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 a 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 into
208 the object file corresponding to the LLVM module. From the linker's
209 perspective, an ``available_externally`` global is equivalent to
210 an external declaration. They exist to allow inlining and other
211 optimizations to take place given knowledge of the definition of the
212 global, which is known to be somewhere outside the module. Globals
213 with ``available_externally`` linkage are allowed to be discarded at
214 will, and allow inlining and other optimizations. This linkage type is
215 only allowed on definitions, not declarations.
217 Globals with "``linkonce``" linkage are merged with other globals of
218 the same name when linkage occurs. This can be used to implement
219 some forms of inline functions, templates, or other code which must
220 be generated in each translation unit that uses it, but where the
221 body may be overridden with a more definitive definition later.
222 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
223 that ``linkonce`` linkage does not actually allow the optimizer to
224 inline the body of this function into callers because it doesn't
225 know if this definition of the function is the definitive definition
226 within the program or whether it will be overridden by a stronger
227 definition. To enable inlining and other optimizations, use
228 "``linkonce_odr``" linkage.
230 "``weak``" linkage has the same merging semantics as ``linkonce``
231 linkage, except that unreferenced globals with ``weak`` linkage may
232 not be discarded. This is used for globals that are declared "weak"
235 "``common``" linkage is most similar to "``weak``" linkage, but they
236 are used for tentative definitions in C, such as "``int X;``" at
237 global scope. Symbols with "``common``" linkage are merged in the
238 same way as ``weak symbols``, and they may not be deleted if
239 unreferenced. ``common`` symbols may not have an explicit section,
240 must have a zero initializer, and may not be marked
241 ':ref:`constant <globalvars>`'. Functions and aliases may not have
244 .. _linkage_appending:
247 "``appending``" linkage may only be applied to global variables of
248 pointer to array type. When two global variables with appending
249 linkage are linked together, the two global arrays are appended
250 together. This is the LLVM, typesafe, equivalent of having the
251 system linker append together "sections" with identical names when
254 Unfortunately this doesn't correspond to any feature in .o files, so it
255 can only be used for variables like ``llvm.global_ctors`` which llvm
256 interprets specially.
259 The semantics of this linkage follow the ELF object file model: the
260 symbol is weak until linked, if not linked, the symbol becomes null
261 instead of being an undefined reference.
262 ``linkonce_odr``, ``weak_odr``
263 Some languages allow differing globals to be merged, such as two
264 functions with different semantics. Other languages, such as
265 ``C++``, ensure that only equivalent globals are ever merged (the
266 "one definition rule" --- "ODR"). Such languages can use the
267 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
268 global will only be merged with equivalent globals. These linkage
269 types are otherwise the same as their non-``odr`` versions.
271 If none of the above identifiers are used, the global is externally
272 visible, meaning that it participates in linkage and can be used to
273 resolve external symbol references.
275 It is illegal for a function *declaration* to have any linkage type
276 other than ``external`` or ``extern_weak``.
283 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
284 :ref:`invokes <i_invoke>` can all have an optional calling convention
285 specified for the call. The calling convention of any pair of dynamic
286 caller/callee must match, or the behavior of the program is undefined.
287 The following calling conventions are supported by LLVM, and more may be
290 "``ccc``" - The C calling convention
291 This calling convention (the default if no other calling convention
292 is specified) matches the target C calling conventions. This calling
293 convention supports varargs function calls and tolerates some
294 mismatch in the declared prototype and implemented declaration of
295 the function (as does normal C).
296 "``fastcc``" - The fast calling convention
297 This calling convention attempts to make calls as fast as possible
298 (e.g. by passing things in registers). This calling convention
299 allows the target to use whatever tricks it wants to produce fast
300 code for the target, without having to conform to an externally
301 specified ABI (Application Binary Interface). `Tail calls can only
302 be optimized when this, the GHC or the HiPE convention is
303 used. <CodeGenerator.html#id80>`_ This calling convention does not
304 support varargs and requires the prototype of all callees to exactly
305 match the prototype of the function definition.
306 "``coldcc``" - The cold calling convention
307 This calling convention attempts to make code in the caller as
308 efficient as possible under the assumption that the call is not
309 commonly executed. As such, these calls often preserve all registers
310 so that the call does not break any live ranges in the caller side.
311 This calling convention does not support varargs and requires the
312 prototype of all callees to exactly match the prototype of the
313 function definition. Furthermore the inliner doesn't consider such function
315 "``cc 10``" - GHC convention
316 This calling convention has been implemented specifically for use by
317 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
318 It passes everything in registers, going to extremes to achieve this
319 by disabling callee save registers. This calling convention should
320 not be used lightly but only for specific situations such as an
321 alternative to the *register pinning* performance technique often
322 used when implementing functional programming languages. At the
323 moment only X86 supports this convention and it has the following
326 - On *X86-32* only supports up to 4 bit type parameters. No
327 floating-point types are supported.
328 - On *X86-64* only supports up to 10 bit type parameters and 6
329 floating-point parameters.
331 This calling convention supports `tail call
332 optimization <CodeGenerator.html#id80>`_ but requires both the
333 caller and callee are using it.
334 "``cc 11``" - The HiPE calling convention
335 This calling convention has been implemented specifically for use by
336 the `High-Performance Erlang
337 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
338 native code compiler of the `Ericsson's Open Source Erlang/OTP
339 system <http://www.erlang.org/download.shtml>`_. It uses more
340 registers for argument passing than the ordinary C calling
341 convention and defines no callee-saved registers. The calling
342 convention properly supports `tail call
343 optimization <CodeGenerator.html#id80>`_ but requires that both the
344 caller and the callee use it. It uses a *register pinning*
345 mechanism, similar to GHC's convention, for keeping frequently
346 accessed runtime components pinned to specific hardware registers.
347 At the moment only X86 supports this convention (both 32 and 64
349 "``webkit_jscc``" - WebKit's JavaScript calling convention
350 This calling convention has been implemented for `WebKit FTL JIT
351 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
352 stack right to left (as cdecl does), and returns a value in the
353 platform's customary return register.
354 "``anyregcc``" - Dynamic calling convention for code patching
355 This is a special convention that supports patching an arbitrary code
356 sequence in place of a call site. This convention forces the call
357 arguments into registers but allows them to be dynamically
358 allocated. This can currently only be used with calls to
359 llvm.experimental.patchpoint because only this intrinsic records
360 the location of its arguments in a side table. See :doc:`StackMaps`.
361 "``preserve_mostcc``" - The `PreserveMost` calling convention
362 This calling convention attempts to make the code in the caller as
363 unintrusive as possible. This convention behaves identically to the `C`
364 calling convention on how arguments and return values are passed, but it
365 uses a different set of caller/callee-saved registers. This alleviates the
366 burden of saving and recovering a large register set before and after the
367 call in the caller. If the arguments are passed in callee-saved registers,
368 then they will be preserved by the callee across the call. This doesn't
369 apply for values returned in callee-saved registers.
371 - On X86-64 the callee preserves all general purpose registers, except for
372 R11. R11 can be used as a scratch register. Floating-point registers
373 (XMMs/YMMs) are not preserved and need to be saved by the caller.
375 The idea behind this convention is to support calls to runtime functions
376 that have a hot path and a cold path. The hot path is usually a small piece
377 of code that doesn't use many registers. The cold path might need to call out to
378 another function and therefore only needs to preserve the caller-saved
379 registers, which haven't already been saved by the caller. The
380 `PreserveMost` calling convention is very similar to the `cold` calling
381 convention in terms of caller/callee-saved registers, but they are used for
382 different types of function calls. `coldcc` is for function calls that are
383 rarely executed, whereas `preserve_mostcc` function calls are intended to be
384 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
385 doesn't prevent the inliner from inlining the function call.
387 This calling convention will be used by a future version of the ObjectiveC
388 runtime and should therefore still be considered experimental at this time.
389 Although this convention was created to optimize certain runtime calls to
390 the ObjectiveC runtime, it is not limited to this runtime and might be used
391 by other runtimes in the future too. The current implementation only
392 supports X86-64, but the intention is to support more architectures in the
394 "``preserve_allcc``" - The `PreserveAll` calling convention
395 This calling convention attempts to make the code in the caller even less
396 intrusive than the `PreserveMost` calling convention. This calling
397 convention also behaves identical to the `C` calling convention on how
398 arguments and return values are passed, but it uses a different set of
399 caller/callee-saved registers. This removes the burden of saving and
400 recovering a large register set before and after the call in the caller. If
401 the arguments are passed in callee-saved registers, then they will be
402 preserved by the callee across the call. This doesn't apply for values
403 returned in callee-saved registers.
405 - On X86-64 the callee preserves all general purpose registers, except for
406 R11. R11 can be used as a scratch register. Furthermore it also preserves
407 all floating-point registers (XMMs/YMMs).
409 The idea behind this convention is to support calls to runtime functions
410 that don't need to call out to any other functions.
412 This calling convention, like the `PreserveMost` calling convention, will be
413 used by a future version of the ObjectiveC runtime and should be considered
414 experimental at this time.
415 "``cxx_fast_tlscc``" - The `CXX_FAST_TLS` calling convention for access functions
416 Clang generates an access function to access C++-style TLS. The access
417 function generally has an entry block, an exit block and an initialization
418 block that is run at the first time. The entry and exit blocks can access
419 a few TLS IR variables, each access will be lowered to a platform-specific
422 This calling convention aims to minimize overhead in the caller by
423 preserving as many registers as possible (all the registers that are
424 perserved on the fast path, composed of the entry and exit blocks).
426 This calling convention behaves identical to the `C` calling convention on
427 how arguments and return values are passed, but it uses a different set of
428 caller/callee-saved registers.
430 Given that each platform has its own lowering sequence, hence its own set
431 of preserved registers, we can't use the existing `PreserveMost`.
433 - On X86-64 the callee preserves all general purpose registers, except for
435 "``swiftcc``" - This calling convention is used for Swift language.
436 - On X86-64 RCX and R8 are available for additional integer returns, and
437 XMM2 and XMM3 are available for additional FP/vector returns.
438 - On iOS platforms, we use AAPCS-VFP calling convention.
439 "``cc <n>``" - Numbered convention
440 Any calling convention may be specified by number, allowing
441 target-specific calling conventions to be used. Target specific
442 calling conventions start at 64.
444 More calling conventions can be added/defined on an as-needed basis, to
445 support Pascal conventions or any other well-known target-independent
448 .. _visibilitystyles:
453 All Global Variables and Functions have one of the following visibility
456 "``default``" - Default style
457 On targets that use the ELF object file format, default visibility
458 means that the declaration is visible to other modules and, in
459 shared libraries, means that the declared entity may be overridden.
460 On Darwin, default visibility means that the declaration is visible
461 to other modules. Default visibility corresponds to "external
462 linkage" in the language.
463 "``hidden``" - Hidden style
464 Two declarations of an object with hidden visibility refer to the
465 same object if they are in the same shared object. Usually, hidden
466 visibility indicates that the symbol will not be placed into the
467 dynamic symbol table, so no other module (executable or shared
468 library) can reference it directly.
469 "``protected``" - Protected style
470 On ELF, protected visibility indicates that the symbol will be
471 placed in the dynamic symbol table, but that references within the
472 defining module will bind to the local symbol. That is, the symbol
473 cannot be overridden by another module.
475 A symbol with ``internal`` or ``private`` linkage must have ``default``
483 All Global Variables, Functions and Aliases can have one of the following
487 "``dllimport``" causes the compiler to reference a function or variable via
488 a global pointer to a pointer that is set up by the DLL exporting the
489 symbol. On Microsoft Windows targets, the pointer name is formed by
490 combining ``__imp_`` and the function or variable name.
492 "``dllexport``" causes the compiler to provide a global pointer to a pointer
493 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
494 Microsoft Windows targets, the pointer name is formed by combining
495 ``__imp_`` and the function or variable name. Since this storage class
496 exists for defining a dll interface, the compiler, assembler and linker know
497 it is externally referenced and must refrain from deleting the symbol.
501 Thread Local Storage Models
502 ---------------------------
504 A variable may be defined as ``thread_local``, which means that it will
505 not be shared by threads (each thread will have a separated copy of the
506 variable). Not all targets support thread-local variables. Optionally, a
507 TLS model may be specified:
510 For variables that are only used within the current shared library.
512 For variables in modules that will not be loaded dynamically.
514 For variables defined in the executable and only used within it.
516 If no explicit model is given, the "general dynamic" model is used.
518 The models correspond to the ELF TLS models; see `ELF Handling For
519 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
520 more information on under which circumstances the different models may
521 be used. The target may choose a different TLS model if the specified
522 model is not supported, or if a better choice of model can be made.
524 A model can also be specified in an alias, but then it only governs how
525 the alias is accessed. It will not have any effect in the aliasee.
527 For platforms without linker support of ELF TLS model, the -femulated-tls
528 flag can be used to generate GCC compatible emulated TLS code.
530 .. _runtime_preemption_model:
532 Runtime Preemption Specifiers
533 -----------------------------
535 Global variables, functions and aliases may have an optional runtime preemption
536 specifier. If a preemption specifier isn't given explicitly, then a
537 symbol is assumed to be ``dso_preemptable``.
540 Indicates that the function or variable may be replaced by a symbol from
541 outside the linkage unit at runtime.
544 The compiler may assume that a function or variable marked as ``dso_local``
545 will resolve to a symbol within the same linkage unit. Direct access will
546 be generated even if the definition is not within this compilation unit.
553 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
554 types <t_struct>`. Literal types are uniqued structurally, but identified types
555 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
556 to forward declare a type that is not yet available.
558 An example of an identified structure specification is:
562 %mytype = type { %mytype*, i32 }
564 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
565 literal types are uniqued in recent versions of LLVM.
569 Non-Integral Pointer Type
570 -------------------------
572 Note: non-integral pointer types are a work in progress, and they should be
573 considered experimental at this time.
575 LLVM IR optionally allows the frontend to denote pointers in certain address
576 spaces as "non-integral" via the :ref:`datalayout string<langref_datalayout>`.
577 Non-integral pointer types represent pointers that have an *unspecified* bitwise
578 representation; that is, the integral representation may be target dependent or
579 unstable (not backed by a fixed integer).
581 ``inttoptr`` instructions converting integers to non-integral pointer types are
582 ill-typed, and so are ``ptrtoint`` instructions converting values of
583 non-integral pointer types to integers. Vector versions of said instructions
584 are ill-typed as well.
591 Global variables define regions of memory allocated at compilation time
594 Global variable definitions must be initialized.
596 Global variables in other translation units can also be declared, in which
597 case they don't have an initializer.
599 Either global variable definitions or declarations may have an explicit section
600 to be placed in and may have an optional explicit alignment specified. If there
601 is a mismatch between the explicit or inferred section information for the
602 variable declaration and its definition the resulting behavior is undefined.
604 A variable may be defined as a global ``constant``, which indicates that
605 the contents of the variable will **never** be modified (enabling better
606 optimization, allowing the global data to be placed in the read-only
607 section of an executable, etc). Note that variables that need runtime
608 initialization cannot be marked ``constant`` as there is a store to the
611 LLVM explicitly allows *declarations* of global variables to be marked
612 constant, even if the final definition of the global is not. This
613 capability can be used to enable slightly better optimization of the
614 program, but requires the language definition to guarantee that
615 optimizations based on the 'constantness' are valid for the translation
616 units that do not include the definition.
618 As SSA values, global variables define pointer values that are in scope
619 (i.e. they dominate) all basic blocks in the program. Global variables
620 always define a pointer to their "content" type because they describe a
621 region of memory, and all memory objects in LLVM are accessed through
624 Global variables can be marked with ``unnamed_addr`` which indicates
625 that the address is not significant, only the content. Constants marked
626 like this can be merged with other constants if they have the same
627 initializer. Note that a constant with significant address *can* be
628 merged with a ``unnamed_addr`` constant, the result being a constant
629 whose address is significant.
631 If the ``local_unnamed_addr`` attribute is given, the address is known to
632 not be significant within the module.
634 A global variable may be declared to reside in a target-specific
635 numbered address space. For targets that support them, address spaces
636 may affect how optimizations are performed and/or what target
637 instructions are used to access the variable. The default address space
638 is zero. The address space qualifier must precede any other attributes.
640 LLVM allows an explicit section to be specified for globals. If the
641 target supports it, it will emit globals to the section specified.
642 Additionally, the global can placed in a comdat if the target has the necessary
645 External declarations may have an explicit section specified. Section
646 information is retained in LLVM IR for targets that make use of this
647 information. Attaching section information to an external declaration is an
648 assertion that its definition is located in the specified section. If the
649 definition is located in a different section, the behavior is undefined.
651 By default, global initializers are optimized by assuming that global
652 variables defined within the module are not modified from their
653 initial values before the start of the global initializer. This is
654 true even for variables potentially accessible from outside the
655 module, including those with external linkage or appearing in
656 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
657 by marking the variable with ``externally_initialized``.
659 An explicit alignment may be specified for a global, which must be a
660 power of 2. If not present, or if the alignment is set to zero, the
661 alignment of the global is set by the target to whatever it feels
662 convenient. If an explicit alignment is specified, the global is forced
663 to have exactly that alignment. Targets and optimizers are not allowed
664 to over-align the global if the global has an assigned section. In this
665 case, the extra alignment could be observable: for example, code could
666 assume that the globals are densely packed in their section and try to
667 iterate over them as an array, alignment padding would break this
668 iteration. The maximum alignment is ``1 << 29``.
670 Globals can also have a :ref:`DLL storage class <dllstorageclass>`,
671 an optional :ref:`runtime preemption specifier <runtime_preemption_model>`,
672 an optional :ref:`global attributes <glattrs>` and
673 an optional list of attached :ref:`metadata <metadata>`.
675 Variables and aliases can have a
676 :ref:`Thread Local Storage Model <tls_model>`.
680 @<GlobalVarName> = [Linkage] [PreemptionSpecifier] [Visibility]
681 [DLLStorageClass] [ThreadLocal]
682 [(unnamed_addr|local_unnamed_addr)] [AddrSpace]
683 [ExternallyInitialized]
684 <global | constant> <Type> [<InitializerConstant>]
685 [, section "name"] [, comdat [($name)]]
686 [, align <Alignment>] (, !name !N)*
688 For example, the following defines a global in a numbered address space
689 with an initializer, section, and alignment:
693 @G = addrspace(5) constant float 1.0, section "foo", align 4
695 The following example just declares a global variable
699 @G = external global i32
701 The following example defines a thread-local global with the
702 ``initialexec`` TLS model:
706 @G = thread_local(initialexec) global i32 0, align 4
708 .. _functionstructure:
713 LLVM function definitions consist of the "``define``" keyword, an
714 optional :ref:`linkage type <linkage>`, an optional :ref:`runtime preemption
715 specifier <runtime_preemption_model>`, an optional :ref:`visibility
716 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
717 an optional :ref:`calling convention <callingconv>`,
718 an optional ``unnamed_addr`` attribute, a return type, an optional
719 :ref:`parameter attribute <paramattrs>` for the return type, a function
720 name, a (possibly empty) argument list (each with optional :ref:`parameter
721 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
722 an optional section, an optional alignment,
723 an optional :ref:`comdat <langref_comdats>`,
724 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
725 an optional :ref:`prologue <prologuedata>`,
726 an optional :ref:`personality <personalityfn>`,
727 an optional list of attached :ref:`metadata <metadata>`,
728 an opening curly brace, a list of basic blocks, and a closing curly brace.
730 LLVM function declarations consist of the "``declare``" keyword, an
731 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility style
732 <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`, an
733 optional :ref:`calling convention <callingconv>`, an optional ``unnamed_addr``
734 or ``local_unnamed_addr`` attribute, a return type, an optional :ref:`parameter
735 attribute <paramattrs>` for the return type, a function name, a possibly
736 empty list of arguments, an optional alignment, an optional :ref:`garbage
737 collector name <gc>`, an optional :ref:`prefix <prefixdata>`, and an optional
738 :ref:`prologue <prologuedata>`.
740 A function definition contains a list of basic blocks, forming the CFG (Control
741 Flow Graph) for the function. Each basic block may optionally start with a label
742 (giving the basic block a symbol table entry), contains a list of instructions,
743 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
744 function return). If an explicit label is not provided, a block is assigned an
745 implicit numbered label, using the next value from the same counter as used for
746 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
747 entry block does not have an explicit label, it will be assigned label "%0",
748 then the first unnamed temporary in that block will be "%1", etc.
750 The first basic block in a function is special in two ways: it is
751 immediately executed on entrance to the function, and it is not allowed
752 to have predecessor basic blocks (i.e. there can not be any branches to
753 the entry block of a function). Because the block can have no
754 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
756 LLVM allows an explicit section to be specified for functions. If the
757 target supports it, it will emit functions to the section specified.
758 Additionally, the function can be placed in a COMDAT.
760 An explicit alignment may be specified for a function. If not present,
761 or if the alignment is set to zero, the alignment of the function is set
762 by the target to whatever it feels convenient. If an explicit alignment
763 is specified, the function is forced to have at least that much
764 alignment. All alignments must be a power of 2.
766 If the ``unnamed_addr`` attribute is given, the address is known to not
767 be significant and two identical functions can be merged.
769 If the ``local_unnamed_addr`` attribute is given, the address is known to
770 not be significant within the module.
774 define [linkage] [PreemptionSpecifier] [visibility] [DLLStorageClass]
776 <ResultType> @<FunctionName> ([argument list])
777 [(unnamed_addr|local_unnamed_addr)] [fn Attrs] [section "name"]
778 [comdat [($name)]] [align N] [gc] [prefix Constant]
779 [prologue Constant] [personality Constant] (!name !N)* { ... }
781 The argument list is a comma separated sequence of arguments where each
782 argument is of the following form:
786 <type> [parameter Attrs] [name]
794 Aliases, unlike function or variables, don't create any new data. They
795 are just a new symbol and metadata for an existing position.
797 Aliases have a name and an aliasee that is either a global value or a
800 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
801 :ref:`runtime preemption specifier <runtime_preemption_model>`, an optional
802 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
803 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
807 @<Name> = [Linkage] [PreemptionSpecifier] [Visibility] [DLLStorageClass] [ThreadLocal] [(unnamed_addr|local_unnamed_addr)] alias <AliaseeTy>, <AliaseeTy>* @<Aliasee>
809 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
810 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
811 might not correctly handle dropping a weak symbol that is aliased.
813 Aliases that are not ``unnamed_addr`` are guaranteed to have the same address as
814 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
817 If the ``local_unnamed_addr`` attribute is given, the address is known to
818 not be significant within the module.
820 Since aliases are only a second name, some restrictions apply, of which
821 some can only be checked when producing an object file:
823 * The expression defining the aliasee must be computable at assembly
824 time. Since it is just a name, no relocations can be used.
826 * No alias in the expression can be weak as the possibility of the
827 intermediate alias being overridden cannot be represented in an
830 * No global value in the expression can be a declaration, since that
831 would require a relocation, which is not possible.
838 IFuncs, like as aliases, don't create any new data or func. They are just a new
839 symbol that dynamic linker resolves at runtime by calling a resolver function.
841 IFuncs have a name and a resolver that is a function called by dynamic linker
842 that returns address of another function associated with the name.
844 IFunc may have an optional :ref:`linkage type <linkage>` and an optional
845 :ref:`visibility style <visibility>`.
849 @<Name> = [Linkage] [Visibility] ifunc <IFuncTy>, <ResolverTy>* @<Resolver>
857 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
859 Comdats have a name which represents the COMDAT key. All global objects that
860 specify this key will only end up in the final object file if the linker chooses
861 that key over some other key. Aliases are placed in the same COMDAT that their
862 aliasee computes to, if any.
864 Comdats have a selection kind to provide input on how the linker should
865 choose between keys in two different object files.
869 $<Name> = comdat SelectionKind
871 The selection kind must be one of the following:
874 The linker may choose any COMDAT key, the choice is arbitrary.
876 The linker may choose any COMDAT key but the sections must contain the
879 The linker will choose the section containing the largest COMDAT key.
881 The linker requires that only section with this COMDAT key exist.
883 The linker may choose any COMDAT key but the sections must contain the
886 Note that the Mach-O platform doesn't support COMDATs, and ELF and WebAssembly
887 only support ``any`` as a selection kind.
889 Here is an example of a COMDAT group where a function will only be selected if
890 the COMDAT key's section is the largest:
894 $foo = comdat largest
895 @foo = global i32 2, comdat($foo)
897 define void @bar() comdat($foo) {
901 As a syntactic sugar the ``$name`` can be omitted if the name is the same as
907 @foo = global i32 2, comdat
910 In a COFF object file, this will create a COMDAT section with selection kind
911 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
912 and another COMDAT section with selection kind
913 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
914 section and contains the contents of the ``@bar`` symbol.
916 There are some restrictions on the properties of the global object.
917 It, or an alias to it, must have the same name as the COMDAT group when
919 The contents and size of this object may be used during link-time to determine
920 which COMDAT groups get selected depending on the selection kind.
921 Because the name of the object must match the name of the COMDAT group, the
922 linkage of the global object must not be local; local symbols can get renamed
923 if a collision occurs in the symbol table.
925 The combined use of COMDATS and section attributes may yield surprising results.
932 @g1 = global i32 42, section "sec", comdat($foo)
933 @g2 = global i32 42, section "sec", comdat($bar)
935 From the object file perspective, this requires the creation of two sections
936 with the same name. This is necessary because both globals belong to different
937 COMDAT groups and COMDATs, at the object file level, are represented by
940 Note that certain IR constructs like global variables and functions may
941 create COMDATs in the object file in addition to any which are specified using
942 COMDAT IR. This arises when the code generator is configured to emit globals
943 in individual sections (e.g. when `-data-sections` or `-function-sections`
944 is supplied to `llc`).
946 .. _namedmetadatastructure:
951 Named metadata is a collection of metadata. :ref:`Metadata
952 nodes <metadata>` (but not metadata strings) are the only valid
953 operands for a named metadata.
955 #. Named metadata are represented as a string of characters with the
956 metadata prefix. The rules for metadata names are the same as for
957 identifiers, but quoted names are not allowed. ``"\xx"`` type escapes
958 are still valid, which allows any character to be part of a name.
962 ; Some unnamed metadata nodes, which are referenced by the named metadata.
967 !name = !{!0, !1, !2}
974 The return type and each parameter of a function type may have a set of
975 *parameter attributes* associated with them. Parameter attributes are
976 used to communicate additional information about the result or
977 parameters of a function. Parameter attributes are considered to be part
978 of the function, not of the function type, so functions with different
979 parameter attributes can have the same function type.
981 Parameter attributes are simple keywords that follow the type specified.
982 If multiple parameter attributes are needed, they are space separated.
987 declare i32 @printf(i8* noalias nocapture, ...)
988 declare i32 @atoi(i8 zeroext)
989 declare signext i8 @returns_signed_char()
991 Note that any attributes for the function result (``nounwind``,
992 ``readonly``) come immediately after the argument list.
994 Currently, only the following parameter attributes are defined:
997 This indicates to the code generator that the parameter or return
998 value should be zero-extended to the extent required by the target's
999 ABI by the caller (for a parameter) or the callee (for a return value).
1001 This indicates to the code generator that the parameter or return
1002 value should be sign-extended to the extent required by the target's
1003 ABI (which is usually 32-bits) by the caller (for a parameter) or
1004 the callee (for a return value).
1006 This indicates that this parameter or return value should be treated
1007 in a special target-dependent fashion while emitting code for
1008 a function call or return (usually, by putting it in a register as
1009 opposed to memory, though some targets use it to distinguish between
1010 two different kinds of registers). Use of this attribute is
1013 This indicates that the pointer parameter should really be passed by
1014 value to the function. The attribute implies that a hidden copy of
1015 the pointee is made between the caller and the callee, so the callee
1016 is unable to modify the value in the caller. This attribute is only
1017 valid on LLVM pointer arguments. It is generally used to pass
1018 structs and arrays by value, but is also valid on pointers to
1019 scalars. The copy is considered to belong to the caller not the
1020 callee (for example, ``readonly`` functions should not write to
1021 ``byval`` parameters). This is not a valid attribute for return
1024 The byval attribute also supports specifying an alignment with the
1025 align attribute. It indicates the alignment of the stack slot to
1026 form and the known alignment of the pointer specified to the call
1027 site. If the alignment is not specified, then the code generator
1028 makes a target-specific assumption.
1034 The ``inalloca`` argument attribute allows the caller to take the
1035 address of outgoing stack arguments. An ``inalloca`` argument must
1036 be a pointer to stack memory produced by an ``alloca`` instruction.
1037 The alloca, or argument allocation, must also be tagged with the
1038 inalloca keyword. Only the last argument may have the ``inalloca``
1039 attribute, and that argument is guaranteed to be passed in memory.
1041 An argument allocation may be used by a call at most once because
1042 the call may deallocate it. The ``inalloca`` attribute cannot be
1043 used in conjunction with other attributes that affect argument
1044 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
1045 ``inalloca`` attribute also disables LLVM's implicit lowering of
1046 large aggregate return values, which means that frontend authors
1047 must lower them with ``sret`` pointers.
1049 When the call site is reached, the argument allocation must have
1050 been the most recent stack allocation that is still live, or the
1051 results are undefined. It is possible to allocate additional stack
1052 space after an argument allocation and before its call site, but it
1053 must be cleared off with :ref:`llvm.stackrestore
1054 <int_stackrestore>`.
1056 See :doc:`InAlloca` for more information on how to use this
1060 This indicates that the pointer parameter specifies the address of a
1061 structure that is the return value of the function in the source
1062 program. This pointer must be guaranteed by the caller to be valid:
1063 loads and stores to the structure may be assumed by the callee not
1064 to trap and to be properly aligned. This is not a valid attribute
1070 This indicates that the pointer value may be assumed by the optimizer to
1071 have the specified alignment.
1073 Note that this attribute has additional semantics when combined with the
1074 ``byval`` attribute.
1079 This indicates that objects accessed via pointer values
1080 :ref:`based <pointeraliasing>` on the argument or return value are not also
1081 accessed, during the execution of the function, via pointer values not
1082 *based* on the argument or return value. The attribute on a return value
1083 also has additional semantics described below. The caller shares the
1084 responsibility with the callee for ensuring that these requirements are met.
1085 For further details, please see the discussion of the NoAlias response in
1086 :ref:`alias analysis <Must, May, or No>`.
1088 Note that this definition of ``noalias`` is intentionally similar
1089 to the definition of ``restrict`` in C99 for function arguments.
1091 For function return values, C99's ``restrict`` is not meaningful,
1092 while LLVM's ``noalias`` is. Furthermore, the semantics of the ``noalias``
1093 attribute on return values are stronger than the semantics of the attribute
1094 when used on function arguments. On function return values, the ``noalias``
1095 attribute indicates that the function acts like a system memory allocation
1096 function, returning a pointer to allocated storage disjoint from the
1097 storage for any other object accessible to the caller.
1100 This indicates that the callee does not make any copies of the
1101 pointer that outlive the callee itself. This is not a valid
1102 attribute for return values. Addresses used in volatile operations
1103 are considered to be captured.
1108 This indicates that the pointer parameter can be excised using the
1109 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
1110 attribute for return values and can only be applied to one parameter.
1113 This indicates that the function always returns the argument as its return
1114 value. This is a hint to the optimizer and code generator used when
1115 generating the caller, allowing value propagation, tail call optimization,
1116 and omission of register saves and restores in some cases; it is not
1117 checked or enforced when generating the callee. The parameter and the
1118 function return type must be valid operands for the
1119 :ref:`bitcast instruction <i_bitcast>`. This is not a valid attribute for
1120 return values and can only be applied to one parameter.
1123 This indicates that the parameter or return pointer is not null. This
1124 attribute may only be applied to pointer typed parameters. This is not
1125 checked or enforced by LLVM, the caller must ensure that the pointer
1126 passed in is non-null, or the callee must ensure that the returned pointer
1129 ``dereferenceable(<n>)``
1130 This indicates that the parameter or return pointer is dereferenceable. This
1131 attribute may only be applied to pointer typed parameters. A pointer that
1132 is dereferenceable can be loaded from speculatively without a risk of
1133 trapping. The number of bytes known to be dereferenceable must be provided
1134 in parentheses. It is legal for the number of bytes to be less than the
1135 size of the pointee type. The ``nonnull`` attribute does not imply
1136 dereferenceability (consider a pointer to one element past the end of an
1137 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
1138 ``addrspace(0)`` (which is the default address space).
1140 ``dereferenceable_or_null(<n>)``
1141 This indicates that the parameter or return value isn't both
1142 non-null and non-dereferenceable (up to ``<n>`` bytes) at the same
1143 time. All non-null pointers tagged with
1144 ``dereferenceable_or_null(<n>)`` are ``dereferenceable(<n>)``.
1145 For address space 0 ``dereferenceable_or_null(<n>)`` implies that
1146 a pointer is exactly one of ``dereferenceable(<n>)`` or ``null``,
1147 and in other address spaces ``dereferenceable_or_null(<n>)``
1148 implies that a pointer is at least one of ``dereferenceable(<n>)``
1149 or ``null`` (i.e. it may be both ``null`` and
1150 ``dereferenceable(<n>)``). This attribute may only be applied to
1151 pointer typed parameters.
1154 This indicates that the parameter is the self/context parameter. This is not
1155 a valid attribute for return values and can only be applied to one
1159 This attribute is motivated to model and optimize Swift error handling. It
1160 can be applied to a parameter with pointer to pointer type or a
1161 pointer-sized alloca. At the call site, the actual argument that corresponds
1162 to a ``swifterror`` parameter has to come from a ``swifterror`` alloca or
1163 the ``swifterror`` parameter of the caller. A ``swifterror`` value (either
1164 the parameter or the alloca) can only be loaded and stored from, or used as
1165 a ``swifterror`` argument. This is not a valid attribute for return values
1166 and can only be applied to one parameter.
1168 These constraints allow the calling convention to optimize access to
1169 ``swifterror`` variables by associating them with a specific register at
1170 call boundaries rather than placing them in memory. Since this does change
1171 the calling convention, a function which uses the ``swifterror`` attribute
1172 on a parameter is not ABI-compatible with one which does not.
1174 These constraints also allow LLVM to assume that a ``swifterror`` argument
1175 does not alias any other memory visible within a function and that a
1176 ``swifterror`` alloca passed as an argument does not escape.
1180 Garbage Collector Strategy Names
1181 --------------------------------
1183 Each function may specify a garbage collector strategy name, which is simply a
1186 .. code-block:: llvm
1188 define void @f() gc "name" { ... }
1190 The supported values of *name* includes those :ref:`built in to LLVM
1191 <builtin-gc-strategies>` and any provided by loaded plugins. Specifying a GC
1192 strategy will cause the compiler to alter its output in order to support the
1193 named garbage collection algorithm. Note that LLVM itself does not contain a
1194 garbage collector, this functionality is restricted to generating machine code
1195 which can interoperate with a collector provided externally.
1202 Prefix data is data associated with a function which the code
1203 generator will emit immediately before the function's entrypoint.
1204 The purpose of this feature is to allow frontends to associate
1205 language-specific runtime metadata with specific functions and make it
1206 available through the function pointer while still allowing the
1207 function pointer to be called.
1209 To access the data for a given function, a program may bitcast the
1210 function pointer to a pointer to the constant's type and dereference
1211 index -1. This implies that the IR symbol points just past the end of
1212 the prefix data. For instance, take the example of a function annotated
1213 with a single ``i32``,
1215 .. code-block:: llvm
1217 define void @f() prefix i32 123 { ... }
1219 The prefix data can be referenced as,
1221 .. code-block:: llvm
1223 %0 = bitcast void* () @f to i32*
1224 %a = getelementptr inbounds i32, i32* %0, i32 -1
1225 %b = load i32, i32* %a
1227 Prefix data is laid out as if it were an initializer for a global variable
1228 of the prefix data's type. The function will be placed such that the
1229 beginning of the prefix data is aligned. This means that if the size
1230 of the prefix data is not a multiple of the alignment size, the
1231 function's entrypoint will not be aligned. If alignment of the
1232 function's entrypoint is desired, padding must be added to the prefix
1235 A function may have prefix data but no body. This has similar semantics
1236 to the ``available_externally`` linkage in that the data may be used by the
1237 optimizers but will not be emitted in the object file.
1244 The ``prologue`` attribute allows arbitrary code (encoded as bytes) to
1245 be inserted prior to the function body. This can be used for enabling
1246 function hot-patching and instrumentation.
1248 To maintain the semantics of ordinary function calls, the prologue data must
1249 have a particular format. Specifically, it must begin with a sequence of
1250 bytes which decode to a sequence of machine instructions, valid for the
1251 module's target, which transfer control to the point immediately succeeding
1252 the prologue data, without performing any other visible action. This allows
1253 the inliner and other passes to reason about the semantics of the function
1254 definition without needing to reason about the prologue data. Obviously this
1255 makes the format of the prologue data highly target dependent.
1257 A trivial example of valid prologue data for the x86 architecture is ``i8 144``,
1258 which encodes the ``nop`` instruction:
1260 .. code-block:: text
1262 define void @f() prologue i8 144 { ... }
1264 Generally prologue data can be formed by encoding a relative branch instruction
1265 which skips the metadata, as in this example of valid prologue data for the
1266 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1268 .. code-block:: text
1270 %0 = type <{ i8, i8, i8* }>
1272 define void @f() prologue %0 <{ i8 235, i8 8, i8* @md}> { ... }
1274 A function may have prologue data but no body. This has similar semantics
1275 to the ``available_externally`` linkage in that the data may be used by the
1276 optimizers but will not be emitted in the object file.
1280 Personality Function
1281 --------------------
1283 The ``personality`` attribute permits functions to specify what function
1284 to use for exception handling.
1291 Attribute groups are groups of attributes that are referenced by objects within
1292 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1293 functions will use the same set of attributes. In the degenerative case of a
1294 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1295 group will capture the important command line flags used to build that file.
1297 An attribute group is a module-level object. To use an attribute group, an
1298 object references the attribute group's ID (e.g. ``#37``). An object may refer
1299 to more than one attribute group. In that situation, the attributes from the
1300 different groups are merged.
1302 Here is an example of attribute groups for a function that should always be
1303 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1305 .. code-block:: llvm
1307 ; Target-independent attributes:
1308 attributes #0 = { alwaysinline alignstack=4 }
1310 ; Target-dependent attributes:
1311 attributes #1 = { "no-sse" }
1313 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1314 define void @f() #0 #1 { ... }
1321 Function attributes are set to communicate additional information about
1322 a function. Function attributes are considered to be part of the
1323 function, not of the function type, so functions with different function
1324 attributes can have the same function type.
1326 Function attributes are simple keywords that follow the type specified.
1327 If multiple attributes are needed, they are space separated. For
1330 .. code-block:: llvm
1332 define void @f() noinline { ... }
1333 define void @f() alwaysinline { ... }
1334 define void @f() alwaysinline optsize { ... }
1335 define void @f() optsize { ... }
1338 This attribute indicates that, when emitting the prologue and
1339 epilogue, the backend should forcibly align the stack pointer.
1340 Specify the desired alignment, which must be a power of two, in
1342 ``allocsize(<EltSizeParam>[, <NumEltsParam>])``
1343 This attribute indicates that the annotated function will always return at
1344 least a given number of bytes (or null). Its arguments are zero-indexed
1345 parameter numbers; if one argument is provided, then it's assumed that at
1346 least ``CallSite.Args[EltSizeParam]`` bytes will be available at the
1347 returned pointer. If two are provided, then it's assumed that
1348 ``CallSite.Args[EltSizeParam] * CallSite.Args[NumEltsParam]`` bytes are
1349 available. The referenced parameters must be integer types. No assumptions
1350 are made about the contents of the returned block of memory.
1352 This attribute indicates that the inliner should attempt to inline
1353 this function into callers whenever possible, ignoring any active
1354 inlining size threshold for this caller.
1356 This indicates that the callee function at a call site should be
1357 recognized as a built-in function, even though the function's declaration
1358 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1359 direct calls to functions that are declared with the ``nobuiltin``
1362 This attribute indicates that this function is rarely called. When
1363 computing edge weights, basic blocks post-dominated by a cold
1364 function call are also considered to be cold; and, thus, given low
1367 In some parallel execution models, there exist operations that cannot be
1368 made control-dependent on any additional values. We call such operations
1369 ``convergent``, and mark them with this attribute.
1371 The ``convergent`` attribute may appear on functions or call/invoke
1372 instructions. When it appears on a function, it indicates that calls to
1373 this function should not be made control-dependent on additional values.
1374 For example, the intrinsic ``llvm.nvvm.barrier0`` is ``convergent``, so
1375 calls to this intrinsic cannot be made control-dependent on additional
1378 When it appears on a call/invoke, the ``convergent`` attribute indicates
1379 that we should treat the call as though we're calling a convergent
1380 function. This is particularly useful on indirect calls; without this we
1381 may treat such calls as though the target is non-convergent.
1383 The optimizer may remove the ``convergent`` attribute on functions when it
1384 can prove that the function does not execute any convergent operations.
1385 Similarly, the optimizer may remove ``convergent`` on calls/invokes when it
1386 can prove that the call/invoke cannot call a convergent function.
1387 ``inaccessiblememonly``
1388 This attribute indicates that the function may only access memory that
1389 is not accessible by the module being compiled. This is a weaker form
1391 ``inaccessiblemem_or_argmemonly``
1392 This attribute indicates that the function may only access memory that is
1393 either not accessible by the module being compiled, or is pointed to
1394 by its pointer arguments. This is a weaker form of ``argmemonly``
1396 This attribute indicates that the source code contained a hint that
1397 inlining this function is desirable (such as the "inline" keyword in
1398 C/C++). It is just a hint; it imposes no requirements on the
1401 This attribute indicates that the function should be added to a
1402 jump-instruction table at code-generation time, and that all address-taken
1403 references to this function should be replaced with a reference to the
1404 appropriate jump-instruction-table function pointer. Note that this creates
1405 a new pointer for the original function, which means that code that depends
1406 on function-pointer identity can break. So, any function annotated with
1407 ``jumptable`` must also be ``unnamed_addr``.
1409 This attribute suggests that optimization passes and code generator
1410 passes make choices that keep the code size of this function as small
1411 as possible and perform optimizations that may sacrifice runtime
1412 performance in order to minimize the size of the generated code.
1414 This attribute disables prologue / epilogue emission for the
1415 function. This can have very system-specific consequences.
1417 When this attribute is set to true, the jump tables and lookup tables that
1418 can be generated from a switch case lowering are disabled.
1420 This indicates that the callee function at a call site is not recognized as
1421 a built-in function. LLVM will retain the original call and not replace it
1422 with equivalent code based on the semantics of the built-in function, unless
1423 the call site uses the ``builtin`` attribute. This is valid at call sites
1424 and on function declarations and definitions.
1426 This attribute indicates that calls to the function cannot be
1427 duplicated. A call to a ``noduplicate`` function may be moved
1428 within its parent function, but may not be duplicated within
1429 its parent function.
1431 A function containing a ``noduplicate`` call may still
1432 be an inlining candidate, provided that the call is not
1433 duplicated by inlining. That implies that the function has
1434 internal linkage and only has one call site, so the original
1435 call is dead after inlining.
1437 This attributes disables implicit floating-point instructions.
1439 This attribute indicates that the inliner should never inline this
1440 function in any situation. This attribute may not be used together
1441 with the ``alwaysinline`` attribute.
1443 This attribute suppresses lazy symbol binding for the function. This
1444 may make calls to the function faster, at the cost of extra program
1445 startup time if the function is not called during program startup.
1447 This attribute indicates that the code generator should not use a
1448 red zone, even if the target-specific ABI normally permits it.
1450 This function attribute indicates that the function never returns
1451 normally. This produces undefined behavior at runtime if the
1452 function ever does dynamically return.
1454 This function attribute indicates that the function does not call itself
1455 either directly or indirectly down any possible call path. This produces
1456 undefined behavior at runtime if the function ever does recurse.
1458 This function attribute indicates that the function never raises an
1459 exception. If the function does raise an exception, its runtime
1460 behavior is undefined. However, functions marked nounwind may still
1461 trap or generate asynchronous exceptions. Exception handling schemes
1462 that are recognized by LLVM to handle asynchronous exceptions, such
1463 as SEH, will still provide their implementation defined semantics.
1465 This attribute indicates that this function should be optimized
1466 for maximum fuzzing signal.
1468 This function attribute indicates that most optimization passes will skip
1469 this function, with the exception of interprocedural optimization passes.
1470 Code generation defaults to the "fast" instruction selector.
1471 This attribute cannot be used together with the ``alwaysinline``
1472 attribute; this attribute is also incompatible
1473 with the ``minsize`` attribute and the ``optsize`` attribute.
1475 This attribute requires the ``noinline`` attribute to be specified on
1476 the function as well, so the function is never inlined into any caller.
1477 Only functions with the ``alwaysinline`` attribute are valid
1478 candidates for inlining into the body of this function.
1480 This attribute suggests that optimization passes and code generator
1481 passes make choices that keep the code size of this function low,
1482 and otherwise do optimizations specifically to reduce code size as
1483 long as they do not significantly impact runtime performance.
1484 ``"patchable-function"``
1485 This attribute tells the code generator that the code
1486 generated for this function needs to follow certain conventions that
1487 make it possible for a runtime function to patch over it later.
1488 The exact effect of this attribute depends on its string value,
1489 for which there currently is one legal possibility:
1491 * ``"prologue-short-redirect"`` - This style of patchable
1492 function is intended to support patching a function prologue to
1493 redirect control away from the function in a thread safe
1494 manner. It guarantees that the first instruction of the
1495 function will be large enough to accommodate a short jump
1496 instruction, and will be sufficiently aligned to allow being
1497 fully changed via an atomic compare-and-swap instruction.
1498 While the first requirement can be satisfied by inserting large
1499 enough NOP, LLVM can and will try to re-purpose an existing
1500 instruction (i.e. one that would have to be emitted anyway) as
1501 the patchable instruction larger than a short jump.
1503 ``"prologue-short-redirect"`` is currently only supported on
1506 This attribute by itself does not imply restrictions on
1507 inter-procedural optimizations. All of the semantic effects the
1508 patching may have to be separately conveyed via the linkage type.
1510 This attribute indicates that the function will trigger a guard region
1511 in the end of the stack. It ensures that accesses to the stack must be
1512 no further apart than the size of the guard region to a previous
1513 access of the stack. It takes one required string value, the name of
1514 the stack probing function that will be called.
1516 If a function that has a ``"probe-stack"`` attribute is inlined into
1517 a function with another ``"probe-stack"`` attribute, the resulting
1518 function has the ``"probe-stack"`` attribute of the caller. If a
1519 function that has a ``"probe-stack"`` attribute is inlined into a
1520 function that has no ``"probe-stack"`` attribute at all, the resulting
1521 function has the ``"probe-stack"`` attribute of the callee.
1523 On a function, this attribute indicates that the function computes its
1524 result (or decides to unwind an exception) based strictly on its arguments,
1525 without dereferencing any pointer arguments or otherwise accessing
1526 any mutable state (e.g. memory, control registers, etc) visible to
1527 caller functions. It does not write through any pointer arguments
1528 (including ``byval`` arguments) and never changes any state visible
1529 to callers. This means while it cannot unwind exceptions by calling
1530 the ``C++`` exception throwing methods (since they write to memory), there may
1531 be non-``C++`` mechanisms that throw exceptions without writing to LLVM
1534 On an argument, this attribute indicates that the function does not
1535 dereference that pointer argument, even though it may read or write the
1536 memory that the pointer points to if accessed through other pointers.
1538 On a function, this attribute indicates that the function does not write
1539 through any pointer arguments (including ``byval`` arguments) or otherwise
1540 modify any state (e.g. memory, control registers, etc) visible to
1541 caller functions. It may dereference pointer arguments and read
1542 state that may be set in the caller. A readonly function always
1543 returns the same value (or unwinds an exception identically) when
1544 called with the same set of arguments and global state. This means while it
1545 cannot unwind exceptions by calling the ``C++`` exception throwing methods
1546 (since they write to memory), there may be non-``C++`` mechanisms that throw
1547 exceptions without writing to LLVM visible memory.
1549 On an argument, this attribute indicates that the function does not write
1550 through this pointer argument, even though it may write to the memory that
1551 the pointer points to.
1552 ``"stack-probe-size"``
1553 This attribute controls the behavior of stack probes: either
1554 the ``"probe-stack"`` attribute, or ABI-required stack probes, if any.
1555 It defines the size of the guard region. It ensures that if the function
1556 may use more stack space than the size of the guard region, stack probing
1557 sequence will be emitted. It takes one required integer value, which
1560 If a function that has a ``"stack-probe-size"`` attribute is inlined into
1561 a function with another ``"stack-probe-size"`` attribute, the resulting
1562 function has the ``"stack-probe-size"`` attribute that has the lower
1563 numeric value. If a function that has a ``"stack-probe-size"`` attribute is
1564 inlined into a function that has no ``"stack-probe-size"`` attribute
1565 at all, the resulting function has the ``"stack-probe-size"`` attribute
1567 ``"no-stack-arg-probe"``
1568 This attribute disables ABI-required stack probes, if any.
1570 On a function, this attribute indicates that the function may write to but
1571 does not read from memory.
1573 On an argument, this attribute indicates that the function may write to but
1574 does not read through this pointer argument (even though it may read from
1575 the memory that the pointer points to).
1577 This attribute indicates that the only memory accesses inside function are
1578 loads and stores from objects pointed to by its pointer-typed arguments,
1579 with arbitrary offsets. Or in other words, all memory operations in the
1580 function can refer to memory only using pointers based on its function
1582 Note that ``argmemonly`` can be used together with ``readonly`` attribute
1583 in order to specify that function reads only from its arguments.
1585 This attribute indicates that this function can return twice. The C
1586 ``setjmp`` is an example of such a function. The compiler disables
1587 some optimizations (like tail calls) in the caller of these
1590 This attribute indicates that
1591 `SafeStack <http://clang.llvm.org/docs/SafeStack.html>`_
1592 protection is enabled for this function.
1594 If a function that has a ``safestack`` attribute is inlined into a
1595 function that doesn't have a ``safestack`` attribute or which has an
1596 ``ssp``, ``sspstrong`` or ``sspreq`` attribute, then the resulting
1597 function will have a ``safestack`` attribute.
1598 ``sanitize_address``
1599 This attribute indicates that AddressSanitizer checks
1600 (dynamic address safety analysis) are enabled for this function.
1602 This attribute indicates that MemorySanitizer checks (dynamic detection
1603 of accesses to uninitialized memory) are enabled for this function.
1605 This attribute indicates that ThreadSanitizer checks
1606 (dynamic thread safety analysis) are enabled for this function.
1607 ``sanitize_hwaddress``
1608 This attribute indicates that HWAddressSanitizer checks
1609 (dynamic address safety analysis based on tagged pointers) are enabled for
1612 This function attribute indicates that the function does not have any
1613 effects besides calculating its result and does not have undefined behavior.
1614 Note that ``speculatable`` is not enough to conclude that along any
1615 particular execution path the number of calls to this function will not be
1616 externally observable. This attribute is only valid on functions
1617 and declarations, not on individual call sites. If a function is
1618 incorrectly marked as speculatable and really does exhibit
1619 undefined behavior, the undefined behavior may be observed even
1620 if the call site is dead code.
1623 This attribute indicates that the function should emit a stack
1624 smashing protector. It is in the form of a "canary" --- a random value
1625 placed on the stack before the local variables that's checked upon
1626 return from the function to see if it has been overwritten. A
1627 heuristic is used to determine if a function needs stack protectors
1628 or not. The heuristic used will enable protectors for functions with:
1630 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1631 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1632 - Calls to alloca() with variable sizes or constant sizes greater than
1633 ``ssp-buffer-size``.
1635 Variables that are identified as requiring a protector will be arranged
1636 on the stack such that they are adjacent to the stack protector guard.
1638 If a function that has an ``ssp`` attribute is inlined into a
1639 function that doesn't have an ``ssp`` attribute, then the resulting
1640 function will have an ``ssp`` attribute.
1642 This attribute indicates that the function should *always* emit a
1643 stack smashing protector. This overrides the ``ssp`` function
1646 Variables that are identified as requiring a protector will be arranged
1647 on the stack such that they are adjacent to the stack protector guard.
1648 The specific layout rules are:
1650 #. Large arrays and structures containing large arrays
1651 (``>= ssp-buffer-size``) are closest to the stack protector.
1652 #. Small arrays and structures containing small arrays
1653 (``< ssp-buffer-size``) are 2nd closest to the protector.
1654 #. Variables that have had their address taken are 3rd closest to the
1657 If a function that has an ``sspreq`` attribute is inlined into a
1658 function that doesn't have an ``sspreq`` attribute or which has an
1659 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1660 an ``sspreq`` attribute.
1662 This attribute indicates that the function should emit a stack smashing
1663 protector. This attribute causes a strong heuristic to be used when
1664 determining if a function needs stack protectors. The strong heuristic
1665 will enable protectors for functions with:
1667 - Arrays of any size and type
1668 - Aggregates containing an array of any size and type.
1669 - Calls to alloca().
1670 - Local variables that have had their address taken.
1672 Variables that are identified as requiring a protector will be arranged
1673 on the stack such that they are adjacent to the stack protector guard.
1674 The specific layout rules are:
1676 #. Large arrays and structures containing large arrays
1677 (``>= ssp-buffer-size``) are closest to the stack protector.
1678 #. Small arrays and structures containing small arrays
1679 (``< ssp-buffer-size``) are 2nd closest to the protector.
1680 #. Variables that have had their address taken are 3rd closest to the
1683 This overrides the ``ssp`` function attribute.
1685 If a function that has an ``sspstrong`` attribute is inlined into a
1686 function that doesn't have an ``sspstrong`` attribute, then the
1687 resulting function will have an ``sspstrong`` attribute.
1689 This attribute indicates that the function was called from a scope that
1690 requires strict floating-point semantics. LLVM will not attempt any
1691 optimizations that require assumptions about the floating-point rounding
1692 mode or that might alter the state of floating-point status flags that
1693 might otherwise be set or cleared by calling this function.
1695 This attribute indicates that the function will delegate to some other
1696 function with a tail call. The prototype of a thunk should not be used for
1697 optimization purposes. The caller is expected to cast the thunk prototype to
1698 match the thunk target prototype.
1700 This attribute indicates that the ABI being targeted requires that
1701 an unwind table entry be produced for this function even if we can
1702 show that no exceptions passes by it. This is normally the case for
1703 the ELF x86-64 abi, but it can be disabled for some compilation
1706 This attribute indicates that no control-flow check will be perfomed on
1707 the attributed entity. It disables -fcf-protection=<> for a specific
1708 entity to fine grain the HW control flow protection mechanism. The flag
1709 is target independant and currently appertains to a function or function
1712 This attribute indicates that the ShadowCallStack checks are enabled for
1713 the function. The instrumentation checks that the return address for the
1714 function has not changed between the function prolog and eiplog. It is
1715 currently x86_64-specific.
1722 Attributes may be set to communicate additional information about a global variable.
1723 Unlike :ref:`function attributes <fnattrs>`, attributes on a global variable
1724 are grouped into a single :ref:`attribute group <attrgrp>`.
1731 Operand bundles are tagged sets of SSA values that can be associated
1732 with certain LLVM instructions (currently only ``call`` s and
1733 ``invoke`` s). In a way they are like metadata, but dropping them is
1734 incorrect and will change program semantics.
1738 operand bundle set ::= '[' operand bundle (, operand bundle )* ']'
1739 operand bundle ::= tag '(' [ bundle operand ] (, bundle operand )* ')'
1740 bundle operand ::= SSA value
1741 tag ::= string constant
1743 Operand bundles are **not** part of a function's signature, and a
1744 given function may be called from multiple places with different kinds
1745 of operand bundles. This reflects the fact that the operand bundles
1746 are conceptually a part of the ``call`` (or ``invoke``), not the
1747 callee being dispatched to.
1749 Operand bundles are a generic mechanism intended to support
1750 runtime-introspection-like functionality for managed languages. While
1751 the exact semantics of an operand bundle depend on the bundle tag,
1752 there are certain limitations to how much the presence of an operand
1753 bundle can influence the semantics of a program. These restrictions
1754 are described as the semantics of an "unknown" operand bundle. As
1755 long as the behavior of an operand bundle is describable within these
1756 restrictions, LLVM does not need to have special knowledge of the
1757 operand bundle to not miscompile programs containing it.
1759 - The bundle operands for an unknown operand bundle escape in unknown
1760 ways before control is transferred to the callee or invokee.
1761 - Calls and invokes with operand bundles have unknown read / write
1762 effect on the heap on entry and exit (even if the call target is
1763 ``readnone`` or ``readonly``), unless they're overridden with
1764 callsite specific attributes.
1765 - An operand bundle at a call site cannot change the implementation
1766 of the called function. Inter-procedural optimizations work as
1767 usual as long as they take into account the first two properties.
1769 More specific types of operand bundles are described below.
1771 .. _deopt_opbundles:
1773 Deoptimization Operand Bundles
1774 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1776 Deoptimization operand bundles are characterized by the ``"deopt"``
1777 operand bundle tag. These operand bundles represent an alternate
1778 "safe" continuation for the call site they're attached to, and can be
1779 used by a suitable runtime to deoptimize the compiled frame at the
1780 specified call site. There can be at most one ``"deopt"`` operand
1781 bundle attached to a call site. Exact details of deoptimization is
1782 out of scope for the language reference, but it usually involves
1783 rewriting a compiled frame into a set of interpreted frames.
1785 From the compiler's perspective, deoptimization operand bundles make
1786 the call sites they're attached to at least ``readonly``. They read
1787 through all of their pointer typed operands (even if they're not
1788 otherwise escaped) and the entire visible heap. Deoptimization
1789 operand bundles do not capture their operands except during
1790 deoptimization, in which case control will not be returned to the
1793 The inliner knows how to inline through calls that have deoptimization
1794 operand bundles. Just like inlining through a normal call site
1795 involves composing the normal and exceptional continuations, inlining
1796 through a call site with a deoptimization operand bundle needs to
1797 appropriately compose the "safe" deoptimization continuation. The
1798 inliner does this by prepending the parent's deoptimization
1799 continuation to every deoptimization continuation in the inlined body.
1800 E.g. inlining ``@f`` into ``@g`` in the following example
1802 .. code-block:: llvm
1805 call void @x() ;; no deopt state
1806 call void @y() [ "deopt"(i32 10) ]
1807 call void @y() [ "deopt"(i32 10), "unknown"(i8* null) ]
1812 call void @f() [ "deopt"(i32 20) ]
1818 .. code-block:: llvm
1821 call void @x() ;; still no deopt state
1822 call void @y() [ "deopt"(i32 20, i32 10) ]
1823 call void @y() [ "deopt"(i32 20, i32 10), "unknown"(i8* null) ]
1827 It is the frontend's responsibility to structure or encode the
1828 deoptimization state in a way that syntactically prepending the
1829 caller's deoptimization state to the callee's deoptimization state is
1830 semantically equivalent to composing the caller's deoptimization
1831 continuation after the callee's deoptimization continuation.
1835 Funclet Operand Bundles
1836 ^^^^^^^^^^^^^^^^^^^^^^^
1838 Funclet operand bundles are characterized by the ``"funclet"``
1839 operand bundle tag. These operand bundles indicate that a call site
1840 is within a particular funclet. There can be at most one
1841 ``"funclet"`` operand bundle attached to a call site and it must have
1842 exactly one bundle operand.
1844 If any funclet EH pads have been "entered" but not "exited" (per the
1845 `description in the EH doc\ <ExceptionHandling.html#wineh-constraints>`_),
1846 it is undefined behavior to execute a ``call`` or ``invoke`` which:
1848 * does not have a ``"funclet"`` bundle and is not a ``call`` to a nounwind
1850 * has a ``"funclet"`` bundle whose operand is not the most-recently-entered
1851 not-yet-exited funclet EH pad.
1853 Similarly, if no funclet EH pads have been entered-but-not-yet-exited,
1854 executing a ``call`` or ``invoke`` with a ``"funclet"`` bundle is undefined behavior.
1856 GC Transition Operand Bundles
1857 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1859 GC transition operand bundles are characterized by the
1860 ``"gc-transition"`` operand bundle tag. These operand bundles mark a
1861 call as a transition between a function with one GC strategy to a
1862 function with a different GC strategy. If coordinating the transition
1863 between GC strategies requires additional code generation at the call
1864 site, these bundles may contain any values that are needed by the
1865 generated code. For more details, see :ref:`GC Transitions
1866 <gc_transition_args>`.
1870 Module-Level Inline Assembly
1871 ----------------------------
1873 Modules may contain "module-level inline asm" blocks, which corresponds
1874 to the GCC "file scope inline asm" blocks. These blocks are internally
1875 concatenated by LLVM and treated as a single unit, but may be separated
1876 in the ``.ll`` file if desired. The syntax is very simple:
1878 .. code-block:: llvm
1880 module asm "inline asm code goes here"
1881 module asm "more can go here"
1883 The strings can contain any character by escaping non-printable
1884 characters. The escape sequence used is simply "\\xx" where "xx" is the
1885 two digit hex code for the number.
1887 Note that the assembly string *must* be parseable by LLVM's integrated assembler
1888 (unless it is disabled), even when emitting a ``.s`` file.
1890 .. _langref_datalayout:
1895 A module may specify a target specific data layout string that specifies
1896 how data is to be laid out in memory. The syntax for the data layout is
1899 .. code-block:: llvm
1901 target datalayout = "layout specification"
1903 The *layout specification* consists of a list of specifications
1904 separated by the minus sign character ('-'). Each specification starts
1905 with a letter and may include other information after the letter to
1906 define some aspect of the data layout. The specifications accepted are
1910 Specifies that the target lays out data in big-endian form. That is,
1911 the bits with the most significance have the lowest address
1914 Specifies that the target lays out data in little-endian form. That
1915 is, the bits with the least significance have the lowest address
1918 Specifies the natural alignment of the stack in bits. Alignment
1919 promotion of stack variables is limited to the natural stack
1920 alignment to avoid dynamic stack realignment. The stack alignment
1921 must be a multiple of 8-bits. If omitted, the natural stack
1922 alignment defaults to "unspecified", which does not prevent any
1923 alignment promotions.
1924 ``P<address space>``
1925 Specifies the address space that corresponds to program memory.
1926 Harvard architectures can use this to specify what space LLVM
1927 should place things such as functions into. If omitted, the
1928 program memory space defaults to the default address space of 0,
1929 which corresponds to a Von Neumann architecture that has code
1930 and data in the same space.
1931 ``A<address space>``
1932 Specifies the address space of objects created by '``alloca``'.
1933 Defaults to the default address space of 0.
1934 ``p[n]:<size>:<abi>:<pref>:<idx>``
1935 This specifies the *size* of a pointer and its ``<abi>`` and
1936 ``<pref>``\erred alignments for address space ``n``. The fourth parameter
1937 ``<idx>`` is a size of index that used for address calculation. If not
1938 specified, the default index size is equal to the pointer size. All sizes
1939 are in bits. The address space, ``n``, is optional, and if not specified,
1940 denotes the default address space 0. The value of ``n`` must be
1941 in the range [1,2^23).
1942 ``i<size>:<abi>:<pref>``
1943 This specifies the alignment for an integer type of a given bit
1944 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1945 ``v<size>:<abi>:<pref>``
1946 This specifies the alignment for a vector type of a given bit
1948 ``f<size>:<abi>:<pref>``
1949 This specifies the alignment for a floating-point type of a given bit
1950 ``<size>``. Only values of ``<size>`` that are supported by the target
1951 will work. 32 (float) and 64 (double) are supported on all targets; 80
1952 or 128 (different flavors of long double) are also supported on some
1955 This specifies the alignment for an object of aggregate type.
1957 If present, specifies that llvm names are mangled in the output. Symbols
1958 prefixed with the mangling escape character ``\01`` are passed through
1959 directly to the assembler without the escape character. The mangling style
1962 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1963 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1964 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1965 symbols get a ``_`` prefix.
1966 * ``x``: Windows x86 COFF mangling: Private symbols get the usual prefix.
1967 Regular C symbols get a ``_`` prefix. Functions with ``__stdcall``,
1968 ``__fastcall``, and ``__vectorcall`` have custom mangling that appends
1969 ``@N`` where N is the number of bytes used to pass parameters. C++ symbols
1970 starting with ``?`` are not mangled in any way.
1971 * ``w``: Windows COFF mangling: Similar to ``x``, except that normal C
1972 symbols do not receive a ``_`` prefix.
1973 ``n<size1>:<size2>:<size3>...``
1974 This specifies a set of native integer widths for the target CPU in
1975 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1976 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1977 this set are considered to support most general arithmetic operations
1979 ``ni:<address space0>:<address space1>:<address space2>...``
1980 This specifies pointer types with the specified address spaces
1981 as :ref:`Non-Integral Pointer Type <nointptrtype>` s. The ``0``
1982 address space cannot be specified as non-integral.
1984 On every specification that takes a ``<abi>:<pref>``, specifying the
1985 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1986 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1988 When constructing the data layout for a given target, LLVM starts with a
1989 default set of specifications which are then (possibly) overridden by
1990 the specifications in the ``datalayout`` keyword. The default
1991 specifications are given in this list:
1993 - ``E`` - big endian
1994 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1995 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1996 same as the default address space.
1997 - ``S0`` - natural stack alignment is unspecified
1998 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1999 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
2000 - ``i16:16:16`` - i16 is 16-bit aligned
2001 - ``i32:32:32`` - i32 is 32-bit aligned
2002 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
2003 alignment of 64-bits
2004 - ``f16:16:16`` - half is 16-bit aligned
2005 - ``f32:32:32`` - float is 32-bit aligned
2006 - ``f64:64:64`` - double is 64-bit aligned
2007 - ``f128:128:128`` - quad is 128-bit aligned
2008 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
2009 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
2010 - ``a:0:64`` - aggregates are 64-bit aligned
2012 When LLVM is determining the alignment for a given type, it uses the
2015 #. If the type sought is an exact match for one of the specifications,
2016 that specification is used.
2017 #. If no match is found, and the type sought is an integer type, then
2018 the smallest integer type that is larger than the bitwidth of the
2019 sought type is used. If none of the specifications are larger than
2020 the bitwidth then the largest integer type is used. For example,
2021 given the default specifications above, the i7 type will use the
2022 alignment of i8 (next largest) while both i65 and i256 will use the
2023 alignment of i64 (largest specified).
2024 #. If no match is found, and the type sought is a vector type, then the
2025 largest vector type that is smaller than the sought vector type will
2026 be used as a fall back. This happens because <128 x double> can be
2027 implemented in terms of 64 <2 x double>, for example.
2029 The function of the data layout string may not be what you expect.
2030 Notably, this is not a specification from the frontend of what alignment
2031 the code generator should use.
2033 Instead, if specified, the target data layout is required to match what
2034 the ultimate *code generator* expects. This string is used by the
2035 mid-level optimizers to improve code, and this only works if it matches
2036 what the ultimate code generator uses. There is no way to generate IR
2037 that does not embed this target-specific detail into the IR. If you
2038 don't specify the string, the default specifications will be used to
2039 generate a Data Layout and the optimization phases will operate
2040 accordingly and introduce target specificity into the IR with respect to
2041 these default specifications.
2048 A module may specify a target triple string that describes the target
2049 host. The syntax for the target triple is simply:
2051 .. code-block:: llvm
2053 target triple = "x86_64-apple-macosx10.7.0"
2055 The *target triple* string consists of a series of identifiers delimited
2056 by the minus sign character ('-'). The canonical forms are:
2060 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
2061 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
2063 This information is passed along to the backend so that it generates
2064 code for the proper architecture. It's possible to override this on the
2065 command line with the ``-mtriple`` command line option.
2067 .. _pointeraliasing:
2069 Pointer Aliasing Rules
2070 ----------------------
2072 Any memory access must be done through a pointer value associated with
2073 an address range of the memory access, otherwise the behavior is
2074 undefined. Pointer values are associated with address ranges according
2075 to the following rules:
2077 - A pointer value is associated with the addresses associated with any
2078 value it is *based* on.
2079 - An address of a global variable is associated with the address range
2080 of the variable's storage.
2081 - The result value of an allocation instruction is associated with the
2082 address range of the allocated storage.
2083 - A null pointer in the default address-space is associated with no
2085 - An integer constant other than zero or a pointer value returned from
2086 a function not defined within LLVM may be associated with address
2087 ranges allocated through mechanisms other than those provided by
2088 LLVM. Such ranges shall not overlap with any ranges of addresses
2089 allocated by mechanisms provided by LLVM.
2091 A pointer value is *based* on another pointer value according to the
2094 - A pointer value formed from a scalar ``getelementptr`` operation is *based* on
2095 the pointer-typed operand of the ``getelementptr``.
2096 - The pointer in lane *l* of the result of a vector ``getelementptr`` operation
2097 is *based* on the pointer in lane *l* of the vector-of-pointers-typed operand
2098 of the ``getelementptr``.
2099 - The result value of a ``bitcast`` is *based* on the operand of the
2101 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
2102 values that contribute (directly or indirectly) to the computation of
2103 the pointer's value.
2104 - The "*based* on" relationship is transitive.
2106 Note that this definition of *"based"* is intentionally similar to the
2107 definition of *"based"* in C99, though it is slightly weaker.
2109 LLVM IR does not associate types with memory. The result type of a
2110 ``load`` merely indicates the size and alignment of the memory from
2111 which to load, as well as the interpretation of the value. The first
2112 operand type of a ``store`` similarly only indicates the size and
2113 alignment of the store.
2115 Consequently, type-based alias analysis, aka TBAA, aka
2116 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
2117 :ref:`Metadata <metadata>` may be used to encode additional information
2118 which specialized optimization passes may use to implement type-based
2123 Volatile Memory Accesses
2124 ------------------------
2126 Certain memory accesses, such as :ref:`load <i_load>`'s,
2127 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
2128 marked ``volatile``. The optimizers must not change the number of
2129 volatile operations or change their order of execution relative to other
2130 volatile operations. The optimizers *may* change the order of volatile
2131 operations relative to non-volatile operations. This is not Java's
2132 "volatile" and has no cross-thread synchronization behavior.
2134 IR-level volatile loads and stores cannot safely be optimized into
2135 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
2136 flagged volatile. Likewise, the backend should never split or merge
2137 target-legal volatile load/store instructions.
2139 .. admonition:: Rationale
2141 Platforms may rely on volatile loads and stores of natively supported
2142 data width to be executed as single instruction. For example, in C
2143 this holds for an l-value of volatile primitive type with native
2144 hardware support, but not necessarily for aggregate types. The
2145 frontend upholds these expectations, which are intentionally
2146 unspecified in the IR. The rules above ensure that IR transformations
2147 do not violate the frontend's contract with the language.
2151 Memory Model for Concurrent Operations
2152 --------------------------------------
2154 The LLVM IR does not define any way to start parallel threads of
2155 execution or to register signal handlers. Nonetheless, there are
2156 platform-specific ways to create them, and we define LLVM IR's behavior
2157 in their presence. This model is inspired by the C++0x memory model.
2159 For a more informal introduction to this model, see the :doc:`Atomics`.
2161 We define a *happens-before* partial order as the least partial order
2164 - Is a superset of single-thread program order, and
2165 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
2166 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
2167 techniques, like pthread locks, thread creation, thread joining,
2168 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
2169 Constraints <ordering>`).
2171 Note that program order does not introduce *happens-before* edges
2172 between a thread and signals executing inside that thread.
2174 Every (defined) read operation (load instructions, memcpy, atomic
2175 loads/read-modify-writes, etc.) R reads a series of bytes written by
2176 (defined) write operations (store instructions, atomic
2177 stores/read-modify-writes, memcpy, etc.). For the purposes of this
2178 section, initialized globals are considered to have a write of the
2179 initializer which is atomic and happens before any other read or write
2180 of the memory in question. For each byte of a read R, R\ :sub:`byte`
2181 may see any write to the same byte, except:
2183 - If write\ :sub:`1` happens before write\ :sub:`2`, and
2184 write\ :sub:`2` happens before R\ :sub:`byte`, then
2185 R\ :sub:`byte` does not see write\ :sub:`1`.
2186 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
2187 R\ :sub:`byte` does not see write\ :sub:`3`.
2189 Given that definition, R\ :sub:`byte` is defined as follows:
2191 - If R is volatile, the result is target-dependent. (Volatile is
2192 supposed to give guarantees which can support ``sig_atomic_t`` in
2193 C/C++, and may be used for accesses to addresses that do not behave
2194 like normal memory. It does not generally provide cross-thread
2196 - Otherwise, if there is no write to the same byte that happens before
2197 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
2198 - Otherwise, if R\ :sub:`byte` may see exactly one write,
2199 R\ :sub:`byte` returns the value written by that write.
2200 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
2201 see are atomic, it chooses one of the values written. See the :ref:`Atomic
2202 Memory Ordering Constraints <ordering>` section for additional
2203 constraints on how the choice is made.
2204 - Otherwise R\ :sub:`byte` returns ``undef``.
2206 R returns the value composed of the series of bytes it read. This
2207 implies that some bytes within the value may be ``undef`` **without**
2208 the entire value being ``undef``. Note that this only defines the
2209 semantics of the operation; it doesn't mean that targets will emit more
2210 than one instruction to read the series of bytes.
2212 Note that in cases where none of the atomic intrinsics are used, this
2213 model places only one restriction on IR transformations on top of what
2214 is required for single-threaded execution: introducing a store to a byte
2215 which might not otherwise be stored is not allowed in general.
2216 (Specifically, in the case where another thread might write to and read
2217 from an address, introducing a store can change a load that may see
2218 exactly one write into a load that may see multiple writes.)
2222 Atomic Memory Ordering Constraints
2223 ----------------------------------
2225 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
2226 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
2227 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
2228 ordering parameters that determine which other atomic instructions on
2229 the same address they *synchronize with*. These semantics are borrowed
2230 from Java and C++0x, but are somewhat more colloquial. If these
2231 descriptions aren't precise enough, check those specs (see spec
2232 references in the :doc:`atomics guide <Atomics>`).
2233 :ref:`fence <i_fence>` instructions treat these orderings somewhat
2234 differently since they don't take an address. See that instruction's
2235 documentation for details.
2237 For a simpler introduction to the ordering constraints, see the
2241 The set of values that can be read is governed by the happens-before
2242 partial order. A value cannot be read unless some operation wrote
2243 it. This is intended to provide a guarantee strong enough to model
2244 Java's non-volatile shared variables. This ordering cannot be
2245 specified for read-modify-write operations; it is not strong enough
2246 to make them atomic in any interesting way.
2248 In addition to the guarantees of ``unordered``, there is a single
2249 total order for modifications by ``monotonic`` operations on each
2250 address. All modification orders must be compatible with the
2251 happens-before order. There is no guarantee that the modification
2252 orders can be combined to a global total order for the whole program
2253 (and this often will not be possible). The read in an atomic
2254 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
2255 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
2256 order immediately before the value it writes. If one atomic read
2257 happens before another atomic read of the same address, the later
2258 read must see the same value or a later value in the address's
2259 modification order. This disallows reordering of ``monotonic`` (or
2260 stronger) operations on the same address. If an address is written
2261 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
2262 read that address repeatedly, the other threads must eventually see
2263 the write. This corresponds to the C++0x/C1x
2264 ``memory_order_relaxed``.
2266 In addition to the guarantees of ``monotonic``, a
2267 *synchronizes-with* edge may be formed with a ``release`` operation.
2268 This is intended to model C++'s ``memory_order_acquire``.
2270 In addition to the guarantees of ``monotonic``, if this operation
2271 writes a value which is subsequently read by an ``acquire``
2272 operation, it *synchronizes-with* that operation. (This isn't a
2273 complete description; see the C++0x definition of a release
2274 sequence.) This corresponds to the C++0x/C1x
2275 ``memory_order_release``.
2276 ``acq_rel`` (acquire+release)
2277 Acts as both an ``acquire`` and ``release`` operation on its
2278 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
2279 ``seq_cst`` (sequentially consistent)
2280 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
2281 operation that only reads, ``release`` for an operation that only
2282 writes), there is a global total order on all
2283 sequentially-consistent operations on all addresses, which is
2284 consistent with the *happens-before* partial order and with the
2285 modification orders of all the affected addresses. Each
2286 sequentially-consistent read sees the last preceding write to the
2287 same address in this global order. This corresponds to the C++0x/C1x
2288 ``memory_order_seq_cst`` and Java volatile.
2292 If an atomic operation is marked ``syncscope("singlethread")``, it only
2293 *synchronizes with* and only participates in the seq\_cst total orderings of
2294 other operations running in the same thread (for example, in signal handlers).
2296 If an atomic operation is marked ``syncscope("<target-scope>")``, where
2297 ``<target-scope>`` is a target specific synchronization scope, then it is target
2298 dependent if it *synchronizes with* and participates in the seq\_cst total
2299 orderings of other operations.
2301 Otherwise, an atomic operation that is not marked ``syncscope("singlethread")``
2302 or ``syncscope("<target-scope>")`` *synchronizes with* and participates in the
2303 seq\_cst total orderings of other operations that are not marked
2304 ``syncscope("singlethread")`` or ``syncscope("<target-scope>")``.
2308 Floating-Point Environment
2309 --------------------------
2311 The default LLVM floating-point environment assumes that floating-point
2312 instructions do not have side effects. Results assume the round-to-nearest
2313 rounding mode. No floating-point exception state is maintained in this
2314 environment. Therefore, there is no attempt to create or preserve invalid
2315 operation (SNaN) or division-by-zero exceptions in these examples:
2317 .. code-block:: llvm
2319 %A = fdiv 0x7ff0000000000001, %X ; 64-bit SNaN hex value
2325 The benefit of this exception-free assumption is that floating-point
2326 operations may be speculated freely without any other fast-math relaxations
2327 to the floating-point model.
2329 Code that requires different behavior than this should use the
2330 :ref:`Constrained Floating-Point Intrinsics <constrainedfp>`.
2337 LLVM IR floating-point operations (:ref:`fadd <i_fadd>`,
2338 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
2339 :ref:`frem <i_frem>`, :ref:`fcmp <i_fcmp>`) and :ref:`call <i_call>`
2340 may use the following flags to enable otherwise unsafe
2341 floating-point transformations.
2344 No NaNs - Allow optimizations to assume the arguments and result are not
2345 NaN. Such optimizations are required to retain defined behavior over
2346 NaNs, but the value of the result is undefined.
2349 No Infs - Allow optimizations to assume the arguments and result are not
2350 +/-Inf. Such optimizations are required to retain defined behavior over
2351 +/-Inf, but the value of the result is undefined.
2354 No Signed Zeros - Allow optimizations to treat the sign of a zero
2355 argument or result as insignificant.
2358 Allow Reciprocal - Allow optimizations to use the reciprocal of an
2359 argument rather than perform division.
2362 Allow floating-point contraction (e.g. fusing a multiply followed by an
2363 addition into a fused multiply-and-add).
2366 Approximate functions - Allow substitution of approximate calculations for
2367 functions (sin, log, sqrt, etc). See floating-point intrinsic definitions
2368 for places where this can apply to LLVM's intrinsic math functions.
2371 Allow reassociation transformations for floating-point instructions.
2372 This may dramatically change results in floating-point.
2375 This flag implies all of the others.
2379 Use-list Order Directives
2380 -------------------------
2382 Use-list directives encode the in-memory order of each use-list, allowing the
2383 order to be recreated. ``<order-indexes>`` is a comma-separated list of
2384 indexes that are assigned to the referenced value's uses. The referenced
2385 value's use-list is immediately sorted by these indexes.
2387 Use-list directives may appear at function scope or global scope. They are not
2388 instructions, and have no effect on the semantics of the IR. When they're at
2389 function scope, they must appear after the terminator of the final basic block.
2391 If basic blocks have their address taken via ``blockaddress()`` expressions,
2392 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
2399 uselistorder <ty> <value>, { <order-indexes> }
2400 uselistorder_bb @function, %block { <order-indexes> }
2406 define void @foo(i32 %arg1, i32 %arg2) {
2408 ; ... instructions ...
2410 ; ... instructions ...
2412 ; At function scope.
2413 uselistorder i32 %arg1, { 1, 0, 2 }
2414 uselistorder label %bb, { 1, 0 }
2418 uselistorder i32* @global, { 1, 2, 0 }
2419 uselistorder i32 7, { 1, 0 }
2420 uselistorder i32 (i32) @bar, { 1, 0 }
2421 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
2423 .. _source_filename:
2428 The *source filename* string is set to the original module identifier,
2429 which will be the name of the compiled source file when compiling from
2430 source through the clang front end, for example. It is then preserved through
2433 This is currently necessary to generate a consistent unique global
2434 identifier for local functions used in profile data, which prepends the
2435 source file name to the local function name.
2437 The syntax for the source file name is simply:
2439 .. code-block:: text
2441 source_filename = "/path/to/source.c"
2448 The LLVM type system is one of the most important features of the
2449 intermediate representation. Being typed enables a number of
2450 optimizations to be performed on the intermediate representation
2451 directly, without having to do extra analyses on the side before the
2452 transformation. A strong type system makes it easier to read the
2453 generated code and enables novel analyses and transformations that are
2454 not feasible to perform on normal three address code representations.
2464 The void type does not represent any value and has no size.
2482 The function type can be thought of as a function signature. It consists of a
2483 return type and a list of formal parameter types. The return type of a function
2484 type is a void type or first class type --- except for :ref:`label <t_label>`
2485 and :ref:`metadata <t_metadata>` types.
2491 <returntype> (<parameter list>)
2493 ...where '``<parameter list>``' is a comma-separated list of type
2494 specifiers. Optionally, the parameter list may include a type ``...``, which
2495 indicates that the function takes a variable number of arguments. Variable
2496 argument functions can access their arguments with the :ref:`variable argument
2497 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
2498 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
2502 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2503 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
2504 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2505 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
2506 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2507 | ``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. |
2508 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2509 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
2510 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2517 The :ref:`first class <t_firstclass>` types are perhaps the most important.
2518 Values of these types are the only ones which can be produced by
2526 These are the types that are valid in registers from CodeGen's perspective.
2535 The integer type is a very simple type that simply specifies an
2536 arbitrary bit width for the integer type desired. Any bit width from 1
2537 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
2545 The number of bits the integer will occupy is specified by the ``N``
2551 +----------------+------------------------------------------------+
2552 | ``i1`` | a single-bit integer. |
2553 +----------------+------------------------------------------------+
2554 | ``i32`` | a 32-bit integer. |
2555 +----------------+------------------------------------------------+
2556 | ``i1942652`` | a really big integer of over 1 million bits. |
2557 +----------------+------------------------------------------------+
2561 Floating-Point Types
2562 """"""""""""""""""""
2571 - 16-bit floating-point value
2574 - 32-bit floating-point value
2577 - 64-bit floating-point value
2580 - 128-bit floating-point value (112-bit mantissa)
2583 - 80-bit floating-point value (X87)
2586 - 128-bit floating-point value (two 64-bits)
2588 The binary format of half, float, double, and fp128 correspond to the
2589 IEEE-754-2008 specifications for binary16, binary32, binary64, and binary128
2597 The x86_mmx type represents a value held in an MMX register on an x86
2598 machine. The operations allowed on it are quite limited: parameters and
2599 return values, load and store, and bitcast. User-specified MMX
2600 instructions are represented as intrinsic or asm calls with arguments
2601 and/or results of this type. There are no arrays, vectors or constants
2618 The pointer type is used to specify memory locations. Pointers are
2619 commonly used to reference objects in memory.
2621 Pointer types may have an optional address space attribute defining the
2622 numbered address space where the pointed-to object resides. The default
2623 address space is number zero. The semantics of non-zero address spaces
2624 are target-specific.
2626 Note that LLVM does not permit pointers to void (``void*``) nor does it
2627 permit pointers to labels (``label*``). Use ``i8*`` instead.
2637 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2638 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2639 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2640 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2641 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2642 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2643 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2652 A vector type is a simple derived type that represents a vector of
2653 elements. Vector types are used when multiple primitive data are
2654 operated in parallel using a single instruction (SIMD). A vector type
2655 requires a size (number of elements) and an underlying primitive data
2656 type. Vector types are considered :ref:`first class <t_firstclass>`.
2662 < <# elements> x <elementtype> >
2664 The number of elements is a constant integer value larger than 0;
2665 elementtype may be any integer, floating-point or pointer type. Vectors
2666 of size zero are not allowed.
2670 +-------------------+--------------------------------------------------+
2671 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2672 +-------------------+--------------------------------------------------+
2673 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2674 +-------------------+--------------------------------------------------+
2675 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2676 +-------------------+--------------------------------------------------+
2677 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2678 +-------------------+--------------------------------------------------+
2687 The label type represents code labels.
2702 The token type is used when a value is associated with an instruction
2703 but all uses of the value must not attempt to introspect or obscure it.
2704 As such, it is not appropriate to have a :ref:`phi <i_phi>` or
2705 :ref:`select <i_select>` of type token.
2722 The metadata type represents embedded metadata. No derived types may be
2723 created from metadata except for :ref:`function <t_function>` arguments.
2736 Aggregate Types are a subset of derived types that can contain multiple
2737 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2738 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2748 The array type is a very simple derived type that arranges elements
2749 sequentially in memory. The array type requires a size (number of
2750 elements) and an underlying data type.
2756 [<# elements> x <elementtype>]
2758 The number of elements is a constant integer value; ``elementtype`` may
2759 be any type with a size.
2763 +------------------+--------------------------------------+
2764 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2765 +------------------+--------------------------------------+
2766 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2767 +------------------+--------------------------------------+
2768 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2769 +------------------+--------------------------------------+
2771 Here are some examples of multidimensional arrays:
2773 +-----------------------------+----------------------------------------------------------+
2774 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2775 +-----------------------------+----------------------------------------------------------+
2776 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating-point values. |
2777 +-----------------------------+----------------------------------------------------------+
2778 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2779 +-----------------------------+----------------------------------------------------------+
2781 There is no restriction on indexing beyond the end of the array implied
2782 by a static type (though there are restrictions on indexing beyond the
2783 bounds of an allocated object in some cases). This means that
2784 single-dimension 'variable sized array' addressing can be implemented in
2785 LLVM with a zero length array type. An implementation of 'pascal style
2786 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2796 The structure type is used to represent a collection of data members
2797 together in memory. The elements of a structure may be any type that has
2800 Structures in memory are accessed using '``load``' and '``store``' by
2801 getting a pointer to a field with the '``getelementptr``' instruction.
2802 Structures in registers are accessed using the '``extractvalue``' and
2803 '``insertvalue``' instructions.
2805 Structures may optionally be "packed" structures, which indicate that
2806 the alignment of the struct is one byte, and that there is no padding
2807 between the elements. In non-packed structs, padding between field types
2808 is inserted as defined by the DataLayout string in the module, which is
2809 required to match what the underlying code generator expects.
2811 Structures can either be "literal" or "identified". A literal structure
2812 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2813 identified types are always defined at the top level with a name.
2814 Literal types are uniqued by their contents and can never be recursive
2815 or opaque since there is no way to write one. Identified types can be
2816 recursive, can be opaqued, and are never uniqued.
2822 %T1 = type { <type list> } ; Identified normal struct type
2823 %T2 = type <{ <type list> }> ; Identified packed struct type
2827 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2828 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2829 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2830 | ``{ 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``. |
2831 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2832 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2833 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2837 Opaque Structure Types
2838 """"""""""""""""""""""
2842 Opaque structure types are used to represent named structure types that
2843 do not have a body specified. This corresponds (for example) to the C
2844 notion of a forward declared structure.
2855 +--------------+-------------------+
2856 | ``opaque`` | An opaque type. |
2857 +--------------+-------------------+
2864 LLVM has several different basic types of constants. This section
2865 describes them all and their syntax.
2870 **Boolean constants**
2871 The two strings '``true``' and '``false``' are both valid constants
2873 **Integer constants**
2874 Standard integers (such as '4') are constants of the
2875 :ref:`integer <t_integer>` type. Negative numbers may be used with
2877 **Floating-point constants**
2878 Floating-point constants use standard decimal notation (e.g.
2879 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2880 hexadecimal notation (see below). The assembler requires the exact
2881 decimal value of a floating-point constant. For example, the
2882 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2883 decimal in binary. Floating-point constants must have a
2884 :ref:`floating-point <t_floating>` type.
2885 **Null pointer constants**
2886 The identifier '``null``' is recognized as a null pointer constant
2887 and must be of :ref:`pointer type <t_pointer>`.
2889 The identifier '``none``' is recognized as an empty token constant
2890 and must be of :ref:`token type <t_token>`.
2892 The one non-intuitive notation for constants is the hexadecimal form of
2893 floating-point constants. For example, the form
2894 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2895 than) '``double 4.5e+15``'. The only time hexadecimal floating-point
2896 constants are required (and the only time that they are generated by the
2897 disassembler) is when a floating-point constant must be emitted but it
2898 cannot be represented as a decimal floating-point number in a reasonable
2899 number of digits. For example, NaN's, infinities, and other special
2900 values are represented in their IEEE hexadecimal format so that assembly
2901 and disassembly do not cause any bits to change in the constants.
2903 When using the hexadecimal form, constants of types half, float, and
2904 double are represented using the 16-digit form shown above (which
2905 matches the IEEE754 representation for double); half and float values
2906 must, however, be exactly representable as IEEE 754 half and single
2907 precision, respectively. Hexadecimal format is always used for long
2908 double, and there are three forms of long double. The 80-bit format used
2909 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2910 128-bit format used by PowerPC (two adjacent doubles) is represented by
2911 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2912 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2913 will only work if they match the long double format on your target.
2914 The IEEE 16-bit format (half precision) is represented by ``0xH``
2915 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2916 (sign bit at the left).
2918 There are no constants of type x86_mmx.
2920 .. _complexconstants:
2925 Complex constants are a (potentially recursive) combination of simple
2926 constants and smaller complex constants.
2928 **Structure constants**
2929 Structure constants are represented with notation similar to
2930 structure type definitions (a comma separated list of elements,
2931 surrounded by braces (``{}``)). For example:
2932 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2933 "``@G = external global i32``". Structure constants must have
2934 :ref:`structure type <t_struct>`, and the number and types of elements
2935 must match those specified by the type.
2937 Array constants are represented with notation similar to array type
2938 definitions (a comma separated list of elements, surrounded by
2939 square brackets (``[]``)). For example:
2940 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2941 :ref:`array type <t_array>`, and the number and types of elements must
2942 match those specified by the type. As a special case, character array
2943 constants may also be represented as a double-quoted string using the ``c``
2944 prefix. For example: "``c"Hello World\0A\00"``".
2945 **Vector constants**
2946 Vector constants are represented with notation similar to vector
2947 type definitions (a comma separated list of elements, surrounded by
2948 less-than/greater-than's (``<>``)). For example:
2949 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2950 must have :ref:`vector type <t_vector>`, and the number and types of
2951 elements must match those specified by the type.
2952 **Zero initialization**
2953 The string '``zeroinitializer``' can be used to zero initialize a
2954 value to zero of *any* type, including scalar and
2955 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2956 having to print large zero initializers (e.g. for large arrays) and
2957 is always exactly equivalent to using explicit zero initializers.
2959 A metadata node is a constant tuple without types. For example:
2960 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2961 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2962 Unlike other typed constants that are meant to be interpreted as part of
2963 the instruction stream, metadata is a place to attach additional
2964 information such as debug info.
2966 Global Variable and Function Addresses
2967 --------------------------------------
2969 The addresses of :ref:`global variables <globalvars>` and
2970 :ref:`functions <functionstructure>` are always implicitly valid
2971 (link-time) constants. These constants are explicitly referenced when
2972 the :ref:`identifier for the global <identifiers>` is used and always have
2973 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2976 .. code-block:: llvm
2980 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2987 The string '``undef``' can be used anywhere a constant is expected, and
2988 indicates that the user of the value may receive an unspecified
2989 bit-pattern. Undefined values may be of any type (other than '``label``'
2990 or '``void``') and be used anywhere a constant is permitted.
2992 Undefined values are useful because they indicate to the compiler that
2993 the program is well defined no matter what value is used. This gives the
2994 compiler more freedom to optimize. Here are some examples of
2995 (potentially surprising) transformations that are valid (in pseudo IR):
2997 .. code-block:: llvm
3007 This is safe because all of the output bits are affected by the undef
3008 bits. Any output bit can have a zero or one depending on the input bits.
3010 .. code-block:: llvm
3018 %A = %X ;; By choosing undef as 0
3019 %B = %X ;; By choosing undef as -1
3024 These logical operations have bits that are not always affected by the
3025 input. For example, if ``%X`` has a zero bit, then the output of the
3026 '``and``' operation will always be a zero for that bit, no matter what
3027 the corresponding bit from the '``undef``' is. As such, it is unsafe to
3028 optimize or assume that the result of the '``and``' is '``undef``'.
3029 However, it is safe to assume that all bits of the '``undef``' could be
3030 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
3031 all the bits of the '``undef``' operand to the '``or``' could be set,
3032 allowing the '``or``' to be folded to -1.
3034 .. code-block:: llvm
3036 %A = select undef, %X, %Y
3037 %B = select undef, 42, %Y
3038 %C = select %X, %Y, undef
3048 This set of examples shows that undefined '``select``' (and conditional
3049 branch) conditions can go *either way*, but they have to come from one
3050 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
3051 both known to have a clear low bit, then ``%A`` would have to have a
3052 cleared low bit. However, in the ``%C`` example, the optimizer is
3053 allowed to assume that the '``undef``' operand could be the same as
3054 ``%Y``, allowing the whole '``select``' to be eliminated.
3056 .. code-block:: text
3058 %A = xor undef, undef
3075 This example points out that two '``undef``' operands are not
3076 necessarily the same. This can be surprising to people (and also matches
3077 C semantics) where they assume that "``X^X``" is always zero, even if
3078 ``X`` is undefined. This isn't true for a number of reasons, but the
3079 short answer is that an '``undef``' "variable" can arbitrarily change
3080 its value over its "live range". This is true because the variable
3081 doesn't actually *have a live range*. Instead, the value is logically
3082 read from arbitrary registers that happen to be around when needed, so
3083 the value is not necessarily consistent over time. In fact, ``%A`` and
3084 ``%C`` need to have the same semantics or the core LLVM "replace all
3085 uses with" concept would not hold.
3087 .. code-block:: llvm
3095 These examples show the crucial difference between an *undefined value*
3096 and *undefined behavior*. An undefined value (like '``undef``') is
3097 allowed to have an arbitrary bit-pattern. This means that the ``%A``
3098 operation can be constant folded to '``0``', because the '``undef``'
3099 could be zero, and zero divided by any value is zero.
3100 However, in the second example, we can make a more aggressive
3101 assumption: because the ``undef`` is allowed to be an arbitrary value,
3102 we are allowed to assume that it could be zero. Since a divide by zero
3103 has *undefined behavior*, we are allowed to assume that the operation
3104 does not execute at all. This allows us to delete the divide and all
3105 code after it. Because the undefined operation "can't happen", the
3106 optimizer can assume that it occurs in dead code.
3108 .. code-block:: text
3110 a: store undef -> %X
3111 b: store %X -> undef
3116 A store *of* an undefined value can be assumed to not have any effect;
3117 we can assume that the value is overwritten with bits that happen to
3118 match what was already there. However, a store *to* an undefined
3119 location could clobber arbitrary memory, therefore, it has undefined
3127 Poison values are similar to :ref:`undef values <undefvalues>`, however
3128 they also represent the fact that an instruction or constant expression
3129 that cannot evoke side effects has nevertheless detected a condition
3130 that results in undefined behavior.
3132 There is currently no way of representing a poison value in the IR; they
3133 only exist when produced by operations such as :ref:`add <i_add>` with
3136 Poison value behavior is defined in terms of value *dependence*:
3138 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
3139 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
3140 their dynamic predecessor basic block.
3141 - Function arguments depend on the corresponding actual argument values
3142 in the dynamic callers of their functions.
3143 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
3144 instructions that dynamically transfer control back to them.
3145 - :ref:`Invoke <i_invoke>` instructions depend on the
3146 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
3147 call instructions that dynamically transfer control back to them.
3148 - Non-volatile loads and stores depend on the most recent stores to all
3149 of the referenced memory addresses, following the order in the IR
3150 (including loads and stores implied by intrinsics such as
3151 :ref:`@llvm.memcpy <int_memcpy>`.)
3152 - An instruction with externally visible side effects depends on the
3153 most recent preceding instruction with externally visible side
3154 effects, following the order in the IR. (This includes :ref:`volatile
3155 operations <volatile>`.)
3156 - An instruction *control-depends* on a :ref:`terminator
3157 instruction <terminators>` if the terminator instruction has
3158 multiple successors and the instruction is always executed when
3159 control transfers to one of the successors, and may not be executed
3160 when control is transferred to another.
3161 - Additionally, an instruction also *control-depends* on a terminator
3162 instruction if the set of instructions it otherwise depends on would
3163 be different if the terminator had transferred control to a different
3165 - Dependence is transitive.
3167 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
3168 with the additional effect that any instruction that has a *dependence*
3169 on a poison value has undefined behavior.
3171 Here are some examples:
3173 .. code-block:: llvm
3176 %poison = sub nuw i32 0, 1 ; Results in a poison value.
3177 %still_poison = and i32 %poison, 0 ; 0, but also poison.
3178 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
3179 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
3181 store i32 %poison, i32* @g ; Poison value stored to memory.
3182 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
3184 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
3186 %narrowaddr = bitcast i32* @g to i16*
3187 %wideaddr = bitcast i32* @g to i64*
3188 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
3189 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
3191 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
3192 br i1 %cmp, label %true, label %end ; Branch to either destination.
3195 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
3196 ; it has undefined behavior.
3200 %p = phi i32 [ 0, %entry ], [ 1, %true ]
3201 ; Both edges into this PHI are
3202 ; control-dependent on %cmp, so this
3203 ; always results in a poison value.
3205 store volatile i32 0, i32* @g ; This would depend on the store in %true
3206 ; if %cmp is true, or the store in %entry
3207 ; otherwise, so this is undefined behavior.
3209 br i1 %cmp, label %second_true, label %second_end
3210 ; The same branch again, but this time the
3211 ; true block doesn't have side effects.
3218 store volatile i32 0, i32* @g ; This time, the instruction always depends
3219 ; on the store in %end. Also, it is
3220 ; control-equivalent to %end, so this is
3221 ; well-defined (ignoring earlier undefined
3222 ; behavior in this example).
3226 Addresses of Basic Blocks
3227 -------------------------
3229 ``blockaddress(@function, %block)``
3231 The '``blockaddress``' constant computes the address of the specified
3232 basic block in the specified function, and always has an ``i8*`` type.
3233 Taking the address of the entry block is illegal.
3235 This value only has defined behavior when used as an operand to the
3236 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
3237 against null. Pointer equality tests between labels addresses results in
3238 undefined behavior --- though, again, comparison against null is ok, and
3239 no label is equal to the null pointer. This may be passed around as an
3240 opaque pointer sized value as long as the bits are not inspected. This
3241 allows ``ptrtoint`` and arithmetic to be performed on these values so
3242 long as the original value is reconstituted before the ``indirectbr``
3245 Finally, some targets may provide defined semantics when using the value
3246 as the operand to an inline assembly, but that is target specific.
3250 Constant Expressions
3251 --------------------
3253 Constant expressions are used to allow expressions involving other
3254 constants to be used as constants. Constant expressions may be of any
3255 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
3256 that does not have side effects (e.g. load and call are not supported).
3257 The following is the syntax for constant expressions:
3259 ``trunc (CST to TYPE)``
3260 Perform the :ref:`trunc operation <i_trunc>` on constants.
3261 ``zext (CST to TYPE)``
3262 Perform the :ref:`zext operation <i_zext>` on constants.
3263 ``sext (CST to TYPE)``
3264 Perform the :ref:`sext operation <i_sext>` on constants.
3265 ``fptrunc (CST to TYPE)``
3266 Truncate a floating-point constant to another floating-point type.
3267 The size of CST must be larger than the size of TYPE. Both types
3268 must be floating-point.
3269 ``fpext (CST to TYPE)``
3270 Floating-point extend a constant to another type. The size of CST
3271 must be smaller or equal to the size of TYPE. Both types must be
3273 ``fptoui (CST to TYPE)``
3274 Convert a floating-point constant to the corresponding unsigned
3275 integer constant. TYPE must be a scalar or vector integer type. CST
3276 must be of scalar or vector floating-point type. Both CST and TYPE
3277 must be scalars, or vectors of the same number of elements. If the
3278 value won't fit in the integer type, the results are undefined.
3279 ``fptosi (CST to TYPE)``
3280 Convert a floating-point constant to the corresponding signed
3281 integer constant. TYPE must be a scalar or vector integer type. CST
3282 must be of scalar or vector floating-point type. Both CST and TYPE
3283 must be scalars, or vectors of the same number of elements. If the
3284 value won't fit in the integer type, the results are undefined.
3285 ``uitofp (CST to TYPE)``
3286 Convert an unsigned integer constant to the corresponding
3287 floating-point constant. TYPE must be a scalar or vector floating-point
3288 type. CST must be of scalar or vector integer type. Both CST and TYPE must
3289 be scalars, or vectors of the same number of elements. If the value
3290 won't fit in the floating-point type, the results are undefined.
3291 ``sitofp (CST to TYPE)``
3292 Convert a signed integer constant to the corresponding floating-point
3293 constant. TYPE must be a scalar or vector floating-point type.
3294 CST must be of scalar or vector integer type. Both CST and TYPE must
3295 be scalars, or vectors of the same number of elements. If the value
3296 won't fit in the floating-point type, the results are undefined.
3297 ``ptrtoint (CST to TYPE)``
3298 Perform the :ref:`ptrtoint operation <i_ptrtoint>` on constants.
3299 ``inttoptr (CST to TYPE)``
3300 Perform the :ref:`inttoptr operation <i_inttoptr>` on constants.
3301 This one is *really* dangerous!
3302 ``bitcast (CST to TYPE)``
3303 Convert a constant, CST, to another TYPE.
3304 The constraints of the operands are the same as those for the
3305 :ref:`bitcast instruction <i_bitcast>`.
3306 ``addrspacecast (CST to TYPE)``
3307 Convert a constant pointer or constant vector of pointer, CST, to another
3308 TYPE in a different address space. The constraints of the operands are the
3309 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
3310 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
3311 Perform the :ref:`getelementptr operation <i_getelementptr>` on
3312 constants. As with the :ref:`getelementptr <i_getelementptr>`
3313 instruction, the index list may have one or more indexes, which are
3314 required to make sense for the type of "pointer to TY".
3315 ``select (COND, VAL1, VAL2)``
3316 Perform the :ref:`select operation <i_select>` on constants.
3317 ``icmp COND (VAL1, VAL2)``
3318 Perform the :ref:`icmp operation <i_icmp>` on constants.
3319 ``fcmp COND (VAL1, VAL2)``
3320 Perform the :ref:`fcmp operation <i_fcmp>` on constants.
3321 ``extractelement (VAL, IDX)``
3322 Perform the :ref:`extractelement operation <i_extractelement>` on
3324 ``insertelement (VAL, ELT, IDX)``
3325 Perform the :ref:`insertelement operation <i_insertelement>` on
3327 ``shufflevector (VEC1, VEC2, IDXMASK)``
3328 Perform the :ref:`shufflevector operation <i_shufflevector>` on
3330 ``extractvalue (VAL, IDX0, IDX1, ...)``
3331 Perform the :ref:`extractvalue operation <i_extractvalue>` on
3332 constants. The index list is interpreted in a similar manner as
3333 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
3334 least one index value must be specified.
3335 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
3336 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
3337 The index list is interpreted in a similar manner as indices in a
3338 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
3339 value must be specified.
3340 ``OPCODE (LHS, RHS)``
3341 Perform the specified operation of the LHS and RHS constants. OPCODE
3342 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
3343 binary <bitwiseops>` operations. The constraints on operands are
3344 the same as those for the corresponding instruction (e.g. no bitwise
3345 operations on floating-point values are allowed).
3352 Inline Assembler Expressions
3353 ----------------------------
3355 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
3356 Inline Assembly <moduleasm>`) through the use of a special value. This value
3357 represents the inline assembler as a template string (containing the
3358 instructions to emit), a list of operand constraints (stored as a string), a
3359 flag that indicates whether or not the inline asm expression has side effects,
3360 and a flag indicating whether the function containing the asm needs to align its
3361 stack conservatively.
3363 The template string supports argument substitution of the operands using "``$``"
3364 followed by a number, to indicate substitution of the given register/memory
3365 location, as specified by the constraint string. "``${NUM:MODIFIER}``" may also
3366 be used, where ``MODIFIER`` is a target-specific annotation for how to print the
3367 operand (See :ref:`inline-asm-modifiers`).
3369 A literal "``$``" may be included by using "``$$``" in the template. To include
3370 other special characters into the output, the usual "``\XX``" escapes may be
3371 used, just as in other strings. Note that after template substitution, the
3372 resulting assembly string is parsed by LLVM's integrated assembler unless it is
3373 disabled -- even when emitting a ``.s`` file -- and thus must contain assembly
3374 syntax known to LLVM.
3376 LLVM also supports a few more substitions useful for writing inline assembly:
3378 - ``${:uid}``: Expands to a decimal integer unique to this inline assembly blob.
3379 This substitution is useful when declaring a local label. Many standard
3380 compiler optimizations, such as inlining, may duplicate an inline asm blob.
3381 Adding a blob-unique identifier ensures that the two labels will not conflict
3382 during assembly. This is used to implement `GCC's %= special format
3383 string <https://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html>`_.
3384 - ``${:comment}``: Expands to the comment character of the current target's
3385 assembly dialect. This is usually ``#``, but many targets use other strings,
3386 such as ``;``, ``//``, or ``!``.
3387 - ``${:private}``: Expands to the assembler private label prefix. Labels with
3388 this prefix will not appear in the symbol table of the assembled object.
3389 Typically the prefix is ``L``, but targets may use other strings. ``.L`` is
3392 LLVM's support for inline asm is modeled closely on the requirements of Clang's
3393 GCC-compatible inline-asm support. Thus, the feature-set and the constraint and
3394 modifier codes listed here are similar or identical to those in GCC's inline asm
3395 support. However, to be clear, the syntax of the template and constraint strings
3396 described here is *not* the same as the syntax accepted by GCC and Clang, and,
3397 while most constraint letters are passed through as-is by Clang, some get
3398 translated to other codes when converting from the C source to the LLVM
3401 An example inline assembler expression is:
3403 .. code-block:: llvm
3405 i32 (i32) asm "bswap $0", "=r,r"
3407 Inline assembler expressions may **only** be used as the callee operand
3408 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
3409 Thus, typically we have:
3411 .. code-block:: llvm
3413 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
3415 Inline asms with side effects not visible in the constraint list must be
3416 marked as having side effects. This is done through the use of the
3417 '``sideeffect``' keyword, like so:
3419 .. code-block:: llvm
3421 call void asm sideeffect "eieio", ""()
3423 In some cases inline asms will contain code that will not work unless
3424 the stack is aligned in some way, such as calls or SSE instructions on
3425 x86, yet will not contain code that does that alignment within the asm.
3426 The compiler should make conservative assumptions about what the asm
3427 might contain and should generate its usual stack alignment code in the
3428 prologue if the '``alignstack``' keyword is present:
3430 .. code-block:: llvm
3432 call void asm alignstack "eieio", ""()
3434 Inline asms also support using non-standard assembly dialects. The
3435 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
3436 the inline asm is using the Intel dialect. Currently, ATT and Intel are
3437 the only supported dialects. An example is:
3439 .. code-block:: llvm
3441 call void asm inteldialect "eieio", ""()
3443 If multiple keywords appear the '``sideeffect``' keyword must come
3444 first, the '``alignstack``' keyword second and the '``inteldialect``'
3447 Inline Asm Constraint String
3448 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3450 The constraint list is a comma-separated string, each element containing one or
3451 more constraint codes.
3453 For each element in the constraint list an appropriate register or memory
3454 operand will be chosen, and it will be made available to assembly template
3455 string expansion as ``$0`` for the first constraint in the list, ``$1`` for the
3458 There are three different types of constraints, which are distinguished by a
3459 prefix symbol in front of the constraint code: Output, Input, and Clobber. The
3460 constraints must always be given in that order: outputs first, then inputs, then
3461 clobbers. They cannot be intermingled.
3463 There are also three different categories of constraint codes:
3465 - Register constraint. This is either a register class, or a fixed physical
3466 register. This kind of constraint will allocate a register, and if necessary,
3467 bitcast the argument or result to the appropriate type.
3468 - Memory constraint. This kind of constraint is for use with an instruction
3469 taking a memory operand. Different constraints allow for different addressing
3470 modes used by the target.
3471 - Immediate value constraint. This kind of constraint is for an integer or other
3472 immediate value which can be rendered directly into an instruction. The
3473 various target-specific constraints allow the selection of a value in the
3474 proper range for the instruction you wish to use it with.
3479 Output constraints are specified by an "``=``" prefix (e.g. "``=r``"). This
3480 indicates that the assembly will write to this operand, and the operand will
3481 then be made available as a return value of the ``asm`` expression. Output
3482 constraints do not consume an argument from the call instruction. (Except, see
3483 below about indirect outputs).
3485 Normally, it is expected that no output locations are written to by the assembly
3486 expression until *all* of the inputs have been read. As such, LLVM may assign
3487 the same register to an output and an input. If this is not safe (e.g. if the
3488 assembly contains two instructions, where the first writes to one output, and
3489 the second reads an input and writes to a second output), then the "``&``"
3490 modifier must be used (e.g. "``=&r``") to specify that the output is an
3491 "early-clobber" output. Marking an output as "early-clobber" ensures that LLVM
3492 will not use the same register for any inputs (other than an input tied to this
3498 Input constraints do not have a prefix -- just the constraint codes. Each input
3499 constraint will consume one argument from the call instruction. It is not
3500 permitted for the asm to write to any input register or memory location (unless
3501 that input is tied to an output). Note also that multiple inputs may all be
3502 assigned to the same register, if LLVM can determine that they necessarily all
3503 contain the same value.
3505 Instead of providing a Constraint Code, input constraints may also "tie"
3506 themselves to an output constraint, by providing an integer as the constraint
3507 string. Tied inputs still consume an argument from the call instruction, and
3508 take up a position in the asm template numbering as is usual -- they will simply
3509 be constrained to always use the same register as the output they've been tied
3510 to. For example, a constraint string of "``=r,0``" says to assign a register for
3511 output, and use that register as an input as well (it being the 0'th
3514 It is permitted to tie an input to an "early-clobber" output. In that case, no
3515 *other* input may share the same register as the input tied to the early-clobber
3516 (even when the other input has the same value).
3518 You may only tie an input to an output which has a register constraint, not a
3519 memory constraint. Only a single input may be tied to an output.
3521 There is also an "interesting" feature which deserves a bit of explanation: if a
3522 register class constraint allocates a register which is too small for the value
3523 type operand provided as input, the input value will be split into multiple
3524 registers, and all of them passed to the inline asm.
3526 However, this feature is often not as useful as you might think.
3528 Firstly, the registers are *not* guaranteed to be consecutive. So, on those
3529 architectures that have instructions which operate on multiple consecutive
3530 instructions, this is not an appropriate way to support them. (e.g. the 32-bit
3531 SparcV8 has a 64-bit load, which instruction takes a single 32-bit register. The
3532 hardware then loads into both the named register, and the next register. This
3533 feature of inline asm would not be useful to support that.)
3535 A few of the targets provide a template string modifier allowing explicit access
3536 to the second register of a two-register operand (e.g. MIPS ``L``, ``M``, and
3537 ``D``). On such an architecture, you can actually access the second allocated
3538 register (yet, still, not any subsequent ones). But, in that case, you're still
3539 probably better off simply splitting the value into two separate operands, for
3540 clarity. (e.g. see the description of the ``A`` constraint on X86, which,
3541 despite existing only for use with this feature, is not really a good idea to
3544 Indirect inputs and outputs
3545 """""""""""""""""""""""""""
3547 Indirect output or input constraints can be specified by the "``*``" modifier
3548 (which goes after the "``=``" in case of an output). This indicates that the asm
3549 will write to or read from the contents of an *address* provided as an input
3550 argument. (Note that in this way, indirect outputs act more like an *input* than
3551 an output: just like an input, they consume an argument of the call expression,
3552 rather than producing a return value. An indirect output constraint is an
3553 "output" only in that the asm is expected to write to the contents of the input
3554 memory location, instead of just read from it).
3556 This is most typically used for memory constraint, e.g. "``=*m``", to pass the
3557 address of a variable as a value.
3559 It is also possible to use an indirect *register* constraint, but only on output
3560 (e.g. "``=*r``"). This will cause LLVM to allocate a register for an output
3561 value normally, and then, separately emit a store to the address provided as
3562 input, after the provided inline asm. (It's not clear what value this
3563 functionality provides, compared to writing the store explicitly after the asm
3564 statement, and it can only produce worse code, since it bypasses many
3565 optimization passes. I would recommend not using it.)
3571 A clobber constraint is indicated by a "``~``" prefix. A clobber does not
3572 consume an input operand, nor generate an output. Clobbers cannot use any of the
3573 general constraint code letters -- they may use only explicit register
3574 constraints, e.g. "``~{eax}``". The one exception is that a clobber string of
3575 "``~{memory}``" indicates that the assembly writes to arbitrary undeclared
3576 memory locations -- not only the memory pointed to by a declared indirect
3579 Note that clobbering named registers that are also present in output
3580 constraints is not legal.
3585 After a potential prefix comes constraint code, or codes.
3587 A Constraint Code is either a single letter (e.g. "``r``"), a "``^``" character
3588 followed by two letters (e.g. "``^wc``"), or "``{``" register-name "``}``"
3591 The one and two letter constraint codes are typically chosen to be the same as
3592 GCC's constraint codes.
3594 A single constraint may include one or more than constraint code in it, leaving
3595 it up to LLVM to choose which one to use. This is included mainly for
3596 compatibility with the translation of GCC inline asm coming from clang.
3598 There are two ways to specify alternatives, and either or both may be used in an
3599 inline asm constraint list:
3601 1) Append the codes to each other, making a constraint code set. E.g. "``im``"
3602 or "``{eax}m``". This means "choose any of the options in the set". The
3603 choice of constraint is made independently for each constraint in the
3606 2) Use "``|``" between constraint code sets, creating alternatives. Every
3607 constraint in the constraint list must have the same number of alternative
3608 sets. With this syntax, the same alternative in *all* of the items in the
3609 constraint list will be chosen together.
3611 Putting those together, you might have a two operand constraint string like
3612 ``"rm|r,ri|rm"``. This indicates that if operand 0 is ``r`` or ``m``, then
3613 operand 1 may be one of ``r`` or ``i``. If operand 0 is ``r``, then operand 1
3614 may be one of ``r`` or ``m``. But, operand 0 and 1 cannot both be of type m.
3616 However, the use of either of the alternatives features is *NOT* recommended, as
3617 LLVM is not able to make an intelligent choice about which one to use. (At the
3618 point it currently needs to choose, not enough information is available to do so
3619 in a smart way.) Thus, it simply tries to make a choice that's most likely to
3620 compile, not one that will be optimal performance. (e.g., given "``rm``", it'll
3621 always choose to use memory, not registers). And, if given multiple registers,
3622 or multiple register classes, it will simply choose the first one. (In fact, it
3623 doesn't currently even ensure explicitly specified physical registers are
3624 unique, so specifying multiple physical registers as alternatives, like
3625 ``{r11}{r12},{r11}{r12}``, will assign r11 to both operands, not at all what was
3628 Supported Constraint Code List
3629 """"""""""""""""""""""""""""""
3631 The constraint codes are, in general, expected to behave the same way they do in
3632 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3633 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3634 and GCC likely indicates a bug in LLVM.
3636 Some constraint codes are typically supported by all targets:
3638 - ``r``: A register in the target's general purpose register class.
3639 - ``m``: A memory address operand. It is target-specific what addressing modes
3640 are supported, typical examples are register, or register + register offset,
3641 or register + immediate offset (of some target-specific size).
3642 - ``i``: An integer constant (of target-specific width). Allows either a simple
3643 immediate, or a relocatable value.
3644 - ``n``: An integer constant -- *not* including relocatable values.
3645 - ``s``: An integer constant, but allowing *only* relocatable values.
3646 - ``X``: Allows an operand of any kind, no constraint whatsoever. Typically
3647 useful to pass a label for an asm branch or call.
3649 .. FIXME: but that surely isn't actually okay to jump out of an asm
3650 block without telling llvm about the control transfer???)
3652 - ``{register-name}``: Requires exactly the named physical register.
3654 Other constraints are target-specific:
3658 - ``z``: An immediate integer 0. Outputs ``WZR`` or ``XZR``, as appropriate.
3659 - ``I``: An immediate integer valid for an ``ADD`` or ``SUB`` instruction,
3660 i.e. 0 to 4095 with optional shift by 12.
3661 - ``J``: An immediate integer that, when negated, is valid for an ``ADD`` or
3662 ``SUB`` instruction, i.e. -1 to -4095 with optional left shift by 12.
3663 - ``K``: An immediate integer that is valid for the 'bitmask immediate 32' of a
3664 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 32-bit register.
3665 - ``L``: An immediate integer that is valid for the 'bitmask immediate 64' of a
3666 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 64-bit register.
3667 - ``M``: An immediate integer for use with the ``MOV`` assembly alias on a
3668 32-bit register. This is a superset of ``K``: in addition to the bitmask
3669 immediate, also allows immediate integers which can be loaded with a single
3670 ``MOVZ`` or ``MOVL`` instruction.
3671 - ``N``: An immediate integer for use with the ``MOV`` assembly alias on a
3672 64-bit register. This is a superset of ``L``.
3673 - ``Q``: Memory address operand must be in a single register (no
3674 offsets). (However, LLVM currently does this for the ``m`` constraint as
3676 - ``r``: A 32 or 64-bit integer register (W* or X*).
3677 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register.
3678 - ``x``: A lower 128-bit floating-point/SIMD register (``V0`` to ``V15``).
3682 - ``r``: A 32 or 64-bit integer register.
3683 - ``[0-9]v``: The 32-bit VGPR register, number 0-9.
3684 - ``[0-9]s``: The 32-bit SGPR register, number 0-9.
3689 - ``Q``, ``Um``, ``Un``, ``Uq``, ``Us``, ``Ut``, ``Uv``, ``Uy``: Memory address
3690 operand. Treated the same as operand ``m``, at the moment.
3692 ARM and ARM's Thumb2 mode:
3694 - ``j``: An immediate integer between 0 and 65535 (valid for ``MOVW``)
3695 - ``I``: An immediate integer valid for a data-processing instruction.
3696 - ``J``: An immediate integer between -4095 and 4095.
3697 - ``K``: An immediate integer whose bitwise inverse is valid for a
3698 data-processing instruction. (Can be used with template modifier "``B``" to
3699 print the inverted value).
3700 - ``L``: An immediate integer whose negation is valid for a data-processing
3701 instruction. (Can be used with template modifier "``n``" to print the negated
3703 - ``M``: A power of two or a integer between 0 and 32.
3704 - ``N``: Invalid immediate constraint.
3705 - ``O``: Invalid immediate constraint.
3706 - ``r``: A general-purpose 32-bit integer register (``r0-r15``).
3707 - ``l``: In Thumb2 mode, low 32-bit GPR registers (``r0-r7``). In ARM mode, same
3709 - ``h``: In Thumb2 mode, a high 32-bit GPR register (``r8-r15``). In ARM mode,
3711 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3712 ``d0-d31``, or ``q0-q15``.
3713 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3714 ``d0-d7``, or ``q0-q3``.
3715 - ``t``: A low floating-point/SIMD register: ``s0-s31``, ``d0-d16``, or
3720 - ``I``: An immediate integer between 0 and 255.
3721 - ``J``: An immediate integer between -255 and -1.
3722 - ``K``: An immediate integer between 0 and 255, with optional left-shift by
3724 - ``L``: An immediate integer between -7 and 7.
3725 - ``M``: An immediate integer which is a multiple of 4 between 0 and 1020.
3726 - ``N``: An immediate integer between 0 and 31.
3727 - ``O``: An immediate integer which is a multiple of 4 between -508 and 508.
3728 - ``r``: A low 32-bit GPR register (``r0-r7``).
3729 - ``l``: A low 32-bit GPR register (``r0-r7``).
3730 - ``h``: A high GPR register (``r0-r7``).
3731 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3732 ``d0-d31``, or ``q0-q15``.
3733 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3734 ``d0-d7``, or ``q0-q3``.
3735 - ``t``: A low floating-point/SIMD register: ``s0-s31``, ``d0-d16``, or
3741 - ``o``, ``v``: A memory address operand, treated the same as constraint ``m``,
3743 - ``r``: A 32 or 64-bit register.
3747 - ``r``: An 8 or 16-bit register.
3751 - ``I``: An immediate signed 16-bit integer.
3752 - ``J``: An immediate integer zero.
3753 - ``K``: An immediate unsigned 16-bit integer.
3754 - ``L``: An immediate 32-bit integer, where the lower 16 bits are 0.
3755 - ``N``: An immediate integer between -65535 and -1.
3756 - ``O``: An immediate signed 15-bit integer.
3757 - ``P``: An immediate integer between 1 and 65535.
3758 - ``m``: A memory address operand. In MIPS-SE mode, allows a base address
3759 register plus 16-bit immediate offset. In MIPS mode, just a base register.
3760 - ``R``: A memory address operand. In MIPS-SE mode, allows a base address
3761 register plus a 9-bit signed offset. In MIPS mode, the same as constraint
3763 - ``ZC``: A memory address operand, suitable for use in a ``pref``, ``ll``, or
3764 ``sc`` instruction on the given subtarget (details vary).
3765 - ``r``, ``d``, ``y``: A 32 or 64-bit GPR register.
3766 - ``f``: A 32 or 64-bit FPU register (``F0-F31``), or a 128-bit MSA register
3767 (``W0-W31``). In the case of MSA registers, it is recommended to use the ``w``
3768 argument modifier for compatibility with GCC.
3769 - ``c``: A 32-bit or 64-bit GPR register suitable for indirect jump (always
3771 - ``l``: The ``lo`` register, 32 or 64-bit.
3776 - ``b``: A 1-bit integer register.
3777 - ``c`` or ``h``: A 16-bit integer register.
3778 - ``r``: A 32-bit integer register.
3779 - ``l`` or ``N``: A 64-bit integer register.
3780 - ``f``: A 32-bit float register.
3781 - ``d``: A 64-bit float register.
3786 - ``I``: An immediate signed 16-bit integer.
3787 - ``J``: An immediate unsigned 16-bit integer, shifted left 16 bits.
3788 - ``K``: An immediate unsigned 16-bit integer.
3789 - ``L``: An immediate signed 16-bit integer, shifted left 16 bits.
3790 - ``M``: An immediate integer greater than 31.
3791 - ``N``: An immediate integer that is an exact power of 2.
3792 - ``O``: The immediate integer constant 0.
3793 - ``P``: An immediate integer constant whose negation is a signed 16-bit
3795 - ``es``, ``o``, ``Q``, ``Z``, ``Zy``: A memory address operand, currently
3796 treated the same as ``m``.
3797 - ``r``: A 32 or 64-bit integer register.
3798 - ``b``: A 32 or 64-bit integer register, excluding ``R0`` (that is:
3800 - ``f``: A 32 or 64-bit float register (``F0-F31``), or when QPX is enabled, a
3801 128 or 256-bit QPX register (``Q0-Q31``; aliases the ``F`` registers).
3802 - ``v``: For ``4 x f32`` or ``4 x f64`` types, when QPX is enabled, a
3803 128 or 256-bit QPX register (``Q0-Q31``), otherwise a 128-bit
3804 altivec vector register (``V0-V31``).
3806 .. FIXME: is this a bug that v accepts QPX registers? I think this
3807 is supposed to only use the altivec vector registers?
3809 - ``y``: Condition register (``CR0-CR7``).
3810 - ``wc``: An individual CR bit in a CR register.
3811 - ``wa``, ``wd``, ``wf``: Any 128-bit VSX vector register, from the full VSX
3812 register set (overlapping both the floating-point and vector register files).
3813 - ``ws``: A 32 or 64-bit floating-point register, from the full VSX register
3818 - ``I``: An immediate 13-bit signed integer.
3819 - ``r``: A 32-bit integer register.
3820 - ``f``: Any floating-point register on SparcV8, or a floating-point
3821 register in the "low" half of the registers on SparcV9.
3822 - ``e``: Any floating-point register. (Same as ``f`` on SparcV8.)
3826 - ``I``: An immediate unsigned 8-bit integer.
3827 - ``J``: An immediate unsigned 12-bit integer.
3828 - ``K``: An immediate signed 16-bit integer.
3829 - ``L``: An immediate signed 20-bit integer.
3830 - ``M``: An immediate integer 0x7fffffff.
3831 - ``Q``: A memory address operand with a base address and a 12-bit immediate
3832 unsigned displacement.
3833 - ``R``: A memory address operand with a base address, a 12-bit immediate
3834 unsigned displacement, and an index register.
3835 - ``S``: A memory address operand with a base address and a 20-bit immediate
3836 signed displacement.
3837 - ``T``: A memory address operand with a base address, a 20-bit immediate
3838 signed displacement, and an index register.
3839 - ``r`` or ``d``: A 32, 64, or 128-bit integer register.
3840 - ``a``: A 32, 64, or 128-bit integer address register (excludes R0, which in an
3841 address context evaluates as zero).
3842 - ``h``: A 32-bit value in the high part of a 64bit data register
3844 - ``f``: A 32, 64, or 128-bit floating-point register.
3848 - ``I``: An immediate integer between 0 and 31.
3849 - ``J``: An immediate integer between 0 and 64.
3850 - ``K``: An immediate signed 8-bit integer.
3851 - ``L``: An immediate integer, 0xff or 0xffff or (in 64-bit mode only)
3853 - ``M``: An immediate integer between 0 and 3.
3854 - ``N``: An immediate unsigned 8-bit integer.
3855 - ``O``: An immediate integer between 0 and 127.
3856 - ``e``: An immediate 32-bit signed integer.
3857 - ``Z``: An immediate 32-bit unsigned integer.
3858 - ``o``, ``v``: Treated the same as ``m``, at the moment.
3859 - ``q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3860 ``l`` integer register. On X86-32, this is the ``a``, ``b``, ``c``, and ``d``
3861 registers, and on X86-64, it is all of the integer registers.
3862 - ``Q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3863 ``h`` integer register. This is the ``a``, ``b``, ``c``, and ``d`` registers.
3864 - ``r`` or ``l``: An 8, 16, 32, or 64-bit integer register.
3865 - ``R``: An 8, 16, 32, or 64-bit "legacy" integer register -- one which has
3866 existed since i386, and can be accessed without the REX prefix.
3867 - ``f``: A 32, 64, or 80-bit '387 FPU stack pseudo-register.
3868 - ``y``: A 64-bit MMX register, if MMX is enabled.
3869 - ``x``: If SSE is enabled: a 32 or 64-bit scalar operand, or 128-bit vector
3870 operand in a SSE register. If AVX is also enabled, can also be a 256-bit
3871 vector operand in an AVX register. If AVX-512 is also enabled, can also be a
3872 512-bit vector operand in an AVX512 register, Otherwise, an error.
3873 - ``Y``: The same as ``x``, if *SSE2* is enabled, otherwise an error.
3874 - ``A``: Special case: allocates EAX first, then EDX, for a single operand (in
3875 32-bit mode, a 64-bit integer operand will get split into two registers). It
3876 is not recommended to use this constraint, as in 64-bit mode, the 64-bit
3877 operand will get allocated only to RAX -- if two 32-bit operands are needed,
3878 you're better off splitting it yourself, before passing it to the asm
3883 - ``r``: A 32-bit integer register.
3886 .. _inline-asm-modifiers:
3888 Asm template argument modifiers
3889 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3891 In the asm template string, modifiers can be used on the operand reference, like
3894 The modifiers are, in general, expected to behave the same way they do in
3895 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3896 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3897 and GCC likely indicates a bug in LLVM.
3901 - ``c``: Print an immediate integer constant unadorned, without
3902 the target-specific immediate punctuation (e.g. no ``$`` prefix).
3903 - ``n``: Negate and print immediate integer constant unadorned, without the
3904 target-specific immediate punctuation (e.g. no ``$`` prefix).
3905 - ``l``: Print as an unadorned label, without the target-specific label
3906 punctuation (e.g. no ``$`` prefix).
3910 - ``w``: Print a GPR register with a ``w*`` name instead of ``x*`` name. E.g.,
3911 instead of ``x30``, print ``w30``.
3912 - ``x``: Print a GPR register with a ``x*`` name. (this is the default, anyhow).
3913 - ``b``, ``h``, ``s``, ``d``, ``q``: Print a floating-point/SIMD register with a
3914 ``b*``, ``h*``, ``s*``, ``d*``, or ``q*`` name, rather than the default of
3923 - ``a``: Print an operand as an address (with ``[`` and ``]`` surrounding a
3927 - ``y``: Print a VFP single-precision register as an indexed double (e.g. print
3928 as ``d4[1]`` instead of ``s9``)
3929 - ``B``: Bitwise invert and print an immediate integer constant without ``#``
3931 - ``L``: Print the low 16-bits of an immediate integer constant.
3932 - ``M``: Print as a register set suitable for ldm/stm. Also prints *all*
3933 register operands subsequent to the specified one (!), so use carefully.
3934 - ``Q``: Print the low-order register of a register-pair, or the low-order
3935 register of a two-register operand.
3936 - ``R``: Print the high-order register of a register-pair, or the high-order
3937 register of a two-register operand.
3938 - ``H``: Print the second register of a register-pair. (On a big-endian system,
3939 ``H`` is equivalent to ``Q``, and on little-endian system, ``H`` is equivalent
3942 .. FIXME: H doesn't currently support printing the second register
3943 of a two-register operand.
3945 - ``e``: Print the low doubleword register of a NEON quad register.
3946 - ``f``: Print the high doubleword register of a NEON quad register.
3947 - ``m``: Print the base register of a memory operand without the ``[`` and ``]``
3952 - ``L``: Print the second register of a two-register operand. Requires that it
3953 has been allocated consecutively to the first.
3955 .. FIXME: why is it restricted to consecutive ones? And there's
3956 nothing that ensures that happens, is there?
3958 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3959 nothing. Used to print 'addi' vs 'add' instructions.
3963 No additional modifiers.
3967 - ``X``: Print an immediate integer as hexadecimal
3968 - ``x``: Print the low 16 bits of an immediate integer as hexadecimal.
3969 - ``d``: Print an immediate integer as decimal.
3970 - ``m``: Subtract one and print an immediate integer as decimal.
3971 - ``z``: Print $0 if an immediate zero, otherwise print normally.
3972 - ``L``: Print the low-order register of a two-register operand, or prints the
3973 address of the low-order word of a double-word memory operand.
3975 .. FIXME: L seems to be missing memory operand support.
3977 - ``M``: Print the high-order register of a two-register operand, or prints the
3978 address of the high-order word of a double-word memory operand.
3980 .. FIXME: M seems to be missing memory operand support.
3982 - ``D``: Print the second register of a two-register operand, or prints the
3983 second word of a double-word memory operand. (On a big-endian system, ``D`` is
3984 equivalent to ``L``, and on little-endian system, ``D`` is equivalent to
3986 - ``w``: No effect. Provided for compatibility with GCC which requires this
3987 modifier in order to print MSA registers (``W0-W31``) with the ``f``
3996 - ``L``: Print the second register of a two-register operand. Requires that it
3997 has been allocated consecutively to the first.
3999 .. FIXME: why is it restricted to consecutive ones? And there's
4000 nothing that ensures that happens, is there?
4002 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
4003 nothing. Used to print 'addi' vs 'add' instructions.
4004 - ``y``: For a memory operand, prints formatter for a two-register X-form
4005 instruction. (Currently always prints ``r0,OPERAND``).
4006 - ``U``: Prints 'u' if the memory operand is an update form, and nothing
4007 otherwise. (NOTE: LLVM does not support update form, so this will currently
4008 always print nothing)
4009 - ``X``: Prints 'x' if the memory operand is an indexed form. (NOTE: LLVM does
4010 not support indexed form, so this will currently always print nothing)
4018 SystemZ implements only ``n``, and does *not* support any of the other
4019 target-independent modifiers.
4023 - ``c``: Print an unadorned integer or symbol name. (The latter is
4024 target-specific behavior for this typically target-independent modifier).
4025 - ``A``: Print a register name with a '``*``' before it.
4026 - ``b``: Print an 8-bit register name (e.g. ``al``); do nothing on a memory
4028 - ``h``: Print the upper 8-bit register name (e.g. ``ah``); do nothing on a
4030 - ``w``: Print the 16-bit register name (e.g. ``ax``); do nothing on a memory
4032 - ``k``: Print the 32-bit register name (e.g. ``eax``); do nothing on a memory
4034 - ``q``: Print the 64-bit register name (e.g. ``rax``), if 64-bit registers are
4035 available, otherwise the 32-bit register name; do nothing on a memory operand.
4036 - ``n``: Negate and print an unadorned integer, or, for operands other than an
4037 immediate integer (e.g. a relocatable symbol expression), print a '-' before
4038 the operand. (The behavior for relocatable symbol expressions is a
4039 target-specific behavior for this typically target-independent modifier)
4040 - ``H``: Print a memory reference with additional offset +8.
4041 - ``P``: Print a memory reference or operand for use as the argument of a call
4042 instruction. (E.g. omit ``(rip)``, even though it's PC-relative.)
4046 No additional modifiers.
4052 The call instructions that wrap inline asm nodes may have a
4053 "``!srcloc``" MDNode attached to it that contains a list of constant
4054 integers. If present, the code generator will use the integer as the
4055 location cookie value when report errors through the ``LLVMContext``
4056 error reporting mechanisms. This allows a front-end to correlate backend
4057 errors that occur with inline asm back to the source code that produced
4060 .. code-block:: llvm
4062 call void asm sideeffect "something bad", ""(), !srcloc !42
4064 !42 = !{ i32 1234567 }
4066 It is up to the front-end to make sense of the magic numbers it places
4067 in the IR. If the MDNode contains multiple constants, the code generator
4068 will use the one that corresponds to the line of the asm that the error
4076 LLVM IR allows metadata to be attached to instructions in the program
4077 that can convey extra information about the code to the optimizers and
4078 code generator. One example application of metadata is source-level
4079 debug information. There are two metadata primitives: strings and nodes.
4081 Metadata does not have a type, and is not a value. If referenced from a
4082 ``call`` instruction, it uses the ``metadata`` type.
4084 All metadata are identified in syntax by a exclamation point ('``!``').
4086 .. _metadata-string:
4088 Metadata Nodes and Metadata Strings
4089 -----------------------------------
4091 A metadata string is a string surrounded by double quotes. It can
4092 contain any character by escaping non-printable characters with
4093 "``\xx``" where "``xx``" is the two digit hex code. For example:
4096 Metadata nodes are represented with notation similar to structure
4097 constants (a comma separated list of elements, surrounded by braces and
4098 preceded by an exclamation point). Metadata nodes can have any values as
4099 their operand. For example:
4101 .. code-block:: llvm
4103 !{ !"test\00", i32 10}
4105 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
4107 .. code-block:: text
4109 !0 = distinct !{!"test\00", i32 10}
4111 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
4112 content. They can also occur when transformations cause uniquing collisions
4113 when metadata operands change.
4115 A :ref:`named metadata <namedmetadatastructure>` is a collection of
4116 metadata nodes, which can be looked up in the module symbol table. For
4119 .. code-block:: llvm
4123 Metadata can be used as function arguments. Here the ``llvm.dbg.value``
4124 intrinsic is using three metadata arguments:
4126 .. code-block:: llvm
4128 call void @llvm.dbg.value(metadata !24, metadata !25, metadata !26)
4130 Metadata can be attached to an instruction. Here metadata ``!21`` is attached
4131 to the ``add`` instruction using the ``!dbg`` identifier:
4133 .. code-block:: llvm
4135 %indvar.next = add i64 %indvar, 1, !dbg !21
4137 Metadata can also be attached to a function or a global variable. Here metadata
4138 ``!22`` is attached to the ``f1`` and ``f2 functions, and the globals ``g1``
4139 and ``g2`` using the ``!dbg`` identifier:
4141 .. code-block:: llvm
4143 declare !dbg !22 void @f1()
4144 define void @f2() !dbg !22 {
4148 @g1 = global i32 0, !dbg !22
4149 @g2 = external global i32, !dbg !22
4151 A transformation is required to drop any metadata attachment that it does not
4152 know or know it can't preserve. Currently there is an exception for metadata
4153 attachment to globals for ``!type`` and ``!absolute_symbol`` which can't be
4154 unconditionally dropped unless the global is itself deleted.
4156 Metadata attached to a module using named metadata may not be dropped, with
4157 the exception of debug metadata (named metadata with the name ``!llvm.dbg.*``).
4159 More information about specific metadata nodes recognized by the
4160 optimizers and code generator is found below.
4162 .. _specialized-metadata:
4164 Specialized Metadata Nodes
4165 ^^^^^^^^^^^^^^^^^^^^^^^^^^
4167 Specialized metadata nodes are custom data structures in metadata (as opposed
4168 to generic tuples). Their fields are labelled, and can be specified in any
4171 These aren't inherently debug info centric, but currently all the specialized
4172 metadata nodes are related to debug info.
4179 ``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
4180 ``retainedTypes:``, ``globals:``, ``imports:`` and ``macros:`` fields are tuples
4181 containing the debug info to be emitted along with the compile unit, regardless
4182 of code optimizations (some nodes are only emitted if there are references to
4183 them from instructions). The ``debugInfoForProfiling:`` field is a boolean
4184 indicating whether or not line-table discriminators are updated to provide
4185 more-accurate debug info for profiling results.
4187 .. code-block:: text
4189 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
4190 isOptimized: true, flags: "-O2", runtimeVersion: 2,
4191 splitDebugFilename: "abc.debug", emissionKind: FullDebug,
4192 enums: !2, retainedTypes: !3, globals: !4, imports: !5,
4193 macros: !6, dwoId: 0x0abcd)
4195 Compile unit descriptors provide the root scope for objects declared in a
4196 specific compilation unit. File descriptors are defined using this scope. These
4197 descriptors are collected by a named metadata node ``!llvm.dbg.cu``. They keep
4198 track of global variables, type information, and imported entities (declarations
4206 ``DIFile`` nodes represent files. The ``filename:`` can include slashes.
4208 .. code-block:: none
4210 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir",
4211 checksumkind: CSK_MD5,
4212 checksum: "000102030405060708090a0b0c0d0e0f")
4214 Files are sometimes used in ``scope:`` fields, and are the only valid target
4215 for ``file:`` fields.
4216 Valid values for ``checksumkind:`` field are: {CSK_None, CSK_MD5, CSK_SHA1}
4223 ``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
4224 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
4226 .. code-block:: text
4228 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
4229 encoding: DW_ATE_unsigned_char)
4230 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
4232 The ``encoding:`` describes the details of the type. Usually it's one of the
4235 .. code-block:: text
4241 DW_ATE_signed_char = 6
4243 DW_ATE_unsigned_char = 8
4245 .. _DISubroutineType:
4250 ``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
4251 refers to a tuple; the first operand is the return type, while the rest are the
4252 types of the formal arguments in order. If the first operand is ``null``, that
4253 represents a function with no return value (such as ``void foo() {}`` in C++).
4255 .. code-block:: text
4257 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
4258 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
4259 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
4266 ``DIDerivedType`` nodes represent types derived from other types, such as
4269 .. code-block:: text
4271 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
4272 encoding: DW_ATE_unsigned_char)
4273 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
4276 The following ``tag:`` values are valid:
4278 .. code-block:: text
4281 DW_TAG_pointer_type = 15
4282 DW_TAG_reference_type = 16
4284 DW_TAG_inheritance = 28
4285 DW_TAG_ptr_to_member_type = 31
4286 DW_TAG_const_type = 38
4288 DW_TAG_volatile_type = 53
4289 DW_TAG_restrict_type = 55
4290 DW_TAG_atomic_type = 71
4292 .. _DIDerivedTypeMember:
4294 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
4295 <DICompositeType>`. The type of the member is the ``baseType:``. The
4296 ``offset:`` is the member's bit offset. If the composite type has an ODR
4297 ``identifier:`` and does not set ``flags: DIFwdDecl``, then the member is
4298 uniqued based only on its ``name:`` and ``scope:``.
4300 ``DW_TAG_inheritance`` and ``DW_TAG_friend`` are used in the ``elements:``
4301 field of :ref:`composite types <DICompositeType>` to describe parents and
4304 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
4306 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
4307 ``DW_TAG_volatile_type``, ``DW_TAG_restrict_type`` and ``DW_TAG_atomic_type``
4308 are used to qualify the ``baseType:``.
4310 Note that the ``void *`` type is expressed as a type derived from NULL.
4312 .. _DICompositeType:
4317 ``DICompositeType`` nodes represent types composed of other types, like
4318 structures and unions. ``elements:`` points to a tuple of the composed types.
4320 If the source language supports ODR, the ``identifier:`` field gives the unique
4321 identifier used for type merging between modules. When specified,
4322 :ref:`subprogram declarations <DISubprogramDeclaration>` and :ref:`member
4323 derived types <DIDerivedTypeMember>` that reference the ODR-type in their
4324 ``scope:`` change uniquing rules.
4326 For a given ``identifier:``, there should only be a single composite type that
4327 does not have ``flags: DIFlagFwdDecl`` set. LLVM tools that link modules
4328 together will unique such definitions at parse time via the ``identifier:``
4329 field, even if the nodes are ``distinct``.
4331 .. code-block:: text
4333 !0 = !DIEnumerator(name: "SixKind", value: 7)
4334 !1 = !DIEnumerator(name: "SevenKind", value: 7)
4335 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
4336 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
4337 line: 2, size: 32, align: 32, identifier: "_M4Enum",
4338 elements: !{!0, !1, !2})
4340 The following ``tag:`` values are valid:
4342 .. code-block:: text
4344 DW_TAG_array_type = 1
4345 DW_TAG_class_type = 2
4346 DW_TAG_enumeration_type = 4
4347 DW_TAG_structure_type = 19
4348 DW_TAG_union_type = 23
4350 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
4351 descriptors <DISubrange>`, each representing the range of subscripts at that
4352 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
4353 array type is a native packed vector.
4355 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
4356 descriptors <DIEnumerator>`, each representing the definition of an enumeration
4357 value for the set. All enumeration type descriptors are collected in the
4358 ``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
4360 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
4361 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
4362 <DIDerivedType>` with ``tag: DW_TAG_member``, ``tag: DW_TAG_inheritance``, or
4363 ``tag: DW_TAG_friend``; or :ref:`subprograms <DISubprogram>` with
4364 ``isDefinition: false``.
4371 ``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
4372 :ref:`DICompositeType`.
4374 - ``count: -1`` indicates an empty array.
4375 - ``count: !9`` describes the count with a :ref:`DILocalVariable`.
4376 - ``count: !11`` describes the count with a :ref:`DIGlobalVariable`.
4378 .. code-block:: llvm
4380 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
4381 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
4382 !2 = !DISubrange(count: -1) ; empty array.
4384 ; Scopes used in rest of example
4385 !6 = !DIFile(filename: "vla.c", directory: "/path/to/file")
4386 !7 = distinct !DICompileUnit(language: DW_LANG_C99, ...
4387 !8 = distinct !DISubprogram(name: "foo", scope: !7, file: !6, line: 5, ...
4389 ; Use of local variable as count value
4390 !9 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
4391 !10 = !DILocalVariable(name: "count", scope: !8, file: !6, line: 42, type: !9)
4392 !11 = !DISubrange(count !10, lowerBound: 0)
4394 ; Use of global variable as count value
4395 !12 = !DIGlobalVariable(name: "count", scope: !8, file: !6, line: 22, type: !9)
4396 !13 = !DISubrange(count !12, lowerBound: 0)
4403 ``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
4404 variants of :ref:`DICompositeType`.
4406 .. code-block:: llvm
4408 !0 = !DIEnumerator(name: "SixKind", value: 7)
4409 !1 = !DIEnumerator(name: "SevenKind", value: 7)
4410 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
4412 DITemplateTypeParameter
4413 """""""""""""""""""""""
4415 ``DITemplateTypeParameter`` nodes represent type parameters to generic source
4416 language constructs. They are used (optionally) in :ref:`DICompositeType` and
4417 :ref:`DISubprogram` ``templateParams:`` fields.
4419 .. code-block:: llvm
4421 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
4423 DITemplateValueParameter
4424 """"""""""""""""""""""""
4426 ``DITemplateValueParameter`` nodes represent value parameters to generic source
4427 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
4428 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
4429 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
4430 :ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
4432 .. code-block:: llvm
4434 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
4439 ``DINamespace`` nodes represent namespaces in the source language.
4441 .. code-block:: llvm
4443 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
4445 .. _DIGlobalVariable:
4450 ``DIGlobalVariable`` nodes represent global variables in the source language.
4452 .. code-block:: llvm
4454 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
4455 file: !2, line: 7, type: !3, isLocal: true,
4456 isDefinition: false, variable: i32* @foo,
4459 All global variables should be referenced by the `globals:` field of a
4460 :ref:`compile unit <DICompileUnit>`.
4467 ``DISubprogram`` nodes represent functions from the source language. A
4468 ``DISubprogram`` may be attached to a function definition using ``!dbg``
4469 metadata. The ``variables:`` field points at :ref:`variables <DILocalVariable>`
4470 that must be retained, even if their IR counterparts are optimized out of
4471 the IR. The ``type:`` field must point at an :ref:`DISubroutineType`.
4473 .. _DISubprogramDeclaration:
4475 When ``isDefinition: false``, subprograms describe a declaration in the type
4476 tree as opposed to a definition of a function. If the scope is a composite
4477 type with an ODR ``identifier:`` and that does not set ``flags: DIFwdDecl``,
4478 then the subprogram declaration is uniqued based only on its ``linkageName:``
4481 .. code-block:: text
4483 define void @_Z3foov() !dbg !0 {
4487 !0 = distinct !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
4488 file: !2, line: 7, type: !3, isLocal: true,
4489 isDefinition: true, scopeLine: 8,
4491 virtuality: DW_VIRTUALITY_pure_virtual,
4492 virtualIndex: 10, flags: DIFlagPrototyped,
4493 isOptimized: true, unit: !5, templateParams: !6,
4494 declaration: !7, variables: !8, thrownTypes: !9)
4501 ``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
4502 <DISubprogram>`. The line number and column numbers are used to distinguish
4503 two lexical blocks at same depth. They are valid targets for ``scope:``
4506 .. code-block:: text
4508 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
4510 Usually lexical blocks are ``distinct`` to prevent node merging based on
4513 .. _DILexicalBlockFile:
4518 ``DILexicalBlockFile`` nodes are used to discriminate between sections of a
4519 :ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
4520 indicate textual inclusion, or the ``discriminator:`` field can be used to
4521 discriminate between control flow within a single block in the source language.
4523 .. code-block:: llvm
4525 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
4526 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
4527 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
4534 ``DILocation`` nodes represent source debug locations. The ``scope:`` field is
4535 mandatory, and points at an :ref:`DILexicalBlockFile`, an
4536 :ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
4538 .. code-block:: llvm
4540 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
4542 .. _DILocalVariable:
4547 ``DILocalVariable`` nodes represent local variables in the source language. If
4548 the ``arg:`` field is set to non-zero, then this variable is a subprogram
4549 parameter, and it will be included in the ``variables:`` field of its
4550 :ref:`DISubprogram`.
4552 .. code-block:: text
4554 !0 = !DILocalVariable(name: "this", arg: 1, scope: !3, file: !2, line: 7,
4555 type: !3, flags: DIFlagArtificial)
4556 !1 = !DILocalVariable(name: "x", arg: 2, scope: !4, file: !2, line: 7,
4558 !2 = !DILocalVariable(name: "y", scope: !5, file: !2, line: 7, type: !3)
4563 ``DIExpression`` nodes represent expressions that are inspired by the DWARF
4564 expression language. They are used in :ref:`debug intrinsics<dbg_intrinsics>`
4565 (such as ``llvm.dbg.declare`` and ``llvm.dbg.value``) to describe how the
4566 referenced LLVM variable relates to the source language variable.
4568 The current supported vocabulary is limited:
4570 - ``DW_OP_deref`` dereferences the top of the expression stack.
4571 - ``DW_OP_plus`` pops the last two entries from the expression stack, adds
4572 them together and appends the result to the expression stack.
4573 - ``DW_OP_minus`` pops the last two entries from the expression stack, subtracts
4574 the last entry from the second last entry and appends the result to the
4576 - ``DW_OP_plus_uconst, 93`` adds ``93`` to the working expression.
4577 - ``DW_OP_LLVM_fragment, 16, 8`` specifies the offset and size (``16`` and ``8``
4578 here, respectively) of the variable fragment from the working expression. Note
4579 that contrary to DW_OP_bit_piece, the offset is describing the location
4580 within the described source variable.
4581 - ``DW_OP_swap`` swaps top two stack entries.
4582 - ``DW_OP_xderef`` provides extended dereference mechanism. The entry at the top
4583 of the stack is treated as an address. The second stack entry is treated as an
4584 address space identifier.
4585 - ``DW_OP_stack_value`` marks a constant value.
4587 DWARF specifies three kinds of simple location descriptions: Register, memory,
4588 and implicit location descriptions. Register and memory location descriptions
4589 describe the *location* of a source variable (in the sense that a debugger might
4590 modify its value), whereas implicit locations describe merely the *value* of a
4591 source variable. DIExpressions also follow this model: A DIExpression that
4592 doesn't have a trailing ``DW_OP_stack_value`` will describe an *address* when
4593 combined with a concrete location.
4595 .. code-block:: text
4597 !0 = !DIExpression(DW_OP_deref)
4598 !1 = !DIExpression(DW_OP_plus_uconst, 3)
4599 !1 = !DIExpression(DW_OP_constu, 3, DW_OP_plus)
4600 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
4601 !3 = !DIExpression(DW_OP_deref, DW_OP_constu, 3, DW_OP_plus, DW_OP_LLVM_fragment, 3, 7)
4602 !4 = !DIExpression(DW_OP_constu, 2, DW_OP_swap, DW_OP_xderef)
4603 !5 = !DIExpression(DW_OP_constu, 42, DW_OP_stack_value)
4608 ``DIObjCProperty`` nodes represent Objective-C property nodes.
4610 .. code-block:: llvm
4612 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
4613 getter: "getFoo", attributes: 7, type: !2)
4618 ``DIImportedEntity`` nodes represent entities (such as modules) imported into a
4621 .. code-block:: text
4623 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
4624 entity: !1, line: 7)
4629 ``DIMacro`` nodes represent definition or undefinition of a macro identifiers.
4630 The ``name:`` field is the macro identifier, followed by macro parameters when
4631 defining a function-like macro, and the ``value`` field is the token-string
4632 used to expand the macro identifier.
4634 .. code-block:: text
4636 !2 = !DIMacro(macinfo: DW_MACINFO_define, line: 7, name: "foo(x)",
4638 !3 = !DIMacro(macinfo: DW_MACINFO_undef, line: 30, name: "foo")
4643 ``DIMacroFile`` nodes represent inclusion of source files.
4644 The ``nodes:`` field is a list of ``DIMacro`` and ``DIMacroFile`` nodes that
4645 appear in the included source file.
4647 .. code-block:: text
4649 !2 = !DIMacroFile(macinfo: DW_MACINFO_start_file, line: 7, file: !2,
4655 In LLVM IR, memory does not have types, so LLVM's own type system is not
4656 suitable for doing type based alias analysis (TBAA). Instead, metadata is
4657 added to the IR to describe a type system of a higher level language. This
4658 can be used to implement C/C++ strict type aliasing rules, but it can also
4659 be used to implement custom alias analysis behavior for other languages.
4661 This description of LLVM's TBAA system is broken into two parts:
4662 :ref:`Semantics<tbaa_node_semantics>` talks about high level issues, and
4663 :ref:`Representation<tbaa_node_representation>` talks about the metadata
4664 encoding of various entities.
4666 It is always possible to trace any TBAA node to a "root" TBAA node (details
4667 in the :ref:`Representation<tbaa_node_representation>` section). TBAA
4668 nodes with different roots have an unknown aliasing relationship, and LLVM
4669 conservatively infers ``MayAlias`` between them. The rules mentioned in
4670 this section only pertain to TBAA nodes living under the same root.
4672 .. _tbaa_node_semantics:
4677 The TBAA metadata system, referred to as "struct path TBAA" (not to be
4678 confused with ``tbaa.struct``), consists of the following high level
4679 concepts: *Type Descriptors*, further subdivided into scalar type
4680 descriptors and struct type descriptors; and *Access Tags*.
4682 **Type descriptors** describe the type system of the higher level language
4683 being compiled. **Scalar type descriptors** describe types that do not
4684 contain other types. Each scalar type has a parent type, which must also
4685 be a scalar type or the TBAA root. Via this parent relation, scalar types
4686 within a TBAA root form a tree. **Struct type descriptors** denote types
4687 that contain a sequence of other type descriptors, at known offsets. These
4688 contained type descriptors can either be struct type descriptors themselves
4689 or scalar type descriptors.
4691 **Access tags** are metadata nodes attached to load and store instructions.
4692 Access tags use type descriptors to describe the *location* being accessed
4693 in terms of the type system of the higher level language. Access tags are
4694 tuples consisting of a base type, an access type and an offset. The base
4695 type is a scalar type descriptor or a struct type descriptor, the access
4696 type is a scalar type descriptor, and the offset is a constant integer.
4698 The access tag ``(BaseTy, AccessTy, Offset)`` can describe one of two
4701 * If ``BaseTy`` is a struct type, the tag describes a memory access (load
4702 or store) of a value of type ``AccessTy`` contained in the struct type
4703 ``BaseTy`` at offset ``Offset``.
4705 * If ``BaseTy`` is a scalar type, ``Offset`` must be 0 and ``BaseTy`` and
4706 ``AccessTy`` must be the same; and the access tag describes a scalar
4707 access with scalar type ``AccessTy``.
4709 We first define an ``ImmediateParent`` relation on ``(BaseTy, Offset)``
4712 * If ``BaseTy`` is a scalar type then ``ImmediateParent(BaseTy, 0)`` is
4713 ``(ParentTy, 0)`` where ``ParentTy`` is the parent of the scalar type as
4714 described in the TBAA metadata. ``ImmediateParent(BaseTy, Offset)`` is
4715 undefined if ``Offset`` is non-zero.
4717 * If ``BaseTy`` is a struct type then ``ImmediateParent(BaseTy, Offset)``
4718 is ``(NewTy, NewOffset)`` where ``NewTy`` is the type contained in
4719 ``BaseTy`` at offset ``Offset`` and ``NewOffset`` is ``Offset`` adjusted
4720 to be relative within that inner type.
4722 A memory access with an access tag ``(BaseTy1, AccessTy1, Offset1)``
4723 aliases a memory access with an access tag ``(BaseTy2, AccessTy2,
4724 Offset2)`` if either ``(BaseTy1, Offset1)`` is reachable from ``(Base2,
4725 Offset2)`` via the ``Parent`` relation or vice versa.
4727 As a concrete example, the type descriptor graph for the following program
4733 float f; // offset 4
4737 float f; // offset 0
4738 double d; // offset 4
4739 struct Inner inner_a; // offset 12
4742 void f(struct Outer* outer, struct Inner* inner, float* f, int* i, char* c) {
4743 outer->f = 0; // tag0: (OuterStructTy, FloatScalarTy, 0)
4744 outer->inner_a.i = 0; // tag1: (OuterStructTy, IntScalarTy, 12)
4745 outer->inner_a.f = 0.0; // tag2: (OuterStructTy, IntScalarTy, 16)
4746 *f = 0.0; // tag3: (FloatScalarTy, FloatScalarTy, 0)
4749 is (note that in C and C++, ``char`` can be used to access any arbitrary
4752 .. code-block:: text
4755 CharScalarTy = ("char", Root, 0)
4756 FloatScalarTy = ("float", CharScalarTy, 0)
4757 DoubleScalarTy = ("double", CharScalarTy, 0)
4758 IntScalarTy = ("int", CharScalarTy, 0)
4759 InnerStructTy = {"Inner" (IntScalarTy, 0), (FloatScalarTy, 4)}
4760 OuterStructTy = {"Outer", (FloatScalarTy, 0), (DoubleScalarTy, 4),
4761 (InnerStructTy, 12)}
4764 with (e.g.) ``ImmediateParent(OuterStructTy, 12)`` = ``(InnerStructTy,
4765 0)``, ``ImmediateParent(InnerStructTy, 0)`` = ``(IntScalarTy, 0)``, and
4766 ``ImmediateParent(IntScalarTy, 0)`` = ``(CharScalarTy, 0)``.
4768 .. _tbaa_node_representation:
4773 The root node of a TBAA type hierarchy is an ``MDNode`` with 0 operands or
4774 with exactly one ``MDString`` operand.
4776 Scalar type descriptors are represented as an ``MDNode`` s with two
4777 operands. The first operand is an ``MDString`` denoting the name of the
4778 struct type. LLVM does not assign meaning to the value of this operand, it
4779 only cares about it being an ``MDString``. The second operand is an
4780 ``MDNode`` which points to the parent for said scalar type descriptor,
4781 which is either another scalar type descriptor or the TBAA root. Scalar
4782 type descriptors can have an optional third argument, but that must be the
4783 constant integer zero.
4785 Struct type descriptors are represented as ``MDNode`` s with an odd number
4786 of operands greater than 1. The first operand is an ``MDString`` denoting
4787 the name of the struct type. Like in scalar type descriptors the actual
4788 value of this name operand is irrelevant to LLVM. After the name operand,
4789 the struct type descriptors have a sequence of alternating ``MDNode`` and
4790 ``ConstantInt`` operands. With N starting from 1, the 2N - 1 th operand,
4791 an ``MDNode``, denotes a contained field, and the 2N th operand, a
4792 ``ConstantInt``, is the offset of the said contained field. The offsets
4793 must be in non-decreasing order.
4795 Access tags are represented as ``MDNode`` s with either 3 or 4 operands.
4796 The first operand is an ``MDNode`` pointing to the node representing the
4797 base type. The second operand is an ``MDNode`` pointing to the node
4798 representing the access type. The third operand is a ``ConstantInt`` that
4799 states the offset of the access. If a fourth field is present, it must be
4800 a ``ConstantInt`` valued at 0 or 1. If it is 1 then the access tag states
4801 that the location being accessed is "constant" (meaning
4802 ``pointsToConstantMemory`` should return true; see `other useful
4803 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_). The TBAA root of
4804 the access type and the base type of an access tag must be the same, and
4805 that is the TBAA root of the access tag.
4807 '``tbaa.struct``' Metadata
4808 ^^^^^^^^^^^^^^^^^^^^^^^^^^
4810 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
4811 aggregate assignment operations in C and similar languages, however it
4812 is defined to copy a contiguous region of memory, which is more than
4813 strictly necessary for aggregate types which contain holes due to
4814 padding. Also, it doesn't contain any TBAA information about the fields
4817 ``!tbaa.struct`` metadata can describe which memory subregions in a
4818 memcpy are padding and what the TBAA tags of the struct are.
4820 The current metadata format is very simple. ``!tbaa.struct`` metadata
4821 nodes are a list of operands which are in conceptual groups of three.
4822 For each group of three, the first operand gives the byte offset of a
4823 field in bytes, the second gives its size in bytes, and the third gives
4826 .. code-block:: llvm
4828 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
4830 This describes a struct with two fields. The first is at offset 0 bytes
4831 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
4832 and has size 4 bytes and has tbaa tag !2.
4834 Note that the fields need not be contiguous. In this example, there is a
4835 4 byte gap between the two fields. This gap represents padding which
4836 does not carry useful data and need not be preserved.
4838 '``noalias``' and '``alias.scope``' Metadata
4839 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4841 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
4842 noalias memory-access sets. This means that some collection of memory access
4843 instructions (loads, stores, memory-accessing calls, etc.) that carry
4844 ``noalias`` metadata can specifically be specified not to alias with some other
4845 collection of memory access instructions that carry ``alias.scope`` metadata.
4846 Each type of metadata specifies a list of scopes where each scope has an id and
4849 When evaluating an aliasing query, if for some domain, the set
4850 of scopes with that domain in one instruction's ``alias.scope`` list is a
4851 subset of (or equal to) the set of scopes for that domain in another
4852 instruction's ``noalias`` list, then the two memory accesses are assumed not to
4855 Because scopes in one domain don't affect scopes in other domains, separate
4856 domains can be used to compose multiple independent noalias sets. This is
4857 used for example during inlining. As the noalias function parameters are
4858 turned into noalias scope metadata, a new domain is used every time the
4859 function is inlined.
4861 The metadata identifying each domain is itself a list containing one or two
4862 entries. The first entry is the name of the domain. Note that if the name is a
4863 string then it can be combined across functions and translation units. A
4864 self-reference can be used to create globally unique domain names. A
4865 descriptive string may optionally be provided as a second list entry.
4867 The metadata identifying each scope is also itself a list containing two or
4868 three entries. The first entry is the name of the scope. Note that if the name
4869 is a string then it can be combined across functions and translation units. A
4870 self-reference can be used to create globally unique scope names. A metadata
4871 reference to the scope's domain is the second entry. A descriptive string may
4872 optionally be provided as a third list entry.
4876 .. code-block:: llvm
4878 ; Two scope domains:
4882 ; Some scopes in these domains:
4888 !5 = !{!4} ; A list containing only scope !4
4892 ; These two instructions don't alias:
4893 %0 = load float, float* %c, align 4, !alias.scope !5
4894 store float %0, float* %arrayidx.i, align 4, !noalias !5
4896 ; These two instructions also don't alias (for domain !1, the set of scopes
4897 ; in the !alias.scope equals that in the !noalias list):
4898 %2 = load float, float* %c, align 4, !alias.scope !5
4899 store float %2, float* %arrayidx.i2, align 4, !noalias !6
4901 ; These two instructions may alias (for domain !0, the set of scopes in
4902 ; the !noalias list is not a superset of, or equal to, the scopes in the
4903 ; !alias.scope list):
4904 %2 = load float, float* %c, align 4, !alias.scope !6
4905 store float %0, float* %arrayidx.i, align 4, !noalias !7
4907 '``fpmath``' Metadata
4908 ^^^^^^^^^^^^^^^^^^^^^
4910 ``fpmath`` metadata may be attached to any instruction of floating-point
4911 type. It can be used to express the maximum acceptable error in the
4912 result of that instruction, in ULPs, thus potentially allowing the
4913 compiler to use a more efficient but less accurate method of computing
4914 it. ULP is defined as follows:
4916 If ``x`` is a real number that lies between two finite consecutive
4917 floating-point numbers ``a`` and ``b``, without being equal to one
4918 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
4919 distance between the two non-equal finite floating-point numbers
4920 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
4922 The metadata node shall consist of a single positive float type number
4923 representing the maximum relative error, for example:
4925 .. code-block:: llvm
4927 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
4931 '``range``' Metadata
4932 ^^^^^^^^^^^^^^^^^^^^
4934 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
4935 integer types. It expresses the possible ranges the loaded value or the value
4936 returned by the called function at this call site is in. The ranges are
4937 represented with a flattened list of integers. The loaded value or the value
4938 returned is known to be in the union of the ranges defined by each consecutive
4939 pair. Each pair has the following properties:
4941 - The type must match the type loaded by the instruction.
4942 - The pair ``a,b`` represents the range ``[a,b)``.
4943 - Both ``a`` and ``b`` are constants.
4944 - The range is allowed to wrap.
4945 - The range should not represent the full or empty set. That is,
4948 In addition, the pairs must be in signed order of the lower bound and
4949 they must be non-contiguous.
4953 .. code-block:: llvm
4955 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
4956 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
4957 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
4958 %d = invoke i8 @bar() to label %cont
4959 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
4961 !0 = !{ i8 0, i8 2 }
4962 !1 = !{ i8 255, i8 2 }
4963 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
4964 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
4966 '``absolute_symbol``' Metadata
4967 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4969 ``absolute_symbol`` metadata may be attached to a global variable
4970 declaration. It marks the declaration as a reference to an absolute symbol,
4971 which causes the backend to use absolute relocations for the symbol even
4972 in position independent code, and expresses the possible ranges that the
4973 global variable's *address* (not its value) is in, in the same format as
4974 ``range`` metadata, with the extension that the pair ``all-ones,all-ones``
4975 may be used to represent the full set.
4977 Example (assuming 64-bit pointers):
4979 .. code-block:: llvm
4981 @a = external global i8, !absolute_symbol !0 ; Absolute symbol in range [0,256)
4982 @b = external global i8, !absolute_symbol !1 ; Absolute symbol in range [0,2^64)
4985 !0 = !{ i64 0, i64 256 }
4986 !1 = !{ i64 -1, i64 -1 }
4988 '``callees``' Metadata
4989 ^^^^^^^^^^^^^^^^^^^^^^
4991 ``callees`` metadata may be attached to indirect call sites. If ``callees``
4992 metadata is attached to a call site, and any callee is not among the set of
4993 functions provided by the metadata, the behavior is undefined. The intent of
4994 this metadata is to facilitate optimizations such as indirect-call promotion.
4995 For example, in the code below, the call instruction may only target the
4996 ``add`` or ``sub`` functions:
4998 .. code-block:: llvm
5000 %result = call i64 %binop(i64 %x, i64 %y), !callees !0
5003 !0 = !{i64 (i64, i64)* @add, i64 (i64, i64)* @sub}
5005 '``unpredictable``' Metadata
5006 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5008 ``unpredictable`` metadata may be attached to any branch or switch
5009 instruction. It can be used to express the unpredictability of control
5010 flow. Similar to the llvm.expect intrinsic, it may be used to alter
5011 optimizations related to compare and branch instructions. The metadata
5012 is treated as a boolean value; if it exists, it signals that the branch
5013 or switch that it is attached to is completely unpredictable.
5018 It is sometimes useful to attach information to loop constructs. Currently,
5019 loop metadata is implemented as metadata attached to the branch instruction
5020 in the loop latch block. This type of metadata refer to a metadata node that is
5021 guaranteed to be separate for each loop. The loop identifier metadata is
5022 specified with the name ``llvm.loop``.
5024 The loop identifier metadata is implemented using a metadata that refers to
5025 itself to avoid merging it with any other identifier metadata, e.g.,
5026 during module linkage or function inlining. That is, each loop should refer
5027 to their own identification metadata even if they reside in separate functions.
5028 The following example contains loop identifier metadata for two separate loop
5031 .. code-block:: llvm
5036 The loop identifier metadata can be used to specify additional
5037 per-loop metadata. Any operands after the first operand can be treated
5038 as user-defined metadata. For example the ``llvm.loop.unroll.count``
5039 suggests an unroll factor to the loop unroller:
5041 .. code-block:: llvm
5043 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
5046 !1 = !{!"llvm.loop.unroll.count", i32 4}
5048 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
5049 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5051 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
5052 used to control per-loop vectorization and interleaving parameters such as
5053 vectorization width and interleave count. These metadata should be used in
5054 conjunction with ``llvm.loop`` loop identification metadata. The
5055 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
5056 optimization hints and the optimizer will only interleave and vectorize loops if
5057 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
5058 which contains information about loop-carried memory dependencies can be helpful
5059 in determining the safety of these transformations.
5061 '``llvm.loop.interleave.count``' Metadata
5062 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5064 This metadata suggests an interleave count to the loop interleaver.
5065 The first operand is the string ``llvm.loop.interleave.count`` and the
5066 second operand is an integer specifying the interleave count. For
5069 .. code-block:: llvm
5071 !0 = !{!"llvm.loop.interleave.count", i32 4}
5073 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
5074 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
5075 then the interleave count will be determined automatically.
5077 '``llvm.loop.vectorize.enable``' Metadata
5078 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5080 This metadata selectively enables or disables vectorization for the loop. The
5081 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
5082 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
5083 0 disables vectorization:
5085 .. code-block:: llvm
5087 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
5088 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
5090 '``llvm.loop.vectorize.width``' Metadata
5091 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5093 This metadata sets the target width of the vectorizer. The first
5094 operand is the string ``llvm.loop.vectorize.width`` and the second
5095 operand is an integer specifying the width. For example:
5097 .. code-block:: llvm
5099 !0 = !{!"llvm.loop.vectorize.width", i32 4}
5101 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
5102 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
5103 0 or if the loop does not have this metadata the width will be
5104 determined automatically.
5106 '``llvm.loop.unroll``'
5107 ^^^^^^^^^^^^^^^^^^^^^^
5109 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
5110 optimization hints such as the unroll factor. ``llvm.loop.unroll``
5111 metadata should be used in conjunction with ``llvm.loop`` loop
5112 identification metadata. The ``llvm.loop.unroll`` metadata are only
5113 optimization hints and the unrolling will only be performed if the
5114 optimizer believes it is safe to do so.
5116 '``llvm.loop.unroll.count``' Metadata
5117 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5119 This metadata suggests an unroll factor to the loop unroller. The
5120 first operand is the string ``llvm.loop.unroll.count`` and the second
5121 operand is a positive integer specifying the unroll factor. For
5124 .. code-block:: llvm
5126 !0 = !{!"llvm.loop.unroll.count", i32 4}
5128 If the trip count of the loop is less than the unroll count the loop
5129 will be partially unrolled.
5131 '``llvm.loop.unroll.disable``' Metadata
5132 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5134 This metadata disables loop unrolling. The metadata has a single operand
5135 which is the string ``llvm.loop.unroll.disable``. For example:
5137 .. code-block:: llvm
5139 !0 = !{!"llvm.loop.unroll.disable"}
5141 '``llvm.loop.unroll.runtime.disable``' Metadata
5142 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5144 This metadata disables runtime loop unrolling. The metadata has a single
5145 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
5147 .. code-block:: llvm
5149 !0 = !{!"llvm.loop.unroll.runtime.disable"}
5151 '``llvm.loop.unroll.enable``' Metadata
5152 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5154 This metadata suggests that the loop should be fully unrolled if the trip count
5155 is known at compile time and partially unrolled if the trip count is not known
5156 at compile time. The metadata has a single operand which is the string
5157 ``llvm.loop.unroll.enable``. For example:
5159 .. code-block:: llvm
5161 !0 = !{!"llvm.loop.unroll.enable"}
5163 '``llvm.loop.unroll.full``' Metadata
5164 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5166 This metadata suggests that the loop should be unrolled fully. The
5167 metadata has a single operand which is the string ``llvm.loop.unroll.full``.
5170 .. code-block:: llvm
5172 !0 = !{!"llvm.loop.unroll.full"}
5174 '``llvm.loop.licm_versioning.disable``' Metadata
5175 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5177 This metadata indicates that the loop should not be versioned for the purpose
5178 of enabling loop-invariant code motion (LICM). The metadata has a single operand
5179 which is the string ``llvm.loop.licm_versioning.disable``. For example:
5181 .. code-block:: llvm
5183 !0 = !{!"llvm.loop.licm_versioning.disable"}
5185 '``llvm.loop.distribute.enable``' Metadata
5186 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5188 Loop distribution allows splitting a loop into multiple loops. Currently,
5189 this is only performed if the entire loop cannot be vectorized due to unsafe
5190 memory dependencies. The transformation will attempt to isolate the unsafe
5191 dependencies into their own loop.
5193 This metadata can be used to selectively enable or disable distribution of the
5194 loop. The first operand is the string ``llvm.loop.distribute.enable`` and the
5195 second operand is a bit. If the bit operand value is 1 distribution is
5196 enabled. A value of 0 disables distribution:
5198 .. code-block:: llvm
5200 !0 = !{!"llvm.loop.distribute.enable", i1 0}
5201 !1 = !{!"llvm.loop.distribute.enable", i1 1}
5203 This metadata should be used in conjunction with ``llvm.loop`` loop
5204 identification metadata.
5209 Metadata types used to annotate memory accesses with information helpful
5210 for optimizations are prefixed with ``llvm.mem``.
5212 '``llvm.mem.parallel_loop_access``' Metadata
5213 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5215 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
5216 or metadata containing a list of loop identifiers for nested loops.
5217 The metadata is attached to memory accessing instructions and denotes that
5218 no loop carried memory dependence exist between it and other instructions denoted
5219 with the same loop identifier. The metadata on memory reads also implies that
5220 if conversion (i.e. speculative execution within a loop iteration) is safe.
5222 Precisely, given two instructions ``m1`` and ``m2`` that both have the
5223 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
5224 set of loops associated with that metadata, respectively, then there is no loop
5225 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
5228 As a special case, if all memory accessing instructions in a loop have
5229 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
5230 loop has no loop carried memory dependences and is considered to be a parallel
5233 Note that if not all memory access instructions have such metadata referring to
5234 the loop, then the loop is considered not being trivially parallel. Additional
5235 memory dependence analysis is required to make that determination. As a fail
5236 safe mechanism, this causes loops that were originally parallel to be considered
5237 sequential (if optimization passes that are unaware of the parallel semantics
5238 insert new memory instructions into the loop body).
5240 Example of a loop that is considered parallel due to its correct use of
5241 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
5242 metadata types that refer to the same loop identifier metadata.
5244 .. code-block:: llvm
5248 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
5250 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
5252 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
5258 It is also possible to have nested parallel loops. In that case the
5259 memory accesses refer to a list of loop identifier metadata nodes instead of
5260 the loop identifier metadata node directly:
5262 .. code-block:: llvm
5266 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
5268 br label %inner.for.body
5272 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
5274 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
5276 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
5280 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
5282 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
5284 outer.for.end: ; preds = %for.body
5286 !0 = !{!1, !2} ; a list of loop identifiers
5287 !1 = !{!1} ; an identifier for the inner loop
5288 !2 = !{!2} ; an identifier for the outer loop
5290 '``irr_loop``' Metadata
5291 ^^^^^^^^^^^^^^^^^^^^^^^
5293 ``irr_loop`` metadata may be attached to the terminator instruction of a basic
5294 block that's an irreducible loop header (note that an irreducible loop has more
5295 than once header basic blocks.) If ``irr_loop`` metadata is attached to the
5296 terminator instruction of a basic block that is not really an irreducible loop
5297 header, the behavior is undefined. The intent of this metadata is to improve the
5298 accuracy of the block frequency propagation. For example, in the code below, the
5299 block ``header0`` may have a loop header weight (relative to the other headers of
5300 the irreducible loop) of 100:
5302 .. code-block:: llvm
5306 br i1 %cmp, label %t1, label %t2, !irr_loop !0
5309 !0 = !{"loop_header_weight", i64 100}
5311 Irreducible loop header weights are typically based on profile data.
5313 '``invariant.group``' Metadata
5314 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5316 The experimental ``invariant.group`` metadata may be attached to
5317 ``load``/``store`` instructions referencing a single metadata with no entries.
5318 The existence of the ``invariant.group`` metadata on the instruction tells
5319 the optimizer that every ``load`` and ``store`` to the same pointer operand
5320 can be assumed to load or store the same
5321 value (but see the ``llvm.launder.invariant.group`` intrinsic which affects
5322 when two pointers are considered the same). Pointers returned by bitcast or
5323 getelementptr with only zero indices are considered the same.
5327 .. code-block:: llvm
5329 @unknownPtr = external global i8
5332 store i8 42, i8* %ptr, !invariant.group !0
5333 call void @foo(i8* %ptr)
5335 %a = load i8, i8* %ptr, !invariant.group !0 ; Can assume that value under %ptr didn't change
5336 call void @foo(i8* %ptr)
5338 %newPtr = call i8* @getPointer(i8* %ptr)
5339 %c = load i8, i8* %newPtr, !invariant.group !0 ; Can't assume anything, because we only have information about %ptr
5341 %unknownValue = load i8, i8* @unknownPtr
5342 store i8 %unknownValue, i8* %ptr, !invariant.group !0 ; Can assume that %unknownValue == 42
5344 call void @foo(i8* %ptr)
5345 %newPtr2 = call i8* @llvm.launder.invariant.group(i8* %ptr)
5346 %d = load i8, i8* %newPtr2, !invariant.group !0 ; Can't step through launder.invariant.group to get value of %ptr
5349 declare void @foo(i8*)
5350 declare i8* @getPointer(i8*)
5351 declare i8* @llvm.launder.invariant.group(i8*)
5355 The invariant.group metadata must be dropped when replacing one pointer by
5356 another based on aliasing information. This is because invariant.group is tied
5357 to the SSA value of the pointer operand.
5359 .. code-block:: llvm
5361 %v = load i8, i8* %x, !invariant.group !0
5362 ; if %x mustalias %y then we can replace the above instruction with
5363 %v = load i8, i8* %y
5365 Note that this is an experimental feature, which means that its semantics might
5366 change in the future.
5371 See :doc:`TypeMetadata`.
5373 '``associated``' Metadata
5374 ^^^^^^^^^^^^^^^^^^^^^^^^^
5376 The ``associated`` metadata may be attached to a global object
5377 declaration with a single argument that references another global object.
5379 This metadata prevents discarding of the global object in linker GC
5380 unless the referenced object is also discarded. The linker support for
5381 this feature is spotty. For best compatibility, globals carrying this
5384 - Be in a comdat with the referenced global.
5385 - Be in @llvm.compiler.used.
5386 - Have an explicit section with a name which is a valid C identifier.
5388 It does not have any effect on non-ELF targets.
5392 .. code-block:: text
5395 @a = global i32 1, comdat $a
5396 @b = internal global i32 2, comdat $a, section "abc", !associated !0
5403 The ``prof`` metadata is used to record profile data in the IR.
5404 The first operand of the metadata node indicates the profile metadata
5405 type. There are currently 3 types:
5406 :ref:`branch_weights<prof_node_branch_weights>`,
5407 :ref:`function_entry_count<prof_node_function_entry_count>`, and
5408 :ref:`VP<prof_node_VP>`.
5410 .. _prof_node_branch_weights:
5415 Branch weight metadata attached to a branch, select, switch or call instruction
5416 represents the likeliness of the associated branch being taken.
5417 For more information, see :doc:`BranchWeightMetadata`.
5419 .. _prof_node_function_entry_count:
5421 function_entry_count
5422 """"""""""""""""""""
5424 Function entry count metadata can be attached to function definitions
5425 to record the number of times the function is called. Used with BFI
5426 information, it is also used to derive the basic block profile count.
5427 For more information, see :doc:`BranchWeightMetadata`.
5434 VP (value profile) metadata can be attached to instructions that have
5435 value profile information. Currently this is indirect calls (where it
5436 records the hottest callees) and calls to memory intrinsics such as memcpy,
5437 memmove, and memset (where it records the hottest byte lengths).
5439 Each VP metadata node contains "VP" string, then a uint32_t value for the value
5440 profiling kind, a uint64_t value for the total number of times the instruction
5441 is executed, followed by uint64_t value and execution count pairs.
5442 The value profiling kind is 0 for indirect call targets and 1 for memory
5443 operations. For indirect call targets, each profile value is a hash
5444 of the callee function name, and for memory operations each value is the
5447 Note that the value counts do not need to add up to the total count
5448 listed in the third operand (in practice only the top hottest values
5449 are tracked and reported).
5451 Indirect call example:
5453 .. code-block:: llvm
5455 call void %f(), !prof !1
5456 !1 = !{!"VP", i32 0, i64 1600, i64 7651369219802541373, i64 1030, i64 -4377547752858689819, i64 410}
5458 Note that the VP type is 0 (the second operand), which indicates this is
5459 an indirect call value profile data. The third operand indicates that the
5460 indirect call executed 1600 times. The 4th and 6th operands give the
5461 hashes of the 2 hottest target functions' names (this is the same hash used
5462 to represent function names in the profile database), and the 5th and 7th
5463 operands give the execution count that each of the respective prior target
5464 functions was called.
5466 Module Flags Metadata
5467 =====================
5469 Information about the module as a whole is difficult to convey to LLVM's
5470 subsystems. The LLVM IR isn't sufficient to transmit this information.
5471 The ``llvm.module.flags`` named metadata exists in order to facilitate
5472 this. These flags are in the form of key / value pairs --- much like a
5473 dictionary --- making it easy for any subsystem who cares about a flag to
5476 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
5477 Each triplet has the following form:
5479 - The first element is a *behavior* flag, which specifies the behavior
5480 when two (or more) modules are merged together, and it encounters two
5481 (or more) metadata with the same ID. The supported behaviors are
5483 - The second element is a metadata string that is a unique ID for the
5484 metadata. Each module may only have one flag entry for each unique ID (not
5485 including entries with the **Require** behavior).
5486 - The third element is the value of the flag.
5488 When two (or more) modules are merged together, the resulting
5489 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
5490 each unique metadata ID string, there will be exactly one entry in the merged
5491 modules ``llvm.module.flags`` metadata table, and the value for that entry will
5492 be determined by the merge behavior flag, as described below. The only exception
5493 is that entries with the *Require* behavior are always preserved.
5495 The following behaviors are supported:
5506 Emits an error if two values disagree, otherwise the resulting value
5507 is that of the operands.
5511 Emits a warning if two values disagree. The result value will be the
5512 operand for the flag from the first module being linked.
5516 Adds a requirement that another module flag be present and have a
5517 specified value after linking is performed. The value must be a
5518 metadata pair, where the first element of the pair is the ID of the
5519 module flag to be restricted, and the second element of the pair is
5520 the value the module flag should be restricted to. This behavior can
5521 be used to restrict the allowable results (via triggering of an
5522 error) of linking IDs with the **Override** behavior.
5526 Uses the specified value, regardless of the behavior or value of the
5527 other module. If both modules specify **Override**, but the values
5528 differ, an error will be emitted.
5532 Appends the two values, which are required to be metadata nodes.
5536 Appends the two values, which are required to be metadata
5537 nodes. However, duplicate entries in the second list are dropped
5538 during the append operation.
5542 Takes the max of the two values, which are required to be integers.
5544 It is an error for a particular unique flag ID to have multiple behaviors,
5545 except in the case of **Require** (which adds restrictions on another metadata
5546 value) or **Override**.
5548 An example of module flags:
5550 .. code-block:: llvm
5552 !0 = !{ i32 1, !"foo", i32 1 }
5553 !1 = !{ i32 4, !"bar", i32 37 }
5554 !2 = !{ i32 2, !"qux", i32 42 }
5555 !3 = !{ i32 3, !"qux",
5560 !llvm.module.flags = !{ !0, !1, !2, !3 }
5562 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
5563 if two or more ``!"foo"`` flags are seen is to emit an error if their
5564 values are not equal.
5566 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
5567 behavior if two or more ``!"bar"`` flags are seen is to use the value
5570 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
5571 behavior if two or more ``!"qux"`` flags are seen is to emit a
5572 warning if their values are not equal.
5574 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
5580 The behavior is to emit an error if the ``llvm.module.flags`` does not
5581 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
5584 Objective-C Garbage Collection Module Flags Metadata
5585 ----------------------------------------------------
5587 On the Mach-O platform, Objective-C stores metadata about garbage
5588 collection in a special section called "image info". The metadata
5589 consists of a version number and a bitmask specifying what types of
5590 garbage collection are supported (if any) by the file. If two or more
5591 modules are linked together their garbage collection metadata needs to
5592 be merged rather than appended together.
5594 The Objective-C garbage collection module flags metadata consists of the
5595 following key-value pairs:
5604 * - ``Objective-C Version``
5605 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
5607 * - ``Objective-C Image Info Version``
5608 - **[Required]** --- The version of the image info section. Currently
5611 * - ``Objective-C Image Info Section``
5612 - **[Required]** --- The section to place the metadata. Valid values are
5613 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
5614 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
5615 Objective-C ABI version 2.
5617 * - ``Objective-C Garbage Collection``
5618 - **[Required]** --- Specifies whether garbage collection is supported or
5619 not. Valid values are 0, for no garbage collection, and 2, for garbage
5620 collection supported.
5622 * - ``Objective-C GC Only``
5623 - **[Optional]** --- Specifies that only garbage collection is supported.
5624 If present, its value must be 6. This flag requires that the
5625 ``Objective-C Garbage Collection`` flag have the value 2.
5627 Some important flag interactions:
5629 - If a module with ``Objective-C Garbage Collection`` set to 0 is
5630 merged with a module with ``Objective-C Garbage Collection`` set to
5631 2, then the resulting module has the
5632 ``Objective-C Garbage Collection`` flag set to 0.
5633 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
5634 merged with a module with ``Objective-C GC Only`` set to 6.
5636 C type width Module Flags Metadata
5637 ----------------------------------
5639 The ARM backend emits a section into each generated object file describing the
5640 options that it was compiled with (in a compiler-independent way) to prevent
5641 linking incompatible objects, and to allow automatic library selection. Some
5642 of these options are not visible at the IR level, namely wchar_t width and enum
5645 To pass this information to the backend, these options are encoded in module
5646 flags metadata, using the following key-value pairs:
5656 - * 0 --- sizeof(wchar_t) == 4
5657 * 1 --- sizeof(wchar_t) == 2
5660 - * 0 --- Enums are at least as large as an ``int``.
5661 * 1 --- Enums are stored in the smallest integer type which can
5662 represent all of its values.
5664 For example, the following metadata section specifies that the module was
5665 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
5666 enum is the smallest type which can represent all of its values::
5668 !llvm.module.flags = !{!0, !1}
5669 !0 = !{i32 1, !"short_wchar", i32 1}
5670 !1 = !{i32 1, !"short_enum", i32 0}
5672 Automatic Linker Flags Named Metadata
5673 =====================================
5675 Some targets support embedding flags to the linker inside individual object
5676 files. Typically this is used in conjunction with language extensions which
5677 allow source files to explicitly declare the libraries they depend on, and have
5678 these automatically be transmitted to the linker via object files.
5680 These flags are encoded in the IR using named metadata with the name
5681 ``!llvm.linker.options``. Each operand is expected to be a metadata node
5682 which should be a list of other metadata nodes, each of which should be a
5683 list of metadata strings defining linker options.
5685 For example, the following metadata section specifies two separate sets of
5686 linker options, presumably to link against ``libz`` and the ``Cocoa``
5690 !1 = !{ !"-framework", !"Cocoa" } } }
5691 !llvm.linker.options = !{ !0, !1 }
5693 The metadata encoding as lists of lists of options, as opposed to a collapsed
5694 list of options, is chosen so that the IR encoding can use multiple option
5695 strings to specify e.g., a single library, while still having that specifier be
5696 preserved as an atomic element that can be recognized by a target specific
5697 assembly writer or object file emitter.
5699 Each individual option is required to be either a valid option for the target's
5700 linker, or an option that is reserved by the target specific assembly writer or
5701 object file emitter. No other aspect of these options is defined by the IR.
5703 .. _intrinsicglobalvariables:
5705 Intrinsic Global Variables
5706 ==========================
5708 LLVM has a number of "magic" global variables that contain data that
5709 affect code generation or other IR semantics. These are documented here.
5710 All globals of this sort should have a section specified as
5711 "``llvm.metadata``". This section and all globals that start with
5712 "``llvm.``" are reserved for use by LLVM.
5716 The '``llvm.used``' Global Variable
5717 -----------------------------------
5719 The ``@llvm.used`` global is an array which has
5720 :ref:`appending linkage <linkage_appending>`. This array contains a list of
5721 pointers to named global variables, functions and aliases which may optionally
5722 have a pointer cast formed of bitcast or getelementptr. For example, a legal
5725 .. code-block:: llvm
5730 @llvm.used = appending global [2 x i8*] [
5732 i8* bitcast (i32* @Y to i8*)
5733 ], section "llvm.metadata"
5735 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
5736 and linker are required to treat the symbol as if there is a reference to the
5737 symbol that it cannot see (which is why they have to be named). For example, if
5738 a variable has internal linkage and no references other than that from the
5739 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
5740 references from inline asms and other things the compiler cannot "see", and
5741 corresponds to "``attribute((used))``" in GNU C.
5743 On some targets, the code generator must emit a directive to the
5744 assembler or object file to prevent the assembler and linker from
5745 molesting the symbol.
5747 .. _gv_llvmcompilerused:
5749 The '``llvm.compiler.used``' Global Variable
5750 --------------------------------------------
5752 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
5753 directive, except that it only prevents the compiler from touching the
5754 symbol. On targets that support it, this allows an intelligent linker to
5755 optimize references to the symbol without being impeded as it would be
5758 This is a rare construct that should only be used in rare circumstances,
5759 and should not be exposed to source languages.
5761 .. _gv_llvmglobalctors:
5763 The '``llvm.global_ctors``' Global Variable
5764 -------------------------------------------
5766 .. code-block:: llvm
5768 %0 = type { i32, void ()*, i8* }
5769 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
5771 The ``@llvm.global_ctors`` array contains a list of constructor
5772 functions, priorities, and an optional associated global or function.
5773 The functions referenced by this array will be called in ascending order
5774 of priority (i.e. lowest first) when the module is loaded. The order of
5775 functions with the same priority is not defined.
5777 If the third field is present, non-null, and points to a global variable
5778 or function, the initializer function will only run if the associated
5779 data from the current module is not discarded.
5781 .. _llvmglobaldtors:
5783 The '``llvm.global_dtors``' Global Variable
5784 -------------------------------------------
5786 .. code-block:: llvm
5788 %0 = type { i32, void ()*, i8* }
5789 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
5791 The ``@llvm.global_dtors`` array contains a list of destructor
5792 functions, priorities, and an optional associated global or function.
5793 The functions referenced by this array will be called in descending
5794 order of priority (i.e. highest first) when the module is unloaded. The
5795 order of functions with the same priority is not defined.
5797 If the third field is present, non-null, and points to a global variable
5798 or function, the destructor function will only run if the associated
5799 data from the current module is not discarded.
5801 Instruction Reference
5802 =====================
5804 The LLVM instruction set consists of several different classifications
5805 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
5806 instructions <binaryops>`, :ref:`bitwise binary
5807 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
5808 :ref:`other instructions <otherops>`.
5812 Terminator Instructions
5813 -----------------------
5815 As mentioned :ref:`previously <functionstructure>`, every basic block in a
5816 program ends with a "Terminator" instruction, which indicates which
5817 block should be executed after the current block is finished. These
5818 terminator instructions typically yield a '``void``' value: they produce
5819 control flow, not values (the one exception being the
5820 ':ref:`invoke <i_invoke>`' instruction).
5822 The terminator instructions are: ':ref:`ret <i_ret>`',
5823 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
5824 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
5825 ':ref:`resume <i_resume>`', ':ref:`catchswitch <i_catchswitch>`',
5826 ':ref:`catchret <i_catchret>`',
5827 ':ref:`cleanupret <i_cleanupret>`',
5828 and ':ref:`unreachable <i_unreachable>`'.
5832 '``ret``' Instruction
5833 ^^^^^^^^^^^^^^^^^^^^^
5840 ret <type> <value> ; Return a value from a non-void function
5841 ret void ; Return from void function
5846 The '``ret``' instruction is used to return control flow (and optionally
5847 a value) from a function back to the caller.
5849 There are two forms of the '``ret``' instruction: one that returns a
5850 value and then causes control flow, and one that just causes control
5856 The '``ret``' instruction optionally accepts a single argument, the
5857 return value. The type of the return value must be a ':ref:`first
5858 class <t_firstclass>`' type.
5860 A function is not :ref:`well formed <wellformed>` if it it has a non-void
5861 return type and contains a '``ret``' instruction with no return value or
5862 a return value with a type that does not match its type, or if it has a
5863 void return type and contains a '``ret``' instruction with a return
5869 When the '``ret``' instruction is executed, control flow returns back to
5870 the calling function's context. If the caller is a
5871 ":ref:`call <i_call>`" instruction, execution continues at the
5872 instruction after the call. If the caller was an
5873 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
5874 beginning of the "normal" destination block. If the instruction returns
5875 a value, that value shall set the call or invoke instruction's return
5881 .. code-block:: llvm
5883 ret i32 5 ; Return an integer value of 5
5884 ret void ; Return from a void function
5885 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
5889 '``br``' Instruction
5890 ^^^^^^^^^^^^^^^^^^^^
5897 br i1 <cond>, label <iftrue>, label <iffalse>
5898 br label <dest> ; Unconditional branch
5903 The '``br``' instruction is used to cause control flow to transfer to a
5904 different basic block in the current function. There are two forms of
5905 this instruction, corresponding to a conditional branch and an
5906 unconditional branch.
5911 The conditional branch form of the '``br``' instruction takes a single
5912 '``i1``' value and two '``label``' values. The unconditional form of the
5913 '``br``' instruction takes a single '``label``' value as a target.
5918 Upon execution of a conditional '``br``' instruction, the '``i1``'
5919 argument is evaluated. If the value is ``true``, control flows to the
5920 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
5921 to the '``iffalse``' ``label`` argument.
5926 .. code-block:: llvm
5929 %cond = icmp eq i32 %a, %b
5930 br i1 %cond, label %IfEqual, label %IfUnequal
5938 '``switch``' Instruction
5939 ^^^^^^^^^^^^^^^^^^^^^^^^
5946 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
5951 The '``switch``' instruction is used to transfer control flow to one of
5952 several different places. It is a generalization of the '``br``'
5953 instruction, allowing a branch to occur to one of many possible
5959 The '``switch``' instruction uses three parameters: an integer
5960 comparison value '``value``', a default '``label``' destination, and an
5961 array of pairs of comparison value constants and '``label``'s. The table
5962 is not allowed to contain duplicate constant entries.
5967 The ``switch`` instruction specifies a table of values and destinations.
5968 When the '``switch``' instruction is executed, this table is searched
5969 for the given value. If the value is found, control flow is transferred
5970 to the corresponding destination; otherwise, control flow is transferred
5971 to the default destination.
5976 Depending on properties of the target machine and the particular
5977 ``switch`` instruction, this instruction may be code generated in
5978 different ways. For example, it could be generated as a series of
5979 chained conditional branches or with a lookup table.
5984 .. code-block:: llvm
5986 ; Emulate a conditional br instruction
5987 %Val = zext i1 %value to i32
5988 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
5990 ; Emulate an unconditional br instruction
5991 switch i32 0, label %dest [ ]
5993 ; Implement a jump table:
5994 switch i32 %val, label %otherwise [ i32 0, label %onzero
5996 i32 2, label %ontwo ]
6000 '``indirectbr``' Instruction
6001 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6008 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
6013 The '``indirectbr``' instruction implements an indirect branch to a
6014 label within the current function, whose address is specified by
6015 "``address``". Address must be derived from a
6016 :ref:`blockaddress <blockaddress>` constant.
6021 The '``address``' argument is the address of the label to jump to. The
6022 rest of the arguments indicate the full set of possible destinations
6023 that the address may point to. Blocks are allowed to occur multiple
6024 times in the destination list, though this isn't particularly useful.
6026 This destination list is required so that dataflow analysis has an
6027 accurate understanding of the CFG.
6032 Control transfers to the block specified in the address argument. All
6033 possible destination blocks must be listed in the label list, otherwise
6034 this instruction has undefined behavior. This implies that jumps to
6035 labels defined in other functions have undefined behavior as well.
6040 This is typically implemented with a jump through a register.
6045 .. code-block:: llvm
6047 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
6051 '``invoke``' Instruction
6052 ^^^^^^^^^^^^^^^^^^^^^^^^
6059 <result> = invoke [cconv] [ret attrs] <ty>|<fnty> <fnptrval>(<function args>) [fn attrs]
6060 [operand bundles] to label <normal label> unwind label <exception label>
6065 The '``invoke``' instruction causes control to transfer to a specified
6066 function, with the possibility of control flow transfer to either the
6067 '``normal``' label or the '``exception``' label. If the callee function
6068 returns with the "``ret``" instruction, control flow will return to the
6069 "normal" label. If the callee (or any indirect callees) returns via the
6070 ":ref:`resume <i_resume>`" instruction or other exception handling
6071 mechanism, control is interrupted and continued at the dynamically
6072 nearest "exception" label.
6074 The '``exception``' label is a `landing
6075 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
6076 '``exception``' label is required to have the
6077 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
6078 information about the behavior of the program after unwinding happens,
6079 as its first non-PHI instruction. The restrictions on the
6080 "``landingpad``" instruction's tightly couples it to the "``invoke``"
6081 instruction, so that the important information contained within the
6082 "``landingpad``" instruction can't be lost through normal code motion.
6087 This instruction requires several arguments:
6089 #. The optional "cconv" marker indicates which :ref:`calling
6090 convention <callingconv>` the call should use. If none is
6091 specified, the call defaults to using C calling conventions.
6092 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6093 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6095 #. '``ty``': the type of the call instruction itself which is also the
6096 type of the return value. Functions that return no value are marked
6098 #. '``fnty``': shall be the signature of the function being invoked. The
6099 argument types must match the types implied by this signature. This
6100 type can be omitted if the function is not varargs.
6101 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6102 be invoked. In most cases, this is a direct function invocation, but
6103 indirect ``invoke``'s are just as possible, calling an arbitrary pointer
6105 #. '``function args``': argument list whose types match the function
6106 signature argument types and parameter attributes. All arguments must
6107 be of :ref:`first class <t_firstclass>` type. If the function signature
6108 indicates the function accepts a variable number of arguments, the
6109 extra arguments can be specified.
6110 #. '``normal label``': the label reached when the called function
6111 executes a '``ret``' instruction.
6112 #. '``exception label``': the label reached when a callee returns via
6113 the :ref:`resume <i_resume>` instruction or other exception handling
6115 #. The optional :ref:`function attributes <fnattrs>` list.
6116 #. The optional :ref:`operand bundles <opbundles>` list.
6121 This instruction is designed to operate as a standard '``call``'
6122 instruction in most regards. The primary difference is that it
6123 establishes an association with a label, which is used by the runtime
6124 library to unwind the stack.
6126 This instruction is used in languages with destructors to ensure that
6127 proper cleanup is performed in the case of either a ``longjmp`` or a
6128 thrown exception. Additionally, this is important for implementation of
6129 '``catch``' clauses in high-level languages that support them.
6131 For the purposes of the SSA form, the definition of the value returned
6132 by the '``invoke``' instruction is deemed to occur on the edge from the
6133 current block to the "normal" label. If the callee unwinds then no
6134 return value is available.
6139 .. code-block:: llvm
6141 %retval = invoke i32 @Test(i32 15) to label %Continue
6142 unwind label %TestCleanup ; i32:retval set
6143 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
6144 unwind label %TestCleanup ; i32:retval set
6148 '``resume``' Instruction
6149 ^^^^^^^^^^^^^^^^^^^^^^^^
6156 resume <type> <value>
6161 The '``resume``' instruction is a terminator instruction that has no
6167 The '``resume``' instruction requires one argument, which must have the
6168 same type as the result of any '``landingpad``' instruction in the same
6174 The '``resume``' instruction resumes propagation of an existing
6175 (in-flight) exception whose unwinding was interrupted with a
6176 :ref:`landingpad <i_landingpad>` instruction.
6181 .. code-block:: llvm
6183 resume { i8*, i32 } %exn
6187 '``catchswitch``' Instruction
6188 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6195 <resultval> = catchswitch within <parent> [ label <handler1>, label <handler2>, ... ] unwind to caller
6196 <resultval> = catchswitch within <parent> [ label <handler1>, label <handler2>, ... ] unwind label <default>
6201 The '``catchswitch``' instruction is used by `LLVM's exception handling system
6202 <ExceptionHandling.html#overview>`_ to describe the set of possible catch handlers
6203 that may be executed by the :ref:`EH personality routine <personalityfn>`.
6208 The ``parent`` argument is the token of the funclet that contains the
6209 ``catchswitch`` instruction. If the ``catchswitch`` is not inside a funclet,
6210 this operand may be the token ``none``.
6212 The ``default`` argument is the label of another basic block beginning with
6213 either a ``cleanuppad`` or ``catchswitch`` instruction. This unwind destination
6214 must be a legal target with respect to the ``parent`` links, as described in
6215 the `exception handling documentation\ <ExceptionHandling.html#wineh-constraints>`_.
6217 The ``handlers`` are a nonempty list of successor blocks that each begin with a
6218 :ref:`catchpad <i_catchpad>` instruction.
6223 Executing this instruction transfers control to one of the successors in
6224 ``handlers``, if appropriate, or continues to unwind via the unwind label if
6227 The ``catchswitch`` is both a terminator and a "pad" instruction, meaning that
6228 it must be both the first non-phi instruction and last instruction in the basic
6229 block. Therefore, it must be the only non-phi instruction in the block.
6234 .. code-block:: text
6237 %cs1 = catchswitch within none [label %handler0, label %handler1] unwind to caller
6239 %cs2 = catchswitch within %parenthandler [label %handler0] unwind label %cleanup
6243 '``catchret``' Instruction
6244 ^^^^^^^^^^^^^^^^^^^^^^^^^^
6251 catchret from <token> to label <normal>
6256 The '``catchret``' instruction is a terminator instruction that has a
6263 The first argument to a '``catchret``' indicates which ``catchpad`` it
6264 exits. It must be a :ref:`catchpad <i_catchpad>`.
6265 The second argument to a '``catchret``' specifies where control will
6271 The '``catchret``' instruction ends an existing (in-flight) exception whose
6272 unwinding was interrupted with a :ref:`catchpad <i_catchpad>` instruction. The
6273 :ref:`personality function <personalityfn>` gets a chance to execute arbitrary
6274 code to, for example, destroy the active exception. Control then transfers to
6277 The ``token`` argument must be a token produced by a ``catchpad`` instruction.
6278 If the specified ``catchpad`` is not the most-recently-entered not-yet-exited
6279 funclet pad (as described in the `EH documentation\ <ExceptionHandling.html#wineh-constraints>`_),
6280 the ``catchret``'s behavior is undefined.
6285 .. code-block:: text
6287 catchret from %catch label %continue
6291 '``cleanupret``' Instruction
6292 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6299 cleanupret from <value> unwind label <continue>
6300 cleanupret from <value> unwind to caller
6305 The '``cleanupret``' instruction is a terminator instruction that has
6306 an optional successor.
6312 The '``cleanupret``' instruction requires one argument, which indicates
6313 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
6314 If the specified ``cleanuppad`` is not the most-recently-entered not-yet-exited
6315 funclet pad (as described in the `EH documentation\ <ExceptionHandling.html#wineh-constraints>`_),
6316 the ``cleanupret``'s behavior is undefined.
6318 The '``cleanupret``' instruction also has an optional successor, ``continue``,
6319 which must be the label of another basic block beginning with either a
6320 ``cleanuppad`` or ``catchswitch`` instruction. This unwind destination must
6321 be a legal target with respect to the ``parent`` links, as described in the
6322 `exception handling documentation\ <ExceptionHandling.html#wineh-constraints>`_.
6327 The '``cleanupret``' instruction indicates to the
6328 :ref:`personality function <personalityfn>` that one
6329 :ref:`cleanuppad <i_cleanuppad>` it transferred control to has ended.
6330 It transfers control to ``continue`` or unwinds out of the function.
6335 .. code-block:: text
6337 cleanupret from %cleanup unwind to caller
6338 cleanupret from %cleanup unwind label %continue
6342 '``unreachable``' Instruction
6343 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6355 The '``unreachable``' instruction has no defined semantics. This
6356 instruction is used to inform the optimizer that a particular portion of
6357 the code is not reachable. This can be used to indicate that the code
6358 after a no-return function cannot be reached, and other facts.
6363 The '``unreachable``' instruction has no defined semantics.
6370 Binary operators are used to do most of the computation in a program.
6371 They require two operands of the same type, execute an operation on
6372 them, and produce a single value. The operands might represent multiple
6373 data, as is the case with the :ref:`vector <t_vector>` data type. The
6374 result value has the same type as its operands.
6376 There are several different binary operators:
6380 '``add``' Instruction
6381 ^^^^^^^^^^^^^^^^^^^^^
6388 <result> = add <ty> <op1>, <op2> ; yields ty:result
6389 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
6390 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
6391 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
6396 The '``add``' instruction returns the sum of its two operands.
6401 The two arguments to the '``add``' instruction must be
6402 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6403 arguments must have identical types.
6408 The value produced is the integer sum of the two operands.
6410 If the sum has unsigned overflow, the result returned is the
6411 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
6414 Because LLVM integers use a two's complement representation, this
6415 instruction is appropriate for both signed and unsigned integers.
6417 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
6418 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
6419 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
6420 unsigned and/or signed overflow, respectively, occurs.
6425 .. code-block:: text
6427 <result> = add i32 4, %var ; yields i32:result = 4 + %var
6431 '``fadd``' Instruction
6432 ^^^^^^^^^^^^^^^^^^^^^^
6439 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6444 The '``fadd``' instruction returns the sum of its two operands.
6449 The two arguments to the '``fadd``' instruction must be
6450 :ref:`floating-point <t_floating>` or :ref:`vector <t_vector>` of
6451 floating-point values. Both arguments must have identical types.
6456 The value produced is the floating-point sum of the two operands.
6457 This instruction is assumed to execute in the default :ref:`floating-point
6458 environment <floatenv>`.
6459 This instruction can also take any number of :ref:`fast-math
6460 flags <fastmath>`, which are optimization hints to enable otherwise
6461 unsafe floating-point optimizations:
6466 .. code-block:: text
6468 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
6470 '``sub``' Instruction
6471 ^^^^^^^^^^^^^^^^^^^^^
6478 <result> = sub <ty> <op1>, <op2> ; yields ty:result
6479 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
6480 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
6481 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
6486 The '``sub``' instruction returns the difference of its two operands.
6488 Note that the '``sub``' instruction is used to represent the '``neg``'
6489 instruction present in most other intermediate representations.
6494 The two arguments to the '``sub``' instruction must be
6495 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6496 arguments must have identical types.
6501 The value produced is the integer difference of the two operands.
6503 If the difference has unsigned overflow, the result returned is the
6504 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
6507 Because LLVM integers use a two's complement representation, this
6508 instruction is appropriate for both signed and unsigned integers.
6510 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
6511 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
6512 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
6513 unsigned and/or signed overflow, respectively, occurs.
6518 .. code-block:: text
6520 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
6521 <result> = sub i32 0, %val ; yields i32:result = -%var
6525 '``fsub``' Instruction
6526 ^^^^^^^^^^^^^^^^^^^^^^
6533 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6538 The '``fsub``' instruction returns the difference of its two operands.
6540 Note that the '``fsub``' instruction is used to represent the '``fneg``'
6541 instruction present in most other intermediate representations.
6546 The two arguments to the '``fsub``' instruction must be
6547 :ref:`floating-point <t_floating>` or :ref:`vector <t_vector>` of
6548 floating-point values. Both arguments must have identical types.
6553 The value produced is the floating-point difference of the two operands.
6554 This instruction is assumed to execute in the default :ref:`floating-point
6555 environment <floatenv>`.
6556 This instruction can also take any number of :ref:`fast-math
6557 flags <fastmath>`, which are optimization hints to enable otherwise
6558 unsafe floating-point optimizations:
6563 .. code-block:: text
6565 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
6566 <result> = fsub float -0.0, %val ; yields float:result = -%var
6568 '``mul``' Instruction
6569 ^^^^^^^^^^^^^^^^^^^^^
6576 <result> = mul <ty> <op1>, <op2> ; yields ty:result
6577 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
6578 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
6579 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
6584 The '``mul``' instruction returns the product of its two operands.
6589 The two arguments to the '``mul``' instruction must be
6590 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6591 arguments must have identical types.
6596 The value produced is the integer product of the two operands.
6598 If the result of the multiplication has unsigned overflow, the result
6599 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
6600 bit width of the result.
6602 Because LLVM integers use a two's complement representation, and the
6603 result is the same width as the operands, this instruction returns the
6604 correct result for both signed and unsigned integers. If a full product
6605 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
6606 sign-extended or zero-extended as appropriate to the width of the full
6609 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
6610 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
6611 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
6612 unsigned and/or signed overflow, respectively, occurs.
6617 .. code-block:: text
6619 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
6623 '``fmul``' Instruction
6624 ^^^^^^^^^^^^^^^^^^^^^^
6631 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6636 The '``fmul``' instruction returns the product of its two operands.
6641 The two arguments to the '``fmul``' instruction must be
6642 :ref:`floating-point <t_floating>` or :ref:`vector <t_vector>` of
6643 floating-point values. Both arguments must have identical types.
6648 The value produced is the floating-point product of the two operands.
6649 This instruction is assumed to execute in the default :ref:`floating-point
6650 environment <floatenv>`.
6651 This instruction can also take any number of :ref:`fast-math
6652 flags <fastmath>`, which are optimization hints to enable otherwise
6653 unsafe floating-point optimizations:
6658 .. code-block:: text
6660 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
6662 '``udiv``' Instruction
6663 ^^^^^^^^^^^^^^^^^^^^^^
6670 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
6671 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
6676 The '``udiv``' instruction returns the quotient of its two operands.
6681 The two arguments to the '``udiv``' instruction must be
6682 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6683 arguments must have identical types.
6688 The value produced is the unsigned integer quotient of the two operands.
6690 Note that unsigned integer division and signed integer division are
6691 distinct operations; for signed integer division, use '``sdiv``'.
6693 Division by zero is undefined behavior. For vectors, if any element
6694 of the divisor is zero, the operation has undefined behavior.
6697 If the ``exact`` keyword is present, the result value of the ``udiv`` is
6698 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
6699 such, "((a udiv exact b) mul b) == a").
6704 .. code-block:: text
6706 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
6708 '``sdiv``' Instruction
6709 ^^^^^^^^^^^^^^^^^^^^^^
6716 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
6717 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
6722 The '``sdiv``' instruction returns the quotient of its two operands.
6727 The two arguments to the '``sdiv``' instruction must be
6728 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6729 arguments must have identical types.
6734 The value produced is the signed integer quotient of the two operands
6735 rounded towards zero.
6737 Note that signed integer division and unsigned integer division are
6738 distinct operations; for unsigned integer division, use '``udiv``'.
6740 Division by zero is undefined behavior. For vectors, if any element
6741 of the divisor is zero, the operation has undefined behavior.
6742 Overflow also leads to undefined behavior; this is a rare case, but can
6743 occur, for example, by doing a 32-bit division of -2147483648 by -1.
6745 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
6746 a :ref:`poison value <poisonvalues>` if the result would be rounded.
6751 .. code-block:: text
6753 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
6757 '``fdiv``' Instruction
6758 ^^^^^^^^^^^^^^^^^^^^^^
6765 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6770 The '``fdiv``' instruction returns the quotient of its two operands.
6775 The two arguments to the '``fdiv``' instruction must be
6776 :ref:`floating-point <t_floating>` or :ref:`vector <t_vector>` of
6777 floating-point values. Both arguments must have identical types.
6782 The value produced is the floating-point quotient of the two operands.
6783 This instruction is assumed to execute in the default :ref:`floating-point
6784 environment <floatenv>`.
6785 This instruction can also take any number of :ref:`fast-math
6786 flags <fastmath>`, which are optimization hints to enable otherwise
6787 unsafe floating-point optimizations:
6792 .. code-block:: text
6794 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
6796 '``urem``' Instruction
6797 ^^^^^^^^^^^^^^^^^^^^^^
6804 <result> = urem <ty> <op1>, <op2> ; yields ty:result
6809 The '``urem``' instruction returns the remainder from the unsigned
6810 division of its two arguments.
6815 The two arguments to the '``urem``' instruction must be
6816 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6817 arguments must have identical types.
6822 This instruction returns the unsigned integer *remainder* of a division.
6823 This instruction always performs an unsigned division to get the
6826 Note that unsigned integer remainder and signed integer remainder are
6827 distinct operations; for signed integer remainder, use '``srem``'.
6829 Taking the remainder of a division by zero is undefined behavior.
6830 For vectors, if any element of the divisor is zero, the operation has
6836 .. code-block:: text
6838 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
6840 '``srem``' Instruction
6841 ^^^^^^^^^^^^^^^^^^^^^^
6848 <result> = srem <ty> <op1>, <op2> ; yields ty:result
6853 The '``srem``' instruction returns the remainder from the signed
6854 division of its two operands. This instruction can also take
6855 :ref:`vector <t_vector>` versions of the values in which case the elements
6861 The two arguments to the '``srem``' instruction must be
6862 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6863 arguments must have identical types.
6868 This instruction returns the *remainder* of a division (where the result
6869 is either zero or has the same sign as the dividend, ``op1``), not the
6870 *modulo* operator (where the result is either zero or has the same sign
6871 as the divisor, ``op2``) of a value. For more information about the
6872 difference, see `The Math
6873 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
6874 table of how this is implemented in various languages, please see
6876 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
6878 Note that signed integer remainder and unsigned integer remainder are
6879 distinct operations; for unsigned integer remainder, use '``urem``'.
6881 Taking the remainder of a division by zero is undefined behavior.
6882 For vectors, if any element of the divisor is zero, the operation has
6884 Overflow also leads to undefined behavior; this is a rare case, but can
6885 occur, for example, by taking the remainder of a 32-bit division of
6886 -2147483648 by -1. (The remainder doesn't actually overflow, but this
6887 rule lets srem be implemented using instructions that return both the
6888 result of the division and the remainder.)
6893 .. code-block:: text
6895 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
6899 '``frem``' Instruction
6900 ^^^^^^^^^^^^^^^^^^^^^^
6907 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6912 The '``frem``' instruction returns the remainder from the division of
6918 The two arguments to the '``frem``' instruction must be
6919 :ref:`floating-point <t_floating>` or :ref:`vector <t_vector>` of
6920 floating-point values. Both arguments must have identical types.
6925 The value produced is the floating-point remainder of the two operands.
6926 This is the same output as a libm '``fmod``' function, but without any
6927 possibility of setting ``errno``. The remainder has the same sign as the
6929 This instruction is assumed to execute in the default :ref:`floating-point
6930 environment <floatenv>`.
6931 This instruction can also take any number of :ref:`fast-math
6932 flags <fastmath>`, which are optimization hints to enable otherwise
6933 unsafe floating-point optimizations:
6938 .. code-block:: text
6940 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
6944 Bitwise Binary Operations
6945 -------------------------
6947 Bitwise binary operators are used to do various forms of bit-twiddling
6948 in a program. They are generally very efficient instructions and can
6949 commonly be strength reduced from other instructions. They require two
6950 operands of the same type, execute an operation on them, and produce a
6951 single value. The resulting value is the same type as its operands.
6953 '``shl``' Instruction
6954 ^^^^^^^^^^^^^^^^^^^^^
6961 <result> = shl <ty> <op1>, <op2> ; yields ty:result
6962 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
6963 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
6964 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
6969 The '``shl``' instruction returns the first operand shifted to the left
6970 a specified number of bits.
6975 Both arguments to the '``shl``' instruction must be the same
6976 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6977 '``op2``' is treated as an unsigned value.
6982 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
6983 where ``n`` is the width of the result. If ``op2`` is (statically or
6984 dynamically) equal to or larger than the number of bits in
6985 ``op1``, this instruction returns a :ref:`poison value <poisonvalues>`.
6986 If the arguments are vectors, each vector element of ``op1`` is shifted
6987 by the corresponding shift amount in ``op2``.
6989 If the ``nuw`` keyword is present, then the shift produces a poison
6990 value if it shifts out any non-zero bits.
6991 If the ``nsw`` keyword is present, then the shift produces a poison
6992 value it shifts out any bits that disagree with the resultant sign bit.
6997 .. code-block:: text
6999 <result> = shl i32 4, %var ; yields i32: 4 << %var
7000 <result> = shl i32 4, 2 ; yields i32: 16
7001 <result> = shl i32 1, 10 ; yields i32: 1024
7002 <result> = shl i32 1, 32 ; undefined
7003 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
7005 '``lshr``' Instruction
7006 ^^^^^^^^^^^^^^^^^^^^^^
7013 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
7014 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
7019 The '``lshr``' instruction (logical shift right) returns the first
7020 operand shifted to the right a specified number of bits with zero fill.
7025 Both arguments to the '``lshr``' instruction must be the same
7026 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
7027 '``op2``' is treated as an unsigned value.
7032 This instruction always performs a logical shift right operation. The
7033 most significant bits of the result will be filled with zero bits after
7034 the shift. If ``op2`` is (statically or dynamically) equal to or larger
7035 than the number of bits in ``op1``, this instruction returns a :ref:`poison
7036 value <poisonvalues>`. If the arguments are vectors, each vector element
7037 of ``op1`` is shifted by the corresponding shift amount in ``op2``.
7039 If the ``exact`` keyword is present, the result value of the ``lshr`` is
7040 a poison value if any of the bits shifted out are non-zero.
7045 .. code-block:: text
7047 <result> = lshr i32 4, 1 ; yields i32:result = 2
7048 <result> = lshr i32 4, 2 ; yields i32:result = 1
7049 <result> = lshr i8 4, 3 ; yields i8:result = 0
7050 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
7051 <result> = lshr i32 1, 32 ; undefined
7052 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
7054 '``ashr``' Instruction
7055 ^^^^^^^^^^^^^^^^^^^^^^
7062 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
7063 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
7068 The '``ashr``' instruction (arithmetic shift right) returns the first
7069 operand shifted to the right a specified number of bits with sign
7075 Both arguments to the '``ashr``' instruction must be the same
7076 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
7077 '``op2``' is treated as an unsigned value.
7082 This instruction always performs an arithmetic shift right operation,
7083 The most significant bits of the result will be filled with the sign bit
7084 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
7085 than the number of bits in ``op1``, this instruction returns a :ref:`poison
7086 value <poisonvalues>`. If the arguments are vectors, each vector element
7087 of ``op1`` is shifted by the corresponding shift amount in ``op2``.
7089 If the ``exact`` keyword is present, the result value of the ``ashr`` is
7090 a poison value if any of the bits shifted out are non-zero.
7095 .. code-block:: text
7097 <result> = ashr i32 4, 1 ; yields i32:result = 2
7098 <result> = ashr i32 4, 2 ; yields i32:result = 1
7099 <result> = ashr i8 4, 3 ; yields i8:result = 0
7100 <result> = ashr i8 -2, 1 ; yields i8:result = -1
7101 <result> = ashr i32 1, 32 ; undefined
7102 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
7104 '``and``' Instruction
7105 ^^^^^^^^^^^^^^^^^^^^^
7112 <result> = and <ty> <op1>, <op2> ; yields ty:result
7117 The '``and``' instruction returns the bitwise logical and of its two
7123 The two arguments to the '``and``' instruction must be
7124 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
7125 arguments must have identical types.
7130 The truth table used for the '``and``' instruction is:
7147 .. code-block:: text
7149 <result> = and i32 4, %var ; yields i32:result = 4 & %var
7150 <result> = and i32 15, 40 ; yields i32:result = 8
7151 <result> = and i32 4, 8 ; yields i32:result = 0
7153 '``or``' Instruction
7154 ^^^^^^^^^^^^^^^^^^^^
7161 <result> = or <ty> <op1>, <op2> ; yields ty:result
7166 The '``or``' instruction returns the bitwise logical inclusive or of its
7172 The two arguments to the '``or``' instruction must be
7173 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
7174 arguments must have identical types.
7179 The truth table used for the '``or``' instruction is:
7198 <result> = or i32 4, %var ; yields i32:result = 4 | %var
7199 <result> = or i32 15, 40 ; yields i32:result = 47
7200 <result> = or i32 4, 8 ; yields i32:result = 12
7202 '``xor``' Instruction
7203 ^^^^^^^^^^^^^^^^^^^^^
7210 <result> = xor <ty> <op1>, <op2> ; yields ty:result
7215 The '``xor``' instruction returns the bitwise logical exclusive or of
7216 its two operands. The ``xor`` is used to implement the "one's
7217 complement" operation, which is the "~" operator in C.
7222 The two arguments to the '``xor``' instruction must be
7223 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
7224 arguments must have identical types.
7229 The truth table used for the '``xor``' instruction is:
7246 .. code-block:: text
7248 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
7249 <result> = xor i32 15, 40 ; yields i32:result = 39
7250 <result> = xor i32 4, 8 ; yields i32:result = 12
7251 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
7256 LLVM supports several instructions to represent vector operations in a
7257 target-independent manner. These instructions cover the element-access
7258 and vector-specific operations needed to process vectors effectively.
7259 While LLVM does directly support these vector operations, many
7260 sophisticated algorithms will want to use target-specific intrinsics to
7261 take full advantage of a specific target.
7263 .. _i_extractelement:
7265 '``extractelement``' Instruction
7266 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7273 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
7278 The '``extractelement``' instruction extracts a single scalar element
7279 from a vector at a specified index.
7284 The first operand of an '``extractelement``' instruction is a value of
7285 :ref:`vector <t_vector>` type. The second operand is an index indicating
7286 the position from which to extract the element. The index may be a
7287 variable of any integer type.
7292 The result is a scalar of the same type as the element type of ``val``.
7293 Its value is the value at position ``idx`` of ``val``. If ``idx``
7294 exceeds the length of ``val``, the results are undefined.
7299 .. code-block:: text
7301 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
7303 .. _i_insertelement:
7305 '``insertelement``' Instruction
7306 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7313 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
7318 The '``insertelement``' instruction inserts a scalar element into a
7319 vector at a specified index.
7324 The first operand of an '``insertelement``' instruction is a value of
7325 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
7326 type must equal the element type of the first operand. The third operand
7327 is an index indicating the position at which to insert the value. The
7328 index may be a variable of any integer type.
7333 The result is a vector of the same type as ``val``. Its element values
7334 are those of ``val`` except at position ``idx``, where it gets the value
7335 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
7341 .. code-block:: text
7343 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
7345 .. _i_shufflevector:
7347 '``shufflevector``' Instruction
7348 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7355 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
7360 The '``shufflevector``' instruction constructs a permutation of elements
7361 from two input vectors, returning a vector with the same element type as
7362 the input and length that is the same as the shuffle mask.
7367 The first two operands of a '``shufflevector``' instruction are vectors
7368 with the same type. The third argument is a shuffle mask whose element
7369 type is always 'i32'. The result of the instruction is a vector whose
7370 length is the same as the shuffle mask and whose element type is the
7371 same as the element type of the first two operands.
7373 The shuffle mask operand is required to be a constant vector with either
7374 constant integer or undef values.
7379 The elements of the two input vectors are numbered from left to right
7380 across both of the vectors. The shuffle mask operand specifies, for each
7381 element of the result vector, which element of the two input vectors the
7382 result element gets. If the shuffle mask is undef, the result vector is
7383 undef. If any element of the mask operand is undef, that element of the
7384 result is undef. If the shuffle mask selects an undef element from one
7385 of the input vectors, the resulting element is undef.
7390 .. code-block:: text
7392 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
7393 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
7394 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
7395 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
7396 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
7397 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
7398 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
7399 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
7401 Aggregate Operations
7402 --------------------
7404 LLVM supports several instructions for working with
7405 :ref:`aggregate <t_aggregate>` values.
7409 '``extractvalue``' Instruction
7410 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7417 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
7422 The '``extractvalue``' instruction extracts the value of a member field
7423 from an :ref:`aggregate <t_aggregate>` value.
7428 The first operand of an '``extractvalue``' instruction is a value of
7429 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The other operands are
7430 constant indices to specify which value to extract in a similar manner
7431 as indices in a '``getelementptr``' instruction.
7433 The major differences to ``getelementptr`` indexing are:
7435 - Since the value being indexed is not a pointer, the first index is
7436 omitted and assumed to be zero.
7437 - At least one index must be specified.
7438 - Not only struct indices but also array indices must be in bounds.
7443 The result is the value at the position in the aggregate specified by
7449 .. code-block:: text
7451 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
7455 '``insertvalue``' Instruction
7456 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7463 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
7468 The '``insertvalue``' instruction inserts a value into a member field in
7469 an :ref:`aggregate <t_aggregate>` value.
7474 The first operand of an '``insertvalue``' instruction is a value of
7475 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
7476 a first-class value to insert. The following operands are constant
7477 indices indicating the position at which to insert the value in a
7478 similar manner as indices in a '``extractvalue``' instruction. The value
7479 to insert must have the same type as the value identified by the
7485 The result is an aggregate of the same type as ``val``. Its value is
7486 that of ``val`` except that the value at the position specified by the
7487 indices is that of ``elt``.
7492 .. code-block:: llvm
7494 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
7495 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
7496 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
7500 Memory Access and Addressing Operations
7501 ---------------------------------------
7503 A key design point of an SSA-based representation is how it represents
7504 memory. In LLVM, no memory locations are in SSA form, which makes things
7505 very simple. This section describes how to read, write, and allocate
7510 '``alloca``' Instruction
7511 ^^^^^^^^^^^^^^^^^^^^^^^^
7518 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] [, addrspace(<num>)] ; yields type addrspace(num)*:result
7523 The '``alloca``' instruction allocates memory on the stack frame of the
7524 currently executing function, to be automatically released when this
7525 function returns to its caller. The object is always allocated in the
7526 address space for allocas indicated in the datalayout.
7531 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
7532 bytes of memory on the runtime stack, returning a pointer of the
7533 appropriate type to the program. If "NumElements" is specified, it is
7534 the number of elements allocated, otherwise "NumElements" is defaulted
7535 to be one. If a constant alignment is specified, the value result of the
7536 allocation is guaranteed to be aligned to at least that boundary. The
7537 alignment may not be greater than ``1 << 29``. If not specified, or if
7538 zero, the target can choose to align the allocation on any convenient
7539 boundary compatible with the type.
7541 '``type``' may be any sized type.
7546 Memory is allocated; a pointer is returned. The operation is undefined
7547 if there is insufficient stack space for the allocation. '``alloca``'d
7548 memory is automatically released when the function returns. The
7549 '``alloca``' instruction is commonly used to represent automatic
7550 variables that must have an address available. When the function returns
7551 (either with the ``ret`` or ``resume`` instructions), the memory is
7552 reclaimed. Allocating zero bytes is legal, but the result is undefined.
7553 The order in which memory is allocated (ie., which way the stack grows)
7559 .. code-block:: llvm
7561 %ptr = alloca i32 ; yields i32*:ptr
7562 %ptr = alloca i32, i32 4 ; yields i32*:ptr
7563 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
7564 %ptr = alloca i32, align 1024 ; yields i32*:ptr
7568 '``load``' Instruction
7569 ^^^^^^^^^^^^^^^^^^^^^^
7576 <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>]
7577 <result> = load atomic [volatile] <ty>, <ty>* <pointer> [syncscope("<target-scope>")] <ordering>, align <alignment> [, !invariant.group !<index>]
7578 !<index> = !{ i32 1 }
7579 !<deref_bytes_node> = !{i64 <dereferenceable_bytes>}
7580 !<align_node> = !{ i64 <value_alignment> }
7585 The '``load``' instruction is used to read from memory.
7590 The argument to the ``load`` instruction specifies the memory address from which
7591 to load. The type specified must be a :ref:`first class <t_firstclass>` type of
7592 known size (i.e. not containing an :ref:`opaque structural type <t_opaque>`). If
7593 the ``load`` is marked as ``volatile``, then the optimizer is not allowed to
7594 modify the number or order of execution of this ``load`` with other
7595 :ref:`volatile operations <volatile>`.
7597 If the ``load`` is marked as ``atomic``, it takes an extra :ref:`ordering
7598 <ordering>` and optional ``syncscope("<target-scope>")`` argument. The
7599 ``release`` and ``acq_rel`` orderings are not valid on ``load`` instructions.
7600 Atomic loads produce :ref:`defined <memmodel>` results when they may see
7601 multiple atomic stores. The type of the pointee must be an integer, pointer, or
7602 floating-point type whose bit width is a power of two greater than or equal to
7603 eight and less than or equal to a target-specific size limit. ``align`` must be
7604 explicitly specified on atomic loads, and the load has undefined behavior if the
7605 alignment is not set to a value which is at least the size in bytes of the
7606 pointee. ``!nontemporal`` does not have any defined semantics for atomic loads.
7608 The optional constant ``align`` argument specifies the alignment of the
7609 operation (that is, the alignment of the memory address). A value of 0
7610 or an omitted ``align`` argument means that the operation has the ABI
7611 alignment for the target. It is the responsibility of the code emitter
7612 to ensure that the alignment information is correct. Overestimating the
7613 alignment results in undefined behavior. Underestimating the alignment
7614 may produce less efficient code. An alignment of 1 is always safe. The
7615 maximum possible alignment is ``1 << 29``. An alignment value higher
7616 than the size of the loaded type implies memory up to the alignment
7617 value bytes can be safely loaded without trapping in the default
7618 address space. Access of the high bytes can interfere with debugging
7619 tools, so should not be accessed if the function has the
7620 ``sanitize_thread`` or ``sanitize_address`` attributes.
7622 The optional ``!nontemporal`` metadata must reference a single
7623 metadata name ``<index>`` corresponding to a metadata node with one
7624 ``i32`` entry of value 1. The existence of the ``!nontemporal``
7625 metadata on the instruction tells the optimizer and code generator
7626 that this load is not expected to be reused in the cache. The code
7627 generator may select special instructions to save cache bandwidth, such
7628 as the ``MOVNT`` instruction on x86.
7630 The optional ``!invariant.load`` metadata must reference a single
7631 metadata name ``<index>`` corresponding to a metadata node with no
7632 entries. If a load instruction tagged with the ``!invariant.load``
7633 metadata is executed, the optimizer may assume the memory location
7634 referenced by the load contains the same value at all points in the
7635 program where the memory location is known to be dereferenceable.
7637 The optional ``!invariant.group`` metadata must reference a single metadata name
7638 ``<index>`` corresponding to a metadata node with no entries.
7639 See ``invariant.group`` metadata.
7641 The optional ``!nonnull`` metadata must reference a single
7642 metadata name ``<index>`` corresponding to a metadata node with no
7643 entries. The existence of the ``!nonnull`` metadata on the
7644 instruction tells the optimizer that the value loaded is known to
7645 never be null. This is analogous to the ``nonnull`` attribute
7646 on parameters and return values. This metadata can only be applied
7647 to loads of a pointer type.
7649 The optional ``!dereferenceable`` metadata must reference a single metadata
7650 name ``<deref_bytes_node>`` corresponding to a metadata node with one ``i64``
7651 entry. The existence of the ``!dereferenceable`` metadata on the instruction
7652 tells the optimizer that the value loaded is known to be dereferenceable.
7653 The number of bytes known to be dereferenceable is specified by the integer
7654 value in the metadata node. This is analogous to the ''dereferenceable''
7655 attribute on parameters and return values. This metadata can only be applied
7656 to loads of a pointer type.
7658 The optional ``!dereferenceable_or_null`` metadata must reference a single
7659 metadata name ``<deref_bytes_node>`` corresponding to a metadata node with one
7660 ``i64`` entry. The existence of the ``!dereferenceable_or_null`` metadata on the
7661 instruction tells the optimizer that the value loaded is known to be either
7662 dereferenceable or null.
7663 The number of bytes known to be dereferenceable is specified by the integer
7664 value in the metadata node. This is analogous to the ''dereferenceable_or_null''
7665 attribute on parameters and return values. This metadata can only be applied
7666 to loads of a pointer type.
7668 The optional ``!align`` metadata must reference a single metadata name
7669 ``<align_node>`` corresponding to a metadata node with one ``i64`` entry.
7670 The existence of the ``!align`` metadata on the instruction tells the
7671 optimizer that the value loaded is known to be aligned to a boundary specified
7672 by the integer value in the metadata node. The alignment must be a power of 2.
7673 This is analogous to the ''align'' attribute on parameters and return values.
7674 This metadata can only be applied to loads of a pointer type.
7679 The location of memory pointed to is loaded. If the value being loaded
7680 is of scalar type then the number of bytes read does not exceed the
7681 minimum number of bytes needed to hold all bits of the type. For
7682 example, loading an ``i24`` reads at most three bytes. When loading a
7683 value of a type like ``i20`` with a size that is not an integral number
7684 of bytes, the result is undefined if the value was not originally
7685 written using a store of the same type.
7690 .. code-block:: llvm
7692 %ptr = alloca i32 ; yields i32*:ptr
7693 store i32 3, i32* %ptr ; yields void
7694 %val = load i32, i32* %ptr ; yields i32:val = i32 3
7698 '``store``' Instruction
7699 ^^^^^^^^^^^^^^^^^^^^^^^
7706 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.group !<index>] ; yields void
7707 store atomic [volatile] <ty> <value>, <ty>* <pointer> [syncscope("<target-scope>")] <ordering>, align <alignment> [, !invariant.group !<index>] ; yields void
7712 The '``store``' instruction is used to write to memory.
7717 There are two arguments to the ``store`` instruction: a value to store and an
7718 address at which to store it. The type of the ``<pointer>`` operand must be a
7719 pointer to the :ref:`first class <t_firstclass>` type of the ``<value>``
7720 operand. If the ``store`` is marked as ``volatile``, then the optimizer is not
7721 allowed to modify the number or order of execution of this ``store`` with other
7722 :ref:`volatile operations <volatile>`. Only values of :ref:`first class
7723 <t_firstclass>` types of known size (i.e. not containing an :ref:`opaque
7724 structural type <t_opaque>`) can be stored.
7726 If the ``store`` is marked as ``atomic``, it takes an extra :ref:`ordering
7727 <ordering>` and optional ``syncscope("<target-scope>")`` argument. The
7728 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store`` instructions.
7729 Atomic loads produce :ref:`defined <memmodel>` results when they may see
7730 multiple atomic stores. The type of the pointee must be an integer, pointer, or
7731 floating-point type whose bit width is a power of two greater than or equal to
7732 eight and less than or equal to a target-specific size limit. ``align`` must be
7733 explicitly specified on atomic stores, and the store has undefined behavior if
7734 the alignment is not set to a value which is at least the size in bytes of the
7735 pointee. ``!nontemporal`` does not have any defined semantics for atomic stores.
7737 The optional constant ``align`` argument specifies the alignment of the
7738 operation (that is, the alignment of the memory address). A value of 0
7739 or an omitted ``align`` argument means that the operation has the ABI
7740 alignment for the target. It is the responsibility of the code emitter
7741 to ensure that the alignment information is correct. Overestimating the
7742 alignment results in undefined behavior. Underestimating the
7743 alignment may produce less efficient code. An alignment of 1 is always
7744 safe. The maximum possible alignment is ``1 << 29``. An alignment
7745 value higher than the size of the stored type implies memory up to the
7746 alignment value bytes can be stored to without trapping in the default
7747 address space. Storing to the higher bytes however may result in data
7748 races if another thread can access the same address. Introducing a
7749 data race is not allowed. Storing to the extra bytes is not allowed
7750 even in situations where a data race is known to not exist if the
7751 function has the ``sanitize_address`` attribute.
7753 The optional ``!nontemporal`` metadata must reference a single metadata
7754 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
7755 value 1. The existence of the ``!nontemporal`` metadata on the instruction
7756 tells the optimizer and code generator that this load is not expected to
7757 be reused in the cache. The code generator may select special
7758 instructions to save cache bandwidth, such as the ``MOVNT`` instruction on
7761 The optional ``!invariant.group`` metadata must reference a
7762 single metadata name ``<index>``. See ``invariant.group`` metadata.
7767 The contents of memory are updated to contain ``<value>`` at the
7768 location specified by the ``<pointer>`` operand. If ``<value>`` is
7769 of scalar type then the number of bytes written does not exceed the
7770 minimum number of bytes needed to hold all bits of the type. For
7771 example, storing an ``i24`` writes at most three bytes. When writing a
7772 value of a type like ``i20`` with a size that is not an integral number
7773 of bytes, it is unspecified what happens to the extra bits that do not
7774 belong to the type, but they will typically be overwritten.
7779 .. code-block:: llvm
7781 %ptr = alloca i32 ; yields i32*:ptr
7782 store i32 3, i32* %ptr ; yields void
7783 %val = load i32, i32* %ptr ; yields i32:val = i32 3
7787 '``fence``' Instruction
7788 ^^^^^^^^^^^^^^^^^^^^^^^
7795 fence [syncscope("<target-scope>")] <ordering> ; yields void
7800 The '``fence``' instruction is used to introduce happens-before edges
7806 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
7807 defines what *synchronizes-with* edges they add. They can only be given
7808 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
7813 A fence A which has (at least) ``release`` ordering semantics
7814 *synchronizes with* a fence B with (at least) ``acquire`` ordering
7815 semantics if and only if there exist atomic operations X and Y, both
7816 operating on some atomic object M, such that A is sequenced before X, X
7817 modifies M (either directly or through some side effect of a sequence
7818 headed by X), Y is sequenced before B, and Y observes M. This provides a
7819 *happens-before* dependency between A and B. Rather than an explicit
7820 ``fence``, one (but not both) of the atomic operations X or Y might
7821 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
7822 still *synchronize-with* the explicit ``fence`` and establish the
7823 *happens-before* edge.
7825 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
7826 ``acquire`` and ``release`` semantics specified above, participates in
7827 the global program order of other ``seq_cst`` operations and/or fences.
7829 A ``fence`` instruction can also take an optional
7830 ":ref:`syncscope <syncscope>`" argument.
7835 .. code-block:: text
7837 fence acquire ; yields void
7838 fence syncscope("singlethread") seq_cst ; yields void
7839 fence syncscope("agent") seq_cst ; yields void
7843 '``cmpxchg``' Instruction
7844 ^^^^^^^^^^^^^^^^^^^^^^^^^
7851 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [syncscope("<target-scope>")] <success ordering> <failure ordering> ; yields { ty, i1 }
7856 The '``cmpxchg``' instruction is used to atomically modify memory. It
7857 loads a value in memory and compares it to a given value. If they are
7858 equal, it tries to store a new value into the memory.
7863 There are three arguments to the '``cmpxchg``' instruction: an address
7864 to operate on, a value to compare to the value currently be at that
7865 address, and a new value to place at that address if the compared values
7866 are equal. The type of '<cmp>' must be an integer or pointer type whose
7867 bit width is a power of two greater than or equal to eight and less
7868 than or equal to a target-specific size limit. '<cmp>' and '<new>' must
7869 have the same type, and the type of '<pointer>' must be a pointer to
7870 that type. If the ``cmpxchg`` is marked as ``volatile``, then the
7871 optimizer is not allowed to modify the number or order of execution of
7872 this ``cmpxchg`` with other :ref:`volatile operations <volatile>`.
7874 The success and failure :ref:`ordering <ordering>` arguments specify how this
7875 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
7876 must be at least ``monotonic``, the ordering constraint on failure must be no
7877 stronger than that on success, and the failure ordering cannot be either
7878 ``release`` or ``acq_rel``.
7880 A ``cmpxchg`` instruction can also take an optional
7881 ":ref:`syncscope <syncscope>`" argument.
7883 The pointer passed into cmpxchg must have alignment greater than or
7884 equal to the size in memory of the operand.
7889 The contents of memory at the location specified by the '``<pointer>``' operand
7890 is read and compared to '``<cmp>``'; if the values are equal, '``<new>``' is
7891 written to the location. The original value at the location is returned,
7892 together with a flag indicating success (true) or failure (false).
7894 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
7895 permitted: the operation may not write ``<new>`` even if the comparison
7898 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
7899 if the value loaded equals ``cmp``.
7901 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
7902 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
7903 load with an ordering parameter determined the second ordering parameter.
7908 .. code-block:: llvm
7911 %orig = load atomic i32, i32* %ptr unordered, align 4 ; yields i32
7915 %cmp = phi i32 [ %orig, %entry ], [%value_loaded, %loop]
7916 %squared = mul i32 %cmp, %cmp
7917 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
7918 %value_loaded = extractvalue { i32, i1 } %val_success, 0
7919 %success = extractvalue { i32, i1 } %val_success, 1
7920 br i1 %success, label %done, label %loop
7927 '``atomicrmw``' Instruction
7928 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7935 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [syncscope("<target-scope>")] <ordering> ; yields ty
7940 The '``atomicrmw``' instruction is used to atomically modify memory.
7945 There are three arguments to the '``atomicrmw``' instruction: an
7946 operation to apply, an address whose value to modify, an argument to the
7947 operation. The operation must be one of the following keywords:
7961 The type of '<value>' must be an integer type whose bit width is a power
7962 of two greater than or equal to eight and less than or equal to a
7963 target-specific size limit. The type of the '``<pointer>``' operand must
7964 be a pointer to that type. If the ``atomicrmw`` is marked as
7965 ``volatile``, then the optimizer is not allowed to modify the number or
7966 order of execution of this ``atomicrmw`` with other :ref:`volatile
7967 operations <volatile>`.
7969 A ``atomicrmw`` instruction can also take an optional
7970 ":ref:`syncscope <syncscope>`" argument.
7975 The contents of memory at the location specified by the '``<pointer>``'
7976 operand are atomically read, modified, and written back. The original
7977 value at the location is returned. The modification is specified by the
7980 - xchg: ``*ptr = val``
7981 - add: ``*ptr = *ptr + val``
7982 - sub: ``*ptr = *ptr - val``
7983 - and: ``*ptr = *ptr & val``
7984 - nand: ``*ptr = ~(*ptr & val)``
7985 - or: ``*ptr = *ptr | val``
7986 - xor: ``*ptr = *ptr ^ val``
7987 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
7988 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
7989 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
7991 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
7997 .. code-block:: llvm
7999 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
8001 .. _i_getelementptr:
8003 '``getelementptr``' Instruction
8004 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8011 <result> = getelementptr <ty>, <ty>* <ptrval>{, [inrange] <ty> <idx>}*
8012 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, [inrange] <ty> <idx>}*
8013 <result> = getelementptr <ty>, <ptr vector> <ptrval>, [inrange] <vector index type> <idx>
8018 The '``getelementptr``' instruction is used to get the address of a
8019 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
8020 address calculation only and does not access memory. The instruction can also
8021 be used to calculate a vector of such addresses.
8026 The first argument is always a type used as the basis for the calculations.
8027 The second argument is always a pointer or a vector of pointers, and is the
8028 base address to start from. The remaining arguments are indices
8029 that indicate which of the elements of the aggregate object are indexed.
8030 The interpretation of each index is dependent on the type being indexed
8031 into. The first index always indexes the pointer value given as the
8032 second argument, the second index indexes a value of the type pointed to
8033 (not necessarily the value directly pointed to, since the first index
8034 can be non-zero), etc. The first type indexed into must be a pointer
8035 value, subsequent types can be arrays, vectors, and structs. Note that
8036 subsequent types being indexed into can never be pointers, since that
8037 would require loading the pointer before continuing calculation.
8039 The type of each index argument depends on the type it is indexing into.
8040 When indexing into a (optionally packed) structure, only ``i32`` integer
8041 **constants** are allowed (when using a vector of indices they must all
8042 be the **same** ``i32`` integer constant). When indexing into an array,
8043 pointer or vector, integers of any width are allowed, and they are not
8044 required to be constant. These integers are treated as signed values
8047 For example, let's consider a C code fragment and how it gets compiled
8063 int *foo(struct ST *s) {
8064 return &s[1].Z.B[5][13];
8067 The LLVM code generated by Clang is:
8069 .. code-block:: llvm
8071 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
8072 %struct.ST = type { i32, double, %struct.RT }
8074 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
8076 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
8083 In the example above, the first index is indexing into the
8084 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
8085 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
8086 indexes into the third element of the structure, yielding a
8087 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
8088 structure. The third index indexes into the second element of the
8089 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
8090 dimensions of the array are subscripted into, yielding an '``i32``'
8091 type. The '``getelementptr``' instruction returns a pointer to this
8092 element, thus computing a value of '``i32*``' type.
8094 Note that it is perfectly legal to index partially through a structure,
8095 returning a pointer to an inner element. Because of this, the LLVM code
8096 for the given testcase is equivalent to:
8098 .. code-block:: llvm
8100 define i32* @foo(%struct.ST* %s) {
8101 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
8102 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
8103 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
8104 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
8105 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
8109 If the ``inbounds`` keyword is present, the result value of the
8110 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
8111 pointer is not an *in bounds* address of an allocated object, or if any
8112 of the addresses that would be formed by successive addition of the
8113 offsets implied by the indices to the base address with infinitely
8114 precise signed arithmetic are not an *in bounds* address of that
8115 allocated object. The *in bounds* addresses for an allocated object are
8116 all the addresses that point into the object, plus the address one byte
8117 past the end. The only *in bounds* address for a null pointer in the
8118 default address-space is the null pointer itself. In cases where the
8119 base is a vector of pointers the ``inbounds`` keyword applies to each
8120 of the computations element-wise.
8122 If the ``inbounds`` keyword is not present, the offsets are added to the
8123 base address with silently-wrapping two's complement arithmetic. If the
8124 offsets have a different width from the pointer, they are sign-extended
8125 or truncated to the width of the pointer. The result value of the
8126 ``getelementptr`` may be outside the object pointed to by the base
8127 pointer. The result value may not necessarily be used to access memory
8128 though, even if it happens to point into allocated storage. See the
8129 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
8132 If the ``inrange`` keyword is present before any index, loading from or
8133 storing to any pointer derived from the ``getelementptr`` has undefined
8134 behavior if the load or store would access memory outside of the bounds of
8135 the element selected by the index marked as ``inrange``. The result of a
8136 pointer comparison or ``ptrtoint`` (including ``ptrtoint``-like operations
8137 involving memory) involving a pointer derived from a ``getelementptr`` with
8138 the ``inrange`` keyword is undefined, with the exception of comparisons
8139 in the case where both operands are in the range of the element selected
8140 by the ``inrange`` keyword, inclusive of the address one past the end of
8141 that element. Note that the ``inrange`` keyword is currently only allowed
8142 in constant ``getelementptr`` expressions.
8144 The getelementptr instruction is often confusing. For some more insight
8145 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
8150 .. code-block:: llvm
8152 ; yields [12 x i8]*:aptr
8153 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
8155 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
8157 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
8159 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
8164 The ``getelementptr`` returns a vector of pointers, instead of a single address,
8165 when one or more of its arguments is a vector. In such cases, all vector
8166 arguments should have the same number of elements, and every scalar argument
8167 will be effectively broadcast into a vector during address calculation.
8169 .. code-block:: llvm
8171 ; All arguments are vectors:
8172 ; A[i] = ptrs[i] + offsets[i]*sizeof(i8)
8173 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets
8175 ; Add the same scalar offset to each pointer of a vector:
8176 ; A[i] = ptrs[i] + offset*sizeof(i8)
8177 %A = getelementptr i8, <4 x i8*> %ptrs, i64 %offset
8179 ; Add distinct offsets to the same pointer:
8180 ; A[i] = ptr + offsets[i]*sizeof(i8)
8181 %A = getelementptr i8, i8* %ptr, <4 x i64> %offsets
8183 ; In all cases described above the type of the result is <4 x i8*>
8185 The two following instructions are equivalent:
8187 .. code-block:: llvm
8189 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
8190 <4 x i32> <i32 2, i32 2, i32 2, i32 2>,
8191 <4 x i32> <i32 1, i32 1, i32 1, i32 1>,
8193 <4 x i64> <i64 13, i64 13, i64 13, i64 13>
8195 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
8196 i32 2, i32 1, <4 x i32> %ind4, i64 13
8198 Let's look at the C code, where the vector version of ``getelementptr``
8203 // Let's assume that we vectorize the following loop:
8204 double *A, *B; int *C;
8205 for (int i = 0; i < size; ++i) {
8209 .. code-block:: llvm
8211 ; get pointers for 8 elements from array B
8212 %ptrs = getelementptr double, double* %B, <8 x i32> %C
8213 ; load 8 elements from array B into A
8214 %A = call <8 x double> @llvm.masked.gather.v8f64.v8p0f64(<8 x double*> %ptrs,
8215 i32 8, <8 x i1> %mask, <8 x double> %passthru)
8217 Conversion Operations
8218 ---------------------
8220 The instructions in this category are the conversion instructions
8221 (casting) which all take a single operand and a type. They perform
8222 various bit conversions on the operand.
8226 '``trunc .. to``' Instruction
8227 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8234 <result> = trunc <ty> <value> to <ty2> ; yields ty2
8239 The '``trunc``' instruction truncates its operand to the type ``ty2``.
8244 The '``trunc``' instruction takes a value to trunc, and a type to trunc
8245 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
8246 of the same number of integers. The bit size of the ``value`` must be
8247 larger than the bit size of the destination type, ``ty2``. Equal sized
8248 types are not allowed.
8253 The '``trunc``' instruction truncates the high order bits in ``value``
8254 and converts the remaining bits to ``ty2``. Since the source size must
8255 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
8256 It will always truncate bits.
8261 .. code-block:: llvm
8263 %X = trunc i32 257 to i8 ; yields i8:1
8264 %Y = trunc i32 123 to i1 ; yields i1:true
8265 %Z = trunc i32 122 to i1 ; yields i1:false
8266 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
8270 '``zext .. to``' Instruction
8271 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8278 <result> = zext <ty> <value> to <ty2> ; yields ty2
8283 The '``zext``' instruction zero extends its operand to type ``ty2``.
8288 The '``zext``' instruction takes a value to cast, and a type to cast it
8289 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
8290 the same number of integers. The bit size of the ``value`` must be
8291 smaller than the bit size of the destination type, ``ty2``.
8296 The ``zext`` fills the high order bits of the ``value`` with zero bits
8297 until it reaches the size of the destination type, ``ty2``.
8299 When zero extending from i1, the result will always be either 0 or 1.
8304 .. code-block:: llvm
8306 %X = zext i32 257 to i64 ; yields i64:257
8307 %Y = zext i1 true to i32 ; yields i32:1
8308 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
8312 '``sext .. to``' Instruction
8313 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8320 <result> = sext <ty> <value> to <ty2> ; yields ty2
8325 The '``sext``' sign extends ``value`` to the type ``ty2``.
8330 The '``sext``' instruction takes a value to cast, and a type to cast it
8331 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
8332 the same number of integers. The bit size of the ``value`` must be
8333 smaller than the bit size of the destination type, ``ty2``.
8338 The '``sext``' instruction performs a sign extension by copying the sign
8339 bit (highest order bit) of the ``value`` until it reaches the bit size
8340 of the type ``ty2``.
8342 When sign extending from i1, the extension always results in -1 or 0.
8347 .. code-block:: llvm
8349 %X = sext i8 -1 to i16 ; yields i16 :65535
8350 %Y = sext i1 true to i32 ; yields i32:-1
8351 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
8353 '``fptrunc .. to``' Instruction
8354 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8361 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
8366 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
8371 The '``fptrunc``' instruction takes a :ref:`floating-point <t_floating>`
8372 value to cast and a :ref:`floating-point <t_floating>` type to cast it to.
8373 The size of ``value`` must be larger than the size of ``ty2``. This
8374 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
8379 The '``fptrunc``' instruction casts a ``value`` from a larger
8380 :ref:`floating-point <t_floating>` type to a smaller :ref:`floating-point
8382 This instruction is assumed to execute in the default :ref:`floating-point
8383 environment <floatenv>`.
8388 .. code-block:: llvm
8390 %X = fptrunc double 16777217.0 to float ; yields float:16777216.0
8391 %Y = fptrunc double 1.0E+300 to half ; yields half:+infinity
8393 '``fpext .. to``' Instruction
8394 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8401 <result> = fpext <ty> <value> to <ty2> ; yields ty2
8406 The '``fpext``' extends a floating-point ``value`` to a larger floating-point
8412 The '``fpext``' instruction takes a :ref:`floating-point <t_floating>`
8413 ``value`` to cast, and a :ref:`floating-point <t_floating>` type to cast it
8414 to. The source type must be smaller than the destination type.
8419 The '``fpext``' instruction extends the ``value`` from a smaller
8420 :ref:`floating-point <t_floating>` type to a larger :ref:`floating-point
8421 <t_floating>` type. The ``fpext`` cannot be used to make a
8422 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
8423 *no-op cast* for a floating-point cast.
8428 .. code-block:: llvm
8430 %X = fpext float 3.125 to double ; yields double:3.125000e+00
8431 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
8433 '``fptoui .. to``' Instruction
8434 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8441 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
8446 The '``fptoui``' converts a floating-point ``value`` to its unsigned
8447 integer equivalent of type ``ty2``.
8452 The '``fptoui``' instruction takes a value to cast, which must be a
8453 scalar or vector :ref:`floating-point <t_floating>` value, and a type to
8454 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
8455 ``ty`` is a vector floating-point type, ``ty2`` must be a vector integer
8456 type with the same number of elements as ``ty``
8461 The '``fptoui``' instruction converts its :ref:`floating-point
8462 <t_floating>` operand into the nearest (rounding towards zero)
8463 unsigned integer value. If the value cannot fit in ``ty2``, the results
8469 .. code-block:: llvm
8471 %X = fptoui double 123.0 to i32 ; yields i32:123
8472 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
8473 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
8475 '``fptosi .. to``' Instruction
8476 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8483 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
8488 The '``fptosi``' instruction converts :ref:`floating-point <t_floating>`
8489 ``value`` to type ``ty2``.
8494 The '``fptosi``' instruction takes a value to cast, which must be a
8495 scalar or vector :ref:`floating-point <t_floating>` value, and a type to
8496 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
8497 ``ty`` is a vector floating-point type, ``ty2`` must be a vector integer
8498 type with the same number of elements as ``ty``
8503 The '``fptosi``' instruction converts its :ref:`floating-point
8504 <t_floating>` operand into the nearest (rounding towards zero)
8505 signed integer value. If the value cannot fit in ``ty2``, the results
8511 .. code-block:: llvm
8513 %X = fptosi double -123.0 to i32 ; yields i32:-123
8514 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
8515 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
8517 '``uitofp .. to``' Instruction
8518 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8525 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
8530 The '``uitofp``' instruction regards ``value`` as an unsigned integer
8531 and converts that value to the ``ty2`` type.
8536 The '``uitofp``' instruction takes a value to cast, which must be a
8537 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
8538 ``ty2``, which must be an :ref:`floating-point <t_floating>` type. If
8539 ``ty`` is a vector integer type, ``ty2`` must be a vector floating-point
8540 type with the same number of elements as ``ty``
8545 The '``uitofp``' instruction interprets its operand as an unsigned
8546 integer quantity and converts it to the corresponding floating-point
8547 value. If the value cannot fit in the floating-point value, the results
8553 .. code-block:: llvm
8555 %X = uitofp i32 257 to float ; yields float:257.0
8556 %Y = uitofp i8 -1 to double ; yields double:255.0
8558 '``sitofp .. to``' Instruction
8559 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8566 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
8571 The '``sitofp``' instruction regards ``value`` as a signed integer and
8572 converts that value to the ``ty2`` type.
8577 The '``sitofp``' instruction takes a value to cast, which must be a
8578 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
8579 ``ty2``, which must be an :ref:`floating-point <t_floating>` type. If
8580 ``ty`` is a vector integer type, ``ty2`` must be a vector floating-point
8581 type with the same number of elements as ``ty``
8586 The '``sitofp``' instruction interprets its operand as a signed integer
8587 quantity and converts it to the corresponding floating-point value. If
8588 the value cannot fit in the floating-point value, the results are
8594 .. code-block:: llvm
8596 %X = sitofp i32 257 to float ; yields float:257.0
8597 %Y = sitofp i8 -1 to double ; yields double:-1.0
8601 '``ptrtoint .. to``' Instruction
8602 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8609 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
8614 The '``ptrtoint``' instruction converts the pointer or a vector of
8615 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
8620 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
8621 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
8622 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
8623 a vector of integers type.
8628 The '``ptrtoint``' instruction converts ``value`` to integer type
8629 ``ty2`` by interpreting the pointer value as an integer and either
8630 truncating or zero extending that value to the size of the integer type.
8631 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
8632 ``value`` is larger than ``ty2`` then a truncation is done. If they are
8633 the same size, then nothing is done (*no-op cast*) other than a type
8639 .. code-block:: llvm
8641 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
8642 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
8643 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
8647 '``inttoptr .. to``' Instruction
8648 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8655 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
8660 The '``inttoptr``' instruction converts an integer ``value`` to a
8661 pointer type, ``ty2``.
8666 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
8667 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
8673 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
8674 applying either a zero extension or a truncation depending on the size
8675 of the integer ``value``. If ``value`` is larger than the size of a
8676 pointer then a truncation is done. If ``value`` is smaller than the size
8677 of a pointer then a zero extension is done. If they are the same size,
8678 nothing is done (*no-op cast*).
8683 .. code-block:: llvm
8685 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
8686 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
8687 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
8688 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
8692 '``bitcast .. to``' Instruction
8693 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8700 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
8705 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
8711 The '``bitcast``' instruction takes a value to cast, which must be a
8712 non-aggregate first class value, and a type to cast it to, which must
8713 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
8714 bit sizes of ``value`` and the destination type, ``ty2``, must be
8715 identical. If the source type is a pointer, the destination type must
8716 also be a pointer of the same size. This instruction supports bitwise
8717 conversion of vectors to integers and to vectors of other types (as
8718 long as they have the same size).
8723 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
8724 is always a *no-op cast* because no bits change with this
8725 conversion. The conversion is done as if the ``value`` had been stored
8726 to memory and read back as type ``ty2``. Pointer (or vector of
8727 pointers) types may only be converted to other pointer (or vector of
8728 pointers) types with the same address space through this instruction.
8729 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
8730 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
8735 .. code-block:: text
8737 %X = bitcast i8 255 to i8 ; yields i8 :-1
8738 %Y = bitcast i32* %x to sint* ; yields sint*:%x
8739 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
8740 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
8742 .. _i_addrspacecast:
8744 '``addrspacecast .. to``' Instruction
8745 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8752 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
8757 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
8758 address space ``n`` to type ``pty2`` in address space ``m``.
8763 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
8764 to cast and a pointer type to cast it to, which must have a different
8770 The '``addrspacecast``' instruction converts the pointer value
8771 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
8772 value modification, depending on the target and the address space
8773 pair. Pointer conversions within the same address space must be
8774 performed with the ``bitcast`` instruction. Note that if the address space
8775 conversion is legal then both result and operand refer to the same memory
8781 .. code-block:: llvm
8783 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
8784 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
8785 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
8792 The instructions in this category are the "miscellaneous" instructions,
8793 which defy better classification.
8797 '``icmp``' Instruction
8798 ^^^^^^^^^^^^^^^^^^^^^^
8805 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8810 The '``icmp``' instruction returns a boolean value or a vector of
8811 boolean values based on comparison of its two integer, integer vector,
8812 pointer, or pointer vector operands.
8817 The '``icmp``' instruction takes three operands. The first operand is
8818 the condition code indicating the kind of comparison to perform. It is
8819 not a value, just a keyword. The possible condition codes are:
8822 #. ``ne``: not equal
8823 #. ``ugt``: unsigned greater than
8824 #. ``uge``: unsigned greater or equal
8825 #. ``ult``: unsigned less than
8826 #. ``ule``: unsigned less or equal
8827 #. ``sgt``: signed greater than
8828 #. ``sge``: signed greater or equal
8829 #. ``slt``: signed less than
8830 #. ``sle``: signed less or equal
8832 The remaining two arguments must be :ref:`integer <t_integer>` or
8833 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
8834 must also be identical types.
8839 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
8840 code given as ``cond``. The comparison performed always yields either an
8841 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
8843 #. ``eq``: yields ``true`` if the operands are equal, ``false``
8844 otherwise. No sign interpretation is necessary or performed.
8845 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
8846 otherwise. No sign interpretation is necessary or performed.
8847 #. ``ugt``: interprets the operands as unsigned values and yields
8848 ``true`` if ``op1`` is greater than ``op2``.
8849 #. ``uge``: interprets the operands as unsigned values and yields
8850 ``true`` if ``op1`` is greater than or equal to ``op2``.
8851 #. ``ult``: interprets the operands as unsigned values and yields
8852 ``true`` if ``op1`` is less than ``op2``.
8853 #. ``ule``: interprets the operands as unsigned values and yields
8854 ``true`` if ``op1`` is less than or equal to ``op2``.
8855 #. ``sgt``: interprets the operands as signed values and yields ``true``
8856 if ``op1`` is greater than ``op2``.
8857 #. ``sge``: interprets the operands as signed values and yields ``true``
8858 if ``op1`` is greater than or equal to ``op2``.
8859 #. ``slt``: interprets the operands as signed values and yields ``true``
8860 if ``op1`` is less than ``op2``.
8861 #. ``sle``: interprets the operands as signed values and yields ``true``
8862 if ``op1`` is less than or equal to ``op2``.
8864 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
8865 are compared as if they were integers.
8867 If the operands are integer vectors, then they are compared element by
8868 element. The result is an ``i1`` vector with the same number of elements
8869 as the values being compared. Otherwise, the result is an ``i1``.
8874 .. code-block:: text
8876 <result> = icmp eq i32 4, 5 ; yields: result=false
8877 <result> = icmp ne float* %X, %X ; yields: result=false
8878 <result> = icmp ult i16 4, 5 ; yields: result=true
8879 <result> = icmp sgt i16 4, 5 ; yields: result=false
8880 <result> = icmp ule i16 -4, 5 ; yields: result=false
8881 <result> = icmp sge i16 4, 5 ; yields: result=false
8885 '``fcmp``' Instruction
8886 ^^^^^^^^^^^^^^^^^^^^^^
8893 <result> = fcmp [fast-math flags]* <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8898 The '``fcmp``' instruction returns a boolean value or vector of boolean
8899 values based on comparison of its operands.
8901 If the operands are floating-point scalars, then the result type is a
8902 boolean (:ref:`i1 <t_integer>`).
8904 If the operands are floating-point vectors, then the result type is a
8905 vector of boolean with the same number of elements as the operands being
8911 The '``fcmp``' instruction takes three operands. The first operand is
8912 the condition code indicating the kind of comparison to perform. It is
8913 not a value, just a keyword. The possible condition codes are:
8915 #. ``false``: no comparison, always returns false
8916 #. ``oeq``: ordered and equal
8917 #. ``ogt``: ordered and greater than
8918 #. ``oge``: ordered and greater than or equal
8919 #. ``olt``: ordered and less than
8920 #. ``ole``: ordered and less than or equal
8921 #. ``one``: ordered and not equal
8922 #. ``ord``: ordered (no nans)
8923 #. ``ueq``: unordered or equal
8924 #. ``ugt``: unordered or greater than
8925 #. ``uge``: unordered or greater than or equal
8926 #. ``ult``: unordered or less than
8927 #. ``ule``: unordered or less than or equal
8928 #. ``une``: unordered or not equal
8929 #. ``uno``: unordered (either nans)
8930 #. ``true``: no comparison, always returns true
8932 *Ordered* means that neither operand is a QNAN while *unordered* means
8933 that either operand may be a QNAN.
8935 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating-point
8936 <t_floating>` type or a :ref:`vector <t_vector>` of floating-point type.
8937 They must have identical types.
8942 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
8943 condition code given as ``cond``. If the operands are vectors, then the
8944 vectors are compared element by element. Each comparison performed
8945 always yields an :ref:`i1 <t_integer>` result, as follows:
8947 #. ``false``: always yields ``false``, regardless of operands.
8948 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
8949 is equal to ``op2``.
8950 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
8951 is greater than ``op2``.
8952 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
8953 is greater than or equal to ``op2``.
8954 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
8955 is less than ``op2``.
8956 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
8957 is less than or equal to ``op2``.
8958 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
8959 is not equal to ``op2``.
8960 #. ``ord``: yields ``true`` if both operands are not a QNAN.
8961 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
8963 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
8964 greater than ``op2``.
8965 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
8966 greater than or equal to ``op2``.
8967 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
8969 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
8970 less than or equal to ``op2``.
8971 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
8972 not equal to ``op2``.
8973 #. ``uno``: yields ``true`` if either operand is a QNAN.
8974 #. ``true``: always yields ``true``, regardless of operands.
8976 The ``fcmp`` instruction can also optionally take any number of
8977 :ref:`fast-math flags <fastmath>`, which are optimization hints to enable
8978 otherwise unsafe floating-point optimizations.
8980 Any set of fast-math flags are legal on an ``fcmp`` instruction, but the
8981 only flags that have any effect on its semantics are those that allow
8982 assumptions to be made about the values of input arguments; namely
8983 ``nnan``, ``ninf``, and ``nsz``. See :ref:`fastmath` for more information.
8988 .. code-block:: text
8990 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
8991 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
8992 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
8993 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
8997 '``phi``' Instruction
8998 ^^^^^^^^^^^^^^^^^^^^^
9005 <result> = phi <ty> [ <val0>, <label0>], ...
9010 The '``phi``' instruction is used to implement the φ node in the SSA
9011 graph representing the function.
9016 The type of the incoming values is specified with the first type field.
9017 After this, the '``phi``' instruction takes a list of pairs as
9018 arguments, with one pair for each predecessor basic block of the current
9019 block. Only values of :ref:`first class <t_firstclass>` type may be used as
9020 the value arguments to the PHI node. Only labels may be used as the
9023 There must be no non-phi instructions between the start of a basic block
9024 and the PHI instructions: i.e. PHI instructions must be first in a basic
9027 For the purposes of the SSA form, the use of each incoming value is
9028 deemed to occur on the edge from the corresponding predecessor block to
9029 the current block (but after any definition of an '``invoke``'
9030 instruction's return value on the same edge).
9035 At runtime, the '``phi``' instruction logically takes on the value
9036 specified by the pair corresponding to the predecessor basic block that
9037 executed just prior to the current block.
9042 .. code-block:: llvm
9044 Loop: ; Infinite loop that counts from 0 on up...
9045 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
9046 %nextindvar = add i32 %indvar, 1
9051 '``select``' Instruction
9052 ^^^^^^^^^^^^^^^^^^^^^^^^
9059 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
9061 selty is either i1 or {<N x i1>}
9066 The '``select``' instruction is used to choose one value based on a
9067 condition, without IR-level branching.
9072 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
9073 values indicating the condition, and two values of the same :ref:`first
9074 class <t_firstclass>` type.
9079 If the condition is an i1 and it evaluates to 1, the instruction returns
9080 the first value argument; otherwise, it returns the second value
9083 If the condition is a vector of i1, then the value arguments must be
9084 vectors of the same size, and the selection is done element by element.
9086 If the condition is an i1 and the value arguments are vectors of the
9087 same size, then an entire vector is selected.
9092 .. code-block:: llvm
9094 %X = select i1 true, i8 17, i8 42 ; yields i8:17
9098 '``call``' Instruction
9099 ^^^^^^^^^^^^^^^^^^^^^^
9106 <result> = [tail | musttail | notail ] call [fast-math flags] [cconv] [ret attrs] <ty>|<fnty> <fnptrval>(<function args>) [fn attrs]
9112 The '``call``' instruction represents a simple function call.
9117 This instruction requires several arguments:
9119 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
9120 should perform tail call optimization. The ``tail`` marker is a hint that
9121 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
9122 means that the call must be tail call optimized in order for the program to
9123 be correct. The ``musttail`` marker provides these guarantees:
9125 #. The call will not cause unbounded stack growth if it is part of a
9126 recursive cycle in the call graph.
9127 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
9130 Both markers imply that the callee does not access allocas from the caller.
9131 The ``tail`` marker additionally implies that the callee does not access
9132 varargs from the caller, while ``musttail`` implies that varargs from the
9133 caller are passed to the callee. Calls marked ``musttail`` must obey the
9134 following additional rules:
9136 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
9137 or a pointer bitcast followed by a ret instruction.
9138 - The ret instruction must return the (possibly bitcasted) value
9139 produced by the call or void.
9140 - The caller and callee prototypes must match. Pointer types of
9141 parameters or return types may differ in pointee type, but not
9143 - The calling conventions of the caller and callee must match.
9144 - All ABI-impacting function attributes, such as sret, byval, inreg,
9145 returned, and inalloca, must match.
9146 - The callee must be varargs iff the caller is varargs. Bitcasting a
9147 non-varargs function to the appropriate varargs type is legal so
9148 long as the non-varargs prefixes obey the other rules.
9150 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
9151 the following conditions are met:
9153 - Caller and callee both have the calling convention ``fastcc``.
9154 - The call is in tail position (ret immediately follows call and ret
9155 uses value of call or is void).
9156 - Option ``-tailcallopt`` is enabled, or
9157 ``llvm::GuaranteedTailCallOpt`` is ``true``.
9158 - `Platform-specific constraints are
9159 met. <CodeGenerator.html#tailcallopt>`_
9161 #. The optional ``notail`` marker indicates that the optimizers should not add
9162 ``tail`` or ``musttail`` markers to the call. It is used to prevent tail
9163 call optimization from being performed on the call.
9165 #. The optional ``fast-math flags`` marker indicates that the call has one or more
9166 :ref:`fast-math flags <fastmath>`, which are optimization hints to enable
9167 otherwise unsafe floating-point optimizations. Fast-math flags are only valid
9168 for calls that return a floating-point scalar or vector type.
9170 #. The optional "cconv" marker indicates which :ref:`calling
9171 convention <callingconv>` the call should use. If none is
9172 specified, the call defaults to using C calling conventions. The
9173 calling convention of the call must match the calling convention of
9174 the target function, or else the behavior is undefined.
9175 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
9176 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
9178 #. '``ty``': the type of the call instruction itself which is also the
9179 type of the return value. Functions that return no value are marked
9181 #. '``fnty``': shall be the signature of the function being called. The
9182 argument types must match the types implied by this signature. This
9183 type can be omitted if the function is not varargs.
9184 #. '``fnptrval``': An LLVM value containing a pointer to a function to
9185 be called. In most cases, this is a direct function call, but
9186 indirect ``call``'s are just as possible, calling an arbitrary pointer
9188 #. '``function args``': argument list whose types match the function
9189 signature argument types and parameter attributes. All arguments must
9190 be of :ref:`first class <t_firstclass>` type. If the function signature
9191 indicates the function accepts a variable number of arguments, the
9192 extra arguments can be specified.
9193 #. The optional :ref:`function attributes <fnattrs>` list.
9194 #. The optional :ref:`operand bundles <opbundles>` list.
9199 The '``call``' instruction is used to cause control flow to transfer to
9200 a specified function, with its incoming arguments bound to the specified
9201 values. Upon a '``ret``' instruction in the called function, control
9202 flow continues with the instruction after the function call, and the
9203 return value of the function is bound to the result argument.
9208 .. code-block:: llvm
9210 %retval = call i32 @test(i32 %argc)
9211 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
9212 %X = tail call i32 @foo() ; yields i32
9213 %Y = tail call fastcc i32 @foo() ; yields i32
9214 call void %foo(i8 97 signext)
9216 %struct.A = type { i32, i8 }
9217 %r = call %struct.A @foo() ; yields { i32, i8 }
9218 %gr = extractvalue %struct.A %r, 0 ; yields i32
9219 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
9220 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
9221 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
9223 llvm treats calls to some functions with names and arguments that match
9224 the standard C99 library as being the C99 library functions, and may
9225 perform optimizations or generate code for them under that assumption.
9226 This is something we'd like to change in the future to provide better
9227 support for freestanding environments and non-C-based languages.
9231 '``va_arg``' Instruction
9232 ^^^^^^^^^^^^^^^^^^^^^^^^
9239 <resultval> = va_arg <va_list*> <arglist>, <argty>
9244 The '``va_arg``' instruction is used to access arguments passed through
9245 the "variable argument" area of a function call. It is used to implement
9246 the ``va_arg`` macro in C.
9251 This instruction takes a ``va_list*`` value and the type of the
9252 argument. It returns a value of the specified argument type and
9253 increments the ``va_list`` to point to the next argument. The actual
9254 type of ``va_list`` is target specific.
9259 The '``va_arg``' instruction loads an argument of the specified type
9260 from the specified ``va_list`` and causes the ``va_list`` to point to
9261 the next argument. For more information, see the variable argument
9262 handling :ref:`Intrinsic Functions <int_varargs>`.
9264 It is legal for this instruction to be called in a function which does
9265 not take a variable number of arguments, for example, the ``vfprintf``
9268 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
9269 function <intrinsics>` because it takes a type as an argument.
9274 See the :ref:`variable argument processing <int_varargs>` section.
9276 Note that the code generator does not yet fully support va\_arg on many
9277 targets. Also, it does not currently support va\_arg with aggregate
9278 types on any target.
9282 '``landingpad``' Instruction
9283 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9290 <resultval> = landingpad <resultty> <clause>+
9291 <resultval> = landingpad <resultty> cleanup <clause>*
9293 <clause> := catch <type> <value>
9294 <clause> := filter <array constant type> <array constant>
9299 The '``landingpad``' instruction is used by `LLVM's exception handling
9300 system <ExceptionHandling.html#overview>`_ to specify that a basic block
9301 is a landing pad --- one where the exception lands, and corresponds to the
9302 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
9303 defines values supplied by the :ref:`personality function <personalityfn>` upon
9304 re-entry to the function. The ``resultval`` has the type ``resultty``.
9310 ``cleanup`` flag indicates that the landing pad block is a cleanup.
9312 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
9313 contains the global variable representing the "type" that may be caught
9314 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
9315 clause takes an array constant as its argument. Use
9316 "``[0 x i8**] undef``" for a filter which cannot throw. The
9317 '``landingpad``' instruction must contain *at least* one ``clause`` or
9318 the ``cleanup`` flag.
9323 The '``landingpad``' instruction defines the values which are set by the
9324 :ref:`personality function <personalityfn>` upon re-entry to the function, and
9325 therefore the "result type" of the ``landingpad`` instruction. As with
9326 calling conventions, how the personality function results are
9327 represented in LLVM IR is target specific.
9329 The clauses are applied in order from top to bottom. If two
9330 ``landingpad`` instructions are merged together through inlining, the
9331 clauses from the calling function are appended to the list of clauses.
9332 When the call stack is being unwound due to an exception being thrown,
9333 the exception is compared against each ``clause`` in turn. If it doesn't
9334 match any of the clauses, and the ``cleanup`` flag is not set, then
9335 unwinding continues further up the call stack.
9337 The ``landingpad`` instruction has several restrictions:
9339 - A landing pad block is a basic block which is the unwind destination
9340 of an '``invoke``' instruction.
9341 - A landing pad block must have a '``landingpad``' instruction as its
9342 first non-PHI instruction.
9343 - There can be only one '``landingpad``' instruction within the landing
9345 - A basic block that is not a landing pad block may not include a
9346 '``landingpad``' instruction.
9351 .. code-block:: llvm
9353 ;; A landing pad which can catch an integer.
9354 %res = landingpad { i8*, i32 }
9356 ;; A landing pad that is a cleanup.
9357 %res = landingpad { i8*, i32 }
9359 ;; A landing pad which can catch an integer and can only throw a double.
9360 %res = landingpad { i8*, i32 }
9362 filter [1 x i8**] [@_ZTId]
9366 '``catchpad``' Instruction
9367 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9374 <resultval> = catchpad within <catchswitch> [<args>*]
9379 The '``catchpad``' instruction is used by `LLVM's exception handling
9380 system <ExceptionHandling.html#overview>`_ to specify that a basic block
9381 begins a catch handler --- one where a personality routine attempts to transfer
9382 control to catch an exception.
9387 The ``catchswitch`` operand must always be a token produced by a
9388 :ref:`catchswitch <i_catchswitch>` instruction in a predecessor block. This
9389 ensures that each ``catchpad`` has exactly one predecessor block, and it always
9390 terminates in a ``catchswitch``.
9392 The ``args`` correspond to whatever information the personality routine
9393 requires to know if this is an appropriate handler for the exception. Control
9394 will transfer to the ``catchpad`` if this is the first appropriate handler for
9397 The ``resultval`` has the type :ref:`token <t_token>` and is used to match the
9398 ``catchpad`` to corresponding :ref:`catchrets <i_catchret>` and other nested EH
9404 When the call stack is being unwound due to an exception being thrown, the
9405 exception is compared against the ``args``. If it doesn't match, control will
9406 not reach the ``catchpad`` instruction. The representation of ``args`` is
9407 entirely target and personality function-specific.
9409 Like the :ref:`landingpad <i_landingpad>` instruction, the ``catchpad``
9410 instruction must be the first non-phi of its parent basic block.
9412 The meaning of the tokens produced and consumed by ``catchpad`` and other "pad"
9413 instructions is described in the
9414 `Windows exception handling documentation\ <ExceptionHandling.html#wineh>`_.
9416 When a ``catchpad`` has been "entered" but not yet "exited" (as
9417 described in the `EH documentation\ <ExceptionHandling.html#wineh-constraints>`_),
9418 it is undefined behavior to execute a :ref:`call <i_call>` or :ref:`invoke <i_invoke>`
9419 that does not carry an appropriate :ref:`"funclet" bundle <ob_funclet>`.
9424 .. code-block:: text
9427 %cs = catchswitch within none [label %handler0] unwind to caller
9428 ;; A catch block which can catch an integer.
9430 %tok = catchpad within %cs [i8** @_ZTIi]
9434 '``cleanuppad``' Instruction
9435 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9442 <resultval> = cleanuppad within <parent> [<args>*]
9447 The '``cleanuppad``' instruction is used by `LLVM's exception handling
9448 system <ExceptionHandling.html#overview>`_ to specify that a basic block
9449 is a cleanup block --- one where a personality routine attempts to
9450 transfer control to run cleanup actions.
9451 The ``args`` correspond to whatever additional
9452 information the :ref:`personality function <personalityfn>` requires to
9453 execute the cleanup.
9454 The ``resultval`` has the type :ref:`token <t_token>` and is used to
9455 match the ``cleanuppad`` to corresponding :ref:`cleanuprets <i_cleanupret>`.
9456 The ``parent`` argument is the token of the funclet that contains the
9457 ``cleanuppad`` instruction. If the ``cleanuppad`` is not inside a funclet,
9458 this operand may be the token ``none``.
9463 The instruction takes a list of arbitrary values which are interpreted
9464 by the :ref:`personality function <personalityfn>`.
9469 When the call stack is being unwound due to an exception being thrown,
9470 the :ref:`personality function <personalityfn>` transfers control to the
9471 ``cleanuppad`` with the aid of the personality-specific arguments.
9472 As with calling conventions, how the personality function results are
9473 represented in LLVM IR is target specific.
9475 The ``cleanuppad`` instruction has several restrictions:
9477 - A cleanup block is a basic block which is the unwind destination of
9478 an exceptional instruction.
9479 - A cleanup block must have a '``cleanuppad``' instruction as its
9480 first non-PHI instruction.
9481 - There can be only one '``cleanuppad``' instruction within the
9483 - A basic block that is not a cleanup block may not include a
9484 '``cleanuppad``' instruction.
9486 When a ``cleanuppad`` has been "entered" but not yet "exited" (as
9487 described in the `EH documentation\ <ExceptionHandling.html#wineh-constraints>`_),
9488 it is undefined behavior to execute a :ref:`call <i_call>` or :ref:`invoke <i_invoke>`
9489 that does not carry an appropriate :ref:`"funclet" bundle <ob_funclet>`.
9494 .. code-block:: text
9496 %tok = cleanuppad within %cs []
9503 LLVM supports the notion of an "intrinsic function". These functions
9504 have well known names and semantics and are required to follow certain
9505 restrictions. Overall, these intrinsics represent an extension mechanism
9506 for the LLVM language that does not require changing all of the
9507 transformations in LLVM when adding to the language (or the bitcode
9508 reader/writer, the parser, etc...).
9510 Intrinsic function names must all start with an "``llvm.``" prefix. This
9511 prefix is reserved in LLVM for intrinsic names; thus, function names may
9512 not begin with this prefix. Intrinsic functions must always be external
9513 functions: you cannot define the body of intrinsic functions. Intrinsic
9514 functions may only be used in call or invoke instructions: it is illegal
9515 to take the address of an intrinsic function. Additionally, because
9516 intrinsic functions are part of the LLVM language, it is required if any
9517 are added that they be documented here.
9519 Some intrinsic functions can be overloaded, i.e., the intrinsic
9520 represents a family of functions that perform the same operation but on
9521 different data types. Because LLVM can represent over 8 million
9522 different integer types, overloading is used commonly to allow an
9523 intrinsic function to operate on any integer type. One or more of the
9524 argument types or the result type can be overloaded to accept any
9525 integer type. Argument types may also be defined as exactly matching a
9526 previous argument's type or the result type. This allows an intrinsic
9527 function which accepts multiple arguments, but needs all of them to be
9528 of the same type, to only be overloaded with respect to a single
9529 argument or the result.
9531 Overloaded intrinsics will have the names of its overloaded argument
9532 types encoded into its function name, each preceded by a period. Only
9533 those types which are overloaded result in a name suffix. Arguments
9534 whose type is matched against another type do not. For example, the
9535 ``llvm.ctpop`` function can take an integer of any width and returns an
9536 integer of exactly the same integer width. This leads to a family of
9537 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
9538 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
9539 overloaded, and only one type suffix is required. Because the argument's
9540 type is matched against the return type, it does not require its own
9543 To learn how to add an intrinsic function, please see the `Extending
9544 LLVM Guide <ExtendingLLVM.html>`_.
9548 Variable Argument Handling Intrinsics
9549 -------------------------------------
9551 Variable argument support is defined in LLVM with the
9552 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
9553 functions. These functions are related to the similarly named macros
9554 defined in the ``<stdarg.h>`` header file.
9556 All of these functions operate on arguments that use a target-specific
9557 value type "``va_list``". The LLVM assembly language reference manual
9558 does not define what this type is, so all transformations should be
9559 prepared to handle these functions regardless of the type used.
9561 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
9562 variable argument handling intrinsic functions are used.
9564 .. code-block:: llvm
9566 ; This struct is different for every platform. For most platforms,
9567 ; it is merely an i8*.
9568 %struct.va_list = type { i8* }
9570 ; For Unix x86_64 platforms, va_list is the following struct:
9571 ; %struct.va_list = type { i32, i32, i8*, i8* }
9573 define i32 @test(i32 %X, ...) {
9574 ; Initialize variable argument processing
9575 %ap = alloca %struct.va_list
9576 %ap2 = bitcast %struct.va_list* %ap to i8*
9577 call void @llvm.va_start(i8* %ap2)
9579 ; Read a single integer argument
9580 %tmp = va_arg i8* %ap2, i32
9582 ; Demonstrate usage of llvm.va_copy and llvm.va_end
9584 %aq2 = bitcast i8** %aq to i8*
9585 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
9586 call void @llvm.va_end(i8* %aq2)
9588 ; Stop processing of arguments.
9589 call void @llvm.va_end(i8* %ap2)
9593 declare void @llvm.va_start(i8*)
9594 declare void @llvm.va_copy(i8*, i8*)
9595 declare void @llvm.va_end(i8*)
9599 '``llvm.va_start``' Intrinsic
9600 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9607 declare void @llvm.va_start(i8* <arglist>)
9612 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
9613 subsequent use by ``va_arg``.
9618 The argument is a pointer to a ``va_list`` element to initialize.
9623 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
9624 available in C. In a target-dependent way, it initializes the
9625 ``va_list`` element to which the argument points, so that the next call
9626 to ``va_arg`` will produce the first variable argument passed to the
9627 function. Unlike the C ``va_start`` macro, this intrinsic does not need
9628 to know the last argument of the function as the compiler can figure
9631 '``llvm.va_end``' Intrinsic
9632 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9639 declare void @llvm.va_end(i8* <arglist>)
9644 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
9645 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
9650 The argument is a pointer to a ``va_list`` to destroy.
9655 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
9656 available in C. In a target-dependent way, it destroys the ``va_list``
9657 element to which the argument points. Calls to
9658 :ref:`llvm.va_start <int_va_start>` and
9659 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
9664 '``llvm.va_copy``' Intrinsic
9665 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9672 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
9677 The '``llvm.va_copy``' intrinsic copies the current argument position
9678 from the source argument list to the destination argument list.
9683 The first argument is a pointer to a ``va_list`` element to initialize.
9684 The second argument is a pointer to a ``va_list`` element to copy from.
9689 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
9690 available in C. In a target-dependent way, it copies the source
9691 ``va_list`` element into the destination ``va_list`` element. This
9692 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
9693 arbitrarily complex and require, for example, memory allocation.
9695 Accurate Garbage Collection Intrinsics
9696 --------------------------------------
9698 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
9699 (GC) requires the frontend to generate code containing appropriate intrinsic
9700 calls and select an appropriate GC strategy which knows how to lower these
9701 intrinsics in a manner which is appropriate for the target collector.
9703 These intrinsics allow identification of :ref:`GC roots on the
9704 stack <int_gcroot>`, as well as garbage collector implementations that
9705 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
9706 Frontends for type-safe garbage collected languages should generate
9707 these intrinsics to make use of the LLVM garbage collectors. For more
9708 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
9710 Experimental Statepoint Intrinsics
9711 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9713 LLVM provides an second experimental set of intrinsics for describing garbage
9714 collection safepoints in compiled code. These intrinsics are an alternative
9715 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
9716 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
9717 differences in approach are covered in the `Garbage Collection with LLVM
9718 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
9719 described in :doc:`Statepoints`.
9723 '``llvm.gcroot``' Intrinsic
9724 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9731 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
9736 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
9737 the code generator, and allows some metadata to be associated with it.
9742 The first argument specifies the address of a stack object that contains
9743 the root pointer. The second pointer (which must be either a constant or
9744 a global value address) contains the meta-data to be associated with the
9750 At runtime, a call to this intrinsic stores a null pointer into the
9751 "ptrloc" location. At compile-time, the code generator generates
9752 information to allow the runtime to find the pointer at GC safe points.
9753 The '``llvm.gcroot``' intrinsic may only be used in a function which
9754 :ref:`specifies a GC algorithm <gc>`.
9758 '``llvm.gcread``' Intrinsic
9759 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9766 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
9771 The '``llvm.gcread``' intrinsic identifies reads of references from heap
9772 locations, allowing garbage collector implementations that require read
9778 The second argument is the address to read from, which should be an
9779 address allocated from the garbage collector. The first object is a
9780 pointer to the start of the referenced object, if needed by the language
9781 runtime (otherwise null).
9786 The '``llvm.gcread``' intrinsic has the same semantics as a load
9787 instruction, but may be replaced with substantially more complex code by
9788 the garbage collector runtime, as needed. The '``llvm.gcread``'
9789 intrinsic may only be used in a function which :ref:`specifies a GC
9794 '``llvm.gcwrite``' Intrinsic
9795 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9802 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
9807 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
9808 locations, allowing garbage collector implementations that require write
9809 barriers (such as generational or reference counting collectors).
9814 The first argument is the reference to store, the second is the start of
9815 the object to store it to, and the third is the address of the field of
9816 Obj to store to. If the runtime does not require a pointer to the
9817 object, Obj may be null.
9822 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
9823 instruction, but may be replaced with substantially more complex code by
9824 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
9825 intrinsic may only be used in a function which :ref:`specifies a GC
9828 Code Generator Intrinsics
9829 -------------------------
9831 These intrinsics are provided by LLVM to expose special features that
9832 may only be implemented with code generator support.
9834 '``llvm.returnaddress``' Intrinsic
9835 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9842 declare i8* @llvm.returnaddress(i32 <level>)
9847 The '``llvm.returnaddress``' intrinsic attempts to compute a
9848 target-specific value indicating the return address of the current
9849 function or one of its callers.
9854 The argument to this intrinsic indicates which function to return the
9855 address for. Zero indicates the calling function, one indicates its
9856 caller, etc. The argument is **required** to be a constant integer
9862 The '``llvm.returnaddress``' intrinsic either returns a pointer
9863 indicating the return address of the specified call frame, or zero if it
9864 cannot be identified. The value returned by this intrinsic is likely to
9865 be incorrect or 0 for arguments other than zero, so it should only be
9866 used for debugging purposes.
9868 Note that calling this intrinsic does not prevent function inlining or
9869 other aggressive transformations, so the value returned may not be that
9870 of the obvious source-language caller.
9872 '``llvm.addressofreturnaddress``' Intrinsic
9873 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9880 declare i8* @llvm.addressofreturnaddress()
9885 The '``llvm.addressofreturnaddress``' intrinsic returns a target-specific
9886 pointer to the place in the stack frame where the return address of the
9887 current function is stored.
9892 Note that calling this intrinsic does not prevent function inlining or
9893 other aggressive transformations, so the value returned may not be that
9894 of the obvious source-language caller.
9896 This intrinsic is only implemented for x86.
9898 '``llvm.frameaddress``' Intrinsic
9899 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9906 declare i8* @llvm.frameaddress(i32 <level>)
9911 The '``llvm.frameaddress``' intrinsic attempts to return the
9912 target-specific frame pointer value for the specified stack frame.
9917 The argument to this intrinsic indicates which function to return the
9918 frame pointer for. Zero indicates the calling function, one indicates
9919 its caller, etc. The argument is **required** to be a constant integer
9925 The '``llvm.frameaddress``' intrinsic either returns a pointer
9926 indicating the frame address of the specified call frame, or zero if it
9927 cannot be identified. The value returned by this intrinsic is likely to
9928 be incorrect or 0 for arguments other than zero, so it should only be
9929 used for debugging purposes.
9931 Note that calling this intrinsic does not prevent function inlining or
9932 other aggressive transformations, so the value returned may not be that
9933 of the obvious source-language caller.
9935 '``llvm.localescape``' and '``llvm.localrecover``' Intrinsics
9936 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9943 declare void @llvm.localescape(...)
9944 declare i8* @llvm.localrecover(i8* %func, i8* %fp, i32 %idx)
9949 The '``llvm.localescape``' intrinsic escapes offsets of a collection of static
9950 allocas, and the '``llvm.localrecover``' intrinsic applies those offsets to a
9951 live frame pointer to recover the address of the allocation. The offset is
9952 computed during frame layout of the caller of ``llvm.localescape``.
9957 All arguments to '``llvm.localescape``' must be pointers to static allocas or
9958 casts of static allocas. Each function can only call '``llvm.localescape``'
9959 once, and it can only do so from the entry block.
9961 The ``func`` argument to '``llvm.localrecover``' must be a constant
9962 bitcasted pointer to a function defined in the current module. The code
9963 generator cannot determine the frame allocation offset of functions defined in
9966 The ``fp`` argument to '``llvm.localrecover``' must be a frame pointer of a
9967 call frame that is currently live. The return value of '``llvm.localaddress``'
9968 is one way to produce such a value, but various runtimes also expose a suitable
9969 pointer in platform-specific ways.
9971 The ``idx`` argument to '``llvm.localrecover``' indicates which alloca passed to
9972 '``llvm.localescape``' to recover. It is zero-indexed.
9977 These intrinsics allow a group of functions to share access to a set of local
9978 stack allocations of a one parent function. The parent function may call the
9979 '``llvm.localescape``' intrinsic once from the function entry block, and the
9980 child functions can use '``llvm.localrecover``' to access the escaped allocas.
9981 The '``llvm.localescape``' intrinsic blocks inlining, as inlining changes where
9982 the escaped allocas are allocated, which would break attempts to use
9983 '``llvm.localrecover``'.
9985 .. _int_read_register:
9986 .. _int_write_register:
9988 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
9989 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9996 declare i32 @llvm.read_register.i32(metadata)
9997 declare i64 @llvm.read_register.i64(metadata)
9998 declare void @llvm.write_register.i32(metadata, i32 @value)
9999 declare void @llvm.write_register.i64(metadata, i64 @value)
10005 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
10006 provides access to the named register. The register must be valid on
10007 the architecture being compiled to. The type needs to be compatible
10008 with the register being read.
10013 The '``llvm.read_register``' intrinsic returns the current value of the
10014 register, where possible. The '``llvm.write_register``' intrinsic sets
10015 the current value of the register, where possible.
10017 This is useful to implement named register global variables that need
10018 to always be mapped to a specific register, as is common practice on
10019 bare-metal programs including OS kernels.
10021 The compiler doesn't check for register availability or use of the used
10022 register in surrounding code, including inline assembly. Because of that,
10023 allocatable registers are not supported.
10025 Warning: So far it only works with the stack pointer on selected
10026 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
10027 work is needed to support other registers and even more so, allocatable
10032 '``llvm.stacksave``' Intrinsic
10033 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10040 declare i8* @llvm.stacksave()
10045 The '``llvm.stacksave``' intrinsic is used to remember the current state
10046 of the function stack, for use with
10047 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
10048 implementing language features like scoped automatic variable sized
10054 This intrinsic returns a opaque pointer value that can be passed to
10055 :ref:`llvm.stackrestore <int_stackrestore>`. When an
10056 ``llvm.stackrestore`` intrinsic is executed with a value saved from
10057 ``llvm.stacksave``, it effectively restores the state of the stack to
10058 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
10059 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
10060 were allocated after the ``llvm.stacksave`` was executed.
10062 .. _int_stackrestore:
10064 '``llvm.stackrestore``' Intrinsic
10065 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10072 declare void @llvm.stackrestore(i8* %ptr)
10077 The '``llvm.stackrestore``' intrinsic is used to restore the state of
10078 the function stack to the state it was in when the corresponding
10079 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
10080 useful for implementing language features like scoped automatic variable
10081 sized arrays in C99.
10086 See the description for :ref:`llvm.stacksave <int_stacksave>`.
10088 .. _int_get_dynamic_area_offset:
10090 '``llvm.get.dynamic.area.offset``' Intrinsic
10091 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10098 declare i32 @llvm.get.dynamic.area.offset.i32()
10099 declare i64 @llvm.get.dynamic.area.offset.i64()
10104 The '``llvm.get.dynamic.area.offset.*``' intrinsic family is used to
10105 get the offset from native stack pointer to the address of the most
10106 recent dynamic alloca on the caller's stack. These intrinsics are
10107 intendend for use in combination with
10108 :ref:`llvm.stacksave <int_stacksave>` to get a
10109 pointer to the most recent dynamic alloca. This is useful, for example,
10110 for AddressSanitizer's stack unpoisoning routines.
10115 These intrinsics return a non-negative integer value that can be used to
10116 get the address of the most recent dynamic alloca, allocated by :ref:`alloca <i_alloca>`
10117 on the caller's stack. In particular, for targets where stack grows downwards,
10118 adding this offset to the native stack pointer would get the address of the most
10119 recent dynamic alloca. For targets where stack grows upwards, the situation is a bit more
10120 complicated, because subtracting this value from stack pointer would get the address
10121 one past the end of the most recent dynamic alloca.
10123 Although for most targets `llvm.get.dynamic.area.offset <int_get_dynamic_area_offset>`
10124 returns just a zero, for others, such as PowerPC and PowerPC64, it returns a
10125 compile-time-known constant value.
10127 The return value type of :ref:`llvm.get.dynamic.area.offset <int_get_dynamic_area_offset>`
10128 must match the target's default address space's (address space 0) pointer type.
10130 '``llvm.prefetch``' Intrinsic
10131 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10138 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
10143 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
10144 insert a prefetch instruction if supported; otherwise, it is a noop.
10145 Prefetches have no effect on the behavior of the program but can change
10146 its performance characteristics.
10151 ``address`` is the address to be prefetched, ``rw`` is the specifier
10152 determining if the fetch should be for a read (0) or write (1), and
10153 ``locality`` is a temporal locality specifier ranging from (0) - no
10154 locality, to (3) - extremely local keep in cache. The ``cache type``
10155 specifies whether the prefetch is performed on the data (1) or
10156 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
10157 arguments must be constant integers.
10162 This intrinsic does not modify the behavior of the program. In
10163 particular, prefetches cannot trap and do not produce a value. On
10164 targets that support this intrinsic, the prefetch can provide hints to
10165 the processor cache for better performance.
10167 '``llvm.pcmarker``' Intrinsic
10168 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10175 declare void @llvm.pcmarker(i32 <id>)
10180 The '``llvm.pcmarker``' intrinsic is a method to export a Program
10181 Counter (PC) in a region of code to simulators and other tools. The
10182 method is target specific, but it is expected that the marker will use
10183 exported symbols to transmit the PC of the marker. The marker makes no
10184 guarantees that it will remain with any specific instruction after
10185 optimizations. It is possible that the presence of a marker will inhibit
10186 optimizations. The intended use is to be inserted after optimizations to
10187 allow correlations of simulation runs.
10192 ``id`` is a numerical id identifying the marker.
10197 This intrinsic does not modify the behavior of the program. Backends
10198 that do not support this intrinsic may ignore it.
10200 '``llvm.readcyclecounter``' Intrinsic
10201 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10208 declare i64 @llvm.readcyclecounter()
10213 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
10214 counter register (or similar low latency, high accuracy clocks) on those
10215 targets that support it. On X86, it should map to RDTSC. On Alpha, it
10216 should map to RPCC. As the backing counters overflow quickly (on the
10217 order of 9 seconds on alpha), this should only be used for small
10223 When directly supported, reading the cycle counter should not modify any
10224 memory. Implementations are allowed to either return a application
10225 specific value or a system wide value. On backends without support, this
10226 is lowered to a constant 0.
10228 Note that runtime support may be conditional on the privilege-level code is
10229 running at and the host platform.
10231 '``llvm.clear_cache``' Intrinsic
10232 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10239 declare void @llvm.clear_cache(i8*, i8*)
10244 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
10245 in the specified range to the execution unit of the processor. On
10246 targets with non-unified instruction and data cache, the implementation
10247 flushes the instruction cache.
10252 On platforms with coherent instruction and data caches (e.g. x86), this
10253 intrinsic is a nop. On platforms with non-coherent instruction and data
10254 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
10255 instructions or a system call, if cache flushing requires special
10258 The default behavior is to emit a call to ``__clear_cache`` from the run
10261 This instrinsic does *not* empty the instruction pipeline. Modifications
10262 of the current function are outside the scope of the intrinsic.
10264 '``llvm.instrprof.increment``' Intrinsic
10265 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10272 declare void @llvm.instrprof.increment(i8* <name>, i64 <hash>,
10273 i32 <num-counters>, i32 <index>)
10278 The '``llvm.instrprof.increment``' intrinsic can be emitted by a
10279 frontend for use with instrumentation based profiling. These will be
10280 lowered by the ``-instrprof`` pass to generate execution counts of a
10281 program at runtime.
10286 The first argument is a pointer to a global variable containing the
10287 name of the entity being instrumented. This should generally be the
10288 (mangled) function name for a set of counters.
10290 The second argument is a hash value that can be used by the consumer
10291 of the profile data to detect changes to the instrumented source, and
10292 the third is the number of counters associated with ``name``. It is an
10293 error if ``hash`` or ``num-counters`` differ between two instances of
10294 ``instrprof.increment`` that refer to the same name.
10296 The last argument refers to which of the counters for ``name`` should
10297 be incremented. It should be a value between 0 and ``num-counters``.
10302 This intrinsic represents an increment of a profiling counter. It will
10303 cause the ``-instrprof`` pass to generate the appropriate data
10304 structures and the code to increment the appropriate value, in a
10305 format that can be written out by a compiler runtime and consumed via
10306 the ``llvm-profdata`` tool.
10308 '``llvm.instrprof.increment.step``' Intrinsic
10309 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10316 declare void @llvm.instrprof.increment.step(i8* <name>, i64 <hash>,
10317 i32 <num-counters>,
10318 i32 <index>, i64 <step>)
10323 The '``llvm.instrprof.increment.step``' intrinsic is an extension to
10324 the '``llvm.instrprof.increment``' intrinsic with an additional fifth
10325 argument to specify the step of the increment.
10329 The first four arguments are the same as '``llvm.instrprof.increment``'
10332 The last argument specifies the value of the increment of the counter variable.
10336 See description of '``llvm.instrprof.increment``' instrinsic.
10339 '``llvm.instrprof.value.profile``' Intrinsic
10340 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10347 declare void @llvm.instrprof.value.profile(i8* <name>, i64 <hash>,
10348 i64 <value>, i32 <value_kind>,
10354 The '``llvm.instrprof.value.profile``' intrinsic can be emitted by a
10355 frontend for use with instrumentation based profiling. This will be
10356 lowered by the ``-instrprof`` pass to find out the target values,
10357 instrumented expressions take in a program at runtime.
10362 The first argument is a pointer to a global variable containing the
10363 name of the entity being instrumented. ``name`` should generally be the
10364 (mangled) function name for a set of counters.
10366 The second argument is a hash value that can be used by the consumer
10367 of the profile data to detect changes to the instrumented source. It
10368 is an error if ``hash`` differs between two instances of
10369 ``llvm.instrprof.*`` that refer to the same name.
10371 The third argument is the value of the expression being profiled. The profiled
10372 expression's value should be representable as an unsigned 64-bit value. The
10373 fourth argument represents the kind of value profiling that is being done. The
10374 supported value profiling kinds are enumerated through the
10375 ``InstrProfValueKind`` type declared in the
10376 ``<include/llvm/ProfileData/InstrProf.h>`` header file. The last argument is the
10377 index of the instrumented expression within ``name``. It should be >= 0.
10382 This intrinsic represents the point where a call to a runtime routine
10383 should be inserted for value profiling of target expressions. ``-instrprof``
10384 pass will generate the appropriate data structures and replace the
10385 ``llvm.instrprof.value.profile`` intrinsic with the call to the profile
10386 runtime library with proper arguments.
10388 '``llvm.thread.pointer``' Intrinsic
10389 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10396 declare i8* @llvm.thread.pointer()
10401 The '``llvm.thread.pointer``' intrinsic returns the value of the thread
10407 The '``llvm.thread.pointer``' intrinsic returns a pointer to the TLS area
10408 for the current thread. The exact semantics of this value are target
10409 specific: it may point to the start of TLS area, to the end, or somewhere
10410 in the middle. Depending on the target, this intrinsic may read a register,
10411 call a helper function, read from an alternate memory space, or perform
10412 other operations necessary to locate the TLS area. Not all targets support
10415 Standard C Library Intrinsics
10416 -----------------------------
10418 LLVM provides intrinsics for a few important standard C library
10419 functions. These intrinsics allow source-language front-ends to pass
10420 information about the alignment of the pointer arguments to the code
10421 generator, providing opportunity for more efficient code generation.
10425 '``llvm.memcpy``' Intrinsic
10426 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10431 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
10432 integer bit width and for different address spaces. Not all targets
10433 support all bit widths however.
10437 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
10438 i32 <len>, i1 <isvolatile>)
10439 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
10440 i64 <len>, i1 <isvolatile>)
10445 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
10446 source location to the destination location.
10448 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
10449 intrinsics do not return a value, takes extra isvolatile
10450 arguments and the pointers can be in specified address spaces.
10455 The first argument is a pointer to the destination, the second is a
10456 pointer to the source. The third argument is an integer argument
10457 specifying the number of bytes to copy, and the fourth is a
10458 boolean indicating a volatile access.
10460 The :ref:`align <attr_align>` parameter attribute can be provided
10461 for the first and second arguments.
10463 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
10464 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
10465 very cleanly specified and it is unwise to depend on it.
10470 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
10471 source location to the destination location, which are not allowed to
10472 overlap. It copies "len" bytes of memory over. If the argument is known
10473 to be aligned to some boundary, this can be specified as the fourth
10474 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
10478 '``llvm.memmove``' Intrinsic
10479 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10484 This is an overloaded intrinsic. You can use llvm.memmove on any integer
10485 bit width and for different address space. Not all targets support all
10486 bit widths however.
10490 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
10491 i32 <len>, i1 <isvolatile>)
10492 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
10493 i64 <len>, i1 <isvolatile>)
10498 The '``llvm.memmove.*``' intrinsics move a block of memory from the
10499 source location to the destination location. It is similar to the
10500 '``llvm.memcpy``' intrinsic but allows the two memory locations to
10503 Note that, unlike the standard libc function, the ``llvm.memmove.*``
10504 intrinsics do not return a value, takes an extra isvolatile
10505 argument and the pointers can be in specified address spaces.
10510 The first argument is a pointer to the destination, the second is a
10511 pointer to the source. The third argument is an integer argument
10512 specifying the number of bytes to copy, and the fourth is a
10513 boolean indicating a volatile access.
10515 The :ref:`align <attr_align>` parameter attribute can be provided
10516 for the first and second arguments.
10518 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
10519 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
10520 not very cleanly specified and it is unwise to depend on it.
10525 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
10526 source location to the destination location, which may overlap. It
10527 copies "len" bytes of memory over. If the argument is known to be
10528 aligned to some boundary, this can be specified as the fourth argument,
10529 otherwise it should be set to 0 or 1 (both meaning no alignment).
10533 '``llvm.memset.*``' Intrinsics
10534 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10539 This is an overloaded intrinsic. You can use llvm.memset on any integer
10540 bit width and for different address spaces. However, not all targets
10541 support all bit widths.
10545 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
10546 i32 <len>, i1 <isvolatile>)
10547 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
10548 i64 <len>, i1 <isvolatile>)
10553 The '``llvm.memset.*``' intrinsics fill a block of memory with a
10554 particular byte value.
10556 Note that, unlike the standard libc function, the ``llvm.memset``
10557 intrinsic does not return a value and takes an extra volatile
10558 argument. Also, the destination can be in an arbitrary address space.
10563 The first argument is a pointer to the destination to fill, the second
10564 is the byte value with which to fill it, the third argument is an
10565 integer argument specifying the number of bytes to fill, and the fourth
10566 is a boolean indicating a volatile access.
10568 The :ref:`align <attr_align>` parameter attribute can be provided
10569 for the first arguments.
10571 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
10572 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
10573 very cleanly specified and it is unwise to depend on it.
10578 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
10579 at the destination location.
10581 '``llvm.sqrt.*``' Intrinsic
10582 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10587 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
10588 floating-point or vector of floating-point type. Not all targets support
10593 declare float @llvm.sqrt.f32(float %Val)
10594 declare double @llvm.sqrt.f64(double %Val)
10595 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
10596 declare fp128 @llvm.sqrt.f128(fp128 %Val)
10597 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
10602 The '``llvm.sqrt``' intrinsics return the square root of the specified value.
10607 The argument and return value are floating-point numbers of the same type.
10612 Return the same value as a corresponding libm '``sqrt``' function but without
10613 trapping or setting ``errno``. For types specified by IEEE-754, the result
10614 matches a conforming libm implementation.
10616 When specified with the fast-math-flag 'afn', the result may be approximated
10617 using a less accurate calculation.
10619 '``llvm.powi.*``' Intrinsic
10620 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10625 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
10626 floating-point or vector of floating-point type. Not all targets support
10631 declare float @llvm.powi.f32(float %Val, i32 %power)
10632 declare double @llvm.powi.f64(double %Val, i32 %power)
10633 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
10634 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
10635 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
10640 The '``llvm.powi.*``' intrinsics return the first operand raised to the
10641 specified (positive or negative) power. The order of evaluation of
10642 multiplications is not defined. When a vector of floating-point type is
10643 used, the second argument remains a scalar integer value.
10648 The second argument is an integer power, and the first is a value to
10649 raise to that power.
10654 This function returns the first value raised to the second power with an
10655 unspecified sequence of rounding operations.
10657 '``llvm.sin.*``' Intrinsic
10658 ^^^^^^^^^^^^^^^^^^^^^^^^^^
10663 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
10664 floating-point or vector of floating-point type. Not all targets support
10669 declare float @llvm.sin.f32(float %Val)
10670 declare double @llvm.sin.f64(double %Val)
10671 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
10672 declare fp128 @llvm.sin.f128(fp128 %Val)
10673 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
10678 The '``llvm.sin.*``' intrinsics return the sine of the operand.
10683 The argument and return value are floating-point numbers of the same type.
10688 Return the same value as a corresponding libm '``sin``' function but without
10689 trapping or setting ``errno``.
10691 When specified with the fast-math-flag 'afn', the result may be approximated
10692 using a less accurate calculation.
10694 '``llvm.cos.*``' Intrinsic
10695 ^^^^^^^^^^^^^^^^^^^^^^^^^^
10700 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
10701 floating-point or vector of floating-point type. Not all targets support
10706 declare float @llvm.cos.f32(float %Val)
10707 declare double @llvm.cos.f64(double %Val)
10708 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
10709 declare fp128 @llvm.cos.f128(fp128 %Val)
10710 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
10715 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
10720 The argument and return value are floating-point numbers of the same type.
10725 Return the same value as a corresponding libm '``cos``' function but without
10726 trapping or setting ``errno``.
10728 When specified with the fast-math-flag 'afn', the result may be approximated
10729 using a less accurate calculation.
10731 '``llvm.pow.*``' Intrinsic
10732 ^^^^^^^^^^^^^^^^^^^^^^^^^^
10737 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
10738 floating-point or vector of floating-point type. Not all targets support
10743 declare float @llvm.pow.f32(float %Val, float %Power)
10744 declare double @llvm.pow.f64(double %Val, double %Power)
10745 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
10746 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
10747 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
10752 The '``llvm.pow.*``' intrinsics return the first operand raised to the
10753 specified (positive or negative) power.
10758 The arguments and return value are floating-point numbers of the same type.
10763 Return the same value as a corresponding libm '``pow``' function but without
10764 trapping or setting ``errno``.
10766 When specified with the fast-math-flag 'afn', the result may be approximated
10767 using a less accurate calculation.
10769 '``llvm.exp.*``' Intrinsic
10770 ^^^^^^^^^^^^^^^^^^^^^^^^^^
10775 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
10776 floating-point or vector of floating-point type. Not all targets support
10781 declare float @llvm.exp.f32(float %Val)
10782 declare double @llvm.exp.f64(double %Val)
10783 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
10784 declare fp128 @llvm.exp.f128(fp128 %Val)
10785 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
10790 The '``llvm.exp.*``' intrinsics compute the base-e exponential of the specified
10796 The argument and return value are floating-point numbers of the same type.
10801 Return the same value as a corresponding libm '``exp``' function but without
10802 trapping or setting ``errno``.
10804 When specified with the fast-math-flag 'afn', the result may be approximated
10805 using a less accurate calculation.
10807 '``llvm.exp2.*``' Intrinsic
10808 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10813 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
10814 floating-point or vector of floating-point type. Not all targets support
10819 declare float @llvm.exp2.f32(float %Val)
10820 declare double @llvm.exp2.f64(double %Val)
10821 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
10822 declare fp128 @llvm.exp2.f128(fp128 %Val)
10823 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
10828 The '``llvm.exp2.*``' intrinsics compute the base-2 exponential of the
10834 The argument and return value are floating-point numbers of the same type.
10839 Return the same value as a corresponding libm '``exp2``' function but without
10840 trapping or setting ``errno``.
10842 When specified with the fast-math-flag 'afn', the result may be approximated
10843 using a less accurate calculation.
10845 '``llvm.log.*``' Intrinsic
10846 ^^^^^^^^^^^^^^^^^^^^^^^^^^
10851 This is an overloaded intrinsic. You can use ``llvm.log`` on any
10852 floating-point or vector of floating-point type. Not all targets support
10857 declare float @llvm.log.f32(float %Val)
10858 declare double @llvm.log.f64(double %Val)
10859 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
10860 declare fp128 @llvm.log.f128(fp128 %Val)
10861 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
10866 The '``llvm.log.*``' intrinsics compute the base-e logarithm of the specified
10872 The argument and return value are floating-point numbers of the same type.
10877 Return the same value as a corresponding libm '``log``' function but without
10878 trapping or setting ``errno``.
10880 When specified with the fast-math-flag 'afn', the result may be approximated
10881 using a less accurate calculation.
10883 '``llvm.log10.*``' Intrinsic
10884 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10889 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
10890 floating-point or vector of floating-point type. Not all targets support
10895 declare float @llvm.log10.f32(float %Val)
10896 declare double @llvm.log10.f64(double %Val)
10897 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
10898 declare fp128 @llvm.log10.f128(fp128 %Val)
10899 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
10904 The '``llvm.log10.*``' intrinsics compute the base-10 logarithm of the
10910 The argument and return value are floating-point numbers of the same type.
10915 Return the same value as a corresponding libm '``log10``' function but without
10916 trapping or setting ``errno``.
10918 When specified with the fast-math-flag 'afn', the result may be approximated
10919 using a less accurate calculation.
10921 '``llvm.log2.*``' Intrinsic
10922 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10927 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
10928 floating-point or vector of floating-point type. Not all targets support
10933 declare float @llvm.log2.f32(float %Val)
10934 declare double @llvm.log2.f64(double %Val)
10935 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
10936 declare fp128 @llvm.log2.f128(fp128 %Val)
10937 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
10942 The '``llvm.log2.*``' intrinsics compute the base-2 logarithm of the specified
10948 The argument and return value are floating-point numbers of the same type.
10953 Return the same value as a corresponding libm '``log2``' function but without
10954 trapping or setting ``errno``.
10956 When specified with the fast-math-flag 'afn', the result may be approximated
10957 using a less accurate calculation.
10959 '``llvm.fma.*``' Intrinsic
10960 ^^^^^^^^^^^^^^^^^^^^^^^^^^
10965 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
10966 floating-point or vector of floating-point type. Not all targets support
10971 declare float @llvm.fma.f32(float %a, float %b, float %c)
10972 declare double @llvm.fma.f64(double %a, double %b, double %c)
10973 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
10974 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
10975 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
10980 The '``llvm.fma.*``' intrinsics perform the fused multiply-add operation.
10985 The arguments and return value are floating-point numbers of the same type.
10990 Return the same value as a corresponding libm '``fma``' function but without
10991 trapping or setting ``errno``.
10993 When specified with the fast-math-flag 'afn', the result may be approximated
10994 using a less accurate calculation.
10996 '``llvm.fabs.*``' Intrinsic
10997 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
11002 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
11003 floating-point or vector of floating-point type. Not all targets support
11008 declare float @llvm.fabs.f32(float %Val)
11009 declare double @llvm.fabs.f64(double %Val)
11010 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
11011 declare fp128 @llvm.fabs.f128(fp128 %Val)
11012 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
11017 The '``llvm.fabs.*``' intrinsics return the absolute value of the
11023 The argument and return value are floating-point numbers of the same
11029 This function returns the same values as the libm ``fabs`` functions
11030 would, and handles error conditions in the same way.
11032 '``llvm.minnum.*``' Intrinsic
11033 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11038 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
11039 floating-point or vector of floating-point type. Not all targets support
11044 declare float @llvm.minnum.f32(float %Val0, float %Val1)
11045 declare double @llvm.minnum.f64(double %Val0, double %Val1)
11046 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
11047 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
11048 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
11053 The '``llvm.minnum.*``' intrinsics return the minimum of the two
11060 The arguments and return value are floating-point numbers of the same
11066 Follows the IEEE-754 semantics for minNum, which also match for libm's
11069 If either operand is a NaN, returns the other non-NaN operand. Returns
11070 NaN only if both operands are NaN. If the operands compare equal,
11071 returns a value that compares equal to both operands. This means that
11072 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
11074 '``llvm.maxnum.*``' Intrinsic
11075 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11080 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
11081 floating-point or vector of floating-point type. Not all targets support
11086 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
11087 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
11088 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
11089 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
11090 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
11095 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
11102 The arguments and return value are floating-point numbers of the same
11107 Follows the IEEE-754 semantics for maxNum, which also match for libm's
11110 If either operand is a NaN, returns the other non-NaN operand. Returns
11111 NaN only if both operands are NaN. If the operands compare equal,
11112 returns a value that compares equal to both operands. This means that
11113 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
11115 '``llvm.copysign.*``' Intrinsic
11116 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11121 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
11122 floating-point or vector of floating-point type. Not all targets support
11127 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
11128 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
11129 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
11130 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
11131 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
11136 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
11137 first operand and the sign of the second operand.
11142 The arguments and return value are floating-point numbers of the same
11148 This function returns the same values as the libm ``copysign``
11149 functions would, and handles error conditions in the same way.
11151 '``llvm.floor.*``' Intrinsic
11152 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11157 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
11158 floating-point or vector of floating-point type. Not all targets support
11163 declare float @llvm.floor.f32(float %Val)
11164 declare double @llvm.floor.f64(double %Val)
11165 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
11166 declare fp128 @llvm.floor.f128(fp128 %Val)
11167 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
11172 The '``llvm.floor.*``' intrinsics return the floor of the operand.
11177 The argument and return value are floating-point numbers of the same
11183 This function returns the same values as the libm ``floor`` functions
11184 would, and handles error conditions in the same way.
11186 '``llvm.ceil.*``' Intrinsic
11187 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
11192 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
11193 floating-point or vector of floating-point type. Not all targets support
11198 declare float @llvm.ceil.f32(float %Val)
11199 declare double @llvm.ceil.f64(double %Val)
11200 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
11201 declare fp128 @llvm.ceil.f128(fp128 %Val)
11202 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
11207 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
11212 The argument and return value are floating-point numbers of the same
11218 This function returns the same values as the libm ``ceil`` functions
11219 would, and handles error conditions in the same way.
11221 '``llvm.trunc.*``' Intrinsic
11222 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11227 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
11228 floating-point or vector of floating-point type. Not all targets support
11233 declare float @llvm.trunc.f32(float %Val)
11234 declare double @llvm.trunc.f64(double %Val)
11235 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
11236 declare fp128 @llvm.trunc.f128(fp128 %Val)
11237 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
11242 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
11243 nearest integer not larger in magnitude than the operand.
11248 The argument and return value are floating-point numbers of the same
11254 This function returns the same values as the libm ``trunc`` functions
11255 would, and handles error conditions in the same way.
11257 '``llvm.rint.*``' Intrinsic
11258 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
11263 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
11264 floating-point or vector of floating-point type. Not all targets support
11269 declare float @llvm.rint.f32(float %Val)
11270 declare double @llvm.rint.f64(double %Val)
11271 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
11272 declare fp128 @llvm.rint.f128(fp128 %Val)
11273 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
11278 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
11279 nearest integer. It may raise an inexact floating-point exception if the
11280 operand isn't an integer.
11285 The argument and return value are floating-point numbers of the same
11291 This function returns the same values as the libm ``rint`` functions
11292 would, and handles error conditions in the same way.
11294 '``llvm.nearbyint.*``' Intrinsic
11295 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11300 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
11301 floating-point or vector of floating-point type. Not all targets support
11306 declare float @llvm.nearbyint.f32(float %Val)
11307 declare double @llvm.nearbyint.f64(double %Val)
11308 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
11309 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
11310 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
11315 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
11321 The argument and return value are floating-point numbers of the same
11327 This function returns the same values as the libm ``nearbyint``
11328 functions would, and handles error conditions in the same way.
11330 '``llvm.round.*``' Intrinsic
11331 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11336 This is an overloaded intrinsic. You can use ``llvm.round`` on any
11337 floating-point or vector of floating-point type. Not all targets support
11342 declare float @llvm.round.f32(float %Val)
11343 declare double @llvm.round.f64(double %Val)
11344 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
11345 declare fp128 @llvm.round.f128(fp128 %Val)
11346 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
11351 The '``llvm.round.*``' intrinsics returns the operand rounded to the
11357 The argument and return value are floating-point numbers of the same
11363 This function returns the same values as the libm ``round``
11364 functions would, and handles error conditions in the same way.
11366 Bit Manipulation Intrinsics
11367 ---------------------------
11369 LLVM provides intrinsics for a few important bit manipulation
11370 operations. These allow efficient code generation for some algorithms.
11372 '``llvm.bitreverse.*``' Intrinsics
11373 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11378 This is an overloaded intrinsic function. You can use bitreverse on any
11383 declare i16 @llvm.bitreverse.i16(i16 <id>)
11384 declare i32 @llvm.bitreverse.i32(i32 <id>)
11385 declare i64 @llvm.bitreverse.i64(i64 <id>)
11390 The '``llvm.bitreverse``' family of intrinsics is used to reverse the
11391 bitpattern of an integer value; for example ``0b10110110`` becomes
11397 The ``llvm.bitreverse.iN`` intrinsic returns an iN value that has bit
11398 ``M`` in the input moved to bit ``N-M`` in the output.
11400 '``llvm.bswap.*``' Intrinsics
11401 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11406 This is an overloaded intrinsic function. You can use bswap on any
11407 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
11411 declare i16 @llvm.bswap.i16(i16 <id>)
11412 declare i32 @llvm.bswap.i32(i32 <id>)
11413 declare i64 @llvm.bswap.i64(i64 <id>)
11418 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
11419 values with an even number of bytes (positive multiple of 16 bits).
11420 These are useful for performing operations on data that is not in the
11421 target's native byte order.
11426 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
11427 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
11428 intrinsic returns an i32 value that has the four bytes of the input i32
11429 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
11430 returned i32 will have its bytes in 3, 2, 1, 0 order. The
11431 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
11432 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
11435 '``llvm.ctpop.*``' Intrinsic
11436 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11441 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
11442 bit width, or on any vector with integer elements. Not all targets
11443 support all bit widths or vector types, however.
11447 declare i8 @llvm.ctpop.i8(i8 <src>)
11448 declare i16 @llvm.ctpop.i16(i16 <src>)
11449 declare i32 @llvm.ctpop.i32(i32 <src>)
11450 declare i64 @llvm.ctpop.i64(i64 <src>)
11451 declare i256 @llvm.ctpop.i256(i256 <src>)
11452 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
11457 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
11463 The only argument is the value to be counted. The argument may be of any
11464 integer type, or a vector with integer elements. The return type must
11465 match the argument type.
11470 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
11471 each element of a vector.
11473 '``llvm.ctlz.*``' Intrinsic
11474 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
11479 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
11480 integer bit width, or any vector whose elements are integers. Not all
11481 targets support all bit widths or vector types, however.
11485 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
11486 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
11487 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
11488 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
11489 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
11490 declare <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
11495 The '``llvm.ctlz``' family of intrinsic functions counts the number of
11496 leading zeros in a variable.
11501 The first argument is the value to be counted. This argument may be of
11502 any integer type, or a vector with integer element type. The return
11503 type must match the first argument type.
11505 The second argument must be a constant and is a flag to indicate whether
11506 the intrinsic should ensure that a zero as the first argument produces a
11507 defined result. Historically some architectures did not provide a
11508 defined result for zero values as efficiently, and many algorithms are
11509 now predicated on avoiding zero-value inputs.
11514 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
11515 zeros in a variable, or within each element of the vector. If
11516 ``src == 0`` then the result is the size in bits of the type of ``src``
11517 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
11518 ``llvm.ctlz(i32 2) = 30``.
11520 '``llvm.cttz.*``' Intrinsic
11521 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
11526 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
11527 integer bit width, or any vector of integer elements. Not all targets
11528 support all bit widths or vector types, however.
11532 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
11533 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
11534 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
11535 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
11536 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
11537 declare <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
11542 The '``llvm.cttz``' family of intrinsic functions counts the number of
11548 The first argument is the value to be counted. This argument may be of
11549 any integer type, or a vector with integer element type. The return
11550 type must match the first argument type.
11552 The second argument must be a constant and is a flag to indicate whether
11553 the intrinsic should ensure that a zero as the first argument produces a
11554 defined result. Historically some architectures did not provide a
11555 defined result for zero values as efficiently, and many algorithms are
11556 now predicated on avoiding zero-value inputs.
11561 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
11562 zeros in a variable, or within each element of a vector. If ``src == 0``
11563 then the result is the size in bits of the type of ``src`` if
11564 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
11565 ``llvm.cttz(2) = 1``.
11569 Arithmetic with Overflow Intrinsics
11570 -----------------------------------
11572 LLVM provides intrinsics for fast arithmetic overflow checking.
11574 Each of these intrinsics returns a two-element struct. The first
11575 element of this struct contains the result of the corresponding
11576 arithmetic operation modulo 2\ :sup:`n`\ , where n is the bit width of
11577 the result. Therefore, for example, the first element of the struct
11578 returned by ``llvm.sadd.with.overflow.i32`` is always the same as the
11579 result of a 32-bit ``add`` instruction with the same operands, where
11580 the ``add`` is *not* modified by an ``nsw`` or ``nuw`` flag.
11582 The second element of the result is an ``i1`` that is 1 if the
11583 arithmetic operation overflowed and 0 otherwise. An operation
11584 overflows if, for any values of its operands ``A`` and ``B`` and for
11585 any ``N`` larger than the operands' width, ``ext(A op B) to iN`` is
11586 not equal to ``(ext(A) to iN) op (ext(B) to iN)`` where ``ext`` is
11587 ``sext`` for signed overflow and ``zext`` for unsigned overflow, and
11588 ``op`` is the underlying arithmetic operation.
11590 The behavior of these intrinsics is well-defined for all argument
11593 '``llvm.sadd.with.overflow.*``' Intrinsics
11594 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11599 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
11600 on any integer bit width.
11604 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
11605 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
11606 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
11611 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
11612 a signed addition of the two arguments, and indicate whether an overflow
11613 occurred during the signed summation.
11618 The arguments (%a and %b) and the first element of the result structure
11619 may be of integer types of any bit width, but they must have the same
11620 bit width. The second element of the result structure must be of type
11621 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
11627 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
11628 a signed addition of the two variables. They return a structure --- the
11629 first element of which is the signed summation, and the second element
11630 of which is a bit specifying if the signed summation resulted in an
11636 .. code-block:: llvm
11638 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
11639 %sum = extractvalue {i32, i1} %res, 0
11640 %obit = extractvalue {i32, i1} %res, 1
11641 br i1 %obit, label %overflow, label %normal
11643 '``llvm.uadd.with.overflow.*``' Intrinsics
11644 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11649 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
11650 on any integer bit width.
11654 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
11655 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
11656 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
11661 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
11662 an unsigned addition of the two arguments, and indicate whether a carry
11663 occurred during the unsigned summation.
11668 The arguments (%a and %b) and the first element of the result structure
11669 may be of integer types of any bit width, but they must have the same
11670 bit width. The second element of the result structure must be of type
11671 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
11677 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
11678 an unsigned addition of the two arguments. They return a structure --- the
11679 first element of which is the sum, and the second element of which is a
11680 bit specifying if the unsigned summation resulted in a carry.
11685 .. code-block:: llvm
11687 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
11688 %sum = extractvalue {i32, i1} %res, 0
11689 %obit = extractvalue {i32, i1} %res, 1
11690 br i1 %obit, label %carry, label %normal
11692 '``llvm.ssub.with.overflow.*``' Intrinsics
11693 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11698 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
11699 on any integer bit width.
11703 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
11704 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
11705 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
11710 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
11711 a signed subtraction of the two arguments, and indicate whether an
11712 overflow occurred during the signed subtraction.
11717 The arguments (%a and %b) and the first element of the result structure
11718 may be of integer types of any bit width, but they must have the same
11719 bit width. The second element of the result structure must be of type
11720 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
11726 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
11727 a signed subtraction of the two arguments. They return a structure --- the
11728 first element of which is the subtraction, and the second element of
11729 which is a bit specifying if the signed subtraction resulted in an
11735 .. code-block:: llvm
11737 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
11738 %sum = extractvalue {i32, i1} %res, 0
11739 %obit = extractvalue {i32, i1} %res, 1
11740 br i1 %obit, label %overflow, label %normal
11742 '``llvm.usub.with.overflow.*``' Intrinsics
11743 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11748 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
11749 on any integer bit width.
11753 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
11754 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
11755 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
11760 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
11761 an unsigned subtraction of the two arguments, and indicate whether an
11762 overflow occurred during the unsigned subtraction.
11767 The arguments (%a and %b) and the first element of the result structure
11768 may be of integer types of any bit width, but they must have the same
11769 bit width. The second element of the result structure must be of type
11770 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
11776 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
11777 an unsigned subtraction of the two arguments. They return a structure ---
11778 the first element of which is the subtraction, and the second element of
11779 which is a bit specifying if the unsigned subtraction resulted in an
11785 .. code-block:: llvm
11787 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
11788 %sum = extractvalue {i32, i1} %res, 0
11789 %obit = extractvalue {i32, i1} %res, 1
11790 br i1 %obit, label %overflow, label %normal
11792 '``llvm.smul.with.overflow.*``' Intrinsics
11793 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11798 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
11799 on any integer bit width.
11803 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
11804 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
11805 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
11810 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
11811 a signed multiplication of the two arguments, and indicate whether an
11812 overflow occurred during the signed multiplication.
11817 The arguments (%a and %b) and the first element of the result structure
11818 may be of integer types of any bit width, but they must have the same
11819 bit width. The second element of the result structure must be of type
11820 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
11826 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
11827 a signed multiplication of the two arguments. They return a structure ---
11828 the first element of which is the multiplication, and the second element
11829 of which is a bit specifying if the signed multiplication resulted in an
11835 .. code-block:: llvm
11837 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
11838 %sum = extractvalue {i32, i1} %res, 0
11839 %obit = extractvalue {i32, i1} %res, 1
11840 br i1 %obit, label %overflow, label %normal
11842 '``llvm.umul.with.overflow.*``' Intrinsics
11843 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11848 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
11849 on any integer bit width.
11853 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
11854 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
11855 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
11860 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
11861 a unsigned multiplication of the two arguments, and indicate whether an
11862 overflow occurred during the unsigned multiplication.
11867 The arguments (%a and %b) and the first element of the result structure
11868 may be of integer types of any bit width, but they must have the same
11869 bit width. The second element of the result structure must be of type
11870 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
11876 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
11877 an unsigned multiplication of the two arguments. They return a structure ---
11878 the first element of which is the multiplication, and the second
11879 element of which is a bit specifying if the unsigned multiplication
11880 resulted in an overflow.
11885 .. code-block:: llvm
11887 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
11888 %sum = extractvalue {i32, i1} %res, 0
11889 %obit = extractvalue {i32, i1} %res, 1
11890 br i1 %obit, label %overflow, label %normal
11892 Specialised Arithmetic Intrinsics
11893 ---------------------------------
11895 '``llvm.canonicalize.*``' Intrinsic
11896 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11903 declare float @llvm.canonicalize.f32(float %a)
11904 declare double @llvm.canonicalize.f64(double %b)
11909 The '``llvm.canonicalize.*``' intrinsic returns the platform specific canonical
11910 encoding of a floating-point number. This canonicalization is useful for
11911 implementing certain numeric primitives such as frexp. The canonical encoding is
11912 defined by IEEE-754-2008 to be:
11916 2.1.8 canonical encoding: The preferred encoding of a floating-point
11917 representation in a format. Applied to declets, significands of finite
11918 numbers, infinities, and NaNs, especially in decimal formats.
11920 This operation can also be considered equivalent to the IEEE-754-2008
11921 conversion of a floating-point value to the same format. NaNs are handled
11922 according to section 6.2.
11924 Examples of non-canonical encodings:
11926 - x87 pseudo denormals, pseudo NaNs, pseudo Infinity, Unnormals. These are
11927 converted to a canonical representation per hardware-specific protocol.
11928 - Many normal decimal floating-point numbers have non-canonical alternative
11930 - Some machines, like GPUs or ARMv7 NEON, do not support subnormal values.
11931 These are treated as non-canonical encodings of zero and will be flushed to
11932 a zero of the same sign by this operation.
11934 Note that per IEEE-754-2008 6.2, systems that support signaling NaNs with
11935 default exception handling must signal an invalid exception, and produce a
11938 This function should always be implementable as multiplication by 1.0, provided
11939 that the compiler does not constant fold the operation. Likewise, division by
11940 1.0 and ``llvm.minnum(x, x)`` are possible implementations. Addition with
11941 -0.0 is also sufficient provided that the rounding mode is not -Infinity.
11943 ``@llvm.canonicalize`` must preserve the equality relation. That is:
11945 - ``(@llvm.canonicalize(x) == x)`` is equivalent to ``(x == x)``
11946 - ``(@llvm.canonicalize(x) == @llvm.canonicalize(y))`` is equivalent to
11949 Additionally, the sign of zero must be conserved:
11950 ``@llvm.canonicalize(-0.0) = -0.0`` and ``@llvm.canonicalize(+0.0) = +0.0``
11952 The payload bits of a NaN must be conserved, with two exceptions.
11953 First, environments which use only a single canonical representation of NaN
11954 must perform said canonicalization. Second, SNaNs must be quieted per the
11957 The canonicalization operation may be optimized away if:
11959 - The input is known to be canonical. For example, it was produced by a
11960 floating-point operation that is required by the standard to be canonical.
11961 - The result is consumed only by (or fused with) other floating-point
11962 operations. That is, the bits of the floating-point value are not examined.
11964 '``llvm.fmuladd.*``' Intrinsic
11965 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11972 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
11973 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
11978 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
11979 expressions that can be fused if the code generator determines that (a) the
11980 target instruction set has support for a fused operation, and (b) that the
11981 fused operation is more efficient than the equivalent, separate pair of mul
11982 and add instructions.
11987 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
11988 multiplicands, a and b, and an addend c.
11997 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
11999 is equivalent to the expression a \* b + c, except that rounding will
12000 not be performed between the multiplication and addition steps if the
12001 code generator fuses the operations. Fusion is not guaranteed, even if
12002 the target platform supports it. If a fused multiply-add is required the
12003 corresponding llvm.fma.\* intrinsic function should be used
12004 instead. This never sets errno, just as '``llvm.fma.*``'.
12009 .. code-block:: llvm
12011 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
12014 Experimental Vector Reduction Intrinsics
12015 ----------------------------------------
12017 Horizontal reductions of vectors can be expressed using the following
12018 intrinsics. Each one takes a vector operand as an input and applies its
12019 respective operation across all elements of the vector, returning a single
12020 scalar result of the same element type.
12023 '``llvm.experimental.vector.reduce.add.*``' Intrinsic
12024 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12031 declare i32 @llvm.experimental.vector.reduce.add.i32.v4i32(<4 x i32> %a)
12032 declare i64 @llvm.experimental.vector.reduce.add.i64.v2i64(<2 x i64> %a)
12037 The '``llvm.experimental.vector.reduce.add.*``' intrinsics do an integer ``ADD``
12038 reduction of a vector, returning the result as a scalar. The return type matches
12039 the element-type of the vector input.
12043 The argument to this intrinsic must be a vector of integer values.
12045 '``llvm.experimental.vector.reduce.fadd.*``' Intrinsic
12046 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12053 declare float @llvm.experimental.vector.reduce.fadd.f32.v4f32(float %acc, <4 x float> %a)
12054 declare double @llvm.experimental.vector.reduce.fadd.f64.v2f64(double %acc, <2 x double> %a)
12059 The '``llvm.experimental.vector.reduce.fadd.*``' intrinsics do a floating-point
12060 ``ADD`` reduction of a vector, returning the result as a scalar. The return type
12061 matches the element-type of the vector input.
12063 If the intrinsic call has fast-math flags, then the reduction will not preserve
12064 the associativity of an equivalent scalarized counterpart. If it does not have
12065 fast-math flags, then the reduction will be *ordered*, implying that the
12066 operation respects the associativity of a scalarized reduction.
12071 The first argument to this intrinsic is a scalar accumulator value, which is
12072 only used when there are no fast-math flags attached. This argument may be undef
12073 when fast-math flags are used.
12075 The second argument must be a vector of floating-point values.
12080 .. code-block:: llvm
12082 %fast = call fast float @llvm.experimental.vector.reduce.fadd.f32.v4f32(float undef, <4 x float> %input) ; fast reduction
12083 %ord = call float @llvm.experimental.vector.reduce.fadd.f32.v4f32(float %acc, <4 x float> %input) ; ordered reduction
12086 '``llvm.experimental.vector.reduce.mul.*``' Intrinsic
12087 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12094 declare i32 @llvm.experimental.vector.reduce.mul.i32.v4i32(<4 x i32> %a)
12095 declare i64 @llvm.experimental.vector.reduce.mul.i64.v2i64(<2 x i64> %a)
12100 The '``llvm.experimental.vector.reduce.mul.*``' intrinsics do an integer ``MUL``
12101 reduction of a vector, returning the result as a scalar. The return type matches
12102 the element-type of the vector input.
12106 The argument to this intrinsic must be a vector of integer values.
12108 '``llvm.experimental.vector.reduce.fmul.*``' Intrinsic
12109 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12116 declare float @llvm.experimental.vector.reduce.fmul.f32.v4f32(float %acc, <4 x float> %a)
12117 declare double @llvm.experimental.vector.reduce.fmul.f64.v2f64(double %acc, <2 x double> %a)
12122 The '``llvm.experimental.vector.reduce.fmul.*``' intrinsics do a floating-point
12123 ``MUL`` reduction of a vector, returning the result as a scalar. The return type
12124 matches the element-type of the vector input.
12126 If the intrinsic call has fast-math flags, then the reduction will not preserve
12127 the associativity of an equivalent scalarized counterpart. If it does not have
12128 fast-math flags, then the reduction will be *ordered*, implying that the
12129 operation respects the associativity of a scalarized reduction.
12134 The first argument to this intrinsic is a scalar accumulator value, which is
12135 only used when there are no fast-math flags attached. This argument may be undef
12136 when fast-math flags are used.
12138 The second argument must be a vector of floating-point values.
12143 .. code-block:: llvm
12145 %fast = call fast float @llvm.experimental.vector.reduce.fmul.f32.v4f32(float undef, <4 x float> %input) ; fast reduction
12146 %ord = call float @llvm.experimental.vector.reduce.fmul.f32.v4f32(float %acc, <4 x float> %input) ; ordered reduction
12148 '``llvm.experimental.vector.reduce.and.*``' Intrinsic
12149 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12156 declare i32 @llvm.experimental.vector.reduce.and.i32.v4i32(<4 x i32> %a)
12161 The '``llvm.experimental.vector.reduce.and.*``' intrinsics do a bitwise ``AND``
12162 reduction of a vector, returning the result as a scalar. The return type matches
12163 the element-type of the vector input.
12167 The argument to this intrinsic must be a vector of integer values.
12169 '``llvm.experimental.vector.reduce.or.*``' Intrinsic
12170 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12177 declare i32 @llvm.experimental.vector.reduce.or.i32.v4i32(<4 x i32> %a)
12182 The '``llvm.experimental.vector.reduce.or.*``' intrinsics do a bitwise ``OR`` reduction
12183 of a vector, returning the result as a scalar. The return type matches the
12184 element-type of the vector input.
12188 The argument to this intrinsic must be a vector of integer values.
12190 '``llvm.experimental.vector.reduce.xor.*``' Intrinsic
12191 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12198 declare i32 @llvm.experimental.vector.reduce.xor.i32.v4i32(<4 x i32> %a)
12203 The '``llvm.experimental.vector.reduce.xor.*``' intrinsics do a bitwise ``XOR``
12204 reduction of a vector, returning the result as a scalar. The return type matches
12205 the element-type of the vector input.
12209 The argument to this intrinsic must be a vector of integer values.
12211 '``llvm.experimental.vector.reduce.smax.*``' Intrinsic
12212 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12219 declare i32 @llvm.experimental.vector.reduce.smax.i32.v4i32(<4 x i32> %a)
12224 The '``llvm.experimental.vector.reduce.smax.*``' intrinsics do a signed integer
12225 ``MAX`` reduction of a vector, returning the result as a scalar. The return type
12226 matches the element-type of the vector input.
12230 The argument to this intrinsic must be a vector of integer values.
12232 '``llvm.experimental.vector.reduce.smin.*``' Intrinsic
12233 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12240 declare i32 @llvm.experimental.vector.reduce.smin.i32.v4i32(<4 x i32> %a)
12245 The '``llvm.experimental.vector.reduce.smin.*``' intrinsics do a signed integer
12246 ``MIN`` reduction of a vector, returning the result as a scalar. The return type
12247 matches the element-type of the vector input.
12251 The argument to this intrinsic must be a vector of integer values.
12253 '``llvm.experimental.vector.reduce.umax.*``' Intrinsic
12254 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12261 declare i32 @llvm.experimental.vector.reduce.umax.i32.v4i32(<4 x i32> %a)
12266 The '``llvm.experimental.vector.reduce.umax.*``' intrinsics do an unsigned
12267 integer ``MAX`` reduction of a vector, returning the result as a scalar. The
12268 return type matches the element-type of the vector input.
12272 The argument to this intrinsic must be a vector of integer values.
12274 '``llvm.experimental.vector.reduce.umin.*``' Intrinsic
12275 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12282 declare i32 @llvm.experimental.vector.reduce.umin.i32.v4i32(<4 x i32> %a)
12287 The '``llvm.experimental.vector.reduce.umin.*``' intrinsics do an unsigned
12288 integer ``MIN`` reduction of a vector, returning the result as a scalar. The
12289 return type matches the element-type of the vector input.
12293 The argument to this intrinsic must be a vector of integer values.
12295 '``llvm.experimental.vector.reduce.fmax.*``' Intrinsic
12296 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12303 declare float @llvm.experimental.vector.reduce.fmax.f32.v4f32(<4 x float> %a)
12304 declare double @llvm.experimental.vector.reduce.fmax.f64.v2f64(<2 x double> %a)
12309 The '``llvm.experimental.vector.reduce.fmax.*``' intrinsics do a floating-point
12310 ``MAX`` reduction of a vector, returning the result as a scalar. The return type
12311 matches the element-type of the vector input.
12313 If the intrinsic call has the ``nnan`` fast-math flag then the operation can
12314 assume that NaNs are not present in the input vector.
12318 The argument to this intrinsic must be a vector of floating-point values.
12320 '``llvm.experimental.vector.reduce.fmin.*``' Intrinsic
12321 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12328 declare float @llvm.experimental.vector.reduce.fmin.f32.v4f32(<4 x float> %a)
12329 declare double @llvm.experimental.vector.reduce.fmin.f64.v2f64(<2 x double> %a)
12334 The '``llvm.experimental.vector.reduce.fmin.*``' intrinsics do a floating-point
12335 ``MIN`` reduction of a vector, returning the result as a scalar. The return type
12336 matches the element-type of the vector input.
12338 If the intrinsic call has the ``nnan`` fast-math flag then the operation can
12339 assume that NaNs are not present in the input vector.
12343 The argument to this intrinsic must be a vector of floating-point values.
12345 Half Precision Floating-Point Intrinsics
12346 ----------------------------------------
12348 For most target platforms, half precision floating-point is a
12349 storage-only format. This means that it is a dense encoding (in memory)
12350 but does not support computation in the format.
12352 This means that code must first load the half-precision floating-point
12353 value as an i16, then convert it to float with
12354 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
12355 then be performed on the float value (including extending to double
12356 etc). To store the value back to memory, it is first converted to float
12357 if needed, then converted to i16 with
12358 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
12361 .. _int_convert_to_fp16:
12363 '``llvm.convert.to.fp16``' Intrinsic
12364 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12371 declare i16 @llvm.convert.to.fp16.f32(float %a)
12372 declare i16 @llvm.convert.to.fp16.f64(double %a)
12377 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
12378 conventional floating-point type to half precision floating-point format.
12383 The intrinsic function contains single argument - the value to be
12389 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
12390 conventional floating-point format to half precision floating-point format. The
12391 return value is an ``i16`` which contains the converted number.
12396 .. code-block:: llvm
12398 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
12399 store i16 %res, i16* @x, align 2
12401 .. _int_convert_from_fp16:
12403 '``llvm.convert.from.fp16``' Intrinsic
12404 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12411 declare float @llvm.convert.from.fp16.f32(i16 %a)
12412 declare double @llvm.convert.from.fp16.f64(i16 %a)
12417 The '``llvm.convert.from.fp16``' intrinsic function performs a
12418 conversion from half precision floating-point format to single precision
12419 floating-point format.
12424 The intrinsic function contains single argument - the value to be
12430 The '``llvm.convert.from.fp16``' intrinsic function performs a
12431 conversion from half single precision floating-point format to single
12432 precision floating-point format. The input half-float value is
12433 represented by an ``i16`` value.
12438 .. code-block:: llvm
12440 %a = load i16, i16* @x, align 2
12441 %res = call float @llvm.convert.from.fp16(i16 %a)
12443 .. _dbg_intrinsics:
12445 Debugger Intrinsics
12446 -------------------
12448 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
12449 prefix), are described in the `LLVM Source Level
12450 Debugging <SourceLevelDebugging.html#format-common-intrinsics>`_
12453 Exception Handling Intrinsics
12454 -----------------------------
12456 The LLVM exception handling intrinsics (which all start with
12457 ``llvm.eh.`` prefix), are described in the `LLVM Exception
12458 Handling <ExceptionHandling.html#format-common-intrinsics>`_ document.
12460 .. _int_trampoline:
12462 Trampoline Intrinsics
12463 ---------------------
12465 These intrinsics make it possible to excise one parameter, marked with
12466 the :ref:`nest <nest>` attribute, from a function. The result is a
12467 callable function pointer lacking the nest parameter - the caller does
12468 not need to provide a value for it. Instead, the value to use is stored
12469 in advance in a "trampoline", a block of memory usually allocated on the
12470 stack, which also contains code to splice the nest value into the
12471 argument list. This is used to implement the GCC nested function address
12474 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
12475 then the resulting function pointer has signature ``i32 (i32, i32)*``.
12476 It can be created as follows:
12478 .. code-block:: llvm
12480 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
12481 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
12482 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
12483 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
12484 %fp = bitcast i8* %p to i32 (i32, i32)*
12486 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
12487 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
12491 '``llvm.init.trampoline``' Intrinsic
12492 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12499 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
12504 This fills the memory pointed to by ``tramp`` with executable code,
12505 turning it into a trampoline.
12510 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
12511 pointers. The ``tramp`` argument must point to a sufficiently large and
12512 sufficiently aligned block of memory; this memory is written to by the
12513 intrinsic. Note that the size and the alignment are target-specific -
12514 LLVM currently provides no portable way of determining them, so a
12515 front-end that generates this intrinsic needs to have some
12516 target-specific knowledge. The ``func`` argument must hold a function
12517 bitcast to an ``i8*``.
12522 The block of memory pointed to by ``tramp`` is filled with target
12523 dependent code, turning it into a function. Then ``tramp`` needs to be
12524 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
12525 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
12526 function's signature is the same as that of ``func`` with any arguments
12527 marked with the ``nest`` attribute removed. At most one such ``nest``
12528 argument is allowed, and it must be of pointer type. Calling the new
12529 function is equivalent to calling ``func`` with the same argument list,
12530 but with ``nval`` used for the missing ``nest`` argument. If, after
12531 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
12532 modified, then the effect of any later call to the returned function
12533 pointer is undefined.
12537 '``llvm.adjust.trampoline``' Intrinsic
12538 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12545 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
12550 This performs any required machine-specific adjustment to the address of
12551 a trampoline (passed as ``tramp``).
12556 ``tramp`` must point to a block of memory which already has trampoline
12557 code filled in by a previous call to
12558 :ref:`llvm.init.trampoline <int_it>`.
12563 On some architectures the address of the code to be executed needs to be
12564 different than the address where the trampoline is actually stored. This
12565 intrinsic returns the executable address corresponding to ``tramp``
12566 after performing the required machine specific adjustments. The pointer
12567 returned can then be :ref:`bitcast and executed <int_trampoline>`.
12569 .. _int_mload_mstore:
12571 Masked Vector Load and Store Intrinsics
12572 ---------------------------------------
12574 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.
12578 '``llvm.masked.load.*``' Intrinsics
12579 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12583 This is an overloaded intrinsic. The loaded data is a vector of any integer, floating-point or pointer data type.
12587 declare <16 x float> @llvm.masked.load.v16f32.p0v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
12588 declare <2 x double> @llvm.masked.load.v2f64.p0v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
12589 ;; The data is a vector of pointers to double
12590 declare <8 x double*> @llvm.masked.load.v8p0f64.p0v8p0f64 (<8 x double*>* <ptr>, i32 <alignment>, <8 x i1> <mask>, <8 x double*> <passthru>)
12591 ;; The data is a vector of function pointers
12592 declare <8 x i32 ()*> @llvm.masked.load.v8p0f_i32f.p0v8p0f_i32f (<8 x i32 ()*>* <ptr>, i32 <alignment>, <8 x i1> <mask>, <8 x i32 ()*> <passthru>)
12597 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.
12603 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.
12609 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.
12610 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.
12615 %res = call <16 x float> @llvm.masked.load.v16f32.p0v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
12617 ;; The result of the two following instructions is identical aside from potential memory access exception
12618 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
12619 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
12623 '``llvm.masked.store.*``' Intrinsics
12624 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12628 This is an overloaded intrinsic. The data stored in memory is a vector of any integer, floating-point or pointer data type.
12632 declare void @llvm.masked.store.v8i32.p0v8i32 (<8 x i32> <value>, <8 x i32>* <ptr>, i32 <alignment>, <8 x i1> <mask>)
12633 declare void @llvm.masked.store.v16f32.p0v16f32 (<16 x float> <value>, <16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
12634 ;; The data is a vector of pointers to double
12635 declare void @llvm.masked.store.v8p0f64.p0v8p0f64 (<8 x double*> <value>, <8 x double*>* <ptr>, i32 <alignment>, <8 x i1> <mask>)
12636 ;; The data is a vector of function pointers
12637 declare void @llvm.masked.store.v4p0f_i32f.p0v4p0f_i32f (<4 x i32 ()*> <value>, <4 x i32 ()*>* <ptr>, i32 <alignment>, <4 x i1> <mask>)
12642 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.
12647 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.
12653 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.
12654 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.
12658 call void @llvm.masked.store.v16f32.p0v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
12660 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
12661 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
12662 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
12663 store <16 x float> %res, <16 x float>* %ptr, align 4
12666 Masked Vector Gather and Scatter Intrinsics
12667 -------------------------------------------
12669 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.
12673 '``llvm.masked.gather.*``' Intrinsics
12674 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12678 This is an overloaded intrinsic. The loaded data are multiple scalar values of any integer, floating-point or pointer data type gathered together into one vector.
12682 declare <16 x float> @llvm.masked.gather.v16f32.v16p0f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
12683 declare <2 x double> @llvm.masked.gather.v2f64.v2p1f64 (<2 x double addrspace(1)*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
12684 declare <8 x float*> @llvm.masked.gather.v8p0f32.v8p0p0f32 (<8 x float**> <ptrs>, i32 <alignment>, <8 x i1> <mask>, <8 x float*> <passthru>)
12689 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.
12695 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.
12701 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.
12702 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.
12707 %res = call <4 x double> @llvm.masked.gather.v4f64.v4p0f64 (<4 x double*> %ptrs, i32 8, <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x double> undef)
12709 ;; The gather with all-true mask is equivalent to the following instruction sequence
12710 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
12711 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
12712 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
12713 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
12715 %val0 = load double, double* %ptr0, align 8
12716 %val1 = load double, double* %ptr1, align 8
12717 %val2 = load double, double* %ptr2, align 8
12718 %val3 = load double, double* %ptr3, align 8
12720 %vec0 = insertelement <4 x double>undef, %val0, 0
12721 %vec01 = insertelement <4 x double>%vec0, %val1, 1
12722 %vec012 = insertelement <4 x double>%vec01, %val2, 2
12723 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
12727 '``llvm.masked.scatter.*``' Intrinsics
12728 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12732 This is an overloaded intrinsic. The data stored in memory is a vector of any integer, floating-point or pointer data type. Each vector element is stored in an arbitrary memory address. Scatter with overlapping addresses is guaranteed to be ordered from least-significant to most-significant element.
12736 declare void @llvm.masked.scatter.v8i32.v8p0i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
12737 declare void @llvm.masked.scatter.v16f32.v16p1f32 (<16 x float> <value>, <16 x float addrspace(1)*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
12738 declare void @llvm.masked.scatter.v4p0f64.v4p0p0f64 (<4 x double*> <value>, <4 x double**> <ptrs>, i32 <alignment>, <4 x i1> <mask>)
12743 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.
12748 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.
12754 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.
12758 ;; This instruction unconditionally stores data vector in multiple addresses
12759 call @llvm.masked.scatter.v8i32.v8p0i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
12761 ;; It is equivalent to a list of scalar stores
12762 %val0 = extractelement <8 x i32> %value, i32 0
12763 %val1 = extractelement <8 x i32> %value, i32 1
12765 %val7 = extractelement <8 x i32> %value, i32 7
12766 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
12767 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
12769 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
12770 ;; Note: the order of the following stores is important when they overlap:
12771 store i32 %val0, i32* %ptr0, align 4
12772 store i32 %val1, i32* %ptr1, align 4
12774 store i32 %val7, i32* %ptr7, align 4
12780 This class of intrinsics provides information about the lifetime of
12781 memory objects and ranges where variables are immutable.
12785 '``llvm.lifetime.start``' Intrinsic
12786 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12793 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
12798 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
12804 The first argument is a constant integer representing the size of the
12805 object, or -1 if it is variable sized. The second argument is a pointer
12811 This intrinsic indicates that before this point in the code, the value
12812 of the memory pointed to by ``ptr`` is dead. This means that it is known
12813 to never be used and has an undefined value. A load from the pointer
12814 that precedes this intrinsic can be replaced with ``'undef'``.
12818 '``llvm.lifetime.end``' Intrinsic
12819 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12826 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
12831 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
12837 The first argument is a constant integer representing the size of the
12838 object, or -1 if it is variable sized. The second argument is a pointer
12844 This intrinsic indicates that after this point in the code, the value of
12845 the memory pointed to by ``ptr`` is dead. This means that it is known to
12846 never be used and has an undefined value. Any stores into the memory
12847 object following this intrinsic may be removed as dead.
12849 '``llvm.invariant.start``' Intrinsic
12850 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12854 This is an overloaded intrinsic. The memory object can belong to any address space.
12858 declare {}* @llvm.invariant.start.p0i8(i64 <size>, i8* nocapture <ptr>)
12863 The '``llvm.invariant.start``' intrinsic specifies that the contents of
12864 a memory object will not change.
12869 The first argument is a constant integer representing the size of the
12870 object, or -1 if it is variable sized. The second argument is a pointer
12876 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
12877 the return value, the referenced memory location is constant and
12880 '``llvm.invariant.end``' Intrinsic
12881 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12885 This is an overloaded intrinsic. The memory object can belong to any address space.
12889 declare void @llvm.invariant.end.p0i8({}* <start>, i64 <size>, i8* nocapture <ptr>)
12894 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
12895 memory object are mutable.
12900 The first argument is the matching ``llvm.invariant.start`` intrinsic.
12901 The second argument is a constant integer representing the size of the
12902 object, or -1 if it is variable sized and the third argument is a
12903 pointer to the object.
12908 This intrinsic indicates that the memory is mutable again.
12910 '``llvm.launder.invariant.group``' Intrinsic
12911 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12915 This is an overloaded intrinsic. The memory object can belong to any address
12916 space. The returned pointer must belong to the same address space as the
12921 declare i8* @llvm.launder.invariant.group.p0i8(i8* <ptr>)
12926 The '``llvm.launder.invariant.group``' intrinsic can be used when an invariant
12927 established by invariant.group metadata no longer holds, to obtain a new pointer
12928 value that does not carry the invariant information. It is an experimental
12929 intrinsic, which means that its semantics might change in the future.
12935 The ``llvm.launder.invariant.group`` takes only one argument, which is
12936 the pointer to the memory for which the ``invariant.group`` no longer holds.
12941 Returns another pointer that aliases its argument but which is considered different
12942 for the purposes of ``load``/``store`` ``invariant.group`` metadata.
12943 It does not read any accessible memory and the execution can be speculated.
12947 Constrained Floating-Point Intrinsics
12948 -------------------------------------
12950 These intrinsics are used to provide special handling of floating-point
12951 operations when specific rounding mode or floating-point exception behavior is
12952 required. By default, LLVM optimization passes assume that the rounding mode is
12953 round-to-nearest and that floating-point exceptions will not be monitored.
12954 Constrained FP intrinsics are used to support non-default rounding modes and
12955 accurately preserve exception behavior without compromising LLVM's ability to
12956 optimize FP code when the default behavior is used.
12958 Each of these intrinsics corresponds to a normal floating-point operation. The
12959 first two arguments and the return value are the same as the corresponding FP
12962 The third argument is a metadata argument specifying the rounding mode to be
12963 assumed. This argument must be one of the following strings:
12973 If this argument is "round.dynamic" optimization passes must assume that the
12974 rounding mode is unknown and may change at runtime. No transformations that
12975 depend on rounding mode may be performed in this case.
12977 The other possible values for the rounding mode argument correspond to the
12978 similarly named IEEE rounding modes. If the argument is any of these values
12979 optimization passes may perform transformations as long as they are consistent
12980 with the specified rounding mode.
12982 For example, 'x-0'->'x' is not a valid transformation if the rounding mode is
12983 "round.downward" or "round.dynamic" because if the value of 'x' is +0 then
12984 'x-0' should evaluate to '-0' when rounding downward. However, this
12985 transformation is legal for all other rounding modes.
12987 For values other than "round.dynamic" optimization passes may assume that the
12988 actual runtime rounding mode (as defined in a target-specific manner) matches
12989 the specified rounding mode, but this is not guaranteed. Using a specific
12990 non-dynamic rounding mode which does not match the actual rounding mode at
12991 runtime results in undefined behavior.
12993 The fourth argument to the constrained floating-point intrinsics specifies the
12994 required exception behavior. This argument must be one of the following
13003 If this argument is "fpexcept.ignore" optimization passes may assume that the
13004 exception status flags will not be read and that floating-point exceptions will
13005 be masked. This allows transformations to be performed that may change the
13006 exception semantics of the original code. For example, FP operations may be
13007 speculatively executed in this case whereas they must not be for either of the
13008 other possible values of this argument.
13010 If the exception behavior argument is "fpexcept.maytrap" optimization passes
13011 must avoid transformations that may raise exceptions that would not have been
13012 raised by the original code (such as speculatively executing FP operations), but
13013 passes are not required to preserve all exceptions that are implied by the
13014 original code. For example, exceptions may be potentially hidden by constant
13017 If the exception behavior argument is "fpexcept.strict" all transformations must
13018 strictly preserve the floating-point exception semantics of the original code.
13019 Any FP exception that would have been raised by the original code must be raised
13020 by the transformed code, and the transformed code must not raise any FP
13021 exceptions that would not have been raised by the original code. This is the
13022 exception behavior argument that will be used if the code being compiled reads
13023 the FP exception status flags, but this mode can also be used with code that
13024 unmasks FP exceptions.
13026 The number and order of floating-point exceptions is NOT guaranteed. For
13027 example, a series of FP operations that each may raise exceptions may be
13028 vectorized into a single instruction that raises each unique exception a single
13032 '``llvm.experimental.constrained.fadd``' Intrinsic
13033 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13041 @llvm.experimental.constrained.fadd(<type> <op1>, <type> <op2>,
13042 metadata <rounding mode>,
13043 metadata <exception behavior>)
13048 The '``llvm.experimental.constrained.fadd``' intrinsic returns the sum of its
13055 The first two arguments to the '``llvm.experimental.constrained.fadd``'
13056 intrinsic must be :ref:`floating-point <t_floating>` or :ref:`vector <t_vector>`
13057 of floating-point values. Both arguments must have identical types.
13059 The third and fourth arguments specify the rounding mode and exception
13060 behavior as described above.
13065 The value produced is the floating-point sum of the two value operands and has
13066 the same type as the operands.
13069 '``llvm.experimental.constrained.fsub``' Intrinsic
13070 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13078 @llvm.experimental.constrained.fsub(<type> <op1>, <type> <op2>,
13079 metadata <rounding mode>,
13080 metadata <exception behavior>)
13085 The '``llvm.experimental.constrained.fsub``' intrinsic returns the difference
13086 of its two operands.
13092 The first two arguments to the '``llvm.experimental.constrained.fsub``'
13093 intrinsic must be :ref:`floating-point <t_floating>` or :ref:`vector <t_vector>`
13094 of floating-point values. Both arguments must have identical types.
13096 The third and fourth arguments specify the rounding mode and exception
13097 behavior as described above.
13102 The value produced is the floating-point difference of the two value operands
13103 and has the same type as the operands.
13106 '``llvm.experimental.constrained.fmul``' Intrinsic
13107 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13115 @llvm.experimental.constrained.fmul(<type> <op1>, <type> <op2>,
13116 metadata <rounding mode>,
13117 metadata <exception behavior>)
13122 The '``llvm.experimental.constrained.fmul``' intrinsic returns the product of
13129 The first two arguments to the '``llvm.experimental.constrained.fmul``'
13130 intrinsic must be :ref:`floating-point <t_floating>` or :ref:`vector <t_vector>`
13131 of floating-point values. Both arguments must have identical types.
13133 The third and fourth arguments specify the rounding mode and exception
13134 behavior as described above.
13139 The value produced is the floating-point product of the two value operands and
13140 has the same type as the operands.
13143 '``llvm.experimental.constrained.fdiv``' Intrinsic
13144 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13152 @llvm.experimental.constrained.fdiv(<type> <op1>, <type> <op2>,
13153 metadata <rounding mode>,
13154 metadata <exception behavior>)
13159 The '``llvm.experimental.constrained.fdiv``' intrinsic returns the quotient of
13166 The first two arguments to the '``llvm.experimental.constrained.fdiv``'
13167 intrinsic must be :ref:`floating-point <t_floating>` or :ref:`vector <t_vector>`
13168 of floating-point values. Both arguments must have identical types.
13170 The third and fourth arguments specify the rounding mode and exception
13171 behavior as described above.
13176 The value produced is the floating-point quotient of the two value operands and
13177 has the same type as the operands.
13180 '``llvm.experimental.constrained.frem``' Intrinsic
13181 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13189 @llvm.experimental.constrained.frem(<type> <op1>, <type> <op2>,
13190 metadata <rounding mode>,
13191 metadata <exception behavior>)
13196 The '``llvm.experimental.constrained.frem``' intrinsic returns the remainder
13197 from the division of its two operands.
13203 The first two arguments to the '``llvm.experimental.constrained.frem``'
13204 intrinsic must be :ref:`floating-point <t_floating>` or :ref:`vector <t_vector>`
13205 of floating-point values. Both arguments must have identical types.
13207 The third and fourth arguments specify the rounding mode and exception
13208 behavior as described above. The rounding mode argument has no effect, since
13209 the result of frem is never rounded, but the argument is included for
13210 consistency with the other constrained floating-point intrinsics.
13215 The value produced is the floating-point remainder from the division of the two
13216 value operands and has the same type as the operands. The remainder has the
13217 same sign as the dividend.
13219 '``llvm.experimental.constrained.fma``' Intrinsic
13220 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13228 @llvm.experimental.constrained.fma(<type> <op1>, <type> <op2>, <type> <op3>,
13229 metadata <rounding mode>,
13230 metadata <exception behavior>)
13235 The '``llvm.experimental.constrained.fma``' intrinsic returns the result of a
13236 fused-multiply-add operation on its operands.
13241 The first three arguments to the '``llvm.experimental.constrained.fma``'
13242 intrinsic must be :ref:`floating-point <t_floating>` or :ref:`vector
13243 <t_vector>` of floating-point values. All arguments must have identical types.
13245 The fourth and fifth arguments specify the rounding mode and exception behavior
13246 as described above.
13251 The result produced is the product of the first two operands added to the third
13252 operand computed with infinite precision, and then rounded to the target
13255 Constrained libm-equivalent Intrinsics
13256 --------------------------------------
13258 In addition to the basic floating-point operations for which constrained
13259 intrinsics are described above, there are constrained versions of various
13260 operations which provide equivalent behavior to a corresponding libm function.
13261 These intrinsics allow the precise behavior of these operations with respect to
13262 rounding mode and exception behavior to be controlled.
13264 As with the basic constrained floating-point intrinsics, the rounding mode
13265 and exception behavior arguments only control the behavior of the optimizer.
13266 They do not change the runtime floating-point environment.
13269 '``llvm.experimental.constrained.sqrt``' Intrinsic
13270 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13278 @llvm.experimental.constrained.sqrt(<type> <op1>,
13279 metadata <rounding mode>,
13280 metadata <exception behavior>)
13285 The '``llvm.experimental.constrained.sqrt``' intrinsic returns the square root
13286 of the specified value, returning the same value as the libm '``sqrt``'
13287 functions would, but without setting ``errno``.
13292 The first argument and the return type are floating-point numbers of the same
13295 The second and third arguments specify the rounding mode and exception
13296 behavior as described above.
13301 This function returns the nonnegative square root of the specified value.
13302 If the value is less than negative zero, a floating-point exception occurs
13303 and the return value is architecture specific.
13306 '``llvm.experimental.constrained.pow``' Intrinsic
13307 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13315 @llvm.experimental.constrained.pow(<type> <op1>, <type> <op2>,
13316 metadata <rounding mode>,
13317 metadata <exception behavior>)
13322 The '``llvm.experimental.constrained.pow``' intrinsic returns the first operand
13323 raised to the (positive or negative) power specified by the second operand.
13328 The first two arguments and the return value are floating-point numbers of the
13329 same type. The second argument specifies the power to which the first argument
13332 The third and fourth arguments specify the rounding mode and exception
13333 behavior as described above.
13338 This function returns the first value raised to the second power,
13339 returning the same values as the libm ``pow`` functions would, and
13340 handles error conditions in the same way.
13343 '``llvm.experimental.constrained.powi``' Intrinsic
13344 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13352 @llvm.experimental.constrained.powi(<type> <op1>, i32 <op2>,
13353 metadata <rounding mode>,
13354 metadata <exception behavior>)
13359 The '``llvm.experimental.constrained.powi``' intrinsic returns the first operand
13360 raised to the (positive or negative) power specified by the second operand. The
13361 order of evaluation of multiplications is not defined. When a vector of
13362 floating-point type is used, the second argument remains a scalar integer value.
13368 The first argument and the return value are floating-point numbers of the same
13369 type. The second argument is a 32-bit signed integer specifying the power to
13370 which the first argument should be raised.
13372 The third and fourth arguments specify the rounding mode and exception
13373 behavior as described above.
13378 This function returns the first value raised to the second power with an
13379 unspecified sequence of rounding operations.
13382 '``llvm.experimental.constrained.sin``' Intrinsic
13383 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13391 @llvm.experimental.constrained.sin(<type> <op1>,
13392 metadata <rounding mode>,
13393 metadata <exception behavior>)
13398 The '``llvm.experimental.constrained.sin``' intrinsic returns the sine of the
13404 The first argument and the return type are floating-point numbers of the same
13407 The second and third arguments specify the rounding mode and exception
13408 behavior as described above.
13413 This function returns the sine of the specified operand, returning the
13414 same values as the libm ``sin`` functions would, and handles error
13415 conditions in the same way.
13418 '``llvm.experimental.constrained.cos``' Intrinsic
13419 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13427 @llvm.experimental.constrained.cos(<type> <op1>,
13428 metadata <rounding mode>,
13429 metadata <exception behavior>)
13434 The '``llvm.experimental.constrained.cos``' intrinsic returns the cosine of the
13440 The first argument and the return type are floating-point numbers of the same
13443 The second and third arguments specify the rounding mode and exception
13444 behavior as described above.
13449 This function returns the cosine of the specified operand, returning the
13450 same values as the libm ``cos`` functions would, and handles error
13451 conditions in the same way.
13454 '``llvm.experimental.constrained.exp``' Intrinsic
13455 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13463 @llvm.experimental.constrained.exp(<type> <op1>,
13464 metadata <rounding mode>,
13465 metadata <exception behavior>)
13470 The '``llvm.experimental.constrained.exp``' intrinsic computes the base-e
13471 exponential of the specified value.
13476 The first argument and the return value are floating-point numbers of the same
13479 The second and third arguments specify the rounding mode and exception
13480 behavior as described above.
13485 This function returns the same values as the libm ``exp`` functions
13486 would, and handles error conditions in the same way.
13489 '``llvm.experimental.constrained.exp2``' Intrinsic
13490 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13498 @llvm.experimental.constrained.exp2(<type> <op1>,
13499 metadata <rounding mode>,
13500 metadata <exception behavior>)
13505 The '``llvm.experimental.constrained.exp2``' intrinsic computes the base-2
13506 exponential of the specified value.
13512 The first argument and the return value are floating-point numbers of the same
13515 The second and third arguments specify the rounding mode and exception
13516 behavior as described above.
13521 This function returns the same values as the libm ``exp2`` functions
13522 would, and handles error conditions in the same way.
13525 '``llvm.experimental.constrained.log``' Intrinsic
13526 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13534 @llvm.experimental.constrained.log(<type> <op1>,
13535 metadata <rounding mode>,
13536 metadata <exception behavior>)
13541 The '``llvm.experimental.constrained.log``' intrinsic computes the base-e
13542 logarithm of the specified value.
13547 The first argument and the return value are floating-point numbers of the same
13550 The second and third arguments specify the rounding mode and exception
13551 behavior as described above.
13557 This function returns the same values as the libm ``log`` functions
13558 would, and handles error conditions in the same way.
13561 '``llvm.experimental.constrained.log10``' Intrinsic
13562 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13570 @llvm.experimental.constrained.log10(<type> <op1>,
13571 metadata <rounding mode>,
13572 metadata <exception behavior>)
13577 The '``llvm.experimental.constrained.log10``' intrinsic computes the base-10
13578 logarithm of the specified value.
13583 The first argument and the return value are floating-point numbers of the same
13586 The second and third arguments specify the rounding mode and exception
13587 behavior as described above.
13592 This function returns the same values as the libm ``log10`` functions
13593 would, and handles error conditions in the same way.
13596 '``llvm.experimental.constrained.log2``' Intrinsic
13597 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13605 @llvm.experimental.constrained.log2(<type> <op1>,
13606 metadata <rounding mode>,
13607 metadata <exception behavior>)
13612 The '``llvm.experimental.constrained.log2``' intrinsic computes the base-2
13613 logarithm of the specified value.
13618 The first argument and the return value are floating-point numbers of the same
13621 The second and third arguments specify the rounding mode and exception
13622 behavior as described above.
13627 This function returns the same values as the libm ``log2`` functions
13628 would, and handles error conditions in the same way.
13631 '``llvm.experimental.constrained.rint``' Intrinsic
13632 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13640 @llvm.experimental.constrained.rint(<type> <op1>,
13641 metadata <rounding mode>,
13642 metadata <exception behavior>)
13647 The '``llvm.experimental.constrained.rint``' intrinsic returns the first
13648 operand rounded to the nearest integer. It may raise an inexact floating-point
13649 exception if the operand is not an integer.
13654 The first argument and the return value are floating-point numbers of the same
13657 The second and third arguments specify the rounding mode and exception
13658 behavior as described above.
13663 This function returns the same values as the libm ``rint`` functions
13664 would, and handles error conditions in the same way. The rounding mode is
13665 described, not determined, by the rounding mode argument. The actual rounding
13666 mode is determined by the runtime floating-point environment. The rounding
13667 mode argument is only intended as information to the compiler.
13670 '``llvm.experimental.constrained.nearbyint``' Intrinsic
13671 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13679 @llvm.experimental.constrained.nearbyint(<type> <op1>,
13680 metadata <rounding mode>,
13681 metadata <exception behavior>)
13686 The '``llvm.experimental.constrained.nearbyint``' intrinsic returns the first
13687 operand rounded to the nearest integer. It will not raise an inexact
13688 floating-point exception if the operand is not an integer.
13694 The first argument and the return value are floating-point numbers of the same
13697 The second and third arguments specify the rounding mode and exception
13698 behavior as described above.
13703 This function returns the same values as the libm ``nearbyint`` functions
13704 would, and handles error conditions in the same way. The rounding mode is
13705 described, not determined, by the rounding mode argument. The actual rounding
13706 mode is determined by the runtime floating-point environment. The rounding
13707 mode argument is only intended as information to the compiler.
13713 This class of intrinsics is designed to be generic and has no specific
13716 '``llvm.var.annotation``' Intrinsic
13717 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13724 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
13729 The '``llvm.var.annotation``' intrinsic.
13734 The first argument is a pointer to a value, the second is a pointer to a
13735 global string, the third is a pointer to a global string which is the
13736 source file name, and the last argument is the line number.
13741 This intrinsic allows annotation of local variables with arbitrary
13742 strings. This can be useful for special purpose optimizations that want
13743 to look for these annotations. These have no other defined use; they are
13744 ignored by code generation and optimization.
13746 '``llvm.ptr.annotation.*``' Intrinsic
13747 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13752 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
13753 pointer to an integer of any width. *NOTE* you must specify an address space for
13754 the pointer. The identifier for the default address space is the integer
13759 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
13760 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
13761 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
13762 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
13763 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
13768 The '``llvm.ptr.annotation``' intrinsic.
13773 The first argument is a pointer to an integer value of arbitrary bitwidth
13774 (result of some expression), the second is a pointer to a global string, the
13775 third is a pointer to a global string which is the source file name, and the
13776 last argument is the line number. It returns the value of the first argument.
13781 This intrinsic allows annotation of a pointer to an integer with arbitrary
13782 strings. This can be useful for special purpose optimizations that want to look
13783 for these annotations. These have no other defined use; they are ignored by code
13784 generation and optimization.
13786 '``llvm.annotation.*``' Intrinsic
13787 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13792 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
13793 any integer bit width.
13797 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
13798 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
13799 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
13800 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
13801 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
13806 The '``llvm.annotation``' intrinsic.
13811 The first argument is an integer value (result of some expression), the
13812 second is a pointer to a global string, the third is a pointer to a
13813 global string which is the source file name, and the last argument is
13814 the line number. It returns the value of the first argument.
13819 This intrinsic allows annotations to be put on arbitrary expressions
13820 with arbitrary strings. This can be useful for special purpose
13821 optimizations that want to look for these annotations. These have no
13822 other defined use; they are ignored by code generation and optimization.
13824 '``llvm.codeview.annotation``' Intrinsic
13825 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13830 This annotation emits a label at its program point and an associated
13831 ``S_ANNOTATION`` codeview record with some additional string metadata. This is
13832 used to implement MSVC's ``__annotation`` intrinsic. It is marked
13833 ``noduplicate``, so calls to this intrinsic prevent inlining and should be
13834 considered expensive.
13838 declare void @llvm.codeview.annotation(metadata)
13843 The argument should be an MDTuple containing any number of MDStrings.
13845 '``llvm.trap``' Intrinsic
13846 ^^^^^^^^^^^^^^^^^^^^^^^^^
13853 declare void @llvm.trap() noreturn nounwind
13858 The '``llvm.trap``' intrinsic.
13868 This intrinsic is lowered to the target dependent trap instruction. If
13869 the target does not have a trap instruction, this intrinsic will be
13870 lowered to a call of the ``abort()`` function.
13872 '``llvm.debugtrap``' Intrinsic
13873 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13880 declare void @llvm.debugtrap() nounwind
13885 The '``llvm.debugtrap``' intrinsic.
13895 This intrinsic is lowered to code which is intended to cause an
13896 execution trap with the intention of requesting the attention of a
13899 '``llvm.stackprotector``' Intrinsic
13900 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13907 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
13912 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
13913 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
13914 is placed on the stack before local variables.
13919 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
13920 The first argument is the value loaded from the stack guard
13921 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
13922 enough space to hold the value of the guard.
13927 This intrinsic causes the prologue/epilogue inserter to force the position of
13928 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
13929 to ensure that if a local variable on the stack is overwritten, it will destroy
13930 the value of the guard. When the function exits, the guard on the stack is
13931 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
13932 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
13933 calling the ``__stack_chk_fail()`` function.
13935 '``llvm.stackguard``' Intrinsic
13936 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13943 declare i8* @llvm.stackguard()
13948 The ``llvm.stackguard`` intrinsic returns the system stack guard value.
13950 It should not be generated by frontends, since it is only for internal usage.
13951 The reason why we create this intrinsic is that we still support IR form Stack
13952 Protector in FastISel.
13962 On some platforms, the value returned by this intrinsic remains unchanged
13963 between loads in the same thread. On other platforms, it returns the same
13964 global variable value, if any, e.g. ``@__stack_chk_guard``.
13966 Currently some platforms have IR-level customized stack guard loading (e.g.
13967 X86 Linux) that is not handled by ``llvm.stackguard()``, while they should be
13970 '``llvm.objectsize``' Intrinsic
13971 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
13978 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>, i1 <nullunknown>)
13979 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>, i1 <nullunknown>)
13984 The ``llvm.objectsize`` intrinsic is designed to provide information to
13985 the optimizers to determine at compile time whether a) an operation
13986 (like memcpy) will overflow a buffer that corresponds to an object, or
13987 b) that a runtime check for overflow isn't necessary. An object in this
13988 context means an allocation of a specific class, structure, array, or
13994 The ``llvm.objectsize`` intrinsic takes three arguments. The first argument is
13995 a pointer to or into the ``object``. The second argument determines whether
13996 ``llvm.objectsize`` returns 0 (if true) or -1 (if false) when the object size
13997 is unknown. The third argument controls how ``llvm.objectsize`` acts when
13998 ``null`` is used as its pointer argument. If it's true and the pointer is in
13999 address space 0, ``null`` is treated as an opaque value with an unknown number
14000 of bytes. Otherwise, ``llvm.objectsize`` reports 0 bytes available when given
14003 The second and third arguments only accept constants.
14008 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
14009 the size of the object concerned. If the size cannot be determined at
14010 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
14011 on the ``min`` argument).
14013 '``llvm.expect``' Intrinsic
14014 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
14019 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
14024 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
14025 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
14026 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
14031 The ``llvm.expect`` intrinsic provides information about expected (the
14032 most probable) value of ``val``, which can be used by optimizers.
14037 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
14038 a value. The second argument is an expected value, this needs to be a
14039 constant value, variables are not allowed.
14044 This intrinsic is lowered to the ``val``.
14048 '``llvm.assume``' Intrinsic
14049 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
14056 declare void @llvm.assume(i1 %cond)
14061 The ``llvm.assume`` allows the optimizer to assume that the provided
14062 condition is true. This information can then be used in simplifying other parts
14068 The condition which the optimizer may assume is always true.
14073 The intrinsic allows the optimizer to assume that the provided condition is
14074 always true whenever the control flow reaches the intrinsic call. No code is
14075 generated for this intrinsic, and instructions that contribute only to the
14076 provided condition are not used for code generation. If the condition is
14077 violated during execution, the behavior is undefined.
14079 Note that the optimizer might limit the transformations performed on values
14080 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
14081 only used to form the intrinsic's input argument. This might prove undesirable
14082 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
14083 sufficient overall improvement in code quality. For this reason,
14084 ``llvm.assume`` should not be used to document basic mathematical invariants
14085 that the optimizer can otherwise deduce or facts that are of little use to the
14090 '``llvm.ssa_copy``' Intrinsic
14091 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
14098 declare type @llvm.ssa_copy(type %operand) returned(1) readnone
14103 The first argument is an operand which is used as the returned value.
14108 The ``llvm.ssa_copy`` intrinsic can be used to attach information to
14109 operations by copying them and giving them new names. For example,
14110 the PredicateInfo utility uses it to build Extended SSA form, and
14111 attach various forms of information to operands that dominate specific
14112 uses. It is not meant for general use, only for building temporary
14113 renaming forms that require value splits at certain points.
14117 '``llvm.type.test``' Intrinsic
14118 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
14125 declare i1 @llvm.type.test(i8* %ptr, metadata %type) nounwind readnone
14131 The first argument is a pointer to be tested. The second argument is a
14132 metadata object representing a :doc:`type identifier <TypeMetadata>`.
14137 The ``llvm.type.test`` intrinsic tests whether the given pointer is associated
14138 with the given type identifier.
14140 '``llvm.type.checked.load``' Intrinsic
14141 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
14148 declare {i8*, i1} @llvm.type.checked.load(i8* %ptr, i32 %offset, metadata %type) argmemonly nounwind readonly
14154 The first argument is a pointer from which to load a function pointer. The
14155 second argument is the byte offset from which to load the function pointer. The
14156 third argument is a metadata object representing a :doc:`type identifier
14162 The ``llvm.type.checked.load`` intrinsic safely loads a function pointer from a
14163 virtual table pointer using type metadata. This intrinsic is used to implement
14164 control flow integrity in conjunction with virtual call optimization. The
14165 virtual call optimization pass will optimize away ``llvm.type.checked.load``
14166 intrinsics associated with devirtualized calls, thereby removing the type
14167 check in cases where it is not needed to enforce the control flow integrity
14170 If the given pointer is associated with a type metadata identifier, this
14171 function returns true as the second element of its return value. (Note that
14172 the function may also return true if the given pointer is not associated
14173 with a type metadata identifier.) If the function's return value's second
14174 element is true, the following rules apply to the first element:
14176 - If the given pointer is associated with the given type metadata identifier,
14177 it is the function pointer loaded from the given byte offset from the given
14180 - If the given pointer is not associated with the given type metadata
14181 identifier, it is one of the following (the choice of which is unspecified):
14183 1. The function pointer that would have been loaded from an arbitrarily chosen
14184 (through an unspecified mechanism) pointer associated with the type
14187 2. If the function has a non-void return type, a pointer to a function that
14188 returns an unspecified value without causing side effects.
14190 If the function's return value's second element is false, the value of the
14191 first element is undefined.
14194 '``llvm.donothing``' Intrinsic
14195 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
14202 declare void @llvm.donothing() nounwind readnone
14207 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
14208 three intrinsics (besides ``llvm.experimental.patchpoint`` and
14209 ``llvm.experimental.gc.statepoint``) that can be called with an invoke
14220 This intrinsic does nothing, and it's removed by optimizers and ignored
14223 '``llvm.experimental.deoptimize``' Intrinsic
14224 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
14231 declare type @llvm.experimental.deoptimize(...) [ "deopt"(...) ]
14236 This intrinsic, together with :ref:`deoptimization operand bundles
14237 <deopt_opbundles>`, allow frontends to express transfer of control and
14238 frame-local state from the currently executing (typically more specialized,
14239 hence faster) version of a function into another (typically more generic, hence
14242 In languages with a fully integrated managed runtime like Java and JavaScript
14243 this intrinsic can be used to implement "uncommon trap" or "side exit" like
14244 functionality. In unmanaged languages like C and C++, this intrinsic can be
14245 used to represent the slow paths of specialized functions.
14251 The intrinsic takes an arbitrary number of arguments, whose meaning is
14252 decided by the :ref:`lowering strategy<deoptimize_lowering>`.
14257 The ``@llvm.experimental.deoptimize`` intrinsic executes an attached
14258 deoptimization continuation (denoted using a :ref:`deoptimization
14259 operand bundle <deopt_opbundles>`) and returns the value returned by
14260 the deoptimization continuation. Defining the semantic properties of
14261 the continuation itself is out of scope of the language reference --
14262 as far as LLVM is concerned, the deoptimization continuation can
14263 invoke arbitrary side effects, including reading from and writing to
14266 Deoptimization continuations expressed using ``"deopt"`` operand bundles always
14267 continue execution to the end of the physical frame containing them, so all
14268 calls to ``@llvm.experimental.deoptimize`` must be in "tail position":
14270 - ``@llvm.experimental.deoptimize`` cannot be invoked.
14271 - The call must immediately precede a :ref:`ret <i_ret>` instruction.
14272 - The ``ret`` instruction must return the value produced by the
14273 ``@llvm.experimental.deoptimize`` call if there is one, or void.
14275 Note that the above restrictions imply that the return type for a call to
14276 ``@llvm.experimental.deoptimize`` will match the return type of its immediate
14279 The inliner composes the ``"deopt"`` continuations of the caller into the
14280 ``"deopt"`` continuations present in the inlinee, and also updates calls to this
14281 intrinsic to return directly from the frame of the function it inlined into.
14283 All declarations of ``@llvm.experimental.deoptimize`` must share the
14284 same calling convention.
14286 .. _deoptimize_lowering:
14291 Calls to ``@llvm.experimental.deoptimize`` are lowered to calls to the
14292 symbol ``__llvm_deoptimize`` (it is the frontend's responsibility to
14293 ensure that this symbol is defined). The call arguments to
14294 ``@llvm.experimental.deoptimize`` are lowered as if they were formal
14295 arguments of the specified types, and not as varargs.
14298 '``llvm.experimental.guard``' Intrinsic
14299 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
14306 declare void @llvm.experimental.guard(i1, ...) [ "deopt"(...) ]
14311 This intrinsic, together with :ref:`deoptimization operand bundles
14312 <deopt_opbundles>`, allows frontends to express guards or checks on
14313 optimistic assumptions made during compilation. The semantics of
14314 ``@llvm.experimental.guard`` is defined in terms of
14315 ``@llvm.experimental.deoptimize`` -- its body is defined to be
14318 .. code-block:: text
14320 define void @llvm.experimental.guard(i1 %pred, <args...>) {
14321 %realPred = and i1 %pred, undef
14322 br i1 %realPred, label %continue, label %leave [, !make.implicit !{}]
14325 call void @llvm.experimental.deoptimize(<args...>) [ "deopt"() ]
14333 with the optional ``[, !make.implicit !{}]`` present if and only if it
14334 is present on the call site. For more details on ``!make.implicit``,
14335 see :doc:`FaultMaps`.
14337 In words, ``@llvm.experimental.guard`` executes the attached
14338 ``"deopt"`` continuation if (but **not** only if) its first argument
14339 is ``false``. Since the optimizer is allowed to replace the ``undef``
14340 with an arbitrary value, it can optimize guard to fail "spuriously",
14341 i.e. without the original condition being false (hence the "not only
14342 if"); and this allows for "check widening" type optimizations.
14344 ``@llvm.experimental.guard`` cannot be invoked.
14347 '``llvm.load.relative``' Intrinsic
14348 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
14355 declare i8* @llvm.load.relative.iN(i8* %ptr, iN %offset) argmemonly nounwind readonly
14360 This intrinsic loads a 32-bit value from the address ``%ptr + %offset``,
14361 adds ``%ptr`` to that value and returns it. The constant folder specifically
14362 recognizes the form of this intrinsic and the constant initializers it may
14363 load from; if a loaded constant initializer is known to have the form
14364 ``i32 trunc(x - %ptr)``, the intrinsic call is folded to ``x``.
14366 LLVM provides that the calculation of such a constant initializer will
14367 not overflow at link time under the medium code model if ``x`` is an
14368 ``unnamed_addr`` function. However, it does not provide this guarantee for
14369 a constant initializer folded into a function body. This intrinsic can be
14370 used to avoid the possibility of overflows when loading from such a constant.
14372 '``llvm.sideeffect``' Intrinsic
14373 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
14380 declare void @llvm.sideeffect() inaccessiblememonly nounwind
14385 The ``llvm.sideeffect`` intrinsic doesn't perform any operation. Optimizers
14386 treat it as having side effects, so it can be inserted into a loop to
14387 indicate that the loop shouldn't be assumed to terminate (which could
14388 potentially lead to the loop being optimized away entirely), even if it's
14389 an infinite loop with no other side effects.
14399 This intrinsic actually does nothing, but optimizers must assume that it
14400 has externally observable side effects.
14402 Stack Map Intrinsics
14403 --------------------
14405 LLVM provides experimental intrinsics to support runtime patching
14406 mechanisms commonly desired in dynamic language JITs. These intrinsics
14407 are described in :doc:`StackMaps`.
14409 Element Wise Atomic Memory Intrinsics
14410 -------------------------------------
14412 These intrinsics are similar to the standard library memory intrinsics except
14413 that they perform memory transfer as a sequence of atomic memory accesses.
14415 .. _int_memcpy_element_unordered_atomic:
14417 '``llvm.memcpy.element.unordered.atomic``' Intrinsic
14418 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
14423 This is an overloaded intrinsic. You can use ``llvm.memcpy.element.unordered.atomic`` on
14424 any integer bit width and for different address spaces. Not all targets
14425 support all bit widths however.
14429 declare void @llvm.memcpy.element.unordered.atomic.p0i8.p0i8.i32(i8* <dest>,
14432 i32 <element_size>)
14433 declare void @llvm.memcpy.element.unordered.atomic.p0i8.p0i8.i64(i8* <dest>,
14436 i32 <element_size>)
14441 The '``llvm.memcpy.element.unordered.atomic.*``' intrinsic is a specialization of the
14442 '``llvm.memcpy.*``' intrinsic. It differs in that the ``dest`` and ``src`` are treated
14443 as arrays with elements that are exactly ``element_size`` bytes, and the copy between
14444 buffers uses a sequence of :ref:`unordered atomic <ordering>` load/store operations
14445 that are a positive integer multiple of the ``element_size`` in size.
14450 The first three arguments are the same as they are in the :ref:`@llvm.memcpy <int_memcpy>`
14451 intrinsic, with the added constraint that ``len`` is required to be a positive integer
14452 multiple of the ``element_size``. If ``len`` is not a positive integer multiple of
14453 ``element_size``, then the behaviour of the intrinsic is undefined.
14455 ``element_size`` must be a compile-time constant positive power of two no greater than
14456 target-specific atomic access size limit.
14458 For each of the input pointers ``align`` parameter attribute must be specified. It
14459 must be a power of two no less than the ``element_size``. Caller guarantees that
14460 both the source and destination pointers are aligned to that boundary.
14465 The '``llvm.memcpy.element.unordered.atomic.*``' intrinsic copies ``len`` bytes of
14466 memory from the source location to the destination location. These locations are not
14467 allowed to overlap. The memory copy is performed as a sequence of load/store operations
14468 where each access is guaranteed to be a multiple of ``element_size`` bytes wide and
14469 aligned at an ``element_size`` boundary.
14471 The order of the copy is unspecified. The same value may be read from the source
14472 buffer many times, but only one write is issued to the destination buffer per
14473 element. It is well defined to have concurrent reads and writes to both source and
14474 destination provided those reads and writes are unordered atomic when specified.
14476 This intrinsic does not provide any additional ordering guarantees over those
14477 provided by a set of unordered loads from the source location and stores to the
14483 In the most general case call to the '``llvm.memcpy.element.unordered.atomic.*``' is
14484 lowered to a call to the symbol ``__llvm_memcpy_element_unordered_atomic_*``. Where '*'
14485 is replaced with an actual element size.
14487 Optimizer is allowed to inline memory copy when it's profitable to do so.
14489 '``llvm.memmove.element.unordered.atomic``' Intrinsic
14490 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
14495 This is an overloaded intrinsic. You can use
14496 ``llvm.memmove.element.unordered.atomic`` on any integer bit width and for
14497 different address spaces. Not all targets support all bit widths however.
14501 declare void @llvm.memmove.element.unordered.atomic.p0i8.p0i8.i32(i8* <dest>,
14504 i32 <element_size>)
14505 declare void @llvm.memmove.element.unordered.atomic.p0i8.p0i8.i64(i8* <dest>,
14508 i32 <element_size>)
14513 The '``llvm.memmove.element.unordered.atomic.*``' intrinsic is a specialization
14514 of the '``llvm.memmove.*``' intrinsic. It differs in that the ``dest`` and
14515 ``src`` are treated as arrays with elements that are exactly ``element_size``
14516 bytes, and the copy between buffers uses a sequence of
14517 :ref:`unordered atomic <ordering>` load/store operations that are a positive
14518 integer multiple of the ``element_size`` in size.
14523 The first three arguments are the same as they are in the
14524 :ref:`@llvm.memmove <int_memmove>` intrinsic, with the added constraint that
14525 ``len`` is required to be a positive integer multiple of the ``element_size``.
14526 If ``len`` is not a positive integer multiple of ``element_size``, then the
14527 behaviour of the intrinsic is undefined.
14529 ``element_size`` must be a compile-time constant positive power of two no
14530 greater than a target-specific atomic access size limit.
14532 For each of the input pointers the ``align`` parameter attribute must be
14533 specified. It must be a power of two no less than the ``element_size``. Caller
14534 guarantees that both the source and destination pointers are aligned to that
14540 The '``llvm.memmove.element.unordered.atomic.*``' intrinsic copies ``len`` bytes
14541 of memory from the source location to the destination location. These locations
14542 are allowed to overlap. The memory copy is performed as a sequence of load/store
14543 operations where each access is guaranteed to be a multiple of ``element_size``
14544 bytes wide and aligned at an ``element_size`` boundary.
14546 The order of the copy is unspecified. The same value may be read from the source
14547 buffer many times, but only one write is issued to the destination buffer per
14548 element. It is well defined to have concurrent reads and writes to both source
14549 and destination provided those reads and writes are unordered atomic when
14552 This intrinsic does not provide any additional ordering guarantees over those
14553 provided by a set of unordered loads from the source location and stores to the
14559 In the most general case call to the
14560 '``llvm.memmove.element.unordered.atomic.*``' is lowered to a call to the symbol
14561 ``__llvm_memmove_element_unordered_atomic_*``. Where '*' is replaced with an
14562 actual element size.
14564 The optimizer is allowed to inline the memory copy when it's profitable to do so.
14566 .. _int_memset_element_unordered_atomic:
14568 '``llvm.memset.element.unordered.atomic``' Intrinsic
14569 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
14574 This is an overloaded intrinsic. You can use ``llvm.memset.element.unordered.atomic`` on
14575 any integer bit width and for different address spaces. Not all targets
14576 support all bit widths however.
14580 declare void @llvm.memset.element.unordered.atomic.p0i8.i32(i8* <dest>,
14583 i32 <element_size>)
14584 declare void @llvm.memset.element.unordered.atomic.p0i8.i64(i8* <dest>,
14587 i32 <element_size>)
14592 The '``llvm.memset.element.unordered.atomic.*``' intrinsic is a specialization of the
14593 '``llvm.memset.*``' intrinsic. It differs in that the ``dest`` is treated as an array
14594 with elements that are exactly ``element_size`` bytes, and the assignment to that array
14595 uses uses a sequence of :ref:`unordered atomic <ordering>` store operations
14596 that are a positive integer multiple of the ``element_size`` in size.
14601 The first three arguments are the same as they are in the :ref:`@llvm.memset <int_memset>`
14602 intrinsic, with the added constraint that ``len`` is required to be a positive integer
14603 multiple of the ``element_size``. If ``len`` is not a positive integer multiple of
14604 ``element_size``, then the behaviour of the intrinsic is undefined.
14606 ``element_size`` must be a compile-time constant positive power of two no greater than
14607 target-specific atomic access size limit.
14609 The ``dest`` input pointer must have the ``align`` parameter attribute specified. It
14610 must be a power of two no less than the ``element_size``. Caller guarantees that
14611 the destination pointer is aligned to that boundary.
14616 The '``llvm.memset.element.unordered.atomic.*``' intrinsic sets the ``len`` bytes of
14617 memory starting at the destination location to the given ``value``. The memory is
14618 set with a sequence of store operations where each access is guaranteed to be a
14619 multiple of ``element_size`` bytes wide and aligned at an ``element_size`` boundary.
14621 The order of the assignment is unspecified. Only one write is issued to the
14622 destination buffer per element. It is well defined to have concurrent reads and
14623 writes to the destination provided those reads and writes are unordered atomic
14626 This intrinsic does not provide any additional ordering guarantees over those
14627 provided by a set of unordered stores to the destination.
14632 In the most general case call to the '``llvm.memset.element.unordered.atomic.*``' is
14633 lowered to a call to the symbol ``__llvm_memset_element_unordered_atomic_*``. Where '*'
14634 is replaced with an actual element size.
14636 The optimizer is allowed to inline the memory assignment when it's profitable to do so.