraction of the two arguments. They return a structure --- the
9366 first element of which is the subtraction, and the second element of
9367 which is a bit specifying if the signed subtraction resulted in an
9373 .. code-block:: llvm
9375 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
9376 %sum = extractvalue {i32, i1} %res, 0
9377 %obit = extractvalue {i32, i1} %res, 1
9378 br i1 %obit, label %overflow, label %normal
9380 '``llvm.usub.with.overflow.*``' Intrinsics
9381 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9386 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
9387 on any integer bit width.
9391 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
9392 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
9393 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
9398 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
9399 an unsigned subtraction of the two arguments, and indicate whether an
9400 overflow occurred during the unsigned subtraction.
9405 The arguments (%a and %b) and the first element of the result structure
9406 may be of integer types of any bit width, but they must have the same
9407 bit width. The second element of the result structure must be of type
9408 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9414 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
9415 an unsigned subtraction of the two arguments. They return a structure ---
9416 the first element of which is the subtraction, and the second element of
9417 which is a bit specifying if the unsigned subtraction resulted in an
9423 .. code-block:: llvm
9425 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
9426 %sum = extractvalue {i32, i1} %res, 0
9427 %obit = extractvalue {i32, i1} %res, 1
9428 br i1 %obit, label %overflow, label %normal
9430 '``llvm.smul.with.overflow.*``' Intrinsics
9431 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9436 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
9437 on any integer bit width.
9441 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
9442 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
9443 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
9448 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
9449 a signed multiplication of the two arguments, and indicate whether an
9450 overflow occurred during the signed multiplication.
9455 The arguments (%a and %b) and the first element of the result structure
9456 may be of integer types of any bit width, but they must have the same
9457 bit width. The second element of the result structure must be of type
9458 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9464 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
9465 a signed multiplication of the two arguments. They return a structure ---
9466 the first element of which is the multiplication, and the second element
9467 of which is a bit specifying if the signed multiplication resulted in an
9473 .. code-block:: llvm
9475 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
9476 %sum = extractvalue {i32, i1} %res, 0
9477 %obit = extractvalue {i32, i1} %res, 1
9478 br i1 %obit, label %overflow, label %normal
9480 '``llvm.umul.with.overflow.*``' Intrinsics
9481 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9486 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
9487 on any integer bit width.
9491 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
9492 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
9493 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
9498 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
9499 a unsigned multiplication of the two arguments, and indicate whether an
9500 overflow occurred during the unsigned multiplication.
9505 The arguments (%a and %b) and the first element of the result structure
9506 may be of integer types of any bit width, but they must have the same
9507 bit width. The second element of the result structure must be of type
9508 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9514 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
9515 an unsigned multiplication of the two arguments. They return a structure ---
9516 the first element of which is the multiplication, and the second
9517 element of which is a bit specifying if the unsigned multiplication
9518 resulted in an overflow.
9523 .. code-block:: llvm
9525 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
9526 %sum = extractvalue {i32, i1} %res, 0
9527 %obit = extractvalue {i32, i1} %res, 1
9528 br i1 %obit, label %overflow, label %normal
9530 Specialised Arithmetic Intrinsics
9531 ---------------------------------
9533 '``llvm.fmuladd.*``' Intrinsic
9534 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9541 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
9542 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
9547 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
9548 expressions that can be fused if the code generator determines that (a) the
9549 target instruction set has support for a fused operation, and (b) that the
9550 fused operation is more efficient than the equivalent, separate pair of mul
9551 and add instructions.
9556 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
9557 multiplicands, a and b, and an addend c.
9566 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
9568 is equivalent to the expression a \* b + c, except that rounding will
9569 not be performed between the multiplication and addition steps if the
9570 code generator fuses the operations. Fusion is not guaranteed, even if
9571 the target platform supports it. If a fused multiply-add is required the
9572 corresponding llvm.fma.\* intrinsic function should be used
9573 instead. This never sets errno, just as '``llvm.fma.*``'.
9578 .. code-block:: llvm
9580 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
9582 Half Precision Floating Point Intrinsics
9583 ----------------------------------------
9585 For most target platforms, half precision floating point is a
9586 storage-only format. This means that it is a dense encoding (in memory)
9587 but does not support computation in the format.
9589 This means that code must first load the half-precision floating point
9590 value as an i16, then convert it to float with
9591 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
9592 then be performed on the float value (including extending to double
9593 etc). To store the value back to memory, it is first converted to float
9594 if needed, then converted to i16 with
9595 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
9598 .. _int_convert_to_fp16:
9600 '``llvm.convert.to.fp16``' Intrinsic
9601 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9608 declare i16 @llvm.convert.to.fp16.f32(float %a)
9609 declare i16 @llvm.convert.to.fp16.f64(double %a)
9614 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
9615 conventional floating point type to half precision floating point format.
9620 The intrinsic function contains single argument - the value to be
9626 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
9627 conventional floating point format to half precision floating point format. The
9628 return value is an ``i16`` which contains the converted number.
9633 .. code-block:: llvm
9635 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
9636 store i16 %res, i16* @x, align 2
9638 .. _int_convert_from_fp16:
9640 '``llvm.convert.from.fp16``' Intrinsic
9641 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9648 declare float @llvm.convert.from.fp16.f32(i16 %a)
9649 declare double @llvm.convert.from.fp16.f64(i16 %a)
9654 The '``llvm.convert.from.fp16``' intrinsic function performs a
9655 conversion from half precision floating point format to single precision
9656 floating point format.
9661 The intrinsic function contains single argument - the value to be
9667 The '``llvm.convert.from.fp16``' intrinsic function performs a
9668 conversion from half single precision floating point format to single
9669 precision floating point format. The input half-float value is
9670 represented by an ``i16`` value.
9675 .. code-block:: llvm
9677 %a = load i16, i16* @x, align 2
9678 %res = call float @llvm.convert.from.fp16(i16 %a)
9685 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
9686 prefix), are described in the `LLVM Source Level
9687 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
9690 Exception Handling Intrinsics
9691 -----------------------------
9693 The LLVM exception handling intrinsics (which all start with
9694 ``llvm.eh.`` prefix), are described in the `LLVM Exception
9695 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
9699 Trampoline Intrinsics
9700 ---------------------
9702 These intrinsics make it possible to excise one parameter, marked with
9703 the :ref:`nest <nest>` attribute, from a function. The result is a
9704 callable function pointer lacking the nest parameter - the caller does
9705 not need to provide a value for it. Instead, the value to use is stored
9706 in advance in a "trampoline", a block of memory usually allocated on the
9707 stack, which also contains code to splice the nest value into the
9708 argument list. This is used to implement the GCC nested function address
9711 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
9712 then the resulting function pointer has signature ``i32 (i32, i32)*``.
9713 It can be created as follows:
9715 .. code-block:: llvm
9717 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
9718 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
9719 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
9720 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
9721 %fp = bitcast i8* %p to i32 (i32, i32)*
9723 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
9724 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
9728 '``llvm.init.trampoline``' Intrinsic
9729 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9736 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
9741 This fills the memory pointed to by ``tramp`` with executable code,
9742 turning it into a trampoline.
9747 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
9748 pointers. The ``tramp`` argument must point to a sufficiently large and
9749 sufficiently aligned block of memory; this memory is written to by the
9750 intrinsic. Note that the size and the alignment are target-specific -
9751 LLVM currently provides no portable way of determining them, so a
9752 front-end that generates this intrinsic needs to have some
9753 target-specific knowledge. The ``func`` argument must hold a function
9754 bitcast to an ``i8*``.
9759 The block of memory pointed to by ``tramp`` is filled with target
9760 dependent code, turning it into a function. Then ``tramp`` needs to be
9761 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
9762 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
9763 function's signature is the same as that of ``func`` with any arguments
9764 marked with the ``nest`` attribute removed. At most one such ``nest``
9765 argument is allowed, and it must be of pointer type. Calling the new
9766 function is equivalent to calling ``func`` with the same argument list,
9767 but with ``nval`` used for the missing ``nest`` argument. If, after
9768 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
9769 modified, then the effect of any later call to the returned function
9770 pointer is undefined.
9774 '``llvm.adjust.trampoline``' Intrinsic
9775 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9782 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
9787 This performs any required machine-specific adjustment to the address of
9788 a trampoline (passed as ``tramp``).
9793 ``tramp`` must point to a block of memory which already has trampoline
9794 code filled in by a previous call to
9795 :ref:`llvm.init.trampoline <int_it>`.
9800 On some architectures the address of the code to be executed needs to be
9801 different than the address where the trampoline is actually stored. This
9802 intrinsic returns the executable address corresponding to ``tramp``
9803 after performing the required machine specific adjustments. The pointer
9804 returned can then be :ref:`bitcast and executed <int_trampoline>`.
9806 .. _int_mload_mstore:
9808 Masked Vector Load and Store Intrinsics
9809 ---------------------------------------
9811 LLVM provides intrinsics for predicated vector load and store operations. The predicate is specified by a mask operand, which holds one bit per vector element, switching the associated vector lane on or off. The memory addresses corresponding to the "off" lanes are not accessed. When all bits of the mask are on, the intrinsic is identical to a regular vector load or store. When all bits are off, no memory is accessed.
9815 '``llvm.masked.load.*``' Intrinsics
9816 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9820 This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
9824 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
9825 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
9830 Reads a vector from memory according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes. The masked-off lanes in the result vector are taken from the corresponding lanes of the '``passthru``' operand.
9836 The first operand is the base pointer for the load. The second operand is the alignment of the source location. It must be a constant integer value. The third operand, mask, is a vector of boolean values with the same number of elements as the return type. The fourth is a pass-through value that is used to fill the masked-off lanes of the result. The return type, underlying type of the base pointer and the type of the '``passthru``' operand are the same vector types.
9842 The '``llvm.masked.load``' intrinsic is designed for conditional reading of selected vector elements in a single IR operation. It is useful for targets that support vector masked loads and allows vectorizing predicated basic blocks on these targets. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar load operations.
9843 The result of this operation is equivalent to a regular vector load instruction followed by a 'select' between the loaded and the passthru values, predicated on the same mask. However, using this intrinsic prevents exceptions on memory access to masked-off lanes.
9848 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
9850 ;; The result of the two following instructions is identical aside from potential memory access exception
9851 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
9852 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
9856 '``llvm.masked.store.*``' Intrinsics
9857 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9861 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
9865 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
9866 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
9871 Writes a vector to memory according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes.
9876 The first operand is the vector value to be written to memory. The second operand is the base pointer for the store, it has the same underlying type as the value operand. The third operand is the alignment of the destination location. The fourth operand, mask, is a vector of boolean values. The types of the mask and the value operand must have the same number of vector elements.
9882 The '``llvm.masked.store``' intrinsics is designed for conditional writing of selected vector elements in a single IR operation. It is useful for targets that support vector masked store and allows vectorizing predicated basic blocks on these targets. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar store operations.
9883 The result of this operation is equivalent to a load-modify-store sequence. However, using this intrinsic prevents exceptions and data races on memory access to masked-off lanes.
9887 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
9889 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
9890 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
9891 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
9892 store <16 x float> %res, <16 x float>* %ptr, align 4
9895 Masked Vector Gather and Scatter Intrinsics
9896 -------------------------------------------
9898 LLVM provides intrinsics for vector gather and scatter operations. They are similar to :ref:`Masked Vector Load and Store <int_mload_mstore>`, except they are designed for arbitrary memory accesses, rather than sequential memory accesses. Gather and scatter also employ a mask operand, which holds one bit per vector element, switching the associated vector lane on or off. The memory addresses corresponding to the "off" lanes are not accessed. When all bits are off, no memory is accessed.
9902 '``llvm.masked.gather.*``' Intrinsics
9903 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9907 This is an overloaded intrinsic. The loaded data are multiple scalar values of any integer or floating point data type gathered together into one vector.
9911 declare <16 x float> @llvm.masked.gather.v16f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
9912 declare <2 x double> @llvm.masked.gather.v2f64 (<2 x double*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
9917 Reads scalar values from arbitrary memory locations and gathers them into one vector. The memory locations are provided in the vector of pointers '``ptrs``'. The memory is accessed according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes. The masked-off lanes in the result vector are taken from the corresponding lanes of the '``passthru``' operand.
9923 The first operand is a vector of pointers which holds all memory addresses to read. The second operand is an alignment of the source addresses. It must be a constant integer value. The third operand, mask, is a vector of boolean values with the same number of elements as the return type. The fourth is a pass-through value that is used to fill the masked-off lanes of the result. The return type, underlying type of the vector of pointers and the type of the '``passthru``' operand are the same vector types.
9929 The '``llvm.masked.gather``' intrinsic is designed for conditional reading of multiple scalar values from arbitrary memory locations in a single IR operation. It is useful for targets that support vector masked gathers and allows vectorizing basic blocks with data and control divergence. Other targets may support this intrinsic differently, for example by lowering it into a sequence of scalar load operations.
9930 The semantics of this operation are equivalent to a sequence of conditional scalar loads with subsequent gathering all loaded values into a single vector. The mask restricts memory access to certain lanes and facilitates vectorization of predicated basic blocks.
9935 %res = call <4 x double> @llvm.masked.gather.v4f64 (<4 x double*> %ptrs, i32 8, <4 x i1>%mask, <4 x double> <true, true, true, true>)
9937 ;; The gather with all-true mask is equivalent to the following instruction sequence
9938 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
9939 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
9940 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
9941 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
9943 %val0 = load double, double* %ptr0, align 8
9944 %val1 = load double, double* %ptr1, align 8
9945 %val2 = load double, double* %ptr2, align 8
9946 %val3 = load double, double* %ptr3, align 8
9948 %vec0 = insertelement <4 x double>undef, %val0, 0
9949 %vec01 = insertelement <4 x double>%vec0, %val1, 1
9950 %vec012 = insertelement <4 x double>%vec01, %val2, 2
9951 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
9955 '``llvm.masked.scatter.*``' Intrinsics
9956 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9960 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type. Each vector element is stored in an arbitrary memory addresses. Scatter with overlapping addresses is guaranteed to be ordered from least-significant to most-significant element.
9964 declare void @llvm.masked.scatter.v8i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
9965 declare void @llvm.masked.scatter.v16f32(<16 x i32> <value>, <16 x i32*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
9970 Writes each element from the value vector to the corresponding memory address. The memory addresses are represented as a vector of pointers. Writing is done according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes.
9975 The first operand is a vector value to be written to memory. The second operand is a vector of pointers, pointing to where the value elements should be stored. It has the same underlying type as the value operand. The third operand is an alignment of the destination addresses. The fourth operand, mask, is a vector of boolean values. The types of the mask and the value operand must have the same number of vector elements.
9981 The '``llvm.masked.scatter``' intrinsics is designed for writing selected vector elements to arbitrary memory addresses in a single IR operation. The operation may be conditional, when not all bits in the mask are switched on. It is useful for targets that support vector masked scatter and allows vectorizing basic blocks with data and control divergency. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar store operations.
9985 ;; This instruction unconditionaly stores data vector in multiple addresses
9986 call @llvm.masked.scatter.v8i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
9988 ;; It is equivalent to a list of scalar stores
9989 %val0 = extractelement <8 x i32> %value, i32 0
9990 %val1 = extractelement <8 x i32> %value, i32 1
9992 %val7 = extractelement <8 x i32> %value, i32 7
9993 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
9994 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
9996 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
9997 ;; Note: the order of the following stores is important when they overlap:
9998 store i32 %val0, i32* %ptr0, align 4
9999 store i32 %val1, i32* %ptr1, align 4
10001 store i32 %val7, i32* %ptr7, align 4
10007 This class of intrinsics provides information about the lifetime of
10008 memory objects and ranges where variables are immutable.
10012 '``llvm.lifetime.start``' Intrinsic
10013 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10020 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
10025 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
10031 The first argument is a constant integer representing the size of the
10032 object, or -1 if it is variable sized. The second argument is a pointer
10038 This intrinsic indicates that before this point in the code, the value
10039 of the memory pointed to by ``ptr`` is dead. This means that it is known
10040 to never be used and has an undefined value. A load from the pointer
10041 that precedes this intrinsic can be replaced with ``'undef'``.
10045 '``llvm.lifetime.end``' Intrinsic
10046 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10053 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
10058 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
10064 The first argument is a constant integer representing the size of the
10065 object, or -1 if it is variable sized. The second argument is a pointer
10071 This intrinsic indicates that after this point in the code, the value of
10072 the memory pointed to by ``ptr`` is dead. This means that it is known to
10073 never be used and has an undefined value. Any stores into the memory
10074 object following this intrinsic may be removed as dead.
10076 '``llvm.invariant.start``' Intrinsic
10077 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10084 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
10089 The '``llvm.invariant.start``' intrinsic specifies that the contents of
10090 a memory object will not change.
10095 The first argument is a constant integer representing the size of the
10096 object, or -1 if it is variable sized. The second argument is a pointer
10102 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
10103 the return value, the referenced memory location is constant and
10106 '``llvm.invariant.end``' Intrinsic
10107 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10114 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
10119 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
10120 memory object are mutable.
10125 The first argument is the matching ``llvm.invariant.start`` intrinsic.
10126 The second argument is a constant integer representing the size of the
10127 object, or -1 if it is variable sized and the third argument is a
10128 pointer to the object.
10133 This intrinsic indicates that the memory is mutable again.
10138 This class of intrinsics is designed to be generic and has no specific
10141 '``llvm.var.annotation``' Intrinsic
10142 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10149 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
10154 The '``llvm.var.annotation``' intrinsic.
10159 The first argument is a pointer to a value, the second is a pointer to a
10160 global string, the third is a pointer to a global string which is the
10161 source file name, and the last argument is the line number.
10166 This intrinsic allows annotation of local variables with arbitrary
10167 strings. This can be useful for special purpose optimizations that want
10168 to look for these annotations. These have no other defined use; they are
10169 ignored by code generation and optimization.
10171 '``llvm.ptr.annotation.*``' Intrinsic
10172 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10177 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
10178 pointer to an integer of any width. *NOTE* you must specify an address space for
10179 the pointer. The identifier for the default address space is the integer
10184 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
10185 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
10186 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
10187 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
10188 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
10193 The '``llvm.ptr.annotation``' intrinsic.
10198 The first argument is a pointer to an integer value of arbitrary bitwidth
10199 (result of some expression), the second is a pointer to a global string, the
10200 third is a pointer to a global string which is the source file name, and the
10201 last argument is the line number. It returns the value of the first argument.
10206 This intrinsic allows annotation of a pointer to an integer with arbitrary
10207 strings. This can be useful for special purpose optimizations that want to look
10208 for these annotations. These have no other defined use; they are ignored by code
10209 generation and optimization.
10211 '``llvm.annotation.*``' Intrinsic
10212 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10217 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
10218 any integer bit width.
10222 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
10223 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
10224 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
10225 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
10226 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
10231 The '``llvm.annotation``' intrinsic.
10236 The first argument is an integer value (result of some expression), the
10237 second is a pointer to a global string, the third is a pointer to a
10238 global string which is the source file name, and the last argument is
10239 the line number. It returns the value of the first argument.
10244 This intrinsic allows annotations to be put on arbitrary expressions
10245 with arbitrary strings. This can be useful for special purpose
10246 optimizations that want to look for these annotations. These have no
10247 other defined use; they are ignored by code generation and optimization.
10249 '``llvm.trap``' Intrinsic
10250 ^^^^^^^^^^^^^^^^^^^^^^^^^
10257 declare void @llvm.trap() noreturn nounwind
10262 The '``llvm.trap``' intrinsic.
10272 This intrinsic is lowered to the target dependent trap instruction. If
10273 the target does not have a trap instruction, this intrinsic will be
10274 lowered to a call of the ``abort()`` function.
10276 '``llvm.debugtrap``' Intrinsic
10277 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10284 declare void @llvm.debugtrap() nounwind
10289 The '``llvm.debugtrap``' intrinsic.
10299 This intrinsic is lowered to code which is intended to cause an
10300 execution trap with the intention of requesting the attention of a
10303 '``llvm.stackprotector``' Intrinsic
10304 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10311 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
10316 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
10317 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
10318 is placed on the stack before local variables.
10323 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
10324 The first argument is the value loaded from the stack guard
10325 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
10326 enough space to hold the value of the guard.
10331 This intrinsic causes the prologue/epilogue inserter to force the position of
10332 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
10333 to ensure that if a local variable on the stack is overwritten, it will destroy
10334 the value of the guard. When the function exits, the guard on the stack is
10335 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
10336 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
10337 calling the ``__stack_chk_fail()`` function.
10339 '``llvm.stackprotectorcheck``' Intrinsic
10340 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10347 declare void @llvm.stackprotectorcheck(i8** <guard>)
10352 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
10353 created stack protector and if they are not equal calls the
10354 ``__stack_chk_fail()`` function.
10359 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
10360 the variable ``@__stack_chk_guard``.
10365 This intrinsic is provided to perform the stack protector check by comparing
10366 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
10367 values do not match call the ``__stack_chk_fail()`` function.
10369 The reason to provide this as an IR level intrinsic instead of implementing it
10370 via other IR operations is that in order to perform this operation at the IR
10371 level without an intrinsic, one would need to create additional basic blocks to
10372 handle the success/failure cases. This makes it difficult to stop the stack
10373 protector check from disrupting sibling tail calls in Codegen. With this
10374 intrinsic, we are able to generate the stack protector basic blocks late in
10375 codegen after the tail call decision has occurred.
10377 '``llvm.objectsize``' Intrinsic
10378 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10385 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
10386 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
10391 The ``llvm.objectsize`` intrinsic is designed to provide information to
10392 the optimizers to determine at compile time whether a) an operation
10393 (like memcpy) will overflow a buffer that corresponds to an object, or
10394 b) that a runtime check for overflow isn't necessary. An object in this
10395 context means an allocation of a specific class, structure, array, or
10401 The ``llvm.objectsize`` intrinsic takes two arguments. The first
10402 argument is a pointer to or into the ``object``. The second argument is
10403 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
10404 or -1 (if false) when the object size is unknown. The second argument
10405 only accepts constants.
10410 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
10411 the size of the object concerned. If the size cannot be determined at
10412 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
10413 on the ``min`` argument).
10415 '``llvm.expect``' Intrinsic
10416 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10421 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
10426 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
10427 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
10428 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
10433 The ``llvm.expect`` intrinsic provides information about expected (the
10434 most probable) value of ``val``, which can be used by optimizers.
10439 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
10440 a value. The second argument is an expected value, this needs to be a
10441 constant value, variables are not allowed.
10446 This intrinsic is lowered to the ``val``.
10450 '``llvm.assume``' Intrinsic
10451 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10458 declare void @llvm.assume(i1 %cond)
10463 The ``llvm.assume`` allows the optimizer to assume that the provided
10464 condition is true. This information can then be used in simplifying other parts
10470 The condition which the optimizer may assume is always true.
10475 The intrinsic allows the optimizer to assume that the provided condition is
10476 always true whenever the control flow reaches the intrinsic call. No code is
10477 generated for this intrinsic, and instructions that contribute only to the
10478 provided condition are not used for code generation. If the condition is
10479 violated during execution, the behavior is undefined.
10481 Note that the optimizer might limit the transformations performed on values
10482 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
10483 only used to form the intrinsic's input argument. This might prove undesirable
10484 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
10485 sufficient overall improvement in code quality. For this reason,
10486 ``llvm.assume`` should not be used to document basic mathematical invariants
10487 that the optimizer can otherwise deduce or facts that are of little use to the
10492 '``llvm.bitset.test``' Intrinsic
10493 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10500 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
10506 The first argument is a pointer to be tested. The second argument is a
10507 metadata string containing the name of a :doc:`bitset <BitSets>`.
10512 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
10513 member of the given bitset.
10515 '``llvm.donothing``' Intrinsic
10516 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10523 declare void @llvm.donothing() nounwind readnone
10528 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
10529 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
10530 with an invoke instruction.
10540 This intrinsic does nothing, and it's removed by optimizers and ignored
10543 Stack Map Intrinsics
10544 --------------------
10546 LLVM provides experimental intrinsics to support runtime patching
10547 mechanisms commonly desired in dynamic language JITs. These intrinsics
10548 are described in :doc:`StackMaps`.