1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][-a-zA-Z$._][-a-zA-Z$._0-9]*``'. Identifiers that require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves. The ``"\01"`` prefix
83 can be used on global variables to suppress mangling.
84 #. Unnamed values are represented as an unsigned numeric value with
85 their prefix. For example, ``%12``, ``@2``, ``%44``.
86 #. Constants, which are described in the section Constants_ below.
88 LLVM requires that values start with a prefix for two reasons: Compilers
89 don't need to worry about name clashes with reserved words, and the set
90 of reserved words may be expanded in the future without penalty.
91 Additionally, unnamed identifiers allow a compiler to quickly come up
92 with a temporary variable without having to avoid symbol table
95 Reserved words in LLVM are very similar to reserved words in other
96 languages. There are keywords for different opcodes ('``add``',
97 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
98 '``i32``', etc...), and others. These reserved words cannot conflict
99 with variable names, because none of them start with a prefix character
100 (``'%'`` or ``'@'``).
102 Here is an example of LLVM code to multiply the integer variable
109 %result = mul i32 %X, 8
111 After strength reduction:
115 %result = shl i32 %X, 3
121 %0 = add i32 %X, %X ; yields i32:%0
122 %1 = add i32 %0, %0 ; yields i32:%1
123 %result = add i32 %1, %1
125 This last way of multiplying ``%X`` by 8 illustrates several important
126 lexical features of LLVM:
128 #. Comments are delimited with a '``;``' and go until the end of line.
129 #. Unnamed temporaries are created when the result of a computation is
130 not assigned to a named value.
131 #. Unnamed temporaries are numbered sequentially (using a per-function
132 incrementing counter, starting with 0). Note that basic blocks and unnamed
133 function parameters are included in this numbering. For example, if the
134 entry basic block is not given a label name and all function parameters are
135 named, then it will get number 0.
137 It also shows a convention that we follow in this document. When
138 demonstrating instructions, we will follow an instruction with a comment
139 that defines the type and name of value produced.
147 LLVM programs are composed of ``Module``'s, each of which is a
148 translation unit of the input programs. Each module consists of
149 functions, global variables, and symbol table entries. Modules may be
150 combined together with the LLVM linker, which merges function (and
151 global variable) definitions, resolves forward declarations, and merges
152 symbol table entries. Here is an example of the "hello world" module:
156 ; Declare the string constant as a global constant.
157 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
159 ; External declaration of the puts function
160 declare i32 @puts(i8* nocapture) nounwind
162 ; Definition of main function
163 define i32 @main() { ; i32()*
164 ; Convert [13 x i8]* to i8 *...
165 %cast210 = getelementptr [13 x i8], [13 x i8]* @.str, i64 0, i64 0
167 ; Call puts function to write out the string to stdout.
168 call i32 @puts(i8* %cast210)
173 !0 = !{i32 42, null, !"string"}
176 This example is made up of a :ref:`global variable <globalvars>` named
177 "``.str``", an external declaration of the "``puts``" function, a
178 :ref:`function definition <functionstructure>` for "``main``" and
179 :ref:`named metadata <namedmetadatastructure>` "``foo``".
181 In general, a module is made up of a list of global values (where both
182 functions and global variables are global values). Global values are
183 represented by a pointer to a memory location (in this case, a pointer
184 to an array of char, and a pointer to a function), and have one of the
185 following :ref:`linkage types <linkage>`.
192 All Global Variables and Functions have one of the following types of
196 Global values with "``private``" linkage are only directly
197 accessible by objects in the current module. In particular, linking
198 code into a module with an private global value may cause the
199 private to be renamed as necessary to avoid collisions. Because the
200 symbol is private to the module, all references can be updated. This
201 doesn't show up in any symbol table in the object file.
203 Similar to private, but the value shows as a local symbol
204 (``STB_LOCAL`` in the case of ELF) in the object file. This
205 corresponds to the notion of the '``static``' keyword in C.
206 ``available_externally``
207 Globals with "``available_externally``" linkage are never emitted 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 The semantics of this linkage follow the ELF object file model: the
255 symbol is weak until linked, if not linked, the symbol becomes null
256 instead of being an undefined reference.
257 ``linkonce_odr``, ``weak_odr``
258 Some languages allow differing globals to be merged, such as two
259 functions with different semantics. Other languages, such as
260 ``C++``, ensure that only equivalent globals are ever merged (the
261 "one definition rule" --- "ODR"). Such languages can use the
262 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
263 global will only be merged with equivalent globals. These linkage
264 types are otherwise the same as their non-``odr`` versions.
266 If none of the above identifiers are used, the global is externally
267 visible, meaning that it participates in linkage and can be used to
268 resolve external symbol references.
270 It is illegal for a function *declaration* to have any linkage type
271 other than ``external`` or ``extern_weak``.
278 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
279 :ref:`invokes <i_invoke>` can all have an optional calling convention
280 specified for the call. The calling convention of any pair of dynamic
281 caller/callee must match, or the behavior of the program is undefined.
282 The following calling conventions are supported by LLVM, and more may be
285 "``ccc``" - The C calling convention
286 This calling convention (the default if no other calling convention
287 is specified) matches the target C calling conventions. This calling
288 convention supports varargs function calls and tolerates some
289 mismatch in the declared prototype and implemented declaration of
290 the function (as does normal C).
291 "``fastcc``" - The fast calling convention
292 This calling convention attempts to make calls as fast as possible
293 (e.g. by passing things in registers). This calling convention
294 allows the target to use whatever tricks it wants to produce fast
295 code for the target, without having to conform to an externally
296 specified ABI (Application Binary Interface). `Tail calls can only
297 be optimized when this, the GHC or the HiPE convention is
298 used. <CodeGenerator.html#id80>`_ This calling convention does not
299 support varargs and requires the prototype of all callees to exactly
300 match the prototype of the function definition.
301 "``coldcc``" - The cold calling convention
302 This calling convention attempts to make code in the caller as
303 efficient as possible under the assumption that the call is not
304 commonly executed. As such, these calls often preserve all registers
305 so that the call does not break any live ranges in the caller side.
306 This calling convention does not support varargs and requires the
307 prototype of all callees to exactly match the prototype of the
308 function definition. Furthermore the inliner doesn't consider such function
310 "``cc 10``" - GHC convention
311 This calling convention has been implemented specifically for use by
312 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
313 It passes everything in registers, going to extremes to achieve this
314 by disabling callee save registers. This calling convention should
315 not be used lightly but only for specific situations such as an
316 alternative to the *register pinning* performance technique often
317 used when implementing functional programming languages. At the
318 moment only X86 supports this convention and it has the following
321 - On *X86-32* only supports up to 4 bit type parameters. No
322 floating point types are supported.
323 - On *X86-64* only supports up to 10 bit type parameters and 6
324 floating point parameters.
326 This calling convention supports `tail call
327 optimization <CodeGenerator.html#id80>`_ but requires both the
328 caller and callee are using it.
329 "``cc 11``" - The HiPE calling convention
330 This calling convention has been implemented specifically for use by
331 the `High-Performance Erlang
332 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
333 native code compiler of the `Ericsson's Open Source Erlang/OTP
334 system <http://www.erlang.org/download.shtml>`_. It uses more
335 registers for argument passing than the ordinary C calling
336 convention and defines no callee-saved registers. The calling
337 convention properly supports `tail call
338 optimization <CodeGenerator.html#id80>`_ but requires that both the
339 caller and the callee use it. It uses a *register pinning*
340 mechanism, similar to GHC's convention, for keeping frequently
341 accessed runtime components pinned to specific hardware registers.
342 At the moment only X86 supports this convention (both 32 and 64
344 "``webkit_jscc``" - WebKit's JavaScript calling convention
345 This calling convention has been implemented for `WebKit FTL JIT
346 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
347 stack right to left (as cdecl does), and returns a value in the
348 platform's customary return register.
349 "``anyregcc``" - Dynamic calling convention for code patching
350 This is a special convention that supports patching an arbitrary code
351 sequence in place of a call site. This convention forces the call
352 arguments into registers but allows them to be dynamically
353 allocated. This can currently only be used with calls to
354 llvm.experimental.patchpoint because only this intrinsic records
355 the location of its arguments in a side table. See :doc:`StackMaps`.
356 "``preserve_mostcc``" - The `PreserveMost` calling convention
357 This calling convention attempts to make the code in the caller as
358 unintrusive as possible. This convention behaves identically to the `C`
359 calling convention on how arguments and return values are passed, but it
360 uses a different set of caller/callee-saved registers. This alleviates the
361 burden of saving and recovering a large register set before and after the
362 call in the caller. If the arguments are passed in callee-saved registers,
363 then they will be preserved by the callee across the call. This doesn't
364 apply for values returned in callee-saved registers.
366 - On X86-64 the callee preserves all general purpose registers, except for
367 R11. R11 can be used as a scratch register. Floating-point registers
368 (XMMs/YMMs) are not preserved and need to be saved by the caller.
370 The idea behind this convention is to support calls to runtime functions
371 that have a hot path and a cold path. The hot path is usually a small piece
372 of code that doesn't use many registers. The cold path might need to call out to
373 another function and therefore only needs to preserve the caller-saved
374 registers, which haven't already been saved by the caller. The
375 `PreserveMost` calling convention is very similar to the `cold` calling
376 convention in terms of caller/callee-saved registers, but they are used for
377 different types of function calls. `coldcc` is for function calls that are
378 rarely executed, whereas `preserve_mostcc` function calls are intended to be
379 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
380 doesn't prevent the inliner from inlining the function call.
382 This calling convention will be used by a future version of the ObjectiveC
383 runtime and should therefore still be considered experimental at this time.
384 Although this convention was created to optimize certain runtime calls to
385 the ObjectiveC runtime, it is not limited to this runtime and might be used
386 by other runtimes in the future too. The current implementation only
387 supports X86-64, but the intention is to support more architectures in the
389 "``preserve_allcc``" - The `PreserveAll` calling convention
390 This calling convention attempts to make the code in the caller even less
391 intrusive than the `PreserveMost` calling convention. This calling
392 convention also behaves identical to the `C` calling convention on how
393 arguments and return values are passed, but it uses a different set of
394 caller/callee-saved registers. This removes the burden of saving and
395 recovering a large register set before and after the call in the caller. If
396 the arguments are passed in callee-saved registers, then they will be
397 preserved by the callee across the call. This doesn't apply for values
398 returned in callee-saved registers.
400 - On X86-64 the callee preserves all general purpose registers, except for
401 R11. R11 can be used as a scratch register. Furthermore it also preserves
402 all floating-point registers (XMMs/YMMs).
404 The idea behind this convention is to support calls to runtime functions
405 that don't need to call out to any other functions.
407 This calling convention, like the `PreserveMost` calling convention, will be
408 used by a future version of the ObjectiveC runtime and should be considered
409 experimental at this time.
410 "``cxx_fast_tlscc``" - The `CXX_FAST_TLS` calling convention for access functions
411 Clang generates an access function to access C++-style TLS. The access
412 function generally has an entry block, an exit block and an initialization
413 block that is run at the first time. The entry and exit blocks can access
414 a few TLS IR variables, each access will be lowered to a platform-specific
417 This calling convention aims to minimize overhead in the caller by
418 preserving as many registers as possible (all the registers that are
419 perserved on the fast path, composed of the entry and exit blocks).
421 This calling convention behaves identical to the `C` calling convention on
422 how arguments and return values are passed, but it uses a different set of
423 caller/callee-saved registers.
425 Given that each platform has its own lowering sequence, hence its own set
426 of preserved registers, we can't use the existing `PreserveMost`.
428 - On X86-64 the callee preserves all general purpose registers, except for
430 "``cc <n>``" - Numbered convention
431 Any calling convention may be specified by number, allowing
432 target-specific calling conventions to be used. Target specific
433 calling conventions start at 64.
435 More calling conventions can be added/defined on an as-needed basis, to
436 support Pascal conventions or any other well-known target-independent
439 .. _visibilitystyles:
444 All Global Variables and Functions have one of the following visibility
447 "``default``" - Default style
448 On targets that use the ELF object file format, default visibility
449 means that the declaration is visible to other modules and, in
450 shared libraries, means that the declared entity may be overridden.
451 On Darwin, default visibility means that the declaration is visible
452 to other modules. Default visibility corresponds to "external
453 linkage" in the language.
454 "``hidden``" - Hidden style
455 Two declarations of an object with hidden visibility refer to the
456 same object if they are in the same shared object. Usually, hidden
457 visibility indicates that the symbol will not be placed into the
458 dynamic symbol table, so no other module (executable or shared
459 library) can reference it directly.
460 "``protected``" - Protected style
461 On ELF, protected visibility indicates that the symbol will be
462 placed in the dynamic symbol table, but that references within the
463 defining module will bind to the local symbol. That is, the symbol
464 cannot be overridden by another module.
466 A symbol with ``internal`` or ``private`` linkage must have ``default``
474 All Global Variables, Functions and Aliases can have one of the following
478 "``dllimport``" causes the compiler to reference a function or variable via
479 a global pointer to a pointer that is set up by the DLL exporting the
480 symbol. On Microsoft Windows targets, the pointer name is formed by
481 combining ``__imp_`` and the function or variable name.
483 "``dllexport``" causes the compiler to provide a global pointer to a pointer
484 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
485 Microsoft Windows targets, the pointer name is formed by combining
486 ``__imp_`` and the function or variable name. Since this storage class
487 exists for defining a dll interface, the compiler, assembler and linker know
488 it is externally referenced and must refrain from deleting the symbol.
492 Thread Local Storage Models
493 ---------------------------
495 A variable may be defined as ``thread_local``, which means that it will
496 not be shared by threads (each thread will have a separated copy of the
497 variable). Not all targets support thread-local variables. Optionally, a
498 TLS model may be specified:
501 For variables that are only used within the current shared library.
503 For variables in modules that will not be loaded dynamically.
505 For variables defined in the executable and only used within it.
507 If no explicit model is given, the "general dynamic" model is used.
509 The models correspond to the ELF TLS models; see `ELF Handling For
510 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
511 more information on under which circumstances the different models may
512 be used. The target may choose a different TLS model if the specified
513 model is not supported, or if a better choice of model can be made.
515 A model can also be specified in an alias, but then it only governs how
516 the alias is accessed. It will not have any effect in the aliasee.
518 For platforms without linker support of ELF TLS model, the -femulated-tls
519 flag can be used to generate GCC compatible emulated TLS code.
526 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
527 types <t_struct>`. Literal types are uniqued structurally, but identified types
528 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
529 to forward declare a type that is not yet available.
531 An example of an identified structure specification is:
535 %mytype = type { %mytype*, i32 }
537 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
538 literal types are uniqued in recent versions of LLVM.
545 Global variables define regions of memory allocated at compilation time
548 Global variable definitions must be initialized.
550 Global variables in other translation units can also be declared, in which
551 case they don't have an initializer.
553 Either global variable definitions or declarations may have an explicit section
554 to be placed in and may have an optional explicit alignment specified.
556 A variable may be defined as a global ``constant``, which indicates that
557 the contents of the variable will **never** be modified (enabling better
558 optimization, allowing the global data to be placed in the read-only
559 section of an executable, etc). Note that variables that need runtime
560 initialization cannot be marked ``constant`` as there is a store to the
563 LLVM explicitly allows *declarations* of global variables to be marked
564 constant, even if the final definition of the global is not. This
565 capability can be used to enable slightly better optimization of the
566 program, but requires the language definition to guarantee that
567 optimizations based on the 'constantness' are valid for the translation
568 units that do not include the definition.
570 As SSA values, global variables define pointer values that are in scope
571 (i.e. they dominate) all basic blocks in the program. Global variables
572 always define a pointer to their "content" type because they describe a
573 region of memory, and all memory objects in LLVM are accessed through
576 Global variables can be marked with ``unnamed_addr`` which indicates
577 that the address is not significant, only the content. Constants marked
578 like this can be merged with other constants if they have the same
579 initializer. Note that a constant with significant address *can* be
580 merged with a ``unnamed_addr`` constant, the result being a constant
581 whose address is significant.
583 A global variable may be declared to reside in a target-specific
584 numbered address space. For targets that support them, address spaces
585 may affect how optimizations are performed and/or what target
586 instructions are used to access the variable. The default address space
587 is zero. The address space qualifier must precede any other attributes.
589 LLVM allows an explicit section to be specified for globals. If the
590 target supports it, it will emit globals to the section specified.
591 Additionally, the global can placed in a comdat if the target has the necessary
594 By default, global initializers are optimized by assuming that global
595 variables defined within the module are not modified from their
596 initial values before the start of the global initializer. This is
597 true even for variables potentially accessible from outside the
598 module, including those with external linkage or appearing in
599 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
600 by marking the variable with ``externally_initialized``.
602 An explicit alignment may be specified for a global, which must be a
603 power of 2. If not present, or if the alignment is set to zero, the
604 alignment of the global is set by the target to whatever it feels
605 convenient. If an explicit alignment is specified, the global is forced
606 to have exactly that alignment. Targets and optimizers are not allowed
607 to over-align the global if the global has an assigned section. In this
608 case, the extra alignment could be observable: for example, code could
609 assume that the globals are densely packed in their section and try to
610 iterate over them as an array, alignment padding would break this
611 iteration. The maximum alignment is ``1 << 29``.
613 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
615 Variables and aliases can have a
616 :ref:`Thread Local Storage Model <tls_model>`.
620 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
621 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
622 <global | constant> <Type> [<InitializerConstant>]
623 [, section "name"] [, comdat [($name)]]
624 [, align <Alignment>]
626 For example, the following defines a global in a numbered address space
627 with an initializer, section, and alignment:
631 @G = addrspace(5) constant float 1.0, section "foo", align 4
633 The following example just declares a global variable
637 @G = external global i32
639 The following example defines a thread-local global with the
640 ``initialexec`` TLS model:
644 @G = thread_local(initialexec) global i32 0, align 4
646 .. _functionstructure:
651 LLVM function definitions consist of the "``define``" keyword, an
652 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
653 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
654 an optional :ref:`calling convention <callingconv>`,
655 an optional ``unnamed_addr`` attribute, a return type, an optional
656 :ref:`parameter attribute <paramattrs>` for the return type, a function
657 name, a (possibly empty) argument list (each with optional :ref:`parameter
658 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
659 an optional section, an optional alignment,
660 an optional :ref:`comdat <langref_comdats>`,
661 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
662 an optional :ref:`prologue <prologuedata>`,
663 an optional :ref:`personality <personalityfn>`,
664 an optional list of attached :ref:`metadata <metadata>`,
665 an opening curly brace, a list of basic blocks, and a closing curly brace.
667 LLVM function declarations consist of the "``declare``" keyword, an
668 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
669 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
670 an optional :ref:`calling convention <callingconv>`,
671 an optional ``unnamed_addr`` attribute, a return type, an optional
672 :ref:`parameter attribute <paramattrs>` for the return type, a function
673 name, a possibly empty list of arguments, an optional alignment, an optional
674 :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
675 and an optional :ref:`prologue <prologuedata>`.
677 A function definition contains a list of basic blocks, forming the CFG (Control
678 Flow Graph) for the function. Each basic block may optionally start with a label
679 (giving the basic block a symbol table entry), contains a list of instructions,
680 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
681 function return). If an explicit label is not provided, a block is assigned an
682 implicit numbered label, using the next value from the same counter as used for
683 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
684 entry block does not have an explicit label, it will be assigned label "%0",
685 then the first unnamed temporary in that block will be "%1", etc.
687 The first basic block in a function is special in two ways: it is
688 immediately executed on entrance to the function, and it is not allowed
689 to have predecessor basic blocks (i.e. there can not be any branches to
690 the entry block of a function). Because the block can have no
691 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
693 LLVM allows an explicit section to be specified for functions. If the
694 target supports it, it will emit functions to the section specified.
695 Additionally, the function can be placed in a COMDAT.
697 An explicit alignment may be specified for a function. If not present,
698 or if the alignment is set to zero, the alignment of the function is set
699 by the target to whatever it feels convenient. If an explicit alignment
700 is specified, the function is forced to have at least that much
701 alignment. All alignments must be a power of 2.
703 If the ``unnamed_addr`` attribute is given, the address is known to not
704 be significant and two identical functions can be merged.
708 define [linkage] [visibility] [DLLStorageClass]
710 <ResultType> @<FunctionName> ([argument list])
711 [unnamed_addr] [fn Attrs] [section "name"] [comdat [($name)]]
712 [align N] [gc] [prefix Constant] [prologue Constant]
713 [personality Constant] (!name !N)* { ... }
715 The argument list is a comma separated sequence of arguments where each
716 argument is of the following form:
720 <type> [parameter Attrs] [name]
728 Aliases, unlike function or variables, don't create any new data. They
729 are just a new symbol and metadata for an existing position.
731 Aliases have a name and an aliasee that is either a global value or a
734 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
735 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
736 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
740 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy>, <AliaseeTy>* @<Aliasee>
742 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
743 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
744 might not correctly handle dropping a weak symbol that is aliased.
746 Aliases that are not ``unnamed_addr`` are guaranteed to have the same address as
747 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
750 Since aliases are only a second name, some restrictions apply, of which
751 some can only be checked when producing an object file:
753 * The expression defining the aliasee must be computable at assembly
754 time. Since it is just a name, no relocations can be used.
756 * No alias in the expression can be weak as the possibility of the
757 intermediate alias being overridden cannot be represented in an
760 * No global value in the expression can be a declaration, since that
761 would require a relocation, which is not possible.
768 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
770 Comdats have a name which represents the COMDAT key. All global objects that
771 specify this key will only end up in the final object file if the linker chooses
772 that key over some other key. Aliases are placed in the same COMDAT that their
773 aliasee computes to, if any.
775 Comdats have a selection kind to provide input on how the linker should
776 choose between keys in two different object files.
780 $<Name> = comdat SelectionKind
782 The selection kind must be one of the following:
785 The linker may choose any COMDAT key, the choice is arbitrary.
787 The linker may choose any COMDAT key but the sections must contain the
790 The linker will choose the section containing the largest COMDAT key.
792 The linker requires that only section with this COMDAT key exist.
794 The linker may choose any COMDAT key but the sections must contain the
797 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
798 ``any`` as a selection kind.
800 Here is an example of a COMDAT group where a function will only be selected if
801 the COMDAT key's section is the largest:
805 $foo = comdat largest
806 @foo = global i32 2, comdat($foo)
808 define void @bar() comdat($foo) {
812 As a syntactic sugar the ``$name`` can be omitted if the name is the same as
818 @foo = global i32 2, comdat
821 In a COFF object file, this will create a COMDAT section with selection kind
822 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
823 and another COMDAT section with selection kind
824 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
825 section and contains the contents of the ``@bar`` symbol.
827 There are some restrictions on the properties of the global object.
828 It, or an alias to it, must have the same name as the COMDAT group when
830 The contents and size of this object may be used during link-time to determine
831 which COMDAT groups get selected depending on the selection kind.
832 Because the name of the object must match the name of the COMDAT group, the
833 linkage of the global object must not be local; local symbols can get renamed
834 if a collision occurs in the symbol table.
836 The combined use of COMDATS and section attributes may yield surprising results.
843 @g1 = global i32 42, section "sec", comdat($foo)
844 @g2 = global i32 42, section "sec", comdat($bar)
846 From the object file perspective, this requires the creation of two sections
847 with the same name. This is necessary because both globals belong to different
848 COMDAT groups and COMDATs, at the object file level, are represented by
851 Note that certain IR constructs like global variables and functions may
852 create COMDATs in the object file in addition to any which are specified using
853 COMDAT IR. This arises when the code generator is configured to emit globals
854 in individual sections (e.g. when `-data-sections` or `-function-sections`
855 is supplied to `llc`).
857 .. _namedmetadatastructure:
862 Named metadata is a collection of metadata. :ref:`Metadata
863 nodes <metadata>` (but not metadata strings) are the only valid
864 operands for a named metadata.
866 #. Named metadata are represented as a string of characters with the
867 metadata prefix. The rules for metadata names are the same as for
868 identifiers, but quoted names are not allowed. ``"\xx"`` type escapes
869 are still valid, which allows any character to be part of a name.
873 ; Some unnamed metadata nodes, which are referenced by the named metadata.
878 !name = !{!0, !1, !2}
885 The return type and each parameter of a function type may have a set of
886 *parameter attributes* associated with them. Parameter attributes are
887 used to communicate additional information about the result or
888 parameters of a function. Parameter attributes are considered to be part
889 of the function, not of the function type, so functions with different
890 parameter attributes can have the same function type.
892 Parameter attributes are simple keywords that follow the type specified.
893 If multiple parameter attributes are needed, they are space separated.
898 declare i32 @printf(i8* noalias nocapture, ...)
899 declare i32 @atoi(i8 zeroext)
900 declare signext i8 @returns_signed_char()
902 Note that any attributes for the function result (``nounwind``,
903 ``readonly``) come immediately after the argument list.
905 Currently, only the following parameter attributes are defined:
908 This indicates to the code generator that the parameter or return
909 value should be zero-extended to the extent required by the target's
910 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
911 the caller (for a parameter) or the callee (for a return value).
913 This indicates to the code generator that the parameter or return
914 value should be sign-extended to the extent required by the target's
915 ABI (which is usually 32-bits) by the caller (for a parameter) or
916 the callee (for a return value).
918 This indicates that this parameter or return value should be treated
919 in a special target-dependent fashion while emitting code for
920 a function call or return (usually, by putting it in a register as
921 opposed to memory, though some targets use it to distinguish between
922 two different kinds of registers). Use of this attribute is
925 This indicates that the pointer parameter should really be passed by
926 value to the function. The attribute implies that a hidden copy of
927 the pointee is made between the caller and the callee, so the callee
928 is unable to modify the value in the caller. This attribute is only
929 valid on LLVM pointer arguments. It is generally used to pass
930 structs and arrays by value, but is also valid on pointers to
931 scalars. The copy is considered to belong to the caller not the
932 callee (for example, ``readonly`` functions should not write to
933 ``byval`` parameters). This is not a valid attribute for return
936 The byval attribute also supports specifying an alignment with the
937 align attribute. It indicates the alignment of the stack slot to
938 form and the known alignment of the pointer specified to the call
939 site. If the alignment is not specified, then the code generator
940 makes a target-specific assumption.
946 The ``inalloca`` argument attribute allows the caller to take the
947 address of outgoing stack arguments. An ``inalloca`` argument must
948 be a pointer to stack memory produced by an ``alloca`` instruction.
949 The alloca, or argument allocation, must also be tagged with the
950 inalloca keyword. Only the last argument may have the ``inalloca``
951 attribute, and that argument is guaranteed to be passed in memory.
953 An argument allocation may be used by a call at most once because
954 the call may deallocate it. The ``inalloca`` attribute cannot be
955 used in conjunction with other attributes that affect argument
956 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
957 ``inalloca`` attribute also disables LLVM's implicit lowering of
958 large aggregate return values, which means that frontend authors
959 must lower them with ``sret`` pointers.
961 When the call site is reached, the argument allocation must have
962 been the most recent stack allocation that is still live, or the
963 results are undefined. It is possible to allocate additional stack
964 space after an argument allocation and before its call site, but it
965 must be cleared off with :ref:`llvm.stackrestore
968 See :doc:`InAlloca` for more information on how to use this
972 This indicates that the pointer parameter specifies the address of a
973 structure that is the return value of the function in the source
974 program. This pointer must be guaranteed by the caller to be valid:
975 loads and stores to the structure may be assumed by the callee
976 not to trap and to be properly aligned. This may only be applied to
977 the first parameter. This is not a valid attribute for return
981 This indicates that the pointer value may be assumed by the optimizer to
982 have the specified alignment.
984 Note that this attribute has additional semantics when combined with the
990 This indicates that objects accessed via pointer values
991 :ref:`based <pointeraliasing>` on the argument or return value are not also
992 accessed, during the execution of the function, via pointer values not
993 *based* on the argument or return value. The attribute on a return value
994 also has additional semantics described below. The caller shares the
995 responsibility with the callee for ensuring that these requirements are met.
996 For further details, please see the discussion of the NoAlias response in
997 :ref:`alias analysis <Must, May, or No>`.
999 Note that this definition of ``noalias`` is intentionally similar
1000 to the definition of ``restrict`` in C99 for function arguments.
1002 For function return values, C99's ``restrict`` is not meaningful,
1003 while LLVM's ``noalias`` is. Furthermore, the semantics of the ``noalias``
1004 attribute on return values are stronger than the semantics of the attribute
1005 when used on function arguments. On function return values, the ``noalias``
1006 attribute indicates that the function acts like a system memory allocation
1007 function, returning a pointer to allocated storage disjoint from the
1008 storage for any other object accessible to the caller.
1011 This indicates that the callee does not make any copies of the
1012 pointer that outlive the callee itself. This is not a valid
1013 attribute for return values.
1018 This indicates that the pointer parameter can be excised using the
1019 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
1020 attribute for return values and can only be applied to one parameter.
1023 This indicates that the function always returns the argument as its return
1024 value. This is an optimization hint to the code generator when generating
1025 the caller, allowing tail call optimization and omission of register saves
1026 and restores in some cases; it is not checked or enforced when generating
1027 the callee. The parameter and the function return type must be valid
1028 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
1029 valid attribute for return values and can only be applied to one parameter.
1032 This indicates that the parameter or return pointer is not null. This
1033 attribute may only be applied to pointer typed parameters. This is not
1034 checked or enforced by LLVM, the caller must ensure that the pointer
1035 passed in is non-null, or the callee must ensure that the returned pointer
1038 ``dereferenceable(<n>)``
1039 This indicates that the parameter or return pointer is dereferenceable. This
1040 attribute may only be applied to pointer typed parameters. A pointer that
1041 is dereferenceable can be loaded from speculatively without a risk of
1042 trapping. The number of bytes known to be dereferenceable must be provided
1043 in parentheses. It is legal for the number of bytes to be less than the
1044 size of the pointee type. The ``nonnull`` attribute does not imply
1045 dereferenceability (consider a pointer to one element past the end of an
1046 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
1047 ``addrspace(0)`` (which is the default address space).
1049 ``dereferenceable_or_null(<n>)``
1050 This indicates that the parameter or return value isn't both
1051 non-null and non-dereferenceable (up to ``<n>`` bytes) at the same
1052 time. All non-null pointers tagged with
1053 ``dereferenceable_or_null(<n>)`` are ``dereferenceable(<n>)``.
1054 For address space 0 ``dereferenceable_or_null(<n>)`` implies that
1055 a pointer is exactly one of ``dereferenceable(<n>)`` or ``null``,
1056 and in other address spaces ``dereferenceable_or_null(<n>)``
1057 implies that a pointer is at least one of ``dereferenceable(<n>)``
1058 or ``null`` (i.e. it may be both ``null`` and
1059 ``dereferenceable(<n>)``). This attribute may only be applied to
1060 pointer typed parameters.
1064 Garbage Collector Strategy Names
1065 --------------------------------
1067 Each function may specify a garbage collector strategy name, which is simply a
1070 .. code-block:: llvm
1072 define void @f() gc "name" { ... }
1074 The supported values of *name* includes those :ref:`built in to LLVM
1075 <builtin-gc-strategies>` and any provided by loaded plugins. Specifying a GC
1076 strategy will cause the compiler to alter its output in order to support the
1077 named garbage collection algorithm. Note that LLVM itself does not contain a
1078 garbage collector, this functionality is restricted to generating machine code
1079 which can interoperate with a collector provided externally.
1086 Prefix data is data associated with a function which the code
1087 generator will emit immediately before the function's entrypoint.
1088 The purpose of this feature is to allow frontends to associate
1089 language-specific runtime metadata with specific functions and make it
1090 available through the function pointer while still allowing the
1091 function pointer to be called.
1093 To access the data for a given function, a program may bitcast the
1094 function pointer to a pointer to the constant's type and dereference
1095 index -1. This implies that the IR symbol points just past the end of
1096 the prefix data. For instance, take the example of a function annotated
1097 with a single ``i32``,
1099 .. code-block:: llvm
1101 define void @f() prefix i32 123 { ... }
1103 The prefix data can be referenced as,
1105 .. code-block:: llvm
1107 %0 = bitcast void* () @f to i32*
1108 %a = getelementptr inbounds i32, i32* %0, i32 -1
1109 %b = load i32, i32* %a
1111 Prefix data is laid out as if it were an initializer for a global variable
1112 of the prefix data's type. The function will be placed such that the
1113 beginning of the prefix data is aligned. This means that if the size
1114 of the prefix data is not a multiple of the alignment size, the
1115 function's entrypoint will not be aligned. If alignment of the
1116 function's entrypoint is desired, padding must be added to the prefix
1119 A function may have prefix data but no body. This has similar semantics
1120 to the ``available_externally`` linkage in that the data may be used by the
1121 optimizers but will not be emitted in the object file.
1128 The ``prologue`` attribute allows arbitrary code (encoded as bytes) to
1129 be inserted prior to the function body. This can be used for enabling
1130 function hot-patching and instrumentation.
1132 To maintain the semantics of ordinary function calls, the prologue data must
1133 have a particular format. Specifically, it must begin with a sequence of
1134 bytes which decode to a sequence of machine instructions, valid for the
1135 module's target, which transfer control to the point immediately succeeding
1136 the prologue data, without performing any other visible action. This allows
1137 the inliner and other passes to reason about the semantics of the function
1138 definition without needing to reason about the prologue data. Obviously this
1139 makes the format of the prologue data highly target dependent.
1141 A trivial example of valid prologue data for the x86 architecture is ``i8 144``,
1142 which encodes the ``nop`` instruction:
1144 .. code-block:: llvm
1146 define void @f() prologue i8 144 { ... }
1148 Generally prologue data can be formed by encoding a relative branch instruction
1149 which skips the metadata, as in this example of valid prologue data for the
1150 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1152 .. code-block:: llvm
1154 %0 = type <{ i8, i8, i8* }>
1156 define void @f() prologue %0 <{ i8 235, i8 8, i8* @md}> { ... }
1158 A function may have prologue data but no body. This has similar semantics
1159 to the ``available_externally`` linkage in that the data may be used by the
1160 optimizers but will not be emitted in the object file.
1164 Personality Function
1165 --------------------
1167 The ``personality`` attribute permits functions to specify what function
1168 to use for exception handling.
1175 Attribute groups are groups of attributes that are referenced by objects within
1176 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1177 functions will use the same set of attributes. In the degenerative case of a
1178 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1179 group will capture the important command line flags used to build that file.
1181 An attribute group is a module-level object. To use an attribute group, an
1182 object references the attribute group's ID (e.g. ``#37``). An object may refer
1183 to more than one attribute group. In that situation, the attributes from the
1184 different groups are merged.
1186 Here is an example of attribute groups for a function that should always be
1187 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1189 .. code-block:: llvm
1191 ; Target-independent attributes:
1192 attributes #0 = { alwaysinline alignstack=4 }
1194 ; Target-dependent attributes:
1195 attributes #1 = { "no-sse" }
1197 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1198 define void @f() #0 #1 { ... }
1205 Function attributes are set to communicate additional information about
1206 a function. Function attributes are considered to be part of the
1207 function, not of the function type, so functions with different function
1208 attributes can have the same function type.
1210 Function attributes are simple keywords that follow the type specified.
1211 If multiple attributes are needed, they are space separated. For
1214 .. code-block:: llvm
1216 define void @f() noinline { ... }
1217 define void @f() alwaysinline { ... }
1218 define void @f() alwaysinline optsize { ... }
1219 define void @f() optsize { ... }
1222 This attribute indicates that, when emitting the prologue and
1223 epilogue, the backend should forcibly align the stack pointer.
1224 Specify the desired alignment, which must be a power of two, in
1227 This attribute indicates that the inliner should attempt to inline
1228 this function into callers whenever possible, ignoring any active
1229 inlining size threshold for this caller.
1231 This indicates that the callee function at a call site should be
1232 recognized as a built-in function, even though the function's declaration
1233 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1234 direct calls to functions that are declared with the ``nobuiltin``
1237 This attribute indicates that this function is rarely called. When
1238 computing edge weights, basic blocks post-dominated by a cold
1239 function call are also considered to be cold; and, thus, given low
1242 This attribute indicates that the callee is dependent on a convergent
1243 thread execution pattern under certain parallel execution models.
1244 Transformations that are execution model agnostic may not make the execution
1245 of a convergent operation control dependent on any additional values.
1246 ``inaccessiblememonly``
1247 This attribute indicates that the function may only access memory that
1248 is not accessible by the module being compiled. This is a weaker form
1250 ``inaccessiblemem_or_argmemonly``
1251 This attribute indicates that the function may only access memory that is
1252 either not accessible by the module being compiled, or is pointed to
1253 by its pointer arguments. This is a weaker form of ``argmemonly``
1255 This attribute indicates that the source code contained a hint that
1256 inlining this function is desirable (such as the "inline" keyword in
1257 C/C++). It is just a hint; it imposes no requirements on the
1260 This attribute indicates that the function should be added to a
1261 jump-instruction table at code-generation time, and that all address-taken
1262 references to this function should be replaced with a reference to the
1263 appropriate jump-instruction-table function pointer. Note that this creates
1264 a new pointer for the original function, which means that code that depends
1265 on function-pointer identity can break. So, any function annotated with
1266 ``jumptable`` must also be ``unnamed_addr``.
1268 This attribute suggests that optimization passes and code generator
1269 passes make choices that keep the code size of this function as small
1270 as possible and perform optimizations that may sacrifice runtime
1271 performance in order to minimize the size of the generated code.
1273 This attribute disables prologue / epilogue emission for the
1274 function. This can have very system-specific consequences.
1276 This indicates that the callee function at a call site is not recognized as
1277 a built-in function. LLVM will retain the original call and not replace it
1278 with equivalent code based on the semantics of the built-in function, unless
1279 the call site uses the ``builtin`` attribute. This is valid at call sites
1280 and on function declarations and definitions.
1282 This attribute indicates that calls to the function cannot be
1283 duplicated. A call to a ``noduplicate`` function may be moved
1284 within its parent function, but may not be duplicated within
1285 its parent function.
1287 A function containing a ``noduplicate`` call may still
1288 be an inlining candidate, provided that the call is not
1289 duplicated by inlining. That implies that the function has
1290 internal linkage and only has one call site, so the original
1291 call is dead after inlining.
1293 This attributes disables implicit floating point instructions.
1295 This attribute indicates that the inliner should never inline this
1296 function in any situation. This attribute may not be used together
1297 with the ``alwaysinline`` attribute.
1299 This attribute suppresses lazy symbol binding for the function. This
1300 may make calls to the function faster, at the cost of extra program
1301 startup time if the function is not called during program startup.
1303 This attribute indicates that the code generator should not use a
1304 red zone, even if the target-specific ABI normally permits it.
1306 This function attribute indicates that the function never returns
1307 normally. This produces undefined behavior at runtime if the
1308 function ever does dynamically return.
1310 This function attribute indicates that the function does not call itself
1311 either directly or indirectly down any possible call path. This produces
1312 undefined behavior at runtime if the function ever does recurse.
1314 This function attribute indicates that the function never raises an
1315 exception. If the function does raise an exception, its runtime
1316 behavior is undefined. However, functions marked nounwind may still
1317 trap or generate asynchronous exceptions. Exception handling schemes
1318 that are recognized by LLVM to handle asynchronous exceptions, such
1319 as SEH, will still provide their implementation defined semantics.
1321 This function attribute indicates that most optimization passes will skip
1322 this function, with the exception of interprocedural optimization passes.
1323 Code generation defaults to the "fast" instruction selector.
1324 This attribute cannot be used together with the ``alwaysinline``
1325 attribute; this attribute is also incompatible
1326 with the ``minsize`` attribute and the ``optsize`` attribute.
1328 This attribute requires the ``noinline`` attribute to be specified on
1329 the function as well, so the function is never inlined into any caller.
1330 Only functions with the ``alwaysinline`` attribute are valid
1331 candidates for inlining into the body of this function.
1333 This attribute suggests that optimization passes and code generator
1334 passes make choices that keep the code size of this function low,
1335 and otherwise do optimizations specifically to reduce code size as
1336 long as they do not significantly impact runtime performance.
1338 On a function, this attribute indicates that the function computes its
1339 result (or decides to unwind an exception) based strictly on its arguments,
1340 without dereferencing any pointer arguments or otherwise accessing
1341 any mutable state (e.g. memory, control registers, etc) visible to
1342 caller functions. It does not write through any pointer arguments
1343 (including ``byval`` arguments) and never changes any state visible
1344 to callers. This means that it cannot unwind exceptions by calling
1345 the ``C++`` exception throwing methods.
1347 On an argument, this attribute indicates that the function does not
1348 dereference that pointer argument, even though it may read or write the
1349 memory that the pointer points to if accessed through other pointers.
1351 On a function, this attribute indicates that the function does not write
1352 through any pointer arguments (including ``byval`` arguments) or otherwise
1353 modify any state (e.g. memory, control registers, etc) visible to
1354 caller functions. It may dereference pointer arguments and read
1355 state that may be set in the caller. A readonly function always
1356 returns the same value (or unwinds an exception identically) when
1357 called with the same set of arguments and global state. It cannot
1358 unwind an exception by calling the ``C++`` exception throwing
1361 On an argument, this attribute indicates that the function does not write
1362 through this pointer argument, even though it may write to the memory that
1363 the pointer points to.
1365 This attribute indicates that the only memory accesses inside function are
1366 loads and stores from objects pointed to by its pointer-typed arguments,
1367 with arbitrary offsets. Or in other words, all memory operations in the
1368 function can refer to memory only using pointers based on its function
1370 Note that ``argmemonly`` can be used together with ``readonly`` attribute
1371 in order to specify that function reads only from its arguments.
1373 This attribute indicates that this function can return twice. The C
1374 ``setjmp`` is an example of such a function. The compiler disables
1375 some optimizations (like tail calls) in the caller of these
1378 This attribute indicates that
1379 `SafeStack <http://clang.llvm.org/docs/SafeStack.html>`_
1380 protection is enabled for this function.
1382 If a function that has a ``safestack`` attribute is inlined into a
1383 function that doesn't have a ``safestack`` attribute or which has an
1384 ``ssp``, ``sspstrong`` or ``sspreq`` attribute, then the resulting
1385 function will have a ``safestack`` attribute.
1386 ``sanitize_address``
1387 This attribute indicates that AddressSanitizer checks
1388 (dynamic address safety analysis) are enabled for this function.
1390 This attribute indicates that MemorySanitizer checks (dynamic detection
1391 of accesses to uninitialized memory) are enabled for this function.
1393 This attribute indicates that ThreadSanitizer checks
1394 (dynamic thread safety analysis) are enabled for this function.
1396 This attribute indicates that the function should emit a stack
1397 smashing protector. It is in the form of a "canary" --- a random value
1398 placed on the stack before the local variables that's checked upon
1399 return from the function to see if it has been overwritten. A
1400 heuristic is used to determine if a function needs stack protectors
1401 or not. The heuristic used will enable protectors for functions with:
1403 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1404 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1405 - Calls to alloca() with variable sizes or constant sizes greater than
1406 ``ssp-buffer-size``.
1408 Variables that are identified as requiring a protector will be arranged
1409 on the stack such that they are adjacent to the stack protector guard.
1411 If a function that has an ``ssp`` attribute is inlined into a
1412 function that doesn't have an ``ssp`` attribute, then the resulting
1413 function will have an ``ssp`` attribute.
1415 This attribute indicates that the function should *always* emit a
1416 stack smashing protector. This overrides the ``ssp`` function
1419 Variables that are identified as requiring a protector will be arranged
1420 on the stack such that they are adjacent to the stack protector guard.
1421 The specific layout rules are:
1423 #. Large arrays and structures containing large arrays
1424 (``>= ssp-buffer-size``) are closest to the stack protector.
1425 #. Small arrays and structures containing small arrays
1426 (``< ssp-buffer-size``) are 2nd closest to the protector.
1427 #. Variables that have had their address taken are 3rd closest to the
1430 If a function that has an ``sspreq`` attribute is inlined into a
1431 function that doesn't have an ``sspreq`` attribute or which has an
1432 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1433 an ``sspreq`` attribute.
1435 This attribute indicates that the function should emit a stack smashing
1436 protector. This attribute causes a strong heuristic to be used when
1437 determining if a function needs stack protectors. The strong heuristic
1438 will enable protectors for functions with:
1440 - Arrays of any size and type
1441 - Aggregates containing an array of any size and type.
1442 - Calls to alloca().
1443 - Local variables that have had their address taken.
1445 Variables that are identified as requiring a protector will be arranged
1446 on the stack such that they are adjacent to the stack protector guard.
1447 The specific layout rules are:
1449 #. Large arrays and structures containing large arrays
1450 (``>= ssp-buffer-size``) are closest to the stack protector.
1451 #. Small arrays and structures containing small arrays
1452 (``< ssp-buffer-size``) are 2nd closest to the protector.
1453 #. Variables that have had their address taken are 3rd closest to the
1456 This overrides the ``ssp`` function attribute.
1458 If a function that has an ``sspstrong`` attribute is inlined into a
1459 function that doesn't have an ``sspstrong`` attribute, then the
1460 resulting function will have an ``sspstrong`` attribute.
1462 This attribute indicates that the function will delegate to some other
1463 function with a tail call. The prototype of a thunk should not be used for
1464 optimization purposes. The caller is expected to cast the thunk prototype to
1465 match the thunk target prototype.
1467 This attribute indicates that the ABI being targeted requires that
1468 an unwind table entry be produced for this function even if we can
1469 show that no exceptions passes by it. This is normally the case for
1470 the ELF x86-64 abi, but it can be disabled for some compilation
1479 Note: operand bundles are a work in progress, and they should be
1480 considered experimental at this time.
1482 Operand bundles are tagged sets of SSA values that can be associated
1483 with certain LLVM instructions (currently only ``call`` s and
1484 ``invoke`` s). In a way they are like metadata, but dropping them is
1485 incorrect and will change program semantics.
1489 operand bundle set ::= '[' operand bundle (, operand bundle )* ']'
1490 operand bundle ::= tag '(' [ bundle operand ] (, bundle operand )* ')'
1491 bundle operand ::= SSA value
1492 tag ::= string constant
1494 Operand bundles are **not** part of a function's signature, and a
1495 given function may be called from multiple places with different kinds
1496 of operand bundles. This reflects the fact that the operand bundles
1497 are conceptually a part of the ``call`` (or ``invoke``), not the
1498 callee being dispatched to.
1500 Operand bundles are a generic mechanism intended to support
1501 runtime-introspection-like functionality for managed languages. While
1502 the exact semantics of an operand bundle depend on the bundle tag,
1503 there are certain limitations to how much the presence of an operand
1504 bundle can influence the semantics of a program. These restrictions
1505 are described as the semantics of an "unknown" operand bundle. As
1506 long as the behavior of an operand bundle is describable within these
1507 restrictions, LLVM does not need to have special knowledge of the
1508 operand bundle to not miscompile programs containing it.
1510 - The bundle operands for an unknown operand bundle escape in unknown
1511 ways before control is transferred to the callee or invokee.
1512 - Calls and invokes with operand bundles have unknown read / write
1513 effect on the heap on entry and exit (even if the call target is
1514 ``readnone`` or ``readonly``), unless they're overriden with
1515 callsite specific attributes.
1516 - An operand bundle at a call site cannot change the implementation
1517 of the called function. Inter-procedural optimizations work as
1518 usual as long as they take into account the first two properties.
1520 More specific types of operand bundles are described below.
1522 Deoptimization Operand Bundles
1523 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1525 Deoptimization operand bundles are characterized by the ``"deopt"``
1526 operand bundle tag. These operand bundles represent an alternate
1527 "safe" continuation for the call site they're attached to, and can be
1528 used by a suitable runtime to deoptimize the compiled frame at the
1529 specified call site. There can be at most one ``"deopt"`` operand
1530 bundle attached to a call site. Exact details of deoptimization is
1531 out of scope for the language reference, but it usually involves
1532 rewriting a compiled frame into a set of interpreted frames.
1534 From the compiler's perspective, deoptimization operand bundles make
1535 the call sites they're attached to at least ``readonly``. They read
1536 through all of their pointer typed operands (even if they're not
1537 otherwise escaped) and the entire visible heap. Deoptimization
1538 operand bundles do not capture their operands except during
1539 deoptimization, in which case control will not be returned to the
1542 The inliner knows how to inline through calls that have deoptimization
1543 operand bundles. Just like inlining through a normal call site
1544 involves composing the normal and exceptional continuations, inlining
1545 through a call site with a deoptimization operand bundle needs to
1546 appropriately compose the "safe" deoptimization continuation. The
1547 inliner does this by prepending the parent's deoptimization
1548 continuation to every deoptimization continuation in the inlined body.
1549 E.g. inlining ``@f`` into ``@g`` in the following example
1551 .. code-block:: llvm
1554 call void @x() ;; no deopt state
1555 call void @y() [ "deopt"(i32 10) ]
1556 call void @y() [ "deopt"(i32 10), "unknown"(i8* null) ]
1561 call void @f() [ "deopt"(i32 20) ]
1567 .. code-block:: llvm
1570 call void @x() ;; still no deopt state
1571 call void @y() [ "deopt"(i32 20, i32 10) ]
1572 call void @y() [ "deopt"(i32 20, i32 10), "unknown"(i8* null) ]
1576 It is the frontend's responsibility to structure or encode the
1577 deoptimization state in a way that syntactically prepending the
1578 caller's deoptimization state to the callee's deoptimization state is
1579 semantically equivalent to composing the caller's deoptimization
1580 continuation after the callee's deoptimization continuation.
1584 Funclet Operand Bundles
1585 ^^^^^^^^^^^^^^^^^^^^^^^
1587 Funclet operand bundles are characterized by the ``"funclet"``
1588 operand bundle tag. These operand bundles indicate that a call site
1589 is within a particular funclet. There can be at most one
1590 ``"funclet"`` operand bundle attached to a call site and it must have
1591 exactly one bundle operand.
1593 If any funclet EH pads have been "entered" but not "exited" (per the
1594 `description in the EH doc\ <ExceptionHandling.html#wineh-constraints>`_),
1595 it is undefined behavior to execute a ``call`` or ``invoke`` which:
1597 * does not have a ``"funclet"`` bundle and is not a ``call`` to a nounwind
1599 * has a ``"funclet"`` bundle whose operand is not the most-recently-entered
1600 not-yet-exited funclet EH pad.
1602 Similarly, if no funclet EH pads have been entered-but-not-yet-exited,
1603 executing a ``call`` or ``invoke`` with a ``"funclet"`` bundle is undefined behavior.
1607 Module-Level Inline Assembly
1608 ----------------------------
1610 Modules may contain "module-level inline asm" blocks, which corresponds
1611 to the GCC "file scope inline asm" blocks. These blocks are internally
1612 concatenated by LLVM and treated as a single unit, but may be separated
1613 in the ``.ll`` file if desired. The syntax is very simple:
1615 .. code-block:: llvm
1617 module asm "inline asm code goes here"
1618 module asm "more can go here"
1620 The strings can contain any character by escaping non-printable
1621 characters. The escape sequence used is simply "\\xx" where "xx" is the
1622 two digit hex code for the number.
1624 Note that the assembly string *must* be parseable by LLVM's integrated assembler
1625 (unless it is disabled), even when emitting a ``.s`` file.
1627 .. _langref_datalayout:
1632 A module may specify a target specific data layout string that specifies
1633 how data is to be laid out in memory. The syntax for the data layout is
1636 .. code-block:: llvm
1638 target datalayout = "layout specification"
1640 The *layout specification* consists of a list of specifications
1641 separated by the minus sign character ('-'). Each specification starts
1642 with a letter and may include other information after the letter to
1643 define some aspect of the data layout. The specifications accepted are
1647 Specifies that the target lays out data in big-endian form. That is,
1648 the bits with the most significance have the lowest address
1651 Specifies that the target lays out data in little-endian form. That
1652 is, the bits with the least significance have the lowest address
1655 Specifies the natural alignment of the stack in bits. Alignment
1656 promotion of stack variables is limited to the natural stack
1657 alignment to avoid dynamic stack realignment. The stack alignment
1658 must be a multiple of 8-bits. If omitted, the natural stack
1659 alignment defaults to "unspecified", which does not prevent any
1660 alignment promotions.
1661 ``p[n]:<size>:<abi>:<pref>``
1662 This specifies the *size* of a pointer and its ``<abi>`` and
1663 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1664 bits. The address space, ``n``, is optional, and if not specified,
1665 denotes the default address space 0. The value of ``n`` must be
1666 in the range [1,2^23).
1667 ``i<size>:<abi>:<pref>``
1668 This specifies the alignment for an integer type of a given bit
1669 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1670 ``v<size>:<abi>:<pref>``
1671 This specifies the alignment for a vector type of a given bit
1673 ``f<size>:<abi>:<pref>``
1674 This specifies the alignment for a floating point type of a given bit
1675 ``<size>``. Only values of ``<size>`` that are supported by the target
1676 will work. 32 (float) and 64 (double) are supported on all targets; 80
1677 or 128 (different flavors of long double) are also supported on some
1680 This specifies the alignment for an object of aggregate type.
1682 If present, specifies that llvm names are mangled in the output. The
1685 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1686 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1687 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1688 symbols get a ``_`` prefix.
1689 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1690 functions also get a suffix based on the frame size.
1691 * ``x``: Windows x86 COFF prefix: Similar to Windows COFF, but use a ``_``
1692 prefix for ``__cdecl`` functions.
1693 ``n<size1>:<size2>:<size3>...``
1694 This specifies a set of native integer widths for the target CPU in
1695 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1696 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1697 this set are considered to support most general arithmetic operations
1700 On every specification that takes a ``<abi>:<pref>``, specifying the
1701 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1702 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1704 When constructing the data layout for a given target, LLVM starts with a
1705 default set of specifications which are then (possibly) overridden by
1706 the specifications in the ``datalayout`` keyword. The default
1707 specifications are given in this list:
1709 - ``E`` - big endian
1710 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1711 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1712 same as the default address space.
1713 - ``S0`` - natural stack alignment is unspecified
1714 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1715 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1716 - ``i16:16:16`` - i16 is 16-bit aligned
1717 - ``i32:32:32`` - i32 is 32-bit aligned
1718 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1719 alignment of 64-bits
1720 - ``f16:16:16`` - half is 16-bit aligned
1721 - ``f32:32:32`` - float is 32-bit aligned
1722 - ``f64:64:64`` - double is 64-bit aligned
1723 - ``f128:128:128`` - quad is 128-bit aligned
1724 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1725 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1726 - ``a:0:64`` - aggregates are 64-bit aligned
1728 When LLVM is determining the alignment for a given type, it uses the
1731 #. If the type sought is an exact match for one of the specifications,
1732 that specification is used.
1733 #. If no match is found, and the type sought is an integer type, then
1734 the smallest integer type that is larger than the bitwidth of the
1735 sought type is used. If none of the specifications are larger than
1736 the bitwidth then the largest integer type is used. For example,
1737 given the default specifications above, the i7 type will use the
1738 alignment of i8 (next largest) while both i65 and i256 will use the
1739 alignment of i64 (largest specified).
1740 #. If no match is found, and the type sought is a vector type, then the
1741 largest vector type that is smaller than the sought vector type will
1742 be used as a fall back. This happens because <128 x double> can be
1743 implemented in terms of 64 <2 x double>, for example.
1745 The function of the data layout string may not be what you expect.
1746 Notably, this is not a specification from the frontend of what alignment
1747 the code generator should use.
1749 Instead, if specified, the target data layout is required to match what
1750 the ultimate *code generator* expects. This string is used by the
1751 mid-level optimizers to improve code, and this only works if it matches
1752 what the ultimate code generator uses. There is no way to generate IR
1753 that does not embed this target-specific detail into the IR. If you
1754 don't specify the string, the default specifications will be used to
1755 generate a Data Layout and the optimization phases will operate
1756 accordingly and introduce target specificity into the IR with respect to
1757 these default specifications.
1764 A module may specify a target triple string that describes the target
1765 host. The syntax for the target triple is simply:
1767 .. code-block:: llvm
1769 target triple = "x86_64-apple-macosx10.7.0"
1771 The *target triple* string consists of a series of identifiers delimited
1772 by the minus sign character ('-'). The canonical forms are:
1776 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1777 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1779 This information is passed along to the backend so that it generates
1780 code for the proper architecture. It's possible to override this on the
1781 command line with the ``-mtriple`` command line option.
1783 .. _pointeraliasing:
1785 Pointer Aliasing Rules
1786 ----------------------
1788 Any memory access must be done through a pointer value associated with
1789 an address range of the memory access, otherwise the behavior is
1790 undefined. Pointer values are associated with address ranges according
1791 to the following rules:
1793 - A pointer value is associated with the addresses associated with any
1794 value it is *based* on.
1795 - An address of a global variable is associated with the address range
1796 of the variable's storage.
1797 - The result value of an allocation instruction is associated with the
1798 address range of the allocated storage.
1799 - A null pointer in the default address-space is associated with no
1801 - An integer constant other than zero or a pointer value returned from
1802 a function not defined within LLVM may be associated with address
1803 ranges allocated through mechanisms other than those provided by
1804 LLVM. Such ranges shall not overlap with any ranges of addresses
1805 allocated by mechanisms provided by LLVM.
1807 A pointer value is *based* on another pointer value according to the
1810 - A pointer value formed from a ``getelementptr`` operation is *based*
1811 on the first value operand of the ``getelementptr``.
1812 - The result value of a ``bitcast`` is *based* on the operand of the
1814 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1815 values that contribute (directly or indirectly) to the computation of
1816 the pointer's value.
1817 - The "*based* on" relationship is transitive.
1819 Note that this definition of *"based"* is intentionally similar to the
1820 definition of *"based"* in C99, though it is slightly weaker.
1822 LLVM IR does not associate types with memory. The result type of a
1823 ``load`` merely indicates the size and alignment of the memory from
1824 which to load, as well as the interpretation of the value. The first
1825 operand type of a ``store`` similarly only indicates the size and
1826 alignment of the store.
1828 Consequently, type-based alias analysis, aka TBAA, aka
1829 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1830 :ref:`Metadata <metadata>` may be used to encode additional information
1831 which specialized optimization passes may use to implement type-based
1836 Volatile Memory Accesses
1837 ------------------------
1839 Certain memory accesses, such as :ref:`load <i_load>`'s,
1840 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1841 marked ``volatile``. The optimizers must not change the number of
1842 volatile operations or change their order of execution relative to other
1843 volatile operations. The optimizers *may* change the order of volatile
1844 operations relative to non-volatile operations. This is not Java's
1845 "volatile" and has no cross-thread synchronization behavior.
1847 IR-level volatile loads and stores cannot safely be optimized into
1848 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1849 flagged volatile. Likewise, the backend should never split or merge
1850 target-legal volatile load/store instructions.
1852 .. admonition:: Rationale
1854 Platforms may rely on volatile loads and stores of natively supported
1855 data width to be executed as single instruction. For example, in C
1856 this holds for an l-value of volatile primitive type with native
1857 hardware support, but not necessarily for aggregate types. The
1858 frontend upholds these expectations, which are intentionally
1859 unspecified in the IR. The rules above ensure that IR transformations
1860 do not violate the frontend's contract with the language.
1864 Memory Model for Concurrent Operations
1865 --------------------------------------
1867 The LLVM IR does not define any way to start parallel threads of
1868 execution or to register signal handlers. Nonetheless, there are
1869 platform-specific ways to create them, and we define LLVM IR's behavior
1870 in their presence. This model is inspired by the C++0x memory model.
1872 For a more informal introduction to this model, see the :doc:`Atomics`.
1874 We define a *happens-before* partial order as the least partial order
1877 - Is a superset of single-thread program order, and
1878 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1879 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1880 techniques, like pthread locks, thread creation, thread joining,
1881 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1882 Constraints <ordering>`).
1884 Note that program order does not introduce *happens-before* edges
1885 between a thread and signals executing inside that thread.
1887 Every (defined) read operation (load instructions, memcpy, atomic
1888 loads/read-modify-writes, etc.) R reads a series of bytes written by
1889 (defined) write operations (store instructions, atomic
1890 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1891 section, initialized globals are considered to have a write of the
1892 initializer which is atomic and happens before any other read or write
1893 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1894 may see any write to the same byte, except:
1896 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1897 write\ :sub:`2` happens before R\ :sub:`byte`, then
1898 R\ :sub:`byte` does not see write\ :sub:`1`.
1899 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1900 R\ :sub:`byte` does not see write\ :sub:`3`.
1902 Given that definition, R\ :sub:`byte` is defined as follows:
1904 - If R is volatile, the result is target-dependent. (Volatile is
1905 supposed to give guarantees which can support ``sig_atomic_t`` in
1906 C/C++, and may be used for accesses to addresses that do not behave
1907 like normal memory. It does not generally provide cross-thread
1909 - Otherwise, if there is no write to the same byte that happens before
1910 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1911 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1912 R\ :sub:`byte` returns the value written by that write.
1913 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1914 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1915 Memory Ordering Constraints <ordering>` section for additional
1916 constraints on how the choice is made.
1917 - Otherwise R\ :sub:`byte` returns ``undef``.
1919 R returns the value composed of the series of bytes it read. This
1920 implies that some bytes within the value may be ``undef`` **without**
1921 the entire value being ``undef``. Note that this only defines the
1922 semantics of the operation; it doesn't mean that targets will emit more
1923 than one instruction to read the series of bytes.
1925 Note that in cases where none of the atomic intrinsics are used, this
1926 model places only one restriction on IR transformations on top of what
1927 is required for single-threaded execution: introducing a store to a byte
1928 which might not otherwise be stored is not allowed in general.
1929 (Specifically, in the case where another thread might write to and read
1930 from an address, introducing a store can change a load that may see
1931 exactly one write into a load that may see multiple writes.)
1935 Atomic Memory Ordering Constraints
1936 ----------------------------------
1938 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1939 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1940 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1941 ordering parameters that determine which other atomic instructions on
1942 the same address they *synchronize with*. These semantics are borrowed
1943 from Java and C++0x, but are somewhat more colloquial. If these
1944 descriptions aren't precise enough, check those specs (see spec
1945 references in the :doc:`atomics guide <Atomics>`).
1946 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1947 differently since they don't take an address. See that instruction's
1948 documentation for details.
1950 For a simpler introduction to the ordering constraints, see the
1954 The set of values that can be read is governed by the happens-before
1955 partial order. A value cannot be read unless some operation wrote
1956 it. This is intended to provide a guarantee strong enough to model
1957 Java's non-volatile shared variables. This ordering cannot be
1958 specified for read-modify-write operations; it is not strong enough
1959 to make them atomic in any interesting way.
1961 In addition to the guarantees of ``unordered``, there is a single
1962 total order for modifications by ``monotonic`` operations on each
1963 address. All modification orders must be compatible with the
1964 happens-before order. There is no guarantee that the modification
1965 orders can be combined to a global total order for the whole program
1966 (and this often will not be possible). The read in an atomic
1967 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1968 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1969 order immediately before the value it writes. If one atomic read
1970 happens before another atomic read of the same address, the later
1971 read must see the same value or a later value in the address's
1972 modification order. This disallows reordering of ``monotonic`` (or
1973 stronger) operations on the same address. If an address is written
1974 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1975 read that address repeatedly, the other threads must eventually see
1976 the write. This corresponds to the C++0x/C1x
1977 ``memory_order_relaxed``.
1979 In addition to the guarantees of ``monotonic``, a
1980 *synchronizes-with* edge may be formed with a ``release`` operation.
1981 This is intended to model C++'s ``memory_order_acquire``.
1983 In addition to the guarantees of ``monotonic``, if this operation
1984 writes a value which is subsequently read by an ``acquire``
1985 operation, it *synchronizes-with* that operation. (This isn't a
1986 complete description; see the C++0x definition of a release
1987 sequence.) This corresponds to the C++0x/C1x
1988 ``memory_order_release``.
1989 ``acq_rel`` (acquire+release)
1990 Acts as both an ``acquire`` and ``release`` operation on its
1991 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1992 ``seq_cst`` (sequentially consistent)
1993 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1994 operation that only reads, ``release`` for an operation that only
1995 writes), there is a global total order on all
1996 sequentially-consistent operations on all addresses, which is
1997 consistent with the *happens-before* partial order and with the
1998 modification orders of all the affected addresses. Each
1999 sequentially-consistent read sees the last preceding write to the
2000 same address in this global order. This corresponds to the C++0x/C1x
2001 ``memory_order_seq_cst`` and Java volatile.
2005 If an atomic operation is marked ``singlethread``, it only *synchronizes
2006 with* or participates in modification and seq\_cst total orderings with
2007 other operations running in the same thread (for example, in signal
2015 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
2016 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
2017 :ref:`frem <i_frem>`, :ref:`fcmp <i_fcmp>`) have the following flags that can
2018 be set to enable otherwise unsafe floating point operations
2021 No NaNs - Allow optimizations to assume the arguments and result are not
2022 NaN. Such optimizations are required to retain defined behavior over
2023 NaNs, but the value of the result is undefined.
2026 No Infs - Allow optimizations to assume the arguments and result are not
2027 +/-Inf. Such optimizations are required to retain defined behavior over
2028 +/-Inf, but the value of the result is undefined.
2031 No Signed Zeros - Allow optimizations to treat the sign of a zero
2032 argument or result as insignificant.
2035 Allow Reciprocal - Allow optimizations to use the reciprocal of an
2036 argument rather than perform division.
2039 Fast - Allow algebraically equivalent transformations that may
2040 dramatically change results in floating point (e.g. reassociate). This
2041 flag implies all the others.
2045 Use-list Order Directives
2046 -------------------------
2048 Use-list directives encode the in-memory order of each use-list, allowing the
2049 order to be recreated. ``<order-indexes>`` is a comma-separated list of
2050 indexes that are assigned to the referenced value's uses. The referenced
2051 value's use-list is immediately sorted by these indexes.
2053 Use-list directives may appear at function scope or global scope. They are not
2054 instructions, and have no effect on the semantics of the IR. When they're at
2055 function scope, they must appear after the terminator of the final basic block.
2057 If basic blocks have their address taken via ``blockaddress()`` expressions,
2058 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
2065 uselistorder <ty> <value>, { <order-indexes> }
2066 uselistorder_bb @function, %block { <order-indexes> }
2072 define void @foo(i32 %arg1, i32 %arg2) {
2074 ; ... instructions ...
2076 ; ... instructions ...
2078 ; At function scope.
2079 uselistorder i32 %arg1, { 1, 0, 2 }
2080 uselistorder label %bb, { 1, 0 }
2084 uselistorder i32* @global, { 1, 2, 0 }
2085 uselistorder i32 7, { 1, 0 }
2086 uselistorder i32 (i32) @bar, { 1, 0 }
2087 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
2094 The LLVM type system is one of the most important features of the
2095 intermediate representation. Being typed enables a number of
2096 optimizations to be performed on the intermediate representation
2097 directly, without having to do extra analyses on the side before the
2098 transformation. A strong type system makes it easier to read the
2099 generated code and enables novel analyses and transformations that are
2100 not feasible to perform on normal three address code representations.
2110 The void type does not represent any value and has no size.
2128 The function type can be thought of as a function signature. It consists of a
2129 return type and a list of formal parameter types. The return type of a function
2130 type is a void type or first class type --- except for :ref:`label <t_label>`
2131 and :ref:`metadata <t_metadata>` types.
2137 <returntype> (<parameter list>)
2139 ...where '``<parameter list>``' is a comma-separated list of type
2140 specifiers. Optionally, the parameter list may include a type ``...``, which
2141 indicates that the function takes a variable number of arguments. Variable
2142 argument functions can access their arguments with the :ref:`variable argument
2143 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
2144 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
2148 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2149 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
2150 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2151 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
2152 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2153 | ``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. |
2154 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2155 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
2156 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2163 The :ref:`first class <t_firstclass>` types are perhaps the most important.
2164 Values of these types are the only ones which can be produced by
2172 These are the types that are valid in registers from CodeGen's perspective.
2181 The integer type is a very simple type that simply specifies an
2182 arbitrary bit width for the integer type desired. Any bit width from 1
2183 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
2191 The number of bits the integer will occupy is specified by the ``N``
2197 +----------------+------------------------------------------------+
2198 | ``i1`` | a single-bit integer. |
2199 +----------------+------------------------------------------------+
2200 | ``i32`` | a 32-bit integer. |
2201 +----------------+------------------------------------------------+
2202 | ``i1942652`` | a really big integer of over 1 million bits. |
2203 +----------------+------------------------------------------------+
2207 Floating Point Types
2208 """"""""""""""""""""
2217 - 16-bit floating point value
2220 - 32-bit floating point value
2223 - 64-bit floating point value
2226 - 128-bit floating point value (112-bit mantissa)
2229 - 80-bit floating point value (X87)
2232 - 128-bit floating point value (two 64-bits)
2239 The x86_mmx type represents a value held in an MMX register on an x86
2240 machine. The operations allowed on it are quite limited: parameters and
2241 return values, load and store, and bitcast. User-specified MMX
2242 instructions are represented as intrinsic or asm calls with arguments
2243 and/or results of this type. There are no arrays, vectors or constants
2260 The pointer type is used to specify memory locations. Pointers are
2261 commonly used to reference objects in memory.
2263 Pointer types may have an optional address space attribute defining the
2264 numbered address space where the pointed-to object resides. The default
2265 address space is number zero. The semantics of non-zero address spaces
2266 are target-specific.
2268 Note that LLVM does not permit pointers to void (``void*``) nor does it
2269 permit pointers to labels (``label*``). Use ``i8*`` instead.
2279 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2280 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2281 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2282 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2283 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2284 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2285 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2294 A vector type is a simple derived type that represents a vector of
2295 elements. Vector types are used when multiple primitive data are
2296 operated in parallel using a single instruction (SIMD). A vector type
2297 requires a size (number of elements) and an underlying primitive data
2298 type. Vector types are considered :ref:`first class <t_firstclass>`.
2304 < <# elements> x <elementtype> >
2306 The number of elements is a constant integer value larger than 0;
2307 elementtype may be any integer, floating point or pointer type. Vectors
2308 of size zero are not allowed.
2312 +-------------------+--------------------------------------------------+
2313 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2314 +-------------------+--------------------------------------------------+
2315 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2316 +-------------------+--------------------------------------------------+
2317 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2318 +-------------------+--------------------------------------------------+
2319 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2320 +-------------------+--------------------------------------------------+
2329 The label type represents code labels.
2344 The token type is used when a value is associated with an instruction
2345 but all uses of the value must not attempt to introspect or obscure it.
2346 As such, it is not appropriate to have a :ref:`phi <i_phi>` or
2347 :ref:`select <i_select>` of type token.
2364 The metadata type represents embedded metadata. No derived types may be
2365 created from metadata except for :ref:`function <t_function>` arguments.
2378 Aggregate Types are a subset of derived types that can contain multiple
2379 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2380 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2390 The array type is a very simple derived type that arranges elements
2391 sequentially in memory. The array type requires a size (number of
2392 elements) and an underlying data type.
2398 [<# elements> x <elementtype>]
2400 The number of elements is a constant integer value; ``elementtype`` may
2401 be any type with a size.
2405 +------------------+--------------------------------------+
2406 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2407 +------------------+--------------------------------------+
2408 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2409 +------------------+--------------------------------------+
2410 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2411 +------------------+--------------------------------------+
2413 Here are some examples of multidimensional arrays:
2415 +-----------------------------+----------------------------------------------------------+
2416 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2417 +-----------------------------+----------------------------------------------------------+
2418 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2419 +-----------------------------+----------------------------------------------------------+
2420 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2421 +-----------------------------+----------------------------------------------------------+
2423 There is no restriction on indexing beyond the end of the array implied
2424 by a static type (though there are restrictions on indexing beyond the
2425 bounds of an allocated object in some cases). This means that
2426 single-dimension 'variable sized array' addressing can be implemented in
2427 LLVM with a zero length array type. An implementation of 'pascal style
2428 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2438 The structure type is used to represent a collection of data members
2439 together in memory. The elements of a structure may be any type that has
2442 Structures in memory are accessed using '``load``' and '``store``' by
2443 getting a pointer to a field with the '``getelementptr``' instruction.
2444 Structures in registers are accessed using the '``extractvalue``' and
2445 '``insertvalue``' instructions.
2447 Structures may optionally be "packed" structures, which indicate that
2448 the alignment of the struct is one byte, and that there is no padding
2449 between the elements. In non-packed structs, padding between field types
2450 is inserted as defined by the DataLayout string in the module, which is
2451 required to match what the underlying code generator expects.
2453 Structures can either be "literal" or "identified". A literal structure
2454 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2455 identified types are always defined at the top level with a name.
2456 Literal types are uniqued by their contents and can never be recursive
2457 or opaque since there is no way to write one. Identified types can be
2458 recursive, can be opaqued, and are never uniqued.
2464 %T1 = type { <type list> } ; Identified normal struct type
2465 %T2 = type <{ <type list> }> ; Identified packed struct type
2469 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2470 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2471 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2472 | ``{ 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``. |
2473 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2474 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2475 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2479 Opaque Structure Types
2480 """"""""""""""""""""""
2484 Opaque structure types are used to represent named structure types that
2485 do not have a body specified. This corresponds (for example) to the C
2486 notion of a forward declared structure.
2497 +--------------+-------------------+
2498 | ``opaque`` | An opaque type. |
2499 +--------------+-------------------+
2506 LLVM has several different basic types of constants. This section
2507 describes them all and their syntax.
2512 **Boolean constants**
2513 The two strings '``true``' and '``false``' are both valid constants
2515 **Integer constants**
2516 Standard integers (such as '4') are constants of the
2517 :ref:`integer <t_integer>` type. Negative numbers may be used with
2519 **Floating point constants**
2520 Floating point constants use standard decimal notation (e.g.
2521 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2522 hexadecimal notation (see below). The assembler requires the exact
2523 decimal value of a floating-point constant. For example, the
2524 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2525 decimal in binary. Floating point constants must have a :ref:`floating
2526 point <t_floating>` type.
2527 **Null pointer constants**
2528 The identifier '``null``' is recognized as a null pointer constant
2529 and must be of :ref:`pointer type <t_pointer>`.
2531 The identifier '``none``' is recognized as an empty token constant
2532 and must be of :ref:`token type <t_token>`.
2534 The one non-intuitive notation for constants is the hexadecimal form of
2535 floating point constants. For example, the form
2536 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2537 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2538 constants are required (and the only time that they are generated by the
2539 disassembler) is when a floating point constant must be emitted but it
2540 cannot be represented as a decimal floating point number in a reasonable
2541 number of digits. For example, NaN's, infinities, and other special
2542 values are represented in their IEEE hexadecimal format so that assembly
2543 and disassembly do not cause any bits to change in the constants.
2545 When using the hexadecimal form, constants of types half, float, and
2546 double are represented using the 16-digit form shown above (which
2547 matches the IEEE754 representation for double); half and float values
2548 must, however, be exactly representable as IEEE 754 half and single
2549 precision, respectively. Hexadecimal format is always used for long
2550 double, and there are three forms of long double. The 80-bit format used
2551 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2552 128-bit format used by PowerPC (two adjacent doubles) is represented by
2553 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2554 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2555 will only work if they match the long double format on your target.
2556 The IEEE 16-bit format (half precision) is represented by ``0xH``
2557 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2558 (sign bit at the left).
2560 There are no constants of type x86_mmx.
2562 .. _complexconstants:
2567 Complex constants are a (potentially recursive) combination of simple
2568 constants and smaller complex constants.
2570 **Structure constants**
2571 Structure constants are represented with notation similar to
2572 structure type definitions (a comma separated list of elements,
2573 surrounded by braces (``{}``)). For example:
2574 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2575 "``@G = external global i32``". Structure constants must have
2576 :ref:`structure type <t_struct>`, and the number and types of elements
2577 must match those specified by the type.
2579 Array constants are represented with notation similar to array type
2580 definitions (a comma separated list of elements, surrounded by
2581 square brackets (``[]``)). For example:
2582 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2583 :ref:`array type <t_array>`, and the number and types of elements must
2584 match those specified by the type. As a special case, character array
2585 constants may also be represented as a double-quoted string using the ``c``
2586 prefix. For example: "``c"Hello World\0A\00"``".
2587 **Vector constants**
2588 Vector constants are represented with notation similar to vector
2589 type definitions (a comma separated list of elements, surrounded by
2590 less-than/greater-than's (``<>``)). For example:
2591 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2592 must have :ref:`vector type <t_vector>`, and the number and types of
2593 elements must match those specified by the type.
2594 **Zero initialization**
2595 The string '``zeroinitializer``' can be used to zero initialize a
2596 value to zero of *any* type, including scalar and
2597 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2598 having to print large zero initializers (e.g. for large arrays) and
2599 is always exactly equivalent to using explicit zero initializers.
2601 A metadata node is a constant tuple without types. For example:
2602 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2603 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2604 Unlike other typed constants that are meant to be interpreted as part of
2605 the instruction stream, metadata is a place to attach additional
2606 information such as debug info.
2608 Global Variable and Function Addresses
2609 --------------------------------------
2611 The addresses of :ref:`global variables <globalvars>` and
2612 :ref:`functions <functionstructure>` are always implicitly valid
2613 (link-time) constants. These constants are explicitly referenced when
2614 the :ref:`identifier for the global <identifiers>` is used and always have
2615 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2618 .. code-block:: llvm
2622 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2629 The string '``undef``' can be used anywhere a constant is expected, and
2630 indicates that the user of the value may receive an unspecified
2631 bit-pattern. Undefined values may be of any type (other than '``label``'
2632 or '``void``') and be used anywhere a constant is permitted.
2634 Undefined values are useful because they indicate to the compiler that
2635 the program is well defined no matter what value is used. This gives the
2636 compiler more freedom to optimize. Here are some examples of
2637 (potentially surprising) transformations that are valid (in pseudo IR):
2639 .. code-block:: llvm
2649 This is safe because all of the output bits are affected by the undef
2650 bits. Any output bit can have a zero or one depending on the input bits.
2652 .. code-block:: llvm
2663 These logical operations have bits that are not always affected by the
2664 input. For example, if ``%X`` has a zero bit, then the output of the
2665 '``and``' operation will always be a zero for that bit, no matter what
2666 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2667 optimize or assume that the result of the '``and``' is '``undef``'.
2668 However, it is safe to assume that all bits of the '``undef``' could be
2669 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2670 all the bits of the '``undef``' operand to the '``or``' could be set,
2671 allowing the '``or``' to be folded to -1.
2673 .. code-block:: llvm
2675 %A = select undef, %X, %Y
2676 %B = select undef, 42, %Y
2677 %C = select %X, %Y, undef
2687 This set of examples shows that undefined '``select``' (and conditional
2688 branch) conditions can go *either way*, but they have to come from one
2689 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2690 both known to have a clear low bit, then ``%A`` would have to have a
2691 cleared low bit. However, in the ``%C`` example, the optimizer is
2692 allowed to assume that the '``undef``' operand could be the same as
2693 ``%Y``, allowing the whole '``select``' to be eliminated.
2695 .. code-block:: llvm
2697 %A = xor undef, undef
2714 This example points out that two '``undef``' operands are not
2715 necessarily the same. This can be surprising to people (and also matches
2716 C semantics) where they assume that "``X^X``" is always zero, even if
2717 ``X`` is undefined. This isn't true for a number of reasons, but the
2718 short answer is that an '``undef``' "variable" can arbitrarily change
2719 its value over its "live range". This is true because the variable
2720 doesn't actually *have a live range*. Instead, the value is logically
2721 read from arbitrary registers that happen to be around when needed, so
2722 the value is not necessarily consistent over time. In fact, ``%A`` and
2723 ``%C`` need to have the same semantics or the core LLVM "replace all
2724 uses with" concept would not hold.
2726 .. code-block:: llvm
2734 These examples show the crucial difference between an *undefined value*
2735 and *undefined behavior*. An undefined value (like '``undef``') is
2736 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2737 operation can be constant folded to '``undef``', because the '``undef``'
2738 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2739 However, in the second example, we can make a more aggressive
2740 assumption: because the ``undef`` is allowed to be an arbitrary value,
2741 we are allowed to assume that it could be zero. Since a divide by zero
2742 has *undefined behavior*, we are allowed to assume that the operation
2743 does not execute at all. This allows us to delete the divide and all
2744 code after it. Because the undefined operation "can't happen", the
2745 optimizer can assume that it occurs in dead code.
2747 .. code-block:: llvm
2749 a: store undef -> %X
2750 b: store %X -> undef
2755 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2756 value can be assumed to not have any effect; we can assume that the
2757 value is overwritten with bits that happen to match what was already
2758 there. However, a store *to* an undefined location could clobber
2759 arbitrary memory, therefore, it has undefined behavior.
2766 Poison values are similar to :ref:`undef values <undefvalues>`, however
2767 they also represent the fact that an instruction or constant expression
2768 that cannot evoke side effects has nevertheless detected a condition
2769 that results in undefined behavior.
2771 There is currently no way of representing a poison value in the IR; they
2772 only exist when produced by operations such as :ref:`add <i_add>` with
2775 Poison value behavior is defined in terms of value *dependence*:
2777 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2778 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2779 their dynamic predecessor basic block.
2780 - Function arguments depend on the corresponding actual argument values
2781 in the dynamic callers of their functions.
2782 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2783 instructions that dynamically transfer control back to them.
2784 - :ref:`Invoke <i_invoke>` instructions depend on the
2785 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2786 call instructions that dynamically transfer control back to them.
2787 - Non-volatile loads and stores depend on the most recent stores to all
2788 of the referenced memory addresses, following the order in the IR
2789 (including loads and stores implied by intrinsics such as
2790 :ref:`@llvm.memcpy <int_memcpy>`.)
2791 - An instruction with externally visible side effects depends on the
2792 most recent preceding instruction with externally visible side
2793 effects, following the order in the IR. (This includes :ref:`volatile
2794 operations <volatile>`.)
2795 - An instruction *control-depends* on a :ref:`terminator
2796 instruction <terminators>` if the terminator instruction has
2797 multiple successors and the instruction is always executed when
2798 control transfers to one of the successors, and may not be executed
2799 when control is transferred to another.
2800 - Additionally, an instruction also *control-depends* on a terminator
2801 instruction if the set of instructions it otherwise depends on would
2802 be different if the terminator had transferred control to a different
2804 - Dependence is transitive.
2806 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2807 with the additional effect that any instruction that has a *dependence*
2808 on a poison value has undefined behavior.
2810 Here are some examples:
2812 .. code-block:: llvm
2815 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2816 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2817 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
2818 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2820 store i32 %poison, i32* @g ; Poison value stored to memory.
2821 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
2823 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2825 %narrowaddr = bitcast i32* @g to i16*
2826 %wideaddr = bitcast i32* @g to i64*
2827 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2828 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
2830 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2831 br i1 %cmp, label %true, label %end ; Branch to either destination.
2834 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2835 ; it has undefined behavior.
2839 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2840 ; Both edges into this PHI are
2841 ; control-dependent on %cmp, so this
2842 ; always results in a poison value.
2844 store volatile i32 0, i32* @g ; This would depend on the store in %true
2845 ; if %cmp is true, or the store in %entry
2846 ; otherwise, so this is undefined behavior.
2848 br i1 %cmp, label %second_true, label %second_end
2849 ; The same branch again, but this time the
2850 ; true block doesn't have side effects.
2857 store volatile i32 0, i32* @g ; This time, the instruction always depends
2858 ; on the store in %end. Also, it is
2859 ; control-equivalent to %end, so this is
2860 ; well-defined (ignoring earlier undefined
2861 ; behavior in this example).
2865 Addresses of Basic Blocks
2866 -------------------------
2868 ``blockaddress(@function, %block)``
2870 The '``blockaddress``' constant computes the address of the specified
2871 basic block in the specified function, and always has an ``i8*`` type.
2872 Taking the address of the entry block is illegal.
2874 This value only has defined behavior when used as an operand to the
2875 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2876 against null. Pointer equality tests between labels addresses results in
2877 undefined behavior --- though, again, comparison against null is ok, and
2878 no label is equal to the null pointer. This may be passed around as an
2879 opaque pointer sized value as long as the bits are not inspected. This
2880 allows ``ptrtoint`` and arithmetic to be performed on these values so
2881 long as the original value is reconstituted before the ``indirectbr``
2884 Finally, some targets may provide defined semantics when using the value
2885 as the operand to an inline assembly, but that is target specific.
2889 Constant Expressions
2890 --------------------
2892 Constant expressions are used to allow expressions involving other
2893 constants to be used as constants. Constant expressions may be of any
2894 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2895 that does not have side effects (e.g. load and call are not supported).
2896 The following is the syntax for constant expressions:
2898 ``trunc (CST to TYPE)``
2899 Truncate a constant to another type. The bit size of CST must be
2900 larger than the bit size of TYPE. Both types must be integers.
2901 ``zext (CST to TYPE)``
2902 Zero extend a constant to another type. The bit size of CST must be
2903 smaller than the bit size of TYPE. Both types must be integers.
2904 ``sext (CST to TYPE)``
2905 Sign extend a constant to another type. The bit size of CST must be
2906 smaller than the bit size of TYPE. Both types must be integers.
2907 ``fptrunc (CST to TYPE)``
2908 Truncate a floating point constant to another floating point type.
2909 The size of CST must be larger than the size of TYPE. Both types
2910 must be floating point.
2911 ``fpext (CST to TYPE)``
2912 Floating point extend a constant to another type. The size of CST
2913 must be smaller or equal to the size of TYPE. Both types must be
2915 ``fptoui (CST to TYPE)``
2916 Convert a floating point constant to the corresponding unsigned
2917 integer constant. TYPE must be a scalar or vector integer type. CST
2918 must be of scalar or vector floating point type. Both CST and TYPE
2919 must be scalars, or vectors of the same number of elements. If the
2920 value won't fit in the integer type, the results are undefined.
2921 ``fptosi (CST to TYPE)``
2922 Convert a floating point constant to the corresponding signed
2923 integer constant. TYPE must be a scalar or vector integer type. CST
2924 must be of scalar or vector floating point type. Both CST and TYPE
2925 must be scalars, or vectors of the same number of elements. If the
2926 value won't fit in the integer type, the results are undefined.
2927 ``uitofp (CST to TYPE)``
2928 Convert an unsigned integer constant to the corresponding floating
2929 point constant. TYPE must be a scalar or vector floating point type.
2930 CST must be of scalar or vector integer type. Both CST and TYPE must
2931 be scalars, or vectors of the same number of elements. If the value
2932 won't fit in the floating point type, the results are undefined.
2933 ``sitofp (CST to TYPE)``
2934 Convert a signed integer constant to the corresponding floating
2935 point constant. TYPE must be a scalar or vector floating point type.
2936 CST must be of scalar or vector integer type. Both CST and TYPE must
2937 be scalars, or vectors of the same number of elements. If the value
2938 won't fit in the floating point type, the results are undefined.
2939 ``ptrtoint (CST to TYPE)``
2940 Convert a pointer typed constant to the corresponding integer
2941 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2942 pointer type. The ``CST`` value is zero extended, truncated, or
2943 unchanged to make it fit in ``TYPE``.
2944 ``inttoptr (CST to TYPE)``
2945 Convert an integer constant to a pointer constant. TYPE must be a
2946 pointer type. CST must be of integer type. The CST value is zero
2947 extended, truncated, or unchanged to make it fit in a pointer size.
2948 This one is *really* dangerous!
2949 ``bitcast (CST to TYPE)``
2950 Convert a constant, CST, to another TYPE. The constraints of the
2951 operands are the same as those for the :ref:`bitcast
2952 instruction <i_bitcast>`.
2953 ``addrspacecast (CST to TYPE)``
2954 Convert a constant pointer or constant vector of pointer, CST, to another
2955 TYPE in a different address space. The constraints of the operands are the
2956 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2957 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
2958 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2959 constants. As with the :ref:`getelementptr <i_getelementptr>`
2960 instruction, the index list may have zero or more indexes, which are
2961 required to make sense for the type of "pointer to TY".
2962 ``select (COND, VAL1, VAL2)``
2963 Perform the :ref:`select operation <i_select>` on constants.
2964 ``icmp COND (VAL1, VAL2)``
2965 Performs the :ref:`icmp operation <i_icmp>` on constants.
2966 ``fcmp COND (VAL1, VAL2)``
2967 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2968 ``extractelement (VAL, IDX)``
2969 Perform the :ref:`extractelement operation <i_extractelement>` on
2971 ``insertelement (VAL, ELT, IDX)``
2972 Perform the :ref:`insertelement operation <i_insertelement>` on
2974 ``shufflevector (VEC1, VEC2, IDXMASK)``
2975 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2977 ``extractvalue (VAL, IDX0, IDX1, ...)``
2978 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2979 constants. The index list is interpreted in a similar manner as
2980 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2981 least one index value must be specified.
2982 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2983 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2984 The index list is interpreted in a similar manner as indices in a
2985 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2986 value must be specified.
2987 ``OPCODE (LHS, RHS)``
2988 Perform the specified operation of the LHS and RHS constants. OPCODE
2989 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2990 binary <bitwiseops>` operations. The constraints on operands are
2991 the same as those for the corresponding instruction (e.g. no bitwise
2992 operations on floating point values are allowed).
2999 Inline Assembler Expressions
3000 ----------------------------
3002 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
3003 Inline Assembly <moduleasm>`) through the use of a special value. This value
3004 represents the inline assembler as a template string (containing the
3005 instructions to emit), a list of operand constraints (stored as a string), a
3006 flag that indicates whether or not the inline asm expression has side effects,
3007 and a flag indicating whether the function containing the asm needs to align its
3008 stack conservatively.
3010 The template string supports argument substitution of the operands using "``$``"
3011 followed by a number, to indicate substitution of the given register/memory
3012 location, as specified by the constraint string. "``${NUM:MODIFIER}``" may also
3013 be used, where ``MODIFIER`` is a target-specific annotation for how to print the
3014 operand (See :ref:`inline-asm-modifiers`).
3016 A literal "``$``" may be included by using "``$$``" in the template. To include
3017 other special characters into the output, the usual "``\XX``" escapes may be
3018 used, just as in other strings. Note that after template substitution, the
3019 resulting assembly string is parsed by LLVM's integrated assembler unless it is
3020 disabled -- even when emitting a ``.s`` file -- and thus must contain assembly
3021 syntax known to LLVM.
3023 LLVM's support for inline asm is modeled closely on the requirements of Clang's
3024 GCC-compatible inline-asm support. Thus, the feature-set and the constraint and
3025 modifier codes listed here are similar or identical to those in GCC's inline asm
3026 support. However, to be clear, the syntax of the template and constraint strings
3027 described here is *not* the same as the syntax accepted by GCC and Clang, and,
3028 while most constraint letters are passed through as-is by Clang, some get
3029 translated to other codes when converting from the C source to the LLVM
3032 An example inline assembler expression is:
3034 .. code-block:: llvm
3036 i32 (i32) asm "bswap $0", "=r,r"
3038 Inline assembler expressions may **only** be used as the callee operand
3039 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
3040 Thus, typically we have:
3042 .. code-block:: llvm
3044 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
3046 Inline asms with side effects not visible in the constraint list must be
3047 marked as having side effects. This is done through the use of the
3048 '``sideeffect``' keyword, like so:
3050 .. code-block:: llvm
3052 call void asm sideeffect "eieio", ""()
3054 In some cases inline asms will contain code that will not work unless
3055 the stack is aligned in some way, such as calls or SSE instructions on
3056 x86, yet will not contain code that does that alignment within the asm.
3057 The compiler should make conservative assumptions about what the asm
3058 might contain and should generate its usual stack alignment code in the
3059 prologue if the '``alignstack``' keyword is present:
3061 .. code-block:: llvm
3063 call void asm alignstack "eieio", ""()
3065 Inline asms also support using non-standard assembly dialects. The
3066 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
3067 the inline asm is using the Intel dialect. Currently, ATT and Intel are
3068 the only supported dialects. An example is:
3070 .. code-block:: llvm
3072 call void asm inteldialect "eieio", ""()
3074 If multiple keywords appear the '``sideeffect``' keyword must come
3075 first, the '``alignstack``' keyword second and the '``inteldialect``'
3078 Inline Asm Constraint String
3079 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3081 The constraint list is a comma-separated string, each element containing one or
3082 more constraint codes.
3084 For each element in the constraint list an appropriate register or memory
3085 operand will be chosen, and it will be made available to assembly template
3086 string expansion as ``$0`` for the first constraint in the list, ``$1`` for the
3089 There are three different types of constraints, which are distinguished by a
3090 prefix symbol in front of the constraint code: Output, Input, and Clobber. The
3091 constraints must always be given in that order: outputs first, then inputs, then
3092 clobbers. They cannot be intermingled.
3094 There are also three different categories of constraint codes:
3096 - Register constraint. This is either a register class, or a fixed physical
3097 register. This kind of constraint will allocate a register, and if necessary,
3098 bitcast the argument or result to the appropriate type.
3099 - Memory constraint. This kind of constraint is for use with an instruction
3100 taking a memory operand. Different constraints allow for different addressing
3101 modes used by the target.
3102 - Immediate value constraint. This kind of constraint is for an integer or other
3103 immediate value which can be rendered directly into an instruction. The
3104 various target-specific constraints allow the selection of a value in the
3105 proper range for the instruction you wish to use it with.
3110 Output constraints are specified by an "``=``" prefix (e.g. "``=r``"). This
3111 indicates that the assembly will write to this operand, and the operand will
3112 then be made available as a return value of the ``asm`` expression. Output
3113 constraints do not consume an argument from the call instruction. (Except, see
3114 below about indirect outputs).
3116 Normally, it is expected that no output locations are written to by the assembly
3117 expression until *all* of the inputs have been read. As such, LLVM may assign
3118 the same register to an output and an input. If this is not safe (e.g. if the
3119 assembly contains two instructions, where the first writes to one output, and
3120 the second reads an input and writes to a second output), then the "``&``"
3121 modifier must be used (e.g. "``=&r``") to specify that the output is an
3122 "early-clobber" output. Marking an ouput as "early-clobber" ensures that LLVM
3123 will not use the same register for any inputs (other than an input tied to this
3129 Input constraints do not have a prefix -- just the constraint codes. Each input
3130 constraint will consume one argument from the call instruction. It is not
3131 permitted for the asm to write to any input register or memory location (unless
3132 that input is tied to an output). Note also that multiple inputs may all be
3133 assigned to the same register, if LLVM can determine that they necessarily all
3134 contain the same value.
3136 Instead of providing a Constraint Code, input constraints may also "tie"
3137 themselves to an output constraint, by providing an integer as the constraint
3138 string. Tied inputs still consume an argument from the call instruction, and
3139 take up a position in the asm template numbering as is usual -- they will simply
3140 be constrained to always use the same register as the output they've been tied
3141 to. For example, a constraint string of "``=r,0``" says to assign a register for
3142 output, and use that register as an input as well (it being the 0'th
3145 It is permitted to tie an input to an "early-clobber" output. In that case, no
3146 *other* input may share the same register as the input tied to the early-clobber
3147 (even when the other input has the same value).
3149 You may only tie an input to an output which has a register constraint, not a
3150 memory constraint. Only a single input may be tied to an output.
3152 There is also an "interesting" feature which deserves a bit of explanation: if a
3153 register class constraint allocates a register which is too small for the value
3154 type operand provided as input, the input value will be split into multiple
3155 registers, and all of them passed to the inline asm.
3157 However, this feature is often not as useful as you might think.
3159 Firstly, the registers are *not* guaranteed to be consecutive. So, on those
3160 architectures that have instructions which operate on multiple consecutive
3161 instructions, this is not an appropriate way to support them. (e.g. the 32-bit
3162 SparcV8 has a 64-bit load, which instruction takes a single 32-bit register. The
3163 hardware then loads into both the named register, and the next register. This
3164 feature of inline asm would not be useful to support that.)
3166 A few of the targets provide a template string modifier allowing explicit access
3167 to the second register of a two-register operand (e.g. MIPS ``L``, ``M``, and
3168 ``D``). On such an architecture, you can actually access the second allocated
3169 register (yet, still, not any subsequent ones). But, in that case, you're still
3170 probably better off simply splitting the value into two separate operands, for
3171 clarity. (e.g. see the description of the ``A`` constraint on X86, which,
3172 despite existing only for use with this feature, is not really a good idea to
3175 Indirect inputs and outputs
3176 """""""""""""""""""""""""""
3178 Indirect output or input constraints can be specified by the "``*``" modifier
3179 (which goes after the "``=``" in case of an output). This indicates that the asm
3180 will write to or read from the contents of an *address* provided as an input
3181 argument. (Note that in this way, indirect outputs act more like an *input* than
3182 an output: just like an input, they consume an argument of the call expression,
3183 rather than producing a return value. An indirect output constraint is an
3184 "output" only in that the asm is expected to write to the contents of the input
3185 memory location, instead of just read from it).
3187 This is most typically used for memory constraint, e.g. "``=*m``", to pass the
3188 address of a variable as a value.
3190 It is also possible to use an indirect *register* constraint, but only on output
3191 (e.g. "``=*r``"). This will cause LLVM to allocate a register for an output
3192 value normally, and then, separately emit a store to the address provided as
3193 input, after the provided inline asm. (It's not clear what value this
3194 functionality provides, compared to writing the store explicitly after the asm
3195 statement, and it can only produce worse code, since it bypasses many
3196 optimization passes. I would recommend not using it.)
3202 A clobber constraint is indicated by a "``~``" prefix. A clobber does not
3203 consume an input operand, nor generate an output. Clobbers cannot use any of the
3204 general constraint code letters -- they may use only explicit register
3205 constraints, e.g. "``~{eax}``". The one exception is that a clobber string of
3206 "``~{memory}``" indicates that the assembly writes to arbitrary undeclared
3207 memory locations -- not only the memory pointed to by a declared indirect
3213 After a potential prefix comes constraint code, or codes.
3215 A Constraint Code is either a single letter (e.g. "``r``"), a "``^``" character
3216 followed by two letters (e.g. "``^wc``"), or "``{``" register-name "``}``"
3219 The one and two letter constraint codes are typically chosen to be the same as
3220 GCC's constraint codes.
3222 A single constraint may include one or more than constraint code in it, leaving
3223 it up to LLVM to choose which one to use. This is included mainly for
3224 compatibility with the translation of GCC inline asm coming from clang.
3226 There are two ways to specify alternatives, and either or both may be used in an
3227 inline asm constraint list:
3229 1) Append the codes to each other, making a constraint code set. E.g. "``im``"
3230 or "``{eax}m``". This means "choose any of the options in the set". The
3231 choice of constraint is made independently for each constraint in the
3234 2) Use "``|``" between constraint code sets, creating alternatives. Every
3235 constraint in the constraint list must have the same number of alternative
3236 sets. With this syntax, the same alternative in *all* of the items in the
3237 constraint list will be chosen together.
3239 Putting those together, you might have a two operand constraint string like
3240 ``"rm|r,ri|rm"``. This indicates that if operand 0 is ``r`` or ``m``, then
3241 operand 1 may be one of ``r`` or ``i``. If operand 0 is ``r``, then operand 1
3242 may be one of ``r`` or ``m``. But, operand 0 and 1 cannot both be of type m.
3244 However, the use of either of the alternatives features is *NOT* recommended, as
3245 LLVM is not able to make an intelligent choice about which one to use. (At the
3246 point it currently needs to choose, not enough information is available to do so
3247 in a smart way.) Thus, it simply tries to make a choice that's most likely to
3248 compile, not one that will be optimal performance. (e.g., given "``rm``", it'll
3249 always choose to use memory, not registers). And, if given multiple registers,
3250 or multiple register classes, it will simply choose the first one. (In fact, it
3251 doesn't currently even ensure explicitly specified physical registers are
3252 unique, so specifying multiple physical registers as alternatives, like
3253 ``{r11}{r12},{r11}{r12}``, will assign r11 to both operands, not at all what was
3256 Supported Constraint Code List
3257 """"""""""""""""""""""""""""""
3259 The constraint codes are, in general, expected to behave the same way they do in
3260 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3261 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3262 and GCC likely indicates a bug in LLVM.
3264 Some constraint codes are typically supported by all targets:
3266 - ``r``: A register in the target's general purpose register class.
3267 - ``m``: A memory address operand. It is target-specific what addressing modes
3268 are supported, typical examples are register, or register + register offset,
3269 or register + immediate offset (of some target-specific size).
3270 - ``i``: An integer constant (of target-specific width). Allows either a simple
3271 immediate, or a relocatable value.
3272 - ``n``: An integer constant -- *not* including relocatable values.
3273 - ``s``: An integer constant, but allowing *only* relocatable values.
3274 - ``X``: Allows an operand of any kind, no constraint whatsoever. Typically
3275 useful to pass a label for an asm branch or call.
3277 .. FIXME: but that surely isn't actually okay to jump out of an asm
3278 block without telling llvm about the control transfer???)
3280 - ``{register-name}``: Requires exactly the named physical register.
3282 Other constraints are target-specific:
3286 - ``z``: An immediate integer 0. Outputs ``WZR`` or ``XZR``, as appropriate.
3287 - ``I``: An immediate integer valid for an ``ADD`` or ``SUB`` instruction,
3288 i.e. 0 to 4095 with optional shift by 12.
3289 - ``J``: An immediate integer that, when negated, is valid for an ``ADD`` or
3290 ``SUB`` instruction, i.e. -1 to -4095 with optional left shift by 12.
3291 - ``K``: An immediate integer that is valid for the 'bitmask immediate 32' of a
3292 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 32-bit register.
3293 - ``L``: An immediate integer that is valid for the 'bitmask immediate 64' of a
3294 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 64-bit register.
3295 - ``M``: An immediate integer for use with the ``MOV`` assembly alias on a
3296 32-bit register. This is a superset of ``K``: in addition to the bitmask
3297 immediate, also allows immediate integers which can be loaded with a single
3298 ``MOVZ`` or ``MOVL`` instruction.
3299 - ``N``: An immediate integer for use with the ``MOV`` assembly alias on a
3300 64-bit register. This is a superset of ``L``.
3301 - ``Q``: Memory address operand must be in a single register (no
3302 offsets). (However, LLVM currently does this for the ``m`` constraint as
3304 - ``r``: A 32 or 64-bit integer register (W* or X*).
3305 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register.
3306 - ``x``: A lower 128-bit floating-point/SIMD register (``V0`` to ``V15``).
3310 - ``r``: A 32 or 64-bit integer register.
3311 - ``[0-9]v``: The 32-bit VGPR register, number 0-9.
3312 - ``[0-9]s``: The 32-bit SGPR register, number 0-9.
3317 - ``Q``, ``Um``, ``Un``, ``Uq``, ``Us``, ``Ut``, ``Uv``, ``Uy``: Memory address
3318 operand. Treated the same as operand ``m``, at the moment.
3320 ARM and ARM's Thumb2 mode:
3322 - ``j``: An immediate integer between 0 and 65535 (valid for ``MOVW``)
3323 - ``I``: An immediate integer valid for a data-processing instruction.
3324 - ``J``: An immediate integer between -4095 and 4095.
3325 - ``K``: An immediate integer whose bitwise inverse is valid for a
3326 data-processing instruction. (Can be used with template modifier "``B``" to
3327 print the inverted value).
3328 - ``L``: An immediate integer whose negation is valid for a data-processing
3329 instruction. (Can be used with template modifier "``n``" to print the negated
3331 - ``M``: A power of two or a integer between 0 and 32.
3332 - ``N``: Invalid immediate constraint.
3333 - ``O``: Invalid immediate constraint.
3334 - ``r``: A general-purpose 32-bit integer register (``r0-r15``).
3335 - ``l``: In Thumb2 mode, low 32-bit GPR registers (``r0-r7``). In ARM mode, same
3337 - ``h``: In Thumb2 mode, a high 32-bit GPR register (``r8-r15``). In ARM mode,
3339 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3340 ``d0-d31``, or ``q0-q15``.
3341 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3342 ``d0-d7``, or ``q0-q3``.
3343 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3348 - ``I``: An immediate integer between 0 and 255.
3349 - ``J``: An immediate integer between -255 and -1.
3350 - ``K``: An immediate integer between 0 and 255, with optional left-shift by
3352 - ``L``: An immediate integer between -7 and 7.
3353 - ``M``: An immediate integer which is a multiple of 4 between 0 and 1020.
3354 - ``N``: An immediate integer between 0 and 31.
3355 - ``O``: An immediate integer which is a multiple of 4 between -508 and 508.
3356 - ``r``: A low 32-bit GPR register (``r0-r7``).
3357 - ``l``: A low 32-bit GPR register (``r0-r7``).
3358 - ``h``: A high GPR register (``r0-r7``).
3359 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3360 ``d0-d31``, or ``q0-q15``.
3361 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3362 ``d0-d7``, or ``q0-q3``.
3363 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3369 - ``o``, ``v``: A memory address operand, treated the same as constraint ``m``,
3371 - ``r``: A 32 or 64-bit register.
3375 - ``r``: An 8 or 16-bit register.
3379 - ``I``: An immediate signed 16-bit integer.
3380 - ``J``: An immediate integer zero.
3381 - ``K``: An immediate unsigned 16-bit integer.
3382 - ``L``: An immediate 32-bit integer, where the lower 16 bits are 0.
3383 - ``N``: An immediate integer between -65535 and -1.
3384 - ``O``: An immediate signed 15-bit integer.
3385 - ``P``: An immediate integer between 1 and 65535.
3386 - ``m``: A memory address operand. In MIPS-SE mode, allows a base address
3387 register plus 16-bit immediate offset. In MIPS mode, just a base register.
3388 - ``R``: A memory address operand. In MIPS-SE mode, allows a base address
3389 register plus a 9-bit signed offset. In MIPS mode, the same as constraint
3391 - ``ZC``: A memory address operand, suitable for use in a ``pref``, ``ll``, or
3392 ``sc`` instruction on the given subtarget (details vary).
3393 - ``r``, ``d``, ``y``: A 32 or 64-bit GPR register.
3394 - ``f``: A 32 or 64-bit FPU register (``F0-F31``), or a 128-bit MSA register
3395 (``W0-W31``). In the case of MSA registers, it is recommended to use the ``w``
3396 argument modifier for compatibility with GCC.
3397 - ``c``: A 32-bit or 64-bit GPR register suitable for indirect jump (always
3399 - ``l``: The ``lo`` register, 32 or 64-bit.
3404 - ``b``: A 1-bit integer register.
3405 - ``c`` or ``h``: A 16-bit integer register.
3406 - ``r``: A 32-bit integer register.
3407 - ``l`` or ``N``: A 64-bit integer register.
3408 - ``f``: A 32-bit float register.
3409 - ``d``: A 64-bit float register.
3414 - ``I``: An immediate signed 16-bit integer.
3415 - ``J``: An immediate unsigned 16-bit integer, shifted left 16 bits.
3416 - ``K``: An immediate unsigned 16-bit integer.
3417 - ``L``: An immediate signed 16-bit integer, shifted left 16 bits.
3418 - ``M``: An immediate integer greater than 31.
3419 - ``N``: An immediate integer that is an exact power of 2.
3420 - ``O``: The immediate integer constant 0.
3421 - ``P``: An immediate integer constant whose negation is a signed 16-bit
3423 - ``es``, ``o``, ``Q``, ``Z``, ``Zy``: A memory address operand, currently
3424 treated the same as ``m``.
3425 - ``r``: A 32 or 64-bit integer register.
3426 - ``b``: A 32 or 64-bit integer register, excluding ``R0`` (that is:
3428 - ``f``: A 32 or 64-bit float register (``F0-F31``), or when QPX is enabled, a
3429 128 or 256-bit QPX register (``Q0-Q31``; aliases the ``F`` registers).
3430 - ``v``: For ``4 x f32`` or ``4 x f64`` types, when QPX is enabled, a
3431 128 or 256-bit QPX register (``Q0-Q31``), otherwise a 128-bit
3432 altivec vector register (``V0-V31``).
3434 .. FIXME: is this a bug that v accepts QPX registers? I think this
3435 is supposed to only use the altivec vector registers?
3437 - ``y``: Condition register (``CR0-CR7``).
3438 - ``wc``: An individual CR bit in a CR register.
3439 - ``wa``, ``wd``, ``wf``: Any 128-bit VSX vector register, from the full VSX
3440 register set (overlapping both the floating-point and vector register files).
3441 - ``ws``: A 32 or 64-bit floating point register, from the full VSX register
3446 - ``I``: An immediate 13-bit signed integer.
3447 - ``r``: A 32-bit integer register.
3451 - ``I``: An immediate unsigned 8-bit integer.
3452 - ``J``: An immediate unsigned 12-bit integer.
3453 - ``K``: An immediate signed 16-bit integer.
3454 - ``L``: An immediate signed 20-bit integer.
3455 - ``M``: An immediate integer 0x7fffffff.
3456 - ``Q``, ``R``, ``S``, ``T``: A memory address operand, treated the same as
3457 ``m``, at the moment.
3458 - ``r`` or ``d``: A 32, 64, or 128-bit integer register.
3459 - ``a``: A 32, 64, or 128-bit integer address register (excludes R0, which in an
3460 address context evaluates as zero).
3461 - ``h``: A 32-bit value in the high part of a 64bit data register
3463 - ``f``: A 32, 64, or 128-bit floating point register.
3467 - ``I``: An immediate integer between 0 and 31.
3468 - ``J``: An immediate integer between 0 and 64.
3469 - ``K``: An immediate signed 8-bit integer.
3470 - ``L``: An immediate integer, 0xff or 0xffff or (in 64-bit mode only)
3472 - ``M``: An immediate integer between 0 and 3.
3473 - ``N``: An immediate unsigned 8-bit integer.
3474 - ``O``: An immediate integer between 0 and 127.
3475 - ``e``: An immediate 32-bit signed integer.
3476 - ``Z``: An immediate 32-bit unsigned integer.
3477 - ``o``, ``v``: Treated the same as ``m``, at the moment.
3478 - ``q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3479 ``l`` integer register. On X86-32, this is the ``a``, ``b``, ``c``, and ``d``
3480 registers, and on X86-64, it is all of the integer registers.
3481 - ``Q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3482 ``h`` integer register. This is the ``a``, ``b``, ``c``, and ``d`` registers.
3483 - ``r`` or ``l``: An 8, 16, 32, or 64-bit integer register.
3484 - ``R``: An 8, 16, 32, or 64-bit "legacy" integer register -- one which has
3485 existed since i386, and can be accessed without the REX prefix.
3486 - ``f``: A 32, 64, or 80-bit '387 FPU stack pseudo-register.
3487 - ``y``: A 64-bit MMX register, if MMX is enabled.
3488 - ``x``: If SSE is enabled: a 32 or 64-bit scalar operand, or 128-bit vector
3489 operand in a SSE register. If AVX is also enabled, can also be a 256-bit
3490 vector operand in an AVX register. If AVX-512 is also enabled, can also be a
3491 512-bit vector operand in an AVX512 register, Otherwise, an error.
3492 - ``Y``: The same as ``x``, if *SSE2* is enabled, otherwise an error.
3493 - ``A``: Special case: allocates EAX first, then EDX, for a single operand (in
3494 32-bit mode, a 64-bit integer operand will get split into two registers). It
3495 is not recommended to use this constraint, as in 64-bit mode, the 64-bit
3496 operand will get allocated only to RAX -- if two 32-bit operands are needed,
3497 you're better off splitting it yourself, before passing it to the asm
3502 - ``r``: A 32-bit integer register.
3505 .. _inline-asm-modifiers:
3507 Asm template argument modifiers
3508 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3510 In the asm template string, modifiers can be used on the operand reference, like
3513 The modifiers are, in general, expected to behave the same way they do in
3514 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3515 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3516 and GCC likely indicates a bug in LLVM.
3520 - ``c``: Print an immediate integer constant unadorned, without
3521 the target-specific immediate punctuation (e.g. no ``$`` prefix).
3522 - ``n``: Negate and print immediate integer constant unadorned, without the
3523 target-specific immediate punctuation (e.g. no ``$`` prefix).
3524 - ``l``: Print as an unadorned label, without the target-specific label
3525 punctuation (e.g. no ``$`` prefix).
3529 - ``w``: Print a GPR register with a ``w*`` name instead of ``x*`` name. E.g.,
3530 instead of ``x30``, print ``w30``.
3531 - ``x``: Print a GPR register with a ``x*`` name. (this is the default, anyhow).
3532 - ``b``, ``h``, ``s``, ``d``, ``q``: Print a floating-point/SIMD register with a
3533 ``b*``, ``h*``, ``s*``, ``d*``, or ``q*`` name, rather than the default of
3542 - ``a``: Print an operand as an address (with ``[`` and ``]`` surrounding a
3546 - ``y``: Print a VFP single-precision register as an indexed double (e.g. print
3547 as ``d4[1]`` instead of ``s9``)
3548 - ``B``: Bitwise invert and print an immediate integer constant without ``#``
3550 - ``L``: Print the low 16-bits of an immediate integer constant.
3551 - ``M``: Print as a register set suitable for ldm/stm. Also prints *all*
3552 register operands subsequent to the specified one (!), so use carefully.
3553 - ``Q``: Print the low-order register of a register-pair, or the low-order
3554 register of a two-register operand.
3555 - ``R``: Print the high-order register of a register-pair, or the high-order
3556 register of a two-register operand.
3557 - ``H``: Print the second register of a register-pair. (On a big-endian system,
3558 ``H`` is equivalent to ``Q``, and on little-endian system, ``H`` is equivalent
3561 .. FIXME: H doesn't currently support printing the second register
3562 of a two-register operand.
3564 - ``e``: Print the low doubleword register of a NEON quad register.
3565 - ``f``: Print the high doubleword register of a NEON quad register.
3566 - ``m``: Print the base register of a memory operand without the ``[`` and ``]``
3571 - ``L``: Print the second register of a two-register operand. Requires that it
3572 has been allocated consecutively to the first.
3574 .. FIXME: why is it restricted to consecutive ones? And there's
3575 nothing that ensures that happens, is there?
3577 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3578 nothing. Used to print 'addi' vs 'add' instructions.
3582 No additional modifiers.
3586 - ``X``: Print an immediate integer as hexadecimal
3587 - ``x``: Print the low 16 bits of an immediate integer as hexadecimal.
3588 - ``d``: Print an immediate integer as decimal.
3589 - ``m``: Subtract one and print an immediate integer as decimal.
3590 - ``z``: Print $0 if an immediate zero, otherwise print normally.
3591 - ``L``: Print the low-order register of a two-register operand, or prints the
3592 address of the low-order word of a double-word memory operand.
3594 .. FIXME: L seems to be missing memory operand support.
3596 - ``M``: Print the high-order register of a two-register operand, or prints the
3597 address of the high-order word of a double-word memory operand.
3599 .. FIXME: M seems to be missing memory operand support.
3601 - ``D``: Print the second register of a two-register operand, or prints the
3602 second word of a double-word memory operand. (On a big-endian system, ``D`` is
3603 equivalent to ``L``, and on little-endian system, ``D`` is equivalent to
3605 - ``w``: No effect. Provided for compatibility with GCC which requires this
3606 modifier in order to print MSA registers (``W0-W31``) with the ``f``
3615 - ``L``: Print the second register of a two-register operand. Requires that it
3616 has been allocated consecutively to the first.
3618 .. FIXME: why is it restricted to consecutive ones? And there's
3619 nothing that ensures that happens, is there?
3621 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3622 nothing. Used to print 'addi' vs 'add' instructions.
3623 - ``y``: For a memory operand, prints formatter for a two-register X-form
3624 instruction. (Currently always prints ``r0,OPERAND``).
3625 - ``U``: Prints 'u' if the memory operand is an update form, and nothing
3626 otherwise. (NOTE: LLVM does not support update form, so this will currently
3627 always print nothing)
3628 - ``X``: Prints 'x' if the memory operand is an indexed form. (NOTE: LLVM does
3629 not support indexed form, so this will currently always print nothing)
3637 SystemZ implements only ``n``, and does *not* support any of the other
3638 target-independent modifiers.
3642 - ``c``: Print an unadorned integer or symbol name. (The latter is
3643 target-specific behavior for this typically target-independent modifier).
3644 - ``A``: Print a register name with a '``*``' before it.
3645 - ``b``: Print an 8-bit register name (e.g. ``al``); do nothing on a memory
3647 - ``h``: Print the upper 8-bit register name (e.g. ``ah``); do nothing on a
3649 - ``w``: Print the 16-bit register name (e.g. ``ax``); do nothing on a memory
3651 - ``k``: Print the 32-bit register name (e.g. ``eax``); do nothing on a memory
3653 - ``q``: Print the 64-bit register name (e.g. ``rax``), if 64-bit registers are
3654 available, otherwise the 32-bit register name; do nothing on a memory operand.
3655 - ``n``: Negate and print an unadorned integer, or, for operands other than an
3656 immediate integer (e.g. a relocatable symbol expression), print a '-' before
3657 the operand. (The behavior for relocatable symbol expressions is a
3658 target-specific behavior for this typically target-independent modifier)
3659 - ``H``: Print a memory reference with additional offset +8.
3660 - ``P``: Print a memory reference or operand for use as the argument of a call
3661 instruction. (E.g. omit ``(rip)``, even though it's PC-relative.)
3665 No additional modifiers.
3671 The call instructions that wrap inline asm nodes may have a
3672 "``!srcloc``" MDNode attached to it that contains a list of constant
3673 integers. If present, the code generator will use the integer as the
3674 location cookie value when report errors through the ``LLVMContext``
3675 error reporting mechanisms. This allows a front-end to correlate backend
3676 errors that occur with inline asm back to the source code that produced
3679 .. code-block:: llvm
3681 call void asm sideeffect "something bad", ""(), !srcloc !42
3683 !42 = !{ i32 1234567 }
3685 It is up to the front-end to make sense of the magic numbers it places
3686 in the IR. If the MDNode contains multiple constants, the code generator
3687 will use the one that corresponds to the line of the asm that the error
3695 LLVM IR allows metadata to be attached to instructions in the program
3696 that can convey extra information about the code to the optimizers and
3697 code generator. One example application of metadata is source-level
3698 debug information. There are two metadata primitives: strings and nodes.
3700 Metadata does not have a type, and is not a value. If referenced from a
3701 ``call`` instruction, it uses the ``metadata`` type.
3703 All metadata are identified in syntax by a exclamation point ('``!``').
3705 .. _metadata-string:
3707 Metadata Nodes and Metadata Strings
3708 -----------------------------------
3710 A metadata string is a string surrounded by double quotes. It can
3711 contain any character by escaping non-printable characters with
3712 "``\xx``" where "``xx``" is the two digit hex code. For example:
3715 Metadata nodes are represented with notation similar to structure
3716 constants (a comma separated list of elements, surrounded by braces and
3717 preceded by an exclamation point). Metadata nodes can have any values as
3718 their operand. For example:
3720 .. code-block:: llvm
3722 !{ !"test\00", i32 10}
3724 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
3726 .. code-block:: llvm
3728 !0 = distinct !{!"test\00", i32 10}
3730 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
3731 content. They can also occur when transformations cause uniquing collisions
3732 when metadata operands change.
3734 A :ref:`named metadata <namedmetadatastructure>` is a collection of
3735 metadata nodes, which can be looked up in the module symbol table. For
3738 .. code-block:: llvm
3742 Metadata can be used as function arguments. Here ``llvm.dbg.value``
3743 function is using two metadata arguments:
3745 .. code-block:: llvm
3747 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
3749 Metadata can be attached to an instruction. Here metadata ``!21`` is attached
3750 to the ``add`` instruction using the ``!dbg`` identifier:
3752 .. code-block:: llvm
3754 %indvar.next = add i64 %indvar, 1, !dbg !21
3756 Metadata can also be attached to a function definition. Here metadata ``!22``
3757 is attached to the ``foo`` function using the ``!dbg`` identifier:
3759 .. code-block:: llvm
3761 define void @foo() !dbg !22 {
3765 More information about specific metadata nodes recognized by the
3766 optimizers and code generator is found below.
3768 .. _specialized-metadata:
3770 Specialized Metadata Nodes
3771 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3773 Specialized metadata nodes are custom data structures in metadata (as opposed
3774 to generic tuples). Their fields are labelled, and can be specified in any
3777 These aren't inherently debug info centric, but currently all the specialized
3778 metadata nodes are related to debug info.
3785 ``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
3786 ``retainedTypes:``, ``subprograms:``, ``globals:``, ``imports:`` and ``macros:``
3787 fields are tuples containing the debug info to be emitted along with the compile
3788 unit, regardless of code optimizations (some nodes are only emitted if there are
3789 references to them from instructions).
3791 .. code-block:: llvm
3793 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
3794 isOptimized: true, flags: "-O2", runtimeVersion: 2,
3795 splitDebugFilename: "abc.debug", emissionKind: 1,
3796 enums: !2, retainedTypes: !3, subprograms: !4,
3797 globals: !5, imports: !6, macros: !7, dwoId: 0x0abcd)
3799 Compile unit descriptors provide the root scope for objects declared in a
3800 specific compilation unit. File descriptors are defined using this scope.
3801 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
3802 keep track of subprograms, global variables, type information, and imported
3803 entities (declarations and namespaces).
3810 ``DIFile`` nodes represent files. The ``filename:`` can include slashes.
3812 .. code-block:: llvm
3814 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
3816 Files are sometimes used in ``scope:`` fields, and are the only valid target
3817 for ``file:`` fields.
3824 ``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
3825 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
3827 .. code-block:: llvm
3829 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3830 encoding: DW_ATE_unsigned_char)
3831 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
3833 The ``encoding:`` describes the details of the type. Usually it's one of the
3836 .. code-block:: llvm
3842 DW_ATE_signed_char = 6
3844 DW_ATE_unsigned_char = 8
3846 .. _DISubroutineType:
3851 ``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
3852 refers to a tuple; the first operand is the return type, while the rest are the
3853 types of the formal arguments in order. If the first operand is ``null``, that
3854 represents a function with no return value (such as ``void foo() {}`` in C++).
3856 .. code-block:: llvm
3858 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
3859 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
3860 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
3867 ``DIDerivedType`` nodes represent types derived from other types, such as
3870 .. code-block:: llvm
3872 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3873 encoding: DW_ATE_unsigned_char)
3874 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
3877 The following ``tag:`` values are valid:
3879 .. code-block:: llvm
3881 DW_TAG_formal_parameter = 5
3883 DW_TAG_pointer_type = 15
3884 DW_TAG_reference_type = 16
3886 DW_TAG_ptr_to_member_type = 31
3887 DW_TAG_const_type = 38
3888 DW_TAG_volatile_type = 53
3889 DW_TAG_restrict_type = 55
3891 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
3892 <DICompositeType>` or :ref:`subprogram <DISubprogram>`. The type of the member
3893 is the ``baseType:``. The ``offset:`` is the member's bit offset.
3894 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
3895 argument of a subprogram.
3897 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3899 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3900 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3903 Note that the ``void *`` type is expressed as a type derived from NULL.
3905 .. _DICompositeType:
3910 ``DICompositeType`` nodes represent types composed of other types, like
3911 structures and unions. ``elements:`` points to a tuple of the composed types.
3913 If the source language supports ODR, the ``identifier:`` field gives the unique
3914 identifier used for type merging between modules. When specified, other types
3915 can refer to composite types indirectly via a :ref:`metadata string
3916 <metadata-string>` that matches their identifier.
3918 .. code-block:: llvm
3920 !0 = !DIEnumerator(name: "SixKind", value: 7)
3921 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3922 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3923 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
3924 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3925 elements: !{!0, !1, !2})
3927 The following ``tag:`` values are valid:
3929 .. code-block:: llvm
3931 DW_TAG_array_type = 1
3932 DW_TAG_class_type = 2
3933 DW_TAG_enumeration_type = 4
3934 DW_TAG_structure_type = 19
3935 DW_TAG_union_type = 23
3936 DW_TAG_subroutine_type = 21
3937 DW_TAG_inheritance = 28
3940 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
3941 descriptors <DISubrange>`, each representing the range of subscripts at that
3942 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3943 array type is a native packed vector.
3945 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
3946 descriptors <DIEnumerator>`, each representing the definition of an enumeration
3947 value for the set. All enumeration type descriptors are collected in the
3948 ``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
3950 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3951 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
3952 <DIDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
3959 ``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3960 :ref:`DICompositeType`. ``count: -1`` indicates an empty array.
3962 .. code-block:: llvm
3964 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
3965 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
3966 !2 = !DISubrange(count: -1) ; empty array.
3973 ``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3974 variants of :ref:`DICompositeType`.
3976 .. code-block:: llvm
3978 !0 = !DIEnumerator(name: "SixKind", value: 7)
3979 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3980 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3982 DITemplateTypeParameter
3983 """""""""""""""""""""""
3985 ``DITemplateTypeParameter`` nodes represent type parameters to generic source
3986 language constructs. They are used (optionally) in :ref:`DICompositeType` and
3987 :ref:`DISubprogram` ``templateParams:`` fields.
3989 .. code-block:: llvm
3991 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
3993 DITemplateValueParameter
3994 """"""""""""""""""""""""
3996 ``DITemplateValueParameter`` nodes represent value parameters to generic source
3997 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3998 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3999 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
4000 :ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
4002 .. code-block:: llvm
4004 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
4009 ``DINamespace`` nodes represent namespaces in the source language.
4011 .. code-block:: llvm
4013 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
4018 ``DIGlobalVariable`` nodes represent global variables in the source language.
4020 .. code-block:: llvm
4022 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
4023 file: !2, line: 7, type: !3, isLocal: true,
4024 isDefinition: false, variable: i32* @foo,
4027 All global variables should be referenced by the `globals:` field of a
4028 :ref:`compile unit <DICompileUnit>`.
4035 ``DISubprogram`` nodes represent functions from the source language. A
4036 ``DISubprogram`` may be attached to a function definition using ``!dbg``
4037 metadata. The ``variables:`` field points at :ref:`variables <DILocalVariable>`
4038 that must be retained, even if their IR counterparts are optimized out of
4039 the IR. The ``type:`` field must point at an :ref:`DISubroutineType`.
4041 .. code-block:: llvm
4043 define void @_Z3foov() !dbg !0 {
4047 !0 = distinct !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
4048 file: !2, line: 7, type: !3, isLocal: true,
4049 isDefinition: false, scopeLine: 8,
4051 virtuality: DW_VIRTUALITY_pure_virtual,
4052 virtualIndex: 10, flags: DIFlagPrototyped,
4053 isOptimized: true, templateParams: !5,
4054 declaration: !6, variables: !7)
4061 ``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
4062 <DISubprogram>`. The line number and column numbers are used to distinguish
4063 two lexical blocks at same depth. They are valid targets for ``scope:``
4066 .. code-block:: llvm
4068 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
4070 Usually lexical blocks are ``distinct`` to prevent node merging based on
4073 .. _DILexicalBlockFile:
4078 ``DILexicalBlockFile`` nodes are used to discriminate between sections of a
4079 :ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
4080 indicate textual inclusion, or the ``discriminator:`` field can be used to
4081 discriminate between control flow within a single block in the source language.
4083 .. code-block:: llvm
4085 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
4086 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
4087 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
4094 ``DILocation`` nodes represent source debug locations. The ``scope:`` field is
4095 mandatory, and points at an :ref:`DILexicalBlockFile`, an
4096 :ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
4098 .. code-block:: llvm
4100 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
4102 .. _DILocalVariable:
4107 ``DILocalVariable`` nodes represent local variables in the source language. If
4108 the ``arg:`` field is set to non-zero, then this variable is a subprogram
4109 parameter, and it will be included in the ``variables:`` field of its
4110 :ref:`DISubprogram`.
4112 .. code-block:: llvm
4114 !0 = !DILocalVariable(name: "this", arg: 1, scope: !3, file: !2, line: 7,
4115 type: !3, flags: DIFlagArtificial)
4116 !1 = !DILocalVariable(name: "x", arg: 2, scope: !4, file: !2, line: 7,
4118 !2 = !DILocalVariable(name: "y", scope: !5, file: !2, line: 7, type: !3)
4123 ``DIExpression`` nodes represent DWARF expression sequences. They are used in
4124 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
4125 describe how the referenced LLVM variable relates to the source language
4128 The current supported vocabulary is limited:
4130 - ``DW_OP_deref`` dereferences the working expression.
4131 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
4132 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
4133 here, respectively) of the variable piece from the working expression.
4135 .. code-block:: llvm
4137 !0 = !DIExpression(DW_OP_deref)
4138 !1 = !DIExpression(DW_OP_plus, 3)
4139 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
4140 !3 = !DIExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
4145 ``DIObjCProperty`` nodes represent Objective-C property nodes.
4147 .. code-block:: llvm
4149 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
4150 getter: "getFoo", attributes: 7, type: !2)
4155 ``DIImportedEntity`` nodes represent entities (such as modules) imported into a
4158 .. code-block:: llvm
4160 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
4161 entity: !1, line: 7)
4166 ``DIMacro`` nodes represent definition or undefinition of a macro identifiers.
4167 The ``name:`` field is the macro identifier, followed by macro parameters when
4168 definining a function-like macro, and the ``value`` field is the token-string
4169 used to expand the macro identifier.
4171 .. code-block:: llvm
4173 !2 = !DIMacro(macinfo: DW_MACINFO_define, line: 7, name: "foo(x)",
4175 !3 = !DIMacro(macinfo: DW_MACINFO_undef, line: 30, name: "foo")
4180 ``DIMacroFile`` nodes represent inclusion of source files.
4181 The ``nodes:`` field is a list of ``DIMacro`` and ``DIMacroFile`` nodes that
4182 appear in the included source file.
4184 .. code-block:: llvm
4186 !2 = !DIMacroFile(macinfo: DW_MACINFO_start_file, line: 7, file: !2,
4192 In LLVM IR, memory does not have types, so LLVM's own type system is not
4193 suitable for doing TBAA. Instead, metadata is added to the IR to
4194 describe a type system of a higher level language. This can be used to
4195 implement typical C/C++ TBAA, but it can also be used to implement
4196 custom alias analysis behavior for other languages.
4198 The current metadata format is very simple. TBAA metadata nodes have up
4199 to three fields, e.g.:
4201 .. code-block:: llvm
4203 !0 = !{ !"an example type tree" }
4204 !1 = !{ !"int", !0 }
4205 !2 = !{ !"float", !0 }
4206 !3 = !{ !"const float", !2, i64 1 }
4208 The first field is an identity field. It can be any value, usually a
4209 metadata string, which uniquely identifies the type. The most important
4210 name in the tree is the name of the root node. Two trees with different
4211 root node names are entirely disjoint, even if they have leaves with
4214 The second field identifies the type's parent node in the tree, or is
4215 null or omitted for a root node. A type is considered to alias all of
4216 its descendants and all of its ancestors in the tree. Also, a type is
4217 considered to alias all types in other trees, so that bitcode produced
4218 from multiple front-ends is handled conservatively.
4220 If the third field is present, it's an integer which if equal to 1
4221 indicates that the type is "constant" (meaning
4222 ``pointsToConstantMemory`` should return true; see `other useful
4223 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
4225 '``tbaa.struct``' Metadata
4226 ^^^^^^^^^^^^^^^^^^^^^^^^^^
4228 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
4229 aggregate assignment operations in C and similar languages, however it
4230 is defined to copy a contiguous region of memory, which is more than
4231 strictly necessary for aggregate types which contain holes due to
4232 padding. Also, it doesn't contain any TBAA information about the fields
4235 ``!tbaa.struct`` metadata can describe which memory subregions in a
4236 memcpy are padding and what the TBAA tags of the struct are.
4238 The current metadata format is very simple. ``!tbaa.struct`` metadata
4239 nodes are a list of operands which are in conceptual groups of three.
4240 For each group of three, the first operand gives the byte offset of a
4241 field in bytes, the second gives its size in bytes, and the third gives
4244 .. code-block:: llvm
4246 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
4248 This describes a struct with two fields. The first is at offset 0 bytes
4249 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
4250 and has size 4 bytes and has tbaa tag !2.
4252 Note that the fields need not be contiguous. In this example, there is a
4253 4 byte gap between the two fields. This gap represents padding which
4254 does not carry useful data and need not be preserved.
4256 '``noalias``' and '``alias.scope``' Metadata
4257 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4259 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
4260 noalias memory-access sets. This means that some collection of memory access
4261 instructions (loads, stores, memory-accessing calls, etc.) that carry
4262 ``noalias`` metadata can specifically be specified not to alias with some other
4263 collection of memory access instructions that carry ``alias.scope`` metadata.
4264 Each type of metadata specifies a list of scopes where each scope has an id and
4265 a domain. When evaluating an aliasing query, if for some domain, the set
4266 of scopes with that domain in one instruction's ``alias.scope`` list is a
4267 subset of (or equal to) the set of scopes for that domain in another
4268 instruction's ``noalias`` list, then the two memory accesses are assumed not to
4271 The metadata identifying each domain is itself a list containing one or two
4272 entries. The first entry is the name of the domain. Note that if the name is a
4273 string then it can be combined across functions and translation units. A
4274 self-reference can be used to create globally unique domain names. A
4275 descriptive string may optionally be provided as a second list entry.
4277 The metadata identifying each scope is also itself a list containing two or
4278 three entries. The first entry is the name of the scope. Note that if the name
4279 is a string then it can be combined across functions and translation units. A
4280 self-reference can be used to create globally unique scope names. A metadata
4281 reference to the scope's domain is the second entry. A descriptive string may
4282 optionally be provided as a third list entry.
4286 .. code-block:: llvm
4288 ; Two scope domains:
4292 ; Some scopes in these domains:
4298 !5 = !{!4} ; A list containing only scope !4
4302 ; These two instructions don't alias:
4303 %0 = load float, float* %c, align 4, !alias.scope !5
4304 store float %0, float* %arrayidx.i, align 4, !noalias !5
4306 ; These two instructions also don't alias (for domain !1, the set of scopes
4307 ; in the !alias.scope equals that in the !noalias list):
4308 %2 = load float, float* %c, align 4, !alias.scope !5
4309 store float %2, float* %arrayidx.i2, align 4, !noalias !6
4311 ; These two instructions may alias (for domain !0, the set of scopes in
4312 ; the !noalias list is not a superset of, or equal to, the scopes in the
4313 ; !alias.scope list):
4314 %2 = load float, float* %c, align 4, !alias.scope !6
4315 store float %0, float* %arrayidx.i, align 4, !noalias !7
4317 '``fpmath``' Metadata
4318 ^^^^^^^^^^^^^^^^^^^^^
4320 ``fpmath`` metadata may be attached to any instruction of floating point
4321 type. It can be used to express the maximum acceptable error in the
4322 result of that instruction, in ULPs, thus potentially allowing the
4323 compiler to use a more efficient but less accurate method of computing
4324 it. ULP is defined as follows:
4326 If ``x`` is a real number that lies between two finite consecutive
4327 floating-point numbers ``a`` and ``b``, without being equal to one
4328 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
4329 distance between the two non-equal finite floating-point numbers
4330 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
4332 The metadata node shall consist of a single positive floating point
4333 number representing the maximum relative error, for example:
4335 .. code-block:: llvm
4337 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
4341 '``range``' Metadata
4342 ^^^^^^^^^^^^^^^^^^^^
4344 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
4345 integer types. It expresses the possible ranges the loaded value or the value
4346 returned by the called function at this call site is in. The ranges are
4347 represented with a flattened list of integers. The loaded value or the value
4348 returned is known to be in the union of the ranges defined by each consecutive
4349 pair. Each pair has the following properties:
4351 - The type must match the type loaded by the instruction.
4352 - The pair ``a,b`` represents the range ``[a,b)``.
4353 - Both ``a`` and ``b`` are constants.
4354 - The range is allowed to wrap.
4355 - The range should not represent the full or empty set. That is,
4358 In addition, the pairs must be in signed order of the lower bound and
4359 they must be non-contiguous.
4363 .. code-block:: llvm
4365 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
4366 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
4367 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
4368 %d = invoke i8 @bar() to label %cont
4369 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
4371 !0 = !{ i8 0, i8 2 }
4372 !1 = !{ i8 255, i8 2 }
4373 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
4374 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
4376 '``unpredictable``' Metadata
4377 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4379 ``unpredictable`` metadata may be attached to any branch or switch
4380 instruction. It can be used to express the unpredictability of control
4381 flow. Similar to the llvm.expect intrinsic, it may be used to alter
4382 optimizations related to compare and branch instructions. The metadata
4383 is treated as a boolean value; if it exists, it signals that the branch
4384 or switch that it is attached to is completely unpredictable.
4389 It is sometimes useful to attach information to loop constructs. Currently,
4390 loop metadata is implemented as metadata attached to the branch instruction
4391 in the loop latch block. This type of metadata refer to a metadata node that is
4392 guaranteed to be separate for each loop. The loop identifier metadata is
4393 specified with the name ``llvm.loop``.
4395 The loop identifier metadata is implemented using a metadata that refers to
4396 itself to avoid merging it with any other identifier metadata, e.g.,
4397 during module linkage or function inlining. That is, each loop should refer
4398 to their own identification metadata even if they reside in separate functions.
4399 The following example contains loop identifier metadata for two separate loop
4402 .. code-block:: llvm
4407 The loop identifier metadata can be used to specify additional
4408 per-loop metadata. Any operands after the first operand can be treated
4409 as user-defined metadata. For example the ``llvm.loop.unroll.count``
4410 suggests an unroll factor to the loop unroller:
4412 .. code-block:: llvm
4414 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
4417 !1 = !{!"llvm.loop.unroll.count", i32 4}
4419 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
4420 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4422 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
4423 used to control per-loop vectorization and interleaving parameters such as
4424 vectorization width and interleave count. These metadata should be used in
4425 conjunction with ``llvm.loop`` loop identification metadata. The
4426 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
4427 optimization hints and the optimizer will only interleave and vectorize loops if
4428 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
4429 which contains information about loop-carried memory dependencies can be helpful
4430 in determining the safety of these transformations.
4432 '``llvm.loop.interleave.count``' Metadata
4433 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4435 This metadata suggests an interleave count to the loop interleaver.
4436 The first operand is the string ``llvm.loop.interleave.count`` and the
4437 second operand is an integer specifying the interleave count. For
4440 .. code-block:: llvm
4442 !0 = !{!"llvm.loop.interleave.count", i32 4}
4444 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
4445 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
4446 then the interleave count will be determined automatically.
4448 '``llvm.loop.vectorize.enable``' Metadata
4449 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4451 This metadata selectively enables or disables vectorization for the loop. The
4452 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
4453 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
4454 0 disables vectorization:
4456 .. code-block:: llvm
4458 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
4459 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
4461 '``llvm.loop.vectorize.width``' Metadata
4462 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4464 This metadata sets the target width of the vectorizer. The first
4465 operand is the string ``llvm.loop.vectorize.width`` and the second
4466 operand is an integer specifying the width. For example:
4468 .. code-block:: llvm
4470 !0 = !{!"llvm.loop.vectorize.width", i32 4}
4472 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
4473 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
4474 0 or if the loop does not have this metadata the width will be
4475 determined automatically.
4477 '``llvm.loop.unroll``'
4478 ^^^^^^^^^^^^^^^^^^^^^^
4480 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
4481 optimization hints such as the unroll factor. ``llvm.loop.unroll``
4482 metadata should be used in conjunction with ``llvm.loop`` loop
4483 identification metadata. The ``llvm.loop.unroll`` metadata are only
4484 optimization hints and the unrolling will only be performed if the
4485 optimizer believes it is safe to do so.
4487 '``llvm.loop.unroll.count``' Metadata
4488 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4490 This metadata suggests an unroll factor to the loop unroller. The
4491 first operand is the string ``llvm.loop.unroll.count`` and the second
4492 operand is a positive integer specifying the unroll factor. For
4495 .. code-block:: llvm
4497 !0 = !{!"llvm.loop.unroll.count", i32 4}
4499 If the trip count of the loop is less than the unroll count the loop
4500 will be partially unrolled.
4502 '``llvm.loop.unroll.disable``' Metadata
4503 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4505 This metadata disables loop unrolling. The metadata has a single operand
4506 which is the string ``llvm.loop.unroll.disable``. For example:
4508 .. code-block:: llvm
4510 !0 = !{!"llvm.loop.unroll.disable"}
4512 '``llvm.loop.unroll.runtime.disable``' Metadata
4513 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4515 This metadata disables runtime loop unrolling. The metadata has a single
4516 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
4518 .. code-block:: llvm
4520 !0 = !{!"llvm.loop.unroll.runtime.disable"}
4522 '``llvm.loop.unroll.enable``' Metadata
4523 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4525 This metadata suggests that the loop should be fully unrolled if the trip count
4526 is known at compile time and partially unrolled if the trip count is not known
4527 at compile time. The metadata has a single operand which is the string
4528 ``llvm.loop.unroll.enable``. For example:
4530 .. code-block:: llvm
4532 !0 = !{!"llvm.loop.unroll.enable"}
4534 '``llvm.loop.unroll.full``' Metadata
4535 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4537 This metadata suggests that the loop should be unrolled fully. The
4538 metadata has a single operand which is the string ``llvm.loop.unroll.full``.
4541 .. code-block:: llvm
4543 !0 = !{!"llvm.loop.unroll.full"}
4548 Metadata types used to annotate memory accesses with information helpful
4549 for optimizations are prefixed with ``llvm.mem``.
4551 '``llvm.mem.parallel_loop_access``' Metadata
4552 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4554 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
4555 or metadata containing a list of loop identifiers for nested loops.
4556 The metadata is attached to memory accessing instructions and denotes that
4557 no loop carried memory dependence exist between it and other instructions denoted
4558 with the same loop identifier.
4560 Precisely, given two instructions ``m1`` and ``m2`` that both have the
4561 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
4562 set of loops associated with that metadata, respectively, then there is no loop
4563 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
4566 As a special case, if all memory accessing instructions in a loop have
4567 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
4568 loop has no loop carried memory dependences and is considered to be a parallel
4571 Note that if not all memory access instructions have such metadata referring to
4572 the loop, then the loop is considered not being trivially parallel. Additional
4573 memory dependence analysis is required to make that determination. As a fail
4574 safe mechanism, this causes loops that were originally parallel to be considered
4575 sequential (if optimization passes that are unaware of the parallel semantics
4576 insert new memory instructions into the loop body).
4578 Example of a loop that is considered parallel due to its correct use of
4579 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
4580 metadata types that refer to the same loop identifier metadata.
4582 .. code-block:: llvm
4586 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
4588 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4590 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
4596 It is also possible to have nested parallel loops. In that case the
4597 memory accesses refer to a list of loop identifier metadata nodes instead of
4598 the loop identifier metadata node directly:
4600 .. code-block:: llvm
4604 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
4606 br label %inner.for.body
4610 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4612 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
4614 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
4618 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
4620 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
4622 outer.for.end: ; preds = %for.body
4624 !0 = !{!1, !2} ; a list of loop identifiers
4625 !1 = !{!1} ; an identifier for the inner loop
4626 !2 = !{!2} ; an identifier for the outer loop
4631 The ``llvm.bitsets`` global metadata is used to implement
4632 :doc:`bitsets <BitSets>`.
4634 '``invariant.group``' Metadata
4635 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4637 The ``invariant.group`` metadata may be attached to ``load``/``store`` instructions.
4638 The existence of the ``invariant.group`` metadata on the instruction tells
4639 the optimizer that every ``load`` and ``store`` to the same pointer operand
4640 within the same invariant group can be assumed to load or store the same
4641 value (but see the ``llvm.invariant.group.barrier`` intrinsic which affects
4642 when two pointers are considered the same).
4646 .. code-block:: llvm
4648 @unknownPtr = external global i8
4651 store i8 42, i8* %ptr, !invariant.group !0
4652 call void @foo(i8* %ptr)
4654 %a = load i8, i8* %ptr, !invariant.group !0 ; Can assume that value under %ptr didn't change
4655 call void @foo(i8* %ptr)
4656 %b = load i8, i8* %ptr, !invariant.group !1 ; Can't assume anything, because group changed
4658 %newPtr = call i8* @getPointer(i8* %ptr)
4659 %c = load i8, i8* %newPtr, !invariant.group !0 ; Can't assume anything, because we only have information about %ptr
4661 %unknownValue = load i8, i8* @unknownPtr
4662 store i8 %unknownValue, i8* %ptr, !invariant.group !0 ; Can assume that %unknownValue == 42
4664 call void @foo(i8* %ptr)
4665 %newPtr2 = call i8* @llvm.invariant.group.barrier(i8* %ptr)
4666 %d = load i8, i8* %newPtr2, !invariant.group !0 ; Can't step through invariant.group.barrier to get value of %ptr
4669 declare void @foo(i8*)
4670 declare i8* @getPointer(i8*)
4671 declare i8* @llvm.invariant.group.barrier(i8*)
4673 !0 = !{!"magic ptr"}
4674 !1 = !{!"other ptr"}
4678 Module Flags Metadata
4679 =====================
4681 Information about the module as a whole is difficult to convey to LLVM's
4682 subsystems. The LLVM IR isn't sufficient to transmit this information.
4683 The ``llvm.module.flags`` named metadata exists in order to facilitate
4684 this. These flags are in the form of key / value pairs --- much like a
4685 dictionary --- making it easy for any subsystem who cares about a flag to
4688 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
4689 Each triplet has the following form:
4691 - The first element is a *behavior* flag, which specifies the behavior
4692 when two (or more) modules are merged together, and it encounters two
4693 (or more) metadata with the same ID. The supported behaviors are
4695 - The second element is a metadata string that is a unique ID for the
4696 metadata. Each module may only have one flag entry for each unique ID (not
4697 including entries with the **Require** behavior).
4698 - The third element is the value of the flag.
4700 When two (or more) modules are merged together, the resulting
4701 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
4702 each unique metadata ID string, there will be exactly one entry in the merged
4703 modules ``llvm.module.flags`` metadata table, and the value for that entry will
4704 be determined by the merge behavior flag, as described below. The only exception
4705 is that entries with the *Require* behavior are always preserved.
4707 The following behaviors are supported:
4718 Emits an error if two values disagree, otherwise the resulting value
4719 is that of the operands.
4723 Emits a warning if two values disagree. The result value will be the
4724 operand for the flag from the first module being linked.
4728 Adds a requirement that another module flag be present and have a
4729 specified value after linking is performed. The value must be a
4730 metadata pair, where the first element of the pair is the ID of the
4731 module flag to be restricted, and the second element of the pair is
4732 the value the module flag should be restricted to. This behavior can
4733 be used to restrict the allowable results (via triggering of an
4734 error) of linking IDs with the **Override** behavior.
4738 Uses the specified value, regardless of the behavior or value of the
4739 other module. If both modules specify **Override**, but the values
4740 differ, an error will be emitted.
4744 Appends the two values, which are required to be metadata nodes.
4748 Appends the two values, which are required to be metadata
4749 nodes. However, duplicate entries in the second list are dropped
4750 during the append operation.
4752 It is an error for a particular unique flag ID to have multiple behaviors,
4753 except in the case of **Require** (which adds restrictions on another metadata
4754 value) or **Override**.
4756 An example of module flags:
4758 .. code-block:: llvm
4760 !0 = !{ i32 1, !"foo", i32 1 }
4761 !1 = !{ i32 4, !"bar", i32 37 }
4762 !2 = !{ i32 2, !"qux", i32 42 }
4763 !3 = !{ i32 3, !"qux",
4768 !llvm.module.flags = !{ !0, !1, !2, !3 }
4770 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
4771 if two or more ``!"foo"`` flags are seen is to emit an error if their
4772 values are not equal.
4774 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
4775 behavior if two or more ``!"bar"`` flags are seen is to use the value
4778 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
4779 behavior if two or more ``!"qux"`` flags are seen is to emit a
4780 warning if their values are not equal.
4782 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
4788 The behavior is to emit an error if the ``llvm.module.flags`` does not
4789 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
4792 Objective-C Garbage Collection Module Flags Metadata
4793 ----------------------------------------------------
4795 On the Mach-O platform, Objective-C stores metadata about garbage
4796 collection in a special section called "image info". The metadata
4797 consists of a version number and a bitmask specifying what types of
4798 garbage collection are supported (if any) by the file. If two or more
4799 modules are linked together their garbage collection metadata needs to
4800 be merged rather than appended together.
4802 The Objective-C garbage collection module flags metadata consists of the
4803 following key-value pairs:
4812 * - ``Objective-C Version``
4813 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
4815 * - ``Objective-C Image Info Version``
4816 - **[Required]** --- The version of the image info section. Currently
4819 * - ``Objective-C Image Info Section``
4820 - **[Required]** --- The section to place the metadata. Valid values are
4821 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
4822 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
4823 Objective-C ABI version 2.
4825 * - ``Objective-C Garbage Collection``
4826 - **[Required]** --- Specifies whether garbage collection is supported or
4827 not. Valid values are 0, for no garbage collection, and 2, for garbage
4828 collection supported.
4830 * - ``Objective-C GC Only``
4831 - **[Optional]** --- Specifies that only garbage collection is supported.
4832 If present, its value must be 6. This flag requires that the
4833 ``Objective-C Garbage Collection`` flag have the value 2.
4835 Some important flag interactions:
4837 - If a module with ``Objective-C Garbage Collection`` set to 0 is
4838 merged with a module with ``Objective-C Garbage Collection`` set to
4839 2, then the resulting module has the
4840 ``Objective-C Garbage Collection`` flag set to 0.
4841 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
4842 merged with a module with ``Objective-C GC Only`` set to 6.
4844 Automatic Linker Flags Module Flags Metadata
4845 --------------------------------------------
4847 Some targets support embedding flags to the linker inside individual object
4848 files. Typically this is used in conjunction with language extensions which
4849 allow source files to explicitly declare the libraries they depend on, and have
4850 these automatically be transmitted to the linker via object files.
4852 These flags are encoded in the IR using metadata in the module flags section,
4853 using the ``Linker Options`` key. The merge behavior for this flag is required
4854 to be ``AppendUnique``, and the value for the key is expected to be a metadata
4855 node which should be a list of other metadata nodes, each of which should be a
4856 list of metadata strings defining linker options.
4858 For example, the following metadata section specifies two separate sets of
4859 linker options, presumably to link against ``libz`` and the ``Cocoa``
4862 !0 = !{ i32 6, !"Linker Options",
4865 !{ !"-framework", !"Cocoa" } } }
4866 !llvm.module.flags = !{ !0 }
4868 The metadata encoding as lists of lists of options, as opposed to a collapsed
4869 list of options, is chosen so that the IR encoding can use multiple option
4870 strings to specify e.g., a single library, while still having that specifier be
4871 preserved as an atomic element that can be recognized by a target specific
4872 assembly writer or object file emitter.
4874 Each individual option is required to be either a valid option for the target's
4875 linker, or an option that is reserved by the target specific assembly writer or
4876 object file emitter. No other aspect of these options is defined by the IR.
4878 C type width Module Flags Metadata
4879 ----------------------------------
4881 The ARM backend emits a section into each generated object file describing the
4882 options that it was compiled with (in a compiler-independent way) to prevent
4883 linking incompatible objects, and to allow automatic library selection. Some
4884 of these options are not visible at the IR level, namely wchar_t width and enum
4887 To pass this information to the backend, these options are encoded in module
4888 flags metadata, using the following key-value pairs:
4898 - * 0 --- sizeof(wchar_t) == 4
4899 * 1 --- sizeof(wchar_t) == 2
4902 - * 0 --- Enums are at least as large as an ``int``.
4903 * 1 --- Enums are stored in the smallest integer type which can
4904 represent all of its values.
4906 For example, the following metadata section specifies that the module was
4907 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
4908 enum is the smallest type which can represent all of its values::
4910 !llvm.module.flags = !{!0, !1}
4911 !0 = !{i32 1, !"short_wchar", i32 1}
4912 !1 = !{i32 1, !"short_enum", i32 0}
4914 .. _intrinsicglobalvariables:
4916 Intrinsic Global Variables
4917 ==========================
4919 LLVM has a number of "magic" global variables that contain data that
4920 affect code generation or other IR semantics. These are documented here.
4921 All globals of this sort should have a section specified as
4922 "``llvm.metadata``". This section and all globals that start with
4923 "``llvm.``" are reserved for use by LLVM.
4927 The '``llvm.used``' Global Variable
4928 -----------------------------------
4930 The ``@llvm.used`` global is an array which has
4931 :ref:`appending linkage <linkage_appending>`. This array contains a list of
4932 pointers to named global variables, functions and aliases which may optionally
4933 have a pointer cast formed of bitcast or getelementptr. For example, a legal
4936 .. code-block:: llvm
4941 @llvm.used = appending global [2 x i8*] [
4943 i8* bitcast (i32* @Y to i8*)
4944 ], section "llvm.metadata"
4946 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
4947 and linker are required to treat the symbol as if there is a reference to the
4948 symbol that it cannot see (which is why they have to be named). For example, if
4949 a variable has internal linkage and no references other than that from the
4950 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
4951 references from inline asms and other things the compiler cannot "see", and
4952 corresponds to "``attribute((used))``" in GNU C.
4954 On some targets, the code generator must emit a directive to the
4955 assembler or object file to prevent the assembler and linker from
4956 molesting the symbol.
4958 .. _gv_llvmcompilerused:
4960 The '``llvm.compiler.used``' Global Variable
4961 --------------------------------------------
4963 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4964 directive, except that it only prevents the compiler from touching the
4965 symbol. On targets that support it, this allows an intelligent linker to
4966 optimize references to the symbol without being impeded as it would be
4969 This is a rare construct that should only be used in rare circumstances,
4970 and should not be exposed to source languages.
4972 .. _gv_llvmglobalctors:
4974 The '``llvm.global_ctors``' Global Variable
4975 -------------------------------------------
4977 .. code-block:: llvm
4979 %0 = type { i32, void ()*, i8* }
4980 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
4982 The ``@llvm.global_ctors`` array contains a list of constructor
4983 functions, priorities, and an optional associated global or function.
4984 The functions referenced by this array will be called in ascending order
4985 of priority (i.e. lowest first) when the module is loaded. The order of
4986 functions with the same priority is not defined.
4988 If the third field is present, non-null, and points to a global variable
4989 or function, the initializer function will only run if the associated
4990 data from the current module is not discarded.
4992 .. _llvmglobaldtors:
4994 The '``llvm.global_dtors``' Global Variable
4995 -------------------------------------------
4997 .. code-block:: llvm
4999 %0 = type { i32, void ()*, i8* }
5000 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
5002 The ``@llvm.global_dtors`` array contains a list of destructor
5003 functions, priorities, and an optional associated global or function.
5004 The functions referenced by this array will be called in descending
5005 order of priority (i.e. highest first) when the module is unloaded. The
5006 order of functions with the same priority is not defined.
5008 If the third field is present, non-null, and points to a global variable
5009 or function, the destructor function will only run if the associated
5010 data from the current module is not discarded.
5012 Instruction Reference
5013 =====================
5015 The LLVM instruction set consists of several different classifications
5016 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
5017 instructions <binaryops>`, :ref:`bitwise binary
5018 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
5019 :ref:`other instructions <otherops>`.
5023 Terminator Instructions
5024 -----------------------
5026 As mentioned :ref:`previously <functionstructure>`, every basic block in a
5027 program ends with a "Terminator" instruction, which indicates which
5028 block should be executed after the current block is finished. These
5029 terminator instructions typically yield a '``void``' value: they produce
5030 control flow, not values (the one exception being the
5031 ':ref:`invoke <i_invoke>`' instruction).
5033 The terminator instructions are: ':ref:`ret <i_ret>`',
5034 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
5035 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
5036 ':ref:`resume <i_resume>`', ':ref:`catchswitch <i_catchswitch>`',
5037 ':ref:`catchret <i_catchret>`',
5038 ':ref:`cleanupret <i_cleanupret>`',
5039 and ':ref:`unreachable <i_unreachable>`'.
5043 '``ret``' Instruction
5044 ^^^^^^^^^^^^^^^^^^^^^
5051 ret <type> <value> ; Return a value from a non-void function
5052 ret void ; Return from void function
5057 The '``ret``' instruction is used to return control flow (and optionally
5058 a value) from a function back to the caller.
5060 There are two forms of the '``ret``' instruction: one that returns a
5061 value and then causes control flow, and one that just causes control
5067 The '``ret``' instruction optionally accepts a single argument, the
5068 return value. The type of the return value must be a ':ref:`first
5069 class <t_firstclass>`' type.
5071 A function is not :ref:`well formed <wellformed>` if it it has a non-void
5072 return type and contains a '``ret``' instruction with no return value or
5073 a return value with a type that does not match its type, or if it has a
5074 void return type and contains a '``ret``' instruction with a return
5080 When the '``ret``' instruction is executed, control flow returns back to
5081 the calling function's context. If the caller is a
5082 ":ref:`call <i_call>`" instruction, execution continues at the
5083 instruction after the call. If the caller was an
5084 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
5085 beginning of the "normal" destination block. If the instruction returns
5086 a value, that value shall set the call or invoke instruction's return
5092 .. code-block:: llvm
5094 ret i32 5 ; Return an integer value of 5
5095 ret void ; Return from a void function
5096 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
5100 '``br``' Instruction
5101 ^^^^^^^^^^^^^^^^^^^^
5108 br i1 <cond>, label <iftrue>, label <iffalse>
5109 br label <dest> ; Unconditional branch
5114 The '``br``' instruction is used to cause control flow to transfer to a
5115 different basic block in the current function. There are two forms of
5116 this instruction, corresponding to a conditional branch and an
5117 unconditional branch.
5122 The conditional branch form of the '``br``' instruction takes a single
5123 '``i1``' value and two '``label``' values. The unconditional form of the
5124 '``br``' instruction takes a single '``label``' value as a target.
5129 Upon execution of a conditional '``br``' instruction, the '``i1``'
5130 argument is evaluated. If the value is ``true``, control flows to the
5131 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
5132 to the '``iffalse``' ``label`` argument.
5137 .. code-block:: llvm
5140 %cond = icmp eq i32 %a, %b
5141 br i1 %cond, label %IfEqual, label %IfUnequal
5149 '``switch``' Instruction
5150 ^^^^^^^^^^^^^^^^^^^^^^^^
5157 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
5162 The '``switch``' instruction is used to transfer control flow to one of
5163 several different places. It is a generalization of the '``br``'
5164 instruction, allowing a branch to occur to one of many possible
5170 The '``switch``' instruction uses three parameters: an integer
5171 comparison value '``value``', a default '``label``' destination, and an
5172 array of pairs of comparison value constants and '``label``'s. The table
5173 is not allowed to contain duplicate constant entries.
5178 The ``switch`` instruction specifies a table of values and destinations.
5179 When the '``switch``' instruction is executed, this table is searched
5180 for the given value. If the value is found, control flow is transferred
5181 to the corresponding destination; otherwise, control flow is transferred
5182 to the default destination.
5187 Depending on properties of the target machine and the particular
5188 ``switch`` instruction, this instruction may be code generated in
5189 different ways. For example, it could be generated as a series of
5190 chained conditional branches or with a lookup table.
5195 .. code-block:: llvm
5197 ; Emulate a conditional br instruction
5198 %Val = zext i1 %value to i32
5199 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
5201 ; Emulate an unconditional br instruction
5202 switch i32 0, label %dest [ ]
5204 ; Implement a jump table:
5205 switch i32 %val, label %otherwise [ i32 0, label %onzero
5207 i32 2, label %ontwo ]
5211 '``indirectbr``' Instruction
5212 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5219 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
5224 The '``indirectbr``' instruction implements an indirect branch to a
5225 label within the current function, whose address is specified by
5226 "``address``". Address must be derived from a
5227 :ref:`blockaddress <blockaddress>` constant.
5232 The '``address``' argument is the address of the label to jump to. The
5233 rest of the arguments indicate the full set of possible destinations
5234 that the address may point to. Blocks are allowed to occur multiple
5235 times in the destination list, though this isn't particularly useful.
5237 This destination list is required so that dataflow analysis has an
5238 accurate understanding of the CFG.
5243 Control transfers to the block specified in the address argument. All
5244 possible destination blocks must be listed in the label list, otherwise
5245 this instruction has undefined behavior. This implies that jumps to
5246 labels defined in other functions have undefined behavior as well.
5251 This is typically implemented with a jump through a register.
5256 .. code-block:: llvm
5258 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
5262 '``invoke``' Instruction
5263 ^^^^^^^^^^^^^^^^^^^^^^^^
5270 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
5271 [operand bundles] to label <normal label> unwind label <exception label>
5276 The '``invoke``' instruction causes control to transfer to a specified
5277 function, with the possibility of control flow transfer to either the
5278 '``normal``' label or the '``exception``' label. If the callee function
5279 returns with the "``ret``" instruction, control flow will return to the
5280 "normal" label. If the callee (or any indirect callees) returns via the
5281 ":ref:`resume <i_resume>`" instruction or other exception handling
5282 mechanism, control is interrupted and continued at the dynamically
5283 nearest "exception" label.
5285 The '``exception``' label is a `landing
5286 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
5287 '``exception``' label is required to have the
5288 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
5289 information about the behavior of the program after unwinding happens,
5290 as its first non-PHI instruction. The restrictions on the
5291 "``landingpad``" instruction's tightly couples it to the "``invoke``"
5292 instruction, so that the important information contained within the
5293 "``landingpad``" instruction can't be lost through normal code motion.
5298 This instruction requires several arguments:
5300 #. The optional "cconv" marker indicates which :ref:`calling
5301 convention <callingconv>` the call should use. If none is
5302 specified, the call defaults to using C calling conventions.
5303 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5304 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5306 #. '``ptr to function ty``': shall be the signature of the pointer to
5307 function value being invoked. In most cases, this is a direct
5308 function invocation, but indirect ``invoke``'s are just as possible,
5309 branching off an arbitrary pointer to function value.
5310 #. '``function ptr val``': An LLVM value containing a pointer to a
5311 function to be invoked.
5312 #. '``function args``': argument list whose types match the function
5313 signature argument types and parameter attributes. All arguments must
5314 be of :ref:`first class <t_firstclass>` type. If the function signature
5315 indicates the function accepts a variable number of arguments, the
5316 extra arguments can be specified.
5317 #. '``normal label``': the label reached when the called function
5318 executes a '``ret``' instruction.
5319 #. '``exception label``': the label reached when a callee returns via
5320 the :ref:`resume <i_resume>` instruction or other exception handling
5322 #. The optional :ref:`function attributes <fnattrs>` list. Only
5323 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5324 attributes are valid here.
5325 #. The optional :ref:`operand bundles <opbundles>` list.
5330 This instruction is designed to operate as a standard '``call``'
5331 instruction in most regards. The primary difference is that it
5332 establishes an association with a label, which is used by the runtime
5333 library to unwind the stack.
5335 This instruction is used in languages with destructors to ensure that
5336 proper cleanup is performed in the case of either a ``longjmp`` or a
5337 thrown exception. Additionally, this is important for implementation of
5338 '``catch``' clauses in high-level languages that support them.
5340 For the purposes of the SSA form, the definition of the value returned
5341 by the '``invoke``' instruction is deemed to occur on the edge from the
5342 current block to the "normal" label. If the callee unwinds then no
5343 return value is available.
5348 .. code-block:: llvm
5350 %retval = invoke i32 @Test(i32 15) to label %Continue
5351 unwind label %TestCleanup ; i32:retval set
5352 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
5353 unwind label %TestCleanup ; i32:retval set
5357 '``resume``' Instruction
5358 ^^^^^^^^^^^^^^^^^^^^^^^^
5365 resume <type> <value>
5370 The '``resume``' instruction is a terminator instruction that has no
5376 The '``resume``' instruction requires one argument, which must have the
5377 same type as the result of any '``landingpad``' instruction in the same
5383 The '``resume``' instruction resumes propagation of an existing
5384 (in-flight) exception whose unwinding was interrupted with a
5385 :ref:`landingpad <i_landingpad>` instruction.
5390 .. code-block:: llvm
5392 resume { i8*, i32 } %exn
5396 '``catchswitch``' Instruction
5397 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5404 <resultval> = catchswitch within <parent> [ label <handler1>, label <handler2>, ... ] unwind to caller
5405 <resultval> = catchswitch within <parent> [ label <handler1>, label <handler2>, ... ] unwind label <default>
5410 The '``catchswitch``' instruction is used by `LLVM's exception handling system
5411 <ExceptionHandling.html#overview>`_ to describe the set of possible catch handlers
5412 that may be executed by the :ref:`EH personality routine <personalityfn>`.
5417 The ``parent`` argument is the token of the funclet that contains the
5418 ``catchswitch`` instruction. If the ``catchswitch`` is not inside a funclet,
5419 this operand may be the token ``none``.
5421 The ``default`` argument is the label of another basic block beginning with
5422 either a ``cleanuppad`` or ``catchswitch`` instruction. This unwind destination
5423 must be a legal target with respect to the ``parent`` links, as described in
5424 the `exception handling documentation\ <ExceptionHandling.html#wineh-constraints>`_.
5426 The ``handlers`` are a nonempty list of successor blocks that each begin with a
5427 :ref:`catchpad <i_catchpad>` instruction.
5432 Executing this instruction transfers control to one of the successors in
5433 ``handlers``, if appropriate, or continues to unwind via the unwind label if
5436 The ``catchswitch`` is both a terminator and a "pad" instruction, meaning that
5437 it must be both the first non-phi instruction and last instruction in the basic
5438 block. Therefore, it must be the only non-phi instruction in the block.
5443 .. code-block:: llvm
5446 %cs1 = catchswitch within none [label %handler0, label %handler1] unwind to caller
5448 %cs2 = catchswitch within %parenthandler [label %handler0] unwind label %cleanup
5452 '``catchret``' Instruction
5453 ^^^^^^^^^^^^^^^^^^^^^^^^^^
5460 catchret from <token> to label <normal>
5465 The '``catchret``' instruction is a terminator instruction that has a
5472 The first argument to a '``catchret``' indicates which ``catchpad`` it
5473 exits. It must be a :ref:`catchpad <i_catchpad>`.
5474 The second argument to a '``catchret``' specifies where control will
5480 The '``catchret``' instruction ends an existing (in-flight) exception whose
5481 unwinding was interrupted with a :ref:`catchpad <i_catchpad>` instruction. The
5482 :ref:`personality function <personalityfn>` gets a chance to execute arbitrary
5483 code to, for example, destroy the active exception. Control then transfers to
5486 The ``token`` argument must be a token produced by a ``catchpad`` instruction.
5487 If the specified ``catchpad`` is not the most-recently-entered not-yet-exited
5488 funclet pad (as described in the `EH documentation\ <ExceptionHandling.html#wineh-constraints>`_),
5489 the ``catchret``'s behavior is undefined.
5494 .. code-block:: llvm
5496 catchret from %catch label %continue
5500 '``cleanupret``' Instruction
5501 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5508 cleanupret from <value> unwind label <continue>
5509 cleanupret from <value> unwind to caller
5514 The '``cleanupret``' instruction is a terminator instruction that has
5515 an optional successor.
5521 The '``cleanupret``' instruction requires one argument, which indicates
5522 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
5523 If the specified ``cleanuppad`` is not the most-recently-entered not-yet-exited
5524 funclet pad (as described in the `EH documentation\ <ExceptionHandling.html#wineh-constraints>`_),
5525 the ``cleanupret``'s behavior is undefined.
5527 The '``cleanupret``' instruction also has an optional successor, ``continue``,
5528 which must be the label of another basic block beginning with either a
5529 ``cleanuppad`` or ``catchswitch`` instruction. This unwind destination must
5530 be a legal target with respect to the ``parent`` links, as described in the
5531 `exception handling documentation\ <ExceptionHandling.html#wineh-constraints>`_.
5536 The '``cleanupret``' instruction indicates to the
5537 :ref:`personality function <personalityfn>` that one
5538 :ref:`cleanuppad <i_cleanuppad>` it transferred control to has ended.
5539 It transfers control to ``continue`` or unwinds out of the function.
5544 .. code-block:: llvm
5546 cleanupret from %cleanup unwind to caller
5547 cleanupret from %cleanup unwind label %continue
5551 '``unreachable``' Instruction
5552 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5564 The '``unreachable``' instruction has no defined semantics. This
5565 instruction is used to inform the optimizer that a particular portion of
5566 the code is not reachable. This can be used to indicate that the code
5567 after a no-return function cannot be reached, and other facts.
5572 The '``unreachable``' instruction has no defined semantics.
5579 Binary operators are used to do most of the computation in a program.
5580 They require two operands of the same type, execute an operation on
5581 them, and produce a single value. The operands might represent multiple
5582 data, as is the case with the :ref:`vector <t_vector>` data type. The
5583 result value has the same type as its operands.
5585 There are several different binary operators:
5589 '``add``' Instruction
5590 ^^^^^^^^^^^^^^^^^^^^^
5597 <result> = add <ty> <op1>, <op2> ; yields ty:result
5598 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
5599 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
5600 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
5605 The '``add``' instruction returns the sum of its two operands.
5610 The two arguments to the '``add``' instruction must be
5611 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5612 arguments must have identical types.
5617 The value produced is the integer sum of the two operands.
5619 If the sum has unsigned overflow, the result returned is the
5620 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5623 Because LLVM integers use a two's complement representation, this
5624 instruction is appropriate for both signed and unsigned integers.
5626 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5627 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5628 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
5629 unsigned and/or signed overflow, respectively, occurs.
5634 .. code-block:: llvm
5636 <result> = add i32 4, %var ; yields i32:result = 4 + %var
5640 '``fadd``' Instruction
5641 ^^^^^^^^^^^^^^^^^^^^^^
5648 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5653 The '``fadd``' instruction returns the sum of its two operands.
5658 The two arguments to the '``fadd``' instruction must be :ref:`floating
5659 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5660 Both arguments must have identical types.
5665 The value produced is the floating point sum of the two operands. This
5666 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
5667 which are optimization hints to enable otherwise unsafe floating point
5673 .. code-block:: llvm
5675 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
5677 '``sub``' Instruction
5678 ^^^^^^^^^^^^^^^^^^^^^
5685 <result> = sub <ty> <op1>, <op2> ; yields ty:result
5686 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
5687 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
5688 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
5693 The '``sub``' instruction returns the difference of its two operands.
5695 Note that the '``sub``' instruction is used to represent the '``neg``'
5696 instruction present in most other intermediate representations.
5701 The two arguments to the '``sub``' instruction must be
5702 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5703 arguments must have identical types.
5708 The value produced is the integer difference of the two operands.
5710 If the difference has unsigned overflow, the result returned is the
5711 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5714 Because LLVM integers use a two's complement representation, this
5715 instruction is appropriate for both signed and unsigned integers.
5717 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5718 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5719 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
5720 unsigned and/or signed overflow, respectively, occurs.
5725 .. code-block:: llvm
5727 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
5728 <result> = sub i32 0, %val ; yields i32:result = -%var
5732 '``fsub``' Instruction
5733 ^^^^^^^^^^^^^^^^^^^^^^
5740 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5745 The '``fsub``' instruction returns the difference of its two operands.
5747 Note that the '``fsub``' instruction is used to represent the '``fneg``'
5748 instruction present in most other intermediate representations.
5753 The two arguments to the '``fsub``' instruction must be :ref:`floating
5754 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5755 Both arguments must have identical types.
5760 The value produced is the floating point difference of the two operands.
5761 This instruction can also take any number of :ref:`fast-math
5762 flags <fastmath>`, which are optimization hints to enable otherwise
5763 unsafe floating point optimizations:
5768 .. code-block:: llvm
5770 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
5771 <result> = fsub float -0.0, %val ; yields float:result = -%var
5773 '``mul``' Instruction
5774 ^^^^^^^^^^^^^^^^^^^^^
5781 <result> = mul <ty> <op1>, <op2> ; yields ty:result
5782 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
5783 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
5784 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
5789 The '``mul``' instruction returns the product of its two operands.
5794 The two arguments to the '``mul``' instruction must be
5795 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5796 arguments must have identical types.
5801 The value produced is the integer product of the two operands.
5803 If the result of the multiplication has unsigned overflow, the result
5804 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
5805 bit width of the result.
5807 Because LLVM integers use a two's complement representation, and the
5808 result is the same width as the operands, this instruction returns the
5809 correct result for both signed and unsigned integers. If a full product
5810 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
5811 sign-extended or zero-extended as appropriate to the width of the full
5814 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5815 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5816 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
5817 unsigned and/or signed overflow, respectively, occurs.
5822 .. code-block:: llvm
5824 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
5828 '``fmul``' Instruction
5829 ^^^^^^^^^^^^^^^^^^^^^^
5836 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5841 The '``fmul``' instruction returns the product of its two operands.
5846 The two arguments to the '``fmul``' instruction must be :ref:`floating
5847 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5848 Both arguments must have identical types.
5853 The value produced is the floating point product of the two operands.
5854 This instruction can also take any number of :ref:`fast-math
5855 flags <fastmath>`, which are optimization hints to enable otherwise
5856 unsafe floating point optimizations:
5861 .. code-block:: llvm
5863 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
5865 '``udiv``' Instruction
5866 ^^^^^^^^^^^^^^^^^^^^^^
5873 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
5874 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
5879 The '``udiv``' instruction returns the quotient of its two operands.
5884 The two arguments to the '``udiv``' instruction must be
5885 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5886 arguments must have identical types.
5891 The value produced is the unsigned integer quotient of the two operands.
5893 Note that unsigned integer division and signed integer division are
5894 distinct operations; for signed integer division, use '``sdiv``'.
5896 Division by zero leads to undefined behavior.
5898 If the ``exact`` keyword is present, the result value of the ``udiv`` is
5899 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
5900 such, "((a udiv exact b) mul b) == a").
5905 .. code-block:: llvm
5907 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
5909 '``sdiv``' Instruction
5910 ^^^^^^^^^^^^^^^^^^^^^^
5917 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
5918 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
5923 The '``sdiv``' instruction returns the quotient of its two operands.
5928 The two arguments to the '``sdiv``' instruction must be
5929 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5930 arguments must have identical types.
5935 The value produced is the signed integer quotient of the two operands
5936 rounded towards zero.
5938 Note that signed integer division and unsigned integer division are
5939 distinct operations; for unsigned integer division, use '``udiv``'.
5941 Division by zero leads to undefined behavior. Overflow also leads to
5942 undefined behavior; this is a rare case, but can occur, for example, by
5943 doing a 32-bit division of -2147483648 by -1.
5945 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
5946 a :ref:`poison value <poisonvalues>` if the result would be rounded.
5951 .. code-block:: llvm
5953 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
5957 '``fdiv``' Instruction
5958 ^^^^^^^^^^^^^^^^^^^^^^
5965 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5970 The '``fdiv``' instruction returns the quotient of its two operands.
5975 The two arguments to the '``fdiv``' instruction must be :ref:`floating
5976 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5977 Both arguments must have identical types.
5982 The value produced is the floating point quotient of the two operands.
5983 This instruction can also take any number of :ref:`fast-math
5984 flags <fastmath>`, which are optimization hints to enable otherwise
5985 unsafe floating point optimizations:
5990 .. code-block:: llvm
5992 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
5994 '``urem``' Instruction
5995 ^^^^^^^^^^^^^^^^^^^^^^
6002 <result> = urem <ty> <op1>, <op2> ; yields ty:result
6007 The '``urem``' instruction returns the remainder from the unsigned
6008 division of its two arguments.
6013 The two arguments to the '``urem``' instruction must be
6014 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6015 arguments must have identical types.
6020 This instruction returns the unsigned integer *remainder* of a division.
6021 This instruction always performs an unsigned division to get the
6024 Note that unsigned integer remainder and signed integer remainder are
6025 distinct operations; for signed integer remainder, use '``srem``'.
6027 Taking the remainder of a division by zero leads to undefined behavior.
6032 .. code-block:: llvm
6034 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
6036 '``srem``' Instruction
6037 ^^^^^^^^^^^^^^^^^^^^^^
6044 <result> = srem <ty> <op1>, <op2> ; yields ty:result
6049 The '``srem``' instruction returns the remainder from the signed
6050 division of its two operands. This instruction can also take
6051 :ref:`vector <t_vector>` versions of the values in which case the elements
6057 The two arguments to the '``srem``' instruction must be
6058 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6059 arguments must have identical types.
6064 This instruction returns the *remainder* of a division (where the result
6065 is either zero or has the same sign as the dividend, ``op1``), not the
6066 *modulo* operator (where the result is either zero or has the same sign
6067 as the divisor, ``op2``) of a value. For more information about the
6068 difference, see `The Math
6069 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
6070 table of how this is implemented in various languages, please see
6072 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
6074 Note that signed integer remainder and unsigned integer remainder are
6075 distinct operations; for unsigned integer remainder, use '``urem``'.
6077 Taking the remainder of a division by zero leads to undefined behavior.
6078 Overflow also leads to undefined behavior; this is a rare case, but can
6079 occur, for example, by taking the remainder of a 32-bit division of
6080 -2147483648 by -1. (The remainder doesn't actually overflow, but this
6081 rule lets srem be implemented using instructions that return both the
6082 result of the division and the remainder.)
6087 .. code-block:: llvm
6089 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
6093 '``frem``' Instruction
6094 ^^^^^^^^^^^^^^^^^^^^^^
6101 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6106 The '``frem``' instruction returns the remainder from the division of
6112 The two arguments to the '``frem``' instruction must be :ref:`floating
6113 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6114 Both arguments must have identical types.
6119 This instruction returns the *remainder* of a division. The remainder
6120 has the same sign as the dividend. This instruction can also take any
6121 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
6122 to enable otherwise unsafe floating point optimizations:
6127 .. code-block:: llvm
6129 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
6133 Bitwise Binary Operations
6134 -------------------------
6136 Bitwise binary operators are used to do various forms of bit-twiddling
6137 in a program. They are generally very efficient instructions and can
6138 commonly be strength reduced from other instructions. They require two
6139 operands of the same type, execute an operation on them, and produce a
6140 single value. The resulting value is the same type as its operands.
6142 '``shl``' Instruction
6143 ^^^^^^^^^^^^^^^^^^^^^
6150 <result> = shl <ty> <op1>, <op2> ; yields ty:result
6151 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
6152 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
6153 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
6158 The '``shl``' instruction returns the first operand shifted to the left
6159 a specified number of bits.
6164 Both arguments to the '``shl``' instruction must be the same
6165 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6166 '``op2``' is treated as an unsigned value.
6171 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
6172 where ``n`` is the width of the result. If ``op2`` is (statically or
6173 dynamically) equal to or larger than the number of bits in
6174 ``op1``, the result is undefined. If the arguments are vectors, each
6175 vector element of ``op1`` is shifted by the corresponding shift amount
6178 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
6179 value <poisonvalues>` if it shifts out any non-zero bits. If the
6180 ``nsw`` keyword is present, then the shift produces a :ref:`poison
6181 value <poisonvalues>` if it shifts out any bits that disagree with the
6182 resultant sign bit. As such, NUW/NSW have the same semantics as they
6183 would if the shift were expressed as a mul instruction with the same
6184 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
6189 .. code-block:: llvm
6191 <result> = shl i32 4, %var ; yields i32: 4 << %var
6192 <result> = shl i32 4, 2 ; yields i32: 16
6193 <result> = shl i32 1, 10 ; yields i32: 1024
6194 <result> = shl i32 1, 32 ; undefined
6195 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
6197 '``lshr``' Instruction
6198 ^^^^^^^^^^^^^^^^^^^^^^
6205 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
6206 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
6211 The '``lshr``' instruction (logical shift right) returns the first
6212 operand shifted to the right a specified number of bits with zero fill.
6217 Both arguments to the '``lshr``' instruction must be the same
6218 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6219 '``op2``' is treated as an unsigned value.
6224 This instruction always performs a logical shift right operation. The
6225 most significant bits of the result will be filled with zero bits after
6226 the shift. If ``op2`` is (statically or dynamically) equal to or larger
6227 than the number of bits in ``op1``, the result is undefined. If the
6228 arguments are vectors, each vector element of ``op1`` is shifted by the
6229 corresponding shift amount in ``op2``.
6231 If the ``exact`` keyword is present, the result value of the ``lshr`` is
6232 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6238 .. code-block:: llvm
6240 <result> = lshr i32 4, 1 ; yields i32:result = 2
6241 <result> = lshr i32 4, 2 ; yields i32:result = 1
6242 <result> = lshr i8 4, 3 ; yields i8:result = 0
6243 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
6244 <result> = lshr i32 1, 32 ; undefined
6245 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
6247 '``ashr``' Instruction
6248 ^^^^^^^^^^^^^^^^^^^^^^
6255 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
6256 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
6261 The '``ashr``' instruction (arithmetic shift right) returns the first
6262 operand shifted to the right a specified number of bits with sign
6268 Both arguments to the '``ashr``' instruction must be the same
6269 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6270 '``op2``' is treated as an unsigned value.
6275 This instruction always performs an arithmetic shift right operation,
6276 The most significant bits of the result will be filled with the sign bit
6277 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
6278 than the number of bits in ``op1``, the result is undefined. If the
6279 arguments are vectors, each vector element of ``op1`` is shifted by the
6280 corresponding shift amount in ``op2``.
6282 If the ``exact`` keyword is present, the result value of the ``ashr`` is
6283 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6289 .. code-block:: llvm
6291 <result> = ashr i32 4, 1 ; yields i32:result = 2
6292 <result> = ashr i32 4, 2 ; yields i32:result = 1
6293 <result> = ashr i8 4, 3 ; yields i8:result = 0
6294 <result> = ashr i8 -2, 1 ; yields i8:result = -1
6295 <result> = ashr i32 1, 32 ; undefined
6296 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
6298 '``and``' Instruction
6299 ^^^^^^^^^^^^^^^^^^^^^
6306 <result> = and <ty> <op1>, <op2> ; yields ty:result
6311 The '``and``' instruction returns the bitwise logical and of its two
6317 The two arguments to the '``and``' instruction must be
6318 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6319 arguments must have identical types.
6324 The truth table used for the '``and``' instruction is:
6341 .. code-block:: llvm
6343 <result> = and i32 4, %var ; yields i32:result = 4 & %var
6344 <result> = and i32 15, 40 ; yields i32:result = 8
6345 <result> = and i32 4, 8 ; yields i32:result = 0
6347 '``or``' Instruction
6348 ^^^^^^^^^^^^^^^^^^^^
6355 <result> = or <ty> <op1>, <op2> ; yields ty:result
6360 The '``or``' instruction returns the bitwise logical inclusive or of its
6366 The two arguments to the '``or``' instruction must be
6367 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6368 arguments must have identical types.
6373 The truth table used for the '``or``' instruction is:
6392 <result> = or i32 4, %var ; yields i32:result = 4 | %var
6393 <result> = or i32 15, 40 ; yields i32:result = 47
6394 <result> = or i32 4, 8 ; yields i32:result = 12
6396 '``xor``' Instruction
6397 ^^^^^^^^^^^^^^^^^^^^^
6404 <result> = xor <ty> <op1>, <op2> ; yields ty:result
6409 The '``xor``' instruction returns the bitwise logical exclusive or of
6410 its two operands. The ``xor`` is used to implement the "one's
6411 complement" operation, which is the "~" operator in C.
6416 The two arguments to the '``xor``' instruction must be
6417 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6418 arguments must have identical types.
6423 The truth table used for the '``xor``' instruction is:
6440 .. code-block:: llvm
6442 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
6443 <result> = xor i32 15, 40 ; yields i32:result = 39
6444 <result> = xor i32 4, 8 ; yields i32:result = 12
6445 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
6450 LLVM supports several instructions to represent vector operations in a
6451 target-independent manner. These instructions cover the element-access
6452 and vector-specific operations needed to process vectors effectively.
6453 While LLVM does directly support these vector operations, many
6454 sophisticated algorithms will want to use target-specific intrinsics to
6455 take full advantage of a specific target.
6457 .. _i_extractelement:
6459 '``extractelement``' Instruction
6460 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6467 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
6472 The '``extractelement``' instruction extracts a single scalar element
6473 from a vector at a specified index.
6478 The first operand of an '``extractelement``' instruction is a value of
6479 :ref:`vector <t_vector>` type. The second operand is an index indicating
6480 the position from which to extract the element. The index may be a
6481 variable of any integer type.
6486 The result is a scalar of the same type as the element type of ``val``.
6487 Its value is the value at position ``idx`` of ``val``. If ``idx``
6488 exceeds the length of ``val``, the results are undefined.
6493 .. code-block:: llvm
6495 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
6497 .. _i_insertelement:
6499 '``insertelement``' Instruction
6500 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6507 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
6512 The '``insertelement``' instruction inserts a scalar element into a
6513 vector at a specified index.
6518 The first operand of an '``insertelement``' instruction is a value of
6519 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
6520 type must equal the element type of the first operand. The third operand
6521 is an index indicating the position at which to insert the value. The
6522 index may be a variable of any integer type.
6527 The result is a vector of the same type as ``val``. Its element values
6528 are those of ``val`` except at position ``idx``, where it gets the value
6529 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
6535 .. code-block:: llvm
6537 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
6539 .. _i_shufflevector:
6541 '``shufflevector``' Instruction
6542 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6549 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
6554 The '``shufflevector``' instruction constructs a permutation of elements
6555 from two input vectors, returning a vector with the same element type as
6556 the input and length that is the same as the shuffle mask.
6561 The first two operands of a '``shufflevector``' instruction are vectors
6562 with the same type. The third argument is a shuffle mask whose element
6563 type is always 'i32'. The result of the instruction is a vector whose
6564 length is the same as the shuffle mask and whose element type is the
6565 same as the element type of the first two operands.
6567 The shuffle mask operand is required to be a constant vector with either
6568 constant integer or undef values.
6573 The elements of the two input vectors are numbered from left to right
6574 across both of the vectors. The shuffle mask operand specifies, for each
6575 element of the result vector, which element of the two input vectors the
6576 result element gets. The element selector may be undef (meaning "don't
6577 care") and the second operand may be undef if performing a shuffle from
6583 .. code-block:: llvm
6585 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6586 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
6587 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
6588 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
6589 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
6590 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
6591 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6592 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
6594 Aggregate Operations
6595 --------------------
6597 LLVM supports several instructions for working with
6598 :ref:`aggregate <t_aggregate>` values.
6602 '``extractvalue``' Instruction
6603 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6610 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
6615 The '``extractvalue``' instruction extracts the value of a member field
6616 from an :ref:`aggregate <t_aggregate>` value.
6621 The first operand of an '``extractvalue``' instruction is a value of
6622 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The other operands are
6623 constant indices to specify which value to extract in a similar manner
6624 as indices in a '``getelementptr``' instruction.
6626 The major differences to ``getelementptr`` indexing are:
6628 - Since the value being indexed is not a pointer, the first index is
6629 omitted and assumed to be zero.
6630 - At least one index must be specified.
6631 - Not only struct indices but also array indices must be in bounds.
6636 The result is the value at the position in the aggregate specified by
6642 .. code-block:: llvm
6644 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
6648 '``insertvalue``' Instruction
6649 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6656 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
6661 The '``insertvalue``' instruction inserts a value into a member field in
6662 an :ref:`aggregate <t_aggregate>` value.
6667 The first operand of an '``insertvalue``' instruction is a value of
6668 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
6669 a first-class value to insert. The following operands are constant
6670 indices indicating the position at which to insert the value in a
6671 similar manner as indices in a '``extractvalue``' instruction. The value
6672 to insert must have the same type as the value identified by the
6678 The result is an aggregate of the same type as ``val``. Its value is
6679 that of ``val`` except that the value at the position specified by the
6680 indices is that of ``elt``.
6685 .. code-block:: llvm
6687 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
6688 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
6689 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
6693 Memory Access and Addressing Operations
6694 ---------------------------------------
6696 A key design point of an SSA-based representation is how it represents
6697 memory. In LLVM, no memory locations are in SSA form, which makes things
6698 very simple. This section describes how to read, write, and allocate
6703 '``alloca``' Instruction
6704 ^^^^^^^^^^^^^^^^^^^^^^^^
6711 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
6716 The '``alloca``' instruction allocates memory on the stack frame of the
6717 currently executing function, to be automatically released when this
6718 function returns to its caller. The object is always allocated in the
6719 generic address space (address space zero).
6724 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
6725 bytes of memory on the runtime stack, returning a pointer of the
6726 appropriate type to the program. If "NumElements" is specified, it is
6727 the number of elements allocated, otherwise "NumElements" is defaulted
6728 to be one. If a constant alignment is specified, the value result of the
6729 allocation is guaranteed to be aligned to at least that boundary. The
6730 alignment may not be greater than ``1 << 29``. If not specified, or if
6731 zero, the target can choose to align the allocation on any convenient
6732 boundary compatible with the type.
6734 '``type``' may be any sized type.
6739 Memory is allocated; a pointer is returned. The operation is undefined
6740 if there is insufficient stack space for the allocation. '``alloca``'d
6741 memory is automatically released when the function returns. The
6742 '``alloca``' instruction is commonly used to represent automatic
6743 variables that must have an address available. When the function returns
6744 (either with the ``ret`` or ``resume`` instructions), the memory is
6745 reclaimed. Allocating zero bytes is legal, but the result is undefined.
6746 The order in which memory is allocated (ie., which way the stack grows)
6752 .. code-block:: llvm
6754 %ptr = alloca i32 ; yields i32*:ptr
6755 %ptr = alloca i32, i32 4 ; yields i32*:ptr
6756 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
6757 %ptr = alloca i32, align 1024 ; yields i32*:ptr
6761 '``load``' Instruction
6762 ^^^^^^^^^^^^^^^^^^^^^^
6769 <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>]
6770 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>]
6771 !<index> = !{ i32 1 }
6772 !<deref_bytes_node> = !{i64 <dereferenceable_bytes>}
6773 !<align_node> = !{ i64 <value_alignment> }
6778 The '``load``' instruction is used to read from memory.
6783 The argument to the ``load`` instruction specifies the memory address
6784 from which to load. The type specified must be a :ref:`first
6785 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
6786 then the optimizer is not allowed to modify the number or order of
6787 execution of this ``load`` with other :ref:`volatile
6788 operations <volatile>`.
6790 If the ``load`` is marked as ``atomic``, it takes an extra :ref:`ordering
6791 <ordering>` and optional ``singlethread`` argument. The ``release`` and
6792 ``acq_rel`` orderings are not valid on ``load`` instructions. Atomic loads
6793 produce :ref:`defined <memmodel>` results when they may see multiple atomic
6794 stores. The type of the pointee must be an integer, pointer, or floating-point
6795 type whose bit width is a power of two greater than or equal to eight and less
6796 than or equal to a target-specific size limit. ``align`` must be explicitly
6797 specified on atomic loads, and the load has undefined behavior if the alignment
6798 is not set to a value which is at least the size in bytes of the
6799 pointee. ``!nontemporal`` does not have any defined semantics for atomic loads.
6801 The optional constant ``align`` argument specifies the alignment of the
6802 operation (that is, the alignment of the memory address). A value of 0
6803 or an omitted ``align`` argument means that the operation has the ABI
6804 alignment for the target. It is the responsibility of the code emitter
6805 to ensure that the alignment information is correct. Overestimating the
6806 alignment results in undefined behavior. Underestimating the alignment
6807 may produce less efficient code. An alignment of 1 is always safe. The
6808 maximum possible alignment is ``1 << 29``.
6810 The optional ``!nontemporal`` metadata must reference a single
6811 metadata name ``<index>`` corresponding to a metadata node with one
6812 ``i32`` entry of value 1. The existence of the ``!nontemporal``
6813 metadata on the instruction tells the optimizer and code generator
6814 that this load is not expected to be reused in the cache. The code
6815 generator may select special instructions to save cache bandwidth, such
6816 as the ``MOVNT`` instruction on x86.
6818 The optional ``!invariant.load`` metadata must reference a single
6819 metadata name ``<index>`` corresponding to a metadata node with no
6820 entries. The existence of the ``!invariant.load`` metadata on the
6821 instruction tells the optimizer and code generator that the address
6822 operand to this load points to memory which can be assumed unchanged.
6823 Being invariant does not imply that a location is dereferenceable,
6824 but it does imply that once the location is known dereferenceable
6825 its value is henceforth unchanging.
6827 The optional ``!invariant.group`` metadata must reference a single metadata name
6828 ``<index>`` corresponding to a metadata node. See ``invariant.group`` metadata.
6830 The optional ``!nonnull`` metadata must reference a single
6831 metadata name ``<index>`` corresponding to a metadata node with no
6832 entries. The existence of the ``!nonnull`` metadata on the
6833 instruction tells the optimizer that the value loaded is known to
6834 never be null. This is analogous to the ``nonnull`` attribute
6835 on parameters and return values. This metadata can only be applied
6836 to loads of a pointer type.
6838 The optional ``!dereferenceable`` metadata must reference a single metadata
6839 name ``<deref_bytes_node>`` corresponding to a metadata node with one ``i64``
6840 entry. The existence of the ``!dereferenceable`` metadata on the instruction
6841 tells the optimizer that the value loaded is known to be dereferenceable.
6842 The number of bytes known to be dereferenceable is specified by the integer
6843 value in the metadata node. This is analogous to the ''dereferenceable''
6844 attribute on parameters and return values. This metadata can only be applied
6845 to loads of a pointer type.
6847 The optional ``!dereferenceable_or_null`` metadata must reference a single
6848 metadata name ``<deref_bytes_node>`` corresponding to a metadata node with one
6849 ``i64`` entry. The existence of the ``!dereferenceable_or_null`` metadata on the
6850 instruction tells the optimizer that the value loaded is known to be either
6851 dereferenceable or null.
6852 The number of bytes known to be dereferenceable is specified by the integer
6853 value in the metadata node. This is analogous to the ''dereferenceable_or_null''
6854 attribute on parameters and return values. This metadata can only be applied
6855 to loads of a pointer type.
6857 The optional ``!align`` metadata must reference a single metadata name
6858 ``<align_node>`` corresponding to a metadata node with one ``i64`` entry.
6859 The existence of the ``!align`` metadata on the instruction tells the
6860 optimizer that the value loaded is known to be aligned to a boundary specified
6861 by the integer value in the metadata node. The alignment must be a power of 2.
6862 This is analogous to the ''align'' attribute on parameters and return values.
6863 This metadata can only be applied to loads of a pointer type.
6868 The location of memory pointed to is loaded. If the value being loaded
6869 is of scalar type then the number of bytes read does not exceed the
6870 minimum number of bytes needed to hold all bits of the type. For
6871 example, loading an ``i24`` reads at most three bytes. When loading a
6872 value of a type like ``i20`` with a size that is not an integral number
6873 of bytes, the result is undefined if the value was not originally
6874 written using a store of the same type.
6879 .. code-block:: llvm
6881 %ptr = alloca i32 ; yields i32*:ptr
6882 store i32 3, i32* %ptr ; yields void
6883 %val = load i32, i32* %ptr ; yields i32:val = i32 3
6887 '``store``' Instruction
6888 ^^^^^^^^^^^^^^^^^^^^^^^
6895 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.group !<index>] ; yields void
6896 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>] ; yields void
6901 The '``store``' instruction is used to write to memory.
6906 There are two arguments to the ``store`` instruction: a value to store
6907 and an address at which to store it. The type of the ``<pointer>``
6908 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
6909 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
6910 then the optimizer is not allowed to modify the number or order of
6911 execution of this ``store`` with other :ref:`volatile
6912 operations <volatile>`.
6914 If the ``store`` is marked as ``atomic``, it takes an extra :ref:`ordering
6915 <ordering>` and optional ``singlethread`` argument. The ``acquire`` and
6916 ``acq_rel`` orderings aren't valid on ``store`` instructions. Atomic loads
6917 produce :ref:`defined <memmodel>` results when they may see multiple atomic
6918 stores. The type of the pointee must be an integer, pointer, or floating-point
6919 type whose bit width is a power of two greater than or equal to eight and less
6920 than or equal to a target-specific size limit. ``align`` must be explicitly
6921 specified on atomic stores, and the store has undefined behavior if the
6922 alignment is not set to a value which is at least the size in bytes of the
6923 pointee. ``!nontemporal`` does not have any defined semantics for atomic stores.
6925 The optional constant ``align`` argument specifies the alignment of the
6926 operation (that is, the alignment of the memory address). A value of 0
6927 or an omitted ``align`` argument means that the operation has the ABI
6928 alignment for the target. It is the responsibility of the code emitter
6929 to ensure that the alignment information is correct. Overestimating the
6930 alignment results in undefined behavior. Underestimating the
6931 alignment may produce less efficient code. An alignment of 1 is always
6932 safe. The maximum possible alignment is ``1 << 29``.
6934 The optional ``!nontemporal`` metadata must reference a single metadata
6935 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
6936 value 1. The existence of the ``!nontemporal`` metadata on the instruction
6937 tells the optimizer and code generator that this load is not expected to
6938 be reused in the cache. The code generator may select special
6939 instructions to save cache bandwidth, such as the ``MOVNT`` instruction on
6942 The optional ``!invariant.group`` metadata must reference a
6943 single metadata name ``<index>``. See ``invariant.group`` metadata.
6948 The contents of memory are updated to contain ``<value>`` at the
6949 location specified by the ``<pointer>`` operand. If ``<value>`` is
6950 of scalar type then the number of bytes written does not exceed the
6951 minimum number of bytes needed to hold all bits of the type. For
6952 example, storing an ``i24`` writes at most three bytes. When writing a
6953 value of a type like ``i20`` with a size that is not an integral number
6954 of bytes, it is unspecified what happens to the extra bits that do not
6955 belong to the type, but they will typically be overwritten.
6960 .. code-block:: llvm
6962 %ptr = alloca i32 ; yields i32*:ptr
6963 store i32 3, i32* %ptr ; yields void
6964 %val = load i32, i32* %ptr ; yields i32:val = i32 3
6968 '``fence``' Instruction
6969 ^^^^^^^^^^^^^^^^^^^^^^^
6976 fence [singlethread] <ordering> ; yields void
6981 The '``fence``' instruction is used to introduce happens-before edges
6987 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
6988 defines what *synchronizes-with* edges they add. They can only be given
6989 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
6994 A fence A which has (at least) ``release`` ordering semantics
6995 *synchronizes with* a fence B with (at least) ``acquire`` ordering
6996 semantics if and only if there exist atomic operations X and Y, both
6997 operating on some atomic object M, such that A is sequenced before X, X
6998 modifies M (either directly or through some side effect of a sequence
6999 headed by X), Y is sequenced before B, and Y observes M. This provides a
7000 *happens-before* dependency between A and B. Rather than an explicit
7001 ``fence``, one (but not both) of the atomic operations X or Y might
7002 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
7003 still *synchronize-with* the explicit ``fence`` and establish the
7004 *happens-before* edge.
7006 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
7007 ``acquire`` and ``release`` semantics specified above, participates in
7008 the global program order of other ``seq_cst`` operations and/or fences.
7010 The optional ":ref:`singlethread <singlethread>`" argument specifies
7011 that the fence only synchronizes with other fences in the same thread.
7012 (This is useful for interacting with signal handlers.)
7017 .. code-block:: llvm
7019 fence acquire ; yields void
7020 fence singlethread seq_cst ; yields void
7024 '``cmpxchg``' Instruction
7025 ^^^^^^^^^^^^^^^^^^^^^^^^^
7032 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
7037 The '``cmpxchg``' instruction is used to atomically modify memory. It
7038 loads a value in memory and compares it to a given value. If they are
7039 equal, it tries to store a new value into the memory.
7044 There are three arguments to the '``cmpxchg``' instruction: an address
7045 to operate on, a value to compare to the value currently be at that
7046 address, and a new value to place at that address if the compared values
7047 are equal. The type of '<cmp>' must be an integer type whose bit width
7048 is a power of two greater than or equal to eight and less than or equal
7049 to a target-specific size limit. '<cmp>' and '<new>' must have the same
7050 type, and the type of '<pointer>' must be a pointer to that type. If the
7051 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
7052 to modify the number or order of execution of this ``cmpxchg`` with
7053 other :ref:`volatile operations <volatile>`.
7055 The success and failure :ref:`ordering <ordering>` arguments specify how this
7056 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
7057 must be at least ``monotonic``, the ordering constraint on failure must be no
7058 stronger than that on success, and the failure ordering cannot be either
7059 ``release`` or ``acq_rel``.
7061 The optional "``singlethread``" argument declares that the ``cmpxchg``
7062 is only atomic with respect to code (usually signal handlers) running in
7063 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
7064 respect to all other code in the system.
7066 The pointer passed into cmpxchg must have alignment greater than or
7067 equal to the size in memory of the operand.
7072 The contents of memory at the location specified by the '``<pointer>``' operand
7073 is read and compared to '``<cmp>``'; if the read value is the equal, the
7074 '``<new>``' is written. The original value at the location is returned, together
7075 with a flag indicating success (true) or failure (false).
7077 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
7078 permitted: the operation may not write ``<new>`` even if the comparison
7081 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
7082 if the value loaded equals ``cmp``.
7084 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
7085 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
7086 load with an ordering parameter determined the second ordering parameter.
7091 .. code-block:: llvm
7094 %orig = atomic load i32, i32* %ptr unordered ; yields i32
7098 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
7099 %squared = mul i32 %cmp, %cmp
7100 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
7101 %value_loaded = extractvalue { i32, i1 } %val_success, 0
7102 %success = extractvalue { i32, i1 } %val_success, 1
7103 br i1 %success, label %done, label %loop
7110 '``atomicrmw``' Instruction
7111 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7118 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
7123 The '``atomicrmw``' instruction is used to atomically modify memory.
7128 There are three arguments to the '``atomicrmw``' instruction: an
7129 operation to apply, an address whose value to modify, an argument to the
7130 operation. The operation must be one of the following keywords:
7144 The type of '<value>' must be an integer type whose bit width is a power
7145 of two greater than or equal to eight and less than or equal to a
7146 target-specific size limit. The type of the '``<pointer>``' operand must
7147 be a pointer to that type. If the ``atomicrmw`` is marked as
7148 ``volatile``, then the optimizer is not allowed to modify the number or
7149 order of execution of this ``atomicrmw`` with other :ref:`volatile
7150 operations <volatile>`.
7155 The contents of memory at the location specified by the '``<pointer>``'
7156 operand are atomically read, modified, and written back. The original
7157 value at the location is returned. The modification is specified by the
7160 - xchg: ``*ptr = val``
7161 - add: ``*ptr = *ptr + val``
7162 - sub: ``*ptr = *ptr - val``
7163 - and: ``*ptr = *ptr & val``
7164 - nand: ``*ptr = ~(*ptr & val)``
7165 - or: ``*ptr = *ptr | val``
7166 - xor: ``*ptr = *ptr ^ val``
7167 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
7168 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
7169 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
7171 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
7177 .. code-block:: llvm
7179 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
7181 .. _i_getelementptr:
7183 '``getelementptr``' Instruction
7184 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7191 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7192 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7193 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
7198 The '``getelementptr``' instruction is used to get the address of a
7199 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
7200 address calculation only and does not access memory. The instruction can also
7201 be used to calculate a vector of such addresses.
7206 The first argument is always a type used as the basis for the calculations.
7207 The second argument is always a pointer or a vector of pointers, and is the
7208 base address to start from. The remaining arguments are indices
7209 that indicate which of the elements of the aggregate object are indexed.
7210 The interpretation of each index is dependent on the type being indexed
7211 into. The first index always indexes the pointer value given as the
7212 first argument, the second index indexes a value of the type pointed to
7213 (not necessarily the value directly pointed to, since the first index
7214 can be non-zero), etc. The first type indexed into must be a pointer
7215 value, subsequent types can be arrays, vectors, and structs. Note that
7216 subsequent types being indexed into can never be pointers, since that
7217 would require loading the pointer before continuing calculation.
7219 The type of each index argument depends on the type it is indexing into.
7220 When indexing into a (optionally packed) structure, only ``i32`` integer
7221 **constants** are allowed (when using a vector of indices they must all
7222 be the **same** ``i32`` integer constant). When indexing into an array,
7223 pointer or vector, integers of any width are allowed, and they are not
7224 required to be constant. These integers are treated as signed values
7227 For example, let's consider a C code fragment and how it gets compiled
7243 int *foo(struct ST *s) {
7244 return &s[1].Z.B[5][13];
7247 The LLVM code generated by Clang is:
7249 .. code-block:: llvm
7251 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
7252 %struct.ST = type { i32, double, %struct.RT }
7254 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
7256 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
7263 In the example above, the first index is indexing into the
7264 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
7265 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
7266 indexes into the third element of the structure, yielding a
7267 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
7268 structure. The third index indexes into the second element of the
7269 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
7270 dimensions of the array are subscripted into, yielding an '``i32``'
7271 type. The '``getelementptr``' instruction returns a pointer to this
7272 element, thus computing a value of '``i32*``' type.
7274 Note that it is perfectly legal to index partially through a structure,
7275 returning a pointer to an inner element. Because of this, the LLVM code
7276 for the given testcase is equivalent to:
7278 .. code-block:: llvm
7280 define i32* @foo(%struct.ST* %s) {
7281 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
7282 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
7283 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
7284 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
7285 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
7289 If the ``inbounds`` keyword is present, the result value of the
7290 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
7291 pointer is not an *in bounds* address of an allocated object, or if any
7292 of the addresses that would be formed by successive addition of the
7293 offsets implied by the indices to the base address with infinitely
7294 precise signed arithmetic are not an *in bounds* address of that
7295 allocated object. The *in bounds* addresses for an allocated object are
7296 all the addresses that point into the object, plus the address one byte
7297 past the end. In cases where the base is a vector of pointers the
7298 ``inbounds`` keyword applies to each of the computations element-wise.
7300 If the ``inbounds`` keyword is not present, the offsets are added to the
7301 base address with silently-wrapping two's complement arithmetic. If the
7302 offsets have a different width from the pointer, they are sign-extended
7303 or truncated to the width of the pointer. The result value of the
7304 ``getelementptr`` may be outside the object pointed to by the base
7305 pointer. The result value may not necessarily be used to access memory
7306 though, even if it happens to point into allocated storage. See the
7307 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
7310 The getelementptr instruction is often confusing. For some more insight
7311 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
7316 .. code-block:: llvm
7318 ; yields [12 x i8]*:aptr
7319 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
7321 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
7323 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
7325 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
7330 The ``getelementptr`` returns a vector of pointers, instead of a single address,
7331 when one or more of its arguments is a vector. In such cases, all vector
7332 arguments should have the same number of elements, and every scalar argument
7333 will be effectively broadcast into a vector during address calculation.
7335 .. code-block:: llvm
7337 ; All arguments are vectors:
7338 ; A[i] = ptrs[i] + offsets[i]*sizeof(i8)
7339 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets
7341 ; Add the same scalar offset to each pointer of a vector:
7342 ; A[i] = ptrs[i] + offset*sizeof(i8)
7343 %A = getelementptr i8, <4 x i8*> %ptrs, i64 %offset
7345 ; Add distinct offsets to the same pointer:
7346 ; A[i] = ptr + offsets[i]*sizeof(i8)
7347 %A = getelementptr i8, i8* %ptr, <4 x i64> %offsets
7349 ; In all cases described above the type of the result is <4 x i8*>
7351 The two following instructions are equivalent:
7353 .. code-block:: llvm
7355 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7356 <4 x i32> <i32 2, i32 2, i32 2, i32 2>,
7357 <4 x i32> <i32 1, i32 1, i32 1, i32 1>,
7359 <4 x i64> <i64 13, i64 13, i64 13, i64 13>
7361 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7362 i32 2, i32 1, <4 x i32> %ind4, i64 13
7364 Let's look at the C code, where the vector version of ``getelementptr``
7369 // Let's assume that we vectorize the following loop:
7370 double *A, B; int *C;
7371 for (int i = 0; i < size; ++i) {
7375 .. code-block:: llvm
7377 ; get pointers for 8 elements from array B
7378 %ptrs = getelementptr double, double* %B, <8 x i32> %C
7379 ; load 8 elements from array B into A
7380 %A = call <8 x double> @llvm.masked.gather.v8f64(<8 x double*> %ptrs,
7381 i32 8, <8 x i1> %mask, <8 x double> %passthru)
7383 Conversion Operations
7384 ---------------------
7386 The instructions in this category are the conversion instructions
7387 (casting) which all take a single operand and a type. They perform
7388 various bit conversions on the operand.
7390 '``trunc .. to``' Instruction
7391 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7398 <result> = trunc <ty> <value> to <ty2> ; yields ty2
7403 The '``trunc``' instruction truncates its operand to the type ``ty2``.
7408 The '``trunc``' instruction takes a value to trunc, and a type to trunc
7409 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
7410 of the same number of integers. The bit size of the ``value`` must be
7411 larger than the bit size of the destination type, ``ty2``. Equal sized
7412 types are not allowed.
7417 The '``trunc``' instruction truncates the high order bits in ``value``
7418 and converts the remaining bits to ``ty2``. Since the source size must
7419 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
7420 It will always truncate bits.
7425 .. code-block:: llvm
7427 %X = trunc i32 257 to i8 ; yields i8:1
7428 %Y = trunc i32 123 to i1 ; yields i1:true
7429 %Z = trunc i32 122 to i1 ; yields i1:false
7430 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
7432 '``zext .. to``' Instruction
7433 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7440 <result> = zext <ty> <value> to <ty2> ; yields ty2
7445 The '``zext``' instruction zero extends its operand to type ``ty2``.
7450 The '``zext``' instruction takes a value to cast, and a type to cast it
7451 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7452 the same number of integers. The bit size of the ``value`` must be
7453 smaller than the bit size of the destination type, ``ty2``.
7458 The ``zext`` fills the high order bits of the ``value`` with zero bits
7459 until it reaches the size of the destination type, ``ty2``.
7461 When zero extending from i1, the result will always be either 0 or 1.
7466 .. code-block:: llvm
7468 %X = zext i32 257 to i64 ; yields i64:257
7469 %Y = zext i1 true to i32 ; yields i32:1
7470 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7472 '``sext .. to``' Instruction
7473 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7480 <result> = sext <ty> <value> to <ty2> ; yields ty2
7485 The '``sext``' sign extends ``value`` to the type ``ty2``.
7490 The '``sext``' instruction takes a value to cast, and a type to cast it
7491 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7492 the same number of integers. The bit size of the ``value`` must be
7493 smaller than the bit size of the destination type, ``ty2``.
7498 The '``sext``' instruction performs a sign extension by copying the sign
7499 bit (highest order bit) of the ``value`` until it reaches the bit size
7500 of the type ``ty2``.
7502 When sign extending from i1, the extension always results in -1 or 0.
7507 .. code-block:: llvm
7509 %X = sext i8 -1 to i16 ; yields i16 :65535
7510 %Y = sext i1 true to i32 ; yields i32:-1
7511 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7513 '``fptrunc .. to``' Instruction
7514 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7521 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
7526 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
7531 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
7532 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
7533 The size of ``value`` must be larger than the size of ``ty2``. This
7534 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
7539 The '``fptrunc``' instruction casts a ``value`` from a larger
7540 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
7541 point <t_floating>` type. If the value cannot fit (i.e. overflows) within the
7542 destination type, ``ty2``, then the results are undefined. If the cast produces
7543 an inexact result, how rounding is performed (e.g. truncation, also known as
7544 round to zero) is undefined.
7549 .. code-block:: llvm
7551 %X = fptrunc double 123.0 to float ; yields float:123.0
7552 %Y = fptrunc double 1.0E+300 to float ; yields undefined
7554 '``fpext .. to``' Instruction
7555 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7562 <result> = fpext <ty> <value> to <ty2> ; yields ty2
7567 The '``fpext``' extends a floating point ``value`` to a larger floating
7573 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
7574 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
7575 to. The source type must be smaller than the destination type.
7580 The '``fpext``' instruction extends the ``value`` from a smaller
7581 :ref:`floating point <t_floating>` type to a larger :ref:`floating
7582 point <t_floating>` type. The ``fpext`` cannot be used to make a
7583 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
7584 *no-op cast* for a floating point cast.
7589 .. code-block:: llvm
7591 %X = fpext float 3.125 to double ; yields double:3.125000e+00
7592 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
7594 '``fptoui .. to``' Instruction
7595 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7602 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
7607 The '``fptoui``' converts a floating point ``value`` to its unsigned
7608 integer equivalent of type ``ty2``.
7613 The '``fptoui``' instruction takes a value to cast, which must be a
7614 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7615 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7616 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7617 type with the same number of elements as ``ty``
7622 The '``fptoui``' instruction converts its :ref:`floating
7623 point <t_floating>` operand into the nearest (rounding towards zero)
7624 unsigned integer value. If the value cannot fit in ``ty2``, the results
7630 .. code-block:: llvm
7632 %X = fptoui double 123.0 to i32 ; yields i32:123
7633 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
7634 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
7636 '``fptosi .. to``' Instruction
7637 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7644 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
7649 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
7650 ``value`` to type ``ty2``.
7655 The '``fptosi``' instruction takes a value to cast, which must be a
7656 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7657 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7658 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7659 type with the same number of elements as ``ty``
7664 The '``fptosi``' instruction converts its :ref:`floating
7665 point <t_floating>` operand into the nearest (rounding towards zero)
7666 signed integer value. If the value cannot fit in ``ty2``, the results
7672 .. code-block:: llvm
7674 %X = fptosi double -123.0 to i32 ; yields i32:-123
7675 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
7676 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
7678 '``uitofp .. to``' Instruction
7679 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7686 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
7691 The '``uitofp``' instruction regards ``value`` as an unsigned integer
7692 and converts that value to the ``ty2`` type.
7697 The '``uitofp``' instruction takes a value to cast, which must be a
7698 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7699 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7700 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7701 type with the same number of elements as ``ty``
7706 The '``uitofp``' instruction interprets its operand as an unsigned
7707 integer quantity and converts it to the corresponding floating point
7708 value. If the value cannot fit in the floating point value, the results
7714 .. code-block:: llvm
7716 %X = uitofp i32 257 to float ; yields float:257.0
7717 %Y = uitofp i8 -1 to double ; yields double:255.0
7719 '``sitofp .. to``' Instruction
7720 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7727 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
7732 The '``sitofp``' instruction regards ``value`` as a signed integer and
7733 converts that value to the ``ty2`` type.
7738 The '``sitofp``' instruction takes a value to cast, which must be a
7739 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7740 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7741 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7742 type with the same number of elements as ``ty``
7747 The '``sitofp``' instruction interprets its operand as a signed integer
7748 quantity and converts it to the corresponding floating point value. If
7749 the value cannot fit in the floating point value, the results are
7755 .. code-block:: llvm
7757 %X = sitofp i32 257 to float ; yields float:257.0
7758 %Y = sitofp i8 -1 to double ; yields double:-1.0
7762 '``ptrtoint .. to``' Instruction
7763 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7770 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
7775 The '``ptrtoint``' instruction converts the pointer or a vector of
7776 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
7781 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
7782 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
7783 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
7784 a vector of integers type.
7789 The '``ptrtoint``' instruction converts ``value`` to integer type
7790 ``ty2`` by interpreting the pointer value as an integer and either
7791 truncating or zero extending that value to the size of the integer type.
7792 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
7793 ``value`` is larger than ``ty2`` then a truncation is done. If they are
7794 the same size, then nothing is done (*no-op cast*) other than a type
7800 .. code-block:: llvm
7802 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
7803 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
7804 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
7808 '``inttoptr .. to``' Instruction
7809 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7816 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
7821 The '``inttoptr``' instruction converts an integer ``value`` to a
7822 pointer type, ``ty2``.
7827 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
7828 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
7834 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
7835 applying either a zero extension or a truncation depending on the size
7836 of the integer ``value``. If ``value`` is larger than the size of a
7837 pointer then a truncation is done. If ``value`` is smaller than the size
7838 of a pointer then a zero extension is done. If they are the same size,
7839 nothing is done (*no-op cast*).
7844 .. code-block:: llvm
7846 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
7847 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
7848 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
7849 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
7853 '``bitcast .. to``' Instruction
7854 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7861 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
7866 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
7872 The '``bitcast``' instruction takes a value to cast, which must be a
7873 non-aggregate first class value, and a type to cast it to, which must
7874 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
7875 bit sizes of ``value`` and the destination type, ``ty2``, must be
7876 identical. If the source type is a pointer, the destination type must
7877 also be a pointer of the same size. This instruction supports bitwise
7878 conversion of vectors to integers and to vectors of other types (as
7879 long as they have the same size).
7884 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
7885 is always a *no-op cast* because no bits change with this
7886 conversion. The conversion is done as if the ``value`` had been stored
7887 to memory and read back as type ``ty2``. Pointer (or vector of
7888 pointers) types may only be converted to other pointer (or vector of
7889 pointers) types with the same address space through this instruction.
7890 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
7891 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
7896 .. code-block:: llvm
7898 %X = bitcast i8 255 to i8 ; yields i8 :-1
7899 %Y = bitcast i32* %x to sint* ; yields sint*:%x
7900 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
7901 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
7903 .. _i_addrspacecast:
7905 '``addrspacecast .. to``' Instruction
7906 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7913 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
7918 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
7919 address space ``n`` to type ``pty2`` in address space ``m``.
7924 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
7925 to cast and a pointer type to cast it to, which must have a different
7931 The '``addrspacecast``' instruction converts the pointer value
7932 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
7933 value modification, depending on the target and the address space
7934 pair. Pointer conversions within the same address space must be
7935 performed with the ``bitcast`` instruction. Note that if the address space
7936 conversion is legal then both result and operand refer to the same memory
7942 .. code-block:: llvm
7944 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
7945 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
7946 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
7953 The instructions in this category are the "miscellaneous" instructions,
7954 which defy better classification.
7958 '``icmp``' Instruction
7959 ^^^^^^^^^^^^^^^^^^^^^^
7966 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
7971 The '``icmp``' instruction returns a boolean value or a vector of
7972 boolean values based on comparison of its two integer, integer vector,
7973 pointer, or pointer vector operands.
7978 The '``icmp``' instruction takes three operands. The first operand is
7979 the condition code indicating the kind of comparison to perform. It is
7980 not a value, just a keyword. The possible condition code are:
7983 #. ``ne``: not equal
7984 #. ``ugt``: unsigned greater than
7985 #. ``uge``: unsigned greater or equal
7986 #. ``ult``: unsigned less than
7987 #. ``ule``: unsigned less or equal
7988 #. ``sgt``: signed greater than
7989 #. ``sge``: signed greater or equal
7990 #. ``slt``: signed less than
7991 #. ``sle``: signed less or equal
7993 The remaining two arguments must be :ref:`integer <t_integer>` or
7994 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
7995 must also be identical types.
8000 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
8001 code given as ``cond``. The comparison performed always yields either an
8002 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
8004 #. ``eq``: yields ``true`` if the operands are equal, ``false``
8005 otherwise. No sign interpretation is necessary or performed.
8006 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
8007 otherwise. No sign interpretation is necessary or performed.
8008 #. ``ugt``: interprets the operands as unsigned values and yields
8009 ``true`` if ``op1`` is greater than ``op2``.
8010 #. ``uge``: interprets the operands as unsigned values and yields
8011 ``true`` if ``op1`` is greater than or equal to ``op2``.
8012 #. ``ult``: interprets the operands as unsigned values and yields
8013 ``true`` if ``op1`` is less than ``op2``.
8014 #. ``ule``: interprets the operands as unsigned values and yields
8015 ``true`` if ``op1`` is less than or equal to ``op2``.
8016 #. ``sgt``: interprets the operands as signed values and yields ``true``
8017 if ``op1`` is greater than ``op2``.
8018 #. ``sge``: interprets the operands as signed values and yields ``true``
8019 if ``op1`` is greater than or equal to ``op2``.
8020 #. ``slt``: interprets the operands as signed values and yields ``true``
8021 if ``op1`` is less than ``op2``.
8022 #. ``sle``: interprets the operands as signed values and yields ``true``
8023 if ``op1`` is less than or equal to ``op2``.
8025 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
8026 are compared as if they were integers.
8028 If the operands are integer vectors, then they are compared element by
8029 element. The result is an ``i1`` vector with the same number of elements
8030 as the values being compared. Otherwise, the result is an ``i1``.
8035 .. code-block:: llvm
8037 <result> = icmp eq i32 4, 5 ; yields: result=false
8038 <result> = icmp ne float* %X, %X ; yields: result=false
8039 <result> = icmp ult i16 4, 5 ; yields: result=true
8040 <result> = icmp sgt i16 4, 5 ; yields: result=false
8041 <result> = icmp ule i16 -4, 5 ; yields: result=false
8042 <result> = icmp sge i16 4, 5 ; yields: result=false
8044 Note that the code generator does not yet support vector types with the
8045 ``icmp`` instruction.
8049 '``fcmp``' Instruction
8050 ^^^^^^^^^^^^^^^^^^^^^^
8057 <result> = fcmp [fast-math flags]* <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8062 The '``fcmp``' instruction returns a boolean value or vector of boolean
8063 values based on comparison of its operands.
8065 If the operands are floating point scalars, then the result type is a
8066 boolean (:ref:`i1 <t_integer>`).
8068 If the operands are floating point vectors, then the result type is a
8069 vector of boolean with the same number of elements as the operands being
8075 The '``fcmp``' instruction takes three operands. The first operand is
8076 the condition code indicating the kind of comparison to perform. It is
8077 not a value, just a keyword. The possible condition code are:
8079 #. ``false``: no comparison, always returns false
8080 #. ``oeq``: ordered and equal
8081 #. ``ogt``: ordered and greater than
8082 #. ``oge``: ordered and greater than or equal
8083 #. ``olt``: ordered and less than
8084 #. ``ole``: ordered and less than or equal
8085 #. ``one``: ordered and not equal
8086 #. ``ord``: ordered (no nans)
8087 #. ``ueq``: unordered or equal
8088 #. ``ugt``: unordered or greater than
8089 #. ``uge``: unordered or greater than or equal
8090 #. ``ult``: unordered or less than
8091 #. ``ule``: unordered or less than or equal
8092 #. ``une``: unordered or not equal
8093 #. ``uno``: unordered (either nans)
8094 #. ``true``: no comparison, always returns true
8096 *Ordered* means that neither operand is a QNAN while *unordered* means
8097 that either operand may be a QNAN.
8099 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
8100 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
8101 type. They must have identical types.
8106 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
8107 condition code given as ``cond``. If the operands are vectors, then the
8108 vectors are compared element by element. Each comparison performed
8109 always yields an :ref:`i1 <t_integer>` result, as follows:
8111 #. ``false``: always yields ``false``, regardless of operands.
8112 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
8113 is equal to ``op2``.
8114 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
8115 is greater than ``op2``.
8116 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
8117 is greater than or equal to ``op2``.
8118 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
8119 is less than ``op2``.
8120 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
8121 is less than or equal to ``op2``.
8122 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
8123 is not equal to ``op2``.
8124 #. ``ord``: yields ``true`` if both operands are not a QNAN.
8125 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
8127 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
8128 greater than ``op2``.
8129 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
8130 greater than or equal to ``op2``.
8131 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
8133 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
8134 less than or equal to ``op2``.
8135 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
8136 not equal to ``op2``.
8137 #. ``uno``: yields ``true`` if either operand is a QNAN.
8138 #. ``true``: always yields ``true``, regardless of operands.
8140 The ``fcmp`` instruction can also optionally take any number of
8141 :ref:`fast-math flags <fastmath>`, which are optimization hints to enable
8142 otherwise unsafe floating point optimizations.
8144 Any set of fast-math flags are legal on an ``fcmp`` instruction, but the
8145 only flags that have any effect on its semantics are those that allow
8146 assumptions to be made about the values of input arguments; namely
8147 ``nnan``, ``ninf``, and ``nsz``. See :ref:`fastmath` for more information.
8152 .. code-block:: llvm
8154 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
8155 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
8156 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
8157 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
8159 Note that the code generator does not yet support vector types with the
8160 ``fcmp`` instruction.
8164 '``phi``' Instruction
8165 ^^^^^^^^^^^^^^^^^^^^^
8172 <result> = phi <ty> [ <val0>, <label0>], ...
8177 The '``phi``' instruction is used to implement the φ node in the SSA
8178 graph representing the function.
8183 The type of the incoming values is specified with the first type field.
8184 After this, the '``phi``' instruction takes a list of pairs as
8185 arguments, with one pair for each predecessor basic block of the current
8186 block. Only values of :ref:`first class <t_firstclass>` type may be used as
8187 the value arguments to the PHI node. Only labels may be used as the
8190 There must be no non-phi instructions between the start of a basic block
8191 and the PHI instructions: i.e. PHI instructions must be first in a basic
8194 For the purposes of the SSA form, the use of each incoming value is
8195 deemed to occur on the edge from the corresponding predecessor block to
8196 the current block (but after any definition of an '``invoke``'
8197 instruction's return value on the same edge).
8202 At runtime, the '``phi``' instruction logically takes on the value
8203 specified by the pair corresponding to the predecessor basic block that
8204 executed just prior to the current block.
8209 .. code-block:: llvm
8211 Loop: ; Infinite loop that counts from 0 on up...
8212 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
8213 %nextindvar = add i32 %indvar, 1
8218 '``select``' Instruction
8219 ^^^^^^^^^^^^^^^^^^^^^^^^
8226 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
8228 selty is either i1 or {<N x i1>}
8233 The '``select``' instruction is used to choose one value based on a
8234 condition, without IR-level branching.
8239 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
8240 values indicating the condition, and two values of the same :ref:`first
8241 class <t_firstclass>` type.
8246 If the condition is an i1 and it evaluates to 1, the instruction returns
8247 the first value argument; otherwise, it returns the second value
8250 If the condition is a vector of i1, then the value arguments must be
8251 vectors of the same size, and the selection is done element by element.
8253 If the condition is an i1 and the value arguments are vectors of the
8254 same size, then an entire vector is selected.
8259 .. code-block:: llvm
8261 %X = select i1 true, i8 17, i8 42 ; yields i8:17
8265 '``call``' Instruction
8266 ^^^^^^^^^^^^^^^^^^^^^^
8273 <result> = [tail | musttail | notail ] call [fast-math flags] [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
8279 The '``call``' instruction represents a simple function call.
8284 This instruction requires several arguments:
8286 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
8287 should perform tail call optimization. The ``tail`` marker is a hint that
8288 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
8289 means that the call must be tail call optimized in order for the program to
8290 be correct. The ``musttail`` marker provides these guarantees:
8292 #. The call will not cause unbounded stack growth if it is part of a
8293 recursive cycle in the call graph.
8294 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
8297 Both markers imply that the callee does not access allocas or varargs from
8298 the caller. Calls marked ``musttail`` must obey the following additional
8301 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
8302 or a pointer bitcast followed by a ret instruction.
8303 - The ret instruction must return the (possibly bitcasted) value
8304 produced by the call or void.
8305 - The caller and callee prototypes must match. Pointer types of
8306 parameters or return types may differ in pointee type, but not
8308 - The calling conventions of the caller and callee must match.
8309 - All ABI-impacting function attributes, such as sret, byval, inreg,
8310 returned, and inalloca, must match.
8311 - The callee must be varargs iff the caller is varargs. Bitcasting a
8312 non-varargs function to the appropriate varargs type is legal so
8313 long as the non-varargs prefixes obey the other rules.
8315 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
8316 the following conditions are met:
8318 - Caller and callee both have the calling convention ``fastcc``.
8319 - The call is in tail position (ret immediately follows call and ret
8320 uses value of call or is void).
8321 - Option ``-tailcallopt`` is enabled, or
8322 ``llvm::GuaranteedTailCallOpt`` is ``true``.
8323 - `Platform-specific constraints are
8324 met. <CodeGenerator.html#tailcallopt>`_
8326 #. The optional ``notail`` marker indicates that the optimizers should not add
8327 ``tail`` or ``musttail`` markers to the call. It is used to prevent tail
8328 call optimization from being performed on the call.
8330 #. The optional ``fast-math flags`` marker indicates that the call has one or more
8331 :ref:`fast-math flags <fastmath>`, which are optimization hints to enable
8332 otherwise unsafe floating-point optimizations. Fast-math flags are only valid
8333 for calls that return a floating-point scalar or vector type.
8335 #. The optional "cconv" marker indicates which :ref:`calling
8336 convention <callingconv>` the call should use. If none is
8337 specified, the call defaults to using C calling conventions. The
8338 calling convention of the call must match the calling convention of
8339 the target function, or else the behavior is undefined.
8340 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
8341 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
8343 #. '``ty``': the type of the call instruction itself which is also the
8344 type of the return value. Functions that return no value are marked
8346 #. '``fnty``': shall be the signature of the pointer to function value
8347 being invoked. The argument types must match the types implied by
8348 this signature. This type can be omitted if the function is not
8349 varargs and if the function type does not return a pointer to a
8351 #. '``fnptrval``': An LLVM value containing a pointer to a function to
8352 be invoked. In most cases, this is a direct function invocation, but
8353 indirect ``call``'s are just as possible, calling an arbitrary pointer
8355 #. '``function args``': argument list whose types match the function
8356 signature argument types and parameter attributes. All arguments must
8357 be of :ref:`first class <t_firstclass>` type. If the function signature
8358 indicates the function accepts a variable number of arguments, the
8359 extra arguments can be specified.
8360 #. The optional :ref:`function attributes <fnattrs>` list. Only
8361 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
8362 attributes are valid here.
8363 #. The optional :ref:`operand bundles <opbundles>` list.
8368 The '``call``' instruction is used to cause control flow to transfer to
8369 a specified function, with its incoming arguments bound to the specified
8370 values. Upon a '``ret``' instruction in the called function, control
8371 flow continues with the instruction after the function call, and the
8372 return value of the function is bound to the result argument.
8377 .. code-block:: llvm
8379 %retval = call i32 @test(i32 %argc)
8380 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
8381 %X = tail call i32 @foo() ; yields i32
8382 %Y = tail call fastcc i32 @foo() ; yields i32
8383 call void %foo(i8 97 signext)
8385 %struct.A = type { i32, i8 }
8386 %r = call %struct.A @foo() ; yields { i32, i8 }
8387 %gr = extractvalue %struct.A %r, 0 ; yields i32
8388 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
8389 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
8390 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
8392 llvm treats calls to some functions with names and arguments that match
8393 the standard C99 library as being the C99 library functions, and may
8394 perform optimizations or generate code for them under that assumption.
8395 This is something we'd like to change in the future to provide better
8396 support for freestanding environments and non-C-based languages.
8400 '``va_arg``' Instruction
8401 ^^^^^^^^^^^^^^^^^^^^^^^^
8408 <resultval> = va_arg <va_list*> <arglist>, <argty>
8413 The '``va_arg``' instruction is used to access arguments passed through
8414 the "variable argument" area of a function call. It is used to implement
8415 the ``va_arg`` macro in C.
8420 This instruction takes a ``va_list*`` value and the type of the
8421 argument. It returns a value of the specified argument type and
8422 increments the ``va_list`` to point to the next argument. The actual
8423 type of ``va_list`` is target specific.
8428 The '``va_arg``' instruction loads an argument of the specified type
8429 from the specified ``va_list`` and causes the ``va_list`` to point to
8430 the next argument. For more information, see the variable argument
8431 handling :ref:`Intrinsic Functions <int_varargs>`.
8433 It is legal for this instruction to be called in a function which does
8434 not take a variable number of arguments, for example, the ``vfprintf``
8437 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
8438 function <intrinsics>` because it takes a type as an argument.
8443 See the :ref:`variable argument processing <int_varargs>` section.
8445 Note that the code generator does not yet fully support va\_arg on many
8446 targets. Also, it does not currently support va\_arg with aggregate
8447 types on any target.
8451 '``landingpad``' Instruction
8452 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8459 <resultval> = landingpad <resultty> <clause>+
8460 <resultval> = landingpad <resultty> cleanup <clause>*
8462 <clause> := catch <type> <value>
8463 <clause> := filter <array constant type> <array constant>
8468 The '``landingpad``' instruction is used by `LLVM's exception handling
8469 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8470 is a landing pad --- one where the exception lands, and corresponds to the
8471 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
8472 defines values supplied by the :ref:`personality function <personalityfn>` upon
8473 re-entry to the function. The ``resultval`` has the type ``resultty``.
8479 ``cleanup`` flag indicates that the landing pad block is a cleanup.
8481 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
8482 contains the global variable representing the "type" that may be caught
8483 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
8484 clause takes an array constant as its argument. Use
8485 "``[0 x i8**] undef``" for a filter which cannot throw. The
8486 '``landingpad``' instruction must contain *at least* one ``clause`` or
8487 the ``cleanup`` flag.
8492 The '``landingpad``' instruction defines the values which are set by the
8493 :ref:`personality function <personalityfn>` upon re-entry to the function, and
8494 therefore the "result type" of the ``landingpad`` instruction. As with
8495 calling conventions, how the personality function results are
8496 represented in LLVM IR is target specific.
8498 The clauses are applied in order from top to bottom. If two
8499 ``landingpad`` instructions are merged together through inlining, the
8500 clauses from the calling function are appended to the list of clauses.
8501 When the call stack is being unwound due to an exception being thrown,
8502 the exception is compared against each ``clause`` in turn. If it doesn't
8503 match any of the clauses, and the ``cleanup`` flag is not set, then
8504 unwinding continues further up the call stack.
8506 The ``landingpad`` instruction has several restrictions:
8508 - A landing pad block is a basic block which is the unwind destination
8509 of an '``invoke``' instruction.
8510 - A landing pad block must have a '``landingpad``' instruction as its
8511 first non-PHI instruction.
8512 - There can be only one '``landingpad``' instruction within the landing
8514 - A basic block that is not a landing pad block may not include a
8515 '``landingpad``' instruction.
8520 .. code-block:: llvm
8522 ;; A landing pad which can catch an integer.
8523 %res = landingpad { i8*, i32 }
8525 ;; A landing pad that is a cleanup.
8526 %res = landingpad { i8*, i32 }
8528 ;; A landing pad which can catch an integer and can only throw a double.
8529 %res = landingpad { i8*, i32 }
8531 filter [1 x i8**] [@_ZTId]
8535 '``catchpad``' Instruction
8536 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8543 <resultval> = catchpad within <catchswitch> [<args>*]
8548 The '``catchpad``' instruction is used by `LLVM's exception handling
8549 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8550 begins a catch handler --- one where a personality routine attempts to transfer
8551 control to catch an exception.
8556 The ``catchswitch`` operand must always be a token produced by a
8557 :ref:`catchswitch <i_catchswitch>` instruction in a predecessor block. This
8558 ensures that each ``catchpad`` has exactly one predecessor block, and it always
8559 terminates in a ``catchswitch``.
8561 The ``args`` correspond to whatever information the personality routine
8562 requires to know if this is an appropriate handler for the exception. Control
8563 will transfer to the ``catchpad`` if this is the first appropriate handler for
8566 The ``resultval`` has the type :ref:`token <t_token>` and is used to match the
8567 ``catchpad`` to corresponding :ref:`catchrets <i_catchret>` and other nested EH
8573 When the call stack is being unwound due to an exception being thrown, the
8574 exception is compared against the ``args``. If it doesn't match, control will
8575 not reach the ``catchpad`` instruction. The representation of ``args`` is
8576 entirely target and personality function-specific.
8578 Like the :ref:`landingpad <i_landingpad>` instruction, the ``catchpad``
8579 instruction must be the first non-phi of its parent basic block.
8581 The meaning of the tokens produced and consumed by ``catchpad`` and other "pad"
8582 instructions is described in the
8583 `Windows exception handling documentation\ <ExceptionHandling.html#wineh>`_.
8585 When a ``catchpad`` has been "entered" but not yet "exited" (as
8586 described in the `EH documentation\ <ExceptionHandling.html#wineh-constraints>`_),
8587 it is undefined behavior to execute a :ref:`call <i_call>` or :ref:`invoke <i_invoke>`
8588 that does not carry an appropriate :ref:`"funclet" bundle <ob_funclet>`.
8593 .. code-block:: llvm
8596 %cs = catchswitch within none [label %handler0] unwind to caller
8597 ;; A catch block which can catch an integer.
8599 %tok = catchpad within %cs [i8** @_ZTIi]
8603 '``cleanuppad``' Instruction
8604 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8611 <resultval> = cleanuppad within <parent> [<args>*]
8616 The '``cleanuppad``' instruction is used by `LLVM's exception handling
8617 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8618 is a cleanup block --- one where a personality routine attempts to
8619 transfer control to run cleanup actions.
8620 The ``args`` correspond to whatever additional
8621 information the :ref:`personality function <personalityfn>` requires to
8622 execute the cleanup.
8623 The ``resultval`` has the type :ref:`token <t_token>` and is used to
8624 match the ``cleanuppad`` to corresponding :ref:`cleanuprets <i_cleanupret>`.
8625 The ``parent`` argument is the token of the funclet that contains the
8626 ``cleanuppad`` instruction. If the ``cleanuppad`` is not inside a funclet,
8627 this operand may be the token ``none``.
8632 The instruction takes a list of arbitrary values which are interpreted
8633 by the :ref:`personality function <personalityfn>`.
8638 When the call stack is being unwound due to an exception being thrown,
8639 the :ref:`personality function <personalityfn>` transfers control to the
8640 ``cleanuppad`` with the aid of the personality-specific arguments.
8641 As with calling conventions, how the personality function results are
8642 represented in LLVM IR is target specific.
8644 The ``cleanuppad`` instruction has several restrictions:
8646 - A cleanup block is a basic block which is the unwind destination of
8647 an exceptional instruction.
8648 - A cleanup block must have a '``cleanuppad``' instruction as its
8649 first non-PHI instruction.
8650 - There can be only one '``cleanuppad``' instruction within the
8652 - A basic block that is not a cleanup block may not include a
8653 '``cleanuppad``' instruction.
8655 When a ``cleanuppad`` has been "entered" but not yet "exited" (as
8656 described in the `EH documentation\ <ExceptionHandling.html#wineh-constraints>`_),
8657 it is undefined behavior to execute a :ref:`call <i_call>` or :ref:`invoke <i_invoke>`
8658 that does not carry an appropriate :ref:`"funclet" bundle <ob_funclet>`.
8663 .. code-block:: llvm
8665 %tok = cleanuppad within %cs []
8672 LLVM supports the notion of an "intrinsic function". These functions
8673 have well known names and semantics and are required to follow certain
8674 restrictions. Overall, these intrinsics represent an extension mechanism
8675 for the LLVM language that does not require changing all of the
8676 transformations in LLVM when adding to the language (or the bitcode
8677 reader/writer, the parser, etc...).
8679 Intrinsic function names must all start with an "``llvm.``" prefix. This
8680 prefix is reserved in LLVM for intrinsic names; thus, function names may
8681 not begin with this prefix. Intrinsic functions must always be external
8682 functions: you cannot define the body of intrinsic functions. Intrinsic
8683 functions may only be used in call or invoke instructions: it is illegal
8684 to take the address of an intrinsic function. Additionally, because
8685 intrinsic functions are part of the LLVM language, it is required if any
8686 are added that they be documented here.
8688 Some intrinsic functions can be overloaded, i.e., the intrinsic
8689 represents a family of functions that perform the same operation but on
8690 different data types. Because LLVM can represent over 8 million
8691 different integer types, overloading is used commonly to allow an
8692 intrinsic function to operate on any integer type. One or more of the
8693 argument types or the result type can be overloaded to accept any
8694 integer type. Argument types may also be defined as exactly matching a
8695 previous argument's type or the result type. This allows an intrinsic
8696 function which accepts multiple arguments, but needs all of them to be
8697 of the same type, to only be overloaded with respect to a single
8698 argument or the result.
8700 Overloaded intrinsics will have the names of its overloaded argument
8701 types encoded into its function name, each preceded by a period. Only
8702 those types which are overloaded result in a name suffix. Arguments
8703 whose type is matched against another type do not. For example, the
8704 ``llvm.ctpop`` function can take an integer of any width and returns an
8705 integer of exactly the same integer width. This leads to a family of
8706 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
8707 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
8708 overloaded, and only one type suffix is required. Because the argument's
8709 type is matched against the return type, it does not require its own
8712 To learn how to add an intrinsic function, please see the `Extending
8713 LLVM Guide <ExtendingLLVM.html>`_.
8717 Variable Argument Handling Intrinsics
8718 -------------------------------------
8720 Variable argument support is defined in LLVM with the
8721 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
8722 functions. These functions are related to the similarly named macros
8723 defined in the ``<stdarg.h>`` header file.
8725 All of these functions operate on arguments that use a target-specific
8726 value type "``va_list``". The LLVM assembly language reference manual
8727 does not define what this type is, so all transformations should be
8728 prepared to handle these functions regardless of the type used.
8730 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
8731 variable argument handling intrinsic functions are used.
8733 .. code-block:: llvm
8735 ; This struct is different for every platform. For most platforms,
8736 ; it is merely an i8*.
8737 %struct.va_list = type { i8* }
8739 ; For Unix x86_64 platforms, va_list is the following struct:
8740 ; %struct.va_list = type { i32, i32, i8*, i8* }
8742 define i32 @test(i32 %X, ...) {
8743 ; Initialize variable argument processing
8744 %ap = alloca %struct.va_list
8745 %ap2 = bitcast %struct.va_list* %ap to i8*
8746 call void @llvm.va_start(i8* %ap2)
8748 ; Read a single integer argument
8749 %tmp = va_arg i8* %ap2, i32
8751 ; Demonstrate usage of llvm.va_copy and llvm.va_end
8753 %aq2 = bitcast i8** %aq to i8*
8754 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
8755 call void @llvm.va_end(i8* %aq2)
8757 ; Stop processing of arguments.
8758 call void @llvm.va_end(i8* %ap2)
8762 declare void @llvm.va_start(i8*)
8763 declare void @llvm.va_copy(i8*, i8*)
8764 declare void @llvm.va_end(i8*)
8768 '``llvm.va_start``' Intrinsic
8769 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8776 declare void @llvm.va_start(i8* <arglist>)
8781 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
8782 subsequent use by ``va_arg``.
8787 The argument is a pointer to a ``va_list`` element to initialize.
8792 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
8793 available in C. In a target-dependent way, it initializes the
8794 ``va_list`` element to which the argument points, so that the next call
8795 to ``va_arg`` will produce the first variable argument passed to the
8796 function. Unlike the C ``va_start`` macro, this intrinsic does not need
8797 to know the last argument of the function as the compiler can figure
8800 '``llvm.va_end``' Intrinsic
8801 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8808 declare void @llvm.va_end(i8* <arglist>)
8813 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
8814 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
8819 The argument is a pointer to a ``va_list`` to destroy.
8824 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
8825 available in C. In a target-dependent way, it destroys the ``va_list``
8826 element to which the argument points. Calls to
8827 :ref:`llvm.va_start <int_va_start>` and
8828 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
8833 '``llvm.va_copy``' Intrinsic
8834 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8841 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
8846 The '``llvm.va_copy``' intrinsic copies the current argument position
8847 from the source argument list to the destination argument list.
8852 The first argument is a pointer to a ``va_list`` element to initialize.
8853 The second argument is a pointer to a ``va_list`` element to copy from.
8858 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
8859 available in C. In a target-dependent way, it copies the source
8860 ``va_list`` element into the destination ``va_list`` element. This
8861 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
8862 arbitrarily complex and require, for example, memory allocation.
8864 Accurate Garbage Collection Intrinsics
8865 --------------------------------------
8867 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
8868 (GC) requires the frontend to generate code containing appropriate intrinsic
8869 calls and select an appropriate GC strategy which knows how to lower these
8870 intrinsics in a manner which is appropriate for the target collector.
8872 These intrinsics allow identification of :ref:`GC roots on the
8873 stack <int_gcroot>`, as well as garbage collector implementations that
8874 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
8875 Frontends for type-safe garbage collected languages should generate
8876 these intrinsics to make use of the LLVM garbage collectors. For more
8877 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
8879 Experimental Statepoint Intrinsics
8880 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8882 LLVM provides an second experimental set of intrinsics for describing garbage
8883 collection safepoints in compiled code. These intrinsics are an alternative
8884 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
8885 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
8886 differences in approach are covered in the `Garbage Collection with LLVM
8887 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
8888 described in :doc:`Statepoints`.
8892 '``llvm.gcroot``' Intrinsic
8893 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8900 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
8905 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
8906 the code generator, and allows some metadata to be associated with it.
8911 The first argument specifies the address of a stack object that contains
8912 the root pointer. The second pointer (which must be either a constant or
8913 a global value address) contains the meta-data to be associated with the
8919 At runtime, a call to this intrinsic stores a null pointer into the
8920 "ptrloc" location. At compile-time, the code generator generates
8921 information to allow the runtime to find the pointer at GC safe points.
8922 The '``llvm.gcroot``' intrinsic may only be used in a function which
8923 :ref:`specifies a GC algorithm <gc>`.
8927 '``llvm.gcread``' Intrinsic
8928 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8935 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
8940 The '``llvm.gcread``' intrinsic identifies reads of references from heap
8941 locations, allowing garbage collector implementations that require read
8947 The second argument is the address to read from, which should be an
8948 address allocated from the garbage collector. The first object is a
8949 pointer to the start of the referenced object, if needed by the language
8950 runtime (otherwise null).
8955 The '``llvm.gcread``' intrinsic has the same semantics as a load
8956 instruction, but may be replaced with substantially more complex code by
8957 the garbage collector runtime, as needed. The '``llvm.gcread``'
8958 intrinsic may only be used in a function which :ref:`specifies a GC
8963 '``llvm.gcwrite``' Intrinsic
8964 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8971 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
8976 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
8977 locations, allowing garbage collector implementations that require write
8978 barriers (such as generational or reference counting collectors).
8983 The first argument is the reference to store, the second is the start of
8984 the object to store it to, and the third is the address of the field of
8985 Obj to store to. If the runtime does not require a pointer to the
8986 object, Obj may be null.
8991 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
8992 instruction, but may be replaced with substantially more complex code by
8993 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
8994 intrinsic may only be used in a function which :ref:`specifies a GC
8997 Code Generator Intrinsics
8998 -------------------------
9000 These intrinsics are provided by LLVM to expose special features that
9001 may only be implemented with code generator support.
9003 '``llvm.returnaddress``' Intrinsic
9004 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9011 declare i8 *@llvm.returnaddress(i32 <level>)
9016 The '``llvm.returnaddress``' intrinsic attempts to compute a
9017 target-specific value indicating the return address of the current
9018 function or one of its callers.
9023 The argument to this intrinsic indicates which function to return the
9024 address for. Zero indicates the calling function, one indicates its
9025 caller, etc. The argument is **required** to be a constant integer
9031 The '``llvm.returnaddress``' intrinsic either returns a pointer
9032 indicating the return address of the specified call frame, or zero if it
9033 cannot be identified. The value returned by this intrinsic is likely to
9034 be incorrect or 0 for arguments other than zero, so it should only be
9035 used for debugging purposes.
9037 Note that calling this intrinsic does not prevent function inlining or
9038 other aggressive transformations, so the value returned may not be that
9039 of the obvious source-language caller.
9041 '``llvm.frameaddress``' Intrinsic
9042 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9049 declare i8* @llvm.frameaddress(i32 <level>)
9054 The '``llvm.frameaddress``' intrinsic attempts to return the
9055 target-specific frame pointer value for the specified stack frame.
9060 The argument to this intrinsic indicates which function to return the
9061 frame pointer for. Zero indicates the calling function, one indicates
9062 its caller, etc. The argument is **required** to be a constant integer
9068 The '``llvm.frameaddress``' intrinsic either returns a pointer
9069 indicating the frame address of the specified call frame, or zero if it
9070 cannot be identified. The value returned by this intrinsic is likely to
9071 be incorrect or 0 for arguments other than zero, so it should only be
9072 used for debugging purposes.
9074 Note that calling this intrinsic does not prevent function inlining or
9075 other aggressive transformations, so the value returned may not be that
9076 of the obvious source-language caller.
9078 '``llvm.localescape``' and '``llvm.localrecover``' Intrinsics
9079 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9086 declare void @llvm.localescape(...)
9087 declare i8* @llvm.localrecover(i8* %func, i8* %fp, i32 %idx)
9092 The '``llvm.localescape``' intrinsic escapes offsets of a collection of static
9093 allocas, and the '``llvm.localrecover``' intrinsic applies those offsets to a
9094 live frame pointer to recover the address of the allocation. The offset is
9095 computed during frame layout of the caller of ``llvm.localescape``.
9100 All arguments to '``llvm.localescape``' must be pointers to static allocas or
9101 casts of static allocas. Each function can only call '``llvm.localescape``'
9102 once, and it can only do so from the entry block.
9104 The ``func`` argument to '``llvm.localrecover``' must be a constant
9105 bitcasted pointer to a function defined in the current module. The code
9106 generator cannot determine the frame allocation offset of functions defined in
9109 The ``fp`` argument to '``llvm.localrecover``' must be a frame pointer of a
9110 call frame that is currently live. The return value of '``llvm.localaddress``'
9111 is one way to produce such a value, but various runtimes also expose a suitable
9112 pointer in platform-specific ways.
9114 The ``idx`` argument to '``llvm.localrecover``' indicates which alloca passed to
9115 '``llvm.localescape``' to recover. It is zero-indexed.
9120 These intrinsics allow a group of functions to share access to a set of local
9121 stack allocations of a one parent function. The parent function may call the
9122 '``llvm.localescape``' intrinsic once from the function entry block, and the
9123 child functions can use '``llvm.localrecover``' to access the escaped allocas.
9124 The '``llvm.localescape``' intrinsic blocks inlining, as inlining changes where
9125 the escaped allocas are allocated, which would break attempts to use
9126 '``llvm.localrecover``'.
9128 .. _int_read_register:
9129 .. _int_write_register:
9131 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
9132 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9139 declare i32 @llvm.read_register.i32(metadata)
9140 declare i64 @llvm.read_register.i64(metadata)
9141 declare void @llvm.write_register.i32(metadata, i32 @value)
9142 declare void @llvm.write_register.i64(metadata, i64 @value)
9148 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
9149 provides access to the named register. The register must be valid on
9150 the architecture being compiled to. The type needs to be compatible
9151 with the register being read.
9156 The '``llvm.read_register``' intrinsic returns the current value of the
9157 register, where possible. The '``llvm.write_register``' intrinsic sets
9158 the current value of the register, where possible.
9160 This is useful to implement named register global variables that need
9161 to always be mapped to a specific register, as is common practice on
9162 bare-metal programs including OS kernels.
9164 The compiler doesn't check for register availability or use of the used
9165 register in surrounding code, including inline assembly. Because of that,
9166 allocatable registers are not supported.
9168 Warning: So far it only works with the stack pointer on selected
9169 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
9170 work is needed to support other registers and even more so, allocatable
9175 '``llvm.stacksave``' Intrinsic
9176 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9183 declare i8* @llvm.stacksave()
9188 The '``llvm.stacksave``' intrinsic is used to remember the current state
9189 of the function stack, for use with
9190 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
9191 implementing language features like scoped automatic variable sized
9197 This intrinsic returns a opaque pointer value that can be passed to
9198 :ref:`llvm.stackrestore <int_stackrestore>`. When an
9199 ``llvm.stackrestore`` intrinsic is executed with a value saved from
9200 ``llvm.stacksave``, it effectively restores the state of the stack to
9201 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
9202 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
9203 were allocated after the ``llvm.stacksave`` was executed.
9205 .. _int_stackrestore:
9207 '``llvm.stackrestore``' Intrinsic
9208 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9215 declare void @llvm.stackrestore(i8* %ptr)
9220 The '``llvm.stackrestore``' intrinsic is used to restore the state of
9221 the function stack to the state it was in when the corresponding
9222 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
9223 useful for implementing language features like scoped automatic variable
9224 sized arrays in C99.
9229 See the description for :ref:`llvm.stacksave <int_stacksave>`.
9231 .. _int_get_dynamic_area_offset:
9233 '``llvm.get.dynamic.area.offset``' Intrinsic
9234 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9241 declare i32 @llvm.get.dynamic.area.offset.i32()
9242 declare i64 @llvm.get.dynamic.area.offset.i64()
9247 The '``llvm.get.dynamic.area.offset.*``' intrinsic family is used to
9248 get the offset from native stack pointer to the address of the most
9249 recent dynamic alloca on the caller's stack. These intrinsics are
9250 intendend for use in combination with
9251 :ref:`llvm.stacksave <int_stacksave>` to get a
9252 pointer to the most recent dynamic alloca. This is useful, for example,
9253 for AddressSanitizer's stack unpoisoning routines.
9258 These intrinsics return a non-negative integer value that can be used to
9259 get the address of the most recent dynamic alloca, allocated by :ref:`alloca <i_alloca>`
9260 on the caller's stack. In particular, for targets where stack grows downwards,
9261 adding this offset to the native stack pointer would get the address of the most
9262 recent dynamic alloca. For targets where stack grows upwards, the situation is a bit more
9263 complicated, because substracting this value from stack pointer would get the address
9264 one past the end of the most recent dynamic alloca.
9266 Although for most targets `llvm.get.dynamic.area.offset <int_get_dynamic_area_offset>`
9267 returns just a zero, for others, such as PowerPC and PowerPC64, it returns a
9268 compile-time-known constant value.
9270 The return value type of :ref:`llvm.get.dynamic.area.offset <int_get_dynamic_area_offset>`
9271 must match the target's generic address space's (address space 0) pointer type.
9273 '``llvm.prefetch``' Intrinsic
9274 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9281 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
9286 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
9287 insert a prefetch instruction if supported; otherwise, it is a noop.
9288 Prefetches have no effect on the behavior of the program but can change
9289 its performance characteristics.
9294 ``address`` is the address to be prefetched, ``rw`` is the specifier
9295 determining if the fetch should be for a read (0) or write (1), and
9296 ``locality`` is a temporal locality specifier ranging from (0) - no
9297 locality, to (3) - extremely local keep in cache. The ``cache type``
9298 specifies whether the prefetch is performed on the data (1) or
9299 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
9300 arguments must be constant integers.
9305 This intrinsic does not modify the behavior of the program. In
9306 particular, prefetches cannot trap and do not produce a value. On
9307 targets that support this intrinsic, the prefetch can provide hints to
9308 the processor cache for better performance.
9310 '``llvm.pcmarker``' Intrinsic
9311 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9318 declare void @llvm.pcmarker(i32 <id>)
9323 The '``llvm.pcmarker``' intrinsic is a method to export a Program
9324 Counter (PC) in a region of code to simulators and other tools. The
9325 method is target specific, but it is expected that the marker will use
9326 exported symbols to transmit the PC of the marker. The marker makes no
9327 guarantees that it will remain with any specific instruction after
9328 optimizations. It is possible that the presence of a marker will inhibit
9329 optimizations. The intended use is to be inserted after optimizations to
9330 allow correlations of simulation runs.
9335 ``id`` is a numerical id identifying the marker.
9340 This intrinsic does not modify the behavior of the program. Backends
9341 that do not support this intrinsic may ignore it.
9343 '``llvm.readcyclecounter``' Intrinsic
9344 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9351 declare i64 @llvm.readcyclecounter()
9356 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
9357 counter register (or similar low latency, high accuracy clocks) on those
9358 targets that support it. On X86, it should map to RDTSC. On Alpha, it
9359 should map to RPCC. As the backing counters overflow quickly (on the
9360 order of 9 seconds on alpha), this should only be used for small
9366 When directly supported, reading the cycle counter should not modify any
9367 memory. Implementations are allowed to either return a application
9368 specific value or a system wide value. On backends without support, this
9369 is lowered to a constant 0.
9371 Note that runtime support may be conditional on the privilege-level code is
9372 running at and the host platform.
9374 '``llvm.clear_cache``' Intrinsic
9375 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9382 declare void @llvm.clear_cache(i8*, i8*)
9387 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
9388 in the specified range to the execution unit of the processor. On
9389 targets with non-unified instruction and data cache, the implementation
9390 flushes the instruction cache.
9395 On platforms with coherent instruction and data caches (e.g. x86), this
9396 intrinsic is a nop. On platforms with non-coherent instruction and data
9397 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
9398 instructions or a system call, if cache flushing requires special
9401 The default behavior is to emit a call to ``__clear_cache`` from the run
9404 This instrinsic does *not* empty the instruction pipeline. Modifications
9405 of the current function are outside the scope of the intrinsic.
9407 '``llvm.instrprof_increment``' Intrinsic
9408 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9415 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
9416 i32 <num-counters>, i32 <index>)
9421 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
9422 frontend for use with instrumentation based profiling. These will be
9423 lowered by the ``-instrprof`` pass to generate execution counts of a
9429 The first argument is a pointer to a global variable containing the
9430 name of the entity being instrumented. This should generally be the
9431 (mangled) function name for a set of counters.
9433 The second argument is a hash value that can be used by the consumer
9434 of the profile data to detect changes to the instrumented source, and
9435 the third is the number of counters associated with ``name``. It is an
9436 error if ``hash`` or ``num-counters`` differ between two instances of
9437 ``instrprof_increment`` that refer to the same name.
9439 The last argument refers to which of the counters for ``name`` should
9440 be incremented. It should be a value between 0 and ``num-counters``.
9445 This intrinsic represents an increment of a profiling counter. It will
9446 cause the ``-instrprof`` pass to generate the appropriate data
9447 structures and the code to increment the appropriate value, in a
9448 format that can be written out by a compiler runtime and consumed via
9449 the ``llvm-profdata`` tool.
9451 '``llvm.instrprof_value_profile``' Intrinsic
9452 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9459 declare void @llvm.instrprof_value_profile(i8* <name>, i64 <hash>,
9460 i64 <value>, i32 <value_kind>,
9466 The '``llvm.instrprof_value_profile``' intrinsic can be emitted by a
9467 frontend for use with instrumentation based profiling. This will be
9468 lowered by the ``-instrprof`` pass to find out the target values,
9469 instrumented expressions take in a program at runtime.
9474 The first argument is a pointer to a global variable containing the
9475 name of the entity being instrumented. ``name`` should generally be the
9476 (mangled) function name for a set of counters.
9478 The second argument is a hash value that can be used by the consumer
9479 of the profile data to detect changes to the instrumented source. It
9480 is an error if ``hash`` differs between two instances of
9481 ``llvm.instrprof_*`` that refer to the same name.
9483 The third argument is the value of the expression being profiled. The profiled
9484 expression's value should be representable as an unsigned 64-bit value. The
9485 fourth argument represents the kind of value profiling that is being done. The
9486 supported value profiling kinds are enumerated through the
9487 ``InstrProfValueKind`` type declared in the
9488 ``<include/llvm/ProfileData/InstrProf.h>`` header file. The last argument is the
9489 index of the instrumented expression within ``name``. It should be >= 0.
9494 This intrinsic represents the point where a call to a runtime routine
9495 should be inserted for value profiling of target expressions. ``-instrprof``
9496 pass will generate the appropriate data structures and replace the
9497 ``llvm.instrprof_value_profile`` intrinsic with the call to the profile
9498 runtime library with proper arguments.
9500 Standard C Library Intrinsics
9501 -----------------------------
9503 LLVM provides intrinsics for a few important standard C library
9504 functions. These intrinsics allow source-language front-ends to pass
9505 information about the alignment of the pointer arguments to the code
9506 generator, providing opportunity for more efficient code generation.
9510 '``llvm.memcpy``' Intrinsic
9511 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9516 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
9517 integer bit width and for different address spaces. Not all targets
9518 support all bit widths however.
9522 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9523 i32 <len>, i32 <align>, i1 <isvolatile>)
9524 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9525 i64 <len>, i32 <align>, i1 <isvolatile>)
9530 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9531 source location to the destination location.
9533 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
9534 intrinsics do not return a value, takes extra alignment/isvolatile
9535 arguments and the pointers can be in specified address spaces.
9540 The first argument is a pointer to the destination, the second is a
9541 pointer to the source. The third argument is an integer argument
9542 specifying the number of bytes to copy, the fourth argument is the
9543 alignment of the source and destination locations, and the fifth is a
9544 boolean indicating a volatile access.
9546 If the call to this intrinsic has an alignment value that is not 0 or 1,
9547 then the caller guarantees that both the source and destination pointers
9548 are aligned to that boundary.
9550 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
9551 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9552 very cleanly specified and it is unwise to depend on it.
9557 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9558 source location to the destination location, which are not allowed to
9559 overlap. It copies "len" bytes of memory over. If the argument is known
9560 to be aligned to some boundary, this can be specified as the fourth
9561 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
9563 '``llvm.memmove``' Intrinsic
9564 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9569 This is an overloaded intrinsic. You can use llvm.memmove on any integer
9570 bit width and for different address space. Not all targets support all
9575 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9576 i32 <len>, i32 <align>, i1 <isvolatile>)
9577 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9578 i64 <len>, i32 <align>, i1 <isvolatile>)
9583 The '``llvm.memmove.*``' intrinsics move a block of memory from the
9584 source location to the destination location. It is similar to the
9585 '``llvm.memcpy``' intrinsic but allows the two memory locations to
9588 Note that, unlike the standard libc function, the ``llvm.memmove.*``
9589 intrinsics do not return a value, takes extra alignment/isvolatile
9590 arguments and the pointers can be in specified address spaces.
9595 The first argument is a pointer to the destination, the second is a
9596 pointer to the source. The third argument is an integer argument
9597 specifying the number of bytes to copy, the fourth argument is the
9598 alignment of the source and destination locations, and the fifth is a
9599 boolean indicating a volatile access.
9601 If the call to this intrinsic has an alignment value that is not 0 or 1,
9602 then the caller guarantees that the source and destination pointers are
9603 aligned to that boundary.
9605 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
9606 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
9607 not very cleanly specified and it is unwise to depend on it.
9612 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
9613 source location to the destination location, which may overlap. It
9614 copies "len" bytes of memory over. If the argument is known to be
9615 aligned to some boundary, this can be specified as the fourth argument,
9616 otherwise it should be set to 0 or 1 (both meaning no alignment).
9618 '``llvm.memset.*``' Intrinsics
9619 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9624 This is an overloaded intrinsic. You can use llvm.memset on any integer
9625 bit width and for different address spaces. However, not all targets
9626 support all bit widths.
9630 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
9631 i32 <len>, i32 <align>, i1 <isvolatile>)
9632 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
9633 i64 <len>, i32 <align>, i1 <isvolatile>)
9638 The '``llvm.memset.*``' intrinsics fill a block of memory with a
9639 particular byte value.
9641 Note that, unlike the standard libc function, the ``llvm.memset``
9642 intrinsic does not return a value and takes extra alignment/volatile
9643 arguments. Also, the destination can be in an arbitrary address space.
9648 The first argument is a pointer to the destination to fill, the second
9649 is the byte value with which to fill it, the third argument is an
9650 integer argument specifying the number of bytes to fill, and the fourth
9651 argument is the known alignment of the destination location.
9653 If the call to this intrinsic has an alignment value that is not 0 or 1,
9654 then the caller guarantees that the destination pointer is aligned to
9657 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
9658 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9659 very cleanly specified and it is unwise to depend on it.
9664 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
9665 at the destination location. If the argument is known to be aligned to
9666 some boundary, this can be specified as the fourth argument, otherwise
9667 it should be set to 0 or 1 (both meaning no alignment).
9669 '``llvm.sqrt.*``' Intrinsic
9670 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9675 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
9676 floating point or vector of floating point type. Not all targets support
9681 declare float @llvm.sqrt.f32(float %Val)
9682 declare double @llvm.sqrt.f64(double %Val)
9683 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
9684 declare fp128 @llvm.sqrt.f128(fp128 %Val)
9685 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
9690 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
9691 returning the same value as the libm '``sqrt``' functions would. Unlike
9692 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
9693 negative numbers other than -0.0 (which allows for better optimization,
9694 because there is no need to worry about errno being set).
9695 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
9700 The argument and return value are floating point numbers of the same
9706 This function returns the sqrt of the specified operand if it is a
9707 nonnegative floating point number.
9709 '``llvm.powi.*``' Intrinsic
9710 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9715 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
9716 floating point or vector of floating point type. Not all targets support
9721 declare float @llvm.powi.f32(float %Val, i32 %power)
9722 declare double @llvm.powi.f64(double %Val, i32 %power)
9723 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
9724 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
9725 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
9730 The '``llvm.powi.*``' intrinsics return the first operand raised to the
9731 specified (positive or negative) power. The order of evaluation of
9732 multiplications is not defined. When a vector of floating point type is
9733 used, the second argument remains a scalar integer value.
9738 The second argument is an integer power, and the first is a value to
9739 raise to that power.
9744 This function returns the first value raised to the second power with an
9745 unspecified sequence of rounding operations.
9747 '``llvm.sin.*``' Intrinsic
9748 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9753 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
9754 floating point or vector of floating point type. Not all targets support
9759 declare float @llvm.sin.f32(float %Val)
9760 declare double @llvm.sin.f64(double %Val)
9761 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
9762 declare fp128 @llvm.sin.f128(fp128 %Val)
9763 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
9768 The '``llvm.sin.*``' intrinsics return the sine of the operand.
9773 The argument and return value are floating point numbers of the same
9779 This function returns the sine of the specified operand, returning the
9780 same values as the libm ``sin`` functions would, and handles error
9781 conditions in the same way.
9783 '``llvm.cos.*``' Intrinsic
9784 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9789 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
9790 floating point or vector of floating point type. Not all targets support
9795 declare float @llvm.cos.f32(float %Val)
9796 declare double @llvm.cos.f64(double %Val)
9797 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
9798 declare fp128 @llvm.cos.f128(fp128 %Val)
9799 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
9804 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
9809 The argument and return value are floating point numbers of the same
9815 This function returns the cosine of the specified operand, returning the
9816 same values as the libm ``cos`` functions would, and handles error
9817 conditions in the same way.
9819 '``llvm.pow.*``' Intrinsic
9820 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9825 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
9826 floating point or vector of floating point type. Not all targets support
9831 declare float @llvm.pow.f32(float %Val, float %Power)
9832 declare double @llvm.pow.f64(double %Val, double %Power)
9833 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
9834 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
9835 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
9840 The '``llvm.pow.*``' intrinsics return the first operand raised to the
9841 specified (positive or negative) power.
9846 The second argument is a floating point power, and the first is a value
9847 to raise to that power.
9852 This function returns the first value raised to the second power,
9853 returning the same values as the libm ``pow`` functions would, and
9854 handles error conditions in the same way.
9856 '``llvm.exp.*``' Intrinsic
9857 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9862 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
9863 floating point or vector of floating point type. Not all targets support
9868 declare float @llvm.exp.f32(float %Val)
9869 declare double @llvm.exp.f64(double %Val)
9870 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
9871 declare fp128 @llvm.exp.f128(fp128 %Val)
9872 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
9877 The '``llvm.exp.*``' intrinsics perform the exp function.
9882 The argument and return value are floating point numbers of the same
9888 This function returns the same values as the libm ``exp`` functions
9889 would, and handles error conditions in the same way.
9891 '``llvm.exp2.*``' Intrinsic
9892 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9897 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
9898 floating point or vector of floating point type. Not all targets support
9903 declare float @llvm.exp2.f32(float %Val)
9904 declare double @llvm.exp2.f64(double %Val)
9905 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
9906 declare fp128 @llvm.exp2.f128(fp128 %Val)
9907 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
9912 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
9917 The argument and return value are floating point numbers of the same
9923 This function returns the same values as the libm ``exp2`` functions
9924 would, and handles error conditions in the same way.
9926 '``llvm.log.*``' Intrinsic
9927 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9932 This is an overloaded intrinsic. You can use ``llvm.log`` on any
9933 floating point or vector of floating point type. Not all targets support
9938 declare float @llvm.log.f32(float %Val)
9939 declare double @llvm.log.f64(double %Val)
9940 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
9941 declare fp128 @llvm.log.f128(fp128 %Val)
9942 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
9947 The '``llvm.log.*``' intrinsics perform the log function.
9952 The argument and return value are floating point numbers of the same
9958 This function returns the same values as the libm ``log`` functions
9959 would, and handles error conditions in the same way.
9961 '``llvm.log10.*``' Intrinsic
9962 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9967 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
9968 floating point or vector of floating point type. Not all targets support
9973 declare float @llvm.log10.f32(float %Val)
9974 declare double @llvm.log10.f64(double %Val)
9975 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
9976 declare fp128 @llvm.log10.f128(fp128 %Val)
9977 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
9982 The '``llvm.log10.*``' intrinsics perform the log10 function.
9987 The argument and return value are floating point numbers of the same
9993 This function returns the same values as the libm ``log10`` functions
9994 would, and handles error conditions in the same way.
9996 '``llvm.log2.*``' Intrinsic
9997 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10002 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
10003 floating point or vector of floating point type. Not all targets support
10008 declare float @llvm.log2.f32(float %Val)
10009 declare double @llvm.log2.f64(double %Val)
10010 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
10011 declare fp128 @llvm.log2.f128(fp128 %Val)
10012 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
10017 The '``llvm.log2.*``' intrinsics perform the log2 function.
10022 The argument and return value are floating point numbers of the same
10028 This function returns the same values as the libm ``log2`` functions
10029 would, and handles error conditions in the same way.
10031 '``llvm.fma.*``' Intrinsic
10032 ^^^^^^^^^^^^^^^^^^^^^^^^^^
10037 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
10038 floating point or vector of floating point type. Not all targets support
10043 declare float @llvm.fma.f32(float %a, float %b, float %c)
10044 declare double @llvm.fma.f64(double %a, double %b, double %c)
10045 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
10046 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
10047 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
10052 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
10058 The argument and return value are floating point numbers of the same
10064 This function returns the same values as the libm ``fma`` functions
10065 would, and does not set errno.
10067 '``llvm.fabs.*``' Intrinsic
10068 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10073 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
10074 floating point or vector of floating point type. Not all targets support
10079 declare float @llvm.fabs.f32(float %Val)
10080 declare double @llvm.fabs.f64(double %Val)
10081 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
10082 declare fp128 @llvm.fabs.f128(fp128 %Val)
10083 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
10088 The '``llvm.fabs.*``' intrinsics return the absolute value of the
10094 The argument and return value are floating point numbers of the same
10100 This function returns the same values as the libm ``fabs`` functions
10101 would, and handles error conditions in the same way.
10103 '``llvm.minnum.*``' Intrinsic
10104 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10109 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
10110 floating point or vector of floating point type. Not all targets support
10115 declare float @llvm.minnum.f32(float %Val0, float %Val1)
10116 declare double @llvm.minnum.f64(double %Val0, double %Val1)
10117 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10118 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
10119 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10124 The '``llvm.minnum.*``' intrinsics return the minimum of the two
10131 The arguments and return value are floating point numbers of the same
10137 Follows the IEEE-754 semantics for minNum, which also match for libm's
10140 If either operand is a NaN, returns the other non-NaN operand. Returns
10141 NaN only if both operands are NaN. If the operands compare equal,
10142 returns a value that compares equal to both operands. This means that
10143 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10145 '``llvm.maxnum.*``' Intrinsic
10146 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10151 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
10152 floating point or vector of floating point type. Not all targets support
10157 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
10158 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
10159 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10160 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
10161 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10166 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
10173 The arguments and return value are floating point numbers of the same
10178 Follows the IEEE-754 semantics for maxNum, which also match for libm's
10181 If either operand is a NaN, returns the other non-NaN operand. Returns
10182 NaN only if both operands are NaN. If the operands compare equal,
10183 returns a value that compares equal to both operands. This means that
10184 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10186 '``llvm.copysign.*``' Intrinsic
10187 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10192 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
10193 floating point or vector of floating point type. Not all targets support
10198 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
10199 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
10200 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
10201 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
10202 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
10207 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
10208 first operand and the sign of the second operand.
10213 The arguments and return value are floating point numbers of the same
10219 This function returns the same values as the libm ``copysign``
10220 functions would, and handles error conditions in the same way.
10222 '``llvm.floor.*``' Intrinsic
10223 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10228 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
10229 floating point or vector of floating point type. Not all targets support
10234 declare float @llvm.floor.f32(float %Val)
10235 declare double @llvm.floor.f64(double %Val)
10236 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
10237 declare fp128 @llvm.floor.f128(fp128 %Val)
10238 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
10243 The '``llvm.floor.*``' intrinsics return the floor of the operand.
10248 The argument and return value are floating point numbers of the same
10254 This function returns the same values as the libm ``floor`` functions
10255 would, and handles error conditions in the same way.
10257 '``llvm.ceil.*``' Intrinsic
10258 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10263 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
10264 floating point or vector of floating point type. Not all targets support
10269 declare float @llvm.ceil.f32(float %Val)
10270 declare double @llvm.ceil.f64(double %Val)
10271 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
10272 declare fp128 @llvm.ceil.f128(fp128 %Val)
10273 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
10278 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
10283 The argument and return value are floating point numbers of the same
10289 This function returns the same values as the libm ``ceil`` functions
10290 would, and handles error conditions in the same way.
10292 '``llvm.trunc.*``' Intrinsic
10293 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10298 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
10299 floating point or vector of floating point type. Not all targets support
10304 declare float @llvm.trunc.f32(float %Val)
10305 declare double @llvm.trunc.f64(double %Val)
10306 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
10307 declare fp128 @llvm.trunc.f128(fp128 %Val)
10308 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
10313 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
10314 nearest integer not larger in magnitude than the operand.
10319 The argument and return value are floating point numbers of the same
10325 This function returns the same values as the libm ``trunc`` functions
10326 would, and handles error conditions in the same way.
10328 '``llvm.rint.*``' Intrinsic
10329 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10334 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
10335 floating point or vector of floating point type. Not all targets support
10340 declare float @llvm.rint.f32(float %Val)
10341 declare double @llvm.rint.f64(double %Val)
10342 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
10343 declare fp128 @llvm.rint.f128(fp128 %Val)
10344 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
10349 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
10350 nearest integer. It may raise an inexact floating-point exception if the
10351 operand isn't an integer.
10356 The argument and return value are floating point numbers of the same
10362 This function returns the same values as the libm ``rint`` functions
10363 would, and handles error conditions in the same way.
10365 '``llvm.nearbyint.*``' Intrinsic
10366 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10371 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
10372 floating point or vector of floating point type. Not all targets support
10377 declare float @llvm.nearbyint.f32(float %Val)
10378 declare double @llvm.nearbyint.f64(double %Val)
10379 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
10380 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
10381 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
10386 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
10392 The argument and return value are floating point numbers of the same
10398 This function returns the same values as the libm ``nearbyint``
10399 functions would, and handles error conditions in the same way.
10401 '``llvm.round.*``' Intrinsic
10402 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10407 This is an overloaded intrinsic. You can use ``llvm.round`` on any
10408 floating point or vector of floating point type. Not all targets support
10413 declare float @llvm.round.f32(float %Val)
10414 declare double @llvm.round.f64(double %Val)
10415 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
10416 declare fp128 @llvm.round.f128(fp128 %Val)
10417 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
10422 The '``llvm.round.*``' intrinsics returns the operand rounded to the
10428 The argument and return value are floating point numbers of the same
10434 This function returns the same values as the libm ``round``
10435 functions would, and handles error conditions in the same way.
10437 Bit Manipulation Intrinsics
10438 ---------------------------
10440 LLVM provides intrinsics for a few important bit manipulation
10441 operations. These allow efficient code generation for some algorithms.
10443 '``llvm.bitreverse.*``' Intrinsics
10444 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10449 This is an overloaded intrinsic function. You can use bitreverse on any
10454 declare i16 @llvm.bitreverse.i16(i16 <id>)
10455 declare i32 @llvm.bitreverse.i32(i32 <id>)
10456 declare i64 @llvm.bitreverse.i64(i64 <id>)
10461 The '``llvm.bitreverse``' family of intrinsics is used to reverse the
10462 bitpattern of an integer value; for example ``0b1234567`` becomes
10468 The ``llvm.bitreverse.iN`` intrinsic returns an i16 value that has bit
10469 ``M`` in the input moved to bit ``N-M`` in the output.
10471 '``llvm.bswap.*``' Intrinsics
10472 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10477 This is an overloaded intrinsic function. You can use bswap on any
10478 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
10482 declare i16 @llvm.bswap.i16(i16 <id>)
10483 declare i32 @llvm.bswap.i32(i32 <id>)
10484 declare i64 @llvm.bswap.i64(i64 <id>)
10489 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
10490 values with an even number of bytes (positive multiple of 16 bits).
10491 These are useful for performing operations on data that is not in the
10492 target's native byte order.
10497 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
10498 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
10499 intrinsic returns an i32 value that has the four bytes of the input i32
10500 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
10501 returned i32 will have its bytes in 3, 2, 1, 0 order. The
10502 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
10503 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
10506 '``llvm.ctpop.*``' Intrinsic
10507 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10512 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
10513 bit width, or on any vector with integer elements. Not all targets
10514 support all bit widths or vector types, however.
10518 declare i8 @llvm.ctpop.i8(i8 <src>)
10519 declare i16 @llvm.ctpop.i16(i16 <src>)
10520 declare i32 @llvm.ctpop.i32(i32 <src>)
10521 declare i64 @llvm.ctpop.i64(i64 <src>)
10522 declare i256 @llvm.ctpop.i256(i256 <src>)
10523 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
10528 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
10534 The only argument is the value to be counted. The argument may be of any
10535 integer type, or a vector with integer elements. The return type must
10536 match the argument type.
10541 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
10542 each element of a vector.
10544 '``llvm.ctlz.*``' Intrinsic
10545 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10550 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
10551 integer bit width, or any vector whose elements are integers. Not all
10552 targets support all bit widths or vector types, however.
10556 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
10557 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
10558 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
10559 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
10560 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
10561 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10566 The '``llvm.ctlz``' family of intrinsic functions counts the number of
10567 leading zeros in a variable.
10572 The first argument is the value to be counted. This argument may be of
10573 any integer type, or a vector with integer element type. The return
10574 type must match the first argument type.
10576 The second argument must be a constant and is a flag to indicate whether
10577 the intrinsic should ensure that a zero as the first argument produces a
10578 defined result. Historically some architectures did not provide a
10579 defined result for zero values as efficiently, and many algorithms are
10580 now predicated on avoiding zero-value inputs.
10585 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
10586 zeros in a variable, or within each element of the vector. If
10587 ``src == 0`` then the result is the size in bits of the type of ``src``
10588 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10589 ``llvm.ctlz(i32 2) = 30``.
10591 '``llvm.cttz.*``' Intrinsic
10592 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10597 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
10598 integer bit width, or any vector of integer elements. Not all targets
10599 support all bit widths or vector types, however.
10603 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
10604 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
10605 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
10606 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
10607 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
10608 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10613 The '``llvm.cttz``' family of intrinsic functions counts the number of
10619 The first argument is the value to be counted. This argument may be of
10620 any integer type, or a vector with integer element type. The return
10621 type must match the first argument type.
10623 The second argument must be a constant and is a flag to indicate whether
10624 the intrinsic should ensure that a zero as the first argument produces a
10625 defined result. Historically some architectures did not provide a
10626 defined result for zero values as efficiently, and many algorithms are
10627 now predicated on avoiding zero-value inputs.
10632 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
10633 zeros in a variable, or within each element of a vector. If ``src == 0``
10634 then the result is the size in bits of the type of ``src`` if
10635 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10636 ``llvm.cttz(2) = 1``.
10640 Arithmetic with Overflow Intrinsics
10641 -----------------------------------
10643 LLVM provides intrinsics for some arithmetic with overflow operations.
10645 '``llvm.sadd.with.overflow.*``' Intrinsics
10646 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10651 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
10652 on any integer bit width.
10656 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
10657 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10658 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
10663 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10664 a signed addition of the two arguments, and indicate whether an overflow
10665 occurred during the signed summation.
10670 The arguments (%a and %b) and the first element of the result structure
10671 may be of integer types of any bit width, but they must have the same
10672 bit width. The second element of the result structure must be of type
10673 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10679 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10680 a signed addition of the two variables. They return a structure --- the
10681 first element of which is the signed summation, and the second element
10682 of which is a bit specifying if the signed summation resulted in an
10688 .. code-block:: llvm
10690 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10691 %sum = extractvalue {i32, i1} %res, 0
10692 %obit = extractvalue {i32, i1} %res, 1
10693 br i1 %obit, label %overflow, label %normal
10695 '``llvm.uadd.with.overflow.*``' Intrinsics
10696 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10701 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
10702 on any integer bit width.
10706 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
10707 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10708 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
10713 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10714 an unsigned addition of the two arguments, and indicate whether a carry
10715 occurred during the unsigned summation.
10720 The arguments (%a and %b) and the first element of the result structure
10721 may be of integer types of any bit width, but they must have the same
10722 bit width. The second element of the result structure must be of type
10723 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10729 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10730 an unsigned addition of the two arguments. They return a structure --- the
10731 first element of which is the sum, and the second element of which is a
10732 bit specifying if the unsigned summation resulted in a carry.
10737 .. code-block:: llvm
10739 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10740 %sum = extractvalue {i32, i1} %res, 0
10741 %obit = extractvalue {i32, i1} %res, 1
10742 br i1 %obit, label %carry, label %normal
10744 '``llvm.ssub.with.overflow.*``' Intrinsics
10745 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10750 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
10751 on any integer bit width.
10755 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
10756 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10757 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
10762 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10763 a signed subtraction of the two arguments, and indicate whether an
10764 overflow occurred during the signed subtraction.
10769 The arguments (%a and %b) and the first element of the result structure
10770 may be of integer types of any bit width, but they must have the same
10771 bit width. The second element of the result structure must be of type
10772 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10778 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10779 a signed subtraction of the two arguments. They return a structure --- the
10780 first element of which is the subtraction, and the second element of
10781 which is a bit specifying if the signed subtraction resulted in an
10787 .. code-block:: llvm
10789 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10790 %sum = extractvalue {i32, i1} %res, 0
10791 %obit = extractvalue {i32, i1} %res, 1
10792 br i1 %obit, label %overflow, label %normal
10794 '``llvm.usub.with.overflow.*``' Intrinsics
10795 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10800 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
10801 on any integer bit width.
10805 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
10806 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10807 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
10812 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10813 an unsigned subtraction of the two arguments, and indicate whether an
10814 overflow occurred during the unsigned subtraction.
10819 The arguments (%a and %b) and the first element of the result structure
10820 may be of integer types of any bit width, but they must have the same
10821 bit width. The second element of the result structure must be of type
10822 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10828 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10829 an unsigned subtraction of the two arguments. They return a structure ---
10830 the first element of which is the subtraction, and the second element of
10831 which is a bit specifying if the unsigned subtraction resulted in an
10837 .. code-block:: llvm
10839 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10840 %sum = extractvalue {i32, i1} %res, 0
10841 %obit = extractvalue {i32, i1} %res, 1
10842 br i1 %obit, label %overflow, label %normal
10844 '``llvm.smul.with.overflow.*``' Intrinsics
10845 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10850 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
10851 on any integer bit width.
10855 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
10856 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10857 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
10862 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10863 a signed multiplication of the two arguments, and indicate whether an
10864 overflow occurred during the signed multiplication.
10869 The arguments (%a and %b) and the first element of the result structure
10870 may be of integer types of any bit width, but they must have the same
10871 bit width. The second element of the result structure must be of type
10872 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10878 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10879 a signed multiplication of the two arguments. They return a structure ---
10880 the first element of which is the multiplication, and the second element
10881 of which is a bit specifying if the signed multiplication resulted in an
10887 .. code-block:: llvm
10889 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10890 %sum = extractvalue {i32, i1} %res, 0
10891 %obit = extractvalue {i32, i1} %res, 1
10892 br i1 %obit, label %overflow, label %normal
10894 '``llvm.umul.with.overflow.*``' Intrinsics
10895 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10900 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
10901 on any integer bit width.
10905 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
10906 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10907 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
10912 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10913 a unsigned multiplication of the two arguments, and indicate whether an
10914 overflow occurred during the unsigned multiplication.
10919 The arguments (%a and %b) and the first element of the result structure
10920 may be of integer types of any bit width, but they must have the same
10921 bit width. The second element of the result structure must be of type
10922 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10928 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10929 an unsigned multiplication of the two arguments. They return a structure ---
10930 the first element of which is the multiplication, and the second
10931 element of which is a bit specifying if the unsigned multiplication
10932 resulted in an overflow.
10937 .. code-block:: llvm
10939 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10940 %sum = extractvalue {i32, i1} %res, 0
10941 %obit = extractvalue {i32, i1} %res, 1
10942 br i1 %obit, label %overflow, label %normal
10944 Specialised Arithmetic Intrinsics
10945 ---------------------------------
10947 '``llvm.canonicalize.*``' Intrinsic
10948 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10955 declare float @llvm.canonicalize.f32(float %a)
10956 declare double @llvm.canonicalize.f64(double %b)
10961 The '``llvm.canonicalize.*``' intrinsic returns the platform specific canonical
10962 encoding of a floating point number. This canonicalization is useful for
10963 implementing certain numeric primitives such as frexp. The canonical encoding is
10964 defined by IEEE-754-2008 to be:
10968 2.1.8 canonical encoding: The preferred encoding of a floating-point
10969 representation in a format. Applied to declets, significands of finite
10970 numbers, infinities, and NaNs, especially in decimal formats.
10972 This operation can also be considered equivalent to the IEEE-754-2008
10973 conversion of a floating-point value to the same format. NaNs are handled
10974 according to section 6.2.
10976 Examples of non-canonical encodings:
10978 - x87 pseudo denormals, pseudo NaNs, pseudo Infinity, Unnormals. These are
10979 converted to a canonical representation per hardware-specific protocol.
10980 - Many normal decimal floating point numbers have non-canonical alternative
10982 - Some machines, like GPUs or ARMv7 NEON, do not support subnormal values.
10983 These are treated as non-canonical encodings of zero and with be flushed to
10984 a zero of the same sign by this operation.
10986 Note that per IEEE-754-2008 6.2, systems that support signaling NaNs with
10987 default exception handling must signal an invalid exception, and produce a
10990 This function should always be implementable as multiplication by 1.0, provided
10991 that the compiler does not constant fold the operation. Likewise, division by
10992 1.0 and ``llvm.minnum(x, x)`` are possible implementations. Addition with
10993 -0.0 is also sufficient provided that the rounding mode is not -Infinity.
10995 ``@llvm.canonicalize`` must preserve the equality relation. That is:
10997 - ``(@llvm.canonicalize(x) == x)`` is equivalent to ``(x == x)``
10998 - ``(@llvm.canonicalize(x) == @llvm.canonicalize(y))`` is equivalent to
11001 Additionally, the sign of zero must be conserved:
11002 ``@llvm.canonicalize(-0.0) = -0.0`` and ``@llvm.canonicalize(+0.0) = +0.0``
11004 The payload bits of a NaN must be conserved, with two exceptions.
11005 First, environments which use only a single canonical representation of NaN
11006 must perform said canonicalization. Second, SNaNs must be quieted per the
11009 The canonicalization operation may be optimized away if:
11011 - The input is known to be canonical. For example, it was produced by a
11012 floating-point operation that is required by the standard to be canonical.
11013 - The result is consumed only by (or fused with) other floating-point
11014 operations. That is, the bits of the floating point value are not examined.
11016 '``llvm.fmuladd.*``' Intrinsic
11017 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11024 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
11025 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
11030 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
11031 expressions that can be fused if the code generator determines that (a) the
11032 target instruction set has support for a fused operation, and (b) that the
11033 fused operation is more efficient than the equivalent, separate pair of mul
11034 and add instructions.
11039 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
11040 multiplicands, a and b, and an addend c.
11049 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
11051 is equivalent to the expression a \* b + c, except that rounding will
11052 not be performed between the multiplication and addition steps if the
11053 code generator fuses the operations. Fusion is not guaranteed, even if
11054 the target platform supports it. If a fused multiply-add is required the
11055 corresponding llvm.fma.\* intrinsic function should be used
11056 instead. This never sets errno, just as '``llvm.fma.*``'.
11061 .. code-block:: llvm
11063 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
11065 Half Precision Floating Point Intrinsics
11066 ----------------------------------------
11068 For most target platforms, half precision floating point is a
11069 storage-only format. This means that it is a dense encoding (in memory)
11070 but does not support computation in the format.
11072 This means that code must first load the half-precision floating point
11073 value as an i16, then convert it to float with
11074 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
11075 then be performed on the float value (including extending to double
11076 etc). To store the value back to memory, it is first converted to float
11077 if needed, then converted to i16 with
11078 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
11081 .. _int_convert_to_fp16:
11083 '``llvm.convert.to.fp16``' Intrinsic
11084 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11091 declare i16 @llvm.convert.to.fp16.f32(float %a)
11092 declare i16 @llvm.convert.to.fp16.f64(double %a)
11097 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11098 conventional floating point type to half precision floating point format.
11103 The intrinsic function contains single argument - the value to be
11109 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11110 conventional floating point format to half precision floating point format. The
11111 return value is an ``i16`` which contains the converted number.
11116 .. code-block:: llvm
11118 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
11119 store i16 %res, i16* @x, align 2
11121 .. _int_convert_from_fp16:
11123 '``llvm.convert.from.fp16``' Intrinsic
11124 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11131 declare float @llvm.convert.from.fp16.f32(i16 %a)
11132 declare double @llvm.convert.from.fp16.f64(i16 %a)
11137 The '``llvm.convert.from.fp16``' intrinsic function performs a
11138 conversion from half precision floating point format to single precision
11139 floating point format.
11144 The intrinsic function contains single argument - the value to be
11150 The '``llvm.convert.from.fp16``' intrinsic function performs a
11151 conversion from half single precision floating point format to single
11152 precision floating point format. The input half-float value is
11153 represented by an ``i16`` value.
11158 .. code-block:: llvm
11160 %a = load i16, i16* @x, align 2
11161 %res = call float @llvm.convert.from.fp16(i16 %a)
11163 .. _dbg_intrinsics:
11165 Debugger Intrinsics
11166 -------------------
11168 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
11169 prefix), are described in the `LLVM Source Level
11170 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
11173 Exception Handling Intrinsics
11174 -----------------------------
11176 The LLVM exception handling intrinsics (which all start with
11177 ``llvm.eh.`` prefix), are described in the `LLVM Exception
11178 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
11180 .. _int_trampoline:
11182 Trampoline Intrinsics
11183 ---------------------
11185 These intrinsics make it possible to excise one parameter, marked with
11186 the :ref:`nest <nest>` attribute, from a function. The result is a
11187 callable function pointer lacking the nest parameter - the caller does
11188 not need to provide a value for it. Instead, the value to use is stored
11189 in advance in a "trampoline", a block of memory usually allocated on the
11190 stack, which also contains code to splice the nest value into the
11191 argument list. This is used to implement the GCC nested function address
11194 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
11195 then the resulting function pointer has signature ``i32 (i32, i32)*``.
11196 It can be created as follows:
11198 .. code-block:: llvm
11200 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
11201 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
11202 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
11203 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
11204 %fp = bitcast i8* %p to i32 (i32, i32)*
11206 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
11207 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
11211 '``llvm.init.trampoline``' Intrinsic
11212 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11219 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
11224 This fills the memory pointed to by ``tramp`` with executable code,
11225 turning it into a trampoline.
11230 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
11231 pointers. The ``tramp`` argument must point to a sufficiently large and
11232 sufficiently aligned block of memory; this memory is written to by the
11233 intrinsic. Note that the size and the alignment are target-specific -
11234 LLVM currently provides no portable way of determining them, so a
11235 front-end that generates this intrinsic needs to have some
11236 target-specific knowledge. The ``func`` argument must hold a function
11237 bitcast to an ``i8*``.
11242 The block of memory pointed to by ``tramp`` is filled with target
11243 dependent code, turning it into a function. Then ``tramp`` needs to be
11244 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
11245 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
11246 function's signature is the same as that of ``func`` with any arguments
11247 marked with the ``nest`` attribute removed. At most one such ``nest``
11248 argument is allowed, and it must be of pointer type. Calling the new
11249 function is equivalent to calling ``func`` with the same argument list,
11250 but with ``nval`` used for the missing ``nest`` argument. If, after
11251 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
11252 modified, then the effect of any later call to the returned function
11253 pointer is undefined.
11257 '``llvm.adjust.trampoline``' Intrinsic
11258 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11265 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
11270 This performs any required machine-specific adjustment to the address of
11271 a trampoline (passed as ``tramp``).
11276 ``tramp`` must point to a block of memory which already has trampoline
11277 code filled in by a previous call to
11278 :ref:`llvm.init.trampoline <int_it>`.
11283 On some architectures the address of the code to be executed needs to be
11284 different than the address where the trampoline is actually stored. This
11285 intrinsic returns the executable address corresponding to ``tramp``
11286 after performing the required machine specific adjustments. The pointer
11287 returned can then be :ref:`bitcast and executed <int_trampoline>`.
11289 .. _int_mload_mstore:
11291 Masked Vector Load and Store Intrinsics
11292 ---------------------------------------
11294 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.
11298 '``llvm.masked.load.*``' Intrinsics
11299 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11303 This is an overloaded intrinsic. The loaded data is a vector of any integer, floating point or pointer data type.
11307 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11308 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11309 ;; The data is a vector of pointers to double
11310 declare <8 x double*> @llvm.masked.load.v8p0f64 (<8 x double*>* <ptr>, i32 <alignment>, <8 x i1> <mask>, <8 x double*> <passthru>)
11311 ;; The data is a vector of function pointers
11312 declare <8 x i32 ()*> @llvm.masked.load.v8p0f_i32f (<8 x i32 ()*>* <ptr>, i32 <alignment>, <8 x i1> <mask>, <8 x i32 ()*> <passthru>)
11317 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.
11323 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.
11329 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.
11330 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.
11335 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
11337 ;; The result of the two following instructions is identical aside from potential memory access exception
11338 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
11339 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
11343 '``llvm.masked.store.*``' Intrinsics
11344 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11348 This is an overloaded intrinsic. The data stored in memory is a vector of any integer, floating point or pointer data type.
11352 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32>* <ptr>, i32 <alignment>, <8 x i1> <mask>)
11353 declare void @llvm.masked.store.v16f32 (<16 x float> <value>, <16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
11354 ;; The data is a vector of pointers to double
11355 declare void @llvm.masked.store.v8p0f64 (<8 x double*> <value>, <8 x double*>* <ptr>, i32 <alignment>, <8 x i1> <mask>)
11356 ;; The data is a vector of function pointers
11357 declare void @llvm.masked.store.v4p0f_i32f (<4 x i32 ()*> <value>, <4 x i32 ()*>* <ptr>, i32 <alignment>, <4 x i1> <mask>)
11362 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.
11367 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.
11373 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.
11374 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.
11378 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
11380 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
11381 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
11382 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
11383 store <16 x float> %res, <16 x float>* %ptr, align 4
11386 Masked Vector Gather and Scatter Intrinsics
11387 -------------------------------------------
11389 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.
11393 '``llvm.masked.gather.*``' Intrinsics
11394 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11398 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.
11402 declare <16 x float> @llvm.masked.gather.v16f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11403 declare <2 x double> @llvm.masked.gather.v2f64 (<2 x double*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11404 declare <8 x float*> @llvm.masked.gather.v8p0f32 (<8 x float**> <ptrs>, i32 <alignment>, <8 x i1> <mask>, <8 x float*> <passthru>)
11409 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.
11415 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.
11421 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.
11422 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.
11427 %res = call <4 x double> @llvm.masked.gather.v4f64 (<4 x double*> %ptrs, i32 8, <4 x i1>%mask, <4 x double> <true, true, true, true>)
11429 ;; The gather with all-true mask is equivalent to the following instruction sequence
11430 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
11431 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
11432 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
11433 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
11435 %val0 = load double, double* %ptr0, align 8
11436 %val1 = load double, double* %ptr1, align 8
11437 %val2 = load double, double* %ptr2, align 8
11438 %val3 = load double, double* %ptr3, align 8
11440 %vec0 = insertelement <4 x double>undef, %val0, 0
11441 %vec01 = insertelement <4 x double>%vec0, %val1, 1
11442 %vec012 = insertelement <4 x double>%vec01, %val2, 2
11443 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
11447 '``llvm.masked.scatter.*``' Intrinsics
11448 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11452 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.
11456 declare void @llvm.masked.scatter.v8i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
11457 declare void @llvm.masked.scatter.v16f32 (<16 x float> <value>, <16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
11458 declare void @llvm.masked.scatter.v4p0f64 (<4 x double*> <value>, <4 x double**> <ptrs>, i32 <alignment>, <4 x i1> <mask>)
11463 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.
11468 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.
11474 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.
11478 ;; This instruction unconditionaly stores data vector in multiple addresses
11479 call @llvm.masked.scatter.v8i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
11481 ;; It is equivalent to a list of scalar stores
11482 %val0 = extractelement <8 x i32> %value, i32 0
11483 %val1 = extractelement <8 x i32> %value, i32 1
11485 %val7 = extractelement <8 x i32> %value, i32 7
11486 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
11487 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
11489 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
11490 ;; Note: the order of the following stores is important when they overlap:
11491 store i32 %val0, i32* %ptr0, align 4
11492 store i32 %val1, i32* %ptr1, align 4
11494 store i32 %val7, i32* %ptr7, align 4
11500 This class of intrinsics provides information about the lifetime of
11501 memory objects and ranges where variables are immutable.
11505 '``llvm.lifetime.start``' Intrinsic
11506 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11513 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
11518 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
11524 The first argument is a constant integer representing the size of the
11525 object, or -1 if it is variable sized. The second argument is a pointer
11531 This intrinsic indicates that before this point in the code, the value
11532 of the memory pointed to by ``ptr`` is dead. This means that it is known
11533 to never be used and has an undefined value. A load from the pointer
11534 that precedes this intrinsic can be replaced with ``'undef'``.
11538 '``llvm.lifetime.end``' Intrinsic
11539 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11546 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
11551 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
11557 The first argument is a constant integer representing the size of the
11558 object, or -1 if it is variable sized. The second argument is a pointer
11564 This intrinsic indicates that after this point in the code, the value of
11565 the memory pointed to by ``ptr`` is dead. This means that it is known to
11566 never be used and has an undefined value. Any stores into the memory
11567 object following this intrinsic may be removed as dead.
11569 '``llvm.invariant.start``' Intrinsic
11570 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11577 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
11582 The '``llvm.invariant.start``' intrinsic specifies that the contents of
11583 a memory object will not change.
11588 The first argument is a constant integer representing the size of the
11589 object, or -1 if it is variable sized. The second argument is a pointer
11595 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
11596 the return value, the referenced memory location is constant and
11599 '``llvm.invariant.end``' Intrinsic
11600 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11607 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
11612 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
11613 memory object are mutable.
11618 The first argument is the matching ``llvm.invariant.start`` intrinsic.
11619 The second argument is a constant integer representing the size of the
11620 object, or -1 if it is variable sized and the third argument is a
11621 pointer to the object.
11626 This intrinsic indicates that the memory is mutable again.
11628 '``llvm.invariant.group.barrier``' Intrinsic
11629 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11636 declare i8* @llvm.invariant.group.barrier(i8* <ptr>)
11641 The '``llvm.invariant.group.barrier``' intrinsic can be used when an invariant
11642 established by invariant.group metadata no longer holds, to obtain a new pointer
11643 value that does not carry the invariant information.
11649 The ``llvm.invariant.group.barrier`` takes only one argument, which is
11650 the pointer to the memory for which the ``invariant.group`` no longer holds.
11655 Returns another pointer that aliases its argument but which is considered different
11656 for the purposes of ``load``/``store`` ``invariant.group`` metadata.
11661 This class of intrinsics is designed to be generic and has no specific
11664 '``llvm.var.annotation``' Intrinsic
11665 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11672 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11677 The '``llvm.var.annotation``' intrinsic.
11682 The first argument is a pointer to a value, the second is a pointer to a
11683 global string, the third is a pointer to a global string which is the
11684 source file name, and the last argument is the line number.
11689 This intrinsic allows annotation of local variables with arbitrary
11690 strings. This can be useful for special purpose optimizations that want
11691 to look for these annotations. These have no other defined use; they are
11692 ignored by code generation and optimization.
11694 '``llvm.ptr.annotation.*``' Intrinsic
11695 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11700 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
11701 pointer to an integer of any width. *NOTE* you must specify an address space for
11702 the pointer. The identifier for the default address space is the integer
11707 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11708 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
11709 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
11710 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
11711 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
11716 The '``llvm.ptr.annotation``' intrinsic.
11721 The first argument is a pointer to an integer value of arbitrary bitwidth
11722 (result of some expression), the second is a pointer to a global string, the
11723 third is a pointer to a global string which is the source file name, and the
11724 last argument is the line number. It returns the value of the first argument.
11729 This intrinsic allows annotation of a pointer to an integer with arbitrary
11730 strings. This can be useful for special purpose optimizations that want to look
11731 for these annotations. These have no other defined use; they are ignored by code
11732 generation and optimization.
11734 '``llvm.annotation.*``' Intrinsic
11735 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11740 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
11741 any integer bit width.
11745 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
11746 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
11747 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
11748 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
11749 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
11754 The '``llvm.annotation``' intrinsic.
11759 The first argument is an integer value (result of some expression), the
11760 second is a pointer to a global string, the third is a pointer to a
11761 global string which is the source file name, and the last argument is
11762 the line number. It returns the value of the first argument.
11767 This intrinsic allows annotations to be put on arbitrary expressions
11768 with arbitrary strings. This can be useful for special purpose
11769 optimizations that want to look for these annotations. These have no
11770 other defined use; they are ignored by code generation and optimization.
11772 '``llvm.trap``' Intrinsic
11773 ^^^^^^^^^^^^^^^^^^^^^^^^^
11780 declare void @llvm.trap() noreturn nounwind
11785 The '``llvm.trap``' intrinsic.
11795 This intrinsic is lowered to the target dependent trap instruction. If
11796 the target does not have a trap instruction, this intrinsic will be
11797 lowered to a call of the ``abort()`` function.
11799 '``llvm.debugtrap``' Intrinsic
11800 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11807 declare void @llvm.debugtrap() nounwind
11812 The '``llvm.debugtrap``' intrinsic.
11822 This intrinsic is lowered to code which is intended to cause an
11823 execution trap with the intention of requesting the attention of a
11826 '``llvm.stackprotector``' Intrinsic
11827 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11834 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
11839 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
11840 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
11841 is placed on the stack before local variables.
11846 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
11847 The first argument is the value loaded from the stack guard
11848 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
11849 enough space to hold the value of the guard.
11854 This intrinsic causes the prologue/epilogue inserter to force the position of
11855 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
11856 to ensure that if a local variable on the stack is overwritten, it will destroy
11857 the value of the guard. When the function exits, the guard on the stack is
11858 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
11859 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
11860 calling the ``__stack_chk_fail()`` function.
11862 '``llvm.stackprotectorcheck``' Intrinsic
11863 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11870 declare void @llvm.stackprotectorcheck(i8** <guard>)
11875 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
11876 created stack protector and if they are not equal calls the
11877 ``__stack_chk_fail()`` function.
11882 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
11883 the variable ``@__stack_chk_guard``.
11888 This intrinsic is provided to perform the stack protector check by comparing
11889 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
11890 values do not match call the ``__stack_chk_fail()`` function.
11892 The reason to provide this as an IR level intrinsic instead of implementing it
11893 via other IR operations is that in order to perform this operation at the IR
11894 level without an intrinsic, one would need to create additional basic blocks to
11895 handle the success/failure cases. This makes it difficult to stop the stack
11896 protector check from disrupting sibling tail calls in Codegen. With this
11897 intrinsic, we are able to generate the stack protector basic blocks late in
11898 codegen after the tail call decision has occurred.
11900 '``llvm.objectsize``' Intrinsic
11901 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11908 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
11909 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
11914 The ``llvm.objectsize`` intrinsic is designed to provide information to
11915 the optimizers to determine at compile time whether a) an operation
11916 (like memcpy) will overflow a buffer that corresponds to an object, or
11917 b) that a runtime check for overflow isn't necessary. An object in this
11918 context means an allocation of a specific class, structure, array, or
11924 The ``llvm.objectsize`` intrinsic takes two arguments. The first
11925 argument is a pointer to or into the ``object``. The second argument is
11926 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
11927 or -1 (if false) when the object size is unknown. The second argument
11928 only accepts constants.
11933 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
11934 the size of the object concerned. If the size cannot be determined at
11935 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
11936 on the ``min`` argument).
11938 '``llvm.expect``' Intrinsic
11939 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
11944 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
11949 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
11950 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
11951 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
11956 The ``llvm.expect`` intrinsic provides information about expected (the
11957 most probable) value of ``val``, which can be used by optimizers.
11962 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
11963 a value. The second argument is an expected value, this needs to be a
11964 constant value, variables are not allowed.
11969 This intrinsic is lowered to the ``val``.
11973 '``llvm.assume``' Intrinsic
11974 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11981 declare void @llvm.assume(i1 %cond)
11986 The ``llvm.assume`` allows the optimizer to assume that the provided
11987 condition is true. This information can then be used in simplifying other parts
11993 The condition which the optimizer may assume is always true.
11998 The intrinsic allows the optimizer to assume that the provided condition is
11999 always true whenever the control flow reaches the intrinsic call. No code is
12000 generated for this intrinsic, and instructions that contribute only to the
12001 provided condition are not used for code generation. If the condition is
12002 violated during execution, the behavior is undefined.
12004 Note that the optimizer might limit the transformations performed on values
12005 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
12006 only used to form the intrinsic's input argument. This might prove undesirable
12007 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
12008 sufficient overall improvement in code quality. For this reason,
12009 ``llvm.assume`` should not be used to document basic mathematical invariants
12010 that the optimizer can otherwise deduce or facts that are of little use to the
12015 '``llvm.bitset.test``' Intrinsic
12016 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12023 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
12029 The first argument is a pointer to be tested. The second argument is a
12030 metadata object representing an identifier for a :doc:`bitset <BitSets>`.
12035 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
12036 member of the given bitset.
12038 '``llvm.donothing``' Intrinsic
12039 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12046 declare void @llvm.donothing() nounwind readnone
12051 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
12052 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
12053 with an invoke instruction.
12063 This intrinsic does nothing, and it's removed by optimizers and ignored
12066 Stack Map Intrinsics
12067 --------------------
12069 LLVM provides experimental intrinsics to support runtime patching
12070 mechanisms commonly desired in dynamic language JITs. These intrinsics
12071 are described in :doc:`StackMaps`.