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 an alias, but then it only governs how
495 the alias is accessed. It will not have any effect in the aliasee.
497 For platforms without linker support of ELF TLS model, the -femulated-tls
498 flag can be used to generate GCC compatible emulated TLS code.
505 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
506 types <t_struct>`. Literal types are uniqued structurally, but identified types
507 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
508 to forward declare a type that is not yet available.
510 An example of an identified structure specification is:
514 %mytype = type { %mytype*, i32 }
516 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
517 literal types are uniqued in recent versions of LLVM.
524 Global variables define regions of memory allocated at compilation time
527 Global variable definitions must be initialized.
529 Global variables in other translation units can also be declared, in which
530 case they don't have an initializer.
532 Either global variable definitions or declarations may have an explicit section
533 to be placed in and may have an optional explicit alignment specified.
535 A variable may be defined as a global ``constant``, which indicates that
536 the contents of the variable will **never** be modified (enabling better
537 optimization, allowing the global data to be placed in the read-only
538 section of an executable, etc). Note that variables that need runtime
539 initialization cannot be marked ``constant`` as there is a store to the
542 LLVM explicitly allows *declarations* of global variables to be marked
543 constant, even if the final definition of the global is not. This
544 capability can be used to enable slightly better optimization of the
545 program, but requires the language definition to guarantee that
546 optimizations based on the 'constantness' are valid for the translation
547 units that do not include the definition.
549 As SSA values, global variables define pointer values that are in scope
550 (i.e. they dominate) all basic blocks in the program. Global variables
551 always define a pointer to their "content" type because they describe a
552 region of memory, and all memory objects in LLVM are accessed through
555 Global variables can be marked with ``unnamed_addr`` which indicates
556 that the address is not significant, only the content. Constants marked
557 like this can be merged with other constants if they have the same
558 initializer. Note that a constant with significant address *can* be
559 merged with a ``unnamed_addr`` constant, the result being a constant
560 whose address is significant.
562 A global variable may be declared to reside in a target-specific
563 numbered address space. For targets that support them, address spaces
564 may affect how optimizations are performed and/or what target
565 instructions are used to access the variable. The default address space
566 is zero. The address space qualifier must precede any other attributes.
568 LLVM allows an explicit section to be specified for globals. If the
569 target supports it, it will emit globals to the section specified.
570 Additionally, the global can placed in a comdat if the target has the necessary
573 By default, global initializers are optimized by assuming that global
574 variables defined within the module are not modified from their
575 initial values before the start of the global initializer. This is
576 true even for variables potentially accessible from outside the
577 module, including those with external linkage or appearing in
578 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
579 by marking the variable with ``externally_initialized``.
581 An explicit alignment may be specified for a global, which must be a
582 power of 2. If not present, or if the alignment is set to zero, the
583 alignment of the global is set by the target to whatever it feels
584 convenient. If an explicit alignment is specified, the global is forced
585 to have exactly that alignment. Targets and optimizers are not allowed
586 to over-align the global if the global has an assigned section. In this
587 case, the extra alignment could be observable: for example, code could
588 assume that the globals are densely packed in their section and try to
589 iterate over them as an array, alignment padding would break this
590 iteration. The maximum alignment is ``1 << 29``.
592 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
594 Variables and aliases can have a
595 :ref:`Thread Local Storage Model <tls_model>`.
599 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
600 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
601 <global | constant> <Type> [<InitializerConstant>]
602 [, section "name"] [, comdat [($name)]]
603 [, align <Alignment>]
605 For example, the following defines a global in a numbered address space
606 with an initializer, section, and alignment:
610 @G = addrspace(5) constant float 1.0, section "foo", align 4
612 The following example just declares a global variable
616 @G = external global i32
618 The following example defines a thread-local global with the
619 ``initialexec`` TLS model:
623 @G = thread_local(initialexec) global i32 0, align 4
625 .. _functionstructure:
630 LLVM function definitions consist of the "``define``" keyword, an
631 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
632 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
633 an optional :ref:`calling convention <callingconv>`,
634 an optional ``unnamed_addr`` attribute, a return type, an optional
635 :ref:`parameter attribute <paramattrs>` for the return type, a function
636 name, a (possibly empty) argument list (each with optional :ref:`parameter
637 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
638 an optional section, an optional alignment,
639 an optional :ref:`comdat <langref_comdats>`,
640 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
641 an optional :ref:`prologue <prologuedata>`,
642 an optional :ref:`personality <personalityfn>`,
643 an optional list of attached :ref:`metadata <metadata>`,
644 an opening curly brace, a list of basic blocks, and a closing curly brace.
646 LLVM function declarations consist of the "``declare``" keyword, an
647 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
648 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
649 an optional :ref:`calling convention <callingconv>`,
650 an optional ``unnamed_addr`` attribute, a return type, an optional
651 :ref:`parameter attribute <paramattrs>` for the return type, a function
652 name, a possibly empty list of arguments, an optional alignment, an optional
653 :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
654 and an optional :ref:`prologue <prologuedata>`.
656 A function definition contains a list of basic blocks, forming the CFG (Control
657 Flow Graph) for the function. Each basic block may optionally start with a label
658 (giving the basic block a symbol table entry), contains a list of instructions,
659 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
660 function return). If an explicit label is not provided, a block is assigned an
661 implicit numbered label, using the next value from the same counter as used for
662 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
663 entry block does not have an explicit label, it will be assigned label "%0",
664 then the first unnamed temporary in that block will be "%1", etc.
666 The first basic block in a function is special in two ways: it is
667 immediately executed on entrance to the function, and it is not allowed
668 to have predecessor basic blocks (i.e. there can not be any branches to
669 the entry block of a function). Because the block can have no
670 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
672 LLVM allows an explicit section to be specified for functions. If the
673 target supports it, it will emit functions to the section specified.
674 Additionally, the function can be placed in a COMDAT.
676 An explicit alignment may be specified for a function. If not present,
677 or if the alignment is set to zero, the alignment of the function is set
678 by the target to whatever it feels convenient. If an explicit alignment
679 is specified, the function is forced to have at least that much
680 alignment. All alignments must be a power of 2.
682 If the ``unnamed_addr`` attribute is given, the address is known to not
683 be significant and two identical functions can be merged.
687 define [linkage] [visibility] [DLLStorageClass]
689 <ResultType> @<FunctionName> ([argument list])
690 [unnamed_addr] [fn Attrs] [section "name"] [comdat [($name)]]
691 [align N] [gc] [prefix Constant] [prologue Constant]
692 [personality Constant] (!name !N)* { ... }
694 The argument list is a comma separated sequence of arguments where each
695 argument is of the following form:
699 <type> [parameter Attrs] [name]
707 Aliases, unlike function or variables, don't create any new data. They
708 are just a new symbol and metadata for an existing position.
710 Aliases have a name and an aliasee that is either a global value or a
713 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
714 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
715 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
719 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy>, <AliaseeTy>* @<Aliasee>
721 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
722 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
723 might not correctly handle dropping a weak symbol that is aliased.
725 Aliases that are not ``unnamed_addr`` are guaranteed to have the same address as
726 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
729 Since aliases are only a second name, some restrictions apply, of which
730 some can only be checked when producing an object file:
732 * The expression defining the aliasee must be computable at assembly
733 time. Since it is just a name, no relocations can be used.
735 * No alias in the expression can be weak as the possibility of the
736 intermediate alias being overridden cannot be represented in an
739 * No global value in the expression can be a declaration, since that
740 would require a relocation, which is not possible.
747 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
749 Comdats have a name which represents the COMDAT key. All global objects that
750 specify this key will only end up in the final object file if the linker chooses
751 that key over some other key. Aliases are placed in the same COMDAT that their
752 aliasee computes to, if any.
754 Comdats have a selection kind to provide input on how the linker should
755 choose between keys in two different object files.
759 $<Name> = comdat SelectionKind
761 The selection kind must be one of the following:
764 The linker may choose any COMDAT key, the choice is arbitrary.
766 The linker may choose any COMDAT key but the sections must contain the
769 The linker will choose the section containing the largest COMDAT key.
771 The linker requires that only section with this COMDAT key exist.
773 The linker may choose any COMDAT key but the sections must contain the
776 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
777 ``any`` as a selection kind.
779 Here is an example of a COMDAT group where a function will only be selected if
780 the COMDAT key's section is the largest:
784 $foo = comdat largest
785 @foo = global i32 2, comdat($foo)
787 define void @bar() comdat($foo) {
791 As a syntactic sugar the ``$name`` can be omitted if the name is the same as
797 @foo = global i32 2, comdat
800 In a COFF object file, this will create a COMDAT section with selection kind
801 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
802 and another COMDAT section with selection kind
803 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
804 section and contains the contents of the ``@bar`` symbol.
806 There are some restrictions on the properties of the global object.
807 It, or an alias to it, must have the same name as the COMDAT group when
809 The contents and size of this object may be used during link-time to determine
810 which COMDAT groups get selected depending on the selection kind.
811 Because the name of the object must match the name of the COMDAT group, the
812 linkage of the global object must not be local; local symbols can get renamed
813 if a collision occurs in the symbol table.
815 The combined use of COMDATS and section attributes may yield surprising results.
822 @g1 = global i32 42, section "sec", comdat($foo)
823 @g2 = global i32 42, section "sec", comdat($bar)
825 From the object file perspective, this requires the creation of two sections
826 with the same name. This is necessary because both globals belong to different
827 COMDAT groups and COMDATs, at the object file level, are represented by
830 Note that certain IR constructs like global variables and functions may
831 create COMDATs in the object file in addition to any which are specified using
832 COMDAT IR. This arises when the code generator is configured to emit globals
833 in individual sections (e.g. when `-data-sections` or `-function-sections`
834 is supplied to `llc`).
836 .. _namedmetadatastructure:
841 Named metadata is a collection of metadata. :ref:`Metadata
842 nodes <metadata>` (but not metadata strings) are the only valid
843 operands for a named metadata.
845 #. Named metadata are represented as a string of characters with the
846 metadata prefix. The rules for metadata names are the same as for
847 identifiers, but quoted names are not allowed. ``"\xx"`` type escapes
848 are still valid, which allows any character to be part of a name.
852 ; Some unnamed metadata nodes, which are referenced by the named metadata.
857 !name = !{!0, !1, !2}
864 The return type and each parameter of a function type may have a set of
865 *parameter attributes* associated with them. Parameter attributes are
866 used to communicate additional information about the result or
867 parameters of a function. Parameter attributes are considered to be part
868 of the function, not of the function type, so functions with different
869 parameter attributes can have the same function type.
871 Parameter attributes are simple keywords that follow the type specified.
872 If multiple parameter attributes are needed, they are space separated.
877 declare i32 @printf(i8* noalias nocapture, ...)
878 declare i32 @atoi(i8 zeroext)
879 declare signext i8 @returns_signed_char()
881 Note that any attributes for the function result (``nounwind``,
882 ``readonly``) come immediately after the argument list.
884 Currently, only the following parameter attributes are defined:
887 This indicates to the code generator that the parameter or return
888 value should be zero-extended to the extent required by the target's
889 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
890 the caller (for a parameter) or the callee (for a return value).
892 This indicates to the code generator that the parameter or return
893 value should be sign-extended to the extent required by the target's
894 ABI (which is usually 32-bits) by the caller (for a parameter) or
895 the callee (for a return value).
897 This indicates that this parameter or return value should be treated
898 in a special target-dependent fashion while emitting code for
899 a function call or return (usually, by putting it in a register as
900 opposed to memory, though some targets use it to distinguish between
901 two different kinds of registers). Use of this attribute is
904 This indicates that the pointer parameter should really be passed by
905 value to the function. The attribute implies that a hidden copy of
906 the pointee is made between the caller and the callee, so the callee
907 is unable to modify the value in the caller. This attribute is only
908 valid on LLVM pointer arguments. It is generally used to pass
909 structs and arrays by value, but is also valid on pointers to
910 scalars. The copy is considered to belong to the caller not the
911 callee (for example, ``readonly`` functions should not write to
912 ``byval`` parameters). This is not a valid attribute for return
915 The byval attribute also supports specifying an alignment with the
916 align attribute. It indicates the alignment of the stack slot to
917 form and the known alignment of the pointer specified to the call
918 site. If the alignment is not specified, then the code generator
919 makes a target-specific assumption.
925 The ``inalloca`` argument attribute allows the caller to take the
926 address of outgoing stack arguments. An ``inalloca`` argument must
927 be a pointer to stack memory produced by an ``alloca`` instruction.
928 The alloca, or argument allocation, must also be tagged with the
929 inalloca keyword. Only the last argument may have the ``inalloca``
930 attribute, and that argument is guaranteed to be passed in memory.
932 An argument allocation may be used by a call at most once because
933 the call may deallocate it. The ``inalloca`` attribute cannot be
934 used in conjunction with other attributes that affect argument
935 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
936 ``inalloca`` attribute also disables LLVM's implicit lowering of
937 large aggregate return values, which means that frontend authors
938 must lower them with ``sret`` pointers.
940 When the call site is reached, the argument allocation must have
941 been the most recent stack allocation that is still live, or the
942 results are undefined. It is possible to allocate additional stack
943 space after an argument allocation and before its call site, but it
944 must be cleared off with :ref:`llvm.stackrestore
947 See :doc:`InAlloca` for more information on how to use this
951 This indicates that the pointer parameter specifies the address of a
952 structure that is the return value of the function in the source
953 program. This pointer must be guaranteed by the caller to be valid:
954 loads and stores to the structure may be assumed by the callee
955 not to trap and to be properly aligned. This may only be applied to
956 the first parameter. This is not a valid attribute for return
960 This indicates that the pointer value may be assumed by the optimizer to
961 have the specified alignment.
963 Note that this attribute has additional semantics when combined with the
969 This indicates that objects accessed via pointer values
970 :ref:`based <pointeraliasing>` on the argument or return value are not also
971 accessed, during the execution of the function, via pointer values not
972 *based* on the argument or return value. The attribute on a return value
973 also has additional semantics described below. The caller shares the
974 responsibility with the callee for ensuring that these requirements are met.
975 For further details, please see the discussion of the NoAlias response in
976 :ref:`alias analysis <Must, May, or No>`.
978 Note that this definition of ``noalias`` is intentionally similar
979 to the definition of ``restrict`` in C99 for function arguments.
981 For function return values, C99's ``restrict`` is not meaningful,
982 while LLVM's ``noalias`` is. Furthermore, the semantics of the ``noalias``
983 attribute on return values are stronger than the semantics of the attribute
984 when used on function arguments. On function return values, the ``noalias``
985 attribute indicates that the function acts like a system memory allocation
986 function, returning a pointer to allocated storage disjoint from the
987 storage for any other object accessible to the caller.
990 This indicates that the callee does not make any copies of the
991 pointer that outlive the callee itself. This is not a valid
992 attribute for return values.
997 This indicates that the pointer parameter can be excised using the
998 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
999 attribute for return values and can only be applied to one parameter.
1002 This indicates that the function always returns the argument as its return
1003 value. This is an optimization hint to the code generator when generating
1004 the caller, allowing tail call optimization and omission of register saves
1005 and restores in some cases; it is not checked or enforced when generating
1006 the callee. The parameter and the function return type must be valid
1007 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
1008 valid attribute for return values and can only be applied to one parameter.
1011 This indicates that the parameter or return pointer is not null. This
1012 attribute may only be applied to pointer typed parameters. This is not
1013 checked or enforced by LLVM, the caller must ensure that the pointer
1014 passed in is non-null, or the callee must ensure that the returned pointer
1017 ``dereferenceable(<n>)``
1018 This indicates that the parameter or return pointer is dereferenceable. This
1019 attribute may only be applied to pointer typed parameters. A pointer that
1020 is dereferenceable can be loaded from speculatively without a risk of
1021 trapping. The number of bytes known to be dereferenceable must be provided
1022 in parentheses. It is legal for the number of bytes to be less than the
1023 size of the pointee type. The ``nonnull`` attribute does not imply
1024 dereferenceability (consider a pointer to one element past the end of an
1025 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
1026 ``addrspace(0)`` (which is the default address space).
1028 ``dereferenceable_or_null(<n>)``
1029 This indicates that the parameter or return value isn't both
1030 non-null and non-dereferenceable (up to ``<n>`` bytes) at the same
1031 time. All non-null pointers tagged with
1032 ``dereferenceable_or_null(<n>)`` are ``dereferenceable(<n>)``.
1033 For address space 0 ``dereferenceable_or_null(<n>)`` implies that
1034 a pointer is exactly one of ``dereferenceable(<n>)`` or ``null``,
1035 and in other address spaces ``dereferenceable_or_null(<n>)``
1036 implies that a pointer is at least one of ``dereferenceable(<n>)``
1037 or ``null`` (i.e. it may be both ``null`` and
1038 ``dereferenceable(<n>)``). This attribute may only be applied to
1039 pointer typed parameters.
1043 Garbage Collector Strategy Names
1044 --------------------------------
1046 Each function may specify a garbage collector strategy name, which is simply a
1049 .. code-block:: llvm
1051 define void @f() gc "name" { ... }
1053 The supported values of *name* includes those :ref:`built in to LLVM
1054 <builtin-gc-strategies>` and any provided by loaded plugins. Specifying a GC
1055 strategy will cause the compiler to alter its output in order to support the
1056 named garbage collection algorithm. Note that LLVM itself does not contain a
1057 garbage collector, this functionality is restricted to generating machine code
1058 which can interoperate with a collector provided externally.
1065 Prefix data is data associated with a function which the code
1066 generator will emit immediately before the function's entrypoint.
1067 The purpose of this feature is to allow frontends to associate
1068 language-specific runtime metadata with specific functions and make it
1069 available through the function pointer while still allowing the
1070 function pointer to be called.
1072 To access the data for a given function, a program may bitcast the
1073 function pointer to a pointer to the constant's type and dereference
1074 index -1. This implies that the IR symbol points just past the end of
1075 the prefix data. For instance, take the example of a function annotated
1076 with a single ``i32``,
1078 .. code-block:: llvm
1080 define void @f() prefix i32 123 { ... }
1082 The prefix data can be referenced as,
1084 .. code-block:: llvm
1086 %0 = bitcast void* () @f to i32*
1087 %a = getelementptr inbounds i32, i32* %0, i32 -1
1088 %b = load i32, i32* %a
1090 Prefix data is laid out as if it were an initializer for a global variable
1091 of the prefix data's type. The function will be placed such that the
1092 beginning of the prefix data is aligned. This means that if the size
1093 of the prefix data is not a multiple of the alignment size, the
1094 function's entrypoint will not be aligned. If alignment of the
1095 function's entrypoint is desired, padding must be added to the prefix
1098 A function may have prefix data but no body. This has similar semantics
1099 to the ``available_externally`` linkage in that the data may be used by the
1100 optimizers but will not be emitted in the object file.
1107 The ``prologue`` attribute allows arbitrary code (encoded as bytes) to
1108 be inserted prior to the function body. This can be used for enabling
1109 function hot-patching and instrumentation.
1111 To maintain the semantics of ordinary function calls, the prologue data must
1112 have a particular format. Specifically, it must begin with a sequence of
1113 bytes which decode to a sequence of machine instructions, valid for the
1114 module's target, which transfer control to the point immediately succeeding
1115 the prologue data, without performing any other visible action. This allows
1116 the inliner and other passes to reason about the semantics of the function
1117 definition without needing to reason about the prologue data. Obviously this
1118 makes the format of the prologue data highly target dependent.
1120 A trivial example of valid prologue data for the x86 architecture is ``i8 144``,
1121 which encodes the ``nop`` instruction:
1123 .. code-block:: llvm
1125 define void @f() prologue i8 144 { ... }
1127 Generally prologue data can be formed by encoding a relative branch instruction
1128 which skips the metadata, as in this example of valid prologue data for the
1129 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1131 .. code-block:: llvm
1133 %0 = type <{ i8, i8, i8* }>
1135 define void @f() prologue %0 <{ i8 235, i8 8, i8* @md}> { ... }
1137 A function may have prologue data but no body. This has similar semantics
1138 to the ``available_externally`` linkage in that the data may be used by the
1139 optimizers but will not be emitted in the object file.
1143 Personality Function
1144 --------------------
1146 The ``personality`` attribute permits functions to specify what function
1147 to use for exception handling.
1154 Attribute groups are groups of attributes that are referenced by objects within
1155 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1156 functions will use the same set of attributes. In the degenerative case of a
1157 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1158 group will capture the important command line flags used to build that file.
1160 An attribute group is a module-level object. To use an attribute group, an
1161 object references the attribute group's ID (e.g. ``#37``). An object may refer
1162 to more than one attribute group. In that situation, the attributes from the
1163 different groups are merged.
1165 Here is an example of attribute groups for a function that should always be
1166 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1168 .. code-block:: llvm
1170 ; Target-independent attributes:
1171 attributes #0 = { alwaysinline alignstack=4 }
1173 ; Target-dependent attributes:
1174 attributes #1 = { "no-sse" }
1176 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1177 define void @f() #0 #1 { ... }
1184 Function attributes are set to communicate additional information about
1185 a function. Function attributes are considered to be part of the
1186 function, not of the function type, so functions with different function
1187 attributes can have the same function type.
1189 Function attributes are simple keywords that follow the type specified.
1190 If multiple attributes are needed, they are space separated. For
1193 .. code-block:: llvm
1195 define void @f() noinline { ... }
1196 define void @f() alwaysinline { ... }
1197 define void @f() alwaysinline optsize { ... }
1198 define void @f() optsize { ... }
1201 This attribute indicates that, when emitting the prologue and
1202 epilogue, the backend should forcibly align the stack pointer.
1203 Specify the desired alignment, which must be a power of two, in
1206 This attribute indicates that the inliner should attempt to inline
1207 this function into callers whenever possible, ignoring any active
1208 inlining size threshold for this caller.
1210 This indicates that the callee function at a call site should be
1211 recognized as a built-in function, even though the function's declaration
1212 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1213 direct calls to functions that are declared with the ``nobuiltin``
1216 This attribute indicates that this function is rarely called. When
1217 computing edge weights, basic blocks post-dominated by a cold
1218 function call are also considered to be cold; and, thus, given low
1221 This attribute indicates that the callee is dependent on a convergent
1222 thread execution pattern under certain parallel execution models.
1223 Transformations that are execution model agnostic may not make the execution
1224 of a convergent operation control dependent on any additional values.
1226 This attribute indicates that the source code contained a hint that
1227 inlining this function is desirable (such as the "inline" keyword in
1228 C/C++). It is just a hint; it imposes no requirements on the
1231 This attribute indicates that the function should be added to a
1232 jump-instruction table at code-generation time, and that all address-taken
1233 references to this function should be replaced with a reference to the
1234 appropriate jump-instruction-table function pointer. Note that this creates
1235 a new pointer for the original function, which means that code that depends
1236 on function-pointer identity can break. So, any function annotated with
1237 ``jumptable`` must also be ``unnamed_addr``.
1239 This attribute suggests that optimization passes and code generator
1240 passes make choices that keep the code size of this function as small
1241 as possible and perform optimizations that may sacrifice runtime
1242 performance in order to minimize the size of the generated code.
1244 This attribute disables prologue / epilogue emission for the
1245 function. This can have very system-specific consequences.
1247 This indicates that the callee function at a call site is not recognized as
1248 a built-in function. LLVM will retain the original call and not replace it
1249 with equivalent code based on the semantics of the built-in function, unless
1250 the call site uses the ``builtin`` attribute. This is valid at call sites
1251 and on function declarations and definitions.
1253 This attribute indicates that calls to the function cannot be
1254 duplicated. A call to a ``noduplicate`` function may be moved
1255 within its parent function, but may not be duplicated within
1256 its parent function.
1258 A function containing a ``noduplicate`` call may still
1259 be an inlining candidate, provided that the call is not
1260 duplicated by inlining. That implies that the function has
1261 internal linkage and only has one call site, so the original
1262 call is dead after inlining.
1264 This attributes disables implicit floating point instructions.
1266 This attribute indicates that the inliner should never inline this
1267 function in any situation. This attribute may not be used together
1268 with the ``alwaysinline`` attribute.
1270 This attribute suppresses lazy symbol binding for the function. This
1271 may make calls to the function faster, at the cost of extra program
1272 startup time if the function is not called during program startup.
1274 This attribute indicates that the code generator should not use a
1275 red zone, even if the target-specific ABI normally permits it.
1277 This function attribute indicates that the function never returns
1278 normally. This produces undefined behavior at runtime if the
1279 function ever does dynamically return.
1281 This function attribute indicates that the function never raises an
1282 exception. If the function does raise an exception, its runtime
1283 behavior is undefined. However, functions marked nounwind may still
1284 trap or generate asynchronous exceptions. Exception handling schemes
1285 that are recognized by LLVM to handle asynchronous exceptions, such
1286 as SEH, will still provide their implementation defined semantics.
1288 This function attribute indicates that the function is not optimized
1289 by any optimization or code generator passes with the
1290 exception of interprocedural optimization passes.
1291 This attribute cannot be used together with the ``alwaysinline``
1292 attribute; this attribute is also incompatible
1293 with the ``minsize`` attribute and the ``optsize`` attribute.
1295 This attribute requires the ``noinline`` attribute to be specified on
1296 the function as well, so the function is never inlined into any caller.
1297 Only functions with the ``alwaysinline`` attribute are valid
1298 candidates for inlining into the body of this function.
1300 This attribute suggests that optimization passes and code generator
1301 passes make choices that keep the code size of this function low,
1302 and otherwise do optimizations specifically to reduce code size as
1303 long as they do not significantly impact runtime performance.
1305 On a function, this attribute indicates that the function computes its
1306 result (or decides to unwind an exception) based strictly on its arguments,
1307 without dereferencing any pointer arguments or otherwise accessing
1308 any mutable state (e.g. memory, control registers, etc) visible to
1309 caller functions. It does not write through any pointer arguments
1310 (including ``byval`` arguments) and never changes any state visible
1311 to callers. This means that it cannot unwind exceptions by calling
1312 the ``C++`` exception throwing methods.
1314 On an argument, this attribute indicates that the function does not
1315 dereference that pointer argument, even though it may read or write the
1316 memory that the pointer points to if accessed through other pointers.
1318 On a function, this attribute indicates that the function does not write
1319 through any pointer arguments (including ``byval`` arguments) or otherwise
1320 modify any state (e.g. memory, control registers, etc) visible to
1321 caller functions. It may dereference pointer arguments and read
1322 state that may be set in the caller. A readonly function always
1323 returns the same value (or unwinds an exception identically) when
1324 called with the same set of arguments and global state. It cannot
1325 unwind an exception by calling the ``C++`` exception throwing
1328 On an argument, this attribute indicates that the function does not write
1329 through this pointer argument, even though it may write to the memory that
1330 the pointer points to.
1332 This attribute indicates that the only memory accesses inside function are
1333 loads and stores from objects pointed to by its pointer-typed arguments,
1334 with arbitrary offsets. Or in other words, all memory operations in the
1335 function can refer to memory only using pointers based on its function
1337 Note that ``argmemonly`` can be used together with ``readonly`` attribute
1338 in order to specify that function reads only from its arguments.
1340 This attribute indicates that this function can return twice. The C
1341 ``setjmp`` is an example of such a function. The compiler disables
1342 some optimizations (like tail calls) in the caller of these
1345 This attribute indicates that
1346 `SafeStack <http://clang.llvm.org/docs/SafeStack.html>`_
1347 protection is enabled for this function.
1349 If a function that has a ``safestack`` attribute is inlined into a
1350 function that doesn't have a ``safestack`` attribute or which has an
1351 ``ssp``, ``sspstrong`` or ``sspreq`` attribute, then the resulting
1352 function will have a ``safestack`` attribute.
1353 ``sanitize_address``
1354 This attribute indicates that AddressSanitizer checks
1355 (dynamic address safety analysis) are enabled for this function.
1357 This attribute indicates that MemorySanitizer checks (dynamic detection
1358 of accesses to uninitialized memory) are enabled for this function.
1360 This attribute indicates that ThreadSanitizer checks
1361 (dynamic thread safety analysis) are enabled for this function.
1363 This attribute indicates that the function should emit a stack
1364 smashing protector. It is in the form of a "canary" --- a random value
1365 placed on the stack before the local variables that's checked upon
1366 return from the function to see if it has been overwritten. A
1367 heuristic is used to determine if a function needs stack protectors
1368 or not. The heuristic used will enable protectors for functions with:
1370 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1371 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1372 - Calls to alloca() with variable sizes or constant sizes greater than
1373 ``ssp-buffer-size``.
1375 Variables that are identified as requiring a protector will be arranged
1376 on the stack such that they are adjacent to the stack protector guard.
1378 If a function that has an ``ssp`` attribute is inlined into a
1379 function that doesn't have an ``ssp`` attribute, then the resulting
1380 function will have an ``ssp`` attribute.
1382 This attribute indicates that the function should *always* emit a
1383 stack smashing protector. This overrides the ``ssp`` function
1386 Variables that are identified as requiring a protector will be arranged
1387 on the stack such that they are adjacent to the stack protector guard.
1388 The specific layout rules are:
1390 #. Large arrays and structures containing large arrays
1391 (``>= ssp-buffer-size``) are closest to the stack protector.
1392 #. Small arrays and structures containing small arrays
1393 (``< ssp-buffer-size``) are 2nd closest to the protector.
1394 #. Variables that have had their address taken are 3rd closest to the
1397 If a function that has an ``sspreq`` attribute is inlined into a
1398 function that doesn't have an ``sspreq`` attribute or which has an
1399 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1400 an ``sspreq`` attribute.
1402 This attribute indicates that the function should emit a stack smashing
1403 protector. This attribute causes a strong heuristic to be used when
1404 determining if a function needs stack protectors. The strong heuristic
1405 will enable protectors for functions with:
1407 - Arrays of any size and type
1408 - Aggregates containing an array of any size and type.
1409 - Calls to alloca().
1410 - Local variables that have had their address taken.
1412 Variables that are identified as requiring a protector will be arranged
1413 on the stack such that they are adjacent to the stack protector guard.
1414 The specific layout rules are:
1416 #. Large arrays and structures containing large arrays
1417 (``>= ssp-buffer-size``) are closest to the stack protector.
1418 #. Small arrays and structures containing small arrays
1419 (``< ssp-buffer-size``) are 2nd closest to the protector.
1420 #. Variables that have had their address taken are 3rd closest to the
1423 This overrides the ``ssp`` function attribute.
1425 If a function that has an ``sspstrong`` attribute is inlined into a
1426 function that doesn't have an ``sspstrong`` attribute, then the
1427 resulting function will have an ``sspstrong`` attribute.
1429 This attribute indicates that the function will delegate to some other
1430 function with a tail call. The prototype of a thunk should not be used for
1431 optimization purposes. The caller is expected to cast the thunk prototype to
1432 match the thunk target prototype.
1434 This attribute indicates that the ABI being targeted requires that
1435 an unwind table entry be produced for this function even if we can
1436 show that no exceptions passes by it. This is normally the case for
1437 the ELF x86-64 abi, but it can be disabled for some compilation
1446 Note: operand bundles are a work in progress, and they should be
1447 considered experimental at this time.
1449 Operand bundles are tagged sets of SSA values that can be associated
1450 with certain LLVM instructions (currently only ``call`` s and
1451 ``invoke`` s). In a way they are like metadata, but dropping them is
1452 incorrect and will change program semantics.
1456 operand bundle set ::= '[' operand bundle ']'
1457 operand bundle ::= tag '(' [ bundle operand ] (, bundle operand )* ')'
1458 bundle operand ::= SSA value
1459 tag ::= string constant
1461 Operand bundles are **not** part of a function's signature, and a
1462 given function may be called from multiple places with different kinds
1463 of operand bundles. This reflects the fact that the operand bundles
1464 are conceptually a part of the ``call`` (or ``invoke``), not the
1465 callee being dispatched to.
1467 Operand bundles are a generic mechanism intended to support
1468 runtime-introspection-like functionality for managed languages. While
1469 the exact semantics of an operand bundle depend on the bundle tag,
1470 there are certain limitations to how much the presence of an operand
1471 bundle can influence the semantics of a program. These restrictions
1472 are described as the semantics of an "unknown" operand bundle. As
1473 long as the behavior of an operand bundle is describable within these
1474 restrictions, LLVM does not need to have special knowledge of the
1475 operand bundle to not miscompile programs containing it.
1477 - The bundle operands for an unknown operand bundle escape in unknown
1478 ways before control is transferred to the callee or invokee.
1479 - Calls and invokes with operand bundles have unknown read / write
1480 effect on the heap on entry and exit (even if the call target is
1481 ``readnone`` or ``readonly``), unless they're overriden with
1482 callsite specific attributes.
1483 - An operand bundle at a call site cannot change the implementation
1484 of the called function. Inter-procedural optimizations work as
1485 usual as long as they take into account the first two properties.
1489 Module-Level Inline Assembly
1490 ----------------------------
1492 Modules may contain "module-level inline asm" blocks, which corresponds
1493 to the GCC "file scope inline asm" blocks. These blocks are internally
1494 concatenated by LLVM and treated as a single unit, but may be separated
1495 in the ``.ll`` file if desired. The syntax is very simple:
1497 .. code-block:: llvm
1499 module asm "inline asm code goes here"
1500 module asm "more can go here"
1502 The strings can contain any character by escaping non-printable
1503 characters. The escape sequence used is simply "\\xx" where "xx" is the
1504 two digit hex code for the number.
1506 Note that the assembly string *must* be parseable by LLVM's integrated assembler
1507 (unless it is disabled), even when emitting a ``.s`` file.
1509 .. _langref_datalayout:
1514 A module may specify a target specific data layout string that specifies
1515 how data is to be laid out in memory. The syntax for the data layout is
1518 .. code-block:: llvm
1520 target datalayout = "layout specification"
1522 The *layout specification* consists of a list of specifications
1523 separated by the minus sign character ('-'). Each specification starts
1524 with a letter and may include other information after the letter to
1525 define some aspect of the data layout. The specifications accepted are
1529 Specifies that the target lays out data in big-endian form. That is,
1530 the bits with the most significance have the lowest address
1533 Specifies that the target lays out data in little-endian form. That
1534 is, the bits with the least significance have the lowest address
1537 Specifies the natural alignment of the stack in bits. Alignment
1538 promotion of stack variables is limited to the natural stack
1539 alignment to avoid dynamic stack realignment. The stack alignment
1540 must be a multiple of 8-bits. If omitted, the natural stack
1541 alignment defaults to "unspecified", which does not prevent any
1542 alignment promotions.
1543 ``p[n]:<size>:<abi>:<pref>``
1544 This specifies the *size* of a pointer and its ``<abi>`` and
1545 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1546 bits. The address space, ``n``, is optional, and if not specified,
1547 denotes the default address space 0. The value of ``n`` must be
1548 in the range [1,2^23).
1549 ``i<size>:<abi>:<pref>``
1550 This specifies the alignment for an integer type of a given bit
1551 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1552 ``v<size>:<abi>:<pref>``
1553 This specifies the alignment for a vector type of a given bit
1555 ``f<size>:<abi>:<pref>``
1556 This specifies the alignment for a floating point type of a given bit
1557 ``<size>``. Only values of ``<size>`` that are supported by the target
1558 will work. 32 (float) and 64 (double) are supported on all targets; 80
1559 or 128 (different flavors of long double) are also supported on some
1562 This specifies the alignment for an object of aggregate type.
1564 If present, specifies that llvm names are mangled in the output. The
1567 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1568 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1569 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1570 symbols get a ``_`` prefix.
1571 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1572 functions also get a suffix based on the frame size.
1573 * ``x``: Windows x86 COFF prefix: Similar to Windows COFF, but use a ``_``
1574 prefix for ``__cdecl`` functions.
1575 ``n<size1>:<size2>:<size3>...``
1576 This specifies a set of native integer widths for the target CPU in
1577 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1578 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1579 this set are considered to support most general arithmetic operations
1582 On every specification that takes a ``<abi>:<pref>``, specifying the
1583 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1584 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1586 When constructing the data layout for a given target, LLVM starts with a
1587 default set of specifications which are then (possibly) overridden by
1588 the specifications in the ``datalayout`` keyword. The default
1589 specifications are given in this list:
1591 - ``E`` - big endian
1592 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1593 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1594 same as the default address space.
1595 - ``S0`` - natural stack alignment is unspecified
1596 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1597 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1598 - ``i16:16:16`` - i16 is 16-bit aligned
1599 - ``i32:32:32`` - i32 is 32-bit aligned
1600 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1601 alignment of 64-bits
1602 - ``f16:16:16`` - half is 16-bit aligned
1603 - ``f32:32:32`` - float is 32-bit aligned
1604 - ``f64:64:64`` - double is 64-bit aligned
1605 - ``f128:128:128`` - quad is 128-bit aligned
1606 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1607 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1608 - ``a:0:64`` - aggregates are 64-bit aligned
1610 When LLVM is determining the alignment for a given type, it uses the
1613 #. If the type sought is an exact match for one of the specifications,
1614 that specification is used.
1615 #. If no match is found, and the type sought is an integer type, then
1616 the smallest integer type that is larger than the bitwidth of the
1617 sought type is used. If none of the specifications are larger than
1618 the bitwidth then the largest integer type is used. For example,
1619 given the default specifications above, the i7 type will use the
1620 alignment of i8 (next largest) while both i65 and i256 will use the
1621 alignment of i64 (largest specified).
1622 #. If no match is found, and the type sought is a vector type, then the
1623 largest vector type that is smaller than the sought vector type will
1624 be used as a fall back. This happens because <128 x double> can be
1625 implemented in terms of 64 <2 x double>, for example.
1627 The function of the data layout string may not be what you expect.
1628 Notably, this is not a specification from the frontend of what alignment
1629 the code generator should use.
1631 Instead, if specified, the target data layout is required to match what
1632 the ultimate *code generator* expects. This string is used by the
1633 mid-level optimizers to improve code, and this only works if it matches
1634 what the ultimate code generator uses. There is no way to generate IR
1635 that does not embed this target-specific detail into the IR. If you
1636 don't specify the string, the default specifications will be used to
1637 generate a Data Layout and the optimization phases will operate
1638 accordingly and introduce target specificity into the IR with respect to
1639 these default specifications.
1646 A module may specify a target triple string that describes the target
1647 host. The syntax for the target triple is simply:
1649 .. code-block:: llvm
1651 target triple = "x86_64-apple-macosx10.7.0"
1653 The *target triple* string consists of a series of identifiers delimited
1654 by the minus sign character ('-'). The canonical forms are:
1658 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1659 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1661 This information is passed along to the backend so that it generates
1662 code for the proper architecture. It's possible to override this on the
1663 command line with the ``-mtriple`` command line option.
1665 .. _pointeraliasing:
1667 Pointer Aliasing Rules
1668 ----------------------
1670 Any memory access must be done through a pointer value associated with
1671 an address range of the memory access, otherwise the behavior is
1672 undefined. Pointer values are associated with address ranges according
1673 to the following rules:
1675 - A pointer value is associated with the addresses associated with any
1676 value it is *based* on.
1677 - An address of a global variable is associated with the address range
1678 of the variable's storage.
1679 - The result value of an allocation instruction is associated with the
1680 address range of the allocated storage.
1681 - A null pointer in the default address-space is associated with no
1683 - An integer constant other than zero or a pointer value returned from
1684 a function not defined within LLVM may be associated with address
1685 ranges allocated through mechanisms other than those provided by
1686 LLVM. Such ranges shall not overlap with any ranges of addresses
1687 allocated by mechanisms provided by LLVM.
1689 A pointer value is *based* on another pointer value according to the
1692 - A pointer value formed from a ``getelementptr`` operation is *based*
1693 on the first value operand of the ``getelementptr``.
1694 - The result value of a ``bitcast`` is *based* on the operand of the
1696 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1697 values that contribute (directly or indirectly) to the computation of
1698 the pointer's value.
1699 - The "*based* on" relationship is transitive.
1701 Note that this definition of *"based"* is intentionally similar to the
1702 definition of *"based"* in C99, though it is slightly weaker.
1704 LLVM IR does not associate types with memory. The result type of a
1705 ``load`` merely indicates the size and alignment of the memory from
1706 which to load, as well as the interpretation of the value. The first
1707 operand type of a ``store`` similarly only indicates the size and
1708 alignment of the store.
1710 Consequently, type-based alias analysis, aka TBAA, aka
1711 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1712 :ref:`Metadata <metadata>` may be used to encode additional information
1713 which specialized optimization passes may use to implement type-based
1718 Volatile Memory Accesses
1719 ------------------------
1721 Certain memory accesses, such as :ref:`load <i_load>`'s,
1722 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1723 marked ``volatile``. The optimizers must not change the number of
1724 volatile operations or change their order of execution relative to other
1725 volatile operations. The optimizers *may* change the order of volatile
1726 operations relative to non-volatile operations. This is not Java's
1727 "volatile" and has no cross-thread synchronization behavior.
1729 IR-level volatile loads and stores cannot safely be optimized into
1730 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1731 flagged volatile. Likewise, the backend should never split or merge
1732 target-legal volatile load/store instructions.
1734 .. admonition:: Rationale
1736 Platforms may rely on volatile loads and stores of natively supported
1737 data width to be executed as single instruction. For example, in C
1738 this holds for an l-value of volatile primitive type with native
1739 hardware support, but not necessarily for aggregate types. The
1740 frontend upholds these expectations, which are intentionally
1741 unspecified in the IR. The rules above ensure that IR transformations
1742 do not violate the frontend's contract with the language.
1746 Memory Model for Concurrent Operations
1747 --------------------------------------
1749 The LLVM IR does not define any way to start parallel threads of
1750 execution or to register signal handlers. Nonetheless, there are
1751 platform-specific ways to create them, and we define LLVM IR's behavior
1752 in their presence. This model is inspired by the C++0x memory model.
1754 For a more informal introduction to this model, see the :doc:`Atomics`.
1756 We define a *happens-before* partial order as the least partial order
1759 - Is a superset of single-thread program order, and
1760 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1761 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1762 techniques, like pthread locks, thread creation, thread joining,
1763 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1764 Constraints <ordering>`).
1766 Note that program order does not introduce *happens-before* edges
1767 between a thread and signals executing inside that thread.
1769 Every (defined) read operation (load instructions, memcpy, atomic
1770 loads/read-modify-writes, etc.) R reads a series of bytes written by
1771 (defined) write operations (store instructions, atomic
1772 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1773 section, initialized globals are considered to have a write of the
1774 initializer which is atomic and happens before any other read or write
1775 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1776 may see any write to the same byte, except:
1778 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1779 write\ :sub:`2` happens before R\ :sub:`byte`, then
1780 R\ :sub:`byte` does not see write\ :sub:`1`.
1781 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1782 R\ :sub:`byte` does not see write\ :sub:`3`.
1784 Given that definition, R\ :sub:`byte` is defined as follows:
1786 - If R is volatile, the result is target-dependent. (Volatile is
1787 supposed to give guarantees which can support ``sig_atomic_t`` in
1788 C/C++, and may be used for accesses to addresses that do not behave
1789 like normal memory. It does not generally provide cross-thread
1791 - Otherwise, if there is no write to the same byte that happens before
1792 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1793 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1794 R\ :sub:`byte` returns the value written by that write.
1795 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1796 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1797 Memory Ordering Constraints <ordering>` section for additional
1798 constraints on how the choice is made.
1799 - Otherwise R\ :sub:`byte` returns ``undef``.
1801 R returns the value composed of the series of bytes it read. This
1802 implies that some bytes within the value may be ``undef`` **without**
1803 the entire value being ``undef``. Note that this only defines the
1804 semantics of the operation; it doesn't mean that targets will emit more
1805 than one instruction to read the series of bytes.
1807 Note that in cases where none of the atomic intrinsics are used, this
1808 model places only one restriction on IR transformations on top of what
1809 is required for single-threaded execution: introducing a store to a byte
1810 which might not otherwise be stored is not allowed in general.
1811 (Specifically, in the case where another thread might write to and read
1812 from an address, introducing a store can change a load that may see
1813 exactly one write into a load that may see multiple writes.)
1817 Atomic Memory Ordering Constraints
1818 ----------------------------------
1820 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1821 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1822 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1823 ordering parameters that determine which other atomic instructions on
1824 the same address they *synchronize with*. These semantics are borrowed
1825 from Java and C++0x, but are somewhat more colloquial. If these
1826 descriptions aren't precise enough, check those specs (see spec
1827 references in the :doc:`atomics guide <Atomics>`).
1828 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1829 differently since they don't take an address. See that instruction's
1830 documentation for details.
1832 For a simpler introduction to the ordering constraints, see the
1836 The set of values that can be read is governed by the happens-before
1837 partial order. A value cannot be read unless some operation wrote
1838 it. This is intended to provide a guarantee strong enough to model
1839 Java's non-volatile shared variables. This ordering cannot be
1840 specified for read-modify-write operations; it is not strong enough
1841 to make them atomic in any interesting way.
1843 In addition to the guarantees of ``unordered``, there is a single
1844 total order for modifications by ``monotonic`` operations on each
1845 address. All modification orders must be compatible with the
1846 happens-before order. There is no guarantee that the modification
1847 orders can be combined to a global total order for the whole program
1848 (and this often will not be possible). The read in an atomic
1849 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1850 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1851 order immediately before the value it writes. If one atomic read
1852 happens before another atomic read of the same address, the later
1853 read must see the same value or a later value in the address's
1854 modification order. This disallows reordering of ``monotonic`` (or
1855 stronger) operations on the same address. If an address is written
1856 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1857 read that address repeatedly, the other threads must eventually see
1858 the write. This corresponds to the C++0x/C1x
1859 ``memory_order_relaxed``.
1861 In addition to the guarantees of ``monotonic``, a
1862 *synchronizes-with* edge may be formed with a ``release`` operation.
1863 This is intended to model C++'s ``memory_order_acquire``.
1865 In addition to the guarantees of ``monotonic``, if this operation
1866 writes a value which is subsequently read by an ``acquire``
1867 operation, it *synchronizes-with* that operation. (This isn't a
1868 complete description; see the C++0x definition of a release
1869 sequence.) This corresponds to the C++0x/C1x
1870 ``memory_order_release``.
1871 ``acq_rel`` (acquire+release)
1872 Acts as both an ``acquire`` and ``release`` operation on its
1873 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1874 ``seq_cst`` (sequentially consistent)
1875 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1876 operation that only reads, ``release`` for an operation that only
1877 writes), there is a global total order on all
1878 sequentially-consistent operations on all addresses, which is
1879 consistent with the *happens-before* partial order and with the
1880 modification orders of all the affected addresses. Each
1881 sequentially-consistent read sees the last preceding write to the
1882 same address in this global order. This corresponds to the C++0x/C1x
1883 ``memory_order_seq_cst`` and Java volatile.
1887 If an atomic operation is marked ``singlethread``, it only *synchronizes
1888 with* or participates in modification and seq\_cst total orderings with
1889 other operations running in the same thread (for example, in signal
1897 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1898 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1899 :ref:`frem <i_frem>`, :ref:`fcmp <i_fcmp>`) have the following flags that can
1900 be set to enable otherwise unsafe floating point operations
1903 No NaNs - Allow optimizations to assume the arguments and result are not
1904 NaN. Such optimizations are required to retain defined behavior over
1905 NaNs, but the value of the result is undefined.
1908 No Infs - Allow optimizations to assume the arguments and result are not
1909 +/-Inf. Such optimizations are required to retain defined behavior over
1910 +/-Inf, but the value of the result is undefined.
1913 No Signed Zeros - Allow optimizations to treat the sign of a zero
1914 argument or result as insignificant.
1917 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1918 argument rather than perform division.
1921 Fast - Allow algebraically equivalent transformations that may
1922 dramatically change results in floating point (e.g. reassociate). This
1923 flag implies all the others.
1927 Use-list Order Directives
1928 -------------------------
1930 Use-list directives encode the in-memory order of each use-list, allowing the
1931 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1932 indexes that are assigned to the referenced value's uses. The referenced
1933 value's use-list is immediately sorted by these indexes.
1935 Use-list directives may appear at function scope or global scope. They are not
1936 instructions, and have no effect on the semantics of the IR. When they're at
1937 function scope, they must appear after the terminator of the final basic block.
1939 If basic blocks have their address taken via ``blockaddress()`` expressions,
1940 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1947 uselistorder <ty> <value>, { <order-indexes> }
1948 uselistorder_bb @function, %block { <order-indexes> }
1954 define void @foo(i32 %arg1, i32 %arg2) {
1956 ; ... instructions ...
1958 ; ... instructions ...
1960 ; At function scope.
1961 uselistorder i32 %arg1, { 1, 0, 2 }
1962 uselistorder label %bb, { 1, 0 }
1966 uselistorder i32* @global, { 1, 2, 0 }
1967 uselistorder i32 7, { 1, 0 }
1968 uselistorder i32 (i32) @bar, { 1, 0 }
1969 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1976 The LLVM type system is one of the most important features of the
1977 intermediate representation. Being typed enables a number of
1978 optimizations to be performed on the intermediate representation
1979 directly, without having to do extra analyses on the side before the
1980 transformation. A strong type system makes it easier to read the
1981 generated code and enables novel analyses and transformations that are
1982 not feasible to perform on normal three address code representations.
1992 The void type does not represent any value and has no size.
2010 The function type can be thought of as a function signature. It consists of a
2011 return type and a list of formal parameter types. The return type of a function
2012 type is a void type or first class type --- except for :ref:`label <t_label>`
2013 and :ref:`metadata <t_metadata>` types.
2019 <returntype> (<parameter list>)
2021 ...where '``<parameter list>``' is a comma-separated list of type
2022 specifiers. Optionally, the parameter list may include a type ``...``, which
2023 indicates that the function takes a variable number of arguments. Variable
2024 argument functions can access their arguments with the :ref:`variable argument
2025 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
2026 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
2030 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2031 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
2032 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2033 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
2034 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2035 | ``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. |
2036 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2037 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
2038 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2045 The :ref:`first class <t_firstclass>` types are perhaps the most important.
2046 Values of these types are the only ones which can be produced by
2054 These are the types that are valid in registers from CodeGen's perspective.
2063 The integer type is a very simple type that simply specifies an
2064 arbitrary bit width for the integer type desired. Any bit width from 1
2065 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
2073 The number of bits the integer will occupy is specified by the ``N``
2079 +----------------+------------------------------------------------+
2080 | ``i1`` | a single-bit integer. |
2081 +----------------+------------------------------------------------+
2082 | ``i32`` | a 32-bit integer. |
2083 +----------------+------------------------------------------------+
2084 | ``i1942652`` | a really big integer of over 1 million bits. |
2085 +----------------+------------------------------------------------+
2089 Floating Point Types
2090 """"""""""""""""""""
2099 - 16-bit floating point value
2102 - 32-bit floating point value
2105 - 64-bit floating point value
2108 - 128-bit floating point value (112-bit mantissa)
2111 - 80-bit floating point value (X87)
2114 - 128-bit floating point value (two 64-bits)
2121 The x86_mmx type represents a value held in an MMX register on an x86
2122 machine. The operations allowed on it are quite limited: parameters and
2123 return values, load and store, and bitcast. User-specified MMX
2124 instructions are represented as intrinsic or asm calls with arguments
2125 and/or results of this type. There are no arrays, vectors or constants
2142 The pointer type is used to specify memory locations. Pointers are
2143 commonly used to reference objects in memory.
2145 Pointer types may have an optional address space attribute defining the
2146 numbered address space where the pointed-to object resides. The default
2147 address space is number zero. The semantics of non-zero address spaces
2148 are target-specific.
2150 Note that LLVM does not permit pointers to void (``void*``) nor does it
2151 permit pointers to labels (``label*``). Use ``i8*`` instead.
2161 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2162 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2163 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2164 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2165 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2166 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2167 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2176 A vector type is a simple derived type that represents a vector of
2177 elements. Vector types are used when multiple primitive data are
2178 operated in parallel using a single instruction (SIMD). A vector type
2179 requires a size (number of elements) and an underlying primitive data
2180 type. Vector types are considered :ref:`first class <t_firstclass>`.
2186 < <# elements> x <elementtype> >
2188 The number of elements is a constant integer value larger than 0;
2189 elementtype may be any integer, floating point or pointer type. Vectors
2190 of size zero are not allowed.
2194 +-------------------+--------------------------------------------------+
2195 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2196 +-------------------+--------------------------------------------------+
2197 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2198 +-------------------+--------------------------------------------------+
2199 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2200 +-------------------+--------------------------------------------------+
2201 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2202 +-------------------+--------------------------------------------------+
2211 The label type represents code labels.
2226 The token type is used when a value is associated with an instruction
2227 but all uses of the value must not attempt to introspect or obscure it.
2228 As such, it is not appropriate to have a :ref:`phi <i_phi>` or
2229 :ref:`select <i_select>` of type token.
2246 The metadata type represents embedded metadata. No derived types may be
2247 created from metadata except for :ref:`function <t_function>` arguments.
2260 Aggregate Types are a subset of derived types that can contain multiple
2261 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2262 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2272 The array type is a very simple derived type that arranges elements
2273 sequentially in memory. The array type requires a size (number of
2274 elements) and an underlying data type.
2280 [<# elements> x <elementtype>]
2282 The number of elements is a constant integer value; ``elementtype`` may
2283 be any type with a size.
2287 +------------------+--------------------------------------+
2288 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2289 +------------------+--------------------------------------+
2290 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2291 +------------------+--------------------------------------+
2292 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2293 +------------------+--------------------------------------+
2295 Here are some examples of multidimensional arrays:
2297 +-----------------------------+----------------------------------------------------------+
2298 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2299 +-----------------------------+----------------------------------------------------------+
2300 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2301 +-----------------------------+----------------------------------------------------------+
2302 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2303 +-----------------------------+----------------------------------------------------------+
2305 There is no restriction on indexing beyond the end of the array implied
2306 by a static type (though there are restrictions on indexing beyond the
2307 bounds of an allocated object in some cases). This means that
2308 single-dimension 'variable sized array' addressing can be implemented in
2309 LLVM with a zero length array type. An implementation of 'pascal style
2310 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2320 The structure type is used to represent a collection of data members
2321 together in memory. The elements of a structure may be any type that has
2324 Structures in memory are accessed using '``load``' and '``store``' by
2325 getting a pointer to a field with the '``getelementptr``' instruction.
2326 Structures in registers are accessed using the '``extractvalue``' and
2327 '``insertvalue``' instructions.
2329 Structures may optionally be "packed" structures, which indicate that
2330 the alignment of the struct is one byte, and that there is no padding
2331 between the elements. In non-packed structs, padding between field types
2332 is inserted as defined by the DataLayout string in the module, which is
2333 required to match what the underlying code generator expects.
2335 Structures can either be "literal" or "identified". A literal structure
2336 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2337 identified types are always defined at the top level with a name.
2338 Literal types are uniqued by their contents and can never be recursive
2339 or opaque since there is no way to write one. Identified types can be
2340 recursive, can be opaqued, and are never uniqued.
2346 %T1 = type { <type list> } ; Identified normal struct type
2347 %T2 = type <{ <type list> }> ; Identified packed struct type
2351 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2352 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2353 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2354 | ``{ 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``. |
2355 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2356 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2357 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2361 Opaque Structure Types
2362 """"""""""""""""""""""
2366 Opaque structure types are used to represent named structure types that
2367 do not have a body specified. This corresponds (for example) to the C
2368 notion of a forward declared structure.
2379 +--------------+-------------------+
2380 | ``opaque`` | An opaque type. |
2381 +--------------+-------------------+
2388 LLVM has several different basic types of constants. This section
2389 describes them all and their syntax.
2394 **Boolean constants**
2395 The two strings '``true``' and '``false``' are both valid constants
2397 **Integer constants**
2398 Standard integers (such as '4') are constants of the
2399 :ref:`integer <t_integer>` type. Negative numbers may be used with
2401 **Floating point constants**
2402 Floating point constants use standard decimal notation (e.g.
2403 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2404 hexadecimal notation (see below). The assembler requires the exact
2405 decimal value of a floating-point constant. For example, the
2406 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2407 decimal in binary. Floating point constants must have a :ref:`floating
2408 point <t_floating>` type.
2409 **Null pointer constants**
2410 The identifier '``null``' is recognized as a null pointer constant
2411 and must be of :ref:`pointer type <t_pointer>`.
2413 The one non-intuitive notation for constants is the hexadecimal form of
2414 floating point constants. For example, the form
2415 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2416 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2417 constants are required (and the only time that they are generated by the
2418 disassembler) is when a floating point constant must be emitted but it
2419 cannot be represented as a decimal floating point number in a reasonable
2420 number of digits. For example, NaN's, infinities, and other special
2421 values are represented in their IEEE hexadecimal format so that assembly
2422 and disassembly do not cause any bits to change in the constants.
2424 When using the hexadecimal form, constants of types half, float, and
2425 double are represented using the 16-digit form shown above (which
2426 matches the IEEE754 representation for double); half and float values
2427 must, however, be exactly representable as IEEE 754 half and single
2428 precision, respectively. Hexadecimal format is always used for long
2429 double, and there are three forms of long double. The 80-bit format used
2430 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2431 128-bit format used by PowerPC (two adjacent doubles) is represented by
2432 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2433 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2434 will only work if they match the long double format on your target.
2435 The IEEE 16-bit format (half precision) is represented by ``0xH``
2436 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2437 (sign bit at the left).
2439 There are no constants of type x86_mmx.
2441 .. _complexconstants:
2446 Complex constants are a (potentially recursive) combination of simple
2447 constants and smaller complex constants.
2449 **Structure constants**
2450 Structure constants are represented with notation similar to
2451 structure type definitions (a comma separated list of elements,
2452 surrounded by braces (``{}``)). For example:
2453 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2454 "``@G = external global i32``". Structure constants must have
2455 :ref:`structure type <t_struct>`, and the number and types of elements
2456 must match those specified by the type.
2458 Array constants are represented with notation similar to array type
2459 definitions (a comma separated list of elements, surrounded by
2460 square brackets (``[]``)). For example:
2461 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2462 :ref:`array type <t_array>`, and the number and types of elements must
2463 match those specified by the type. As a special case, character array
2464 constants may also be represented as a double-quoted string using the ``c``
2465 prefix. For example: "``c"Hello World\0A\00"``".
2466 **Vector constants**
2467 Vector constants are represented with notation similar to vector
2468 type definitions (a comma separated list of elements, surrounded by
2469 less-than/greater-than's (``<>``)). For example:
2470 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2471 must have :ref:`vector type <t_vector>`, and the number and types of
2472 elements must match those specified by the type.
2473 **Zero initialization**
2474 The string '``zeroinitializer``' can be used to zero initialize a
2475 value to zero of *any* type, including scalar and
2476 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2477 having to print large zero initializers (e.g. for large arrays) and
2478 is always exactly equivalent to using explicit zero initializers.
2480 A metadata node is a constant tuple without types. For example:
2481 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2482 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2483 Unlike other typed constants that are meant to be interpreted as part of
2484 the instruction stream, metadata is a place to attach additional
2485 information such as debug info.
2487 Global Variable and Function Addresses
2488 --------------------------------------
2490 The addresses of :ref:`global variables <globalvars>` and
2491 :ref:`functions <functionstructure>` are always implicitly valid
2492 (link-time) constants. These constants are explicitly referenced when
2493 the :ref:`identifier for the global <identifiers>` is used and always have
2494 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2497 .. code-block:: llvm
2501 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2508 The string '``undef``' can be used anywhere a constant is expected, and
2509 indicates that the user of the value may receive an unspecified
2510 bit-pattern. Undefined values may be of any type (other than '``label``'
2511 or '``void``') and be used anywhere a constant is permitted.
2513 Undefined values are useful because they indicate to the compiler that
2514 the program is well defined no matter what value is used. This gives the
2515 compiler more freedom to optimize. Here are some examples of
2516 (potentially surprising) transformations that are valid (in pseudo IR):
2518 .. code-block:: llvm
2528 This is safe because all of the output bits are affected by the undef
2529 bits. Any output bit can have a zero or one depending on the input bits.
2531 .. code-block:: llvm
2542 These logical operations have bits that are not always affected by the
2543 input. For example, if ``%X`` has a zero bit, then the output of the
2544 '``and``' operation will always be a zero for that bit, no matter what
2545 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2546 optimize or assume that the result of the '``and``' is '``undef``'.
2547 However, it is safe to assume that all bits of the '``undef``' could be
2548 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2549 all the bits of the '``undef``' operand to the '``or``' could be set,
2550 allowing the '``or``' to be folded to -1.
2552 .. code-block:: llvm
2554 %A = select undef, %X, %Y
2555 %B = select undef, 42, %Y
2556 %C = select %X, %Y, undef
2566 This set of examples shows that undefined '``select``' (and conditional
2567 branch) conditions can go *either way*, but they have to come from one
2568 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2569 both known to have a clear low bit, then ``%A`` would have to have a
2570 cleared low bit. However, in the ``%C`` example, the optimizer is
2571 allowed to assume that the '``undef``' operand could be the same as
2572 ``%Y``, allowing the whole '``select``' to be eliminated.
2574 .. code-block:: llvm
2576 %A = xor undef, undef
2593 This example points out that two '``undef``' operands are not
2594 necessarily the same. This can be surprising to people (and also matches
2595 C semantics) where they assume that "``X^X``" is always zero, even if
2596 ``X`` is undefined. This isn't true for a number of reasons, but the
2597 short answer is that an '``undef``' "variable" can arbitrarily change
2598 its value over its "live range". This is true because the variable
2599 doesn't actually *have a live range*. Instead, the value is logically
2600 read from arbitrary registers that happen to be around when needed, so
2601 the value is not necessarily consistent over time. In fact, ``%A`` and
2602 ``%C`` need to have the same semantics or the core LLVM "replace all
2603 uses with" concept would not hold.
2605 .. code-block:: llvm
2613 These examples show the crucial difference between an *undefined value*
2614 and *undefined behavior*. An undefined value (like '``undef``') is
2615 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2616 operation can be constant folded to '``undef``', because the '``undef``'
2617 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2618 However, in the second example, we can make a more aggressive
2619 assumption: because the ``undef`` is allowed to be an arbitrary value,
2620 we are allowed to assume that it could be zero. Since a divide by zero
2621 has *undefined behavior*, we are allowed to assume that the operation
2622 does not execute at all. This allows us to delete the divide and all
2623 code after it. Because the undefined operation "can't happen", the
2624 optimizer can assume that it occurs in dead code.
2626 .. code-block:: llvm
2628 a: store undef -> %X
2629 b: store %X -> undef
2634 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2635 value can be assumed to not have any effect; we can assume that the
2636 value is overwritten with bits that happen to match what was already
2637 there. However, a store *to* an undefined location could clobber
2638 arbitrary memory, therefore, it has undefined behavior.
2645 Poison values are similar to :ref:`undef values <undefvalues>`, however
2646 they also represent the fact that an instruction or constant expression
2647 that cannot evoke side effects has nevertheless detected a condition
2648 that results in undefined behavior.
2650 There is currently no way of representing a poison value in the IR; they
2651 only exist when produced by operations such as :ref:`add <i_add>` with
2654 Poison value behavior is defined in terms of value *dependence*:
2656 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2657 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2658 their dynamic predecessor basic block.
2659 - Function arguments depend on the corresponding actual argument values
2660 in the dynamic callers of their functions.
2661 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2662 instructions that dynamically transfer control back to them.
2663 - :ref:`Invoke <i_invoke>` instructions depend on the
2664 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2665 call instructions that dynamically transfer control back to them.
2666 - Non-volatile loads and stores depend on the most recent stores to all
2667 of the referenced memory addresses, following the order in the IR
2668 (including loads and stores implied by intrinsics such as
2669 :ref:`@llvm.memcpy <int_memcpy>`.)
2670 - An instruction with externally visible side effects depends on the
2671 most recent preceding instruction with externally visible side
2672 effects, following the order in the IR. (This includes :ref:`volatile
2673 operations <volatile>`.)
2674 - An instruction *control-depends* on a :ref:`terminator
2675 instruction <terminators>` if the terminator instruction has
2676 multiple successors and the instruction is always executed when
2677 control transfers to one of the successors, and may not be executed
2678 when control is transferred to another.
2679 - Additionally, an instruction also *control-depends* on a terminator
2680 instruction if the set of instructions it otherwise depends on would
2681 be different if the terminator had transferred control to a different
2683 - Dependence is transitive.
2685 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2686 with the additional effect that any instruction that has a *dependence*
2687 on a poison value has undefined behavior.
2689 Here are some examples:
2691 .. code-block:: llvm
2694 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2695 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2696 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
2697 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2699 store i32 %poison, i32* @g ; Poison value stored to memory.
2700 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
2702 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2704 %narrowaddr = bitcast i32* @g to i16*
2705 %wideaddr = bitcast i32* @g to i64*
2706 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2707 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
2709 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2710 br i1 %cmp, label %true, label %end ; Branch to either destination.
2713 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2714 ; it has undefined behavior.
2718 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2719 ; Both edges into this PHI are
2720 ; control-dependent on %cmp, so this
2721 ; always results in a poison value.
2723 store volatile i32 0, i32* @g ; This would depend on the store in %true
2724 ; if %cmp is true, or the store in %entry
2725 ; otherwise, so this is undefined behavior.
2727 br i1 %cmp, label %second_true, label %second_end
2728 ; The same branch again, but this time the
2729 ; true block doesn't have side effects.
2736 store volatile i32 0, i32* @g ; This time, the instruction always depends
2737 ; on the store in %end. Also, it is
2738 ; control-equivalent to %end, so this is
2739 ; well-defined (ignoring earlier undefined
2740 ; behavior in this example).
2744 Addresses of Basic Blocks
2745 -------------------------
2747 ``blockaddress(@function, %block)``
2749 The '``blockaddress``' constant computes the address of the specified
2750 basic block in the specified function, and always has an ``i8*`` type.
2751 Taking the address of the entry block is illegal.
2753 This value only has defined behavior when used as an operand to the
2754 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2755 against null. Pointer equality tests between labels addresses results in
2756 undefined behavior --- though, again, comparison against null is ok, and
2757 no label is equal to the null pointer. This may be passed around as an
2758 opaque pointer sized value as long as the bits are not inspected. This
2759 allows ``ptrtoint`` and arithmetic to be performed on these values so
2760 long as the original value is reconstituted before the ``indirectbr``
2763 Finally, some targets may provide defined semantics when using the value
2764 as the operand to an inline assembly, but that is target specific.
2768 Constant Expressions
2769 --------------------
2771 Constant expressions are used to allow expressions involving other
2772 constants to be used as constants. Constant expressions may be of any
2773 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2774 that does not have side effects (e.g. load and call are not supported).
2775 The following is the syntax for constant expressions:
2777 ``trunc (CST to TYPE)``
2778 Truncate a constant to another type. The bit size of CST must be
2779 larger than the bit size of TYPE. Both types must be integers.
2780 ``zext (CST to TYPE)``
2781 Zero extend a constant to another type. The bit size of CST must be
2782 smaller than the bit size of TYPE. Both types must be integers.
2783 ``sext (CST to TYPE)``
2784 Sign extend a constant to another type. The bit size of CST must be
2785 smaller than the bit size of TYPE. Both types must be integers.
2786 ``fptrunc (CST to TYPE)``
2787 Truncate a floating point constant to another floating point type.
2788 The size of CST must be larger than the size of TYPE. Both types
2789 must be floating point.
2790 ``fpext (CST to TYPE)``
2791 Floating point extend a constant to another type. The size of CST
2792 must be smaller or equal to the size of TYPE. Both types must be
2794 ``fptoui (CST to TYPE)``
2795 Convert a floating point constant to the corresponding unsigned
2796 integer constant. TYPE must be a scalar or vector integer type. CST
2797 must be of scalar or vector floating point type. Both CST and TYPE
2798 must be scalars, or vectors of the same number of elements. If the
2799 value won't fit in the integer type, the results are undefined.
2800 ``fptosi (CST to TYPE)``
2801 Convert a floating point constant to the corresponding signed
2802 integer constant. TYPE must be a scalar or vector integer type. CST
2803 must be of scalar or vector floating point type. Both CST and TYPE
2804 must be scalars, or vectors of the same number of elements. If the
2805 value won't fit in the integer type, the results are undefined.
2806 ``uitofp (CST to TYPE)``
2807 Convert an unsigned integer constant to the corresponding floating
2808 point constant. TYPE must be a scalar or vector floating point type.
2809 CST must be of scalar or vector integer type. Both CST and TYPE must
2810 be scalars, or vectors of the same number of elements. If the value
2811 won't fit in the floating point type, the results are undefined.
2812 ``sitofp (CST to TYPE)``
2813 Convert a signed integer constant to the corresponding floating
2814 point constant. TYPE must be a scalar or vector floating point type.
2815 CST must be of scalar or vector integer type. Both CST and TYPE must
2816 be scalars, or vectors of the same number of elements. If the value
2817 won't fit in the floating point type, the results are undefined.
2818 ``ptrtoint (CST to TYPE)``
2819 Convert a pointer typed constant to the corresponding integer
2820 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2821 pointer type. The ``CST`` value is zero extended, truncated, or
2822 unchanged to make it fit in ``TYPE``.
2823 ``inttoptr (CST to TYPE)``
2824 Convert an integer constant to a pointer constant. TYPE must be a
2825 pointer type. CST must be of integer type. The CST value is zero
2826 extended, truncated, or unchanged to make it fit in a pointer size.
2827 This one is *really* dangerous!
2828 ``bitcast (CST to TYPE)``
2829 Convert a constant, CST, to another TYPE. The constraints of the
2830 operands are the same as those for the :ref:`bitcast
2831 instruction <i_bitcast>`.
2832 ``addrspacecast (CST to TYPE)``
2833 Convert a constant pointer or constant vector of pointer, CST, to another
2834 TYPE in a different address space. The constraints of the operands are the
2835 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2836 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
2837 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2838 constants. As with the :ref:`getelementptr <i_getelementptr>`
2839 instruction, the index list may have zero or more indexes, which are
2840 required to make sense for the type of "pointer to TY".
2841 ``select (COND, VAL1, VAL2)``
2842 Perform the :ref:`select operation <i_select>` on constants.
2843 ``icmp COND (VAL1, VAL2)``
2844 Performs the :ref:`icmp operation <i_icmp>` on constants.
2845 ``fcmp COND (VAL1, VAL2)``
2846 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2847 ``extractelement (VAL, IDX)``
2848 Perform the :ref:`extractelement operation <i_extractelement>` on
2850 ``insertelement (VAL, ELT, IDX)``
2851 Perform the :ref:`insertelement operation <i_insertelement>` on
2853 ``shufflevector (VEC1, VEC2, IDXMASK)``
2854 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2856 ``extractvalue (VAL, IDX0, IDX1, ...)``
2857 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2858 constants. The index list is interpreted in a similar manner as
2859 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2860 least one index value must be specified.
2861 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2862 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2863 The index list is interpreted in a similar manner as indices in a
2864 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2865 value must be specified.
2866 ``OPCODE (LHS, RHS)``
2867 Perform the specified operation of the LHS and RHS constants. OPCODE
2868 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2869 binary <bitwiseops>` operations. The constraints on operands are
2870 the same as those for the corresponding instruction (e.g. no bitwise
2871 operations on floating point values are allowed).
2878 Inline Assembler Expressions
2879 ----------------------------
2881 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2882 Inline Assembly <moduleasm>`) through the use of a special value. This value
2883 represents the inline assembler as a template string (containing the
2884 instructions to emit), a list of operand constraints (stored as a string), a
2885 flag that indicates whether or not the inline asm expression has side effects,
2886 and a flag indicating whether the function containing the asm needs to align its
2887 stack conservatively.
2889 The template string supports argument substitution of the operands using "``$``"
2890 followed by a number, to indicate substitution of the given register/memory
2891 location, as specified by the constraint string. "``${NUM:MODIFIER}``" may also
2892 be used, where ``MODIFIER`` is a target-specific annotation for how to print the
2893 operand (See :ref:`inline-asm-modifiers`).
2895 A literal "``$``" may be included by using "``$$``" in the template. To include
2896 other special characters into the output, the usual "``\XX``" escapes may be
2897 used, just as in other strings. Note that after template substitution, the
2898 resulting assembly string is parsed by LLVM's integrated assembler unless it is
2899 disabled -- even when emitting a ``.s`` file -- and thus must contain assembly
2900 syntax known to LLVM.
2902 LLVM's support for inline asm is modeled closely on the requirements of Clang's
2903 GCC-compatible inline-asm support. Thus, the feature-set and the constraint and
2904 modifier codes listed here are similar or identical to those in GCC's inline asm
2905 support. However, to be clear, the syntax of the template and constraint strings
2906 described here is *not* the same as the syntax accepted by GCC and Clang, and,
2907 while most constraint letters are passed through as-is by Clang, some get
2908 translated to other codes when converting from the C source to the LLVM
2911 An example inline assembler expression is:
2913 .. code-block:: llvm
2915 i32 (i32) asm "bswap $0", "=r,r"
2917 Inline assembler expressions may **only** be used as the callee operand
2918 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2919 Thus, typically we have:
2921 .. code-block:: llvm
2923 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2925 Inline asms with side effects not visible in the constraint list must be
2926 marked as having side effects. This is done through the use of the
2927 '``sideeffect``' keyword, like so:
2929 .. code-block:: llvm
2931 call void asm sideeffect "eieio", ""()
2933 In some cases inline asms will contain code that will not work unless
2934 the stack is aligned in some way, such as calls or SSE instructions on
2935 x86, yet will not contain code that does that alignment within the asm.
2936 The compiler should make conservative assumptions about what the asm
2937 might contain and should generate its usual stack alignment code in the
2938 prologue if the '``alignstack``' keyword is present:
2940 .. code-block:: llvm
2942 call void asm alignstack "eieio", ""()
2944 Inline asms also support using non-standard assembly dialects. The
2945 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2946 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2947 the only supported dialects. An example is:
2949 .. code-block:: llvm
2951 call void asm inteldialect "eieio", ""()
2953 If multiple keywords appear the '``sideeffect``' keyword must come
2954 first, the '``alignstack``' keyword second and the '``inteldialect``'
2957 Inline Asm Constraint String
2958 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2960 The constraint list is a comma-separated string, each element containing one or
2961 more constraint codes.
2963 For each element in the constraint list an appropriate register or memory
2964 operand will be chosen, and it will be made available to assembly template
2965 string expansion as ``$0`` for the first constraint in the list, ``$1`` for the
2968 There are three different types of constraints, which are distinguished by a
2969 prefix symbol in front of the constraint code: Output, Input, and Clobber. The
2970 constraints must always be given in that order: outputs first, then inputs, then
2971 clobbers. They cannot be intermingled.
2973 There are also three different categories of constraint codes:
2975 - Register constraint. This is either a register class, or a fixed physical
2976 register. This kind of constraint will allocate a register, and if necessary,
2977 bitcast the argument or result to the appropriate type.
2978 - Memory constraint. This kind of constraint is for use with an instruction
2979 taking a memory operand. Different constraints allow for different addressing
2980 modes used by the target.
2981 - Immediate value constraint. This kind of constraint is for an integer or other
2982 immediate value which can be rendered directly into an instruction. The
2983 various target-specific constraints allow the selection of a value in the
2984 proper range for the instruction you wish to use it with.
2989 Output constraints are specified by an "``=``" prefix (e.g. "``=r``"). This
2990 indicates that the assembly will write to this operand, and the operand will
2991 then be made available as a return value of the ``asm`` expression. Output
2992 constraints do not consume an argument from the call instruction. (Except, see
2993 below about indirect outputs).
2995 Normally, it is expected that no output locations are written to by the assembly
2996 expression until *all* of the inputs have been read. As such, LLVM may assign
2997 the same register to an output and an input. If this is not safe (e.g. if the
2998 assembly contains two instructions, where the first writes to one output, and
2999 the second reads an input and writes to a second output), then the "``&``"
3000 modifier must be used (e.g. "``=&r``") to specify that the output is an
3001 "early-clobber" output. Marking an ouput as "early-clobber" ensures that LLVM
3002 will not use the same register for any inputs (other than an input tied to this
3008 Input constraints do not have a prefix -- just the constraint codes. Each input
3009 constraint will consume one argument from the call instruction. It is not
3010 permitted for the asm to write to any input register or memory location (unless
3011 that input is tied to an output). Note also that multiple inputs may all be
3012 assigned to the same register, if LLVM can determine that they necessarily all
3013 contain the same value.
3015 Instead of providing a Constraint Code, input constraints may also "tie"
3016 themselves to an output constraint, by providing an integer as the constraint
3017 string. Tied inputs still consume an argument from the call instruction, and
3018 take up a position in the asm template numbering as is usual -- they will simply
3019 be constrained to always use the same register as the output they've been tied
3020 to. For example, a constraint string of "``=r,0``" says to assign a register for
3021 output, and use that register as an input as well (it being the 0'th
3024 It is permitted to tie an input to an "early-clobber" output. In that case, no
3025 *other* input may share the same register as the input tied to the early-clobber
3026 (even when the other input has the same value).
3028 You may only tie an input to an output which has a register constraint, not a
3029 memory constraint. Only a single input may be tied to an output.
3031 There is also an "interesting" feature which deserves a bit of explanation: if a
3032 register class constraint allocates a register which is too small for the value
3033 type operand provided as input, the input value will be split into multiple
3034 registers, and all of them passed to the inline asm.
3036 However, this feature is often not as useful as you might think.
3038 Firstly, the registers are *not* guaranteed to be consecutive. So, on those
3039 architectures that have instructions which operate on multiple consecutive
3040 instructions, this is not an appropriate way to support them. (e.g. the 32-bit
3041 SparcV8 has a 64-bit load, which instruction takes a single 32-bit register. The
3042 hardware then loads into both the named register, and the next register. This
3043 feature of inline asm would not be useful to support that.)
3045 A few of the targets provide a template string modifier allowing explicit access
3046 to the second register of a two-register operand (e.g. MIPS ``L``, ``M``, and
3047 ``D``). On such an architecture, you can actually access the second allocated
3048 register (yet, still, not any subsequent ones). But, in that case, you're still
3049 probably better off simply splitting the value into two separate operands, for
3050 clarity. (e.g. see the description of the ``A`` constraint on X86, which,
3051 despite existing only for use with this feature, is not really a good idea to
3054 Indirect inputs and outputs
3055 """""""""""""""""""""""""""
3057 Indirect output or input constraints can be specified by the "``*``" modifier
3058 (which goes after the "``=``" in case of an output). This indicates that the asm
3059 will write to or read from the contents of an *address* provided as an input
3060 argument. (Note that in this way, indirect outputs act more like an *input* than
3061 an output: just like an input, they consume an argument of the call expression,
3062 rather than producing a return value. An indirect output constraint is an
3063 "output" only in that the asm is expected to write to the contents of the input
3064 memory location, instead of just read from it).
3066 This is most typically used for memory constraint, e.g. "``=*m``", to pass the
3067 address of a variable as a value.
3069 It is also possible to use an indirect *register* constraint, but only on output
3070 (e.g. "``=*r``"). This will cause LLVM to allocate a register for an output
3071 value normally, and then, separately emit a store to the address provided as
3072 input, after the provided inline asm. (It's not clear what value this
3073 functionality provides, compared to writing the store explicitly after the asm
3074 statement, and it can only produce worse code, since it bypasses many
3075 optimization passes. I would recommend not using it.)
3081 A clobber constraint is indicated by a "``~``" prefix. A clobber does not
3082 consume an input operand, nor generate an output. Clobbers cannot use any of the
3083 general constraint code letters -- they may use only explicit register
3084 constraints, e.g. "``~{eax}``". The one exception is that a clobber string of
3085 "``~{memory}``" indicates that the assembly writes to arbitrary undeclared
3086 memory locations -- not only the memory pointed to by a declared indirect
3092 After a potential prefix comes constraint code, or codes.
3094 A Constraint Code is either a single letter (e.g. "``r``"), a "``^``" character
3095 followed by two letters (e.g. "``^wc``"), or "``{``" register-name "``}``"
3098 The one and two letter constraint codes are typically chosen to be the same as
3099 GCC's constraint codes.
3101 A single constraint may include one or more than constraint code in it, leaving
3102 it up to LLVM to choose which one to use. This is included mainly for
3103 compatibility with the translation of GCC inline asm coming from clang.
3105 There are two ways to specify alternatives, and either or both may be used in an
3106 inline asm constraint list:
3108 1) Append the codes to each other, making a constraint code set. E.g. "``im``"
3109 or "``{eax}m``". This means "choose any of the options in the set". The
3110 choice of constraint is made independently for each constraint in the
3113 2) Use "``|``" between constraint code sets, creating alternatives. Every
3114 constraint in the constraint list must have the same number of alternative
3115 sets. With this syntax, the same alternative in *all* of the items in the
3116 constraint list will be chosen together.
3118 Putting those together, you might have a two operand constraint string like
3119 ``"rm|r,ri|rm"``. This indicates that if operand 0 is ``r`` or ``m``, then
3120 operand 1 may be one of ``r`` or ``i``. If operand 0 is ``r``, then operand 1
3121 may be one of ``r`` or ``m``. But, operand 0 and 1 cannot both be of type m.
3123 However, the use of either of the alternatives features is *NOT* recommended, as
3124 LLVM is not able to make an intelligent choice about which one to use. (At the
3125 point it currently needs to choose, not enough information is available to do so
3126 in a smart way.) Thus, it simply tries to make a choice that's most likely to
3127 compile, not one that will be optimal performance. (e.g., given "``rm``", it'll
3128 always choose to use memory, not registers). And, if given multiple registers,
3129 or multiple register classes, it will simply choose the first one. (In fact, it
3130 doesn't currently even ensure explicitly specified physical registers are
3131 unique, so specifying multiple physical registers as alternatives, like
3132 ``{r11}{r12},{r11}{r12}``, will assign r11 to both operands, not at all what was
3135 Supported Constraint Code List
3136 """"""""""""""""""""""""""""""
3138 The constraint codes are, in general, expected to behave the same way they do in
3139 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3140 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3141 and GCC likely indicates a bug in LLVM.
3143 Some constraint codes are typically supported by all targets:
3145 - ``r``: A register in the target's general purpose register class.
3146 - ``m``: A memory address operand. It is target-specific what addressing modes
3147 are supported, typical examples are register, or register + register offset,
3148 or register + immediate offset (of some target-specific size).
3149 - ``i``: An integer constant (of target-specific width). Allows either a simple
3150 immediate, or a relocatable value.
3151 - ``n``: An integer constant -- *not* including relocatable values.
3152 - ``s``: An integer constant, but allowing *only* relocatable values.
3153 - ``X``: Allows an operand of any kind, no constraint whatsoever. Typically
3154 useful to pass a label for an asm branch or call.
3156 .. FIXME: but that surely isn't actually okay to jump out of an asm
3157 block without telling llvm about the control transfer???)
3159 - ``{register-name}``: Requires exactly the named physical register.
3161 Other constraints are target-specific:
3165 - ``z``: An immediate integer 0. Outputs ``WZR`` or ``XZR``, as appropriate.
3166 - ``I``: An immediate integer valid for an ``ADD`` or ``SUB`` instruction,
3167 i.e. 0 to 4095 with optional shift by 12.
3168 - ``J``: An immediate integer that, when negated, is valid for an ``ADD`` or
3169 ``SUB`` instruction, i.e. -1 to -4095 with optional left shift by 12.
3170 - ``K``: An immediate integer that is valid for the 'bitmask immediate 32' of a
3171 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 32-bit register.
3172 - ``L``: An immediate integer that is valid for the 'bitmask immediate 64' of a
3173 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 64-bit register.
3174 - ``M``: An immediate integer for use with the ``MOV`` assembly alias on a
3175 32-bit register. This is a superset of ``K``: in addition to the bitmask
3176 immediate, also allows immediate integers which can be loaded with a single
3177 ``MOVZ`` or ``MOVL`` instruction.
3178 - ``N``: An immediate integer for use with the ``MOV`` assembly alias on a
3179 64-bit register. This is a superset of ``L``.
3180 - ``Q``: Memory address operand must be in a single register (no
3181 offsets). (However, LLVM currently does this for the ``m`` constraint as
3183 - ``r``: A 32 or 64-bit integer register (W* or X*).
3184 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register.
3185 - ``x``: A lower 128-bit floating-point/SIMD register (``V0`` to ``V15``).
3189 - ``r``: A 32 or 64-bit integer register.
3190 - ``[0-9]v``: The 32-bit VGPR register, number 0-9.
3191 - ``[0-9]s``: The 32-bit SGPR register, number 0-9.
3196 - ``Q``, ``Um``, ``Un``, ``Uq``, ``Us``, ``Ut``, ``Uv``, ``Uy``: Memory address
3197 operand. Treated the same as operand ``m``, at the moment.
3199 ARM and ARM's Thumb2 mode:
3201 - ``j``: An immediate integer between 0 and 65535 (valid for ``MOVW``)
3202 - ``I``: An immediate integer valid for a data-processing instruction.
3203 - ``J``: An immediate integer between -4095 and 4095.
3204 - ``K``: An immediate integer whose bitwise inverse is valid for a
3205 data-processing instruction. (Can be used with template modifier "``B``" to
3206 print the inverted value).
3207 - ``L``: An immediate integer whose negation is valid for a data-processing
3208 instruction. (Can be used with template modifier "``n``" to print the negated
3210 - ``M``: A power of two or a integer between 0 and 32.
3211 - ``N``: Invalid immediate constraint.
3212 - ``O``: Invalid immediate constraint.
3213 - ``r``: A general-purpose 32-bit integer register (``r0-r15``).
3214 - ``l``: In Thumb2 mode, low 32-bit GPR registers (``r0-r7``). In ARM mode, same
3216 - ``h``: In Thumb2 mode, a high 32-bit GPR register (``r8-r15``). In ARM mode,
3218 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3219 ``d0-d31``, or ``q0-q15``.
3220 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3221 ``d0-d7``, or ``q0-q3``.
3222 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3227 - ``I``: An immediate integer between 0 and 255.
3228 - ``J``: An immediate integer between -255 and -1.
3229 - ``K``: An immediate integer between 0 and 255, with optional left-shift by
3231 - ``L``: An immediate integer between -7 and 7.
3232 - ``M``: An immediate integer which is a multiple of 4 between 0 and 1020.
3233 - ``N``: An immediate integer between 0 and 31.
3234 - ``O``: An immediate integer which is a multiple of 4 between -508 and 508.
3235 - ``r``: A low 32-bit GPR register (``r0-r7``).
3236 - ``l``: A low 32-bit GPR register (``r0-r7``).
3237 - ``h``: A high GPR register (``r0-r7``).
3238 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3239 ``d0-d31``, or ``q0-q15``.
3240 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3241 ``d0-d7``, or ``q0-q3``.
3242 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3248 - ``o``, ``v``: A memory address operand, treated the same as constraint ``m``,
3250 - ``r``: A 32 or 64-bit register.
3254 - ``r``: An 8 or 16-bit register.
3258 - ``I``: An immediate signed 16-bit integer.
3259 - ``J``: An immediate integer zero.
3260 - ``K``: An immediate unsigned 16-bit integer.
3261 - ``L``: An immediate 32-bit integer, where the lower 16 bits are 0.
3262 - ``N``: An immediate integer between -65535 and -1.
3263 - ``O``: An immediate signed 15-bit integer.
3264 - ``P``: An immediate integer between 1 and 65535.
3265 - ``m``: A memory address operand. In MIPS-SE mode, allows a base address
3266 register plus 16-bit immediate offset. In MIPS mode, just a base register.
3267 - ``R``: A memory address operand. In MIPS-SE mode, allows a base address
3268 register plus a 9-bit signed offset. In MIPS mode, the same as constraint
3270 - ``ZC``: A memory address operand, suitable for use in a ``pref``, ``ll``, or
3271 ``sc`` instruction on the given subtarget (details vary).
3272 - ``r``, ``d``, ``y``: A 32 or 64-bit GPR register.
3273 - ``f``: A 32 or 64-bit FPU register (``F0-F31``), or a 128-bit MSA register
3274 (``W0-W31``). In the case of MSA registers, it is recommended to use the ``w``
3275 argument modifier for compatibility with GCC.
3276 - ``c``: A 32-bit or 64-bit GPR register suitable for indirect jump (always
3278 - ``l``: The ``lo`` register, 32 or 64-bit.
3283 - ``b``: A 1-bit integer register.
3284 - ``c`` or ``h``: A 16-bit integer register.
3285 - ``r``: A 32-bit integer register.
3286 - ``l`` or ``N``: A 64-bit integer register.
3287 - ``f``: A 32-bit float register.
3288 - ``d``: A 64-bit float register.
3293 - ``I``: An immediate signed 16-bit integer.
3294 - ``J``: An immediate unsigned 16-bit integer, shifted left 16 bits.
3295 - ``K``: An immediate unsigned 16-bit integer.
3296 - ``L``: An immediate signed 16-bit integer, shifted left 16 bits.
3297 - ``M``: An immediate integer greater than 31.
3298 - ``N``: An immediate integer that is an exact power of 2.
3299 - ``O``: The immediate integer constant 0.
3300 - ``P``: An immediate integer constant whose negation is a signed 16-bit
3302 - ``es``, ``o``, ``Q``, ``Z``, ``Zy``: A memory address operand, currently
3303 treated the same as ``m``.
3304 - ``r``: A 32 or 64-bit integer register.
3305 - ``b``: A 32 or 64-bit integer register, excluding ``R0`` (that is:
3307 - ``f``: A 32 or 64-bit float register (``F0-F31``), or when QPX is enabled, a
3308 128 or 256-bit QPX register (``Q0-Q31``; aliases the ``F`` registers).
3309 - ``v``: For ``4 x f32`` or ``4 x f64`` types, when QPX is enabled, a
3310 128 or 256-bit QPX register (``Q0-Q31``), otherwise a 128-bit
3311 altivec vector register (``V0-V31``).
3313 .. FIXME: is this a bug that v accepts QPX registers? I think this
3314 is supposed to only use the altivec vector registers?
3316 - ``y``: Condition register (``CR0-CR7``).
3317 - ``wc``: An individual CR bit in a CR register.
3318 - ``wa``, ``wd``, ``wf``: Any 128-bit VSX vector register, from the full VSX
3319 register set (overlapping both the floating-point and vector register files).
3320 - ``ws``: A 32 or 64-bit floating point register, from the full VSX register
3325 - ``I``: An immediate 13-bit signed integer.
3326 - ``r``: A 32-bit integer register.
3330 - ``I``: An immediate unsigned 8-bit integer.
3331 - ``J``: An immediate unsigned 12-bit integer.
3332 - ``K``: An immediate signed 16-bit integer.
3333 - ``L``: An immediate signed 20-bit integer.
3334 - ``M``: An immediate integer 0x7fffffff.
3335 - ``Q``, ``R``, ``S``, ``T``: A memory address operand, treated the same as
3336 ``m``, at the moment.
3337 - ``r`` or ``d``: A 32, 64, or 128-bit integer register.
3338 - ``a``: A 32, 64, or 128-bit integer address register (excludes R0, which in an
3339 address context evaluates as zero).
3340 - ``h``: A 32-bit value in the high part of a 64bit data register
3342 - ``f``: A 32, 64, or 128-bit floating point register.
3346 - ``I``: An immediate integer between 0 and 31.
3347 - ``J``: An immediate integer between 0 and 64.
3348 - ``K``: An immediate signed 8-bit integer.
3349 - ``L``: An immediate integer, 0xff or 0xffff or (in 64-bit mode only)
3351 - ``M``: An immediate integer between 0 and 3.
3352 - ``N``: An immediate unsigned 8-bit integer.
3353 - ``O``: An immediate integer between 0 and 127.
3354 - ``e``: An immediate 32-bit signed integer.
3355 - ``Z``: An immediate 32-bit unsigned integer.
3356 - ``o``, ``v``: Treated the same as ``m``, at the moment.
3357 - ``q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3358 ``l`` integer register. On X86-32, this is the ``a``, ``b``, ``c``, and ``d``
3359 registers, and on X86-64, it is all of the integer registers.
3360 - ``Q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3361 ``h`` integer register. This is the ``a``, ``b``, ``c``, and ``d`` registers.
3362 - ``r`` or ``l``: An 8, 16, 32, or 64-bit integer register.
3363 - ``R``: An 8, 16, 32, or 64-bit "legacy" integer register -- one which has
3364 existed since i386, and can be accessed without the REX prefix.
3365 - ``f``: A 32, 64, or 80-bit '387 FPU stack pseudo-register.
3366 - ``y``: A 64-bit MMX register, if MMX is enabled.
3367 - ``x``: If SSE is enabled: a 32 or 64-bit scalar operand, or 128-bit vector
3368 operand in a SSE register. If AVX is also enabled, can also be a 256-bit
3369 vector operand in an AVX register. If AVX-512 is also enabled, can also be a
3370 512-bit vector operand in an AVX512 register, Otherwise, an error.
3371 - ``Y``: The same as ``x``, if *SSE2* is enabled, otherwise an error.
3372 - ``A``: Special case: allocates EAX first, then EDX, for a single operand (in
3373 32-bit mode, a 64-bit integer operand will get split into two registers). It
3374 is not recommended to use this constraint, as in 64-bit mode, the 64-bit
3375 operand will get allocated only to RAX -- if two 32-bit operands are needed,
3376 you're better off splitting it yourself, before passing it to the asm
3381 - ``r``: A 32-bit integer register.
3384 .. _inline-asm-modifiers:
3386 Asm template argument modifiers
3387 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3389 In the asm template string, modifiers can be used on the operand reference, like
3392 The modifiers are, in general, expected to behave the same way they do in
3393 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3394 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3395 and GCC likely indicates a bug in LLVM.
3399 - ``c``: Print an immediate integer constant unadorned, without
3400 the target-specific immediate punctuation (e.g. no ``$`` prefix).
3401 - ``n``: Negate and print immediate integer constant unadorned, without the
3402 target-specific immediate punctuation (e.g. no ``$`` prefix).
3403 - ``l``: Print as an unadorned label, without the target-specific label
3404 punctuation (e.g. no ``$`` prefix).
3408 - ``w``: Print a GPR register with a ``w*`` name instead of ``x*`` name. E.g.,
3409 instead of ``x30``, print ``w30``.
3410 - ``x``: Print a GPR register with a ``x*`` name. (this is the default, anyhow).
3411 - ``b``, ``h``, ``s``, ``d``, ``q``: Print a floating-point/SIMD register with a
3412 ``b*``, ``h*``, ``s*``, ``d*``, or ``q*`` name, rather than the default of
3421 - ``a``: Print an operand as an address (with ``[`` and ``]`` surrounding a
3425 - ``y``: Print a VFP single-precision register as an indexed double (e.g. print
3426 as ``d4[1]`` instead of ``s9``)
3427 - ``B``: Bitwise invert and print an immediate integer constant without ``#``
3429 - ``L``: Print the low 16-bits of an immediate integer constant.
3430 - ``M``: Print as a register set suitable for ldm/stm. Also prints *all*
3431 register operands subsequent to the specified one (!), so use carefully.
3432 - ``Q``: Print the low-order register of a register-pair, or the low-order
3433 register of a two-register operand.
3434 - ``R``: Print the high-order register of a register-pair, or the high-order
3435 register of a two-register operand.
3436 - ``H``: Print the second register of a register-pair. (On a big-endian system,
3437 ``H`` is equivalent to ``Q``, and on little-endian system, ``H`` is equivalent
3440 .. FIXME: H doesn't currently support printing the second register
3441 of a two-register operand.
3443 - ``e``: Print the low doubleword register of a NEON quad register.
3444 - ``f``: Print the high doubleword register of a NEON quad register.
3445 - ``m``: Print the base register of a memory operand without the ``[`` and ``]``
3450 - ``L``: Print the second register of a two-register operand. Requires that it
3451 has been allocated consecutively to the first.
3453 .. FIXME: why is it restricted to consecutive ones? And there's
3454 nothing that ensures that happens, is there?
3456 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3457 nothing. Used to print 'addi' vs 'add' instructions.
3461 No additional modifiers.
3465 - ``X``: Print an immediate integer as hexadecimal
3466 - ``x``: Print the low 16 bits of an immediate integer as hexadecimal.
3467 - ``d``: Print an immediate integer as decimal.
3468 - ``m``: Subtract one and print an immediate integer as decimal.
3469 - ``z``: Print $0 if an immediate zero, otherwise print normally.
3470 - ``L``: Print the low-order register of a two-register operand, or prints the
3471 address of the low-order word of a double-word memory operand.
3473 .. FIXME: L seems to be missing memory operand support.
3475 - ``M``: Print the high-order register of a two-register operand, or prints the
3476 address of the high-order word of a double-word memory operand.
3478 .. FIXME: M seems to be missing memory operand support.
3480 - ``D``: Print the second register of a two-register operand, or prints the
3481 second word of a double-word memory operand. (On a big-endian system, ``D`` is
3482 equivalent to ``L``, and on little-endian system, ``D`` is equivalent to
3484 - ``w``: No effect. Provided for compatibility with GCC which requires this
3485 modifier in order to print MSA registers (``W0-W31``) with the ``f``
3494 - ``L``: Print the second register of a two-register operand. Requires that it
3495 has been allocated consecutively to the first.
3497 .. FIXME: why is it restricted to consecutive ones? And there's
3498 nothing that ensures that happens, is there?
3500 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3501 nothing. Used to print 'addi' vs 'add' instructions.
3502 - ``y``: For a memory operand, prints formatter for a two-register X-form
3503 instruction. (Currently always prints ``r0,OPERAND``).
3504 - ``U``: Prints 'u' if the memory operand is an update form, and nothing
3505 otherwise. (NOTE: LLVM does not support update form, so this will currently
3506 always print nothing)
3507 - ``X``: Prints 'x' if the memory operand is an indexed form. (NOTE: LLVM does
3508 not support indexed form, so this will currently always print nothing)
3516 SystemZ implements only ``n``, and does *not* support any of the other
3517 target-independent modifiers.
3521 - ``c``: Print an unadorned integer or symbol name. (The latter is
3522 target-specific behavior for this typically target-independent modifier).
3523 - ``A``: Print a register name with a '``*``' before it.
3524 - ``b``: Print an 8-bit register name (e.g. ``al``); do nothing on a memory
3526 - ``h``: Print the upper 8-bit register name (e.g. ``ah``); do nothing on a
3528 - ``w``: Print the 16-bit register name (e.g. ``ax``); do nothing on a memory
3530 - ``k``: Print the 32-bit register name (e.g. ``eax``); do nothing on a memory
3532 - ``q``: Print the 64-bit register name (e.g. ``rax``), if 64-bit registers are
3533 available, otherwise the 32-bit register name; do nothing on a memory operand.
3534 - ``n``: Negate and print an unadorned integer, or, for operands other than an
3535 immediate integer (e.g. a relocatable symbol expression), print a '-' before
3536 the operand. (The behavior for relocatable symbol expressions is a
3537 target-specific behavior for this typically target-independent modifier)
3538 - ``H``: Print a memory reference with additional offset +8.
3539 - ``P``: Print a memory reference or operand for use as the argument of a call
3540 instruction. (E.g. omit ``(rip)``, even though it's PC-relative.)
3544 No additional modifiers.
3550 The call instructions that wrap inline asm nodes may have a
3551 "``!srcloc``" MDNode attached to it that contains a list of constant
3552 integers. If present, the code generator will use the integer as the
3553 location cookie value when report errors through the ``LLVMContext``
3554 error reporting mechanisms. This allows a front-end to correlate backend
3555 errors that occur with inline asm back to the source code that produced
3558 .. code-block:: llvm
3560 call void asm sideeffect "something bad", ""(), !srcloc !42
3562 !42 = !{ i32 1234567 }
3564 It is up to the front-end to make sense of the magic numbers it places
3565 in the IR. If the MDNode contains multiple constants, the code generator
3566 will use the one that corresponds to the line of the asm that the error
3574 LLVM IR allows metadata to be attached to instructions in the program
3575 that can convey extra information about the code to the optimizers and
3576 code generator. One example application of metadata is source-level
3577 debug information. There are two metadata primitives: strings and nodes.
3579 Metadata does not have a type, and is not a value. If referenced from a
3580 ``call`` instruction, it uses the ``metadata`` type.
3582 All metadata are identified in syntax by a exclamation point ('``!``').
3584 .. _metadata-string:
3586 Metadata Nodes and Metadata Strings
3587 -----------------------------------
3589 A metadata string is a string surrounded by double quotes. It can
3590 contain any character by escaping non-printable characters with
3591 "``\xx``" where "``xx``" is the two digit hex code. For example:
3594 Metadata nodes are represented with notation similar to structure
3595 constants (a comma separated list of elements, surrounded by braces and
3596 preceded by an exclamation point). Metadata nodes can have any values as
3597 their operand. For example:
3599 .. code-block:: llvm
3601 !{ !"test\00", i32 10}
3603 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
3605 .. code-block:: llvm
3607 !0 = distinct !{!"test\00", i32 10}
3609 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
3610 content. They can also occur when transformations cause uniquing collisions
3611 when metadata operands change.
3613 A :ref:`named metadata <namedmetadatastructure>` is a collection of
3614 metadata nodes, which can be looked up in the module symbol table. For
3617 .. code-block:: llvm
3621 Metadata can be used as function arguments. Here ``llvm.dbg.value``
3622 function is using two metadata arguments:
3624 .. code-block:: llvm
3626 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
3628 Metadata can be attached to an instruction. Here metadata ``!21`` is attached
3629 to the ``add`` instruction using the ``!dbg`` identifier:
3631 .. code-block:: llvm
3633 %indvar.next = add i64 %indvar, 1, !dbg !21
3635 Metadata can also be attached to a function definition. Here metadata ``!22``
3636 is attached to the ``foo`` function using the ``!dbg`` identifier:
3638 .. code-block:: llvm
3640 define void @foo() !dbg !22 {
3644 More information about specific metadata nodes recognized by the
3645 optimizers and code generator is found below.
3647 .. _specialized-metadata:
3649 Specialized Metadata Nodes
3650 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3652 Specialized metadata nodes are custom data structures in metadata (as opposed
3653 to generic tuples). Their fields are labelled, and can be specified in any
3656 These aren't inherently debug info centric, but currently all the specialized
3657 metadata nodes are related to debug info.
3664 ``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
3665 ``retainedTypes:``, ``subprograms:``, ``globals:`` and ``imports:`` fields are
3666 tuples containing the debug info to be emitted along with the compile unit,
3667 regardless of code optimizations (some nodes are only emitted if there are
3668 references to them from instructions).
3670 .. code-block:: llvm
3672 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
3673 isOptimized: true, flags: "-O2", runtimeVersion: 2,
3674 splitDebugFilename: "abc.debug", emissionKind: 1,
3675 enums: !2, retainedTypes: !3, subprograms: !4,
3676 globals: !5, imports: !6)
3678 Compile unit descriptors provide the root scope for objects declared in a
3679 specific compilation unit. File descriptors are defined using this scope.
3680 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
3681 keep track of subprograms, global variables, type information, and imported
3682 entities (declarations and namespaces).
3689 ``DIFile`` nodes represent files. The ``filename:`` can include slashes.
3691 .. code-block:: llvm
3693 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
3695 Files are sometimes used in ``scope:`` fields, and are the only valid target
3696 for ``file:`` fields.
3703 ``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
3704 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
3706 .. code-block:: llvm
3708 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3709 encoding: DW_ATE_unsigned_char)
3710 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
3712 The ``encoding:`` describes the details of the type. Usually it's one of the
3715 .. code-block:: llvm
3721 DW_ATE_signed_char = 6
3723 DW_ATE_unsigned_char = 8
3725 .. _DISubroutineType:
3730 ``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
3731 refers to a tuple; the first operand is the return type, while the rest are the
3732 types of the formal arguments in order. If the first operand is ``null``, that
3733 represents a function with no return value (such as ``void foo() {}`` in C++).
3735 .. code-block:: llvm
3737 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
3738 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
3739 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
3746 ``DIDerivedType`` nodes represent types derived from other types, such as
3749 .. code-block:: llvm
3751 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3752 encoding: DW_ATE_unsigned_char)
3753 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
3756 The following ``tag:`` values are valid:
3758 .. code-block:: llvm
3760 DW_TAG_formal_parameter = 5
3762 DW_TAG_pointer_type = 15
3763 DW_TAG_reference_type = 16
3765 DW_TAG_ptr_to_member_type = 31
3766 DW_TAG_const_type = 38
3767 DW_TAG_volatile_type = 53
3768 DW_TAG_restrict_type = 55
3770 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
3771 <DICompositeType>` or :ref:`subprogram <DISubprogram>`. The type of the member
3772 is the ``baseType:``. The ``offset:`` is the member's bit offset.
3773 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
3774 argument of a subprogram.
3776 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3778 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3779 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3782 Note that the ``void *`` type is expressed as a type derived from NULL.
3784 .. _DICompositeType:
3789 ``DICompositeType`` nodes represent types composed of other types, like
3790 structures and unions. ``elements:`` points to a tuple of the composed types.
3792 If the source language supports ODR, the ``identifier:`` field gives the unique
3793 identifier used for type merging between modules. When specified, other types
3794 can refer to composite types indirectly via a :ref:`metadata string
3795 <metadata-string>` that matches their identifier.
3797 .. code-block:: llvm
3799 !0 = !DIEnumerator(name: "SixKind", value: 7)
3800 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3801 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3802 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
3803 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3804 elements: !{!0, !1, !2})
3806 The following ``tag:`` values are valid:
3808 .. code-block:: llvm
3810 DW_TAG_array_type = 1
3811 DW_TAG_class_type = 2
3812 DW_TAG_enumeration_type = 4
3813 DW_TAG_structure_type = 19
3814 DW_TAG_union_type = 23
3815 DW_TAG_subroutine_type = 21
3816 DW_TAG_inheritance = 28
3819 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
3820 descriptors <DISubrange>`, each representing the range of subscripts at that
3821 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3822 array type is a native packed vector.
3824 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
3825 descriptors <DIEnumerator>`, each representing the definition of an enumeration
3826 value for the set. All enumeration type descriptors are collected in the
3827 ``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
3829 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3830 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
3831 <DIDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
3838 ``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3839 :ref:`DICompositeType`. ``count: -1`` indicates an empty array.
3841 .. code-block:: llvm
3843 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
3844 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
3845 !2 = !DISubrange(count: -1) ; empty array.
3852 ``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3853 variants of :ref:`DICompositeType`.
3855 .. code-block:: llvm
3857 !0 = !DIEnumerator(name: "SixKind", value: 7)
3858 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3859 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3861 DITemplateTypeParameter
3862 """""""""""""""""""""""
3864 ``DITemplateTypeParameter`` nodes represent type parameters to generic source
3865 language constructs. They are used (optionally) in :ref:`DICompositeType` and
3866 :ref:`DISubprogram` ``templateParams:`` fields.
3868 .. code-block:: llvm
3870 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
3872 DITemplateValueParameter
3873 """"""""""""""""""""""""
3875 ``DITemplateValueParameter`` nodes represent value parameters to generic source
3876 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3877 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3878 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
3879 :ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
3881 .. code-block:: llvm
3883 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
3888 ``DINamespace`` nodes represent namespaces in the source language.
3890 .. code-block:: llvm
3892 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
3897 ``DIGlobalVariable`` nodes represent global variables in the source language.
3899 .. code-block:: llvm
3901 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
3902 file: !2, line: 7, type: !3, isLocal: true,
3903 isDefinition: false, variable: i32* @foo,
3906 All global variables should be referenced by the `globals:` field of a
3907 :ref:`compile unit <DICompileUnit>`.
3914 ``DISubprogram`` nodes represent functions from the source language. A
3915 ``DISubprogram`` may be attached to a function definition using ``!dbg``
3916 metadata. The ``variables:`` field points at :ref:`variables <DILocalVariable>`
3917 that must be retained, even if their IR counterparts are optimized out of
3918 the IR. The ``type:`` field must point at an :ref:`DISubroutineType`.
3920 .. code-block:: llvm
3922 define void @_Z3foov() !dbg !0 {
3926 !0 = distinct !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
3927 file: !2, line: 7, type: !3, isLocal: true,
3928 isDefinition: false, scopeLine: 8,
3930 virtuality: DW_VIRTUALITY_pure_virtual,
3931 virtualIndex: 10, flags: DIFlagPrototyped,
3932 isOptimized: true, templateParams: !5,
3933 declaration: !6, variables: !7)
3940 ``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
3941 <DISubprogram>`. The line number and column numbers are used to distinguish
3942 two lexical blocks at same depth. They are valid targets for ``scope:``
3945 .. code-block:: llvm
3947 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
3949 Usually lexical blocks are ``distinct`` to prevent node merging based on
3952 .. _DILexicalBlockFile:
3957 ``DILexicalBlockFile`` nodes are used to discriminate between sections of a
3958 :ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
3959 indicate textual inclusion, or the ``discriminator:`` field can be used to
3960 discriminate between control flow within a single block in the source language.
3962 .. code-block:: llvm
3964 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
3965 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
3966 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
3973 ``DILocation`` nodes represent source debug locations. The ``scope:`` field is
3974 mandatory, and points at an :ref:`DILexicalBlockFile`, an
3975 :ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
3977 .. code-block:: llvm
3979 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
3981 .. _DILocalVariable:
3986 ``DILocalVariable`` nodes represent local variables in the source language. If
3987 the ``arg:`` field is set to non-zero, then this variable is a subprogram
3988 parameter, and it will be included in the ``variables:`` field of its
3989 :ref:`DISubprogram`.
3991 .. code-block:: llvm
3993 !0 = !DILocalVariable(name: "this", arg: 1, scope: !3, file: !2, line: 7,
3994 type: !3, flags: DIFlagArtificial)
3995 !1 = !DILocalVariable(name: "x", arg: 2, scope: !4, file: !2, line: 7,
3997 !2 = !DILocalVariable(name: "y", scope: !5, file: !2, line: 7, type: !3)
4002 ``DIExpression`` nodes represent DWARF expression sequences. They are used in
4003 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
4004 describe how the referenced LLVM variable relates to the source language
4007 The current supported vocabulary is limited:
4009 - ``DW_OP_deref`` dereferences the working expression.
4010 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
4011 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
4012 here, respectively) of the variable piece from the working expression.
4014 .. code-block:: llvm
4016 !0 = !DIExpression(DW_OP_deref)
4017 !1 = !DIExpression(DW_OP_plus, 3)
4018 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
4019 !3 = !DIExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
4024 ``DIObjCProperty`` nodes represent Objective-C property nodes.
4026 .. code-block:: llvm
4028 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
4029 getter: "getFoo", attributes: 7, type: !2)
4034 ``DIImportedEntity`` nodes represent entities (such as modules) imported into a
4037 .. code-block:: llvm
4039 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
4040 entity: !1, line: 7)
4045 In LLVM IR, memory does not have types, so LLVM's own type system is not
4046 suitable for doing TBAA. Instead, metadata is added to the IR to
4047 describe a type system of a higher level language. This can be used to
4048 implement typical C/C++ TBAA, but it can also be used to implement
4049 custom alias analysis behavior for other languages.
4051 The current metadata format is very simple. TBAA metadata nodes have up
4052 to three fields, e.g.:
4054 .. code-block:: llvm
4056 !0 = !{ !"an example type tree" }
4057 !1 = !{ !"int", !0 }
4058 !2 = !{ !"float", !0 }
4059 !3 = !{ !"const float", !2, i64 1 }
4061 The first field is an identity field. It can be any value, usually a
4062 metadata string, which uniquely identifies the type. The most important
4063 name in the tree is the name of the root node. Two trees with different
4064 root node names are entirely disjoint, even if they have leaves with
4067 The second field identifies the type's parent node in the tree, or is
4068 null or omitted for a root node. A type is considered to alias all of
4069 its descendants and all of its ancestors in the tree. Also, a type is
4070 considered to alias all types in other trees, so that bitcode produced
4071 from multiple front-ends is handled conservatively.
4073 If the third field is present, it's an integer which if equal to 1
4074 indicates that the type is "constant" (meaning
4075 ``pointsToConstantMemory`` should return true; see `other useful
4076 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
4078 '``tbaa.struct``' Metadata
4079 ^^^^^^^^^^^^^^^^^^^^^^^^^^
4081 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
4082 aggregate assignment operations in C and similar languages, however it
4083 is defined to copy a contiguous region of memory, which is more than
4084 strictly necessary for aggregate types which contain holes due to
4085 padding. Also, it doesn't contain any TBAA information about the fields
4088 ``!tbaa.struct`` metadata can describe which memory subregions in a
4089 memcpy are padding and what the TBAA tags of the struct are.
4091 The current metadata format is very simple. ``!tbaa.struct`` metadata
4092 nodes are a list of operands which are in conceptual groups of three.
4093 For each group of three, the first operand gives the byte offset of a
4094 field in bytes, the second gives its size in bytes, and the third gives
4097 .. code-block:: llvm
4099 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
4101 This describes a struct with two fields. The first is at offset 0 bytes
4102 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
4103 and has size 4 bytes and has tbaa tag !2.
4105 Note that the fields need not be contiguous. In this example, there is a
4106 4 byte gap between the two fields. This gap represents padding which
4107 does not carry useful data and need not be preserved.
4109 '``noalias``' and '``alias.scope``' Metadata
4110 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4112 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
4113 noalias memory-access sets. This means that some collection of memory access
4114 instructions (loads, stores, memory-accessing calls, etc.) that carry
4115 ``noalias`` metadata can specifically be specified not to alias with some other
4116 collection of memory access instructions that carry ``alias.scope`` metadata.
4117 Each type of metadata specifies a list of scopes where each scope has an id and
4118 a domain. When evaluating an aliasing query, if for some domain, the set
4119 of scopes with that domain in one instruction's ``alias.scope`` list is a
4120 subset of (or equal to) the set of scopes for that domain in another
4121 instruction's ``noalias`` list, then the two memory accesses are assumed not to
4124 The metadata identifying each domain is itself a list containing one or two
4125 entries. The first entry is the name of the domain. Note that if the name is a
4126 string then it can be combined across functions and translation units. A
4127 self-reference can be used to create globally unique domain names. A
4128 descriptive string may optionally be provided as a second list entry.
4130 The metadata identifying each scope is also itself a list containing two or
4131 three entries. The first entry is the name of the scope. Note that if the name
4132 is a string then it can be combined across functions and translation units. A
4133 self-reference can be used to create globally unique scope names. A metadata
4134 reference to the scope's domain is the second entry. A descriptive string may
4135 optionally be provided as a third list entry.
4139 .. code-block:: llvm
4141 ; Two scope domains:
4145 ; Some scopes in these domains:
4151 !5 = !{!4} ; A list containing only scope !4
4155 ; These two instructions don't alias:
4156 %0 = load float, float* %c, align 4, !alias.scope !5
4157 store float %0, float* %arrayidx.i, align 4, !noalias !5
4159 ; These two instructions also don't alias (for domain !1, the set of scopes
4160 ; in the !alias.scope equals that in the !noalias list):
4161 %2 = load float, float* %c, align 4, !alias.scope !5
4162 store float %2, float* %arrayidx.i2, align 4, !noalias !6
4164 ; These two instructions may alias (for domain !0, the set of scopes in
4165 ; the !noalias list is not a superset of, or equal to, the scopes in the
4166 ; !alias.scope list):
4167 %2 = load float, float* %c, align 4, !alias.scope !6
4168 store float %0, float* %arrayidx.i, align 4, !noalias !7
4170 '``fpmath``' Metadata
4171 ^^^^^^^^^^^^^^^^^^^^^
4173 ``fpmath`` metadata may be attached to any instruction of floating point
4174 type. It can be used to express the maximum acceptable error in the
4175 result of that instruction, in ULPs, thus potentially allowing the
4176 compiler to use a more efficient but less accurate method of computing
4177 it. ULP is defined as follows:
4179 If ``x`` is a real number that lies between two finite consecutive
4180 floating-point numbers ``a`` and ``b``, without being equal to one
4181 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
4182 distance between the two non-equal finite floating-point numbers
4183 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
4185 The metadata node shall consist of a single positive floating point
4186 number representing the maximum relative error, for example:
4188 .. code-block:: llvm
4190 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
4194 '``range``' Metadata
4195 ^^^^^^^^^^^^^^^^^^^^
4197 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
4198 integer types. It expresses the possible ranges the loaded value or the value
4199 returned by the called function at this call site is in. The ranges are
4200 represented with a flattened list of integers. The loaded value or the value
4201 returned is known to be in the union of the ranges defined by each consecutive
4202 pair. Each pair has the following properties:
4204 - The type must match the type loaded by the instruction.
4205 - The pair ``a,b`` represents the range ``[a,b)``.
4206 - Both ``a`` and ``b`` are constants.
4207 - The range is allowed to wrap.
4208 - The range should not represent the full or empty set. That is,
4211 In addition, the pairs must be in signed order of the lower bound and
4212 they must be non-contiguous.
4216 .. code-block:: llvm
4218 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
4219 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
4220 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
4221 %d = invoke i8 @bar() to label %cont
4222 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
4224 !0 = !{ i8 0, i8 2 }
4225 !1 = !{ i8 255, i8 2 }
4226 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
4227 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
4229 '``unpredictable``' Metadata
4230 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4232 ``unpredictable`` metadata may be attached to any branch or switch
4233 instruction. It can be used to express the unpredictability of control
4234 flow. Similar to the llvm.expect intrinsic, it may be used to alter
4235 optimizations related to compare and branch instructions. The metadata
4236 is treated as a boolean value; if it exists, it signals that the branch
4237 or switch that it is attached to is completely unpredictable.
4242 It is sometimes useful to attach information to loop constructs. Currently,
4243 loop metadata is implemented as metadata attached to the branch instruction
4244 in the loop latch block. This type of metadata refer to a metadata node that is
4245 guaranteed to be separate for each loop. The loop identifier metadata is
4246 specified with the name ``llvm.loop``.
4248 The loop identifier metadata is implemented using a metadata that refers to
4249 itself to avoid merging it with any other identifier metadata, e.g.,
4250 during module linkage or function inlining. That is, each loop should refer
4251 to their own identification metadata even if they reside in separate functions.
4252 The following example contains loop identifier metadata for two separate loop
4255 .. code-block:: llvm
4260 The loop identifier metadata can be used to specify additional
4261 per-loop metadata. Any operands after the first operand can be treated
4262 as user-defined metadata. For example the ``llvm.loop.unroll.count``
4263 suggests an unroll factor to the loop unroller:
4265 .. code-block:: llvm
4267 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
4270 !1 = !{!"llvm.loop.unroll.count", i32 4}
4272 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
4273 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4275 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
4276 used to control per-loop vectorization and interleaving parameters such as
4277 vectorization width and interleave count. These metadata should be used in
4278 conjunction with ``llvm.loop`` loop identification metadata. The
4279 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
4280 optimization hints and the optimizer will only interleave and vectorize loops if
4281 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
4282 which contains information about loop-carried memory dependencies can be helpful
4283 in determining the safety of these transformations.
4285 '``llvm.loop.interleave.count``' Metadata
4286 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4288 This metadata suggests an interleave count to the loop interleaver.
4289 The first operand is the string ``llvm.loop.interleave.count`` and the
4290 second operand is an integer specifying the interleave count. For
4293 .. code-block:: llvm
4295 !0 = !{!"llvm.loop.interleave.count", i32 4}
4297 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
4298 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
4299 then the interleave count will be determined automatically.
4301 '``llvm.loop.vectorize.enable``' Metadata
4302 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4304 This metadata selectively enables or disables vectorization for the loop. The
4305 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
4306 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
4307 0 disables vectorization:
4309 .. code-block:: llvm
4311 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
4312 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
4314 '``llvm.loop.vectorize.width``' Metadata
4315 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4317 This metadata sets the target width of the vectorizer. The first
4318 operand is the string ``llvm.loop.vectorize.width`` and the second
4319 operand is an integer specifying the width. For example:
4321 .. code-block:: llvm
4323 !0 = !{!"llvm.loop.vectorize.width", i32 4}
4325 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
4326 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
4327 0 or if the loop does not have this metadata the width will be
4328 determined automatically.
4330 '``llvm.loop.unroll``'
4331 ^^^^^^^^^^^^^^^^^^^^^^
4333 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
4334 optimization hints such as the unroll factor. ``llvm.loop.unroll``
4335 metadata should be used in conjunction with ``llvm.loop`` loop
4336 identification metadata. The ``llvm.loop.unroll`` metadata are only
4337 optimization hints and the unrolling will only be performed if the
4338 optimizer believes it is safe to do so.
4340 '``llvm.loop.unroll.count``' Metadata
4341 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4343 This metadata suggests an unroll factor to the loop unroller. The
4344 first operand is the string ``llvm.loop.unroll.count`` and the second
4345 operand is a positive integer specifying the unroll factor. For
4348 .. code-block:: llvm
4350 !0 = !{!"llvm.loop.unroll.count", i32 4}
4352 If the trip count of the loop is less than the unroll count the loop
4353 will be partially unrolled.
4355 '``llvm.loop.unroll.disable``' Metadata
4356 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4358 This metadata disables loop unrolling. The metadata has a single operand
4359 which is the string ``llvm.loop.unroll.disable``. For example:
4361 .. code-block:: llvm
4363 !0 = !{!"llvm.loop.unroll.disable"}
4365 '``llvm.loop.unroll.runtime.disable``' Metadata
4366 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4368 This metadata disables runtime loop unrolling. The metadata has a single
4369 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
4371 .. code-block:: llvm
4373 !0 = !{!"llvm.loop.unroll.runtime.disable"}
4375 '``llvm.loop.unroll.enable``' Metadata
4376 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4378 This metadata suggests that the loop should be fully unrolled if the trip count
4379 is known at compile time and partially unrolled if the trip count is not known
4380 at compile time. The metadata has a single operand which is the string
4381 ``llvm.loop.unroll.enable``. For example:
4383 .. code-block:: llvm
4385 !0 = !{!"llvm.loop.unroll.enable"}
4387 '``llvm.loop.unroll.full``' Metadata
4388 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4390 This metadata suggests that the loop should be unrolled fully. The
4391 metadata has a single operand which is the string ``llvm.loop.unroll.full``.
4394 .. code-block:: llvm
4396 !0 = !{!"llvm.loop.unroll.full"}
4401 Metadata types used to annotate memory accesses with information helpful
4402 for optimizations are prefixed with ``llvm.mem``.
4404 '``llvm.mem.parallel_loop_access``' Metadata
4405 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4407 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
4408 or metadata containing a list of loop identifiers for nested loops.
4409 The metadata is attached to memory accessing instructions and denotes that
4410 no loop carried memory dependence exist between it and other instructions denoted
4411 with the same loop identifier.
4413 Precisely, given two instructions ``m1`` and ``m2`` that both have the
4414 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
4415 set of loops associated with that metadata, respectively, then there is no loop
4416 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
4419 As a special case, if all memory accessing instructions in a loop have
4420 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
4421 loop has no loop carried memory dependences and is considered to be a parallel
4424 Note that if not all memory access instructions have such metadata referring to
4425 the loop, then the loop is considered not being trivially parallel. Additional
4426 memory dependence analysis is required to make that determination. As a fail
4427 safe mechanism, this causes loops that were originally parallel to be considered
4428 sequential (if optimization passes that are unaware of the parallel semantics
4429 insert new memory instructions into the loop body).
4431 Example of a loop that is considered parallel due to its correct use of
4432 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
4433 metadata types that refer to the same loop identifier metadata.
4435 .. code-block:: llvm
4439 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
4441 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4443 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
4449 It is also possible to have nested parallel loops. In that case the
4450 memory accesses refer to a list of loop identifier metadata nodes instead of
4451 the loop identifier metadata node directly:
4453 .. code-block:: llvm
4457 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
4459 br label %inner.for.body
4463 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4465 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
4467 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
4471 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
4473 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
4475 outer.for.end: ; preds = %for.body
4477 !0 = !{!1, !2} ; a list of loop identifiers
4478 !1 = !{!1} ; an identifier for the inner loop
4479 !2 = !{!2} ; an identifier for the outer loop
4484 The ``llvm.bitsets`` global metadata is used to implement
4485 :doc:`bitsets <BitSets>`.
4487 '``invariant.group``' Metadata
4488 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4490 The ``invariant.group`` metadata may be attached to ``load``/``store`` instructions.
4491 The existence of the ``invariant.group`` metadata on the instruction tells
4492 the optimizer that every ``load`` and ``store`` to the same pointer operand
4493 within the same invariant group can be assumed to load or store the same
4494 value (but see the ``llvm.invariant.group.barrier`` intrinsic which affects
4495 when two pointers are considered the same).
4499 .. code-block:: llvm
4501 @unknownPtr = external global i8
4504 store i8 42, i8* %ptr, !invariant.group !0
4505 call void @foo(i8* %ptr)
4507 %a = load i8, i8* %ptr, !invariant.group !0 ; Can assume that value under %ptr didn't change
4508 call void @foo(i8* %ptr)
4509 %b = load i8, i8* %ptr, !invariant.group !1 ; Can't assume anything, because group changed
4511 %newPtr = call i8* @getPointer(i8* %ptr)
4512 %c = load i8, i8* %newPtr, !invariant.group !0 ; Can't assume anything, because we only have information about %ptr
4514 %unknownValue = load i8, i8* @unknownPtr
4515 store i8 %unknownValue, i8* %ptr, !invariant.group !0 ; Can assume that %unknownValue == 42
4517 call void @foo(i8* %ptr)
4518 %newPtr2 = call i8* @llvm.invariant.group.barrier(i8* %ptr)
4519 %d = load i8, i8* %newPtr2, !invariant.group !0 ; Can't step through invariant.group.barrier to get value of %ptr
4522 declare void @foo(i8*)
4523 declare i8* @getPointer(i8*)
4524 declare i8* @llvm.invariant.group.barrier(i8*)
4526 !0 = !{!"magic ptr"}
4527 !1 = !{!"other ptr"}
4531 Module Flags Metadata
4532 =====================
4534 Information about the module as a whole is difficult to convey to LLVM's
4535 subsystems. The LLVM IR isn't sufficient to transmit this information.
4536 The ``llvm.module.flags`` named metadata exists in order to facilitate
4537 this. These flags are in the form of key / value pairs --- much like a
4538 dictionary --- making it easy for any subsystem who cares about a flag to
4541 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
4542 Each triplet has the following form:
4544 - The first element is a *behavior* flag, which specifies the behavior
4545 when two (or more) modules are merged together, and it encounters two
4546 (or more) metadata with the same ID. The supported behaviors are
4548 - The second element is a metadata string that is a unique ID for the
4549 metadata. Each module may only have one flag entry for each unique ID (not
4550 including entries with the **Require** behavior).
4551 - The third element is the value of the flag.
4553 When two (or more) modules are merged together, the resulting
4554 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
4555 each unique metadata ID string, there will be exactly one entry in the merged
4556 modules ``llvm.module.flags`` metadata table, and the value for that entry will
4557 be determined by the merge behavior flag, as described below. The only exception
4558 is that entries with the *Require* behavior are always preserved.
4560 The following behaviors are supported:
4571 Emits an error if two values disagree, otherwise the resulting value
4572 is that of the operands.
4576 Emits a warning if two values disagree. The result value will be the
4577 operand for the flag from the first module being linked.
4581 Adds a requirement that another module flag be present and have a
4582 specified value after linking is performed. The value must be a
4583 metadata pair, where the first element of the pair is the ID of the
4584 module flag to be restricted, and the second element of the pair is
4585 the value the module flag should be restricted to. This behavior can
4586 be used to restrict the allowable results (via triggering of an
4587 error) of linking IDs with the **Override** behavior.
4591 Uses the specified value, regardless of the behavior or value of the
4592 other module. If both modules specify **Override**, but the values
4593 differ, an error will be emitted.
4597 Appends the two values, which are required to be metadata nodes.
4601 Appends the two values, which are required to be metadata
4602 nodes. However, duplicate entries in the second list are dropped
4603 during the append operation.
4605 It is an error for a particular unique flag ID to have multiple behaviors,
4606 except in the case of **Require** (which adds restrictions on another metadata
4607 value) or **Override**.
4609 An example of module flags:
4611 .. code-block:: llvm
4613 !0 = !{ i32 1, !"foo", i32 1 }
4614 !1 = !{ i32 4, !"bar", i32 37 }
4615 !2 = !{ i32 2, !"qux", i32 42 }
4616 !3 = !{ i32 3, !"qux",
4621 !llvm.module.flags = !{ !0, !1, !2, !3 }
4623 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
4624 if two or more ``!"foo"`` flags are seen is to emit an error if their
4625 values are not equal.
4627 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
4628 behavior if two or more ``!"bar"`` flags are seen is to use the value
4631 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
4632 behavior if two or more ``!"qux"`` flags are seen is to emit a
4633 warning if their values are not equal.
4635 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
4641 The behavior is to emit an error if the ``llvm.module.flags`` does not
4642 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
4645 Objective-C Garbage Collection Module Flags Metadata
4646 ----------------------------------------------------
4648 On the Mach-O platform, Objective-C stores metadata about garbage
4649 collection in a special section called "image info". The metadata
4650 consists of a version number and a bitmask specifying what types of
4651 garbage collection are supported (if any) by the file. If two or more
4652 modules are linked together their garbage collection metadata needs to
4653 be merged rather than appended together.
4655 The Objective-C garbage collection module flags metadata consists of the
4656 following key-value pairs:
4665 * - ``Objective-C Version``
4666 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
4668 * - ``Objective-C Image Info Version``
4669 - **[Required]** --- The version of the image info section. Currently
4672 * - ``Objective-C Image Info Section``
4673 - **[Required]** --- The section to place the metadata. Valid values are
4674 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
4675 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
4676 Objective-C ABI version 2.
4678 * - ``Objective-C Garbage Collection``
4679 - **[Required]** --- Specifies whether garbage collection is supported or
4680 not. Valid values are 0, for no garbage collection, and 2, for garbage
4681 collection supported.
4683 * - ``Objective-C GC Only``
4684 - **[Optional]** --- Specifies that only garbage collection is supported.
4685 If present, its value must be 6. This flag requires that the
4686 ``Objective-C Garbage Collection`` flag have the value 2.
4688 Some important flag interactions:
4690 - If a module with ``Objective-C Garbage Collection`` set to 0 is
4691 merged with a module with ``Objective-C Garbage Collection`` set to
4692 2, then the resulting module has the
4693 ``Objective-C Garbage Collection`` flag set to 0.
4694 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
4695 merged with a module with ``Objective-C GC Only`` set to 6.
4697 Automatic Linker Flags Module Flags Metadata
4698 --------------------------------------------
4700 Some targets support embedding flags to the linker inside individual object
4701 files. Typically this is used in conjunction with language extensions which
4702 allow source files to explicitly declare the libraries they depend on, and have
4703 these automatically be transmitted to the linker via object files.
4705 These flags are encoded in the IR using metadata in the module flags section,
4706 using the ``Linker Options`` key. The merge behavior for this flag is required
4707 to be ``AppendUnique``, and the value for the key is expected to be a metadata
4708 node which should be a list of other metadata nodes, each of which should be a
4709 list of metadata strings defining linker options.
4711 For example, the following metadata section specifies two separate sets of
4712 linker options, presumably to link against ``libz`` and the ``Cocoa``
4715 !0 = !{ i32 6, !"Linker Options",
4718 !{ !"-framework", !"Cocoa" } } }
4719 !llvm.module.flags = !{ !0 }
4721 The metadata encoding as lists of lists of options, as opposed to a collapsed
4722 list of options, is chosen so that the IR encoding can use multiple option
4723 strings to specify e.g., a single library, while still having that specifier be
4724 preserved as an atomic element that can be recognized by a target specific
4725 assembly writer or object file emitter.
4727 Each individual option is required to be either a valid option for the target's
4728 linker, or an option that is reserved by the target specific assembly writer or
4729 object file emitter. No other aspect of these options is defined by the IR.
4731 C type width Module Flags Metadata
4732 ----------------------------------
4734 The ARM backend emits a section into each generated object file describing the
4735 options that it was compiled with (in a compiler-independent way) to prevent
4736 linking incompatible objects, and to allow automatic library selection. Some
4737 of these options are not visible at the IR level, namely wchar_t width and enum
4740 To pass this information to the backend, these options are encoded in module
4741 flags metadata, using the following key-value pairs:
4751 - * 0 --- sizeof(wchar_t) == 4
4752 * 1 --- sizeof(wchar_t) == 2
4755 - * 0 --- Enums are at least as large as an ``int``.
4756 * 1 --- Enums are stored in the smallest integer type which can
4757 represent all of its values.
4759 For example, the following metadata section specifies that the module was
4760 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
4761 enum is the smallest type which can represent all of its values::
4763 !llvm.module.flags = !{!0, !1}
4764 !0 = !{i32 1, !"short_wchar", i32 1}
4765 !1 = !{i32 1, !"short_enum", i32 0}
4767 .. _intrinsicglobalvariables:
4769 Intrinsic Global Variables
4770 ==========================
4772 LLVM has a number of "magic" global variables that contain data that
4773 affect code generation or other IR semantics. These are documented here.
4774 All globals of this sort should have a section specified as
4775 "``llvm.metadata``". This section and all globals that start with
4776 "``llvm.``" are reserved for use by LLVM.
4780 The '``llvm.used``' Global Variable
4781 -----------------------------------
4783 The ``@llvm.used`` global is an array which has
4784 :ref:`appending linkage <linkage_appending>`. This array contains a list of
4785 pointers to named global variables, functions and aliases which may optionally
4786 have a pointer cast formed of bitcast or getelementptr. For example, a legal
4789 .. code-block:: llvm
4794 @llvm.used = appending global [2 x i8*] [
4796 i8* bitcast (i32* @Y to i8*)
4797 ], section "llvm.metadata"
4799 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
4800 and linker are required to treat the symbol as if there is a reference to the
4801 symbol that it cannot see (which is why they have to be named). For example, if
4802 a variable has internal linkage and no references other than that from the
4803 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
4804 references from inline asms and other things the compiler cannot "see", and
4805 corresponds to "``attribute((used))``" in GNU C.
4807 On some targets, the code generator must emit a directive to the
4808 assembler or object file to prevent the assembler and linker from
4809 molesting the symbol.
4811 .. _gv_llvmcompilerused:
4813 The '``llvm.compiler.used``' Global Variable
4814 --------------------------------------------
4816 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4817 directive, except that it only prevents the compiler from touching the
4818 symbol. On targets that support it, this allows an intelligent linker to
4819 optimize references to the symbol without being impeded as it would be
4822 This is a rare construct that should only be used in rare circumstances,
4823 and should not be exposed to source languages.
4825 .. _gv_llvmglobalctors:
4827 The '``llvm.global_ctors``' Global Variable
4828 -------------------------------------------
4830 .. code-block:: llvm
4832 %0 = type { i32, void ()*, i8* }
4833 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
4835 The ``@llvm.global_ctors`` array contains a list of constructor
4836 functions, priorities, and an optional associated global or function.
4837 The functions referenced by this array will be called in ascending order
4838 of priority (i.e. lowest first) when the module is loaded. The order of
4839 functions with the same priority is not defined.
4841 If the third field is present, non-null, and points to a global variable
4842 or function, the initializer function will only run if the associated
4843 data from the current module is not discarded.
4845 .. _llvmglobaldtors:
4847 The '``llvm.global_dtors``' Global Variable
4848 -------------------------------------------
4850 .. code-block:: llvm
4852 %0 = type { i32, void ()*, i8* }
4853 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
4855 The ``@llvm.global_dtors`` array contains a list of destructor
4856 functions, priorities, and an optional associated global or function.
4857 The functions referenced by this array will be called in descending
4858 order of priority (i.e. highest first) when the module is unloaded. The
4859 order of functions with the same priority is not defined.
4861 If the third field is present, non-null, and points to a global variable
4862 or function, the destructor function will only run if the associated
4863 data from the current module is not discarded.
4865 Instruction Reference
4866 =====================
4868 The LLVM instruction set consists of several different classifications
4869 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
4870 instructions <binaryops>`, :ref:`bitwise binary
4871 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
4872 :ref:`other instructions <otherops>`.
4876 Terminator Instructions
4877 -----------------------
4879 As mentioned :ref:`previously <functionstructure>`, every basic block in a
4880 program ends with a "Terminator" instruction, which indicates which
4881 block should be executed after the current block is finished. These
4882 terminator instructions typically yield a '``void``' value: they produce
4883 control flow, not values (the one exception being the
4884 ':ref:`invoke <i_invoke>`' instruction).
4886 The terminator instructions are: ':ref:`ret <i_ret>`',
4887 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
4888 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
4889 ':ref:`resume <i_resume>`', ':ref:`catchpad <i_catchpad>`',
4890 ':ref:`catchendpad <i_catchendpad>`',
4891 ':ref:`catchret <i_catchret>`',
4892 ':ref:`cleanupendpad <i_cleanupendpad>`',
4893 ':ref:`cleanupret <i_cleanupret>`',
4894 ':ref:`terminatepad <i_terminatepad>`',
4895 and ':ref:`unreachable <i_unreachable>`'.
4899 '``ret``' Instruction
4900 ^^^^^^^^^^^^^^^^^^^^^
4907 ret <type> <value> ; Return a value from a non-void function
4908 ret void ; Return from void function
4913 The '``ret``' instruction is used to return control flow (and optionally
4914 a value) from a function back to the caller.
4916 There are two forms of the '``ret``' instruction: one that returns a
4917 value and then causes control flow, and one that just causes control
4923 The '``ret``' instruction optionally accepts a single argument, the
4924 return value. The type of the return value must be a ':ref:`first
4925 class <t_firstclass>`' type.
4927 A function is not :ref:`well formed <wellformed>` if it it has a non-void
4928 return type and contains a '``ret``' instruction with no return value or
4929 a return value with a type that does not match its type, or if it has a
4930 void return type and contains a '``ret``' instruction with a return
4936 When the '``ret``' instruction is executed, control flow returns back to
4937 the calling function's context. If the caller is a
4938 ":ref:`call <i_call>`" instruction, execution continues at the
4939 instruction after the call. If the caller was an
4940 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
4941 beginning of the "normal" destination block. If the instruction returns
4942 a value, that value shall set the call or invoke instruction's return
4948 .. code-block:: llvm
4950 ret i32 5 ; Return an integer value of 5
4951 ret void ; Return from a void function
4952 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
4956 '``br``' Instruction
4957 ^^^^^^^^^^^^^^^^^^^^
4964 br i1 <cond>, label <iftrue>, label <iffalse>
4965 br label <dest> ; Unconditional branch
4970 The '``br``' instruction is used to cause control flow to transfer to a
4971 different basic block in the current function. There are two forms of
4972 this instruction, corresponding to a conditional branch and an
4973 unconditional branch.
4978 The conditional branch form of the '``br``' instruction takes a single
4979 '``i1``' value and two '``label``' values. The unconditional form of the
4980 '``br``' instruction takes a single '``label``' value as a target.
4985 Upon execution of a conditional '``br``' instruction, the '``i1``'
4986 argument is evaluated. If the value is ``true``, control flows to the
4987 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
4988 to the '``iffalse``' ``label`` argument.
4993 .. code-block:: llvm
4996 %cond = icmp eq i32 %a, %b
4997 br i1 %cond, label %IfEqual, label %IfUnequal
5005 '``switch``' Instruction
5006 ^^^^^^^^^^^^^^^^^^^^^^^^
5013 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
5018 The '``switch``' instruction is used to transfer control flow to one of
5019 several different places. It is a generalization of the '``br``'
5020 instruction, allowing a branch to occur to one of many possible
5026 The '``switch``' instruction uses three parameters: an integer
5027 comparison value '``value``', a default '``label``' destination, and an
5028 array of pairs of comparison value constants and '``label``'s. The table
5029 is not allowed to contain duplicate constant entries.
5034 The ``switch`` instruction specifies a table of values and destinations.
5035 When the '``switch``' instruction is executed, this table is searched
5036 for the given value. If the value is found, control flow is transferred
5037 to the corresponding destination; otherwise, control flow is transferred
5038 to the default destination.
5043 Depending on properties of the target machine and the particular
5044 ``switch`` instruction, this instruction may be code generated in
5045 different ways. For example, it could be generated as a series of
5046 chained conditional branches or with a lookup table.
5051 .. code-block:: llvm
5053 ; Emulate a conditional br instruction
5054 %Val = zext i1 %value to i32
5055 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
5057 ; Emulate an unconditional br instruction
5058 switch i32 0, label %dest [ ]
5060 ; Implement a jump table:
5061 switch i32 %val, label %otherwise [ i32 0, label %onzero
5063 i32 2, label %ontwo ]
5067 '``indirectbr``' Instruction
5068 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5075 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
5080 The '``indirectbr``' instruction implements an indirect branch to a
5081 label within the current function, whose address is specified by
5082 "``address``". Address must be derived from a
5083 :ref:`blockaddress <blockaddress>` constant.
5088 The '``address``' argument is the address of the label to jump to. The
5089 rest of the arguments indicate the full set of possible destinations
5090 that the address may point to. Blocks are allowed to occur multiple
5091 times in the destination list, though this isn't particularly useful.
5093 This destination list is required so that dataflow analysis has an
5094 accurate understanding of the CFG.
5099 Control transfers to the block specified in the address argument. All
5100 possible destination blocks must be listed in the label list, otherwise
5101 this instruction has undefined behavior. This implies that jumps to
5102 labels defined in other functions have undefined behavior as well.
5107 This is typically implemented with a jump through a register.
5112 .. code-block:: llvm
5114 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
5118 '``invoke``' Instruction
5119 ^^^^^^^^^^^^^^^^^^^^^^^^
5126 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
5127 [operand bundles] to label <normal label> unwind label <exception label>
5132 The '``invoke``' instruction causes control to transfer to a specified
5133 function, with the possibility of control flow transfer to either the
5134 '``normal``' label or the '``exception``' label. If the callee function
5135 returns with the "``ret``" instruction, control flow will return to the
5136 "normal" label. If the callee (or any indirect callees) returns via the
5137 ":ref:`resume <i_resume>`" instruction or other exception handling
5138 mechanism, control is interrupted and continued at the dynamically
5139 nearest "exception" label.
5141 The '``exception``' label is a `landing
5142 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
5143 '``exception``' label is required to have the
5144 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
5145 information about the behavior of the program after unwinding happens,
5146 as its first non-PHI instruction. The restrictions on the
5147 "``landingpad``" instruction's tightly couples it to the "``invoke``"
5148 instruction, so that the important information contained within the
5149 "``landingpad``" instruction can't be lost through normal code motion.
5154 This instruction requires several arguments:
5156 #. The optional "cconv" marker indicates which :ref:`calling
5157 convention <callingconv>` the call should use. If none is
5158 specified, the call defaults to using C calling conventions.
5159 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5160 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5162 #. '``ptr to function ty``': shall be the signature of the pointer to
5163 function value being invoked. In most cases, this is a direct
5164 function invocation, but indirect ``invoke``'s are just as possible,
5165 branching off an arbitrary pointer to function value.
5166 #. '``function ptr val``': An LLVM value containing a pointer to a
5167 function to be invoked.
5168 #. '``function args``': argument list whose types match the function
5169 signature argument types and parameter attributes. All arguments must
5170 be of :ref:`first class <t_firstclass>` type. If the function signature
5171 indicates the function accepts a variable number of arguments, the
5172 extra arguments can be specified.
5173 #. '``normal label``': the label reached when the called function
5174 executes a '``ret``' instruction.
5175 #. '``exception label``': the label reached when a callee returns via
5176 the :ref:`resume <i_resume>` instruction or other exception handling
5178 #. The optional :ref:`function attributes <fnattrs>` list. Only
5179 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5180 attributes are valid here.
5181 #. The optional :ref:`operand bundles <opbundles>` list.
5186 This instruction is designed to operate as a standard '``call``'
5187 instruction in most regards. The primary difference is that it
5188 establishes an association with a label, which is used by the runtime
5189 library to unwind the stack.
5191 This instruction is used in languages with destructors to ensure that
5192 proper cleanup is performed in the case of either a ``longjmp`` or a
5193 thrown exception. Additionally, this is important for implementation of
5194 '``catch``' clauses in high-level languages that support them.
5196 For the purposes of the SSA form, the definition of the value returned
5197 by the '``invoke``' instruction is deemed to occur on the edge from the
5198 current block to the "normal" label. If the callee unwinds then no
5199 return value is available.
5204 .. code-block:: llvm
5206 %retval = invoke i32 @Test(i32 15) to label %Continue
5207 unwind label %TestCleanup ; i32:retval set
5208 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
5209 unwind label %TestCleanup ; i32:retval set
5213 '``resume``' Instruction
5214 ^^^^^^^^^^^^^^^^^^^^^^^^
5221 resume <type> <value>
5226 The '``resume``' instruction is a terminator instruction that has no
5232 The '``resume``' instruction requires one argument, which must have the
5233 same type as the result of any '``landingpad``' instruction in the same
5239 The '``resume``' instruction resumes propagation of an existing
5240 (in-flight) exception whose unwinding was interrupted with a
5241 :ref:`landingpad <i_landingpad>` instruction.
5246 .. code-block:: llvm
5248 resume { i8*, i32 } %exn
5252 '``catchpad``' Instruction
5253 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5260 <resultval> = catchpad [<args>*]
5261 to label <normal label> unwind label <exception label>
5266 The '``catchpad``' instruction is used by `LLVM's exception handling
5267 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5268 is a catch block --- one where a personality routine attempts to transfer
5269 control to catch an exception.
5270 The ``args`` correspond to whatever information the personality
5271 routine requires to know if this is an appropriate place to catch the
5272 exception. Control is transfered to the ``exception`` label if the
5273 ``catchpad`` is not an appropriate handler for the in-flight exception.
5274 The ``normal`` label should contain the code found in the ``catch``
5275 portion of a ``try``/``catch`` sequence. The ``resultval`` has the type
5276 :ref:`token <t_token>` and is used to match the ``catchpad`` to
5277 corresponding :ref:`catchrets <i_catchret>`.
5282 The instruction takes a list of arbitrary values which are interpreted
5283 by the :ref:`personality function <personalityfn>`.
5285 The ``catchpad`` must be provided a ``normal`` label to transfer control
5286 to if the ``catchpad`` matches the exception and an ``exception``
5287 label to transfer control to if it doesn't.
5292 When the call stack is being unwound due to an exception being thrown,
5293 the exception is compared against the ``args``. If it doesn't match,
5294 then control is transfered to the ``exception`` basic block.
5295 As with calling conventions, how the personality function results are
5296 represented in LLVM IR is target specific.
5298 The ``catchpad`` instruction has several restrictions:
5300 - A catch block is a basic block which is the unwind destination of
5301 an exceptional instruction.
5302 - A catch block must have a '``catchpad``' instruction as its
5303 first non-PHI instruction.
5304 - A catch block's ``exception`` edge must refer to a catch block or a
5306 - There can be only one '``catchpad``' instruction within the
5308 - A basic block that is not a catch block may not include a
5309 '``catchpad``' instruction.
5310 - A catch block which has another catch block as a predecessor may not have
5311 any other predecessors.
5312 - It is undefined behavior for control to transfer from a ``catchpad`` to a
5313 ``ret`` without first executing a ``catchret`` that consumes the
5314 ``catchpad`` or unwinding through its ``catchendpad``.
5315 - It is undefined behavior for control to transfer from a ``catchpad`` to
5316 itself without first executing a ``catchret`` that consumes the
5317 ``catchpad`` or unwinding through its ``catchendpad``.
5322 .. code-block:: llvm
5324 ;; A catch block which can catch an integer.
5325 %tok = catchpad [i8** @_ZTIi]
5326 to label %int.handler unwind label %terminate
5330 '``catchendpad``' Instruction
5331 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5338 catchendpad unwind label <nextaction>
5339 catchendpad unwind to caller
5344 The '``catchendpad``' instruction is used by `LLVM's exception handling
5345 system <ExceptionHandling.html#overview>`_ to communicate to the
5346 :ref:`personality function <personalityfn>` which invokes are associated
5347 with a chain of :ref:`catchpad <i_catchpad>` instructions; propagating an
5348 exception out of a catch handler is represented by unwinding through its
5349 ``catchendpad``. Unwinding to the outer scope when a chain of catch handlers
5350 do not handle an exception is also represented by unwinding through their
5353 The ``nextaction`` label indicates where control should transfer to if
5354 none of the ``catchpad`` instructions are suitable for catching the
5355 in-flight exception.
5357 If a ``nextaction`` label is not present, the instruction unwinds out of
5358 its parent function. The
5359 :ref:`personality function <personalityfn>` will continue processing
5360 exception handling actions in the caller.
5365 The instruction optionally takes a label, ``nextaction``, indicating
5366 where control should transfer to if none of the preceding
5367 ``catchpad`` instructions are suitable for the in-flight exception.
5372 When the call stack is being unwound due to an exception being thrown
5373 and none of the constituent ``catchpad`` instructions match, then
5374 control is transfered to ``nextaction`` if it is present. If it is not
5375 present, control is transfered to the caller.
5377 The ``catchendpad`` instruction has several restrictions:
5379 - A catch-end block is a basic block which is the unwind destination of
5380 an exceptional instruction.
5381 - A catch-end block must have a '``catchendpad``' instruction as its
5382 first non-PHI instruction.
5383 - There can be only one '``catchendpad``' instruction within the
5385 - A basic block that is not a catch-end block may not include a
5386 '``catchendpad``' instruction.
5387 - Exactly one catch block may unwind to a ``catchendpad``.
5388 - It is undefined behavior to execute a ``catchendpad`` if none of the
5389 '``catchpad``'s chained to it have been executed.
5390 - It is undefined behavior to execute a ``catchendpad`` twice without an
5391 intervening execution of one or more of the '``catchpad``'s chained to it.
5392 - It is undefined behavior to execute a ``catchendpad`` if, after the most
5393 recent execution of the normal successor edge of any ``catchpad`` chained
5394 to it, some ``catchret`` consuming that ``catchpad`` has already been
5396 - It is undefined behavior to execute a ``catchendpad`` if, after the most
5397 recent execution of the normal successor edge of any ``catchpad`` chained
5398 to it, any other ``catchpad`` or ``cleanuppad`` has been executed but has
5399 not had a corresponding
5400 ``catchret``/``cleanupret``/``catchendpad``/``cleanupendpad`` executed.
5405 .. code-block:: llvm
5407 catchendpad unwind label %terminate
5408 catchendpad unwind to caller
5412 '``catchret``' Instruction
5413 ^^^^^^^^^^^^^^^^^^^^^^^^^^
5420 catchret <value> to label <normal>
5425 The '``catchret``' instruction is a terminator instruction that has a
5432 The first argument to a '``catchret``' indicates which ``catchpad`` it
5433 exits. It must be a :ref:`catchpad <i_catchpad>`.
5434 The second argument to a '``catchret``' specifies where control will
5440 The '``catchret``' instruction ends the existing (in-flight) exception
5441 whose unwinding was interrupted with a
5442 :ref:`catchpad <i_catchpad>` instruction.
5443 The :ref:`personality function <personalityfn>` gets a chance to execute
5444 arbitrary code to, for example, run a C++ destructor.
5445 Control then transfers to ``normal``.
5446 It may be passed an optional, personality specific, value.
5448 It is undefined behavior to execute a ``catchret`` whose ``catchpad`` has
5451 It is undefined behavior to execute a ``catchret`` if, after the most recent
5452 execution of its ``catchpad``, some ``catchret`` or ``catchendpad`` linked
5453 to the same ``catchpad`` has already been executed.
5455 It is undefined behavior to execute a ``catchret`` if, after the most recent
5456 execution of its ``catchpad``, any other ``catchpad`` or ``cleanuppad`` has
5457 been executed but has not had a corresponding
5458 ``catchret``/``cleanupret``/``catchendpad``/``cleanupendpad`` executed.
5463 .. code-block:: llvm
5465 catchret %catch label %continue
5467 .. _i_cleanupendpad:
5469 '``cleanupendpad``' Instruction
5470 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5477 cleanupendpad <value> unwind label <nextaction>
5478 cleanupendpad <value> unwind to caller
5483 The '``cleanupendpad``' instruction is used by `LLVM's exception handling
5484 system <ExceptionHandling.html#overview>`_ to communicate to the
5485 :ref:`personality function <personalityfn>` which invokes are associated
5486 with a :ref:`cleanuppad <i_cleanuppad>` instructions; propagating an exception
5487 out of a cleanup is represented by unwinding through its ``cleanupendpad``.
5489 The ``nextaction`` label indicates where control should unwind to next, in the
5490 event that a cleanup is exited by means of an(other) exception being raised.
5492 If a ``nextaction`` label is not present, the instruction unwinds out of
5493 its parent function. The
5494 :ref:`personality function <personalityfn>` will continue processing
5495 exception handling actions in the caller.
5500 The '``cleanupendpad``' instruction requires one argument, which indicates
5501 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
5502 It also has an optional successor, ``nextaction``, indicating where control
5508 When and exception propagates to a ``cleanupendpad``, control is transfered to
5509 ``nextaction`` if it is present. If it is not present, control is transfered to
5512 The ``cleanupendpad`` instruction has several restrictions:
5514 - A cleanup-end block is a basic block which is the unwind destination of
5515 an exceptional instruction.
5516 - A cleanup-end block must have a '``cleanupendpad``' instruction as its
5517 first non-PHI instruction.
5518 - There can be only one '``cleanupendpad``' instruction within the
5520 - A basic block that is not a cleanup-end block may not include a
5521 '``cleanupendpad``' instruction.
5522 - It is undefined behavior to execute a ``cleanupendpad`` whose ``cleanuppad``
5523 has not been executed.
5524 - It is undefined behavior to execute a ``cleanupendpad`` if, after the most
5525 recent execution of its ``cleanuppad``, some ``cleanupret`` or ``cleanupendpad``
5526 consuming the same ``cleanuppad`` has already been executed.
5527 - It is undefined behavior to execute a ``cleanupendpad`` if, after the most
5528 recent execution of its ``cleanuppad``, any other ``cleanuppad`` or
5529 ``catchpad`` has been executed but has not had a corresponding
5530 ``cleanupret``/``catchret``/``cleanupendpad``/``catchendpad`` executed.
5535 .. code-block:: llvm
5537 cleanupendpad %cleanup unwind label %terminate
5538 cleanupendpad %cleanup unwind to caller
5542 '``cleanupret``' Instruction
5543 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5550 cleanupret <value> unwind label <continue>
5551 cleanupret <value> unwind to caller
5556 The '``cleanupret``' instruction is a terminator instruction that has
5557 an optional successor.
5563 The '``cleanupret``' instruction requires one argument, which indicates
5564 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
5565 It also has an optional successor, ``continue``.
5570 The '``cleanupret``' instruction indicates to the
5571 :ref:`personality function <personalityfn>` that one
5572 :ref:`cleanuppad <i_cleanuppad>` it transferred control to has ended.
5573 It transfers control to ``continue`` or unwinds out of the function.
5575 It is undefined behavior to execute a ``cleanupret`` whose ``cleanuppad`` has
5578 It is undefined behavior to execute a ``cleanupret`` if, after the most recent
5579 execution of its ``cleanuppad``, some ``cleanupret`` or ``cleanupendpad``
5580 consuming the same ``cleanuppad`` has already been executed.
5582 It is undefined behavior to execute a ``cleanupret`` if, after the most recent
5583 execution of its ``cleanuppad``, any other ``cleanuppad`` or ``catchpad`` has
5584 been executed but has not had a corresponding
5585 ``cleanupret``/``catchret``/``cleanupendpad``/``catchendpad`` executed.
5590 .. code-block:: llvm
5592 cleanupret %cleanup unwind to caller
5593 cleanupret %cleanup unwind label %continue
5597 '``terminatepad``' Instruction
5598 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5605 terminatepad [<args>*] unwind label <exception label>
5606 terminatepad [<args>*] unwind to caller
5611 The '``terminatepad``' instruction is used by `LLVM's exception handling
5612 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5613 is a terminate block --- one where a personality routine may decide to
5614 terminate the program.
5615 The ``args`` correspond to whatever information the personality
5616 routine requires to know if this is an appropriate place to terminate the
5617 program. Control is transferred to the ``exception`` label if the
5618 personality routine decides not to terminate the program for the
5619 in-flight exception.
5624 The instruction takes a list of arbitrary values which are interpreted
5625 by the :ref:`personality function <personalityfn>`.
5627 The ``terminatepad`` may be given an ``exception`` label to
5628 transfer control to if the in-flight exception matches the ``args``.
5633 When the call stack is being unwound due to an exception being thrown,
5634 the exception is compared against the ``args``. If it matches,
5635 then control is transfered to the ``exception`` basic block. Otherwise,
5636 the program is terminated via personality-specific means. Typically,
5637 the first argument to ``terminatepad`` specifies what function the
5638 personality should defer to in order to terminate the program.
5640 The ``terminatepad`` instruction has several restrictions:
5642 - A terminate block is a basic block which is the unwind destination of
5643 an exceptional instruction.
5644 - A terminate block must have a '``terminatepad``' instruction as its
5645 first non-PHI instruction.
5646 - There can be only one '``terminatepad``' instruction within the
5648 - A basic block that is not a terminate block may not include a
5649 '``terminatepad``' instruction.
5654 .. code-block:: llvm
5656 ;; A terminate block which only permits integers.
5657 terminatepad [i8** @_ZTIi] unwind label %continue
5661 '``unreachable``' Instruction
5662 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5674 The '``unreachable``' instruction has no defined semantics. This
5675 instruction is used to inform the optimizer that a particular portion of
5676 the code is not reachable. This can be used to indicate that the code
5677 after a no-return function cannot be reached, and other facts.
5682 The '``unreachable``' instruction has no defined semantics.
5689 Binary operators are used to do most of the computation in a program.
5690 They require two operands of the same type, execute an operation on
5691 them, and produce a single value. The operands might represent multiple
5692 data, as is the case with the :ref:`vector <t_vector>` data type. The
5693 result value has the same type as its operands.
5695 There are several different binary operators:
5699 '``add``' Instruction
5700 ^^^^^^^^^^^^^^^^^^^^^
5707 <result> = add <ty> <op1>, <op2> ; yields ty:result
5708 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
5709 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
5710 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
5715 The '``add``' instruction returns the sum of its two operands.
5720 The two arguments to the '``add``' instruction must be
5721 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5722 arguments must have identical types.
5727 The value produced is the integer sum of the two operands.
5729 If the sum has unsigned overflow, the result returned is the
5730 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5733 Because LLVM integers use a two's complement representation, this
5734 instruction is appropriate for both signed and unsigned integers.
5736 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5737 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5738 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
5739 unsigned and/or signed overflow, respectively, occurs.
5744 .. code-block:: llvm
5746 <result> = add i32 4, %var ; yields i32:result = 4 + %var
5750 '``fadd``' Instruction
5751 ^^^^^^^^^^^^^^^^^^^^^^
5758 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5763 The '``fadd``' instruction returns the sum of its two operands.
5768 The two arguments to the '``fadd``' instruction must be :ref:`floating
5769 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5770 Both arguments must have identical types.
5775 The value produced is the floating point sum of the two operands. This
5776 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
5777 which are optimization hints to enable otherwise unsafe floating point
5783 .. code-block:: llvm
5785 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
5787 '``sub``' Instruction
5788 ^^^^^^^^^^^^^^^^^^^^^
5795 <result> = sub <ty> <op1>, <op2> ; yields ty:result
5796 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
5797 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
5798 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
5803 The '``sub``' instruction returns the difference of its two operands.
5805 Note that the '``sub``' instruction is used to represent the '``neg``'
5806 instruction present in most other intermediate representations.
5811 The two arguments to the '``sub``' instruction must be
5812 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5813 arguments must have identical types.
5818 The value produced is the integer difference of the two operands.
5820 If the difference has unsigned overflow, the result returned is the
5821 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5824 Because LLVM integers use a two's complement representation, this
5825 instruction is appropriate for both signed and unsigned integers.
5827 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5828 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5829 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
5830 unsigned and/or signed overflow, respectively, occurs.
5835 .. code-block:: llvm
5837 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
5838 <result> = sub i32 0, %val ; yields i32:result = -%var
5842 '``fsub``' Instruction
5843 ^^^^^^^^^^^^^^^^^^^^^^
5850 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5855 The '``fsub``' instruction returns the difference of its two operands.
5857 Note that the '``fsub``' instruction is used to represent the '``fneg``'
5858 instruction present in most other intermediate representations.
5863 The two arguments to the '``fsub``' instruction must be :ref:`floating
5864 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5865 Both arguments must have identical types.
5870 The value produced is the floating point difference of the two operands.
5871 This instruction can also take any number of :ref:`fast-math
5872 flags <fastmath>`, which are optimization hints to enable otherwise
5873 unsafe floating point optimizations:
5878 .. code-block:: llvm
5880 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
5881 <result> = fsub float -0.0, %val ; yields float:result = -%var
5883 '``mul``' Instruction
5884 ^^^^^^^^^^^^^^^^^^^^^
5891 <result> = mul <ty> <op1>, <op2> ; yields ty:result
5892 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
5893 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
5894 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
5899 The '``mul``' instruction returns the product of its two operands.
5904 The two arguments to the '``mul``' instruction must be
5905 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5906 arguments must have identical types.
5911 The value produced is the integer product of the two operands.
5913 If the result of the multiplication has unsigned overflow, the result
5914 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
5915 bit width of the result.
5917 Because LLVM integers use a two's complement representation, and the
5918 result is the same width as the operands, this instruction returns the
5919 correct result for both signed and unsigned integers. If a full product
5920 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
5921 sign-extended or zero-extended as appropriate to the width of the full
5924 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5925 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5926 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
5927 unsigned and/or signed overflow, respectively, occurs.
5932 .. code-block:: llvm
5934 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
5938 '``fmul``' Instruction
5939 ^^^^^^^^^^^^^^^^^^^^^^
5946 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5951 The '``fmul``' instruction returns the product of its two operands.
5956 The two arguments to the '``fmul``' instruction must be :ref:`floating
5957 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5958 Both arguments must have identical types.
5963 The value produced is the floating point product of the two operands.
5964 This instruction can also take any number of :ref:`fast-math
5965 flags <fastmath>`, which are optimization hints to enable otherwise
5966 unsafe floating point optimizations:
5971 .. code-block:: llvm
5973 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
5975 '``udiv``' Instruction
5976 ^^^^^^^^^^^^^^^^^^^^^^
5983 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
5984 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
5989 The '``udiv``' instruction returns the quotient of its two operands.
5994 The two arguments to the '``udiv``' instruction must be
5995 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5996 arguments must have identical types.
6001 The value produced is the unsigned integer quotient of the two operands.
6003 Note that unsigned integer division and signed integer division are
6004 distinct operations; for signed integer division, use '``sdiv``'.
6006 Division by zero leads to undefined behavior.
6008 If the ``exact`` keyword is present, the result value of the ``udiv`` is
6009 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
6010 such, "((a udiv exact b) mul b) == a").
6015 .. code-block:: llvm
6017 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
6019 '``sdiv``' Instruction
6020 ^^^^^^^^^^^^^^^^^^^^^^
6027 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
6028 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
6033 The '``sdiv``' instruction returns the quotient of its two operands.
6038 The two arguments to the '``sdiv``' instruction must be
6039 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6040 arguments must have identical types.
6045 The value produced is the signed integer quotient of the two operands
6046 rounded towards zero.
6048 Note that signed integer division and unsigned integer division are
6049 distinct operations; for unsigned integer division, use '``udiv``'.
6051 Division by zero leads to undefined behavior. Overflow also leads to
6052 undefined behavior; this is a rare case, but can occur, for example, by
6053 doing a 32-bit division of -2147483648 by -1.
6055 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
6056 a :ref:`poison value <poisonvalues>` if the result would be rounded.
6061 .. code-block:: llvm
6063 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
6067 '``fdiv``' Instruction
6068 ^^^^^^^^^^^^^^^^^^^^^^
6075 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6080 The '``fdiv``' instruction returns the quotient of its two operands.
6085 The two arguments to the '``fdiv``' instruction must be :ref:`floating
6086 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6087 Both arguments must have identical types.
6092 The value produced is the floating point quotient of the two operands.
6093 This instruction can also take any number of :ref:`fast-math
6094 flags <fastmath>`, which are optimization hints to enable otherwise
6095 unsafe floating point optimizations:
6100 .. code-block:: llvm
6102 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
6104 '``urem``' Instruction
6105 ^^^^^^^^^^^^^^^^^^^^^^
6112 <result> = urem <ty> <op1>, <op2> ; yields ty:result
6117 The '``urem``' instruction returns the remainder from the unsigned
6118 division of its two arguments.
6123 The two arguments to the '``urem``' instruction must be
6124 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6125 arguments must have identical types.
6130 This instruction returns the unsigned integer *remainder* of a division.
6131 This instruction always performs an unsigned division to get the
6134 Note that unsigned integer remainder and signed integer remainder are
6135 distinct operations; for signed integer remainder, use '``srem``'.
6137 Taking the remainder of a division by zero leads to undefined behavior.
6142 .. code-block:: llvm
6144 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
6146 '``srem``' Instruction
6147 ^^^^^^^^^^^^^^^^^^^^^^
6154 <result> = srem <ty> <op1>, <op2> ; yields ty:result
6159 The '``srem``' instruction returns the remainder from the signed
6160 division of its two operands. This instruction can also take
6161 :ref:`vector <t_vector>` versions of the values in which case the elements
6167 The two arguments to the '``srem``' instruction must be
6168 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6169 arguments must have identical types.
6174 This instruction returns the *remainder* of a division (where the result
6175 is either zero or has the same sign as the dividend, ``op1``), not the
6176 *modulo* operator (where the result is either zero or has the same sign
6177 as the divisor, ``op2``) of a value. For more information about the
6178 difference, see `The Math
6179 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
6180 table of how this is implemented in various languages, please see
6182 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
6184 Note that signed integer remainder and unsigned integer remainder are
6185 distinct operations; for unsigned integer remainder, use '``urem``'.
6187 Taking the remainder of a division by zero leads to undefined behavior.
6188 Overflow also leads to undefined behavior; this is a rare case, but can
6189 occur, for example, by taking the remainder of a 32-bit division of
6190 -2147483648 by -1. (The remainder doesn't actually overflow, but this
6191 rule lets srem be implemented using instructions that return both the
6192 result of the division and the remainder.)
6197 .. code-block:: llvm
6199 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
6203 '``frem``' Instruction
6204 ^^^^^^^^^^^^^^^^^^^^^^
6211 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6216 The '``frem``' instruction returns the remainder from the division of
6222 The two arguments to the '``frem``' instruction must be :ref:`floating
6223 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6224 Both arguments must have identical types.
6229 This instruction returns the *remainder* of a division. The remainder
6230 has the same sign as the dividend. This instruction can also take any
6231 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
6232 to enable otherwise unsafe floating point optimizations:
6237 .. code-block:: llvm
6239 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
6243 Bitwise Binary Operations
6244 -------------------------
6246 Bitwise binary operators are used to do various forms of bit-twiddling
6247 in a program. They are generally very efficient instructions and can
6248 commonly be strength reduced from other instructions. They require two
6249 operands of the same type, execute an operation on them, and produce a
6250 single value. The resulting value is the same type as its operands.
6252 '``shl``' Instruction
6253 ^^^^^^^^^^^^^^^^^^^^^
6260 <result> = shl <ty> <op1>, <op2> ; yields ty:result
6261 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
6262 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
6263 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
6268 The '``shl``' instruction returns the first operand shifted to the left
6269 a specified number of bits.
6274 Both arguments to the '``shl``' instruction must be the same
6275 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6276 '``op2``' is treated as an unsigned value.
6281 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
6282 where ``n`` is the width of the result. If ``op2`` is (statically or
6283 dynamically) equal to or larger than the number of bits in
6284 ``op1``, the result is undefined. If the arguments are vectors, each
6285 vector element of ``op1`` is shifted by the corresponding shift amount
6288 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
6289 value <poisonvalues>` if it shifts out any non-zero bits. If the
6290 ``nsw`` keyword is present, then the shift produces a :ref:`poison
6291 value <poisonvalues>` if it shifts out any bits that disagree with the
6292 resultant sign bit. As such, NUW/NSW have the same semantics as they
6293 would if the shift were expressed as a mul instruction with the same
6294 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
6299 .. code-block:: llvm
6301 <result> = shl i32 4, %var ; yields i32: 4 << %var
6302 <result> = shl i32 4, 2 ; yields i32: 16
6303 <result> = shl i32 1, 10 ; yields i32: 1024
6304 <result> = shl i32 1, 32 ; undefined
6305 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
6307 '``lshr``' Instruction
6308 ^^^^^^^^^^^^^^^^^^^^^^
6315 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
6316 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
6321 The '``lshr``' instruction (logical shift right) returns the first
6322 operand shifted to the right a specified number of bits with zero fill.
6327 Both arguments to the '``lshr``' instruction must be the same
6328 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6329 '``op2``' is treated as an unsigned value.
6334 This instruction always performs a logical shift right operation. The
6335 most significant bits of the result will be filled with zero bits after
6336 the shift. If ``op2`` is (statically or dynamically) equal to or larger
6337 than the number of bits in ``op1``, the result is undefined. If the
6338 arguments are vectors, each vector element of ``op1`` is shifted by the
6339 corresponding shift amount in ``op2``.
6341 If the ``exact`` keyword is present, the result value of the ``lshr`` is
6342 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6348 .. code-block:: llvm
6350 <result> = lshr i32 4, 1 ; yields i32:result = 2
6351 <result> = lshr i32 4, 2 ; yields i32:result = 1
6352 <result> = lshr i8 4, 3 ; yields i8:result = 0
6353 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
6354 <result> = lshr i32 1, 32 ; undefined
6355 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
6357 '``ashr``' Instruction
6358 ^^^^^^^^^^^^^^^^^^^^^^
6365 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
6366 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
6371 The '``ashr``' instruction (arithmetic shift right) returns the first
6372 operand shifted to the right a specified number of bits with sign
6378 Both arguments to the '``ashr``' instruction must be the same
6379 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6380 '``op2``' is treated as an unsigned value.
6385 This instruction always performs an arithmetic shift right operation,
6386 The most significant bits of the result will be filled with the sign bit
6387 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
6388 than the number of bits in ``op1``, the result is undefined. If the
6389 arguments are vectors, each vector element of ``op1`` is shifted by the
6390 corresponding shift amount in ``op2``.
6392 If the ``exact`` keyword is present, the result value of the ``ashr`` is
6393 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6399 .. code-block:: llvm
6401 <result> = ashr i32 4, 1 ; yields i32:result = 2
6402 <result> = ashr i32 4, 2 ; yields i32:result = 1
6403 <result> = ashr i8 4, 3 ; yields i8:result = 0
6404 <result> = ashr i8 -2, 1 ; yields i8:result = -1
6405 <result> = ashr i32 1, 32 ; undefined
6406 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
6408 '``and``' Instruction
6409 ^^^^^^^^^^^^^^^^^^^^^
6416 <result> = and <ty> <op1>, <op2> ; yields ty:result
6421 The '``and``' instruction returns the bitwise logical and of its two
6427 The two arguments to the '``and``' instruction must be
6428 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6429 arguments must have identical types.
6434 The truth table used for the '``and``' instruction is:
6451 .. code-block:: llvm
6453 <result> = and i32 4, %var ; yields i32:result = 4 & %var
6454 <result> = and i32 15, 40 ; yields i32:result = 8
6455 <result> = and i32 4, 8 ; yields i32:result = 0
6457 '``or``' Instruction
6458 ^^^^^^^^^^^^^^^^^^^^
6465 <result> = or <ty> <op1>, <op2> ; yields ty:result
6470 The '``or``' instruction returns the bitwise logical inclusive or of its
6476 The two arguments to the '``or``' instruction must be
6477 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6478 arguments must have identical types.
6483 The truth table used for the '``or``' instruction is:
6502 <result> = or i32 4, %var ; yields i32:result = 4 | %var
6503 <result> = or i32 15, 40 ; yields i32:result = 47
6504 <result> = or i32 4, 8 ; yields i32:result = 12
6506 '``xor``' Instruction
6507 ^^^^^^^^^^^^^^^^^^^^^
6514 <result> = xor <ty> <op1>, <op2> ; yields ty:result
6519 The '``xor``' instruction returns the bitwise logical exclusive or of
6520 its two operands. The ``xor`` is used to implement the "one's
6521 complement" operation, which is the "~" operator in C.
6526 The two arguments to the '``xor``' instruction must be
6527 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6528 arguments must have identical types.
6533 The truth table used for the '``xor``' instruction is:
6550 .. code-block:: llvm
6552 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
6553 <result> = xor i32 15, 40 ; yields i32:result = 39
6554 <result> = xor i32 4, 8 ; yields i32:result = 12
6555 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
6560 LLVM supports several instructions to represent vector operations in a
6561 target-independent manner. These instructions cover the element-access
6562 and vector-specific operations needed to process vectors effectively.
6563 While LLVM does directly support these vector operations, many
6564 sophisticated algorithms will want to use target-specific intrinsics to
6565 take full advantage of a specific target.
6567 .. _i_extractelement:
6569 '``extractelement``' Instruction
6570 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6577 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
6582 The '``extractelement``' instruction extracts a single scalar element
6583 from a vector at a specified index.
6588 The first operand of an '``extractelement``' instruction is a value of
6589 :ref:`vector <t_vector>` type. The second operand is an index indicating
6590 the position from which to extract the element. The index may be a
6591 variable of any integer type.
6596 The result is a scalar of the same type as the element type of ``val``.
6597 Its value is the value at position ``idx`` of ``val``. If ``idx``
6598 exceeds the length of ``val``, the results are undefined.
6603 .. code-block:: llvm
6605 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
6607 .. _i_insertelement:
6609 '``insertelement``' Instruction
6610 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6617 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
6622 The '``insertelement``' instruction inserts a scalar element into a
6623 vector at a specified index.
6628 The first operand of an '``insertelement``' instruction is a value of
6629 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
6630 type must equal the element type of the first operand. The third operand
6631 is an index indicating the position at which to insert the value. The
6632 index may be a variable of any integer type.
6637 The result is a vector of the same type as ``val``. Its element values
6638 are those of ``val`` except at position ``idx``, where it gets the value
6639 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
6645 .. code-block:: llvm
6647 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
6649 .. _i_shufflevector:
6651 '``shufflevector``' Instruction
6652 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6659 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
6664 The '``shufflevector``' instruction constructs a permutation of elements
6665 from two input vectors, returning a vector with the same element type as
6666 the input and length that is the same as the shuffle mask.
6671 The first two operands of a '``shufflevector``' instruction are vectors
6672 with the same type. The third argument is a shuffle mask whose element
6673 type is always 'i32'. The result of the instruction is a vector whose
6674 length is the same as the shuffle mask and whose element type is the
6675 same as the element type of the first two operands.
6677 The shuffle mask operand is required to be a constant vector with either
6678 constant integer or undef values.
6683 The elements of the two input vectors are numbered from left to right
6684 across both of the vectors. The shuffle mask operand specifies, for each
6685 element of the result vector, which element of the two input vectors the
6686 result element gets. The element selector may be undef (meaning "don't
6687 care") and the second operand may be undef if performing a shuffle from
6693 .. code-block:: llvm
6695 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6696 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
6697 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
6698 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
6699 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
6700 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
6701 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6702 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
6704 Aggregate Operations
6705 --------------------
6707 LLVM supports several instructions for working with
6708 :ref:`aggregate <t_aggregate>` values.
6712 '``extractvalue``' Instruction
6713 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6720 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
6725 The '``extractvalue``' instruction extracts the value of a member field
6726 from an :ref:`aggregate <t_aggregate>` value.
6731 The first operand of an '``extractvalue``' instruction is a value of
6732 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The other operands are
6733 constant indices to specify which value to extract in a similar manner
6734 as indices in a '``getelementptr``' instruction.
6736 The major differences to ``getelementptr`` indexing are:
6738 - Since the value being indexed is not a pointer, the first index is
6739 omitted and assumed to be zero.
6740 - At least one index must be specified.
6741 - Not only struct indices but also array indices must be in bounds.
6746 The result is the value at the position in the aggregate specified by
6752 .. code-block:: llvm
6754 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
6758 '``insertvalue``' Instruction
6759 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6766 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
6771 The '``insertvalue``' instruction inserts a value into a member field in
6772 an :ref:`aggregate <t_aggregate>` value.
6777 The first operand of an '``insertvalue``' instruction is a value of
6778 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
6779 a first-class value to insert. The following operands are constant
6780 indices indicating the position at which to insert the value in a
6781 similar manner as indices in a '``extractvalue``' instruction. The value
6782 to insert must have the same type as the value identified by the
6788 The result is an aggregate of the same type as ``val``. Its value is
6789 that of ``val`` except that the value at the position specified by the
6790 indices is that of ``elt``.
6795 .. code-block:: llvm
6797 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
6798 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
6799 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
6803 Memory Access and Addressing Operations
6804 ---------------------------------------
6806 A key design point of an SSA-based representation is how it represents
6807 memory. In LLVM, no memory locations are in SSA form, which makes things
6808 very simple. This section describes how to read, write, and allocate
6813 '``alloca``' Instruction
6814 ^^^^^^^^^^^^^^^^^^^^^^^^
6821 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
6826 The '``alloca``' instruction allocates memory on the stack frame of the
6827 currently executing function, to be automatically released when this
6828 function returns to its caller. The object is always allocated in the
6829 generic address space (address space zero).
6834 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
6835 bytes of memory on the runtime stack, returning a pointer of the
6836 appropriate type to the program. If "NumElements" is specified, it is
6837 the number of elements allocated, otherwise "NumElements" is defaulted
6838 to be one. If a constant alignment is specified, the value result of the
6839 allocation is guaranteed to be aligned to at least that boundary. The
6840 alignment may not be greater than ``1 << 29``. If not specified, or if
6841 zero, the target can choose to align the allocation on any convenient
6842 boundary compatible with the type.
6844 '``type``' may be any sized type.
6849 Memory is allocated; a pointer is returned. The operation is undefined
6850 if there is insufficient stack space for the allocation. '``alloca``'d
6851 memory is automatically released when the function returns. The
6852 '``alloca``' instruction is commonly used to represent automatic
6853 variables that must have an address available. When the function returns
6854 (either with the ``ret`` or ``resume`` instructions), the memory is
6855 reclaimed. Allocating zero bytes is legal, but the result is undefined.
6856 The order in which memory is allocated (ie., which way the stack grows)
6862 .. code-block:: llvm
6864 %ptr = alloca i32 ; yields i32*:ptr
6865 %ptr = alloca i32, i32 4 ; yields i32*:ptr
6866 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
6867 %ptr = alloca i32, align 1024 ; yields i32*:ptr
6871 '``load``' Instruction
6872 ^^^^^^^^^^^^^^^^^^^^^^
6879 <result> = load [volatile] <ty>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>][, !invariant.group !<index>][, !nonnull !<index>][, !dereferenceable !<deref_bytes_node>][, !dereferenceable_or_null !<deref_bytes_node>][, !align !<align_node>]
6880 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>]
6881 !<index> = !{ i32 1 }
6882 !<deref_bytes_node> = !{i64 <dereferenceable_bytes>}
6883 !<align_node> = !{ i64 <value_alignment> }
6888 The '``load``' instruction is used to read from memory.
6893 The argument to the ``load`` instruction specifies the memory address
6894 from which to load. The type specified must be a :ref:`first
6895 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
6896 then the optimizer is not allowed to modify the number or order of
6897 execution of this ``load`` with other :ref:`volatile
6898 operations <volatile>`.
6900 If the ``load`` is marked as ``atomic``, it takes an extra
6901 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
6902 ``release`` and ``acq_rel`` orderings are not valid on ``load``
6903 instructions. Atomic loads produce :ref:`defined <memmodel>` results
6904 when they may see multiple atomic stores. The type of the pointee must
6905 be an integer type whose bit width is a power of two greater than or
6906 equal to eight and less than or equal to a target-specific size limit.
6907 ``align`` must be explicitly specified on atomic loads, and the load has
6908 undefined behavior if the alignment is not set to a value which is at
6909 least the size in bytes of the pointee. ``!nontemporal`` does not have
6910 any defined semantics for atomic loads.
6912 The optional constant ``align`` argument specifies the alignment of the
6913 operation (that is, the alignment of the memory address). A value of 0
6914 or an omitted ``align`` argument means that the operation has the ABI
6915 alignment for the target. It is the responsibility of the code emitter
6916 to ensure that the alignment information is correct. Overestimating the
6917 alignment results in undefined behavior. Underestimating the alignment
6918 may produce less efficient code. An alignment of 1 is always safe. The
6919 maximum possible alignment is ``1 << 29``.
6921 The optional ``!nontemporal`` metadata must reference a single
6922 metadata name ``<index>`` corresponding to a metadata node with one
6923 ``i32`` entry of value 1. The existence of the ``!nontemporal``
6924 metadata on the instruction tells the optimizer and code generator
6925 that this load is not expected to be reused in the cache. The code
6926 generator may select special instructions to save cache bandwidth, such
6927 as the ``MOVNT`` instruction on x86.
6929 The optional ``!invariant.load`` metadata must reference a single
6930 metadata name ``<index>`` corresponding to a metadata node with no
6931 entries. The existence of the ``!invariant.load`` metadata on the
6932 instruction tells the optimizer and code generator that the address
6933 operand to this load points to memory which can be assumed unchanged.
6934 Being invariant does not imply that a location is dereferenceable,
6935 but it does imply that once the location is known dereferenceable
6936 its value is henceforth unchanging.
6938 The optional ``!invariant.group`` metadata must reference a single metadata name
6939 ``<index>`` corresponding to a metadata node. See ``invariant.group`` metadata.
6941 The optional ``!nonnull`` metadata must reference a single
6942 metadata name ``<index>`` corresponding to a metadata node with no
6943 entries. The existence of the ``!nonnull`` metadata on the
6944 instruction tells the optimizer that the value loaded is known to
6945 never be null. This is analogous to the ``nonnull`` attribute
6946 on parameters and return values. This metadata can only be applied
6947 to loads of a pointer type.
6949 The optional ``!dereferenceable`` metadata must reference a single metadata
6950 name ``<deref_bytes_node>`` corresponding to a metadata node with one ``i64``
6951 entry. The existence of the ``!dereferenceable`` metadata on the instruction
6952 tells the optimizer that the value loaded is known to be dereferenceable.
6953 The number of bytes known to be dereferenceable is specified by the integer
6954 value in the metadata node. This is analogous to the ''dereferenceable''
6955 attribute on parameters and return values. This metadata can only be applied
6956 to loads of a pointer type.
6958 The optional ``!dereferenceable_or_null`` metadata must reference a single
6959 metadata name ``<deref_bytes_node>`` corresponding to a metadata node with one
6960 ``i64`` entry. The existence of the ``!dereferenceable_or_null`` metadata on the
6961 instruction tells the optimizer that the value loaded is known to be either
6962 dereferenceable or null.
6963 The number of bytes known to be dereferenceable is specified by the integer
6964 value in the metadata node. This is analogous to the ''dereferenceable_or_null''
6965 attribute on parameters and return values. This metadata can only be applied
6966 to loads of a pointer type.
6968 The optional ``!align`` metadata must reference a single metadata name
6969 ``<align_node>`` corresponding to a metadata node with one ``i64`` entry.
6970 The existence of the ``!align`` metadata on the instruction tells the
6971 optimizer that the value loaded is known to be aligned to a boundary specified
6972 by the integer value in the metadata node. The alignment must be a power of 2.
6973 This is analogous to the ''align'' attribute on parameters and return values.
6974 This metadata can only be applied to loads of a pointer type.
6979 The location of memory pointed to is loaded. If the value being loaded
6980 is of scalar type then the number of bytes read does not exceed the
6981 minimum number of bytes needed to hold all bits of the type. For
6982 example, loading an ``i24`` reads at most three bytes. When loading a
6983 value of a type like ``i20`` with a size that is not an integral number
6984 of bytes, the result is undefined if the value was not originally
6985 written using a store of the same type.
6990 .. code-block:: llvm
6992 %ptr = alloca i32 ; yields i32*:ptr
6993 store i32 3, i32* %ptr ; yields void
6994 %val = load i32, i32* %ptr ; yields i32:val = i32 3
6998 '``store``' Instruction
6999 ^^^^^^^^^^^^^^^^^^^^^^^
7006 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.group !<index>] ; yields void
7007 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>] ; yields void
7012 The '``store``' instruction is used to write to memory.
7017 There are two arguments to the ``store`` instruction: a value to store
7018 and an address at which to store it. The type of the ``<pointer>``
7019 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
7020 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
7021 then the optimizer is not allowed to modify the number or order of
7022 execution of this ``store`` with other :ref:`volatile
7023 operations <volatile>`.
7025 If the ``store`` is marked as ``atomic``, it takes an extra
7026 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
7027 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
7028 instructions. Atomic loads produce :ref:`defined <memmodel>` results
7029 when they may see multiple atomic stores. The type of the pointee must
7030 be an integer type whose bit width is a power of two greater than or
7031 equal to eight and less than or equal to a target-specific size limit.
7032 ``align`` must be explicitly specified on atomic stores, and the store
7033 has undefined behavior if the alignment is not set to a value which is
7034 at least the size in bytes of the pointee. ``!nontemporal`` does not
7035 have any defined semantics for atomic stores.
7037 The optional constant ``align`` argument specifies the alignment of the
7038 operation (that is, the alignment of the memory address). A value of 0
7039 or an omitted ``align`` argument means that the operation has the ABI
7040 alignment for the target. It is the responsibility of the code emitter
7041 to ensure that the alignment information is correct. Overestimating the
7042 alignment results in undefined behavior. Underestimating the
7043 alignment may produce less efficient code. An alignment of 1 is always
7044 safe. The maximum possible alignment is ``1 << 29``.
7046 The optional ``!nontemporal`` metadata must reference a single metadata
7047 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
7048 value 1. The existence of the ``!nontemporal`` metadata on the instruction
7049 tells the optimizer and code generator that this load is not expected to
7050 be reused in the cache. The code generator may select special
7051 instructions to save cache bandwidth, such as the MOVNT instruction on
7054 The optional ``!invariant.group`` metadata must reference a
7055 single metadata name ``<index>``. See ``invariant.group`` metadata.
7060 The contents of memory are updated to contain ``<value>`` at the
7061 location specified by the ``<pointer>`` operand. If ``<value>`` is
7062 of scalar type then the number of bytes written does not exceed the
7063 minimum number of bytes needed to hold all bits of the type. For
7064 example, storing an ``i24`` writes at most three bytes. When writing a
7065 value of a type like ``i20`` with a size that is not an integral number
7066 of bytes, it is unspecified what happens to the extra bits that do not
7067 belong to the type, but they will typically be overwritten.
7072 .. code-block:: llvm
7074 %ptr = alloca i32 ; yields i32*:ptr
7075 store i32 3, i32* %ptr ; yields void
7076 %val = load i32, i32* %ptr ; yields i32:val = i32 3
7080 '``fence``' Instruction
7081 ^^^^^^^^^^^^^^^^^^^^^^^
7088 fence [singlethread] <ordering> ; yields void
7093 The '``fence``' instruction is used to introduce happens-before edges
7099 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
7100 defines what *synchronizes-with* edges they add. They can only be given
7101 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
7106 A fence A which has (at least) ``release`` ordering semantics
7107 *synchronizes with* a fence B with (at least) ``acquire`` ordering
7108 semantics if and only if there exist atomic operations X and Y, both
7109 operating on some atomic object M, such that A is sequenced before X, X
7110 modifies M (either directly or through some side effect of a sequence
7111 headed by X), Y is sequenced before B, and Y observes M. This provides a
7112 *happens-before* dependency between A and B. Rather than an explicit
7113 ``fence``, one (but not both) of the atomic operations X or Y might
7114 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
7115 still *synchronize-with* the explicit ``fence`` and establish the
7116 *happens-before* edge.
7118 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
7119 ``acquire`` and ``release`` semantics specified above, participates in
7120 the global program order of other ``seq_cst`` operations and/or fences.
7122 The optional ":ref:`singlethread <singlethread>`" argument specifies
7123 that the fence only synchronizes with other fences in the same thread.
7124 (This is useful for interacting with signal handlers.)
7129 .. code-block:: llvm
7131 fence acquire ; yields void
7132 fence singlethread seq_cst ; yields void
7136 '``cmpxchg``' Instruction
7137 ^^^^^^^^^^^^^^^^^^^^^^^^^
7144 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
7149 The '``cmpxchg``' instruction is used to atomically modify memory. It
7150 loads a value in memory and compares it to a given value. If they are
7151 equal, it tries to store a new value into the memory.
7156 There are three arguments to the '``cmpxchg``' instruction: an address
7157 to operate on, a value to compare to the value currently be at that
7158 address, and a new value to place at that address if the compared values
7159 are equal. The type of '<cmp>' must be an integer type whose bit width
7160 is a power of two greater than or equal to eight and less than or equal
7161 to a target-specific size limit. '<cmp>' and '<new>' must have the same
7162 type, and the type of '<pointer>' must be a pointer to that type. If the
7163 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
7164 to modify the number or order of execution of this ``cmpxchg`` with
7165 other :ref:`volatile operations <volatile>`.
7167 The success and failure :ref:`ordering <ordering>` arguments specify how this
7168 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
7169 must be at least ``monotonic``, the ordering constraint on failure must be no
7170 stronger than that on success, and the failure ordering cannot be either
7171 ``release`` or ``acq_rel``.
7173 The optional "``singlethread``" argument declares that the ``cmpxchg``
7174 is only atomic with respect to code (usually signal handlers) running in
7175 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
7176 respect to all other code in the system.
7178 The pointer passed into cmpxchg must have alignment greater than or
7179 equal to the size in memory of the operand.
7184 The contents of memory at the location specified by the '``<pointer>``' operand
7185 is read and compared to '``<cmp>``'; if the read value is the equal, the
7186 '``<new>``' is written. The original value at the location is returned, together
7187 with a flag indicating success (true) or failure (false).
7189 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
7190 permitted: the operation may not write ``<new>`` even if the comparison
7193 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
7194 if the value loaded equals ``cmp``.
7196 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
7197 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
7198 load with an ordering parameter determined the second ordering parameter.
7203 .. code-block:: llvm
7206 %orig = atomic load i32, i32* %ptr unordered ; yields i32
7210 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
7211 %squared = mul i32 %cmp, %cmp
7212 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
7213 %value_loaded = extractvalue { i32, i1 } %val_success, 0
7214 %success = extractvalue { i32, i1 } %val_success, 1
7215 br i1 %success, label %done, label %loop
7222 '``atomicrmw``' Instruction
7223 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7230 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
7235 The '``atomicrmw``' instruction is used to atomically modify memory.
7240 There are three arguments to the '``atomicrmw``' instruction: an
7241 operation to apply, an address whose value to modify, an argument to the
7242 operation. The operation must be one of the following keywords:
7256 The type of '<value>' must be an integer type whose bit width is a power
7257 of two greater than or equal to eight and less than or equal to a
7258 target-specific size limit. The type of the '``<pointer>``' operand must
7259 be a pointer to that type. If the ``atomicrmw`` is marked as
7260 ``volatile``, then the optimizer is not allowed to modify the number or
7261 order of execution of this ``atomicrmw`` with other :ref:`volatile
7262 operations <volatile>`.
7267 The contents of memory at the location specified by the '``<pointer>``'
7268 operand are atomically read, modified, and written back. The original
7269 value at the location is returned. The modification is specified by the
7272 - xchg: ``*ptr = val``
7273 - add: ``*ptr = *ptr + val``
7274 - sub: ``*ptr = *ptr - val``
7275 - and: ``*ptr = *ptr & val``
7276 - nand: ``*ptr = ~(*ptr & val)``
7277 - or: ``*ptr = *ptr | val``
7278 - xor: ``*ptr = *ptr ^ val``
7279 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
7280 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
7281 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
7283 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
7289 .. code-block:: llvm
7291 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
7293 .. _i_getelementptr:
7295 '``getelementptr``' Instruction
7296 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7303 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7304 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7305 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
7310 The '``getelementptr``' instruction is used to get the address of a
7311 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
7312 address calculation only and does not access memory. The instruction can also
7313 be used to calculate a vector of such addresses.
7318 The first argument is always a type used as the basis for the calculations.
7319 The second argument is always a pointer or a vector of pointers, and is the
7320 base address to start from. The remaining arguments are indices
7321 that indicate which of the elements of the aggregate object are indexed.
7322 The interpretation of each index is dependent on the type being indexed
7323 into. The first index always indexes the pointer value given as the
7324 first argument, the second index indexes a value of the type pointed to
7325 (not necessarily the value directly pointed to, since the first index
7326 can be non-zero), etc. The first type indexed into must be a pointer
7327 value, subsequent types can be arrays, vectors, and structs. Note that
7328 subsequent types being indexed into can never be pointers, since that
7329 would require loading the pointer before continuing calculation.
7331 The type of each index argument depends on the type it is indexing into.
7332 When indexing into a (optionally packed) structure, only ``i32`` integer
7333 **constants** are allowed (when using a vector of indices they must all
7334 be the **same** ``i32`` integer constant). When indexing into an array,
7335 pointer or vector, integers of any width are allowed, and they are not
7336 required to be constant. These integers are treated as signed values
7339 For example, let's consider a C code fragment and how it gets compiled
7355 int *foo(struct ST *s) {
7356 return &s[1].Z.B[5][13];
7359 The LLVM code generated by Clang is:
7361 .. code-block:: llvm
7363 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
7364 %struct.ST = type { i32, double, %struct.RT }
7366 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
7368 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
7375 In the example above, the first index is indexing into the
7376 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
7377 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
7378 indexes into the third element of the structure, yielding a
7379 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
7380 structure. The third index indexes into the second element of the
7381 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
7382 dimensions of the array are subscripted into, yielding an '``i32``'
7383 type. The '``getelementptr``' instruction returns a pointer to this
7384 element, thus computing a value of '``i32*``' type.
7386 Note that it is perfectly legal to index partially through a structure,
7387 returning a pointer to an inner element. Because of this, the LLVM code
7388 for the given testcase is equivalent to:
7390 .. code-block:: llvm
7392 define i32* @foo(%struct.ST* %s) {
7393 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
7394 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
7395 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
7396 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
7397 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
7401 If the ``inbounds`` keyword is present, the result value of the
7402 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
7403 pointer is not an *in bounds* address of an allocated object, or if any
7404 of the addresses that would be formed by successive addition of the
7405 offsets implied by the indices to the base address with infinitely
7406 precise signed arithmetic are not an *in bounds* address of that
7407 allocated object. The *in bounds* addresses for an allocated object are
7408 all the addresses that point into the object, plus the address one byte
7409 past the end. In cases where the base is a vector of pointers the
7410 ``inbounds`` keyword applies to each of the computations element-wise.
7412 If the ``inbounds`` keyword is not present, the offsets are added to the
7413 base address with silently-wrapping two's complement arithmetic. If the
7414 offsets have a different width from the pointer, they are sign-extended
7415 or truncated to the width of the pointer. The result value of the
7416 ``getelementptr`` may be outside the object pointed to by the base
7417 pointer. The result value may not necessarily be used to access memory
7418 though, even if it happens to point into allocated storage. See the
7419 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
7422 The getelementptr instruction is often confusing. For some more insight
7423 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
7428 .. code-block:: llvm
7430 ; yields [12 x i8]*:aptr
7431 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
7433 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
7435 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
7437 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
7442 The ``getelementptr`` returns a vector of pointers, instead of a single address,
7443 when one or more of its arguments is a vector. In such cases, all vector
7444 arguments should have the same number of elements, and every scalar argument
7445 will be effectively broadcast into a vector during address calculation.
7447 .. code-block:: llvm
7449 ; All arguments are vectors:
7450 ; A[i] = ptrs[i] + offsets[i]*sizeof(i8)
7451 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets
7453 ; Add the same scalar offset to each pointer of a vector:
7454 ; A[i] = ptrs[i] + offset*sizeof(i8)
7455 %A = getelementptr i8, <4 x i8*> %ptrs, i64 %offset
7457 ; Add distinct offsets to the same pointer:
7458 ; A[i] = ptr + offsets[i]*sizeof(i8)
7459 %A = getelementptr i8, i8* %ptr, <4 x i64> %offsets
7461 ; In all cases described above the type of the result is <4 x i8*>
7463 The two following instructions are equivalent:
7465 .. code-block:: llvm
7467 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7468 <4 x i32> <i32 2, i32 2, i32 2, i32 2>,
7469 <4 x i32> <i32 1, i32 1, i32 1, i32 1>,
7471 <4 x i64> <i64 13, i64 13, i64 13, i64 13>
7473 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7474 i32 2, i32 1, <4 x i32> %ind4, i64 13
7476 Let's look at the C code, where the vector version of ``getelementptr``
7481 // Let's assume that we vectorize the following loop:
7482 double *A, B; int *C;
7483 for (int i = 0; i < size; ++i) {
7487 .. code-block:: llvm
7489 ; get pointers for 8 elements from array B
7490 %ptrs = getelementptr double, double* %B, <8 x i32> %C
7491 ; load 8 elements from array B into A
7492 %A = call <8 x double> @llvm.masked.gather.v8f64(<8 x double*> %ptrs,
7493 i32 8, <8 x i1> %mask, <8 x double> %passthru)
7495 Conversion Operations
7496 ---------------------
7498 The instructions in this category are the conversion instructions
7499 (casting) which all take a single operand and a type. They perform
7500 various bit conversions on the operand.
7502 '``trunc .. to``' Instruction
7503 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7510 <result> = trunc <ty> <value> to <ty2> ; yields ty2
7515 The '``trunc``' instruction truncates its operand to the type ``ty2``.
7520 The '``trunc``' instruction takes a value to trunc, and a type to trunc
7521 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
7522 of the same number of integers. The bit size of the ``value`` must be
7523 larger than the bit size of the destination type, ``ty2``. Equal sized
7524 types are not allowed.
7529 The '``trunc``' instruction truncates the high order bits in ``value``
7530 and converts the remaining bits to ``ty2``. Since the source size must
7531 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
7532 It will always truncate bits.
7537 .. code-block:: llvm
7539 %X = trunc i32 257 to i8 ; yields i8:1
7540 %Y = trunc i32 123 to i1 ; yields i1:true
7541 %Z = trunc i32 122 to i1 ; yields i1:false
7542 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
7544 '``zext .. to``' Instruction
7545 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7552 <result> = zext <ty> <value> to <ty2> ; yields ty2
7557 The '``zext``' instruction zero extends its operand to type ``ty2``.
7562 The '``zext``' instruction takes a value to cast, and a type to cast it
7563 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7564 the same number of integers. The bit size of the ``value`` must be
7565 smaller than the bit size of the destination type, ``ty2``.
7570 The ``zext`` fills the high order bits of the ``value`` with zero bits
7571 until it reaches the size of the destination type, ``ty2``.
7573 When zero extending from i1, the result will always be either 0 or 1.
7578 .. code-block:: llvm
7580 %X = zext i32 257 to i64 ; yields i64:257
7581 %Y = zext i1 true to i32 ; yields i32:1
7582 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7584 '``sext .. to``' Instruction
7585 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7592 <result> = sext <ty> <value> to <ty2> ; yields ty2
7597 The '``sext``' sign extends ``value`` to the type ``ty2``.
7602 The '``sext``' instruction takes a value to cast, and a type to cast it
7603 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7604 the same number of integers. The bit size of the ``value`` must be
7605 smaller than the bit size of the destination type, ``ty2``.
7610 The '``sext``' instruction performs a sign extension by copying the sign
7611 bit (highest order bit) of the ``value`` until it reaches the bit size
7612 of the type ``ty2``.
7614 When sign extending from i1, the extension always results in -1 or 0.
7619 .. code-block:: llvm
7621 %X = sext i8 -1 to i16 ; yields i16 :65535
7622 %Y = sext i1 true to i32 ; yields i32:-1
7623 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7625 '``fptrunc .. to``' Instruction
7626 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7633 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
7638 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
7643 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
7644 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
7645 The size of ``value`` must be larger than the size of ``ty2``. This
7646 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
7651 The '``fptrunc``' instruction casts a ``value`` from a larger
7652 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
7653 point <t_floating>` type. If the value cannot fit (i.e. overflows) within the
7654 destination type, ``ty2``, then the results are undefined. If the cast produces
7655 an inexact result, how rounding is performed (e.g. truncation, also known as
7656 round to zero) is undefined.
7661 .. code-block:: llvm
7663 %X = fptrunc double 123.0 to float ; yields float:123.0
7664 %Y = fptrunc double 1.0E+300 to float ; yields undefined
7666 '``fpext .. to``' Instruction
7667 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7674 <result> = fpext <ty> <value> to <ty2> ; yields ty2
7679 The '``fpext``' extends a floating point ``value`` to a larger floating
7685 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
7686 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
7687 to. The source type must be smaller than the destination type.
7692 The '``fpext``' instruction extends the ``value`` from a smaller
7693 :ref:`floating point <t_floating>` type to a larger :ref:`floating
7694 point <t_floating>` type. The ``fpext`` cannot be used to make a
7695 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
7696 *no-op cast* for a floating point cast.
7701 .. code-block:: llvm
7703 %X = fpext float 3.125 to double ; yields double:3.125000e+00
7704 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
7706 '``fptoui .. to``' Instruction
7707 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7714 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
7719 The '``fptoui``' converts a floating point ``value`` to its unsigned
7720 integer equivalent of type ``ty2``.
7725 The '``fptoui``' instruction takes a value to cast, which must be a
7726 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7727 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7728 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7729 type with the same number of elements as ``ty``
7734 The '``fptoui``' instruction converts its :ref:`floating
7735 point <t_floating>` operand into the nearest (rounding towards zero)
7736 unsigned integer value. If the value cannot fit in ``ty2``, the results
7742 .. code-block:: llvm
7744 %X = fptoui double 123.0 to i32 ; yields i32:123
7745 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
7746 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
7748 '``fptosi .. to``' Instruction
7749 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7756 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
7761 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
7762 ``value`` to type ``ty2``.
7767 The '``fptosi``' instruction takes a value to cast, which must be a
7768 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7769 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7770 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7771 type with the same number of elements as ``ty``
7776 The '``fptosi``' instruction converts its :ref:`floating
7777 point <t_floating>` operand into the nearest (rounding towards zero)
7778 signed integer value. If the value cannot fit in ``ty2``, the results
7784 .. code-block:: llvm
7786 %X = fptosi double -123.0 to i32 ; yields i32:-123
7787 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
7788 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
7790 '``uitofp .. to``' Instruction
7791 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7798 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
7803 The '``uitofp``' instruction regards ``value`` as an unsigned integer
7804 and converts that value to the ``ty2`` type.
7809 The '``uitofp``' instruction takes a value to cast, which must be a
7810 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7811 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7812 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7813 type with the same number of elements as ``ty``
7818 The '``uitofp``' instruction interprets its operand as an unsigned
7819 integer quantity and converts it to the corresponding floating point
7820 value. If the value cannot fit in the floating point value, the results
7826 .. code-block:: llvm
7828 %X = uitofp i32 257 to float ; yields float:257.0
7829 %Y = uitofp i8 -1 to double ; yields double:255.0
7831 '``sitofp .. to``' Instruction
7832 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7839 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
7844 The '``sitofp``' instruction regards ``value`` as a signed integer and
7845 converts that value to the ``ty2`` type.
7850 The '``sitofp``' instruction takes a value to cast, which must be a
7851 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7852 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7853 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7854 type with the same number of elements as ``ty``
7859 The '``sitofp``' instruction interprets its operand as a signed integer
7860 quantity and converts it to the corresponding floating point value. If
7861 the value cannot fit in the floating point value, the results are
7867 .. code-block:: llvm
7869 %X = sitofp i32 257 to float ; yields float:257.0
7870 %Y = sitofp i8 -1 to double ; yields double:-1.0
7874 '``ptrtoint .. to``' Instruction
7875 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7882 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
7887 The '``ptrtoint``' instruction converts the pointer or a vector of
7888 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
7893 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
7894 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
7895 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
7896 a vector of integers type.
7901 The '``ptrtoint``' instruction converts ``value`` to integer type
7902 ``ty2`` by interpreting the pointer value as an integer and either
7903 truncating or zero extending that value to the size of the integer type.
7904 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
7905 ``value`` is larger than ``ty2`` then a truncation is done. If they are
7906 the same size, then nothing is done (*no-op cast*) other than a type
7912 .. code-block:: llvm
7914 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
7915 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
7916 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
7920 '``inttoptr .. to``' Instruction
7921 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7928 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
7933 The '``inttoptr``' instruction converts an integer ``value`` to a
7934 pointer type, ``ty2``.
7939 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
7940 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
7946 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
7947 applying either a zero extension or a truncation depending on the size
7948 of the integer ``value``. If ``value`` is larger than the size of a
7949 pointer then a truncation is done. If ``value`` is smaller than the size
7950 of a pointer then a zero extension is done. If they are the same size,
7951 nothing is done (*no-op cast*).
7956 .. code-block:: llvm
7958 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
7959 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
7960 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
7961 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
7965 '``bitcast .. to``' Instruction
7966 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7973 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
7978 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
7984 The '``bitcast``' instruction takes a value to cast, which must be a
7985 non-aggregate first class value, and a type to cast it to, which must
7986 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
7987 bit sizes of ``value`` and the destination type, ``ty2``, must be
7988 identical. If the source type is a pointer, the destination type must
7989 also be a pointer of the same size. This instruction supports bitwise
7990 conversion of vectors to integers and to vectors of other types (as
7991 long as they have the same size).
7996 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
7997 is always a *no-op cast* because no bits change with this
7998 conversion. The conversion is done as if the ``value`` had been stored
7999 to memory and read back as type ``ty2``. Pointer (or vector of
8000 pointers) types may only be converted to other pointer (or vector of
8001 pointers) types with the same address space through this instruction.
8002 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
8003 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
8008 .. code-block:: llvm
8010 %X = bitcast i8 255 to i8 ; yields i8 :-1
8011 %Y = bitcast i32* %x to sint* ; yields sint*:%x
8012 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
8013 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
8015 .. _i_addrspacecast:
8017 '``addrspacecast .. to``' Instruction
8018 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8025 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
8030 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
8031 address space ``n`` to type ``pty2`` in address space ``m``.
8036 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
8037 to cast and a pointer type to cast it to, which must have a different
8043 The '``addrspacecast``' instruction converts the pointer value
8044 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
8045 value modification, depending on the target and the address space
8046 pair. Pointer conversions within the same address space must be
8047 performed with the ``bitcast`` instruction. Note that if the address space
8048 conversion is legal then both result and operand refer to the same memory
8054 .. code-block:: llvm
8056 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
8057 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
8058 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
8065 The instructions in this category are the "miscellaneous" instructions,
8066 which defy better classification.
8070 '``icmp``' Instruction
8071 ^^^^^^^^^^^^^^^^^^^^^^
8078 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8083 The '``icmp``' instruction returns a boolean value or a vector of
8084 boolean values based on comparison of its two integer, integer vector,
8085 pointer, or pointer vector operands.
8090 The '``icmp``' instruction takes three operands. The first operand is
8091 the condition code indicating the kind of comparison to perform. It is
8092 not a value, just a keyword. The possible condition code are:
8095 #. ``ne``: not equal
8096 #. ``ugt``: unsigned greater than
8097 #. ``uge``: unsigned greater or equal
8098 #. ``ult``: unsigned less than
8099 #. ``ule``: unsigned less or equal
8100 #. ``sgt``: signed greater than
8101 #. ``sge``: signed greater or equal
8102 #. ``slt``: signed less than
8103 #. ``sle``: signed less or equal
8105 The remaining two arguments must be :ref:`integer <t_integer>` or
8106 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
8107 must also be identical types.
8112 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
8113 code given as ``cond``. The comparison performed always yields either an
8114 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
8116 #. ``eq``: yields ``true`` if the operands are equal, ``false``
8117 otherwise. No sign interpretation is necessary or performed.
8118 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
8119 otherwise. No sign interpretation is necessary or performed.
8120 #. ``ugt``: interprets the operands as unsigned values and yields
8121 ``true`` if ``op1`` is greater than ``op2``.
8122 #. ``uge``: interprets the operands as unsigned values and yields
8123 ``true`` if ``op1`` is greater than or equal to ``op2``.
8124 #. ``ult``: interprets the operands as unsigned values and yields
8125 ``true`` if ``op1`` is less than ``op2``.
8126 #. ``ule``: interprets the operands as unsigned values and yields
8127 ``true`` if ``op1`` is less than or equal to ``op2``.
8128 #. ``sgt``: interprets the operands as signed values and yields ``true``
8129 if ``op1`` is greater than ``op2``.
8130 #. ``sge``: interprets the operands as signed values and yields ``true``
8131 if ``op1`` is greater than or equal to ``op2``.
8132 #. ``slt``: interprets the operands as signed values and yields ``true``
8133 if ``op1`` is less than ``op2``.
8134 #. ``sle``: interprets the operands as signed values and yields ``true``
8135 if ``op1`` is less than or equal to ``op2``.
8137 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
8138 are compared as if they were integers.
8140 If the operands are integer vectors, then they are compared element by
8141 element. The result is an ``i1`` vector with the same number of elements
8142 as the values being compared. Otherwise, the result is an ``i1``.
8147 .. code-block:: llvm
8149 <result> = icmp eq i32 4, 5 ; yields: result=false
8150 <result> = icmp ne float* %X, %X ; yields: result=false
8151 <result> = icmp ult i16 4, 5 ; yields: result=true
8152 <result> = icmp sgt i16 4, 5 ; yields: result=false
8153 <result> = icmp ule i16 -4, 5 ; yields: result=false
8154 <result> = icmp sge i16 4, 5 ; yields: result=false
8156 Note that the code generator does not yet support vector types with the
8157 ``icmp`` instruction.
8161 '``fcmp``' Instruction
8162 ^^^^^^^^^^^^^^^^^^^^^^
8169 <result> = fcmp [fast-math flags]* <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8174 The '``fcmp``' instruction returns a boolean value or vector of boolean
8175 values based on comparison of its operands.
8177 If the operands are floating point scalars, then the result type is a
8178 boolean (:ref:`i1 <t_integer>`).
8180 If the operands are floating point vectors, then the result type is a
8181 vector of boolean with the same number of elements as the operands being
8187 The '``fcmp``' instruction takes three operands. The first operand is
8188 the condition code indicating the kind of comparison to perform. It is
8189 not a value, just a keyword. The possible condition code are:
8191 #. ``false``: no comparison, always returns false
8192 #. ``oeq``: ordered and equal
8193 #. ``ogt``: ordered and greater than
8194 #. ``oge``: ordered and greater than or equal
8195 #. ``olt``: ordered and less than
8196 #. ``ole``: ordered and less than or equal
8197 #. ``one``: ordered and not equal
8198 #. ``ord``: ordered (no nans)
8199 #. ``ueq``: unordered or equal
8200 #. ``ugt``: unordered or greater than
8201 #. ``uge``: unordered or greater than or equal
8202 #. ``ult``: unordered or less than
8203 #. ``ule``: unordered or less than or equal
8204 #. ``une``: unordered or not equal
8205 #. ``uno``: unordered (either nans)
8206 #. ``true``: no comparison, always returns true
8208 *Ordered* means that neither operand is a QNAN while *unordered* means
8209 that either operand may be a QNAN.
8211 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
8212 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
8213 type. They must have identical types.
8218 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
8219 condition code given as ``cond``. If the operands are vectors, then the
8220 vectors are compared element by element. Each comparison performed
8221 always yields an :ref:`i1 <t_integer>` result, as follows:
8223 #. ``false``: always yields ``false``, regardless of operands.
8224 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
8225 is equal to ``op2``.
8226 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
8227 is greater than ``op2``.
8228 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
8229 is greater than or equal to ``op2``.
8230 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
8231 is less than ``op2``.
8232 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
8233 is less than or equal to ``op2``.
8234 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
8235 is not equal to ``op2``.
8236 #. ``ord``: yields ``true`` if both operands are not a QNAN.
8237 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
8239 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
8240 greater than ``op2``.
8241 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
8242 greater than or equal to ``op2``.
8243 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
8245 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
8246 less than or equal to ``op2``.
8247 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
8248 not equal to ``op2``.
8249 #. ``uno``: yields ``true`` if either operand is a QNAN.
8250 #. ``true``: always yields ``true``, regardless of operands.
8252 The ``fcmp`` instruction can also optionally take any number of
8253 :ref:`fast-math flags <fastmath>`, which are optimization hints to enable
8254 otherwise unsafe floating point optimizations.
8256 Any set of fast-math flags are legal on an ``fcmp`` instruction, but the
8257 only flags that have any effect on its semantics are those that allow
8258 assumptions to be made about the values of input arguments; namely
8259 ``nnan``, ``ninf``, and ``nsz``. See :ref:`fastmath` for more information.
8264 .. code-block:: llvm
8266 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
8267 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
8268 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
8269 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
8271 Note that the code generator does not yet support vector types with the
8272 ``fcmp`` instruction.
8276 '``phi``' Instruction
8277 ^^^^^^^^^^^^^^^^^^^^^
8284 <result> = phi <ty> [ <val0>, <label0>], ...
8289 The '``phi``' instruction is used to implement the φ node in the SSA
8290 graph representing the function.
8295 The type of the incoming values is specified with the first type field.
8296 After this, the '``phi``' instruction takes a list of pairs as
8297 arguments, with one pair for each predecessor basic block of the current
8298 block. Only values of :ref:`first class <t_firstclass>` type may be used as
8299 the value arguments to the PHI node. Only labels may be used as the
8302 There must be no non-phi instructions between the start of a basic block
8303 and the PHI instructions: i.e. PHI instructions must be first in a basic
8306 For the purposes of the SSA form, the use of each incoming value is
8307 deemed to occur on the edge from the corresponding predecessor block to
8308 the current block (but after any definition of an '``invoke``'
8309 instruction's return value on the same edge).
8314 At runtime, the '``phi``' instruction logically takes on the value
8315 specified by the pair corresponding to the predecessor basic block that
8316 executed just prior to the current block.
8321 .. code-block:: llvm
8323 Loop: ; Infinite loop that counts from 0 on up...
8324 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
8325 %nextindvar = add i32 %indvar, 1
8330 '``select``' Instruction
8331 ^^^^^^^^^^^^^^^^^^^^^^^^
8338 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
8340 selty is either i1 or {<N x i1>}
8345 The '``select``' instruction is used to choose one value based on a
8346 condition, without IR-level branching.
8351 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
8352 values indicating the condition, and two values of the same :ref:`first
8353 class <t_firstclass>` type.
8358 If the condition is an i1 and it evaluates to 1, the instruction returns
8359 the first value argument; otherwise, it returns the second value
8362 If the condition is a vector of i1, then the value arguments must be
8363 vectors of the same size, and the selection is done element by element.
8365 If the condition is an i1 and the value arguments are vectors of the
8366 same size, then an entire vector is selected.
8371 .. code-block:: llvm
8373 %X = select i1 true, i8 17, i8 42 ; yields i8:17
8377 '``call``' Instruction
8378 ^^^^^^^^^^^^^^^^^^^^^^
8385 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
8391 The '``call``' instruction represents a simple function call.
8396 This instruction requires several arguments:
8398 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
8399 should perform tail call optimization. The ``tail`` marker is a hint that
8400 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
8401 means that the call must be tail call optimized in order for the program to
8402 be correct. The ``musttail`` marker provides these guarantees:
8404 #. The call will not cause unbounded stack growth if it is part of a
8405 recursive cycle in the call graph.
8406 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
8409 Both markers imply that the callee does not access allocas or varargs from
8410 the caller. Calls marked ``musttail`` must obey the following additional
8413 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
8414 or a pointer bitcast followed by a ret instruction.
8415 - The ret instruction must return the (possibly bitcasted) value
8416 produced by the call or void.
8417 - The caller and callee prototypes must match. Pointer types of
8418 parameters or return types may differ in pointee type, but not
8420 - The calling conventions of the caller and callee must match.
8421 - All ABI-impacting function attributes, such as sret, byval, inreg,
8422 returned, and inalloca, must match.
8423 - The callee must be varargs iff the caller is varargs. Bitcasting a
8424 non-varargs function to the appropriate varargs type is legal so
8425 long as the non-varargs prefixes obey the other rules.
8427 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
8428 the following conditions are met:
8430 - Caller and callee both have the calling convention ``fastcc``.
8431 - The call is in tail position (ret immediately follows call and ret
8432 uses value of call or is void).
8433 - Option ``-tailcallopt`` is enabled, or
8434 ``llvm::GuaranteedTailCallOpt`` is ``true``.
8435 - `Platform-specific constraints are
8436 met. <CodeGenerator.html#tailcallopt>`_
8438 #. The optional "cconv" marker indicates which :ref:`calling
8439 convention <callingconv>` the call should use. If none is
8440 specified, the call defaults to using C calling conventions. The
8441 calling convention of the call must match the calling convention of
8442 the target function, or else the behavior is undefined.
8443 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
8444 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
8446 #. '``ty``': the type of the call instruction itself which is also the
8447 type of the return value. Functions that return no value are marked
8449 #. '``fnty``': shall be the signature of the pointer to function value
8450 being invoked. The argument types must match the types implied by
8451 this signature. This type can be omitted if the function is not
8452 varargs and if the function type does not return a pointer to a
8454 #. '``fnptrval``': An LLVM value containing a pointer to a function to
8455 be invoked. In most cases, this is a direct function invocation, but
8456 indirect ``call``'s are just as possible, calling an arbitrary pointer
8458 #. '``function args``': argument list whose types match the function
8459 signature argument types and parameter attributes. All arguments must
8460 be of :ref:`first class <t_firstclass>` type. If the function signature
8461 indicates the function accepts a variable number of arguments, the
8462 extra arguments can be specified.
8463 #. The optional :ref:`function attributes <fnattrs>` list. Only
8464 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
8465 attributes are valid here.
8466 #. The optional :ref:`operand bundles <opbundles>` list.
8471 The '``call``' instruction is used to cause control flow to transfer to
8472 a specified function, with its incoming arguments bound to the specified
8473 values. Upon a '``ret``' instruction in the called function, control
8474 flow continues with the instruction after the function call, and the
8475 return value of the function is bound to the result argument.
8480 .. code-block:: llvm
8482 %retval = call i32 @test(i32 %argc)
8483 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
8484 %X = tail call i32 @foo() ; yields i32
8485 %Y = tail call fastcc i32 @foo() ; yields i32
8486 call void %foo(i8 97 signext)
8488 %struct.A = type { i32, i8 }
8489 %r = call %struct.A @foo() ; yields { i32, i8 }
8490 %gr = extractvalue %struct.A %r, 0 ; yields i32
8491 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
8492 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
8493 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
8495 llvm treats calls to some functions with names and arguments that match
8496 the standard C99 library as being the C99 library functions, and may
8497 perform optimizations or generate code for them under that assumption.
8498 This is something we'd like to change in the future to provide better
8499 support for freestanding environments and non-C-based languages.
8503 '``va_arg``' Instruction
8504 ^^^^^^^^^^^^^^^^^^^^^^^^
8511 <resultval> = va_arg <va_list*> <arglist>, <argty>
8516 The '``va_arg``' instruction is used to access arguments passed through
8517 the "variable argument" area of a function call. It is used to implement
8518 the ``va_arg`` macro in C.
8523 This instruction takes a ``va_list*`` value and the type of the
8524 argument. It returns a value of the specified argument type and
8525 increments the ``va_list`` to point to the next argument. The actual
8526 type of ``va_list`` is target specific.
8531 The '``va_arg``' instruction loads an argument of the specified type
8532 from the specified ``va_list`` and causes the ``va_list`` to point to
8533 the next argument. For more information, see the variable argument
8534 handling :ref:`Intrinsic Functions <int_varargs>`.
8536 It is legal for this instruction to be called in a function which does
8537 not take a variable number of arguments, for example, the ``vfprintf``
8540 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
8541 function <intrinsics>` because it takes a type as an argument.
8546 See the :ref:`variable argument processing <int_varargs>` section.
8548 Note that the code generator does not yet fully support va\_arg on many
8549 targets. Also, it does not currently support va\_arg with aggregate
8550 types on any target.
8554 '``landingpad``' Instruction
8555 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8562 <resultval> = landingpad <resultty> <clause>+
8563 <resultval> = landingpad <resultty> cleanup <clause>*
8565 <clause> := catch <type> <value>
8566 <clause> := filter <array constant type> <array constant>
8571 The '``landingpad``' instruction is used by `LLVM's exception handling
8572 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8573 is a landing pad --- one where the exception lands, and corresponds to the
8574 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
8575 defines values supplied by the :ref:`personality function <personalityfn>` upon
8576 re-entry to the function. The ``resultval`` has the type ``resultty``.
8582 ``cleanup`` flag indicates that the landing pad block is a cleanup.
8584 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
8585 contains the global variable representing the "type" that may be caught
8586 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
8587 clause takes an array constant as its argument. Use
8588 "``[0 x i8**] undef``" for a filter which cannot throw. The
8589 '``landingpad``' instruction must contain *at least* one ``clause`` or
8590 the ``cleanup`` flag.
8595 The '``landingpad``' instruction defines the values which are set by the
8596 :ref:`personality function <personalityfn>` upon re-entry to the function, and
8597 therefore the "result type" of the ``landingpad`` instruction. As with
8598 calling conventions, how the personality function results are
8599 represented in LLVM IR is target specific.
8601 The clauses are applied in order from top to bottom. If two
8602 ``landingpad`` instructions are merged together through inlining, the
8603 clauses from the calling function are appended to the list of clauses.
8604 When the call stack is being unwound due to an exception being thrown,
8605 the exception is compared against each ``clause`` in turn. If it doesn't
8606 match any of the clauses, and the ``cleanup`` flag is not set, then
8607 unwinding continues further up the call stack.
8609 The ``landingpad`` instruction has several restrictions:
8611 - A landing pad block is a basic block which is the unwind destination
8612 of an '``invoke``' instruction.
8613 - A landing pad block must have a '``landingpad``' instruction as its
8614 first non-PHI instruction.
8615 - There can be only one '``landingpad``' instruction within the landing
8617 - A basic block that is not a landing pad block may not include a
8618 '``landingpad``' instruction.
8623 .. code-block:: llvm
8625 ;; A landing pad which can catch an integer.
8626 %res = landingpad { i8*, i32 }
8628 ;; A landing pad that is a cleanup.
8629 %res = landingpad { i8*, i32 }
8631 ;; A landing pad which can catch an integer and can only throw a double.
8632 %res = landingpad { i8*, i32 }
8634 filter [1 x i8**] [@_ZTId]
8638 '``cleanuppad``' Instruction
8639 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8646 <resultval> = cleanuppad [<args>*]
8651 The '``cleanuppad``' instruction is used by `LLVM's exception handling
8652 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8653 is a cleanup block --- one where a personality routine attempts to
8654 transfer control to run cleanup actions.
8655 The ``args`` correspond to whatever additional
8656 information the :ref:`personality function <personalityfn>` requires to
8657 execute the cleanup.
8658 The ``resultval`` has the type :ref:`token <t_token>` and is used to
8659 match the ``cleanuppad`` to corresponding :ref:`cleanuprets <i_cleanupret>`
8660 and :ref:`cleanupendpads <i_cleanupendpad>`.
8665 The instruction takes a list of arbitrary values which are interpreted
8666 by the :ref:`personality function <personalityfn>`.
8671 When the call stack is being unwound due to an exception being thrown,
8672 the :ref:`personality function <personalityfn>` transfers control to the
8673 ``cleanuppad`` with the aid of the personality-specific arguments.
8674 As with calling conventions, how the personality function results are
8675 represented in LLVM IR is target specific.
8677 The ``cleanuppad`` instruction has several restrictions:
8679 - A cleanup block is a basic block which is the unwind destination of
8680 an exceptional instruction.
8681 - A cleanup block must have a '``cleanuppad``' instruction as its
8682 first non-PHI instruction.
8683 - There can be only one '``cleanuppad``' instruction within the
8685 - A basic block that is not a cleanup block may not include a
8686 '``cleanuppad``' instruction.
8687 - All '``cleanupret``'s and '``cleanupendpad``'s which consume a ``cleanuppad``
8688 must have the same exceptional successor.
8689 - It is undefined behavior for control to transfer from a ``cleanuppad`` to a
8690 ``ret`` without first executing a ``cleanupret`` or ``cleanupendpad`` that
8691 consumes the ``cleanuppad``.
8692 - It is undefined behavior for control to transfer from a ``cleanuppad`` to
8693 itself without first executing a ``cleanupret`` or ``cleanupendpad`` that
8694 consumes the ``cleanuppad``.
8699 .. code-block:: llvm
8701 %tok = cleanuppad []
8708 LLVM supports the notion of an "intrinsic function". These functions
8709 have well known names and semantics and are required to follow certain
8710 restrictions. Overall, these intrinsics represent an extension mechanism
8711 for the LLVM language that does not require changing all of the
8712 transformations in LLVM when adding to the language (or the bitcode
8713 reader/writer, the parser, etc...).
8715 Intrinsic function names must all start with an "``llvm.``" prefix. This
8716 prefix is reserved in LLVM for intrinsic names; thus, function names may
8717 not begin with this prefix. Intrinsic functions must always be external
8718 functions: you cannot define the body of intrinsic functions. Intrinsic
8719 functions may only be used in call or invoke instructions: it is illegal
8720 to take the address of an intrinsic function. Additionally, because
8721 intrinsic functions are part of the LLVM language, it is required if any
8722 are added that they be documented here.
8724 Some intrinsic functions can be overloaded, i.e., the intrinsic
8725 represents a family of functions that perform the same operation but on
8726 different data types. Because LLVM can represent over 8 million
8727 different integer types, overloading is used commonly to allow an
8728 intrinsic function to operate on any integer type. One or more of the
8729 argument types or the result type can be overloaded to accept any
8730 integer type. Argument types may also be defined as exactly matching a
8731 previous argument's type or the result type. This allows an intrinsic
8732 function which accepts multiple arguments, but needs all of them to be
8733 of the same type, to only be overloaded with respect to a single
8734 argument or the result.
8736 Overloaded intrinsics will have the names of its overloaded argument
8737 types encoded into its function name, each preceded by a period. Only
8738 those types which are overloaded result in a name suffix. Arguments
8739 whose type is matched against another type do not. For example, the
8740 ``llvm.ctpop`` function can take an integer of any width and returns an
8741 integer of exactly the same integer width. This leads to a family of
8742 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
8743 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
8744 overloaded, and only one type suffix is required. Because the argument's
8745 type is matched against the return type, it does not require its own
8748 To learn how to add an intrinsic function, please see the `Extending
8749 LLVM Guide <ExtendingLLVM.html>`_.
8753 Variable Argument Handling Intrinsics
8754 -------------------------------------
8756 Variable argument support is defined in LLVM with the
8757 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
8758 functions. These functions are related to the similarly named macros
8759 defined in the ``<stdarg.h>`` header file.
8761 All of these functions operate on arguments that use a target-specific
8762 value type "``va_list``". The LLVM assembly language reference manual
8763 does not define what this type is, so all transformations should be
8764 prepared to handle these functions regardless of the type used.
8766 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
8767 variable argument handling intrinsic functions are used.
8769 .. code-block:: llvm
8771 ; This struct is different for every platform. For most platforms,
8772 ; it is merely an i8*.
8773 %struct.va_list = type { i8* }
8775 ; For Unix x86_64 platforms, va_list is the following struct:
8776 ; %struct.va_list = type { i32, i32, i8*, i8* }
8778 define i32 @test(i32 %X, ...) {
8779 ; Initialize variable argument processing
8780 %ap = alloca %struct.va_list
8781 %ap2 = bitcast %struct.va_list* %ap to i8*
8782 call void @llvm.va_start(i8* %ap2)
8784 ; Read a single integer argument
8785 %tmp = va_arg i8* %ap2, i32
8787 ; Demonstrate usage of llvm.va_copy and llvm.va_end
8789 %aq2 = bitcast i8** %aq to i8*
8790 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
8791 call void @llvm.va_end(i8* %aq2)
8793 ; Stop processing of arguments.
8794 call void @llvm.va_end(i8* %ap2)
8798 declare void @llvm.va_start(i8*)
8799 declare void @llvm.va_copy(i8*, i8*)
8800 declare void @llvm.va_end(i8*)
8804 '``llvm.va_start``' Intrinsic
8805 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8812 declare void @llvm.va_start(i8* <arglist>)
8817 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
8818 subsequent use by ``va_arg``.
8823 The argument is a pointer to a ``va_list`` element to initialize.
8828 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
8829 available in C. In a target-dependent way, it initializes the
8830 ``va_list`` element to which the argument points, so that the next call
8831 to ``va_arg`` will produce the first variable argument passed to the
8832 function. Unlike the C ``va_start`` macro, this intrinsic does not need
8833 to know the last argument of the function as the compiler can figure
8836 '``llvm.va_end``' Intrinsic
8837 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8844 declare void @llvm.va_end(i8* <arglist>)
8849 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
8850 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
8855 The argument is a pointer to a ``va_list`` to destroy.
8860 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
8861 available in C. In a target-dependent way, it destroys the ``va_list``
8862 element to which the argument points. Calls to
8863 :ref:`llvm.va_start <int_va_start>` and
8864 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
8869 '``llvm.va_copy``' Intrinsic
8870 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8877 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
8882 The '``llvm.va_copy``' intrinsic copies the current argument position
8883 from the source argument list to the destination argument list.
8888 The first argument is a pointer to a ``va_list`` element to initialize.
8889 The second argument is a pointer to a ``va_list`` element to copy from.
8894 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
8895 available in C. In a target-dependent way, it copies the source
8896 ``va_list`` element into the destination ``va_list`` element. This
8897 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
8898 arbitrarily complex and require, for example, memory allocation.
8900 Accurate Garbage Collection Intrinsics
8901 --------------------------------------
8903 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
8904 (GC) requires the frontend to generate code containing appropriate intrinsic
8905 calls and select an appropriate GC strategy which knows how to lower these
8906 intrinsics in a manner which is appropriate for the target collector.
8908 These intrinsics allow identification of :ref:`GC roots on the
8909 stack <int_gcroot>`, as well as garbage collector implementations that
8910 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
8911 Frontends for type-safe garbage collected languages should generate
8912 these intrinsics to make use of the LLVM garbage collectors. For more
8913 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
8915 Experimental Statepoint Intrinsics
8916 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8918 LLVM provides an second experimental set of intrinsics for describing garbage
8919 collection safepoints in compiled code. These intrinsics are an alternative
8920 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
8921 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
8922 differences in approach are covered in the `Garbage Collection with LLVM
8923 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
8924 described in :doc:`Statepoints`.
8928 '``llvm.gcroot``' Intrinsic
8929 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8936 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
8941 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
8942 the code generator, and allows some metadata to be associated with it.
8947 The first argument specifies the address of a stack object that contains
8948 the root pointer. The second pointer (which must be either a constant or
8949 a global value address) contains the meta-data to be associated with the
8955 At runtime, a call to this intrinsic stores a null pointer into the
8956 "ptrloc" location. At compile-time, the code generator generates
8957 information to allow the runtime to find the pointer at GC safe points.
8958 The '``llvm.gcroot``' intrinsic may only be used in a function which
8959 :ref:`specifies a GC algorithm <gc>`.
8963 '``llvm.gcread``' Intrinsic
8964 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8971 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
8976 The '``llvm.gcread``' intrinsic identifies reads of references from heap
8977 locations, allowing garbage collector implementations that require read
8983 The second argument is the address to read from, which should be an
8984 address allocated from the garbage collector. The first object is a
8985 pointer to the start of the referenced object, if needed by the language
8986 runtime (otherwise null).
8991 The '``llvm.gcread``' intrinsic has the same semantics as a load
8992 instruction, but may be replaced with substantially more complex code by
8993 the garbage collector runtime, as needed. The '``llvm.gcread``'
8994 intrinsic may only be used in a function which :ref:`specifies a GC
8999 '``llvm.gcwrite``' Intrinsic
9000 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9007 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
9012 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
9013 locations, allowing garbage collector implementations that require write
9014 barriers (such as generational or reference counting collectors).
9019 The first argument is the reference to store, the second is the start of
9020 the object to store it to, and the third is the address of the field of
9021 Obj to store to. If the runtime does not require a pointer to the
9022 object, Obj may be null.
9027 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
9028 instruction, but may be replaced with substantially more complex code by
9029 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
9030 intrinsic may only be used in a function which :ref:`specifies a GC
9033 Code Generator Intrinsics
9034 -------------------------
9036 These intrinsics are provided by LLVM to expose special features that
9037 may only be implemented with code generator support.
9039 '``llvm.returnaddress``' Intrinsic
9040 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9047 declare i8 *@llvm.returnaddress(i32 <level>)
9052 The '``llvm.returnaddress``' intrinsic attempts to compute a
9053 target-specific value indicating the return address of the current
9054 function or one of its callers.
9059 The argument to this intrinsic indicates which function to return the
9060 address for. Zero indicates the calling function, one indicates its
9061 caller, etc. The argument is **required** to be a constant integer
9067 The '``llvm.returnaddress``' intrinsic either returns a pointer
9068 indicating the return address of the specified call frame, or zero if it
9069 cannot be identified. The value returned by this intrinsic is likely to
9070 be incorrect or 0 for arguments other than zero, so it should only be
9071 used for debugging purposes.
9073 Note that calling this intrinsic does not prevent function inlining or
9074 other aggressive transformations, so the value returned may not be that
9075 of the obvious source-language caller.
9077 '``llvm.frameaddress``' Intrinsic
9078 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9085 declare i8* @llvm.frameaddress(i32 <level>)
9090 The '``llvm.frameaddress``' intrinsic attempts to return the
9091 target-specific frame pointer value for the specified stack frame.
9096 The argument to this intrinsic indicates which function to return the
9097 frame pointer for. Zero indicates the calling function, one indicates
9098 its caller, etc. The argument is **required** to be a constant integer
9104 The '``llvm.frameaddress``' intrinsic either returns a pointer
9105 indicating the frame address of the specified call frame, or zero if it
9106 cannot be identified. The value returned by this intrinsic is likely to
9107 be incorrect or 0 for arguments other than zero, so it should only be
9108 used for debugging purposes.
9110 Note that calling this intrinsic does not prevent function inlining or
9111 other aggressive transformations, so the value returned may not be that
9112 of the obvious source-language caller.
9114 '``llvm.localescape``' and '``llvm.localrecover``' Intrinsics
9115 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9122 declare void @llvm.localescape(...)
9123 declare i8* @llvm.localrecover(i8* %func, i8* %fp, i32 %idx)
9128 The '``llvm.localescape``' intrinsic escapes offsets of a collection of static
9129 allocas, and the '``llvm.localrecover``' intrinsic applies those offsets to a
9130 live frame pointer to recover the address of the allocation. The offset is
9131 computed during frame layout of the caller of ``llvm.localescape``.
9136 All arguments to '``llvm.localescape``' must be pointers to static allocas or
9137 casts of static allocas. Each function can only call '``llvm.localescape``'
9138 once, and it can only do so from the entry block.
9140 The ``func`` argument to '``llvm.localrecover``' must be a constant
9141 bitcasted pointer to a function defined in the current module. The code
9142 generator cannot determine the frame allocation offset of functions defined in
9145 The ``fp`` argument to '``llvm.localrecover``' must be a frame pointer of a
9146 call frame that is currently live. The return value of '``llvm.localaddress``'
9147 is one way to produce such a value, but various runtimes also expose a suitable
9148 pointer in platform-specific ways.
9150 The ``idx`` argument to '``llvm.localrecover``' indicates which alloca passed to
9151 '``llvm.localescape``' to recover. It is zero-indexed.
9156 These intrinsics allow a group of functions to share access to a set of local
9157 stack allocations of a one parent function. The parent function may call the
9158 '``llvm.localescape``' intrinsic once from the function entry block, and the
9159 child functions can use '``llvm.localrecover``' to access the escaped allocas.
9160 The '``llvm.localescape``' intrinsic blocks inlining, as inlining changes where
9161 the escaped allocas are allocated, which would break attempts to use
9162 '``llvm.localrecover``'.
9164 .. _int_read_register:
9165 .. _int_write_register:
9167 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
9168 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9175 declare i32 @llvm.read_register.i32(metadata)
9176 declare i64 @llvm.read_register.i64(metadata)
9177 declare void @llvm.write_register.i32(metadata, i32 @value)
9178 declare void @llvm.write_register.i64(metadata, i64 @value)
9184 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
9185 provides access to the named register. The register must be valid on
9186 the architecture being compiled to. The type needs to be compatible
9187 with the register being read.
9192 The '``llvm.read_register``' intrinsic returns the current value of the
9193 register, where possible. The '``llvm.write_register``' intrinsic sets
9194 the current value of the register, where possible.
9196 This is useful to implement named register global variables that need
9197 to always be mapped to a specific register, as is common practice on
9198 bare-metal programs including OS kernels.
9200 The compiler doesn't check for register availability or use of the used
9201 register in surrounding code, including inline assembly. Because of that,
9202 allocatable registers are not supported.
9204 Warning: So far it only works with the stack pointer on selected
9205 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
9206 work is needed to support other registers and even more so, allocatable
9211 '``llvm.stacksave``' Intrinsic
9212 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9219 declare i8* @llvm.stacksave()
9224 The '``llvm.stacksave``' intrinsic is used to remember the current state
9225 of the function stack, for use with
9226 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
9227 implementing language features like scoped automatic variable sized
9233 This intrinsic returns a opaque pointer value that can be passed to
9234 :ref:`llvm.stackrestore <int_stackrestore>`. When an
9235 ``llvm.stackrestore`` intrinsic is executed with a value saved from
9236 ``llvm.stacksave``, it effectively restores the state of the stack to
9237 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
9238 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
9239 were allocated after the ``llvm.stacksave`` was executed.
9241 .. _int_stackrestore:
9243 '``llvm.stackrestore``' Intrinsic
9244 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9251 declare void @llvm.stackrestore(i8* %ptr)
9256 The '``llvm.stackrestore``' intrinsic is used to restore the state of
9257 the function stack to the state it was in when the corresponding
9258 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
9259 useful for implementing language features like scoped automatic variable
9260 sized arrays in C99.
9265 See the description for :ref:`llvm.stacksave <int_stacksave>`.
9267 '``llvm.prefetch``' Intrinsic
9268 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9275 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
9280 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
9281 insert a prefetch instruction if supported; otherwise, it is a noop.
9282 Prefetches have no effect on the behavior of the program but can change
9283 its performance characteristics.
9288 ``address`` is the address to be prefetched, ``rw`` is the specifier
9289 determining if the fetch should be for a read (0) or write (1), and
9290 ``locality`` is a temporal locality specifier ranging from (0) - no
9291 locality, to (3) - extremely local keep in cache. The ``cache type``
9292 specifies whether the prefetch is performed on the data (1) or
9293 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
9294 arguments must be constant integers.
9299 This intrinsic does not modify the behavior of the program. In
9300 particular, prefetches cannot trap and do not produce a value. On
9301 targets that support this intrinsic, the prefetch can provide hints to
9302 the processor cache for better performance.
9304 '``llvm.pcmarker``' Intrinsic
9305 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9312 declare void @llvm.pcmarker(i32 <id>)
9317 The '``llvm.pcmarker``' intrinsic is a method to export a Program
9318 Counter (PC) in a region of code to simulators and other tools. The
9319 method is target specific, but it is expected that the marker will use
9320 exported symbols to transmit the PC of the marker. The marker makes no
9321 guarantees that it will remain with any specific instruction after
9322 optimizations. It is possible that the presence of a marker will inhibit
9323 optimizations. The intended use is to be inserted after optimizations to
9324 allow correlations of simulation runs.
9329 ``id`` is a numerical id identifying the marker.
9334 This intrinsic does not modify the behavior of the program. Backends
9335 that do not support this intrinsic may ignore it.
9337 '``llvm.readcyclecounter``' Intrinsic
9338 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9345 declare i64 @llvm.readcyclecounter()
9350 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
9351 counter register (or similar low latency, high accuracy clocks) on those
9352 targets that support it. On X86, it should map to RDTSC. On Alpha, it
9353 should map to RPCC. As the backing counters overflow quickly (on the
9354 order of 9 seconds on alpha), this should only be used for small
9360 When directly supported, reading the cycle counter should not modify any
9361 memory. Implementations are allowed to either return a application
9362 specific value or a system wide value. On backends without support, this
9363 is lowered to a constant 0.
9365 Note that runtime support may be conditional on the privilege-level code is
9366 running at and the host platform.
9368 '``llvm.clear_cache``' Intrinsic
9369 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9376 declare void @llvm.clear_cache(i8*, i8*)
9381 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
9382 in the specified range to the execution unit of the processor. On
9383 targets with non-unified instruction and data cache, the implementation
9384 flushes the instruction cache.
9389 On platforms with coherent instruction and data caches (e.g. x86), this
9390 intrinsic is a nop. On platforms with non-coherent instruction and data
9391 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
9392 instructions or a system call, if cache flushing requires special
9395 The default behavior is to emit a call to ``__clear_cache`` from the run
9398 This instrinsic does *not* empty the instruction pipeline. Modifications
9399 of the current function are outside the scope of the intrinsic.
9401 '``llvm.instrprof_increment``' Intrinsic
9402 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9409 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
9410 i32 <num-counters>, i32 <index>)
9415 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
9416 frontend for use with instrumentation based profiling. These will be
9417 lowered by the ``-instrprof`` pass to generate execution counts of a
9423 The first argument is a pointer to a global variable containing the
9424 name of the entity being instrumented. This should generally be the
9425 (mangled) function name for a set of counters.
9427 The second argument is a hash value that can be used by the consumer
9428 of the profile data to detect changes to the instrumented source, and
9429 the third is the number of counters associated with ``name``. It is an
9430 error if ``hash`` or ``num-counters`` differ between two instances of
9431 ``instrprof_increment`` that refer to the same name.
9433 The last argument refers to which of the counters for ``name`` should
9434 be incremented. It should be a value between 0 and ``num-counters``.
9439 This intrinsic represents an increment of a profiling counter. It will
9440 cause the ``-instrprof`` pass to generate the appropriate data
9441 structures and the code to increment the appropriate value, in a
9442 format that can be written out by a compiler runtime and consumed via
9443 the ``llvm-profdata`` tool.
9445 Standard C Library Intrinsics
9446 -----------------------------
9448 LLVM provides intrinsics for a few important standard C library
9449 functions. These intrinsics allow source-language front-ends to pass
9450 information about the alignment of the pointer arguments to the code
9451 generator, providing opportunity for more efficient code generation.
9455 '``llvm.memcpy``' Intrinsic
9456 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9461 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
9462 integer bit width and for different address spaces. Not all targets
9463 support all bit widths however.
9467 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9468 i32 <len>, i32 <align>, i1 <isvolatile>)
9469 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9470 i64 <len>, i32 <align>, i1 <isvolatile>)
9475 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9476 source location to the destination location.
9478 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
9479 intrinsics do not return a value, takes extra alignment/isvolatile
9480 arguments and the pointers can be in specified address spaces.
9485 The first argument is a pointer to the destination, the second is a
9486 pointer to the source. The third argument is an integer argument
9487 specifying the number of bytes to copy, the fourth argument is the
9488 alignment of the source and destination locations, and the fifth is a
9489 boolean indicating a volatile access.
9491 If the call to this intrinsic has an alignment value that is not 0 or 1,
9492 then the caller guarantees that both the source and destination pointers
9493 are aligned to that boundary.
9495 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
9496 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9497 very cleanly specified and it is unwise to depend on it.
9502 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9503 source location to the destination location, which are not allowed to
9504 overlap. It copies "len" bytes of memory over. If the argument is known
9505 to be aligned to some boundary, this can be specified as the fourth
9506 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
9508 '``llvm.memmove``' Intrinsic
9509 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9514 This is an overloaded intrinsic. You can use llvm.memmove on any integer
9515 bit width and for different address space. Not all targets support all
9520 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9521 i32 <len>, i32 <align>, i1 <isvolatile>)
9522 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9523 i64 <len>, i32 <align>, i1 <isvolatile>)
9528 The '``llvm.memmove.*``' intrinsics move a block of memory from the
9529 source location to the destination location. It is similar to the
9530 '``llvm.memcpy``' intrinsic but allows the two memory locations to
9533 Note that, unlike the standard libc function, the ``llvm.memmove.*``
9534 intrinsics do not return a value, takes extra alignment/isvolatile
9535 arguments and the pointers can be in specified address spaces.
9540 The first argument is a pointer to the destination, the second is a
9541 pointer to the source. The third argument is an integer argument
9542 specifying the number of bytes to copy, the fourth argument is the
9543 alignment of the source and destination locations, and the fifth is a
9544 boolean indicating a volatile access.
9546 If the call to this intrinsic has an alignment value that is not 0 or 1,
9547 then the caller guarantees that the source and destination pointers are
9548 aligned to that boundary.
9550 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
9551 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
9552 not very cleanly specified and it is unwise to depend on it.
9557 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
9558 source location to the destination location, which may overlap. It
9559 copies "len" bytes of memory over. If the argument is known to be
9560 aligned to some boundary, this can be specified as the fourth argument,
9561 otherwise it should be set to 0 or 1 (both meaning no alignment).
9563 '``llvm.memset.*``' Intrinsics
9564 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9569 This is an overloaded intrinsic. You can use llvm.memset on any integer
9570 bit width and for different address spaces. However, not all targets
9571 support all bit widths.
9575 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
9576 i32 <len>, i32 <align>, i1 <isvolatile>)
9577 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
9578 i64 <len>, i32 <align>, i1 <isvolatile>)
9583 The '``llvm.memset.*``' intrinsics fill a block of memory with a
9584 particular byte value.
9586 Note that, unlike the standard libc function, the ``llvm.memset``
9587 intrinsic does not return a value and takes extra alignment/volatile
9588 arguments. Also, the destination can be in an arbitrary address space.
9593 The first argument is a pointer to the destination to fill, the second
9594 is the byte value with which to fill it, the third argument is an
9595 integer argument specifying the number of bytes to fill, and the fourth
9596 argument is the known alignment of the destination location.
9598 If the call to this intrinsic has an alignment value that is not 0 or 1,
9599 then the caller guarantees that the destination pointer is aligned to
9602 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
9603 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9604 very cleanly specified and it is unwise to depend on it.
9609 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
9610 at the destination location. If the argument is known to be aligned to
9611 some boundary, this can be specified as the fourth argument, otherwise
9612 it should be set to 0 or 1 (both meaning no alignment).
9614 '``llvm.sqrt.*``' Intrinsic
9615 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9620 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
9621 floating point or vector of floating point type. Not all targets support
9626 declare float @llvm.sqrt.f32(float %Val)
9627 declare double @llvm.sqrt.f64(double %Val)
9628 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
9629 declare fp128 @llvm.sqrt.f128(fp128 %Val)
9630 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
9635 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
9636 returning the same value as the libm '``sqrt``' functions would. Unlike
9637 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
9638 negative numbers other than -0.0 (which allows for better optimization,
9639 because there is no need to worry about errno being set).
9640 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
9645 The argument and return value are floating point numbers of the same
9651 This function returns the sqrt of the specified operand if it is a
9652 nonnegative floating point number.
9654 '``llvm.powi.*``' Intrinsic
9655 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9660 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
9661 floating point or vector of floating point type. Not all targets support
9666 declare float @llvm.powi.f32(float %Val, i32 %power)
9667 declare double @llvm.powi.f64(double %Val, i32 %power)
9668 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
9669 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
9670 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
9675 The '``llvm.powi.*``' intrinsics return the first operand raised to the
9676 specified (positive or negative) power. The order of evaluation of
9677 multiplications is not defined. When a vector of floating point type is
9678 used, the second argument remains a scalar integer value.
9683 The second argument is an integer power, and the first is a value to
9684 raise to that power.
9689 This function returns the first value raised to the second power with an
9690 unspecified sequence of rounding operations.
9692 '``llvm.sin.*``' Intrinsic
9693 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9698 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
9699 floating point or vector of floating point type. Not all targets support
9704 declare float @llvm.sin.f32(float %Val)
9705 declare double @llvm.sin.f64(double %Val)
9706 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
9707 declare fp128 @llvm.sin.f128(fp128 %Val)
9708 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
9713 The '``llvm.sin.*``' intrinsics return the sine of the operand.
9718 The argument and return value are floating point numbers of the same
9724 This function returns the sine of the specified operand, returning the
9725 same values as the libm ``sin`` functions would, and handles error
9726 conditions in the same way.
9728 '``llvm.cos.*``' Intrinsic
9729 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9734 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
9735 floating point or vector of floating point type. Not all targets support
9740 declare float @llvm.cos.f32(float %Val)
9741 declare double @llvm.cos.f64(double %Val)
9742 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
9743 declare fp128 @llvm.cos.f128(fp128 %Val)
9744 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
9749 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
9754 The argument and return value are floating point numbers of the same
9760 This function returns the cosine of the specified operand, returning the
9761 same values as the libm ``cos`` functions would, and handles error
9762 conditions in the same way.
9764 '``llvm.pow.*``' Intrinsic
9765 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9770 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
9771 floating point or vector of floating point type. Not all targets support
9776 declare float @llvm.pow.f32(float %Val, float %Power)
9777 declare double @llvm.pow.f64(double %Val, double %Power)
9778 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
9779 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
9780 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
9785 The '``llvm.pow.*``' intrinsics return the first operand raised to the
9786 specified (positive or negative) power.
9791 The second argument is a floating point power, and the first is a value
9792 to raise to that power.
9797 This function returns the first value raised to the second power,
9798 returning the same values as the libm ``pow`` functions would, and
9799 handles error conditions in the same way.
9801 '``llvm.exp.*``' Intrinsic
9802 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9807 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
9808 floating point or vector of floating point type. Not all targets support
9813 declare float @llvm.exp.f32(float %Val)
9814 declare double @llvm.exp.f64(double %Val)
9815 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
9816 declare fp128 @llvm.exp.f128(fp128 %Val)
9817 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
9822 The '``llvm.exp.*``' intrinsics perform the exp function.
9827 The argument and return value are floating point numbers of the same
9833 This function returns the same values as the libm ``exp`` functions
9834 would, and handles error conditions in the same way.
9836 '``llvm.exp2.*``' Intrinsic
9837 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9842 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
9843 floating point or vector of floating point type. Not all targets support
9848 declare float @llvm.exp2.f32(float %Val)
9849 declare double @llvm.exp2.f64(double %Val)
9850 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
9851 declare fp128 @llvm.exp2.f128(fp128 %Val)
9852 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
9857 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
9862 The argument and return value are floating point numbers of the same
9868 This function returns the same values as the libm ``exp2`` functions
9869 would, and handles error conditions in the same way.
9871 '``llvm.log.*``' Intrinsic
9872 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9877 This is an overloaded intrinsic. You can use ``llvm.log`` on any
9878 floating point or vector of floating point type. Not all targets support
9883 declare float @llvm.log.f32(float %Val)
9884 declare double @llvm.log.f64(double %Val)
9885 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
9886 declare fp128 @llvm.log.f128(fp128 %Val)
9887 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
9892 The '``llvm.log.*``' intrinsics perform the log function.
9897 The argument and return value are floating point numbers of the same
9903 This function returns the same values as the libm ``log`` functions
9904 would, and handles error conditions in the same way.
9906 '``llvm.log10.*``' Intrinsic
9907 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9912 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
9913 floating point or vector of floating point type. Not all targets support
9918 declare float @llvm.log10.f32(float %Val)
9919 declare double @llvm.log10.f64(double %Val)
9920 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
9921 declare fp128 @llvm.log10.f128(fp128 %Val)
9922 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
9927 The '``llvm.log10.*``' intrinsics perform the log10 function.
9932 The argument and return value are floating point numbers of the same
9938 This function returns the same values as the libm ``log10`` functions
9939 would, and handles error conditions in the same way.
9941 '``llvm.log2.*``' Intrinsic
9942 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9947 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
9948 floating point or vector of floating point type. Not all targets support
9953 declare float @llvm.log2.f32(float %Val)
9954 declare double @llvm.log2.f64(double %Val)
9955 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
9956 declare fp128 @llvm.log2.f128(fp128 %Val)
9957 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
9962 The '``llvm.log2.*``' intrinsics perform the log2 function.
9967 The argument and return value are floating point numbers of the same
9973 This function returns the same values as the libm ``log2`` functions
9974 would, and handles error conditions in the same way.
9976 '``llvm.fma.*``' Intrinsic
9977 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9982 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
9983 floating point or vector of floating point type. Not all targets support
9988 declare float @llvm.fma.f32(float %a, float %b, float %c)
9989 declare double @llvm.fma.f64(double %a, double %b, double %c)
9990 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
9991 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
9992 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
9997 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
10003 The argument and return value are floating point numbers of the same
10009 This function returns the same values as the libm ``fma`` functions
10010 would, and does not set errno.
10012 '``llvm.fabs.*``' Intrinsic
10013 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10018 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
10019 floating point or vector of floating point type. Not all targets support
10024 declare float @llvm.fabs.f32(float %Val)
10025 declare double @llvm.fabs.f64(double %Val)
10026 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
10027 declare fp128 @llvm.fabs.f128(fp128 %Val)
10028 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
10033 The '``llvm.fabs.*``' intrinsics return the absolute value of the
10039 The argument and return value are floating point numbers of the same
10045 This function returns the same values as the libm ``fabs`` functions
10046 would, and handles error conditions in the same way.
10048 '``llvm.minnum.*``' Intrinsic
10049 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10054 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
10055 floating point or vector of floating point type. Not all targets support
10060 declare float @llvm.minnum.f32(float %Val0, float %Val1)
10061 declare double @llvm.minnum.f64(double %Val0, double %Val1)
10062 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10063 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
10064 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10069 The '``llvm.minnum.*``' intrinsics return the minimum of the two
10076 The arguments and return value are floating point numbers of the same
10082 Follows the IEEE-754 semantics for minNum, which also match for libm's
10085 If either operand is a NaN, returns the other non-NaN operand. Returns
10086 NaN only if both operands are NaN. If the operands compare equal,
10087 returns a value that compares equal to both operands. This means that
10088 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10090 '``llvm.maxnum.*``' Intrinsic
10091 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10096 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
10097 floating point or vector of floating point type. Not all targets support
10102 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
10103 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
10104 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10105 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
10106 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10111 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
10118 The arguments and return value are floating point numbers of the same
10123 Follows the IEEE-754 semantics for maxNum, which also match for libm's
10126 If either operand is a NaN, returns the other non-NaN operand. Returns
10127 NaN only if both operands are NaN. If the operands compare equal,
10128 returns a value that compares equal to both operands. This means that
10129 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10131 '``llvm.copysign.*``' Intrinsic
10132 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10137 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
10138 floating point or vector of floating point type. Not all targets support
10143 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
10144 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
10145 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
10146 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
10147 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
10152 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
10153 first operand and the sign of the second operand.
10158 The arguments and return value are floating point numbers of the same
10164 This function returns the same values as the libm ``copysign``
10165 functions would, and handles error conditions in the same way.
10167 '``llvm.floor.*``' Intrinsic
10168 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10173 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
10174 floating point or vector of floating point type. Not all targets support
10179 declare float @llvm.floor.f32(float %Val)
10180 declare double @llvm.floor.f64(double %Val)
10181 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
10182 declare fp128 @llvm.floor.f128(fp128 %Val)
10183 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
10188 The '``llvm.floor.*``' intrinsics return the floor of the operand.
10193 The argument and return value are floating point numbers of the same
10199 This function returns the same values as the libm ``floor`` functions
10200 would, and handles error conditions in the same way.
10202 '``llvm.ceil.*``' Intrinsic
10203 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10208 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
10209 floating point or vector of floating point type. Not all targets support
10214 declare float @llvm.ceil.f32(float %Val)
10215 declare double @llvm.ceil.f64(double %Val)
10216 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
10217 declare fp128 @llvm.ceil.f128(fp128 %Val)
10218 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
10223 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
10228 The argument and return value are floating point numbers of the same
10234 This function returns the same values as the libm ``ceil`` functions
10235 would, and handles error conditions in the same way.
10237 '``llvm.trunc.*``' Intrinsic
10238 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10243 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
10244 floating point or vector of floating point type. Not all targets support
10249 declare float @llvm.trunc.f32(float %Val)
10250 declare double @llvm.trunc.f64(double %Val)
10251 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
10252 declare fp128 @llvm.trunc.f128(fp128 %Val)
10253 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
10258 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
10259 nearest integer not larger in magnitude than the operand.
10264 The argument and return value are floating point numbers of the same
10270 This function returns the same values as the libm ``trunc`` functions
10271 would, and handles error conditions in the same way.
10273 '``llvm.rint.*``' Intrinsic
10274 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10279 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
10280 floating point or vector of floating point type. Not all targets support
10285 declare float @llvm.rint.f32(float %Val)
10286 declare double @llvm.rint.f64(double %Val)
10287 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
10288 declare fp128 @llvm.rint.f128(fp128 %Val)
10289 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
10294 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
10295 nearest integer. It may raise an inexact floating-point exception if the
10296 operand isn't an integer.
10301 The argument and return value are floating point numbers of the same
10307 This function returns the same values as the libm ``rint`` functions
10308 would, and handles error conditions in the same way.
10310 '``llvm.nearbyint.*``' Intrinsic
10311 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10316 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
10317 floating point or vector of floating point type. Not all targets support
10322 declare float @llvm.nearbyint.f32(float %Val)
10323 declare double @llvm.nearbyint.f64(double %Val)
10324 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
10325 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
10326 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
10331 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
10337 The argument and return value are floating point numbers of the same
10343 This function returns the same values as the libm ``nearbyint``
10344 functions would, and handles error conditions in the same way.
10346 '``llvm.round.*``' Intrinsic
10347 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10352 This is an overloaded intrinsic. You can use ``llvm.round`` on any
10353 floating point or vector of floating point type. Not all targets support
10358 declare float @llvm.round.f32(float %Val)
10359 declare double @llvm.round.f64(double %Val)
10360 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
10361 declare fp128 @llvm.round.f128(fp128 %Val)
10362 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
10367 The '``llvm.round.*``' intrinsics returns the operand rounded to the
10373 The argument and return value are floating point numbers of the same
10379 This function returns the same values as the libm ``round``
10380 functions would, and handles error conditions in the same way.
10382 Bit Manipulation Intrinsics
10383 ---------------------------
10385 LLVM provides intrinsics for a few important bit manipulation
10386 operations. These allow efficient code generation for some algorithms.
10388 '``llvm.bswap.*``' Intrinsics
10389 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10394 This is an overloaded intrinsic function. You can use bswap on any
10395 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
10399 declare i16 @llvm.bswap.i16(i16 <id>)
10400 declare i32 @llvm.bswap.i32(i32 <id>)
10401 declare i64 @llvm.bswap.i64(i64 <id>)
10406 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
10407 values with an even number of bytes (positive multiple of 16 bits).
10408 These are useful for performing operations on data that is not in the
10409 target's native byte order.
10414 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
10415 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
10416 intrinsic returns an i32 value that has the four bytes of the input i32
10417 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
10418 returned i32 will have its bytes in 3, 2, 1, 0 order. The
10419 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
10420 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
10423 '``llvm.ctpop.*``' Intrinsic
10424 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10429 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
10430 bit width, or on any vector with integer elements. Not all targets
10431 support all bit widths or vector types, however.
10435 declare i8 @llvm.ctpop.i8(i8 <src>)
10436 declare i16 @llvm.ctpop.i16(i16 <src>)
10437 declare i32 @llvm.ctpop.i32(i32 <src>)
10438 declare i64 @llvm.ctpop.i64(i64 <src>)
10439 declare i256 @llvm.ctpop.i256(i256 <src>)
10440 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
10445 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
10451 The only argument is the value to be counted. The argument may be of any
10452 integer type, or a vector with integer elements. The return type must
10453 match the argument type.
10458 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
10459 each element of a vector.
10461 '``llvm.ctlz.*``' Intrinsic
10462 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10467 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
10468 integer bit width, or any vector whose elements are integers. Not all
10469 targets support all bit widths or vector types, however.
10473 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
10474 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
10475 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
10476 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
10477 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
10478 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10483 The '``llvm.ctlz``' family of intrinsic functions counts the number of
10484 leading zeros in a variable.
10489 The first argument is the value to be counted. This argument may be of
10490 any integer type, or a vector with integer element type. The return
10491 type must match the first argument type.
10493 The second argument must be a constant and is a flag to indicate whether
10494 the intrinsic should ensure that a zero as the first argument produces a
10495 defined result. Historically some architectures did not provide a
10496 defined result for zero values as efficiently, and many algorithms are
10497 now predicated on avoiding zero-value inputs.
10502 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
10503 zeros in a variable, or within each element of the vector. If
10504 ``src == 0`` then the result is the size in bits of the type of ``src``
10505 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10506 ``llvm.ctlz(i32 2) = 30``.
10508 '``llvm.cttz.*``' Intrinsic
10509 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10514 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
10515 integer bit width, or any vector of integer elements. Not all targets
10516 support all bit widths or vector types, however.
10520 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
10521 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
10522 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
10523 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
10524 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
10525 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10530 The '``llvm.cttz``' family of intrinsic functions counts the number of
10536 The first argument is the value to be counted. This argument may be of
10537 any integer type, or a vector with integer element type. The return
10538 type must match the first argument type.
10540 The second argument must be a constant and is a flag to indicate whether
10541 the intrinsic should ensure that a zero as the first argument produces a
10542 defined result. Historically some architectures did not provide a
10543 defined result for zero values as efficiently, and many algorithms are
10544 now predicated on avoiding zero-value inputs.
10549 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
10550 zeros in a variable, or within each element of a vector. If ``src == 0``
10551 then the result is the size in bits of the type of ``src`` if
10552 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10553 ``llvm.cttz(2) = 1``.
10557 Arithmetic with Overflow Intrinsics
10558 -----------------------------------
10560 LLVM provides intrinsics for some arithmetic with overflow operations.
10562 '``llvm.sadd.with.overflow.*``' Intrinsics
10563 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10568 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
10569 on any integer bit width.
10573 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
10574 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10575 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
10580 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10581 a signed addition of the two arguments, and indicate whether an overflow
10582 occurred during the signed summation.
10587 The arguments (%a and %b) and the first element of the result structure
10588 may be of integer types of any bit width, but they must have the same
10589 bit width. The second element of the result structure must be of type
10590 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10596 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10597 a signed addition of the two variables. They return a structure --- the
10598 first element of which is the signed summation, and the second element
10599 of which is a bit specifying if the signed summation resulted in an
10605 .. code-block:: llvm
10607 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10608 %sum = extractvalue {i32, i1} %res, 0
10609 %obit = extractvalue {i32, i1} %res, 1
10610 br i1 %obit, label %overflow, label %normal
10612 '``llvm.uadd.with.overflow.*``' Intrinsics
10613 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10618 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
10619 on any integer bit width.
10623 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
10624 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10625 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
10630 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10631 an unsigned addition of the two arguments, and indicate whether a carry
10632 occurred during the unsigned summation.
10637 The arguments (%a and %b) and the first element of the result structure
10638 may be of integer types of any bit width, but they must have the same
10639 bit width. The second element of the result structure must be of type
10640 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10646 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10647 an unsigned addition of the two arguments. They return a structure --- the
10648 first element of which is the sum, and the second element of which is a
10649 bit specifying if the unsigned summation resulted in a carry.
10654 .. code-block:: llvm
10656 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10657 %sum = extractvalue {i32, i1} %res, 0
10658 %obit = extractvalue {i32, i1} %res, 1
10659 br i1 %obit, label %carry, label %normal
10661 '``llvm.ssub.with.overflow.*``' Intrinsics
10662 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10667 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
10668 on any integer bit width.
10672 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
10673 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10674 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
10679 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10680 a signed subtraction of the two arguments, and indicate whether an
10681 overflow occurred during the signed subtraction.
10686 The arguments (%a and %b) and the first element of the result structure
10687 may be of integer types of any bit width, but they must have the same
10688 bit width. The second element of the result structure must be of type
10689 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10695 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10696 a signed subtraction of the two arguments. They return a structure --- the
10697 first element of which is the subtraction, and the second element of
10698 which is a bit specifying if the signed subtraction resulted in an
10704 .. code-block:: llvm
10706 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10707 %sum = extractvalue {i32, i1} %res, 0
10708 %obit = extractvalue {i32, i1} %res, 1
10709 br i1 %obit, label %overflow, label %normal
10711 '``llvm.usub.with.overflow.*``' Intrinsics
10712 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10717 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
10718 on any integer bit width.
10722 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
10723 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10724 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
10729 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10730 an unsigned subtraction of the two arguments, and indicate whether an
10731 overflow occurred during the unsigned subtraction.
10736 The arguments (%a and %b) and the first element of the result structure
10737 may be of integer types of any bit width, but they must have the same
10738 bit width. The second element of the result structure must be of type
10739 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10745 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10746 an unsigned subtraction of the two arguments. They return a structure ---
10747 the first element of which is the subtraction, and the second element of
10748 which is a bit specifying if the unsigned subtraction resulted in an
10754 .. code-block:: llvm
10756 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10757 %sum = extractvalue {i32, i1} %res, 0
10758 %obit = extractvalue {i32, i1} %res, 1
10759 br i1 %obit, label %overflow, label %normal
10761 '``llvm.smul.with.overflow.*``' Intrinsics
10762 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10767 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
10768 on any integer bit width.
10772 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
10773 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10774 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
10779 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10780 a signed multiplication of the two arguments, and indicate whether an
10781 overflow occurred during the signed multiplication.
10786 The arguments (%a and %b) and the first element of the result structure
10787 may be of integer types of any bit width, but they must have the same
10788 bit width. The second element of the result structure must be of type
10789 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10795 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10796 a signed multiplication of the two arguments. They return a structure ---
10797 the first element of which is the multiplication, and the second element
10798 of which is a bit specifying if the signed multiplication resulted in an
10804 .. code-block:: llvm
10806 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10807 %sum = extractvalue {i32, i1} %res, 0
10808 %obit = extractvalue {i32, i1} %res, 1
10809 br i1 %obit, label %overflow, label %normal
10811 '``llvm.umul.with.overflow.*``' Intrinsics
10812 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10817 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
10818 on any integer bit width.
10822 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
10823 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10824 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
10829 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10830 a unsigned multiplication of the two arguments, and indicate whether an
10831 overflow occurred during the unsigned multiplication.
10836 The arguments (%a and %b) and the first element of the result structure
10837 may be of integer types of any bit width, but they must have the same
10838 bit width. The second element of the result structure must be of type
10839 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10845 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10846 an unsigned multiplication of the two arguments. They return a structure ---
10847 the first element of which is the multiplication, and the second
10848 element of which is a bit specifying if the unsigned multiplication
10849 resulted in an overflow.
10854 .. code-block:: llvm
10856 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10857 %sum = extractvalue {i32, i1} %res, 0
10858 %obit = extractvalue {i32, i1} %res, 1
10859 br i1 %obit, label %overflow, label %normal
10861 Specialised Arithmetic Intrinsics
10862 ---------------------------------
10864 '``llvm.canonicalize.*``' Intrinsic
10865 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10872 declare float @llvm.canonicalize.f32(float %a)
10873 declare double @llvm.canonicalize.f64(double %b)
10878 The '``llvm.canonicalize.*``' intrinsic returns the platform specific canonical
10879 encoding of a floating point number. This canonicalization is useful for
10880 implementing certain numeric primitives such as frexp. The canonical encoding is
10881 defined by IEEE-754-2008 to be:
10885 2.1.8 canonical encoding: The preferred encoding of a floating-point
10886 representation in a format. Applied to declets, significands of finite
10887 numbers, infinities, and NaNs, especially in decimal formats.
10889 This operation can also be considered equivalent to the IEEE-754-2008
10890 conversion of a floating-point value to the same format. NaNs are handled
10891 according to section 6.2.
10893 Examples of non-canonical encodings:
10895 - x87 pseudo denormals, pseudo NaNs, pseudo Infinity, Unnormals. These are
10896 converted to a canonical representation per hardware-specific protocol.
10897 - Many normal decimal floating point numbers have non-canonical alternative
10899 - Some machines, like GPUs or ARMv7 NEON, do not support subnormal values.
10900 These are treated as non-canonical encodings of zero and with be flushed to
10901 a zero of the same sign by this operation.
10903 Note that per IEEE-754-2008 6.2, systems that support signaling NaNs with
10904 default exception handling must signal an invalid exception, and produce a
10907 This function should always be implementable as multiplication by 1.0, provided
10908 that the compiler does not constant fold the operation. Likewise, division by
10909 1.0 and ``llvm.minnum(x, x)`` are possible implementations. Addition with
10910 -0.0 is also sufficient provided that the rounding mode is not -Infinity.
10912 ``@llvm.canonicalize`` must preserve the equality relation. That is:
10914 - ``(@llvm.canonicalize(x) == x)`` is equivalent to ``(x == x)``
10915 - ``(@llvm.canonicalize(x) == @llvm.canonicalize(y))`` is equivalent to
10918 Additionally, the sign of zero must be conserved:
10919 ``@llvm.canonicalize(-0.0) = -0.0`` and ``@llvm.canonicalize(+0.0) = +0.0``
10921 The payload bits of a NaN must be conserved, with two exceptions.
10922 First, environments which use only a single canonical representation of NaN
10923 must perform said canonicalization. Second, SNaNs must be quieted per the
10926 The canonicalization operation may be optimized away if:
10928 - The input is known to be canonical. For example, it was produced by a
10929 floating-point operation that is required by the standard to be canonical.
10930 - The result is consumed only by (or fused with) other floating-point
10931 operations. That is, the bits of the floating point value are not examined.
10933 '``llvm.fmuladd.*``' Intrinsic
10934 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10941 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
10942 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
10947 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
10948 expressions that can be fused if the code generator determines that (a) the
10949 target instruction set has support for a fused operation, and (b) that the
10950 fused operation is more efficient than the equivalent, separate pair of mul
10951 and add instructions.
10956 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
10957 multiplicands, a and b, and an addend c.
10966 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
10968 is equivalent to the expression a \* b + c, except that rounding will
10969 not be performed between the multiplication and addition steps if the
10970 code generator fuses the operations. Fusion is not guaranteed, even if
10971 the target platform supports it. If a fused multiply-add is required the
10972 corresponding llvm.fma.\* intrinsic function should be used
10973 instead. This never sets errno, just as '``llvm.fma.*``'.
10978 .. code-block:: llvm
10980 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
10983 '``llvm.uabsdiff.*``' and '``llvm.sabsdiff.*``' Intrinsics
10984 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10988 This is an overloaded intrinsic. The loaded data is a vector of any integer bit width.
10990 .. code-block:: llvm
10992 declare <4 x integer> @llvm.uabsdiff.v4i32(<4 x integer> %a, <4 x integer> %b)
10998 The ``llvm.uabsdiff`` intrinsic returns a vector result of the absolute difference
10999 of the two operands, treating them both as unsigned integers. The intermediate
11000 calculations are computed using infinitely precise unsigned arithmetic. The final
11001 result will be truncated to the given type.
11003 The ``llvm.sabsdiff`` intrinsic returns a vector result of the absolute difference of
11004 the two operands, treating them both as signed integers. If the result overflows, the
11005 behavior is undefined.
11009 These intrinsics are primarily used during the code generation stage of compilation.
11010 They are generated by compiler passes such as the Loop and SLP vectorizers. It is not
11011 recommended for users to create them manually.
11016 Both intrinsics take two integer of the same bitwidth.
11023 call <4 x i32> @llvm.uabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
11027 %1 = zext <4 x i32> %a to <4 x i64>
11028 %2 = zext <4 x i32> %b to <4 x i64>
11029 %sub = sub <4 x i64> %1, %2
11030 %trunc = trunc <4 x i64> to <4 x i32>
11032 and the expression::
11034 call <4 x i32> @llvm.sabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
11038 %sub = sub nsw <4 x i32> %a, %b
11039 %ispos = icmp sge <4 x i32> %sub, zeroinitializer
11040 %neg = sub nsw <4 x i32> zeroinitializer, %sub
11041 %1 = select <4 x i1> %ispos, <4 x i32> %sub, <4 x i32> %neg
11044 Half Precision Floating Point Intrinsics
11045 ----------------------------------------
11047 For most target platforms, half precision floating point is a
11048 storage-only format. This means that it is a dense encoding (in memory)
11049 but does not support computation in the format.
11051 This means that code must first load the half-precision floating point
11052 value as an i16, then convert it to float with
11053 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
11054 then be performed on the float value (including extending to double
11055 etc). To store the value back to memory, it is first converted to float
11056 if needed, then converted to i16 with
11057 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
11060 .. _int_convert_to_fp16:
11062 '``llvm.convert.to.fp16``' Intrinsic
11063 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11070 declare i16 @llvm.convert.to.fp16.f32(float %a)
11071 declare i16 @llvm.convert.to.fp16.f64(double %a)
11076 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11077 conventional floating point type to half precision floating point format.
11082 The intrinsic function contains single argument - the value to be
11088 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11089 conventional floating point format to half precision floating point format. The
11090 return value is an ``i16`` which contains the converted number.
11095 .. code-block:: llvm
11097 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
11098 store i16 %res, i16* @x, align 2
11100 .. _int_convert_from_fp16:
11102 '``llvm.convert.from.fp16``' Intrinsic
11103 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11110 declare float @llvm.convert.from.fp16.f32(i16 %a)
11111 declare double @llvm.convert.from.fp16.f64(i16 %a)
11116 The '``llvm.convert.from.fp16``' intrinsic function performs a
11117 conversion from half precision floating point format to single precision
11118 floating point format.
11123 The intrinsic function contains single argument - the value to be
11129 The '``llvm.convert.from.fp16``' intrinsic function performs a
11130 conversion from half single precision floating point format to single
11131 precision floating point format. The input half-float value is
11132 represented by an ``i16`` value.
11137 .. code-block:: llvm
11139 %a = load i16, i16* @x, align 2
11140 %res = call float @llvm.convert.from.fp16(i16 %a)
11142 .. _dbg_intrinsics:
11144 Debugger Intrinsics
11145 -------------------
11147 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
11148 prefix), are described in the `LLVM Source Level
11149 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
11152 Exception Handling Intrinsics
11153 -----------------------------
11155 The LLVM exception handling intrinsics (which all start with
11156 ``llvm.eh.`` prefix), are described in the `LLVM Exception
11157 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
11159 .. _int_trampoline:
11161 Trampoline Intrinsics
11162 ---------------------
11164 These intrinsics make it possible to excise one parameter, marked with
11165 the :ref:`nest <nest>` attribute, from a function. The result is a
11166 callable function pointer lacking the nest parameter - the caller does
11167 not need to provide a value for it. Instead, the value to use is stored
11168 in advance in a "trampoline", a block of memory usually allocated on the
11169 stack, which also contains code to splice the nest value into the
11170 argument list. This is used to implement the GCC nested function address
11173 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
11174 then the resulting function pointer has signature ``i32 (i32, i32)*``.
11175 It can be created as follows:
11177 .. code-block:: llvm
11179 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
11180 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
11181 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
11182 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
11183 %fp = bitcast i8* %p to i32 (i32, i32)*
11185 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
11186 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
11190 '``llvm.init.trampoline``' Intrinsic
11191 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11198 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
11203 This fills the memory pointed to by ``tramp`` with executable code,
11204 turning it into a trampoline.
11209 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
11210 pointers. The ``tramp`` argument must point to a sufficiently large and
11211 sufficiently aligned block of memory; this memory is written to by the
11212 intrinsic. Note that the size and the alignment are target-specific -
11213 LLVM currently provides no portable way of determining them, so a
11214 front-end that generates this intrinsic needs to have some
11215 target-specific knowledge. The ``func`` argument must hold a function
11216 bitcast to an ``i8*``.
11221 The block of memory pointed to by ``tramp`` is filled with target
11222 dependent code, turning it into a function. Then ``tramp`` needs to be
11223 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
11224 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
11225 function's signature is the same as that of ``func`` with any arguments
11226 marked with the ``nest`` attribute removed. At most one such ``nest``
11227 argument is allowed, and it must be of pointer type. Calling the new
11228 function is equivalent to calling ``func`` with the same argument list,
11229 but with ``nval`` used for the missing ``nest`` argument. If, after
11230 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
11231 modified, then the effect of any later call to the returned function
11232 pointer is undefined.
11236 '``llvm.adjust.trampoline``' Intrinsic
11237 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11244 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
11249 This performs any required machine-specific adjustment to the address of
11250 a trampoline (passed as ``tramp``).
11255 ``tramp`` must point to a block of memory which already has trampoline
11256 code filled in by a previous call to
11257 :ref:`llvm.init.trampoline <int_it>`.
11262 On some architectures the address of the code to be executed needs to be
11263 different than the address where the trampoline is actually stored. This
11264 intrinsic returns the executable address corresponding to ``tramp``
11265 after performing the required machine specific adjustments. The pointer
11266 returned can then be :ref:`bitcast and executed <int_trampoline>`.
11268 .. _int_mload_mstore:
11270 Masked Vector Load and Store Intrinsics
11271 ---------------------------------------
11273 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.
11277 '``llvm.masked.load.*``' Intrinsics
11278 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11282 This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
11286 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11287 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11292 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.
11298 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.
11304 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.
11305 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.
11310 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
11312 ;; The result of the two following instructions is identical aside from potential memory access exception
11313 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
11314 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
11318 '``llvm.masked.store.*``' Intrinsics
11319 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11323 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
11327 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
11328 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
11333 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.
11338 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.
11344 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.
11345 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.
11349 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
11351 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
11352 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
11353 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
11354 store <16 x float> %res, <16 x float>* %ptr, align 4
11357 Masked Vector Gather and Scatter Intrinsics
11358 -------------------------------------------
11360 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.
11364 '``llvm.masked.gather.*``' Intrinsics
11365 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11369 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.
11373 declare <16 x float> @llvm.masked.gather.v16f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11374 declare <2 x double> @llvm.masked.gather.v2f64 (<2 x double*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11379 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.
11385 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.
11391 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.
11392 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.
11397 %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>)
11399 ;; The gather with all-true mask is equivalent to the following instruction sequence
11400 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
11401 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
11402 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
11403 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
11405 %val0 = load double, double* %ptr0, align 8
11406 %val1 = load double, double* %ptr1, align 8
11407 %val2 = load double, double* %ptr2, align 8
11408 %val3 = load double, double* %ptr3, align 8
11410 %vec0 = insertelement <4 x double>undef, %val0, 0
11411 %vec01 = insertelement <4 x double>%vec0, %val1, 1
11412 %vec012 = insertelement <4 x double>%vec01, %val2, 2
11413 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
11417 '``llvm.masked.scatter.*``' Intrinsics
11418 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11422 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.
11426 declare void @llvm.masked.scatter.v8i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
11427 declare void @llvm.masked.scatter.v16f32(<16 x i32> <value>, <16 x i32*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
11432 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.
11437 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.
11443 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 divergence. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar store operations.
11447 ;; This instruction unconditionaly stores data vector in multiple addresses
11448 call @llvm.masked.scatter.v8i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
11450 ;; It is equivalent to a list of scalar stores
11451 %val0 = extractelement <8 x i32> %value, i32 0
11452 %val1 = extractelement <8 x i32> %value, i32 1
11454 %val7 = extractelement <8 x i32> %value, i32 7
11455 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
11456 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
11458 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
11459 ;; Note: the order of the following stores is important when they overlap:
11460 store i32 %val0, i32* %ptr0, align 4
11461 store i32 %val1, i32* %ptr1, align 4
11463 store i32 %val7, i32* %ptr7, align 4
11469 This class of intrinsics provides information about the lifetime of
11470 memory objects and ranges where variables are immutable.
11474 '``llvm.lifetime.start``' Intrinsic
11475 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11482 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
11487 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
11493 The first argument is a constant integer representing the size of the
11494 object, or -1 if it is variable sized. The second argument is a pointer
11500 This intrinsic indicates that before this point in the code, the value
11501 of the memory pointed to by ``ptr`` is dead. This means that it is known
11502 to never be used and has an undefined value. A load from the pointer
11503 that precedes this intrinsic can be replaced with ``'undef'``.
11507 '``llvm.lifetime.end``' Intrinsic
11508 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11515 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
11520 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
11526 The first argument is a constant integer representing the size of the
11527 object, or -1 if it is variable sized. The second argument is a pointer
11533 This intrinsic indicates that after this point in the code, the value of
11534 the memory pointed to by ``ptr`` is dead. This means that it is known to
11535 never be used and has an undefined value. Any stores into the memory
11536 object following this intrinsic may be removed as dead.
11538 '``llvm.invariant.start``' Intrinsic
11539 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11546 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
11551 The '``llvm.invariant.start``' intrinsic specifies that the contents of
11552 a memory object will not change.
11557 The first argument is a constant integer representing the size of the
11558 object, or -1 if it is variable sized. The second argument is a pointer
11564 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
11565 the return value, the referenced memory location is constant and
11568 '``llvm.invariant.end``' Intrinsic
11569 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11576 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
11581 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
11582 memory object are mutable.
11587 The first argument is the matching ``llvm.invariant.start`` intrinsic.
11588 The second argument is a constant integer representing the size of the
11589 object, or -1 if it is variable sized and the third argument is a
11590 pointer to the object.
11595 This intrinsic indicates that the memory is mutable again.
11597 '``llvm.invariant.group.barrier``' Intrinsic
11598 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11605 declare i8* @llvm.invariant.group.barrier(i8* <ptr>)
11610 The '``llvm.invariant.group.barrier``' intrinsic can be used when an invariant
11611 established by invariant.group metadata no longer holds, to obtain a new pointer
11612 value that does not carry the invariant information.
11618 The ``llvm.invariant.group.barrier`` takes only one argument, which is
11619 the pointer to the memory for which the ``invariant.group`` no longer holds.
11624 Returns another pointer that aliases its argument but which is considered different
11625 for the purposes of ``load``/``store`` ``invariant.group`` metadata.
11630 This class of intrinsics is designed to be generic and has no specific
11633 '``llvm.var.annotation``' Intrinsic
11634 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11641 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11646 The '``llvm.var.annotation``' intrinsic.
11651 The first argument is a pointer to a value, the second is a pointer to a
11652 global string, the third is a pointer to a global string which is the
11653 source file name, and the last argument is the line number.
11658 This intrinsic allows annotation of local variables with arbitrary
11659 strings. This can be useful for special purpose optimizations that want
11660 to look for these annotations. These have no other defined use; they are
11661 ignored by code generation and optimization.
11663 '``llvm.ptr.annotation.*``' Intrinsic
11664 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11669 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
11670 pointer to an integer of any width. *NOTE* you must specify an address space for
11671 the pointer. The identifier for the default address space is the integer
11676 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11677 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
11678 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
11679 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
11680 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
11685 The '``llvm.ptr.annotation``' intrinsic.
11690 The first argument is a pointer to an integer value of arbitrary bitwidth
11691 (result of some expression), the second is a pointer to a global string, the
11692 third is a pointer to a global string which is the source file name, and the
11693 last argument is the line number. It returns the value of the first argument.
11698 This intrinsic allows annotation of a pointer to an integer with arbitrary
11699 strings. This can be useful for special purpose optimizations that want to look
11700 for these annotations. These have no other defined use; they are ignored by code
11701 generation and optimization.
11703 '``llvm.annotation.*``' Intrinsic
11704 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11709 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
11710 any integer bit width.
11714 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
11715 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
11716 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
11717 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
11718 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
11723 The '``llvm.annotation``' intrinsic.
11728 The first argument is an integer value (result of some expression), the
11729 second is a pointer to a global string, the third is a pointer to a
11730 global string which is the source file name, and the last argument is
11731 the line number. It returns the value of the first argument.
11736 This intrinsic allows annotations to be put on arbitrary expressions
11737 with arbitrary strings. This can be useful for special purpose
11738 optimizations that want to look for these annotations. These have no
11739 other defined use; they are ignored by code generation and optimization.
11741 '``llvm.trap``' Intrinsic
11742 ^^^^^^^^^^^^^^^^^^^^^^^^^
11749 declare void @llvm.trap() noreturn nounwind
11754 The '``llvm.trap``' intrinsic.
11764 This intrinsic is lowered to the target dependent trap instruction. If
11765 the target does not have a trap instruction, this intrinsic will be
11766 lowered to a call of the ``abort()`` function.
11768 '``llvm.debugtrap``' Intrinsic
11769 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11776 declare void @llvm.debugtrap() nounwind
11781 The '``llvm.debugtrap``' intrinsic.
11791 This intrinsic is lowered to code which is intended to cause an
11792 execution trap with the intention of requesting the attention of a
11795 '``llvm.stackprotector``' Intrinsic
11796 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11803 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
11808 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
11809 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
11810 is placed on the stack before local variables.
11815 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
11816 The first argument is the value loaded from the stack guard
11817 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
11818 enough space to hold the value of the guard.
11823 This intrinsic causes the prologue/epilogue inserter to force the position of
11824 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
11825 to ensure that if a local variable on the stack is overwritten, it will destroy
11826 the value of the guard. When the function exits, the guard on the stack is
11827 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
11828 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
11829 calling the ``__stack_chk_fail()`` function.
11831 '``llvm.stackprotectorcheck``' Intrinsic
11832 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11839 declare void @llvm.stackprotectorcheck(i8** <guard>)
11844 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
11845 created stack protector and if they are not equal calls the
11846 ``__stack_chk_fail()`` function.
11851 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
11852 the variable ``@__stack_chk_guard``.
11857 This intrinsic is provided to perform the stack protector check by comparing
11858 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
11859 values do not match call the ``__stack_chk_fail()`` function.
11861 The reason to provide this as an IR level intrinsic instead of implementing it
11862 via other IR operations is that in order to perform this operation at the IR
11863 level without an intrinsic, one would need to create additional basic blocks to
11864 handle the success/failure cases. This makes it difficult to stop the stack
11865 protector check from disrupting sibling tail calls in Codegen. With this
11866 intrinsic, we are able to generate the stack protector basic blocks late in
11867 codegen after the tail call decision has occurred.
11869 '``llvm.objectsize``' Intrinsic
11870 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11877 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
11878 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
11883 The ``llvm.objectsize`` intrinsic is designed to provide information to
11884 the optimizers to determine at compile time whether a) an operation
11885 (like memcpy) will overflow a buffer that corresponds to an object, or
11886 b) that a runtime check for overflow isn't necessary. An object in this
11887 context means an allocation of a specific class, structure, array, or
11893 The ``llvm.objectsize`` intrinsic takes two arguments. The first
11894 argument is a pointer to or into the ``object``. The second argument is
11895 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
11896 or -1 (if false) when the object size is unknown. The second argument
11897 only accepts constants.
11902 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
11903 the size of the object concerned. If the size cannot be determined at
11904 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
11905 on the ``min`` argument).
11907 '``llvm.expect``' Intrinsic
11908 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
11913 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
11918 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
11919 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
11920 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
11925 The ``llvm.expect`` intrinsic provides information about expected (the
11926 most probable) value of ``val``, which can be used by optimizers.
11931 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
11932 a value. The second argument is an expected value, this needs to be a
11933 constant value, variables are not allowed.
11938 This intrinsic is lowered to the ``val``.
11942 '``llvm.assume``' Intrinsic
11943 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11950 declare void @llvm.assume(i1 %cond)
11955 The ``llvm.assume`` allows the optimizer to assume that the provided
11956 condition is true. This information can then be used in simplifying other parts
11962 The condition which the optimizer may assume is always true.
11967 The intrinsic allows the optimizer to assume that the provided condition is
11968 always true whenever the control flow reaches the intrinsic call. No code is
11969 generated for this intrinsic, and instructions that contribute only to the
11970 provided condition are not used for code generation. If the condition is
11971 violated during execution, the behavior is undefined.
11973 Note that the optimizer might limit the transformations performed on values
11974 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
11975 only used to form the intrinsic's input argument. This might prove undesirable
11976 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
11977 sufficient overall improvement in code quality. For this reason,
11978 ``llvm.assume`` should not be used to document basic mathematical invariants
11979 that the optimizer can otherwise deduce or facts that are of little use to the
11984 '``llvm.bitset.test``' Intrinsic
11985 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11992 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
11998 The first argument is a pointer to be tested. The second argument is a
11999 metadata object representing an identifier for a :doc:`bitset <BitSets>`.
12004 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
12005 member of the given bitset.
12007 '``llvm.donothing``' Intrinsic
12008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12015 declare void @llvm.donothing() nounwind readnone
12020 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
12021 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
12022 with an invoke instruction.
12032 This intrinsic does nothing, and it's removed by optimizers and ignored
12035 Stack Map Intrinsics
12036 --------------------
12038 LLVM provides experimental intrinsics to support runtime patching
12039 mechanisms commonly desired in dynamic language JITs. These intrinsics
12040 are described in :doc:`StackMaps`.