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.
1582 Funclet Operand Bundles
1583 ^^^^^^^^^^^^^^^^^^^^^^^
1585 Funclet operand bundles are characterized by the ``"funclet"``
1586 operand bundle tag. These operand bundles indicate that a call site
1587 is within a particular funclet. There can be at most one
1588 ``"funclet"`` operand bundle attached to a call site and it must have
1589 exactly one bundle operand.
1593 Module-Level Inline Assembly
1594 ----------------------------
1596 Modules may contain "module-level inline asm" blocks, which corresponds
1597 to the GCC "file scope inline asm" blocks. These blocks are internally
1598 concatenated by LLVM and treated as a single unit, but may be separated
1599 in the ``.ll`` file if desired. The syntax is very simple:
1601 .. code-block:: llvm
1603 module asm "inline asm code goes here"
1604 module asm "more can go here"
1606 The strings can contain any character by escaping non-printable
1607 characters. The escape sequence used is simply "\\xx" where "xx" is the
1608 two digit hex code for the number.
1610 Note that the assembly string *must* be parseable by LLVM's integrated assembler
1611 (unless it is disabled), even when emitting a ``.s`` file.
1613 .. _langref_datalayout:
1618 A module may specify a target specific data layout string that specifies
1619 how data is to be laid out in memory. The syntax for the data layout is
1622 .. code-block:: llvm
1624 target datalayout = "layout specification"
1626 The *layout specification* consists of a list of specifications
1627 separated by the minus sign character ('-'). Each specification starts
1628 with a letter and may include other information after the letter to
1629 define some aspect of the data layout. The specifications accepted are
1633 Specifies that the target lays out data in big-endian form. That is,
1634 the bits with the most significance have the lowest address
1637 Specifies that the target lays out data in little-endian form. That
1638 is, the bits with the least significance have the lowest address
1641 Specifies the natural alignment of the stack in bits. Alignment
1642 promotion of stack variables is limited to the natural stack
1643 alignment to avoid dynamic stack realignment. The stack alignment
1644 must be a multiple of 8-bits. If omitted, the natural stack
1645 alignment defaults to "unspecified", which does not prevent any
1646 alignment promotions.
1647 ``p[n]:<size>:<abi>:<pref>``
1648 This specifies the *size* of a pointer and its ``<abi>`` and
1649 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1650 bits. The address space, ``n``, is optional, and if not specified,
1651 denotes the default address space 0. The value of ``n`` must be
1652 in the range [1,2^23).
1653 ``i<size>:<abi>:<pref>``
1654 This specifies the alignment for an integer type of a given bit
1655 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1656 ``v<size>:<abi>:<pref>``
1657 This specifies the alignment for a vector type of a given bit
1659 ``f<size>:<abi>:<pref>``
1660 This specifies the alignment for a floating point type of a given bit
1661 ``<size>``. Only values of ``<size>`` that are supported by the target
1662 will work. 32 (float) and 64 (double) are supported on all targets; 80
1663 or 128 (different flavors of long double) are also supported on some
1666 This specifies the alignment for an object of aggregate type.
1668 If present, specifies that llvm names are mangled in the output. The
1671 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1672 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1673 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1674 symbols get a ``_`` prefix.
1675 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1676 functions also get a suffix based on the frame size.
1677 * ``x``: Windows x86 COFF prefix: Similar to Windows COFF, but use a ``_``
1678 prefix for ``__cdecl`` functions.
1679 ``n<size1>:<size2>:<size3>...``
1680 This specifies a set of native integer widths for the target CPU in
1681 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1682 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1683 this set are considered to support most general arithmetic operations
1686 On every specification that takes a ``<abi>:<pref>``, specifying the
1687 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1688 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1690 When constructing the data layout for a given target, LLVM starts with a
1691 default set of specifications which are then (possibly) overridden by
1692 the specifications in the ``datalayout`` keyword. The default
1693 specifications are given in this list:
1695 - ``E`` - big endian
1696 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1697 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1698 same as the default address space.
1699 - ``S0`` - natural stack alignment is unspecified
1700 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1701 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1702 - ``i16:16:16`` - i16 is 16-bit aligned
1703 - ``i32:32:32`` - i32 is 32-bit aligned
1704 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1705 alignment of 64-bits
1706 - ``f16:16:16`` - half is 16-bit aligned
1707 - ``f32:32:32`` - float is 32-bit aligned
1708 - ``f64:64:64`` - double is 64-bit aligned
1709 - ``f128:128:128`` - quad is 128-bit aligned
1710 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1711 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1712 - ``a:0:64`` - aggregates are 64-bit aligned
1714 When LLVM is determining the alignment for a given type, it uses the
1717 #. If the type sought is an exact match for one of the specifications,
1718 that specification is used.
1719 #. If no match is found, and the type sought is an integer type, then
1720 the smallest integer type that is larger than the bitwidth of the
1721 sought type is used. If none of the specifications are larger than
1722 the bitwidth then the largest integer type is used. For example,
1723 given the default specifications above, the i7 type will use the
1724 alignment of i8 (next largest) while both i65 and i256 will use the
1725 alignment of i64 (largest specified).
1726 #. If no match is found, and the type sought is a vector type, then the
1727 largest vector type that is smaller than the sought vector type will
1728 be used as a fall back. This happens because <128 x double> can be
1729 implemented in terms of 64 <2 x double>, for example.
1731 The function of the data layout string may not be what you expect.
1732 Notably, this is not a specification from the frontend of what alignment
1733 the code generator should use.
1735 Instead, if specified, the target data layout is required to match what
1736 the ultimate *code generator* expects. This string is used by the
1737 mid-level optimizers to improve code, and this only works if it matches
1738 what the ultimate code generator uses. There is no way to generate IR
1739 that does not embed this target-specific detail into the IR. If you
1740 don't specify the string, the default specifications will be used to
1741 generate a Data Layout and the optimization phases will operate
1742 accordingly and introduce target specificity into the IR with respect to
1743 these default specifications.
1750 A module may specify a target triple string that describes the target
1751 host. The syntax for the target triple is simply:
1753 .. code-block:: llvm
1755 target triple = "x86_64-apple-macosx10.7.0"
1757 The *target triple* string consists of a series of identifiers delimited
1758 by the minus sign character ('-'). The canonical forms are:
1762 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1763 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1765 This information is passed along to the backend so that it generates
1766 code for the proper architecture. It's possible to override this on the
1767 command line with the ``-mtriple`` command line option.
1769 .. _pointeraliasing:
1771 Pointer Aliasing Rules
1772 ----------------------
1774 Any memory access must be done through a pointer value associated with
1775 an address range of the memory access, otherwise the behavior is
1776 undefined. Pointer values are associated with address ranges according
1777 to the following rules:
1779 - A pointer value is associated with the addresses associated with any
1780 value it is *based* on.
1781 - An address of a global variable is associated with the address range
1782 of the variable's storage.
1783 - The result value of an allocation instruction is associated with the
1784 address range of the allocated storage.
1785 - A null pointer in the default address-space is associated with no
1787 - An integer constant other than zero or a pointer value returned from
1788 a function not defined within LLVM may be associated with address
1789 ranges allocated through mechanisms other than those provided by
1790 LLVM. Such ranges shall not overlap with any ranges of addresses
1791 allocated by mechanisms provided by LLVM.
1793 A pointer value is *based* on another pointer value according to the
1796 - A pointer value formed from a ``getelementptr`` operation is *based*
1797 on the first value operand of the ``getelementptr``.
1798 - The result value of a ``bitcast`` is *based* on the operand of the
1800 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1801 values that contribute (directly or indirectly) to the computation of
1802 the pointer's value.
1803 - The "*based* on" relationship is transitive.
1805 Note that this definition of *"based"* is intentionally similar to the
1806 definition of *"based"* in C99, though it is slightly weaker.
1808 LLVM IR does not associate types with memory. The result type of a
1809 ``load`` merely indicates the size and alignment of the memory from
1810 which to load, as well as the interpretation of the value. The first
1811 operand type of a ``store`` similarly only indicates the size and
1812 alignment of the store.
1814 Consequently, type-based alias analysis, aka TBAA, aka
1815 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1816 :ref:`Metadata <metadata>` may be used to encode additional information
1817 which specialized optimization passes may use to implement type-based
1822 Volatile Memory Accesses
1823 ------------------------
1825 Certain memory accesses, such as :ref:`load <i_load>`'s,
1826 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1827 marked ``volatile``. The optimizers must not change the number of
1828 volatile operations or change their order of execution relative to other
1829 volatile operations. The optimizers *may* change the order of volatile
1830 operations relative to non-volatile operations. This is not Java's
1831 "volatile" and has no cross-thread synchronization behavior.
1833 IR-level volatile loads and stores cannot safely be optimized into
1834 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1835 flagged volatile. Likewise, the backend should never split or merge
1836 target-legal volatile load/store instructions.
1838 .. admonition:: Rationale
1840 Platforms may rely on volatile loads and stores of natively supported
1841 data width to be executed as single instruction. For example, in C
1842 this holds for an l-value of volatile primitive type with native
1843 hardware support, but not necessarily for aggregate types. The
1844 frontend upholds these expectations, which are intentionally
1845 unspecified in the IR. The rules above ensure that IR transformations
1846 do not violate the frontend's contract with the language.
1850 Memory Model for Concurrent Operations
1851 --------------------------------------
1853 The LLVM IR does not define any way to start parallel threads of
1854 execution or to register signal handlers. Nonetheless, there are
1855 platform-specific ways to create them, and we define LLVM IR's behavior
1856 in their presence. This model is inspired by the C++0x memory model.
1858 For a more informal introduction to this model, see the :doc:`Atomics`.
1860 We define a *happens-before* partial order as the least partial order
1863 - Is a superset of single-thread program order, and
1864 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1865 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1866 techniques, like pthread locks, thread creation, thread joining,
1867 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1868 Constraints <ordering>`).
1870 Note that program order does not introduce *happens-before* edges
1871 between a thread and signals executing inside that thread.
1873 Every (defined) read operation (load instructions, memcpy, atomic
1874 loads/read-modify-writes, etc.) R reads a series of bytes written by
1875 (defined) write operations (store instructions, atomic
1876 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1877 section, initialized globals are considered to have a write of the
1878 initializer which is atomic and happens before any other read or write
1879 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1880 may see any write to the same byte, except:
1882 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1883 write\ :sub:`2` happens before R\ :sub:`byte`, then
1884 R\ :sub:`byte` does not see write\ :sub:`1`.
1885 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1886 R\ :sub:`byte` does not see write\ :sub:`3`.
1888 Given that definition, R\ :sub:`byte` is defined as follows:
1890 - If R is volatile, the result is target-dependent. (Volatile is
1891 supposed to give guarantees which can support ``sig_atomic_t`` in
1892 C/C++, and may be used for accesses to addresses that do not behave
1893 like normal memory. It does not generally provide cross-thread
1895 - Otherwise, if there is no write to the same byte that happens before
1896 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1897 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1898 R\ :sub:`byte` returns the value written by that write.
1899 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1900 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1901 Memory Ordering Constraints <ordering>` section for additional
1902 constraints on how the choice is made.
1903 - Otherwise R\ :sub:`byte` returns ``undef``.
1905 R returns the value composed of the series of bytes it read. This
1906 implies that some bytes within the value may be ``undef`` **without**
1907 the entire value being ``undef``. Note that this only defines the
1908 semantics of the operation; it doesn't mean that targets will emit more
1909 than one instruction to read the series of bytes.
1911 Note that in cases where none of the atomic intrinsics are used, this
1912 model places only one restriction on IR transformations on top of what
1913 is required for single-threaded execution: introducing a store to a byte
1914 which might not otherwise be stored is not allowed in general.
1915 (Specifically, in the case where another thread might write to and read
1916 from an address, introducing a store can change a load that may see
1917 exactly one write into a load that may see multiple writes.)
1921 Atomic Memory Ordering Constraints
1922 ----------------------------------
1924 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1925 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1926 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1927 ordering parameters that determine which other atomic instructions on
1928 the same address they *synchronize with*. These semantics are borrowed
1929 from Java and C++0x, but are somewhat more colloquial. If these
1930 descriptions aren't precise enough, check those specs (see spec
1931 references in the :doc:`atomics guide <Atomics>`).
1932 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1933 differently since they don't take an address. See that instruction's
1934 documentation for details.
1936 For a simpler introduction to the ordering constraints, see the
1940 The set of values that can be read is governed by the happens-before
1941 partial order. A value cannot be read unless some operation wrote
1942 it. This is intended to provide a guarantee strong enough to model
1943 Java's non-volatile shared variables. This ordering cannot be
1944 specified for read-modify-write operations; it is not strong enough
1945 to make them atomic in any interesting way.
1947 In addition to the guarantees of ``unordered``, there is a single
1948 total order for modifications by ``monotonic`` operations on each
1949 address. All modification orders must be compatible with the
1950 happens-before order. There is no guarantee that the modification
1951 orders can be combined to a global total order for the whole program
1952 (and this often will not be possible). The read in an atomic
1953 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1954 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1955 order immediately before the value it writes. If one atomic read
1956 happens before another atomic read of the same address, the later
1957 read must see the same value or a later value in the address's
1958 modification order. This disallows reordering of ``monotonic`` (or
1959 stronger) operations on the same address. If an address is written
1960 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1961 read that address repeatedly, the other threads must eventually see
1962 the write. This corresponds to the C++0x/C1x
1963 ``memory_order_relaxed``.
1965 In addition to the guarantees of ``monotonic``, a
1966 *synchronizes-with* edge may be formed with a ``release`` operation.
1967 This is intended to model C++'s ``memory_order_acquire``.
1969 In addition to the guarantees of ``monotonic``, if this operation
1970 writes a value which is subsequently read by an ``acquire``
1971 operation, it *synchronizes-with* that operation. (This isn't a
1972 complete description; see the C++0x definition of a release
1973 sequence.) This corresponds to the C++0x/C1x
1974 ``memory_order_release``.
1975 ``acq_rel`` (acquire+release)
1976 Acts as both an ``acquire`` and ``release`` operation on its
1977 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1978 ``seq_cst`` (sequentially consistent)
1979 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1980 operation that only reads, ``release`` for an operation that only
1981 writes), there is a global total order on all
1982 sequentially-consistent operations on all addresses, which is
1983 consistent with the *happens-before* partial order and with the
1984 modification orders of all the affected addresses. Each
1985 sequentially-consistent read sees the last preceding write to the
1986 same address in this global order. This corresponds to the C++0x/C1x
1987 ``memory_order_seq_cst`` and Java volatile.
1991 If an atomic operation is marked ``singlethread``, it only *synchronizes
1992 with* or participates in modification and seq\_cst total orderings with
1993 other operations running in the same thread (for example, in signal
2001 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
2002 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
2003 :ref:`frem <i_frem>`, :ref:`fcmp <i_fcmp>`) have the following flags that can
2004 be set to enable otherwise unsafe floating point operations
2007 No NaNs - Allow optimizations to assume the arguments and result are not
2008 NaN. Such optimizations are required to retain defined behavior over
2009 NaNs, but the value of the result is undefined.
2012 No Infs - Allow optimizations to assume the arguments and result are not
2013 +/-Inf. Such optimizations are required to retain defined behavior over
2014 +/-Inf, but the value of the result is undefined.
2017 No Signed Zeros - Allow optimizations to treat the sign of a zero
2018 argument or result as insignificant.
2021 Allow Reciprocal - Allow optimizations to use the reciprocal of an
2022 argument rather than perform division.
2025 Fast - Allow algebraically equivalent transformations that may
2026 dramatically change results in floating point (e.g. reassociate). This
2027 flag implies all the others.
2031 Use-list Order Directives
2032 -------------------------
2034 Use-list directives encode the in-memory order of each use-list, allowing the
2035 order to be recreated. ``<order-indexes>`` is a comma-separated list of
2036 indexes that are assigned to the referenced value's uses. The referenced
2037 value's use-list is immediately sorted by these indexes.
2039 Use-list directives may appear at function scope or global scope. They are not
2040 instructions, and have no effect on the semantics of the IR. When they're at
2041 function scope, they must appear after the terminator of the final basic block.
2043 If basic blocks have their address taken via ``blockaddress()`` expressions,
2044 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
2051 uselistorder <ty> <value>, { <order-indexes> }
2052 uselistorder_bb @function, %block { <order-indexes> }
2058 define void @foo(i32 %arg1, i32 %arg2) {
2060 ; ... instructions ...
2062 ; ... instructions ...
2064 ; At function scope.
2065 uselistorder i32 %arg1, { 1, 0, 2 }
2066 uselistorder label %bb, { 1, 0 }
2070 uselistorder i32* @global, { 1, 2, 0 }
2071 uselistorder i32 7, { 1, 0 }
2072 uselistorder i32 (i32) @bar, { 1, 0 }
2073 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
2080 The LLVM type system is one of the most important features of the
2081 intermediate representation. Being typed enables a number of
2082 optimizations to be performed on the intermediate representation
2083 directly, without having to do extra analyses on the side before the
2084 transformation. A strong type system makes it easier to read the
2085 generated code and enables novel analyses and transformations that are
2086 not feasible to perform on normal three address code representations.
2096 The void type does not represent any value and has no size.
2114 The function type can be thought of as a function signature. It consists of a
2115 return type and a list of formal parameter types. The return type of a function
2116 type is a void type or first class type --- except for :ref:`label <t_label>`
2117 and :ref:`metadata <t_metadata>` types.
2123 <returntype> (<parameter list>)
2125 ...where '``<parameter list>``' is a comma-separated list of type
2126 specifiers. Optionally, the parameter list may include a type ``...``, which
2127 indicates that the function takes a variable number of arguments. Variable
2128 argument functions can access their arguments with the :ref:`variable argument
2129 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
2130 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
2134 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2135 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
2136 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2137 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
2138 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2139 | ``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. |
2140 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2141 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
2142 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2149 The :ref:`first class <t_firstclass>` types are perhaps the most important.
2150 Values of these types are the only ones which can be produced by
2158 These are the types that are valid in registers from CodeGen's perspective.
2167 The integer type is a very simple type that simply specifies an
2168 arbitrary bit width for the integer type desired. Any bit width from 1
2169 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
2177 The number of bits the integer will occupy is specified by the ``N``
2183 +----------------+------------------------------------------------+
2184 | ``i1`` | a single-bit integer. |
2185 +----------------+------------------------------------------------+
2186 | ``i32`` | a 32-bit integer. |
2187 +----------------+------------------------------------------------+
2188 | ``i1942652`` | a really big integer of over 1 million bits. |
2189 +----------------+------------------------------------------------+
2193 Floating Point Types
2194 """"""""""""""""""""
2203 - 16-bit floating point value
2206 - 32-bit floating point value
2209 - 64-bit floating point value
2212 - 128-bit floating point value (112-bit mantissa)
2215 - 80-bit floating point value (X87)
2218 - 128-bit floating point value (two 64-bits)
2225 The x86_mmx type represents a value held in an MMX register on an x86
2226 machine. The operations allowed on it are quite limited: parameters and
2227 return values, load and store, and bitcast. User-specified MMX
2228 instructions are represented as intrinsic or asm calls with arguments
2229 and/or results of this type. There are no arrays, vectors or constants
2246 The pointer type is used to specify memory locations. Pointers are
2247 commonly used to reference objects in memory.
2249 Pointer types may have an optional address space attribute defining the
2250 numbered address space where the pointed-to object resides. The default
2251 address space is number zero. The semantics of non-zero address spaces
2252 are target-specific.
2254 Note that LLVM does not permit pointers to void (``void*``) nor does it
2255 permit pointers to labels (``label*``). Use ``i8*`` instead.
2265 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2266 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2267 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2268 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2269 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2270 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2271 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2280 A vector type is a simple derived type that represents a vector of
2281 elements. Vector types are used when multiple primitive data are
2282 operated in parallel using a single instruction (SIMD). A vector type
2283 requires a size (number of elements) and an underlying primitive data
2284 type. Vector types are considered :ref:`first class <t_firstclass>`.
2290 < <# elements> x <elementtype> >
2292 The number of elements is a constant integer value larger than 0;
2293 elementtype may be any integer, floating point or pointer type. Vectors
2294 of size zero are not allowed.
2298 +-------------------+--------------------------------------------------+
2299 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2300 +-------------------+--------------------------------------------------+
2301 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2302 +-------------------+--------------------------------------------------+
2303 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2304 +-------------------+--------------------------------------------------+
2305 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2306 +-------------------+--------------------------------------------------+
2315 The label type represents code labels.
2330 The token type is used when a value is associated with an instruction
2331 but all uses of the value must not attempt to introspect or obscure it.
2332 As such, it is not appropriate to have a :ref:`phi <i_phi>` or
2333 :ref:`select <i_select>` of type token.
2350 The metadata type represents embedded metadata. No derived types may be
2351 created from metadata except for :ref:`function <t_function>` arguments.
2364 Aggregate Types are a subset of derived types that can contain multiple
2365 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2366 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2376 The array type is a very simple derived type that arranges elements
2377 sequentially in memory. The array type requires a size (number of
2378 elements) and an underlying data type.
2384 [<# elements> x <elementtype>]
2386 The number of elements is a constant integer value; ``elementtype`` may
2387 be any type with a size.
2391 +------------------+--------------------------------------+
2392 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2393 +------------------+--------------------------------------+
2394 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2395 +------------------+--------------------------------------+
2396 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2397 +------------------+--------------------------------------+
2399 Here are some examples of multidimensional arrays:
2401 +-----------------------------+----------------------------------------------------------+
2402 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2403 +-----------------------------+----------------------------------------------------------+
2404 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2405 +-----------------------------+----------------------------------------------------------+
2406 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2407 +-----------------------------+----------------------------------------------------------+
2409 There is no restriction on indexing beyond the end of the array implied
2410 by a static type (though there are restrictions on indexing beyond the
2411 bounds of an allocated object in some cases). This means that
2412 single-dimension 'variable sized array' addressing can be implemented in
2413 LLVM with a zero length array type. An implementation of 'pascal style
2414 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2424 The structure type is used to represent a collection of data members
2425 together in memory. The elements of a structure may be any type that has
2428 Structures in memory are accessed using '``load``' and '``store``' by
2429 getting a pointer to a field with the '``getelementptr``' instruction.
2430 Structures in registers are accessed using the '``extractvalue``' and
2431 '``insertvalue``' instructions.
2433 Structures may optionally be "packed" structures, which indicate that
2434 the alignment of the struct is one byte, and that there is no padding
2435 between the elements. In non-packed structs, padding between field types
2436 is inserted as defined by the DataLayout string in the module, which is
2437 required to match what the underlying code generator expects.
2439 Structures can either be "literal" or "identified". A literal structure
2440 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2441 identified types are always defined at the top level with a name.
2442 Literal types are uniqued by their contents and can never be recursive
2443 or opaque since there is no way to write one. Identified types can be
2444 recursive, can be opaqued, and are never uniqued.
2450 %T1 = type { <type list> } ; Identified normal struct type
2451 %T2 = type <{ <type list> }> ; Identified packed struct type
2455 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2456 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2457 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2458 | ``{ 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``. |
2459 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2460 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2461 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2465 Opaque Structure Types
2466 """"""""""""""""""""""
2470 Opaque structure types are used to represent named structure types that
2471 do not have a body specified. This corresponds (for example) to the C
2472 notion of a forward declared structure.
2483 +--------------+-------------------+
2484 | ``opaque`` | An opaque type. |
2485 +--------------+-------------------+
2492 LLVM has several different basic types of constants. This section
2493 describes them all and their syntax.
2498 **Boolean constants**
2499 The two strings '``true``' and '``false``' are both valid constants
2501 **Integer constants**
2502 Standard integers (such as '4') are constants of the
2503 :ref:`integer <t_integer>` type. Negative numbers may be used with
2505 **Floating point constants**
2506 Floating point constants use standard decimal notation (e.g.
2507 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2508 hexadecimal notation (see below). The assembler requires the exact
2509 decimal value of a floating-point constant. For example, the
2510 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2511 decimal in binary. Floating point constants must have a :ref:`floating
2512 point <t_floating>` type.
2513 **Null pointer constants**
2514 The identifier '``null``' is recognized as a null pointer constant
2515 and must be of :ref:`pointer type <t_pointer>`.
2517 The identifier '``none``' is recognized as an empty token constant
2518 and must be of :ref:`token type <t_token>`.
2520 The one non-intuitive notation for constants is the hexadecimal form of
2521 floating point constants. For example, the form
2522 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2523 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2524 constants are required (and the only time that they are generated by the
2525 disassembler) is when a floating point constant must be emitted but it
2526 cannot be represented as a decimal floating point number in a reasonable
2527 number of digits. For example, NaN's, infinities, and other special
2528 values are represented in their IEEE hexadecimal format so that assembly
2529 and disassembly do not cause any bits to change in the constants.
2531 When using the hexadecimal form, constants of types half, float, and
2532 double are represented using the 16-digit form shown above (which
2533 matches the IEEE754 representation for double); half and float values
2534 must, however, be exactly representable as IEEE 754 half and single
2535 precision, respectively. Hexadecimal format is always used for long
2536 double, and there are three forms of long double. The 80-bit format used
2537 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2538 128-bit format used by PowerPC (two adjacent doubles) is represented by
2539 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2540 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2541 will only work if they match the long double format on your target.
2542 The IEEE 16-bit format (half precision) is represented by ``0xH``
2543 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2544 (sign bit at the left).
2546 There are no constants of type x86_mmx.
2548 .. _complexconstants:
2553 Complex constants are a (potentially recursive) combination of simple
2554 constants and smaller complex constants.
2556 **Structure constants**
2557 Structure constants are represented with notation similar to
2558 structure type definitions (a comma separated list of elements,
2559 surrounded by braces (``{}``)). For example:
2560 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2561 "``@G = external global i32``". Structure constants must have
2562 :ref:`structure type <t_struct>`, and the number and types of elements
2563 must match those specified by the type.
2565 Array constants are represented with notation similar to array type
2566 definitions (a comma separated list of elements, surrounded by
2567 square brackets (``[]``)). For example:
2568 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2569 :ref:`array type <t_array>`, and the number and types of elements must
2570 match those specified by the type. As a special case, character array
2571 constants may also be represented as a double-quoted string using the ``c``
2572 prefix. For example: "``c"Hello World\0A\00"``".
2573 **Vector constants**
2574 Vector constants are represented with notation similar to vector
2575 type definitions (a comma separated list of elements, surrounded by
2576 less-than/greater-than's (``<>``)). For example:
2577 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2578 must have :ref:`vector type <t_vector>`, and the number and types of
2579 elements must match those specified by the type.
2580 **Zero initialization**
2581 The string '``zeroinitializer``' can be used to zero initialize a
2582 value to zero of *any* type, including scalar and
2583 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2584 having to print large zero initializers (e.g. for large arrays) and
2585 is always exactly equivalent to using explicit zero initializers.
2587 A metadata node is a constant tuple without types. For example:
2588 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2589 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2590 Unlike other typed constants that are meant to be interpreted as part of
2591 the instruction stream, metadata is a place to attach additional
2592 information such as debug info.
2594 Global Variable and Function Addresses
2595 --------------------------------------
2597 The addresses of :ref:`global variables <globalvars>` and
2598 :ref:`functions <functionstructure>` are always implicitly valid
2599 (link-time) constants. These constants are explicitly referenced when
2600 the :ref:`identifier for the global <identifiers>` is used and always have
2601 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2604 .. code-block:: llvm
2608 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2615 The string '``undef``' can be used anywhere a constant is expected, and
2616 indicates that the user of the value may receive an unspecified
2617 bit-pattern. Undefined values may be of any type (other than '``label``'
2618 or '``void``') and be used anywhere a constant is permitted.
2620 Undefined values are useful because they indicate to the compiler that
2621 the program is well defined no matter what value is used. This gives the
2622 compiler more freedom to optimize. Here are some examples of
2623 (potentially surprising) transformations that are valid (in pseudo IR):
2625 .. code-block:: llvm
2635 This is safe because all of the output bits are affected by the undef
2636 bits. Any output bit can have a zero or one depending on the input bits.
2638 .. code-block:: llvm
2649 These logical operations have bits that are not always affected by the
2650 input. For example, if ``%X`` has a zero bit, then the output of the
2651 '``and``' operation will always be a zero for that bit, no matter what
2652 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2653 optimize or assume that the result of the '``and``' is '``undef``'.
2654 However, it is safe to assume that all bits of the '``undef``' could be
2655 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2656 all the bits of the '``undef``' operand to the '``or``' could be set,
2657 allowing the '``or``' to be folded to -1.
2659 .. code-block:: llvm
2661 %A = select undef, %X, %Y
2662 %B = select undef, 42, %Y
2663 %C = select %X, %Y, undef
2673 This set of examples shows that undefined '``select``' (and conditional
2674 branch) conditions can go *either way*, but they have to come from one
2675 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2676 both known to have a clear low bit, then ``%A`` would have to have a
2677 cleared low bit. However, in the ``%C`` example, the optimizer is
2678 allowed to assume that the '``undef``' operand could be the same as
2679 ``%Y``, allowing the whole '``select``' to be eliminated.
2681 .. code-block:: llvm
2683 %A = xor undef, undef
2700 This example points out that two '``undef``' operands are not
2701 necessarily the same. This can be surprising to people (and also matches
2702 C semantics) where they assume that "``X^X``" is always zero, even if
2703 ``X`` is undefined. This isn't true for a number of reasons, but the
2704 short answer is that an '``undef``' "variable" can arbitrarily change
2705 its value over its "live range". This is true because the variable
2706 doesn't actually *have a live range*. Instead, the value is logically
2707 read from arbitrary registers that happen to be around when needed, so
2708 the value is not necessarily consistent over time. In fact, ``%A`` and
2709 ``%C`` need to have the same semantics or the core LLVM "replace all
2710 uses with" concept would not hold.
2712 .. code-block:: llvm
2720 These examples show the crucial difference between an *undefined value*
2721 and *undefined behavior*. An undefined value (like '``undef``') is
2722 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2723 operation can be constant folded to '``undef``', because the '``undef``'
2724 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2725 However, in the second example, we can make a more aggressive
2726 assumption: because the ``undef`` is allowed to be an arbitrary value,
2727 we are allowed to assume that it could be zero. Since a divide by zero
2728 has *undefined behavior*, we are allowed to assume that the operation
2729 does not execute at all. This allows us to delete the divide and all
2730 code after it. Because the undefined operation "can't happen", the
2731 optimizer can assume that it occurs in dead code.
2733 .. code-block:: llvm
2735 a: store undef -> %X
2736 b: store %X -> undef
2741 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2742 value can be assumed to not have any effect; we can assume that the
2743 value is overwritten with bits that happen to match what was already
2744 there. However, a store *to* an undefined location could clobber
2745 arbitrary memory, therefore, it has undefined behavior.
2752 Poison values are similar to :ref:`undef values <undefvalues>`, however
2753 they also represent the fact that an instruction or constant expression
2754 that cannot evoke side effects has nevertheless detected a condition
2755 that results in undefined behavior.
2757 There is currently no way of representing a poison value in the IR; they
2758 only exist when produced by operations such as :ref:`add <i_add>` with
2761 Poison value behavior is defined in terms of value *dependence*:
2763 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2764 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2765 their dynamic predecessor basic block.
2766 - Function arguments depend on the corresponding actual argument values
2767 in the dynamic callers of their functions.
2768 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2769 instructions that dynamically transfer control back to them.
2770 - :ref:`Invoke <i_invoke>` instructions depend on the
2771 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2772 call instructions that dynamically transfer control back to them.
2773 - Non-volatile loads and stores depend on the most recent stores to all
2774 of the referenced memory addresses, following the order in the IR
2775 (including loads and stores implied by intrinsics such as
2776 :ref:`@llvm.memcpy <int_memcpy>`.)
2777 - An instruction with externally visible side effects depends on the
2778 most recent preceding instruction with externally visible side
2779 effects, following the order in the IR. (This includes :ref:`volatile
2780 operations <volatile>`.)
2781 - An instruction *control-depends* on a :ref:`terminator
2782 instruction <terminators>` if the terminator instruction has
2783 multiple successors and the instruction is always executed when
2784 control transfers to one of the successors, and may not be executed
2785 when control is transferred to another.
2786 - Additionally, an instruction also *control-depends* on a terminator
2787 instruction if the set of instructions it otherwise depends on would
2788 be different if the terminator had transferred control to a different
2790 - Dependence is transitive.
2792 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2793 with the additional effect that any instruction that has a *dependence*
2794 on a poison value has undefined behavior.
2796 Here are some examples:
2798 .. code-block:: llvm
2801 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2802 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2803 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
2804 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2806 store i32 %poison, i32* @g ; Poison value stored to memory.
2807 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
2809 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2811 %narrowaddr = bitcast i32* @g to i16*
2812 %wideaddr = bitcast i32* @g to i64*
2813 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2814 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
2816 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2817 br i1 %cmp, label %true, label %end ; Branch to either destination.
2820 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2821 ; it has undefined behavior.
2825 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2826 ; Both edges into this PHI are
2827 ; control-dependent on %cmp, so this
2828 ; always results in a poison value.
2830 store volatile i32 0, i32* @g ; This would depend on the store in %true
2831 ; if %cmp is true, or the store in %entry
2832 ; otherwise, so this is undefined behavior.
2834 br i1 %cmp, label %second_true, label %second_end
2835 ; The same branch again, but this time the
2836 ; true block doesn't have side effects.
2843 store volatile i32 0, i32* @g ; This time, the instruction always depends
2844 ; on the store in %end. Also, it is
2845 ; control-equivalent to %end, so this is
2846 ; well-defined (ignoring earlier undefined
2847 ; behavior in this example).
2851 Addresses of Basic Blocks
2852 -------------------------
2854 ``blockaddress(@function, %block)``
2856 The '``blockaddress``' constant computes the address of the specified
2857 basic block in the specified function, and always has an ``i8*`` type.
2858 Taking the address of the entry block is illegal.
2860 This value only has defined behavior when used as an operand to the
2861 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2862 against null. Pointer equality tests between labels addresses results in
2863 undefined behavior --- though, again, comparison against null is ok, and
2864 no label is equal to the null pointer. This may be passed around as an
2865 opaque pointer sized value as long as the bits are not inspected. This
2866 allows ``ptrtoint`` and arithmetic to be performed on these values so
2867 long as the original value is reconstituted before the ``indirectbr``
2870 Finally, some targets may provide defined semantics when using the value
2871 as the operand to an inline assembly, but that is target specific.
2875 Constant Expressions
2876 --------------------
2878 Constant expressions are used to allow expressions involving other
2879 constants to be used as constants. Constant expressions may be of any
2880 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2881 that does not have side effects (e.g. load and call are not supported).
2882 The following is the syntax for constant expressions:
2884 ``trunc (CST to TYPE)``
2885 Truncate a constant to another type. The bit size of CST must be
2886 larger than the bit size of TYPE. Both types must be integers.
2887 ``zext (CST to TYPE)``
2888 Zero extend a constant to another type. The bit size of CST must be
2889 smaller than the bit size of TYPE. Both types must be integers.
2890 ``sext (CST to TYPE)``
2891 Sign extend a constant to another type. The bit size of CST must be
2892 smaller than the bit size of TYPE. Both types must be integers.
2893 ``fptrunc (CST to TYPE)``
2894 Truncate a floating point constant to another floating point type.
2895 The size of CST must be larger than the size of TYPE. Both types
2896 must be floating point.
2897 ``fpext (CST to TYPE)``
2898 Floating point extend a constant to another type. The size of CST
2899 must be smaller or equal to the size of TYPE. Both types must be
2901 ``fptoui (CST to TYPE)``
2902 Convert a floating point constant to the corresponding unsigned
2903 integer constant. TYPE must be a scalar or vector integer type. CST
2904 must be of scalar or vector floating point type. Both CST and TYPE
2905 must be scalars, or vectors of the same number of elements. If the
2906 value won't fit in the integer type, the results are undefined.
2907 ``fptosi (CST to TYPE)``
2908 Convert a floating point constant to the corresponding signed
2909 integer constant. TYPE must be a scalar or vector integer type. CST
2910 must be of scalar or vector floating point type. Both CST and TYPE
2911 must be scalars, or vectors of the same number of elements. If the
2912 value won't fit in the integer type, the results are undefined.
2913 ``uitofp (CST to TYPE)``
2914 Convert an unsigned integer constant to the corresponding floating
2915 point constant. TYPE must be a scalar or vector floating point type.
2916 CST must be of scalar or vector integer type. Both CST and TYPE must
2917 be scalars, or vectors of the same number of elements. If the value
2918 won't fit in the floating point type, the results are undefined.
2919 ``sitofp (CST to TYPE)``
2920 Convert a signed integer constant to the corresponding floating
2921 point constant. TYPE must be a scalar or vector floating point type.
2922 CST must be of scalar or vector integer type. Both CST and TYPE must
2923 be scalars, or vectors of the same number of elements. If the value
2924 won't fit in the floating point type, the results are undefined.
2925 ``ptrtoint (CST to TYPE)``
2926 Convert a pointer typed constant to the corresponding integer
2927 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2928 pointer type. The ``CST`` value is zero extended, truncated, or
2929 unchanged to make it fit in ``TYPE``.
2930 ``inttoptr (CST to TYPE)``
2931 Convert an integer constant to a pointer constant. TYPE must be a
2932 pointer type. CST must be of integer type. The CST value is zero
2933 extended, truncated, or unchanged to make it fit in a pointer size.
2934 This one is *really* dangerous!
2935 ``bitcast (CST to TYPE)``
2936 Convert a constant, CST, to another TYPE. The constraints of the
2937 operands are the same as those for the :ref:`bitcast
2938 instruction <i_bitcast>`.
2939 ``addrspacecast (CST to TYPE)``
2940 Convert a constant pointer or constant vector of pointer, CST, to another
2941 TYPE in a different address space. The constraints of the operands are the
2942 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2943 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
2944 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2945 constants. As with the :ref:`getelementptr <i_getelementptr>`
2946 instruction, the index list may have zero or more indexes, which are
2947 required to make sense for the type of "pointer to TY".
2948 ``select (COND, VAL1, VAL2)``
2949 Perform the :ref:`select operation <i_select>` on constants.
2950 ``icmp COND (VAL1, VAL2)``
2951 Performs the :ref:`icmp operation <i_icmp>` on constants.
2952 ``fcmp COND (VAL1, VAL2)``
2953 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2954 ``extractelement (VAL, IDX)``
2955 Perform the :ref:`extractelement operation <i_extractelement>` on
2957 ``insertelement (VAL, ELT, IDX)``
2958 Perform the :ref:`insertelement operation <i_insertelement>` on
2960 ``shufflevector (VEC1, VEC2, IDXMASK)``
2961 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2963 ``extractvalue (VAL, IDX0, IDX1, ...)``
2964 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2965 constants. The index list is interpreted in a similar manner as
2966 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2967 least one index value must be specified.
2968 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2969 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2970 The index list is interpreted in a similar manner as indices in a
2971 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2972 value must be specified.
2973 ``OPCODE (LHS, RHS)``
2974 Perform the specified operation of the LHS and RHS constants. OPCODE
2975 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2976 binary <bitwiseops>` operations. The constraints on operands are
2977 the same as those for the corresponding instruction (e.g. no bitwise
2978 operations on floating point values are allowed).
2985 Inline Assembler Expressions
2986 ----------------------------
2988 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2989 Inline Assembly <moduleasm>`) through the use of a special value. This value
2990 represents the inline assembler as a template string (containing the
2991 instructions to emit), a list of operand constraints (stored as a string), a
2992 flag that indicates whether or not the inline asm expression has side effects,
2993 and a flag indicating whether the function containing the asm needs to align its
2994 stack conservatively.
2996 The template string supports argument substitution of the operands using "``$``"
2997 followed by a number, to indicate substitution of the given register/memory
2998 location, as specified by the constraint string. "``${NUM:MODIFIER}``" may also
2999 be used, where ``MODIFIER`` is a target-specific annotation for how to print the
3000 operand (See :ref:`inline-asm-modifiers`).
3002 A literal "``$``" may be included by using "``$$``" in the template. To include
3003 other special characters into the output, the usual "``\XX``" escapes may be
3004 used, just as in other strings. Note that after template substitution, the
3005 resulting assembly string is parsed by LLVM's integrated assembler unless it is
3006 disabled -- even when emitting a ``.s`` file -- and thus must contain assembly
3007 syntax known to LLVM.
3009 LLVM's support for inline asm is modeled closely on the requirements of Clang's
3010 GCC-compatible inline-asm support. Thus, the feature-set and the constraint and
3011 modifier codes listed here are similar or identical to those in GCC's inline asm
3012 support. However, to be clear, the syntax of the template and constraint strings
3013 described here is *not* the same as the syntax accepted by GCC and Clang, and,
3014 while most constraint letters are passed through as-is by Clang, some get
3015 translated to other codes when converting from the C source to the LLVM
3018 An example inline assembler expression is:
3020 .. code-block:: llvm
3022 i32 (i32) asm "bswap $0", "=r,r"
3024 Inline assembler expressions may **only** be used as the callee operand
3025 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
3026 Thus, typically we have:
3028 .. code-block:: llvm
3030 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
3032 Inline asms with side effects not visible in the constraint list must be
3033 marked as having side effects. This is done through the use of the
3034 '``sideeffect``' keyword, like so:
3036 .. code-block:: llvm
3038 call void asm sideeffect "eieio", ""()
3040 In some cases inline asms will contain code that will not work unless
3041 the stack is aligned in some way, such as calls or SSE instructions on
3042 x86, yet will not contain code that does that alignment within the asm.
3043 The compiler should make conservative assumptions about what the asm
3044 might contain and should generate its usual stack alignment code in the
3045 prologue if the '``alignstack``' keyword is present:
3047 .. code-block:: llvm
3049 call void asm alignstack "eieio", ""()
3051 Inline asms also support using non-standard assembly dialects. The
3052 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
3053 the inline asm is using the Intel dialect. Currently, ATT and Intel are
3054 the only supported dialects. An example is:
3056 .. code-block:: llvm
3058 call void asm inteldialect "eieio", ""()
3060 If multiple keywords appear the '``sideeffect``' keyword must come
3061 first, the '``alignstack``' keyword second and the '``inteldialect``'
3064 Inline Asm Constraint String
3065 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3067 The constraint list is a comma-separated string, each element containing one or
3068 more constraint codes.
3070 For each element in the constraint list an appropriate register or memory
3071 operand will be chosen, and it will be made available to assembly template
3072 string expansion as ``$0`` for the first constraint in the list, ``$1`` for the
3075 There are three different types of constraints, which are distinguished by a
3076 prefix symbol in front of the constraint code: Output, Input, and Clobber. The
3077 constraints must always be given in that order: outputs first, then inputs, then
3078 clobbers. They cannot be intermingled.
3080 There are also three different categories of constraint codes:
3082 - Register constraint. This is either a register class, or a fixed physical
3083 register. This kind of constraint will allocate a register, and if necessary,
3084 bitcast the argument or result to the appropriate type.
3085 - Memory constraint. This kind of constraint is for use with an instruction
3086 taking a memory operand. Different constraints allow for different addressing
3087 modes used by the target.
3088 - Immediate value constraint. This kind of constraint is for an integer or other
3089 immediate value which can be rendered directly into an instruction. The
3090 various target-specific constraints allow the selection of a value in the
3091 proper range for the instruction you wish to use it with.
3096 Output constraints are specified by an "``=``" prefix (e.g. "``=r``"). This
3097 indicates that the assembly will write to this operand, and the operand will
3098 then be made available as a return value of the ``asm`` expression. Output
3099 constraints do not consume an argument from the call instruction. (Except, see
3100 below about indirect outputs).
3102 Normally, it is expected that no output locations are written to by the assembly
3103 expression until *all* of the inputs have been read. As such, LLVM may assign
3104 the same register to an output and an input. If this is not safe (e.g. if the
3105 assembly contains two instructions, where the first writes to one output, and
3106 the second reads an input and writes to a second output), then the "``&``"
3107 modifier must be used (e.g. "``=&r``") to specify that the output is an
3108 "early-clobber" output. Marking an ouput as "early-clobber" ensures that LLVM
3109 will not use the same register for any inputs (other than an input tied to this
3115 Input constraints do not have a prefix -- just the constraint codes. Each input
3116 constraint will consume one argument from the call instruction. It is not
3117 permitted for the asm to write to any input register or memory location (unless
3118 that input is tied to an output). Note also that multiple inputs may all be
3119 assigned to the same register, if LLVM can determine that they necessarily all
3120 contain the same value.
3122 Instead of providing a Constraint Code, input constraints may also "tie"
3123 themselves to an output constraint, by providing an integer as the constraint
3124 string. Tied inputs still consume an argument from the call instruction, and
3125 take up a position in the asm template numbering as is usual -- they will simply
3126 be constrained to always use the same register as the output they've been tied
3127 to. For example, a constraint string of "``=r,0``" says to assign a register for
3128 output, and use that register as an input as well (it being the 0'th
3131 It is permitted to tie an input to an "early-clobber" output. In that case, no
3132 *other* input may share the same register as the input tied to the early-clobber
3133 (even when the other input has the same value).
3135 You may only tie an input to an output which has a register constraint, not a
3136 memory constraint. Only a single input may be tied to an output.
3138 There is also an "interesting" feature which deserves a bit of explanation: if a
3139 register class constraint allocates a register which is too small for the value
3140 type operand provided as input, the input value will be split into multiple
3141 registers, and all of them passed to the inline asm.
3143 However, this feature is often not as useful as you might think.
3145 Firstly, the registers are *not* guaranteed to be consecutive. So, on those
3146 architectures that have instructions which operate on multiple consecutive
3147 instructions, this is not an appropriate way to support them. (e.g. the 32-bit
3148 SparcV8 has a 64-bit load, which instruction takes a single 32-bit register. The
3149 hardware then loads into both the named register, and the next register. This
3150 feature of inline asm would not be useful to support that.)
3152 A few of the targets provide a template string modifier allowing explicit access
3153 to the second register of a two-register operand (e.g. MIPS ``L``, ``M``, and
3154 ``D``). On such an architecture, you can actually access the second allocated
3155 register (yet, still, not any subsequent ones). But, in that case, you're still
3156 probably better off simply splitting the value into two separate operands, for
3157 clarity. (e.g. see the description of the ``A`` constraint on X86, which,
3158 despite existing only for use with this feature, is not really a good idea to
3161 Indirect inputs and outputs
3162 """""""""""""""""""""""""""
3164 Indirect output or input constraints can be specified by the "``*``" modifier
3165 (which goes after the "``=``" in case of an output). This indicates that the asm
3166 will write to or read from the contents of an *address* provided as an input
3167 argument. (Note that in this way, indirect outputs act more like an *input* than
3168 an output: just like an input, they consume an argument of the call expression,
3169 rather than producing a return value. An indirect output constraint is an
3170 "output" only in that the asm is expected to write to the contents of the input
3171 memory location, instead of just read from it).
3173 This is most typically used for memory constraint, e.g. "``=*m``", to pass the
3174 address of a variable as a value.
3176 It is also possible to use an indirect *register* constraint, but only on output
3177 (e.g. "``=*r``"). This will cause LLVM to allocate a register for an output
3178 value normally, and then, separately emit a store to the address provided as
3179 input, after the provided inline asm. (It's not clear what value this
3180 functionality provides, compared to writing the store explicitly after the asm
3181 statement, and it can only produce worse code, since it bypasses many
3182 optimization passes. I would recommend not using it.)
3188 A clobber constraint is indicated by a "``~``" prefix. A clobber does not
3189 consume an input operand, nor generate an output. Clobbers cannot use any of the
3190 general constraint code letters -- they may use only explicit register
3191 constraints, e.g. "``~{eax}``". The one exception is that a clobber string of
3192 "``~{memory}``" indicates that the assembly writes to arbitrary undeclared
3193 memory locations -- not only the memory pointed to by a declared indirect
3199 After a potential prefix comes constraint code, or codes.
3201 A Constraint Code is either a single letter (e.g. "``r``"), a "``^``" character
3202 followed by two letters (e.g. "``^wc``"), or "``{``" register-name "``}``"
3205 The one and two letter constraint codes are typically chosen to be the same as
3206 GCC's constraint codes.
3208 A single constraint may include one or more than constraint code in it, leaving
3209 it up to LLVM to choose which one to use. This is included mainly for
3210 compatibility with the translation of GCC inline asm coming from clang.
3212 There are two ways to specify alternatives, and either or both may be used in an
3213 inline asm constraint list:
3215 1) Append the codes to each other, making a constraint code set. E.g. "``im``"
3216 or "``{eax}m``". This means "choose any of the options in the set". The
3217 choice of constraint is made independently for each constraint in the
3220 2) Use "``|``" between constraint code sets, creating alternatives. Every
3221 constraint in the constraint list must have the same number of alternative
3222 sets. With this syntax, the same alternative in *all* of the items in the
3223 constraint list will be chosen together.
3225 Putting those together, you might have a two operand constraint string like
3226 ``"rm|r,ri|rm"``. This indicates that if operand 0 is ``r`` or ``m``, then
3227 operand 1 may be one of ``r`` or ``i``. If operand 0 is ``r``, then operand 1
3228 may be one of ``r`` or ``m``. But, operand 0 and 1 cannot both be of type m.
3230 However, the use of either of the alternatives features is *NOT* recommended, as
3231 LLVM is not able to make an intelligent choice about which one to use. (At the
3232 point it currently needs to choose, not enough information is available to do so
3233 in a smart way.) Thus, it simply tries to make a choice that's most likely to
3234 compile, not one that will be optimal performance. (e.g., given "``rm``", it'll
3235 always choose to use memory, not registers). And, if given multiple registers,
3236 or multiple register classes, it will simply choose the first one. (In fact, it
3237 doesn't currently even ensure explicitly specified physical registers are
3238 unique, so specifying multiple physical registers as alternatives, like
3239 ``{r11}{r12},{r11}{r12}``, will assign r11 to both operands, not at all what was
3242 Supported Constraint Code List
3243 """"""""""""""""""""""""""""""
3245 The constraint codes are, in general, expected to behave the same way they do in
3246 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3247 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3248 and GCC likely indicates a bug in LLVM.
3250 Some constraint codes are typically supported by all targets:
3252 - ``r``: A register in the target's general purpose register class.
3253 - ``m``: A memory address operand. It is target-specific what addressing modes
3254 are supported, typical examples are register, or register + register offset,
3255 or register + immediate offset (of some target-specific size).
3256 - ``i``: An integer constant (of target-specific width). Allows either a simple
3257 immediate, or a relocatable value.
3258 - ``n``: An integer constant -- *not* including relocatable values.
3259 - ``s``: An integer constant, but allowing *only* relocatable values.
3260 - ``X``: Allows an operand of any kind, no constraint whatsoever. Typically
3261 useful to pass a label for an asm branch or call.
3263 .. FIXME: but that surely isn't actually okay to jump out of an asm
3264 block without telling llvm about the control transfer???)
3266 - ``{register-name}``: Requires exactly the named physical register.
3268 Other constraints are target-specific:
3272 - ``z``: An immediate integer 0. Outputs ``WZR`` or ``XZR``, as appropriate.
3273 - ``I``: An immediate integer valid for an ``ADD`` or ``SUB`` instruction,
3274 i.e. 0 to 4095 with optional shift by 12.
3275 - ``J``: An immediate integer that, when negated, is valid for an ``ADD`` or
3276 ``SUB`` instruction, i.e. -1 to -4095 with optional left shift by 12.
3277 - ``K``: An immediate integer that is valid for the 'bitmask immediate 32' of a
3278 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 32-bit register.
3279 - ``L``: An immediate integer that is valid for the 'bitmask immediate 64' of a
3280 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 64-bit register.
3281 - ``M``: An immediate integer for use with the ``MOV`` assembly alias on a
3282 32-bit register. This is a superset of ``K``: in addition to the bitmask
3283 immediate, also allows immediate integers which can be loaded with a single
3284 ``MOVZ`` or ``MOVL`` instruction.
3285 - ``N``: An immediate integer for use with the ``MOV`` assembly alias on a
3286 64-bit register. This is a superset of ``L``.
3287 - ``Q``: Memory address operand must be in a single register (no
3288 offsets). (However, LLVM currently does this for the ``m`` constraint as
3290 - ``r``: A 32 or 64-bit integer register (W* or X*).
3291 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register.
3292 - ``x``: A lower 128-bit floating-point/SIMD register (``V0`` to ``V15``).
3296 - ``r``: A 32 or 64-bit integer register.
3297 - ``[0-9]v``: The 32-bit VGPR register, number 0-9.
3298 - ``[0-9]s``: The 32-bit SGPR register, number 0-9.
3303 - ``Q``, ``Um``, ``Un``, ``Uq``, ``Us``, ``Ut``, ``Uv``, ``Uy``: Memory address
3304 operand. Treated the same as operand ``m``, at the moment.
3306 ARM and ARM's Thumb2 mode:
3308 - ``j``: An immediate integer between 0 and 65535 (valid for ``MOVW``)
3309 - ``I``: An immediate integer valid for a data-processing instruction.
3310 - ``J``: An immediate integer between -4095 and 4095.
3311 - ``K``: An immediate integer whose bitwise inverse is valid for a
3312 data-processing instruction. (Can be used with template modifier "``B``" to
3313 print the inverted value).
3314 - ``L``: An immediate integer whose negation is valid for a data-processing
3315 instruction. (Can be used with template modifier "``n``" to print the negated
3317 - ``M``: A power of two or a integer between 0 and 32.
3318 - ``N``: Invalid immediate constraint.
3319 - ``O``: Invalid immediate constraint.
3320 - ``r``: A general-purpose 32-bit integer register (``r0-r15``).
3321 - ``l``: In Thumb2 mode, low 32-bit GPR registers (``r0-r7``). In ARM mode, same
3323 - ``h``: In Thumb2 mode, a high 32-bit GPR register (``r8-r15``). In ARM mode,
3325 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3326 ``d0-d31``, or ``q0-q15``.
3327 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3328 ``d0-d7``, or ``q0-q3``.
3329 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3334 - ``I``: An immediate integer between 0 and 255.
3335 - ``J``: An immediate integer between -255 and -1.
3336 - ``K``: An immediate integer between 0 and 255, with optional left-shift by
3338 - ``L``: An immediate integer between -7 and 7.
3339 - ``M``: An immediate integer which is a multiple of 4 between 0 and 1020.
3340 - ``N``: An immediate integer between 0 and 31.
3341 - ``O``: An immediate integer which is a multiple of 4 between -508 and 508.
3342 - ``r``: A low 32-bit GPR register (``r0-r7``).
3343 - ``l``: A low 32-bit GPR register (``r0-r7``).
3344 - ``h``: A high GPR register (``r0-r7``).
3345 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3346 ``d0-d31``, or ``q0-q15``.
3347 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3348 ``d0-d7``, or ``q0-q3``.
3349 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3355 - ``o``, ``v``: A memory address operand, treated the same as constraint ``m``,
3357 - ``r``: A 32 or 64-bit register.
3361 - ``r``: An 8 or 16-bit register.
3365 - ``I``: An immediate signed 16-bit integer.
3366 - ``J``: An immediate integer zero.
3367 - ``K``: An immediate unsigned 16-bit integer.
3368 - ``L``: An immediate 32-bit integer, where the lower 16 bits are 0.
3369 - ``N``: An immediate integer between -65535 and -1.
3370 - ``O``: An immediate signed 15-bit integer.
3371 - ``P``: An immediate integer between 1 and 65535.
3372 - ``m``: A memory address operand. In MIPS-SE mode, allows a base address
3373 register plus 16-bit immediate offset. In MIPS mode, just a base register.
3374 - ``R``: A memory address operand. In MIPS-SE mode, allows a base address
3375 register plus a 9-bit signed offset. In MIPS mode, the same as constraint
3377 - ``ZC``: A memory address operand, suitable for use in a ``pref``, ``ll``, or
3378 ``sc`` instruction on the given subtarget (details vary).
3379 - ``r``, ``d``, ``y``: A 32 or 64-bit GPR register.
3380 - ``f``: A 32 or 64-bit FPU register (``F0-F31``), or a 128-bit MSA register
3381 (``W0-W31``). In the case of MSA registers, it is recommended to use the ``w``
3382 argument modifier for compatibility with GCC.
3383 - ``c``: A 32-bit or 64-bit GPR register suitable for indirect jump (always
3385 - ``l``: The ``lo`` register, 32 or 64-bit.
3390 - ``b``: A 1-bit integer register.
3391 - ``c`` or ``h``: A 16-bit integer register.
3392 - ``r``: A 32-bit integer register.
3393 - ``l`` or ``N``: A 64-bit integer register.
3394 - ``f``: A 32-bit float register.
3395 - ``d``: A 64-bit float register.
3400 - ``I``: An immediate signed 16-bit integer.
3401 - ``J``: An immediate unsigned 16-bit integer, shifted left 16 bits.
3402 - ``K``: An immediate unsigned 16-bit integer.
3403 - ``L``: An immediate signed 16-bit integer, shifted left 16 bits.
3404 - ``M``: An immediate integer greater than 31.
3405 - ``N``: An immediate integer that is an exact power of 2.
3406 - ``O``: The immediate integer constant 0.
3407 - ``P``: An immediate integer constant whose negation is a signed 16-bit
3409 - ``es``, ``o``, ``Q``, ``Z``, ``Zy``: A memory address operand, currently
3410 treated the same as ``m``.
3411 - ``r``: A 32 or 64-bit integer register.
3412 - ``b``: A 32 or 64-bit integer register, excluding ``R0`` (that is:
3414 - ``f``: A 32 or 64-bit float register (``F0-F31``), or when QPX is enabled, a
3415 128 or 256-bit QPX register (``Q0-Q31``; aliases the ``F`` registers).
3416 - ``v``: For ``4 x f32`` or ``4 x f64`` types, when QPX is enabled, a
3417 128 or 256-bit QPX register (``Q0-Q31``), otherwise a 128-bit
3418 altivec vector register (``V0-V31``).
3420 .. FIXME: is this a bug that v accepts QPX registers? I think this
3421 is supposed to only use the altivec vector registers?
3423 - ``y``: Condition register (``CR0-CR7``).
3424 - ``wc``: An individual CR bit in a CR register.
3425 - ``wa``, ``wd``, ``wf``: Any 128-bit VSX vector register, from the full VSX
3426 register set (overlapping both the floating-point and vector register files).
3427 - ``ws``: A 32 or 64-bit floating point register, from the full VSX register
3432 - ``I``: An immediate 13-bit signed integer.
3433 - ``r``: A 32-bit integer register.
3437 - ``I``: An immediate unsigned 8-bit integer.
3438 - ``J``: An immediate unsigned 12-bit integer.
3439 - ``K``: An immediate signed 16-bit integer.
3440 - ``L``: An immediate signed 20-bit integer.
3441 - ``M``: An immediate integer 0x7fffffff.
3442 - ``Q``, ``R``, ``S``, ``T``: A memory address operand, treated the same as
3443 ``m``, at the moment.
3444 - ``r`` or ``d``: A 32, 64, or 128-bit integer register.
3445 - ``a``: A 32, 64, or 128-bit integer address register (excludes R0, which in an
3446 address context evaluates as zero).
3447 - ``h``: A 32-bit value in the high part of a 64bit data register
3449 - ``f``: A 32, 64, or 128-bit floating point register.
3453 - ``I``: An immediate integer between 0 and 31.
3454 - ``J``: An immediate integer between 0 and 64.
3455 - ``K``: An immediate signed 8-bit integer.
3456 - ``L``: An immediate integer, 0xff or 0xffff or (in 64-bit mode only)
3458 - ``M``: An immediate integer between 0 and 3.
3459 - ``N``: An immediate unsigned 8-bit integer.
3460 - ``O``: An immediate integer between 0 and 127.
3461 - ``e``: An immediate 32-bit signed integer.
3462 - ``Z``: An immediate 32-bit unsigned integer.
3463 - ``o``, ``v``: Treated the same as ``m``, at the moment.
3464 - ``q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3465 ``l`` integer register. On X86-32, this is the ``a``, ``b``, ``c``, and ``d``
3466 registers, and on X86-64, it is all of the integer registers.
3467 - ``Q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3468 ``h`` integer register. This is the ``a``, ``b``, ``c``, and ``d`` registers.
3469 - ``r`` or ``l``: An 8, 16, 32, or 64-bit integer register.
3470 - ``R``: An 8, 16, 32, or 64-bit "legacy" integer register -- one which has
3471 existed since i386, and can be accessed without the REX prefix.
3472 - ``f``: A 32, 64, or 80-bit '387 FPU stack pseudo-register.
3473 - ``y``: A 64-bit MMX register, if MMX is enabled.
3474 - ``x``: If SSE is enabled: a 32 or 64-bit scalar operand, or 128-bit vector
3475 operand in a SSE register. If AVX is also enabled, can also be a 256-bit
3476 vector operand in an AVX register. If AVX-512 is also enabled, can also be a
3477 512-bit vector operand in an AVX512 register, Otherwise, an error.
3478 - ``Y``: The same as ``x``, if *SSE2* is enabled, otherwise an error.
3479 - ``A``: Special case: allocates EAX first, then EDX, for a single operand (in
3480 32-bit mode, a 64-bit integer operand will get split into two registers). It
3481 is not recommended to use this constraint, as in 64-bit mode, the 64-bit
3482 operand will get allocated only to RAX -- if two 32-bit operands are needed,
3483 you're better off splitting it yourself, before passing it to the asm
3488 - ``r``: A 32-bit integer register.
3491 .. _inline-asm-modifiers:
3493 Asm template argument modifiers
3494 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3496 In the asm template string, modifiers can be used on the operand reference, like
3499 The modifiers are, in general, expected to behave the same way they do in
3500 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3501 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3502 and GCC likely indicates a bug in LLVM.
3506 - ``c``: Print an immediate integer constant unadorned, without
3507 the target-specific immediate punctuation (e.g. no ``$`` prefix).
3508 - ``n``: Negate and print immediate integer constant unadorned, without the
3509 target-specific immediate punctuation (e.g. no ``$`` prefix).
3510 - ``l``: Print as an unadorned label, without the target-specific label
3511 punctuation (e.g. no ``$`` prefix).
3515 - ``w``: Print a GPR register with a ``w*`` name instead of ``x*`` name. E.g.,
3516 instead of ``x30``, print ``w30``.
3517 - ``x``: Print a GPR register with a ``x*`` name. (this is the default, anyhow).
3518 - ``b``, ``h``, ``s``, ``d``, ``q``: Print a floating-point/SIMD register with a
3519 ``b*``, ``h*``, ``s*``, ``d*``, or ``q*`` name, rather than the default of
3528 - ``a``: Print an operand as an address (with ``[`` and ``]`` surrounding a
3532 - ``y``: Print a VFP single-precision register as an indexed double (e.g. print
3533 as ``d4[1]`` instead of ``s9``)
3534 - ``B``: Bitwise invert and print an immediate integer constant without ``#``
3536 - ``L``: Print the low 16-bits of an immediate integer constant.
3537 - ``M``: Print as a register set suitable for ldm/stm. Also prints *all*
3538 register operands subsequent to the specified one (!), so use carefully.
3539 - ``Q``: Print the low-order register of a register-pair, or the low-order
3540 register of a two-register operand.
3541 - ``R``: Print the high-order register of a register-pair, or the high-order
3542 register of a two-register operand.
3543 - ``H``: Print the second register of a register-pair. (On a big-endian system,
3544 ``H`` is equivalent to ``Q``, and on little-endian system, ``H`` is equivalent
3547 .. FIXME: H doesn't currently support printing the second register
3548 of a two-register operand.
3550 - ``e``: Print the low doubleword register of a NEON quad register.
3551 - ``f``: Print the high doubleword register of a NEON quad register.
3552 - ``m``: Print the base register of a memory operand without the ``[`` and ``]``
3557 - ``L``: Print the second register of a two-register operand. Requires that it
3558 has been allocated consecutively to the first.
3560 .. FIXME: why is it restricted to consecutive ones? And there's
3561 nothing that ensures that happens, is there?
3563 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3564 nothing. Used to print 'addi' vs 'add' instructions.
3568 No additional modifiers.
3572 - ``X``: Print an immediate integer as hexadecimal
3573 - ``x``: Print the low 16 bits of an immediate integer as hexadecimal.
3574 - ``d``: Print an immediate integer as decimal.
3575 - ``m``: Subtract one and print an immediate integer as decimal.
3576 - ``z``: Print $0 if an immediate zero, otherwise print normally.
3577 - ``L``: Print the low-order register of a two-register operand, or prints the
3578 address of the low-order word of a double-word memory operand.
3580 .. FIXME: L seems to be missing memory operand support.
3582 - ``M``: Print the high-order register of a two-register operand, or prints the
3583 address of the high-order word of a double-word memory operand.
3585 .. FIXME: M seems to be missing memory operand support.
3587 - ``D``: Print the second register of a two-register operand, or prints the
3588 second word of a double-word memory operand. (On a big-endian system, ``D`` is
3589 equivalent to ``L``, and on little-endian system, ``D`` is equivalent to
3591 - ``w``: No effect. Provided for compatibility with GCC which requires this
3592 modifier in order to print MSA registers (``W0-W31``) with the ``f``
3601 - ``L``: Print the second register of a two-register operand. Requires that it
3602 has been allocated consecutively to the first.
3604 .. FIXME: why is it restricted to consecutive ones? And there's
3605 nothing that ensures that happens, is there?
3607 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3608 nothing. Used to print 'addi' vs 'add' instructions.
3609 - ``y``: For a memory operand, prints formatter for a two-register X-form
3610 instruction. (Currently always prints ``r0,OPERAND``).
3611 - ``U``: Prints 'u' if the memory operand is an update form, and nothing
3612 otherwise. (NOTE: LLVM does not support update form, so this will currently
3613 always print nothing)
3614 - ``X``: Prints 'x' if the memory operand is an indexed form. (NOTE: LLVM does
3615 not support indexed form, so this will currently always print nothing)
3623 SystemZ implements only ``n``, and does *not* support any of the other
3624 target-independent modifiers.
3628 - ``c``: Print an unadorned integer or symbol name. (The latter is
3629 target-specific behavior for this typically target-independent modifier).
3630 - ``A``: Print a register name with a '``*``' before it.
3631 - ``b``: Print an 8-bit register name (e.g. ``al``); do nothing on a memory
3633 - ``h``: Print the upper 8-bit register name (e.g. ``ah``); do nothing on a
3635 - ``w``: Print the 16-bit register name (e.g. ``ax``); do nothing on a memory
3637 - ``k``: Print the 32-bit register name (e.g. ``eax``); do nothing on a memory
3639 - ``q``: Print the 64-bit register name (e.g. ``rax``), if 64-bit registers are
3640 available, otherwise the 32-bit register name; do nothing on a memory operand.
3641 - ``n``: Negate and print an unadorned integer, or, for operands other than an
3642 immediate integer (e.g. a relocatable symbol expression), print a '-' before
3643 the operand. (The behavior for relocatable symbol expressions is a
3644 target-specific behavior for this typically target-independent modifier)
3645 - ``H``: Print a memory reference with additional offset +8.
3646 - ``P``: Print a memory reference or operand for use as the argument of a call
3647 instruction. (E.g. omit ``(rip)``, even though it's PC-relative.)
3651 No additional modifiers.
3657 The call instructions that wrap inline asm nodes may have a
3658 "``!srcloc``" MDNode attached to it that contains a list of constant
3659 integers. If present, the code generator will use the integer as the
3660 location cookie value when report errors through the ``LLVMContext``
3661 error reporting mechanisms. This allows a front-end to correlate backend
3662 errors that occur with inline asm back to the source code that produced
3665 .. code-block:: llvm
3667 call void asm sideeffect "something bad", ""(), !srcloc !42
3669 !42 = !{ i32 1234567 }
3671 It is up to the front-end to make sense of the magic numbers it places
3672 in the IR. If the MDNode contains multiple constants, the code generator
3673 will use the one that corresponds to the line of the asm that the error
3681 LLVM IR allows metadata to be attached to instructions in the program
3682 that can convey extra information about the code to the optimizers and
3683 code generator. One example application of metadata is source-level
3684 debug information. There are two metadata primitives: strings and nodes.
3686 Metadata does not have a type, and is not a value. If referenced from a
3687 ``call`` instruction, it uses the ``metadata`` type.
3689 All metadata are identified in syntax by a exclamation point ('``!``').
3691 .. _metadata-string:
3693 Metadata Nodes and Metadata Strings
3694 -----------------------------------
3696 A metadata string is a string surrounded by double quotes. It can
3697 contain any character by escaping non-printable characters with
3698 "``\xx``" where "``xx``" is the two digit hex code. For example:
3701 Metadata nodes are represented with notation similar to structure
3702 constants (a comma separated list of elements, surrounded by braces and
3703 preceded by an exclamation point). Metadata nodes can have any values as
3704 their operand. For example:
3706 .. code-block:: llvm
3708 !{ !"test\00", i32 10}
3710 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
3712 .. code-block:: llvm
3714 !0 = distinct !{!"test\00", i32 10}
3716 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
3717 content. They can also occur when transformations cause uniquing collisions
3718 when metadata operands change.
3720 A :ref:`named metadata <namedmetadatastructure>` is a collection of
3721 metadata nodes, which can be looked up in the module symbol table. For
3724 .. code-block:: llvm
3728 Metadata can be used as function arguments. Here ``llvm.dbg.value``
3729 function is using two metadata arguments:
3731 .. code-block:: llvm
3733 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
3735 Metadata can be attached to an instruction. Here metadata ``!21`` is attached
3736 to the ``add`` instruction using the ``!dbg`` identifier:
3738 .. code-block:: llvm
3740 %indvar.next = add i64 %indvar, 1, !dbg !21
3742 Metadata can also be attached to a function definition. Here metadata ``!22``
3743 is attached to the ``foo`` function using the ``!dbg`` identifier:
3745 .. code-block:: llvm
3747 define void @foo() !dbg !22 {
3751 More information about specific metadata nodes recognized by the
3752 optimizers and code generator is found below.
3754 .. _specialized-metadata:
3756 Specialized Metadata Nodes
3757 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3759 Specialized metadata nodes are custom data structures in metadata (as opposed
3760 to generic tuples). Their fields are labelled, and can be specified in any
3763 These aren't inherently debug info centric, but currently all the specialized
3764 metadata nodes are related to debug info.
3771 ``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
3772 ``retainedTypes:``, ``subprograms:``, ``globals:``, ``imports:`` and ``macros:``
3773 fields are tuples containing the debug info to be emitted along with the compile
3774 unit, regardless of code optimizations (some nodes are only emitted if there are
3775 references to them from instructions).
3777 .. code-block:: llvm
3779 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
3780 isOptimized: true, flags: "-O2", runtimeVersion: 2,
3781 splitDebugFilename: "abc.debug", emissionKind: 1,
3782 enums: !2, retainedTypes: !3, subprograms: !4,
3783 globals: !5, imports: !6, macros: !7, dwoId: 0x0abcd)
3785 Compile unit descriptors provide the root scope for objects declared in a
3786 specific compilation unit. File descriptors are defined using this scope.
3787 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
3788 keep track of subprograms, global variables, type information, and imported
3789 entities (declarations and namespaces).
3796 ``DIFile`` nodes represent files. The ``filename:`` can include slashes.
3798 .. code-block:: llvm
3800 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
3802 Files are sometimes used in ``scope:`` fields, and are the only valid target
3803 for ``file:`` fields.
3810 ``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
3811 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
3813 .. code-block:: llvm
3815 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3816 encoding: DW_ATE_unsigned_char)
3817 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
3819 The ``encoding:`` describes the details of the type. Usually it's one of the
3822 .. code-block:: llvm
3828 DW_ATE_signed_char = 6
3830 DW_ATE_unsigned_char = 8
3832 .. _DISubroutineType:
3837 ``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
3838 refers to a tuple; the first operand is the return type, while the rest are the
3839 types of the formal arguments in order. If the first operand is ``null``, that
3840 represents a function with no return value (such as ``void foo() {}`` in C++).
3842 .. code-block:: llvm
3844 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
3845 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
3846 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
3853 ``DIDerivedType`` nodes represent types derived from other types, such as
3856 .. code-block:: llvm
3858 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3859 encoding: DW_ATE_unsigned_char)
3860 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
3863 The following ``tag:`` values are valid:
3865 .. code-block:: llvm
3867 DW_TAG_formal_parameter = 5
3869 DW_TAG_pointer_type = 15
3870 DW_TAG_reference_type = 16
3872 DW_TAG_ptr_to_member_type = 31
3873 DW_TAG_const_type = 38
3874 DW_TAG_volatile_type = 53
3875 DW_TAG_restrict_type = 55
3877 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
3878 <DICompositeType>` or :ref:`subprogram <DISubprogram>`. The type of the member
3879 is the ``baseType:``. The ``offset:`` is the member's bit offset.
3880 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
3881 argument of a subprogram.
3883 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3885 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3886 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3889 Note that the ``void *`` type is expressed as a type derived from NULL.
3891 .. _DICompositeType:
3896 ``DICompositeType`` nodes represent types composed of other types, like
3897 structures and unions. ``elements:`` points to a tuple of the composed types.
3899 If the source language supports ODR, the ``identifier:`` field gives the unique
3900 identifier used for type merging between modules. When specified, other types
3901 can refer to composite types indirectly via a :ref:`metadata string
3902 <metadata-string>` that matches their identifier.
3904 .. code-block:: llvm
3906 !0 = !DIEnumerator(name: "SixKind", value: 7)
3907 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3908 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3909 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
3910 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3911 elements: !{!0, !1, !2})
3913 The following ``tag:`` values are valid:
3915 .. code-block:: llvm
3917 DW_TAG_array_type = 1
3918 DW_TAG_class_type = 2
3919 DW_TAG_enumeration_type = 4
3920 DW_TAG_structure_type = 19
3921 DW_TAG_union_type = 23
3922 DW_TAG_subroutine_type = 21
3923 DW_TAG_inheritance = 28
3926 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
3927 descriptors <DISubrange>`, each representing the range of subscripts at that
3928 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3929 array type is a native packed vector.
3931 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
3932 descriptors <DIEnumerator>`, each representing the definition of an enumeration
3933 value for the set. All enumeration type descriptors are collected in the
3934 ``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
3936 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3937 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
3938 <DIDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
3945 ``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3946 :ref:`DICompositeType`. ``count: -1`` indicates an empty array.
3948 .. code-block:: llvm
3950 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
3951 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
3952 !2 = !DISubrange(count: -1) ; empty array.
3959 ``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3960 variants of :ref:`DICompositeType`.
3962 .. code-block:: llvm
3964 !0 = !DIEnumerator(name: "SixKind", value: 7)
3965 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3966 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3968 DITemplateTypeParameter
3969 """""""""""""""""""""""
3971 ``DITemplateTypeParameter`` nodes represent type parameters to generic source
3972 language constructs. They are used (optionally) in :ref:`DICompositeType` and
3973 :ref:`DISubprogram` ``templateParams:`` fields.
3975 .. code-block:: llvm
3977 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
3979 DITemplateValueParameter
3980 """"""""""""""""""""""""
3982 ``DITemplateValueParameter`` nodes represent value parameters to generic source
3983 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3984 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3985 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
3986 :ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
3988 .. code-block:: llvm
3990 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
3995 ``DINamespace`` nodes represent namespaces in the source language.
3997 .. code-block:: llvm
3999 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
4004 ``DIGlobalVariable`` nodes represent global variables in the source language.
4006 .. code-block:: llvm
4008 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
4009 file: !2, line: 7, type: !3, isLocal: true,
4010 isDefinition: false, variable: i32* @foo,
4013 All global variables should be referenced by the `globals:` field of a
4014 :ref:`compile unit <DICompileUnit>`.
4021 ``DISubprogram`` nodes represent functions from the source language. A
4022 ``DISubprogram`` may be attached to a function definition using ``!dbg``
4023 metadata. The ``variables:`` field points at :ref:`variables <DILocalVariable>`
4024 that must be retained, even if their IR counterparts are optimized out of
4025 the IR. The ``type:`` field must point at an :ref:`DISubroutineType`.
4027 .. code-block:: llvm
4029 define void @_Z3foov() !dbg !0 {
4033 !0 = distinct !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
4034 file: !2, line: 7, type: !3, isLocal: true,
4035 isDefinition: false, scopeLine: 8,
4037 virtuality: DW_VIRTUALITY_pure_virtual,
4038 virtualIndex: 10, flags: DIFlagPrototyped,
4039 isOptimized: true, templateParams: !5,
4040 declaration: !6, variables: !7)
4047 ``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
4048 <DISubprogram>`. The line number and column numbers are used to distinguish
4049 two lexical blocks at same depth. They are valid targets for ``scope:``
4052 .. code-block:: llvm
4054 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
4056 Usually lexical blocks are ``distinct`` to prevent node merging based on
4059 .. _DILexicalBlockFile:
4064 ``DILexicalBlockFile`` nodes are used to discriminate between sections of a
4065 :ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
4066 indicate textual inclusion, or the ``discriminator:`` field can be used to
4067 discriminate between control flow within a single block in the source language.
4069 .. code-block:: llvm
4071 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
4072 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
4073 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
4080 ``DILocation`` nodes represent source debug locations. The ``scope:`` field is
4081 mandatory, and points at an :ref:`DILexicalBlockFile`, an
4082 :ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
4084 .. code-block:: llvm
4086 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
4088 .. _DILocalVariable:
4093 ``DILocalVariable`` nodes represent local variables in the source language. If
4094 the ``arg:`` field is set to non-zero, then this variable is a subprogram
4095 parameter, and it will be included in the ``variables:`` field of its
4096 :ref:`DISubprogram`.
4098 .. code-block:: llvm
4100 !0 = !DILocalVariable(name: "this", arg: 1, scope: !3, file: !2, line: 7,
4101 type: !3, flags: DIFlagArtificial)
4102 !1 = !DILocalVariable(name: "x", arg: 2, scope: !4, file: !2, line: 7,
4104 !2 = !DILocalVariable(name: "y", scope: !5, file: !2, line: 7, type: !3)
4109 ``DIExpression`` nodes represent DWARF expression sequences. They are used in
4110 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
4111 describe how the referenced LLVM variable relates to the source language
4114 The current supported vocabulary is limited:
4116 - ``DW_OP_deref`` dereferences the working expression.
4117 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
4118 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
4119 here, respectively) of the variable piece from the working expression.
4121 .. code-block:: llvm
4123 !0 = !DIExpression(DW_OP_deref)
4124 !1 = !DIExpression(DW_OP_plus, 3)
4125 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
4126 !3 = !DIExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
4131 ``DIObjCProperty`` nodes represent Objective-C property nodes.
4133 .. code-block:: llvm
4135 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
4136 getter: "getFoo", attributes: 7, type: !2)
4141 ``DIImportedEntity`` nodes represent entities (such as modules) imported into a
4144 .. code-block:: llvm
4146 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
4147 entity: !1, line: 7)
4152 ``DIMacro`` nodes represent definition or undefinition of a macro identifiers.
4153 The ``name:`` field is the macro identifier, followed by macro parameters when
4154 definining a function-like macro, and the ``value`` field is the token-string
4155 used to expand the macro identifier.
4157 .. code-block:: llvm
4159 !2 = !DIMacro(macinfo: DW_MACINFO_define, line: 7, name: "foo(x)",
4161 !3 = !DIMacro(macinfo: DW_MACINFO_undef, line: 30, name: "foo")
4166 ``DIMacroFile`` nodes represent inclusion of source files.
4167 The ``nodes:`` field is a list of ``DIMacro`` and ``DIMacroFile`` nodes that
4168 appear in the included source file.
4170 .. code-block:: llvm
4172 !2 = !DIMacroFile(macinfo: DW_MACINFO_start_file, line: 7, file: !2,
4178 In LLVM IR, memory does not have types, so LLVM's own type system is not
4179 suitable for doing TBAA. Instead, metadata is added to the IR to
4180 describe a type system of a higher level language. This can be used to
4181 implement typical C/C++ TBAA, but it can also be used to implement
4182 custom alias analysis behavior for other languages.
4184 The current metadata format is very simple. TBAA metadata nodes have up
4185 to three fields, e.g.:
4187 .. code-block:: llvm
4189 !0 = !{ !"an example type tree" }
4190 !1 = !{ !"int", !0 }
4191 !2 = !{ !"float", !0 }
4192 !3 = !{ !"const float", !2, i64 1 }
4194 The first field is an identity field. It can be any value, usually a
4195 metadata string, which uniquely identifies the type. The most important
4196 name in the tree is the name of the root node. Two trees with different
4197 root node names are entirely disjoint, even if they have leaves with
4200 The second field identifies the type's parent node in the tree, or is
4201 null or omitted for a root node. A type is considered to alias all of
4202 its descendants and all of its ancestors in the tree. Also, a type is
4203 considered to alias all types in other trees, so that bitcode produced
4204 from multiple front-ends is handled conservatively.
4206 If the third field is present, it's an integer which if equal to 1
4207 indicates that the type is "constant" (meaning
4208 ``pointsToConstantMemory`` should return true; see `other useful
4209 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
4211 '``tbaa.struct``' Metadata
4212 ^^^^^^^^^^^^^^^^^^^^^^^^^^
4214 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
4215 aggregate assignment operations in C and similar languages, however it
4216 is defined to copy a contiguous region of memory, which is more than
4217 strictly necessary for aggregate types which contain holes due to
4218 padding. Also, it doesn't contain any TBAA information about the fields
4221 ``!tbaa.struct`` metadata can describe which memory subregions in a
4222 memcpy are padding and what the TBAA tags of the struct are.
4224 The current metadata format is very simple. ``!tbaa.struct`` metadata
4225 nodes are a list of operands which are in conceptual groups of three.
4226 For each group of three, the first operand gives the byte offset of a
4227 field in bytes, the second gives its size in bytes, and the third gives
4230 .. code-block:: llvm
4232 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
4234 This describes a struct with two fields. The first is at offset 0 bytes
4235 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
4236 and has size 4 bytes and has tbaa tag !2.
4238 Note that the fields need not be contiguous. In this example, there is a
4239 4 byte gap between the two fields. This gap represents padding which
4240 does not carry useful data and need not be preserved.
4242 '``noalias``' and '``alias.scope``' Metadata
4243 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4245 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
4246 noalias memory-access sets. This means that some collection of memory access
4247 instructions (loads, stores, memory-accessing calls, etc.) that carry
4248 ``noalias`` metadata can specifically be specified not to alias with some other
4249 collection of memory access instructions that carry ``alias.scope`` metadata.
4250 Each type of metadata specifies a list of scopes where each scope has an id and
4251 a domain. When evaluating an aliasing query, if for some domain, the set
4252 of scopes with that domain in one instruction's ``alias.scope`` list is a
4253 subset of (or equal to) the set of scopes for that domain in another
4254 instruction's ``noalias`` list, then the two memory accesses are assumed not to
4257 The metadata identifying each domain is itself a list containing one or two
4258 entries. The first entry is the name of the domain. Note that if the name is a
4259 string then it can be combined across functions and translation units. A
4260 self-reference can be used to create globally unique domain names. A
4261 descriptive string may optionally be provided as a second list entry.
4263 The metadata identifying each scope is also itself a list containing two or
4264 three entries. The first entry is the name of the scope. Note that if the name
4265 is a string then it can be combined across functions and translation units. A
4266 self-reference can be used to create globally unique scope names. A metadata
4267 reference to the scope's domain is the second entry. A descriptive string may
4268 optionally be provided as a third list entry.
4272 .. code-block:: llvm
4274 ; Two scope domains:
4278 ; Some scopes in these domains:
4284 !5 = !{!4} ; A list containing only scope !4
4288 ; These two instructions don't alias:
4289 %0 = load float, float* %c, align 4, !alias.scope !5
4290 store float %0, float* %arrayidx.i, align 4, !noalias !5
4292 ; These two instructions also don't alias (for domain !1, the set of scopes
4293 ; in the !alias.scope equals that in the !noalias list):
4294 %2 = load float, float* %c, align 4, !alias.scope !5
4295 store float %2, float* %arrayidx.i2, align 4, !noalias !6
4297 ; These two instructions may alias (for domain !0, the set of scopes in
4298 ; the !noalias list is not a superset of, or equal to, the scopes in the
4299 ; !alias.scope list):
4300 %2 = load float, float* %c, align 4, !alias.scope !6
4301 store float %0, float* %arrayidx.i, align 4, !noalias !7
4303 '``fpmath``' Metadata
4304 ^^^^^^^^^^^^^^^^^^^^^
4306 ``fpmath`` metadata may be attached to any instruction of floating point
4307 type. It can be used to express the maximum acceptable error in the
4308 result of that instruction, in ULPs, thus potentially allowing the
4309 compiler to use a more efficient but less accurate method of computing
4310 it. ULP is defined as follows:
4312 If ``x`` is a real number that lies between two finite consecutive
4313 floating-point numbers ``a`` and ``b``, without being equal to one
4314 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
4315 distance between the two non-equal finite floating-point numbers
4316 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
4318 The metadata node shall consist of a single positive floating point
4319 number representing the maximum relative error, for example:
4321 .. code-block:: llvm
4323 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
4327 '``range``' Metadata
4328 ^^^^^^^^^^^^^^^^^^^^
4330 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
4331 integer types. It expresses the possible ranges the loaded value or the value
4332 returned by the called function at this call site is in. The ranges are
4333 represented with a flattened list of integers. The loaded value or the value
4334 returned is known to be in the union of the ranges defined by each consecutive
4335 pair. Each pair has the following properties:
4337 - The type must match the type loaded by the instruction.
4338 - The pair ``a,b`` represents the range ``[a,b)``.
4339 - Both ``a`` and ``b`` are constants.
4340 - The range is allowed to wrap.
4341 - The range should not represent the full or empty set. That is,
4344 In addition, the pairs must be in signed order of the lower bound and
4345 they must be non-contiguous.
4349 .. code-block:: llvm
4351 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
4352 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
4353 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
4354 %d = invoke i8 @bar() to label %cont
4355 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
4357 !0 = !{ i8 0, i8 2 }
4358 !1 = !{ i8 255, i8 2 }
4359 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
4360 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
4362 '``unpredictable``' Metadata
4363 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4365 ``unpredictable`` metadata may be attached to any branch or switch
4366 instruction. It can be used to express the unpredictability of control
4367 flow. Similar to the llvm.expect intrinsic, it may be used to alter
4368 optimizations related to compare and branch instructions. The metadata
4369 is treated as a boolean value; if it exists, it signals that the branch
4370 or switch that it is attached to is completely unpredictable.
4375 It is sometimes useful to attach information to loop constructs. Currently,
4376 loop metadata is implemented as metadata attached to the branch instruction
4377 in the loop latch block. This type of metadata refer to a metadata node that is
4378 guaranteed to be separate for each loop. The loop identifier metadata is
4379 specified with the name ``llvm.loop``.
4381 The loop identifier metadata is implemented using a metadata that refers to
4382 itself to avoid merging it with any other identifier metadata, e.g.,
4383 during module linkage or function inlining. That is, each loop should refer
4384 to their own identification metadata even if they reside in separate functions.
4385 The following example contains loop identifier metadata for two separate loop
4388 .. code-block:: llvm
4393 The loop identifier metadata can be used to specify additional
4394 per-loop metadata. Any operands after the first operand can be treated
4395 as user-defined metadata. For example the ``llvm.loop.unroll.count``
4396 suggests an unroll factor to the loop unroller:
4398 .. code-block:: llvm
4400 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
4403 !1 = !{!"llvm.loop.unroll.count", i32 4}
4405 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
4406 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4408 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
4409 used to control per-loop vectorization and interleaving parameters such as
4410 vectorization width and interleave count. These metadata should be used in
4411 conjunction with ``llvm.loop`` loop identification metadata. The
4412 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
4413 optimization hints and the optimizer will only interleave and vectorize loops if
4414 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
4415 which contains information about loop-carried memory dependencies can be helpful
4416 in determining the safety of these transformations.
4418 '``llvm.loop.interleave.count``' Metadata
4419 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4421 This metadata suggests an interleave count to the loop interleaver.
4422 The first operand is the string ``llvm.loop.interleave.count`` and the
4423 second operand is an integer specifying the interleave count. For
4426 .. code-block:: llvm
4428 !0 = !{!"llvm.loop.interleave.count", i32 4}
4430 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
4431 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
4432 then the interleave count will be determined automatically.
4434 '``llvm.loop.vectorize.enable``' Metadata
4435 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4437 This metadata selectively enables or disables vectorization for the loop. The
4438 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
4439 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
4440 0 disables vectorization:
4442 .. code-block:: llvm
4444 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
4445 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
4447 '``llvm.loop.vectorize.width``' Metadata
4448 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4450 This metadata sets the target width of the vectorizer. The first
4451 operand is the string ``llvm.loop.vectorize.width`` and the second
4452 operand is an integer specifying the width. For example:
4454 .. code-block:: llvm
4456 !0 = !{!"llvm.loop.vectorize.width", i32 4}
4458 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
4459 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
4460 0 or if the loop does not have this metadata the width will be
4461 determined automatically.
4463 '``llvm.loop.unroll``'
4464 ^^^^^^^^^^^^^^^^^^^^^^
4466 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
4467 optimization hints such as the unroll factor. ``llvm.loop.unroll``
4468 metadata should be used in conjunction with ``llvm.loop`` loop
4469 identification metadata. The ``llvm.loop.unroll`` metadata are only
4470 optimization hints and the unrolling will only be performed if the
4471 optimizer believes it is safe to do so.
4473 '``llvm.loop.unroll.count``' Metadata
4474 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4476 This metadata suggests an unroll factor to the loop unroller. The
4477 first operand is the string ``llvm.loop.unroll.count`` and the second
4478 operand is a positive integer specifying the unroll factor. For
4481 .. code-block:: llvm
4483 !0 = !{!"llvm.loop.unroll.count", i32 4}
4485 If the trip count of the loop is less than the unroll count the loop
4486 will be partially unrolled.
4488 '``llvm.loop.unroll.disable``' Metadata
4489 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4491 This metadata disables loop unrolling. The metadata has a single operand
4492 which is the string ``llvm.loop.unroll.disable``. For example:
4494 .. code-block:: llvm
4496 !0 = !{!"llvm.loop.unroll.disable"}
4498 '``llvm.loop.unroll.runtime.disable``' Metadata
4499 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4501 This metadata disables runtime loop unrolling. The metadata has a single
4502 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
4504 .. code-block:: llvm
4506 !0 = !{!"llvm.loop.unroll.runtime.disable"}
4508 '``llvm.loop.unroll.enable``' Metadata
4509 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4511 This metadata suggests that the loop should be fully unrolled if the trip count
4512 is known at compile time and partially unrolled if the trip count is not known
4513 at compile time. The metadata has a single operand which is the string
4514 ``llvm.loop.unroll.enable``. For example:
4516 .. code-block:: llvm
4518 !0 = !{!"llvm.loop.unroll.enable"}
4520 '``llvm.loop.unroll.full``' Metadata
4521 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4523 This metadata suggests that the loop should be unrolled fully. The
4524 metadata has a single operand which is the string ``llvm.loop.unroll.full``.
4527 .. code-block:: llvm
4529 !0 = !{!"llvm.loop.unroll.full"}
4534 Metadata types used to annotate memory accesses with information helpful
4535 for optimizations are prefixed with ``llvm.mem``.
4537 '``llvm.mem.parallel_loop_access``' Metadata
4538 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4540 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
4541 or metadata containing a list of loop identifiers for nested loops.
4542 The metadata is attached to memory accessing instructions and denotes that
4543 no loop carried memory dependence exist between it and other instructions denoted
4544 with the same loop identifier.
4546 Precisely, given two instructions ``m1`` and ``m2`` that both have the
4547 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
4548 set of loops associated with that metadata, respectively, then there is no loop
4549 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
4552 As a special case, if all memory accessing instructions in a loop have
4553 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
4554 loop has no loop carried memory dependences and is considered to be a parallel
4557 Note that if not all memory access instructions have such metadata referring to
4558 the loop, then the loop is considered not being trivially parallel. Additional
4559 memory dependence analysis is required to make that determination. As a fail
4560 safe mechanism, this causes loops that were originally parallel to be considered
4561 sequential (if optimization passes that are unaware of the parallel semantics
4562 insert new memory instructions into the loop body).
4564 Example of a loop that is considered parallel due to its correct use of
4565 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
4566 metadata types that refer to the same loop identifier metadata.
4568 .. code-block:: llvm
4572 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
4574 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4576 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
4582 It is also possible to have nested parallel loops. In that case the
4583 memory accesses refer to a list of loop identifier metadata nodes instead of
4584 the loop identifier metadata node directly:
4586 .. code-block:: llvm
4590 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
4592 br label %inner.for.body
4596 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4598 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
4600 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
4604 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
4606 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
4608 outer.for.end: ; preds = %for.body
4610 !0 = !{!1, !2} ; a list of loop identifiers
4611 !1 = !{!1} ; an identifier for the inner loop
4612 !2 = !{!2} ; an identifier for the outer loop
4617 The ``llvm.bitsets`` global metadata is used to implement
4618 :doc:`bitsets <BitSets>`.
4620 '``invariant.group``' Metadata
4621 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4623 The ``invariant.group`` metadata may be attached to ``load``/``store`` instructions.
4624 The existence of the ``invariant.group`` metadata on the instruction tells
4625 the optimizer that every ``load`` and ``store`` to the same pointer operand
4626 within the same invariant group can be assumed to load or store the same
4627 value (but see the ``llvm.invariant.group.barrier`` intrinsic which affects
4628 when two pointers are considered the same).
4632 .. code-block:: llvm
4634 @unknownPtr = external global i8
4637 store i8 42, i8* %ptr, !invariant.group !0
4638 call void @foo(i8* %ptr)
4640 %a = load i8, i8* %ptr, !invariant.group !0 ; Can assume that value under %ptr didn't change
4641 call void @foo(i8* %ptr)
4642 %b = load i8, i8* %ptr, !invariant.group !1 ; Can't assume anything, because group changed
4644 %newPtr = call i8* @getPointer(i8* %ptr)
4645 %c = load i8, i8* %newPtr, !invariant.group !0 ; Can't assume anything, because we only have information about %ptr
4647 %unknownValue = load i8, i8* @unknownPtr
4648 store i8 %unknownValue, i8* %ptr, !invariant.group !0 ; Can assume that %unknownValue == 42
4650 call void @foo(i8* %ptr)
4651 %newPtr2 = call i8* @llvm.invariant.group.barrier(i8* %ptr)
4652 %d = load i8, i8* %newPtr2, !invariant.group !0 ; Can't step through invariant.group.barrier to get value of %ptr
4655 declare void @foo(i8*)
4656 declare i8* @getPointer(i8*)
4657 declare i8* @llvm.invariant.group.barrier(i8*)
4659 !0 = !{!"magic ptr"}
4660 !1 = !{!"other ptr"}
4664 Module Flags Metadata
4665 =====================
4667 Information about the module as a whole is difficult to convey to LLVM's
4668 subsystems. The LLVM IR isn't sufficient to transmit this information.
4669 The ``llvm.module.flags`` named metadata exists in order to facilitate
4670 this. These flags are in the form of key / value pairs --- much like a
4671 dictionary --- making it easy for any subsystem who cares about a flag to
4674 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
4675 Each triplet has the following form:
4677 - The first element is a *behavior* flag, which specifies the behavior
4678 when two (or more) modules are merged together, and it encounters two
4679 (or more) metadata with the same ID. The supported behaviors are
4681 - The second element is a metadata string that is a unique ID for the
4682 metadata. Each module may only have one flag entry for each unique ID (not
4683 including entries with the **Require** behavior).
4684 - The third element is the value of the flag.
4686 When two (or more) modules are merged together, the resulting
4687 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
4688 each unique metadata ID string, there will be exactly one entry in the merged
4689 modules ``llvm.module.flags`` metadata table, and the value for that entry will
4690 be determined by the merge behavior flag, as described below. The only exception
4691 is that entries with the *Require* behavior are always preserved.
4693 The following behaviors are supported:
4704 Emits an error if two values disagree, otherwise the resulting value
4705 is that of the operands.
4709 Emits a warning if two values disagree. The result value will be the
4710 operand for the flag from the first module being linked.
4714 Adds a requirement that another module flag be present and have a
4715 specified value after linking is performed. The value must be a
4716 metadata pair, where the first element of the pair is the ID of the
4717 module flag to be restricted, and the second element of the pair is
4718 the value the module flag should be restricted to. This behavior can
4719 be used to restrict the allowable results (via triggering of an
4720 error) of linking IDs with the **Override** behavior.
4724 Uses the specified value, regardless of the behavior or value of the
4725 other module. If both modules specify **Override**, but the values
4726 differ, an error will be emitted.
4730 Appends the two values, which are required to be metadata nodes.
4734 Appends the two values, which are required to be metadata
4735 nodes. However, duplicate entries in the second list are dropped
4736 during the append operation.
4738 It is an error for a particular unique flag ID to have multiple behaviors,
4739 except in the case of **Require** (which adds restrictions on another metadata
4740 value) or **Override**.
4742 An example of module flags:
4744 .. code-block:: llvm
4746 !0 = !{ i32 1, !"foo", i32 1 }
4747 !1 = !{ i32 4, !"bar", i32 37 }
4748 !2 = !{ i32 2, !"qux", i32 42 }
4749 !3 = !{ i32 3, !"qux",
4754 !llvm.module.flags = !{ !0, !1, !2, !3 }
4756 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
4757 if two or more ``!"foo"`` flags are seen is to emit an error if their
4758 values are not equal.
4760 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
4761 behavior if two or more ``!"bar"`` flags are seen is to use the value
4764 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
4765 behavior if two or more ``!"qux"`` flags are seen is to emit a
4766 warning if their values are not equal.
4768 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
4774 The behavior is to emit an error if the ``llvm.module.flags`` does not
4775 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
4778 Objective-C Garbage Collection Module Flags Metadata
4779 ----------------------------------------------------
4781 On the Mach-O platform, Objective-C stores metadata about garbage
4782 collection in a special section called "image info". The metadata
4783 consists of a version number and a bitmask specifying what types of
4784 garbage collection are supported (if any) by the file. If two or more
4785 modules are linked together their garbage collection metadata needs to
4786 be merged rather than appended together.
4788 The Objective-C garbage collection module flags metadata consists of the
4789 following key-value pairs:
4798 * - ``Objective-C Version``
4799 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
4801 * - ``Objective-C Image Info Version``
4802 - **[Required]** --- The version of the image info section. Currently
4805 * - ``Objective-C Image Info Section``
4806 - **[Required]** --- The section to place the metadata. Valid values are
4807 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
4808 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
4809 Objective-C ABI version 2.
4811 * - ``Objective-C Garbage Collection``
4812 - **[Required]** --- Specifies whether garbage collection is supported or
4813 not. Valid values are 0, for no garbage collection, and 2, for garbage
4814 collection supported.
4816 * - ``Objective-C GC Only``
4817 - **[Optional]** --- Specifies that only garbage collection is supported.
4818 If present, its value must be 6. This flag requires that the
4819 ``Objective-C Garbage Collection`` flag have the value 2.
4821 Some important flag interactions:
4823 - If a module with ``Objective-C Garbage Collection`` set to 0 is
4824 merged with a module with ``Objective-C Garbage Collection`` set to
4825 2, then the resulting module has the
4826 ``Objective-C Garbage Collection`` flag set to 0.
4827 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
4828 merged with a module with ``Objective-C GC Only`` set to 6.
4830 Automatic Linker Flags Module Flags Metadata
4831 --------------------------------------------
4833 Some targets support embedding flags to the linker inside individual object
4834 files. Typically this is used in conjunction with language extensions which
4835 allow source files to explicitly declare the libraries they depend on, and have
4836 these automatically be transmitted to the linker via object files.
4838 These flags are encoded in the IR using metadata in the module flags section,
4839 using the ``Linker Options`` key. The merge behavior for this flag is required
4840 to be ``AppendUnique``, and the value for the key is expected to be a metadata
4841 node which should be a list of other metadata nodes, each of which should be a
4842 list of metadata strings defining linker options.
4844 For example, the following metadata section specifies two separate sets of
4845 linker options, presumably to link against ``libz`` and the ``Cocoa``
4848 !0 = !{ i32 6, !"Linker Options",
4851 !{ !"-framework", !"Cocoa" } } }
4852 !llvm.module.flags = !{ !0 }
4854 The metadata encoding as lists of lists of options, as opposed to a collapsed
4855 list of options, is chosen so that the IR encoding can use multiple option
4856 strings to specify e.g., a single library, while still having that specifier be
4857 preserved as an atomic element that can be recognized by a target specific
4858 assembly writer or object file emitter.
4860 Each individual option is required to be either a valid option for the target's
4861 linker, or an option that is reserved by the target specific assembly writer or
4862 object file emitter. No other aspect of these options is defined by the IR.
4864 C type width Module Flags Metadata
4865 ----------------------------------
4867 The ARM backend emits a section into each generated object file describing the
4868 options that it was compiled with (in a compiler-independent way) to prevent
4869 linking incompatible objects, and to allow automatic library selection. Some
4870 of these options are not visible at the IR level, namely wchar_t width and enum
4873 To pass this information to the backend, these options are encoded in module
4874 flags metadata, using the following key-value pairs:
4884 - * 0 --- sizeof(wchar_t) == 4
4885 * 1 --- sizeof(wchar_t) == 2
4888 - * 0 --- Enums are at least as large as an ``int``.
4889 * 1 --- Enums are stored in the smallest integer type which can
4890 represent all of its values.
4892 For example, the following metadata section specifies that the module was
4893 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
4894 enum is the smallest type which can represent all of its values::
4896 !llvm.module.flags = !{!0, !1}
4897 !0 = !{i32 1, !"short_wchar", i32 1}
4898 !1 = !{i32 1, !"short_enum", i32 0}
4900 .. _intrinsicglobalvariables:
4902 Intrinsic Global Variables
4903 ==========================
4905 LLVM has a number of "magic" global variables that contain data that
4906 affect code generation or other IR semantics. These are documented here.
4907 All globals of this sort should have a section specified as
4908 "``llvm.metadata``". This section and all globals that start with
4909 "``llvm.``" are reserved for use by LLVM.
4913 The '``llvm.used``' Global Variable
4914 -----------------------------------
4916 The ``@llvm.used`` global is an array which has
4917 :ref:`appending linkage <linkage_appending>`. This array contains a list of
4918 pointers to named global variables, functions and aliases which may optionally
4919 have a pointer cast formed of bitcast or getelementptr. For example, a legal
4922 .. code-block:: llvm
4927 @llvm.used = appending global [2 x i8*] [
4929 i8* bitcast (i32* @Y to i8*)
4930 ], section "llvm.metadata"
4932 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
4933 and linker are required to treat the symbol as if there is a reference to the
4934 symbol that it cannot see (which is why they have to be named). For example, if
4935 a variable has internal linkage and no references other than that from the
4936 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
4937 references from inline asms and other things the compiler cannot "see", and
4938 corresponds to "``attribute((used))``" in GNU C.
4940 On some targets, the code generator must emit a directive to the
4941 assembler or object file to prevent the assembler and linker from
4942 molesting the symbol.
4944 .. _gv_llvmcompilerused:
4946 The '``llvm.compiler.used``' Global Variable
4947 --------------------------------------------
4949 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4950 directive, except that it only prevents the compiler from touching the
4951 symbol. On targets that support it, this allows an intelligent linker to
4952 optimize references to the symbol without being impeded as it would be
4955 This is a rare construct that should only be used in rare circumstances,
4956 and should not be exposed to source languages.
4958 .. _gv_llvmglobalctors:
4960 The '``llvm.global_ctors``' Global Variable
4961 -------------------------------------------
4963 .. code-block:: llvm
4965 %0 = type { i32, void ()*, i8* }
4966 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
4968 The ``@llvm.global_ctors`` array contains a list of constructor
4969 functions, priorities, and an optional associated global or function.
4970 The functions referenced by this array will be called in ascending order
4971 of priority (i.e. lowest first) when the module is loaded. The order of
4972 functions with the same priority is not defined.
4974 If the third field is present, non-null, and points to a global variable
4975 or function, the initializer function will only run if the associated
4976 data from the current module is not discarded.
4978 .. _llvmglobaldtors:
4980 The '``llvm.global_dtors``' Global Variable
4981 -------------------------------------------
4983 .. code-block:: llvm
4985 %0 = type { i32, void ()*, i8* }
4986 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
4988 The ``@llvm.global_dtors`` array contains a list of destructor
4989 functions, priorities, and an optional associated global or function.
4990 The functions referenced by this array will be called in descending
4991 order of priority (i.e. highest first) when the module is unloaded. The
4992 order of functions with the same priority is not defined.
4994 If the third field is present, non-null, and points to a global variable
4995 or function, the destructor function will only run if the associated
4996 data from the current module is not discarded.
4998 Instruction Reference
4999 =====================
5001 The LLVM instruction set consists of several different classifications
5002 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
5003 instructions <binaryops>`, :ref:`bitwise binary
5004 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
5005 :ref:`other instructions <otherops>`.
5009 Terminator Instructions
5010 -----------------------
5012 As mentioned :ref:`previously <functionstructure>`, every basic block in a
5013 program ends with a "Terminator" instruction, which indicates which
5014 block should be executed after the current block is finished. These
5015 terminator instructions typically yield a '``void``' value: they produce
5016 control flow, not values (the one exception being the
5017 ':ref:`invoke <i_invoke>`' instruction).
5019 The terminator instructions are: ':ref:`ret <i_ret>`',
5020 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
5021 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
5022 ':ref:`resume <i_resume>`', ':ref:`catchswitch <i_catchswitch>`',
5023 ':ref:`catchret <i_catchret>`',
5024 ':ref:`cleanupret <i_cleanupret>`',
5025 and ':ref:`unreachable <i_unreachable>`'.
5029 '``ret``' Instruction
5030 ^^^^^^^^^^^^^^^^^^^^^
5037 ret <type> <value> ; Return a value from a non-void function
5038 ret void ; Return from void function
5043 The '``ret``' instruction is used to return control flow (and optionally
5044 a value) from a function back to the caller.
5046 There are two forms of the '``ret``' instruction: one that returns a
5047 value and then causes control flow, and one that just causes control
5053 The '``ret``' instruction optionally accepts a single argument, the
5054 return value. The type of the return value must be a ':ref:`first
5055 class <t_firstclass>`' type.
5057 A function is not :ref:`well formed <wellformed>` if it it has a non-void
5058 return type and contains a '``ret``' instruction with no return value or
5059 a return value with a type that does not match its type, or if it has a
5060 void return type and contains a '``ret``' instruction with a return
5066 When the '``ret``' instruction is executed, control flow returns back to
5067 the calling function's context. If the caller is a
5068 ":ref:`call <i_call>`" instruction, execution continues at the
5069 instruction after the call. If the caller was an
5070 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
5071 beginning of the "normal" destination block. If the instruction returns
5072 a value, that value shall set the call or invoke instruction's return
5078 .. code-block:: llvm
5080 ret i32 5 ; Return an integer value of 5
5081 ret void ; Return from a void function
5082 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
5086 '``br``' Instruction
5087 ^^^^^^^^^^^^^^^^^^^^
5094 br i1 <cond>, label <iftrue>, label <iffalse>
5095 br label <dest> ; Unconditional branch
5100 The '``br``' instruction is used to cause control flow to transfer to a
5101 different basic block in the current function. There are two forms of
5102 this instruction, corresponding to a conditional branch and an
5103 unconditional branch.
5108 The conditional branch form of the '``br``' instruction takes a single
5109 '``i1``' value and two '``label``' values. The unconditional form of the
5110 '``br``' instruction takes a single '``label``' value as a target.
5115 Upon execution of a conditional '``br``' instruction, the '``i1``'
5116 argument is evaluated. If the value is ``true``, control flows to the
5117 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
5118 to the '``iffalse``' ``label`` argument.
5123 .. code-block:: llvm
5126 %cond = icmp eq i32 %a, %b
5127 br i1 %cond, label %IfEqual, label %IfUnequal
5135 '``switch``' Instruction
5136 ^^^^^^^^^^^^^^^^^^^^^^^^
5143 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
5148 The '``switch``' instruction is used to transfer control flow to one of
5149 several different places. It is a generalization of the '``br``'
5150 instruction, allowing a branch to occur to one of many possible
5156 The '``switch``' instruction uses three parameters: an integer
5157 comparison value '``value``', a default '``label``' destination, and an
5158 array of pairs of comparison value constants and '``label``'s. The table
5159 is not allowed to contain duplicate constant entries.
5164 The ``switch`` instruction specifies a table of values and destinations.
5165 When the '``switch``' instruction is executed, this table is searched
5166 for the given value. If the value is found, control flow is transferred
5167 to the corresponding destination; otherwise, control flow is transferred
5168 to the default destination.
5173 Depending on properties of the target machine and the particular
5174 ``switch`` instruction, this instruction may be code generated in
5175 different ways. For example, it could be generated as a series of
5176 chained conditional branches or with a lookup table.
5181 .. code-block:: llvm
5183 ; Emulate a conditional br instruction
5184 %Val = zext i1 %value to i32
5185 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
5187 ; Emulate an unconditional br instruction
5188 switch i32 0, label %dest [ ]
5190 ; Implement a jump table:
5191 switch i32 %val, label %otherwise [ i32 0, label %onzero
5193 i32 2, label %ontwo ]
5197 '``indirectbr``' Instruction
5198 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5205 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
5210 The '``indirectbr``' instruction implements an indirect branch to a
5211 label within the current function, whose address is specified by
5212 "``address``". Address must be derived from a
5213 :ref:`blockaddress <blockaddress>` constant.
5218 The '``address``' argument is the address of the label to jump to. The
5219 rest of the arguments indicate the full set of possible destinations
5220 that the address may point to. Blocks are allowed to occur multiple
5221 times in the destination list, though this isn't particularly useful.
5223 This destination list is required so that dataflow analysis has an
5224 accurate understanding of the CFG.
5229 Control transfers to the block specified in the address argument. All
5230 possible destination blocks must be listed in the label list, otherwise
5231 this instruction has undefined behavior. This implies that jumps to
5232 labels defined in other functions have undefined behavior as well.
5237 This is typically implemented with a jump through a register.
5242 .. code-block:: llvm
5244 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
5248 '``invoke``' Instruction
5249 ^^^^^^^^^^^^^^^^^^^^^^^^
5256 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
5257 [operand bundles] to label <normal label> unwind label <exception label>
5262 The '``invoke``' instruction causes control to transfer to a specified
5263 function, with the possibility of control flow transfer to either the
5264 '``normal``' label or the '``exception``' label. If the callee function
5265 returns with the "``ret``" instruction, control flow will return to the
5266 "normal" label. If the callee (or any indirect callees) returns via the
5267 ":ref:`resume <i_resume>`" instruction or other exception handling
5268 mechanism, control is interrupted and continued at the dynamically
5269 nearest "exception" label.
5271 The '``exception``' label is a `landing
5272 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
5273 '``exception``' label is required to have the
5274 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
5275 information about the behavior of the program after unwinding happens,
5276 as its first non-PHI instruction. The restrictions on the
5277 "``landingpad``" instruction's tightly couples it to the "``invoke``"
5278 instruction, so that the important information contained within the
5279 "``landingpad``" instruction can't be lost through normal code motion.
5284 This instruction requires several arguments:
5286 #. The optional "cconv" marker indicates which :ref:`calling
5287 convention <callingconv>` the call should use. If none is
5288 specified, the call defaults to using C calling conventions.
5289 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5290 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5292 #. '``ptr to function ty``': shall be the signature of the pointer to
5293 function value being invoked. In most cases, this is a direct
5294 function invocation, but indirect ``invoke``'s are just as possible,
5295 branching off an arbitrary pointer to function value.
5296 #. '``function ptr val``': An LLVM value containing a pointer to a
5297 function to be invoked.
5298 #. '``function args``': argument list whose types match the function
5299 signature argument types and parameter attributes. All arguments must
5300 be of :ref:`first class <t_firstclass>` type. If the function signature
5301 indicates the function accepts a variable number of arguments, the
5302 extra arguments can be specified.
5303 #. '``normal label``': the label reached when the called function
5304 executes a '``ret``' instruction.
5305 #. '``exception label``': the label reached when a callee returns via
5306 the :ref:`resume <i_resume>` instruction or other exception handling
5308 #. The optional :ref:`function attributes <fnattrs>` list. Only
5309 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5310 attributes are valid here.
5311 #. The optional :ref:`operand bundles <opbundles>` list.
5316 This instruction is designed to operate as a standard '``call``'
5317 instruction in most regards. The primary difference is that it
5318 establishes an association with a label, which is used by the runtime
5319 library to unwind the stack.
5321 This instruction is used in languages with destructors to ensure that
5322 proper cleanup is performed in the case of either a ``longjmp`` or a
5323 thrown exception. Additionally, this is important for implementation of
5324 '``catch``' clauses in high-level languages that support them.
5326 For the purposes of the SSA form, the definition of the value returned
5327 by the '``invoke``' instruction is deemed to occur on the edge from the
5328 current block to the "normal" label. If the callee unwinds then no
5329 return value is available.
5334 .. code-block:: llvm
5336 %retval = invoke i32 @Test(i32 15) to label %Continue
5337 unwind label %TestCleanup ; i32:retval set
5338 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
5339 unwind label %TestCleanup ; i32:retval set
5343 '``resume``' Instruction
5344 ^^^^^^^^^^^^^^^^^^^^^^^^
5351 resume <type> <value>
5356 The '``resume``' instruction is a terminator instruction that has no
5362 The '``resume``' instruction requires one argument, which must have the
5363 same type as the result of any '``landingpad``' instruction in the same
5369 The '``resume``' instruction resumes propagation of an existing
5370 (in-flight) exception whose unwinding was interrupted with a
5371 :ref:`landingpad <i_landingpad>` instruction.
5376 .. code-block:: llvm
5378 resume { i8*, i32 } %exn
5382 '``catchswitch``' Instruction
5383 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5390 <resultval> = catchswitch within <parent> [ label <handler1>, label <handler2>, ... ] unwind to caller
5391 <resultval> = catchswitch within <parent> [ label <handler1>, label <handler2>, ... ] unwind label <default>
5396 The '``catchswitch``' instruction is used by `LLVM's exception handling system
5397 <ExceptionHandling.html#overview>`_ to describe the set of possible catch handlers
5398 that may be executed by the :ref:`EH personality routine <personalityfn>`.
5403 The ``parent`` argument is the token of the funclet that contains the
5404 ``catchswitch`` instruction. If the ``catchswitch`` is not inside a funclet,
5405 this operand may be the token ``none``.
5407 The ``default`` argument is the label of another basic block beginning with a
5408 "pad" instruction, one of ``cleanuppad`` or ``catchswitch``.
5410 The ``handlers`` are a list of successor blocks that each begin with a
5411 :ref:`catchpad <i_catchpad>` instruction.
5416 Executing this instruction transfers control to one of the successors in
5417 ``handlers``, if appropriate, or continues to unwind via the unwind label if
5420 The ``catchswitch`` is both a terminator and a "pad" instruction, meaning that
5421 it must be both the first non-phi instruction and last instruction in the basic
5422 block. Therefore, it must be the only non-phi instruction in the block.
5427 .. code-block:: llvm
5430 %cs1 = catchswitch within none [label %handler0, label %handler1] unwind to caller
5432 %cs2 = catchswitch within %parenthandler [label %handler0] unwind label %cleanup
5436 '``catchpad``' Instruction
5437 ^^^^^^^^^^^^^^^^^^^^^^^^^^
5444 <resultval> = catchpad within <catchswitch> [<args>*]
5449 The '``catchpad``' instruction is used by `LLVM's exception handling
5450 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5451 begins a catch handler --- one where a personality routine attempts to transfer
5452 control to catch an exception.
5457 The ``catchswitch`` operand must always be a token produced by a
5458 :ref:`catchswitch <i_catchswitch>` instruction in a predecessor block. This
5459 ensures that each ``catchpad`` has exactly one predecessor block, and it always
5460 terminates in a ``catchswitch``.
5462 The ``args`` correspond to whatever information the personality routine
5463 requires to know if this is an appropriate handler for the exception. Control
5464 will transfer to the ``catchpad`` if this is the first appropriate handler for
5467 The ``resultval`` has the type :ref:`token <t_token>` and is used to match the
5468 ``catchpad`` to corresponding :ref:`catchrets <i_catchret>` and other nested EH
5474 When the call stack is being unwound due to an exception being thrown, the
5475 exception is compared against the ``args``. If it doesn't match, control will
5476 not reach the ``catchpad`` instruction. The representation of ``args`` is
5477 entirely target and personality function-specific.
5479 Like the :ref:`landingpad <i_landingpad>` instruction, the ``catchpad``
5480 instruction must be the first non-phi of its parent basic block.
5482 The meaning of the tokens produced and consumed by ``catchpad`` and other "pad"
5483 instructions is described in the
5484 `Windows exception handling documentation <ExceptionHandling.html#wineh>`.
5486 Executing a ``catchpad`` instruction constitutes "entering" that pad.
5487 The pad may then be "exited" in one of three ways:
5489 1) explicitly via a ``catchret`` that consumes it. Executing such a ``catchret``
5490 is undefined behavior if any descendant pads have been entered but not yet
5492 2) implicitly via a call (which unwinds all the way to the current function's caller),
5493 or via a ``catchswitch`` or a ``cleanupret`` that unwinds to caller.
5494 3) implicitly via an unwind edge whose destination EH pad isn't a descendant of
5495 the ``catchpad``. When the ``catchpad`` is exited in this manner, it is
5496 undefined behavior if the destination EH pad has a parent which is not an
5497 ancestor of the ``catchpad`` being exited.
5502 .. code-block:: llvm
5505 %cs = catchswitch within none [label %handler0] unwind to caller
5506 ;; A catch block which can catch an integer.
5508 %tok = catchpad within %cs [i8** @_ZTIi]
5512 '``catchret``' Instruction
5513 ^^^^^^^^^^^^^^^^^^^^^^^^^^
5520 catchret from <token> to label <normal>
5525 The '``catchret``' instruction is a terminator instruction that has a
5532 The first argument to a '``catchret``' indicates which ``catchpad`` it
5533 exits. It must be a :ref:`catchpad <i_catchpad>`.
5534 The second argument to a '``catchret``' specifies where control will
5540 The '``catchret``' instruction ends an existing (in-flight) exception whose
5541 unwinding was interrupted with a :ref:`catchpad <i_catchpad>` instruction. The
5542 :ref:`personality function <personalityfn>` gets a chance to execute arbitrary
5543 code to, for example, destroy the active exception. Control then transfers to
5546 The ``token`` argument must be a token produced by a dominating ``catchpad``
5547 instruction. The ``catchret`` destroys the physical frame established by
5548 ``catchpad``, so executing multiple returns on the same token without
5549 re-executing the ``catchpad`` will result in undefined behavior.
5550 See :ref:`catchpad <i_catchpad>` for more details.
5555 .. code-block:: llvm
5557 catchret from %catch label %continue
5561 '``cleanupret``' Instruction
5562 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5569 cleanupret from <value> unwind label <continue>
5570 cleanupret from <value> unwind to caller
5575 The '``cleanupret``' instruction is a terminator instruction that has
5576 an optional successor.
5582 The '``cleanupret``' instruction requires one argument, which indicates
5583 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
5584 It also has an optional successor, ``continue``.
5589 The '``cleanupret``' instruction indicates to the
5590 :ref:`personality function <personalityfn>` that one
5591 :ref:`cleanuppad <i_cleanuppad>` it transferred control to has ended.
5592 It transfers control to ``continue`` or unwinds out of the function.
5594 The unwind destination ``continue``, if present, must be an EH pad
5595 whose parent is either ``none`` or an ancestor of the ``cleanuppad``
5596 being returned from. This constitutes an exceptional exit from all
5597 ancestors of the completed ``cleanuppad``, up to but not including
5598 the parent of ``continue``.
5599 See :ref:`cleanuppad <i_cleanuppad>` for more details.
5604 .. code-block:: llvm
5606 cleanupret from %cleanup unwind to caller
5607 cleanupret from %cleanup unwind label %continue
5611 '``unreachable``' Instruction
5612 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5624 The '``unreachable``' instruction has no defined semantics. This
5625 instruction is used to inform the optimizer that a particular portion of
5626 the code is not reachable. This can be used to indicate that the code
5627 after a no-return function cannot be reached, and other facts.
5632 The '``unreachable``' instruction has no defined semantics.
5639 Binary operators are used to do most of the computation in a program.
5640 They require two operands of the same type, execute an operation on
5641 them, and produce a single value. The operands might represent multiple
5642 data, as is the case with the :ref:`vector <t_vector>` data type. The
5643 result value has the same type as its operands.
5645 There are several different binary operators:
5649 '``add``' Instruction
5650 ^^^^^^^^^^^^^^^^^^^^^
5657 <result> = add <ty> <op1>, <op2> ; yields ty:result
5658 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
5659 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
5660 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
5665 The '``add``' instruction returns the sum of its two operands.
5670 The two arguments to the '``add``' instruction must be
5671 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5672 arguments must have identical types.
5677 The value produced is the integer sum of the two operands.
5679 If the sum has unsigned overflow, the result returned is the
5680 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5683 Because LLVM integers use a two's complement representation, this
5684 instruction is appropriate for both signed and unsigned integers.
5686 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5687 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5688 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
5689 unsigned and/or signed overflow, respectively, occurs.
5694 .. code-block:: llvm
5696 <result> = add i32 4, %var ; yields i32:result = 4 + %var
5700 '``fadd``' Instruction
5701 ^^^^^^^^^^^^^^^^^^^^^^
5708 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5713 The '``fadd``' instruction returns the sum of its two operands.
5718 The two arguments to the '``fadd``' instruction must be :ref:`floating
5719 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5720 Both arguments must have identical types.
5725 The value produced is the floating point sum of the two operands. This
5726 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
5727 which are optimization hints to enable otherwise unsafe floating point
5733 .. code-block:: llvm
5735 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
5737 '``sub``' Instruction
5738 ^^^^^^^^^^^^^^^^^^^^^
5745 <result> = sub <ty> <op1>, <op2> ; yields ty:result
5746 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
5747 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
5748 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
5753 The '``sub``' instruction returns the difference of its two operands.
5755 Note that the '``sub``' instruction is used to represent the '``neg``'
5756 instruction present in most other intermediate representations.
5761 The two arguments to the '``sub``' instruction must be
5762 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5763 arguments must have identical types.
5768 The value produced is the integer difference of the two operands.
5770 If the difference has unsigned overflow, the result returned is the
5771 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5774 Because LLVM integers use a two's complement representation, this
5775 instruction is appropriate for both signed and unsigned integers.
5777 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5778 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5779 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
5780 unsigned and/or signed overflow, respectively, occurs.
5785 .. code-block:: llvm
5787 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
5788 <result> = sub i32 0, %val ; yields i32:result = -%var
5792 '``fsub``' Instruction
5793 ^^^^^^^^^^^^^^^^^^^^^^
5800 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5805 The '``fsub``' instruction returns the difference of its two operands.
5807 Note that the '``fsub``' instruction is used to represent the '``fneg``'
5808 instruction present in most other intermediate representations.
5813 The two arguments to the '``fsub``' instruction must be :ref:`floating
5814 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5815 Both arguments must have identical types.
5820 The value produced is the floating point difference of the two operands.
5821 This instruction can also take any number of :ref:`fast-math
5822 flags <fastmath>`, which are optimization hints to enable otherwise
5823 unsafe floating point optimizations:
5828 .. code-block:: llvm
5830 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
5831 <result> = fsub float -0.0, %val ; yields float:result = -%var
5833 '``mul``' Instruction
5834 ^^^^^^^^^^^^^^^^^^^^^
5841 <result> = mul <ty> <op1>, <op2> ; yields ty:result
5842 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
5843 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
5844 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
5849 The '``mul``' instruction returns the product of its two operands.
5854 The two arguments to the '``mul``' instruction must be
5855 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5856 arguments must have identical types.
5861 The value produced is the integer product of the two operands.
5863 If the result of the multiplication has unsigned overflow, the result
5864 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
5865 bit width of the result.
5867 Because LLVM integers use a two's complement representation, and the
5868 result is the same width as the operands, this instruction returns the
5869 correct result for both signed and unsigned integers. If a full product
5870 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
5871 sign-extended or zero-extended as appropriate to the width of the full
5874 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5875 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5876 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
5877 unsigned and/or signed overflow, respectively, occurs.
5882 .. code-block:: llvm
5884 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
5888 '``fmul``' Instruction
5889 ^^^^^^^^^^^^^^^^^^^^^^
5896 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5901 The '``fmul``' instruction returns the product of its two operands.
5906 The two arguments to the '``fmul``' instruction must be :ref:`floating
5907 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5908 Both arguments must have identical types.
5913 The value produced is the floating point product of the two operands.
5914 This instruction can also take any number of :ref:`fast-math
5915 flags <fastmath>`, which are optimization hints to enable otherwise
5916 unsafe floating point optimizations:
5921 .. code-block:: llvm
5923 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
5925 '``udiv``' Instruction
5926 ^^^^^^^^^^^^^^^^^^^^^^
5933 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
5934 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
5939 The '``udiv``' instruction returns the quotient of its two operands.
5944 The two arguments to the '``udiv``' instruction must be
5945 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5946 arguments must have identical types.
5951 The value produced is the unsigned integer quotient of the two operands.
5953 Note that unsigned integer division and signed integer division are
5954 distinct operations; for signed integer division, use '``sdiv``'.
5956 Division by zero leads to undefined behavior.
5958 If the ``exact`` keyword is present, the result value of the ``udiv`` is
5959 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
5960 such, "((a udiv exact b) mul b) == a").
5965 .. code-block:: llvm
5967 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
5969 '``sdiv``' Instruction
5970 ^^^^^^^^^^^^^^^^^^^^^^
5977 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
5978 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
5983 The '``sdiv``' instruction returns the quotient of its two operands.
5988 The two arguments to the '``sdiv``' instruction must be
5989 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5990 arguments must have identical types.
5995 The value produced is the signed integer quotient of the two operands
5996 rounded towards zero.
5998 Note that signed integer division and unsigned integer division are
5999 distinct operations; for unsigned integer division, use '``udiv``'.
6001 Division by zero leads to undefined behavior. Overflow also leads to
6002 undefined behavior; this is a rare case, but can occur, for example, by
6003 doing a 32-bit division of -2147483648 by -1.
6005 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
6006 a :ref:`poison value <poisonvalues>` if the result would be rounded.
6011 .. code-block:: llvm
6013 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
6017 '``fdiv``' Instruction
6018 ^^^^^^^^^^^^^^^^^^^^^^
6025 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6030 The '``fdiv``' instruction returns the quotient of its two operands.
6035 The two arguments to the '``fdiv``' instruction must be :ref:`floating
6036 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6037 Both arguments must have identical types.
6042 The value produced is the floating point quotient of the two operands.
6043 This instruction can also take any number of :ref:`fast-math
6044 flags <fastmath>`, which are optimization hints to enable otherwise
6045 unsafe floating point optimizations:
6050 .. code-block:: llvm
6052 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
6054 '``urem``' Instruction
6055 ^^^^^^^^^^^^^^^^^^^^^^
6062 <result> = urem <ty> <op1>, <op2> ; yields ty:result
6067 The '``urem``' instruction returns the remainder from the unsigned
6068 division of its two arguments.
6073 The two arguments to the '``urem``' instruction must be
6074 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6075 arguments must have identical types.
6080 This instruction returns the unsigned integer *remainder* of a division.
6081 This instruction always performs an unsigned division to get the
6084 Note that unsigned integer remainder and signed integer remainder are
6085 distinct operations; for signed integer remainder, use '``srem``'.
6087 Taking the remainder of a division by zero leads to undefined behavior.
6092 .. code-block:: llvm
6094 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
6096 '``srem``' Instruction
6097 ^^^^^^^^^^^^^^^^^^^^^^
6104 <result> = srem <ty> <op1>, <op2> ; yields ty:result
6109 The '``srem``' instruction returns the remainder from the signed
6110 division of its two operands. This instruction can also take
6111 :ref:`vector <t_vector>` versions of the values in which case the elements
6117 The two arguments to the '``srem``' instruction must be
6118 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6119 arguments must have identical types.
6124 This instruction returns the *remainder* of a division (where the result
6125 is either zero or has the same sign as the dividend, ``op1``), not the
6126 *modulo* operator (where the result is either zero or has the same sign
6127 as the divisor, ``op2``) of a value. For more information about the
6128 difference, see `The Math
6129 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
6130 table of how this is implemented in various languages, please see
6132 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
6134 Note that signed integer remainder and unsigned integer remainder are
6135 distinct operations; for unsigned integer remainder, use '``urem``'.
6137 Taking the remainder of a division by zero leads to undefined behavior.
6138 Overflow also leads to undefined behavior; this is a rare case, but can
6139 occur, for example, by taking the remainder of a 32-bit division of
6140 -2147483648 by -1. (The remainder doesn't actually overflow, but this
6141 rule lets srem be implemented using instructions that return both the
6142 result of the division and the remainder.)
6147 .. code-block:: llvm
6149 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
6153 '``frem``' Instruction
6154 ^^^^^^^^^^^^^^^^^^^^^^
6161 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6166 The '``frem``' instruction returns the remainder from the division of
6172 The two arguments to the '``frem``' instruction must be :ref:`floating
6173 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6174 Both arguments must have identical types.
6179 This instruction returns the *remainder* of a division. The remainder
6180 has the same sign as the dividend. This instruction can also take any
6181 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
6182 to enable otherwise unsafe floating point optimizations:
6187 .. code-block:: llvm
6189 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
6193 Bitwise Binary Operations
6194 -------------------------
6196 Bitwise binary operators are used to do various forms of bit-twiddling
6197 in a program. They are generally very efficient instructions and can
6198 commonly be strength reduced from other instructions. They require two
6199 operands of the same type, execute an operation on them, and produce a
6200 single value. The resulting value is the same type as its operands.
6202 '``shl``' Instruction
6203 ^^^^^^^^^^^^^^^^^^^^^
6210 <result> = shl <ty> <op1>, <op2> ; yields ty:result
6211 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
6212 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
6213 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
6218 The '``shl``' instruction returns the first operand shifted to the left
6219 a specified number of bits.
6224 Both arguments to the '``shl``' instruction must be the same
6225 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6226 '``op2``' is treated as an unsigned value.
6231 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
6232 where ``n`` is the width of the result. If ``op2`` is (statically or
6233 dynamically) equal to or larger than the number of bits in
6234 ``op1``, the result is undefined. If the arguments are vectors, each
6235 vector element of ``op1`` is shifted by the corresponding shift amount
6238 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
6239 value <poisonvalues>` if it shifts out any non-zero bits. If the
6240 ``nsw`` keyword is present, then the shift produces a :ref:`poison
6241 value <poisonvalues>` if it shifts out any bits that disagree with the
6242 resultant sign bit. As such, NUW/NSW have the same semantics as they
6243 would if the shift were expressed as a mul instruction with the same
6244 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
6249 .. code-block:: llvm
6251 <result> = shl i32 4, %var ; yields i32: 4 << %var
6252 <result> = shl i32 4, 2 ; yields i32: 16
6253 <result> = shl i32 1, 10 ; yields i32: 1024
6254 <result> = shl i32 1, 32 ; undefined
6255 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
6257 '``lshr``' Instruction
6258 ^^^^^^^^^^^^^^^^^^^^^^
6265 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
6266 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
6271 The '``lshr``' instruction (logical shift right) returns the first
6272 operand shifted to the right a specified number of bits with zero fill.
6277 Both arguments to the '``lshr``' instruction must be the same
6278 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6279 '``op2``' is treated as an unsigned value.
6284 This instruction always performs a logical shift right operation. The
6285 most significant bits of the result will be filled with zero bits after
6286 the shift. If ``op2`` is (statically or dynamically) equal to or larger
6287 than the number of bits in ``op1``, the result is undefined. If the
6288 arguments are vectors, each vector element of ``op1`` is shifted by the
6289 corresponding shift amount in ``op2``.
6291 If the ``exact`` keyword is present, the result value of the ``lshr`` is
6292 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6298 .. code-block:: llvm
6300 <result> = lshr i32 4, 1 ; yields i32:result = 2
6301 <result> = lshr i32 4, 2 ; yields i32:result = 1
6302 <result> = lshr i8 4, 3 ; yields i8:result = 0
6303 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
6304 <result> = lshr i32 1, 32 ; undefined
6305 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
6307 '``ashr``' Instruction
6308 ^^^^^^^^^^^^^^^^^^^^^^
6315 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
6316 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
6321 The '``ashr``' instruction (arithmetic shift right) returns the first
6322 operand shifted to the right a specified number of bits with sign
6328 Both arguments to the '``ashr``' instruction must be the same
6329 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6330 '``op2``' is treated as an unsigned value.
6335 This instruction always performs an arithmetic shift right operation,
6336 The most significant bits of the result will be filled with the sign bit
6337 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
6338 than the number of bits in ``op1``, the result is undefined. If the
6339 arguments are vectors, each vector element of ``op1`` is shifted by the
6340 corresponding shift amount in ``op2``.
6342 If the ``exact`` keyword is present, the result value of the ``ashr`` is
6343 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6349 .. code-block:: llvm
6351 <result> = ashr i32 4, 1 ; yields i32:result = 2
6352 <result> = ashr i32 4, 2 ; yields i32:result = 1
6353 <result> = ashr i8 4, 3 ; yields i8:result = 0
6354 <result> = ashr i8 -2, 1 ; yields i8:result = -1
6355 <result> = ashr i32 1, 32 ; undefined
6356 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
6358 '``and``' Instruction
6359 ^^^^^^^^^^^^^^^^^^^^^
6366 <result> = and <ty> <op1>, <op2> ; yields ty:result
6371 The '``and``' instruction returns the bitwise logical and of its two
6377 The two arguments to the '``and``' instruction must be
6378 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6379 arguments must have identical types.
6384 The truth table used for the '``and``' instruction is:
6401 .. code-block:: llvm
6403 <result> = and i32 4, %var ; yields i32:result = 4 & %var
6404 <result> = and i32 15, 40 ; yields i32:result = 8
6405 <result> = and i32 4, 8 ; yields i32:result = 0
6407 '``or``' Instruction
6408 ^^^^^^^^^^^^^^^^^^^^
6415 <result> = or <ty> <op1>, <op2> ; yields ty:result
6420 The '``or``' instruction returns the bitwise logical inclusive or of its
6426 The two arguments to the '``or``' instruction must be
6427 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6428 arguments must have identical types.
6433 The truth table used for the '``or``' instruction is:
6452 <result> = or i32 4, %var ; yields i32:result = 4 | %var
6453 <result> = or i32 15, 40 ; yields i32:result = 47
6454 <result> = or i32 4, 8 ; yields i32:result = 12
6456 '``xor``' Instruction
6457 ^^^^^^^^^^^^^^^^^^^^^
6464 <result> = xor <ty> <op1>, <op2> ; yields ty:result
6469 The '``xor``' instruction returns the bitwise logical exclusive or of
6470 its two operands. The ``xor`` is used to implement the "one's
6471 complement" operation, which is the "~" operator in C.
6476 The two arguments to the '``xor``' instruction must be
6477 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6478 arguments must have identical types.
6483 The truth table used for the '``xor``' instruction is:
6500 .. code-block:: llvm
6502 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
6503 <result> = xor i32 15, 40 ; yields i32:result = 39
6504 <result> = xor i32 4, 8 ; yields i32:result = 12
6505 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
6510 LLVM supports several instructions to represent vector operations in a
6511 target-independent manner. These instructions cover the element-access
6512 and vector-specific operations needed to process vectors effectively.
6513 While LLVM does directly support these vector operations, many
6514 sophisticated algorithms will want to use target-specific intrinsics to
6515 take full advantage of a specific target.
6517 .. _i_extractelement:
6519 '``extractelement``' Instruction
6520 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6527 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
6532 The '``extractelement``' instruction extracts a single scalar element
6533 from a vector at a specified index.
6538 The first operand of an '``extractelement``' instruction is a value of
6539 :ref:`vector <t_vector>` type. The second operand is an index indicating
6540 the position from which to extract the element. The index may be a
6541 variable of any integer type.
6546 The result is a scalar of the same type as the element type of ``val``.
6547 Its value is the value at position ``idx`` of ``val``. If ``idx``
6548 exceeds the length of ``val``, the results are undefined.
6553 .. code-block:: llvm
6555 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
6557 .. _i_insertelement:
6559 '``insertelement``' Instruction
6560 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6567 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
6572 The '``insertelement``' instruction inserts a scalar element into a
6573 vector at a specified index.
6578 The first operand of an '``insertelement``' instruction is a value of
6579 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
6580 type must equal the element type of the first operand. The third operand
6581 is an index indicating the position at which to insert the value. The
6582 index may be a variable of any integer type.
6587 The result is a vector of the same type as ``val``. Its element values
6588 are those of ``val`` except at position ``idx``, where it gets the value
6589 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
6595 .. code-block:: llvm
6597 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
6599 .. _i_shufflevector:
6601 '``shufflevector``' Instruction
6602 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6609 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
6614 The '``shufflevector``' instruction constructs a permutation of elements
6615 from two input vectors, returning a vector with the same element type as
6616 the input and length that is the same as the shuffle mask.
6621 The first two operands of a '``shufflevector``' instruction are vectors
6622 with the same type. The third argument is a shuffle mask whose element
6623 type is always 'i32'. The result of the instruction is a vector whose
6624 length is the same as the shuffle mask and whose element type is the
6625 same as the element type of the first two operands.
6627 The shuffle mask operand is required to be a constant vector with either
6628 constant integer or undef values.
6633 The elements of the two input vectors are numbered from left to right
6634 across both of the vectors. The shuffle mask operand specifies, for each
6635 element of the result vector, which element of the two input vectors the
6636 result element gets. The element selector may be undef (meaning "don't
6637 care") and the second operand may be undef if performing a shuffle from
6643 .. code-block:: llvm
6645 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6646 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
6647 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
6648 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
6649 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
6650 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
6651 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6652 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
6654 Aggregate Operations
6655 --------------------
6657 LLVM supports several instructions for working with
6658 :ref:`aggregate <t_aggregate>` values.
6662 '``extractvalue``' Instruction
6663 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6670 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
6675 The '``extractvalue``' instruction extracts the value of a member field
6676 from an :ref:`aggregate <t_aggregate>` value.
6681 The first operand of an '``extractvalue``' instruction is a value of
6682 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The other operands are
6683 constant indices to specify which value to extract in a similar manner
6684 as indices in a '``getelementptr``' instruction.
6686 The major differences to ``getelementptr`` indexing are:
6688 - Since the value being indexed is not a pointer, the first index is
6689 omitted and assumed to be zero.
6690 - At least one index must be specified.
6691 - Not only struct indices but also array indices must be in bounds.
6696 The result is the value at the position in the aggregate specified by
6702 .. code-block:: llvm
6704 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
6708 '``insertvalue``' Instruction
6709 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6716 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
6721 The '``insertvalue``' instruction inserts a value into a member field in
6722 an :ref:`aggregate <t_aggregate>` value.
6727 The first operand of an '``insertvalue``' instruction is a value of
6728 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
6729 a first-class value to insert. The following operands are constant
6730 indices indicating the position at which to insert the value in a
6731 similar manner as indices in a '``extractvalue``' instruction. The value
6732 to insert must have the same type as the value identified by the
6738 The result is an aggregate of the same type as ``val``. Its value is
6739 that of ``val`` except that the value at the position specified by the
6740 indices is that of ``elt``.
6745 .. code-block:: llvm
6747 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
6748 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
6749 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
6753 Memory Access and Addressing Operations
6754 ---------------------------------------
6756 A key design point of an SSA-based representation is how it represents
6757 memory. In LLVM, no memory locations are in SSA form, which makes things
6758 very simple. This section describes how to read, write, and allocate
6763 '``alloca``' Instruction
6764 ^^^^^^^^^^^^^^^^^^^^^^^^
6771 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
6776 The '``alloca``' instruction allocates memory on the stack frame of the
6777 currently executing function, to be automatically released when this
6778 function returns to its caller. The object is always allocated in the
6779 generic address space (address space zero).
6784 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
6785 bytes of memory on the runtime stack, returning a pointer of the
6786 appropriate type to the program. If "NumElements" is specified, it is
6787 the number of elements allocated, otherwise "NumElements" is defaulted
6788 to be one. If a constant alignment is specified, the value result of the
6789 allocation is guaranteed to be aligned to at least that boundary. The
6790 alignment may not be greater than ``1 << 29``. If not specified, or if
6791 zero, the target can choose to align the allocation on any convenient
6792 boundary compatible with the type.
6794 '``type``' may be any sized type.
6799 Memory is allocated; a pointer is returned. The operation is undefined
6800 if there is insufficient stack space for the allocation. '``alloca``'d
6801 memory is automatically released when the function returns. The
6802 '``alloca``' instruction is commonly used to represent automatic
6803 variables that must have an address available. When the function returns
6804 (either with the ``ret`` or ``resume`` instructions), the memory is
6805 reclaimed. Allocating zero bytes is legal, but the result is undefined.
6806 The order in which memory is allocated (ie., which way the stack grows)
6812 .. code-block:: llvm
6814 %ptr = alloca i32 ; yields i32*:ptr
6815 %ptr = alloca i32, i32 4 ; yields i32*:ptr
6816 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
6817 %ptr = alloca i32, align 1024 ; yields i32*:ptr
6821 '``load``' Instruction
6822 ^^^^^^^^^^^^^^^^^^^^^^
6829 <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>]
6830 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>]
6831 !<index> = !{ i32 1 }
6832 !<deref_bytes_node> = !{i64 <dereferenceable_bytes>}
6833 !<align_node> = !{ i64 <value_alignment> }
6838 The '``load``' instruction is used to read from memory.
6843 The argument to the ``load`` instruction specifies the memory address
6844 from which to load. The type specified must be a :ref:`first
6845 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
6846 then the optimizer is not allowed to modify the number or order of
6847 execution of this ``load`` with other :ref:`volatile
6848 operations <volatile>`.
6850 If the ``load`` is marked as ``atomic``, it takes an extra :ref:`ordering
6851 <ordering>` and optional ``singlethread`` argument. The ``release`` and
6852 ``acq_rel`` orderings are not valid on ``load`` instructions. Atomic loads
6853 produce :ref:`defined <memmodel>` results when they may see multiple atomic
6854 stores. The type of the pointee must be an integer, pointer, or floating-point
6855 type whose bit width is a power of two greater than or equal to eight and less
6856 than or equal to a target-specific size limit. ``align`` must be explicitly
6857 specified on atomic loads, and the load has undefined behavior if the alignment
6858 is not set to a value which is at least the size in bytes of the
6859 pointee. ``!nontemporal`` does not have any defined semantics for atomic loads.
6861 The optional constant ``align`` argument specifies the alignment of the
6862 operation (that is, the alignment of the memory address). A value of 0
6863 or an omitted ``align`` argument means that the operation has the ABI
6864 alignment for the target. It is the responsibility of the code emitter
6865 to ensure that the alignment information is correct. Overestimating the
6866 alignment results in undefined behavior. Underestimating the alignment
6867 may produce less efficient code. An alignment of 1 is always safe. The
6868 maximum possible alignment is ``1 << 29``.
6870 The optional ``!nontemporal`` metadata must reference a single
6871 metadata name ``<index>`` corresponding to a metadata node with one
6872 ``i32`` entry of value 1. The existence of the ``!nontemporal``
6873 metadata on the instruction tells the optimizer and code generator
6874 that this load is not expected to be reused in the cache. The code
6875 generator may select special instructions to save cache bandwidth, such
6876 as the ``MOVNT`` instruction on x86.
6878 The optional ``!invariant.load`` metadata must reference a single
6879 metadata name ``<index>`` corresponding to a metadata node with no
6880 entries. The existence of the ``!invariant.load`` metadata on the
6881 instruction tells the optimizer and code generator that the address
6882 operand to this load points to memory which can be assumed unchanged.
6883 Being invariant does not imply that a location is dereferenceable,
6884 but it does imply that once the location is known dereferenceable
6885 its value is henceforth unchanging.
6887 The optional ``!invariant.group`` metadata must reference a single metadata name
6888 ``<index>`` corresponding to a metadata node. See ``invariant.group`` metadata.
6890 The optional ``!nonnull`` metadata must reference a single
6891 metadata name ``<index>`` corresponding to a metadata node with no
6892 entries. The existence of the ``!nonnull`` metadata on the
6893 instruction tells the optimizer that the value loaded is known to
6894 never be null. This is analogous to the ``nonnull`` attribute
6895 on parameters and return values. This metadata can only be applied
6896 to loads of a pointer type.
6898 The optional ``!dereferenceable`` metadata must reference a single metadata
6899 name ``<deref_bytes_node>`` corresponding to a metadata node with one ``i64``
6900 entry. The existence of the ``!dereferenceable`` metadata on the instruction
6901 tells the optimizer that the value loaded is known to be dereferenceable.
6902 The number of bytes known to be dereferenceable is specified by the integer
6903 value in the metadata node. This is analogous to the ''dereferenceable''
6904 attribute on parameters and return values. This metadata can only be applied
6905 to loads of a pointer type.
6907 The optional ``!dereferenceable_or_null`` metadata must reference a single
6908 metadata name ``<deref_bytes_node>`` corresponding to a metadata node with one
6909 ``i64`` entry. The existence of the ``!dereferenceable_or_null`` metadata on the
6910 instruction tells the optimizer that the value loaded is known to be either
6911 dereferenceable or null.
6912 The number of bytes known to be dereferenceable is specified by the integer
6913 value in the metadata node. This is analogous to the ''dereferenceable_or_null''
6914 attribute on parameters and return values. This metadata can only be applied
6915 to loads of a pointer type.
6917 The optional ``!align`` metadata must reference a single metadata name
6918 ``<align_node>`` corresponding to a metadata node with one ``i64`` entry.
6919 The existence of the ``!align`` metadata on the instruction tells the
6920 optimizer that the value loaded is known to be aligned to a boundary specified
6921 by the integer value in the metadata node. The alignment must be a power of 2.
6922 This is analogous to the ''align'' attribute on parameters and return values.
6923 This metadata can only be applied to loads of a pointer type.
6928 The location of memory pointed to is loaded. If the value being loaded
6929 is of scalar type then the number of bytes read does not exceed the
6930 minimum number of bytes needed to hold all bits of the type. For
6931 example, loading an ``i24`` reads at most three bytes. When loading a
6932 value of a type like ``i20`` with a size that is not an integral number
6933 of bytes, the result is undefined if the value was not originally
6934 written using a store of the same type.
6939 .. code-block:: llvm
6941 %ptr = alloca i32 ; yields i32*:ptr
6942 store i32 3, i32* %ptr ; yields void
6943 %val = load i32, i32* %ptr ; yields i32:val = i32 3
6947 '``store``' Instruction
6948 ^^^^^^^^^^^^^^^^^^^^^^^
6955 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.group !<index>] ; yields void
6956 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>] ; yields void
6961 The '``store``' instruction is used to write to memory.
6966 There are two arguments to the ``store`` instruction: a value to store
6967 and an address at which to store it. The type of the ``<pointer>``
6968 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
6969 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
6970 then the optimizer is not allowed to modify the number or order of
6971 execution of this ``store`` with other :ref:`volatile
6972 operations <volatile>`.
6974 If the ``store`` is marked as ``atomic``, it takes an extra :ref:`ordering
6975 <ordering>` and optional ``singlethread`` argument. The ``acquire`` and
6976 ``acq_rel`` orderings aren't valid on ``store`` instructions. Atomic loads
6977 produce :ref:`defined <memmodel>` results when they may see multiple atomic
6978 stores. The type of the pointee must be an integer, pointer, or floating-point
6979 type whose bit width is a power of two greater than or equal to eight and less
6980 than or equal to a target-specific size limit. ``align`` must be explicitly
6981 specified on atomic stores, and the store has undefined behavior if the
6982 alignment is not set to a value which is at least the size in bytes of the
6983 pointee. ``!nontemporal`` does not have any defined semantics for atomic stores.
6985 The optional constant ``align`` argument specifies the alignment of the
6986 operation (that is, the alignment of the memory address). A value of 0
6987 or an omitted ``align`` argument means that the operation has the ABI
6988 alignment for the target. It is the responsibility of the code emitter
6989 to ensure that the alignment information is correct. Overestimating the
6990 alignment results in undefined behavior. Underestimating the
6991 alignment may produce less efficient code. An alignment of 1 is always
6992 safe. The maximum possible alignment is ``1 << 29``.
6994 The optional ``!nontemporal`` metadata must reference a single metadata
6995 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
6996 value 1. The existence of the ``!nontemporal`` metadata on the instruction
6997 tells the optimizer and code generator that this load is not expected to
6998 be reused in the cache. The code generator may select special
6999 instructions to save cache bandwidth, such as the MOVNT instruction on
7002 The optional ``!invariant.group`` metadata must reference a
7003 single metadata name ``<index>``. See ``invariant.group`` metadata.
7008 The contents of memory are updated to contain ``<value>`` at the
7009 location specified by the ``<pointer>`` operand. If ``<value>`` is
7010 of scalar type then the number of bytes written does not exceed the
7011 minimum number of bytes needed to hold all bits of the type. For
7012 example, storing an ``i24`` writes at most three bytes. When writing a
7013 value of a type like ``i20`` with a size that is not an integral number
7014 of bytes, it is unspecified what happens to the extra bits that do not
7015 belong to the type, but they will typically be overwritten.
7020 .. code-block:: llvm
7022 %ptr = alloca i32 ; yields i32*:ptr
7023 store i32 3, i32* %ptr ; yields void
7024 %val = load i32, i32* %ptr ; yields i32:val = i32 3
7028 '``fence``' Instruction
7029 ^^^^^^^^^^^^^^^^^^^^^^^
7036 fence [singlethread] <ordering> ; yields void
7041 The '``fence``' instruction is used to introduce happens-before edges
7047 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
7048 defines what *synchronizes-with* edges they add. They can only be given
7049 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
7054 A fence A which has (at least) ``release`` ordering semantics
7055 *synchronizes with* a fence B with (at least) ``acquire`` ordering
7056 semantics if and only if there exist atomic operations X and Y, both
7057 operating on some atomic object M, such that A is sequenced before X, X
7058 modifies M (either directly or through some side effect of a sequence
7059 headed by X), Y is sequenced before B, and Y observes M. This provides a
7060 *happens-before* dependency between A and B. Rather than an explicit
7061 ``fence``, one (but not both) of the atomic operations X or Y might
7062 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
7063 still *synchronize-with* the explicit ``fence`` and establish the
7064 *happens-before* edge.
7066 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
7067 ``acquire`` and ``release`` semantics specified above, participates in
7068 the global program order of other ``seq_cst`` operations and/or fences.
7070 The optional ":ref:`singlethread <singlethread>`" argument specifies
7071 that the fence only synchronizes with other fences in the same thread.
7072 (This is useful for interacting with signal handlers.)
7077 .. code-block:: llvm
7079 fence acquire ; yields void
7080 fence singlethread seq_cst ; yields void
7084 '``cmpxchg``' Instruction
7085 ^^^^^^^^^^^^^^^^^^^^^^^^^
7092 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
7097 The '``cmpxchg``' instruction is used to atomically modify memory. It
7098 loads a value in memory and compares it to a given value. If they are
7099 equal, it tries to store a new value into the memory.
7104 There are three arguments to the '``cmpxchg``' instruction: an address
7105 to operate on, a value to compare to the value currently be at that
7106 address, and a new value to place at that address if the compared values
7107 are equal. The type of '<cmp>' must be an integer type whose bit width
7108 is a power of two greater than or equal to eight and less than or equal
7109 to a target-specific size limit. '<cmp>' and '<new>' must have the same
7110 type, and the type of '<pointer>' must be a pointer to that type. If the
7111 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
7112 to modify the number or order of execution of this ``cmpxchg`` with
7113 other :ref:`volatile operations <volatile>`.
7115 The success and failure :ref:`ordering <ordering>` arguments specify how this
7116 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
7117 must be at least ``monotonic``, the ordering constraint on failure must be no
7118 stronger than that on success, and the failure ordering cannot be either
7119 ``release`` or ``acq_rel``.
7121 The optional "``singlethread``" argument declares that the ``cmpxchg``
7122 is only atomic with respect to code (usually signal handlers) running in
7123 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
7124 respect to all other code in the system.
7126 The pointer passed into cmpxchg must have alignment greater than or
7127 equal to the size in memory of the operand.
7132 The contents of memory at the location specified by the '``<pointer>``' operand
7133 is read and compared to '``<cmp>``'; if the read value is the equal, the
7134 '``<new>``' is written. The original value at the location is returned, together
7135 with a flag indicating success (true) or failure (false).
7137 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
7138 permitted: the operation may not write ``<new>`` even if the comparison
7141 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
7142 if the value loaded equals ``cmp``.
7144 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
7145 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
7146 load with an ordering parameter determined the second ordering parameter.
7151 .. code-block:: llvm
7154 %orig = atomic load i32, i32* %ptr unordered ; yields i32
7158 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
7159 %squared = mul i32 %cmp, %cmp
7160 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
7161 %value_loaded = extractvalue { i32, i1 } %val_success, 0
7162 %success = extractvalue { i32, i1 } %val_success, 1
7163 br i1 %success, label %done, label %loop
7170 '``atomicrmw``' Instruction
7171 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7178 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
7183 The '``atomicrmw``' instruction is used to atomically modify memory.
7188 There are three arguments to the '``atomicrmw``' instruction: an
7189 operation to apply, an address whose value to modify, an argument to the
7190 operation. The operation must be one of the following keywords:
7204 The type of '<value>' must be an integer type whose bit width is a power
7205 of two greater than or equal to eight and less than or equal to a
7206 target-specific size limit. The type of the '``<pointer>``' operand must
7207 be a pointer to that type. If the ``atomicrmw`` is marked as
7208 ``volatile``, then the optimizer is not allowed to modify the number or
7209 order of execution of this ``atomicrmw`` with other :ref:`volatile
7210 operations <volatile>`.
7215 The contents of memory at the location specified by the '``<pointer>``'
7216 operand are atomically read, modified, and written back. The original
7217 value at the location is returned. The modification is specified by the
7220 - xchg: ``*ptr = val``
7221 - add: ``*ptr = *ptr + val``
7222 - sub: ``*ptr = *ptr - val``
7223 - and: ``*ptr = *ptr & val``
7224 - nand: ``*ptr = ~(*ptr & val)``
7225 - or: ``*ptr = *ptr | val``
7226 - xor: ``*ptr = *ptr ^ val``
7227 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
7228 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
7229 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
7231 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
7237 .. code-block:: llvm
7239 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
7241 .. _i_getelementptr:
7243 '``getelementptr``' Instruction
7244 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7251 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7252 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7253 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
7258 The '``getelementptr``' instruction is used to get the address of a
7259 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
7260 address calculation only and does not access memory. The instruction can also
7261 be used to calculate a vector of such addresses.
7266 The first argument is always a type used as the basis for the calculations.
7267 The second argument is always a pointer or a vector of pointers, and is the
7268 base address to start from. The remaining arguments are indices
7269 that indicate which of the elements of the aggregate object are indexed.
7270 The interpretation of each index is dependent on the type being indexed
7271 into. The first index always indexes the pointer value given as the
7272 first argument, the second index indexes a value of the type pointed to
7273 (not necessarily the value directly pointed to, since the first index
7274 can be non-zero), etc. The first type indexed into must be a pointer
7275 value, subsequent types can be arrays, vectors, and structs. Note that
7276 subsequent types being indexed into can never be pointers, since that
7277 would require loading the pointer before continuing calculation.
7279 The type of each index argument depends on the type it is indexing into.
7280 When indexing into a (optionally packed) structure, only ``i32`` integer
7281 **constants** are allowed (when using a vector of indices they must all
7282 be the **same** ``i32`` integer constant). When indexing into an array,
7283 pointer or vector, integers of any width are allowed, and they are not
7284 required to be constant. These integers are treated as signed values
7287 For example, let's consider a C code fragment and how it gets compiled
7303 int *foo(struct ST *s) {
7304 return &s[1].Z.B[5][13];
7307 The LLVM code generated by Clang is:
7309 .. code-block:: llvm
7311 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
7312 %struct.ST = type { i32, double, %struct.RT }
7314 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
7316 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
7323 In the example above, the first index is indexing into the
7324 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
7325 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
7326 indexes into the third element of the structure, yielding a
7327 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
7328 structure. The third index indexes into the second element of the
7329 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
7330 dimensions of the array are subscripted into, yielding an '``i32``'
7331 type. The '``getelementptr``' instruction returns a pointer to this
7332 element, thus computing a value of '``i32*``' type.
7334 Note that it is perfectly legal to index partially through a structure,
7335 returning a pointer to an inner element. Because of this, the LLVM code
7336 for the given testcase is equivalent to:
7338 .. code-block:: llvm
7340 define i32* @foo(%struct.ST* %s) {
7341 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
7342 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
7343 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
7344 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
7345 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
7349 If the ``inbounds`` keyword is present, the result value of the
7350 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
7351 pointer is not an *in bounds* address of an allocated object, or if any
7352 of the addresses that would be formed by successive addition of the
7353 offsets implied by the indices to the base address with infinitely
7354 precise signed arithmetic are not an *in bounds* address of that
7355 allocated object. The *in bounds* addresses for an allocated object are
7356 all the addresses that point into the object, plus the address one byte
7357 past the end. In cases where the base is a vector of pointers the
7358 ``inbounds`` keyword applies to each of the computations element-wise.
7360 If the ``inbounds`` keyword is not present, the offsets are added to the
7361 base address with silently-wrapping two's complement arithmetic. If the
7362 offsets have a different width from the pointer, they are sign-extended
7363 or truncated to the width of the pointer. The result value of the
7364 ``getelementptr`` may be outside the object pointed to by the base
7365 pointer. The result value may not necessarily be used to access memory
7366 though, even if it happens to point into allocated storage. See the
7367 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
7370 The getelementptr instruction is often confusing. For some more insight
7371 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
7376 .. code-block:: llvm
7378 ; yields [12 x i8]*:aptr
7379 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
7381 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
7383 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
7385 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
7390 The ``getelementptr`` returns a vector of pointers, instead of a single address,
7391 when one or more of its arguments is a vector. In such cases, all vector
7392 arguments should have the same number of elements, and every scalar argument
7393 will be effectively broadcast into a vector during address calculation.
7395 .. code-block:: llvm
7397 ; All arguments are vectors:
7398 ; A[i] = ptrs[i] + offsets[i]*sizeof(i8)
7399 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets
7401 ; Add the same scalar offset to each pointer of a vector:
7402 ; A[i] = ptrs[i] + offset*sizeof(i8)
7403 %A = getelementptr i8, <4 x i8*> %ptrs, i64 %offset
7405 ; Add distinct offsets to the same pointer:
7406 ; A[i] = ptr + offsets[i]*sizeof(i8)
7407 %A = getelementptr i8, i8* %ptr, <4 x i64> %offsets
7409 ; In all cases described above the type of the result is <4 x i8*>
7411 The two following instructions are equivalent:
7413 .. code-block:: llvm
7415 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7416 <4 x i32> <i32 2, i32 2, i32 2, i32 2>,
7417 <4 x i32> <i32 1, i32 1, i32 1, i32 1>,
7419 <4 x i64> <i64 13, i64 13, i64 13, i64 13>
7421 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7422 i32 2, i32 1, <4 x i32> %ind4, i64 13
7424 Let's look at the C code, where the vector version of ``getelementptr``
7429 // Let's assume that we vectorize the following loop:
7430 double *A, B; int *C;
7431 for (int i = 0; i < size; ++i) {
7435 .. code-block:: llvm
7437 ; get pointers for 8 elements from array B
7438 %ptrs = getelementptr double, double* %B, <8 x i32> %C
7439 ; load 8 elements from array B into A
7440 %A = call <8 x double> @llvm.masked.gather.v8f64(<8 x double*> %ptrs,
7441 i32 8, <8 x i1> %mask, <8 x double> %passthru)
7443 Conversion Operations
7444 ---------------------
7446 The instructions in this category are the conversion instructions
7447 (casting) which all take a single operand and a type. They perform
7448 various bit conversions on the operand.
7450 '``trunc .. to``' Instruction
7451 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7458 <result> = trunc <ty> <value> to <ty2> ; yields ty2
7463 The '``trunc``' instruction truncates its operand to the type ``ty2``.
7468 The '``trunc``' instruction takes a value to trunc, and a type to trunc
7469 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
7470 of the same number of integers. The bit size of the ``value`` must be
7471 larger than the bit size of the destination type, ``ty2``. Equal sized
7472 types are not allowed.
7477 The '``trunc``' instruction truncates the high order bits in ``value``
7478 and converts the remaining bits to ``ty2``. Since the source size must
7479 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
7480 It will always truncate bits.
7485 .. code-block:: llvm
7487 %X = trunc i32 257 to i8 ; yields i8:1
7488 %Y = trunc i32 123 to i1 ; yields i1:true
7489 %Z = trunc i32 122 to i1 ; yields i1:false
7490 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
7492 '``zext .. to``' Instruction
7493 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7500 <result> = zext <ty> <value> to <ty2> ; yields ty2
7505 The '``zext``' instruction zero extends its operand to type ``ty2``.
7510 The '``zext``' instruction takes a value to cast, and a type to cast it
7511 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7512 the same number of integers. The bit size of the ``value`` must be
7513 smaller than the bit size of the destination type, ``ty2``.
7518 The ``zext`` fills the high order bits of the ``value`` with zero bits
7519 until it reaches the size of the destination type, ``ty2``.
7521 When zero extending from i1, the result will always be either 0 or 1.
7526 .. code-block:: llvm
7528 %X = zext i32 257 to i64 ; yields i64:257
7529 %Y = zext i1 true to i32 ; yields i32:1
7530 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7532 '``sext .. to``' Instruction
7533 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7540 <result> = sext <ty> <value> to <ty2> ; yields ty2
7545 The '``sext``' sign extends ``value`` to the type ``ty2``.
7550 The '``sext``' instruction takes a value to cast, and a type to cast it
7551 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7552 the same number of integers. The bit size of the ``value`` must be
7553 smaller than the bit size of the destination type, ``ty2``.
7558 The '``sext``' instruction performs a sign extension by copying the sign
7559 bit (highest order bit) of the ``value`` until it reaches the bit size
7560 of the type ``ty2``.
7562 When sign extending from i1, the extension always results in -1 or 0.
7567 .. code-block:: llvm
7569 %X = sext i8 -1 to i16 ; yields i16 :65535
7570 %Y = sext i1 true to i32 ; yields i32:-1
7571 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7573 '``fptrunc .. to``' Instruction
7574 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7581 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
7586 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
7591 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
7592 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
7593 The size of ``value`` must be larger than the size of ``ty2``. This
7594 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
7599 The '``fptrunc``' instruction casts a ``value`` from a larger
7600 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
7601 point <t_floating>` type. If the value cannot fit (i.e. overflows) within the
7602 destination type, ``ty2``, then the results are undefined. If the cast produces
7603 an inexact result, how rounding is performed (e.g. truncation, also known as
7604 round to zero) is undefined.
7609 .. code-block:: llvm
7611 %X = fptrunc double 123.0 to float ; yields float:123.0
7612 %Y = fptrunc double 1.0E+300 to float ; yields undefined
7614 '``fpext .. to``' Instruction
7615 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7622 <result> = fpext <ty> <value> to <ty2> ; yields ty2
7627 The '``fpext``' extends a floating point ``value`` to a larger floating
7633 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
7634 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
7635 to. The source type must be smaller than the destination type.
7640 The '``fpext``' instruction extends the ``value`` from a smaller
7641 :ref:`floating point <t_floating>` type to a larger :ref:`floating
7642 point <t_floating>` type. The ``fpext`` cannot be used to make a
7643 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
7644 *no-op cast* for a floating point cast.
7649 .. code-block:: llvm
7651 %X = fpext float 3.125 to double ; yields double:3.125000e+00
7652 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
7654 '``fptoui .. to``' Instruction
7655 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7662 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
7667 The '``fptoui``' converts a floating point ``value`` to its unsigned
7668 integer equivalent of type ``ty2``.
7673 The '``fptoui``' instruction takes a value to cast, which must be a
7674 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7675 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7676 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7677 type with the same number of elements as ``ty``
7682 The '``fptoui``' instruction converts its :ref:`floating
7683 point <t_floating>` operand into the nearest (rounding towards zero)
7684 unsigned integer value. If the value cannot fit in ``ty2``, the results
7690 .. code-block:: llvm
7692 %X = fptoui double 123.0 to i32 ; yields i32:123
7693 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
7694 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
7696 '``fptosi .. to``' Instruction
7697 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7704 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
7709 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
7710 ``value`` to type ``ty2``.
7715 The '``fptosi``' instruction takes a value to cast, which must be a
7716 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7717 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7718 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7719 type with the same number of elements as ``ty``
7724 The '``fptosi``' instruction converts its :ref:`floating
7725 point <t_floating>` operand into the nearest (rounding towards zero)
7726 signed integer value. If the value cannot fit in ``ty2``, the results
7732 .. code-block:: llvm
7734 %X = fptosi double -123.0 to i32 ; yields i32:-123
7735 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
7736 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
7738 '``uitofp .. to``' Instruction
7739 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7746 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
7751 The '``uitofp``' instruction regards ``value`` as an unsigned integer
7752 and converts that value to the ``ty2`` type.
7757 The '``uitofp``' instruction takes a value to cast, which must be a
7758 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7759 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7760 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7761 type with the same number of elements as ``ty``
7766 The '``uitofp``' instruction interprets its operand as an unsigned
7767 integer quantity and converts it to the corresponding floating point
7768 value. If the value cannot fit in the floating point value, the results
7774 .. code-block:: llvm
7776 %X = uitofp i32 257 to float ; yields float:257.0
7777 %Y = uitofp i8 -1 to double ; yields double:255.0
7779 '``sitofp .. to``' Instruction
7780 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7787 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
7792 The '``sitofp``' instruction regards ``value`` as a signed integer and
7793 converts that value to the ``ty2`` type.
7798 The '``sitofp``' instruction takes a value to cast, which must be a
7799 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7800 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7801 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7802 type with the same number of elements as ``ty``
7807 The '``sitofp``' instruction interprets its operand as a signed integer
7808 quantity and converts it to the corresponding floating point value. If
7809 the value cannot fit in the floating point value, the results are
7815 .. code-block:: llvm
7817 %X = sitofp i32 257 to float ; yields float:257.0
7818 %Y = sitofp i8 -1 to double ; yields double:-1.0
7822 '``ptrtoint .. to``' Instruction
7823 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7830 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
7835 The '``ptrtoint``' instruction converts the pointer or a vector of
7836 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
7841 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
7842 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
7843 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
7844 a vector of integers type.
7849 The '``ptrtoint``' instruction converts ``value`` to integer type
7850 ``ty2`` by interpreting the pointer value as an integer and either
7851 truncating or zero extending that value to the size of the integer type.
7852 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
7853 ``value`` is larger than ``ty2`` then a truncation is done. If they are
7854 the same size, then nothing is done (*no-op cast*) other than a type
7860 .. code-block:: llvm
7862 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
7863 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
7864 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
7868 '``inttoptr .. to``' Instruction
7869 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7876 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
7881 The '``inttoptr``' instruction converts an integer ``value`` to a
7882 pointer type, ``ty2``.
7887 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
7888 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
7894 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
7895 applying either a zero extension or a truncation depending on the size
7896 of the integer ``value``. If ``value`` is larger than the size of a
7897 pointer then a truncation is done. If ``value`` is smaller than the size
7898 of a pointer then a zero extension is done. If they are the same size,
7899 nothing is done (*no-op cast*).
7904 .. code-block:: llvm
7906 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
7907 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
7908 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
7909 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
7913 '``bitcast .. to``' Instruction
7914 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7921 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
7926 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
7932 The '``bitcast``' instruction takes a value to cast, which must be a
7933 non-aggregate first class value, and a type to cast it to, which must
7934 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
7935 bit sizes of ``value`` and the destination type, ``ty2``, must be
7936 identical. If the source type is a pointer, the destination type must
7937 also be a pointer of the same size. This instruction supports bitwise
7938 conversion of vectors to integers and to vectors of other types (as
7939 long as they have the same size).
7944 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
7945 is always a *no-op cast* because no bits change with this
7946 conversion. The conversion is done as if the ``value`` had been stored
7947 to memory and read back as type ``ty2``. Pointer (or vector of
7948 pointers) types may only be converted to other pointer (or vector of
7949 pointers) types with the same address space through this instruction.
7950 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
7951 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
7956 .. code-block:: llvm
7958 %X = bitcast i8 255 to i8 ; yields i8 :-1
7959 %Y = bitcast i32* %x to sint* ; yields sint*:%x
7960 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
7961 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
7963 .. _i_addrspacecast:
7965 '``addrspacecast .. to``' Instruction
7966 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7973 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
7978 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
7979 address space ``n`` to type ``pty2`` in address space ``m``.
7984 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
7985 to cast and a pointer type to cast it to, which must have a different
7991 The '``addrspacecast``' instruction converts the pointer value
7992 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
7993 value modification, depending on the target and the address space
7994 pair. Pointer conversions within the same address space must be
7995 performed with the ``bitcast`` instruction. Note that if the address space
7996 conversion is legal then both result and operand refer to the same memory
8002 .. code-block:: llvm
8004 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
8005 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
8006 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
8013 The instructions in this category are the "miscellaneous" instructions,
8014 which defy better classification.
8018 '``icmp``' Instruction
8019 ^^^^^^^^^^^^^^^^^^^^^^
8026 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8031 The '``icmp``' instruction returns a boolean value or a vector of
8032 boolean values based on comparison of its two integer, integer vector,
8033 pointer, or pointer vector operands.
8038 The '``icmp``' instruction takes three operands. The first operand is
8039 the condition code indicating the kind of comparison to perform. It is
8040 not a value, just a keyword. The possible condition code are:
8043 #. ``ne``: not equal
8044 #. ``ugt``: unsigned greater than
8045 #. ``uge``: unsigned greater or equal
8046 #. ``ult``: unsigned less than
8047 #. ``ule``: unsigned less or equal
8048 #. ``sgt``: signed greater than
8049 #. ``sge``: signed greater or equal
8050 #. ``slt``: signed less than
8051 #. ``sle``: signed less or equal
8053 The remaining two arguments must be :ref:`integer <t_integer>` or
8054 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
8055 must also be identical types.
8060 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
8061 code given as ``cond``. The comparison performed always yields either an
8062 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
8064 #. ``eq``: yields ``true`` if the operands are equal, ``false``
8065 otherwise. No sign interpretation is necessary or performed.
8066 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
8067 otherwise. No sign interpretation is necessary or performed.
8068 #. ``ugt``: interprets the operands as unsigned values and yields
8069 ``true`` if ``op1`` is greater than ``op2``.
8070 #. ``uge``: interprets the operands as unsigned values and yields
8071 ``true`` if ``op1`` is greater than or equal to ``op2``.
8072 #. ``ult``: interprets the operands as unsigned values and yields
8073 ``true`` if ``op1`` is less than ``op2``.
8074 #. ``ule``: interprets the operands as unsigned values and yields
8075 ``true`` if ``op1`` is less than or equal to ``op2``.
8076 #. ``sgt``: interprets the operands as signed values and yields ``true``
8077 if ``op1`` is greater than ``op2``.
8078 #. ``sge``: interprets the operands as signed values and yields ``true``
8079 if ``op1`` is greater than or equal to ``op2``.
8080 #. ``slt``: interprets the operands as signed values and yields ``true``
8081 if ``op1`` is less than ``op2``.
8082 #. ``sle``: interprets the operands as signed values and yields ``true``
8083 if ``op1`` is less than or equal to ``op2``.
8085 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
8086 are compared as if they were integers.
8088 If the operands are integer vectors, then they are compared element by
8089 element. The result is an ``i1`` vector with the same number of elements
8090 as the values being compared. Otherwise, the result is an ``i1``.
8095 .. code-block:: llvm
8097 <result> = icmp eq i32 4, 5 ; yields: result=false
8098 <result> = icmp ne float* %X, %X ; yields: result=false
8099 <result> = icmp ult i16 4, 5 ; yields: result=true
8100 <result> = icmp sgt i16 4, 5 ; yields: result=false
8101 <result> = icmp ule i16 -4, 5 ; yields: result=false
8102 <result> = icmp sge i16 4, 5 ; yields: result=false
8104 Note that the code generator does not yet support vector types with the
8105 ``icmp`` instruction.
8109 '``fcmp``' Instruction
8110 ^^^^^^^^^^^^^^^^^^^^^^
8117 <result> = fcmp [fast-math flags]* <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8122 The '``fcmp``' instruction returns a boolean value or vector of boolean
8123 values based on comparison of its operands.
8125 If the operands are floating point scalars, then the result type is a
8126 boolean (:ref:`i1 <t_integer>`).
8128 If the operands are floating point vectors, then the result type is a
8129 vector of boolean with the same number of elements as the operands being
8135 The '``fcmp``' instruction takes three operands. The first operand is
8136 the condition code indicating the kind of comparison to perform. It is
8137 not a value, just a keyword. The possible condition code are:
8139 #. ``false``: no comparison, always returns false
8140 #. ``oeq``: ordered and equal
8141 #. ``ogt``: ordered and greater than
8142 #. ``oge``: ordered and greater than or equal
8143 #. ``olt``: ordered and less than
8144 #. ``ole``: ordered and less than or equal
8145 #. ``one``: ordered and not equal
8146 #. ``ord``: ordered (no nans)
8147 #. ``ueq``: unordered or equal
8148 #. ``ugt``: unordered or greater than
8149 #. ``uge``: unordered or greater than or equal
8150 #. ``ult``: unordered or less than
8151 #. ``ule``: unordered or less than or equal
8152 #. ``une``: unordered or not equal
8153 #. ``uno``: unordered (either nans)
8154 #. ``true``: no comparison, always returns true
8156 *Ordered* means that neither operand is a QNAN while *unordered* means
8157 that either operand may be a QNAN.
8159 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
8160 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
8161 type. They must have identical types.
8166 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
8167 condition code given as ``cond``. If the operands are vectors, then the
8168 vectors are compared element by element. Each comparison performed
8169 always yields an :ref:`i1 <t_integer>` result, as follows:
8171 #. ``false``: always yields ``false``, regardless of operands.
8172 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
8173 is equal to ``op2``.
8174 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
8175 is greater than ``op2``.
8176 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
8177 is greater than or equal to ``op2``.
8178 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
8179 is less than ``op2``.
8180 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
8181 is less than or equal to ``op2``.
8182 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
8183 is not equal to ``op2``.
8184 #. ``ord``: yields ``true`` if both operands are not a QNAN.
8185 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
8187 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
8188 greater than ``op2``.
8189 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
8190 greater than or equal to ``op2``.
8191 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
8193 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
8194 less than or equal to ``op2``.
8195 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
8196 not equal to ``op2``.
8197 #. ``uno``: yields ``true`` if either operand is a QNAN.
8198 #. ``true``: always yields ``true``, regardless of operands.
8200 The ``fcmp`` instruction can also optionally take any number of
8201 :ref:`fast-math flags <fastmath>`, which are optimization hints to enable
8202 otherwise unsafe floating point optimizations.
8204 Any set of fast-math flags are legal on an ``fcmp`` instruction, but the
8205 only flags that have any effect on its semantics are those that allow
8206 assumptions to be made about the values of input arguments; namely
8207 ``nnan``, ``ninf``, and ``nsz``. See :ref:`fastmath` for more information.
8212 .. code-block:: llvm
8214 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
8215 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
8216 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
8217 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
8219 Note that the code generator does not yet support vector types with the
8220 ``fcmp`` instruction.
8224 '``phi``' Instruction
8225 ^^^^^^^^^^^^^^^^^^^^^
8232 <result> = phi <ty> [ <val0>, <label0>], ...
8237 The '``phi``' instruction is used to implement the φ node in the SSA
8238 graph representing the function.
8243 The type of the incoming values is specified with the first type field.
8244 After this, the '``phi``' instruction takes a list of pairs as
8245 arguments, with one pair for each predecessor basic block of the current
8246 block. Only values of :ref:`first class <t_firstclass>` type may be used as
8247 the value arguments to the PHI node. Only labels may be used as the
8250 There must be no non-phi instructions between the start of a basic block
8251 and the PHI instructions: i.e. PHI instructions must be first in a basic
8254 For the purposes of the SSA form, the use of each incoming value is
8255 deemed to occur on the edge from the corresponding predecessor block to
8256 the current block (but after any definition of an '``invoke``'
8257 instruction's return value on the same edge).
8262 At runtime, the '``phi``' instruction logically takes on the value
8263 specified by the pair corresponding to the predecessor basic block that
8264 executed just prior to the current block.
8269 .. code-block:: llvm
8271 Loop: ; Infinite loop that counts from 0 on up...
8272 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
8273 %nextindvar = add i32 %indvar, 1
8278 '``select``' Instruction
8279 ^^^^^^^^^^^^^^^^^^^^^^^^
8286 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
8288 selty is either i1 or {<N x i1>}
8293 The '``select``' instruction is used to choose one value based on a
8294 condition, without IR-level branching.
8299 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
8300 values indicating the condition, and two values of the same :ref:`first
8301 class <t_firstclass>` type.
8306 If the condition is an i1 and it evaluates to 1, the instruction returns
8307 the first value argument; otherwise, it returns the second value
8310 If the condition is a vector of i1, then the value arguments must be
8311 vectors of the same size, and the selection is done element by element.
8313 If the condition is an i1 and the value arguments are vectors of the
8314 same size, then an entire vector is selected.
8319 .. code-block:: llvm
8321 %X = select i1 true, i8 17, i8 42 ; yields i8:17
8325 '``call``' Instruction
8326 ^^^^^^^^^^^^^^^^^^^^^^
8333 <result> = [tail | musttail | notail ] call [fast-math flags] [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
8339 The '``call``' instruction represents a simple function call.
8344 This instruction requires several arguments:
8346 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
8347 should perform tail call optimization. The ``tail`` marker is a hint that
8348 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
8349 means that the call must be tail call optimized in order for the program to
8350 be correct. The ``musttail`` marker provides these guarantees:
8352 #. The call will not cause unbounded stack growth if it is part of a
8353 recursive cycle in the call graph.
8354 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
8357 Both markers imply that the callee does not access allocas or varargs from
8358 the caller. Calls marked ``musttail`` must obey the following additional
8361 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
8362 or a pointer bitcast followed by a ret instruction.
8363 - The ret instruction must return the (possibly bitcasted) value
8364 produced by the call or void.
8365 - The caller and callee prototypes must match. Pointer types of
8366 parameters or return types may differ in pointee type, but not
8368 - The calling conventions of the caller and callee must match.
8369 - All ABI-impacting function attributes, such as sret, byval, inreg,
8370 returned, and inalloca, must match.
8371 - The callee must be varargs iff the caller is varargs. Bitcasting a
8372 non-varargs function to the appropriate varargs type is legal so
8373 long as the non-varargs prefixes obey the other rules.
8375 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
8376 the following conditions are met:
8378 - Caller and callee both have the calling convention ``fastcc``.
8379 - The call is in tail position (ret immediately follows call and ret
8380 uses value of call or is void).
8381 - Option ``-tailcallopt`` is enabled, or
8382 ``llvm::GuaranteedTailCallOpt`` is ``true``.
8383 - `Platform-specific constraints are
8384 met. <CodeGenerator.html#tailcallopt>`_
8386 #. The optional ``notail`` marker indicates that the optimizers should not add
8387 ``tail`` or ``musttail`` markers to the call. It is used to prevent tail
8388 call optimization from being performed on the call.
8390 #. The optional ``fast-math flags`` marker indicates that the call has one or more
8391 :ref:`fast-math flags <fastmath>`, which are optimization hints to enable
8392 otherwise unsafe floating-point optimizations. Fast-math flags are only valid
8393 for calls that return a floating-point scalar or vector type.
8395 #. The optional "cconv" marker indicates which :ref:`calling
8396 convention <callingconv>` the call should use. If none is
8397 specified, the call defaults to using C calling conventions. The
8398 calling convention of the call must match the calling convention of
8399 the target function, or else the behavior is undefined.
8400 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
8401 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
8403 #. '``ty``': the type of the call instruction itself which is also the
8404 type of the return value. Functions that return no value are marked
8406 #. '``fnty``': shall be the signature of the pointer to function value
8407 being invoked. The argument types must match the types implied by
8408 this signature. This type can be omitted if the function is not
8409 varargs and if the function type does not return a pointer to a
8411 #. '``fnptrval``': An LLVM value containing a pointer to a function to
8412 be invoked. In most cases, this is a direct function invocation, but
8413 indirect ``call``'s are just as possible, calling an arbitrary pointer
8415 #. '``function args``': argument list whose types match the function
8416 signature argument types and parameter attributes. All arguments must
8417 be of :ref:`first class <t_firstclass>` type. If the function signature
8418 indicates the function accepts a variable number of arguments, the
8419 extra arguments can be specified.
8420 #. The optional :ref:`function attributes <fnattrs>` list. Only
8421 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
8422 attributes are valid here.
8423 #. The optional :ref:`operand bundles <opbundles>` list.
8428 The '``call``' instruction is used to cause control flow to transfer to
8429 a specified function, with its incoming arguments bound to the specified
8430 values. Upon a '``ret``' instruction in the called function, control
8431 flow continues with the instruction after the function call, and the
8432 return value of the function is bound to the result argument.
8437 .. code-block:: llvm
8439 %retval = call i32 @test(i32 %argc)
8440 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
8441 %X = tail call i32 @foo() ; yields i32
8442 %Y = tail call fastcc i32 @foo() ; yields i32
8443 call void %foo(i8 97 signext)
8445 %struct.A = type { i32, i8 }
8446 %r = call %struct.A @foo() ; yields { i32, i8 }
8447 %gr = extractvalue %struct.A %r, 0 ; yields i32
8448 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
8449 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
8450 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
8452 llvm treats calls to some functions with names and arguments that match
8453 the standard C99 library as being the C99 library functions, and may
8454 perform optimizations or generate code for them under that assumption.
8455 This is something we'd like to change in the future to provide better
8456 support for freestanding environments and non-C-based languages.
8460 '``va_arg``' Instruction
8461 ^^^^^^^^^^^^^^^^^^^^^^^^
8468 <resultval> = va_arg <va_list*> <arglist>, <argty>
8473 The '``va_arg``' instruction is used to access arguments passed through
8474 the "variable argument" area of a function call. It is used to implement
8475 the ``va_arg`` macro in C.
8480 This instruction takes a ``va_list*`` value and the type of the
8481 argument. It returns a value of the specified argument type and
8482 increments the ``va_list`` to point to the next argument. The actual
8483 type of ``va_list`` is target specific.
8488 The '``va_arg``' instruction loads an argument of the specified type
8489 from the specified ``va_list`` and causes the ``va_list`` to point to
8490 the next argument. For more information, see the variable argument
8491 handling :ref:`Intrinsic Functions <int_varargs>`.
8493 It is legal for this instruction to be called in a function which does
8494 not take a variable number of arguments, for example, the ``vfprintf``
8497 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
8498 function <intrinsics>` because it takes a type as an argument.
8503 See the :ref:`variable argument processing <int_varargs>` section.
8505 Note that the code generator does not yet fully support va\_arg on many
8506 targets. Also, it does not currently support va\_arg with aggregate
8507 types on any target.
8511 '``landingpad``' Instruction
8512 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8519 <resultval> = landingpad <resultty> <clause>+
8520 <resultval> = landingpad <resultty> cleanup <clause>*
8522 <clause> := catch <type> <value>
8523 <clause> := filter <array constant type> <array constant>
8528 The '``landingpad``' instruction is used by `LLVM's exception handling
8529 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8530 is a landing pad --- one where the exception lands, and corresponds to the
8531 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
8532 defines values supplied by the :ref:`personality function <personalityfn>` upon
8533 re-entry to the function. The ``resultval`` has the type ``resultty``.
8539 ``cleanup`` flag indicates that the landing pad block is a cleanup.
8541 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
8542 contains the global variable representing the "type" that may be caught
8543 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
8544 clause takes an array constant as its argument. Use
8545 "``[0 x i8**] undef``" for a filter which cannot throw. The
8546 '``landingpad``' instruction must contain *at least* one ``clause`` or
8547 the ``cleanup`` flag.
8552 The '``landingpad``' instruction defines the values which are set by the
8553 :ref:`personality function <personalityfn>` upon re-entry to the function, and
8554 therefore the "result type" of the ``landingpad`` instruction. As with
8555 calling conventions, how the personality function results are
8556 represented in LLVM IR is target specific.
8558 The clauses are applied in order from top to bottom. If two
8559 ``landingpad`` instructions are merged together through inlining, the
8560 clauses from the calling function are appended to the list of clauses.
8561 When the call stack is being unwound due to an exception being thrown,
8562 the exception is compared against each ``clause`` in turn. If it doesn't
8563 match any of the clauses, and the ``cleanup`` flag is not set, then
8564 unwinding continues further up the call stack.
8566 The ``landingpad`` instruction has several restrictions:
8568 - A landing pad block is a basic block which is the unwind destination
8569 of an '``invoke``' instruction.
8570 - A landing pad block must have a '``landingpad``' instruction as its
8571 first non-PHI instruction.
8572 - There can be only one '``landingpad``' instruction within the landing
8574 - A basic block that is not a landing pad block may not include a
8575 '``landingpad``' instruction.
8580 .. code-block:: llvm
8582 ;; A landing pad which can catch an integer.
8583 %res = landingpad { i8*, i32 }
8585 ;; A landing pad that is a cleanup.
8586 %res = landingpad { i8*, i32 }
8588 ;; A landing pad which can catch an integer and can only throw a double.
8589 %res = landingpad { i8*, i32 }
8591 filter [1 x i8**] [@_ZTId]
8595 '``cleanuppad``' Instruction
8596 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8603 <resultval> = cleanuppad within <parent> [<args>*]
8608 The '``cleanuppad``' instruction is used by `LLVM's exception handling
8609 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8610 is a cleanup block --- one where a personality routine attempts to
8611 transfer control to run cleanup actions.
8612 The ``args`` correspond to whatever additional
8613 information the :ref:`personality function <personalityfn>` requires to
8614 execute the cleanup.
8615 The ``resultval`` has the type :ref:`token <t_token>` and is used to
8616 match the ``cleanuppad`` to corresponding :ref:`cleanuprets <i_cleanupret>`.
8617 The ``parent`` argument is the token of the funclet that contains the
8618 ``cleanuppad`` instruction. If the ``cleanuppad`` is not inside a funclet,
8619 this operand may be the token ``none``.
8624 The instruction takes a list of arbitrary values which are interpreted
8625 by the :ref:`personality function <personalityfn>`.
8630 When the call stack is being unwound due to an exception being thrown,
8631 the :ref:`personality function <personalityfn>` transfers control to the
8632 ``cleanuppad`` with the aid of the personality-specific arguments.
8633 As with calling conventions, how the personality function results are
8634 represented in LLVM IR is target specific.
8636 The ``cleanuppad`` instruction has several restrictions:
8638 - A cleanup block is a basic block which is the unwind destination of
8639 an exceptional instruction.
8640 - A cleanup block must have a '``cleanuppad``' instruction as its
8641 first non-PHI instruction.
8642 - There can be only one '``cleanuppad``' instruction within the
8644 - A basic block that is not a cleanup block may not include a
8645 '``cleanuppad``' instruction.
8647 Executing a ``cleanuppad`` instruction constitutes "entering" that pad.
8648 The pad may then be "exited" in one of three ways:
8650 1) explicitly via a ``cleanupret`` that consumes it. Executing such a ``cleanupret``
8651 is undefined behavior if any descendant pads have been entered but not yet
8653 2) implicitly via a call (which unwinds all the way to the current function's caller),
8654 or via a ``catchswitch`` or a ``cleanupret`` that unwinds to caller.
8655 3) implicitly via an unwind edge whose destination EH pad isn't a descendant of
8656 the ``cleanuppad``. When the ``cleanuppad`` is exited in this manner, it is
8657 undefined behavior if the destination EH pad has a parent which is not an
8658 ancestor of the ``cleanuppad`` being exited.
8660 It is undefined behavior for the ``cleanuppad`` to exit via an unwind edge which
8661 does not transitively unwind to the same destination as a constituent
8667 .. code-block:: llvm
8669 %tok = cleanuppad within %cs []
8676 LLVM supports the notion of an "intrinsic function". These functions
8677 have well known names and semantics and are required to follow certain
8678 restrictions. Overall, these intrinsics represent an extension mechanism
8679 for the LLVM language that does not require changing all of the
8680 transformations in LLVM when adding to the language (or the bitcode
8681 reader/writer, the parser, etc...).
8683 Intrinsic function names must all start with an "``llvm.``" prefix. This
8684 prefix is reserved in LLVM for intrinsic names; thus, function names may
8685 not begin with this prefix. Intrinsic functions must always be external
8686 functions: you cannot define the body of intrinsic functions. Intrinsic
8687 functions may only be used in call or invoke instructions: it is illegal
8688 to take the address of an intrinsic function. Additionally, because
8689 intrinsic functions are part of the LLVM language, it is required if any
8690 are added that they be documented here.
8692 Some intrinsic functions can be overloaded, i.e., the intrinsic
8693 represents a family of functions that perform the same operation but on
8694 different data types. Because LLVM can represent over 8 million
8695 different integer types, overloading is used commonly to allow an
8696 intrinsic function to operate on any integer type. One or more of the
8697 argument types or the result type can be overloaded to accept any
8698 integer type. Argument types may also be defined as exactly matching a
8699 previous argument's type or the result type. This allows an intrinsic
8700 function which accepts multiple arguments, but needs all of them to be
8701 of the same type, to only be overloaded with respect to a single
8702 argument or the result.
8704 Overloaded intrinsics will have the names of its overloaded argument
8705 types encoded into its function name, each preceded by a period. Only
8706 those types which are overloaded result in a name suffix. Arguments
8707 whose type is matched against another type do not. For example, the
8708 ``llvm.ctpop`` function can take an integer of any width and returns an
8709 integer of exactly the same integer width. This leads to a family of
8710 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
8711 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
8712 overloaded, and only one type suffix is required. Because the argument's
8713 type is matched against the return type, it does not require its own
8716 To learn how to add an intrinsic function, please see the `Extending
8717 LLVM Guide <ExtendingLLVM.html>`_.
8721 Variable Argument Handling Intrinsics
8722 -------------------------------------
8724 Variable argument support is defined in LLVM with the
8725 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
8726 functions. These functions are related to the similarly named macros
8727 defined in the ``<stdarg.h>`` header file.
8729 All of these functions operate on arguments that use a target-specific
8730 value type "``va_list``". The LLVM assembly language reference manual
8731 does not define what this type is, so all transformations should be
8732 prepared to handle these functions regardless of the type used.
8734 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
8735 variable argument handling intrinsic functions are used.
8737 .. code-block:: llvm
8739 ; This struct is different for every platform. For most platforms,
8740 ; it is merely an i8*.
8741 %struct.va_list = type { i8* }
8743 ; For Unix x86_64 platforms, va_list is the following struct:
8744 ; %struct.va_list = type { i32, i32, i8*, i8* }
8746 define i32 @test(i32 %X, ...) {
8747 ; Initialize variable argument processing
8748 %ap = alloca %struct.va_list
8749 %ap2 = bitcast %struct.va_list* %ap to i8*
8750 call void @llvm.va_start(i8* %ap2)
8752 ; Read a single integer argument
8753 %tmp = va_arg i8* %ap2, i32
8755 ; Demonstrate usage of llvm.va_copy and llvm.va_end
8757 %aq2 = bitcast i8** %aq to i8*
8758 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
8759 call void @llvm.va_end(i8* %aq2)
8761 ; Stop processing of arguments.
8762 call void @llvm.va_end(i8* %ap2)
8766 declare void @llvm.va_start(i8*)
8767 declare void @llvm.va_copy(i8*, i8*)
8768 declare void @llvm.va_end(i8*)
8772 '``llvm.va_start``' Intrinsic
8773 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8780 declare void @llvm.va_start(i8* <arglist>)
8785 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
8786 subsequent use by ``va_arg``.
8791 The argument is a pointer to a ``va_list`` element to initialize.
8796 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
8797 available in C. In a target-dependent way, it initializes the
8798 ``va_list`` element to which the argument points, so that the next call
8799 to ``va_arg`` will produce the first variable argument passed to the
8800 function. Unlike the C ``va_start`` macro, this intrinsic does not need
8801 to know the last argument of the function as the compiler can figure
8804 '``llvm.va_end``' Intrinsic
8805 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8812 declare void @llvm.va_end(i8* <arglist>)
8817 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
8818 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
8823 The argument is a pointer to a ``va_list`` to destroy.
8828 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
8829 available in C. In a target-dependent way, it destroys the ``va_list``
8830 element to which the argument points. Calls to
8831 :ref:`llvm.va_start <int_va_start>` and
8832 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
8837 '``llvm.va_copy``' Intrinsic
8838 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8845 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
8850 The '``llvm.va_copy``' intrinsic copies the current argument position
8851 from the source argument list to the destination argument list.
8856 The first argument is a pointer to a ``va_list`` element to initialize.
8857 The second argument is a pointer to a ``va_list`` element to copy from.
8862 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
8863 available in C. In a target-dependent way, it copies the source
8864 ``va_list`` element into the destination ``va_list`` element. This
8865 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
8866 arbitrarily complex and require, for example, memory allocation.
8868 Accurate Garbage Collection Intrinsics
8869 --------------------------------------
8871 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
8872 (GC) requires the frontend to generate code containing appropriate intrinsic
8873 calls and select an appropriate GC strategy which knows how to lower these
8874 intrinsics in a manner which is appropriate for the target collector.
8876 These intrinsics allow identification of :ref:`GC roots on the
8877 stack <int_gcroot>`, as well as garbage collector implementations that
8878 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
8879 Frontends for type-safe garbage collected languages should generate
8880 these intrinsics to make use of the LLVM garbage collectors. For more
8881 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
8883 Experimental Statepoint Intrinsics
8884 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8886 LLVM provides an second experimental set of intrinsics for describing garbage
8887 collection safepoints in compiled code. These intrinsics are an alternative
8888 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
8889 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
8890 differences in approach are covered in the `Garbage Collection with LLVM
8891 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
8892 described in :doc:`Statepoints`.
8896 '``llvm.gcroot``' Intrinsic
8897 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8904 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
8909 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
8910 the code generator, and allows some metadata to be associated with it.
8915 The first argument specifies the address of a stack object that contains
8916 the root pointer. The second pointer (which must be either a constant or
8917 a global value address) contains the meta-data to be associated with the
8923 At runtime, a call to this intrinsic stores a null pointer into the
8924 "ptrloc" location. At compile-time, the code generator generates
8925 information to allow the runtime to find the pointer at GC safe points.
8926 The '``llvm.gcroot``' intrinsic may only be used in a function which
8927 :ref:`specifies a GC algorithm <gc>`.
8931 '``llvm.gcread``' Intrinsic
8932 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8939 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
8944 The '``llvm.gcread``' intrinsic identifies reads of references from heap
8945 locations, allowing garbage collector implementations that require read
8951 The second argument is the address to read from, which should be an
8952 address allocated from the garbage collector. The first object is a
8953 pointer to the start of the referenced object, if needed by the language
8954 runtime (otherwise null).
8959 The '``llvm.gcread``' intrinsic has the same semantics as a load
8960 instruction, but may be replaced with substantially more complex code by
8961 the garbage collector runtime, as needed. The '``llvm.gcread``'
8962 intrinsic may only be used in a function which :ref:`specifies a GC
8967 '``llvm.gcwrite``' Intrinsic
8968 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8975 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
8980 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
8981 locations, allowing garbage collector implementations that require write
8982 barriers (such as generational or reference counting collectors).
8987 The first argument is the reference to store, the second is the start of
8988 the object to store it to, and the third is the address of the field of
8989 Obj to store to. If the runtime does not require a pointer to the
8990 object, Obj may be null.
8995 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
8996 instruction, but may be replaced with substantially more complex code by
8997 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
8998 intrinsic may only be used in a function which :ref:`specifies a GC
9001 Code Generator Intrinsics
9002 -------------------------
9004 These intrinsics are provided by LLVM to expose special features that
9005 may only be implemented with code generator support.
9007 '``llvm.returnaddress``' Intrinsic
9008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9015 declare i8 *@llvm.returnaddress(i32 <level>)
9020 The '``llvm.returnaddress``' intrinsic attempts to compute a
9021 target-specific value indicating the return address of the current
9022 function or one of its callers.
9027 The argument to this intrinsic indicates which function to return the
9028 address for. Zero indicates the calling function, one indicates its
9029 caller, etc. The argument is **required** to be a constant integer
9035 The '``llvm.returnaddress``' intrinsic either returns a pointer
9036 indicating the return address of the specified call frame, or zero if it
9037 cannot be identified. The value returned by this intrinsic is likely to
9038 be incorrect or 0 for arguments other than zero, so it should only be
9039 used for debugging purposes.
9041 Note that calling this intrinsic does not prevent function inlining or
9042 other aggressive transformations, so the value returned may not be that
9043 of the obvious source-language caller.
9045 '``llvm.frameaddress``' Intrinsic
9046 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9053 declare i8* @llvm.frameaddress(i32 <level>)
9058 The '``llvm.frameaddress``' intrinsic attempts to return the
9059 target-specific frame pointer value for the specified stack frame.
9064 The argument to this intrinsic indicates which function to return the
9065 frame pointer for. Zero indicates the calling function, one indicates
9066 its caller, etc. The argument is **required** to be a constant integer
9072 The '``llvm.frameaddress``' intrinsic either returns a pointer
9073 indicating the frame address of the specified call frame, or zero if it
9074 cannot be identified. The value returned by this intrinsic is likely to
9075 be incorrect or 0 for arguments other than zero, so it should only be
9076 used for debugging purposes.
9078 Note that calling this intrinsic does not prevent function inlining or
9079 other aggressive transformations, so the value returned may not be that
9080 of the obvious source-language caller.
9082 '``llvm.localescape``' and '``llvm.localrecover``' Intrinsics
9083 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9090 declare void @llvm.localescape(...)
9091 declare i8* @llvm.localrecover(i8* %func, i8* %fp, i32 %idx)
9096 The '``llvm.localescape``' intrinsic escapes offsets of a collection of static
9097 allocas, and the '``llvm.localrecover``' intrinsic applies those offsets to a
9098 live frame pointer to recover the address of the allocation. The offset is
9099 computed during frame layout of the caller of ``llvm.localescape``.
9104 All arguments to '``llvm.localescape``' must be pointers to static allocas or
9105 casts of static allocas. Each function can only call '``llvm.localescape``'
9106 once, and it can only do so from the entry block.
9108 The ``func`` argument to '``llvm.localrecover``' must be a constant
9109 bitcasted pointer to a function defined in the current module. The code
9110 generator cannot determine the frame allocation offset of functions defined in
9113 The ``fp`` argument to '``llvm.localrecover``' must be a frame pointer of a
9114 call frame that is currently live. The return value of '``llvm.localaddress``'
9115 is one way to produce such a value, but various runtimes also expose a suitable
9116 pointer in platform-specific ways.
9118 The ``idx`` argument to '``llvm.localrecover``' indicates which alloca passed to
9119 '``llvm.localescape``' to recover. It is zero-indexed.
9124 These intrinsics allow a group of functions to share access to a set of local
9125 stack allocations of a one parent function. The parent function may call the
9126 '``llvm.localescape``' intrinsic once from the function entry block, and the
9127 child functions can use '``llvm.localrecover``' to access the escaped allocas.
9128 The '``llvm.localescape``' intrinsic blocks inlining, as inlining changes where
9129 the escaped allocas are allocated, which would break attempts to use
9130 '``llvm.localrecover``'.
9132 .. _int_read_register:
9133 .. _int_write_register:
9135 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
9136 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9143 declare i32 @llvm.read_register.i32(metadata)
9144 declare i64 @llvm.read_register.i64(metadata)
9145 declare void @llvm.write_register.i32(metadata, i32 @value)
9146 declare void @llvm.write_register.i64(metadata, i64 @value)
9152 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
9153 provides access to the named register. The register must be valid on
9154 the architecture being compiled to. The type needs to be compatible
9155 with the register being read.
9160 The '``llvm.read_register``' intrinsic returns the current value of the
9161 register, where possible. The '``llvm.write_register``' intrinsic sets
9162 the current value of the register, where possible.
9164 This is useful to implement named register global variables that need
9165 to always be mapped to a specific register, as is common practice on
9166 bare-metal programs including OS kernels.
9168 The compiler doesn't check for register availability or use of the used
9169 register in surrounding code, including inline assembly. Because of that,
9170 allocatable registers are not supported.
9172 Warning: So far it only works with the stack pointer on selected
9173 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
9174 work is needed to support other registers and even more so, allocatable
9179 '``llvm.stacksave``' Intrinsic
9180 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9187 declare i8* @llvm.stacksave()
9192 The '``llvm.stacksave``' intrinsic is used to remember the current state
9193 of the function stack, for use with
9194 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
9195 implementing language features like scoped automatic variable sized
9201 This intrinsic returns a opaque pointer value that can be passed to
9202 :ref:`llvm.stackrestore <int_stackrestore>`. When an
9203 ``llvm.stackrestore`` intrinsic is executed with a value saved from
9204 ``llvm.stacksave``, it effectively restores the state of the stack to
9205 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
9206 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
9207 were allocated after the ``llvm.stacksave`` was executed.
9209 .. _int_stackrestore:
9211 '``llvm.stackrestore``' Intrinsic
9212 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9219 declare void @llvm.stackrestore(i8* %ptr)
9224 The '``llvm.stackrestore``' intrinsic is used to restore the state of
9225 the function stack to the state it was in when the corresponding
9226 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
9227 useful for implementing language features like scoped automatic variable
9228 sized arrays in C99.
9233 See the description for :ref:`llvm.stacksave <int_stacksave>`.
9235 .. _int_get_dynamic_area_offset:
9237 '``llvm.get.dynamic.area.offset``' Intrinsic
9238 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9245 declare i32 @llvm.get.dynamic.area.offset.i32()
9246 declare i64 @llvm.get.dynamic.area.offset.i64()
9251 The '``llvm.get.dynamic.area.offset.*``' intrinsic family is used to
9252 get the offset from native stack pointer to the address of the most
9253 recent dynamic alloca on the caller's stack. These intrinsics are
9254 intendend for use in combination with
9255 :ref:`llvm.stacksave <int_stacksave>` to get a
9256 pointer to the most recent dynamic alloca. This is useful, for example,
9257 for AddressSanitizer's stack unpoisoning routines.
9262 These intrinsics return a non-negative integer value that can be used to
9263 get the address of the most recent dynamic alloca, allocated by :ref:`alloca <i_alloca>`
9264 on the caller's stack. In particular, for targets where stack grows downwards,
9265 adding this offset to the native stack pointer would get the address of the most
9266 recent dynamic alloca. For targets where stack grows upwards, the situation is a bit more
9267 complicated, because substracting this value from stack pointer would get the address
9268 one past the end of the most recent dynamic alloca.
9270 Although for most targets `llvm.get.dynamic.area.offset <int_get_dynamic_area_offset>`
9271 returns just a zero, for others, such as PowerPC and PowerPC64, it returns a
9272 compile-time-known constant value.
9274 The return value type of :ref:`llvm.get.dynamic.area.offset <int_get_dynamic_area_offset>`
9275 must match the target's generic address space's (address space 0) pointer type.
9277 '``llvm.prefetch``' Intrinsic
9278 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9285 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
9290 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
9291 insert a prefetch instruction if supported; otherwise, it is a noop.
9292 Prefetches have no effect on the behavior of the program but can change
9293 its performance characteristics.
9298 ``address`` is the address to be prefetched, ``rw`` is the specifier
9299 determining if the fetch should be for a read (0) or write (1), and
9300 ``locality`` is a temporal locality specifier ranging from (0) - no
9301 locality, to (3) - extremely local keep in cache. The ``cache type``
9302 specifies whether the prefetch is performed on the data (1) or
9303 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
9304 arguments must be constant integers.
9309 This intrinsic does not modify the behavior of the program. In
9310 particular, prefetches cannot trap and do not produce a value. On
9311 targets that support this intrinsic, the prefetch can provide hints to
9312 the processor cache for better performance.
9314 '``llvm.pcmarker``' Intrinsic
9315 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9322 declare void @llvm.pcmarker(i32 <id>)
9327 The '``llvm.pcmarker``' intrinsic is a method to export a Program
9328 Counter (PC) in a region of code to simulators and other tools. The
9329 method is target specific, but it is expected that the marker will use
9330 exported symbols to transmit the PC of the marker. The marker makes no
9331 guarantees that it will remain with any specific instruction after
9332 optimizations. It is possible that the presence of a marker will inhibit
9333 optimizations. The intended use is to be inserted after optimizations to
9334 allow correlations of simulation runs.
9339 ``id`` is a numerical id identifying the marker.
9344 This intrinsic does not modify the behavior of the program. Backends
9345 that do not support this intrinsic may ignore it.
9347 '``llvm.readcyclecounter``' Intrinsic
9348 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9355 declare i64 @llvm.readcyclecounter()
9360 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
9361 counter register (or similar low latency, high accuracy clocks) on those
9362 targets that support it. On X86, it should map to RDTSC. On Alpha, it
9363 should map to RPCC. As the backing counters overflow quickly (on the
9364 order of 9 seconds on alpha), this should only be used for small
9370 When directly supported, reading the cycle counter should not modify any
9371 memory. Implementations are allowed to either return a application
9372 specific value or a system wide value. On backends without support, this
9373 is lowered to a constant 0.
9375 Note that runtime support may be conditional on the privilege-level code is
9376 running at and the host platform.
9378 '``llvm.clear_cache``' Intrinsic
9379 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9386 declare void @llvm.clear_cache(i8*, i8*)
9391 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
9392 in the specified range to the execution unit of the processor. On
9393 targets with non-unified instruction and data cache, the implementation
9394 flushes the instruction cache.
9399 On platforms with coherent instruction and data caches (e.g. x86), this
9400 intrinsic is a nop. On platforms with non-coherent instruction and data
9401 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
9402 instructions or a system call, if cache flushing requires special
9405 The default behavior is to emit a call to ``__clear_cache`` from the run
9408 This instrinsic does *not* empty the instruction pipeline. Modifications
9409 of the current function are outside the scope of the intrinsic.
9411 '``llvm.instrprof_increment``' Intrinsic
9412 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9419 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
9420 i32 <num-counters>, i32 <index>)
9425 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
9426 frontend for use with instrumentation based profiling. These will be
9427 lowered by the ``-instrprof`` pass to generate execution counts of a
9433 The first argument is a pointer to a global variable containing the
9434 name of the entity being instrumented. This should generally be the
9435 (mangled) function name for a set of counters.
9437 The second argument is a hash value that can be used by the consumer
9438 of the profile data to detect changes to the instrumented source, and
9439 the third is the number of counters associated with ``name``. It is an
9440 error if ``hash`` or ``num-counters`` differ between two instances of
9441 ``instrprof_increment`` that refer to the same name.
9443 The last argument refers to which of the counters for ``name`` should
9444 be incremented. It should be a value between 0 and ``num-counters``.
9449 This intrinsic represents an increment of a profiling counter. It will
9450 cause the ``-instrprof`` pass to generate the appropriate data
9451 structures and the code to increment the appropriate value, in a
9452 format that can be written out by a compiler runtime and consumed via
9453 the ``llvm-profdata`` tool.
9455 '``llvm.instrprof_value_profile``' Intrinsic
9456 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9463 declare void @llvm.instrprof_value_profile(i8* <name>, i64 <hash>,
9464 i64 <value>, i32 <value_kind>,
9470 The '``llvm.instrprof_value_profile``' intrinsic can be emitted by a
9471 frontend for use with instrumentation based profiling. This will be
9472 lowered by the ``-instrprof`` pass to find out the target values,
9473 instrumented expressions take in a program at runtime.
9478 The first argument is a pointer to a global variable containing the
9479 name of the entity being instrumented. ``name`` should generally be the
9480 (mangled) function name for a set of counters.
9482 The second argument is a hash value that can be used by the consumer
9483 of the profile data to detect changes to the instrumented source. It
9484 is an error if ``hash`` differs between two instances of
9485 ``llvm.instrprof_*`` that refer to the same name.
9487 The third argument is the value of the expression being profiled. The profiled
9488 expression's value should be representable as an unsigned 64-bit value. The
9489 fourth argument represents the kind of value profiling that is being done. The
9490 supported value profiling kinds are enumerated through the
9491 ``InstrProfValueKind`` type declared in the
9492 ``<include/llvm/ProfileData/InstrProf.h>`` header file. The last argument is the
9493 index of the instrumented expression within ``name``. It should be >= 0.
9498 This intrinsic represents the point where a call to a runtime routine
9499 should be inserted for value profiling of target expressions. ``-instrprof``
9500 pass will generate the appropriate data structures and replace the
9501 ``llvm.instrprof_value_profile`` intrinsic with the call to the profile
9502 runtime library with proper arguments.
9504 Standard C Library Intrinsics
9505 -----------------------------
9507 LLVM provides intrinsics for a few important standard C library
9508 functions. These intrinsics allow source-language front-ends to pass
9509 information about the alignment of the pointer arguments to the code
9510 generator, providing opportunity for more efficient code generation.
9514 '``llvm.memcpy``' Intrinsic
9515 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9520 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
9521 integer bit width and for different address spaces. Not all targets
9522 support all bit widths however.
9526 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9527 i32 <len>, i32 <align>, i1 <isvolatile>)
9528 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9529 i64 <len>, i32 <align>, i1 <isvolatile>)
9534 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9535 source location to the destination location.
9537 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
9538 intrinsics do not return a value, takes extra alignment/isvolatile
9539 arguments and the pointers can be in specified address spaces.
9544 The first argument is a pointer to the destination, the second is a
9545 pointer to the source. The third argument is an integer argument
9546 specifying the number of bytes to copy, the fourth argument is the
9547 alignment of the source and destination locations, and the fifth is a
9548 boolean indicating a volatile access.
9550 If the call to this intrinsic has an alignment value that is not 0 or 1,
9551 then the caller guarantees that both the source and destination pointers
9552 are aligned to that boundary.
9554 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
9555 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9556 very cleanly specified and it is unwise to depend on it.
9561 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9562 source location to the destination location, which are not allowed to
9563 overlap. It copies "len" bytes of memory over. If the argument is known
9564 to be aligned to some boundary, this can be specified as the fourth
9565 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
9567 '``llvm.memmove``' Intrinsic
9568 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9573 This is an overloaded intrinsic. You can use llvm.memmove on any integer
9574 bit width and for different address space. Not all targets support all
9579 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9580 i32 <len>, i32 <align>, i1 <isvolatile>)
9581 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9582 i64 <len>, i32 <align>, i1 <isvolatile>)
9587 The '``llvm.memmove.*``' intrinsics move a block of memory from the
9588 source location to the destination location. It is similar to the
9589 '``llvm.memcpy``' intrinsic but allows the two memory locations to
9592 Note that, unlike the standard libc function, the ``llvm.memmove.*``
9593 intrinsics do not return a value, takes extra alignment/isvolatile
9594 arguments and the pointers can be in specified address spaces.
9599 The first argument is a pointer to the destination, the second is a
9600 pointer to the source. The third argument is an integer argument
9601 specifying the number of bytes to copy, the fourth argument is the
9602 alignment of the source and destination locations, and the fifth is a
9603 boolean indicating a volatile access.
9605 If the call to this intrinsic has an alignment value that is not 0 or 1,
9606 then the caller guarantees that the source and destination pointers are
9607 aligned to that boundary.
9609 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
9610 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
9611 not very cleanly specified and it is unwise to depend on it.
9616 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
9617 source location to the destination location, which may overlap. It
9618 copies "len" bytes of memory over. If the argument is known to be
9619 aligned to some boundary, this can be specified as the fourth argument,
9620 otherwise it should be set to 0 or 1 (both meaning no alignment).
9622 '``llvm.memset.*``' Intrinsics
9623 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9628 This is an overloaded intrinsic. You can use llvm.memset on any integer
9629 bit width and for different address spaces. However, not all targets
9630 support all bit widths.
9634 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
9635 i32 <len>, i32 <align>, i1 <isvolatile>)
9636 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
9637 i64 <len>, i32 <align>, i1 <isvolatile>)
9642 The '``llvm.memset.*``' intrinsics fill a block of memory with a
9643 particular byte value.
9645 Note that, unlike the standard libc function, the ``llvm.memset``
9646 intrinsic does not return a value and takes extra alignment/volatile
9647 arguments. Also, the destination can be in an arbitrary address space.
9652 The first argument is a pointer to the destination to fill, the second
9653 is the byte value with which to fill it, the third argument is an
9654 integer argument specifying the number of bytes to fill, and the fourth
9655 argument is the known alignment of the destination location.
9657 If the call to this intrinsic has an alignment value that is not 0 or 1,
9658 then the caller guarantees that the destination pointer is aligned to
9661 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
9662 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9663 very cleanly specified and it is unwise to depend on it.
9668 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
9669 at the destination location. If the argument is known to be aligned to
9670 some boundary, this can be specified as the fourth argument, otherwise
9671 it should be set to 0 or 1 (both meaning no alignment).
9673 '``llvm.sqrt.*``' Intrinsic
9674 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9679 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
9680 floating point or vector of floating point type. Not all targets support
9685 declare float @llvm.sqrt.f32(float %Val)
9686 declare double @llvm.sqrt.f64(double %Val)
9687 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
9688 declare fp128 @llvm.sqrt.f128(fp128 %Val)
9689 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
9694 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
9695 returning the same value as the libm '``sqrt``' functions would. Unlike
9696 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
9697 negative numbers other than -0.0 (which allows for better optimization,
9698 because there is no need to worry about errno being set).
9699 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
9704 The argument and return value are floating point numbers of the same
9710 This function returns the sqrt of the specified operand if it is a
9711 nonnegative floating point number.
9713 '``llvm.powi.*``' Intrinsic
9714 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9719 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
9720 floating point or vector of floating point type. Not all targets support
9725 declare float @llvm.powi.f32(float %Val, i32 %power)
9726 declare double @llvm.powi.f64(double %Val, i32 %power)
9727 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
9728 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
9729 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
9734 The '``llvm.powi.*``' intrinsics return the first operand raised to the
9735 specified (positive or negative) power. The order of evaluation of
9736 multiplications is not defined. When a vector of floating point type is
9737 used, the second argument remains a scalar integer value.
9742 The second argument is an integer power, and the first is a value to
9743 raise to that power.
9748 This function returns the first value raised to the second power with an
9749 unspecified sequence of rounding operations.
9751 '``llvm.sin.*``' Intrinsic
9752 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9757 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
9758 floating point or vector of floating point type. Not all targets support
9763 declare float @llvm.sin.f32(float %Val)
9764 declare double @llvm.sin.f64(double %Val)
9765 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
9766 declare fp128 @llvm.sin.f128(fp128 %Val)
9767 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
9772 The '``llvm.sin.*``' intrinsics return the sine of the operand.
9777 The argument and return value are floating point numbers of the same
9783 This function returns the sine of the specified operand, returning the
9784 same values as the libm ``sin`` functions would, and handles error
9785 conditions in the same way.
9787 '``llvm.cos.*``' Intrinsic
9788 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9793 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
9794 floating point or vector of floating point type. Not all targets support
9799 declare float @llvm.cos.f32(float %Val)
9800 declare double @llvm.cos.f64(double %Val)
9801 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
9802 declare fp128 @llvm.cos.f128(fp128 %Val)
9803 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
9808 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
9813 The argument and return value are floating point numbers of the same
9819 This function returns the cosine of the specified operand, returning the
9820 same values as the libm ``cos`` functions would, and handles error
9821 conditions in the same way.
9823 '``llvm.pow.*``' Intrinsic
9824 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9829 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
9830 floating point or vector of floating point type. Not all targets support
9835 declare float @llvm.pow.f32(float %Val, float %Power)
9836 declare double @llvm.pow.f64(double %Val, double %Power)
9837 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
9838 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
9839 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
9844 The '``llvm.pow.*``' intrinsics return the first operand raised to the
9845 specified (positive or negative) power.
9850 The second argument is a floating point power, and the first is a value
9851 to raise to that power.
9856 This function returns the first value raised to the second power,
9857 returning the same values as the libm ``pow`` functions would, and
9858 handles error conditions in the same way.
9860 '``llvm.exp.*``' Intrinsic
9861 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9866 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
9867 floating point or vector of floating point type. Not all targets support
9872 declare float @llvm.exp.f32(float %Val)
9873 declare double @llvm.exp.f64(double %Val)
9874 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
9875 declare fp128 @llvm.exp.f128(fp128 %Val)
9876 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
9881 The '``llvm.exp.*``' intrinsics perform the exp function.
9886 The argument and return value are floating point numbers of the same
9892 This function returns the same values as the libm ``exp`` functions
9893 would, and handles error conditions in the same way.
9895 '``llvm.exp2.*``' Intrinsic
9896 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9901 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
9902 floating point or vector of floating point type. Not all targets support
9907 declare float @llvm.exp2.f32(float %Val)
9908 declare double @llvm.exp2.f64(double %Val)
9909 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
9910 declare fp128 @llvm.exp2.f128(fp128 %Val)
9911 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
9916 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
9921 The argument and return value are floating point numbers of the same
9927 This function returns the same values as the libm ``exp2`` functions
9928 would, and handles error conditions in the same way.
9930 '``llvm.log.*``' Intrinsic
9931 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9936 This is an overloaded intrinsic. You can use ``llvm.log`` on any
9937 floating point or vector of floating point type. Not all targets support
9942 declare float @llvm.log.f32(float %Val)
9943 declare double @llvm.log.f64(double %Val)
9944 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
9945 declare fp128 @llvm.log.f128(fp128 %Val)
9946 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
9951 The '``llvm.log.*``' intrinsics perform the log function.
9956 The argument and return value are floating point numbers of the same
9962 This function returns the same values as the libm ``log`` functions
9963 would, and handles error conditions in the same way.
9965 '``llvm.log10.*``' Intrinsic
9966 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9971 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
9972 floating point or vector of floating point type. Not all targets support
9977 declare float @llvm.log10.f32(float %Val)
9978 declare double @llvm.log10.f64(double %Val)
9979 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
9980 declare fp128 @llvm.log10.f128(fp128 %Val)
9981 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
9986 The '``llvm.log10.*``' intrinsics perform the log10 function.
9991 The argument and return value are floating point numbers of the same
9997 This function returns the same values as the libm ``log10`` functions
9998 would, and handles error conditions in the same way.
10000 '``llvm.log2.*``' Intrinsic
10001 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10006 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
10007 floating point or vector of floating point type. Not all targets support
10012 declare float @llvm.log2.f32(float %Val)
10013 declare double @llvm.log2.f64(double %Val)
10014 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
10015 declare fp128 @llvm.log2.f128(fp128 %Val)
10016 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
10021 The '``llvm.log2.*``' intrinsics perform the log2 function.
10026 The argument and return value are floating point numbers of the same
10032 This function returns the same values as the libm ``log2`` functions
10033 would, and handles error conditions in the same way.
10035 '``llvm.fma.*``' Intrinsic
10036 ^^^^^^^^^^^^^^^^^^^^^^^^^^
10041 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
10042 floating point or vector of floating point type. Not all targets support
10047 declare float @llvm.fma.f32(float %a, float %b, float %c)
10048 declare double @llvm.fma.f64(double %a, double %b, double %c)
10049 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
10050 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
10051 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
10056 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
10062 The argument and return value are floating point numbers of the same
10068 This function returns the same values as the libm ``fma`` functions
10069 would, and does not set errno.
10071 '``llvm.fabs.*``' Intrinsic
10072 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10077 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
10078 floating point or vector of floating point type. Not all targets support
10083 declare float @llvm.fabs.f32(float %Val)
10084 declare double @llvm.fabs.f64(double %Val)
10085 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
10086 declare fp128 @llvm.fabs.f128(fp128 %Val)
10087 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
10092 The '``llvm.fabs.*``' intrinsics return the absolute value of the
10098 The argument and return value are floating point numbers of the same
10104 This function returns the same values as the libm ``fabs`` functions
10105 would, and handles error conditions in the same way.
10107 '``llvm.minnum.*``' Intrinsic
10108 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10113 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
10114 floating point or vector of floating point type. Not all targets support
10119 declare float @llvm.minnum.f32(float %Val0, float %Val1)
10120 declare double @llvm.minnum.f64(double %Val0, double %Val1)
10121 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10122 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
10123 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10128 The '``llvm.minnum.*``' intrinsics return the minimum of the two
10135 The arguments and return value are floating point numbers of the same
10141 Follows the IEEE-754 semantics for minNum, which also match for libm's
10144 If either operand is a NaN, returns the other non-NaN operand. Returns
10145 NaN only if both operands are NaN. If the operands compare equal,
10146 returns a value that compares equal to both operands. This means that
10147 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10149 '``llvm.maxnum.*``' Intrinsic
10150 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10155 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
10156 floating point or vector of floating point type. Not all targets support
10161 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
10162 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
10163 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10164 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
10165 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10170 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
10177 The arguments and return value are floating point numbers of the same
10182 Follows the IEEE-754 semantics for maxNum, which also match for libm's
10185 If either operand is a NaN, returns the other non-NaN operand. Returns
10186 NaN only if both operands are NaN. If the operands compare equal,
10187 returns a value that compares equal to both operands. This means that
10188 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10190 '``llvm.copysign.*``' Intrinsic
10191 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10196 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
10197 floating point or vector of floating point type. Not all targets support
10202 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
10203 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
10204 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
10205 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
10206 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
10211 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
10212 first operand and the sign of the second operand.
10217 The arguments and return value are floating point numbers of the same
10223 This function returns the same values as the libm ``copysign``
10224 functions would, and handles error conditions in the same way.
10226 '``llvm.floor.*``' Intrinsic
10227 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10232 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
10233 floating point or vector of floating point type. Not all targets support
10238 declare float @llvm.floor.f32(float %Val)
10239 declare double @llvm.floor.f64(double %Val)
10240 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
10241 declare fp128 @llvm.floor.f128(fp128 %Val)
10242 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
10247 The '``llvm.floor.*``' intrinsics return the floor of the operand.
10252 The argument and return value are floating point numbers of the same
10258 This function returns the same values as the libm ``floor`` functions
10259 would, and handles error conditions in the same way.
10261 '``llvm.ceil.*``' Intrinsic
10262 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10267 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
10268 floating point or vector of floating point type. Not all targets support
10273 declare float @llvm.ceil.f32(float %Val)
10274 declare double @llvm.ceil.f64(double %Val)
10275 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
10276 declare fp128 @llvm.ceil.f128(fp128 %Val)
10277 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
10282 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
10287 The argument and return value are floating point numbers of the same
10293 This function returns the same values as the libm ``ceil`` functions
10294 would, and handles error conditions in the same way.
10296 '``llvm.trunc.*``' Intrinsic
10297 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10302 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
10303 floating point or vector of floating point type. Not all targets support
10308 declare float @llvm.trunc.f32(float %Val)
10309 declare double @llvm.trunc.f64(double %Val)
10310 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
10311 declare fp128 @llvm.trunc.f128(fp128 %Val)
10312 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
10317 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
10318 nearest integer not larger in magnitude than the operand.
10323 The argument and return value are floating point numbers of the same
10329 This function returns the same values as the libm ``trunc`` functions
10330 would, and handles error conditions in the same way.
10332 '``llvm.rint.*``' Intrinsic
10333 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10338 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
10339 floating point or vector of floating point type. Not all targets support
10344 declare float @llvm.rint.f32(float %Val)
10345 declare double @llvm.rint.f64(double %Val)
10346 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
10347 declare fp128 @llvm.rint.f128(fp128 %Val)
10348 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
10353 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
10354 nearest integer. It may raise an inexact floating-point exception if the
10355 operand isn't an integer.
10360 The argument and return value are floating point numbers of the same
10366 This function returns the same values as the libm ``rint`` functions
10367 would, and handles error conditions in the same way.
10369 '``llvm.nearbyint.*``' Intrinsic
10370 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10375 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
10376 floating point or vector of floating point type. Not all targets support
10381 declare float @llvm.nearbyint.f32(float %Val)
10382 declare double @llvm.nearbyint.f64(double %Val)
10383 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
10384 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
10385 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
10390 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
10396 The argument and return value are floating point numbers of the same
10402 This function returns the same values as the libm ``nearbyint``
10403 functions would, and handles error conditions in the same way.
10405 '``llvm.round.*``' Intrinsic
10406 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10411 This is an overloaded intrinsic. You can use ``llvm.round`` on any
10412 floating point or vector of floating point type. Not all targets support
10417 declare float @llvm.round.f32(float %Val)
10418 declare double @llvm.round.f64(double %Val)
10419 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
10420 declare fp128 @llvm.round.f128(fp128 %Val)
10421 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
10426 The '``llvm.round.*``' intrinsics returns the operand rounded to the
10432 The argument and return value are floating point numbers of the same
10438 This function returns the same values as the libm ``round``
10439 functions would, and handles error conditions in the same way.
10441 Bit Manipulation Intrinsics
10442 ---------------------------
10444 LLVM provides intrinsics for a few important bit manipulation
10445 operations. These allow efficient code generation for some algorithms.
10447 '``llvm.bitreverse.*``' Intrinsics
10448 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10453 This is an overloaded intrinsic function. You can use bitreverse on any
10458 declare i16 @llvm.bitreverse.i16(i16 <id>)
10459 declare i32 @llvm.bitreverse.i32(i32 <id>)
10460 declare i64 @llvm.bitreverse.i64(i64 <id>)
10465 The '``llvm.bitreverse``' family of intrinsics is used to reverse the
10466 bitpattern of an integer value; for example ``0b1234567`` becomes
10472 The ``llvm.bitreverse.iN`` intrinsic returns an i16 value that has bit
10473 ``M`` in the input moved to bit ``N-M`` in the output.
10475 '``llvm.bswap.*``' Intrinsics
10476 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10481 This is an overloaded intrinsic function. You can use bswap on any
10482 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
10486 declare i16 @llvm.bswap.i16(i16 <id>)
10487 declare i32 @llvm.bswap.i32(i32 <id>)
10488 declare i64 @llvm.bswap.i64(i64 <id>)
10493 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
10494 values with an even number of bytes (positive multiple of 16 bits).
10495 These are useful for performing operations on data that is not in the
10496 target's native byte order.
10501 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
10502 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
10503 intrinsic returns an i32 value that has the four bytes of the input i32
10504 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
10505 returned i32 will have its bytes in 3, 2, 1, 0 order. The
10506 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
10507 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
10510 '``llvm.ctpop.*``' Intrinsic
10511 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10516 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
10517 bit width, or on any vector with integer elements. Not all targets
10518 support all bit widths or vector types, however.
10522 declare i8 @llvm.ctpop.i8(i8 <src>)
10523 declare i16 @llvm.ctpop.i16(i16 <src>)
10524 declare i32 @llvm.ctpop.i32(i32 <src>)
10525 declare i64 @llvm.ctpop.i64(i64 <src>)
10526 declare i256 @llvm.ctpop.i256(i256 <src>)
10527 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
10532 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
10538 The only argument is the value to be counted. The argument may be of any
10539 integer type, or a vector with integer elements. The return type must
10540 match the argument type.
10545 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
10546 each element of a vector.
10548 '``llvm.ctlz.*``' Intrinsic
10549 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10554 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
10555 integer bit width, or any vector whose elements are integers. Not all
10556 targets support all bit widths or vector types, however.
10560 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
10561 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
10562 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
10563 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
10564 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
10565 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10570 The '``llvm.ctlz``' family of intrinsic functions counts the number of
10571 leading zeros in a variable.
10576 The first argument is the value to be counted. This argument may be of
10577 any integer type, or a vector with integer element type. The return
10578 type must match the first argument type.
10580 The second argument must be a constant and is a flag to indicate whether
10581 the intrinsic should ensure that a zero as the first argument produces a
10582 defined result. Historically some architectures did not provide a
10583 defined result for zero values as efficiently, and many algorithms are
10584 now predicated on avoiding zero-value inputs.
10589 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
10590 zeros in a variable, or within each element of the vector. If
10591 ``src == 0`` then the result is the size in bits of the type of ``src``
10592 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10593 ``llvm.ctlz(i32 2) = 30``.
10595 '``llvm.cttz.*``' Intrinsic
10596 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10601 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
10602 integer bit width, or any vector of integer elements. Not all targets
10603 support all bit widths or vector types, however.
10607 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
10608 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
10609 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
10610 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
10611 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
10612 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10617 The '``llvm.cttz``' family of intrinsic functions counts the number of
10623 The first argument is the value to be counted. This argument may be of
10624 any integer type, or a vector with integer element type. The return
10625 type must match the first argument type.
10627 The second argument must be a constant and is a flag to indicate whether
10628 the intrinsic should ensure that a zero as the first argument produces a
10629 defined result. Historically some architectures did not provide a
10630 defined result for zero values as efficiently, and many algorithms are
10631 now predicated on avoiding zero-value inputs.
10636 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
10637 zeros in a variable, or within each element of a vector. If ``src == 0``
10638 then the result is the size in bits of the type of ``src`` if
10639 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10640 ``llvm.cttz(2) = 1``.
10644 Arithmetic with Overflow Intrinsics
10645 -----------------------------------
10647 LLVM provides intrinsics for some arithmetic with overflow operations.
10649 '``llvm.sadd.with.overflow.*``' Intrinsics
10650 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10655 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
10656 on any integer bit width.
10660 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
10661 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10662 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
10667 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10668 a signed addition of the two arguments, and indicate whether an overflow
10669 occurred during the signed summation.
10674 The arguments (%a and %b) and the first element of the result structure
10675 may be of integer types of any bit width, but they must have the same
10676 bit width. The second element of the result structure must be of type
10677 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10683 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10684 a signed addition of the two variables. They return a structure --- the
10685 first element of which is the signed summation, and the second element
10686 of which is a bit specifying if the signed summation resulted in an
10692 .. code-block:: llvm
10694 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10695 %sum = extractvalue {i32, i1} %res, 0
10696 %obit = extractvalue {i32, i1} %res, 1
10697 br i1 %obit, label %overflow, label %normal
10699 '``llvm.uadd.with.overflow.*``' Intrinsics
10700 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10705 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
10706 on any integer bit width.
10710 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
10711 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10712 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
10717 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10718 an unsigned addition of the two arguments, and indicate whether a carry
10719 occurred during the unsigned summation.
10724 The arguments (%a and %b) and the first element of the result structure
10725 may be of integer types of any bit width, but they must have the same
10726 bit width. The second element of the result structure must be of type
10727 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10733 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10734 an unsigned addition of the two arguments. They return a structure --- the
10735 first element of which is the sum, and the second element of which is a
10736 bit specifying if the unsigned summation resulted in a carry.
10741 .. code-block:: llvm
10743 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10744 %sum = extractvalue {i32, i1} %res, 0
10745 %obit = extractvalue {i32, i1} %res, 1
10746 br i1 %obit, label %carry, label %normal
10748 '``llvm.ssub.with.overflow.*``' Intrinsics
10749 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10754 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
10755 on any integer bit width.
10759 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
10760 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10761 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
10766 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10767 a signed subtraction of the two arguments, and indicate whether an
10768 overflow occurred during the signed subtraction.
10773 The arguments (%a and %b) and the first element of the result structure
10774 may be of integer types of any bit width, but they must have the same
10775 bit width. The second element of the result structure must be of type
10776 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10782 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10783 a signed subtraction of the two arguments. They return a structure --- the
10784 first element of which is the subtraction, and the second element of
10785 which is a bit specifying if the signed subtraction resulted in an
10791 .. code-block:: llvm
10793 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10794 %sum = extractvalue {i32, i1} %res, 0
10795 %obit = extractvalue {i32, i1} %res, 1
10796 br i1 %obit, label %overflow, label %normal
10798 '``llvm.usub.with.overflow.*``' Intrinsics
10799 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10804 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
10805 on any integer bit width.
10809 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
10810 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10811 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
10816 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10817 an unsigned subtraction of the two arguments, and indicate whether an
10818 overflow occurred during the unsigned subtraction.
10823 The arguments (%a and %b) and the first element of the result structure
10824 may be of integer types of any bit width, but they must have the same
10825 bit width. The second element of the result structure must be of type
10826 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10832 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10833 an unsigned subtraction of the two arguments. They return a structure ---
10834 the first element of which is the subtraction, and the second element of
10835 which is a bit specifying if the unsigned subtraction resulted in an
10841 .. code-block:: llvm
10843 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10844 %sum = extractvalue {i32, i1} %res, 0
10845 %obit = extractvalue {i32, i1} %res, 1
10846 br i1 %obit, label %overflow, label %normal
10848 '``llvm.smul.with.overflow.*``' Intrinsics
10849 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10854 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
10855 on any integer bit width.
10859 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
10860 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10861 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
10866 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10867 a signed multiplication of the two arguments, and indicate whether an
10868 overflow occurred during the signed multiplication.
10873 The arguments (%a and %b) and the first element of the result structure
10874 may be of integer types of any bit width, but they must have the same
10875 bit width. The second element of the result structure must be of type
10876 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10882 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10883 a signed multiplication of the two arguments. They return a structure ---
10884 the first element of which is the multiplication, and the second element
10885 of which is a bit specifying if the signed multiplication resulted in an
10891 .. code-block:: llvm
10893 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10894 %sum = extractvalue {i32, i1} %res, 0
10895 %obit = extractvalue {i32, i1} %res, 1
10896 br i1 %obit, label %overflow, label %normal
10898 '``llvm.umul.with.overflow.*``' Intrinsics
10899 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10904 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
10905 on any integer bit width.
10909 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
10910 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10911 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
10916 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10917 a unsigned multiplication of the two arguments, and indicate whether an
10918 overflow occurred during the unsigned multiplication.
10923 The arguments (%a and %b) and the first element of the result structure
10924 may be of integer types of any bit width, but they must have the same
10925 bit width. The second element of the result structure must be of type
10926 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10932 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10933 an unsigned multiplication of the two arguments. They return a structure ---
10934 the first element of which is the multiplication, and the second
10935 element of which is a bit specifying if the unsigned multiplication
10936 resulted in an overflow.
10941 .. code-block:: llvm
10943 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10944 %sum = extractvalue {i32, i1} %res, 0
10945 %obit = extractvalue {i32, i1} %res, 1
10946 br i1 %obit, label %overflow, label %normal
10948 Specialised Arithmetic Intrinsics
10949 ---------------------------------
10951 '``llvm.canonicalize.*``' Intrinsic
10952 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10959 declare float @llvm.canonicalize.f32(float %a)
10960 declare double @llvm.canonicalize.f64(double %b)
10965 The '``llvm.canonicalize.*``' intrinsic returns the platform specific canonical
10966 encoding of a floating point number. This canonicalization is useful for
10967 implementing certain numeric primitives such as frexp. The canonical encoding is
10968 defined by IEEE-754-2008 to be:
10972 2.1.8 canonical encoding: The preferred encoding of a floating-point
10973 representation in a format. Applied to declets, significands of finite
10974 numbers, infinities, and NaNs, especially in decimal formats.
10976 This operation can also be considered equivalent to the IEEE-754-2008
10977 conversion of a floating-point value to the same format. NaNs are handled
10978 according to section 6.2.
10980 Examples of non-canonical encodings:
10982 - x87 pseudo denormals, pseudo NaNs, pseudo Infinity, Unnormals. These are
10983 converted to a canonical representation per hardware-specific protocol.
10984 - Many normal decimal floating point numbers have non-canonical alternative
10986 - Some machines, like GPUs or ARMv7 NEON, do not support subnormal values.
10987 These are treated as non-canonical encodings of zero and with be flushed to
10988 a zero of the same sign by this operation.
10990 Note that per IEEE-754-2008 6.2, systems that support signaling NaNs with
10991 default exception handling must signal an invalid exception, and produce a
10994 This function should always be implementable as multiplication by 1.0, provided
10995 that the compiler does not constant fold the operation. Likewise, division by
10996 1.0 and ``llvm.minnum(x, x)`` are possible implementations. Addition with
10997 -0.0 is also sufficient provided that the rounding mode is not -Infinity.
10999 ``@llvm.canonicalize`` must preserve the equality relation. That is:
11001 - ``(@llvm.canonicalize(x) == x)`` is equivalent to ``(x == x)``
11002 - ``(@llvm.canonicalize(x) == @llvm.canonicalize(y))`` is equivalent to
11005 Additionally, the sign of zero must be conserved:
11006 ``@llvm.canonicalize(-0.0) = -0.0`` and ``@llvm.canonicalize(+0.0) = +0.0``
11008 The payload bits of a NaN must be conserved, with two exceptions.
11009 First, environments which use only a single canonical representation of NaN
11010 must perform said canonicalization. Second, SNaNs must be quieted per the
11013 The canonicalization operation may be optimized away if:
11015 - The input is known to be canonical. For example, it was produced by a
11016 floating-point operation that is required by the standard to be canonical.
11017 - The result is consumed only by (or fused with) other floating-point
11018 operations. That is, the bits of the floating point value are not examined.
11020 '``llvm.fmuladd.*``' Intrinsic
11021 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11028 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
11029 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
11034 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
11035 expressions that can be fused if the code generator determines that (a) the
11036 target instruction set has support for a fused operation, and (b) that the
11037 fused operation is more efficient than the equivalent, separate pair of mul
11038 and add instructions.
11043 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
11044 multiplicands, a and b, and an addend c.
11053 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
11055 is equivalent to the expression a \* b + c, except that rounding will
11056 not be performed between the multiplication and addition steps if the
11057 code generator fuses the operations. Fusion is not guaranteed, even if
11058 the target platform supports it. If a fused multiply-add is required the
11059 corresponding llvm.fma.\* intrinsic function should be used
11060 instead. This never sets errno, just as '``llvm.fma.*``'.
11065 .. code-block:: llvm
11067 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
11069 Half Precision Floating Point Intrinsics
11070 ----------------------------------------
11072 For most target platforms, half precision floating point is a
11073 storage-only format. This means that it is a dense encoding (in memory)
11074 but does not support computation in the format.
11076 This means that code must first load the half-precision floating point
11077 value as an i16, then convert it to float with
11078 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
11079 then be performed on the float value (including extending to double
11080 etc). To store the value back to memory, it is first converted to float
11081 if needed, then converted to i16 with
11082 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
11085 .. _int_convert_to_fp16:
11087 '``llvm.convert.to.fp16``' Intrinsic
11088 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11095 declare i16 @llvm.convert.to.fp16.f32(float %a)
11096 declare i16 @llvm.convert.to.fp16.f64(double %a)
11101 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11102 conventional floating point type to half precision floating point format.
11107 The intrinsic function contains single argument - the value to be
11113 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11114 conventional floating point format to half precision floating point format. The
11115 return value is an ``i16`` which contains the converted number.
11120 .. code-block:: llvm
11122 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
11123 store i16 %res, i16* @x, align 2
11125 .. _int_convert_from_fp16:
11127 '``llvm.convert.from.fp16``' Intrinsic
11128 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11135 declare float @llvm.convert.from.fp16.f32(i16 %a)
11136 declare double @llvm.convert.from.fp16.f64(i16 %a)
11141 The '``llvm.convert.from.fp16``' intrinsic function performs a
11142 conversion from half precision floating point format to single precision
11143 floating point format.
11148 The intrinsic function contains single argument - the value to be
11154 The '``llvm.convert.from.fp16``' intrinsic function performs a
11155 conversion from half single precision floating point format to single
11156 precision floating point format. The input half-float value is
11157 represented by an ``i16`` value.
11162 .. code-block:: llvm
11164 %a = load i16, i16* @x, align 2
11165 %res = call float @llvm.convert.from.fp16(i16 %a)
11167 .. _dbg_intrinsics:
11169 Debugger Intrinsics
11170 -------------------
11172 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
11173 prefix), are described in the `LLVM Source Level
11174 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
11177 Exception Handling Intrinsics
11178 -----------------------------
11180 The LLVM exception handling intrinsics (which all start with
11181 ``llvm.eh.`` prefix), are described in the `LLVM Exception
11182 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
11184 .. _int_trampoline:
11186 Trampoline Intrinsics
11187 ---------------------
11189 These intrinsics make it possible to excise one parameter, marked with
11190 the :ref:`nest <nest>` attribute, from a function. The result is a
11191 callable function pointer lacking the nest parameter - the caller does
11192 not need to provide a value for it. Instead, the value to use is stored
11193 in advance in a "trampoline", a block of memory usually allocated on the
11194 stack, which also contains code to splice the nest value into the
11195 argument list. This is used to implement the GCC nested function address
11198 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
11199 then the resulting function pointer has signature ``i32 (i32, i32)*``.
11200 It can be created as follows:
11202 .. code-block:: llvm
11204 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
11205 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
11206 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
11207 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
11208 %fp = bitcast i8* %p to i32 (i32, i32)*
11210 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
11211 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
11215 '``llvm.init.trampoline``' Intrinsic
11216 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11223 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
11228 This fills the memory pointed to by ``tramp`` with executable code,
11229 turning it into a trampoline.
11234 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
11235 pointers. The ``tramp`` argument must point to a sufficiently large and
11236 sufficiently aligned block of memory; this memory is written to by the
11237 intrinsic. Note that the size and the alignment are target-specific -
11238 LLVM currently provides no portable way of determining them, so a
11239 front-end that generates this intrinsic needs to have some
11240 target-specific knowledge. The ``func`` argument must hold a function
11241 bitcast to an ``i8*``.
11246 The block of memory pointed to by ``tramp`` is filled with target
11247 dependent code, turning it into a function. Then ``tramp`` needs to be
11248 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
11249 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
11250 function's signature is the same as that of ``func`` with any arguments
11251 marked with the ``nest`` attribute removed. At most one such ``nest``
11252 argument is allowed, and it must be of pointer type. Calling the new
11253 function is equivalent to calling ``func`` with the same argument list,
11254 but with ``nval`` used for the missing ``nest`` argument. If, after
11255 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
11256 modified, then the effect of any later call to the returned function
11257 pointer is undefined.
11261 '``llvm.adjust.trampoline``' Intrinsic
11262 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11269 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
11274 This performs any required machine-specific adjustment to the address of
11275 a trampoline (passed as ``tramp``).
11280 ``tramp`` must point to a block of memory which already has trampoline
11281 code filled in by a previous call to
11282 :ref:`llvm.init.trampoline <int_it>`.
11287 On some architectures the address of the code to be executed needs to be
11288 different than the address where the trampoline is actually stored. This
11289 intrinsic returns the executable address corresponding to ``tramp``
11290 after performing the required machine specific adjustments. The pointer
11291 returned can then be :ref:`bitcast and executed <int_trampoline>`.
11293 .. _int_mload_mstore:
11295 Masked Vector Load and Store Intrinsics
11296 ---------------------------------------
11298 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.
11302 '``llvm.masked.load.*``' Intrinsics
11303 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11307 This is an overloaded intrinsic. The loaded data is a vector of any integer, floating point or pointer data type.
11311 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11312 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11313 ;; The data is a vector of pointers to double
11314 declare <8 x double*> @llvm.masked.load.v8p0f64 (<8 x double*>* <ptr>, i32 <alignment>, <8 x i1> <mask>, <8 x double*> <passthru>)
11315 ;; The data is a vector of function pointers
11316 declare <8 x i32 ()*> @llvm.masked.load.v8p0f_i32f (<8 x i32 ()*>* <ptr>, i32 <alignment>, <8 x i1> <mask>, <8 x i32 ()*> <passthru>)
11321 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.
11327 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.
11333 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.
11334 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.
11339 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
11341 ;; The result of the two following instructions is identical aside from potential memory access exception
11342 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
11343 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
11347 '``llvm.masked.store.*``' Intrinsics
11348 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11352 This is an overloaded intrinsic. The data stored in memory is a vector of any integer, floating point or pointer data type.
11356 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32>* <ptr>, i32 <alignment>, <8 x i1> <mask>)
11357 declare void @llvm.masked.store.v16f32 (<16 x float> <value>, <16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
11358 ;; The data is a vector of pointers to double
11359 declare void @llvm.masked.store.v8p0f64 (<8 x double*> <value>, <8 x double*>* <ptr>, i32 <alignment>, <8 x i1> <mask>)
11360 ;; The data is a vector of function pointers
11361 declare void @llvm.masked.store.v4p0f_i32f (<4 x i32 ()*> <value>, <4 x i32 ()*>* <ptr>, i32 <alignment>, <4 x i1> <mask>)
11366 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.
11371 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.
11377 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.
11378 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.
11382 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
11384 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
11385 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
11386 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
11387 store <16 x float> %res, <16 x float>* %ptr, align 4
11390 Masked Vector Gather and Scatter Intrinsics
11391 -------------------------------------------
11393 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.
11397 '``llvm.masked.gather.*``' Intrinsics
11398 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11402 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.
11406 declare <16 x float> @llvm.masked.gather.v16f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11407 declare <2 x double> @llvm.masked.gather.v2f64 (<2 x double*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11408 declare <8 x float*> @llvm.masked.gather.v8p0f32 (<8 x float**> <ptrs>, i32 <alignment>, <8 x i1> <mask>, <8 x float*> <passthru>)
11413 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.
11419 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.
11425 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.
11426 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.
11431 %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>)
11433 ;; The gather with all-true mask is equivalent to the following instruction sequence
11434 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
11435 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
11436 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
11437 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
11439 %val0 = load double, double* %ptr0, align 8
11440 %val1 = load double, double* %ptr1, align 8
11441 %val2 = load double, double* %ptr2, align 8
11442 %val3 = load double, double* %ptr3, align 8
11444 %vec0 = insertelement <4 x double>undef, %val0, 0
11445 %vec01 = insertelement <4 x double>%vec0, %val1, 1
11446 %vec012 = insertelement <4 x double>%vec01, %val2, 2
11447 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
11451 '``llvm.masked.scatter.*``' Intrinsics
11452 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11456 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.
11460 declare void @llvm.masked.scatter.v8i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
11461 declare void @llvm.masked.scatter.v16f32 (<16 x float> <value>, <16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
11462 declare void @llvm.masked.scatter.v4p0f64 (<4 x double*> <value>, <4 x double**> <ptrs>, i32 <alignment>, <4 x i1> <mask>)
11467 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.
11472 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.
11478 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.
11482 ;; This instruction unconditionaly stores data vector in multiple addresses
11483 call @llvm.masked.scatter.v8i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
11485 ;; It is equivalent to a list of scalar stores
11486 %val0 = extractelement <8 x i32> %value, i32 0
11487 %val1 = extractelement <8 x i32> %value, i32 1
11489 %val7 = extractelement <8 x i32> %value, i32 7
11490 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
11491 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
11493 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
11494 ;; Note: the order of the following stores is important when they overlap:
11495 store i32 %val0, i32* %ptr0, align 4
11496 store i32 %val1, i32* %ptr1, align 4
11498 store i32 %val7, i32* %ptr7, align 4
11504 This class of intrinsics provides information about the lifetime of
11505 memory objects and ranges where variables are immutable.
11509 '``llvm.lifetime.start``' Intrinsic
11510 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11517 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
11522 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
11528 The first argument is a constant integer representing the size of the
11529 object, or -1 if it is variable sized. The second argument is a pointer
11535 This intrinsic indicates that before this point in the code, the value
11536 of the memory pointed to by ``ptr`` is dead. This means that it is known
11537 to never be used and has an undefined value. A load from the pointer
11538 that precedes this intrinsic can be replaced with ``'undef'``.
11542 '``llvm.lifetime.end``' Intrinsic
11543 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11550 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
11555 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
11561 The first argument is a constant integer representing the size of the
11562 object, or -1 if it is variable sized. The second argument is a pointer
11568 This intrinsic indicates that after this point in the code, the value of
11569 the memory pointed to by ``ptr`` is dead. This means that it is known to
11570 never be used and has an undefined value. Any stores into the memory
11571 object following this intrinsic may be removed as dead.
11573 '``llvm.invariant.start``' Intrinsic
11574 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11581 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
11586 The '``llvm.invariant.start``' intrinsic specifies that the contents of
11587 a memory object will not change.
11592 The first argument is a constant integer representing the size of the
11593 object, or -1 if it is variable sized. The second argument is a pointer
11599 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
11600 the return value, the referenced memory location is constant and
11603 '``llvm.invariant.end``' Intrinsic
11604 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11611 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
11616 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
11617 memory object are mutable.
11622 The first argument is the matching ``llvm.invariant.start`` intrinsic.
11623 The second argument is a constant integer representing the size of the
11624 object, or -1 if it is variable sized and the third argument is a
11625 pointer to the object.
11630 This intrinsic indicates that the memory is mutable again.
11632 '``llvm.invariant.group.barrier``' Intrinsic
11633 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11640 declare i8* @llvm.invariant.group.barrier(i8* <ptr>)
11645 The '``llvm.invariant.group.barrier``' intrinsic can be used when an invariant
11646 established by invariant.group metadata no longer holds, to obtain a new pointer
11647 value that does not carry the invariant information.
11653 The ``llvm.invariant.group.barrier`` takes only one argument, which is
11654 the pointer to the memory for which the ``invariant.group`` no longer holds.
11659 Returns another pointer that aliases its argument but which is considered different
11660 for the purposes of ``load``/``store`` ``invariant.group`` metadata.
11665 This class of intrinsics is designed to be generic and has no specific
11668 '``llvm.var.annotation``' Intrinsic
11669 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11676 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11681 The '``llvm.var.annotation``' intrinsic.
11686 The first argument is a pointer to a value, the second is a pointer to a
11687 global string, the third is a pointer to a global string which is the
11688 source file name, and the last argument is the line number.
11693 This intrinsic allows annotation of local variables with arbitrary
11694 strings. This can be useful for special purpose optimizations that want
11695 to look for these annotations. These have no other defined use; they are
11696 ignored by code generation and optimization.
11698 '``llvm.ptr.annotation.*``' Intrinsic
11699 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11704 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
11705 pointer to an integer of any width. *NOTE* you must specify an address space for
11706 the pointer. The identifier for the default address space is the integer
11711 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11712 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
11713 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
11714 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
11715 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
11720 The '``llvm.ptr.annotation``' intrinsic.
11725 The first argument is a pointer to an integer value of arbitrary bitwidth
11726 (result of some expression), the second is a pointer to a global string, the
11727 third is a pointer to a global string which is the source file name, and the
11728 last argument is the line number. It returns the value of the first argument.
11733 This intrinsic allows annotation of a pointer to an integer with arbitrary
11734 strings. This can be useful for special purpose optimizations that want to look
11735 for these annotations. These have no other defined use; they are ignored by code
11736 generation and optimization.
11738 '``llvm.annotation.*``' Intrinsic
11739 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11744 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
11745 any integer bit width.
11749 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
11750 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
11751 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
11752 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
11753 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
11758 The '``llvm.annotation``' intrinsic.
11763 The first argument is an integer value (result of some expression), the
11764 second is a pointer to a global string, the third is a pointer to a
11765 global string which is the source file name, and the last argument is
11766 the line number. It returns the value of the first argument.
11771 This intrinsic allows annotations to be put on arbitrary expressions
11772 with arbitrary strings. This can be useful for special purpose
11773 optimizations that want to look for these annotations. These have no
11774 other defined use; they are ignored by code generation and optimization.
11776 '``llvm.trap``' Intrinsic
11777 ^^^^^^^^^^^^^^^^^^^^^^^^^
11784 declare void @llvm.trap() noreturn nounwind
11789 The '``llvm.trap``' intrinsic.
11799 This intrinsic is lowered to the target dependent trap instruction. If
11800 the target does not have a trap instruction, this intrinsic will be
11801 lowered to a call of the ``abort()`` function.
11803 '``llvm.debugtrap``' Intrinsic
11804 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11811 declare void @llvm.debugtrap() nounwind
11816 The '``llvm.debugtrap``' intrinsic.
11826 This intrinsic is lowered to code which is intended to cause an
11827 execution trap with the intention of requesting the attention of a
11830 '``llvm.stackprotector``' Intrinsic
11831 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11838 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
11843 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
11844 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
11845 is placed on the stack before local variables.
11850 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
11851 The first argument is the value loaded from the stack guard
11852 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
11853 enough space to hold the value of the guard.
11858 This intrinsic causes the prologue/epilogue inserter to force the position of
11859 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
11860 to ensure that if a local variable on the stack is overwritten, it will destroy
11861 the value of the guard. When the function exits, the guard on the stack is
11862 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
11863 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
11864 calling the ``__stack_chk_fail()`` function.
11866 '``llvm.stackprotectorcheck``' Intrinsic
11867 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11874 declare void @llvm.stackprotectorcheck(i8** <guard>)
11879 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
11880 created stack protector and if they are not equal calls the
11881 ``__stack_chk_fail()`` function.
11886 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
11887 the variable ``@__stack_chk_guard``.
11892 This intrinsic is provided to perform the stack protector check by comparing
11893 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
11894 values do not match call the ``__stack_chk_fail()`` function.
11896 The reason to provide this as an IR level intrinsic instead of implementing it
11897 via other IR operations is that in order to perform this operation at the IR
11898 level without an intrinsic, one would need to create additional basic blocks to
11899 handle the success/failure cases. This makes it difficult to stop the stack
11900 protector check from disrupting sibling tail calls in Codegen. With this
11901 intrinsic, we are able to generate the stack protector basic blocks late in
11902 codegen after the tail call decision has occurred.
11904 '``llvm.objectsize``' Intrinsic
11905 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11912 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
11913 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
11918 The ``llvm.objectsize`` intrinsic is designed to provide information to
11919 the optimizers to determine at compile time whether a) an operation
11920 (like memcpy) will overflow a buffer that corresponds to an object, or
11921 b) that a runtime check for overflow isn't necessary. An object in this
11922 context means an allocation of a specific class, structure, array, or
11928 The ``llvm.objectsize`` intrinsic takes two arguments. The first
11929 argument is a pointer to or into the ``object``. The second argument is
11930 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
11931 or -1 (if false) when the object size is unknown. The second argument
11932 only accepts constants.
11937 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
11938 the size of the object concerned. If the size cannot be determined at
11939 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
11940 on the ``min`` argument).
11942 '``llvm.expect``' Intrinsic
11943 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
11948 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
11953 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
11954 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
11955 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
11960 The ``llvm.expect`` intrinsic provides information about expected (the
11961 most probable) value of ``val``, which can be used by optimizers.
11966 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
11967 a value. The second argument is an expected value, this needs to be a
11968 constant value, variables are not allowed.
11973 This intrinsic is lowered to the ``val``.
11977 '``llvm.assume``' Intrinsic
11978 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11985 declare void @llvm.assume(i1 %cond)
11990 The ``llvm.assume`` allows the optimizer to assume that the provided
11991 condition is true. This information can then be used in simplifying other parts
11997 The condition which the optimizer may assume is always true.
12002 The intrinsic allows the optimizer to assume that the provided condition is
12003 always true whenever the control flow reaches the intrinsic call. No code is
12004 generated for this intrinsic, and instructions that contribute only to the
12005 provided condition are not used for code generation. If the condition is
12006 violated during execution, the behavior is undefined.
12008 Note that the optimizer might limit the transformations performed on values
12009 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
12010 only used to form the intrinsic's input argument. This might prove undesirable
12011 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
12012 sufficient overall improvement in code quality. For this reason,
12013 ``llvm.assume`` should not be used to document basic mathematical invariants
12014 that the optimizer can otherwise deduce or facts that are of little use to the
12019 '``llvm.bitset.test``' Intrinsic
12020 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12027 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
12033 The first argument is a pointer to be tested. The second argument is a
12034 metadata object representing an identifier for a :doc:`bitset <BitSets>`.
12039 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
12040 member of the given bitset.
12042 '``llvm.donothing``' Intrinsic
12043 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12050 declare void @llvm.donothing() nounwind readnone
12055 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
12056 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
12057 with an invoke instruction.
12067 This intrinsic does nothing, and it's removed by optimizers and ignored
12070 Stack Map Intrinsics
12071 --------------------
12073 LLVM provides experimental intrinsics to support runtime patching
12074 mechanisms commonly desired in dynamic language JITs. These intrinsics
12075 are described in :doc:`StackMaps`.