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 does not call itself
1282 either directly or indirectly down any possible call path. This produces
1283 undefined behavior at runtime if the function ever does recurse.
1285 This function attribute indicates that the function never raises an
1286 exception. If the function does raise an exception, its runtime
1287 behavior is undefined. However, functions marked nounwind may still
1288 trap or generate asynchronous exceptions. Exception handling schemes
1289 that are recognized by LLVM to handle asynchronous exceptions, such
1290 as SEH, will still provide their implementation defined semantics.
1292 This function attribute indicates that the function is not optimized
1293 by any optimization or code generator passes with the
1294 exception of interprocedural optimization passes.
1295 This attribute cannot be used together with the ``alwaysinline``
1296 attribute; this attribute is also incompatible
1297 with the ``minsize`` attribute and the ``optsize`` attribute.
1299 This attribute requires the ``noinline`` attribute to be specified on
1300 the function as well, so the function is never inlined into any caller.
1301 Only functions with the ``alwaysinline`` attribute are valid
1302 candidates for inlining into the body of this function.
1304 This attribute suggests that optimization passes and code generator
1305 passes make choices that keep the code size of this function low,
1306 and otherwise do optimizations specifically to reduce code size as
1307 long as they do not significantly impact runtime performance.
1309 On a function, this attribute indicates that the function computes its
1310 result (or decides to unwind an exception) based strictly on its arguments,
1311 without dereferencing any pointer arguments or otherwise accessing
1312 any mutable state (e.g. memory, control registers, etc) visible to
1313 caller functions. It does not write through any pointer arguments
1314 (including ``byval`` arguments) and never changes any state visible
1315 to callers. This means that it cannot unwind exceptions by calling
1316 the ``C++`` exception throwing methods.
1318 On an argument, this attribute indicates that the function does not
1319 dereference that pointer argument, even though it may read or write the
1320 memory that the pointer points to if accessed through other pointers.
1322 On a function, this attribute indicates that the function does not write
1323 through any pointer arguments (including ``byval`` arguments) or otherwise
1324 modify any state (e.g. memory, control registers, etc) visible to
1325 caller functions. It may dereference pointer arguments and read
1326 state that may be set in the caller. A readonly function always
1327 returns the same value (or unwinds an exception identically) when
1328 called with the same set of arguments and global state. It cannot
1329 unwind an exception by calling the ``C++`` exception throwing
1332 On an argument, this attribute indicates that the function does not write
1333 through this pointer argument, even though it may write to the memory that
1334 the pointer points to.
1336 This attribute indicates that the only memory accesses inside function are
1337 loads and stores from objects pointed to by its pointer-typed arguments,
1338 with arbitrary offsets. Or in other words, all memory operations in the
1339 function can refer to memory only using pointers based on its function
1341 Note that ``argmemonly`` can be used together with ``readonly`` attribute
1342 in order to specify that function reads only from its arguments.
1344 This attribute indicates that this function can return twice. The C
1345 ``setjmp`` is an example of such a function. The compiler disables
1346 some optimizations (like tail calls) in the caller of these
1349 This attribute indicates that
1350 `SafeStack <http://clang.llvm.org/docs/SafeStack.html>`_
1351 protection is enabled for this function.
1353 If a function that has a ``safestack`` attribute is inlined into a
1354 function that doesn't have a ``safestack`` attribute or which has an
1355 ``ssp``, ``sspstrong`` or ``sspreq`` attribute, then the resulting
1356 function will have a ``safestack`` attribute.
1357 ``sanitize_address``
1358 This attribute indicates that AddressSanitizer checks
1359 (dynamic address safety analysis) are enabled for this function.
1361 This attribute indicates that MemorySanitizer checks (dynamic detection
1362 of accesses to uninitialized memory) are enabled for this function.
1364 This attribute indicates that ThreadSanitizer checks
1365 (dynamic thread safety analysis) are enabled for this function.
1367 This attribute indicates that the function should emit a stack
1368 smashing protector. It is in the form of a "canary" --- a random value
1369 placed on the stack before the local variables that's checked upon
1370 return from the function to see if it has been overwritten. A
1371 heuristic is used to determine if a function needs stack protectors
1372 or not. The heuristic used will enable protectors for functions with:
1374 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1375 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1376 - Calls to alloca() with variable sizes or constant sizes greater than
1377 ``ssp-buffer-size``.
1379 Variables that are identified as requiring a protector will be arranged
1380 on the stack such that they are adjacent to the stack protector guard.
1382 If a function that has an ``ssp`` attribute is inlined into a
1383 function that doesn't have an ``ssp`` attribute, then the resulting
1384 function will have an ``ssp`` attribute.
1386 This attribute indicates that the function should *always* emit a
1387 stack smashing protector. This overrides the ``ssp`` function
1390 Variables that are identified as requiring a protector will be arranged
1391 on the stack such that they are adjacent to the stack protector guard.
1392 The specific layout rules are:
1394 #. Large arrays and structures containing large arrays
1395 (``>= ssp-buffer-size``) are closest to the stack protector.
1396 #. Small arrays and structures containing small arrays
1397 (``< ssp-buffer-size``) are 2nd closest to the protector.
1398 #. Variables that have had their address taken are 3rd closest to the
1401 If a function that has an ``sspreq`` attribute is inlined into a
1402 function that doesn't have an ``sspreq`` attribute or which has an
1403 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1404 an ``sspreq`` attribute.
1406 This attribute indicates that the function should emit a stack smashing
1407 protector. This attribute causes a strong heuristic to be used when
1408 determining if a function needs stack protectors. The strong heuristic
1409 will enable protectors for functions with:
1411 - Arrays of any size and type
1412 - Aggregates containing an array of any size and type.
1413 - Calls to alloca().
1414 - Local variables that have had their address taken.
1416 Variables that are identified as requiring a protector will be arranged
1417 on the stack such that they are adjacent to the stack protector guard.
1418 The specific layout rules are:
1420 #. Large arrays and structures containing large arrays
1421 (``>= ssp-buffer-size``) are closest to the stack protector.
1422 #. Small arrays and structures containing small arrays
1423 (``< ssp-buffer-size``) are 2nd closest to the protector.
1424 #. Variables that have had their address taken are 3rd closest to the
1427 This overrides the ``ssp`` function attribute.
1429 If a function that has an ``sspstrong`` attribute is inlined into a
1430 function that doesn't have an ``sspstrong`` attribute, then the
1431 resulting function will have an ``sspstrong`` attribute.
1433 This attribute indicates that the function will delegate to some other
1434 function with a tail call. The prototype of a thunk should not be used for
1435 optimization purposes. The caller is expected to cast the thunk prototype to
1436 match the thunk target prototype.
1438 This attribute indicates that the ABI being targeted requires that
1439 an unwind table entry be produced for this function even if we can
1440 show that no exceptions passes by it. This is normally the case for
1441 the ELF x86-64 abi, but it can be disabled for some compilation
1450 Note: operand bundles are a work in progress, and they should be
1451 considered experimental at this time.
1453 Operand bundles are tagged sets of SSA values that can be associated
1454 with certain LLVM instructions (currently only ``call`` s and
1455 ``invoke`` s). In a way they are like metadata, but dropping them is
1456 incorrect and will change program semantics.
1460 operand bundle set ::= '[' operand bundle ']'
1461 operand bundle ::= tag '(' [ bundle operand ] (, bundle operand )* ')'
1462 bundle operand ::= SSA value
1463 tag ::= string constant
1465 Operand bundles are **not** part of a function's signature, and a
1466 given function may be called from multiple places with different kinds
1467 of operand bundles. This reflects the fact that the operand bundles
1468 are conceptually a part of the ``call`` (or ``invoke``), not the
1469 callee being dispatched to.
1471 Operand bundles are a generic mechanism intended to support
1472 runtime-introspection-like functionality for managed languages. While
1473 the exact semantics of an operand bundle depend on the bundle tag,
1474 there are certain limitations to how much the presence of an operand
1475 bundle can influence the semantics of a program. These restrictions
1476 are described as the semantics of an "unknown" operand bundle. As
1477 long as the behavior of an operand bundle is describable within these
1478 restrictions, LLVM does not need to have special knowledge of the
1479 operand bundle to not miscompile programs containing it.
1481 - The bundle operands for an unknown operand bundle escape in unknown
1482 ways before control is transferred to the callee or invokee.
1483 - Calls and invokes with operand bundles have unknown read / write
1484 effect on the heap on entry and exit (even if the call target is
1485 ``readnone`` or ``readonly``), unless they're overriden with
1486 callsite specific attributes.
1487 - An operand bundle at a call site cannot change the implementation
1488 of the called function. Inter-procedural optimizations work as
1489 usual as long as they take into account the first two properties.
1493 Module-Level Inline Assembly
1494 ----------------------------
1496 Modules may contain "module-level inline asm" blocks, which corresponds
1497 to the GCC "file scope inline asm" blocks. These blocks are internally
1498 concatenated by LLVM and treated as a single unit, but may be separated
1499 in the ``.ll`` file if desired. The syntax is very simple:
1501 .. code-block:: llvm
1503 module asm "inline asm code goes here"
1504 module asm "more can go here"
1506 The strings can contain any character by escaping non-printable
1507 characters. The escape sequence used is simply "\\xx" where "xx" is the
1508 two digit hex code for the number.
1510 Note that the assembly string *must* be parseable by LLVM's integrated assembler
1511 (unless it is disabled), even when emitting a ``.s`` file.
1513 .. _langref_datalayout:
1518 A module may specify a target specific data layout string that specifies
1519 how data is to be laid out in memory. The syntax for the data layout is
1522 .. code-block:: llvm
1524 target datalayout = "layout specification"
1526 The *layout specification* consists of a list of specifications
1527 separated by the minus sign character ('-'). Each specification starts
1528 with a letter and may include other information after the letter to
1529 define some aspect of the data layout. The specifications accepted are
1533 Specifies that the target lays out data in big-endian form. That is,
1534 the bits with the most significance have the lowest address
1537 Specifies that the target lays out data in little-endian form. That
1538 is, the bits with the least significance have the lowest address
1541 Specifies the natural alignment of the stack in bits. Alignment
1542 promotion of stack variables is limited to the natural stack
1543 alignment to avoid dynamic stack realignment. The stack alignment
1544 must be a multiple of 8-bits. If omitted, the natural stack
1545 alignment defaults to "unspecified", which does not prevent any
1546 alignment promotions.
1547 ``p[n]:<size>:<abi>:<pref>``
1548 This specifies the *size* of a pointer and its ``<abi>`` and
1549 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1550 bits. The address space, ``n``, is optional, and if not specified,
1551 denotes the default address space 0. The value of ``n`` must be
1552 in the range [1,2^23).
1553 ``i<size>:<abi>:<pref>``
1554 This specifies the alignment for an integer type of a given bit
1555 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1556 ``v<size>:<abi>:<pref>``
1557 This specifies the alignment for a vector type of a given bit
1559 ``f<size>:<abi>:<pref>``
1560 This specifies the alignment for a floating point type of a given bit
1561 ``<size>``. Only values of ``<size>`` that are supported by the target
1562 will work. 32 (float) and 64 (double) are supported on all targets; 80
1563 or 128 (different flavors of long double) are also supported on some
1566 This specifies the alignment for an object of aggregate type.
1568 If present, specifies that llvm names are mangled in the output. The
1571 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1572 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1573 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1574 symbols get a ``_`` prefix.
1575 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1576 functions also get a suffix based on the frame size.
1577 * ``x``: Windows x86 COFF prefix: Similar to Windows COFF, but use a ``_``
1578 prefix for ``__cdecl`` functions.
1579 ``n<size1>:<size2>:<size3>...``
1580 This specifies a set of native integer widths for the target CPU in
1581 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1582 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1583 this set are considered to support most general arithmetic operations
1586 On every specification that takes a ``<abi>:<pref>``, specifying the
1587 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1588 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1590 When constructing the data layout for a given target, LLVM starts with a
1591 default set of specifications which are then (possibly) overridden by
1592 the specifications in the ``datalayout`` keyword. The default
1593 specifications are given in this list:
1595 - ``E`` - big endian
1596 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1597 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1598 same as the default address space.
1599 - ``S0`` - natural stack alignment is unspecified
1600 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1601 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1602 - ``i16:16:16`` - i16 is 16-bit aligned
1603 - ``i32:32:32`` - i32 is 32-bit aligned
1604 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1605 alignment of 64-bits
1606 - ``f16:16:16`` - half is 16-bit aligned
1607 - ``f32:32:32`` - float is 32-bit aligned
1608 - ``f64:64:64`` - double is 64-bit aligned
1609 - ``f128:128:128`` - quad is 128-bit aligned
1610 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1611 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1612 - ``a:0:64`` - aggregates are 64-bit aligned
1614 When LLVM is determining the alignment for a given type, it uses the
1617 #. If the type sought is an exact match for one of the specifications,
1618 that specification is used.
1619 #. If no match is found, and the type sought is an integer type, then
1620 the smallest integer type that is larger than the bitwidth of the
1621 sought type is used. If none of the specifications are larger than
1622 the bitwidth then the largest integer type is used. For example,
1623 given the default specifications above, the i7 type will use the
1624 alignment of i8 (next largest) while both i65 and i256 will use the
1625 alignment of i64 (largest specified).
1626 #. If no match is found, and the type sought is a vector type, then the
1627 largest vector type that is smaller than the sought vector type will
1628 be used as a fall back. This happens because <128 x double> can be
1629 implemented in terms of 64 <2 x double>, for example.
1631 The function of the data layout string may not be what you expect.
1632 Notably, this is not a specification from the frontend of what alignment
1633 the code generator should use.
1635 Instead, if specified, the target data layout is required to match what
1636 the ultimate *code generator* expects. This string is used by the
1637 mid-level optimizers to improve code, and this only works if it matches
1638 what the ultimate code generator uses. There is no way to generate IR
1639 that does not embed this target-specific detail into the IR. If you
1640 don't specify the string, the default specifications will be used to
1641 generate a Data Layout and the optimization phases will operate
1642 accordingly and introduce target specificity into the IR with respect to
1643 these default specifications.
1650 A module may specify a target triple string that describes the target
1651 host. The syntax for the target triple is simply:
1653 .. code-block:: llvm
1655 target triple = "x86_64-apple-macosx10.7.0"
1657 The *target triple* string consists of a series of identifiers delimited
1658 by the minus sign character ('-'). The canonical forms are:
1662 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1663 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1665 This information is passed along to the backend so that it generates
1666 code for the proper architecture. It's possible to override this on the
1667 command line with the ``-mtriple`` command line option.
1669 .. _pointeraliasing:
1671 Pointer Aliasing Rules
1672 ----------------------
1674 Any memory access must be done through a pointer value associated with
1675 an address range of the memory access, otherwise the behavior is
1676 undefined. Pointer values are associated with address ranges according
1677 to the following rules:
1679 - A pointer value is associated with the addresses associated with any
1680 value it is *based* on.
1681 - An address of a global variable is associated with the address range
1682 of the variable's storage.
1683 - The result value of an allocation instruction is associated with the
1684 address range of the allocated storage.
1685 - A null pointer in the default address-space is associated with no
1687 - An integer constant other than zero or a pointer value returned from
1688 a function not defined within LLVM may be associated with address
1689 ranges allocated through mechanisms other than those provided by
1690 LLVM. Such ranges shall not overlap with any ranges of addresses
1691 allocated by mechanisms provided by LLVM.
1693 A pointer value is *based* on another pointer value according to the
1696 - A pointer value formed from a ``getelementptr`` operation is *based*
1697 on the first value operand of the ``getelementptr``.
1698 - The result value of a ``bitcast`` is *based* on the operand of the
1700 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1701 values that contribute (directly or indirectly) to the computation of
1702 the pointer's value.
1703 - The "*based* on" relationship is transitive.
1705 Note that this definition of *"based"* is intentionally similar to the
1706 definition of *"based"* in C99, though it is slightly weaker.
1708 LLVM IR does not associate types with memory. The result type of a
1709 ``load`` merely indicates the size and alignment of the memory from
1710 which to load, as well as the interpretation of the value. The first
1711 operand type of a ``store`` similarly only indicates the size and
1712 alignment of the store.
1714 Consequently, type-based alias analysis, aka TBAA, aka
1715 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1716 :ref:`Metadata <metadata>` may be used to encode additional information
1717 which specialized optimization passes may use to implement type-based
1722 Volatile Memory Accesses
1723 ------------------------
1725 Certain memory accesses, such as :ref:`load <i_load>`'s,
1726 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1727 marked ``volatile``. The optimizers must not change the number of
1728 volatile operations or change their order of execution relative to other
1729 volatile operations. The optimizers *may* change the order of volatile
1730 operations relative to non-volatile operations. This is not Java's
1731 "volatile" and has no cross-thread synchronization behavior.
1733 IR-level volatile loads and stores cannot safely be optimized into
1734 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1735 flagged volatile. Likewise, the backend should never split or merge
1736 target-legal volatile load/store instructions.
1738 .. admonition:: Rationale
1740 Platforms may rely on volatile loads and stores of natively supported
1741 data width to be executed as single instruction. For example, in C
1742 this holds for an l-value of volatile primitive type with native
1743 hardware support, but not necessarily for aggregate types. The
1744 frontend upholds these expectations, which are intentionally
1745 unspecified in the IR. The rules above ensure that IR transformations
1746 do not violate the frontend's contract with the language.
1750 Memory Model for Concurrent Operations
1751 --------------------------------------
1753 The LLVM IR does not define any way to start parallel threads of
1754 execution or to register signal handlers. Nonetheless, there are
1755 platform-specific ways to create them, and we define LLVM IR's behavior
1756 in their presence. This model is inspired by the C++0x memory model.
1758 For a more informal introduction to this model, see the :doc:`Atomics`.
1760 We define a *happens-before* partial order as the least partial order
1763 - Is a superset of single-thread program order, and
1764 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1765 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1766 techniques, like pthread locks, thread creation, thread joining,
1767 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1768 Constraints <ordering>`).
1770 Note that program order does not introduce *happens-before* edges
1771 between a thread and signals executing inside that thread.
1773 Every (defined) read operation (load instructions, memcpy, atomic
1774 loads/read-modify-writes, etc.) R reads a series of bytes written by
1775 (defined) write operations (store instructions, atomic
1776 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1777 section, initialized globals are considered to have a write of the
1778 initializer which is atomic and happens before any other read or write
1779 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1780 may see any write to the same byte, except:
1782 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1783 write\ :sub:`2` happens before R\ :sub:`byte`, then
1784 R\ :sub:`byte` does not see write\ :sub:`1`.
1785 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1786 R\ :sub:`byte` does not see write\ :sub:`3`.
1788 Given that definition, R\ :sub:`byte` is defined as follows:
1790 - If R is volatile, the result is target-dependent. (Volatile is
1791 supposed to give guarantees which can support ``sig_atomic_t`` in
1792 C/C++, and may be used for accesses to addresses that do not behave
1793 like normal memory. It does not generally provide cross-thread
1795 - Otherwise, if there is no write to the same byte that happens before
1796 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1797 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1798 R\ :sub:`byte` returns the value written by that write.
1799 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1800 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1801 Memory Ordering Constraints <ordering>` section for additional
1802 constraints on how the choice is made.
1803 - Otherwise R\ :sub:`byte` returns ``undef``.
1805 R returns the value composed of the series of bytes it read. This
1806 implies that some bytes within the value may be ``undef`` **without**
1807 the entire value being ``undef``. Note that this only defines the
1808 semantics of the operation; it doesn't mean that targets will emit more
1809 than one instruction to read the series of bytes.
1811 Note that in cases where none of the atomic intrinsics are used, this
1812 model places only one restriction on IR transformations on top of what
1813 is required for single-threaded execution: introducing a store to a byte
1814 which might not otherwise be stored is not allowed in general.
1815 (Specifically, in the case where another thread might write to and read
1816 from an address, introducing a store can change a load that may see
1817 exactly one write into a load that may see multiple writes.)
1821 Atomic Memory Ordering Constraints
1822 ----------------------------------
1824 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1825 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1826 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1827 ordering parameters that determine which other atomic instructions on
1828 the same address they *synchronize with*. These semantics are borrowed
1829 from Java and C++0x, but are somewhat more colloquial. If these
1830 descriptions aren't precise enough, check those specs (see spec
1831 references in the :doc:`atomics guide <Atomics>`).
1832 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1833 differently since they don't take an address. See that instruction's
1834 documentation for details.
1836 For a simpler introduction to the ordering constraints, see the
1840 The set of values that can be read is governed by the happens-before
1841 partial order. A value cannot be read unless some operation wrote
1842 it. This is intended to provide a guarantee strong enough to model
1843 Java's non-volatile shared variables. This ordering cannot be
1844 specified for read-modify-write operations; it is not strong enough
1845 to make them atomic in any interesting way.
1847 In addition to the guarantees of ``unordered``, there is a single
1848 total order for modifications by ``monotonic`` operations on each
1849 address. All modification orders must be compatible with the
1850 happens-before order. There is no guarantee that the modification
1851 orders can be combined to a global total order for the whole program
1852 (and this often will not be possible). The read in an atomic
1853 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1854 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1855 order immediately before the value it writes. If one atomic read
1856 happens before another atomic read of the same address, the later
1857 read must see the same value or a later value in the address's
1858 modification order. This disallows reordering of ``monotonic`` (or
1859 stronger) operations on the same address. If an address is written
1860 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1861 read that address repeatedly, the other threads must eventually see
1862 the write. This corresponds to the C++0x/C1x
1863 ``memory_order_relaxed``.
1865 In addition to the guarantees of ``monotonic``, a
1866 *synchronizes-with* edge may be formed with a ``release`` operation.
1867 This is intended to model C++'s ``memory_order_acquire``.
1869 In addition to the guarantees of ``monotonic``, if this operation
1870 writes a value which is subsequently read by an ``acquire``
1871 operation, it *synchronizes-with* that operation. (This isn't a
1872 complete description; see the C++0x definition of a release
1873 sequence.) This corresponds to the C++0x/C1x
1874 ``memory_order_release``.
1875 ``acq_rel`` (acquire+release)
1876 Acts as both an ``acquire`` and ``release`` operation on its
1877 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1878 ``seq_cst`` (sequentially consistent)
1879 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1880 operation that only reads, ``release`` for an operation that only
1881 writes), there is a global total order on all
1882 sequentially-consistent operations on all addresses, which is
1883 consistent with the *happens-before* partial order and with the
1884 modification orders of all the affected addresses. Each
1885 sequentially-consistent read sees the last preceding write to the
1886 same address in this global order. This corresponds to the C++0x/C1x
1887 ``memory_order_seq_cst`` and Java volatile.
1891 If an atomic operation is marked ``singlethread``, it only *synchronizes
1892 with* or participates in modification and seq\_cst total orderings with
1893 other operations running in the same thread (for example, in signal
1901 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1902 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1903 :ref:`frem <i_frem>`, :ref:`fcmp <i_fcmp>`) have the following flags that can
1904 be set to enable otherwise unsafe floating point operations
1907 No NaNs - Allow optimizations to assume the arguments and result are not
1908 NaN. Such optimizations are required to retain defined behavior over
1909 NaNs, but the value of the result is undefined.
1912 No Infs - Allow optimizations to assume the arguments and result are not
1913 +/-Inf. Such optimizations are required to retain defined behavior over
1914 +/-Inf, but the value of the result is undefined.
1917 No Signed Zeros - Allow optimizations to treat the sign of a zero
1918 argument or result as insignificant.
1921 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1922 argument rather than perform division.
1925 Fast - Allow algebraically equivalent transformations that may
1926 dramatically change results in floating point (e.g. reassociate). This
1927 flag implies all the others.
1931 Use-list Order Directives
1932 -------------------------
1934 Use-list directives encode the in-memory order of each use-list, allowing the
1935 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1936 indexes that are assigned to the referenced value's uses. The referenced
1937 value's use-list is immediately sorted by these indexes.
1939 Use-list directives may appear at function scope or global scope. They are not
1940 instructions, and have no effect on the semantics of the IR. When they're at
1941 function scope, they must appear after the terminator of the final basic block.
1943 If basic blocks have their address taken via ``blockaddress()`` expressions,
1944 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1951 uselistorder <ty> <value>, { <order-indexes> }
1952 uselistorder_bb @function, %block { <order-indexes> }
1958 define void @foo(i32 %arg1, i32 %arg2) {
1960 ; ... instructions ...
1962 ; ... instructions ...
1964 ; At function scope.
1965 uselistorder i32 %arg1, { 1, 0, 2 }
1966 uselistorder label %bb, { 1, 0 }
1970 uselistorder i32* @global, { 1, 2, 0 }
1971 uselistorder i32 7, { 1, 0 }
1972 uselistorder i32 (i32) @bar, { 1, 0 }
1973 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1980 The LLVM type system is one of the most important features of the
1981 intermediate representation. Being typed enables a number of
1982 optimizations to be performed on the intermediate representation
1983 directly, without having to do extra analyses on the side before the
1984 transformation. A strong type system makes it easier to read the
1985 generated code and enables novel analyses and transformations that are
1986 not feasible to perform on normal three address code representations.
1996 The void type does not represent any value and has no size.
2014 The function type can be thought of as a function signature. It consists of a
2015 return type and a list of formal parameter types. The return type of a function
2016 type is a void type or first class type --- except for :ref:`label <t_label>`
2017 and :ref:`metadata <t_metadata>` types.
2023 <returntype> (<parameter list>)
2025 ...where '``<parameter list>``' is a comma-separated list of type
2026 specifiers. Optionally, the parameter list may include a type ``...``, which
2027 indicates that the function takes a variable number of arguments. Variable
2028 argument functions can access their arguments with the :ref:`variable argument
2029 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
2030 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
2034 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2035 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
2036 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2037 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
2038 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2039 | ``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. |
2040 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2041 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
2042 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2049 The :ref:`first class <t_firstclass>` types are perhaps the most important.
2050 Values of these types are the only ones which can be produced by
2058 These are the types that are valid in registers from CodeGen's perspective.
2067 The integer type is a very simple type that simply specifies an
2068 arbitrary bit width for the integer type desired. Any bit width from 1
2069 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
2077 The number of bits the integer will occupy is specified by the ``N``
2083 +----------------+------------------------------------------------+
2084 | ``i1`` | a single-bit integer. |
2085 +----------------+------------------------------------------------+
2086 | ``i32`` | a 32-bit integer. |
2087 +----------------+------------------------------------------------+
2088 | ``i1942652`` | a really big integer of over 1 million bits. |
2089 +----------------+------------------------------------------------+
2093 Floating Point Types
2094 """"""""""""""""""""
2103 - 16-bit floating point value
2106 - 32-bit floating point value
2109 - 64-bit floating point value
2112 - 128-bit floating point value (112-bit mantissa)
2115 - 80-bit floating point value (X87)
2118 - 128-bit floating point value (two 64-bits)
2125 The x86_mmx type represents a value held in an MMX register on an x86
2126 machine. The operations allowed on it are quite limited: parameters and
2127 return values, load and store, and bitcast. User-specified MMX
2128 instructions are represented as intrinsic or asm calls with arguments
2129 and/or results of this type. There are no arrays, vectors or constants
2146 The pointer type is used to specify memory locations. Pointers are
2147 commonly used to reference objects in memory.
2149 Pointer types may have an optional address space attribute defining the
2150 numbered address space where the pointed-to object resides. The default
2151 address space is number zero. The semantics of non-zero address spaces
2152 are target-specific.
2154 Note that LLVM does not permit pointers to void (``void*``) nor does it
2155 permit pointers to labels (``label*``). Use ``i8*`` instead.
2165 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2166 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2167 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2168 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2169 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2170 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2171 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2180 A vector type is a simple derived type that represents a vector of
2181 elements. Vector types are used when multiple primitive data are
2182 operated in parallel using a single instruction (SIMD). A vector type
2183 requires a size (number of elements) and an underlying primitive data
2184 type. Vector types are considered :ref:`first class <t_firstclass>`.
2190 < <# elements> x <elementtype> >
2192 The number of elements is a constant integer value larger than 0;
2193 elementtype may be any integer, floating point or pointer type. Vectors
2194 of size zero are not allowed.
2198 +-------------------+--------------------------------------------------+
2199 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2200 +-------------------+--------------------------------------------------+
2201 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2202 +-------------------+--------------------------------------------------+
2203 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2204 +-------------------+--------------------------------------------------+
2205 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2206 +-------------------+--------------------------------------------------+
2215 The label type represents code labels.
2230 The token type is used when a value is associated with an instruction
2231 but all uses of the value must not attempt to introspect or obscure it.
2232 As such, it is not appropriate to have a :ref:`phi <i_phi>` or
2233 :ref:`select <i_select>` of type token.
2250 The metadata type represents embedded metadata. No derived types may be
2251 created from metadata except for :ref:`function <t_function>` arguments.
2264 Aggregate Types are a subset of derived types that can contain multiple
2265 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2266 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2276 The array type is a very simple derived type that arranges elements
2277 sequentially in memory. The array type requires a size (number of
2278 elements) and an underlying data type.
2284 [<# elements> x <elementtype>]
2286 The number of elements is a constant integer value; ``elementtype`` may
2287 be any type with a size.
2291 +------------------+--------------------------------------+
2292 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2293 +------------------+--------------------------------------+
2294 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2295 +------------------+--------------------------------------+
2296 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2297 +------------------+--------------------------------------+
2299 Here are some examples of multidimensional arrays:
2301 +-----------------------------+----------------------------------------------------------+
2302 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2303 +-----------------------------+----------------------------------------------------------+
2304 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2305 +-----------------------------+----------------------------------------------------------+
2306 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2307 +-----------------------------+----------------------------------------------------------+
2309 There is no restriction on indexing beyond the end of the array implied
2310 by a static type (though there are restrictions on indexing beyond the
2311 bounds of an allocated object in some cases). This means that
2312 single-dimension 'variable sized array' addressing can be implemented in
2313 LLVM with a zero length array type. An implementation of 'pascal style
2314 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2324 The structure type is used to represent a collection of data members
2325 together in memory. The elements of a structure may be any type that has
2328 Structures in memory are accessed using '``load``' and '``store``' by
2329 getting a pointer to a field with the '``getelementptr``' instruction.
2330 Structures in registers are accessed using the '``extractvalue``' and
2331 '``insertvalue``' instructions.
2333 Structures may optionally be "packed" structures, which indicate that
2334 the alignment of the struct is one byte, and that there is no padding
2335 between the elements. In non-packed structs, padding between field types
2336 is inserted as defined by the DataLayout string in the module, which is
2337 required to match what the underlying code generator expects.
2339 Structures can either be "literal" or "identified". A literal structure
2340 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2341 identified types are always defined at the top level with a name.
2342 Literal types are uniqued by their contents and can never be recursive
2343 or opaque since there is no way to write one. Identified types can be
2344 recursive, can be opaqued, and are never uniqued.
2350 %T1 = type { <type list> } ; Identified normal struct type
2351 %T2 = type <{ <type list> }> ; Identified packed struct type
2355 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2356 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2357 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2358 | ``{ 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``. |
2359 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2360 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2361 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2365 Opaque Structure Types
2366 """"""""""""""""""""""
2370 Opaque structure types are used to represent named structure types that
2371 do not have a body specified. This corresponds (for example) to the C
2372 notion of a forward declared structure.
2383 +--------------+-------------------+
2384 | ``opaque`` | An opaque type. |
2385 +--------------+-------------------+
2392 LLVM has several different basic types of constants. This section
2393 describes them all and their syntax.
2398 **Boolean constants**
2399 The two strings '``true``' and '``false``' are both valid constants
2401 **Integer constants**
2402 Standard integers (such as '4') are constants of the
2403 :ref:`integer <t_integer>` type. Negative numbers may be used with
2405 **Floating point constants**
2406 Floating point constants use standard decimal notation (e.g.
2407 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2408 hexadecimal notation (see below). The assembler requires the exact
2409 decimal value of a floating-point constant. For example, the
2410 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2411 decimal in binary. Floating point constants must have a :ref:`floating
2412 point <t_floating>` type.
2413 **Null pointer constants**
2414 The identifier '``null``' is recognized as a null pointer constant
2415 and must be of :ref:`pointer type <t_pointer>`.
2417 The one non-intuitive notation for constants is the hexadecimal form of
2418 floating point constants. For example, the form
2419 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2420 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2421 constants are required (and the only time that they are generated by the
2422 disassembler) is when a floating point constant must be emitted but it
2423 cannot be represented as a decimal floating point number in a reasonable
2424 number of digits. For example, NaN's, infinities, and other special
2425 values are represented in their IEEE hexadecimal format so that assembly
2426 and disassembly do not cause any bits to change in the constants.
2428 When using the hexadecimal form, constants of types half, float, and
2429 double are represented using the 16-digit form shown above (which
2430 matches the IEEE754 representation for double); half and float values
2431 must, however, be exactly representable as IEEE 754 half and single
2432 precision, respectively. Hexadecimal format is always used for long
2433 double, and there are three forms of long double. The 80-bit format used
2434 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2435 128-bit format used by PowerPC (two adjacent doubles) is represented by
2436 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2437 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2438 will only work if they match the long double format on your target.
2439 The IEEE 16-bit format (half precision) is represented by ``0xH``
2440 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2441 (sign bit at the left).
2443 There are no constants of type x86_mmx.
2445 .. _complexconstants:
2450 Complex constants are a (potentially recursive) combination of simple
2451 constants and smaller complex constants.
2453 **Structure constants**
2454 Structure constants are represented with notation similar to
2455 structure type definitions (a comma separated list of elements,
2456 surrounded by braces (``{}``)). For example:
2457 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2458 "``@G = external global i32``". Structure constants must have
2459 :ref:`structure type <t_struct>`, and the number and types of elements
2460 must match those specified by the type.
2462 Array constants are represented with notation similar to array type
2463 definitions (a comma separated list of elements, surrounded by
2464 square brackets (``[]``)). For example:
2465 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2466 :ref:`array type <t_array>`, and the number and types of elements must
2467 match those specified by the type. As a special case, character array
2468 constants may also be represented as a double-quoted string using the ``c``
2469 prefix. For example: "``c"Hello World\0A\00"``".
2470 **Vector constants**
2471 Vector constants are represented with notation similar to vector
2472 type definitions (a comma separated list of elements, surrounded by
2473 less-than/greater-than's (``<>``)). For example:
2474 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2475 must have :ref:`vector type <t_vector>`, and the number and types of
2476 elements must match those specified by the type.
2477 **Zero initialization**
2478 The string '``zeroinitializer``' can be used to zero initialize a
2479 value to zero of *any* type, including scalar and
2480 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2481 having to print large zero initializers (e.g. for large arrays) and
2482 is always exactly equivalent to using explicit zero initializers.
2484 A metadata node is a constant tuple without types. For example:
2485 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2486 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2487 Unlike other typed constants that are meant to be interpreted as part of
2488 the instruction stream, metadata is a place to attach additional
2489 information such as debug info.
2491 Global Variable and Function Addresses
2492 --------------------------------------
2494 The addresses of :ref:`global variables <globalvars>` and
2495 :ref:`functions <functionstructure>` are always implicitly valid
2496 (link-time) constants. These constants are explicitly referenced when
2497 the :ref:`identifier for the global <identifiers>` is used and always have
2498 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2501 .. code-block:: llvm
2505 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2512 The string '``undef``' can be used anywhere a constant is expected, and
2513 indicates that the user of the value may receive an unspecified
2514 bit-pattern. Undefined values may be of any type (other than '``label``'
2515 or '``void``') and be used anywhere a constant is permitted.
2517 Undefined values are useful because they indicate to the compiler that
2518 the program is well defined no matter what value is used. This gives the
2519 compiler more freedom to optimize. Here are some examples of
2520 (potentially surprising) transformations that are valid (in pseudo IR):
2522 .. code-block:: llvm
2532 This is safe because all of the output bits are affected by the undef
2533 bits. Any output bit can have a zero or one depending on the input bits.
2535 .. code-block:: llvm
2546 These logical operations have bits that are not always affected by the
2547 input. For example, if ``%X`` has a zero bit, then the output of the
2548 '``and``' operation will always be a zero for that bit, no matter what
2549 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2550 optimize or assume that the result of the '``and``' is '``undef``'.
2551 However, it is safe to assume that all bits of the '``undef``' could be
2552 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2553 all the bits of the '``undef``' operand to the '``or``' could be set,
2554 allowing the '``or``' to be folded to -1.
2556 .. code-block:: llvm
2558 %A = select undef, %X, %Y
2559 %B = select undef, 42, %Y
2560 %C = select %X, %Y, undef
2570 This set of examples shows that undefined '``select``' (and conditional
2571 branch) conditions can go *either way*, but they have to come from one
2572 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2573 both known to have a clear low bit, then ``%A`` would have to have a
2574 cleared low bit. However, in the ``%C`` example, the optimizer is
2575 allowed to assume that the '``undef``' operand could be the same as
2576 ``%Y``, allowing the whole '``select``' to be eliminated.
2578 .. code-block:: llvm
2580 %A = xor undef, undef
2597 This example points out that two '``undef``' operands are not
2598 necessarily the same. This can be surprising to people (and also matches
2599 C semantics) where they assume that "``X^X``" is always zero, even if
2600 ``X`` is undefined. This isn't true for a number of reasons, but the
2601 short answer is that an '``undef``' "variable" can arbitrarily change
2602 its value over its "live range". This is true because the variable
2603 doesn't actually *have a live range*. Instead, the value is logically
2604 read from arbitrary registers that happen to be around when needed, so
2605 the value is not necessarily consistent over time. In fact, ``%A`` and
2606 ``%C`` need to have the same semantics or the core LLVM "replace all
2607 uses with" concept would not hold.
2609 .. code-block:: llvm
2617 These examples show the crucial difference between an *undefined value*
2618 and *undefined behavior*. An undefined value (like '``undef``') is
2619 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2620 operation can be constant folded to '``undef``', because the '``undef``'
2621 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2622 However, in the second example, we can make a more aggressive
2623 assumption: because the ``undef`` is allowed to be an arbitrary value,
2624 we are allowed to assume that it could be zero. Since a divide by zero
2625 has *undefined behavior*, we are allowed to assume that the operation
2626 does not execute at all. This allows us to delete the divide and all
2627 code after it. Because the undefined operation "can't happen", the
2628 optimizer can assume that it occurs in dead code.
2630 .. code-block:: llvm
2632 a: store undef -> %X
2633 b: store %X -> undef
2638 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2639 value can be assumed to not have any effect; we can assume that the
2640 value is overwritten with bits that happen to match what was already
2641 there. However, a store *to* an undefined location could clobber
2642 arbitrary memory, therefore, it has undefined behavior.
2649 Poison values are similar to :ref:`undef values <undefvalues>`, however
2650 they also represent the fact that an instruction or constant expression
2651 that cannot evoke side effects has nevertheless detected a condition
2652 that results in undefined behavior.
2654 There is currently no way of representing a poison value in the IR; they
2655 only exist when produced by operations such as :ref:`add <i_add>` with
2658 Poison value behavior is defined in terms of value *dependence*:
2660 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2661 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2662 their dynamic predecessor basic block.
2663 - Function arguments depend on the corresponding actual argument values
2664 in the dynamic callers of their functions.
2665 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2666 instructions that dynamically transfer control back to them.
2667 - :ref:`Invoke <i_invoke>` instructions depend on the
2668 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2669 call instructions that dynamically transfer control back to them.
2670 - Non-volatile loads and stores depend on the most recent stores to all
2671 of the referenced memory addresses, following the order in the IR
2672 (including loads and stores implied by intrinsics such as
2673 :ref:`@llvm.memcpy <int_memcpy>`.)
2674 - An instruction with externally visible side effects depends on the
2675 most recent preceding instruction with externally visible side
2676 effects, following the order in the IR. (This includes :ref:`volatile
2677 operations <volatile>`.)
2678 - An instruction *control-depends* on a :ref:`terminator
2679 instruction <terminators>` if the terminator instruction has
2680 multiple successors and the instruction is always executed when
2681 control transfers to one of the successors, and may not be executed
2682 when control is transferred to another.
2683 - Additionally, an instruction also *control-depends* on a terminator
2684 instruction if the set of instructions it otherwise depends on would
2685 be different if the terminator had transferred control to a different
2687 - Dependence is transitive.
2689 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2690 with the additional effect that any instruction that has a *dependence*
2691 on a poison value has undefined behavior.
2693 Here are some examples:
2695 .. code-block:: llvm
2698 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2699 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2700 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
2701 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2703 store i32 %poison, i32* @g ; Poison value stored to memory.
2704 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
2706 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2708 %narrowaddr = bitcast i32* @g to i16*
2709 %wideaddr = bitcast i32* @g to i64*
2710 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2711 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
2713 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2714 br i1 %cmp, label %true, label %end ; Branch to either destination.
2717 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2718 ; it has undefined behavior.
2722 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2723 ; Both edges into this PHI are
2724 ; control-dependent on %cmp, so this
2725 ; always results in a poison value.
2727 store volatile i32 0, i32* @g ; This would depend on the store in %true
2728 ; if %cmp is true, or the store in %entry
2729 ; otherwise, so this is undefined behavior.
2731 br i1 %cmp, label %second_true, label %second_end
2732 ; The same branch again, but this time the
2733 ; true block doesn't have side effects.
2740 store volatile i32 0, i32* @g ; This time, the instruction always depends
2741 ; on the store in %end. Also, it is
2742 ; control-equivalent to %end, so this is
2743 ; well-defined (ignoring earlier undefined
2744 ; behavior in this example).
2748 Addresses of Basic Blocks
2749 -------------------------
2751 ``blockaddress(@function, %block)``
2753 The '``blockaddress``' constant computes the address of the specified
2754 basic block in the specified function, and always has an ``i8*`` type.
2755 Taking the address of the entry block is illegal.
2757 This value only has defined behavior when used as an operand to the
2758 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2759 against null. Pointer equality tests between labels addresses results in
2760 undefined behavior --- though, again, comparison against null is ok, and
2761 no label is equal to the null pointer. This may be passed around as an
2762 opaque pointer sized value as long as the bits are not inspected. This
2763 allows ``ptrtoint`` and arithmetic to be performed on these values so
2764 long as the original value is reconstituted before the ``indirectbr``
2767 Finally, some targets may provide defined semantics when using the value
2768 as the operand to an inline assembly, but that is target specific.
2772 Constant Expressions
2773 --------------------
2775 Constant expressions are used to allow expressions involving other
2776 constants to be used as constants. Constant expressions may be of any
2777 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2778 that does not have side effects (e.g. load and call are not supported).
2779 The following is the syntax for constant expressions:
2781 ``trunc (CST to TYPE)``
2782 Truncate a constant to another type. The bit size of CST must be
2783 larger than the bit size of TYPE. Both types must be integers.
2784 ``zext (CST to TYPE)``
2785 Zero extend a constant to another type. The bit size of CST must be
2786 smaller than the bit size of TYPE. Both types must be integers.
2787 ``sext (CST to TYPE)``
2788 Sign extend a constant to another type. The bit size of CST must be
2789 smaller than the bit size of TYPE. Both types must be integers.
2790 ``fptrunc (CST to TYPE)``
2791 Truncate a floating point constant to another floating point type.
2792 The size of CST must be larger than the size of TYPE. Both types
2793 must be floating point.
2794 ``fpext (CST to TYPE)``
2795 Floating point extend a constant to another type. The size of CST
2796 must be smaller or equal to the size of TYPE. Both types must be
2798 ``fptoui (CST to TYPE)``
2799 Convert a floating point constant to the corresponding unsigned
2800 integer constant. TYPE must be a scalar or vector integer type. CST
2801 must be of scalar or vector floating point type. Both CST and TYPE
2802 must be scalars, or vectors of the same number of elements. If the
2803 value won't fit in the integer type, the results are undefined.
2804 ``fptosi (CST to TYPE)``
2805 Convert a floating point constant to the corresponding signed
2806 integer constant. TYPE must be a scalar or vector integer type. CST
2807 must be of scalar or vector floating point type. Both CST and TYPE
2808 must be scalars, or vectors of the same number of elements. If the
2809 value won't fit in the integer type, the results are undefined.
2810 ``uitofp (CST to TYPE)``
2811 Convert an unsigned integer constant to the corresponding floating
2812 point constant. TYPE must be a scalar or vector floating point type.
2813 CST must be of scalar or vector integer type. Both CST and TYPE must
2814 be scalars, or vectors of the same number of elements. If the value
2815 won't fit in the floating point type, the results are undefined.
2816 ``sitofp (CST to TYPE)``
2817 Convert a signed integer constant to the corresponding floating
2818 point constant. TYPE must be a scalar or vector floating point type.
2819 CST must be of scalar or vector integer type. Both CST and TYPE must
2820 be scalars, or vectors of the same number of elements. If the value
2821 won't fit in the floating point type, the results are undefined.
2822 ``ptrtoint (CST to TYPE)``
2823 Convert a pointer typed constant to the corresponding integer
2824 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2825 pointer type. The ``CST`` value is zero extended, truncated, or
2826 unchanged to make it fit in ``TYPE``.
2827 ``inttoptr (CST to TYPE)``
2828 Convert an integer constant to a pointer constant. TYPE must be a
2829 pointer type. CST must be of integer type. The CST value is zero
2830 extended, truncated, or unchanged to make it fit in a pointer size.
2831 This one is *really* dangerous!
2832 ``bitcast (CST to TYPE)``
2833 Convert a constant, CST, to another TYPE. The constraints of the
2834 operands are the same as those for the :ref:`bitcast
2835 instruction <i_bitcast>`.
2836 ``addrspacecast (CST to TYPE)``
2837 Convert a constant pointer or constant vector of pointer, CST, to another
2838 TYPE in a different address space. The constraints of the operands are the
2839 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2840 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
2841 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2842 constants. As with the :ref:`getelementptr <i_getelementptr>`
2843 instruction, the index list may have zero or more indexes, which are
2844 required to make sense for the type of "pointer to TY".
2845 ``select (COND, VAL1, VAL2)``
2846 Perform the :ref:`select operation <i_select>` on constants.
2847 ``icmp COND (VAL1, VAL2)``
2848 Performs the :ref:`icmp operation <i_icmp>` on constants.
2849 ``fcmp COND (VAL1, VAL2)``
2850 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2851 ``extractelement (VAL, IDX)``
2852 Perform the :ref:`extractelement operation <i_extractelement>` on
2854 ``insertelement (VAL, ELT, IDX)``
2855 Perform the :ref:`insertelement operation <i_insertelement>` on
2857 ``shufflevector (VEC1, VEC2, IDXMASK)``
2858 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2860 ``extractvalue (VAL, IDX0, IDX1, ...)``
2861 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2862 constants. The index list is interpreted in a similar manner as
2863 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2864 least one index value must be specified.
2865 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2866 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2867 The index list is interpreted in a similar manner as indices in a
2868 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2869 value must be specified.
2870 ``OPCODE (LHS, RHS)``
2871 Perform the specified operation of the LHS and RHS constants. OPCODE
2872 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2873 binary <bitwiseops>` operations. The constraints on operands are
2874 the same as those for the corresponding instruction (e.g. no bitwise
2875 operations on floating point values are allowed).
2882 Inline Assembler Expressions
2883 ----------------------------
2885 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2886 Inline Assembly <moduleasm>`) through the use of a special value. This value
2887 represents the inline assembler as a template string (containing the
2888 instructions to emit), a list of operand constraints (stored as a string), a
2889 flag that indicates whether or not the inline asm expression has side effects,
2890 and a flag indicating whether the function containing the asm needs to align its
2891 stack conservatively.
2893 The template string supports argument substitution of the operands using "``$``"
2894 followed by a number, to indicate substitution of the given register/memory
2895 location, as specified by the constraint string. "``${NUM:MODIFIER}``" may also
2896 be used, where ``MODIFIER`` is a target-specific annotation for how to print the
2897 operand (See :ref:`inline-asm-modifiers`).
2899 A literal "``$``" may be included by using "``$$``" in the template. To include
2900 other special characters into the output, the usual "``\XX``" escapes may be
2901 used, just as in other strings. Note that after template substitution, the
2902 resulting assembly string is parsed by LLVM's integrated assembler unless it is
2903 disabled -- even when emitting a ``.s`` file -- and thus must contain assembly
2904 syntax known to LLVM.
2906 LLVM's support for inline asm is modeled closely on the requirements of Clang's
2907 GCC-compatible inline-asm support. Thus, the feature-set and the constraint and
2908 modifier codes listed here are similar or identical to those in GCC's inline asm
2909 support. However, to be clear, the syntax of the template and constraint strings
2910 described here is *not* the same as the syntax accepted by GCC and Clang, and,
2911 while most constraint letters are passed through as-is by Clang, some get
2912 translated to other codes when converting from the C source to the LLVM
2915 An example inline assembler expression is:
2917 .. code-block:: llvm
2919 i32 (i32) asm "bswap $0", "=r,r"
2921 Inline assembler expressions may **only** be used as the callee operand
2922 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2923 Thus, typically we have:
2925 .. code-block:: llvm
2927 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2929 Inline asms with side effects not visible in the constraint list must be
2930 marked as having side effects. This is done through the use of the
2931 '``sideeffect``' keyword, like so:
2933 .. code-block:: llvm
2935 call void asm sideeffect "eieio", ""()
2937 In some cases inline asms will contain code that will not work unless
2938 the stack is aligned in some way, such as calls or SSE instructions on
2939 x86, yet will not contain code that does that alignment within the asm.
2940 The compiler should make conservative assumptions about what the asm
2941 might contain and should generate its usual stack alignment code in the
2942 prologue if the '``alignstack``' keyword is present:
2944 .. code-block:: llvm
2946 call void asm alignstack "eieio", ""()
2948 Inline asms also support using non-standard assembly dialects. The
2949 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2950 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2951 the only supported dialects. An example is:
2953 .. code-block:: llvm
2955 call void asm inteldialect "eieio", ""()
2957 If multiple keywords appear the '``sideeffect``' keyword must come
2958 first, the '``alignstack``' keyword second and the '``inteldialect``'
2961 Inline Asm Constraint String
2962 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2964 The constraint list is a comma-separated string, each element containing one or
2965 more constraint codes.
2967 For each element in the constraint list an appropriate register or memory
2968 operand will be chosen, and it will be made available to assembly template
2969 string expansion as ``$0`` for the first constraint in the list, ``$1`` for the
2972 There are three different types of constraints, which are distinguished by a
2973 prefix symbol in front of the constraint code: Output, Input, and Clobber. The
2974 constraints must always be given in that order: outputs first, then inputs, then
2975 clobbers. They cannot be intermingled.
2977 There are also three different categories of constraint codes:
2979 - Register constraint. This is either a register class, or a fixed physical
2980 register. This kind of constraint will allocate a register, and if necessary,
2981 bitcast the argument or result to the appropriate type.
2982 - Memory constraint. This kind of constraint is for use with an instruction
2983 taking a memory operand. Different constraints allow for different addressing
2984 modes used by the target.
2985 - Immediate value constraint. This kind of constraint is for an integer or other
2986 immediate value which can be rendered directly into an instruction. The
2987 various target-specific constraints allow the selection of a value in the
2988 proper range for the instruction you wish to use it with.
2993 Output constraints are specified by an "``=``" prefix (e.g. "``=r``"). This
2994 indicates that the assembly will write to this operand, and the operand will
2995 then be made available as a return value of the ``asm`` expression. Output
2996 constraints do not consume an argument from the call instruction. (Except, see
2997 below about indirect outputs).
2999 Normally, it is expected that no output locations are written to by the assembly
3000 expression until *all* of the inputs have been read. As such, LLVM may assign
3001 the same register to an output and an input. If this is not safe (e.g. if the
3002 assembly contains two instructions, where the first writes to one output, and
3003 the second reads an input and writes to a second output), then the "``&``"
3004 modifier must be used (e.g. "``=&r``") to specify that the output is an
3005 "early-clobber" output. Marking an ouput as "early-clobber" ensures that LLVM
3006 will not use the same register for any inputs (other than an input tied to this
3012 Input constraints do not have a prefix -- just the constraint codes. Each input
3013 constraint will consume one argument from the call instruction. It is not
3014 permitted for the asm to write to any input register or memory location (unless
3015 that input is tied to an output). Note also that multiple inputs may all be
3016 assigned to the same register, if LLVM can determine that they necessarily all
3017 contain the same value.
3019 Instead of providing a Constraint Code, input constraints may also "tie"
3020 themselves to an output constraint, by providing an integer as the constraint
3021 string. Tied inputs still consume an argument from the call instruction, and
3022 take up a position in the asm template numbering as is usual -- they will simply
3023 be constrained to always use the same register as the output they've been tied
3024 to. For example, a constraint string of "``=r,0``" says to assign a register for
3025 output, and use that register as an input as well (it being the 0'th
3028 It is permitted to tie an input to an "early-clobber" output. In that case, no
3029 *other* input may share the same register as the input tied to the early-clobber
3030 (even when the other input has the same value).
3032 You may only tie an input to an output which has a register constraint, not a
3033 memory constraint. Only a single input may be tied to an output.
3035 There is also an "interesting" feature which deserves a bit of explanation: if a
3036 register class constraint allocates a register which is too small for the value
3037 type operand provided as input, the input value will be split into multiple
3038 registers, and all of them passed to the inline asm.
3040 However, this feature is often not as useful as you might think.
3042 Firstly, the registers are *not* guaranteed to be consecutive. So, on those
3043 architectures that have instructions which operate on multiple consecutive
3044 instructions, this is not an appropriate way to support them. (e.g. the 32-bit
3045 SparcV8 has a 64-bit load, which instruction takes a single 32-bit register. The
3046 hardware then loads into both the named register, and the next register. This
3047 feature of inline asm would not be useful to support that.)
3049 A few of the targets provide a template string modifier allowing explicit access
3050 to the second register of a two-register operand (e.g. MIPS ``L``, ``M``, and
3051 ``D``). On such an architecture, you can actually access the second allocated
3052 register (yet, still, not any subsequent ones). But, in that case, you're still
3053 probably better off simply splitting the value into two separate operands, for
3054 clarity. (e.g. see the description of the ``A`` constraint on X86, which,
3055 despite existing only for use with this feature, is not really a good idea to
3058 Indirect inputs and outputs
3059 """""""""""""""""""""""""""
3061 Indirect output or input constraints can be specified by the "``*``" modifier
3062 (which goes after the "``=``" in case of an output). This indicates that the asm
3063 will write to or read from the contents of an *address* provided as an input
3064 argument. (Note that in this way, indirect outputs act more like an *input* than
3065 an output: just like an input, they consume an argument of the call expression,
3066 rather than producing a return value. An indirect output constraint is an
3067 "output" only in that the asm is expected to write to the contents of the input
3068 memory location, instead of just read from it).
3070 This is most typically used for memory constraint, e.g. "``=*m``", to pass the
3071 address of a variable as a value.
3073 It is also possible to use an indirect *register* constraint, but only on output
3074 (e.g. "``=*r``"). This will cause LLVM to allocate a register for an output
3075 value normally, and then, separately emit a store to the address provided as
3076 input, after the provided inline asm. (It's not clear what value this
3077 functionality provides, compared to writing the store explicitly after the asm
3078 statement, and it can only produce worse code, since it bypasses many
3079 optimization passes. I would recommend not using it.)
3085 A clobber constraint is indicated by a "``~``" prefix. A clobber does not
3086 consume an input operand, nor generate an output. Clobbers cannot use any of the
3087 general constraint code letters -- they may use only explicit register
3088 constraints, e.g. "``~{eax}``". The one exception is that a clobber string of
3089 "``~{memory}``" indicates that the assembly writes to arbitrary undeclared
3090 memory locations -- not only the memory pointed to by a declared indirect
3096 After a potential prefix comes constraint code, or codes.
3098 A Constraint Code is either a single letter (e.g. "``r``"), a "``^``" character
3099 followed by two letters (e.g. "``^wc``"), or "``{``" register-name "``}``"
3102 The one and two letter constraint codes are typically chosen to be the same as
3103 GCC's constraint codes.
3105 A single constraint may include one or more than constraint code in it, leaving
3106 it up to LLVM to choose which one to use. This is included mainly for
3107 compatibility with the translation of GCC inline asm coming from clang.
3109 There are two ways to specify alternatives, and either or both may be used in an
3110 inline asm constraint list:
3112 1) Append the codes to each other, making a constraint code set. E.g. "``im``"
3113 or "``{eax}m``". This means "choose any of the options in the set". The
3114 choice of constraint is made independently for each constraint in the
3117 2) Use "``|``" between constraint code sets, creating alternatives. Every
3118 constraint in the constraint list must have the same number of alternative
3119 sets. With this syntax, the same alternative in *all* of the items in the
3120 constraint list will be chosen together.
3122 Putting those together, you might have a two operand constraint string like
3123 ``"rm|r,ri|rm"``. This indicates that if operand 0 is ``r`` or ``m``, then
3124 operand 1 may be one of ``r`` or ``i``. If operand 0 is ``r``, then operand 1
3125 may be one of ``r`` or ``m``. But, operand 0 and 1 cannot both be of type m.
3127 However, the use of either of the alternatives features is *NOT* recommended, as
3128 LLVM is not able to make an intelligent choice about which one to use. (At the
3129 point it currently needs to choose, not enough information is available to do so
3130 in a smart way.) Thus, it simply tries to make a choice that's most likely to
3131 compile, not one that will be optimal performance. (e.g., given "``rm``", it'll
3132 always choose to use memory, not registers). And, if given multiple registers,
3133 or multiple register classes, it will simply choose the first one. (In fact, it
3134 doesn't currently even ensure explicitly specified physical registers are
3135 unique, so specifying multiple physical registers as alternatives, like
3136 ``{r11}{r12},{r11}{r12}``, will assign r11 to both operands, not at all what was
3139 Supported Constraint Code List
3140 """"""""""""""""""""""""""""""
3142 The constraint codes are, in general, expected to behave the same way they do in
3143 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3144 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3145 and GCC likely indicates a bug in LLVM.
3147 Some constraint codes are typically supported by all targets:
3149 - ``r``: A register in the target's general purpose register class.
3150 - ``m``: A memory address operand. It is target-specific what addressing modes
3151 are supported, typical examples are register, or register + register offset,
3152 or register + immediate offset (of some target-specific size).
3153 - ``i``: An integer constant (of target-specific width). Allows either a simple
3154 immediate, or a relocatable value.
3155 - ``n``: An integer constant -- *not* including relocatable values.
3156 - ``s``: An integer constant, but allowing *only* relocatable values.
3157 - ``X``: Allows an operand of any kind, no constraint whatsoever. Typically
3158 useful to pass a label for an asm branch or call.
3160 .. FIXME: but that surely isn't actually okay to jump out of an asm
3161 block without telling llvm about the control transfer???)
3163 - ``{register-name}``: Requires exactly the named physical register.
3165 Other constraints are target-specific:
3169 - ``z``: An immediate integer 0. Outputs ``WZR`` or ``XZR``, as appropriate.
3170 - ``I``: An immediate integer valid for an ``ADD`` or ``SUB`` instruction,
3171 i.e. 0 to 4095 with optional shift by 12.
3172 - ``J``: An immediate integer that, when negated, is valid for an ``ADD`` or
3173 ``SUB`` instruction, i.e. -1 to -4095 with optional left shift by 12.
3174 - ``K``: An immediate integer that is valid for the 'bitmask immediate 32' of a
3175 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 32-bit register.
3176 - ``L``: An immediate integer that is valid for the 'bitmask immediate 64' of a
3177 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 64-bit register.
3178 - ``M``: An immediate integer for use with the ``MOV`` assembly alias on a
3179 32-bit register. This is a superset of ``K``: in addition to the bitmask
3180 immediate, also allows immediate integers which can be loaded with a single
3181 ``MOVZ`` or ``MOVL`` instruction.
3182 - ``N``: An immediate integer for use with the ``MOV`` assembly alias on a
3183 64-bit register. This is a superset of ``L``.
3184 - ``Q``: Memory address operand must be in a single register (no
3185 offsets). (However, LLVM currently does this for the ``m`` constraint as
3187 - ``r``: A 32 or 64-bit integer register (W* or X*).
3188 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register.
3189 - ``x``: A lower 128-bit floating-point/SIMD register (``V0`` to ``V15``).
3193 - ``r``: A 32 or 64-bit integer register.
3194 - ``[0-9]v``: The 32-bit VGPR register, number 0-9.
3195 - ``[0-9]s``: The 32-bit SGPR register, number 0-9.
3200 - ``Q``, ``Um``, ``Un``, ``Uq``, ``Us``, ``Ut``, ``Uv``, ``Uy``: Memory address
3201 operand. Treated the same as operand ``m``, at the moment.
3203 ARM and ARM's Thumb2 mode:
3205 - ``j``: An immediate integer between 0 and 65535 (valid for ``MOVW``)
3206 - ``I``: An immediate integer valid for a data-processing instruction.
3207 - ``J``: An immediate integer between -4095 and 4095.
3208 - ``K``: An immediate integer whose bitwise inverse is valid for a
3209 data-processing instruction. (Can be used with template modifier "``B``" to
3210 print the inverted value).
3211 - ``L``: An immediate integer whose negation is valid for a data-processing
3212 instruction. (Can be used with template modifier "``n``" to print the negated
3214 - ``M``: A power of two or a integer between 0 and 32.
3215 - ``N``: Invalid immediate constraint.
3216 - ``O``: Invalid immediate constraint.
3217 - ``r``: A general-purpose 32-bit integer register (``r0-r15``).
3218 - ``l``: In Thumb2 mode, low 32-bit GPR registers (``r0-r7``). In ARM mode, same
3220 - ``h``: In Thumb2 mode, a high 32-bit GPR register (``r8-r15``). In ARM mode,
3222 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3223 ``d0-d31``, or ``q0-q15``.
3224 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3225 ``d0-d7``, or ``q0-q3``.
3226 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3231 - ``I``: An immediate integer between 0 and 255.
3232 - ``J``: An immediate integer between -255 and -1.
3233 - ``K``: An immediate integer between 0 and 255, with optional left-shift by
3235 - ``L``: An immediate integer between -7 and 7.
3236 - ``M``: An immediate integer which is a multiple of 4 between 0 and 1020.
3237 - ``N``: An immediate integer between 0 and 31.
3238 - ``O``: An immediate integer which is a multiple of 4 between -508 and 508.
3239 - ``r``: A low 32-bit GPR register (``r0-r7``).
3240 - ``l``: A low 32-bit GPR register (``r0-r7``).
3241 - ``h``: A high GPR register (``r0-r7``).
3242 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3243 ``d0-d31``, or ``q0-q15``.
3244 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3245 ``d0-d7``, or ``q0-q3``.
3246 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3252 - ``o``, ``v``: A memory address operand, treated the same as constraint ``m``,
3254 - ``r``: A 32 or 64-bit register.
3258 - ``r``: An 8 or 16-bit register.
3262 - ``I``: An immediate signed 16-bit integer.
3263 - ``J``: An immediate integer zero.
3264 - ``K``: An immediate unsigned 16-bit integer.
3265 - ``L``: An immediate 32-bit integer, where the lower 16 bits are 0.
3266 - ``N``: An immediate integer between -65535 and -1.
3267 - ``O``: An immediate signed 15-bit integer.
3268 - ``P``: An immediate integer between 1 and 65535.
3269 - ``m``: A memory address operand. In MIPS-SE mode, allows a base address
3270 register plus 16-bit immediate offset. In MIPS mode, just a base register.
3271 - ``R``: A memory address operand. In MIPS-SE mode, allows a base address
3272 register plus a 9-bit signed offset. In MIPS mode, the same as constraint
3274 - ``ZC``: A memory address operand, suitable for use in a ``pref``, ``ll``, or
3275 ``sc`` instruction on the given subtarget (details vary).
3276 - ``r``, ``d``, ``y``: A 32 or 64-bit GPR register.
3277 - ``f``: A 32 or 64-bit FPU register (``F0-F31``), or a 128-bit MSA register
3278 (``W0-W31``). In the case of MSA registers, it is recommended to use the ``w``
3279 argument modifier for compatibility with GCC.
3280 - ``c``: A 32-bit or 64-bit GPR register suitable for indirect jump (always
3282 - ``l``: The ``lo`` register, 32 or 64-bit.
3287 - ``b``: A 1-bit integer register.
3288 - ``c`` or ``h``: A 16-bit integer register.
3289 - ``r``: A 32-bit integer register.
3290 - ``l`` or ``N``: A 64-bit integer register.
3291 - ``f``: A 32-bit float register.
3292 - ``d``: A 64-bit float register.
3297 - ``I``: An immediate signed 16-bit integer.
3298 - ``J``: An immediate unsigned 16-bit integer, shifted left 16 bits.
3299 - ``K``: An immediate unsigned 16-bit integer.
3300 - ``L``: An immediate signed 16-bit integer, shifted left 16 bits.
3301 - ``M``: An immediate integer greater than 31.
3302 - ``N``: An immediate integer that is an exact power of 2.
3303 - ``O``: The immediate integer constant 0.
3304 - ``P``: An immediate integer constant whose negation is a signed 16-bit
3306 - ``es``, ``o``, ``Q``, ``Z``, ``Zy``: A memory address operand, currently
3307 treated the same as ``m``.
3308 - ``r``: A 32 or 64-bit integer register.
3309 - ``b``: A 32 or 64-bit integer register, excluding ``R0`` (that is:
3311 - ``f``: A 32 or 64-bit float register (``F0-F31``), or when QPX is enabled, a
3312 128 or 256-bit QPX register (``Q0-Q31``; aliases the ``F`` registers).
3313 - ``v``: For ``4 x f32`` or ``4 x f64`` types, when QPX is enabled, a
3314 128 or 256-bit QPX register (``Q0-Q31``), otherwise a 128-bit
3315 altivec vector register (``V0-V31``).
3317 .. FIXME: is this a bug that v accepts QPX registers? I think this
3318 is supposed to only use the altivec vector registers?
3320 - ``y``: Condition register (``CR0-CR7``).
3321 - ``wc``: An individual CR bit in a CR register.
3322 - ``wa``, ``wd``, ``wf``: Any 128-bit VSX vector register, from the full VSX
3323 register set (overlapping both the floating-point and vector register files).
3324 - ``ws``: A 32 or 64-bit floating point register, from the full VSX register
3329 - ``I``: An immediate 13-bit signed integer.
3330 - ``r``: A 32-bit integer register.
3334 - ``I``: An immediate unsigned 8-bit integer.
3335 - ``J``: An immediate unsigned 12-bit integer.
3336 - ``K``: An immediate signed 16-bit integer.
3337 - ``L``: An immediate signed 20-bit integer.
3338 - ``M``: An immediate integer 0x7fffffff.
3339 - ``Q``, ``R``, ``S``, ``T``: A memory address operand, treated the same as
3340 ``m``, at the moment.
3341 - ``r`` or ``d``: A 32, 64, or 128-bit integer register.
3342 - ``a``: A 32, 64, or 128-bit integer address register (excludes R0, which in an
3343 address context evaluates as zero).
3344 - ``h``: A 32-bit value in the high part of a 64bit data register
3346 - ``f``: A 32, 64, or 128-bit floating point register.
3350 - ``I``: An immediate integer between 0 and 31.
3351 - ``J``: An immediate integer between 0 and 64.
3352 - ``K``: An immediate signed 8-bit integer.
3353 - ``L``: An immediate integer, 0xff or 0xffff or (in 64-bit mode only)
3355 - ``M``: An immediate integer between 0 and 3.
3356 - ``N``: An immediate unsigned 8-bit integer.
3357 - ``O``: An immediate integer between 0 and 127.
3358 - ``e``: An immediate 32-bit signed integer.
3359 - ``Z``: An immediate 32-bit unsigned integer.
3360 - ``o``, ``v``: Treated the same as ``m``, at the moment.
3361 - ``q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3362 ``l`` integer register. On X86-32, this is the ``a``, ``b``, ``c``, and ``d``
3363 registers, and on X86-64, it is all of the integer registers.
3364 - ``Q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3365 ``h`` integer register. This is the ``a``, ``b``, ``c``, and ``d`` registers.
3366 - ``r`` or ``l``: An 8, 16, 32, or 64-bit integer register.
3367 - ``R``: An 8, 16, 32, or 64-bit "legacy" integer register -- one which has
3368 existed since i386, and can be accessed without the REX prefix.
3369 - ``f``: A 32, 64, or 80-bit '387 FPU stack pseudo-register.
3370 - ``y``: A 64-bit MMX register, if MMX is enabled.
3371 - ``x``: If SSE is enabled: a 32 or 64-bit scalar operand, or 128-bit vector
3372 operand in a SSE register. If AVX is also enabled, can also be a 256-bit
3373 vector operand in an AVX register. If AVX-512 is also enabled, can also be a
3374 512-bit vector operand in an AVX512 register, Otherwise, an error.
3375 - ``Y``: The same as ``x``, if *SSE2* is enabled, otherwise an error.
3376 - ``A``: Special case: allocates EAX first, then EDX, for a single operand (in
3377 32-bit mode, a 64-bit integer operand will get split into two registers). It
3378 is not recommended to use this constraint, as in 64-bit mode, the 64-bit
3379 operand will get allocated only to RAX -- if two 32-bit operands are needed,
3380 you're better off splitting it yourself, before passing it to the asm
3385 - ``r``: A 32-bit integer register.
3388 .. _inline-asm-modifiers:
3390 Asm template argument modifiers
3391 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3393 In the asm template string, modifiers can be used on the operand reference, like
3396 The modifiers are, in general, expected to behave the same way they do in
3397 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3398 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3399 and GCC likely indicates a bug in LLVM.
3403 - ``c``: Print an immediate integer constant unadorned, without
3404 the target-specific immediate punctuation (e.g. no ``$`` prefix).
3405 - ``n``: Negate and print immediate integer constant unadorned, without the
3406 target-specific immediate punctuation (e.g. no ``$`` prefix).
3407 - ``l``: Print as an unadorned label, without the target-specific label
3408 punctuation (e.g. no ``$`` prefix).
3412 - ``w``: Print a GPR register with a ``w*`` name instead of ``x*`` name. E.g.,
3413 instead of ``x30``, print ``w30``.
3414 - ``x``: Print a GPR register with a ``x*`` name. (this is the default, anyhow).
3415 - ``b``, ``h``, ``s``, ``d``, ``q``: Print a floating-point/SIMD register with a
3416 ``b*``, ``h*``, ``s*``, ``d*``, or ``q*`` name, rather than the default of
3425 - ``a``: Print an operand as an address (with ``[`` and ``]`` surrounding a
3429 - ``y``: Print a VFP single-precision register as an indexed double (e.g. print
3430 as ``d4[1]`` instead of ``s9``)
3431 - ``B``: Bitwise invert and print an immediate integer constant without ``#``
3433 - ``L``: Print the low 16-bits of an immediate integer constant.
3434 - ``M``: Print as a register set suitable for ldm/stm. Also prints *all*
3435 register operands subsequent to the specified one (!), so use carefully.
3436 - ``Q``: Print the low-order register of a register-pair, or the low-order
3437 register of a two-register operand.
3438 - ``R``: Print the high-order register of a register-pair, or the high-order
3439 register of a two-register operand.
3440 - ``H``: Print the second register of a register-pair. (On a big-endian system,
3441 ``H`` is equivalent to ``Q``, and on little-endian system, ``H`` is equivalent
3444 .. FIXME: H doesn't currently support printing the second register
3445 of a two-register operand.
3447 - ``e``: Print the low doubleword register of a NEON quad register.
3448 - ``f``: Print the high doubleword register of a NEON quad register.
3449 - ``m``: Print the base register of a memory operand without the ``[`` and ``]``
3454 - ``L``: Print the second register of a two-register operand. Requires that it
3455 has been allocated consecutively to the first.
3457 .. FIXME: why is it restricted to consecutive ones? And there's
3458 nothing that ensures that happens, is there?
3460 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3461 nothing. Used to print 'addi' vs 'add' instructions.
3465 No additional modifiers.
3469 - ``X``: Print an immediate integer as hexadecimal
3470 - ``x``: Print the low 16 bits of an immediate integer as hexadecimal.
3471 - ``d``: Print an immediate integer as decimal.
3472 - ``m``: Subtract one and print an immediate integer as decimal.
3473 - ``z``: Print $0 if an immediate zero, otherwise print normally.
3474 - ``L``: Print the low-order register of a two-register operand, or prints the
3475 address of the low-order word of a double-word memory operand.
3477 .. FIXME: L seems to be missing memory operand support.
3479 - ``M``: Print the high-order register of a two-register operand, or prints the
3480 address of the high-order word of a double-word memory operand.
3482 .. FIXME: M seems to be missing memory operand support.
3484 - ``D``: Print the second register of a two-register operand, or prints the
3485 second word of a double-word memory operand. (On a big-endian system, ``D`` is
3486 equivalent to ``L``, and on little-endian system, ``D`` is equivalent to
3488 - ``w``: No effect. Provided for compatibility with GCC which requires this
3489 modifier in order to print MSA registers (``W0-W31``) with the ``f``
3498 - ``L``: Print the second register of a two-register operand. Requires that it
3499 has been allocated consecutively to the first.
3501 .. FIXME: why is it restricted to consecutive ones? And there's
3502 nothing that ensures that happens, is there?
3504 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3505 nothing. Used to print 'addi' vs 'add' instructions.
3506 - ``y``: For a memory operand, prints formatter for a two-register X-form
3507 instruction. (Currently always prints ``r0,OPERAND``).
3508 - ``U``: Prints 'u' if the memory operand is an update form, and nothing
3509 otherwise. (NOTE: LLVM does not support update form, so this will currently
3510 always print nothing)
3511 - ``X``: Prints 'x' if the memory operand is an indexed form. (NOTE: LLVM does
3512 not support indexed form, so this will currently always print nothing)
3520 SystemZ implements only ``n``, and does *not* support any of the other
3521 target-independent modifiers.
3525 - ``c``: Print an unadorned integer or symbol name. (The latter is
3526 target-specific behavior for this typically target-independent modifier).
3527 - ``A``: Print a register name with a '``*``' before it.
3528 - ``b``: Print an 8-bit register name (e.g. ``al``); do nothing on a memory
3530 - ``h``: Print the upper 8-bit register name (e.g. ``ah``); do nothing on a
3532 - ``w``: Print the 16-bit register name (e.g. ``ax``); do nothing on a memory
3534 - ``k``: Print the 32-bit register name (e.g. ``eax``); do nothing on a memory
3536 - ``q``: Print the 64-bit register name (e.g. ``rax``), if 64-bit registers are
3537 available, otherwise the 32-bit register name; do nothing on a memory operand.
3538 - ``n``: Negate and print an unadorned integer, or, for operands other than an
3539 immediate integer (e.g. a relocatable symbol expression), print a '-' before
3540 the operand. (The behavior for relocatable symbol expressions is a
3541 target-specific behavior for this typically target-independent modifier)
3542 - ``H``: Print a memory reference with additional offset +8.
3543 - ``P``: Print a memory reference or operand for use as the argument of a call
3544 instruction. (E.g. omit ``(rip)``, even though it's PC-relative.)
3548 No additional modifiers.
3554 The call instructions that wrap inline asm nodes may have a
3555 "``!srcloc``" MDNode attached to it that contains a list of constant
3556 integers. If present, the code generator will use the integer as the
3557 location cookie value when report errors through the ``LLVMContext``
3558 error reporting mechanisms. This allows a front-end to correlate backend
3559 errors that occur with inline asm back to the source code that produced
3562 .. code-block:: llvm
3564 call void asm sideeffect "something bad", ""(), !srcloc !42
3566 !42 = !{ i32 1234567 }
3568 It is up to the front-end to make sense of the magic numbers it places
3569 in the IR. If the MDNode contains multiple constants, the code generator
3570 will use the one that corresponds to the line of the asm that the error
3578 LLVM IR allows metadata to be attached to instructions in the program
3579 that can convey extra information about the code to the optimizers and
3580 code generator. One example application of metadata is source-level
3581 debug information. There are two metadata primitives: strings and nodes.
3583 Metadata does not have a type, and is not a value. If referenced from a
3584 ``call`` instruction, it uses the ``metadata`` type.
3586 All metadata are identified in syntax by a exclamation point ('``!``').
3588 .. _metadata-string:
3590 Metadata Nodes and Metadata Strings
3591 -----------------------------------
3593 A metadata string is a string surrounded by double quotes. It can
3594 contain any character by escaping non-printable characters with
3595 "``\xx``" where "``xx``" is the two digit hex code. For example:
3598 Metadata nodes are represented with notation similar to structure
3599 constants (a comma separated list of elements, surrounded by braces and
3600 preceded by an exclamation point). Metadata nodes can have any values as
3601 their operand. For example:
3603 .. code-block:: llvm
3605 !{ !"test\00", i32 10}
3607 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
3609 .. code-block:: llvm
3611 !0 = distinct !{!"test\00", i32 10}
3613 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
3614 content. They can also occur when transformations cause uniquing collisions
3615 when metadata operands change.
3617 A :ref:`named metadata <namedmetadatastructure>` is a collection of
3618 metadata nodes, which can be looked up in the module symbol table. For
3621 .. code-block:: llvm
3625 Metadata can be used as function arguments. Here ``llvm.dbg.value``
3626 function is using two metadata arguments:
3628 .. code-block:: llvm
3630 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
3632 Metadata can be attached to an instruction. Here metadata ``!21`` is attached
3633 to the ``add`` instruction using the ``!dbg`` identifier:
3635 .. code-block:: llvm
3637 %indvar.next = add i64 %indvar, 1, !dbg !21
3639 Metadata can also be attached to a function definition. Here metadata ``!22``
3640 is attached to the ``foo`` function using the ``!dbg`` identifier:
3642 .. code-block:: llvm
3644 define void @foo() !dbg !22 {
3648 More information about specific metadata nodes recognized by the
3649 optimizers and code generator is found below.
3651 .. _specialized-metadata:
3653 Specialized Metadata Nodes
3654 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3656 Specialized metadata nodes are custom data structures in metadata (as opposed
3657 to generic tuples). Their fields are labelled, and can be specified in any
3660 These aren't inherently debug info centric, but currently all the specialized
3661 metadata nodes are related to debug info.
3668 ``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
3669 ``retainedTypes:``, ``subprograms:``, ``globals:`` and ``imports:`` fields are
3670 tuples containing the debug info to be emitted along with the compile unit,
3671 regardless of code optimizations (some nodes are only emitted if there are
3672 references to them from instructions).
3674 .. code-block:: llvm
3676 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
3677 isOptimized: true, flags: "-O2", runtimeVersion: 2,
3678 splitDebugFilename: "abc.debug", emissionKind: 1,
3679 enums: !2, retainedTypes: !3, subprograms: !4,
3680 globals: !5, imports: !6)
3682 Compile unit descriptors provide the root scope for objects declared in a
3683 specific compilation unit. File descriptors are defined using this scope.
3684 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
3685 keep track of subprograms, global variables, type information, and imported
3686 entities (declarations and namespaces).
3693 ``DIFile`` nodes represent files. The ``filename:`` can include slashes.
3695 .. code-block:: llvm
3697 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
3699 Files are sometimes used in ``scope:`` fields, and are the only valid target
3700 for ``file:`` fields.
3707 ``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
3708 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
3710 .. code-block:: llvm
3712 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3713 encoding: DW_ATE_unsigned_char)
3714 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
3716 The ``encoding:`` describes the details of the type. Usually it's one of the
3719 .. code-block:: llvm
3725 DW_ATE_signed_char = 6
3727 DW_ATE_unsigned_char = 8
3729 .. _DISubroutineType:
3734 ``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
3735 refers to a tuple; the first operand is the return type, while the rest are the
3736 types of the formal arguments in order. If the first operand is ``null``, that
3737 represents a function with no return value (such as ``void foo() {}`` in C++).
3739 .. code-block:: llvm
3741 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
3742 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
3743 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
3750 ``DIDerivedType`` nodes represent types derived from other types, such as
3753 .. code-block:: llvm
3755 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3756 encoding: DW_ATE_unsigned_char)
3757 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
3760 The following ``tag:`` values are valid:
3762 .. code-block:: llvm
3764 DW_TAG_formal_parameter = 5
3766 DW_TAG_pointer_type = 15
3767 DW_TAG_reference_type = 16
3769 DW_TAG_ptr_to_member_type = 31
3770 DW_TAG_const_type = 38
3771 DW_TAG_volatile_type = 53
3772 DW_TAG_restrict_type = 55
3774 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
3775 <DICompositeType>` or :ref:`subprogram <DISubprogram>`. The type of the member
3776 is the ``baseType:``. The ``offset:`` is the member's bit offset.
3777 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
3778 argument of a subprogram.
3780 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3782 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3783 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3786 Note that the ``void *`` type is expressed as a type derived from NULL.
3788 .. _DICompositeType:
3793 ``DICompositeType`` nodes represent types composed of other types, like
3794 structures and unions. ``elements:`` points to a tuple of the composed types.
3796 If the source language supports ODR, the ``identifier:`` field gives the unique
3797 identifier used for type merging between modules. When specified, other types
3798 can refer to composite types indirectly via a :ref:`metadata string
3799 <metadata-string>` that matches their identifier.
3801 .. code-block:: llvm
3803 !0 = !DIEnumerator(name: "SixKind", value: 7)
3804 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3805 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3806 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
3807 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3808 elements: !{!0, !1, !2})
3810 The following ``tag:`` values are valid:
3812 .. code-block:: llvm
3814 DW_TAG_array_type = 1
3815 DW_TAG_class_type = 2
3816 DW_TAG_enumeration_type = 4
3817 DW_TAG_structure_type = 19
3818 DW_TAG_union_type = 23
3819 DW_TAG_subroutine_type = 21
3820 DW_TAG_inheritance = 28
3823 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
3824 descriptors <DISubrange>`, each representing the range of subscripts at that
3825 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3826 array type is a native packed vector.
3828 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
3829 descriptors <DIEnumerator>`, each representing the definition of an enumeration
3830 value for the set. All enumeration type descriptors are collected in the
3831 ``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
3833 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3834 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
3835 <DIDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
3842 ``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3843 :ref:`DICompositeType`. ``count: -1`` indicates an empty array.
3845 .. code-block:: llvm
3847 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
3848 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
3849 !2 = !DISubrange(count: -1) ; empty array.
3856 ``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3857 variants of :ref:`DICompositeType`.
3859 .. code-block:: llvm
3861 !0 = !DIEnumerator(name: "SixKind", value: 7)
3862 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3863 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3865 DITemplateTypeParameter
3866 """""""""""""""""""""""
3868 ``DITemplateTypeParameter`` nodes represent type parameters to generic source
3869 language constructs. They are used (optionally) in :ref:`DICompositeType` and
3870 :ref:`DISubprogram` ``templateParams:`` fields.
3872 .. code-block:: llvm
3874 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
3876 DITemplateValueParameter
3877 """"""""""""""""""""""""
3879 ``DITemplateValueParameter`` nodes represent value parameters to generic source
3880 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3881 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3882 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
3883 :ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
3885 .. code-block:: llvm
3887 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
3892 ``DINamespace`` nodes represent namespaces in the source language.
3894 .. code-block:: llvm
3896 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
3901 ``DIGlobalVariable`` nodes represent global variables in the source language.
3903 .. code-block:: llvm
3905 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
3906 file: !2, line: 7, type: !3, isLocal: true,
3907 isDefinition: false, variable: i32* @foo,
3910 All global variables should be referenced by the `globals:` field of a
3911 :ref:`compile unit <DICompileUnit>`.
3918 ``DISubprogram`` nodes represent functions from the source language. A
3919 ``DISubprogram`` may be attached to a function definition using ``!dbg``
3920 metadata. The ``variables:`` field points at :ref:`variables <DILocalVariable>`
3921 that must be retained, even if their IR counterparts are optimized out of
3922 the IR. The ``type:`` field must point at an :ref:`DISubroutineType`.
3924 .. code-block:: llvm
3926 define void @_Z3foov() !dbg !0 {
3930 !0 = distinct !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
3931 file: !2, line: 7, type: !3, isLocal: true,
3932 isDefinition: false, scopeLine: 8,
3934 virtuality: DW_VIRTUALITY_pure_virtual,
3935 virtualIndex: 10, flags: DIFlagPrototyped,
3936 isOptimized: true, templateParams: !5,
3937 declaration: !6, variables: !7)
3944 ``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
3945 <DISubprogram>`. The line number and column numbers are used to distinguish
3946 two lexical blocks at same depth. They are valid targets for ``scope:``
3949 .. code-block:: llvm
3951 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
3953 Usually lexical blocks are ``distinct`` to prevent node merging based on
3956 .. _DILexicalBlockFile:
3961 ``DILexicalBlockFile`` nodes are used to discriminate between sections of a
3962 :ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
3963 indicate textual inclusion, or the ``discriminator:`` field can be used to
3964 discriminate between control flow within a single block in the source language.
3966 .. code-block:: llvm
3968 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
3969 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
3970 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
3977 ``DILocation`` nodes represent source debug locations. The ``scope:`` field is
3978 mandatory, and points at an :ref:`DILexicalBlockFile`, an
3979 :ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
3981 .. code-block:: llvm
3983 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
3985 .. _DILocalVariable:
3990 ``DILocalVariable`` nodes represent local variables in the source language. If
3991 the ``arg:`` field is set to non-zero, then this variable is a subprogram
3992 parameter, and it will be included in the ``variables:`` field of its
3993 :ref:`DISubprogram`.
3995 .. code-block:: llvm
3997 !0 = !DILocalVariable(name: "this", arg: 1, scope: !3, file: !2, line: 7,
3998 type: !3, flags: DIFlagArtificial)
3999 !1 = !DILocalVariable(name: "x", arg: 2, scope: !4, file: !2, line: 7,
4001 !2 = !DILocalVariable(name: "y", scope: !5, file: !2, line: 7, type: !3)
4006 ``DIExpression`` nodes represent DWARF expression sequences. They are used in
4007 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
4008 describe how the referenced LLVM variable relates to the source language
4011 The current supported vocabulary is limited:
4013 - ``DW_OP_deref`` dereferences the working expression.
4014 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
4015 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
4016 here, respectively) of the variable piece from the working expression.
4018 .. code-block:: llvm
4020 !0 = !DIExpression(DW_OP_deref)
4021 !1 = !DIExpression(DW_OP_plus, 3)
4022 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
4023 !3 = !DIExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
4028 ``DIObjCProperty`` nodes represent Objective-C property nodes.
4030 .. code-block:: llvm
4032 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
4033 getter: "getFoo", attributes: 7, type: !2)
4038 ``DIImportedEntity`` nodes represent entities (such as modules) imported into a
4041 .. code-block:: llvm
4043 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
4044 entity: !1, line: 7)
4049 In LLVM IR, memory does not have types, so LLVM's own type system is not
4050 suitable for doing TBAA. Instead, metadata is added to the IR to
4051 describe a type system of a higher level language. This can be used to
4052 implement typical C/C++ TBAA, but it can also be used to implement
4053 custom alias analysis behavior for other languages.
4055 The current metadata format is very simple. TBAA metadata nodes have up
4056 to three fields, e.g.:
4058 .. code-block:: llvm
4060 !0 = !{ !"an example type tree" }
4061 !1 = !{ !"int", !0 }
4062 !2 = !{ !"float", !0 }
4063 !3 = !{ !"const float", !2, i64 1 }
4065 The first field is an identity field. It can be any value, usually a
4066 metadata string, which uniquely identifies the type. The most important
4067 name in the tree is the name of the root node. Two trees with different
4068 root node names are entirely disjoint, even if they have leaves with
4071 The second field identifies the type's parent node in the tree, or is
4072 null or omitted for a root node. A type is considered to alias all of
4073 its descendants and all of its ancestors in the tree. Also, a type is
4074 considered to alias all types in other trees, so that bitcode produced
4075 from multiple front-ends is handled conservatively.
4077 If the third field is present, it's an integer which if equal to 1
4078 indicates that the type is "constant" (meaning
4079 ``pointsToConstantMemory`` should return true; see `other useful
4080 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
4082 '``tbaa.struct``' Metadata
4083 ^^^^^^^^^^^^^^^^^^^^^^^^^^
4085 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
4086 aggregate assignment operations in C and similar languages, however it
4087 is defined to copy a contiguous region of memory, which is more than
4088 strictly necessary for aggregate types which contain holes due to
4089 padding. Also, it doesn't contain any TBAA information about the fields
4092 ``!tbaa.struct`` metadata can describe which memory subregions in a
4093 memcpy are padding and what the TBAA tags of the struct are.
4095 The current metadata format is very simple. ``!tbaa.struct`` metadata
4096 nodes are a list of operands which are in conceptual groups of three.
4097 For each group of three, the first operand gives the byte offset of a
4098 field in bytes, the second gives its size in bytes, and the third gives
4101 .. code-block:: llvm
4103 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
4105 This describes a struct with two fields. The first is at offset 0 bytes
4106 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
4107 and has size 4 bytes and has tbaa tag !2.
4109 Note that the fields need not be contiguous. In this example, there is a
4110 4 byte gap between the two fields. This gap represents padding which
4111 does not carry useful data and need not be preserved.
4113 '``noalias``' and '``alias.scope``' Metadata
4114 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4116 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
4117 noalias memory-access sets. This means that some collection of memory access
4118 instructions (loads, stores, memory-accessing calls, etc.) that carry
4119 ``noalias`` metadata can specifically be specified not to alias with some other
4120 collection of memory access instructions that carry ``alias.scope`` metadata.
4121 Each type of metadata specifies a list of scopes where each scope has an id and
4122 a domain. When evaluating an aliasing query, if for some domain, the set
4123 of scopes with that domain in one instruction's ``alias.scope`` list is a
4124 subset of (or equal to) the set of scopes for that domain in another
4125 instruction's ``noalias`` list, then the two memory accesses are assumed not to
4128 The metadata identifying each domain is itself a list containing one or two
4129 entries. The first entry is the name of the domain. Note that if the name is a
4130 string then it can be combined across functions and translation units. A
4131 self-reference can be used to create globally unique domain names. A
4132 descriptive string may optionally be provided as a second list entry.
4134 The metadata identifying each scope is also itself a list containing two or
4135 three entries. The first entry is the name of the scope. Note that if the name
4136 is a string then it can be combined across functions and translation units. A
4137 self-reference can be used to create globally unique scope names. A metadata
4138 reference to the scope's domain is the second entry. A descriptive string may
4139 optionally be provided as a third list entry.
4143 .. code-block:: llvm
4145 ; Two scope domains:
4149 ; Some scopes in these domains:
4155 !5 = !{!4} ; A list containing only scope !4
4159 ; These two instructions don't alias:
4160 %0 = load float, float* %c, align 4, !alias.scope !5
4161 store float %0, float* %arrayidx.i, align 4, !noalias !5
4163 ; These two instructions also don't alias (for domain !1, the set of scopes
4164 ; in the !alias.scope equals that in the !noalias list):
4165 %2 = load float, float* %c, align 4, !alias.scope !5
4166 store float %2, float* %arrayidx.i2, align 4, !noalias !6
4168 ; These two instructions may alias (for domain !0, the set of scopes in
4169 ; the !noalias list is not a superset of, or equal to, the scopes in the
4170 ; !alias.scope list):
4171 %2 = load float, float* %c, align 4, !alias.scope !6
4172 store float %0, float* %arrayidx.i, align 4, !noalias !7
4174 '``fpmath``' Metadata
4175 ^^^^^^^^^^^^^^^^^^^^^
4177 ``fpmath`` metadata may be attached to any instruction of floating point
4178 type. It can be used to express the maximum acceptable error in the
4179 result of that instruction, in ULPs, thus potentially allowing the
4180 compiler to use a more efficient but less accurate method of computing
4181 it. ULP is defined as follows:
4183 If ``x`` is a real number that lies between two finite consecutive
4184 floating-point numbers ``a`` and ``b``, without being equal to one
4185 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
4186 distance between the two non-equal finite floating-point numbers
4187 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
4189 The metadata node shall consist of a single positive floating point
4190 number representing the maximum relative error, for example:
4192 .. code-block:: llvm
4194 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
4198 '``range``' Metadata
4199 ^^^^^^^^^^^^^^^^^^^^
4201 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
4202 integer types. It expresses the possible ranges the loaded value or the value
4203 returned by the called function at this call site is in. The ranges are
4204 represented with a flattened list of integers. The loaded value or the value
4205 returned is known to be in the union of the ranges defined by each consecutive
4206 pair. Each pair has the following properties:
4208 - The type must match the type loaded by the instruction.
4209 - The pair ``a,b`` represents the range ``[a,b)``.
4210 - Both ``a`` and ``b`` are constants.
4211 - The range is allowed to wrap.
4212 - The range should not represent the full or empty set. That is,
4215 In addition, the pairs must be in signed order of the lower bound and
4216 they must be non-contiguous.
4220 .. code-block:: llvm
4222 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
4223 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
4224 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
4225 %d = invoke i8 @bar() to label %cont
4226 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
4228 !0 = !{ i8 0, i8 2 }
4229 !1 = !{ i8 255, i8 2 }
4230 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
4231 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
4233 '``unpredictable``' Metadata
4234 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4236 ``unpredictable`` metadata may be attached to any branch or switch
4237 instruction. It can be used to express the unpredictability of control
4238 flow. Similar to the llvm.expect intrinsic, it may be used to alter
4239 optimizations related to compare and branch instructions. The metadata
4240 is treated as a boolean value; if it exists, it signals that the branch
4241 or switch that it is attached to is completely unpredictable.
4246 It is sometimes useful to attach information to loop constructs. Currently,
4247 loop metadata is implemented as metadata attached to the branch instruction
4248 in the loop latch block. This type of metadata refer to a metadata node that is
4249 guaranteed to be separate for each loop. The loop identifier metadata is
4250 specified with the name ``llvm.loop``.
4252 The loop identifier metadata is implemented using a metadata that refers to
4253 itself to avoid merging it with any other identifier metadata, e.g.,
4254 during module linkage or function inlining. That is, each loop should refer
4255 to their own identification metadata even if they reside in separate functions.
4256 The following example contains loop identifier metadata for two separate loop
4259 .. code-block:: llvm
4264 The loop identifier metadata can be used to specify additional
4265 per-loop metadata. Any operands after the first operand can be treated
4266 as user-defined metadata. For example the ``llvm.loop.unroll.count``
4267 suggests an unroll factor to the loop unroller:
4269 .. code-block:: llvm
4271 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
4274 !1 = !{!"llvm.loop.unroll.count", i32 4}
4276 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
4277 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4279 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
4280 used to control per-loop vectorization and interleaving parameters such as
4281 vectorization width and interleave count. These metadata should be used in
4282 conjunction with ``llvm.loop`` loop identification metadata. The
4283 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
4284 optimization hints and the optimizer will only interleave and vectorize loops if
4285 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
4286 which contains information about loop-carried memory dependencies can be helpful
4287 in determining the safety of these transformations.
4289 '``llvm.loop.interleave.count``' Metadata
4290 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4292 This metadata suggests an interleave count to the loop interleaver.
4293 The first operand is the string ``llvm.loop.interleave.count`` and the
4294 second operand is an integer specifying the interleave count. For
4297 .. code-block:: llvm
4299 !0 = !{!"llvm.loop.interleave.count", i32 4}
4301 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
4302 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
4303 then the interleave count will be determined automatically.
4305 '``llvm.loop.vectorize.enable``' Metadata
4306 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4308 This metadata selectively enables or disables vectorization for the loop. The
4309 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
4310 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
4311 0 disables vectorization:
4313 .. code-block:: llvm
4315 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
4316 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
4318 '``llvm.loop.vectorize.width``' Metadata
4319 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4321 This metadata sets the target width of the vectorizer. The first
4322 operand is the string ``llvm.loop.vectorize.width`` and the second
4323 operand is an integer specifying the width. For example:
4325 .. code-block:: llvm
4327 !0 = !{!"llvm.loop.vectorize.width", i32 4}
4329 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
4330 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
4331 0 or if the loop does not have this metadata the width will be
4332 determined automatically.
4334 '``llvm.loop.unroll``'
4335 ^^^^^^^^^^^^^^^^^^^^^^
4337 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
4338 optimization hints such as the unroll factor. ``llvm.loop.unroll``
4339 metadata should be used in conjunction with ``llvm.loop`` loop
4340 identification metadata. The ``llvm.loop.unroll`` metadata are only
4341 optimization hints and the unrolling will only be performed if the
4342 optimizer believes it is safe to do so.
4344 '``llvm.loop.unroll.count``' Metadata
4345 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4347 This metadata suggests an unroll factor to the loop unroller. The
4348 first operand is the string ``llvm.loop.unroll.count`` and the second
4349 operand is a positive integer specifying the unroll factor. For
4352 .. code-block:: llvm
4354 !0 = !{!"llvm.loop.unroll.count", i32 4}
4356 If the trip count of the loop is less than the unroll count the loop
4357 will be partially unrolled.
4359 '``llvm.loop.unroll.disable``' Metadata
4360 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4362 This metadata disables loop unrolling. The metadata has a single operand
4363 which is the string ``llvm.loop.unroll.disable``. For example:
4365 .. code-block:: llvm
4367 !0 = !{!"llvm.loop.unroll.disable"}
4369 '``llvm.loop.unroll.runtime.disable``' Metadata
4370 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4372 This metadata disables runtime loop unrolling. The metadata has a single
4373 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
4375 .. code-block:: llvm
4377 !0 = !{!"llvm.loop.unroll.runtime.disable"}
4379 '``llvm.loop.unroll.enable``' Metadata
4380 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4382 This metadata suggests that the loop should be fully unrolled if the trip count
4383 is known at compile time and partially unrolled if the trip count is not known
4384 at compile time. The metadata has a single operand which is the string
4385 ``llvm.loop.unroll.enable``. For example:
4387 .. code-block:: llvm
4389 !0 = !{!"llvm.loop.unroll.enable"}
4391 '``llvm.loop.unroll.full``' Metadata
4392 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4394 This metadata suggests that the loop should be unrolled fully. The
4395 metadata has a single operand which is the string ``llvm.loop.unroll.full``.
4398 .. code-block:: llvm
4400 !0 = !{!"llvm.loop.unroll.full"}
4405 Metadata types used to annotate memory accesses with information helpful
4406 for optimizations are prefixed with ``llvm.mem``.
4408 '``llvm.mem.parallel_loop_access``' Metadata
4409 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4411 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
4412 or metadata containing a list of loop identifiers for nested loops.
4413 The metadata is attached to memory accessing instructions and denotes that
4414 no loop carried memory dependence exist between it and other instructions denoted
4415 with the same loop identifier.
4417 Precisely, given two instructions ``m1`` and ``m2`` that both have the
4418 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
4419 set of loops associated with that metadata, respectively, then there is no loop
4420 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
4423 As a special case, if all memory accessing instructions in a loop have
4424 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
4425 loop has no loop carried memory dependences and is considered to be a parallel
4428 Note that if not all memory access instructions have such metadata referring to
4429 the loop, then the loop is considered not being trivially parallel. Additional
4430 memory dependence analysis is required to make that determination. As a fail
4431 safe mechanism, this causes loops that were originally parallel to be considered
4432 sequential (if optimization passes that are unaware of the parallel semantics
4433 insert new memory instructions into the loop body).
4435 Example of a loop that is considered parallel due to its correct use of
4436 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
4437 metadata types that refer to the same loop identifier metadata.
4439 .. code-block:: llvm
4443 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
4445 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4447 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
4453 It is also possible to have nested parallel loops. In that case the
4454 memory accesses refer to a list of loop identifier metadata nodes instead of
4455 the loop identifier metadata node directly:
4457 .. code-block:: llvm
4461 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
4463 br label %inner.for.body
4467 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4469 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
4471 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
4475 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
4477 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
4479 outer.for.end: ; preds = %for.body
4481 !0 = !{!1, !2} ; a list of loop identifiers
4482 !1 = !{!1} ; an identifier for the inner loop
4483 !2 = !{!2} ; an identifier for the outer loop
4488 The ``llvm.bitsets`` global metadata is used to implement
4489 :doc:`bitsets <BitSets>`.
4491 '``invariant.group``' Metadata
4492 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4494 The ``invariant.group`` metadata may be attached to ``load``/``store`` instructions.
4495 The existence of the ``invariant.group`` metadata on the instruction tells
4496 the optimizer that every ``load`` and ``store`` to the same pointer operand
4497 within the same invariant group can be assumed to load or store the same
4498 value (but see the ``llvm.invariant.group.barrier`` intrinsic which affects
4499 when two pointers are considered the same).
4503 .. code-block:: llvm
4505 @unknownPtr = external global i8
4508 store i8 42, i8* %ptr, !invariant.group !0
4509 call void @foo(i8* %ptr)
4511 %a = load i8, i8* %ptr, !invariant.group !0 ; Can assume that value under %ptr didn't change
4512 call void @foo(i8* %ptr)
4513 %b = load i8, i8* %ptr, !invariant.group !1 ; Can't assume anything, because group changed
4515 %newPtr = call i8* @getPointer(i8* %ptr)
4516 %c = load i8, i8* %newPtr, !invariant.group !0 ; Can't assume anything, because we only have information about %ptr
4518 %unknownValue = load i8, i8* @unknownPtr
4519 store i8 %unknownValue, i8* %ptr, !invariant.group !0 ; Can assume that %unknownValue == 42
4521 call void @foo(i8* %ptr)
4522 %newPtr2 = call i8* @llvm.invariant.group.barrier(i8* %ptr)
4523 %d = load i8, i8* %newPtr2, !invariant.group !0 ; Can't step through invariant.group.barrier to get value of %ptr
4526 declare void @foo(i8*)
4527 declare i8* @getPointer(i8*)
4528 declare i8* @llvm.invariant.group.barrier(i8*)
4530 !0 = !{!"magic ptr"}
4531 !1 = !{!"other ptr"}
4535 Module Flags Metadata
4536 =====================
4538 Information about the module as a whole is difficult to convey to LLVM's
4539 subsystems. The LLVM IR isn't sufficient to transmit this information.
4540 The ``llvm.module.flags`` named metadata exists in order to facilitate
4541 this. These flags are in the form of key / value pairs --- much like a
4542 dictionary --- making it easy for any subsystem who cares about a flag to
4545 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
4546 Each triplet has the following form:
4548 - The first element is a *behavior* flag, which specifies the behavior
4549 when two (or more) modules are merged together, and it encounters two
4550 (or more) metadata with the same ID. The supported behaviors are
4552 - The second element is a metadata string that is a unique ID for the
4553 metadata. Each module may only have one flag entry for each unique ID (not
4554 including entries with the **Require** behavior).
4555 - The third element is the value of the flag.
4557 When two (or more) modules are merged together, the resulting
4558 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
4559 each unique metadata ID string, there will be exactly one entry in the merged
4560 modules ``llvm.module.flags`` metadata table, and the value for that entry will
4561 be determined by the merge behavior flag, as described below. The only exception
4562 is that entries with the *Require* behavior are always preserved.
4564 The following behaviors are supported:
4575 Emits an error if two values disagree, otherwise the resulting value
4576 is that of the operands.
4580 Emits a warning if two values disagree. The result value will be the
4581 operand for the flag from the first module being linked.
4585 Adds a requirement that another module flag be present and have a
4586 specified value after linking is performed. The value must be a
4587 metadata pair, where the first element of the pair is the ID of the
4588 module flag to be restricted, and the second element of the pair is
4589 the value the module flag should be restricted to. This behavior can
4590 be used to restrict the allowable results (via triggering of an
4591 error) of linking IDs with the **Override** behavior.
4595 Uses the specified value, regardless of the behavior or value of the
4596 other module. If both modules specify **Override**, but the values
4597 differ, an error will be emitted.
4601 Appends the two values, which are required to be metadata nodes.
4605 Appends the two values, which are required to be metadata
4606 nodes. However, duplicate entries in the second list are dropped
4607 during the append operation.
4609 It is an error for a particular unique flag ID to have multiple behaviors,
4610 except in the case of **Require** (which adds restrictions on another metadata
4611 value) or **Override**.
4613 An example of module flags:
4615 .. code-block:: llvm
4617 !0 = !{ i32 1, !"foo", i32 1 }
4618 !1 = !{ i32 4, !"bar", i32 37 }
4619 !2 = !{ i32 2, !"qux", i32 42 }
4620 !3 = !{ i32 3, !"qux",
4625 !llvm.module.flags = !{ !0, !1, !2, !3 }
4627 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
4628 if two or more ``!"foo"`` flags are seen is to emit an error if their
4629 values are not equal.
4631 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
4632 behavior if two or more ``!"bar"`` flags are seen is to use the value
4635 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
4636 behavior if two or more ``!"qux"`` flags are seen is to emit a
4637 warning if their values are not equal.
4639 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
4645 The behavior is to emit an error if the ``llvm.module.flags`` does not
4646 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
4649 Objective-C Garbage Collection Module Flags Metadata
4650 ----------------------------------------------------
4652 On the Mach-O platform, Objective-C stores metadata about garbage
4653 collection in a special section called "image info". The metadata
4654 consists of a version number and a bitmask specifying what types of
4655 garbage collection are supported (if any) by the file. If two or more
4656 modules are linked together their garbage collection metadata needs to
4657 be merged rather than appended together.
4659 The Objective-C garbage collection module flags metadata consists of the
4660 following key-value pairs:
4669 * - ``Objective-C Version``
4670 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
4672 * - ``Objective-C Image Info Version``
4673 - **[Required]** --- The version of the image info section. Currently
4676 * - ``Objective-C Image Info Section``
4677 - **[Required]** --- The section to place the metadata. Valid values are
4678 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
4679 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
4680 Objective-C ABI version 2.
4682 * - ``Objective-C Garbage Collection``
4683 - **[Required]** --- Specifies whether garbage collection is supported or
4684 not. Valid values are 0, for no garbage collection, and 2, for garbage
4685 collection supported.
4687 * - ``Objective-C GC Only``
4688 - **[Optional]** --- Specifies that only garbage collection is supported.
4689 If present, its value must be 6. This flag requires that the
4690 ``Objective-C Garbage Collection`` flag have the value 2.
4692 Some important flag interactions:
4694 - If a module with ``Objective-C Garbage Collection`` set to 0 is
4695 merged with a module with ``Objective-C Garbage Collection`` set to
4696 2, then the resulting module has the
4697 ``Objective-C Garbage Collection`` flag set to 0.
4698 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
4699 merged with a module with ``Objective-C GC Only`` set to 6.
4701 Automatic Linker Flags Module Flags Metadata
4702 --------------------------------------------
4704 Some targets support embedding flags to the linker inside individual object
4705 files. Typically this is used in conjunction with language extensions which
4706 allow source files to explicitly declare the libraries they depend on, and have
4707 these automatically be transmitted to the linker via object files.
4709 These flags are encoded in the IR using metadata in the module flags section,
4710 using the ``Linker Options`` key. The merge behavior for this flag is required
4711 to be ``AppendUnique``, and the value for the key is expected to be a metadata
4712 node which should be a list of other metadata nodes, each of which should be a
4713 list of metadata strings defining linker options.
4715 For example, the following metadata section specifies two separate sets of
4716 linker options, presumably to link against ``libz`` and the ``Cocoa``
4719 !0 = !{ i32 6, !"Linker Options",
4722 !{ !"-framework", !"Cocoa" } } }
4723 !llvm.module.flags = !{ !0 }
4725 The metadata encoding as lists of lists of options, as opposed to a collapsed
4726 list of options, is chosen so that the IR encoding can use multiple option
4727 strings to specify e.g., a single library, while still having that specifier be
4728 preserved as an atomic element that can be recognized by a target specific
4729 assembly writer or object file emitter.
4731 Each individual option is required to be either a valid option for the target's
4732 linker, or an option that is reserved by the target specific assembly writer or
4733 object file emitter. No other aspect of these options is defined by the IR.
4735 C type width Module Flags Metadata
4736 ----------------------------------
4738 The ARM backend emits a section into each generated object file describing the
4739 options that it was compiled with (in a compiler-independent way) to prevent
4740 linking incompatible objects, and to allow automatic library selection. Some
4741 of these options are not visible at the IR level, namely wchar_t width and enum
4744 To pass this information to the backend, these options are encoded in module
4745 flags metadata, using the following key-value pairs:
4755 - * 0 --- sizeof(wchar_t) == 4
4756 * 1 --- sizeof(wchar_t) == 2
4759 - * 0 --- Enums are at least as large as an ``int``.
4760 * 1 --- Enums are stored in the smallest integer type which can
4761 represent all of its values.
4763 For example, the following metadata section specifies that the module was
4764 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
4765 enum is the smallest type which can represent all of its values::
4767 !llvm.module.flags = !{!0, !1}
4768 !0 = !{i32 1, !"short_wchar", i32 1}
4769 !1 = !{i32 1, !"short_enum", i32 0}
4771 .. _intrinsicglobalvariables:
4773 Intrinsic Global Variables
4774 ==========================
4776 LLVM has a number of "magic" global variables that contain data that
4777 affect code generation or other IR semantics. These are documented here.
4778 All globals of this sort should have a section specified as
4779 "``llvm.metadata``". This section and all globals that start with
4780 "``llvm.``" are reserved for use by LLVM.
4784 The '``llvm.used``' Global Variable
4785 -----------------------------------
4787 The ``@llvm.used`` global is an array which has
4788 :ref:`appending linkage <linkage_appending>`. This array contains a list of
4789 pointers to named global variables, functions and aliases which may optionally
4790 have a pointer cast formed of bitcast or getelementptr. For example, a legal
4793 .. code-block:: llvm
4798 @llvm.used = appending global [2 x i8*] [
4800 i8* bitcast (i32* @Y to i8*)
4801 ], section "llvm.metadata"
4803 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
4804 and linker are required to treat the symbol as if there is a reference to the
4805 symbol that it cannot see (which is why they have to be named). For example, if
4806 a variable has internal linkage and no references other than that from the
4807 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
4808 references from inline asms and other things the compiler cannot "see", and
4809 corresponds to "``attribute((used))``" in GNU C.
4811 On some targets, the code generator must emit a directive to the
4812 assembler or object file to prevent the assembler and linker from
4813 molesting the symbol.
4815 .. _gv_llvmcompilerused:
4817 The '``llvm.compiler.used``' Global Variable
4818 --------------------------------------------
4820 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4821 directive, except that it only prevents the compiler from touching the
4822 symbol. On targets that support it, this allows an intelligent linker to
4823 optimize references to the symbol without being impeded as it would be
4826 This is a rare construct that should only be used in rare circumstances,
4827 and should not be exposed to source languages.
4829 .. _gv_llvmglobalctors:
4831 The '``llvm.global_ctors``' Global Variable
4832 -------------------------------------------
4834 .. code-block:: llvm
4836 %0 = type { i32, void ()*, i8* }
4837 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
4839 The ``@llvm.global_ctors`` array contains a list of constructor
4840 functions, priorities, and an optional associated global or function.
4841 The functions referenced by this array will be called in ascending order
4842 of priority (i.e. lowest first) when the module is loaded. The order of
4843 functions with the same priority is not defined.
4845 If the third field is present, non-null, and points to a global variable
4846 or function, the initializer function will only run if the associated
4847 data from the current module is not discarded.
4849 .. _llvmglobaldtors:
4851 The '``llvm.global_dtors``' Global Variable
4852 -------------------------------------------
4854 .. code-block:: llvm
4856 %0 = type { i32, void ()*, i8* }
4857 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
4859 The ``@llvm.global_dtors`` array contains a list of destructor
4860 functions, priorities, and an optional associated global or function.
4861 The functions referenced by this array will be called in descending
4862 order of priority (i.e. highest first) when the module is unloaded. The
4863 order of functions with the same priority is not defined.
4865 If the third field is present, non-null, and points to a global variable
4866 or function, the destructor function will only run if the associated
4867 data from the current module is not discarded.
4869 Instruction Reference
4870 =====================
4872 The LLVM instruction set consists of several different classifications
4873 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
4874 instructions <binaryops>`, :ref:`bitwise binary
4875 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
4876 :ref:`other instructions <otherops>`.
4880 Terminator Instructions
4881 -----------------------
4883 As mentioned :ref:`previously <functionstructure>`, every basic block in a
4884 program ends with a "Terminator" instruction, which indicates which
4885 block should be executed after the current block is finished. These
4886 terminator instructions typically yield a '``void``' value: they produce
4887 control flow, not values (the one exception being the
4888 ':ref:`invoke <i_invoke>`' instruction).
4890 The terminator instructions are: ':ref:`ret <i_ret>`',
4891 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
4892 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
4893 ':ref:`resume <i_resume>`', ':ref:`catchpad <i_catchpad>`',
4894 ':ref:`catchendpad <i_catchendpad>`',
4895 ':ref:`catchret <i_catchret>`',
4896 ':ref:`cleanupendpad <i_cleanupendpad>`',
4897 ':ref:`cleanupret <i_cleanupret>`',
4898 ':ref:`terminatepad <i_terminatepad>`',
4899 and ':ref:`unreachable <i_unreachable>`'.
4903 '``ret``' Instruction
4904 ^^^^^^^^^^^^^^^^^^^^^
4911 ret <type> <value> ; Return a value from a non-void function
4912 ret void ; Return from void function
4917 The '``ret``' instruction is used to return control flow (and optionally
4918 a value) from a function back to the caller.
4920 There are two forms of the '``ret``' instruction: one that returns a
4921 value and then causes control flow, and one that just causes control
4927 The '``ret``' instruction optionally accepts a single argument, the
4928 return value. The type of the return value must be a ':ref:`first
4929 class <t_firstclass>`' type.
4931 A function is not :ref:`well formed <wellformed>` if it it has a non-void
4932 return type and contains a '``ret``' instruction with no return value or
4933 a return value with a type that does not match its type, or if it has a
4934 void return type and contains a '``ret``' instruction with a return
4940 When the '``ret``' instruction is executed, control flow returns back to
4941 the calling function's context. If the caller is a
4942 ":ref:`call <i_call>`" instruction, execution continues at the
4943 instruction after the call. If the caller was an
4944 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
4945 beginning of the "normal" destination block. If the instruction returns
4946 a value, that value shall set the call or invoke instruction's return
4952 .. code-block:: llvm
4954 ret i32 5 ; Return an integer value of 5
4955 ret void ; Return from a void function
4956 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
4960 '``br``' Instruction
4961 ^^^^^^^^^^^^^^^^^^^^
4968 br i1 <cond>, label <iftrue>, label <iffalse>
4969 br label <dest> ; Unconditional branch
4974 The '``br``' instruction is used to cause control flow to transfer to a
4975 different basic block in the current function. There are two forms of
4976 this instruction, corresponding to a conditional branch and an
4977 unconditional branch.
4982 The conditional branch form of the '``br``' instruction takes a single
4983 '``i1``' value and two '``label``' values. The unconditional form of the
4984 '``br``' instruction takes a single '``label``' value as a target.
4989 Upon execution of a conditional '``br``' instruction, the '``i1``'
4990 argument is evaluated. If the value is ``true``, control flows to the
4991 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
4992 to the '``iffalse``' ``label`` argument.
4997 .. code-block:: llvm
5000 %cond = icmp eq i32 %a, %b
5001 br i1 %cond, label %IfEqual, label %IfUnequal
5009 '``switch``' Instruction
5010 ^^^^^^^^^^^^^^^^^^^^^^^^
5017 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
5022 The '``switch``' instruction is used to transfer control flow to one of
5023 several different places. It is a generalization of the '``br``'
5024 instruction, allowing a branch to occur to one of many possible
5030 The '``switch``' instruction uses three parameters: an integer
5031 comparison value '``value``', a default '``label``' destination, and an
5032 array of pairs of comparison value constants and '``label``'s. The table
5033 is not allowed to contain duplicate constant entries.
5038 The ``switch`` instruction specifies a table of values and destinations.
5039 When the '``switch``' instruction is executed, this table is searched
5040 for the given value. If the value is found, control flow is transferred
5041 to the corresponding destination; otherwise, control flow is transferred
5042 to the default destination.
5047 Depending on properties of the target machine and the particular
5048 ``switch`` instruction, this instruction may be code generated in
5049 different ways. For example, it could be generated as a series of
5050 chained conditional branches or with a lookup table.
5055 .. code-block:: llvm
5057 ; Emulate a conditional br instruction
5058 %Val = zext i1 %value to i32
5059 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
5061 ; Emulate an unconditional br instruction
5062 switch i32 0, label %dest [ ]
5064 ; Implement a jump table:
5065 switch i32 %val, label %otherwise [ i32 0, label %onzero
5067 i32 2, label %ontwo ]
5071 '``indirectbr``' Instruction
5072 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5079 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
5084 The '``indirectbr``' instruction implements an indirect branch to a
5085 label within the current function, whose address is specified by
5086 "``address``". Address must be derived from a
5087 :ref:`blockaddress <blockaddress>` constant.
5092 The '``address``' argument is the address of the label to jump to. The
5093 rest of the arguments indicate the full set of possible destinations
5094 that the address may point to. Blocks are allowed to occur multiple
5095 times in the destination list, though this isn't particularly useful.
5097 This destination list is required so that dataflow analysis has an
5098 accurate understanding of the CFG.
5103 Control transfers to the block specified in the address argument. All
5104 possible destination blocks must be listed in the label list, otherwise
5105 this instruction has undefined behavior. This implies that jumps to
5106 labels defined in other functions have undefined behavior as well.
5111 This is typically implemented with a jump through a register.
5116 .. code-block:: llvm
5118 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
5122 '``invoke``' Instruction
5123 ^^^^^^^^^^^^^^^^^^^^^^^^
5130 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
5131 [operand bundles] to label <normal label> unwind label <exception label>
5136 The '``invoke``' instruction causes control to transfer to a specified
5137 function, with the possibility of control flow transfer to either the
5138 '``normal``' label or the '``exception``' label. If the callee function
5139 returns with the "``ret``" instruction, control flow will return to the
5140 "normal" label. If the callee (or any indirect callees) returns via the
5141 ":ref:`resume <i_resume>`" instruction or other exception handling
5142 mechanism, control is interrupted and continued at the dynamically
5143 nearest "exception" label.
5145 The '``exception``' label is a `landing
5146 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
5147 '``exception``' label is required to have the
5148 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
5149 information about the behavior of the program after unwinding happens,
5150 as its first non-PHI instruction. The restrictions on the
5151 "``landingpad``" instruction's tightly couples it to the "``invoke``"
5152 instruction, so that the important information contained within the
5153 "``landingpad``" instruction can't be lost through normal code motion.
5158 This instruction requires several arguments:
5160 #. The optional "cconv" marker indicates which :ref:`calling
5161 convention <callingconv>` the call should use. If none is
5162 specified, the call defaults to using C calling conventions.
5163 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5164 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5166 #. '``ptr to function ty``': shall be the signature of the pointer to
5167 function value being invoked. In most cases, this is a direct
5168 function invocation, but indirect ``invoke``'s are just as possible,
5169 branching off an arbitrary pointer to function value.
5170 #. '``function ptr val``': An LLVM value containing a pointer to a
5171 function to be invoked.
5172 #. '``function args``': argument list whose types match the function
5173 signature argument types and parameter attributes. All arguments must
5174 be of :ref:`first class <t_firstclass>` type. If the function signature
5175 indicates the function accepts a variable number of arguments, the
5176 extra arguments can be specified.
5177 #. '``normal label``': the label reached when the called function
5178 executes a '``ret``' instruction.
5179 #. '``exception label``': the label reached when a callee returns via
5180 the :ref:`resume <i_resume>` instruction or other exception handling
5182 #. The optional :ref:`function attributes <fnattrs>` list. Only
5183 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5184 attributes are valid here.
5185 #. The optional :ref:`operand bundles <opbundles>` list.
5190 This instruction is designed to operate as a standard '``call``'
5191 instruction in most regards. The primary difference is that it
5192 establishes an association with a label, which is used by the runtime
5193 library to unwind the stack.
5195 This instruction is used in languages with destructors to ensure that
5196 proper cleanup is performed in the case of either a ``longjmp`` or a
5197 thrown exception. Additionally, this is important for implementation of
5198 '``catch``' clauses in high-level languages that support them.
5200 For the purposes of the SSA form, the definition of the value returned
5201 by the '``invoke``' instruction is deemed to occur on the edge from the
5202 current block to the "normal" label. If the callee unwinds then no
5203 return value is available.
5208 .. code-block:: llvm
5210 %retval = invoke i32 @Test(i32 15) to label %Continue
5211 unwind label %TestCleanup ; i32:retval set
5212 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
5213 unwind label %TestCleanup ; i32:retval set
5217 '``resume``' Instruction
5218 ^^^^^^^^^^^^^^^^^^^^^^^^
5225 resume <type> <value>
5230 The '``resume``' instruction is a terminator instruction that has no
5236 The '``resume``' instruction requires one argument, which must have the
5237 same type as the result of any '``landingpad``' instruction in the same
5243 The '``resume``' instruction resumes propagation of an existing
5244 (in-flight) exception whose unwinding was interrupted with a
5245 :ref:`landingpad <i_landingpad>` instruction.
5250 .. code-block:: llvm
5252 resume { i8*, i32 } %exn
5256 '``catchpad``' Instruction
5257 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5264 <resultval> = catchpad [<args>*]
5265 to label <normal label> unwind label <exception label>
5270 The '``catchpad``' instruction is used by `LLVM's exception handling
5271 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5272 is a catch block --- one where a personality routine attempts to transfer
5273 control to catch an exception.
5274 The ``args`` correspond to whatever information the personality
5275 routine requires to know if this is an appropriate place to catch the
5276 exception. Control is transfered to the ``exception`` label if the
5277 ``catchpad`` is not an appropriate handler for the in-flight exception.
5278 The ``normal`` label should contain the code found in the ``catch``
5279 portion of a ``try``/``catch`` sequence. The ``resultval`` has the type
5280 :ref:`token <t_token>` and is used to match the ``catchpad`` to
5281 corresponding :ref:`catchrets <i_catchret>`.
5286 The instruction takes a list of arbitrary values which are interpreted
5287 by the :ref:`personality function <personalityfn>`.
5289 The ``catchpad`` must be provided a ``normal`` label to transfer control
5290 to if the ``catchpad`` matches the exception and an ``exception``
5291 label to transfer control to if it doesn't.
5296 When the call stack is being unwound due to an exception being thrown,
5297 the exception is compared against the ``args``. If it doesn't match,
5298 then control is transfered to the ``exception`` basic block.
5299 As with calling conventions, how the personality function results are
5300 represented in LLVM IR is target specific.
5302 The ``catchpad`` instruction has several restrictions:
5304 - A catch block is a basic block which is the unwind destination of
5305 an exceptional instruction.
5306 - A catch block must have a '``catchpad``' instruction as its
5307 first non-PHI instruction.
5308 - A catch block's ``exception`` edge must refer to a catch block or a
5310 - There can be only one '``catchpad``' instruction within the
5312 - A basic block that is not a catch block may not include a
5313 '``catchpad``' instruction.
5314 - A catch block which has another catch block as a predecessor may not have
5315 any other predecessors.
5316 - It is undefined behavior for control to transfer from a ``catchpad`` to a
5317 ``ret`` without first executing a ``catchret`` that consumes the
5318 ``catchpad`` or unwinding through its ``catchendpad``.
5319 - It is undefined behavior for control to transfer from a ``catchpad`` to
5320 itself without first executing a ``catchret`` that consumes the
5321 ``catchpad`` or unwinding through its ``catchendpad``.
5326 .. code-block:: llvm
5328 ;; A catch block which can catch an integer.
5329 %tok = catchpad [i8** @_ZTIi]
5330 to label %int.handler unwind label %terminate
5334 '``catchendpad``' Instruction
5335 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5342 catchendpad unwind label <nextaction>
5343 catchendpad unwind to caller
5348 The '``catchendpad``' instruction is used by `LLVM's exception handling
5349 system <ExceptionHandling.html#overview>`_ to communicate to the
5350 :ref:`personality function <personalityfn>` which invokes are associated
5351 with a chain of :ref:`catchpad <i_catchpad>` instructions; propagating an
5352 exception out of a catch handler is represented by unwinding through its
5353 ``catchendpad``. Unwinding to the outer scope when a chain of catch handlers
5354 do not handle an exception is also represented by unwinding through their
5357 The ``nextaction`` label indicates where control should transfer to if
5358 none of the ``catchpad`` instructions are suitable for catching the
5359 in-flight exception.
5361 If a ``nextaction`` label is not present, the instruction unwinds out of
5362 its parent function. The
5363 :ref:`personality function <personalityfn>` will continue processing
5364 exception handling actions in the caller.
5369 The instruction optionally takes a label, ``nextaction``, indicating
5370 where control should transfer to if none of the preceding
5371 ``catchpad`` instructions are suitable for the in-flight exception.
5376 When the call stack is being unwound due to an exception being thrown
5377 and none of the constituent ``catchpad`` instructions match, then
5378 control is transfered to ``nextaction`` if it is present. If it is not
5379 present, control is transfered to the caller.
5381 The ``catchendpad`` instruction has several restrictions:
5383 - A catch-end block is a basic block which is the unwind destination of
5384 an exceptional instruction.
5385 - A catch-end block must have a '``catchendpad``' instruction as its
5386 first non-PHI instruction.
5387 - There can be only one '``catchendpad``' instruction within the
5389 - A basic block that is not a catch-end block may not include a
5390 '``catchendpad``' instruction.
5391 - Exactly one catch block may unwind to a ``catchendpad``.
5392 - It is undefined behavior to execute a ``catchendpad`` if none of the
5393 '``catchpad``'s chained to it have been executed.
5394 - It is undefined behavior to execute a ``catchendpad`` twice without an
5395 intervening execution of one or more of the '``catchpad``'s chained to it.
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, some ``catchret`` consuming that ``catchpad`` has already been
5400 - It is undefined behavior to execute a ``catchendpad`` if, after the most
5401 recent execution of the normal successor edge of any ``catchpad`` chained
5402 to it, any other ``catchpad`` or ``cleanuppad`` has been executed but has
5403 not had a corresponding
5404 ``catchret``/``cleanupret``/``catchendpad``/``cleanupendpad`` executed.
5409 .. code-block:: llvm
5411 catchendpad unwind label %terminate
5412 catchendpad unwind to caller
5416 '``catchret``' Instruction
5417 ^^^^^^^^^^^^^^^^^^^^^^^^^^
5424 catchret <value> to label <normal>
5429 The '``catchret``' instruction is a terminator instruction that has a
5436 The first argument to a '``catchret``' indicates which ``catchpad`` it
5437 exits. It must be a :ref:`catchpad <i_catchpad>`.
5438 The second argument to a '``catchret``' specifies where control will
5444 The '``catchret``' instruction ends the existing (in-flight) exception
5445 whose unwinding was interrupted with a
5446 :ref:`catchpad <i_catchpad>` instruction.
5447 The :ref:`personality function <personalityfn>` gets a chance to execute
5448 arbitrary code to, for example, run a C++ destructor.
5449 Control then transfers to ``normal``.
5450 It may be passed an optional, personality specific, value.
5452 It is undefined behavior to execute a ``catchret`` whose ``catchpad`` has
5455 It is undefined behavior to execute a ``catchret`` if, after the most recent
5456 execution of its ``catchpad``, some ``catchret`` or ``catchendpad`` linked
5457 to the same ``catchpad`` has already been executed.
5459 It is undefined behavior to execute a ``catchret`` if, after the most recent
5460 execution of its ``catchpad``, any other ``catchpad`` or ``cleanuppad`` has
5461 been executed but has not had a corresponding
5462 ``catchret``/``cleanupret``/``catchendpad``/``cleanupendpad`` executed.
5467 .. code-block:: llvm
5469 catchret %catch label %continue
5471 .. _i_cleanupendpad:
5473 '``cleanupendpad``' Instruction
5474 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5481 cleanupendpad <value> unwind label <nextaction>
5482 cleanupendpad <value> unwind to caller
5487 The '``cleanupendpad``' instruction is used by `LLVM's exception handling
5488 system <ExceptionHandling.html#overview>`_ to communicate to the
5489 :ref:`personality function <personalityfn>` which invokes are associated
5490 with a :ref:`cleanuppad <i_cleanuppad>` instructions; propagating an exception
5491 out of a cleanup is represented by unwinding through its ``cleanupendpad``.
5493 The ``nextaction`` label indicates where control should unwind to next, in the
5494 event that a cleanup is exited by means of an(other) exception being raised.
5496 If a ``nextaction`` label is not present, the instruction unwinds out of
5497 its parent function. The
5498 :ref:`personality function <personalityfn>` will continue processing
5499 exception handling actions in the caller.
5504 The '``cleanupendpad``' instruction requires one argument, which indicates
5505 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
5506 It also has an optional successor, ``nextaction``, indicating where control
5512 When and exception propagates to a ``cleanupendpad``, control is transfered to
5513 ``nextaction`` if it is present. If it is not present, control is transfered to
5516 The ``cleanupendpad`` instruction has several restrictions:
5518 - A cleanup-end block is a basic block which is the unwind destination of
5519 an exceptional instruction.
5520 - A cleanup-end block must have a '``cleanupendpad``' instruction as its
5521 first non-PHI instruction.
5522 - There can be only one '``cleanupendpad``' instruction within the
5524 - A basic block that is not a cleanup-end block may not include a
5525 '``cleanupendpad``' instruction.
5526 - It is undefined behavior to execute a ``cleanupendpad`` whose ``cleanuppad``
5527 has not been executed.
5528 - It is undefined behavior to execute a ``cleanupendpad`` if, after the most
5529 recent execution of its ``cleanuppad``, some ``cleanupret`` or ``cleanupendpad``
5530 consuming the same ``cleanuppad`` has already been executed.
5531 - It is undefined behavior to execute a ``cleanupendpad`` if, after the most
5532 recent execution of its ``cleanuppad``, any other ``cleanuppad`` or
5533 ``catchpad`` has been executed but has not had a corresponding
5534 ``cleanupret``/``catchret``/``cleanupendpad``/``catchendpad`` executed.
5539 .. code-block:: llvm
5541 cleanupendpad %cleanup unwind label %terminate
5542 cleanupendpad %cleanup unwind to caller
5546 '``cleanupret``' Instruction
5547 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5554 cleanupret <value> unwind label <continue>
5555 cleanupret <value> unwind to caller
5560 The '``cleanupret``' instruction is a terminator instruction that has
5561 an optional successor.
5567 The '``cleanupret``' instruction requires one argument, which indicates
5568 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
5569 It also has an optional successor, ``continue``.
5574 The '``cleanupret``' instruction indicates to the
5575 :ref:`personality function <personalityfn>` that one
5576 :ref:`cleanuppad <i_cleanuppad>` it transferred control to has ended.
5577 It transfers control to ``continue`` or unwinds out of the function.
5579 It is undefined behavior to execute a ``cleanupret`` whose ``cleanuppad`` has
5582 It is undefined behavior to execute a ``cleanupret`` if, after the most recent
5583 execution of its ``cleanuppad``, some ``cleanupret`` or ``cleanupendpad``
5584 consuming the same ``cleanuppad`` has already been executed.
5586 It is undefined behavior to execute a ``cleanupret`` if, after the most recent
5587 execution of its ``cleanuppad``, any other ``cleanuppad`` or ``catchpad`` has
5588 been executed but has not had a corresponding
5589 ``cleanupret``/``catchret``/``cleanupendpad``/``catchendpad`` executed.
5594 .. code-block:: llvm
5596 cleanupret %cleanup unwind to caller
5597 cleanupret %cleanup unwind label %continue
5601 '``terminatepad``' Instruction
5602 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5609 terminatepad [<args>*] unwind label <exception label>
5610 terminatepad [<args>*] unwind to caller
5615 The '``terminatepad``' instruction is used by `LLVM's exception handling
5616 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5617 is a terminate block --- one where a personality routine may decide to
5618 terminate the program.
5619 The ``args`` correspond to whatever information the personality
5620 routine requires to know if this is an appropriate place to terminate the
5621 program. Control is transferred to the ``exception`` label if the
5622 personality routine decides not to terminate the program for the
5623 in-flight exception.
5628 The instruction takes a list of arbitrary values which are interpreted
5629 by the :ref:`personality function <personalityfn>`.
5631 The ``terminatepad`` may be given an ``exception`` label to
5632 transfer control to if the in-flight exception matches the ``args``.
5637 When the call stack is being unwound due to an exception being thrown,
5638 the exception is compared against the ``args``. If it matches,
5639 then control is transfered to the ``exception`` basic block. Otherwise,
5640 the program is terminated via personality-specific means. Typically,
5641 the first argument to ``terminatepad`` specifies what function the
5642 personality should defer to in order to terminate the program.
5644 The ``terminatepad`` instruction has several restrictions:
5646 - A terminate block is a basic block which is the unwind destination of
5647 an exceptional instruction.
5648 - A terminate block must have a '``terminatepad``' instruction as its
5649 first non-PHI instruction.
5650 - There can be only one '``terminatepad``' instruction within the
5652 - A basic block that is not a terminate block may not include a
5653 '``terminatepad``' instruction.
5658 .. code-block:: llvm
5660 ;; A terminate block which only permits integers.
5661 terminatepad [i8** @_ZTIi] unwind label %continue
5665 '``unreachable``' Instruction
5666 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5678 The '``unreachable``' instruction has no defined semantics. This
5679 instruction is used to inform the optimizer that a particular portion of
5680 the code is not reachable. This can be used to indicate that the code
5681 after a no-return function cannot be reached, and other facts.
5686 The '``unreachable``' instruction has no defined semantics.
5693 Binary operators are used to do most of the computation in a program.
5694 They require two operands of the same type, execute an operation on
5695 them, and produce a single value. The operands might represent multiple
5696 data, as is the case with the :ref:`vector <t_vector>` data type. The
5697 result value has the same type as its operands.
5699 There are several different binary operators:
5703 '``add``' Instruction
5704 ^^^^^^^^^^^^^^^^^^^^^
5711 <result> = add <ty> <op1>, <op2> ; yields ty:result
5712 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
5713 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
5714 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
5719 The '``add``' instruction returns the sum of its two operands.
5724 The two arguments to the '``add``' instruction must be
5725 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5726 arguments must have identical types.
5731 The value produced is the integer sum of the two operands.
5733 If the sum has unsigned overflow, the result returned is the
5734 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5737 Because LLVM integers use a two's complement representation, this
5738 instruction is appropriate for both signed and unsigned integers.
5740 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5741 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5742 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
5743 unsigned and/or signed overflow, respectively, occurs.
5748 .. code-block:: llvm
5750 <result> = add i32 4, %var ; yields i32:result = 4 + %var
5754 '``fadd``' Instruction
5755 ^^^^^^^^^^^^^^^^^^^^^^
5762 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5767 The '``fadd``' instruction returns the sum of its two operands.
5772 The two arguments to the '``fadd``' instruction must be :ref:`floating
5773 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5774 Both arguments must have identical types.
5779 The value produced is the floating point sum of the two operands. This
5780 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
5781 which are optimization hints to enable otherwise unsafe floating point
5787 .. code-block:: llvm
5789 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
5791 '``sub``' Instruction
5792 ^^^^^^^^^^^^^^^^^^^^^
5799 <result> = sub <ty> <op1>, <op2> ; yields ty:result
5800 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
5801 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
5802 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
5807 The '``sub``' instruction returns the difference of its two operands.
5809 Note that the '``sub``' instruction is used to represent the '``neg``'
5810 instruction present in most other intermediate representations.
5815 The two arguments to the '``sub``' instruction must be
5816 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5817 arguments must have identical types.
5822 The value produced is the integer difference of the two operands.
5824 If the difference has unsigned overflow, the result returned is the
5825 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5828 Because LLVM integers use a two's complement representation, this
5829 instruction is appropriate for both signed and unsigned integers.
5831 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5832 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5833 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
5834 unsigned and/or signed overflow, respectively, occurs.
5839 .. code-block:: llvm
5841 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
5842 <result> = sub i32 0, %val ; yields i32:result = -%var
5846 '``fsub``' Instruction
5847 ^^^^^^^^^^^^^^^^^^^^^^
5854 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5859 The '``fsub``' instruction returns the difference of its two operands.
5861 Note that the '``fsub``' instruction is used to represent the '``fneg``'
5862 instruction present in most other intermediate representations.
5867 The two arguments to the '``fsub``' instruction must be :ref:`floating
5868 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5869 Both arguments must have identical types.
5874 The value produced is the floating point difference of the two operands.
5875 This instruction can also take any number of :ref:`fast-math
5876 flags <fastmath>`, which are optimization hints to enable otherwise
5877 unsafe floating point optimizations:
5882 .. code-block:: llvm
5884 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
5885 <result> = fsub float -0.0, %val ; yields float:result = -%var
5887 '``mul``' Instruction
5888 ^^^^^^^^^^^^^^^^^^^^^
5895 <result> = mul <ty> <op1>, <op2> ; yields ty:result
5896 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
5897 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
5898 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
5903 The '``mul``' instruction returns the product of its two operands.
5908 The two arguments to the '``mul``' instruction must be
5909 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5910 arguments must have identical types.
5915 The value produced is the integer product of the two operands.
5917 If the result of the multiplication has unsigned overflow, the result
5918 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
5919 bit width of the result.
5921 Because LLVM integers use a two's complement representation, and the
5922 result is the same width as the operands, this instruction returns the
5923 correct result for both signed and unsigned integers. If a full product
5924 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
5925 sign-extended or zero-extended as appropriate to the width of the full
5928 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5929 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5930 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
5931 unsigned and/or signed overflow, respectively, occurs.
5936 .. code-block:: llvm
5938 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
5942 '``fmul``' Instruction
5943 ^^^^^^^^^^^^^^^^^^^^^^
5950 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5955 The '``fmul``' instruction returns the product of its two operands.
5960 The two arguments to the '``fmul``' instruction must be :ref:`floating
5961 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5962 Both arguments must have identical types.
5967 The value produced is the floating point product of the two operands.
5968 This instruction can also take any number of :ref:`fast-math
5969 flags <fastmath>`, which are optimization hints to enable otherwise
5970 unsafe floating point optimizations:
5975 .. code-block:: llvm
5977 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
5979 '``udiv``' Instruction
5980 ^^^^^^^^^^^^^^^^^^^^^^
5987 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
5988 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
5993 The '``udiv``' instruction returns the quotient of its two operands.
5998 The two arguments to the '``udiv``' instruction must be
5999 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6000 arguments must have identical types.
6005 The value produced is the unsigned integer quotient of the two operands.
6007 Note that unsigned integer division and signed integer division are
6008 distinct operations; for signed integer division, use '``sdiv``'.
6010 Division by zero leads to undefined behavior.
6012 If the ``exact`` keyword is present, the result value of the ``udiv`` is
6013 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
6014 such, "((a udiv exact b) mul b) == a").
6019 .. code-block:: llvm
6021 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
6023 '``sdiv``' Instruction
6024 ^^^^^^^^^^^^^^^^^^^^^^
6031 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
6032 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
6037 The '``sdiv``' instruction returns the quotient of its two operands.
6042 The two arguments to the '``sdiv``' instruction must be
6043 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6044 arguments must have identical types.
6049 The value produced is the signed integer quotient of the two operands
6050 rounded towards zero.
6052 Note that signed integer division and unsigned integer division are
6053 distinct operations; for unsigned integer division, use '``udiv``'.
6055 Division by zero leads to undefined behavior. Overflow also leads to
6056 undefined behavior; this is a rare case, but can occur, for example, by
6057 doing a 32-bit division of -2147483648 by -1.
6059 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
6060 a :ref:`poison value <poisonvalues>` if the result would be rounded.
6065 .. code-block:: llvm
6067 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
6071 '``fdiv``' Instruction
6072 ^^^^^^^^^^^^^^^^^^^^^^
6079 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6084 The '``fdiv``' instruction returns the quotient of its two operands.
6089 The two arguments to the '``fdiv``' instruction must be :ref:`floating
6090 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6091 Both arguments must have identical types.
6096 The value produced is the floating point quotient of the two operands.
6097 This instruction can also take any number of :ref:`fast-math
6098 flags <fastmath>`, which are optimization hints to enable otherwise
6099 unsafe floating point optimizations:
6104 .. code-block:: llvm
6106 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
6108 '``urem``' Instruction
6109 ^^^^^^^^^^^^^^^^^^^^^^
6116 <result> = urem <ty> <op1>, <op2> ; yields ty:result
6121 The '``urem``' instruction returns the remainder from the unsigned
6122 division of its two arguments.
6127 The two arguments to the '``urem``' instruction must be
6128 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6129 arguments must have identical types.
6134 This instruction returns the unsigned integer *remainder* of a division.
6135 This instruction always performs an unsigned division to get the
6138 Note that unsigned integer remainder and signed integer remainder are
6139 distinct operations; for signed integer remainder, use '``srem``'.
6141 Taking the remainder of a division by zero leads to undefined behavior.
6146 .. code-block:: llvm
6148 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
6150 '``srem``' Instruction
6151 ^^^^^^^^^^^^^^^^^^^^^^
6158 <result> = srem <ty> <op1>, <op2> ; yields ty:result
6163 The '``srem``' instruction returns the remainder from the signed
6164 division of its two operands. This instruction can also take
6165 :ref:`vector <t_vector>` versions of the values in which case the elements
6171 The two arguments to the '``srem``' instruction must be
6172 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6173 arguments must have identical types.
6178 This instruction returns the *remainder* of a division (where the result
6179 is either zero or has the same sign as the dividend, ``op1``), not the
6180 *modulo* operator (where the result is either zero or has the same sign
6181 as the divisor, ``op2``) of a value. For more information about the
6182 difference, see `The Math
6183 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
6184 table of how this is implemented in various languages, please see
6186 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
6188 Note that signed integer remainder and unsigned integer remainder are
6189 distinct operations; for unsigned integer remainder, use '``urem``'.
6191 Taking the remainder of a division by zero leads to undefined behavior.
6192 Overflow also leads to undefined behavior; this is a rare case, but can
6193 occur, for example, by taking the remainder of a 32-bit division of
6194 -2147483648 by -1. (The remainder doesn't actually overflow, but this
6195 rule lets srem be implemented using instructions that return both the
6196 result of the division and the remainder.)
6201 .. code-block:: llvm
6203 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
6207 '``frem``' Instruction
6208 ^^^^^^^^^^^^^^^^^^^^^^
6215 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6220 The '``frem``' instruction returns the remainder from the division of
6226 The two arguments to the '``frem``' instruction must be :ref:`floating
6227 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6228 Both arguments must have identical types.
6233 This instruction returns the *remainder* of a division. The remainder
6234 has the same sign as the dividend. This instruction can also take any
6235 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
6236 to enable otherwise unsafe floating point optimizations:
6241 .. code-block:: llvm
6243 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
6247 Bitwise Binary Operations
6248 -------------------------
6250 Bitwise binary operators are used to do various forms of bit-twiddling
6251 in a program. They are generally very efficient instructions and can
6252 commonly be strength reduced from other instructions. They require two
6253 operands of the same type, execute an operation on them, and produce a
6254 single value. The resulting value is the same type as its operands.
6256 '``shl``' Instruction
6257 ^^^^^^^^^^^^^^^^^^^^^
6264 <result> = shl <ty> <op1>, <op2> ; yields ty:result
6265 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
6266 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
6267 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
6272 The '``shl``' instruction returns the first operand shifted to the left
6273 a specified number of bits.
6278 Both arguments to the '``shl``' instruction must be the same
6279 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6280 '``op2``' is treated as an unsigned value.
6285 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
6286 where ``n`` is the width of the result. If ``op2`` is (statically or
6287 dynamically) equal to or larger than the number of bits in
6288 ``op1``, the result is undefined. If the arguments are vectors, each
6289 vector element of ``op1`` is shifted by the corresponding shift amount
6292 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
6293 value <poisonvalues>` if it shifts out any non-zero bits. If the
6294 ``nsw`` keyword is present, then the shift produces a :ref:`poison
6295 value <poisonvalues>` if it shifts out any bits that disagree with the
6296 resultant sign bit. As such, NUW/NSW have the same semantics as they
6297 would if the shift were expressed as a mul instruction with the same
6298 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
6303 .. code-block:: llvm
6305 <result> = shl i32 4, %var ; yields i32: 4 << %var
6306 <result> = shl i32 4, 2 ; yields i32: 16
6307 <result> = shl i32 1, 10 ; yields i32: 1024
6308 <result> = shl i32 1, 32 ; undefined
6309 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
6311 '``lshr``' Instruction
6312 ^^^^^^^^^^^^^^^^^^^^^^
6319 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
6320 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
6325 The '``lshr``' instruction (logical shift right) returns the first
6326 operand shifted to the right a specified number of bits with zero fill.
6331 Both arguments to the '``lshr``' instruction must be the same
6332 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6333 '``op2``' is treated as an unsigned value.
6338 This instruction always performs a logical shift right operation. The
6339 most significant bits of the result will be filled with zero bits after
6340 the shift. If ``op2`` is (statically or dynamically) equal to or larger
6341 than the number of bits in ``op1``, the result is undefined. If the
6342 arguments are vectors, each vector element of ``op1`` is shifted by the
6343 corresponding shift amount in ``op2``.
6345 If the ``exact`` keyword is present, the result value of the ``lshr`` is
6346 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6352 .. code-block:: llvm
6354 <result> = lshr i32 4, 1 ; yields i32:result = 2
6355 <result> = lshr i32 4, 2 ; yields i32:result = 1
6356 <result> = lshr i8 4, 3 ; yields i8:result = 0
6357 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
6358 <result> = lshr i32 1, 32 ; undefined
6359 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
6361 '``ashr``' Instruction
6362 ^^^^^^^^^^^^^^^^^^^^^^
6369 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
6370 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
6375 The '``ashr``' instruction (arithmetic shift right) returns the first
6376 operand shifted to the right a specified number of bits with sign
6382 Both arguments to the '``ashr``' instruction must be the same
6383 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6384 '``op2``' is treated as an unsigned value.
6389 This instruction always performs an arithmetic shift right operation,
6390 The most significant bits of the result will be filled with the sign bit
6391 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
6392 than the number of bits in ``op1``, the result is undefined. If the
6393 arguments are vectors, each vector element of ``op1`` is shifted by the
6394 corresponding shift amount in ``op2``.
6396 If the ``exact`` keyword is present, the result value of the ``ashr`` is
6397 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6403 .. code-block:: llvm
6405 <result> = ashr i32 4, 1 ; yields i32:result = 2
6406 <result> = ashr i32 4, 2 ; yields i32:result = 1
6407 <result> = ashr i8 4, 3 ; yields i8:result = 0
6408 <result> = ashr i8 -2, 1 ; yields i8:result = -1
6409 <result> = ashr i32 1, 32 ; undefined
6410 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
6412 '``and``' Instruction
6413 ^^^^^^^^^^^^^^^^^^^^^
6420 <result> = and <ty> <op1>, <op2> ; yields ty:result
6425 The '``and``' instruction returns the bitwise logical and of its two
6431 The two arguments to the '``and``' instruction must be
6432 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6433 arguments must have identical types.
6438 The truth table used for the '``and``' instruction is:
6455 .. code-block:: llvm
6457 <result> = and i32 4, %var ; yields i32:result = 4 & %var
6458 <result> = and i32 15, 40 ; yields i32:result = 8
6459 <result> = and i32 4, 8 ; yields i32:result = 0
6461 '``or``' Instruction
6462 ^^^^^^^^^^^^^^^^^^^^
6469 <result> = or <ty> <op1>, <op2> ; yields ty:result
6474 The '``or``' instruction returns the bitwise logical inclusive or of its
6480 The two arguments to the '``or``' instruction must be
6481 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6482 arguments must have identical types.
6487 The truth table used for the '``or``' instruction is:
6506 <result> = or i32 4, %var ; yields i32:result = 4 | %var
6507 <result> = or i32 15, 40 ; yields i32:result = 47
6508 <result> = or i32 4, 8 ; yields i32:result = 12
6510 '``xor``' Instruction
6511 ^^^^^^^^^^^^^^^^^^^^^
6518 <result> = xor <ty> <op1>, <op2> ; yields ty:result
6523 The '``xor``' instruction returns the bitwise logical exclusive or of
6524 its two operands. The ``xor`` is used to implement the "one's
6525 complement" operation, which is the "~" operator in C.
6530 The two arguments to the '``xor``' instruction must be
6531 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6532 arguments must have identical types.
6537 The truth table used for the '``xor``' instruction is:
6554 .. code-block:: llvm
6556 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
6557 <result> = xor i32 15, 40 ; yields i32:result = 39
6558 <result> = xor i32 4, 8 ; yields i32:result = 12
6559 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
6564 LLVM supports several instructions to represent vector operations in a
6565 target-independent manner. These instructions cover the element-access
6566 and vector-specific operations needed to process vectors effectively.
6567 While LLVM does directly support these vector operations, many
6568 sophisticated algorithms will want to use target-specific intrinsics to
6569 take full advantage of a specific target.
6571 .. _i_extractelement:
6573 '``extractelement``' Instruction
6574 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6581 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
6586 The '``extractelement``' instruction extracts a single scalar element
6587 from a vector at a specified index.
6592 The first operand of an '``extractelement``' instruction is a value of
6593 :ref:`vector <t_vector>` type. The second operand is an index indicating
6594 the position from which to extract the element. The index may be a
6595 variable of any integer type.
6600 The result is a scalar of the same type as the element type of ``val``.
6601 Its value is the value at position ``idx`` of ``val``. If ``idx``
6602 exceeds the length of ``val``, the results are undefined.
6607 .. code-block:: llvm
6609 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
6611 .. _i_insertelement:
6613 '``insertelement``' Instruction
6614 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6621 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
6626 The '``insertelement``' instruction inserts a scalar element into a
6627 vector at a specified index.
6632 The first operand of an '``insertelement``' instruction is a value of
6633 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
6634 type must equal the element type of the first operand. The third operand
6635 is an index indicating the position at which to insert the value. The
6636 index may be a variable of any integer type.
6641 The result is a vector of the same type as ``val``. Its element values
6642 are those of ``val`` except at position ``idx``, where it gets the value
6643 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
6649 .. code-block:: llvm
6651 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
6653 .. _i_shufflevector:
6655 '``shufflevector``' Instruction
6656 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6663 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
6668 The '``shufflevector``' instruction constructs a permutation of elements
6669 from two input vectors, returning a vector with the same element type as
6670 the input and length that is the same as the shuffle mask.
6675 The first two operands of a '``shufflevector``' instruction are vectors
6676 with the same type. The third argument is a shuffle mask whose element
6677 type is always 'i32'. The result of the instruction is a vector whose
6678 length is the same as the shuffle mask and whose element type is the
6679 same as the element type of the first two operands.
6681 The shuffle mask operand is required to be a constant vector with either
6682 constant integer or undef values.
6687 The elements of the two input vectors are numbered from left to right
6688 across both of the vectors. The shuffle mask operand specifies, for each
6689 element of the result vector, which element of the two input vectors the
6690 result element gets. The element selector may be undef (meaning "don't
6691 care") and the second operand may be undef if performing a shuffle from
6697 .. code-block:: llvm
6699 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6700 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
6701 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
6702 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
6703 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
6704 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
6705 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6706 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
6708 Aggregate Operations
6709 --------------------
6711 LLVM supports several instructions for working with
6712 :ref:`aggregate <t_aggregate>` values.
6716 '``extractvalue``' Instruction
6717 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6724 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
6729 The '``extractvalue``' instruction extracts the value of a member field
6730 from an :ref:`aggregate <t_aggregate>` value.
6735 The first operand of an '``extractvalue``' instruction is a value of
6736 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The other operands are
6737 constant indices to specify which value to extract in a similar manner
6738 as indices in a '``getelementptr``' instruction.
6740 The major differences to ``getelementptr`` indexing are:
6742 - Since the value being indexed is not a pointer, the first index is
6743 omitted and assumed to be zero.
6744 - At least one index must be specified.
6745 - Not only struct indices but also array indices must be in bounds.
6750 The result is the value at the position in the aggregate specified by
6756 .. code-block:: llvm
6758 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
6762 '``insertvalue``' Instruction
6763 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6770 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
6775 The '``insertvalue``' instruction inserts a value into a member field in
6776 an :ref:`aggregate <t_aggregate>` value.
6781 The first operand of an '``insertvalue``' instruction is a value of
6782 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
6783 a first-class value to insert. The following operands are constant
6784 indices indicating the position at which to insert the value in a
6785 similar manner as indices in a '``extractvalue``' instruction. The value
6786 to insert must have the same type as the value identified by the
6792 The result is an aggregate of the same type as ``val``. Its value is
6793 that of ``val`` except that the value at the position specified by the
6794 indices is that of ``elt``.
6799 .. code-block:: llvm
6801 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
6802 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
6803 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
6807 Memory Access and Addressing Operations
6808 ---------------------------------------
6810 A key design point of an SSA-based representation is how it represents
6811 memory. In LLVM, no memory locations are in SSA form, which makes things
6812 very simple. This section describes how to read, write, and allocate
6817 '``alloca``' Instruction
6818 ^^^^^^^^^^^^^^^^^^^^^^^^
6825 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
6830 The '``alloca``' instruction allocates memory on the stack frame of the
6831 currently executing function, to be automatically released when this
6832 function returns to its caller. The object is always allocated in the
6833 generic address space (address space zero).
6838 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
6839 bytes of memory on the runtime stack, returning a pointer of the
6840 appropriate type to the program. If "NumElements" is specified, it is
6841 the number of elements allocated, otherwise "NumElements" is defaulted
6842 to be one. If a constant alignment is specified, the value result of the
6843 allocation is guaranteed to be aligned to at least that boundary. The
6844 alignment may not be greater than ``1 << 29``. If not specified, or if
6845 zero, the target can choose to align the allocation on any convenient
6846 boundary compatible with the type.
6848 '``type``' may be any sized type.
6853 Memory is allocated; a pointer is returned. The operation is undefined
6854 if there is insufficient stack space for the allocation. '``alloca``'d
6855 memory is automatically released when the function returns. The
6856 '``alloca``' instruction is commonly used to represent automatic
6857 variables that must have an address available. When the function returns
6858 (either with the ``ret`` or ``resume`` instructions), the memory is
6859 reclaimed. Allocating zero bytes is legal, but the result is undefined.
6860 The order in which memory is allocated (ie., which way the stack grows)
6866 .. code-block:: llvm
6868 %ptr = alloca i32 ; yields i32*:ptr
6869 %ptr = alloca i32, i32 4 ; yields i32*:ptr
6870 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
6871 %ptr = alloca i32, align 1024 ; yields i32*:ptr
6875 '``load``' Instruction
6876 ^^^^^^^^^^^^^^^^^^^^^^
6883 <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>]
6884 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>]
6885 !<index> = !{ i32 1 }
6886 !<deref_bytes_node> = !{i64 <dereferenceable_bytes>}
6887 !<align_node> = !{ i64 <value_alignment> }
6892 The '``load``' instruction is used to read from memory.
6897 The argument to the ``load`` instruction specifies the memory address
6898 from which to load. The type specified must be a :ref:`first
6899 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
6900 then the optimizer is not allowed to modify the number or order of
6901 execution of this ``load`` with other :ref:`volatile
6902 operations <volatile>`.
6904 If the ``load`` is marked as ``atomic``, it takes an extra
6905 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
6906 ``release`` and ``acq_rel`` orderings are not valid on ``load``
6907 instructions. Atomic loads produce :ref:`defined <memmodel>` results
6908 when they may see multiple atomic stores. The type of the pointee must
6909 be an integer type whose bit width is a power of two greater than or
6910 equal to eight and less than or equal to a target-specific size limit.
6911 ``align`` must be explicitly specified on atomic loads, and the load has
6912 undefined behavior if the alignment is not set to a value which is at
6913 least the size in bytes of the pointee. ``!nontemporal`` does not have
6914 any defined semantics for atomic loads.
6916 The optional constant ``align`` argument specifies the alignment of the
6917 operation (that is, the alignment of the memory address). A value of 0
6918 or an omitted ``align`` argument means that the operation has the ABI
6919 alignment for the target. It is the responsibility of the code emitter
6920 to ensure that the alignment information is correct. Overestimating the
6921 alignment results in undefined behavior. Underestimating the alignment
6922 may produce less efficient code. An alignment of 1 is always safe. The
6923 maximum possible alignment is ``1 << 29``.
6925 The optional ``!nontemporal`` metadata must reference a single
6926 metadata name ``<index>`` corresponding to a metadata node with one
6927 ``i32`` entry of value 1. The existence of the ``!nontemporal``
6928 metadata on the instruction tells the optimizer and code generator
6929 that this load is not expected to be reused in the cache. The code
6930 generator may select special instructions to save cache bandwidth, such
6931 as the ``MOVNT`` instruction on x86.
6933 The optional ``!invariant.load`` metadata must reference a single
6934 metadata name ``<index>`` corresponding to a metadata node with no
6935 entries. The existence of the ``!invariant.load`` metadata on the
6936 instruction tells the optimizer and code generator that the address
6937 operand to this load points to memory which can be assumed unchanged.
6938 Being invariant does not imply that a location is dereferenceable,
6939 but it does imply that once the location is known dereferenceable
6940 its value is henceforth unchanging.
6942 The optional ``!invariant.group`` metadata must reference a single metadata name
6943 ``<index>`` corresponding to a metadata node. See ``invariant.group`` metadata.
6945 The optional ``!nonnull`` metadata must reference a single
6946 metadata name ``<index>`` corresponding to a metadata node with no
6947 entries. The existence of the ``!nonnull`` metadata on the
6948 instruction tells the optimizer that the value loaded is known to
6949 never be null. This is analogous to the ``nonnull`` attribute
6950 on parameters and return values. This metadata can only be applied
6951 to loads of a pointer type.
6953 The optional ``!dereferenceable`` metadata must reference a single metadata
6954 name ``<deref_bytes_node>`` corresponding to a metadata node with one ``i64``
6955 entry. The existence of the ``!dereferenceable`` metadata on the instruction
6956 tells the optimizer that the value loaded is known to be dereferenceable.
6957 The number of bytes known to be dereferenceable is specified by the integer
6958 value in the metadata node. This is analogous to the ''dereferenceable''
6959 attribute on parameters and return values. This metadata can only be applied
6960 to loads of a pointer type.
6962 The optional ``!dereferenceable_or_null`` metadata must reference a single
6963 metadata name ``<deref_bytes_node>`` corresponding to a metadata node with one
6964 ``i64`` entry. The existence of the ``!dereferenceable_or_null`` metadata on the
6965 instruction tells the optimizer that the value loaded is known to be either
6966 dereferenceable or null.
6967 The number of bytes known to be dereferenceable is specified by the integer
6968 value in the metadata node. This is analogous to the ''dereferenceable_or_null''
6969 attribute on parameters and return values. This metadata can only be applied
6970 to loads of a pointer type.
6972 The optional ``!align`` metadata must reference a single metadata name
6973 ``<align_node>`` corresponding to a metadata node with one ``i64`` entry.
6974 The existence of the ``!align`` metadata on the instruction tells the
6975 optimizer that the value loaded is known to be aligned to a boundary specified
6976 by the integer value in the metadata node. The alignment must be a power of 2.
6977 This is analogous to the ''align'' attribute on parameters and return values.
6978 This metadata can only be applied to loads of a pointer type.
6983 The location of memory pointed to is loaded. If the value being loaded
6984 is of scalar type then the number of bytes read does not exceed the
6985 minimum number of bytes needed to hold all bits of the type. For
6986 example, loading an ``i24`` reads at most three bytes. When loading a
6987 value of a type like ``i20`` with a size that is not an integral number
6988 of bytes, the result is undefined if the value was not originally
6989 written using a store of the same type.
6994 .. code-block:: llvm
6996 %ptr = alloca i32 ; yields i32*:ptr
6997 store i32 3, i32* %ptr ; yields void
6998 %val = load i32, i32* %ptr ; yields i32:val = i32 3
7002 '``store``' Instruction
7003 ^^^^^^^^^^^^^^^^^^^^^^^
7010 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.group !<index>] ; yields void
7011 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>] ; yields void
7016 The '``store``' instruction is used to write to memory.
7021 There are two arguments to the ``store`` instruction: a value to store
7022 and an address at which to store it. The type of the ``<pointer>``
7023 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
7024 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
7025 then the optimizer is not allowed to modify the number or order of
7026 execution of this ``store`` with other :ref:`volatile
7027 operations <volatile>`.
7029 If the ``store`` is marked as ``atomic``, it takes an extra
7030 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
7031 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
7032 instructions. Atomic loads produce :ref:`defined <memmodel>` results
7033 when they may see multiple atomic stores. The type of the pointee must
7034 be an integer type whose bit width is a power of two greater than or
7035 equal to eight and less than or equal to a target-specific size limit.
7036 ``align`` must be explicitly specified on atomic stores, and the store
7037 has undefined behavior if the alignment is not set to a value which is
7038 at least the size in bytes of the pointee. ``!nontemporal`` does not
7039 have any defined semantics for atomic stores.
7041 The optional constant ``align`` argument specifies the alignment of the
7042 operation (that is, the alignment of the memory address). A value of 0
7043 or an omitted ``align`` argument means that the operation has the ABI
7044 alignment for the target. It is the responsibility of the code emitter
7045 to ensure that the alignment information is correct. Overestimating the
7046 alignment results in undefined behavior. Underestimating the
7047 alignment may produce less efficient code. An alignment of 1 is always
7048 safe. The maximum possible alignment is ``1 << 29``.
7050 The optional ``!nontemporal`` metadata must reference a single metadata
7051 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
7052 value 1. The existence of the ``!nontemporal`` metadata on the instruction
7053 tells the optimizer and code generator that this load is not expected to
7054 be reused in the cache. The code generator may select special
7055 instructions to save cache bandwidth, such as the MOVNT instruction on
7058 The optional ``!invariant.group`` metadata must reference a
7059 single metadata name ``<index>``. See ``invariant.group`` metadata.
7064 The contents of memory are updated to contain ``<value>`` at the
7065 location specified by the ``<pointer>`` operand. If ``<value>`` is
7066 of scalar type then the number of bytes written does not exceed the
7067 minimum number of bytes needed to hold all bits of the type. For
7068 example, storing an ``i24`` writes at most three bytes. When writing a
7069 value of a type like ``i20`` with a size that is not an integral number
7070 of bytes, it is unspecified what happens to the extra bits that do not
7071 belong to the type, but they will typically be overwritten.
7076 .. code-block:: llvm
7078 %ptr = alloca i32 ; yields i32*:ptr
7079 store i32 3, i32* %ptr ; yields void
7080 %val = load i32, i32* %ptr ; yields i32:val = i32 3
7084 '``fence``' Instruction
7085 ^^^^^^^^^^^^^^^^^^^^^^^
7092 fence [singlethread] <ordering> ; yields void
7097 The '``fence``' instruction is used to introduce happens-before edges
7103 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
7104 defines what *synchronizes-with* edges they add. They can only be given
7105 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
7110 A fence A which has (at least) ``release`` ordering semantics
7111 *synchronizes with* a fence B with (at least) ``acquire`` ordering
7112 semantics if and only if there exist atomic operations X and Y, both
7113 operating on some atomic object M, such that A is sequenced before X, X
7114 modifies M (either directly or through some side effect of a sequence
7115 headed by X), Y is sequenced before B, and Y observes M. This provides a
7116 *happens-before* dependency between A and B. Rather than an explicit
7117 ``fence``, one (but not both) of the atomic operations X or Y might
7118 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
7119 still *synchronize-with* the explicit ``fence`` and establish the
7120 *happens-before* edge.
7122 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
7123 ``acquire`` and ``release`` semantics specified above, participates in
7124 the global program order of other ``seq_cst`` operations and/or fences.
7126 The optional ":ref:`singlethread <singlethread>`" argument specifies
7127 that the fence only synchronizes with other fences in the same thread.
7128 (This is useful for interacting with signal handlers.)
7133 .. code-block:: llvm
7135 fence acquire ; yields void
7136 fence singlethread seq_cst ; yields void
7140 '``cmpxchg``' Instruction
7141 ^^^^^^^^^^^^^^^^^^^^^^^^^
7148 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
7153 The '``cmpxchg``' instruction is used to atomically modify memory. It
7154 loads a value in memory and compares it to a given value. If they are
7155 equal, it tries to store a new value into the memory.
7160 There are three arguments to the '``cmpxchg``' instruction: an address
7161 to operate on, a value to compare to the value currently be at that
7162 address, and a new value to place at that address if the compared values
7163 are equal. The type of '<cmp>' must be an integer type whose bit width
7164 is a power of two greater than or equal to eight and less than or equal
7165 to a target-specific size limit. '<cmp>' and '<new>' must have the same
7166 type, and the type of '<pointer>' must be a pointer to that type. If the
7167 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
7168 to modify the number or order of execution of this ``cmpxchg`` with
7169 other :ref:`volatile operations <volatile>`.
7171 The success and failure :ref:`ordering <ordering>` arguments specify how this
7172 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
7173 must be at least ``monotonic``, the ordering constraint on failure must be no
7174 stronger than that on success, and the failure ordering cannot be either
7175 ``release`` or ``acq_rel``.
7177 The optional "``singlethread``" argument declares that the ``cmpxchg``
7178 is only atomic with respect to code (usually signal handlers) running in
7179 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
7180 respect to all other code in the system.
7182 The pointer passed into cmpxchg must have alignment greater than or
7183 equal to the size in memory of the operand.
7188 The contents of memory at the location specified by the '``<pointer>``' operand
7189 is read and compared to '``<cmp>``'; if the read value is the equal, the
7190 '``<new>``' is written. The original value at the location is returned, together
7191 with a flag indicating success (true) or failure (false).
7193 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
7194 permitted: the operation may not write ``<new>`` even if the comparison
7197 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
7198 if the value loaded equals ``cmp``.
7200 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
7201 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
7202 load with an ordering parameter determined the second ordering parameter.
7207 .. code-block:: llvm
7210 %orig = atomic load i32, i32* %ptr unordered ; yields i32
7214 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
7215 %squared = mul i32 %cmp, %cmp
7216 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
7217 %value_loaded = extractvalue { i32, i1 } %val_success, 0
7218 %success = extractvalue { i32, i1 } %val_success, 1
7219 br i1 %success, label %done, label %loop
7226 '``atomicrmw``' Instruction
7227 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7234 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
7239 The '``atomicrmw``' instruction is used to atomically modify memory.
7244 There are three arguments to the '``atomicrmw``' instruction: an
7245 operation to apply, an address whose value to modify, an argument to the
7246 operation. The operation must be one of the following keywords:
7260 The type of '<value>' must be an integer type whose bit width is a power
7261 of two greater than or equal to eight and less than or equal to a
7262 target-specific size limit. The type of the '``<pointer>``' operand must
7263 be a pointer to that type. If the ``atomicrmw`` is marked as
7264 ``volatile``, then the optimizer is not allowed to modify the number or
7265 order of execution of this ``atomicrmw`` with other :ref:`volatile
7266 operations <volatile>`.
7271 The contents of memory at the location specified by the '``<pointer>``'
7272 operand are atomically read, modified, and written back. The original
7273 value at the location is returned. The modification is specified by the
7276 - xchg: ``*ptr = val``
7277 - add: ``*ptr = *ptr + val``
7278 - sub: ``*ptr = *ptr - val``
7279 - and: ``*ptr = *ptr & val``
7280 - nand: ``*ptr = ~(*ptr & val)``
7281 - or: ``*ptr = *ptr | val``
7282 - xor: ``*ptr = *ptr ^ val``
7283 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
7284 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
7285 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
7287 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
7293 .. code-block:: llvm
7295 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
7297 .. _i_getelementptr:
7299 '``getelementptr``' Instruction
7300 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7307 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7308 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7309 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
7314 The '``getelementptr``' instruction is used to get the address of a
7315 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
7316 address calculation only and does not access memory. The instruction can also
7317 be used to calculate a vector of such addresses.
7322 The first argument is always a type used as the basis for the calculations.
7323 The second argument is always a pointer or a vector of pointers, and is the
7324 base address to start from. The remaining arguments are indices
7325 that indicate which of the elements of the aggregate object are indexed.
7326 The interpretation of each index is dependent on the type being indexed
7327 into. The first index always indexes the pointer value given as the
7328 first argument, the second index indexes a value of the type pointed to
7329 (not necessarily the value directly pointed to, since the first index
7330 can be non-zero), etc. The first type indexed into must be a pointer
7331 value, subsequent types can be arrays, vectors, and structs. Note that
7332 subsequent types being indexed into can never be pointers, since that
7333 would require loading the pointer before continuing calculation.
7335 The type of each index argument depends on the type it is indexing into.
7336 When indexing into a (optionally packed) structure, only ``i32`` integer
7337 **constants** are allowed (when using a vector of indices they must all
7338 be the **same** ``i32`` integer constant). When indexing into an array,
7339 pointer or vector, integers of any width are allowed, and they are not
7340 required to be constant. These integers are treated as signed values
7343 For example, let's consider a C code fragment and how it gets compiled
7359 int *foo(struct ST *s) {
7360 return &s[1].Z.B[5][13];
7363 The LLVM code generated by Clang is:
7365 .. code-block:: llvm
7367 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
7368 %struct.ST = type { i32, double, %struct.RT }
7370 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
7372 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
7379 In the example above, the first index is indexing into the
7380 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
7381 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
7382 indexes into the third element of the structure, yielding a
7383 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
7384 structure. The third index indexes into the second element of the
7385 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
7386 dimensions of the array are subscripted into, yielding an '``i32``'
7387 type. The '``getelementptr``' instruction returns a pointer to this
7388 element, thus computing a value of '``i32*``' type.
7390 Note that it is perfectly legal to index partially through a structure,
7391 returning a pointer to an inner element. Because of this, the LLVM code
7392 for the given testcase is equivalent to:
7394 .. code-block:: llvm
7396 define i32* @foo(%struct.ST* %s) {
7397 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
7398 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
7399 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
7400 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
7401 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
7405 If the ``inbounds`` keyword is present, the result value of the
7406 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
7407 pointer is not an *in bounds* address of an allocated object, or if any
7408 of the addresses that would be formed by successive addition of the
7409 offsets implied by the indices to the base address with infinitely
7410 precise signed arithmetic are not an *in bounds* address of that
7411 allocated object. The *in bounds* addresses for an allocated object are
7412 all the addresses that point into the object, plus the address one byte
7413 past the end. In cases where the base is a vector of pointers the
7414 ``inbounds`` keyword applies to each of the computations element-wise.
7416 If the ``inbounds`` keyword is not present, the offsets are added to the
7417 base address with silently-wrapping two's complement arithmetic. If the
7418 offsets have a different width from the pointer, they are sign-extended
7419 or truncated to the width of the pointer. The result value of the
7420 ``getelementptr`` may be outside the object pointed to by the base
7421 pointer. The result value may not necessarily be used to access memory
7422 though, even if it happens to point into allocated storage. See the
7423 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
7426 The getelementptr instruction is often confusing. For some more insight
7427 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
7432 .. code-block:: llvm
7434 ; yields [12 x i8]*:aptr
7435 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
7437 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
7439 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
7441 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
7446 The ``getelementptr`` returns a vector of pointers, instead of a single address,
7447 when one or more of its arguments is a vector. In such cases, all vector
7448 arguments should have the same number of elements, and every scalar argument
7449 will be effectively broadcast into a vector during address calculation.
7451 .. code-block:: llvm
7453 ; All arguments are vectors:
7454 ; A[i] = ptrs[i] + offsets[i]*sizeof(i8)
7455 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets
7457 ; Add the same scalar offset to each pointer of a vector:
7458 ; A[i] = ptrs[i] + offset*sizeof(i8)
7459 %A = getelementptr i8, <4 x i8*> %ptrs, i64 %offset
7461 ; Add distinct offsets to the same pointer:
7462 ; A[i] = ptr + offsets[i]*sizeof(i8)
7463 %A = getelementptr i8, i8* %ptr, <4 x i64> %offsets
7465 ; In all cases described above the type of the result is <4 x i8*>
7467 The two following instructions are equivalent:
7469 .. code-block:: llvm
7471 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7472 <4 x i32> <i32 2, i32 2, i32 2, i32 2>,
7473 <4 x i32> <i32 1, i32 1, i32 1, i32 1>,
7475 <4 x i64> <i64 13, i64 13, i64 13, i64 13>
7477 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7478 i32 2, i32 1, <4 x i32> %ind4, i64 13
7480 Let's look at the C code, where the vector version of ``getelementptr``
7485 // Let's assume that we vectorize the following loop:
7486 double *A, B; int *C;
7487 for (int i = 0; i < size; ++i) {
7491 .. code-block:: llvm
7493 ; get pointers for 8 elements from array B
7494 %ptrs = getelementptr double, double* %B, <8 x i32> %C
7495 ; load 8 elements from array B into A
7496 %A = call <8 x double> @llvm.masked.gather.v8f64(<8 x double*> %ptrs,
7497 i32 8, <8 x i1> %mask, <8 x double> %passthru)
7499 Conversion Operations
7500 ---------------------
7502 The instructions in this category are the conversion instructions
7503 (casting) which all take a single operand and a type. They perform
7504 various bit conversions on the operand.
7506 '``trunc .. to``' Instruction
7507 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7514 <result> = trunc <ty> <value> to <ty2> ; yields ty2
7519 The '``trunc``' instruction truncates its operand to the type ``ty2``.
7524 The '``trunc``' instruction takes a value to trunc, and a type to trunc
7525 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
7526 of the same number of integers. The bit size of the ``value`` must be
7527 larger than the bit size of the destination type, ``ty2``. Equal sized
7528 types are not allowed.
7533 The '``trunc``' instruction truncates the high order bits in ``value``
7534 and converts the remaining bits to ``ty2``. Since the source size must
7535 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
7536 It will always truncate bits.
7541 .. code-block:: llvm
7543 %X = trunc i32 257 to i8 ; yields i8:1
7544 %Y = trunc i32 123 to i1 ; yields i1:true
7545 %Z = trunc i32 122 to i1 ; yields i1:false
7546 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
7548 '``zext .. to``' Instruction
7549 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7556 <result> = zext <ty> <value> to <ty2> ; yields ty2
7561 The '``zext``' instruction zero extends its operand to type ``ty2``.
7566 The '``zext``' instruction takes a value to cast, and a type to cast it
7567 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7568 the same number of integers. The bit size of the ``value`` must be
7569 smaller than the bit size of the destination type, ``ty2``.
7574 The ``zext`` fills the high order bits of the ``value`` with zero bits
7575 until it reaches the size of the destination type, ``ty2``.
7577 When zero extending from i1, the result will always be either 0 or 1.
7582 .. code-block:: llvm
7584 %X = zext i32 257 to i64 ; yields i64:257
7585 %Y = zext i1 true to i32 ; yields i32:1
7586 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7588 '``sext .. to``' Instruction
7589 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7596 <result> = sext <ty> <value> to <ty2> ; yields ty2
7601 The '``sext``' sign extends ``value`` to the type ``ty2``.
7606 The '``sext``' instruction takes a value to cast, and a type to cast it
7607 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7608 the same number of integers. The bit size of the ``value`` must be
7609 smaller than the bit size of the destination type, ``ty2``.
7614 The '``sext``' instruction performs a sign extension by copying the sign
7615 bit (highest order bit) of the ``value`` until it reaches the bit size
7616 of the type ``ty2``.
7618 When sign extending from i1, the extension always results in -1 or 0.
7623 .. code-block:: llvm
7625 %X = sext i8 -1 to i16 ; yields i16 :65535
7626 %Y = sext i1 true to i32 ; yields i32:-1
7627 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7629 '``fptrunc .. to``' Instruction
7630 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7637 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
7642 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
7647 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
7648 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
7649 The size of ``value`` must be larger than the size of ``ty2``. This
7650 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
7655 The '``fptrunc``' instruction casts a ``value`` from a larger
7656 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
7657 point <t_floating>` type. If the value cannot fit (i.e. overflows) within the
7658 destination type, ``ty2``, then the results are undefined. If the cast produces
7659 an inexact result, how rounding is performed (e.g. truncation, also known as
7660 round to zero) is undefined.
7665 .. code-block:: llvm
7667 %X = fptrunc double 123.0 to float ; yields float:123.0
7668 %Y = fptrunc double 1.0E+300 to float ; yields undefined
7670 '``fpext .. to``' Instruction
7671 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7678 <result> = fpext <ty> <value> to <ty2> ; yields ty2
7683 The '``fpext``' extends a floating point ``value`` to a larger floating
7689 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
7690 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
7691 to. The source type must be smaller than the destination type.
7696 The '``fpext``' instruction extends the ``value`` from a smaller
7697 :ref:`floating point <t_floating>` type to a larger :ref:`floating
7698 point <t_floating>` type. The ``fpext`` cannot be used to make a
7699 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
7700 *no-op cast* for a floating point cast.
7705 .. code-block:: llvm
7707 %X = fpext float 3.125 to double ; yields double:3.125000e+00
7708 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
7710 '``fptoui .. to``' Instruction
7711 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7718 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
7723 The '``fptoui``' converts a floating point ``value`` to its unsigned
7724 integer equivalent of type ``ty2``.
7729 The '``fptoui``' instruction takes a value to cast, which must be a
7730 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7731 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7732 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7733 type with the same number of elements as ``ty``
7738 The '``fptoui``' instruction converts its :ref:`floating
7739 point <t_floating>` operand into the nearest (rounding towards zero)
7740 unsigned integer value. If the value cannot fit in ``ty2``, the results
7746 .. code-block:: llvm
7748 %X = fptoui double 123.0 to i32 ; yields i32:123
7749 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
7750 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
7752 '``fptosi .. to``' Instruction
7753 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7760 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
7765 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
7766 ``value`` to type ``ty2``.
7771 The '``fptosi``' instruction takes a value to cast, which must be a
7772 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7773 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7774 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7775 type with the same number of elements as ``ty``
7780 The '``fptosi``' instruction converts its :ref:`floating
7781 point <t_floating>` operand into the nearest (rounding towards zero)
7782 signed integer value. If the value cannot fit in ``ty2``, the results
7788 .. code-block:: llvm
7790 %X = fptosi double -123.0 to i32 ; yields i32:-123
7791 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
7792 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
7794 '``uitofp .. to``' Instruction
7795 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7802 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
7807 The '``uitofp``' instruction regards ``value`` as an unsigned integer
7808 and converts that value to the ``ty2`` type.
7813 The '``uitofp``' instruction takes a value to cast, which must be a
7814 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7815 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7816 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7817 type with the same number of elements as ``ty``
7822 The '``uitofp``' instruction interprets its operand as an unsigned
7823 integer quantity and converts it to the corresponding floating point
7824 value. If the value cannot fit in the floating point value, the results
7830 .. code-block:: llvm
7832 %X = uitofp i32 257 to float ; yields float:257.0
7833 %Y = uitofp i8 -1 to double ; yields double:255.0
7835 '``sitofp .. to``' Instruction
7836 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7843 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
7848 The '``sitofp``' instruction regards ``value`` as a signed integer and
7849 converts that value to the ``ty2`` type.
7854 The '``sitofp``' instruction takes a value to cast, which must be a
7855 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7856 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7857 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7858 type with the same number of elements as ``ty``
7863 The '``sitofp``' instruction interprets its operand as a signed integer
7864 quantity and converts it to the corresponding floating point value. If
7865 the value cannot fit in the floating point value, the results are
7871 .. code-block:: llvm
7873 %X = sitofp i32 257 to float ; yields float:257.0
7874 %Y = sitofp i8 -1 to double ; yields double:-1.0
7878 '``ptrtoint .. to``' Instruction
7879 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7886 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
7891 The '``ptrtoint``' instruction converts the pointer or a vector of
7892 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
7897 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
7898 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
7899 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
7900 a vector of integers type.
7905 The '``ptrtoint``' instruction converts ``value`` to integer type
7906 ``ty2`` by interpreting the pointer value as an integer and either
7907 truncating or zero extending that value to the size of the integer type.
7908 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
7909 ``value`` is larger than ``ty2`` then a truncation is done. If they are
7910 the same size, then nothing is done (*no-op cast*) other than a type
7916 .. code-block:: llvm
7918 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
7919 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
7920 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
7924 '``inttoptr .. to``' Instruction
7925 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7932 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
7937 The '``inttoptr``' instruction converts an integer ``value`` to a
7938 pointer type, ``ty2``.
7943 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
7944 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
7950 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
7951 applying either a zero extension or a truncation depending on the size
7952 of the integer ``value``. If ``value`` is larger than the size of a
7953 pointer then a truncation is done. If ``value`` is smaller than the size
7954 of a pointer then a zero extension is done. If they are the same size,
7955 nothing is done (*no-op cast*).
7960 .. code-block:: llvm
7962 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
7963 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
7964 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
7965 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
7969 '``bitcast .. to``' Instruction
7970 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7977 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
7982 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
7988 The '``bitcast``' instruction takes a value to cast, which must be a
7989 non-aggregate first class value, and a type to cast it to, which must
7990 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
7991 bit sizes of ``value`` and the destination type, ``ty2``, must be
7992 identical. If the source type is a pointer, the destination type must
7993 also be a pointer of the same size. This instruction supports bitwise
7994 conversion of vectors to integers and to vectors of other types (as
7995 long as they have the same size).
8000 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
8001 is always a *no-op cast* because no bits change with this
8002 conversion. The conversion is done as if the ``value`` had been stored
8003 to memory and read back as type ``ty2``. Pointer (or vector of
8004 pointers) types may only be converted to other pointer (or vector of
8005 pointers) types with the same address space through this instruction.
8006 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
8007 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
8012 .. code-block:: llvm
8014 %X = bitcast i8 255 to i8 ; yields i8 :-1
8015 %Y = bitcast i32* %x to sint* ; yields sint*:%x
8016 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
8017 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
8019 .. _i_addrspacecast:
8021 '``addrspacecast .. to``' Instruction
8022 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8029 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
8034 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
8035 address space ``n`` to type ``pty2`` in address space ``m``.
8040 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
8041 to cast and a pointer type to cast it to, which must have a different
8047 The '``addrspacecast``' instruction converts the pointer value
8048 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
8049 value modification, depending on the target and the address space
8050 pair. Pointer conversions within the same address space must be
8051 performed with the ``bitcast`` instruction. Note that if the address space
8052 conversion is legal then both result and operand refer to the same memory
8058 .. code-block:: llvm
8060 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
8061 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
8062 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
8069 The instructions in this category are the "miscellaneous" instructions,
8070 which defy better classification.
8074 '``icmp``' Instruction
8075 ^^^^^^^^^^^^^^^^^^^^^^
8082 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8087 The '``icmp``' instruction returns a boolean value or a vector of
8088 boolean values based on comparison of its two integer, integer vector,
8089 pointer, or pointer vector operands.
8094 The '``icmp``' instruction takes three operands. The first operand is
8095 the condition code indicating the kind of comparison to perform. It is
8096 not a value, just a keyword. The possible condition code are:
8099 #. ``ne``: not equal
8100 #. ``ugt``: unsigned greater than
8101 #. ``uge``: unsigned greater or equal
8102 #. ``ult``: unsigned less than
8103 #. ``ule``: unsigned less or equal
8104 #. ``sgt``: signed greater than
8105 #. ``sge``: signed greater or equal
8106 #. ``slt``: signed less than
8107 #. ``sle``: signed less or equal
8109 The remaining two arguments must be :ref:`integer <t_integer>` or
8110 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
8111 must also be identical types.
8116 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
8117 code given as ``cond``. The comparison performed always yields either an
8118 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
8120 #. ``eq``: yields ``true`` if the operands are equal, ``false``
8121 otherwise. No sign interpretation is necessary or performed.
8122 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
8123 otherwise. No sign interpretation is necessary or performed.
8124 #. ``ugt``: interprets the operands as unsigned values and yields
8125 ``true`` if ``op1`` is greater than ``op2``.
8126 #. ``uge``: interprets the operands as unsigned values and yields
8127 ``true`` if ``op1`` is greater than or equal to ``op2``.
8128 #. ``ult``: interprets the operands as unsigned values and yields
8129 ``true`` if ``op1`` is less than ``op2``.
8130 #. ``ule``: interprets the operands as unsigned values and yields
8131 ``true`` if ``op1`` is less than or equal to ``op2``.
8132 #. ``sgt``: interprets the operands as signed values and yields ``true``
8133 if ``op1`` is greater than ``op2``.
8134 #. ``sge``: interprets the operands as signed values and yields ``true``
8135 if ``op1`` is greater than or equal to ``op2``.
8136 #. ``slt``: interprets the operands as signed values and yields ``true``
8137 if ``op1`` is less than ``op2``.
8138 #. ``sle``: interprets the operands as signed values and yields ``true``
8139 if ``op1`` is less than or equal to ``op2``.
8141 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
8142 are compared as if they were integers.
8144 If the operands are integer vectors, then they are compared element by
8145 element. The result is an ``i1`` vector with the same number of elements
8146 as the values being compared. Otherwise, the result is an ``i1``.
8151 .. code-block:: llvm
8153 <result> = icmp eq i32 4, 5 ; yields: result=false
8154 <result> = icmp ne float* %X, %X ; yields: result=false
8155 <result> = icmp ult i16 4, 5 ; yields: result=true
8156 <result> = icmp sgt i16 4, 5 ; yields: result=false
8157 <result> = icmp ule i16 -4, 5 ; yields: result=false
8158 <result> = icmp sge i16 4, 5 ; yields: result=false
8160 Note that the code generator does not yet support vector types with the
8161 ``icmp`` instruction.
8165 '``fcmp``' Instruction
8166 ^^^^^^^^^^^^^^^^^^^^^^
8173 <result> = fcmp [fast-math flags]* <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8178 The '``fcmp``' instruction returns a boolean value or vector of boolean
8179 values based on comparison of its operands.
8181 If the operands are floating point scalars, then the result type is a
8182 boolean (:ref:`i1 <t_integer>`).
8184 If the operands are floating point vectors, then the result type is a
8185 vector of boolean with the same number of elements as the operands being
8191 The '``fcmp``' instruction takes three operands. The first operand is
8192 the condition code indicating the kind of comparison to perform. It is
8193 not a value, just a keyword. The possible condition code are:
8195 #. ``false``: no comparison, always returns false
8196 #. ``oeq``: ordered and equal
8197 #. ``ogt``: ordered and greater than
8198 #. ``oge``: ordered and greater than or equal
8199 #. ``olt``: ordered and less than
8200 #. ``ole``: ordered and less than or equal
8201 #. ``one``: ordered and not equal
8202 #. ``ord``: ordered (no nans)
8203 #. ``ueq``: unordered or equal
8204 #. ``ugt``: unordered or greater than
8205 #. ``uge``: unordered or greater than or equal
8206 #. ``ult``: unordered or less than
8207 #. ``ule``: unordered or less than or equal
8208 #. ``une``: unordered or not equal
8209 #. ``uno``: unordered (either nans)
8210 #. ``true``: no comparison, always returns true
8212 *Ordered* means that neither operand is a QNAN while *unordered* means
8213 that either operand may be a QNAN.
8215 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
8216 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
8217 type. They must have identical types.
8222 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
8223 condition code given as ``cond``. If the operands are vectors, then the
8224 vectors are compared element by element. Each comparison performed
8225 always yields an :ref:`i1 <t_integer>` result, as follows:
8227 #. ``false``: always yields ``false``, regardless of operands.
8228 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
8229 is equal to ``op2``.
8230 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
8231 is greater than ``op2``.
8232 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
8233 is greater than or equal to ``op2``.
8234 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
8235 is less than ``op2``.
8236 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
8237 is less than or equal to ``op2``.
8238 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
8239 is not equal to ``op2``.
8240 #. ``ord``: yields ``true`` if both operands are not a QNAN.
8241 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
8243 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
8244 greater than ``op2``.
8245 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
8246 greater than or equal to ``op2``.
8247 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
8249 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
8250 less than or equal to ``op2``.
8251 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
8252 not equal to ``op2``.
8253 #. ``uno``: yields ``true`` if either operand is a QNAN.
8254 #. ``true``: always yields ``true``, regardless of operands.
8256 The ``fcmp`` instruction can also optionally take any number of
8257 :ref:`fast-math flags <fastmath>`, which are optimization hints to enable
8258 otherwise unsafe floating point optimizations.
8260 Any set of fast-math flags are legal on an ``fcmp`` instruction, but the
8261 only flags that have any effect on its semantics are those that allow
8262 assumptions to be made about the values of input arguments; namely
8263 ``nnan``, ``ninf``, and ``nsz``. See :ref:`fastmath` for more information.
8268 .. code-block:: llvm
8270 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
8271 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
8272 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
8273 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
8275 Note that the code generator does not yet support vector types with the
8276 ``fcmp`` instruction.
8280 '``phi``' Instruction
8281 ^^^^^^^^^^^^^^^^^^^^^
8288 <result> = phi <ty> [ <val0>, <label0>], ...
8293 The '``phi``' instruction is used to implement the φ node in the SSA
8294 graph representing the function.
8299 The type of the incoming values is specified with the first type field.
8300 After this, the '``phi``' instruction takes a list of pairs as
8301 arguments, with one pair for each predecessor basic block of the current
8302 block. Only values of :ref:`first class <t_firstclass>` type may be used as
8303 the value arguments to the PHI node. Only labels may be used as the
8306 There must be no non-phi instructions between the start of a basic block
8307 and the PHI instructions: i.e. PHI instructions must be first in a basic
8310 For the purposes of the SSA form, the use of each incoming value is
8311 deemed to occur on the edge from the corresponding predecessor block to
8312 the current block (but after any definition of an '``invoke``'
8313 instruction's return value on the same edge).
8318 At runtime, the '``phi``' instruction logically takes on the value
8319 specified by the pair corresponding to the predecessor basic block that
8320 executed just prior to the current block.
8325 .. code-block:: llvm
8327 Loop: ; Infinite loop that counts from 0 on up...
8328 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
8329 %nextindvar = add i32 %indvar, 1
8334 '``select``' Instruction
8335 ^^^^^^^^^^^^^^^^^^^^^^^^
8342 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
8344 selty is either i1 or {<N x i1>}
8349 The '``select``' instruction is used to choose one value based on a
8350 condition, without IR-level branching.
8355 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
8356 values indicating the condition, and two values of the same :ref:`first
8357 class <t_firstclass>` type.
8362 If the condition is an i1 and it evaluates to 1, the instruction returns
8363 the first value argument; otherwise, it returns the second value
8366 If the condition is a vector of i1, then the value arguments must be
8367 vectors of the same size, and the selection is done element by element.
8369 If the condition is an i1 and the value arguments are vectors of the
8370 same size, then an entire vector is selected.
8375 .. code-block:: llvm
8377 %X = select i1 true, i8 17, i8 42 ; yields i8:17
8381 '``call``' Instruction
8382 ^^^^^^^^^^^^^^^^^^^^^^
8389 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
8395 The '``call``' instruction represents a simple function call.
8400 This instruction requires several arguments:
8402 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
8403 should perform tail call optimization. The ``tail`` marker is a hint that
8404 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
8405 means that the call must be tail call optimized in order for the program to
8406 be correct. The ``musttail`` marker provides these guarantees:
8408 #. The call will not cause unbounded stack growth if it is part of a
8409 recursive cycle in the call graph.
8410 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
8413 Both markers imply that the callee does not access allocas or varargs from
8414 the caller. Calls marked ``musttail`` must obey the following additional
8417 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
8418 or a pointer bitcast followed by a ret instruction.
8419 - The ret instruction must return the (possibly bitcasted) value
8420 produced by the call or void.
8421 - The caller and callee prototypes must match. Pointer types of
8422 parameters or return types may differ in pointee type, but not
8424 - The calling conventions of the caller and callee must match.
8425 - All ABI-impacting function attributes, such as sret, byval, inreg,
8426 returned, and inalloca, must match.
8427 - The callee must be varargs iff the caller is varargs. Bitcasting a
8428 non-varargs function to the appropriate varargs type is legal so
8429 long as the non-varargs prefixes obey the other rules.
8431 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
8432 the following conditions are met:
8434 - Caller and callee both have the calling convention ``fastcc``.
8435 - The call is in tail position (ret immediately follows call and ret
8436 uses value of call or is void).
8437 - Option ``-tailcallopt`` is enabled, or
8438 ``llvm::GuaranteedTailCallOpt`` is ``true``.
8439 - `Platform-specific constraints are
8440 met. <CodeGenerator.html#tailcallopt>`_
8442 #. The optional "cconv" marker indicates which :ref:`calling
8443 convention <callingconv>` the call should use. If none is
8444 specified, the call defaults to using C calling conventions. The
8445 calling convention of the call must match the calling convention of
8446 the target function, or else the behavior is undefined.
8447 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
8448 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
8450 #. '``ty``': the type of the call instruction itself which is also the
8451 type of the return value. Functions that return no value are marked
8453 #. '``fnty``': shall be the signature of the pointer to function value
8454 being invoked. The argument types must match the types implied by
8455 this signature. This type can be omitted if the function is not
8456 varargs and if the function type does not return a pointer to a
8458 #. '``fnptrval``': An LLVM value containing a pointer to a function to
8459 be invoked. In most cases, this is a direct function invocation, but
8460 indirect ``call``'s are just as possible, calling an arbitrary pointer
8462 #. '``function args``': argument list whose types match the function
8463 signature argument types and parameter attributes. All arguments must
8464 be of :ref:`first class <t_firstclass>` type. If the function signature
8465 indicates the function accepts a variable number of arguments, the
8466 extra arguments can be specified.
8467 #. The optional :ref:`function attributes <fnattrs>` list. Only
8468 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
8469 attributes are valid here.
8470 #. The optional :ref:`operand bundles <opbundles>` list.
8475 The '``call``' instruction is used to cause control flow to transfer to
8476 a specified function, with its incoming arguments bound to the specified
8477 values. Upon a '``ret``' instruction in the called function, control
8478 flow continues with the instruction after the function call, and the
8479 return value of the function is bound to the result argument.
8484 .. code-block:: llvm
8486 %retval = call i32 @test(i32 %argc)
8487 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
8488 %X = tail call i32 @foo() ; yields i32
8489 %Y = tail call fastcc i32 @foo() ; yields i32
8490 call void %foo(i8 97 signext)
8492 %struct.A = type { i32, i8 }
8493 %r = call %struct.A @foo() ; yields { i32, i8 }
8494 %gr = extractvalue %struct.A %r, 0 ; yields i32
8495 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
8496 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
8497 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
8499 llvm treats calls to some functions with names and arguments that match
8500 the standard C99 library as being the C99 library functions, and may
8501 perform optimizations or generate code for them under that assumption.
8502 This is something we'd like to change in the future to provide better
8503 support for freestanding environments and non-C-based languages.
8507 '``va_arg``' Instruction
8508 ^^^^^^^^^^^^^^^^^^^^^^^^
8515 <resultval> = va_arg <va_list*> <arglist>, <argty>
8520 The '``va_arg``' instruction is used to access arguments passed through
8521 the "variable argument" area of a function call. It is used to implement
8522 the ``va_arg`` macro in C.
8527 This instruction takes a ``va_list*`` value and the type of the
8528 argument. It returns a value of the specified argument type and
8529 increments the ``va_list`` to point to the next argument. The actual
8530 type of ``va_list`` is target specific.
8535 The '``va_arg``' instruction loads an argument of the specified type
8536 from the specified ``va_list`` and causes the ``va_list`` to point to
8537 the next argument. For more information, see the variable argument
8538 handling :ref:`Intrinsic Functions <int_varargs>`.
8540 It is legal for this instruction to be called in a function which does
8541 not take a variable number of arguments, for example, the ``vfprintf``
8544 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
8545 function <intrinsics>` because it takes a type as an argument.
8550 See the :ref:`variable argument processing <int_varargs>` section.
8552 Note that the code generator does not yet fully support va\_arg on many
8553 targets. Also, it does not currently support va\_arg with aggregate
8554 types on any target.
8558 '``landingpad``' Instruction
8559 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8566 <resultval> = landingpad <resultty> <clause>+
8567 <resultval> = landingpad <resultty> cleanup <clause>*
8569 <clause> := catch <type> <value>
8570 <clause> := filter <array constant type> <array constant>
8575 The '``landingpad``' instruction is used by `LLVM's exception handling
8576 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8577 is a landing pad --- one where the exception lands, and corresponds to the
8578 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
8579 defines values supplied by the :ref:`personality function <personalityfn>` upon
8580 re-entry to the function. The ``resultval`` has the type ``resultty``.
8586 ``cleanup`` flag indicates that the landing pad block is a cleanup.
8588 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
8589 contains the global variable representing the "type" that may be caught
8590 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
8591 clause takes an array constant as its argument. Use
8592 "``[0 x i8**] undef``" for a filter which cannot throw. The
8593 '``landingpad``' instruction must contain *at least* one ``clause`` or
8594 the ``cleanup`` flag.
8599 The '``landingpad``' instruction defines the values which are set by the
8600 :ref:`personality function <personalityfn>` upon re-entry to the function, and
8601 therefore the "result type" of the ``landingpad`` instruction. As with
8602 calling conventions, how the personality function results are
8603 represented in LLVM IR is target specific.
8605 The clauses are applied in order from top to bottom. If two
8606 ``landingpad`` instructions are merged together through inlining, the
8607 clauses from the calling function are appended to the list of clauses.
8608 When the call stack is being unwound due to an exception being thrown,
8609 the exception is compared against each ``clause`` in turn. If it doesn't
8610 match any of the clauses, and the ``cleanup`` flag is not set, then
8611 unwinding continues further up the call stack.
8613 The ``landingpad`` instruction has several restrictions:
8615 - A landing pad block is a basic block which is the unwind destination
8616 of an '``invoke``' instruction.
8617 - A landing pad block must have a '``landingpad``' instruction as its
8618 first non-PHI instruction.
8619 - There can be only one '``landingpad``' instruction within the landing
8621 - A basic block that is not a landing pad block may not include a
8622 '``landingpad``' instruction.
8627 .. code-block:: llvm
8629 ;; A landing pad which can catch an integer.
8630 %res = landingpad { i8*, i32 }
8632 ;; A landing pad that is a cleanup.
8633 %res = landingpad { i8*, i32 }
8635 ;; A landing pad which can catch an integer and can only throw a double.
8636 %res = landingpad { i8*, i32 }
8638 filter [1 x i8**] [@_ZTId]
8642 '``cleanuppad``' Instruction
8643 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8650 <resultval> = cleanuppad [<args>*]
8655 The '``cleanuppad``' instruction is used by `LLVM's exception handling
8656 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8657 is a cleanup block --- one where a personality routine attempts to
8658 transfer control to run cleanup actions.
8659 The ``args`` correspond to whatever additional
8660 information the :ref:`personality function <personalityfn>` requires to
8661 execute the cleanup.
8662 The ``resultval`` has the type :ref:`token <t_token>` and is used to
8663 match the ``cleanuppad`` to corresponding :ref:`cleanuprets <i_cleanupret>`
8664 and :ref:`cleanupendpads <i_cleanupendpad>`.
8669 The instruction takes a list of arbitrary values which are interpreted
8670 by the :ref:`personality function <personalityfn>`.
8675 When the call stack is being unwound due to an exception being thrown,
8676 the :ref:`personality function <personalityfn>` transfers control to the
8677 ``cleanuppad`` with the aid of the personality-specific arguments.
8678 As with calling conventions, how the personality function results are
8679 represented in LLVM IR is target specific.
8681 The ``cleanuppad`` instruction has several restrictions:
8683 - A cleanup block is a basic block which is the unwind destination of
8684 an exceptional instruction.
8685 - A cleanup block must have a '``cleanuppad``' instruction as its
8686 first non-PHI instruction.
8687 - There can be only one '``cleanuppad``' instruction within the
8689 - A basic block that is not a cleanup block may not include a
8690 '``cleanuppad``' instruction.
8691 - All '``cleanupret``'s and '``cleanupendpad``'s which consume a ``cleanuppad``
8692 must have the same exceptional successor.
8693 - It is undefined behavior for control to transfer from a ``cleanuppad`` to a
8694 ``ret`` without first executing a ``cleanupret`` or ``cleanupendpad`` that
8695 consumes the ``cleanuppad``.
8696 - It is undefined behavior for control to transfer from a ``cleanuppad`` to
8697 itself without first executing a ``cleanupret`` or ``cleanupendpad`` that
8698 consumes the ``cleanuppad``.
8703 .. code-block:: llvm
8705 %tok = cleanuppad []
8712 LLVM supports the notion of an "intrinsic function". These functions
8713 have well known names and semantics and are required to follow certain
8714 restrictions. Overall, these intrinsics represent an extension mechanism
8715 for the LLVM language that does not require changing all of the
8716 transformations in LLVM when adding to the language (or the bitcode
8717 reader/writer, the parser, etc...).
8719 Intrinsic function names must all start with an "``llvm.``" prefix. This
8720 prefix is reserved in LLVM for intrinsic names; thus, function names may
8721 not begin with this prefix. Intrinsic functions must always be external
8722 functions: you cannot define the body of intrinsic functions. Intrinsic
8723 functions may only be used in call or invoke instructions: it is illegal
8724 to take the address of an intrinsic function. Additionally, because
8725 intrinsic functions are part of the LLVM language, it is required if any
8726 are added that they be documented here.
8728 Some intrinsic functions can be overloaded, i.e., the intrinsic
8729 represents a family of functions that perform the same operation but on
8730 different data types. Because LLVM can represent over 8 million
8731 different integer types, overloading is used commonly to allow an
8732 intrinsic function to operate on any integer type. One or more of the
8733 argument types or the result type can be overloaded to accept any
8734 integer type. Argument types may also be defined as exactly matching a
8735 previous argument's type or the result type. This allows an intrinsic
8736 function which accepts multiple arguments, but needs all of them to be
8737 of the same type, to only be overloaded with respect to a single
8738 argument or the result.
8740 Overloaded intrinsics will have the names of its overloaded argument
8741 types encoded into its function name, each preceded by a period. Only
8742 those types which are overloaded result in a name suffix. Arguments
8743 whose type is matched against another type do not. For example, the
8744 ``llvm.ctpop`` function can take an integer of any width and returns an
8745 integer of exactly the same integer width. This leads to a family of
8746 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
8747 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
8748 overloaded, and only one type suffix is required. Because the argument's
8749 type is matched against the return type, it does not require its own
8752 To learn how to add an intrinsic function, please see the `Extending
8753 LLVM Guide <ExtendingLLVM.html>`_.
8757 Variable Argument Handling Intrinsics
8758 -------------------------------------
8760 Variable argument support is defined in LLVM with the
8761 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
8762 functions. These functions are related to the similarly named macros
8763 defined in the ``<stdarg.h>`` header file.
8765 All of these functions operate on arguments that use a target-specific
8766 value type "``va_list``". The LLVM assembly language reference manual
8767 does not define what this type is, so all transformations should be
8768 prepared to handle these functions regardless of the type used.
8770 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
8771 variable argument handling intrinsic functions are used.
8773 .. code-block:: llvm
8775 ; This struct is different for every platform. For most platforms,
8776 ; it is merely an i8*.
8777 %struct.va_list = type { i8* }
8779 ; For Unix x86_64 platforms, va_list is the following struct:
8780 ; %struct.va_list = type { i32, i32, i8*, i8* }
8782 define i32 @test(i32 %X, ...) {
8783 ; Initialize variable argument processing
8784 %ap = alloca %struct.va_list
8785 %ap2 = bitcast %struct.va_list* %ap to i8*
8786 call void @llvm.va_start(i8* %ap2)
8788 ; Read a single integer argument
8789 %tmp = va_arg i8* %ap2, i32
8791 ; Demonstrate usage of llvm.va_copy and llvm.va_end
8793 %aq2 = bitcast i8** %aq to i8*
8794 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
8795 call void @llvm.va_end(i8* %aq2)
8797 ; Stop processing of arguments.
8798 call void @llvm.va_end(i8* %ap2)
8802 declare void @llvm.va_start(i8*)
8803 declare void @llvm.va_copy(i8*, i8*)
8804 declare void @llvm.va_end(i8*)
8808 '``llvm.va_start``' Intrinsic
8809 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8816 declare void @llvm.va_start(i8* <arglist>)
8821 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
8822 subsequent use by ``va_arg``.
8827 The argument is a pointer to a ``va_list`` element to initialize.
8832 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
8833 available in C. In a target-dependent way, it initializes the
8834 ``va_list`` element to which the argument points, so that the next call
8835 to ``va_arg`` will produce the first variable argument passed to the
8836 function. Unlike the C ``va_start`` macro, this intrinsic does not need
8837 to know the last argument of the function as the compiler can figure
8840 '``llvm.va_end``' Intrinsic
8841 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8848 declare void @llvm.va_end(i8* <arglist>)
8853 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
8854 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
8859 The argument is a pointer to a ``va_list`` to destroy.
8864 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
8865 available in C. In a target-dependent way, it destroys the ``va_list``
8866 element to which the argument points. Calls to
8867 :ref:`llvm.va_start <int_va_start>` and
8868 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
8873 '``llvm.va_copy``' Intrinsic
8874 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8881 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
8886 The '``llvm.va_copy``' intrinsic copies the current argument position
8887 from the source argument list to the destination argument list.
8892 The first argument is a pointer to a ``va_list`` element to initialize.
8893 The second argument is a pointer to a ``va_list`` element to copy from.
8898 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
8899 available in C. In a target-dependent way, it copies the source
8900 ``va_list`` element into the destination ``va_list`` element. This
8901 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
8902 arbitrarily complex and require, for example, memory allocation.
8904 Accurate Garbage Collection Intrinsics
8905 --------------------------------------
8907 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
8908 (GC) requires the frontend to generate code containing appropriate intrinsic
8909 calls and select an appropriate GC strategy which knows how to lower these
8910 intrinsics in a manner which is appropriate for the target collector.
8912 These intrinsics allow identification of :ref:`GC roots on the
8913 stack <int_gcroot>`, as well as garbage collector implementations that
8914 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
8915 Frontends for type-safe garbage collected languages should generate
8916 these intrinsics to make use of the LLVM garbage collectors. For more
8917 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
8919 Experimental Statepoint Intrinsics
8920 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8922 LLVM provides an second experimental set of intrinsics for describing garbage
8923 collection safepoints in compiled code. These intrinsics are an alternative
8924 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
8925 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
8926 differences in approach are covered in the `Garbage Collection with LLVM
8927 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
8928 described in :doc:`Statepoints`.
8932 '``llvm.gcroot``' Intrinsic
8933 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8940 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
8945 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
8946 the code generator, and allows some metadata to be associated with it.
8951 The first argument specifies the address of a stack object that contains
8952 the root pointer. The second pointer (which must be either a constant or
8953 a global value address) contains the meta-data to be associated with the
8959 At runtime, a call to this intrinsic stores a null pointer into the
8960 "ptrloc" location. At compile-time, the code generator generates
8961 information to allow the runtime to find the pointer at GC safe points.
8962 The '``llvm.gcroot``' intrinsic may only be used in a function which
8963 :ref:`specifies a GC algorithm <gc>`.
8967 '``llvm.gcread``' Intrinsic
8968 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8975 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
8980 The '``llvm.gcread``' intrinsic identifies reads of references from heap
8981 locations, allowing garbage collector implementations that require read
8987 The second argument is the address to read from, which should be an
8988 address allocated from the garbage collector. The first object is a
8989 pointer to the start of the referenced object, if needed by the language
8990 runtime (otherwise null).
8995 The '``llvm.gcread``' intrinsic has the same semantics as a load
8996 instruction, but may be replaced with substantially more complex code by
8997 the garbage collector runtime, as needed. The '``llvm.gcread``'
8998 intrinsic may only be used in a function which :ref:`specifies a GC
9003 '``llvm.gcwrite``' Intrinsic
9004 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9011 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
9016 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
9017 locations, allowing garbage collector implementations that require write
9018 barriers (such as generational or reference counting collectors).
9023 The first argument is the reference to store, the second is the start of
9024 the object to store it to, and the third is the address of the field of
9025 Obj to store to. If the runtime does not require a pointer to the
9026 object, Obj may be null.
9031 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
9032 instruction, but may be replaced with substantially more complex code by
9033 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
9034 intrinsic may only be used in a function which :ref:`specifies a GC
9037 Code Generator Intrinsics
9038 -------------------------
9040 These intrinsics are provided by LLVM to expose special features that
9041 may only be implemented with code generator support.
9043 '``llvm.returnaddress``' Intrinsic
9044 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9051 declare i8 *@llvm.returnaddress(i32 <level>)
9056 The '``llvm.returnaddress``' intrinsic attempts to compute a
9057 target-specific value indicating the return address of the current
9058 function or one of its callers.
9063 The argument to this intrinsic indicates which function to return the
9064 address for. Zero indicates the calling function, one indicates its
9065 caller, etc. The argument is **required** to be a constant integer
9071 The '``llvm.returnaddress``' intrinsic either returns a pointer
9072 indicating the return address of the specified call frame, or zero if it
9073 cannot be identified. The value returned by this intrinsic is likely to
9074 be incorrect or 0 for arguments other than zero, so it should only be
9075 used for debugging purposes.
9077 Note that calling this intrinsic does not prevent function inlining or
9078 other aggressive transformations, so the value returned may not be that
9079 of the obvious source-language caller.
9081 '``llvm.frameaddress``' Intrinsic
9082 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9089 declare i8* @llvm.frameaddress(i32 <level>)
9094 The '``llvm.frameaddress``' intrinsic attempts to return the
9095 target-specific frame pointer value for the specified stack frame.
9100 The argument to this intrinsic indicates which function to return the
9101 frame pointer for. Zero indicates the calling function, one indicates
9102 its caller, etc. The argument is **required** to be a constant integer
9108 The '``llvm.frameaddress``' intrinsic either returns a pointer
9109 indicating the frame address of the specified call frame, or zero if it
9110 cannot be identified. The value returned by this intrinsic is likely to
9111 be incorrect or 0 for arguments other than zero, so it should only be
9112 used for debugging purposes.
9114 Note that calling this intrinsic does not prevent function inlining or
9115 other aggressive transformations, so the value returned may not be that
9116 of the obvious source-language caller.
9118 '``llvm.localescape``' and '``llvm.localrecover``' Intrinsics
9119 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9126 declare void @llvm.localescape(...)
9127 declare i8* @llvm.localrecover(i8* %func, i8* %fp, i32 %idx)
9132 The '``llvm.localescape``' intrinsic escapes offsets of a collection of static
9133 allocas, and the '``llvm.localrecover``' intrinsic applies those offsets to a
9134 live frame pointer to recover the address of the allocation. The offset is
9135 computed during frame layout of the caller of ``llvm.localescape``.
9140 All arguments to '``llvm.localescape``' must be pointers to static allocas or
9141 casts of static allocas. Each function can only call '``llvm.localescape``'
9142 once, and it can only do so from the entry block.
9144 The ``func`` argument to '``llvm.localrecover``' must be a constant
9145 bitcasted pointer to a function defined in the current module. The code
9146 generator cannot determine the frame allocation offset of functions defined in
9149 The ``fp`` argument to '``llvm.localrecover``' must be a frame pointer of a
9150 call frame that is currently live. The return value of '``llvm.localaddress``'
9151 is one way to produce such a value, but various runtimes also expose a suitable
9152 pointer in platform-specific ways.
9154 The ``idx`` argument to '``llvm.localrecover``' indicates which alloca passed to
9155 '``llvm.localescape``' to recover. It is zero-indexed.
9160 These intrinsics allow a group of functions to share access to a set of local
9161 stack allocations of a one parent function. The parent function may call the
9162 '``llvm.localescape``' intrinsic once from the function entry block, and the
9163 child functions can use '``llvm.localrecover``' to access the escaped allocas.
9164 The '``llvm.localescape``' intrinsic blocks inlining, as inlining changes where
9165 the escaped allocas are allocated, which would break attempts to use
9166 '``llvm.localrecover``'.
9168 .. _int_read_register:
9169 .. _int_write_register:
9171 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
9172 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9179 declare i32 @llvm.read_register.i32(metadata)
9180 declare i64 @llvm.read_register.i64(metadata)
9181 declare void @llvm.write_register.i32(metadata, i32 @value)
9182 declare void @llvm.write_register.i64(metadata, i64 @value)
9188 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
9189 provides access to the named register. The register must be valid on
9190 the architecture being compiled to. The type needs to be compatible
9191 with the register being read.
9196 The '``llvm.read_register``' intrinsic returns the current value of the
9197 register, where possible. The '``llvm.write_register``' intrinsic sets
9198 the current value of the register, where possible.
9200 This is useful to implement named register global variables that need
9201 to always be mapped to a specific register, as is common practice on
9202 bare-metal programs including OS kernels.
9204 The compiler doesn't check for register availability or use of the used
9205 register in surrounding code, including inline assembly. Because of that,
9206 allocatable registers are not supported.
9208 Warning: So far it only works with the stack pointer on selected
9209 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
9210 work is needed to support other registers and even more so, allocatable
9215 '``llvm.stacksave``' Intrinsic
9216 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9223 declare i8* @llvm.stacksave()
9228 The '``llvm.stacksave``' intrinsic is used to remember the current state
9229 of the function stack, for use with
9230 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
9231 implementing language features like scoped automatic variable sized
9237 This intrinsic returns a opaque pointer value that can be passed to
9238 :ref:`llvm.stackrestore <int_stackrestore>`. When an
9239 ``llvm.stackrestore`` intrinsic is executed with a value saved from
9240 ``llvm.stacksave``, it effectively restores the state of the stack to
9241 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
9242 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
9243 were allocated after the ``llvm.stacksave`` was executed.
9245 .. _int_stackrestore:
9247 '``llvm.stackrestore``' Intrinsic
9248 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9255 declare void @llvm.stackrestore(i8* %ptr)
9260 The '``llvm.stackrestore``' intrinsic is used to restore the state of
9261 the function stack to the state it was in when the corresponding
9262 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
9263 useful for implementing language features like scoped automatic variable
9264 sized arrays in C99.
9269 See the description for :ref:`llvm.stacksave <int_stacksave>`.
9271 '``llvm.prefetch``' Intrinsic
9272 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9279 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
9284 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
9285 insert a prefetch instruction if supported; otherwise, it is a noop.
9286 Prefetches have no effect on the behavior of the program but can change
9287 its performance characteristics.
9292 ``address`` is the address to be prefetched, ``rw`` is the specifier
9293 determining if the fetch should be for a read (0) or write (1), and
9294 ``locality`` is a temporal locality specifier ranging from (0) - no
9295 locality, to (3) - extremely local keep in cache. The ``cache type``
9296 specifies whether the prefetch is performed on the data (1) or
9297 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
9298 arguments must be constant integers.
9303 This intrinsic does not modify the behavior of the program. In
9304 particular, prefetches cannot trap and do not produce a value. On
9305 targets that support this intrinsic, the prefetch can provide hints to
9306 the processor cache for better performance.
9308 '``llvm.pcmarker``' Intrinsic
9309 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9316 declare void @llvm.pcmarker(i32 <id>)
9321 The '``llvm.pcmarker``' intrinsic is a method to export a Program
9322 Counter (PC) in a region of code to simulators and other tools. The
9323 method is target specific, but it is expected that the marker will use
9324 exported symbols to transmit the PC of the marker. The marker makes no
9325 guarantees that it will remain with any specific instruction after
9326 optimizations. It is possible that the presence of a marker will inhibit
9327 optimizations. The intended use is to be inserted after optimizations to
9328 allow correlations of simulation runs.
9333 ``id`` is a numerical id identifying the marker.
9338 This intrinsic does not modify the behavior of the program. Backends
9339 that do not support this intrinsic may ignore it.
9341 '``llvm.readcyclecounter``' Intrinsic
9342 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9349 declare i64 @llvm.readcyclecounter()
9354 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
9355 counter register (or similar low latency, high accuracy clocks) on those
9356 targets that support it. On X86, it should map to RDTSC. On Alpha, it
9357 should map to RPCC. As the backing counters overflow quickly (on the
9358 order of 9 seconds on alpha), this should only be used for small
9364 When directly supported, reading the cycle counter should not modify any
9365 memory. Implementations are allowed to either return a application
9366 specific value or a system wide value. On backends without support, this
9367 is lowered to a constant 0.
9369 Note that runtime support may be conditional on the privilege-level code is
9370 running at and the host platform.
9372 '``llvm.clear_cache``' Intrinsic
9373 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9380 declare void @llvm.clear_cache(i8*, i8*)
9385 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
9386 in the specified range to the execution unit of the processor. On
9387 targets with non-unified instruction and data cache, the implementation
9388 flushes the instruction cache.
9393 On platforms with coherent instruction and data caches (e.g. x86), this
9394 intrinsic is a nop. On platforms with non-coherent instruction and data
9395 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
9396 instructions or a system call, if cache flushing requires special
9399 The default behavior is to emit a call to ``__clear_cache`` from the run
9402 This instrinsic does *not* empty the instruction pipeline. Modifications
9403 of the current function are outside the scope of the intrinsic.
9405 '``llvm.instrprof_increment``' Intrinsic
9406 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9413 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
9414 i32 <num-counters>, i32 <index>)
9419 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
9420 frontend for use with instrumentation based profiling. These will be
9421 lowered by the ``-instrprof`` pass to generate execution counts of a
9427 The first argument is a pointer to a global variable containing the
9428 name of the entity being instrumented. This should generally be the
9429 (mangled) function name for a set of counters.
9431 The second argument is a hash value that can be used by the consumer
9432 of the profile data to detect changes to the instrumented source, and
9433 the third is the number of counters associated with ``name``. It is an
9434 error if ``hash`` or ``num-counters`` differ between two instances of
9435 ``instrprof_increment`` that refer to the same name.
9437 The last argument refers to which of the counters for ``name`` should
9438 be incremented. It should be a value between 0 and ``num-counters``.
9443 This intrinsic represents an increment of a profiling counter. It will
9444 cause the ``-instrprof`` pass to generate the appropriate data
9445 structures and the code to increment the appropriate value, in a
9446 format that can be written out by a compiler runtime and consumed via
9447 the ``llvm-profdata`` tool.
9449 Standard C Library Intrinsics
9450 -----------------------------
9452 LLVM provides intrinsics for a few important standard C library
9453 functions. These intrinsics allow source-language front-ends to pass
9454 information about the alignment of the pointer arguments to the code
9455 generator, providing opportunity for more efficient code generation.
9459 '``llvm.memcpy``' Intrinsic
9460 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9465 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
9466 integer bit width and for different address spaces. Not all targets
9467 support all bit widths however.
9471 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9472 i32 <len>, i32 <align>, i1 <isvolatile>)
9473 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9474 i64 <len>, i32 <align>, i1 <isvolatile>)
9479 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9480 source location to the destination location.
9482 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
9483 intrinsics do not return a value, takes extra alignment/isvolatile
9484 arguments and the pointers can be in specified address spaces.
9489 The first argument is a pointer to the destination, the second is a
9490 pointer to the source. The third argument is an integer argument
9491 specifying the number of bytes to copy, the fourth argument is the
9492 alignment of the source and destination locations, and the fifth is a
9493 boolean indicating a volatile access.
9495 If the call to this intrinsic has an alignment value that is not 0 or 1,
9496 then the caller guarantees that both the source and destination pointers
9497 are aligned to that boundary.
9499 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
9500 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9501 very cleanly specified and it is unwise to depend on it.
9506 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9507 source location to the destination location, which are not allowed to
9508 overlap. It copies "len" bytes of memory over. If the argument is known
9509 to be aligned to some boundary, this can be specified as the fourth
9510 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
9512 '``llvm.memmove``' Intrinsic
9513 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9518 This is an overloaded intrinsic. You can use llvm.memmove on any integer
9519 bit width and for different address space. Not all targets support all
9524 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9525 i32 <len>, i32 <align>, i1 <isvolatile>)
9526 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9527 i64 <len>, i32 <align>, i1 <isvolatile>)
9532 The '``llvm.memmove.*``' intrinsics move a block of memory from the
9533 source location to the destination location. It is similar to the
9534 '``llvm.memcpy``' intrinsic but allows the two memory locations to
9537 Note that, unlike the standard libc function, the ``llvm.memmove.*``
9538 intrinsics do not return a value, takes extra alignment/isvolatile
9539 arguments and the pointers can be in specified address spaces.
9544 The first argument is a pointer to the destination, the second is a
9545 pointer to the source. The third argument is an integer argument
9546 specifying the number of bytes to copy, the fourth argument is the
9547 alignment of the source and destination locations, and the fifth is a
9548 boolean indicating a volatile access.
9550 If the call to this intrinsic has an alignment value that is not 0 or 1,
9551 then the caller guarantees that the source and destination pointers are
9552 aligned to that boundary.
9554 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
9555 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
9556 not very cleanly specified and it is unwise to depend on it.
9561 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
9562 source location to the destination location, which may overlap. It
9563 copies "len" bytes of memory over. If the argument is known to be
9564 aligned to some boundary, this can be specified as the fourth argument,
9565 otherwise it should be set to 0 or 1 (both meaning no alignment).
9567 '``llvm.memset.*``' Intrinsics
9568 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9573 This is an overloaded intrinsic. You can use llvm.memset on any integer
9574 bit width and for different address spaces. However, not all targets
9575 support all bit widths.
9579 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
9580 i32 <len>, i32 <align>, i1 <isvolatile>)
9581 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
9582 i64 <len>, i32 <align>, i1 <isvolatile>)
9587 The '``llvm.memset.*``' intrinsics fill a block of memory with a
9588 particular byte value.
9590 Note that, unlike the standard libc function, the ``llvm.memset``
9591 intrinsic does not return a value and takes extra alignment/volatile
9592 arguments. Also, the destination can be in an arbitrary address space.
9597 The first argument is a pointer to the destination to fill, the second
9598 is the byte value with which to fill it, the third argument is an
9599 integer argument specifying the number of bytes to fill, and the fourth
9600 argument is the known alignment of the destination location.
9602 If the call to this intrinsic has an alignment value that is not 0 or 1,
9603 then the caller guarantees that the destination pointer is aligned to
9606 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
9607 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9608 very cleanly specified and it is unwise to depend on it.
9613 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
9614 at the destination location. If the argument is known to be aligned to
9615 some boundary, this can be specified as the fourth argument, otherwise
9616 it should be set to 0 or 1 (both meaning no alignment).
9618 '``llvm.sqrt.*``' Intrinsic
9619 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9624 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
9625 floating point or vector of floating point type. Not all targets support
9630 declare float @llvm.sqrt.f32(float %Val)
9631 declare double @llvm.sqrt.f64(double %Val)
9632 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
9633 declare fp128 @llvm.sqrt.f128(fp128 %Val)
9634 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
9639 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
9640 returning the same value as the libm '``sqrt``' functions would. Unlike
9641 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
9642 negative numbers other than -0.0 (which allows for better optimization,
9643 because there is no need to worry about errno being set).
9644 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
9649 The argument and return value are floating point numbers of the same
9655 This function returns the sqrt of the specified operand if it is a
9656 nonnegative floating point number.
9658 '``llvm.powi.*``' Intrinsic
9659 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9664 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
9665 floating point or vector of floating point type. Not all targets support
9670 declare float @llvm.powi.f32(float %Val, i32 %power)
9671 declare double @llvm.powi.f64(double %Val, i32 %power)
9672 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
9673 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
9674 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
9679 The '``llvm.powi.*``' intrinsics return the first operand raised to the
9680 specified (positive or negative) power. The order of evaluation of
9681 multiplications is not defined. When a vector of floating point type is
9682 used, the second argument remains a scalar integer value.
9687 The second argument is an integer power, and the first is a value to
9688 raise to that power.
9693 This function returns the first value raised to the second power with an
9694 unspecified sequence of rounding operations.
9696 '``llvm.sin.*``' Intrinsic
9697 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9702 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
9703 floating point or vector of floating point type. Not all targets support
9708 declare float @llvm.sin.f32(float %Val)
9709 declare double @llvm.sin.f64(double %Val)
9710 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
9711 declare fp128 @llvm.sin.f128(fp128 %Val)
9712 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
9717 The '``llvm.sin.*``' intrinsics return the sine of the operand.
9722 The argument and return value are floating point numbers of the same
9728 This function returns the sine of the specified operand, returning the
9729 same values as the libm ``sin`` functions would, and handles error
9730 conditions in the same way.
9732 '``llvm.cos.*``' Intrinsic
9733 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9738 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
9739 floating point or vector of floating point type. Not all targets support
9744 declare float @llvm.cos.f32(float %Val)
9745 declare double @llvm.cos.f64(double %Val)
9746 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
9747 declare fp128 @llvm.cos.f128(fp128 %Val)
9748 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
9753 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
9758 The argument and return value are floating point numbers of the same
9764 This function returns the cosine of the specified operand, returning the
9765 same values as the libm ``cos`` functions would, and handles error
9766 conditions in the same way.
9768 '``llvm.pow.*``' Intrinsic
9769 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9774 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
9775 floating point or vector of floating point type. Not all targets support
9780 declare float @llvm.pow.f32(float %Val, float %Power)
9781 declare double @llvm.pow.f64(double %Val, double %Power)
9782 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
9783 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
9784 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
9789 The '``llvm.pow.*``' intrinsics return the first operand raised to the
9790 specified (positive or negative) power.
9795 The second argument is a floating point power, and the first is a value
9796 to raise to that power.
9801 This function returns the first value raised to the second power,
9802 returning the same values as the libm ``pow`` functions would, and
9803 handles error conditions in the same way.
9805 '``llvm.exp.*``' Intrinsic
9806 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9811 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
9812 floating point or vector of floating point type. Not all targets support
9817 declare float @llvm.exp.f32(float %Val)
9818 declare double @llvm.exp.f64(double %Val)
9819 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
9820 declare fp128 @llvm.exp.f128(fp128 %Val)
9821 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
9826 The '``llvm.exp.*``' intrinsics perform the exp function.
9831 The argument and return value are floating point numbers of the same
9837 This function returns the same values as the libm ``exp`` functions
9838 would, and handles error conditions in the same way.
9840 '``llvm.exp2.*``' Intrinsic
9841 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9846 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
9847 floating point or vector of floating point type. Not all targets support
9852 declare float @llvm.exp2.f32(float %Val)
9853 declare double @llvm.exp2.f64(double %Val)
9854 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
9855 declare fp128 @llvm.exp2.f128(fp128 %Val)
9856 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
9861 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
9866 The argument and return value are floating point numbers of the same
9872 This function returns the same values as the libm ``exp2`` functions
9873 would, and handles error conditions in the same way.
9875 '``llvm.log.*``' Intrinsic
9876 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9881 This is an overloaded intrinsic. You can use ``llvm.log`` on any
9882 floating point or vector of floating point type. Not all targets support
9887 declare float @llvm.log.f32(float %Val)
9888 declare double @llvm.log.f64(double %Val)
9889 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
9890 declare fp128 @llvm.log.f128(fp128 %Val)
9891 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
9896 The '``llvm.log.*``' intrinsics perform the log function.
9901 The argument and return value are floating point numbers of the same
9907 This function returns the same values as the libm ``log`` functions
9908 would, and handles error conditions in the same way.
9910 '``llvm.log10.*``' Intrinsic
9911 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9916 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
9917 floating point or vector of floating point type. Not all targets support
9922 declare float @llvm.log10.f32(float %Val)
9923 declare double @llvm.log10.f64(double %Val)
9924 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
9925 declare fp128 @llvm.log10.f128(fp128 %Val)
9926 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
9931 The '``llvm.log10.*``' intrinsics perform the log10 function.
9936 The argument and return value are floating point numbers of the same
9942 This function returns the same values as the libm ``log10`` functions
9943 would, and handles error conditions in the same way.
9945 '``llvm.log2.*``' Intrinsic
9946 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9951 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
9952 floating point or vector of floating point type. Not all targets support
9957 declare float @llvm.log2.f32(float %Val)
9958 declare double @llvm.log2.f64(double %Val)
9959 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
9960 declare fp128 @llvm.log2.f128(fp128 %Val)
9961 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
9966 The '``llvm.log2.*``' intrinsics perform the log2 function.
9971 The argument and return value are floating point numbers of the same
9977 This function returns the same values as the libm ``log2`` functions
9978 would, and handles error conditions in the same way.
9980 '``llvm.fma.*``' Intrinsic
9981 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9986 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
9987 floating point or vector of floating point type. Not all targets support
9992 declare float @llvm.fma.f32(float %a, float %b, float %c)
9993 declare double @llvm.fma.f64(double %a, double %b, double %c)
9994 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
9995 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
9996 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
10001 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
10007 The argument and return value are floating point numbers of the same
10013 This function returns the same values as the libm ``fma`` functions
10014 would, and does not set errno.
10016 '``llvm.fabs.*``' Intrinsic
10017 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10022 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
10023 floating point or vector of floating point type. Not all targets support
10028 declare float @llvm.fabs.f32(float %Val)
10029 declare double @llvm.fabs.f64(double %Val)
10030 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
10031 declare fp128 @llvm.fabs.f128(fp128 %Val)
10032 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
10037 The '``llvm.fabs.*``' intrinsics return the absolute value of the
10043 The argument and return value are floating point numbers of the same
10049 This function returns the same values as the libm ``fabs`` functions
10050 would, and handles error conditions in the same way.
10052 '``llvm.minnum.*``' Intrinsic
10053 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10058 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
10059 floating point or vector of floating point type. Not all targets support
10064 declare float @llvm.minnum.f32(float %Val0, float %Val1)
10065 declare double @llvm.minnum.f64(double %Val0, double %Val1)
10066 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10067 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
10068 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10073 The '``llvm.minnum.*``' intrinsics return the minimum of the two
10080 The arguments and return value are floating point numbers of the same
10086 Follows the IEEE-754 semantics for minNum, which also match for libm's
10089 If either operand is a NaN, returns the other non-NaN operand. Returns
10090 NaN only if both operands are NaN. If the operands compare equal,
10091 returns a value that compares equal to both operands. This means that
10092 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10094 '``llvm.maxnum.*``' Intrinsic
10095 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10100 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
10101 floating point or vector of floating point type. Not all targets support
10106 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
10107 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
10108 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10109 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
10110 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10115 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
10122 The arguments and return value are floating point numbers of the same
10127 Follows the IEEE-754 semantics for maxNum, which also match for libm's
10130 If either operand is a NaN, returns the other non-NaN operand. Returns
10131 NaN only if both operands are NaN. If the operands compare equal,
10132 returns a value that compares equal to both operands. This means that
10133 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10135 '``llvm.copysign.*``' Intrinsic
10136 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10141 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
10142 floating point or vector of floating point type. Not all targets support
10147 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
10148 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
10149 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
10150 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
10151 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
10156 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
10157 first operand and the sign of the second operand.
10162 The arguments and return value are floating point numbers of the same
10168 This function returns the same values as the libm ``copysign``
10169 functions would, and handles error conditions in the same way.
10171 '``llvm.floor.*``' Intrinsic
10172 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10177 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
10178 floating point or vector of floating point type. Not all targets support
10183 declare float @llvm.floor.f32(float %Val)
10184 declare double @llvm.floor.f64(double %Val)
10185 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
10186 declare fp128 @llvm.floor.f128(fp128 %Val)
10187 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
10192 The '``llvm.floor.*``' intrinsics return the floor of the operand.
10197 The argument and return value are floating point numbers of the same
10203 This function returns the same values as the libm ``floor`` functions
10204 would, and handles error conditions in the same way.
10206 '``llvm.ceil.*``' Intrinsic
10207 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10212 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
10213 floating point or vector of floating point type. Not all targets support
10218 declare float @llvm.ceil.f32(float %Val)
10219 declare double @llvm.ceil.f64(double %Val)
10220 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
10221 declare fp128 @llvm.ceil.f128(fp128 %Val)
10222 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
10227 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
10232 The argument and return value are floating point numbers of the same
10238 This function returns the same values as the libm ``ceil`` functions
10239 would, and handles error conditions in the same way.
10241 '``llvm.trunc.*``' Intrinsic
10242 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10247 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
10248 floating point or vector of floating point type. Not all targets support
10253 declare float @llvm.trunc.f32(float %Val)
10254 declare double @llvm.trunc.f64(double %Val)
10255 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
10256 declare fp128 @llvm.trunc.f128(fp128 %Val)
10257 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
10262 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
10263 nearest integer not larger in magnitude than the operand.
10268 The argument and return value are floating point numbers of the same
10274 This function returns the same values as the libm ``trunc`` functions
10275 would, and handles error conditions in the same way.
10277 '``llvm.rint.*``' Intrinsic
10278 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10283 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
10284 floating point or vector of floating point type. Not all targets support
10289 declare float @llvm.rint.f32(float %Val)
10290 declare double @llvm.rint.f64(double %Val)
10291 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
10292 declare fp128 @llvm.rint.f128(fp128 %Val)
10293 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
10298 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
10299 nearest integer. It may raise an inexact floating-point exception if the
10300 operand isn't an integer.
10305 The argument and return value are floating point numbers of the same
10311 This function returns the same values as the libm ``rint`` functions
10312 would, and handles error conditions in the same way.
10314 '``llvm.nearbyint.*``' Intrinsic
10315 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10320 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
10321 floating point or vector of floating point type. Not all targets support
10326 declare float @llvm.nearbyint.f32(float %Val)
10327 declare double @llvm.nearbyint.f64(double %Val)
10328 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
10329 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
10330 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
10335 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
10341 The argument and return value are floating point numbers of the same
10347 This function returns the same values as the libm ``nearbyint``
10348 functions would, and handles error conditions in the same way.
10350 '``llvm.round.*``' Intrinsic
10351 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10356 This is an overloaded intrinsic. You can use ``llvm.round`` on any
10357 floating point or vector of floating point type. Not all targets support
10362 declare float @llvm.round.f32(float %Val)
10363 declare double @llvm.round.f64(double %Val)
10364 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
10365 declare fp128 @llvm.round.f128(fp128 %Val)
10366 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
10371 The '``llvm.round.*``' intrinsics returns the operand rounded to the
10377 The argument and return value are floating point numbers of the same
10383 This function returns the same values as the libm ``round``
10384 functions would, and handles error conditions in the same way.
10386 Bit Manipulation Intrinsics
10387 ---------------------------
10389 LLVM provides intrinsics for a few important bit manipulation
10390 operations. These allow efficient code generation for some algorithms.
10392 '``llvm.bswap.*``' Intrinsics
10393 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10398 This is an overloaded intrinsic function. You can use bswap on any
10399 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
10403 declare i16 @llvm.bswap.i16(i16 <id>)
10404 declare i32 @llvm.bswap.i32(i32 <id>)
10405 declare i64 @llvm.bswap.i64(i64 <id>)
10410 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
10411 values with an even number of bytes (positive multiple of 16 bits).
10412 These are useful for performing operations on data that is not in the
10413 target's native byte order.
10418 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
10419 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
10420 intrinsic returns an i32 value that has the four bytes of the input i32
10421 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
10422 returned i32 will have its bytes in 3, 2, 1, 0 order. The
10423 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
10424 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
10427 '``llvm.ctpop.*``' Intrinsic
10428 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10433 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
10434 bit width, or on any vector with integer elements. Not all targets
10435 support all bit widths or vector types, however.
10439 declare i8 @llvm.ctpop.i8(i8 <src>)
10440 declare i16 @llvm.ctpop.i16(i16 <src>)
10441 declare i32 @llvm.ctpop.i32(i32 <src>)
10442 declare i64 @llvm.ctpop.i64(i64 <src>)
10443 declare i256 @llvm.ctpop.i256(i256 <src>)
10444 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
10449 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
10455 The only argument is the value to be counted. The argument may be of any
10456 integer type, or a vector with integer elements. The return type must
10457 match the argument type.
10462 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
10463 each element of a vector.
10465 '``llvm.ctlz.*``' Intrinsic
10466 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10471 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
10472 integer bit width, or any vector whose elements are integers. Not all
10473 targets support all bit widths or vector types, however.
10477 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
10478 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
10479 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
10480 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
10481 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
10482 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10487 The '``llvm.ctlz``' family of intrinsic functions counts the number of
10488 leading zeros in a variable.
10493 The first argument is the value to be counted. This argument may be of
10494 any integer type, or a vector with integer element type. The return
10495 type must match the first argument type.
10497 The second argument must be a constant and is a flag to indicate whether
10498 the intrinsic should ensure that a zero as the first argument produces a
10499 defined result. Historically some architectures did not provide a
10500 defined result for zero values as efficiently, and many algorithms are
10501 now predicated on avoiding zero-value inputs.
10506 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
10507 zeros in a variable, or within each element of the vector. If
10508 ``src == 0`` then the result is the size in bits of the type of ``src``
10509 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10510 ``llvm.ctlz(i32 2) = 30``.
10512 '``llvm.cttz.*``' Intrinsic
10513 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10518 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
10519 integer bit width, or any vector of integer elements. Not all targets
10520 support all bit widths or vector types, however.
10524 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
10525 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
10526 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
10527 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
10528 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
10529 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10534 The '``llvm.cttz``' family of intrinsic functions counts the number of
10540 The first argument is the value to be counted. This argument may be of
10541 any integer type, or a vector with integer element type. The return
10542 type must match the first argument type.
10544 The second argument must be a constant and is a flag to indicate whether
10545 the intrinsic should ensure that a zero as the first argument produces a
10546 defined result. Historically some architectures did not provide a
10547 defined result for zero values as efficiently, and many algorithms are
10548 now predicated on avoiding zero-value inputs.
10553 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
10554 zeros in a variable, or within each element of a vector. If ``src == 0``
10555 then the result is the size in bits of the type of ``src`` if
10556 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10557 ``llvm.cttz(2) = 1``.
10561 Arithmetic with Overflow Intrinsics
10562 -----------------------------------
10564 LLVM provides intrinsics for some arithmetic with overflow operations.
10566 '``llvm.sadd.with.overflow.*``' Intrinsics
10567 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10572 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
10573 on any integer bit width.
10577 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
10578 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10579 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
10584 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10585 a signed addition of the two arguments, and indicate whether an overflow
10586 occurred during the signed summation.
10591 The arguments (%a and %b) and the first element of the result structure
10592 may be of integer types of any bit width, but they must have the same
10593 bit width. The second element of the result structure must be of type
10594 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10600 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10601 a signed addition of the two variables. They return a structure --- the
10602 first element of which is the signed summation, and the second element
10603 of which is a bit specifying if the signed summation resulted in an
10609 .. code-block:: llvm
10611 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10612 %sum = extractvalue {i32, i1} %res, 0
10613 %obit = extractvalue {i32, i1} %res, 1
10614 br i1 %obit, label %overflow, label %normal
10616 '``llvm.uadd.with.overflow.*``' Intrinsics
10617 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10622 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
10623 on any integer bit width.
10627 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
10628 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10629 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
10634 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10635 an unsigned addition of the two arguments, and indicate whether a carry
10636 occurred during the unsigned summation.
10641 The arguments (%a and %b) and the first element of the result structure
10642 may be of integer types of any bit width, but they must have the same
10643 bit width. The second element of the result structure must be of type
10644 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10650 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10651 an unsigned addition of the two arguments. They return a structure --- the
10652 first element of which is the sum, and the second element of which is a
10653 bit specifying if the unsigned summation resulted in a carry.
10658 .. code-block:: llvm
10660 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10661 %sum = extractvalue {i32, i1} %res, 0
10662 %obit = extractvalue {i32, i1} %res, 1
10663 br i1 %obit, label %carry, label %normal
10665 '``llvm.ssub.with.overflow.*``' Intrinsics
10666 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10671 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
10672 on any integer bit width.
10676 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
10677 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10678 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
10683 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10684 a signed subtraction of the two arguments, and indicate whether an
10685 overflow occurred during the signed subtraction.
10690 The arguments (%a and %b) and the first element of the result structure
10691 may be of integer types of any bit width, but they must have the same
10692 bit width. The second element of the result structure must be of type
10693 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10699 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10700 a signed subtraction of the two arguments. They return a structure --- the
10701 first element of which is the subtraction, and the second element of
10702 which is a bit specifying if the signed subtraction resulted in an
10708 .. code-block:: llvm
10710 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10711 %sum = extractvalue {i32, i1} %res, 0
10712 %obit = extractvalue {i32, i1} %res, 1
10713 br i1 %obit, label %overflow, label %normal
10715 '``llvm.usub.with.overflow.*``' Intrinsics
10716 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10721 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
10722 on any integer bit width.
10726 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
10727 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10728 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
10733 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10734 an unsigned subtraction of the two arguments, and indicate whether an
10735 overflow occurred during the unsigned subtraction.
10740 The arguments (%a and %b) and the first element of the result structure
10741 may be of integer types of any bit width, but they must have the same
10742 bit width. The second element of the result structure must be of type
10743 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10749 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10750 an unsigned subtraction of the two arguments. They return a structure ---
10751 the first element of which is the subtraction, and the second element of
10752 which is a bit specifying if the unsigned subtraction resulted in an
10758 .. code-block:: llvm
10760 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10761 %sum = extractvalue {i32, i1} %res, 0
10762 %obit = extractvalue {i32, i1} %res, 1
10763 br i1 %obit, label %overflow, label %normal
10765 '``llvm.smul.with.overflow.*``' Intrinsics
10766 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10771 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
10772 on any integer bit width.
10776 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
10777 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10778 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
10783 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10784 a signed multiplication of the two arguments, and indicate whether an
10785 overflow occurred during the signed multiplication.
10790 The arguments (%a and %b) and the first element of the result structure
10791 may be of integer types of any bit width, but they must have the same
10792 bit width. The second element of the result structure must be of type
10793 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10799 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10800 a signed multiplication of the two arguments. They return a structure ---
10801 the first element of which is the multiplication, and the second element
10802 of which is a bit specifying if the signed multiplication resulted in an
10808 .. code-block:: llvm
10810 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10811 %sum = extractvalue {i32, i1} %res, 0
10812 %obit = extractvalue {i32, i1} %res, 1
10813 br i1 %obit, label %overflow, label %normal
10815 '``llvm.umul.with.overflow.*``' Intrinsics
10816 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10821 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
10822 on any integer bit width.
10826 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
10827 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10828 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
10833 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10834 a unsigned multiplication of the two arguments, and indicate whether an
10835 overflow occurred during the unsigned multiplication.
10840 The arguments (%a and %b) and the first element of the result structure
10841 may be of integer types of any bit width, but they must have the same
10842 bit width. The second element of the result structure must be of type
10843 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10849 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10850 an unsigned multiplication of the two arguments. They return a structure ---
10851 the first element of which is the multiplication, and the second
10852 element of which is a bit specifying if the unsigned multiplication
10853 resulted in an overflow.
10858 .. code-block:: llvm
10860 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10861 %sum = extractvalue {i32, i1} %res, 0
10862 %obit = extractvalue {i32, i1} %res, 1
10863 br i1 %obit, label %overflow, label %normal
10865 Specialised Arithmetic Intrinsics
10866 ---------------------------------
10868 '``llvm.canonicalize.*``' Intrinsic
10869 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10876 declare float @llvm.canonicalize.f32(float %a)
10877 declare double @llvm.canonicalize.f64(double %b)
10882 The '``llvm.canonicalize.*``' intrinsic returns the platform specific canonical
10883 encoding of a floating point number. This canonicalization is useful for
10884 implementing certain numeric primitives such as frexp. The canonical encoding is
10885 defined by IEEE-754-2008 to be:
10889 2.1.8 canonical encoding: The preferred encoding of a floating-point
10890 representation in a format. Applied to declets, significands of finite
10891 numbers, infinities, and NaNs, especially in decimal formats.
10893 This operation can also be considered equivalent to the IEEE-754-2008
10894 conversion of a floating-point value to the same format. NaNs are handled
10895 according to section 6.2.
10897 Examples of non-canonical encodings:
10899 - x87 pseudo denormals, pseudo NaNs, pseudo Infinity, Unnormals. These are
10900 converted to a canonical representation per hardware-specific protocol.
10901 - Many normal decimal floating point numbers have non-canonical alternative
10903 - Some machines, like GPUs or ARMv7 NEON, do not support subnormal values.
10904 These are treated as non-canonical encodings of zero and with be flushed to
10905 a zero of the same sign by this operation.
10907 Note that per IEEE-754-2008 6.2, systems that support signaling NaNs with
10908 default exception handling must signal an invalid exception, and produce a
10911 This function should always be implementable as multiplication by 1.0, provided
10912 that the compiler does not constant fold the operation. Likewise, division by
10913 1.0 and ``llvm.minnum(x, x)`` are possible implementations. Addition with
10914 -0.0 is also sufficient provided that the rounding mode is not -Infinity.
10916 ``@llvm.canonicalize`` must preserve the equality relation. That is:
10918 - ``(@llvm.canonicalize(x) == x)`` is equivalent to ``(x == x)``
10919 - ``(@llvm.canonicalize(x) == @llvm.canonicalize(y))`` is equivalent to
10922 Additionally, the sign of zero must be conserved:
10923 ``@llvm.canonicalize(-0.0) = -0.0`` and ``@llvm.canonicalize(+0.0) = +0.0``
10925 The payload bits of a NaN must be conserved, with two exceptions.
10926 First, environments which use only a single canonical representation of NaN
10927 must perform said canonicalization. Second, SNaNs must be quieted per the
10930 The canonicalization operation may be optimized away if:
10932 - The input is known to be canonical. For example, it was produced by a
10933 floating-point operation that is required by the standard to be canonical.
10934 - The result is consumed only by (or fused with) other floating-point
10935 operations. That is, the bits of the floating point value are not examined.
10937 '``llvm.fmuladd.*``' Intrinsic
10938 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10945 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
10946 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
10951 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
10952 expressions that can be fused if the code generator determines that (a) the
10953 target instruction set has support for a fused operation, and (b) that the
10954 fused operation is more efficient than the equivalent, separate pair of mul
10955 and add instructions.
10960 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
10961 multiplicands, a and b, and an addend c.
10970 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
10972 is equivalent to the expression a \* b + c, except that rounding will
10973 not be performed between the multiplication and addition steps if the
10974 code generator fuses the operations. Fusion is not guaranteed, even if
10975 the target platform supports it. If a fused multiply-add is required the
10976 corresponding llvm.fma.\* intrinsic function should be used
10977 instead. This never sets errno, just as '``llvm.fma.*``'.
10982 .. code-block:: llvm
10984 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
10987 '``llvm.uabsdiff.*``' and '``llvm.sabsdiff.*``' Intrinsics
10988 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10992 This is an overloaded intrinsic. The loaded data is a vector of any integer bit width.
10994 .. code-block:: llvm
10996 declare <4 x integer> @llvm.uabsdiff.v4i32(<4 x integer> %a, <4 x integer> %b)
11002 The ``llvm.uabsdiff`` intrinsic returns a vector result of the absolute difference
11003 of the two operands, treating them both as unsigned integers. The intermediate
11004 calculations are computed using infinitely precise unsigned arithmetic. The final
11005 result will be truncated to the given type.
11007 The ``llvm.sabsdiff`` intrinsic returns a vector result of the absolute difference of
11008 the two operands, treating them both as signed integers. If the result overflows, the
11009 behavior is undefined.
11013 These intrinsics are primarily used during the code generation stage of compilation.
11014 They are generated by compiler passes such as the Loop and SLP vectorizers. It is not
11015 recommended for users to create them manually.
11020 Both intrinsics take two integer of the same bitwidth.
11027 call <4 x i32> @llvm.uabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
11031 %1 = zext <4 x i32> %a to <4 x i64>
11032 %2 = zext <4 x i32> %b to <4 x i64>
11033 %sub = sub <4 x i64> %1, %2
11034 %trunc = trunc <4 x i64> to <4 x i32>
11036 and the expression::
11038 call <4 x i32> @llvm.sabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
11042 %sub = sub nsw <4 x i32> %a, %b
11043 %ispos = icmp sge <4 x i32> %sub, zeroinitializer
11044 %neg = sub nsw <4 x i32> zeroinitializer, %sub
11045 %1 = select <4 x i1> %ispos, <4 x i32> %sub, <4 x i32> %neg
11048 Half Precision Floating Point Intrinsics
11049 ----------------------------------------
11051 For most target platforms, half precision floating point is a
11052 storage-only format. This means that it is a dense encoding (in memory)
11053 but does not support computation in the format.
11055 This means that code must first load the half-precision floating point
11056 value as an i16, then convert it to float with
11057 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
11058 then be performed on the float value (including extending to double
11059 etc). To store the value back to memory, it is first converted to float
11060 if needed, then converted to i16 with
11061 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
11064 .. _int_convert_to_fp16:
11066 '``llvm.convert.to.fp16``' Intrinsic
11067 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11074 declare i16 @llvm.convert.to.fp16.f32(float %a)
11075 declare i16 @llvm.convert.to.fp16.f64(double %a)
11080 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11081 conventional floating point type to half precision floating point format.
11086 The intrinsic function contains single argument - the value to be
11092 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11093 conventional floating point format to half precision floating point format. The
11094 return value is an ``i16`` which contains the converted number.
11099 .. code-block:: llvm
11101 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
11102 store i16 %res, i16* @x, align 2
11104 .. _int_convert_from_fp16:
11106 '``llvm.convert.from.fp16``' Intrinsic
11107 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11114 declare float @llvm.convert.from.fp16.f32(i16 %a)
11115 declare double @llvm.convert.from.fp16.f64(i16 %a)
11120 The '``llvm.convert.from.fp16``' intrinsic function performs a
11121 conversion from half precision floating point format to single precision
11122 floating point format.
11127 The intrinsic function contains single argument - the value to be
11133 The '``llvm.convert.from.fp16``' intrinsic function performs a
11134 conversion from half single precision floating point format to single
11135 precision floating point format. The input half-float value is
11136 represented by an ``i16`` value.
11141 .. code-block:: llvm
11143 %a = load i16, i16* @x, align 2
11144 %res = call float @llvm.convert.from.fp16(i16 %a)
11146 .. _dbg_intrinsics:
11148 Debugger Intrinsics
11149 -------------------
11151 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
11152 prefix), are described in the `LLVM Source Level
11153 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
11156 Exception Handling Intrinsics
11157 -----------------------------
11159 The LLVM exception handling intrinsics (which all start with
11160 ``llvm.eh.`` prefix), are described in the `LLVM Exception
11161 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
11163 .. _int_trampoline:
11165 Trampoline Intrinsics
11166 ---------------------
11168 These intrinsics make it possible to excise one parameter, marked with
11169 the :ref:`nest <nest>` attribute, from a function. The result is a
11170 callable function pointer lacking the nest parameter - the caller does
11171 not need to provide a value for it. Instead, the value to use is stored
11172 in advance in a "trampoline", a block of memory usually allocated on the
11173 stack, which also contains code to splice the nest value into the
11174 argument list. This is used to implement the GCC nested function address
11177 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
11178 then the resulting function pointer has signature ``i32 (i32, i32)*``.
11179 It can be created as follows:
11181 .. code-block:: llvm
11183 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
11184 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
11185 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
11186 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
11187 %fp = bitcast i8* %p to i32 (i32, i32)*
11189 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
11190 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
11194 '``llvm.init.trampoline``' Intrinsic
11195 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11202 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
11207 This fills the memory pointed to by ``tramp`` with executable code,
11208 turning it into a trampoline.
11213 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
11214 pointers. The ``tramp`` argument must point to a sufficiently large and
11215 sufficiently aligned block of memory; this memory is written to by the
11216 intrinsic. Note that the size and the alignment are target-specific -
11217 LLVM currently provides no portable way of determining them, so a
11218 front-end that generates this intrinsic needs to have some
11219 target-specific knowledge. The ``func`` argument must hold a function
11220 bitcast to an ``i8*``.
11225 The block of memory pointed to by ``tramp`` is filled with target
11226 dependent code, turning it into a function. Then ``tramp`` needs to be
11227 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
11228 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
11229 function's signature is the same as that of ``func`` with any arguments
11230 marked with the ``nest`` attribute removed. At most one such ``nest``
11231 argument is allowed, and it must be of pointer type. Calling the new
11232 function is equivalent to calling ``func`` with the same argument list,
11233 but with ``nval`` used for the missing ``nest`` argument. If, after
11234 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
11235 modified, then the effect of any later call to the returned function
11236 pointer is undefined.
11240 '``llvm.adjust.trampoline``' Intrinsic
11241 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11248 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
11253 This performs any required machine-specific adjustment to the address of
11254 a trampoline (passed as ``tramp``).
11259 ``tramp`` must point to a block of memory which already has trampoline
11260 code filled in by a previous call to
11261 :ref:`llvm.init.trampoline <int_it>`.
11266 On some architectures the address of the code to be executed needs to be
11267 different than the address where the trampoline is actually stored. This
11268 intrinsic returns the executable address corresponding to ``tramp``
11269 after performing the required machine specific adjustments. The pointer
11270 returned can then be :ref:`bitcast and executed <int_trampoline>`.
11272 .. _int_mload_mstore:
11274 Masked Vector Load and Store Intrinsics
11275 ---------------------------------------
11277 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.
11281 '``llvm.masked.load.*``' Intrinsics
11282 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11286 This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
11290 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11291 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11296 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.
11302 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.
11308 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.
11309 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.
11314 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
11316 ;; The result of the two following instructions is identical aside from potential memory access exception
11317 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
11318 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
11322 '``llvm.masked.store.*``' Intrinsics
11323 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11327 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
11331 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
11332 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
11337 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.
11342 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.
11348 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.
11349 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.
11353 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
11355 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
11356 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
11357 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
11358 store <16 x float> %res, <16 x float>* %ptr, align 4
11361 Masked Vector Gather and Scatter Intrinsics
11362 -------------------------------------------
11364 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.
11368 '``llvm.masked.gather.*``' Intrinsics
11369 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11373 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.
11377 declare <16 x float> @llvm.masked.gather.v16f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11378 declare <2 x double> @llvm.masked.gather.v2f64 (<2 x double*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11383 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.
11389 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.
11395 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.
11396 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.
11401 %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>)
11403 ;; The gather with all-true mask is equivalent to the following instruction sequence
11404 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
11405 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
11406 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
11407 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
11409 %val0 = load double, double* %ptr0, align 8
11410 %val1 = load double, double* %ptr1, align 8
11411 %val2 = load double, double* %ptr2, align 8
11412 %val3 = load double, double* %ptr3, align 8
11414 %vec0 = insertelement <4 x double>undef, %val0, 0
11415 %vec01 = insertelement <4 x double>%vec0, %val1, 1
11416 %vec012 = insertelement <4 x double>%vec01, %val2, 2
11417 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
11421 '``llvm.masked.scatter.*``' Intrinsics
11422 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11426 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.
11430 declare void @llvm.masked.scatter.v8i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
11431 declare void @llvm.masked.scatter.v16f32(<16 x i32> <value>, <16 x i32*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
11436 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.
11441 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.
11447 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.
11451 ;; This instruction unconditionaly stores data vector in multiple addresses
11452 call @llvm.masked.scatter.v8i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
11454 ;; It is equivalent to a list of scalar stores
11455 %val0 = extractelement <8 x i32> %value, i32 0
11456 %val1 = extractelement <8 x i32> %value, i32 1
11458 %val7 = extractelement <8 x i32> %value, i32 7
11459 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
11460 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
11462 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
11463 ;; Note: the order of the following stores is important when they overlap:
11464 store i32 %val0, i32* %ptr0, align 4
11465 store i32 %val1, i32* %ptr1, align 4
11467 store i32 %val7, i32* %ptr7, align 4
11473 This class of intrinsics provides information about the lifetime of
11474 memory objects and ranges where variables are immutable.
11478 '``llvm.lifetime.start``' Intrinsic
11479 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11486 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
11491 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
11497 The first argument is a constant integer representing the size of the
11498 object, or -1 if it is variable sized. The second argument is a pointer
11504 This intrinsic indicates that before this point in the code, the value
11505 of the memory pointed to by ``ptr`` is dead. This means that it is known
11506 to never be used and has an undefined value. A load from the pointer
11507 that precedes this intrinsic can be replaced with ``'undef'``.
11511 '``llvm.lifetime.end``' Intrinsic
11512 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11519 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
11524 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
11530 The first argument is a constant integer representing the size of the
11531 object, or -1 if it is variable sized. The second argument is a pointer
11537 This intrinsic indicates that after this point in the code, the value of
11538 the memory pointed to by ``ptr`` is dead. This means that it is known to
11539 never be used and has an undefined value. Any stores into the memory
11540 object following this intrinsic may be removed as dead.
11542 '``llvm.invariant.start``' Intrinsic
11543 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11550 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
11555 The '``llvm.invariant.start``' intrinsic specifies that the contents of
11556 a memory object will not change.
11561 The first argument is a constant integer representing the size of the
11562 object, or -1 if it is variable sized. The second argument is a pointer
11568 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
11569 the return value, the referenced memory location is constant and
11572 '``llvm.invariant.end``' Intrinsic
11573 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11580 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
11585 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
11586 memory object are mutable.
11591 The first argument is the matching ``llvm.invariant.start`` intrinsic.
11592 The second argument is a constant integer representing the size of the
11593 object, or -1 if it is variable sized and the third argument is a
11594 pointer to the object.
11599 This intrinsic indicates that the memory is mutable again.
11601 '``llvm.invariant.group.barrier``' Intrinsic
11602 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11609 declare i8* @llvm.invariant.group.barrier(i8* <ptr>)
11614 The '``llvm.invariant.group.barrier``' intrinsic can be used when an invariant
11615 established by invariant.group metadata no longer holds, to obtain a new pointer
11616 value that does not carry the invariant information.
11622 The ``llvm.invariant.group.barrier`` takes only one argument, which is
11623 the pointer to the memory for which the ``invariant.group`` no longer holds.
11628 Returns another pointer that aliases its argument but which is considered different
11629 for the purposes of ``load``/``store`` ``invariant.group`` metadata.
11634 This class of intrinsics is designed to be generic and has no specific
11637 '``llvm.var.annotation``' Intrinsic
11638 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11645 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11650 The '``llvm.var.annotation``' intrinsic.
11655 The first argument is a pointer to a value, the second is a pointer to a
11656 global string, the third is a pointer to a global string which is the
11657 source file name, and the last argument is the line number.
11662 This intrinsic allows annotation of local variables with arbitrary
11663 strings. This can be useful for special purpose optimizations that want
11664 to look for these annotations. These have no other defined use; they are
11665 ignored by code generation and optimization.
11667 '``llvm.ptr.annotation.*``' Intrinsic
11668 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11673 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
11674 pointer to an integer of any width. *NOTE* you must specify an address space for
11675 the pointer. The identifier for the default address space is the integer
11680 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11681 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
11682 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
11683 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
11684 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
11689 The '``llvm.ptr.annotation``' intrinsic.
11694 The first argument is a pointer to an integer value of arbitrary bitwidth
11695 (result of some expression), the second is a pointer to a global string, the
11696 third is a pointer to a global string which is the source file name, and the
11697 last argument is the line number. It returns the value of the first argument.
11702 This intrinsic allows annotation of a pointer to an integer with arbitrary
11703 strings. This can be useful for special purpose optimizations that want to look
11704 for these annotations. These have no other defined use; they are ignored by code
11705 generation and optimization.
11707 '``llvm.annotation.*``' Intrinsic
11708 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11713 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
11714 any integer bit width.
11718 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
11719 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
11720 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
11721 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
11722 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
11727 The '``llvm.annotation``' intrinsic.
11732 The first argument is an integer value (result of some expression), the
11733 second is a pointer to a global string, the third is a pointer to a
11734 global string which is the source file name, and the last argument is
11735 the line number. It returns the value of the first argument.
11740 This intrinsic allows annotations to be put on arbitrary expressions
11741 with arbitrary strings. This can be useful for special purpose
11742 optimizations that want to look for these annotations. These have no
11743 other defined use; they are ignored by code generation and optimization.
11745 '``llvm.trap``' Intrinsic
11746 ^^^^^^^^^^^^^^^^^^^^^^^^^
11753 declare void @llvm.trap() noreturn nounwind
11758 The '``llvm.trap``' intrinsic.
11768 This intrinsic is lowered to the target dependent trap instruction. If
11769 the target does not have a trap instruction, this intrinsic will be
11770 lowered to a call of the ``abort()`` function.
11772 '``llvm.debugtrap``' Intrinsic
11773 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11780 declare void @llvm.debugtrap() nounwind
11785 The '``llvm.debugtrap``' intrinsic.
11795 This intrinsic is lowered to code which is intended to cause an
11796 execution trap with the intention of requesting the attention of a
11799 '``llvm.stackprotector``' Intrinsic
11800 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11807 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
11812 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
11813 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
11814 is placed on the stack before local variables.
11819 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
11820 The first argument is the value loaded from the stack guard
11821 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
11822 enough space to hold the value of the guard.
11827 This intrinsic causes the prologue/epilogue inserter to force the position of
11828 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
11829 to ensure that if a local variable on the stack is overwritten, it will destroy
11830 the value of the guard. When the function exits, the guard on the stack is
11831 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
11832 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
11833 calling the ``__stack_chk_fail()`` function.
11835 '``llvm.stackprotectorcheck``' Intrinsic
11836 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11843 declare void @llvm.stackprotectorcheck(i8** <guard>)
11848 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
11849 created stack protector and if they are not equal calls the
11850 ``__stack_chk_fail()`` function.
11855 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
11856 the variable ``@__stack_chk_guard``.
11861 This intrinsic is provided to perform the stack protector check by comparing
11862 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
11863 values do not match call the ``__stack_chk_fail()`` function.
11865 The reason to provide this as an IR level intrinsic instead of implementing it
11866 via other IR operations is that in order to perform this operation at the IR
11867 level without an intrinsic, one would need to create additional basic blocks to
11868 handle the success/failure cases. This makes it difficult to stop the stack
11869 protector check from disrupting sibling tail calls in Codegen. With this
11870 intrinsic, we are able to generate the stack protector basic blocks late in
11871 codegen after the tail call decision has occurred.
11873 '``llvm.objectsize``' Intrinsic
11874 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11881 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
11882 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
11887 The ``llvm.objectsize`` intrinsic is designed to provide information to
11888 the optimizers to determine at compile time whether a) an operation
11889 (like memcpy) will overflow a buffer that corresponds to an object, or
11890 b) that a runtime check for overflow isn't necessary. An object in this
11891 context means an allocation of a specific class, structure, array, or
11897 The ``llvm.objectsize`` intrinsic takes two arguments. The first
11898 argument is a pointer to or into the ``object``. The second argument is
11899 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
11900 or -1 (if false) when the object size is unknown. The second argument
11901 only accepts constants.
11906 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
11907 the size of the object concerned. If the size cannot be determined at
11908 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
11909 on the ``min`` argument).
11911 '``llvm.expect``' Intrinsic
11912 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
11917 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
11922 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
11923 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
11924 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
11929 The ``llvm.expect`` intrinsic provides information about expected (the
11930 most probable) value of ``val``, which can be used by optimizers.
11935 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
11936 a value. The second argument is an expected value, this needs to be a
11937 constant value, variables are not allowed.
11942 This intrinsic is lowered to the ``val``.
11946 '``llvm.assume``' Intrinsic
11947 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11954 declare void @llvm.assume(i1 %cond)
11959 The ``llvm.assume`` allows the optimizer to assume that the provided
11960 condition is true. This information can then be used in simplifying other parts
11966 The condition which the optimizer may assume is always true.
11971 The intrinsic allows the optimizer to assume that the provided condition is
11972 always true whenever the control flow reaches the intrinsic call. No code is
11973 generated for this intrinsic, and instructions that contribute only to the
11974 provided condition are not used for code generation. If the condition is
11975 violated during execution, the behavior is undefined.
11977 Note that the optimizer might limit the transformations performed on values
11978 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
11979 only used to form the intrinsic's input argument. This might prove undesirable
11980 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
11981 sufficient overall improvement in code quality. For this reason,
11982 ``llvm.assume`` should not be used to document basic mathematical invariants
11983 that the optimizer can otherwise deduce or facts that are of little use to the
11988 '``llvm.bitset.test``' Intrinsic
11989 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11996 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
12002 The first argument is a pointer to be tested. The second argument is a
12003 metadata object representing an identifier for a :doc:`bitset <BitSets>`.
12008 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
12009 member of the given bitset.
12011 '``llvm.donothing``' Intrinsic
12012 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12019 declare void @llvm.donothing() nounwind readnone
12024 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
12025 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
12026 with an invoke instruction.
12036 This intrinsic does nothing, and it's removed by optimizers and ignored
12039 Stack Map Intrinsics
12040 --------------------
12042 LLVM provides experimental intrinsics to support runtime patching
12043 mechanisms commonly desired in dynamic language JITs. These intrinsics
12044 are described in :doc:`StackMaps`.