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.
1247 This attribute indicates that the source code contained a hint that
1248 inlining this function is desirable (such as the "inline" keyword in
1249 C/C++). It is just a hint; it imposes no requirements on the
1252 This attribute indicates that the function should be added to a
1253 jump-instruction table at code-generation time, and that all address-taken
1254 references to this function should be replaced with a reference to the
1255 appropriate jump-instruction-table function pointer. Note that this creates
1256 a new pointer for the original function, which means that code that depends
1257 on function-pointer identity can break. So, any function annotated with
1258 ``jumptable`` must also be ``unnamed_addr``.
1260 This attribute suggests that optimization passes and code generator
1261 passes make choices that keep the code size of this function as small
1262 as possible and perform optimizations that may sacrifice runtime
1263 performance in order to minimize the size of the generated code.
1265 This attribute disables prologue / epilogue emission for the
1266 function. This can have very system-specific consequences.
1268 This indicates that the callee function at a call site is not recognized as
1269 a built-in function. LLVM will retain the original call and not replace it
1270 with equivalent code based on the semantics of the built-in function, unless
1271 the call site uses the ``builtin`` attribute. This is valid at call sites
1272 and on function declarations and definitions.
1274 This attribute indicates that calls to the function cannot be
1275 duplicated. A call to a ``noduplicate`` function may be moved
1276 within its parent function, but may not be duplicated within
1277 its parent function.
1279 A function containing a ``noduplicate`` call may still
1280 be an inlining candidate, provided that the call is not
1281 duplicated by inlining. That implies that the function has
1282 internal linkage and only has one call site, so the original
1283 call is dead after inlining.
1285 This attributes disables implicit floating point instructions.
1287 This attribute indicates that the inliner should never inline this
1288 function in any situation. This attribute may not be used together
1289 with the ``alwaysinline`` attribute.
1291 This attribute suppresses lazy symbol binding for the function. This
1292 may make calls to the function faster, at the cost of extra program
1293 startup time if the function is not called during program startup.
1295 This attribute indicates that the code generator should not use a
1296 red zone, even if the target-specific ABI normally permits it.
1298 This function attribute indicates that the function never returns
1299 normally. This produces undefined behavior at runtime if the
1300 function ever does dynamically return.
1302 This function attribute indicates that the function does not call itself
1303 either directly or indirectly down any possible call path. This produces
1304 undefined behavior at runtime if the function ever does recurse.
1306 This function attribute indicates that the function never raises an
1307 exception. If the function does raise an exception, its runtime
1308 behavior is undefined. However, functions marked nounwind may still
1309 trap or generate asynchronous exceptions. Exception handling schemes
1310 that are recognized by LLVM to handle asynchronous exceptions, such
1311 as SEH, will still provide their implementation defined semantics.
1313 This function attribute indicates that most optimization passes will skip
1314 this function, with the exception of interprocedural optimization passes.
1315 Code generation defaults to the "fast" instruction selector.
1316 This attribute cannot be used together with the ``alwaysinline``
1317 attribute; this attribute is also incompatible
1318 with the ``minsize`` attribute and the ``optsize`` attribute.
1320 This attribute requires the ``noinline`` attribute to be specified on
1321 the function as well, so the function is never inlined into any caller.
1322 Only functions with the ``alwaysinline`` attribute are valid
1323 candidates for inlining into the body of this function.
1325 This attribute suggests that optimization passes and code generator
1326 passes make choices that keep the code size of this function low,
1327 and otherwise do optimizations specifically to reduce code size as
1328 long as they do not significantly impact runtime performance.
1330 On a function, this attribute indicates that the function computes its
1331 result (or decides to unwind an exception) based strictly on its arguments,
1332 without dereferencing any pointer arguments or otherwise accessing
1333 any mutable state (e.g. memory, control registers, etc) visible to
1334 caller functions. It does not write through any pointer arguments
1335 (including ``byval`` arguments) and never changes any state visible
1336 to callers. This means that it cannot unwind exceptions by calling
1337 the ``C++`` exception throwing methods.
1339 On an argument, this attribute indicates that the function does not
1340 dereference that pointer argument, even though it may read or write the
1341 memory that the pointer points to if accessed through other pointers.
1343 On a function, this attribute indicates that the function does not write
1344 through any pointer arguments (including ``byval`` arguments) or otherwise
1345 modify any state (e.g. memory, control registers, etc) visible to
1346 caller functions. It may dereference pointer arguments and read
1347 state that may be set in the caller. A readonly function always
1348 returns the same value (or unwinds an exception identically) when
1349 called with the same set of arguments and global state. It cannot
1350 unwind an exception by calling the ``C++`` exception throwing
1353 On an argument, this attribute indicates that the function does not write
1354 through this pointer argument, even though it may write to the memory that
1355 the pointer points to.
1357 This attribute indicates that the only memory accesses inside function are
1358 loads and stores from objects pointed to by its pointer-typed arguments,
1359 with arbitrary offsets. Or in other words, all memory operations in the
1360 function can refer to memory only using pointers based on its function
1362 Note that ``argmemonly`` can be used together with ``readonly`` attribute
1363 in order to specify that function reads only from its arguments.
1365 This attribute indicates that this function can return twice. The C
1366 ``setjmp`` is an example of such a function. The compiler disables
1367 some optimizations (like tail calls) in the caller of these
1370 This attribute indicates that
1371 `SafeStack <http://clang.llvm.org/docs/SafeStack.html>`_
1372 protection is enabled for this function.
1374 If a function that has a ``safestack`` attribute is inlined into a
1375 function that doesn't have a ``safestack`` attribute or which has an
1376 ``ssp``, ``sspstrong`` or ``sspreq`` attribute, then the resulting
1377 function will have a ``safestack`` attribute.
1378 ``sanitize_address``
1379 This attribute indicates that AddressSanitizer checks
1380 (dynamic address safety analysis) are enabled for this function.
1382 This attribute indicates that MemorySanitizer checks (dynamic detection
1383 of accesses to uninitialized memory) are enabled for this function.
1385 This attribute indicates that ThreadSanitizer checks
1386 (dynamic thread safety analysis) are enabled for this function.
1388 This attribute indicates that the function should emit a stack
1389 smashing protector. It is in the form of a "canary" --- a random value
1390 placed on the stack before the local variables that's checked upon
1391 return from the function to see if it has been overwritten. A
1392 heuristic is used to determine if a function needs stack protectors
1393 or not. The heuristic used will enable protectors for functions with:
1395 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1396 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1397 - Calls to alloca() with variable sizes or constant sizes greater than
1398 ``ssp-buffer-size``.
1400 Variables that are identified as requiring a protector will be arranged
1401 on the stack such that they are adjacent to the stack protector guard.
1403 If a function that has an ``ssp`` attribute is inlined into a
1404 function that doesn't have an ``ssp`` attribute, then the resulting
1405 function will have an ``ssp`` attribute.
1407 This attribute indicates that the function should *always* emit a
1408 stack smashing protector. This overrides the ``ssp`` function
1411 Variables that are identified as requiring a protector will be arranged
1412 on the stack such that they are adjacent to the stack protector guard.
1413 The specific layout rules are:
1415 #. Large arrays and structures containing large arrays
1416 (``>= ssp-buffer-size``) are closest to the stack protector.
1417 #. Small arrays and structures containing small arrays
1418 (``< ssp-buffer-size``) are 2nd closest to the protector.
1419 #. Variables that have had their address taken are 3rd closest to the
1422 If a function that has an ``sspreq`` attribute is inlined into a
1423 function that doesn't have an ``sspreq`` attribute or which has an
1424 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1425 an ``sspreq`` attribute.
1427 This attribute indicates that the function should emit a stack smashing
1428 protector. This attribute causes a strong heuristic to be used when
1429 determining if a function needs stack protectors. The strong heuristic
1430 will enable protectors for functions with:
1432 - Arrays of any size and type
1433 - Aggregates containing an array of any size and type.
1434 - Calls to alloca().
1435 - Local variables that have had their address taken.
1437 Variables that are identified as requiring a protector will be arranged
1438 on the stack such that they are adjacent to the stack protector guard.
1439 The specific layout rules are:
1441 #. Large arrays and structures containing large arrays
1442 (``>= ssp-buffer-size``) are closest to the stack protector.
1443 #. Small arrays and structures containing small arrays
1444 (``< ssp-buffer-size``) are 2nd closest to the protector.
1445 #. Variables that have had their address taken are 3rd closest to the
1448 This overrides the ``ssp`` function attribute.
1450 If a function that has an ``sspstrong`` attribute is inlined into a
1451 function that doesn't have an ``sspstrong`` attribute, then the
1452 resulting function will have an ``sspstrong`` attribute.
1454 This attribute indicates that the function will delegate to some other
1455 function with a tail call. The prototype of a thunk should not be used for
1456 optimization purposes. The caller is expected to cast the thunk prototype to
1457 match the thunk target prototype.
1459 This attribute indicates that the ABI being targeted requires that
1460 an unwind table entry be produced for this function even if we can
1461 show that no exceptions passes by it. This is normally the case for
1462 the ELF x86-64 abi, but it can be disabled for some compilation
1471 Note: operand bundles are a work in progress, and they should be
1472 considered experimental at this time.
1474 Operand bundles are tagged sets of SSA values that can be associated
1475 with certain LLVM instructions (currently only ``call`` s and
1476 ``invoke`` s). In a way they are like metadata, but dropping them is
1477 incorrect and will change program semantics.
1481 operand bundle set ::= '[' operand bundle (, operand bundle )* ']'
1482 operand bundle ::= tag '(' [ bundle operand ] (, bundle operand )* ')'
1483 bundle operand ::= SSA value
1484 tag ::= string constant
1486 Operand bundles are **not** part of a function's signature, and a
1487 given function may be called from multiple places with different kinds
1488 of operand bundles. This reflects the fact that the operand bundles
1489 are conceptually a part of the ``call`` (or ``invoke``), not the
1490 callee being dispatched to.
1492 Operand bundles are a generic mechanism intended to support
1493 runtime-introspection-like functionality for managed languages. While
1494 the exact semantics of an operand bundle depend on the bundle tag,
1495 there are certain limitations to how much the presence of an operand
1496 bundle can influence the semantics of a program. These restrictions
1497 are described as the semantics of an "unknown" operand bundle. As
1498 long as the behavior of an operand bundle is describable within these
1499 restrictions, LLVM does not need to have special knowledge of the
1500 operand bundle to not miscompile programs containing it.
1502 - The bundle operands for an unknown operand bundle escape in unknown
1503 ways before control is transferred to the callee or invokee.
1504 - Calls and invokes with operand bundles have unknown read / write
1505 effect on the heap on entry and exit (even if the call target is
1506 ``readnone`` or ``readonly``), unless they're overriden with
1507 callsite specific attributes.
1508 - An operand bundle at a call site cannot change the implementation
1509 of the called function. Inter-procedural optimizations work as
1510 usual as long as they take into account the first two properties.
1512 More specific types of operand bundles are described below.
1514 Deoptimization Operand Bundles
1515 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1517 Deoptimization operand bundles are characterized by the ``"deopt"``
1518 operand bundle tag. These operand bundles represent an alternate
1519 "safe" continuation for the call site they're attached to, and can be
1520 used by a suitable runtime to deoptimize the compiled frame at the
1521 specified call site. There can be at most one ``"deopt"`` operand
1522 bundle attached to a call site. Exact details of deoptimization is
1523 out of scope for the language reference, but it usually involves
1524 rewriting a compiled frame into a set of interpreted frames.
1526 From the compiler's perspective, deoptimization operand bundles make
1527 the call sites they're attached to at least ``readonly``. They read
1528 through all of their pointer typed operands (even if they're not
1529 otherwise escaped) and the entire visible heap. Deoptimization
1530 operand bundles do not capture their operands except during
1531 deoptimization, in which case control will not be returned to the
1534 The inliner knows how to inline through calls that have deoptimization
1535 operand bundles. Just like inlining through a normal call site
1536 involves composing the normal and exceptional continuations, inlining
1537 through a call site with a deoptimization operand bundle needs to
1538 appropriately compose the "safe" deoptimization continuation. The
1539 inliner does this by prepending the parent's deoptimization
1540 continuation to every deoptimization continuation in the inlined body.
1541 E.g. inlining ``@f`` into ``@g`` in the following example
1543 .. code-block:: llvm
1546 call void @x() ;; no deopt state
1547 call void @y() [ "deopt"(i32 10) ]
1548 call void @y() [ "deopt"(i32 10), "unknown"(i8* null) ]
1553 call void @f() [ "deopt"(i32 20) ]
1559 .. code-block:: llvm
1562 call void @x() ;; still no deopt state
1563 call void @y() [ "deopt"(i32 20, i32 10) ]
1564 call void @y() [ "deopt"(i32 20, i32 10), "unknown"(i8* null) ]
1568 It is the frontend's responsibility to structure or encode the
1569 deoptimization state in a way that syntactically prepending the
1570 caller's deoptimization state to the callee's deoptimization state is
1571 semantically equivalent to composing the caller's deoptimization
1572 continuation after the callee's deoptimization continuation.
1574 Funclet Operand Bundles
1575 ^^^^^^^^^^^^^^^^^^^^^^^
1577 Funclet operand bundles are characterized by the ``"funclet"``
1578 operand bundle tag. These operand bundles indicate that a call site
1579 is within a particular funclet. There can be at most one
1580 ``"funclet"`` operand bundle attached to a call site and it must have
1581 exactly one bundle operand.
1585 Module-Level Inline Assembly
1586 ----------------------------
1588 Modules may contain "module-level inline asm" blocks, which corresponds
1589 to the GCC "file scope inline asm" blocks. These blocks are internally
1590 concatenated by LLVM and treated as a single unit, but may be separated
1591 in the ``.ll`` file if desired. The syntax is very simple:
1593 .. code-block:: llvm
1595 module asm "inline asm code goes here"
1596 module asm "more can go here"
1598 The strings can contain any character by escaping non-printable
1599 characters. The escape sequence used is simply "\\xx" where "xx" is the
1600 two digit hex code for the number.
1602 Note that the assembly string *must* be parseable by LLVM's integrated assembler
1603 (unless it is disabled), even when emitting a ``.s`` file.
1605 .. _langref_datalayout:
1610 A module may specify a target specific data layout string that specifies
1611 how data is to be laid out in memory. The syntax for the data layout is
1614 .. code-block:: llvm
1616 target datalayout = "layout specification"
1618 The *layout specification* consists of a list of specifications
1619 separated by the minus sign character ('-'). Each specification starts
1620 with a letter and may include other information after the letter to
1621 define some aspect of the data layout. The specifications accepted are
1625 Specifies that the target lays out data in big-endian form. That is,
1626 the bits with the most significance have the lowest address
1629 Specifies that the target lays out data in little-endian form. That
1630 is, the bits with the least significance have the lowest address
1633 Specifies the natural alignment of the stack in bits. Alignment
1634 promotion of stack variables is limited to the natural stack
1635 alignment to avoid dynamic stack realignment. The stack alignment
1636 must be a multiple of 8-bits. If omitted, the natural stack
1637 alignment defaults to "unspecified", which does not prevent any
1638 alignment promotions.
1639 ``p[n]:<size>:<abi>:<pref>``
1640 This specifies the *size* of a pointer and its ``<abi>`` and
1641 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1642 bits. The address space, ``n``, is optional, and if not specified,
1643 denotes the default address space 0. The value of ``n`` must be
1644 in the range [1,2^23).
1645 ``i<size>:<abi>:<pref>``
1646 This specifies the alignment for an integer type of a given bit
1647 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1648 ``v<size>:<abi>:<pref>``
1649 This specifies the alignment for a vector type of a given bit
1651 ``f<size>:<abi>:<pref>``
1652 This specifies the alignment for a floating point type of a given bit
1653 ``<size>``. Only values of ``<size>`` that are supported by the target
1654 will work. 32 (float) and 64 (double) are supported on all targets; 80
1655 or 128 (different flavors of long double) are also supported on some
1658 This specifies the alignment for an object of aggregate type.
1660 If present, specifies that llvm names are mangled in the output. The
1663 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1664 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1665 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1666 symbols get a ``_`` prefix.
1667 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1668 functions also get a suffix based on the frame size.
1669 * ``x``: Windows x86 COFF prefix: Similar to Windows COFF, but use a ``_``
1670 prefix for ``__cdecl`` functions.
1671 ``n<size1>:<size2>:<size3>...``
1672 This specifies a set of native integer widths for the target CPU in
1673 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1674 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1675 this set are considered to support most general arithmetic operations
1678 On every specification that takes a ``<abi>:<pref>``, specifying the
1679 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1680 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1682 When constructing the data layout for a given target, LLVM starts with a
1683 default set of specifications which are then (possibly) overridden by
1684 the specifications in the ``datalayout`` keyword. The default
1685 specifications are given in this list:
1687 - ``E`` - big endian
1688 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1689 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1690 same as the default address space.
1691 - ``S0`` - natural stack alignment is unspecified
1692 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1693 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1694 - ``i16:16:16`` - i16 is 16-bit aligned
1695 - ``i32:32:32`` - i32 is 32-bit aligned
1696 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1697 alignment of 64-bits
1698 - ``f16:16:16`` - half is 16-bit aligned
1699 - ``f32:32:32`` - float is 32-bit aligned
1700 - ``f64:64:64`` - double is 64-bit aligned
1701 - ``f128:128:128`` - quad is 128-bit aligned
1702 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1703 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1704 - ``a:0:64`` - aggregates are 64-bit aligned
1706 When LLVM is determining the alignment for a given type, it uses the
1709 #. If the type sought is an exact match for one of the specifications,
1710 that specification is used.
1711 #. If no match is found, and the type sought is an integer type, then
1712 the smallest integer type that is larger than the bitwidth of the
1713 sought type is used. If none of the specifications are larger than
1714 the bitwidth then the largest integer type is used. For example,
1715 given the default specifications above, the i7 type will use the
1716 alignment of i8 (next largest) while both i65 and i256 will use the
1717 alignment of i64 (largest specified).
1718 #. If no match is found, and the type sought is a vector type, then the
1719 largest vector type that is smaller than the sought vector type will
1720 be used as a fall back. This happens because <128 x double> can be
1721 implemented in terms of 64 <2 x double>, for example.
1723 The function of the data layout string may not be what you expect.
1724 Notably, this is not a specification from the frontend of what alignment
1725 the code generator should use.
1727 Instead, if specified, the target data layout is required to match what
1728 the ultimate *code generator* expects. This string is used by the
1729 mid-level optimizers to improve code, and this only works if it matches
1730 what the ultimate code generator uses. There is no way to generate IR
1731 that does not embed this target-specific detail into the IR. If you
1732 don't specify the string, the default specifications will be used to
1733 generate a Data Layout and the optimization phases will operate
1734 accordingly and introduce target specificity into the IR with respect to
1735 these default specifications.
1742 A module may specify a target triple string that describes the target
1743 host. The syntax for the target triple is simply:
1745 .. code-block:: llvm
1747 target triple = "x86_64-apple-macosx10.7.0"
1749 The *target triple* string consists of a series of identifiers delimited
1750 by the minus sign character ('-'). The canonical forms are:
1754 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1755 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1757 This information is passed along to the backend so that it generates
1758 code for the proper architecture. It's possible to override this on the
1759 command line with the ``-mtriple`` command line option.
1761 .. _pointeraliasing:
1763 Pointer Aliasing Rules
1764 ----------------------
1766 Any memory access must be done through a pointer value associated with
1767 an address range of the memory access, otherwise the behavior is
1768 undefined. Pointer values are associated with address ranges according
1769 to the following rules:
1771 - A pointer value is associated with the addresses associated with any
1772 value it is *based* on.
1773 - An address of a global variable is associated with the address range
1774 of the variable's storage.
1775 - The result value of an allocation instruction is associated with the
1776 address range of the allocated storage.
1777 - A null pointer in the default address-space is associated with no
1779 - An integer constant other than zero or a pointer value returned from
1780 a function not defined within LLVM may be associated with address
1781 ranges allocated through mechanisms other than those provided by
1782 LLVM. Such ranges shall not overlap with any ranges of addresses
1783 allocated by mechanisms provided by LLVM.
1785 A pointer value is *based* on another pointer value according to the
1788 - A pointer value formed from a ``getelementptr`` operation is *based*
1789 on the first value operand of the ``getelementptr``.
1790 - The result value of a ``bitcast`` is *based* on the operand of the
1792 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1793 values that contribute (directly or indirectly) to the computation of
1794 the pointer's value.
1795 - The "*based* on" relationship is transitive.
1797 Note that this definition of *"based"* is intentionally similar to the
1798 definition of *"based"* in C99, though it is slightly weaker.
1800 LLVM IR does not associate types with memory. The result type of a
1801 ``load`` merely indicates the size and alignment of the memory from
1802 which to load, as well as the interpretation of the value. The first
1803 operand type of a ``store`` similarly only indicates the size and
1804 alignment of the store.
1806 Consequently, type-based alias analysis, aka TBAA, aka
1807 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1808 :ref:`Metadata <metadata>` may be used to encode additional information
1809 which specialized optimization passes may use to implement type-based
1814 Volatile Memory Accesses
1815 ------------------------
1817 Certain memory accesses, such as :ref:`load <i_load>`'s,
1818 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1819 marked ``volatile``. The optimizers must not change the number of
1820 volatile operations or change their order of execution relative to other
1821 volatile operations. The optimizers *may* change the order of volatile
1822 operations relative to non-volatile operations. This is not Java's
1823 "volatile" and has no cross-thread synchronization behavior.
1825 IR-level volatile loads and stores cannot safely be optimized into
1826 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1827 flagged volatile. Likewise, the backend should never split or merge
1828 target-legal volatile load/store instructions.
1830 .. admonition:: Rationale
1832 Platforms may rely on volatile loads and stores of natively supported
1833 data width to be executed as single instruction. For example, in C
1834 this holds for an l-value of volatile primitive type with native
1835 hardware support, but not necessarily for aggregate types. The
1836 frontend upholds these expectations, which are intentionally
1837 unspecified in the IR. The rules above ensure that IR transformations
1838 do not violate the frontend's contract with the language.
1842 Memory Model for Concurrent Operations
1843 --------------------------------------
1845 The LLVM IR does not define any way to start parallel threads of
1846 execution or to register signal handlers. Nonetheless, there are
1847 platform-specific ways to create them, and we define LLVM IR's behavior
1848 in their presence. This model is inspired by the C++0x memory model.
1850 For a more informal introduction to this model, see the :doc:`Atomics`.
1852 We define a *happens-before* partial order as the least partial order
1855 - Is a superset of single-thread program order, and
1856 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1857 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1858 techniques, like pthread locks, thread creation, thread joining,
1859 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1860 Constraints <ordering>`).
1862 Note that program order does not introduce *happens-before* edges
1863 between a thread and signals executing inside that thread.
1865 Every (defined) read operation (load instructions, memcpy, atomic
1866 loads/read-modify-writes, etc.) R reads a series of bytes written by
1867 (defined) write operations (store instructions, atomic
1868 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1869 section, initialized globals are considered to have a write of the
1870 initializer which is atomic and happens before any other read or write
1871 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1872 may see any write to the same byte, except:
1874 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1875 write\ :sub:`2` happens before R\ :sub:`byte`, then
1876 R\ :sub:`byte` does not see write\ :sub:`1`.
1877 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1878 R\ :sub:`byte` does not see write\ :sub:`3`.
1880 Given that definition, R\ :sub:`byte` is defined as follows:
1882 - If R is volatile, the result is target-dependent. (Volatile is
1883 supposed to give guarantees which can support ``sig_atomic_t`` in
1884 C/C++, and may be used for accesses to addresses that do not behave
1885 like normal memory. It does not generally provide cross-thread
1887 - Otherwise, if there is no write to the same byte that happens before
1888 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1889 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1890 R\ :sub:`byte` returns the value written by that write.
1891 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1892 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1893 Memory Ordering Constraints <ordering>` section for additional
1894 constraints on how the choice is made.
1895 - Otherwise R\ :sub:`byte` returns ``undef``.
1897 R returns the value composed of the series of bytes it read. This
1898 implies that some bytes within the value may be ``undef`` **without**
1899 the entire value being ``undef``. Note that this only defines the
1900 semantics of the operation; it doesn't mean that targets will emit more
1901 than one instruction to read the series of bytes.
1903 Note that in cases where none of the atomic intrinsics are used, this
1904 model places only one restriction on IR transformations on top of what
1905 is required for single-threaded execution: introducing a store to a byte
1906 which might not otherwise be stored is not allowed in general.
1907 (Specifically, in the case where another thread might write to and read
1908 from an address, introducing a store can change a load that may see
1909 exactly one write into a load that may see multiple writes.)
1913 Atomic Memory Ordering Constraints
1914 ----------------------------------
1916 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1917 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1918 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1919 ordering parameters that determine which other atomic instructions on
1920 the same address they *synchronize with*. These semantics are borrowed
1921 from Java and C++0x, but are somewhat more colloquial. If these
1922 descriptions aren't precise enough, check those specs (see spec
1923 references in the :doc:`atomics guide <Atomics>`).
1924 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1925 differently since they don't take an address. See that instruction's
1926 documentation for details.
1928 For a simpler introduction to the ordering constraints, see the
1932 The set of values that can be read is governed by the happens-before
1933 partial order. A value cannot be read unless some operation wrote
1934 it. This is intended to provide a guarantee strong enough to model
1935 Java's non-volatile shared variables. This ordering cannot be
1936 specified for read-modify-write operations; it is not strong enough
1937 to make them atomic in any interesting way.
1939 In addition to the guarantees of ``unordered``, there is a single
1940 total order for modifications by ``monotonic`` operations on each
1941 address. All modification orders must be compatible with the
1942 happens-before order. There is no guarantee that the modification
1943 orders can be combined to a global total order for the whole program
1944 (and this often will not be possible). The read in an atomic
1945 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1946 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1947 order immediately before the value it writes. If one atomic read
1948 happens before another atomic read of the same address, the later
1949 read must see the same value or a later value in the address's
1950 modification order. This disallows reordering of ``monotonic`` (or
1951 stronger) operations on the same address. If an address is written
1952 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1953 read that address repeatedly, the other threads must eventually see
1954 the write. This corresponds to the C++0x/C1x
1955 ``memory_order_relaxed``.
1957 In addition to the guarantees of ``monotonic``, a
1958 *synchronizes-with* edge may be formed with a ``release`` operation.
1959 This is intended to model C++'s ``memory_order_acquire``.
1961 In addition to the guarantees of ``monotonic``, if this operation
1962 writes a value which is subsequently read by an ``acquire``
1963 operation, it *synchronizes-with* that operation. (This isn't a
1964 complete description; see the C++0x definition of a release
1965 sequence.) This corresponds to the C++0x/C1x
1966 ``memory_order_release``.
1967 ``acq_rel`` (acquire+release)
1968 Acts as both an ``acquire`` and ``release`` operation on its
1969 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1970 ``seq_cst`` (sequentially consistent)
1971 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1972 operation that only reads, ``release`` for an operation that only
1973 writes), there is a global total order on all
1974 sequentially-consistent operations on all addresses, which is
1975 consistent with the *happens-before* partial order and with the
1976 modification orders of all the affected addresses. Each
1977 sequentially-consistent read sees the last preceding write to the
1978 same address in this global order. This corresponds to the C++0x/C1x
1979 ``memory_order_seq_cst`` and Java volatile.
1983 If an atomic operation is marked ``singlethread``, it only *synchronizes
1984 with* or participates in modification and seq\_cst total orderings with
1985 other operations running in the same thread (for example, in signal
1993 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1994 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1995 :ref:`frem <i_frem>`, :ref:`fcmp <i_fcmp>`) have the following flags that can
1996 be set to enable otherwise unsafe floating point operations
1999 No NaNs - Allow optimizations to assume the arguments and result are not
2000 NaN. Such optimizations are required to retain defined behavior over
2001 NaNs, but the value of the result is undefined.
2004 No Infs - Allow optimizations to assume the arguments and result are not
2005 +/-Inf. Such optimizations are required to retain defined behavior over
2006 +/-Inf, but the value of the result is undefined.
2009 No Signed Zeros - Allow optimizations to treat the sign of a zero
2010 argument or result as insignificant.
2013 Allow Reciprocal - Allow optimizations to use the reciprocal of an
2014 argument rather than perform division.
2017 Fast - Allow algebraically equivalent transformations that may
2018 dramatically change results in floating point (e.g. reassociate). This
2019 flag implies all the others.
2023 Use-list Order Directives
2024 -------------------------
2026 Use-list directives encode the in-memory order of each use-list, allowing the
2027 order to be recreated. ``<order-indexes>`` is a comma-separated list of
2028 indexes that are assigned to the referenced value's uses. The referenced
2029 value's use-list is immediately sorted by these indexes.
2031 Use-list directives may appear at function scope or global scope. They are not
2032 instructions, and have no effect on the semantics of the IR. When they're at
2033 function scope, they must appear after the terminator of the final basic block.
2035 If basic blocks have their address taken via ``blockaddress()`` expressions,
2036 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
2043 uselistorder <ty> <value>, { <order-indexes> }
2044 uselistorder_bb @function, %block { <order-indexes> }
2050 define void @foo(i32 %arg1, i32 %arg2) {
2052 ; ... instructions ...
2054 ; ... instructions ...
2056 ; At function scope.
2057 uselistorder i32 %arg1, { 1, 0, 2 }
2058 uselistorder label %bb, { 1, 0 }
2062 uselistorder i32* @global, { 1, 2, 0 }
2063 uselistorder i32 7, { 1, 0 }
2064 uselistorder i32 (i32) @bar, { 1, 0 }
2065 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
2072 The LLVM type system is one of the most important features of the
2073 intermediate representation. Being typed enables a number of
2074 optimizations to be performed on the intermediate representation
2075 directly, without having to do extra analyses on the side before the
2076 transformation. A strong type system makes it easier to read the
2077 generated code and enables novel analyses and transformations that are
2078 not feasible to perform on normal three address code representations.
2088 The void type does not represent any value and has no size.
2106 The function type can be thought of as a function signature. It consists of a
2107 return type and a list of formal parameter types. The return type of a function
2108 type is a void type or first class type --- except for :ref:`label <t_label>`
2109 and :ref:`metadata <t_metadata>` types.
2115 <returntype> (<parameter list>)
2117 ...where '``<parameter list>``' is a comma-separated list of type
2118 specifiers. Optionally, the parameter list may include a type ``...``, which
2119 indicates that the function takes a variable number of arguments. Variable
2120 argument functions can access their arguments with the :ref:`variable argument
2121 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
2122 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
2126 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2127 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
2128 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2129 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
2130 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2131 | ``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. |
2132 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2133 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
2134 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2141 The :ref:`first class <t_firstclass>` types are perhaps the most important.
2142 Values of these types are the only ones which can be produced by
2150 These are the types that are valid in registers from CodeGen's perspective.
2159 The integer type is a very simple type that simply specifies an
2160 arbitrary bit width for the integer type desired. Any bit width from 1
2161 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
2169 The number of bits the integer will occupy is specified by the ``N``
2175 +----------------+------------------------------------------------+
2176 | ``i1`` | a single-bit integer. |
2177 +----------------+------------------------------------------------+
2178 | ``i32`` | a 32-bit integer. |
2179 +----------------+------------------------------------------------+
2180 | ``i1942652`` | a really big integer of over 1 million bits. |
2181 +----------------+------------------------------------------------+
2185 Floating Point Types
2186 """"""""""""""""""""
2195 - 16-bit floating point value
2198 - 32-bit floating point value
2201 - 64-bit floating point value
2204 - 128-bit floating point value (112-bit mantissa)
2207 - 80-bit floating point value (X87)
2210 - 128-bit floating point value (two 64-bits)
2217 The x86_mmx type represents a value held in an MMX register on an x86
2218 machine. The operations allowed on it are quite limited: parameters and
2219 return values, load and store, and bitcast. User-specified MMX
2220 instructions are represented as intrinsic or asm calls with arguments
2221 and/or results of this type. There are no arrays, vectors or constants
2238 The pointer type is used to specify memory locations. Pointers are
2239 commonly used to reference objects in memory.
2241 Pointer types may have an optional address space attribute defining the
2242 numbered address space where the pointed-to object resides. The default
2243 address space is number zero. The semantics of non-zero address spaces
2244 are target-specific.
2246 Note that LLVM does not permit pointers to void (``void*``) nor does it
2247 permit pointers to labels (``label*``). Use ``i8*`` instead.
2257 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2258 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2259 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2260 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2261 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2262 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2263 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2272 A vector type is a simple derived type that represents a vector of
2273 elements. Vector types are used when multiple primitive data are
2274 operated in parallel using a single instruction (SIMD). A vector type
2275 requires a size (number of elements) and an underlying primitive data
2276 type. Vector types are considered :ref:`first class <t_firstclass>`.
2282 < <# elements> x <elementtype> >
2284 The number of elements is a constant integer value larger than 0;
2285 elementtype may be any integer, floating point or pointer type. Vectors
2286 of size zero are not allowed.
2290 +-------------------+--------------------------------------------------+
2291 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2292 +-------------------+--------------------------------------------------+
2293 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2294 +-------------------+--------------------------------------------------+
2295 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2296 +-------------------+--------------------------------------------------+
2297 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2298 +-------------------+--------------------------------------------------+
2307 The label type represents code labels.
2322 The token type is used when a value is associated with an instruction
2323 but all uses of the value must not attempt to introspect or obscure it.
2324 As such, it is not appropriate to have a :ref:`phi <i_phi>` or
2325 :ref:`select <i_select>` of type token.
2342 The metadata type represents embedded metadata. No derived types may be
2343 created from metadata except for :ref:`function <t_function>` arguments.
2356 Aggregate Types are a subset of derived types that can contain multiple
2357 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2358 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2368 The array type is a very simple derived type that arranges elements
2369 sequentially in memory. The array type requires a size (number of
2370 elements) and an underlying data type.
2376 [<# elements> x <elementtype>]
2378 The number of elements is a constant integer value; ``elementtype`` may
2379 be any type with a size.
2383 +------------------+--------------------------------------+
2384 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2385 +------------------+--------------------------------------+
2386 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2387 +------------------+--------------------------------------+
2388 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2389 +------------------+--------------------------------------+
2391 Here are some examples of multidimensional arrays:
2393 +-----------------------------+----------------------------------------------------------+
2394 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2395 +-----------------------------+----------------------------------------------------------+
2396 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2397 +-----------------------------+----------------------------------------------------------+
2398 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2399 +-----------------------------+----------------------------------------------------------+
2401 There is no restriction on indexing beyond the end of the array implied
2402 by a static type (though there are restrictions on indexing beyond the
2403 bounds of an allocated object in some cases). This means that
2404 single-dimension 'variable sized array' addressing can be implemented in
2405 LLVM with a zero length array type. An implementation of 'pascal style
2406 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2416 The structure type is used to represent a collection of data members
2417 together in memory. The elements of a structure may be any type that has
2420 Structures in memory are accessed using '``load``' and '``store``' by
2421 getting a pointer to a field with the '``getelementptr``' instruction.
2422 Structures in registers are accessed using the '``extractvalue``' and
2423 '``insertvalue``' instructions.
2425 Structures may optionally be "packed" structures, which indicate that
2426 the alignment of the struct is one byte, and that there is no padding
2427 between the elements. In non-packed structs, padding between field types
2428 is inserted as defined by the DataLayout string in the module, which is
2429 required to match what the underlying code generator expects.
2431 Structures can either be "literal" or "identified". A literal structure
2432 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2433 identified types are always defined at the top level with a name.
2434 Literal types are uniqued by their contents and can never be recursive
2435 or opaque since there is no way to write one. Identified types can be
2436 recursive, can be opaqued, and are never uniqued.
2442 %T1 = type { <type list> } ; Identified normal struct type
2443 %T2 = type <{ <type list> }> ; Identified packed struct type
2447 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2448 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2449 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2450 | ``{ 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``. |
2451 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2452 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2453 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2457 Opaque Structure Types
2458 """"""""""""""""""""""
2462 Opaque structure types are used to represent named structure types that
2463 do not have a body specified. This corresponds (for example) to the C
2464 notion of a forward declared structure.
2475 +--------------+-------------------+
2476 | ``opaque`` | An opaque type. |
2477 +--------------+-------------------+
2484 LLVM has several different basic types of constants. This section
2485 describes them all and their syntax.
2490 **Boolean constants**
2491 The two strings '``true``' and '``false``' are both valid constants
2493 **Integer constants**
2494 Standard integers (such as '4') are constants of the
2495 :ref:`integer <t_integer>` type. Negative numbers may be used with
2497 **Floating point constants**
2498 Floating point constants use standard decimal notation (e.g.
2499 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2500 hexadecimal notation (see below). The assembler requires the exact
2501 decimal value of a floating-point constant. For example, the
2502 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2503 decimal in binary. Floating point constants must have a :ref:`floating
2504 point <t_floating>` type.
2505 **Null pointer constants**
2506 The identifier '``null``' is recognized as a null pointer constant
2507 and must be of :ref:`pointer type <t_pointer>`.
2509 The identifier '``none``' is recognized as an empty token constant
2510 and must be of :ref:`token type <t_token>`.
2512 The one non-intuitive notation for constants is the hexadecimal form of
2513 floating point constants. For example, the form
2514 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2515 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2516 constants are required (and the only time that they are generated by the
2517 disassembler) is when a floating point constant must be emitted but it
2518 cannot be represented as a decimal floating point number in a reasonable
2519 number of digits. For example, NaN's, infinities, and other special
2520 values are represented in their IEEE hexadecimal format so that assembly
2521 and disassembly do not cause any bits to change in the constants.
2523 When using the hexadecimal form, constants of types half, float, and
2524 double are represented using the 16-digit form shown above (which
2525 matches the IEEE754 representation for double); half and float values
2526 must, however, be exactly representable as IEEE 754 half and single
2527 precision, respectively. Hexadecimal format is always used for long
2528 double, and there are three forms of long double. The 80-bit format used
2529 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2530 128-bit format used by PowerPC (two adjacent doubles) is represented by
2531 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2532 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2533 will only work if they match the long double format on your target.
2534 The IEEE 16-bit format (half precision) is represented by ``0xH``
2535 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2536 (sign bit at the left).
2538 There are no constants of type x86_mmx.
2540 .. _complexconstants:
2545 Complex constants are a (potentially recursive) combination of simple
2546 constants and smaller complex constants.
2548 **Structure constants**
2549 Structure constants are represented with notation similar to
2550 structure type definitions (a comma separated list of elements,
2551 surrounded by braces (``{}``)). For example:
2552 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2553 "``@G = external global i32``". Structure constants must have
2554 :ref:`structure type <t_struct>`, and the number and types of elements
2555 must match those specified by the type.
2557 Array constants are represented with notation similar to array type
2558 definitions (a comma separated list of elements, surrounded by
2559 square brackets (``[]``)). For example:
2560 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2561 :ref:`array type <t_array>`, and the number and types of elements must
2562 match those specified by the type. As a special case, character array
2563 constants may also be represented as a double-quoted string using the ``c``
2564 prefix. For example: "``c"Hello World\0A\00"``".
2565 **Vector constants**
2566 Vector constants are represented with notation similar to vector
2567 type definitions (a comma separated list of elements, surrounded by
2568 less-than/greater-than's (``<>``)). For example:
2569 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2570 must have :ref:`vector type <t_vector>`, and the number and types of
2571 elements must match those specified by the type.
2572 **Zero initialization**
2573 The string '``zeroinitializer``' can be used to zero initialize a
2574 value to zero of *any* type, including scalar and
2575 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2576 having to print large zero initializers (e.g. for large arrays) and
2577 is always exactly equivalent to using explicit zero initializers.
2579 A metadata node is a constant tuple without types. For example:
2580 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2581 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2582 Unlike other typed constants that are meant to be interpreted as part of
2583 the instruction stream, metadata is a place to attach additional
2584 information such as debug info.
2586 Global Variable and Function Addresses
2587 --------------------------------------
2589 The addresses of :ref:`global variables <globalvars>` and
2590 :ref:`functions <functionstructure>` are always implicitly valid
2591 (link-time) constants. These constants are explicitly referenced when
2592 the :ref:`identifier for the global <identifiers>` is used and always have
2593 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2596 .. code-block:: llvm
2600 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2607 The string '``undef``' can be used anywhere a constant is expected, and
2608 indicates that the user of the value may receive an unspecified
2609 bit-pattern. Undefined values may be of any type (other than '``label``'
2610 or '``void``') and be used anywhere a constant is permitted.
2612 Undefined values are useful because they indicate to the compiler that
2613 the program is well defined no matter what value is used. This gives the
2614 compiler more freedom to optimize. Here are some examples of
2615 (potentially surprising) transformations that are valid (in pseudo IR):
2617 .. code-block:: llvm
2627 This is safe because all of the output bits are affected by the undef
2628 bits. Any output bit can have a zero or one depending on the input bits.
2630 .. code-block:: llvm
2641 These logical operations have bits that are not always affected by the
2642 input. For example, if ``%X`` has a zero bit, then the output of the
2643 '``and``' operation will always be a zero for that bit, no matter what
2644 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2645 optimize or assume that the result of the '``and``' is '``undef``'.
2646 However, it is safe to assume that all bits of the '``undef``' could be
2647 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2648 all the bits of the '``undef``' operand to the '``or``' could be set,
2649 allowing the '``or``' to be folded to -1.
2651 .. code-block:: llvm
2653 %A = select undef, %X, %Y
2654 %B = select undef, 42, %Y
2655 %C = select %X, %Y, undef
2665 This set of examples shows that undefined '``select``' (and conditional
2666 branch) conditions can go *either way*, but they have to come from one
2667 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2668 both known to have a clear low bit, then ``%A`` would have to have a
2669 cleared low bit. However, in the ``%C`` example, the optimizer is
2670 allowed to assume that the '``undef``' operand could be the same as
2671 ``%Y``, allowing the whole '``select``' to be eliminated.
2673 .. code-block:: llvm
2675 %A = xor undef, undef
2692 This example points out that two '``undef``' operands are not
2693 necessarily the same. This can be surprising to people (and also matches
2694 C semantics) where they assume that "``X^X``" is always zero, even if
2695 ``X`` is undefined. This isn't true for a number of reasons, but the
2696 short answer is that an '``undef``' "variable" can arbitrarily change
2697 its value over its "live range". This is true because the variable
2698 doesn't actually *have a live range*. Instead, the value is logically
2699 read from arbitrary registers that happen to be around when needed, so
2700 the value is not necessarily consistent over time. In fact, ``%A`` and
2701 ``%C`` need to have the same semantics or the core LLVM "replace all
2702 uses with" concept would not hold.
2704 .. code-block:: llvm
2712 These examples show the crucial difference between an *undefined value*
2713 and *undefined behavior*. An undefined value (like '``undef``') is
2714 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2715 operation can be constant folded to '``undef``', because the '``undef``'
2716 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2717 However, in the second example, we can make a more aggressive
2718 assumption: because the ``undef`` is allowed to be an arbitrary value,
2719 we are allowed to assume that it could be zero. Since a divide by zero
2720 has *undefined behavior*, we are allowed to assume that the operation
2721 does not execute at all. This allows us to delete the divide and all
2722 code after it. Because the undefined operation "can't happen", the
2723 optimizer can assume that it occurs in dead code.
2725 .. code-block:: llvm
2727 a: store undef -> %X
2728 b: store %X -> undef
2733 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2734 value can be assumed to not have any effect; we can assume that the
2735 value is overwritten with bits that happen to match what was already
2736 there. However, a store *to* an undefined location could clobber
2737 arbitrary memory, therefore, it has undefined behavior.
2744 Poison values are similar to :ref:`undef values <undefvalues>`, however
2745 they also represent the fact that an instruction or constant expression
2746 that cannot evoke side effects has nevertheless detected a condition
2747 that results in undefined behavior.
2749 There is currently no way of representing a poison value in the IR; they
2750 only exist when produced by operations such as :ref:`add <i_add>` with
2753 Poison value behavior is defined in terms of value *dependence*:
2755 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2756 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2757 their dynamic predecessor basic block.
2758 - Function arguments depend on the corresponding actual argument values
2759 in the dynamic callers of their functions.
2760 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2761 instructions that dynamically transfer control back to them.
2762 - :ref:`Invoke <i_invoke>` instructions depend on the
2763 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2764 call instructions that dynamically transfer control back to them.
2765 - Non-volatile loads and stores depend on the most recent stores to all
2766 of the referenced memory addresses, following the order in the IR
2767 (including loads and stores implied by intrinsics such as
2768 :ref:`@llvm.memcpy <int_memcpy>`.)
2769 - An instruction with externally visible side effects depends on the
2770 most recent preceding instruction with externally visible side
2771 effects, following the order in the IR. (This includes :ref:`volatile
2772 operations <volatile>`.)
2773 - An instruction *control-depends* on a :ref:`terminator
2774 instruction <terminators>` if the terminator instruction has
2775 multiple successors and the instruction is always executed when
2776 control transfers to one of the successors, and may not be executed
2777 when control is transferred to another.
2778 - Additionally, an instruction also *control-depends* on a terminator
2779 instruction if the set of instructions it otherwise depends on would
2780 be different if the terminator had transferred control to a different
2782 - Dependence is transitive.
2784 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2785 with the additional effect that any instruction that has a *dependence*
2786 on a poison value has undefined behavior.
2788 Here are some examples:
2790 .. code-block:: llvm
2793 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2794 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2795 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
2796 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2798 store i32 %poison, i32* @g ; Poison value stored to memory.
2799 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
2801 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2803 %narrowaddr = bitcast i32* @g to i16*
2804 %wideaddr = bitcast i32* @g to i64*
2805 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2806 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
2808 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2809 br i1 %cmp, label %true, label %end ; Branch to either destination.
2812 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2813 ; it has undefined behavior.
2817 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2818 ; Both edges into this PHI are
2819 ; control-dependent on %cmp, so this
2820 ; always results in a poison value.
2822 store volatile i32 0, i32* @g ; This would depend on the store in %true
2823 ; if %cmp is true, or the store in %entry
2824 ; otherwise, so this is undefined behavior.
2826 br i1 %cmp, label %second_true, label %second_end
2827 ; The same branch again, but this time the
2828 ; true block doesn't have side effects.
2835 store volatile i32 0, i32* @g ; This time, the instruction always depends
2836 ; on the store in %end. Also, it is
2837 ; control-equivalent to %end, so this is
2838 ; well-defined (ignoring earlier undefined
2839 ; behavior in this example).
2843 Addresses of Basic Blocks
2844 -------------------------
2846 ``blockaddress(@function, %block)``
2848 The '``blockaddress``' constant computes the address of the specified
2849 basic block in the specified function, and always has an ``i8*`` type.
2850 Taking the address of the entry block is illegal.
2852 This value only has defined behavior when used as an operand to the
2853 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2854 against null. Pointer equality tests between labels addresses results in
2855 undefined behavior --- though, again, comparison against null is ok, and
2856 no label is equal to the null pointer. This may be passed around as an
2857 opaque pointer sized value as long as the bits are not inspected. This
2858 allows ``ptrtoint`` and arithmetic to be performed on these values so
2859 long as the original value is reconstituted before the ``indirectbr``
2862 Finally, some targets may provide defined semantics when using the value
2863 as the operand to an inline assembly, but that is target specific.
2867 Constant Expressions
2868 --------------------
2870 Constant expressions are used to allow expressions involving other
2871 constants to be used as constants. Constant expressions may be of any
2872 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2873 that does not have side effects (e.g. load and call are not supported).
2874 The following is the syntax for constant expressions:
2876 ``trunc (CST to TYPE)``
2877 Truncate a constant to another type. The bit size of CST must be
2878 larger than the bit size of TYPE. Both types must be integers.
2879 ``zext (CST to TYPE)``
2880 Zero extend a constant to another type. The bit size of CST must be
2881 smaller than the bit size of TYPE. Both types must be integers.
2882 ``sext (CST to TYPE)``
2883 Sign extend a constant to another type. The bit size of CST must be
2884 smaller than the bit size of TYPE. Both types must be integers.
2885 ``fptrunc (CST to TYPE)``
2886 Truncate a floating point constant to another floating point type.
2887 The size of CST must be larger than the size of TYPE. Both types
2888 must be floating point.
2889 ``fpext (CST to TYPE)``
2890 Floating point extend a constant to another type. The size of CST
2891 must be smaller or equal to the size of TYPE. Both types must be
2893 ``fptoui (CST to TYPE)``
2894 Convert a floating point constant to the corresponding unsigned
2895 integer constant. TYPE must be a scalar or vector integer type. CST
2896 must be of scalar or vector floating point type. Both CST and TYPE
2897 must be scalars, or vectors of the same number of elements. If the
2898 value won't fit in the integer type, the results are undefined.
2899 ``fptosi (CST to TYPE)``
2900 Convert a floating point constant to the corresponding signed
2901 integer constant. TYPE must be a scalar or vector integer type. CST
2902 must be of scalar or vector floating point type. Both CST and TYPE
2903 must be scalars, or vectors of the same number of elements. If the
2904 value won't fit in the integer type, the results are undefined.
2905 ``uitofp (CST to TYPE)``
2906 Convert an unsigned integer constant to the corresponding floating
2907 point constant. TYPE must be a scalar or vector floating point type.
2908 CST must be of scalar or vector integer type. Both CST and TYPE must
2909 be scalars, or vectors of the same number of elements. If the value
2910 won't fit in the floating point type, the results are undefined.
2911 ``sitofp (CST to TYPE)``
2912 Convert a signed integer constant to the corresponding floating
2913 point constant. TYPE must be a scalar or vector floating point type.
2914 CST must be of scalar or vector integer type. Both CST and TYPE must
2915 be scalars, or vectors of the same number of elements. If the value
2916 won't fit in the floating point type, the results are undefined.
2917 ``ptrtoint (CST to TYPE)``
2918 Convert a pointer typed constant to the corresponding integer
2919 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2920 pointer type. The ``CST`` value is zero extended, truncated, or
2921 unchanged to make it fit in ``TYPE``.
2922 ``inttoptr (CST to TYPE)``
2923 Convert an integer constant to a pointer constant. TYPE must be a
2924 pointer type. CST must be of integer type. The CST value is zero
2925 extended, truncated, or unchanged to make it fit in a pointer size.
2926 This one is *really* dangerous!
2927 ``bitcast (CST to TYPE)``
2928 Convert a constant, CST, to another TYPE. The constraints of the
2929 operands are the same as those for the :ref:`bitcast
2930 instruction <i_bitcast>`.
2931 ``addrspacecast (CST to TYPE)``
2932 Convert a constant pointer or constant vector of pointer, CST, to another
2933 TYPE in a different address space. The constraints of the operands are the
2934 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2935 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
2936 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2937 constants. As with the :ref:`getelementptr <i_getelementptr>`
2938 instruction, the index list may have zero or more indexes, which are
2939 required to make sense for the type of "pointer to TY".
2940 ``select (COND, VAL1, VAL2)``
2941 Perform the :ref:`select operation <i_select>` on constants.
2942 ``icmp COND (VAL1, VAL2)``
2943 Performs the :ref:`icmp operation <i_icmp>` on constants.
2944 ``fcmp COND (VAL1, VAL2)``
2945 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2946 ``extractelement (VAL, IDX)``
2947 Perform the :ref:`extractelement operation <i_extractelement>` on
2949 ``insertelement (VAL, ELT, IDX)``
2950 Perform the :ref:`insertelement operation <i_insertelement>` on
2952 ``shufflevector (VEC1, VEC2, IDXMASK)``
2953 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2955 ``extractvalue (VAL, IDX0, IDX1, ...)``
2956 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2957 constants. The index list is interpreted in a similar manner as
2958 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2959 least one index value must be specified.
2960 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2961 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2962 The index list is interpreted in a similar manner as indices in a
2963 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2964 value must be specified.
2965 ``OPCODE (LHS, RHS)``
2966 Perform the specified operation of the LHS and RHS constants. OPCODE
2967 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2968 binary <bitwiseops>` operations. The constraints on operands are
2969 the same as those for the corresponding instruction (e.g. no bitwise
2970 operations on floating point values are allowed).
2977 Inline Assembler Expressions
2978 ----------------------------
2980 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2981 Inline Assembly <moduleasm>`) through the use of a special value. This value
2982 represents the inline assembler as a template string (containing the
2983 instructions to emit), a list of operand constraints (stored as a string), a
2984 flag that indicates whether or not the inline asm expression has side effects,
2985 and a flag indicating whether the function containing the asm needs to align its
2986 stack conservatively.
2988 The template string supports argument substitution of the operands using "``$``"
2989 followed by a number, to indicate substitution of the given register/memory
2990 location, as specified by the constraint string. "``${NUM:MODIFIER}``" may also
2991 be used, where ``MODIFIER`` is a target-specific annotation for how to print the
2992 operand (See :ref:`inline-asm-modifiers`).
2994 A literal "``$``" may be included by using "``$$``" in the template. To include
2995 other special characters into the output, the usual "``\XX``" escapes may be
2996 used, just as in other strings. Note that after template substitution, the
2997 resulting assembly string is parsed by LLVM's integrated assembler unless it is
2998 disabled -- even when emitting a ``.s`` file -- and thus must contain assembly
2999 syntax known to LLVM.
3001 LLVM's support for inline asm is modeled closely on the requirements of Clang's
3002 GCC-compatible inline-asm support. Thus, the feature-set and the constraint and
3003 modifier codes listed here are similar or identical to those in GCC's inline asm
3004 support. However, to be clear, the syntax of the template and constraint strings
3005 described here is *not* the same as the syntax accepted by GCC and Clang, and,
3006 while most constraint letters are passed through as-is by Clang, some get
3007 translated to other codes when converting from the C source to the LLVM
3010 An example inline assembler expression is:
3012 .. code-block:: llvm
3014 i32 (i32) asm "bswap $0", "=r,r"
3016 Inline assembler expressions may **only** be used as the callee operand
3017 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
3018 Thus, typically we have:
3020 .. code-block:: llvm
3022 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
3024 Inline asms with side effects not visible in the constraint list must be
3025 marked as having side effects. This is done through the use of the
3026 '``sideeffect``' keyword, like so:
3028 .. code-block:: llvm
3030 call void asm sideeffect "eieio", ""()
3032 In some cases inline asms will contain code that will not work unless
3033 the stack is aligned in some way, such as calls or SSE instructions on
3034 x86, yet will not contain code that does that alignment within the asm.
3035 The compiler should make conservative assumptions about what the asm
3036 might contain and should generate its usual stack alignment code in the
3037 prologue if the '``alignstack``' keyword is present:
3039 .. code-block:: llvm
3041 call void asm alignstack "eieio", ""()
3043 Inline asms also support using non-standard assembly dialects. The
3044 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
3045 the inline asm is using the Intel dialect. Currently, ATT and Intel are
3046 the only supported dialects. An example is:
3048 .. code-block:: llvm
3050 call void asm inteldialect "eieio", ""()
3052 If multiple keywords appear the '``sideeffect``' keyword must come
3053 first, the '``alignstack``' keyword second and the '``inteldialect``'
3056 Inline Asm Constraint String
3057 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3059 The constraint list is a comma-separated string, each element containing one or
3060 more constraint codes.
3062 For each element in the constraint list an appropriate register or memory
3063 operand will be chosen, and it will be made available to assembly template
3064 string expansion as ``$0`` for the first constraint in the list, ``$1`` for the
3067 There are three different types of constraints, which are distinguished by a
3068 prefix symbol in front of the constraint code: Output, Input, and Clobber. The
3069 constraints must always be given in that order: outputs first, then inputs, then
3070 clobbers. They cannot be intermingled.
3072 There are also three different categories of constraint codes:
3074 - Register constraint. This is either a register class, or a fixed physical
3075 register. This kind of constraint will allocate a register, and if necessary,
3076 bitcast the argument or result to the appropriate type.
3077 - Memory constraint. This kind of constraint is for use with an instruction
3078 taking a memory operand. Different constraints allow for different addressing
3079 modes used by the target.
3080 - Immediate value constraint. This kind of constraint is for an integer or other
3081 immediate value which can be rendered directly into an instruction. The
3082 various target-specific constraints allow the selection of a value in the
3083 proper range for the instruction you wish to use it with.
3088 Output constraints are specified by an "``=``" prefix (e.g. "``=r``"). This
3089 indicates that the assembly will write to this operand, and the operand will
3090 then be made available as a return value of the ``asm`` expression. Output
3091 constraints do not consume an argument from the call instruction. (Except, see
3092 below about indirect outputs).
3094 Normally, it is expected that no output locations are written to by the assembly
3095 expression until *all* of the inputs have been read. As such, LLVM may assign
3096 the same register to an output and an input. If this is not safe (e.g. if the
3097 assembly contains two instructions, where the first writes to one output, and
3098 the second reads an input and writes to a second output), then the "``&``"
3099 modifier must be used (e.g. "``=&r``") to specify that the output is an
3100 "early-clobber" output. Marking an ouput as "early-clobber" ensures that LLVM
3101 will not use the same register for any inputs (other than an input tied to this
3107 Input constraints do not have a prefix -- just the constraint codes. Each input
3108 constraint will consume one argument from the call instruction. It is not
3109 permitted for the asm to write to any input register or memory location (unless
3110 that input is tied to an output). Note also that multiple inputs may all be
3111 assigned to the same register, if LLVM can determine that they necessarily all
3112 contain the same value.
3114 Instead of providing a Constraint Code, input constraints may also "tie"
3115 themselves to an output constraint, by providing an integer as the constraint
3116 string. Tied inputs still consume an argument from the call instruction, and
3117 take up a position in the asm template numbering as is usual -- they will simply
3118 be constrained to always use the same register as the output they've been tied
3119 to. For example, a constraint string of "``=r,0``" says to assign a register for
3120 output, and use that register as an input as well (it being the 0'th
3123 It is permitted to tie an input to an "early-clobber" output. In that case, no
3124 *other* input may share the same register as the input tied to the early-clobber
3125 (even when the other input has the same value).
3127 You may only tie an input to an output which has a register constraint, not a
3128 memory constraint. Only a single input may be tied to an output.
3130 There is also an "interesting" feature which deserves a bit of explanation: if a
3131 register class constraint allocates a register which is too small for the value
3132 type operand provided as input, the input value will be split into multiple
3133 registers, and all of them passed to the inline asm.
3135 However, this feature is often not as useful as you might think.
3137 Firstly, the registers are *not* guaranteed to be consecutive. So, on those
3138 architectures that have instructions which operate on multiple consecutive
3139 instructions, this is not an appropriate way to support them. (e.g. the 32-bit
3140 SparcV8 has a 64-bit load, which instruction takes a single 32-bit register. The
3141 hardware then loads into both the named register, and the next register. This
3142 feature of inline asm would not be useful to support that.)
3144 A few of the targets provide a template string modifier allowing explicit access
3145 to the second register of a two-register operand (e.g. MIPS ``L``, ``M``, and
3146 ``D``). On such an architecture, you can actually access the second allocated
3147 register (yet, still, not any subsequent ones). But, in that case, you're still
3148 probably better off simply splitting the value into two separate operands, for
3149 clarity. (e.g. see the description of the ``A`` constraint on X86, which,
3150 despite existing only for use with this feature, is not really a good idea to
3153 Indirect inputs and outputs
3154 """""""""""""""""""""""""""
3156 Indirect output or input constraints can be specified by the "``*``" modifier
3157 (which goes after the "``=``" in case of an output). This indicates that the asm
3158 will write to or read from the contents of an *address* provided as an input
3159 argument. (Note that in this way, indirect outputs act more like an *input* than
3160 an output: just like an input, they consume an argument of the call expression,
3161 rather than producing a return value. An indirect output constraint is an
3162 "output" only in that the asm is expected to write to the contents of the input
3163 memory location, instead of just read from it).
3165 This is most typically used for memory constraint, e.g. "``=*m``", to pass the
3166 address of a variable as a value.
3168 It is also possible to use an indirect *register* constraint, but only on output
3169 (e.g. "``=*r``"). This will cause LLVM to allocate a register for an output
3170 value normally, and then, separately emit a store to the address provided as
3171 input, after the provided inline asm. (It's not clear what value this
3172 functionality provides, compared to writing the store explicitly after the asm
3173 statement, and it can only produce worse code, since it bypasses many
3174 optimization passes. I would recommend not using it.)
3180 A clobber constraint is indicated by a "``~``" prefix. A clobber does not
3181 consume an input operand, nor generate an output. Clobbers cannot use any of the
3182 general constraint code letters -- they may use only explicit register
3183 constraints, e.g. "``~{eax}``". The one exception is that a clobber string of
3184 "``~{memory}``" indicates that the assembly writes to arbitrary undeclared
3185 memory locations -- not only the memory pointed to by a declared indirect
3191 After a potential prefix comes constraint code, or codes.
3193 A Constraint Code is either a single letter (e.g. "``r``"), a "``^``" character
3194 followed by two letters (e.g. "``^wc``"), or "``{``" register-name "``}``"
3197 The one and two letter constraint codes are typically chosen to be the same as
3198 GCC's constraint codes.
3200 A single constraint may include one or more than constraint code in it, leaving
3201 it up to LLVM to choose which one to use. This is included mainly for
3202 compatibility with the translation of GCC inline asm coming from clang.
3204 There are two ways to specify alternatives, and either or both may be used in an
3205 inline asm constraint list:
3207 1) Append the codes to each other, making a constraint code set. E.g. "``im``"
3208 or "``{eax}m``". This means "choose any of the options in the set". The
3209 choice of constraint is made independently for each constraint in the
3212 2) Use "``|``" between constraint code sets, creating alternatives. Every
3213 constraint in the constraint list must have the same number of alternative
3214 sets. With this syntax, the same alternative in *all* of the items in the
3215 constraint list will be chosen together.
3217 Putting those together, you might have a two operand constraint string like
3218 ``"rm|r,ri|rm"``. This indicates that if operand 0 is ``r`` or ``m``, then
3219 operand 1 may be one of ``r`` or ``i``. If operand 0 is ``r``, then operand 1
3220 may be one of ``r`` or ``m``. But, operand 0 and 1 cannot both be of type m.
3222 However, the use of either of the alternatives features is *NOT* recommended, as
3223 LLVM is not able to make an intelligent choice about which one to use. (At the
3224 point it currently needs to choose, not enough information is available to do so
3225 in a smart way.) Thus, it simply tries to make a choice that's most likely to
3226 compile, not one that will be optimal performance. (e.g., given "``rm``", it'll
3227 always choose to use memory, not registers). And, if given multiple registers,
3228 or multiple register classes, it will simply choose the first one. (In fact, it
3229 doesn't currently even ensure explicitly specified physical registers are
3230 unique, so specifying multiple physical registers as alternatives, like
3231 ``{r11}{r12},{r11}{r12}``, will assign r11 to both operands, not at all what was
3234 Supported Constraint Code List
3235 """"""""""""""""""""""""""""""
3237 The constraint codes are, in general, expected to behave the same way they do in
3238 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3239 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3240 and GCC likely indicates a bug in LLVM.
3242 Some constraint codes are typically supported by all targets:
3244 - ``r``: A register in the target's general purpose register class.
3245 - ``m``: A memory address operand. It is target-specific what addressing modes
3246 are supported, typical examples are register, or register + register offset,
3247 or register + immediate offset (of some target-specific size).
3248 - ``i``: An integer constant (of target-specific width). Allows either a simple
3249 immediate, or a relocatable value.
3250 - ``n``: An integer constant -- *not* including relocatable values.
3251 - ``s``: An integer constant, but allowing *only* relocatable values.
3252 - ``X``: Allows an operand of any kind, no constraint whatsoever. Typically
3253 useful to pass a label for an asm branch or call.
3255 .. FIXME: but that surely isn't actually okay to jump out of an asm
3256 block without telling llvm about the control transfer???)
3258 - ``{register-name}``: Requires exactly the named physical register.
3260 Other constraints are target-specific:
3264 - ``z``: An immediate integer 0. Outputs ``WZR`` or ``XZR``, as appropriate.
3265 - ``I``: An immediate integer valid for an ``ADD`` or ``SUB`` instruction,
3266 i.e. 0 to 4095 with optional shift by 12.
3267 - ``J``: An immediate integer that, when negated, is valid for an ``ADD`` or
3268 ``SUB`` instruction, i.e. -1 to -4095 with optional left shift by 12.
3269 - ``K``: An immediate integer that is valid for the 'bitmask immediate 32' of a
3270 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 32-bit register.
3271 - ``L``: An immediate integer that is valid for the 'bitmask immediate 64' of a
3272 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 64-bit register.
3273 - ``M``: An immediate integer for use with the ``MOV`` assembly alias on a
3274 32-bit register. This is a superset of ``K``: in addition to the bitmask
3275 immediate, also allows immediate integers which can be loaded with a single
3276 ``MOVZ`` or ``MOVL`` instruction.
3277 - ``N``: An immediate integer for use with the ``MOV`` assembly alias on a
3278 64-bit register. This is a superset of ``L``.
3279 - ``Q``: Memory address operand must be in a single register (no
3280 offsets). (However, LLVM currently does this for the ``m`` constraint as
3282 - ``r``: A 32 or 64-bit integer register (W* or X*).
3283 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register.
3284 - ``x``: A lower 128-bit floating-point/SIMD register (``V0`` to ``V15``).
3288 - ``r``: A 32 or 64-bit integer register.
3289 - ``[0-9]v``: The 32-bit VGPR register, number 0-9.
3290 - ``[0-9]s``: The 32-bit SGPR register, number 0-9.
3295 - ``Q``, ``Um``, ``Un``, ``Uq``, ``Us``, ``Ut``, ``Uv``, ``Uy``: Memory address
3296 operand. Treated the same as operand ``m``, at the moment.
3298 ARM and ARM's Thumb2 mode:
3300 - ``j``: An immediate integer between 0 and 65535 (valid for ``MOVW``)
3301 - ``I``: An immediate integer valid for a data-processing instruction.
3302 - ``J``: An immediate integer between -4095 and 4095.
3303 - ``K``: An immediate integer whose bitwise inverse is valid for a
3304 data-processing instruction. (Can be used with template modifier "``B``" to
3305 print the inverted value).
3306 - ``L``: An immediate integer whose negation is valid for a data-processing
3307 instruction. (Can be used with template modifier "``n``" to print the negated
3309 - ``M``: A power of two or a integer between 0 and 32.
3310 - ``N``: Invalid immediate constraint.
3311 - ``O``: Invalid immediate constraint.
3312 - ``r``: A general-purpose 32-bit integer register (``r0-r15``).
3313 - ``l``: In Thumb2 mode, low 32-bit GPR registers (``r0-r7``). In ARM mode, same
3315 - ``h``: In Thumb2 mode, a high 32-bit GPR register (``r8-r15``). In ARM mode,
3317 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3318 ``d0-d31``, or ``q0-q15``.
3319 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3320 ``d0-d7``, or ``q0-q3``.
3321 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3326 - ``I``: An immediate integer between 0 and 255.
3327 - ``J``: An immediate integer between -255 and -1.
3328 - ``K``: An immediate integer between 0 and 255, with optional left-shift by
3330 - ``L``: An immediate integer between -7 and 7.
3331 - ``M``: An immediate integer which is a multiple of 4 between 0 and 1020.
3332 - ``N``: An immediate integer between 0 and 31.
3333 - ``O``: An immediate integer which is a multiple of 4 between -508 and 508.
3334 - ``r``: A low 32-bit GPR register (``r0-r7``).
3335 - ``l``: A low 32-bit GPR register (``r0-r7``).
3336 - ``h``: A high GPR register (``r0-r7``).
3337 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3338 ``d0-d31``, or ``q0-q15``.
3339 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3340 ``d0-d7``, or ``q0-q3``.
3341 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3347 - ``o``, ``v``: A memory address operand, treated the same as constraint ``m``,
3349 - ``r``: A 32 or 64-bit register.
3353 - ``r``: An 8 or 16-bit register.
3357 - ``I``: An immediate signed 16-bit integer.
3358 - ``J``: An immediate integer zero.
3359 - ``K``: An immediate unsigned 16-bit integer.
3360 - ``L``: An immediate 32-bit integer, where the lower 16 bits are 0.
3361 - ``N``: An immediate integer between -65535 and -1.
3362 - ``O``: An immediate signed 15-bit integer.
3363 - ``P``: An immediate integer between 1 and 65535.
3364 - ``m``: A memory address operand. In MIPS-SE mode, allows a base address
3365 register plus 16-bit immediate offset. In MIPS mode, just a base register.
3366 - ``R``: A memory address operand. In MIPS-SE mode, allows a base address
3367 register plus a 9-bit signed offset. In MIPS mode, the same as constraint
3369 - ``ZC``: A memory address operand, suitable for use in a ``pref``, ``ll``, or
3370 ``sc`` instruction on the given subtarget (details vary).
3371 - ``r``, ``d``, ``y``: A 32 or 64-bit GPR register.
3372 - ``f``: A 32 or 64-bit FPU register (``F0-F31``), or a 128-bit MSA register
3373 (``W0-W31``). In the case of MSA registers, it is recommended to use the ``w``
3374 argument modifier for compatibility with GCC.
3375 - ``c``: A 32-bit or 64-bit GPR register suitable for indirect jump (always
3377 - ``l``: The ``lo`` register, 32 or 64-bit.
3382 - ``b``: A 1-bit integer register.
3383 - ``c`` or ``h``: A 16-bit integer register.
3384 - ``r``: A 32-bit integer register.
3385 - ``l`` or ``N``: A 64-bit integer register.
3386 - ``f``: A 32-bit float register.
3387 - ``d``: A 64-bit float register.
3392 - ``I``: An immediate signed 16-bit integer.
3393 - ``J``: An immediate unsigned 16-bit integer, shifted left 16 bits.
3394 - ``K``: An immediate unsigned 16-bit integer.
3395 - ``L``: An immediate signed 16-bit integer, shifted left 16 bits.
3396 - ``M``: An immediate integer greater than 31.
3397 - ``N``: An immediate integer that is an exact power of 2.
3398 - ``O``: The immediate integer constant 0.
3399 - ``P``: An immediate integer constant whose negation is a signed 16-bit
3401 - ``es``, ``o``, ``Q``, ``Z``, ``Zy``: A memory address operand, currently
3402 treated the same as ``m``.
3403 - ``r``: A 32 or 64-bit integer register.
3404 - ``b``: A 32 or 64-bit integer register, excluding ``R0`` (that is:
3406 - ``f``: A 32 or 64-bit float register (``F0-F31``), or when QPX is enabled, a
3407 128 or 256-bit QPX register (``Q0-Q31``; aliases the ``F`` registers).
3408 - ``v``: For ``4 x f32`` or ``4 x f64`` types, when QPX is enabled, a
3409 128 or 256-bit QPX register (``Q0-Q31``), otherwise a 128-bit
3410 altivec vector register (``V0-V31``).
3412 .. FIXME: is this a bug that v accepts QPX registers? I think this
3413 is supposed to only use the altivec vector registers?
3415 - ``y``: Condition register (``CR0-CR7``).
3416 - ``wc``: An individual CR bit in a CR register.
3417 - ``wa``, ``wd``, ``wf``: Any 128-bit VSX vector register, from the full VSX
3418 register set (overlapping both the floating-point and vector register files).
3419 - ``ws``: A 32 or 64-bit floating point register, from the full VSX register
3424 - ``I``: An immediate 13-bit signed integer.
3425 - ``r``: A 32-bit integer register.
3429 - ``I``: An immediate unsigned 8-bit integer.
3430 - ``J``: An immediate unsigned 12-bit integer.
3431 - ``K``: An immediate signed 16-bit integer.
3432 - ``L``: An immediate signed 20-bit integer.
3433 - ``M``: An immediate integer 0x7fffffff.
3434 - ``Q``, ``R``, ``S``, ``T``: A memory address operand, treated the same as
3435 ``m``, at the moment.
3436 - ``r`` or ``d``: A 32, 64, or 128-bit integer register.
3437 - ``a``: A 32, 64, or 128-bit integer address register (excludes R0, which in an
3438 address context evaluates as zero).
3439 - ``h``: A 32-bit value in the high part of a 64bit data register
3441 - ``f``: A 32, 64, or 128-bit floating point register.
3445 - ``I``: An immediate integer between 0 and 31.
3446 - ``J``: An immediate integer between 0 and 64.
3447 - ``K``: An immediate signed 8-bit integer.
3448 - ``L``: An immediate integer, 0xff or 0xffff or (in 64-bit mode only)
3450 - ``M``: An immediate integer between 0 and 3.
3451 - ``N``: An immediate unsigned 8-bit integer.
3452 - ``O``: An immediate integer between 0 and 127.
3453 - ``e``: An immediate 32-bit signed integer.
3454 - ``Z``: An immediate 32-bit unsigned integer.
3455 - ``o``, ``v``: Treated the same as ``m``, at the moment.
3456 - ``q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3457 ``l`` integer register. On X86-32, this is the ``a``, ``b``, ``c``, and ``d``
3458 registers, and on X86-64, it is all of the integer registers.
3459 - ``Q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3460 ``h`` integer register. This is the ``a``, ``b``, ``c``, and ``d`` registers.
3461 - ``r`` or ``l``: An 8, 16, 32, or 64-bit integer register.
3462 - ``R``: An 8, 16, 32, or 64-bit "legacy" integer register -- one which has
3463 existed since i386, and can be accessed without the REX prefix.
3464 - ``f``: A 32, 64, or 80-bit '387 FPU stack pseudo-register.
3465 - ``y``: A 64-bit MMX register, if MMX is enabled.
3466 - ``x``: If SSE is enabled: a 32 or 64-bit scalar operand, or 128-bit vector
3467 operand in a SSE register. If AVX is also enabled, can also be a 256-bit
3468 vector operand in an AVX register. If AVX-512 is also enabled, can also be a
3469 512-bit vector operand in an AVX512 register, Otherwise, an error.
3470 - ``Y``: The same as ``x``, if *SSE2* is enabled, otherwise an error.
3471 - ``A``: Special case: allocates EAX first, then EDX, for a single operand (in
3472 32-bit mode, a 64-bit integer operand will get split into two registers). It
3473 is not recommended to use this constraint, as in 64-bit mode, the 64-bit
3474 operand will get allocated only to RAX -- if two 32-bit operands are needed,
3475 you're better off splitting it yourself, before passing it to the asm
3480 - ``r``: A 32-bit integer register.
3483 .. _inline-asm-modifiers:
3485 Asm template argument modifiers
3486 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3488 In the asm template string, modifiers can be used on the operand reference, like
3491 The modifiers are, in general, expected to behave the same way they do in
3492 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3493 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3494 and GCC likely indicates a bug in LLVM.
3498 - ``c``: Print an immediate integer constant unadorned, without
3499 the target-specific immediate punctuation (e.g. no ``$`` prefix).
3500 - ``n``: Negate and print immediate integer constant unadorned, without the
3501 target-specific immediate punctuation (e.g. no ``$`` prefix).
3502 - ``l``: Print as an unadorned label, without the target-specific label
3503 punctuation (e.g. no ``$`` prefix).
3507 - ``w``: Print a GPR register with a ``w*`` name instead of ``x*`` name. E.g.,
3508 instead of ``x30``, print ``w30``.
3509 - ``x``: Print a GPR register with a ``x*`` name. (this is the default, anyhow).
3510 - ``b``, ``h``, ``s``, ``d``, ``q``: Print a floating-point/SIMD register with a
3511 ``b*``, ``h*``, ``s*``, ``d*``, or ``q*`` name, rather than the default of
3520 - ``a``: Print an operand as an address (with ``[`` and ``]`` surrounding a
3524 - ``y``: Print a VFP single-precision register as an indexed double (e.g. print
3525 as ``d4[1]`` instead of ``s9``)
3526 - ``B``: Bitwise invert and print an immediate integer constant without ``#``
3528 - ``L``: Print the low 16-bits of an immediate integer constant.
3529 - ``M``: Print as a register set suitable for ldm/stm. Also prints *all*
3530 register operands subsequent to the specified one (!), so use carefully.
3531 - ``Q``: Print the low-order register of a register-pair, or the low-order
3532 register of a two-register operand.
3533 - ``R``: Print the high-order register of a register-pair, or the high-order
3534 register of a two-register operand.
3535 - ``H``: Print the second register of a register-pair. (On a big-endian system,
3536 ``H`` is equivalent to ``Q``, and on little-endian system, ``H`` is equivalent
3539 .. FIXME: H doesn't currently support printing the second register
3540 of a two-register operand.
3542 - ``e``: Print the low doubleword register of a NEON quad register.
3543 - ``f``: Print the high doubleword register of a NEON quad register.
3544 - ``m``: Print the base register of a memory operand without the ``[`` and ``]``
3549 - ``L``: Print the second register of a two-register operand. Requires that it
3550 has been allocated consecutively to the first.
3552 .. FIXME: why is it restricted to consecutive ones? And there's
3553 nothing that ensures that happens, is there?
3555 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3556 nothing. Used to print 'addi' vs 'add' instructions.
3560 No additional modifiers.
3564 - ``X``: Print an immediate integer as hexadecimal
3565 - ``x``: Print the low 16 bits of an immediate integer as hexadecimal.
3566 - ``d``: Print an immediate integer as decimal.
3567 - ``m``: Subtract one and print an immediate integer as decimal.
3568 - ``z``: Print $0 if an immediate zero, otherwise print normally.
3569 - ``L``: Print the low-order register of a two-register operand, or prints the
3570 address of the low-order word of a double-word memory operand.
3572 .. FIXME: L seems to be missing memory operand support.
3574 - ``M``: Print the high-order register of a two-register operand, or prints the
3575 address of the high-order word of a double-word memory operand.
3577 .. FIXME: M seems to be missing memory operand support.
3579 - ``D``: Print the second register of a two-register operand, or prints the
3580 second word of a double-word memory operand. (On a big-endian system, ``D`` is
3581 equivalent to ``L``, and on little-endian system, ``D`` is equivalent to
3583 - ``w``: No effect. Provided for compatibility with GCC which requires this
3584 modifier in order to print MSA registers (``W0-W31``) with the ``f``
3593 - ``L``: Print the second register of a two-register operand. Requires that it
3594 has been allocated consecutively to the first.
3596 .. FIXME: why is it restricted to consecutive ones? And there's
3597 nothing that ensures that happens, is there?
3599 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3600 nothing. Used to print 'addi' vs 'add' instructions.
3601 - ``y``: For a memory operand, prints formatter for a two-register X-form
3602 instruction. (Currently always prints ``r0,OPERAND``).
3603 - ``U``: Prints 'u' if the memory operand is an update form, and nothing
3604 otherwise. (NOTE: LLVM does not support update form, so this will currently
3605 always print nothing)
3606 - ``X``: Prints 'x' if the memory operand is an indexed form. (NOTE: LLVM does
3607 not support indexed form, so this will currently always print nothing)
3615 SystemZ implements only ``n``, and does *not* support any of the other
3616 target-independent modifiers.
3620 - ``c``: Print an unadorned integer or symbol name. (The latter is
3621 target-specific behavior for this typically target-independent modifier).
3622 - ``A``: Print a register name with a '``*``' before it.
3623 - ``b``: Print an 8-bit register name (e.g. ``al``); do nothing on a memory
3625 - ``h``: Print the upper 8-bit register name (e.g. ``ah``); do nothing on a
3627 - ``w``: Print the 16-bit register name (e.g. ``ax``); do nothing on a memory
3629 - ``k``: Print the 32-bit register name (e.g. ``eax``); do nothing on a memory
3631 - ``q``: Print the 64-bit register name (e.g. ``rax``), if 64-bit registers are
3632 available, otherwise the 32-bit register name; do nothing on a memory operand.
3633 - ``n``: Negate and print an unadorned integer, or, for operands other than an
3634 immediate integer (e.g. a relocatable symbol expression), print a '-' before
3635 the operand. (The behavior for relocatable symbol expressions is a
3636 target-specific behavior for this typically target-independent modifier)
3637 - ``H``: Print a memory reference with additional offset +8.
3638 - ``P``: Print a memory reference or operand for use as the argument of a call
3639 instruction. (E.g. omit ``(rip)``, even though it's PC-relative.)
3643 No additional modifiers.
3649 The call instructions that wrap inline asm nodes may have a
3650 "``!srcloc``" MDNode attached to it that contains a list of constant
3651 integers. If present, the code generator will use the integer as the
3652 location cookie value when report errors through the ``LLVMContext``
3653 error reporting mechanisms. This allows a front-end to correlate backend
3654 errors that occur with inline asm back to the source code that produced
3657 .. code-block:: llvm
3659 call void asm sideeffect "something bad", ""(), !srcloc !42
3661 !42 = !{ i32 1234567 }
3663 It is up to the front-end to make sense of the magic numbers it places
3664 in the IR. If the MDNode contains multiple constants, the code generator
3665 will use the one that corresponds to the line of the asm that the error
3673 LLVM IR allows metadata to be attached to instructions in the program
3674 that can convey extra information about the code to the optimizers and
3675 code generator. One example application of metadata is source-level
3676 debug information. There are two metadata primitives: strings and nodes.
3678 Metadata does not have a type, and is not a value. If referenced from a
3679 ``call`` instruction, it uses the ``metadata`` type.
3681 All metadata are identified in syntax by a exclamation point ('``!``').
3683 .. _metadata-string:
3685 Metadata Nodes and Metadata Strings
3686 -----------------------------------
3688 A metadata string is a string surrounded by double quotes. It can
3689 contain any character by escaping non-printable characters with
3690 "``\xx``" where "``xx``" is the two digit hex code. For example:
3693 Metadata nodes are represented with notation similar to structure
3694 constants (a comma separated list of elements, surrounded by braces and
3695 preceded by an exclamation point). Metadata nodes can have any values as
3696 their operand. For example:
3698 .. code-block:: llvm
3700 !{ !"test\00", i32 10}
3702 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
3704 .. code-block:: llvm
3706 !0 = distinct !{!"test\00", i32 10}
3708 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
3709 content. They can also occur when transformations cause uniquing collisions
3710 when metadata operands change.
3712 A :ref:`named metadata <namedmetadatastructure>` is a collection of
3713 metadata nodes, which can be looked up in the module symbol table. For
3716 .. code-block:: llvm
3720 Metadata can be used as function arguments. Here ``llvm.dbg.value``
3721 function is using two metadata arguments:
3723 .. code-block:: llvm
3725 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
3727 Metadata can be attached to an instruction. Here metadata ``!21`` is attached
3728 to the ``add`` instruction using the ``!dbg`` identifier:
3730 .. code-block:: llvm
3732 %indvar.next = add i64 %indvar, 1, !dbg !21
3734 Metadata can also be attached to a function definition. Here metadata ``!22``
3735 is attached to the ``foo`` function using the ``!dbg`` identifier:
3737 .. code-block:: llvm
3739 define void @foo() !dbg !22 {
3743 More information about specific metadata nodes recognized by the
3744 optimizers and code generator is found below.
3746 .. _specialized-metadata:
3748 Specialized Metadata Nodes
3749 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3751 Specialized metadata nodes are custom data structures in metadata (as opposed
3752 to generic tuples). Their fields are labelled, and can be specified in any
3755 These aren't inherently debug info centric, but currently all the specialized
3756 metadata nodes are related to debug info.
3763 ``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
3764 ``retainedTypes:``, ``subprograms:``, ``globals:``, ``imports:`` and ``macros:``
3765 fields are tuples containing the debug info to be emitted along with the compile
3766 unit, regardless of code optimizations (some nodes are only emitted if there are
3767 references to them from instructions).
3769 .. code-block:: llvm
3771 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
3772 isOptimized: true, flags: "-O2", runtimeVersion: 2,
3773 splitDebugFilename: "abc.debug", emissionKind: 1,
3774 enums: !2, retainedTypes: !3, subprograms: !4,
3775 globals: !5, imports: !6, macros: !7, dwoId: 0x0abcd)
3777 Compile unit descriptors provide the root scope for objects declared in a
3778 specific compilation unit. File descriptors are defined using this scope.
3779 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
3780 keep track of subprograms, global variables, type information, and imported
3781 entities (declarations and namespaces).
3788 ``DIFile`` nodes represent files. The ``filename:`` can include slashes.
3790 .. code-block:: llvm
3792 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
3794 Files are sometimes used in ``scope:`` fields, and are the only valid target
3795 for ``file:`` fields.
3802 ``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
3803 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
3805 .. code-block:: llvm
3807 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3808 encoding: DW_ATE_unsigned_char)
3809 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
3811 The ``encoding:`` describes the details of the type. Usually it's one of the
3814 .. code-block:: llvm
3820 DW_ATE_signed_char = 6
3822 DW_ATE_unsigned_char = 8
3824 .. _DISubroutineType:
3829 ``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
3830 refers to a tuple; the first operand is the return type, while the rest are the
3831 types of the formal arguments in order. If the first operand is ``null``, that
3832 represents a function with no return value (such as ``void foo() {}`` in C++).
3834 .. code-block:: llvm
3836 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
3837 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
3838 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
3845 ``DIDerivedType`` nodes represent types derived from other types, such as
3848 .. code-block:: llvm
3850 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3851 encoding: DW_ATE_unsigned_char)
3852 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
3855 The following ``tag:`` values are valid:
3857 .. code-block:: llvm
3859 DW_TAG_formal_parameter = 5
3861 DW_TAG_pointer_type = 15
3862 DW_TAG_reference_type = 16
3864 DW_TAG_ptr_to_member_type = 31
3865 DW_TAG_const_type = 38
3866 DW_TAG_volatile_type = 53
3867 DW_TAG_restrict_type = 55
3869 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
3870 <DICompositeType>` or :ref:`subprogram <DISubprogram>`. The type of the member
3871 is the ``baseType:``. The ``offset:`` is the member's bit offset.
3872 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
3873 argument of a subprogram.
3875 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3877 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3878 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3881 Note that the ``void *`` type is expressed as a type derived from NULL.
3883 .. _DICompositeType:
3888 ``DICompositeType`` nodes represent types composed of other types, like
3889 structures and unions. ``elements:`` points to a tuple of the composed types.
3891 If the source language supports ODR, the ``identifier:`` field gives the unique
3892 identifier used for type merging between modules. When specified, other types
3893 can refer to composite types indirectly via a :ref:`metadata string
3894 <metadata-string>` that matches their identifier.
3896 .. code-block:: llvm
3898 !0 = !DIEnumerator(name: "SixKind", value: 7)
3899 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3900 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3901 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
3902 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3903 elements: !{!0, !1, !2})
3905 The following ``tag:`` values are valid:
3907 .. code-block:: llvm
3909 DW_TAG_array_type = 1
3910 DW_TAG_class_type = 2
3911 DW_TAG_enumeration_type = 4
3912 DW_TAG_structure_type = 19
3913 DW_TAG_union_type = 23
3914 DW_TAG_subroutine_type = 21
3915 DW_TAG_inheritance = 28
3918 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
3919 descriptors <DISubrange>`, each representing the range of subscripts at that
3920 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3921 array type is a native packed vector.
3923 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
3924 descriptors <DIEnumerator>`, each representing the definition of an enumeration
3925 value for the set. All enumeration type descriptors are collected in the
3926 ``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
3928 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3929 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
3930 <DIDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
3937 ``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3938 :ref:`DICompositeType`. ``count: -1`` indicates an empty array.
3940 .. code-block:: llvm
3942 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
3943 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
3944 !2 = !DISubrange(count: -1) ; empty array.
3951 ``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3952 variants of :ref:`DICompositeType`.
3954 .. code-block:: llvm
3956 !0 = !DIEnumerator(name: "SixKind", value: 7)
3957 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3958 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3960 DITemplateTypeParameter
3961 """""""""""""""""""""""
3963 ``DITemplateTypeParameter`` nodes represent type parameters to generic source
3964 language constructs. They are used (optionally) in :ref:`DICompositeType` and
3965 :ref:`DISubprogram` ``templateParams:`` fields.
3967 .. code-block:: llvm
3969 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
3971 DITemplateValueParameter
3972 """"""""""""""""""""""""
3974 ``DITemplateValueParameter`` nodes represent value parameters to generic source
3975 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3976 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3977 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
3978 :ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
3980 .. code-block:: llvm
3982 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
3987 ``DINamespace`` nodes represent namespaces in the source language.
3989 .. code-block:: llvm
3991 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
3996 ``DIGlobalVariable`` nodes represent global variables in the source language.
3998 .. code-block:: llvm
4000 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
4001 file: !2, line: 7, type: !3, isLocal: true,
4002 isDefinition: false, variable: i32* @foo,
4005 All global variables should be referenced by the `globals:` field of a
4006 :ref:`compile unit <DICompileUnit>`.
4013 ``DISubprogram`` nodes represent functions from the source language. A
4014 ``DISubprogram`` may be attached to a function definition using ``!dbg``
4015 metadata. The ``variables:`` field points at :ref:`variables <DILocalVariable>`
4016 that must be retained, even if their IR counterparts are optimized out of
4017 the IR. The ``type:`` field must point at an :ref:`DISubroutineType`.
4019 .. code-block:: llvm
4021 define void @_Z3foov() !dbg !0 {
4025 !0 = distinct !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
4026 file: !2, line: 7, type: !3, isLocal: true,
4027 isDefinition: false, scopeLine: 8,
4029 virtuality: DW_VIRTUALITY_pure_virtual,
4030 virtualIndex: 10, flags: DIFlagPrototyped,
4031 isOptimized: true, templateParams: !5,
4032 declaration: !6, variables: !7)
4039 ``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
4040 <DISubprogram>`. The line number and column numbers are used to distinguish
4041 two lexical blocks at same depth. They are valid targets for ``scope:``
4044 .. code-block:: llvm
4046 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
4048 Usually lexical blocks are ``distinct`` to prevent node merging based on
4051 .. _DILexicalBlockFile:
4056 ``DILexicalBlockFile`` nodes are used to discriminate between sections of a
4057 :ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
4058 indicate textual inclusion, or the ``discriminator:`` field can be used to
4059 discriminate between control flow within a single block in the source language.
4061 .. code-block:: llvm
4063 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
4064 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
4065 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
4072 ``DILocation`` nodes represent source debug locations. The ``scope:`` field is
4073 mandatory, and points at an :ref:`DILexicalBlockFile`, an
4074 :ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
4076 .. code-block:: llvm
4078 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
4080 .. _DILocalVariable:
4085 ``DILocalVariable`` nodes represent local variables in the source language. If
4086 the ``arg:`` field is set to non-zero, then this variable is a subprogram
4087 parameter, and it will be included in the ``variables:`` field of its
4088 :ref:`DISubprogram`.
4090 .. code-block:: llvm
4092 !0 = !DILocalVariable(name: "this", arg: 1, scope: !3, file: !2, line: 7,
4093 type: !3, flags: DIFlagArtificial)
4094 !1 = !DILocalVariable(name: "x", arg: 2, scope: !4, file: !2, line: 7,
4096 !2 = !DILocalVariable(name: "y", scope: !5, file: !2, line: 7, type: !3)
4101 ``DIExpression`` nodes represent DWARF expression sequences. They are used in
4102 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
4103 describe how the referenced LLVM variable relates to the source language
4106 The current supported vocabulary is limited:
4108 - ``DW_OP_deref`` dereferences the working expression.
4109 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
4110 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
4111 here, respectively) of the variable piece from the working expression.
4113 .. code-block:: llvm
4115 !0 = !DIExpression(DW_OP_deref)
4116 !1 = !DIExpression(DW_OP_plus, 3)
4117 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
4118 !3 = !DIExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
4123 ``DIObjCProperty`` nodes represent Objective-C property nodes.
4125 .. code-block:: llvm
4127 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
4128 getter: "getFoo", attributes: 7, type: !2)
4133 ``DIImportedEntity`` nodes represent entities (such as modules) imported into a
4136 .. code-block:: llvm
4138 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
4139 entity: !1, line: 7)
4144 ``DIMacro`` nodes represent definition or undefinition of a macro identifiers.
4145 The ``name:`` field is the macro identifier, followed by macro parameters when
4146 definining a function-like macro, and the ``value`` field is the token-string
4147 used to expand the macro identifier.
4149 .. code-block:: llvm
4151 !2 = !DIMacro(macinfo: DW_MACINFO_define, line: 7, name: "foo(x)",
4153 !3 = !DIMacro(macinfo: DW_MACINFO_undef, line: 30, name: "foo")
4158 ``DIMacroFile`` nodes represent inclusion of source files.
4159 The ``nodes:`` field is a list of ``DIMacro`` and ``DIMacroFile`` nodes that
4160 appear in the included source file.
4162 .. code-block:: llvm
4164 !2 = !DIMacroFile(macinfo: DW_MACINFO_start_file, line: 7, file: !2,
4170 In LLVM IR, memory does not have types, so LLVM's own type system is not
4171 suitable for doing TBAA. Instead, metadata is added to the IR to
4172 describe a type system of a higher level language. This can be used to
4173 implement typical C/C++ TBAA, but it can also be used to implement
4174 custom alias analysis behavior for other languages.
4176 The current metadata format is very simple. TBAA metadata nodes have up
4177 to three fields, e.g.:
4179 .. code-block:: llvm
4181 !0 = !{ !"an example type tree" }
4182 !1 = !{ !"int", !0 }
4183 !2 = !{ !"float", !0 }
4184 !3 = !{ !"const float", !2, i64 1 }
4186 The first field is an identity field. It can be any value, usually a
4187 metadata string, which uniquely identifies the type. The most important
4188 name in the tree is the name of the root node. Two trees with different
4189 root node names are entirely disjoint, even if they have leaves with
4192 The second field identifies the type's parent node in the tree, or is
4193 null or omitted for a root node. A type is considered to alias all of
4194 its descendants and all of its ancestors in the tree. Also, a type is
4195 considered to alias all types in other trees, so that bitcode produced
4196 from multiple front-ends is handled conservatively.
4198 If the third field is present, it's an integer which if equal to 1
4199 indicates that the type is "constant" (meaning
4200 ``pointsToConstantMemory`` should return true; see `other useful
4201 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
4203 '``tbaa.struct``' Metadata
4204 ^^^^^^^^^^^^^^^^^^^^^^^^^^
4206 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
4207 aggregate assignment operations in C and similar languages, however it
4208 is defined to copy a contiguous region of memory, which is more than
4209 strictly necessary for aggregate types which contain holes due to
4210 padding. Also, it doesn't contain any TBAA information about the fields
4213 ``!tbaa.struct`` metadata can describe which memory subregions in a
4214 memcpy are padding and what the TBAA tags of the struct are.
4216 The current metadata format is very simple. ``!tbaa.struct`` metadata
4217 nodes are a list of operands which are in conceptual groups of three.
4218 For each group of three, the first operand gives the byte offset of a
4219 field in bytes, the second gives its size in bytes, and the third gives
4222 .. code-block:: llvm
4224 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
4226 This describes a struct with two fields. The first is at offset 0 bytes
4227 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
4228 and has size 4 bytes and has tbaa tag !2.
4230 Note that the fields need not be contiguous. In this example, there is a
4231 4 byte gap between the two fields. This gap represents padding which
4232 does not carry useful data and need not be preserved.
4234 '``noalias``' and '``alias.scope``' Metadata
4235 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4237 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
4238 noalias memory-access sets. This means that some collection of memory access
4239 instructions (loads, stores, memory-accessing calls, etc.) that carry
4240 ``noalias`` metadata can specifically be specified not to alias with some other
4241 collection of memory access instructions that carry ``alias.scope`` metadata.
4242 Each type of metadata specifies a list of scopes where each scope has an id and
4243 a domain. When evaluating an aliasing query, if for some domain, the set
4244 of scopes with that domain in one instruction's ``alias.scope`` list is a
4245 subset of (or equal to) the set of scopes for that domain in another
4246 instruction's ``noalias`` list, then the two memory accesses are assumed not to
4249 The metadata identifying each domain is itself a list containing one or two
4250 entries. The first entry is the name of the domain. Note that if the name is a
4251 string then it can be combined across functions and translation units. A
4252 self-reference can be used to create globally unique domain names. A
4253 descriptive string may optionally be provided as a second list entry.
4255 The metadata identifying each scope is also itself a list containing two or
4256 three entries. The first entry is the name of the scope. Note that if the name
4257 is a string then it can be combined across functions and translation units. A
4258 self-reference can be used to create globally unique scope names. A metadata
4259 reference to the scope's domain is the second entry. A descriptive string may
4260 optionally be provided as a third list entry.
4264 .. code-block:: llvm
4266 ; Two scope domains:
4270 ; Some scopes in these domains:
4276 !5 = !{!4} ; A list containing only scope !4
4280 ; These two instructions don't alias:
4281 %0 = load float, float* %c, align 4, !alias.scope !5
4282 store float %0, float* %arrayidx.i, align 4, !noalias !5
4284 ; These two instructions also don't alias (for domain !1, the set of scopes
4285 ; in the !alias.scope equals that in the !noalias list):
4286 %2 = load float, float* %c, align 4, !alias.scope !5
4287 store float %2, float* %arrayidx.i2, align 4, !noalias !6
4289 ; These two instructions may alias (for domain !0, the set of scopes in
4290 ; the !noalias list is not a superset of, or equal to, the scopes in the
4291 ; !alias.scope list):
4292 %2 = load float, float* %c, align 4, !alias.scope !6
4293 store float %0, float* %arrayidx.i, align 4, !noalias !7
4295 '``fpmath``' Metadata
4296 ^^^^^^^^^^^^^^^^^^^^^
4298 ``fpmath`` metadata may be attached to any instruction of floating point
4299 type. It can be used to express the maximum acceptable error in the
4300 result of that instruction, in ULPs, thus potentially allowing the
4301 compiler to use a more efficient but less accurate method of computing
4302 it. ULP is defined as follows:
4304 If ``x`` is a real number that lies between two finite consecutive
4305 floating-point numbers ``a`` and ``b``, without being equal to one
4306 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
4307 distance between the two non-equal finite floating-point numbers
4308 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
4310 The metadata node shall consist of a single positive floating point
4311 number representing the maximum relative error, for example:
4313 .. code-block:: llvm
4315 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
4319 '``range``' Metadata
4320 ^^^^^^^^^^^^^^^^^^^^
4322 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
4323 integer types. It expresses the possible ranges the loaded value or the value
4324 returned by the called function at this call site is in. The ranges are
4325 represented with a flattened list of integers. The loaded value or the value
4326 returned is known to be in the union of the ranges defined by each consecutive
4327 pair. Each pair has the following properties:
4329 - The type must match the type loaded by the instruction.
4330 - The pair ``a,b`` represents the range ``[a,b)``.
4331 - Both ``a`` and ``b`` are constants.
4332 - The range is allowed to wrap.
4333 - The range should not represent the full or empty set. That is,
4336 In addition, the pairs must be in signed order of the lower bound and
4337 they must be non-contiguous.
4341 .. code-block:: llvm
4343 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
4344 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
4345 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
4346 %d = invoke i8 @bar() to label %cont
4347 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
4349 !0 = !{ i8 0, i8 2 }
4350 !1 = !{ i8 255, i8 2 }
4351 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
4352 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
4354 '``unpredictable``' Metadata
4355 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4357 ``unpredictable`` metadata may be attached to any branch or switch
4358 instruction. It can be used to express the unpredictability of control
4359 flow. Similar to the llvm.expect intrinsic, it may be used to alter
4360 optimizations related to compare and branch instructions. The metadata
4361 is treated as a boolean value; if it exists, it signals that the branch
4362 or switch that it is attached to is completely unpredictable.
4367 It is sometimes useful to attach information to loop constructs. Currently,
4368 loop metadata is implemented as metadata attached to the branch instruction
4369 in the loop latch block. This type of metadata refer to a metadata node that is
4370 guaranteed to be separate for each loop. The loop identifier metadata is
4371 specified with the name ``llvm.loop``.
4373 The loop identifier metadata is implemented using a metadata that refers to
4374 itself to avoid merging it with any other identifier metadata, e.g.,
4375 during module linkage or function inlining. That is, each loop should refer
4376 to their own identification metadata even if they reside in separate functions.
4377 The following example contains loop identifier metadata for two separate loop
4380 .. code-block:: llvm
4385 The loop identifier metadata can be used to specify additional
4386 per-loop metadata. Any operands after the first operand can be treated
4387 as user-defined metadata. For example the ``llvm.loop.unroll.count``
4388 suggests an unroll factor to the loop unroller:
4390 .. code-block:: llvm
4392 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
4395 !1 = !{!"llvm.loop.unroll.count", i32 4}
4397 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
4398 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4400 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
4401 used to control per-loop vectorization and interleaving parameters such as
4402 vectorization width and interleave count. These metadata should be used in
4403 conjunction with ``llvm.loop`` loop identification metadata. The
4404 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
4405 optimization hints and the optimizer will only interleave and vectorize loops if
4406 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
4407 which contains information about loop-carried memory dependencies can be helpful
4408 in determining the safety of these transformations.
4410 '``llvm.loop.interleave.count``' Metadata
4411 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4413 This metadata suggests an interleave count to the loop interleaver.
4414 The first operand is the string ``llvm.loop.interleave.count`` and the
4415 second operand is an integer specifying the interleave count. For
4418 .. code-block:: llvm
4420 !0 = !{!"llvm.loop.interleave.count", i32 4}
4422 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
4423 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
4424 then the interleave count will be determined automatically.
4426 '``llvm.loop.vectorize.enable``' Metadata
4427 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4429 This metadata selectively enables or disables vectorization for the loop. The
4430 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
4431 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
4432 0 disables vectorization:
4434 .. code-block:: llvm
4436 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
4437 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
4439 '``llvm.loop.vectorize.width``' Metadata
4440 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4442 This metadata sets the target width of the vectorizer. The first
4443 operand is the string ``llvm.loop.vectorize.width`` and the second
4444 operand is an integer specifying the width. For example:
4446 .. code-block:: llvm
4448 !0 = !{!"llvm.loop.vectorize.width", i32 4}
4450 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
4451 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
4452 0 or if the loop does not have this metadata the width will be
4453 determined automatically.
4455 '``llvm.loop.unroll``'
4456 ^^^^^^^^^^^^^^^^^^^^^^
4458 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
4459 optimization hints such as the unroll factor. ``llvm.loop.unroll``
4460 metadata should be used in conjunction with ``llvm.loop`` loop
4461 identification metadata. The ``llvm.loop.unroll`` metadata are only
4462 optimization hints and the unrolling will only be performed if the
4463 optimizer believes it is safe to do so.
4465 '``llvm.loop.unroll.count``' Metadata
4466 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4468 This metadata suggests an unroll factor to the loop unroller. The
4469 first operand is the string ``llvm.loop.unroll.count`` and the second
4470 operand is a positive integer specifying the unroll factor. For
4473 .. code-block:: llvm
4475 !0 = !{!"llvm.loop.unroll.count", i32 4}
4477 If the trip count of the loop is less than the unroll count the loop
4478 will be partially unrolled.
4480 '``llvm.loop.unroll.disable``' Metadata
4481 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4483 This metadata disables loop unrolling. The metadata has a single operand
4484 which is the string ``llvm.loop.unroll.disable``. For example:
4486 .. code-block:: llvm
4488 !0 = !{!"llvm.loop.unroll.disable"}
4490 '``llvm.loop.unroll.runtime.disable``' Metadata
4491 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4493 This metadata disables runtime loop unrolling. The metadata has a single
4494 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
4496 .. code-block:: llvm
4498 !0 = !{!"llvm.loop.unroll.runtime.disable"}
4500 '``llvm.loop.unroll.enable``' Metadata
4501 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4503 This metadata suggests that the loop should be fully unrolled if the trip count
4504 is known at compile time and partially unrolled if the trip count is not known
4505 at compile time. The metadata has a single operand which is the string
4506 ``llvm.loop.unroll.enable``. For example:
4508 .. code-block:: llvm
4510 !0 = !{!"llvm.loop.unroll.enable"}
4512 '``llvm.loop.unroll.full``' Metadata
4513 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4515 This metadata suggests that the loop should be unrolled fully. The
4516 metadata has a single operand which is the string ``llvm.loop.unroll.full``.
4519 .. code-block:: llvm
4521 !0 = !{!"llvm.loop.unroll.full"}
4526 Metadata types used to annotate memory accesses with information helpful
4527 for optimizations are prefixed with ``llvm.mem``.
4529 '``llvm.mem.parallel_loop_access``' Metadata
4530 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4532 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
4533 or metadata containing a list of loop identifiers for nested loops.
4534 The metadata is attached to memory accessing instructions and denotes that
4535 no loop carried memory dependence exist between it and other instructions denoted
4536 with the same loop identifier.
4538 Precisely, given two instructions ``m1`` and ``m2`` that both have the
4539 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
4540 set of loops associated with that metadata, respectively, then there is no loop
4541 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
4544 As a special case, if all memory accessing instructions in a loop have
4545 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
4546 loop has no loop carried memory dependences and is considered to be a parallel
4549 Note that if not all memory access instructions have such metadata referring to
4550 the loop, then the loop is considered not being trivially parallel. Additional
4551 memory dependence analysis is required to make that determination. As a fail
4552 safe mechanism, this causes loops that were originally parallel to be considered
4553 sequential (if optimization passes that are unaware of the parallel semantics
4554 insert new memory instructions into the loop body).
4556 Example of a loop that is considered parallel due to its correct use of
4557 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
4558 metadata types that refer to the same loop identifier metadata.
4560 .. code-block:: llvm
4564 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
4566 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4568 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
4574 It is also possible to have nested parallel loops. In that case the
4575 memory accesses refer to a list of loop identifier metadata nodes instead of
4576 the loop identifier metadata node directly:
4578 .. code-block:: llvm
4582 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
4584 br label %inner.for.body
4588 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4590 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
4592 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
4596 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
4598 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
4600 outer.for.end: ; preds = %for.body
4602 !0 = !{!1, !2} ; a list of loop identifiers
4603 !1 = !{!1} ; an identifier for the inner loop
4604 !2 = !{!2} ; an identifier for the outer loop
4609 The ``llvm.bitsets`` global metadata is used to implement
4610 :doc:`bitsets <BitSets>`.
4612 '``invariant.group``' Metadata
4613 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4615 The ``invariant.group`` metadata may be attached to ``load``/``store`` instructions.
4616 The existence of the ``invariant.group`` metadata on the instruction tells
4617 the optimizer that every ``load`` and ``store`` to the same pointer operand
4618 within the same invariant group can be assumed to load or store the same
4619 value (but see the ``llvm.invariant.group.barrier`` intrinsic which affects
4620 when two pointers are considered the same).
4624 .. code-block:: llvm
4626 @unknownPtr = external global i8
4629 store i8 42, i8* %ptr, !invariant.group !0
4630 call void @foo(i8* %ptr)
4632 %a = load i8, i8* %ptr, !invariant.group !0 ; Can assume that value under %ptr didn't change
4633 call void @foo(i8* %ptr)
4634 %b = load i8, i8* %ptr, !invariant.group !1 ; Can't assume anything, because group changed
4636 %newPtr = call i8* @getPointer(i8* %ptr)
4637 %c = load i8, i8* %newPtr, !invariant.group !0 ; Can't assume anything, because we only have information about %ptr
4639 %unknownValue = load i8, i8* @unknownPtr
4640 store i8 %unknownValue, i8* %ptr, !invariant.group !0 ; Can assume that %unknownValue == 42
4642 call void @foo(i8* %ptr)
4643 %newPtr2 = call i8* @llvm.invariant.group.barrier(i8* %ptr)
4644 %d = load i8, i8* %newPtr2, !invariant.group !0 ; Can't step through invariant.group.barrier to get value of %ptr
4647 declare void @foo(i8*)
4648 declare i8* @getPointer(i8*)
4649 declare i8* @llvm.invariant.group.barrier(i8*)
4651 !0 = !{!"magic ptr"}
4652 !1 = !{!"other ptr"}
4656 Module Flags Metadata
4657 =====================
4659 Information about the module as a whole is difficult to convey to LLVM's
4660 subsystems. The LLVM IR isn't sufficient to transmit this information.
4661 The ``llvm.module.flags`` named metadata exists in order to facilitate
4662 this. These flags are in the form of key / value pairs --- much like a
4663 dictionary --- making it easy for any subsystem who cares about a flag to
4666 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
4667 Each triplet has the following form:
4669 - The first element is a *behavior* flag, which specifies the behavior
4670 when two (or more) modules are merged together, and it encounters two
4671 (or more) metadata with the same ID. The supported behaviors are
4673 - The second element is a metadata string that is a unique ID for the
4674 metadata. Each module may only have one flag entry for each unique ID (not
4675 including entries with the **Require** behavior).
4676 - The third element is the value of the flag.
4678 When two (or more) modules are merged together, the resulting
4679 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
4680 each unique metadata ID string, there will be exactly one entry in the merged
4681 modules ``llvm.module.flags`` metadata table, and the value for that entry will
4682 be determined by the merge behavior flag, as described below. The only exception
4683 is that entries with the *Require* behavior are always preserved.
4685 The following behaviors are supported:
4696 Emits an error if two values disagree, otherwise the resulting value
4697 is that of the operands.
4701 Emits a warning if two values disagree. The result value will be the
4702 operand for the flag from the first module being linked.
4706 Adds a requirement that another module flag be present and have a
4707 specified value after linking is performed. The value must be a
4708 metadata pair, where the first element of the pair is the ID of the
4709 module flag to be restricted, and the second element of the pair is
4710 the value the module flag should be restricted to. This behavior can
4711 be used to restrict the allowable results (via triggering of an
4712 error) of linking IDs with the **Override** behavior.
4716 Uses the specified value, regardless of the behavior or value of the
4717 other module. If both modules specify **Override**, but the values
4718 differ, an error will be emitted.
4722 Appends the two values, which are required to be metadata nodes.
4726 Appends the two values, which are required to be metadata
4727 nodes. However, duplicate entries in the second list are dropped
4728 during the append operation.
4730 It is an error for a particular unique flag ID to have multiple behaviors,
4731 except in the case of **Require** (which adds restrictions on another metadata
4732 value) or **Override**.
4734 An example of module flags:
4736 .. code-block:: llvm
4738 !0 = !{ i32 1, !"foo", i32 1 }
4739 !1 = !{ i32 4, !"bar", i32 37 }
4740 !2 = !{ i32 2, !"qux", i32 42 }
4741 !3 = !{ i32 3, !"qux",
4746 !llvm.module.flags = !{ !0, !1, !2, !3 }
4748 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
4749 if two or more ``!"foo"`` flags are seen is to emit an error if their
4750 values are not equal.
4752 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
4753 behavior if two or more ``!"bar"`` flags are seen is to use the value
4756 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
4757 behavior if two or more ``!"qux"`` flags are seen is to emit a
4758 warning if their values are not equal.
4760 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
4766 The behavior is to emit an error if the ``llvm.module.flags`` does not
4767 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
4770 Objective-C Garbage Collection Module Flags Metadata
4771 ----------------------------------------------------
4773 On the Mach-O platform, Objective-C stores metadata about garbage
4774 collection in a special section called "image info". The metadata
4775 consists of a version number and a bitmask specifying what types of
4776 garbage collection are supported (if any) by the file. If two or more
4777 modules are linked together their garbage collection metadata needs to
4778 be merged rather than appended together.
4780 The Objective-C garbage collection module flags metadata consists of the
4781 following key-value pairs:
4790 * - ``Objective-C Version``
4791 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
4793 * - ``Objective-C Image Info Version``
4794 - **[Required]** --- The version of the image info section. Currently
4797 * - ``Objective-C Image Info Section``
4798 - **[Required]** --- The section to place the metadata. Valid values are
4799 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
4800 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
4801 Objective-C ABI version 2.
4803 * - ``Objective-C Garbage Collection``
4804 - **[Required]** --- Specifies whether garbage collection is supported or
4805 not. Valid values are 0, for no garbage collection, and 2, for garbage
4806 collection supported.
4808 * - ``Objective-C GC Only``
4809 - **[Optional]** --- Specifies that only garbage collection is supported.
4810 If present, its value must be 6. This flag requires that the
4811 ``Objective-C Garbage Collection`` flag have the value 2.
4813 Some important flag interactions:
4815 - If a module with ``Objective-C Garbage Collection`` set to 0 is
4816 merged with a module with ``Objective-C Garbage Collection`` set to
4817 2, then the resulting module has the
4818 ``Objective-C Garbage Collection`` flag set to 0.
4819 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
4820 merged with a module with ``Objective-C GC Only`` set to 6.
4822 Automatic Linker Flags Module Flags Metadata
4823 --------------------------------------------
4825 Some targets support embedding flags to the linker inside individual object
4826 files. Typically this is used in conjunction with language extensions which
4827 allow source files to explicitly declare the libraries they depend on, and have
4828 these automatically be transmitted to the linker via object files.
4830 These flags are encoded in the IR using metadata in the module flags section,
4831 using the ``Linker Options`` key. The merge behavior for this flag is required
4832 to be ``AppendUnique``, and the value for the key is expected to be a metadata
4833 node which should be a list of other metadata nodes, each of which should be a
4834 list of metadata strings defining linker options.
4836 For example, the following metadata section specifies two separate sets of
4837 linker options, presumably to link against ``libz`` and the ``Cocoa``
4840 !0 = !{ i32 6, !"Linker Options",
4843 !{ !"-framework", !"Cocoa" } } }
4844 !llvm.module.flags = !{ !0 }
4846 The metadata encoding as lists of lists of options, as opposed to a collapsed
4847 list of options, is chosen so that the IR encoding can use multiple option
4848 strings to specify e.g., a single library, while still having that specifier be
4849 preserved as an atomic element that can be recognized by a target specific
4850 assembly writer or object file emitter.
4852 Each individual option is required to be either a valid option for the target's
4853 linker, or an option that is reserved by the target specific assembly writer or
4854 object file emitter. No other aspect of these options is defined by the IR.
4856 C type width Module Flags Metadata
4857 ----------------------------------
4859 The ARM backend emits a section into each generated object file describing the
4860 options that it was compiled with (in a compiler-independent way) to prevent
4861 linking incompatible objects, and to allow automatic library selection. Some
4862 of these options are not visible at the IR level, namely wchar_t width and enum
4865 To pass this information to the backend, these options are encoded in module
4866 flags metadata, using the following key-value pairs:
4876 - * 0 --- sizeof(wchar_t) == 4
4877 * 1 --- sizeof(wchar_t) == 2
4880 - * 0 --- Enums are at least as large as an ``int``.
4881 * 1 --- Enums are stored in the smallest integer type which can
4882 represent all of its values.
4884 For example, the following metadata section specifies that the module was
4885 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
4886 enum is the smallest type which can represent all of its values::
4888 !llvm.module.flags = !{!0, !1}
4889 !0 = !{i32 1, !"short_wchar", i32 1}
4890 !1 = !{i32 1, !"short_enum", i32 0}
4892 .. _intrinsicglobalvariables:
4894 Intrinsic Global Variables
4895 ==========================
4897 LLVM has a number of "magic" global variables that contain data that
4898 affect code generation or other IR semantics. These are documented here.
4899 All globals of this sort should have a section specified as
4900 "``llvm.metadata``". This section and all globals that start with
4901 "``llvm.``" are reserved for use by LLVM.
4905 The '``llvm.used``' Global Variable
4906 -----------------------------------
4908 The ``@llvm.used`` global is an array which has
4909 :ref:`appending linkage <linkage_appending>`. This array contains a list of
4910 pointers to named global variables, functions and aliases which may optionally
4911 have a pointer cast formed of bitcast or getelementptr. For example, a legal
4914 .. code-block:: llvm
4919 @llvm.used = appending global [2 x i8*] [
4921 i8* bitcast (i32* @Y to i8*)
4922 ], section "llvm.metadata"
4924 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
4925 and linker are required to treat the symbol as if there is a reference to the
4926 symbol that it cannot see (which is why they have to be named). For example, if
4927 a variable has internal linkage and no references other than that from the
4928 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
4929 references from inline asms and other things the compiler cannot "see", and
4930 corresponds to "``attribute((used))``" in GNU C.
4932 On some targets, the code generator must emit a directive to the
4933 assembler or object file to prevent the assembler and linker from
4934 molesting the symbol.
4936 .. _gv_llvmcompilerused:
4938 The '``llvm.compiler.used``' Global Variable
4939 --------------------------------------------
4941 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4942 directive, except that it only prevents the compiler from touching the
4943 symbol. On targets that support it, this allows an intelligent linker to
4944 optimize references to the symbol without being impeded as it would be
4947 This is a rare construct that should only be used in rare circumstances,
4948 and should not be exposed to source languages.
4950 .. _gv_llvmglobalctors:
4952 The '``llvm.global_ctors``' Global Variable
4953 -------------------------------------------
4955 .. code-block:: llvm
4957 %0 = type { i32, void ()*, i8* }
4958 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
4960 The ``@llvm.global_ctors`` array contains a list of constructor
4961 functions, priorities, and an optional associated global or function.
4962 The functions referenced by this array will be called in ascending order
4963 of priority (i.e. lowest first) when the module is loaded. The order of
4964 functions with the same priority is not defined.
4966 If the third field is present, non-null, and points to a global variable
4967 or function, the initializer function will only run if the associated
4968 data from the current module is not discarded.
4970 .. _llvmglobaldtors:
4972 The '``llvm.global_dtors``' Global Variable
4973 -------------------------------------------
4975 .. code-block:: llvm
4977 %0 = type { i32, void ()*, i8* }
4978 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
4980 The ``@llvm.global_dtors`` array contains a list of destructor
4981 functions, priorities, and an optional associated global or function.
4982 The functions referenced by this array will be called in descending
4983 order of priority (i.e. highest first) when the module is unloaded. The
4984 order of functions with the same priority is not defined.
4986 If the third field is present, non-null, and points to a global variable
4987 or function, the destructor function will only run if the associated
4988 data from the current module is not discarded.
4990 Instruction Reference
4991 =====================
4993 The LLVM instruction set consists of several different classifications
4994 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
4995 instructions <binaryops>`, :ref:`bitwise binary
4996 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
4997 :ref:`other instructions <otherops>`.
5001 Terminator Instructions
5002 -----------------------
5004 As mentioned :ref:`previously <functionstructure>`, every basic block in a
5005 program ends with a "Terminator" instruction, which indicates which
5006 block should be executed after the current block is finished. These
5007 terminator instructions typically yield a '``void``' value: they produce
5008 control flow, not values (the one exception being the
5009 ':ref:`invoke <i_invoke>`' instruction).
5011 The terminator instructions are: ':ref:`ret <i_ret>`',
5012 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
5013 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
5014 ':ref:`resume <i_resume>`', ':ref:`catchswitch <i_catchswitch>`',
5015 ':ref:`catchret <i_catchret>`',
5016 ':ref:`cleanupret <i_cleanupret>`',
5017 and ':ref:`unreachable <i_unreachable>`'.
5021 '``ret``' Instruction
5022 ^^^^^^^^^^^^^^^^^^^^^
5029 ret <type> <value> ; Return a value from a non-void function
5030 ret void ; Return from void function
5035 The '``ret``' instruction is used to return control flow (and optionally
5036 a value) from a function back to the caller.
5038 There are two forms of the '``ret``' instruction: one that returns a
5039 value and then causes control flow, and one that just causes control
5045 The '``ret``' instruction optionally accepts a single argument, the
5046 return value. The type of the return value must be a ':ref:`first
5047 class <t_firstclass>`' type.
5049 A function is not :ref:`well formed <wellformed>` if it it has a non-void
5050 return type and contains a '``ret``' instruction with no return value or
5051 a return value with a type that does not match its type, or if it has a
5052 void return type and contains a '``ret``' instruction with a return
5058 When the '``ret``' instruction is executed, control flow returns back to
5059 the calling function's context. If the caller is a
5060 ":ref:`call <i_call>`" instruction, execution continues at the
5061 instruction after the call. If the caller was an
5062 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
5063 beginning of the "normal" destination block. If the instruction returns
5064 a value, that value shall set the call or invoke instruction's return
5070 .. code-block:: llvm
5072 ret i32 5 ; Return an integer value of 5
5073 ret void ; Return from a void function
5074 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
5078 '``br``' Instruction
5079 ^^^^^^^^^^^^^^^^^^^^
5086 br i1 <cond>, label <iftrue>, label <iffalse>
5087 br label <dest> ; Unconditional branch
5092 The '``br``' instruction is used to cause control flow to transfer to a
5093 different basic block in the current function. There are two forms of
5094 this instruction, corresponding to a conditional branch and an
5095 unconditional branch.
5100 The conditional branch form of the '``br``' instruction takes a single
5101 '``i1``' value and two '``label``' values. The unconditional form of the
5102 '``br``' instruction takes a single '``label``' value as a target.
5107 Upon execution of a conditional '``br``' instruction, the '``i1``'
5108 argument is evaluated. If the value is ``true``, control flows to the
5109 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
5110 to the '``iffalse``' ``label`` argument.
5115 .. code-block:: llvm
5118 %cond = icmp eq i32 %a, %b
5119 br i1 %cond, label %IfEqual, label %IfUnequal
5127 '``switch``' Instruction
5128 ^^^^^^^^^^^^^^^^^^^^^^^^
5135 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
5140 The '``switch``' instruction is used to transfer control flow to one of
5141 several different places. It is a generalization of the '``br``'
5142 instruction, allowing a branch to occur to one of many possible
5148 The '``switch``' instruction uses three parameters: an integer
5149 comparison value '``value``', a default '``label``' destination, and an
5150 array of pairs of comparison value constants and '``label``'s. The table
5151 is not allowed to contain duplicate constant entries.
5156 The ``switch`` instruction specifies a table of values and destinations.
5157 When the '``switch``' instruction is executed, this table is searched
5158 for the given value. If the value is found, control flow is transferred
5159 to the corresponding destination; otherwise, control flow is transferred
5160 to the default destination.
5165 Depending on properties of the target machine and the particular
5166 ``switch`` instruction, this instruction may be code generated in
5167 different ways. For example, it could be generated as a series of
5168 chained conditional branches or with a lookup table.
5173 .. code-block:: llvm
5175 ; Emulate a conditional br instruction
5176 %Val = zext i1 %value to i32
5177 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
5179 ; Emulate an unconditional br instruction
5180 switch i32 0, label %dest [ ]
5182 ; Implement a jump table:
5183 switch i32 %val, label %otherwise [ i32 0, label %onzero
5185 i32 2, label %ontwo ]
5189 '``indirectbr``' Instruction
5190 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5197 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
5202 The '``indirectbr``' instruction implements an indirect branch to a
5203 label within the current function, whose address is specified by
5204 "``address``". Address must be derived from a
5205 :ref:`blockaddress <blockaddress>` constant.
5210 The '``address``' argument is the address of the label to jump to. The
5211 rest of the arguments indicate the full set of possible destinations
5212 that the address may point to. Blocks are allowed to occur multiple
5213 times in the destination list, though this isn't particularly useful.
5215 This destination list is required so that dataflow analysis has an
5216 accurate understanding of the CFG.
5221 Control transfers to the block specified in the address argument. All
5222 possible destination blocks must be listed in the label list, otherwise
5223 this instruction has undefined behavior. This implies that jumps to
5224 labels defined in other functions have undefined behavior as well.
5229 This is typically implemented with a jump through a register.
5234 .. code-block:: llvm
5236 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
5240 '``invoke``' Instruction
5241 ^^^^^^^^^^^^^^^^^^^^^^^^
5248 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
5249 [operand bundles] to label <normal label> unwind label <exception label>
5254 The '``invoke``' instruction causes control to transfer to a specified
5255 function, with the possibility of control flow transfer to either the
5256 '``normal``' label or the '``exception``' label. If the callee function
5257 returns with the "``ret``" instruction, control flow will return to the
5258 "normal" label. If the callee (or any indirect callees) returns via the
5259 ":ref:`resume <i_resume>`" instruction or other exception handling
5260 mechanism, control is interrupted and continued at the dynamically
5261 nearest "exception" label.
5263 The '``exception``' label is a `landing
5264 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
5265 '``exception``' label is required to have the
5266 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
5267 information about the behavior of the program after unwinding happens,
5268 as its first non-PHI instruction. The restrictions on the
5269 "``landingpad``" instruction's tightly couples it to the "``invoke``"
5270 instruction, so that the important information contained within the
5271 "``landingpad``" instruction can't be lost through normal code motion.
5276 This instruction requires several arguments:
5278 #. The optional "cconv" marker indicates which :ref:`calling
5279 convention <callingconv>` the call should use. If none is
5280 specified, the call defaults to using C calling conventions.
5281 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5282 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5284 #. '``ptr to function ty``': shall be the signature of the pointer to
5285 function value being invoked. In most cases, this is a direct
5286 function invocation, but indirect ``invoke``'s are just as possible,
5287 branching off an arbitrary pointer to function value.
5288 #. '``function ptr val``': An LLVM value containing a pointer to a
5289 function to be invoked.
5290 #. '``function args``': argument list whose types match the function
5291 signature argument types and parameter attributes. All arguments must
5292 be of :ref:`first class <t_firstclass>` type. If the function signature
5293 indicates the function accepts a variable number of arguments, the
5294 extra arguments can be specified.
5295 #. '``normal label``': the label reached when the called function
5296 executes a '``ret``' instruction.
5297 #. '``exception label``': the label reached when a callee returns via
5298 the :ref:`resume <i_resume>` instruction or other exception handling
5300 #. The optional :ref:`function attributes <fnattrs>` list. Only
5301 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5302 attributes are valid here.
5303 #. The optional :ref:`operand bundles <opbundles>` list.
5308 This instruction is designed to operate as a standard '``call``'
5309 instruction in most regards. The primary difference is that it
5310 establishes an association with a label, which is used by the runtime
5311 library to unwind the stack.
5313 This instruction is used in languages with destructors to ensure that
5314 proper cleanup is performed in the case of either a ``longjmp`` or a
5315 thrown exception. Additionally, this is important for implementation of
5316 '``catch``' clauses in high-level languages that support them.
5318 For the purposes of the SSA form, the definition of the value returned
5319 by the '``invoke``' instruction is deemed to occur on the edge from the
5320 current block to the "normal" label. If the callee unwinds then no
5321 return value is available.
5326 .. code-block:: llvm
5328 %retval = invoke i32 @Test(i32 15) to label %Continue
5329 unwind label %TestCleanup ; i32:retval set
5330 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
5331 unwind label %TestCleanup ; i32:retval set
5335 '``resume``' Instruction
5336 ^^^^^^^^^^^^^^^^^^^^^^^^
5343 resume <type> <value>
5348 The '``resume``' instruction is a terminator instruction that has no
5354 The '``resume``' instruction requires one argument, which must have the
5355 same type as the result of any '``landingpad``' instruction in the same
5361 The '``resume``' instruction resumes propagation of an existing
5362 (in-flight) exception whose unwinding was interrupted with a
5363 :ref:`landingpad <i_landingpad>` instruction.
5368 .. code-block:: llvm
5370 resume { i8*, i32 } %exn
5374 '``catchswitch``' Instruction
5375 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5382 <resultval> = catchswitch within <parent> [ label <handler1>, label <handler2>, ... ] unwind to caller
5383 <resultval> = catchswitch within <parent> [ label <handler1>, label <handler2>, ... ] unwind label <default>
5388 The '``catchswitch``' instruction is used by `LLVM's exception handling system
5389 <ExceptionHandling.html#overview>`_ to describe the set of possible catch handlers
5390 that may be executed by the :ref:`EH personality routine <personalityfn>`.
5395 The ``parent`` argument is the token of the funclet that contains the
5396 ``catchswitch`` instruction. If the ``catchswitch`` is not inside a funclet,
5397 this operand may be the token ``none``.
5399 The ``default`` argument is the label of another basic block beginning with a
5400 "pad" instruction, one of ``cleanuppad`` or ``catchswitch``.
5402 The ``handlers`` are a list of successor blocks that each begin with a
5403 :ref:`catchpad <i_catchpad>` instruction.
5408 Executing this instruction transfers control to one of the successors in
5409 ``handlers``, if appropriate, or continues to unwind via the unwind label if
5412 The ``catchswitch`` is both a terminator and a "pad" instruction, meaning that
5413 it must be both the first non-phi instruction and last instruction in the basic
5414 block. Therefore, it must be the only non-phi instruction in the block.
5419 .. code-block:: llvm
5422 %cs1 = catchswitch within none [label %handler0, label %handler1] unwind to caller
5424 %cs2 = catchswitch within %parenthandler [label %handler0] unwind label %cleanup
5428 '``catchpad``' Instruction
5429 ^^^^^^^^^^^^^^^^^^^^^^^^^^
5436 <resultval> = catchpad within <catchswitch> [<args>*]
5441 The '``catchpad``' instruction is used by `LLVM's exception handling
5442 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5443 begins a catch handler --- one where a personality routine attempts to transfer
5444 control to catch an exception.
5449 The ``catchswitch`` operand must always be a token produced by a
5450 :ref:`catchswitch <i_catchswitch>` instruction in a predecessor block. This
5451 ensures that each ``catchpad`` has exactly one predecessor block, and it always
5452 terminates in a ``catchswitch``.
5454 The ``args`` correspond to whatever information the personality routine
5455 requires to know if this is an appropriate handler for the exception. Control
5456 will transfer to the ``catchpad`` if this is the first appropriate handler for
5459 The ``resultval`` has the type :ref:`token <t_token>` and is used to match the
5460 ``catchpad`` to corresponding :ref:`catchrets <i_catchret>` and other nested EH
5466 When the call stack is being unwound due to an exception being thrown, the
5467 exception is compared against the ``args``. If it doesn't match, control will
5468 not reach the ``catchpad`` instruction. The representation of ``args`` is
5469 entirely target and personality function-specific.
5471 Like the :ref:`landingpad <i_landingpad>` instruction, the ``catchpad``
5472 instruction must be the first non-phi of its parent basic block.
5474 The meaning of the tokens produced and consumed by ``catchpad`` and other "pad"
5475 instructions is described in the
5476 `Windows exception handling documentation <ExceptionHandling.html#wineh>`.
5478 Executing a ``catchpad`` instruction constitutes "entering" that pad.
5479 The pad may then be "exited" in one of three ways:
5481 1) explicitly via a ``catchret`` that consumes it. Executing such a ``catchret``
5482 is undefined behavior if any descendant pads have been entered but not yet
5484 2) implicitly via a call (which unwinds all the way to the current function's caller),
5485 or via a ``catchswitch`` or a ``cleanupret`` that unwinds to caller.
5486 3) implicitly via an unwind edge whose destination EH pad isn't a descendant of
5487 the ``catchpad``. When the ``catchpad`` is exited in this manner, it is
5488 undefined behavior if the destination EH pad has a parent which is not an
5489 ancestor of the ``catchpad`` being exited.
5494 .. code-block:: llvm
5497 %cs = catchswitch within none [label %handler0] unwind to caller
5498 ;; A catch block which can catch an integer.
5500 %tok = catchpad within %cs [i8** @_ZTIi]
5504 '``catchret``' Instruction
5505 ^^^^^^^^^^^^^^^^^^^^^^^^^^
5512 catchret from <token> to label <normal>
5517 The '``catchret``' instruction is a terminator instruction that has a
5524 The first argument to a '``catchret``' indicates which ``catchpad`` it
5525 exits. It must be a :ref:`catchpad <i_catchpad>`.
5526 The second argument to a '``catchret``' specifies where control will
5532 The '``catchret``' instruction ends an existing (in-flight) exception whose
5533 unwinding was interrupted with a :ref:`catchpad <i_catchpad>` instruction. The
5534 :ref:`personality function <personalityfn>` gets a chance to execute arbitrary
5535 code to, for example, destroy the active exception. Control then transfers to
5538 The ``token`` argument must be a token produced by a dominating ``catchpad``
5539 instruction. The ``catchret`` destroys the physical frame established by
5540 ``catchpad``, so executing multiple returns on the same token without
5541 re-executing the ``catchpad`` will result in undefined behavior.
5542 See :ref:`catchpad <i_catchpad>` for more details.
5547 .. code-block:: llvm
5549 catchret from %catch label %continue
5553 '``cleanupret``' Instruction
5554 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5561 cleanupret from <value> unwind label <continue>
5562 cleanupret from <value> unwind to caller
5567 The '``cleanupret``' instruction is a terminator instruction that has
5568 an optional successor.
5574 The '``cleanupret``' instruction requires one argument, which indicates
5575 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
5576 It also has an optional successor, ``continue``.
5581 The '``cleanupret``' instruction indicates to the
5582 :ref:`personality function <personalityfn>` that one
5583 :ref:`cleanuppad <i_cleanuppad>` it transferred control to has ended.
5584 It transfers control to ``continue`` or unwinds out of the function.
5586 The unwind destination ``continue``, if present, must be an EH pad
5587 whose parent is either ``none`` or an ancestor of the ``cleanuppad``
5588 being returned from. This constitutes an exceptional exit from all
5589 ancestors of the completed ``cleanuppad``, up to but not including
5590 the parent of ``continue``.
5591 See :ref:`cleanuppad <i_cleanuppad>` for more details.
5596 .. code-block:: llvm
5598 cleanupret from %cleanup unwind to caller
5599 cleanupret from %cleanup unwind label %continue
5603 '``unreachable``' Instruction
5604 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5616 The '``unreachable``' instruction has no defined semantics. This
5617 instruction is used to inform the optimizer that a particular portion of
5618 the code is not reachable. This can be used to indicate that the code
5619 after a no-return function cannot be reached, and other facts.
5624 The '``unreachable``' instruction has no defined semantics.
5631 Binary operators are used to do most of the computation in a program.
5632 They require two operands of the same type, execute an operation on
5633 them, and produce a single value. The operands might represent multiple
5634 data, as is the case with the :ref:`vector <t_vector>` data type. The
5635 result value has the same type as its operands.
5637 There are several different binary operators:
5641 '``add``' Instruction
5642 ^^^^^^^^^^^^^^^^^^^^^
5649 <result> = add <ty> <op1>, <op2> ; yields ty:result
5650 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
5651 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
5652 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
5657 The '``add``' instruction returns the sum of its two operands.
5662 The two arguments to the '``add``' instruction must be
5663 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5664 arguments must have identical types.
5669 The value produced is the integer sum of the two operands.
5671 If the sum has unsigned overflow, the result returned is the
5672 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5675 Because LLVM integers use a two's complement representation, this
5676 instruction is appropriate for both signed and unsigned integers.
5678 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5679 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5680 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
5681 unsigned and/or signed overflow, respectively, occurs.
5686 .. code-block:: llvm
5688 <result> = add i32 4, %var ; yields i32:result = 4 + %var
5692 '``fadd``' Instruction
5693 ^^^^^^^^^^^^^^^^^^^^^^
5700 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5705 The '``fadd``' instruction returns the sum of its two operands.
5710 The two arguments to the '``fadd``' instruction must be :ref:`floating
5711 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5712 Both arguments must have identical types.
5717 The value produced is the floating point sum of the two operands. This
5718 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
5719 which are optimization hints to enable otherwise unsafe floating point
5725 .. code-block:: llvm
5727 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
5729 '``sub``' Instruction
5730 ^^^^^^^^^^^^^^^^^^^^^
5737 <result> = sub <ty> <op1>, <op2> ; yields ty:result
5738 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
5739 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
5740 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
5745 The '``sub``' instruction returns the difference of its two operands.
5747 Note that the '``sub``' instruction is used to represent the '``neg``'
5748 instruction present in most other intermediate representations.
5753 The two arguments to the '``sub``' instruction must be
5754 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5755 arguments must have identical types.
5760 The value produced is the integer difference of the two operands.
5762 If the difference has unsigned overflow, the result returned is the
5763 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5766 Because LLVM integers use a two's complement representation, this
5767 instruction is appropriate for both signed and unsigned integers.
5769 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5770 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5771 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
5772 unsigned and/or signed overflow, respectively, occurs.
5777 .. code-block:: llvm
5779 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
5780 <result> = sub i32 0, %val ; yields i32:result = -%var
5784 '``fsub``' Instruction
5785 ^^^^^^^^^^^^^^^^^^^^^^
5792 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5797 The '``fsub``' instruction returns the difference of its two operands.
5799 Note that the '``fsub``' instruction is used to represent the '``fneg``'
5800 instruction present in most other intermediate representations.
5805 The two arguments to the '``fsub``' instruction must be :ref:`floating
5806 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5807 Both arguments must have identical types.
5812 The value produced is the floating point difference of the two operands.
5813 This instruction can also take any number of :ref:`fast-math
5814 flags <fastmath>`, which are optimization hints to enable otherwise
5815 unsafe floating point optimizations:
5820 .. code-block:: llvm
5822 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
5823 <result> = fsub float -0.0, %val ; yields float:result = -%var
5825 '``mul``' Instruction
5826 ^^^^^^^^^^^^^^^^^^^^^
5833 <result> = mul <ty> <op1>, <op2> ; yields ty:result
5834 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
5835 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
5836 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
5841 The '``mul``' instruction returns the product of its two operands.
5846 The two arguments to the '``mul``' instruction must be
5847 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5848 arguments must have identical types.
5853 The value produced is the integer product of the two operands.
5855 If the result of the multiplication has unsigned overflow, the result
5856 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
5857 bit width of the result.
5859 Because LLVM integers use a two's complement representation, and the
5860 result is the same width as the operands, this instruction returns the
5861 correct result for both signed and unsigned integers. If a full product
5862 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
5863 sign-extended or zero-extended as appropriate to the width of the full
5866 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5867 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5868 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
5869 unsigned and/or signed overflow, respectively, occurs.
5874 .. code-block:: llvm
5876 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
5880 '``fmul``' Instruction
5881 ^^^^^^^^^^^^^^^^^^^^^^
5888 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5893 The '``fmul``' instruction returns the product of its two operands.
5898 The two arguments to the '``fmul``' instruction must be :ref:`floating
5899 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5900 Both arguments must have identical types.
5905 The value produced is the floating point product of the two operands.
5906 This instruction can also take any number of :ref:`fast-math
5907 flags <fastmath>`, which are optimization hints to enable otherwise
5908 unsafe floating point optimizations:
5913 .. code-block:: llvm
5915 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
5917 '``udiv``' Instruction
5918 ^^^^^^^^^^^^^^^^^^^^^^
5925 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
5926 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
5931 The '``udiv``' instruction returns the quotient of its two operands.
5936 The two arguments to the '``udiv``' instruction must be
5937 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5938 arguments must have identical types.
5943 The value produced is the unsigned integer quotient of the two operands.
5945 Note that unsigned integer division and signed integer division are
5946 distinct operations; for signed integer division, use '``sdiv``'.
5948 Division by zero leads to undefined behavior.
5950 If the ``exact`` keyword is present, the result value of the ``udiv`` is
5951 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
5952 such, "((a udiv exact b) mul b) == a").
5957 .. code-block:: llvm
5959 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
5961 '``sdiv``' Instruction
5962 ^^^^^^^^^^^^^^^^^^^^^^
5969 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
5970 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
5975 The '``sdiv``' instruction returns the quotient of its two operands.
5980 The two arguments to the '``sdiv``' instruction must be
5981 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5982 arguments must have identical types.
5987 The value produced is the signed integer quotient of the two operands
5988 rounded towards zero.
5990 Note that signed integer division and unsigned integer division are
5991 distinct operations; for unsigned integer division, use '``udiv``'.
5993 Division by zero leads to undefined behavior. Overflow also leads to
5994 undefined behavior; this is a rare case, but can occur, for example, by
5995 doing a 32-bit division of -2147483648 by -1.
5997 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
5998 a :ref:`poison value <poisonvalues>` if the result would be rounded.
6003 .. code-block:: llvm
6005 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
6009 '``fdiv``' Instruction
6010 ^^^^^^^^^^^^^^^^^^^^^^
6017 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6022 The '``fdiv``' instruction returns the quotient of its two operands.
6027 The two arguments to the '``fdiv``' instruction must be :ref:`floating
6028 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6029 Both arguments must have identical types.
6034 The value produced is the floating point quotient of the two operands.
6035 This instruction can also take any number of :ref:`fast-math
6036 flags <fastmath>`, which are optimization hints to enable otherwise
6037 unsafe floating point optimizations:
6042 .. code-block:: llvm
6044 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
6046 '``urem``' Instruction
6047 ^^^^^^^^^^^^^^^^^^^^^^
6054 <result> = urem <ty> <op1>, <op2> ; yields ty:result
6059 The '``urem``' instruction returns the remainder from the unsigned
6060 division of its two arguments.
6065 The two arguments to the '``urem``' instruction must be
6066 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6067 arguments must have identical types.
6072 This instruction returns the unsigned integer *remainder* of a division.
6073 This instruction always performs an unsigned division to get the
6076 Note that unsigned integer remainder and signed integer remainder are
6077 distinct operations; for signed integer remainder, use '``srem``'.
6079 Taking the remainder of a division by zero leads to undefined behavior.
6084 .. code-block:: llvm
6086 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
6088 '``srem``' Instruction
6089 ^^^^^^^^^^^^^^^^^^^^^^
6096 <result> = srem <ty> <op1>, <op2> ; yields ty:result
6101 The '``srem``' instruction returns the remainder from the signed
6102 division of its two operands. This instruction can also take
6103 :ref:`vector <t_vector>` versions of the values in which case the elements
6109 The two arguments to the '``srem``' instruction must be
6110 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6111 arguments must have identical types.
6116 This instruction returns the *remainder* of a division (where the result
6117 is either zero or has the same sign as the dividend, ``op1``), not the
6118 *modulo* operator (where the result is either zero or has the same sign
6119 as the divisor, ``op2``) of a value. For more information about the
6120 difference, see `The Math
6121 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
6122 table of how this is implemented in various languages, please see
6124 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
6126 Note that signed integer remainder and unsigned integer remainder are
6127 distinct operations; for unsigned integer remainder, use '``urem``'.
6129 Taking the remainder of a division by zero leads to undefined behavior.
6130 Overflow also leads to undefined behavior; this is a rare case, but can
6131 occur, for example, by taking the remainder of a 32-bit division of
6132 -2147483648 by -1. (The remainder doesn't actually overflow, but this
6133 rule lets srem be implemented using instructions that return both the
6134 result of the division and the remainder.)
6139 .. code-block:: llvm
6141 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
6145 '``frem``' Instruction
6146 ^^^^^^^^^^^^^^^^^^^^^^
6153 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6158 The '``frem``' instruction returns the remainder from the division of
6164 The two arguments to the '``frem``' instruction must be :ref:`floating
6165 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6166 Both arguments must have identical types.
6171 This instruction returns the *remainder* of a division. The remainder
6172 has the same sign as the dividend. This instruction can also take any
6173 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
6174 to enable otherwise unsafe floating point optimizations:
6179 .. code-block:: llvm
6181 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
6185 Bitwise Binary Operations
6186 -------------------------
6188 Bitwise binary operators are used to do various forms of bit-twiddling
6189 in a program. They are generally very efficient instructions and can
6190 commonly be strength reduced from other instructions. They require two
6191 operands of the same type, execute an operation on them, and produce a
6192 single value. The resulting value is the same type as its operands.
6194 '``shl``' Instruction
6195 ^^^^^^^^^^^^^^^^^^^^^
6202 <result> = shl <ty> <op1>, <op2> ; yields ty:result
6203 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
6204 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
6205 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
6210 The '``shl``' instruction returns the first operand shifted to the left
6211 a specified number of bits.
6216 Both arguments to the '``shl``' instruction must be the same
6217 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6218 '``op2``' is treated as an unsigned value.
6223 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
6224 where ``n`` is the width of the result. If ``op2`` is (statically or
6225 dynamically) equal to or larger than the number of bits in
6226 ``op1``, the result is undefined. If the arguments are vectors, each
6227 vector element of ``op1`` is shifted by the corresponding shift amount
6230 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
6231 value <poisonvalues>` if it shifts out any non-zero bits. If the
6232 ``nsw`` keyword is present, then the shift produces a :ref:`poison
6233 value <poisonvalues>` if it shifts out any bits that disagree with the
6234 resultant sign bit. As such, NUW/NSW have the same semantics as they
6235 would if the shift were expressed as a mul instruction with the same
6236 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
6241 .. code-block:: llvm
6243 <result> = shl i32 4, %var ; yields i32: 4 << %var
6244 <result> = shl i32 4, 2 ; yields i32: 16
6245 <result> = shl i32 1, 10 ; yields i32: 1024
6246 <result> = shl i32 1, 32 ; undefined
6247 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
6249 '``lshr``' Instruction
6250 ^^^^^^^^^^^^^^^^^^^^^^
6257 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
6258 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
6263 The '``lshr``' instruction (logical shift right) returns the first
6264 operand shifted to the right a specified number of bits with zero fill.
6269 Both arguments to the '``lshr``' instruction must be the same
6270 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6271 '``op2``' is treated as an unsigned value.
6276 This instruction always performs a logical shift right operation. The
6277 most significant bits of the result will be filled with zero bits after
6278 the shift. If ``op2`` is (statically or dynamically) equal to or larger
6279 than the number of bits in ``op1``, the result is undefined. If the
6280 arguments are vectors, each vector element of ``op1`` is shifted by the
6281 corresponding shift amount in ``op2``.
6283 If the ``exact`` keyword is present, the result value of the ``lshr`` is
6284 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6290 .. code-block:: llvm
6292 <result> = lshr i32 4, 1 ; yields i32:result = 2
6293 <result> = lshr i32 4, 2 ; yields i32:result = 1
6294 <result> = lshr i8 4, 3 ; yields i8:result = 0
6295 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
6296 <result> = lshr i32 1, 32 ; undefined
6297 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
6299 '``ashr``' Instruction
6300 ^^^^^^^^^^^^^^^^^^^^^^
6307 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
6308 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
6313 The '``ashr``' instruction (arithmetic shift right) returns the first
6314 operand shifted to the right a specified number of bits with sign
6320 Both arguments to the '``ashr``' instruction must be the same
6321 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6322 '``op2``' is treated as an unsigned value.
6327 This instruction always performs an arithmetic shift right operation,
6328 The most significant bits of the result will be filled with the sign bit
6329 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
6330 than the number of bits in ``op1``, the result is undefined. If the
6331 arguments are vectors, each vector element of ``op1`` is shifted by the
6332 corresponding shift amount in ``op2``.
6334 If the ``exact`` keyword is present, the result value of the ``ashr`` is
6335 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6341 .. code-block:: llvm
6343 <result> = ashr i32 4, 1 ; yields i32:result = 2
6344 <result> = ashr i32 4, 2 ; yields i32:result = 1
6345 <result> = ashr i8 4, 3 ; yields i8:result = 0
6346 <result> = ashr i8 -2, 1 ; yields i8:result = -1
6347 <result> = ashr i32 1, 32 ; undefined
6348 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
6350 '``and``' Instruction
6351 ^^^^^^^^^^^^^^^^^^^^^
6358 <result> = and <ty> <op1>, <op2> ; yields ty:result
6363 The '``and``' instruction returns the bitwise logical and of its two
6369 The two arguments to the '``and``' instruction must be
6370 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6371 arguments must have identical types.
6376 The truth table used for the '``and``' instruction is:
6393 .. code-block:: llvm
6395 <result> = and i32 4, %var ; yields i32:result = 4 & %var
6396 <result> = and i32 15, 40 ; yields i32:result = 8
6397 <result> = and i32 4, 8 ; yields i32:result = 0
6399 '``or``' Instruction
6400 ^^^^^^^^^^^^^^^^^^^^
6407 <result> = or <ty> <op1>, <op2> ; yields ty:result
6412 The '``or``' instruction returns the bitwise logical inclusive or of its
6418 The two arguments to the '``or``' instruction must be
6419 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6420 arguments must have identical types.
6425 The truth table used for the '``or``' instruction is:
6444 <result> = or i32 4, %var ; yields i32:result = 4 | %var
6445 <result> = or i32 15, 40 ; yields i32:result = 47
6446 <result> = or i32 4, 8 ; yields i32:result = 12
6448 '``xor``' Instruction
6449 ^^^^^^^^^^^^^^^^^^^^^
6456 <result> = xor <ty> <op1>, <op2> ; yields ty:result
6461 The '``xor``' instruction returns the bitwise logical exclusive or of
6462 its two operands. The ``xor`` is used to implement the "one's
6463 complement" operation, which is the "~" operator in C.
6468 The two arguments to the '``xor``' instruction must be
6469 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6470 arguments must have identical types.
6475 The truth table used for the '``xor``' instruction is:
6492 .. code-block:: llvm
6494 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
6495 <result> = xor i32 15, 40 ; yields i32:result = 39
6496 <result> = xor i32 4, 8 ; yields i32:result = 12
6497 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
6502 LLVM supports several instructions to represent vector operations in a
6503 target-independent manner. These instructions cover the element-access
6504 and vector-specific operations needed to process vectors effectively.
6505 While LLVM does directly support these vector operations, many
6506 sophisticated algorithms will want to use target-specific intrinsics to
6507 take full advantage of a specific target.
6509 .. _i_extractelement:
6511 '``extractelement``' Instruction
6512 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6519 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
6524 The '``extractelement``' instruction extracts a single scalar element
6525 from a vector at a specified index.
6530 The first operand of an '``extractelement``' instruction is a value of
6531 :ref:`vector <t_vector>` type. The second operand is an index indicating
6532 the position from which to extract the element. The index may be a
6533 variable of any integer type.
6538 The result is a scalar of the same type as the element type of ``val``.
6539 Its value is the value at position ``idx`` of ``val``. If ``idx``
6540 exceeds the length of ``val``, the results are undefined.
6545 .. code-block:: llvm
6547 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
6549 .. _i_insertelement:
6551 '``insertelement``' Instruction
6552 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6559 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
6564 The '``insertelement``' instruction inserts a scalar element into a
6565 vector at a specified index.
6570 The first operand of an '``insertelement``' instruction is a value of
6571 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
6572 type must equal the element type of the first operand. The third operand
6573 is an index indicating the position at which to insert the value. The
6574 index may be a variable of any integer type.
6579 The result is a vector of the same type as ``val``. Its element values
6580 are those of ``val`` except at position ``idx``, where it gets the value
6581 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
6587 .. code-block:: llvm
6589 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
6591 .. _i_shufflevector:
6593 '``shufflevector``' Instruction
6594 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6601 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
6606 The '``shufflevector``' instruction constructs a permutation of elements
6607 from two input vectors, returning a vector with the same element type as
6608 the input and length that is the same as the shuffle mask.
6613 The first two operands of a '``shufflevector``' instruction are vectors
6614 with the same type. The third argument is a shuffle mask whose element
6615 type is always 'i32'. The result of the instruction is a vector whose
6616 length is the same as the shuffle mask and whose element type is the
6617 same as the element type of the first two operands.
6619 The shuffle mask operand is required to be a constant vector with either
6620 constant integer or undef values.
6625 The elements of the two input vectors are numbered from left to right
6626 across both of the vectors. The shuffle mask operand specifies, for each
6627 element of the result vector, which element of the two input vectors the
6628 result element gets. The element selector may be undef (meaning "don't
6629 care") and the second operand may be undef if performing a shuffle from
6635 .. code-block:: llvm
6637 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6638 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
6639 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
6640 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
6641 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
6642 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
6643 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6644 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
6646 Aggregate Operations
6647 --------------------
6649 LLVM supports several instructions for working with
6650 :ref:`aggregate <t_aggregate>` values.
6654 '``extractvalue``' Instruction
6655 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6662 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
6667 The '``extractvalue``' instruction extracts the value of a member field
6668 from an :ref:`aggregate <t_aggregate>` value.
6673 The first operand of an '``extractvalue``' instruction is a value of
6674 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The other operands are
6675 constant indices to specify which value to extract in a similar manner
6676 as indices in a '``getelementptr``' instruction.
6678 The major differences to ``getelementptr`` indexing are:
6680 - Since the value being indexed is not a pointer, the first index is
6681 omitted and assumed to be zero.
6682 - At least one index must be specified.
6683 - Not only struct indices but also array indices must be in bounds.
6688 The result is the value at the position in the aggregate specified by
6694 .. code-block:: llvm
6696 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
6700 '``insertvalue``' Instruction
6701 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6708 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
6713 The '``insertvalue``' instruction inserts a value into a member field in
6714 an :ref:`aggregate <t_aggregate>` value.
6719 The first operand of an '``insertvalue``' instruction is a value of
6720 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
6721 a first-class value to insert. The following operands are constant
6722 indices indicating the position at which to insert the value in a
6723 similar manner as indices in a '``extractvalue``' instruction. The value
6724 to insert must have the same type as the value identified by the
6730 The result is an aggregate of the same type as ``val``. Its value is
6731 that of ``val`` except that the value at the position specified by the
6732 indices is that of ``elt``.
6737 .. code-block:: llvm
6739 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
6740 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
6741 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
6745 Memory Access and Addressing Operations
6746 ---------------------------------------
6748 A key design point of an SSA-based representation is how it represents
6749 memory. In LLVM, no memory locations are in SSA form, which makes things
6750 very simple. This section describes how to read, write, and allocate
6755 '``alloca``' Instruction
6756 ^^^^^^^^^^^^^^^^^^^^^^^^
6763 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
6768 The '``alloca``' instruction allocates memory on the stack frame of the
6769 currently executing function, to be automatically released when this
6770 function returns to its caller. The object is always allocated in the
6771 generic address space (address space zero).
6776 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
6777 bytes of memory on the runtime stack, returning a pointer of the
6778 appropriate type to the program. If "NumElements" is specified, it is
6779 the number of elements allocated, otherwise "NumElements" is defaulted
6780 to be one. If a constant alignment is specified, the value result of the
6781 allocation is guaranteed to be aligned to at least that boundary. The
6782 alignment may not be greater than ``1 << 29``. If not specified, or if
6783 zero, the target can choose to align the allocation on any convenient
6784 boundary compatible with the type.
6786 '``type``' may be any sized type.
6791 Memory is allocated; a pointer is returned. The operation is undefined
6792 if there is insufficient stack space for the allocation. '``alloca``'d
6793 memory is automatically released when the function returns. The
6794 '``alloca``' instruction is commonly used to represent automatic
6795 variables that must have an address available. When the function returns
6796 (either with the ``ret`` or ``resume`` instructions), the memory is
6797 reclaimed. Allocating zero bytes is legal, but the result is undefined.
6798 The order in which memory is allocated (ie., which way the stack grows)
6804 .. code-block:: llvm
6806 %ptr = alloca i32 ; yields i32*:ptr
6807 %ptr = alloca i32, i32 4 ; yields i32*:ptr
6808 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
6809 %ptr = alloca i32, align 1024 ; yields i32*:ptr
6813 '``load``' Instruction
6814 ^^^^^^^^^^^^^^^^^^^^^^
6821 <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>]
6822 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>]
6823 !<index> = !{ i32 1 }
6824 !<deref_bytes_node> = !{i64 <dereferenceable_bytes>}
6825 !<align_node> = !{ i64 <value_alignment> }
6830 The '``load``' instruction is used to read from memory.
6835 The argument to the ``load`` instruction specifies the memory address
6836 from which to load. The type specified must be a :ref:`first
6837 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
6838 then the optimizer is not allowed to modify the number or order of
6839 execution of this ``load`` with other :ref:`volatile
6840 operations <volatile>`.
6842 If the ``load`` is marked as ``atomic``, it takes an extra
6843 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
6844 ``release`` and ``acq_rel`` orderings are not valid on ``load``
6845 instructions. Atomic loads produce :ref:`defined <memmodel>` results
6846 when they may see multiple atomic stores. The type of the pointee must
6847 be an integer type whose bit width is a power of two greater than or
6848 equal to eight and less than or equal to a target-specific size limit.
6849 ``align`` must be explicitly specified on atomic loads, and the load has
6850 undefined behavior if the alignment is not set to a value which is at
6851 least the size in bytes of the pointee. ``!nontemporal`` does not have
6852 any defined semantics for atomic loads.
6854 The optional constant ``align`` argument specifies the alignment of the
6855 operation (that is, the alignment of the memory address). A value of 0
6856 or an omitted ``align`` argument means that the operation has the ABI
6857 alignment for the target. It is the responsibility of the code emitter
6858 to ensure that the alignment information is correct. Overestimating the
6859 alignment results in undefined behavior. Underestimating the alignment
6860 may produce less efficient code. An alignment of 1 is always safe. The
6861 maximum possible alignment is ``1 << 29``.
6863 The optional ``!nontemporal`` metadata must reference a single
6864 metadata name ``<index>`` corresponding to a metadata node with one
6865 ``i32`` entry of value 1. The existence of the ``!nontemporal``
6866 metadata on the instruction tells the optimizer and code generator
6867 that this load is not expected to be reused in the cache. The code
6868 generator may select special instructions to save cache bandwidth, such
6869 as the ``MOVNT`` instruction on x86.
6871 The optional ``!invariant.load`` metadata must reference a single
6872 metadata name ``<index>`` corresponding to a metadata node with no
6873 entries. The existence of the ``!invariant.load`` metadata on the
6874 instruction tells the optimizer and code generator that the address
6875 operand to this load points to memory which can be assumed unchanged.
6876 Being invariant does not imply that a location is dereferenceable,
6877 but it does imply that once the location is known dereferenceable
6878 its value is henceforth unchanging.
6880 The optional ``!invariant.group`` metadata must reference a single metadata name
6881 ``<index>`` corresponding to a metadata node. See ``invariant.group`` metadata.
6883 The optional ``!nonnull`` metadata must reference a single
6884 metadata name ``<index>`` corresponding to a metadata node with no
6885 entries. The existence of the ``!nonnull`` metadata on the
6886 instruction tells the optimizer that the value loaded is known to
6887 never be null. This is analogous to the ``nonnull`` attribute
6888 on parameters and return values. This metadata can only be applied
6889 to loads of a pointer type.
6891 The optional ``!dereferenceable`` metadata must reference a single metadata
6892 name ``<deref_bytes_node>`` corresponding to a metadata node with one ``i64``
6893 entry. The existence of the ``!dereferenceable`` metadata on the instruction
6894 tells the optimizer that the value loaded is known to be dereferenceable.
6895 The number of bytes known to be dereferenceable is specified by the integer
6896 value in the metadata node. This is analogous to the ''dereferenceable''
6897 attribute on parameters and return values. This metadata can only be applied
6898 to loads of a pointer type.
6900 The optional ``!dereferenceable_or_null`` metadata must reference a single
6901 metadata name ``<deref_bytes_node>`` corresponding to a metadata node with one
6902 ``i64`` entry. The existence of the ``!dereferenceable_or_null`` metadata on the
6903 instruction tells the optimizer that the value loaded is known to be either
6904 dereferenceable or null.
6905 The number of bytes known to be dereferenceable is specified by the integer
6906 value in the metadata node. This is analogous to the ''dereferenceable_or_null''
6907 attribute on parameters and return values. This metadata can only be applied
6908 to loads of a pointer type.
6910 The optional ``!align`` metadata must reference a single metadata name
6911 ``<align_node>`` corresponding to a metadata node with one ``i64`` entry.
6912 The existence of the ``!align`` metadata on the instruction tells the
6913 optimizer that the value loaded is known to be aligned to a boundary specified
6914 by the integer value in the metadata node. The alignment must be a power of 2.
6915 This is analogous to the ''align'' attribute on parameters and return values.
6916 This metadata can only be applied to loads of a pointer type.
6921 The location of memory pointed to is loaded. If the value being loaded
6922 is of scalar type then the number of bytes read does not exceed the
6923 minimum number of bytes needed to hold all bits of the type. For
6924 example, loading an ``i24`` reads at most three bytes. When loading a
6925 value of a type like ``i20`` with a size that is not an integral number
6926 of bytes, the result is undefined if the value was not originally
6927 written using a store of the same type.
6932 .. code-block:: llvm
6934 %ptr = alloca i32 ; yields i32*:ptr
6935 store i32 3, i32* %ptr ; yields void
6936 %val = load i32, i32* %ptr ; yields i32:val = i32 3
6940 '``store``' Instruction
6941 ^^^^^^^^^^^^^^^^^^^^^^^
6948 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.group !<index>] ; yields void
6949 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>] ; yields void
6954 The '``store``' instruction is used to write to memory.
6959 There are two arguments to the ``store`` instruction: a value to store
6960 and an address at which to store it. The type of the ``<pointer>``
6961 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
6962 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
6963 then the optimizer is not allowed to modify the number or order of
6964 execution of this ``store`` with other :ref:`volatile
6965 operations <volatile>`.
6967 If the ``store`` is marked as ``atomic``, it takes an extra
6968 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
6969 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
6970 instructions. Atomic loads produce :ref:`defined <memmodel>` results
6971 when they may see multiple atomic stores. The type of the pointee must
6972 be an integer type whose bit width is a power of two greater than or
6973 equal to eight and less than or equal to a target-specific size limit.
6974 ``align`` must be explicitly specified on atomic stores, and the store
6975 has undefined behavior if the alignment is not set to a value which is
6976 at least the size in bytes of the pointee. ``!nontemporal`` does not
6977 have any defined semantics for atomic stores.
6979 The optional constant ``align`` argument specifies the alignment of the
6980 operation (that is, the alignment of the memory address). A value of 0
6981 or an omitted ``align`` argument means that the operation has the ABI
6982 alignment for the target. It is the responsibility of the code emitter
6983 to ensure that the alignment information is correct. Overestimating the
6984 alignment results in undefined behavior. Underestimating the
6985 alignment may produce less efficient code. An alignment of 1 is always
6986 safe. The maximum possible alignment is ``1 << 29``.
6988 The optional ``!nontemporal`` metadata must reference a single metadata
6989 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
6990 value 1. The existence of the ``!nontemporal`` metadata on the instruction
6991 tells the optimizer and code generator that this load is not expected to
6992 be reused in the cache. The code generator may select special
6993 instructions to save cache bandwidth, such as the MOVNT instruction on
6996 The optional ``!invariant.group`` metadata must reference a
6997 single metadata name ``<index>``. See ``invariant.group`` metadata.
7002 The contents of memory are updated to contain ``<value>`` at the
7003 location specified by the ``<pointer>`` operand. If ``<value>`` is
7004 of scalar type then the number of bytes written does not exceed the
7005 minimum number of bytes needed to hold all bits of the type. For
7006 example, storing an ``i24`` writes at most three bytes. When writing a
7007 value of a type like ``i20`` with a size that is not an integral number
7008 of bytes, it is unspecified what happens to the extra bits that do not
7009 belong to the type, but they will typically be overwritten.
7014 .. code-block:: llvm
7016 %ptr = alloca i32 ; yields i32*:ptr
7017 store i32 3, i32* %ptr ; yields void
7018 %val = load i32, i32* %ptr ; yields i32:val = i32 3
7022 '``fence``' Instruction
7023 ^^^^^^^^^^^^^^^^^^^^^^^
7030 fence [singlethread] <ordering> ; yields void
7035 The '``fence``' instruction is used to introduce happens-before edges
7041 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
7042 defines what *synchronizes-with* edges they add. They can only be given
7043 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
7048 A fence A which has (at least) ``release`` ordering semantics
7049 *synchronizes with* a fence B with (at least) ``acquire`` ordering
7050 semantics if and only if there exist atomic operations X and Y, both
7051 operating on some atomic object M, such that A is sequenced before X, X
7052 modifies M (either directly or through some side effect of a sequence
7053 headed by X), Y is sequenced before B, and Y observes M. This provides a
7054 *happens-before* dependency between A and B. Rather than an explicit
7055 ``fence``, one (but not both) of the atomic operations X or Y might
7056 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
7057 still *synchronize-with* the explicit ``fence`` and establish the
7058 *happens-before* edge.
7060 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
7061 ``acquire`` and ``release`` semantics specified above, participates in
7062 the global program order of other ``seq_cst`` operations and/or fences.
7064 The optional ":ref:`singlethread <singlethread>`" argument specifies
7065 that the fence only synchronizes with other fences in the same thread.
7066 (This is useful for interacting with signal handlers.)
7071 .. code-block:: llvm
7073 fence acquire ; yields void
7074 fence singlethread seq_cst ; yields void
7078 '``cmpxchg``' Instruction
7079 ^^^^^^^^^^^^^^^^^^^^^^^^^
7086 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
7091 The '``cmpxchg``' instruction is used to atomically modify memory. It
7092 loads a value in memory and compares it to a given value. If they are
7093 equal, it tries to store a new value into the memory.
7098 There are three arguments to the '``cmpxchg``' instruction: an address
7099 to operate on, a value to compare to the value currently be at that
7100 address, and a new value to place at that address if the compared values
7101 are equal. The type of '<cmp>' must be an integer type whose bit width
7102 is a power of two greater than or equal to eight and less than or equal
7103 to a target-specific size limit. '<cmp>' and '<new>' must have the same
7104 type, and the type of '<pointer>' must be a pointer to that type. If the
7105 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
7106 to modify the number or order of execution of this ``cmpxchg`` with
7107 other :ref:`volatile operations <volatile>`.
7109 The success and failure :ref:`ordering <ordering>` arguments specify how this
7110 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
7111 must be at least ``monotonic``, the ordering constraint on failure must be no
7112 stronger than that on success, and the failure ordering cannot be either
7113 ``release`` or ``acq_rel``.
7115 The optional "``singlethread``" argument declares that the ``cmpxchg``
7116 is only atomic with respect to code (usually signal handlers) running in
7117 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
7118 respect to all other code in the system.
7120 The pointer passed into cmpxchg must have alignment greater than or
7121 equal to the size in memory of the operand.
7126 The contents of memory at the location specified by the '``<pointer>``' operand
7127 is read and compared to '``<cmp>``'; if the read value is the equal, the
7128 '``<new>``' is written. The original value at the location is returned, together
7129 with a flag indicating success (true) or failure (false).
7131 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
7132 permitted: the operation may not write ``<new>`` even if the comparison
7135 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
7136 if the value loaded equals ``cmp``.
7138 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
7139 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
7140 load with an ordering parameter determined the second ordering parameter.
7145 .. code-block:: llvm
7148 %orig = atomic load i32, i32* %ptr unordered ; yields i32
7152 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
7153 %squared = mul i32 %cmp, %cmp
7154 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
7155 %value_loaded = extractvalue { i32, i1 } %val_success, 0
7156 %success = extractvalue { i32, i1 } %val_success, 1
7157 br i1 %success, label %done, label %loop
7164 '``atomicrmw``' Instruction
7165 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7172 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
7177 The '``atomicrmw``' instruction is used to atomically modify memory.
7182 There are three arguments to the '``atomicrmw``' instruction: an
7183 operation to apply, an address whose value to modify, an argument to the
7184 operation. The operation must be one of the following keywords:
7198 The type of '<value>' must be an integer type whose bit width is a power
7199 of two greater than or equal to eight and less than or equal to a
7200 target-specific size limit. The type of the '``<pointer>``' operand must
7201 be a pointer to that type. If the ``atomicrmw`` is marked as
7202 ``volatile``, then the optimizer is not allowed to modify the number or
7203 order of execution of this ``atomicrmw`` with other :ref:`volatile
7204 operations <volatile>`.
7209 The contents of memory at the location specified by the '``<pointer>``'
7210 operand are atomically read, modified, and written back. The original
7211 value at the location is returned. The modification is specified by the
7214 - xchg: ``*ptr = val``
7215 - add: ``*ptr = *ptr + val``
7216 - sub: ``*ptr = *ptr - val``
7217 - and: ``*ptr = *ptr & val``
7218 - nand: ``*ptr = ~(*ptr & val)``
7219 - or: ``*ptr = *ptr | val``
7220 - xor: ``*ptr = *ptr ^ val``
7221 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
7222 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
7223 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
7225 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
7231 .. code-block:: llvm
7233 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
7235 .. _i_getelementptr:
7237 '``getelementptr``' Instruction
7238 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7245 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7246 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7247 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
7252 The '``getelementptr``' instruction is used to get the address of a
7253 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
7254 address calculation only and does not access memory. The instruction can also
7255 be used to calculate a vector of such addresses.
7260 The first argument is always a type used as the basis for the calculations.
7261 The second argument is always a pointer or a vector of pointers, and is the
7262 base address to start from. The remaining arguments are indices
7263 that indicate which of the elements of the aggregate object are indexed.
7264 The interpretation of each index is dependent on the type being indexed
7265 into. The first index always indexes the pointer value given as the
7266 first argument, the second index indexes a value of the type pointed to
7267 (not necessarily the value directly pointed to, since the first index
7268 can be non-zero), etc. The first type indexed into must be a pointer
7269 value, subsequent types can be arrays, vectors, and structs. Note that
7270 subsequent types being indexed into can never be pointers, since that
7271 would require loading the pointer before continuing calculation.
7273 The type of each index argument depends on the type it is indexing into.
7274 When indexing into a (optionally packed) structure, only ``i32`` integer
7275 **constants** are allowed (when using a vector of indices they must all
7276 be the **same** ``i32`` integer constant). When indexing into an array,
7277 pointer or vector, integers of any width are allowed, and they are not
7278 required to be constant. These integers are treated as signed values
7281 For example, let's consider a C code fragment and how it gets compiled
7297 int *foo(struct ST *s) {
7298 return &s[1].Z.B[5][13];
7301 The LLVM code generated by Clang is:
7303 .. code-block:: llvm
7305 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
7306 %struct.ST = type { i32, double, %struct.RT }
7308 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
7310 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
7317 In the example above, the first index is indexing into the
7318 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
7319 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
7320 indexes into the third element of the structure, yielding a
7321 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
7322 structure. The third index indexes into the second element of the
7323 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
7324 dimensions of the array are subscripted into, yielding an '``i32``'
7325 type. The '``getelementptr``' instruction returns a pointer to this
7326 element, thus computing a value of '``i32*``' type.
7328 Note that it is perfectly legal to index partially through a structure,
7329 returning a pointer to an inner element. Because of this, the LLVM code
7330 for the given testcase is equivalent to:
7332 .. code-block:: llvm
7334 define i32* @foo(%struct.ST* %s) {
7335 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
7336 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
7337 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
7338 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
7339 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
7343 If the ``inbounds`` keyword is present, the result value of the
7344 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
7345 pointer is not an *in bounds* address of an allocated object, or if any
7346 of the addresses that would be formed by successive addition of the
7347 offsets implied by the indices to the base address with infinitely
7348 precise signed arithmetic are not an *in bounds* address of that
7349 allocated object. The *in bounds* addresses for an allocated object are
7350 all the addresses that point into the object, plus the address one byte
7351 past the end. In cases where the base is a vector of pointers the
7352 ``inbounds`` keyword applies to each of the computations element-wise.
7354 If the ``inbounds`` keyword is not present, the offsets are added to the
7355 base address with silently-wrapping two's complement arithmetic. If the
7356 offsets have a different width from the pointer, they are sign-extended
7357 or truncated to the width of the pointer. The result value of the
7358 ``getelementptr`` may be outside the object pointed to by the base
7359 pointer. The result value may not necessarily be used to access memory
7360 though, even if it happens to point into allocated storage. See the
7361 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
7364 The getelementptr instruction is often confusing. For some more insight
7365 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
7370 .. code-block:: llvm
7372 ; yields [12 x i8]*:aptr
7373 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
7375 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
7377 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
7379 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
7384 The ``getelementptr`` returns a vector of pointers, instead of a single address,
7385 when one or more of its arguments is a vector. In such cases, all vector
7386 arguments should have the same number of elements, and every scalar argument
7387 will be effectively broadcast into a vector during address calculation.
7389 .. code-block:: llvm
7391 ; All arguments are vectors:
7392 ; A[i] = ptrs[i] + offsets[i]*sizeof(i8)
7393 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets
7395 ; Add the same scalar offset to each pointer of a vector:
7396 ; A[i] = ptrs[i] + offset*sizeof(i8)
7397 %A = getelementptr i8, <4 x i8*> %ptrs, i64 %offset
7399 ; Add distinct offsets to the same pointer:
7400 ; A[i] = ptr + offsets[i]*sizeof(i8)
7401 %A = getelementptr i8, i8* %ptr, <4 x i64> %offsets
7403 ; In all cases described above the type of the result is <4 x i8*>
7405 The two following instructions are equivalent:
7407 .. code-block:: llvm
7409 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7410 <4 x i32> <i32 2, i32 2, i32 2, i32 2>,
7411 <4 x i32> <i32 1, i32 1, i32 1, i32 1>,
7413 <4 x i64> <i64 13, i64 13, i64 13, i64 13>
7415 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7416 i32 2, i32 1, <4 x i32> %ind4, i64 13
7418 Let's look at the C code, where the vector version of ``getelementptr``
7423 // Let's assume that we vectorize the following loop:
7424 double *A, B; int *C;
7425 for (int i = 0; i < size; ++i) {
7429 .. code-block:: llvm
7431 ; get pointers for 8 elements from array B
7432 %ptrs = getelementptr double, double* %B, <8 x i32> %C
7433 ; load 8 elements from array B into A
7434 %A = call <8 x double> @llvm.masked.gather.v8f64(<8 x double*> %ptrs,
7435 i32 8, <8 x i1> %mask, <8 x double> %passthru)
7437 Conversion Operations
7438 ---------------------
7440 The instructions in this category are the conversion instructions
7441 (casting) which all take a single operand and a type. They perform
7442 various bit conversions on the operand.
7444 '``trunc .. to``' Instruction
7445 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7452 <result> = trunc <ty> <value> to <ty2> ; yields ty2
7457 The '``trunc``' instruction truncates its operand to the type ``ty2``.
7462 The '``trunc``' instruction takes a value to trunc, and a type to trunc
7463 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
7464 of the same number of integers. The bit size of the ``value`` must be
7465 larger than the bit size of the destination type, ``ty2``. Equal sized
7466 types are not allowed.
7471 The '``trunc``' instruction truncates the high order bits in ``value``
7472 and converts the remaining bits to ``ty2``. Since the source size must
7473 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
7474 It will always truncate bits.
7479 .. code-block:: llvm
7481 %X = trunc i32 257 to i8 ; yields i8:1
7482 %Y = trunc i32 123 to i1 ; yields i1:true
7483 %Z = trunc i32 122 to i1 ; yields i1:false
7484 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
7486 '``zext .. to``' Instruction
7487 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7494 <result> = zext <ty> <value> to <ty2> ; yields ty2
7499 The '``zext``' instruction zero extends its operand to type ``ty2``.
7504 The '``zext``' instruction takes a value to cast, and a type to cast it
7505 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7506 the same number of integers. The bit size of the ``value`` must be
7507 smaller than the bit size of the destination type, ``ty2``.
7512 The ``zext`` fills the high order bits of the ``value`` with zero bits
7513 until it reaches the size of the destination type, ``ty2``.
7515 When zero extending from i1, the result will always be either 0 or 1.
7520 .. code-block:: llvm
7522 %X = zext i32 257 to i64 ; yields i64:257
7523 %Y = zext i1 true to i32 ; yields i32:1
7524 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7526 '``sext .. to``' Instruction
7527 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7534 <result> = sext <ty> <value> to <ty2> ; yields ty2
7539 The '``sext``' sign extends ``value`` to the type ``ty2``.
7544 The '``sext``' instruction takes a value to cast, and a type to cast it
7545 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7546 the same number of integers. The bit size of the ``value`` must be
7547 smaller than the bit size of the destination type, ``ty2``.
7552 The '``sext``' instruction performs a sign extension by copying the sign
7553 bit (highest order bit) of the ``value`` until it reaches the bit size
7554 of the type ``ty2``.
7556 When sign extending from i1, the extension always results in -1 or 0.
7561 .. code-block:: llvm
7563 %X = sext i8 -1 to i16 ; yields i16 :65535
7564 %Y = sext i1 true to i32 ; yields i32:-1
7565 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7567 '``fptrunc .. to``' Instruction
7568 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7575 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
7580 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
7585 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
7586 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
7587 The size of ``value`` must be larger than the size of ``ty2``. This
7588 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
7593 The '``fptrunc``' instruction casts a ``value`` from a larger
7594 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
7595 point <t_floating>` type. If the value cannot fit (i.e. overflows) within the
7596 destination type, ``ty2``, then the results are undefined. If the cast produces
7597 an inexact result, how rounding is performed (e.g. truncation, also known as
7598 round to zero) is undefined.
7603 .. code-block:: llvm
7605 %X = fptrunc double 123.0 to float ; yields float:123.0
7606 %Y = fptrunc double 1.0E+300 to float ; yields undefined
7608 '``fpext .. to``' Instruction
7609 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7616 <result> = fpext <ty> <value> to <ty2> ; yields ty2
7621 The '``fpext``' extends a floating point ``value`` to a larger floating
7627 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
7628 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
7629 to. The source type must be smaller than the destination type.
7634 The '``fpext``' instruction extends the ``value`` from a smaller
7635 :ref:`floating point <t_floating>` type to a larger :ref:`floating
7636 point <t_floating>` type. The ``fpext`` cannot be used to make a
7637 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
7638 *no-op cast* for a floating point cast.
7643 .. code-block:: llvm
7645 %X = fpext float 3.125 to double ; yields double:3.125000e+00
7646 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
7648 '``fptoui .. to``' Instruction
7649 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7656 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
7661 The '``fptoui``' converts a floating point ``value`` to its unsigned
7662 integer equivalent of type ``ty2``.
7667 The '``fptoui``' instruction takes a value to cast, which must be a
7668 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7669 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7670 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7671 type with the same number of elements as ``ty``
7676 The '``fptoui``' instruction converts its :ref:`floating
7677 point <t_floating>` operand into the nearest (rounding towards zero)
7678 unsigned integer value. If the value cannot fit in ``ty2``, the results
7684 .. code-block:: llvm
7686 %X = fptoui double 123.0 to i32 ; yields i32:123
7687 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
7688 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
7690 '``fptosi .. to``' Instruction
7691 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7698 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
7703 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
7704 ``value`` to type ``ty2``.
7709 The '``fptosi``' instruction takes a value to cast, which must be a
7710 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7711 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7712 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7713 type with the same number of elements as ``ty``
7718 The '``fptosi``' instruction converts its :ref:`floating
7719 point <t_floating>` operand into the nearest (rounding towards zero)
7720 signed integer value. If the value cannot fit in ``ty2``, the results
7726 .. code-block:: llvm
7728 %X = fptosi double -123.0 to i32 ; yields i32:-123
7729 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
7730 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
7732 '``uitofp .. to``' Instruction
7733 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7740 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
7745 The '``uitofp``' instruction regards ``value`` as an unsigned integer
7746 and converts that value to the ``ty2`` type.
7751 The '``uitofp``' instruction takes a value to cast, which must be a
7752 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7753 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7754 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7755 type with the same number of elements as ``ty``
7760 The '``uitofp``' instruction interprets its operand as an unsigned
7761 integer quantity and converts it to the corresponding floating point
7762 value. If the value cannot fit in the floating point value, the results
7768 .. code-block:: llvm
7770 %X = uitofp i32 257 to float ; yields float:257.0
7771 %Y = uitofp i8 -1 to double ; yields double:255.0
7773 '``sitofp .. to``' Instruction
7774 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7781 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
7786 The '``sitofp``' instruction regards ``value`` as a signed integer and
7787 converts that value to the ``ty2`` type.
7792 The '``sitofp``' instruction takes a value to cast, which must be a
7793 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7794 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7795 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7796 type with the same number of elements as ``ty``
7801 The '``sitofp``' instruction interprets its operand as a signed integer
7802 quantity and converts it to the corresponding floating point value. If
7803 the value cannot fit in the floating point value, the results are
7809 .. code-block:: llvm
7811 %X = sitofp i32 257 to float ; yields float:257.0
7812 %Y = sitofp i8 -1 to double ; yields double:-1.0
7816 '``ptrtoint .. to``' Instruction
7817 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7824 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
7829 The '``ptrtoint``' instruction converts the pointer or a vector of
7830 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
7835 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
7836 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
7837 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
7838 a vector of integers type.
7843 The '``ptrtoint``' instruction converts ``value`` to integer type
7844 ``ty2`` by interpreting the pointer value as an integer and either
7845 truncating or zero extending that value to the size of the integer type.
7846 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
7847 ``value`` is larger than ``ty2`` then a truncation is done. If they are
7848 the same size, then nothing is done (*no-op cast*) other than a type
7854 .. code-block:: llvm
7856 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
7857 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
7858 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
7862 '``inttoptr .. to``' Instruction
7863 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7870 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
7875 The '``inttoptr``' instruction converts an integer ``value`` to a
7876 pointer type, ``ty2``.
7881 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
7882 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
7888 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
7889 applying either a zero extension or a truncation depending on the size
7890 of the integer ``value``. If ``value`` is larger than the size of a
7891 pointer then a truncation is done. If ``value`` is smaller than the size
7892 of a pointer then a zero extension is done. If they are the same size,
7893 nothing is done (*no-op cast*).
7898 .. code-block:: llvm
7900 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
7901 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
7902 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
7903 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
7907 '``bitcast .. to``' Instruction
7908 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7915 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
7920 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
7926 The '``bitcast``' instruction takes a value to cast, which must be a
7927 non-aggregate first class value, and a type to cast it to, which must
7928 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
7929 bit sizes of ``value`` and the destination type, ``ty2``, must be
7930 identical. If the source type is a pointer, the destination type must
7931 also be a pointer of the same size. This instruction supports bitwise
7932 conversion of vectors to integers and to vectors of other types (as
7933 long as they have the same size).
7938 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
7939 is always a *no-op cast* because no bits change with this
7940 conversion. The conversion is done as if the ``value`` had been stored
7941 to memory and read back as type ``ty2``. Pointer (or vector of
7942 pointers) types may only be converted to other pointer (or vector of
7943 pointers) types with the same address space through this instruction.
7944 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
7945 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
7950 .. code-block:: llvm
7952 %X = bitcast i8 255 to i8 ; yields i8 :-1
7953 %Y = bitcast i32* %x to sint* ; yields sint*:%x
7954 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
7955 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
7957 .. _i_addrspacecast:
7959 '``addrspacecast .. to``' Instruction
7960 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7967 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
7972 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
7973 address space ``n`` to type ``pty2`` in address space ``m``.
7978 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
7979 to cast and a pointer type to cast it to, which must have a different
7985 The '``addrspacecast``' instruction converts the pointer value
7986 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
7987 value modification, depending on the target and the address space
7988 pair. Pointer conversions within the same address space must be
7989 performed with the ``bitcast`` instruction. Note that if the address space
7990 conversion is legal then both result and operand refer to the same memory
7996 .. code-block:: llvm
7998 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
7999 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
8000 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
8007 The instructions in this category are the "miscellaneous" instructions,
8008 which defy better classification.
8012 '``icmp``' Instruction
8013 ^^^^^^^^^^^^^^^^^^^^^^
8020 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8025 The '``icmp``' instruction returns a boolean value or a vector of
8026 boolean values based on comparison of its two integer, integer vector,
8027 pointer, or pointer vector operands.
8032 The '``icmp``' instruction takes three operands. The first operand is
8033 the condition code indicating the kind of comparison to perform. It is
8034 not a value, just a keyword. The possible condition code are:
8037 #. ``ne``: not equal
8038 #. ``ugt``: unsigned greater than
8039 #. ``uge``: unsigned greater or equal
8040 #. ``ult``: unsigned less than
8041 #. ``ule``: unsigned less or equal
8042 #. ``sgt``: signed greater than
8043 #. ``sge``: signed greater or equal
8044 #. ``slt``: signed less than
8045 #. ``sle``: signed less or equal
8047 The remaining two arguments must be :ref:`integer <t_integer>` or
8048 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
8049 must also be identical types.
8054 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
8055 code given as ``cond``. The comparison performed always yields either an
8056 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
8058 #. ``eq``: yields ``true`` if the operands are equal, ``false``
8059 otherwise. No sign interpretation is necessary or performed.
8060 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
8061 otherwise. No sign interpretation is necessary or performed.
8062 #. ``ugt``: interprets the operands as unsigned values and yields
8063 ``true`` if ``op1`` is greater than ``op2``.
8064 #. ``uge``: interprets the operands as unsigned values and yields
8065 ``true`` if ``op1`` is greater than or equal to ``op2``.
8066 #. ``ult``: interprets the operands as unsigned values and yields
8067 ``true`` if ``op1`` is less than ``op2``.
8068 #. ``ule``: interprets the operands as unsigned values and yields
8069 ``true`` if ``op1`` is less than or equal to ``op2``.
8070 #. ``sgt``: interprets the operands as signed values and yields ``true``
8071 if ``op1`` is greater than ``op2``.
8072 #. ``sge``: interprets the operands as signed values and yields ``true``
8073 if ``op1`` is greater than or equal to ``op2``.
8074 #. ``slt``: interprets the operands as signed values and yields ``true``
8075 if ``op1`` is less than ``op2``.
8076 #. ``sle``: interprets the operands as signed values and yields ``true``
8077 if ``op1`` is less than or equal to ``op2``.
8079 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
8080 are compared as if they were integers.
8082 If the operands are integer vectors, then they are compared element by
8083 element. The result is an ``i1`` vector with the same number of elements
8084 as the values being compared. Otherwise, the result is an ``i1``.
8089 .. code-block:: llvm
8091 <result> = icmp eq i32 4, 5 ; yields: result=false
8092 <result> = icmp ne float* %X, %X ; yields: result=false
8093 <result> = icmp ult i16 4, 5 ; yields: result=true
8094 <result> = icmp sgt i16 4, 5 ; yields: result=false
8095 <result> = icmp ule i16 -4, 5 ; yields: result=false
8096 <result> = icmp sge i16 4, 5 ; yields: result=false
8098 Note that the code generator does not yet support vector types with the
8099 ``icmp`` instruction.
8103 '``fcmp``' Instruction
8104 ^^^^^^^^^^^^^^^^^^^^^^
8111 <result> = fcmp [fast-math flags]* <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8116 The '``fcmp``' instruction returns a boolean value or vector of boolean
8117 values based on comparison of its operands.
8119 If the operands are floating point scalars, then the result type is a
8120 boolean (:ref:`i1 <t_integer>`).
8122 If the operands are floating point vectors, then the result type is a
8123 vector of boolean with the same number of elements as the operands being
8129 The '``fcmp``' instruction takes three operands. The first operand is
8130 the condition code indicating the kind of comparison to perform. It is
8131 not a value, just a keyword. The possible condition code are:
8133 #. ``false``: no comparison, always returns false
8134 #. ``oeq``: ordered and equal
8135 #. ``ogt``: ordered and greater than
8136 #. ``oge``: ordered and greater than or equal
8137 #. ``olt``: ordered and less than
8138 #. ``ole``: ordered and less than or equal
8139 #. ``one``: ordered and not equal
8140 #. ``ord``: ordered (no nans)
8141 #. ``ueq``: unordered or equal
8142 #. ``ugt``: unordered or greater than
8143 #. ``uge``: unordered or greater than or equal
8144 #. ``ult``: unordered or less than
8145 #. ``ule``: unordered or less than or equal
8146 #. ``une``: unordered or not equal
8147 #. ``uno``: unordered (either nans)
8148 #. ``true``: no comparison, always returns true
8150 *Ordered* means that neither operand is a QNAN while *unordered* means
8151 that either operand may be a QNAN.
8153 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
8154 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
8155 type. They must have identical types.
8160 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
8161 condition code given as ``cond``. If the operands are vectors, then the
8162 vectors are compared element by element. Each comparison performed
8163 always yields an :ref:`i1 <t_integer>` result, as follows:
8165 #. ``false``: always yields ``false``, regardless of operands.
8166 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
8167 is equal to ``op2``.
8168 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
8169 is greater than ``op2``.
8170 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
8171 is greater than or equal to ``op2``.
8172 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
8173 is less than ``op2``.
8174 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
8175 is less than or equal to ``op2``.
8176 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
8177 is not equal to ``op2``.
8178 #. ``ord``: yields ``true`` if both operands are not a QNAN.
8179 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
8181 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
8182 greater than ``op2``.
8183 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
8184 greater than or equal to ``op2``.
8185 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
8187 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
8188 less than or equal to ``op2``.
8189 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
8190 not equal to ``op2``.
8191 #. ``uno``: yields ``true`` if either operand is a QNAN.
8192 #. ``true``: always yields ``true``, regardless of operands.
8194 The ``fcmp`` instruction can also optionally take any number of
8195 :ref:`fast-math flags <fastmath>`, which are optimization hints to enable
8196 otherwise unsafe floating point optimizations.
8198 Any set of fast-math flags are legal on an ``fcmp`` instruction, but the
8199 only flags that have any effect on its semantics are those that allow
8200 assumptions to be made about the values of input arguments; namely
8201 ``nnan``, ``ninf``, and ``nsz``. See :ref:`fastmath` for more information.
8206 .. code-block:: llvm
8208 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
8209 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
8210 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
8211 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
8213 Note that the code generator does not yet support vector types with the
8214 ``fcmp`` instruction.
8218 '``phi``' Instruction
8219 ^^^^^^^^^^^^^^^^^^^^^
8226 <result> = phi <ty> [ <val0>, <label0>], ...
8231 The '``phi``' instruction is used to implement the φ node in the SSA
8232 graph representing the function.
8237 The type of the incoming values is specified with the first type field.
8238 After this, the '``phi``' instruction takes a list of pairs as
8239 arguments, with one pair for each predecessor basic block of the current
8240 block. Only values of :ref:`first class <t_firstclass>` type may be used as
8241 the value arguments to the PHI node. Only labels may be used as the
8244 There must be no non-phi instructions between the start of a basic block
8245 and the PHI instructions: i.e. PHI instructions must be first in a basic
8248 For the purposes of the SSA form, the use of each incoming value is
8249 deemed to occur on the edge from the corresponding predecessor block to
8250 the current block (but after any definition of an '``invoke``'
8251 instruction's return value on the same edge).
8256 At runtime, the '``phi``' instruction logically takes on the value
8257 specified by the pair corresponding to the predecessor basic block that
8258 executed just prior to the current block.
8263 .. code-block:: llvm
8265 Loop: ; Infinite loop that counts from 0 on up...
8266 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
8267 %nextindvar = add i32 %indvar, 1
8272 '``select``' Instruction
8273 ^^^^^^^^^^^^^^^^^^^^^^^^
8280 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
8282 selty is either i1 or {<N x i1>}
8287 The '``select``' instruction is used to choose one value based on a
8288 condition, without IR-level branching.
8293 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
8294 values indicating the condition, and two values of the same :ref:`first
8295 class <t_firstclass>` type.
8300 If the condition is an i1 and it evaluates to 1, the instruction returns
8301 the first value argument; otherwise, it returns the second value
8304 If the condition is a vector of i1, then the value arguments must be
8305 vectors of the same size, and the selection is done element by element.
8307 If the condition is an i1 and the value arguments are vectors of the
8308 same size, then an entire vector is selected.
8313 .. code-block:: llvm
8315 %X = select i1 true, i8 17, i8 42 ; yields i8:17
8319 '``call``' Instruction
8320 ^^^^^^^^^^^^^^^^^^^^^^
8327 <result> = [tail | musttail | notail ] call [fast-math flags] [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
8333 The '``call``' instruction represents a simple function call.
8338 This instruction requires several arguments:
8340 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
8341 should perform tail call optimization. The ``tail`` marker is a hint that
8342 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
8343 means that the call must be tail call optimized in order for the program to
8344 be correct. The ``musttail`` marker provides these guarantees:
8346 #. The call will not cause unbounded stack growth if it is part of a
8347 recursive cycle in the call graph.
8348 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
8351 Both markers imply that the callee does not access allocas or varargs from
8352 the caller. Calls marked ``musttail`` must obey the following additional
8355 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
8356 or a pointer bitcast followed by a ret instruction.
8357 - The ret instruction must return the (possibly bitcasted) value
8358 produced by the call or void.
8359 - The caller and callee prototypes must match. Pointer types of
8360 parameters or return types may differ in pointee type, but not
8362 - The calling conventions of the caller and callee must match.
8363 - All ABI-impacting function attributes, such as sret, byval, inreg,
8364 returned, and inalloca, must match.
8365 - The callee must be varargs iff the caller is varargs. Bitcasting a
8366 non-varargs function to the appropriate varargs type is legal so
8367 long as the non-varargs prefixes obey the other rules.
8369 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
8370 the following conditions are met:
8372 - Caller and callee both have the calling convention ``fastcc``.
8373 - The call is in tail position (ret immediately follows call and ret
8374 uses value of call or is void).
8375 - Option ``-tailcallopt`` is enabled, or
8376 ``llvm::GuaranteedTailCallOpt`` is ``true``.
8377 - `Platform-specific constraints are
8378 met. <CodeGenerator.html#tailcallopt>`_
8380 #. The optional ``notail`` marker indicates that the optimizers should not add
8381 ``tail`` or ``musttail`` markers to the call. It is used to prevent tail
8382 call optimization from being performed on the call.
8384 #. The optional ``fast-math flags`` marker indicates that the call has one or more
8385 :ref:`fast-math flags <fastmath>`, which are optimization hints to enable
8386 otherwise unsafe floating-point optimizations. Fast-math flags are only valid
8387 for calls that return a floating-point scalar or vector type.
8389 #. The optional "cconv" marker indicates which :ref:`calling
8390 convention <callingconv>` the call should use. If none is
8391 specified, the call defaults to using C calling conventions. The
8392 calling convention of the call must match the calling convention of
8393 the target function, or else the behavior is undefined.
8394 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
8395 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
8397 #. '``ty``': the type of the call instruction itself which is also the
8398 type of the return value. Functions that return no value are marked
8400 #. '``fnty``': shall be the signature of the pointer to function value
8401 being invoked. The argument types must match the types implied by
8402 this signature. This type can be omitted if the function is not
8403 varargs and if the function type does not return a pointer to a
8405 #. '``fnptrval``': An LLVM value containing a pointer to a function to
8406 be invoked. In most cases, this is a direct function invocation, but
8407 indirect ``call``'s are just as possible, calling an arbitrary pointer
8409 #. '``function args``': argument list whose types match the function
8410 signature argument types and parameter attributes. All arguments must
8411 be of :ref:`first class <t_firstclass>` type. If the function signature
8412 indicates the function accepts a variable number of arguments, the
8413 extra arguments can be specified.
8414 #. The optional :ref:`function attributes <fnattrs>` list. Only
8415 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
8416 attributes are valid here.
8417 #. The optional :ref:`operand bundles <opbundles>` list.
8422 The '``call``' instruction is used to cause control flow to transfer to
8423 a specified function, with its incoming arguments bound to the specified
8424 values. Upon a '``ret``' instruction in the called function, control
8425 flow continues with the instruction after the function call, and the
8426 return value of the function is bound to the result argument.
8431 .. code-block:: llvm
8433 %retval = call i32 @test(i32 %argc)
8434 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
8435 %X = tail call i32 @foo() ; yields i32
8436 %Y = tail call fastcc i32 @foo() ; yields i32
8437 call void %foo(i8 97 signext)
8439 %struct.A = type { i32, i8 }
8440 %r = call %struct.A @foo() ; yields { i32, i8 }
8441 %gr = extractvalue %struct.A %r, 0 ; yields i32
8442 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
8443 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
8444 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
8446 llvm treats calls to some functions with names and arguments that match
8447 the standard C99 library as being the C99 library functions, and may
8448 perform optimizations or generate code for them under that assumption.
8449 This is something we'd like to change in the future to provide better
8450 support for freestanding environments and non-C-based languages.
8454 '``va_arg``' Instruction
8455 ^^^^^^^^^^^^^^^^^^^^^^^^
8462 <resultval> = va_arg <va_list*> <arglist>, <argty>
8467 The '``va_arg``' instruction is used to access arguments passed through
8468 the "variable argument" area of a function call. It is used to implement
8469 the ``va_arg`` macro in C.
8474 This instruction takes a ``va_list*`` value and the type of the
8475 argument. It returns a value of the specified argument type and
8476 increments the ``va_list`` to point to the next argument. The actual
8477 type of ``va_list`` is target specific.
8482 The '``va_arg``' instruction loads an argument of the specified type
8483 from the specified ``va_list`` and causes the ``va_list`` to point to
8484 the next argument. For more information, see the variable argument
8485 handling :ref:`Intrinsic Functions <int_varargs>`.
8487 It is legal for this instruction to be called in a function which does
8488 not take a variable number of arguments, for example, the ``vfprintf``
8491 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
8492 function <intrinsics>` because it takes a type as an argument.
8497 See the :ref:`variable argument processing <int_varargs>` section.
8499 Note that the code generator does not yet fully support va\_arg on many
8500 targets. Also, it does not currently support va\_arg with aggregate
8501 types on any target.
8505 '``landingpad``' Instruction
8506 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8513 <resultval> = landingpad <resultty> <clause>+
8514 <resultval> = landingpad <resultty> cleanup <clause>*
8516 <clause> := catch <type> <value>
8517 <clause> := filter <array constant type> <array constant>
8522 The '``landingpad``' instruction is used by `LLVM's exception handling
8523 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8524 is a landing pad --- one where the exception lands, and corresponds to the
8525 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
8526 defines values supplied by the :ref:`personality function <personalityfn>` upon
8527 re-entry to the function. The ``resultval`` has the type ``resultty``.
8533 ``cleanup`` flag indicates that the landing pad block is a cleanup.
8535 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
8536 contains the global variable representing the "type" that may be caught
8537 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
8538 clause takes an array constant as its argument. Use
8539 "``[0 x i8**] undef``" for a filter which cannot throw. The
8540 '``landingpad``' instruction must contain *at least* one ``clause`` or
8541 the ``cleanup`` flag.
8546 The '``landingpad``' instruction defines the values which are set by the
8547 :ref:`personality function <personalityfn>` upon re-entry to the function, and
8548 therefore the "result type" of the ``landingpad`` instruction. As with
8549 calling conventions, how the personality function results are
8550 represented in LLVM IR is target specific.
8552 The clauses are applied in order from top to bottom. If two
8553 ``landingpad`` instructions are merged together through inlining, the
8554 clauses from the calling function are appended to the list of clauses.
8555 When the call stack is being unwound due to an exception being thrown,
8556 the exception is compared against each ``clause`` in turn. If it doesn't
8557 match any of the clauses, and the ``cleanup`` flag is not set, then
8558 unwinding continues further up the call stack.
8560 The ``landingpad`` instruction has several restrictions:
8562 - A landing pad block is a basic block which is the unwind destination
8563 of an '``invoke``' instruction.
8564 - A landing pad block must have a '``landingpad``' instruction as its
8565 first non-PHI instruction.
8566 - There can be only one '``landingpad``' instruction within the landing
8568 - A basic block that is not a landing pad block may not include a
8569 '``landingpad``' instruction.
8574 .. code-block:: llvm
8576 ;; A landing pad which can catch an integer.
8577 %res = landingpad { i8*, i32 }
8579 ;; A landing pad that is a cleanup.
8580 %res = landingpad { i8*, i32 }
8582 ;; A landing pad which can catch an integer and can only throw a double.
8583 %res = landingpad { i8*, i32 }
8585 filter [1 x i8**] [@_ZTId]
8589 '``cleanuppad``' Instruction
8590 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8597 <resultval> = cleanuppad within <parent> [<args>*]
8602 The '``cleanuppad``' instruction is used by `LLVM's exception handling
8603 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8604 is a cleanup block --- one where a personality routine attempts to
8605 transfer control to run cleanup actions.
8606 The ``args`` correspond to whatever additional
8607 information the :ref:`personality function <personalityfn>` requires to
8608 execute the cleanup.
8609 The ``resultval`` has the type :ref:`token <t_token>` and is used to
8610 match the ``cleanuppad`` to corresponding :ref:`cleanuprets <i_cleanupret>`.
8611 The ``parent`` argument is the token of the funclet that contains the
8612 ``cleanuppad`` instruction. If the ``cleanuppad`` is not inside a funclet,
8613 this operand may be the token ``none``.
8618 The instruction takes a list of arbitrary values which are interpreted
8619 by the :ref:`personality function <personalityfn>`.
8624 When the call stack is being unwound due to an exception being thrown,
8625 the :ref:`personality function <personalityfn>` transfers control to the
8626 ``cleanuppad`` with the aid of the personality-specific arguments.
8627 As with calling conventions, how the personality function results are
8628 represented in LLVM IR is target specific.
8630 The ``cleanuppad`` instruction has several restrictions:
8632 - A cleanup block is a basic block which is the unwind destination of
8633 an exceptional instruction.
8634 - A cleanup block must have a '``cleanuppad``' instruction as its
8635 first non-PHI instruction.
8636 - There can be only one '``cleanuppad``' instruction within the
8638 - A basic block that is not a cleanup block may not include a
8639 '``cleanuppad``' instruction.
8641 Executing a ``cleanuppad`` instruction constitutes "entering" that pad.
8642 The pad may then be "exited" in one of three ways:
8644 1) explicitly via a ``cleanupret`` that consumes it. Executing such a ``cleanupret``
8645 is undefined behavior if any descendant pads have been entered but not yet
8647 2) implicitly via a call (which unwinds all the way to the current function's caller),
8648 or via a ``catchswitch`` or a ``cleanupret`` that unwinds to caller.
8649 3) implicitly via an unwind edge whose destination EH pad isn't a descendant of
8650 the ``cleanuppad``. When the ``cleanuppad`` is exited in this manner, it is
8651 undefined behavior if the destination EH pad has a parent which is not an
8652 ancestor of the ``cleanuppad`` being exited.
8654 It is undefined behavior for the ``cleanuppad`` to exit via an unwind edge which
8655 does not transitively unwind to the same destination as a constituent
8661 .. code-block:: llvm
8663 %tok = cleanuppad within %cs []
8670 LLVM supports the notion of an "intrinsic function". These functions
8671 have well known names and semantics and are required to follow certain
8672 restrictions. Overall, these intrinsics represent an extension mechanism
8673 for the LLVM language that does not require changing all of the
8674 transformations in LLVM when adding to the language (or the bitcode
8675 reader/writer, the parser, etc...).
8677 Intrinsic function names must all start with an "``llvm.``" prefix. This
8678 prefix is reserved in LLVM for intrinsic names; thus, function names may
8679 not begin with this prefix. Intrinsic functions must always be external
8680 functions: you cannot define the body of intrinsic functions. Intrinsic
8681 functions may only be used in call or invoke instructions: it is illegal
8682 to take the address of an intrinsic function. Additionally, because
8683 intrinsic functions are part of the LLVM language, it is required if any
8684 are added that they be documented here.
8686 Some intrinsic functions can be overloaded, i.e., the intrinsic
8687 represents a family of functions that perform the same operation but on
8688 different data types. Because LLVM can represent over 8 million
8689 different integer types, overloading is used commonly to allow an
8690 intrinsic function to operate on any integer type. One or more of the
8691 argument types or the result type can be overloaded to accept any
8692 integer type. Argument types may also be defined as exactly matching a
8693 previous argument's type or the result type. This allows an intrinsic
8694 function which accepts multiple arguments, but needs all of them to be
8695 of the same type, to only be overloaded with respect to a single
8696 argument or the result.
8698 Overloaded intrinsics will have the names of its overloaded argument
8699 types encoded into its function name, each preceded by a period. Only
8700 those types which are overloaded result in a name suffix. Arguments
8701 whose type is matched against another type do not. For example, the
8702 ``llvm.ctpop`` function can take an integer of any width and returns an
8703 integer of exactly the same integer width. This leads to a family of
8704 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
8705 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
8706 overloaded, and only one type suffix is required. Because the argument's
8707 type is matched against the return type, it does not require its own
8710 To learn how to add an intrinsic function, please see the `Extending
8711 LLVM Guide <ExtendingLLVM.html>`_.
8715 Variable Argument Handling Intrinsics
8716 -------------------------------------
8718 Variable argument support is defined in LLVM with the
8719 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
8720 functions. These functions are related to the similarly named macros
8721 defined in the ``<stdarg.h>`` header file.
8723 All of these functions operate on arguments that use a target-specific
8724 value type "``va_list``". The LLVM assembly language reference manual
8725 does not define what this type is, so all transformations should be
8726 prepared to handle these functions regardless of the type used.
8728 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
8729 variable argument handling intrinsic functions are used.
8731 .. code-block:: llvm
8733 ; This struct is different for every platform. For most platforms,
8734 ; it is merely an i8*.
8735 %struct.va_list = type { i8* }
8737 ; For Unix x86_64 platforms, va_list is the following struct:
8738 ; %struct.va_list = type { i32, i32, i8*, i8* }
8740 define i32 @test(i32 %X, ...) {
8741 ; Initialize variable argument processing
8742 %ap = alloca %struct.va_list
8743 %ap2 = bitcast %struct.va_list* %ap to i8*
8744 call void @llvm.va_start(i8* %ap2)
8746 ; Read a single integer argument
8747 %tmp = va_arg i8* %ap2, i32
8749 ; Demonstrate usage of llvm.va_copy and llvm.va_end
8751 %aq2 = bitcast i8** %aq to i8*
8752 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
8753 call void @llvm.va_end(i8* %aq2)
8755 ; Stop processing of arguments.
8756 call void @llvm.va_end(i8* %ap2)
8760 declare void @llvm.va_start(i8*)
8761 declare void @llvm.va_copy(i8*, i8*)
8762 declare void @llvm.va_end(i8*)
8766 '``llvm.va_start``' Intrinsic
8767 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8774 declare void @llvm.va_start(i8* <arglist>)
8779 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
8780 subsequent use by ``va_arg``.
8785 The argument is a pointer to a ``va_list`` element to initialize.
8790 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
8791 available in C. In a target-dependent way, it initializes the
8792 ``va_list`` element to which the argument points, so that the next call
8793 to ``va_arg`` will produce the first variable argument passed to the
8794 function. Unlike the C ``va_start`` macro, this intrinsic does not need
8795 to know the last argument of the function as the compiler can figure
8798 '``llvm.va_end``' Intrinsic
8799 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8806 declare void @llvm.va_end(i8* <arglist>)
8811 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
8812 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
8817 The argument is a pointer to a ``va_list`` to destroy.
8822 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
8823 available in C. In a target-dependent way, it destroys the ``va_list``
8824 element to which the argument points. Calls to
8825 :ref:`llvm.va_start <int_va_start>` and
8826 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
8831 '``llvm.va_copy``' Intrinsic
8832 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8839 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
8844 The '``llvm.va_copy``' intrinsic copies the current argument position
8845 from the source argument list to the destination argument list.
8850 The first argument is a pointer to a ``va_list`` element to initialize.
8851 The second argument is a pointer to a ``va_list`` element to copy from.
8856 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
8857 available in C. In a target-dependent way, it copies the source
8858 ``va_list`` element into the destination ``va_list`` element. This
8859 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
8860 arbitrarily complex and require, for example, memory allocation.
8862 Accurate Garbage Collection Intrinsics
8863 --------------------------------------
8865 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
8866 (GC) requires the frontend to generate code containing appropriate intrinsic
8867 calls and select an appropriate GC strategy which knows how to lower these
8868 intrinsics in a manner which is appropriate for the target collector.
8870 These intrinsics allow identification of :ref:`GC roots on the
8871 stack <int_gcroot>`, as well as garbage collector implementations that
8872 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
8873 Frontends for type-safe garbage collected languages should generate
8874 these intrinsics to make use of the LLVM garbage collectors. For more
8875 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
8877 Experimental Statepoint Intrinsics
8878 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8880 LLVM provides an second experimental set of intrinsics for describing garbage
8881 collection safepoints in compiled code. These intrinsics are an alternative
8882 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
8883 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
8884 differences in approach are covered in the `Garbage Collection with LLVM
8885 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
8886 described in :doc:`Statepoints`.
8890 '``llvm.gcroot``' Intrinsic
8891 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8898 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
8903 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
8904 the code generator, and allows some metadata to be associated with it.
8909 The first argument specifies the address of a stack object that contains
8910 the root pointer. The second pointer (which must be either a constant or
8911 a global value address) contains the meta-data to be associated with the
8917 At runtime, a call to this intrinsic stores a null pointer into the
8918 "ptrloc" location. At compile-time, the code generator generates
8919 information to allow the runtime to find the pointer at GC safe points.
8920 The '``llvm.gcroot``' intrinsic may only be used in a function which
8921 :ref:`specifies a GC algorithm <gc>`.
8925 '``llvm.gcread``' Intrinsic
8926 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8933 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
8938 The '``llvm.gcread``' intrinsic identifies reads of references from heap
8939 locations, allowing garbage collector implementations that require read
8945 The second argument is the address to read from, which should be an
8946 address allocated from the garbage collector. The first object is a
8947 pointer to the start of the referenced object, if needed by the language
8948 runtime (otherwise null).
8953 The '``llvm.gcread``' intrinsic has the same semantics as a load
8954 instruction, but may be replaced with substantially more complex code by
8955 the garbage collector runtime, as needed. The '``llvm.gcread``'
8956 intrinsic may only be used in a function which :ref:`specifies a GC
8961 '``llvm.gcwrite``' Intrinsic
8962 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8969 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
8974 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
8975 locations, allowing garbage collector implementations that require write
8976 barriers (such as generational or reference counting collectors).
8981 The first argument is the reference to store, the second is the start of
8982 the object to store it to, and the third is the address of the field of
8983 Obj to store to. If the runtime does not require a pointer to the
8984 object, Obj may be null.
8989 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
8990 instruction, but may be replaced with substantially more complex code by
8991 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
8992 intrinsic may only be used in a function which :ref:`specifies a GC
8995 Code Generator Intrinsics
8996 -------------------------
8998 These intrinsics are provided by LLVM to expose special features that
8999 may only be implemented with code generator support.
9001 '``llvm.returnaddress``' Intrinsic
9002 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9009 declare i8 *@llvm.returnaddress(i32 <level>)
9014 The '``llvm.returnaddress``' intrinsic attempts to compute a
9015 target-specific value indicating the return address of the current
9016 function or one of its callers.
9021 The argument to this intrinsic indicates which function to return the
9022 address for. Zero indicates the calling function, one indicates its
9023 caller, etc. The argument is **required** to be a constant integer
9029 The '``llvm.returnaddress``' intrinsic either returns a pointer
9030 indicating the return address of the specified call frame, or zero if it
9031 cannot be identified. The value returned by this intrinsic is likely to
9032 be incorrect or 0 for arguments other than zero, so it should only be
9033 used for debugging purposes.
9035 Note that calling this intrinsic does not prevent function inlining or
9036 other aggressive transformations, so the value returned may not be that
9037 of the obvious source-language caller.
9039 '``llvm.frameaddress``' Intrinsic
9040 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9047 declare i8* @llvm.frameaddress(i32 <level>)
9052 The '``llvm.frameaddress``' intrinsic attempts to return the
9053 target-specific frame pointer value for the specified stack frame.
9058 The argument to this intrinsic indicates which function to return the
9059 frame pointer for. Zero indicates the calling function, one indicates
9060 its caller, etc. The argument is **required** to be a constant integer
9066 The '``llvm.frameaddress``' intrinsic either returns a pointer
9067 indicating the frame address of the specified call frame, or zero if it
9068 cannot be identified. The value returned by this intrinsic is likely to
9069 be incorrect or 0 for arguments other than zero, so it should only be
9070 used for debugging purposes.
9072 Note that calling this intrinsic does not prevent function inlining or
9073 other aggressive transformations, so the value returned may not be that
9074 of the obvious source-language caller.
9076 '``llvm.localescape``' and '``llvm.localrecover``' Intrinsics
9077 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9084 declare void @llvm.localescape(...)
9085 declare i8* @llvm.localrecover(i8* %func, i8* %fp, i32 %idx)
9090 The '``llvm.localescape``' intrinsic escapes offsets of a collection of static
9091 allocas, and the '``llvm.localrecover``' intrinsic applies those offsets to a
9092 live frame pointer to recover the address of the allocation. The offset is
9093 computed during frame layout of the caller of ``llvm.localescape``.
9098 All arguments to '``llvm.localescape``' must be pointers to static allocas or
9099 casts of static allocas. Each function can only call '``llvm.localescape``'
9100 once, and it can only do so from the entry block.
9102 The ``func`` argument to '``llvm.localrecover``' must be a constant
9103 bitcasted pointer to a function defined in the current module. The code
9104 generator cannot determine the frame allocation offset of functions defined in
9107 The ``fp`` argument to '``llvm.localrecover``' must be a frame pointer of a
9108 call frame that is currently live. The return value of '``llvm.localaddress``'
9109 is one way to produce such a value, but various runtimes also expose a suitable
9110 pointer in platform-specific ways.
9112 The ``idx`` argument to '``llvm.localrecover``' indicates which alloca passed to
9113 '``llvm.localescape``' to recover. It is zero-indexed.
9118 These intrinsics allow a group of functions to share access to a set of local
9119 stack allocations of a one parent function. The parent function may call the
9120 '``llvm.localescape``' intrinsic once from the function entry block, and the
9121 child functions can use '``llvm.localrecover``' to access the escaped allocas.
9122 The '``llvm.localescape``' intrinsic blocks inlining, as inlining changes where
9123 the escaped allocas are allocated, which would break attempts to use
9124 '``llvm.localrecover``'.
9126 .. _int_read_register:
9127 .. _int_write_register:
9129 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
9130 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9137 declare i32 @llvm.read_register.i32(metadata)
9138 declare i64 @llvm.read_register.i64(metadata)
9139 declare void @llvm.write_register.i32(metadata, i32 @value)
9140 declare void @llvm.write_register.i64(metadata, i64 @value)
9146 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
9147 provides access to the named register. The register must be valid on
9148 the architecture being compiled to. The type needs to be compatible
9149 with the register being read.
9154 The '``llvm.read_register``' intrinsic returns the current value of the
9155 register, where possible. The '``llvm.write_register``' intrinsic sets
9156 the current value of the register, where possible.
9158 This is useful to implement named register global variables that need
9159 to always be mapped to a specific register, as is common practice on
9160 bare-metal programs including OS kernels.
9162 The compiler doesn't check for register availability or use of the used
9163 register in surrounding code, including inline assembly. Because of that,
9164 allocatable registers are not supported.
9166 Warning: So far it only works with the stack pointer on selected
9167 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
9168 work is needed to support other registers and even more so, allocatable
9173 '``llvm.stacksave``' Intrinsic
9174 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9181 declare i8* @llvm.stacksave()
9186 The '``llvm.stacksave``' intrinsic is used to remember the current state
9187 of the function stack, for use with
9188 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
9189 implementing language features like scoped automatic variable sized
9195 This intrinsic returns a opaque pointer value that can be passed to
9196 :ref:`llvm.stackrestore <int_stackrestore>`. When an
9197 ``llvm.stackrestore`` intrinsic is executed with a value saved from
9198 ``llvm.stacksave``, it effectively restores the state of the stack to
9199 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
9200 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
9201 were allocated after the ``llvm.stacksave`` was executed.
9203 .. _int_stackrestore:
9205 '``llvm.stackrestore``' Intrinsic
9206 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9213 declare void @llvm.stackrestore(i8* %ptr)
9218 The '``llvm.stackrestore``' intrinsic is used to restore the state of
9219 the function stack to the state it was in when the corresponding
9220 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
9221 useful for implementing language features like scoped automatic variable
9222 sized arrays in C99.
9227 See the description for :ref:`llvm.stacksave <int_stacksave>`.
9229 .. _int_get_dynamic_area_offset:
9231 '``llvm.get.dynamic.area.offset``' Intrinsic
9232 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9239 declare i32 @llvm.get.dynamic.area.offset.i32()
9240 declare i64 @llvm.get.dynamic.area.offset.i64()
9245 The '``llvm.get.dynamic.area.offset.*``' intrinsic family is used to
9246 get the offset from native stack pointer to the address of the most
9247 recent dynamic alloca on the caller's stack. These intrinsics are
9248 intendend for use in combination with
9249 :ref:`llvm.stacksave <int_stacksave>` to get a
9250 pointer to the most recent dynamic alloca. This is useful, for example,
9251 for AddressSanitizer's stack unpoisoning routines.
9256 These intrinsics return a non-negative integer value that can be used to
9257 get the address of the most recent dynamic alloca, allocated by :ref:`alloca <i_alloca>`
9258 on the caller's stack. In particular, for targets where stack grows downwards,
9259 adding this offset to the native stack pointer would get the address of the most
9260 recent dynamic alloca. For targets where stack grows upwards, the situation is a bit more
9261 complicated, because substracting this value from stack pointer would get the address
9262 one past the end of the most recent dynamic alloca.
9264 Although for most targets `llvm.get.dynamic.area.offset <int_get_dynamic_area_offset>`
9265 returns just a zero, for others, such as PowerPC and PowerPC64, it returns a
9266 compile-time-known constant value.
9268 The return value type of :ref:`llvm.get.dynamic.area.offset <int_get_dynamic_area_offset>`
9269 must match the target's generic address space's (address space 0) pointer type.
9271 '``llvm.prefetch``' Intrinsic
9272 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9279 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
9284 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
9285 insert a prefetch instruction if supported; otherwise, it is a noop.
9286 Prefetches have no effect on the behavior of the program but can change
9287 its performance characteristics.
9292 ``address`` is the address to be prefetched, ``rw`` is the specifier
9293 determining if the fetch should be for a read (0) or write (1), and
9294 ``locality`` is a temporal locality specifier ranging from (0) - no
9295 locality, to (3) - extremely local keep in cache. The ``cache type``
9296 specifies whether the prefetch is performed on the data (1) or
9297 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
9298 arguments must be constant integers.
9303 This intrinsic does not modify the behavior of the program. In
9304 particular, prefetches cannot trap and do not produce a value. On
9305 targets that support this intrinsic, the prefetch can provide hints to
9306 the processor cache for better performance.
9308 '``llvm.pcmarker``' Intrinsic
9309 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9316 declare void @llvm.pcmarker(i32 <id>)
9321 The '``llvm.pcmarker``' intrinsic is a method to export a Program
9322 Counter (PC) in a region of code to simulators and other tools. The
9323 method is target specific, but it is expected that the marker will use
9324 exported symbols to transmit the PC of the marker. The marker makes no
9325 guarantees that it will remain with any specific instruction after
9326 optimizations. It is possible that the presence of a marker will inhibit
9327 optimizations. The intended use is to be inserted after optimizations to
9328 allow correlations of simulation runs.
9333 ``id`` is a numerical id identifying the marker.
9338 This intrinsic does not modify the behavior of the program. Backends
9339 that do not support this intrinsic may ignore it.
9341 '``llvm.readcyclecounter``' Intrinsic
9342 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9349 declare i64 @llvm.readcyclecounter()
9354 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
9355 counter register (or similar low latency, high accuracy clocks) on those
9356 targets that support it. On X86, it should map to RDTSC. On Alpha, it
9357 should map to RPCC. As the backing counters overflow quickly (on the
9358 order of 9 seconds on alpha), this should only be used for small
9364 When directly supported, reading the cycle counter should not modify any
9365 memory. Implementations are allowed to either return a application
9366 specific value or a system wide value. On backends without support, this
9367 is lowered to a constant 0.
9369 Note that runtime support may be conditional on the privilege-level code is
9370 running at and the host platform.
9372 '``llvm.clear_cache``' Intrinsic
9373 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9380 declare void @llvm.clear_cache(i8*, i8*)
9385 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
9386 in the specified range to the execution unit of the processor. On
9387 targets with non-unified instruction and data cache, the implementation
9388 flushes the instruction cache.
9393 On platforms with coherent instruction and data caches (e.g. x86), this
9394 intrinsic is a nop. On platforms with non-coherent instruction and data
9395 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
9396 instructions or a system call, if cache flushing requires special
9399 The default behavior is to emit a call to ``__clear_cache`` from the run
9402 This instrinsic does *not* empty the instruction pipeline. Modifications
9403 of the current function are outside the scope of the intrinsic.
9405 '``llvm.instrprof_increment``' Intrinsic
9406 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9413 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
9414 i32 <num-counters>, i32 <index>)
9419 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
9420 frontend for use with instrumentation based profiling. These will be
9421 lowered by the ``-instrprof`` pass to generate execution counts of a
9427 The first argument is a pointer to a global variable containing the
9428 name of the entity being instrumented. This should generally be the
9429 (mangled) function name for a set of counters.
9431 The second argument is a hash value that can be used by the consumer
9432 of the profile data to detect changes to the instrumented source, and
9433 the third is the number of counters associated with ``name``. It is an
9434 error if ``hash`` or ``num-counters`` differ between two instances of
9435 ``instrprof_increment`` that refer to the same name.
9437 The last argument refers to which of the counters for ``name`` should
9438 be incremented. It should be a value between 0 and ``num-counters``.
9443 This intrinsic represents an increment of a profiling counter. It will
9444 cause the ``-instrprof`` pass to generate the appropriate data
9445 structures and the code to increment the appropriate value, in a
9446 format that can be written out by a compiler runtime and consumed via
9447 the ``llvm-profdata`` tool.
9449 '``llvm.instrprof_value_profile``' Intrinsic
9450 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9457 declare void @llvm.instrprof_value_profile(i8* <name>, i64 <hash>,
9458 i64 <value>, i32 <value_kind>,
9464 The '``llvm.instrprof_value_profile``' intrinsic can be emitted by a
9465 frontend for use with instrumentation based profiling. This will be
9466 lowered by the ``-instrprof`` pass to find out the target values,
9467 instrumented expressions take in a program at runtime.
9472 The first argument is a pointer to a global variable containing the
9473 name of the entity being instrumented. ``name`` should generally be the
9474 (mangled) function name for a set of counters.
9476 The second argument is a hash value that can be used by the consumer
9477 of the profile data to detect changes to the instrumented source. It
9478 is an error if ``hash`` differs between two instances of
9479 ``llvm.instrprof_*`` that refer to the same name.
9481 The third argument is the value of the expression being profiled. The profiled
9482 expression's value should be representable as an unsigned 64-bit value. The
9483 fourth argument represents the kind of value profiling that is being done. The
9484 supported value profiling kinds are enumerated through the
9485 ``InstrProfValueKind`` type declared in the
9486 ``<include/llvm/ProfileData/InstrProf.h>`` header file. The last argument is the
9487 index of the instrumented expression within ``name``. It should be >= 0.
9492 This intrinsic represents the point where a call to a runtime routine
9493 should be inserted for value profiling of target expressions. ``-instrprof``
9494 pass will generate the appropriate data structures and replace the
9495 ``llvm.instrprof_value_profile`` intrinsic with the call to the profile
9496 runtime library with proper arguments.
9498 Standard C Library Intrinsics
9499 -----------------------------
9501 LLVM provides intrinsics for a few important standard C library
9502 functions. These intrinsics allow source-language front-ends to pass
9503 information about the alignment of the pointer arguments to the code
9504 generator, providing opportunity for more efficient code generation.
9508 '``llvm.memcpy``' Intrinsic
9509 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9514 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
9515 integer bit width and for different address spaces. Not all targets
9516 support all bit widths however.
9520 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9521 i32 <len>, i32 <align>, i1 <isvolatile>)
9522 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9523 i64 <len>, i32 <align>, i1 <isvolatile>)
9528 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9529 source location to the destination location.
9531 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
9532 intrinsics do not return a value, takes extra alignment/isvolatile
9533 arguments and the pointers can be in specified address spaces.
9538 The first argument is a pointer to the destination, the second is a
9539 pointer to the source. The third argument is an integer argument
9540 specifying the number of bytes to copy, the fourth argument is the
9541 alignment of the source and destination locations, and the fifth is a
9542 boolean indicating a volatile access.
9544 If the call to this intrinsic has an alignment value that is not 0 or 1,
9545 then the caller guarantees that both the source and destination pointers
9546 are aligned to that boundary.
9548 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
9549 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9550 very cleanly specified and it is unwise to depend on it.
9555 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9556 source location to the destination location, which are not allowed to
9557 overlap. It copies "len" bytes of memory over. If the argument is known
9558 to be aligned to some boundary, this can be specified as the fourth
9559 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
9561 '``llvm.memmove``' Intrinsic
9562 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9567 This is an overloaded intrinsic. You can use llvm.memmove on any integer
9568 bit width and for different address space. Not all targets support all
9573 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9574 i32 <len>, i32 <align>, i1 <isvolatile>)
9575 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9576 i64 <len>, i32 <align>, i1 <isvolatile>)
9581 The '``llvm.memmove.*``' intrinsics move a block of memory from the
9582 source location to the destination location. It is similar to the
9583 '``llvm.memcpy``' intrinsic but allows the two memory locations to
9586 Note that, unlike the standard libc function, the ``llvm.memmove.*``
9587 intrinsics do not return a value, takes extra alignment/isvolatile
9588 arguments and the pointers can be in specified address spaces.
9593 The first argument is a pointer to the destination, the second is a
9594 pointer to the source. The third argument is an integer argument
9595 specifying the number of bytes to copy, the fourth argument is the
9596 alignment of the source and destination locations, and the fifth is a
9597 boolean indicating a volatile access.
9599 If the call to this intrinsic has an alignment value that is not 0 or 1,
9600 then the caller guarantees that the source and destination pointers are
9601 aligned to that boundary.
9603 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
9604 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
9605 not very cleanly specified and it is unwise to depend on it.
9610 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
9611 source location to the destination location, which may overlap. It
9612 copies "len" bytes of memory over. If the argument is known to be
9613 aligned to some boundary, this can be specified as the fourth argument,
9614 otherwise it should be set to 0 or 1 (both meaning no alignment).
9616 '``llvm.memset.*``' Intrinsics
9617 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9622 This is an overloaded intrinsic. You can use llvm.memset on any integer
9623 bit width and for different address spaces. However, not all targets
9624 support all bit widths.
9628 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
9629 i32 <len>, i32 <align>, i1 <isvolatile>)
9630 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
9631 i64 <len>, i32 <align>, i1 <isvolatile>)
9636 The '``llvm.memset.*``' intrinsics fill a block of memory with a
9637 particular byte value.
9639 Note that, unlike the standard libc function, the ``llvm.memset``
9640 intrinsic does not return a value and takes extra alignment/volatile
9641 arguments. Also, the destination can be in an arbitrary address space.
9646 The first argument is a pointer to the destination to fill, the second
9647 is the byte value with which to fill it, the third argument is an
9648 integer argument specifying the number of bytes to fill, and the fourth
9649 argument is the known alignment of the destination location.
9651 If the call to this intrinsic has an alignment value that is not 0 or 1,
9652 then the caller guarantees that the destination pointer is aligned to
9655 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
9656 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9657 very cleanly specified and it is unwise to depend on it.
9662 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
9663 at the destination location. If the argument is known to be aligned to
9664 some boundary, this can be specified as the fourth argument, otherwise
9665 it should be set to 0 or 1 (both meaning no alignment).
9667 '``llvm.sqrt.*``' Intrinsic
9668 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9673 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
9674 floating point or vector of floating point type. Not all targets support
9679 declare float @llvm.sqrt.f32(float %Val)
9680 declare double @llvm.sqrt.f64(double %Val)
9681 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
9682 declare fp128 @llvm.sqrt.f128(fp128 %Val)
9683 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
9688 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
9689 returning the same value as the libm '``sqrt``' functions would. Unlike
9690 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
9691 negative numbers other than -0.0 (which allows for better optimization,
9692 because there is no need to worry about errno being set).
9693 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
9698 The argument and return value are floating point numbers of the same
9704 This function returns the sqrt of the specified operand if it is a
9705 nonnegative floating point number.
9707 '``llvm.powi.*``' Intrinsic
9708 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9713 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
9714 floating point or vector of floating point type. Not all targets support
9719 declare float @llvm.powi.f32(float %Val, i32 %power)
9720 declare double @llvm.powi.f64(double %Val, i32 %power)
9721 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
9722 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
9723 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
9728 The '``llvm.powi.*``' intrinsics return the first operand raised to the
9729 specified (positive or negative) power. The order of evaluation of
9730 multiplications is not defined. When a vector of floating point type is
9731 used, the second argument remains a scalar integer value.
9736 The second argument is an integer power, and the first is a value to
9737 raise to that power.
9742 This function returns the first value raised to the second power with an
9743 unspecified sequence of rounding operations.
9745 '``llvm.sin.*``' Intrinsic
9746 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9751 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
9752 floating point or vector of floating point type. Not all targets support
9757 declare float @llvm.sin.f32(float %Val)
9758 declare double @llvm.sin.f64(double %Val)
9759 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
9760 declare fp128 @llvm.sin.f128(fp128 %Val)
9761 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
9766 The '``llvm.sin.*``' intrinsics return the sine of the operand.
9771 The argument and return value are floating point numbers of the same
9777 This function returns the sine of the specified operand, returning the
9778 same values as the libm ``sin`` functions would, and handles error
9779 conditions in the same way.
9781 '``llvm.cos.*``' Intrinsic
9782 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9787 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
9788 floating point or vector of floating point type. Not all targets support
9793 declare float @llvm.cos.f32(float %Val)
9794 declare double @llvm.cos.f64(double %Val)
9795 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
9796 declare fp128 @llvm.cos.f128(fp128 %Val)
9797 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
9802 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
9807 The argument and return value are floating point numbers of the same
9813 This function returns the cosine of the specified operand, returning the
9814 same values as the libm ``cos`` functions would, and handles error
9815 conditions in the same way.
9817 '``llvm.pow.*``' Intrinsic
9818 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9823 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
9824 floating point or vector of floating point type. Not all targets support
9829 declare float @llvm.pow.f32(float %Val, float %Power)
9830 declare double @llvm.pow.f64(double %Val, double %Power)
9831 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
9832 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
9833 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
9838 The '``llvm.pow.*``' intrinsics return the first operand raised to the
9839 specified (positive or negative) power.
9844 The second argument is a floating point power, and the first is a value
9845 to raise to that power.
9850 This function returns the first value raised to the second power,
9851 returning the same values as the libm ``pow`` functions would, and
9852 handles error conditions in the same way.
9854 '``llvm.exp.*``' Intrinsic
9855 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9860 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
9861 floating point or vector of floating point type. Not all targets support
9866 declare float @llvm.exp.f32(float %Val)
9867 declare double @llvm.exp.f64(double %Val)
9868 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
9869 declare fp128 @llvm.exp.f128(fp128 %Val)
9870 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
9875 The '``llvm.exp.*``' intrinsics perform the exp function.
9880 The argument and return value are floating point numbers of the same
9886 This function returns the same values as the libm ``exp`` functions
9887 would, and handles error conditions in the same way.
9889 '``llvm.exp2.*``' Intrinsic
9890 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9895 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
9896 floating point or vector of floating point type. Not all targets support
9901 declare float @llvm.exp2.f32(float %Val)
9902 declare double @llvm.exp2.f64(double %Val)
9903 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
9904 declare fp128 @llvm.exp2.f128(fp128 %Val)
9905 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
9910 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
9915 The argument and return value are floating point numbers of the same
9921 This function returns the same values as the libm ``exp2`` functions
9922 would, and handles error conditions in the same way.
9924 '``llvm.log.*``' Intrinsic
9925 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9930 This is an overloaded intrinsic. You can use ``llvm.log`` on any
9931 floating point or vector of floating point type. Not all targets support
9936 declare float @llvm.log.f32(float %Val)
9937 declare double @llvm.log.f64(double %Val)
9938 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
9939 declare fp128 @llvm.log.f128(fp128 %Val)
9940 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
9945 The '``llvm.log.*``' intrinsics perform the log function.
9950 The argument and return value are floating point numbers of the same
9956 This function returns the same values as the libm ``log`` functions
9957 would, and handles error conditions in the same way.
9959 '``llvm.log10.*``' Intrinsic
9960 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9965 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
9966 floating point or vector of floating point type. Not all targets support
9971 declare float @llvm.log10.f32(float %Val)
9972 declare double @llvm.log10.f64(double %Val)
9973 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
9974 declare fp128 @llvm.log10.f128(fp128 %Val)
9975 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
9980 The '``llvm.log10.*``' intrinsics perform the log10 function.
9985 The argument and return value are floating point numbers of the same
9991 This function returns the same values as the libm ``log10`` functions
9992 would, and handles error conditions in the same way.
9994 '``llvm.log2.*``' Intrinsic
9995 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10000 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
10001 floating point or vector of floating point type. Not all targets support
10006 declare float @llvm.log2.f32(float %Val)
10007 declare double @llvm.log2.f64(double %Val)
10008 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
10009 declare fp128 @llvm.log2.f128(fp128 %Val)
10010 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
10015 The '``llvm.log2.*``' intrinsics perform the log2 function.
10020 The argument and return value are floating point numbers of the same
10026 This function returns the same values as the libm ``log2`` functions
10027 would, and handles error conditions in the same way.
10029 '``llvm.fma.*``' Intrinsic
10030 ^^^^^^^^^^^^^^^^^^^^^^^^^^
10035 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
10036 floating point or vector of floating point type. Not all targets support
10041 declare float @llvm.fma.f32(float %a, float %b, float %c)
10042 declare double @llvm.fma.f64(double %a, double %b, double %c)
10043 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
10044 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
10045 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
10050 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
10056 The argument and return value are floating point numbers of the same
10062 This function returns the same values as the libm ``fma`` functions
10063 would, and does not set errno.
10065 '``llvm.fabs.*``' Intrinsic
10066 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10071 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
10072 floating point or vector of floating point type. Not all targets support
10077 declare float @llvm.fabs.f32(float %Val)
10078 declare double @llvm.fabs.f64(double %Val)
10079 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
10080 declare fp128 @llvm.fabs.f128(fp128 %Val)
10081 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
10086 The '``llvm.fabs.*``' intrinsics return the absolute value of the
10092 The argument and return value are floating point numbers of the same
10098 This function returns the same values as the libm ``fabs`` functions
10099 would, and handles error conditions in the same way.
10101 '``llvm.minnum.*``' Intrinsic
10102 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10107 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
10108 floating point or vector of floating point type. Not all targets support
10113 declare float @llvm.minnum.f32(float %Val0, float %Val1)
10114 declare double @llvm.minnum.f64(double %Val0, double %Val1)
10115 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10116 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
10117 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10122 The '``llvm.minnum.*``' intrinsics return the minimum of the two
10129 The arguments and return value are floating point numbers of the same
10135 Follows the IEEE-754 semantics for minNum, which also match for libm's
10138 If either operand is a NaN, returns the other non-NaN operand. Returns
10139 NaN only if both operands are NaN. If the operands compare equal,
10140 returns a value that compares equal to both operands. This means that
10141 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10143 '``llvm.maxnum.*``' Intrinsic
10144 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10149 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
10150 floating point or vector of floating point type. Not all targets support
10155 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
10156 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
10157 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10158 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
10159 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10164 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
10171 The arguments and return value are floating point numbers of the same
10176 Follows the IEEE-754 semantics for maxNum, which also match for libm's
10179 If either operand is a NaN, returns the other non-NaN operand. Returns
10180 NaN only if both operands are NaN. If the operands compare equal,
10181 returns a value that compares equal to both operands. This means that
10182 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10184 '``llvm.copysign.*``' Intrinsic
10185 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10190 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
10191 floating point or vector of floating point type. Not all targets support
10196 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
10197 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
10198 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
10199 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
10200 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
10205 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
10206 first operand and the sign of the second operand.
10211 The arguments and return value are floating point numbers of the same
10217 This function returns the same values as the libm ``copysign``
10218 functions would, and handles error conditions in the same way.
10220 '``llvm.floor.*``' Intrinsic
10221 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10226 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
10227 floating point or vector of floating point type. Not all targets support
10232 declare float @llvm.floor.f32(float %Val)
10233 declare double @llvm.floor.f64(double %Val)
10234 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
10235 declare fp128 @llvm.floor.f128(fp128 %Val)
10236 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
10241 The '``llvm.floor.*``' intrinsics return the floor of the operand.
10246 The argument and return value are floating point numbers of the same
10252 This function returns the same values as the libm ``floor`` functions
10253 would, and handles error conditions in the same way.
10255 '``llvm.ceil.*``' Intrinsic
10256 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10261 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
10262 floating point or vector of floating point type. Not all targets support
10267 declare float @llvm.ceil.f32(float %Val)
10268 declare double @llvm.ceil.f64(double %Val)
10269 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
10270 declare fp128 @llvm.ceil.f128(fp128 %Val)
10271 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
10276 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
10281 The argument and return value are floating point numbers of the same
10287 This function returns the same values as the libm ``ceil`` functions
10288 would, and handles error conditions in the same way.
10290 '``llvm.trunc.*``' Intrinsic
10291 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10296 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
10297 floating point or vector of floating point type. Not all targets support
10302 declare float @llvm.trunc.f32(float %Val)
10303 declare double @llvm.trunc.f64(double %Val)
10304 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
10305 declare fp128 @llvm.trunc.f128(fp128 %Val)
10306 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
10311 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
10312 nearest integer not larger in magnitude than the operand.
10317 The argument and return value are floating point numbers of the same
10323 This function returns the same values as the libm ``trunc`` functions
10324 would, and handles error conditions in the same way.
10326 '``llvm.rint.*``' Intrinsic
10327 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10332 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
10333 floating point or vector of floating point type. Not all targets support
10338 declare float @llvm.rint.f32(float %Val)
10339 declare double @llvm.rint.f64(double %Val)
10340 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
10341 declare fp128 @llvm.rint.f128(fp128 %Val)
10342 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
10347 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
10348 nearest integer. It may raise an inexact floating-point exception if the
10349 operand isn't an integer.
10354 The argument and return value are floating point numbers of the same
10360 This function returns the same values as the libm ``rint`` functions
10361 would, and handles error conditions in the same way.
10363 '``llvm.nearbyint.*``' Intrinsic
10364 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10369 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
10370 floating point or vector of floating point type. Not all targets support
10375 declare float @llvm.nearbyint.f32(float %Val)
10376 declare double @llvm.nearbyint.f64(double %Val)
10377 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
10378 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
10379 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
10384 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
10390 The argument and return value are floating point numbers of the same
10396 This function returns the same values as the libm ``nearbyint``
10397 functions would, and handles error conditions in the same way.
10399 '``llvm.round.*``' Intrinsic
10400 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10405 This is an overloaded intrinsic. You can use ``llvm.round`` on any
10406 floating point or vector of floating point type. Not all targets support
10411 declare float @llvm.round.f32(float %Val)
10412 declare double @llvm.round.f64(double %Val)
10413 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
10414 declare fp128 @llvm.round.f128(fp128 %Val)
10415 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
10420 The '``llvm.round.*``' intrinsics returns the operand rounded to the
10426 The argument and return value are floating point numbers of the same
10432 This function returns the same values as the libm ``round``
10433 functions would, and handles error conditions in the same way.
10435 Bit Manipulation Intrinsics
10436 ---------------------------
10438 LLVM provides intrinsics for a few important bit manipulation
10439 operations. These allow efficient code generation for some algorithms.
10441 '``llvm.bitreverse.*``' Intrinsics
10442 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10447 This is an overloaded intrinsic function. You can use bitreverse on any
10452 declare i16 @llvm.bitreverse.i16(i16 <id>)
10453 declare i32 @llvm.bitreverse.i32(i32 <id>)
10454 declare i64 @llvm.bitreverse.i64(i64 <id>)
10459 The '``llvm.bitreverse``' family of intrinsics is used to reverse the
10460 bitpattern of an integer value; for example ``0b1234567`` becomes
10466 The ``llvm.bitreverse.iN`` intrinsic returns an i16 value that has bit
10467 ``M`` in the input moved to bit ``N-M`` in the output.
10469 '``llvm.bswap.*``' Intrinsics
10470 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10475 This is an overloaded intrinsic function. You can use bswap on any
10476 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
10480 declare i16 @llvm.bswap.i16(i16 <id>)
10481 declare i32 @llvm.bswap.i32(i32 <id>)
10482 declare i64 @llvm.bswap.i64(i64 <id>)
10487 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
10488 values with an even number of bytes (positive multiple of 16 bits).
10489 These are useful for performing operations on data that is not in the
10490 target's native byte order.
10495 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
10496 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
10497 intrinsic returns an i32 value that has the four bytes of the input i32
10498 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
10499 returned i32 will have its bytes in 3, 2, 1, 0 order. The
10500 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
10501 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
10504 '``llvm.ctpop.*``' Intrinsic
10505 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10510 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
10511 bit width, or on any vector with integer elements. Not all targets
10512 support all bit widths or vector types, however.
10516 declare i8 @llvm.ctpop.i8(i8 <src>)
10517 declare i16 @llvm.ctpop.i16(i16 <src>)
10518 declare i32 @llvm.ctpop.i32(i32 <src>)
10519 declare i64 @llvm.ctpop.i64(i64 <src>)
10520 declare i256 @llvm.ctpop.i256(i256 <src>)
10521 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
10526 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
10532 The only argument is the value to be counted. The argument may be of any
10533 integer type, or a vector with integer elements. The return type must
10534 match the argument type.
10539 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
10540 each element of a vector.
10542 '``llvm.ctlz.*``' Intrinsic
10543 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10548 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
10549 integer bit width, or any vector whose elements are integers. Not all
10550 targets support all bit widths or vector types, however.
10554 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
10555 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
10556 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
10557 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
10558 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
10559 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10564 The '``llvm.ctlz``' family of intrinsic functions counts the number of
10565 leading zeros in a variable.
10570 The first argument is the value to be counted. This argument may be of
10571 any integer type, or a vector with integer element type. The return
10572 type must match the first argument type.
10574 The second argument must be a constant and is a flag to indicate whether
10575 the intrinsic should ensure that a zero as the first argument produces a
10576 defined result. Historically some architectures did not provide a
10577 defined result for zero values as efficiently, and many algorithms are
10578 now predicated on avoiding zero-value inputs.
10583 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
10584 zeros in a variable, or within each element of the vector. If
10585 ``src == 0`` then the result is the size in bits of the type of ``src``
10586 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10587 ``llvm.ctlz(i32 2) = 30``.
10589 '``llvm.cttz.*``' Intrinsic
10590 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10595 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
10596 integer bit width, or any vector of integer elements. Not all targets
10597 support all bit widths or vector types, however.
10601 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
10602 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
10603 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
10604 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
10605 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
10606 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10611 The '``llvm.cttz``' family of intrinsic functions counts the number of
10617 The first argument is the value to be counted. This argument may be of
10618 any integer type, or a vector with integer element type. The return
10619 type must match the first argument type.
10621 The second argument must be a constant and is a flag to indicate whether
10622 the intrinsic should ensure that a zero as the first argument produces a
10623 defined result. Historically some architectures did not provide a
10624 defined result for zero values as efficiently, and many algorithms are
10625 now predicated on avoiding zero-value inputs.
10630 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
10631 zeros in a variable, or within each element of a vector. If ``src == 0``
10632 then the result is the size in bits of the type of ``src`` if
10633 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10634 ``llvm.cttz(2) = 1``.
10638 Arithmetic with Overflow Intrinsics
10639 -----------------------------------
10641 LLVM provides intrinsics for some arithmetic with overflow operations.
10643 '``llvm.sadd.with.overflow.*``' Intrinsics
10644 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10649 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
10650 on any integer bit width.
10654 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
10655 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10656 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
10661 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10662 a signed addition of the two arguments, and indicate whether an overflow
10663 occurred during the signed summation.
10668 The arguments (%a and %b) and the first element of the result structure
10669 may be of integer types of any bit width, but they must have the same
10670 bit width. The second element of the result structure must be of type
10671 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10677 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10678 a signed addition of the two variables. They return a structure --- the
10679 first element of which is the signed summation, and the second element
10680 of which is a bit specifying if the signed summation resulted in an
10686 .. code-block:: llvm
10688 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10689 %sum = extractvalue {i32, i1} %res, 0
10690 %obit = extractvalue {i32, i1} %res, 1
10691 br i1 %obit, label %overflow, label %normal
10693 '``llvm.uadd.with.overflow.*``' Intrinsics
10694 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10699 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
10700 on any integer bit width.
10704 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
10705 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10706 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
10711 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10712 an unsigned addition of the two arguments, and indicate whether a carry
10713 occurred during the unsigned summation.
10718 The arguments (%a and %b) and the first element of the result structure
10719 may be of integer types of any bit width, but they must have the same
10720 bit width. The second element of the result structure must be of type
10721 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10727 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10728 an unsigned addition of the two arguments. They return a structure --- the
10729 first element of which is the sum, and the second element of which is a
10730 bit specifying if the unsigned summation resulted in a carry.
10735 .. code-block:: llvm
10737 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10738 %sum = extractvalue {i32, i1} %res, 0
10739 %obit = extractvalue {i32, i1} %res, 1
10740 br i1 %obit, label %carry, label %normal
10742 '``llvm.ssub.with.overflow.*``' Intrinsics
10743 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10748 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
10749 on any integer bit width.
10753 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
10754 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10755 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
10760 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10761 a signed subtraction of the two arguments, and indicate whether an
10762 overflow occurred during the signed subtraction.
10767 The arguments (%a and %b) and the first element of the result structure
10768 may be of integer types of any bit width, but they must have the same
10769 bit width. The second element of the result structure must be of type
10770 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10776 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10777 a signed subtraction of the two arguments. They return a structure --- the
10778 first element of which is the subtraction, and the second element of
10779 which is a bit specifying if the signed subtraction resulted in an
10785 .. code-block:: llvm
10787 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10788 %sum = extractvalue {i32, i1} %res, 0
10789 %obit = extractvalue {i32, i1} %res, 1
10790 br i1 %obit, label %overflow, label %normal
10792 '``llvm.usub.with.overflow.*``' Intrinsics
10793 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10798 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
10799 on any integer bit width.
10803 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
10804 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10805 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
10810 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10811 an unsigned subtraction of the two arguments, and indicate whether an
10812 overflow occurred during the unsigned subtraction.
10817 The arguments (%a and %b) and the first element of the result structure
10818 may be of integer types of any bit width, but they must have the same
10819 bit width. The second element of the result structure must be of type
10820 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10826 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10827 an unsigned subtraction of the two arguments. They return a structure ---
10828 the first element of which is the subtraction, and the second element of
10829 which is a bit specifying if the unsigned subtraction resulted in an
10835 .. code-block:: llvm
10837 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10838 %sum = extractvalue {i32, i1} %res, 0
10839 %obit = extractvalue {i32, i1} %res, 1
10840 br i1 %obit, label %overflow, label %normal
10842 '``llvm.smul.with.overflow.*``' Intrinsics
10843 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10848 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
10849 on any integer bit width.
10853 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
10854 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10855 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
10860 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10861 a signed multiplication of the two arguments, and indicate whether an
10862 overflow occurred during the signed multiplication.
10867 The arguments (%a and %b) and the first element of the result structure
10868 may be of integer types of any bit width, but they must have the same
10869 bit width. The second element of the result structure must be of type
10870 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10876 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10877 a signed multiplication of the two arguments. They return a structure ---
10878 the first element of which is the multiplication, and the second element
10879 of which is a bit specifying if the signed multiplication resulted in an
10885 .. code-block:: llvm
10887 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10888 %sum = extractvalue {i32, i1} %res, 0
10889 %obit = extractvalue {i32, i1} %res, 1
10890 br i1 %obit, label %overflow, label %normal
10892 '``llvm.umul.with.overflow.*``' Intrinsics
10893 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10898 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
10899 on any integer bit width.
10903 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
10904 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10905 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
10910 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10911 a unsigned multiplication of the two arguments, and indicate whether an
10912 overflow occurred during the unsigned multiplication.
10917 The arguments (%a and %b) and the first element of the result structure
10918 may be of integer types of any bit width, but they must have the same
10919 bit width. The second element of the result structure must be of type
10920 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10926 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10927 an unsigned multiplication of the two arguments. They return a structure ---
10928 the first element of which is the multiplication, and the second
10929 element of which is a bit specifying if the unsigned multiplication
10930 resulted in an overflow.
10935 .. code-block:: llvm
10937 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10938 %sum = extractvalue {i32, i1} %res, 0
10939 %obit = extractvalue {i32, i1} %res, 1
10940 br i1 %obit, label %overflow, label %normal
10942 Specialised Arithmetic Intrinsics
10943 ---------------------------------
10945 '``llvm.canonicalize.*``' Intrinsic
10946 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10953 declare float @llvm.canonicalize.f32(float %a)
10954 declare double @llvm.canonicalize.f64(double %b)
10959 The '``llvm.canonicalize.*``' intrinsic returns the platform specific canonical
10960 encoding of a floating point number. This canonicalization is useful for
10961 implementing certain numeric primitives such as frexp. The canonical encoding is
10962 defined by IEEE-754-2008 to be:
10966 2.1.8 canonical encoding: The preferred encoding of a floating-point
10967 representation in a format. Applied to declets, significands of finite
10968 numbers, infinities, and NaNs, especially in decimal formats.
10970 This operation can also be considered equivalent to the IEEE-754-2008
10971 conversion of a floating-point value to the same format. NaNs are handled
10972 according to section 6.2.
10974 Examples of non-canonical encodings:
10976 - x87 pseudo denormals, pseudo NaNs, pseudo Infinity, Unnormals. These are
10977 converted to a canonical representation per hardware-specific protocol.
10978 - Many normal decimal floating point numbers have non-canonical alternative
10980 - Some machines, like GPUs or ARMv7 NEON, do not support subnormal values.
10981 These are treated as non-canonical encodings of zero and with be flushed to
10982 a zero of the same sign by this operation.
10984 Note that per IEEE-754-2008 6.2, systems that support signaling NaNs with
10985 default exception handling must signal an invalid exception, and produce a
10988 This function should always be implementable as multiplication by 1.0, provided
10989 that the compiler does not constant fold the operation. Likewise, division by
10990 1.0 and ``llvm.minnum(x, x)`` are possible implementations. Addition with
10991 -0.0 is also sufficient provided that the rounding mode is not -Infinity.
10993 ``@llvm.canonicalize`` must preserve the equality relation. That is:
10995 - ``(@llvm.canonicalize(x) == x)`` is equivalent to ``(x == x)``
10996 - ``(@llvm.canonicalize(x) == @llvm.canonicalize(y))`` is equivalent to
10999 Additionally, the sign of zero must be conserved:
11000 ``@llvm.canonicalize(-0.0) = -0.0`` and ``@llvm.canonicalize(+0.0) = +0.0``
11002 The payload bits of a NaN must be conserved, with two exceptions.
11003 First, environments which use only a single canonical representation of NaN
11004 must perform said canonicalization. Second, SNaNs must be quieted per the
11007 The canonicalization operation may be optimized away if:
11009 - The input is known to be canonical. For example, it was produced by a
11010 floating-point operation that is required by the standard to be canonical.
11011 - The result is consumed only by (or fused with) other floating-point
11012 operations. That is, the bits of the floating point value are not examined.
11014 '``llvm.fmuladd.*``' Intrinsic
11015 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11022 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
11023 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
11028 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
11029 expressions that can be fused if the code generator determines that (a) the
11030 target instruction set has support for a fused operation, and (b) that the
11031 fused operation is more efficient than the equivalent, separate pair of mul
11032 and add instructions.
11037 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
11038 multiplicands, a and b, and an addend c.
11047 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
11049 is equivalent to the expression a \* b + c, except that rounding will
11050 not be performed between the multiplication and addition steps if the
11051 code generator fuses the operations. Fusion is not guaranteed, even if
11052 the target platform supports it. If a fused multiply-add is required the
11053 corresponding llvm.fma.\* intrinsic function should be used
11054 instead. This never sets errno, just as '``llvm.fma.*``'.
11059 .. code-block:: llvm
11061 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
11063 Half Precision Floating Point Intrinsics
11064 ----------------------------------------
11066 For most target platforms, half precision floating point is a
11067 storage-only format. This means that it is a dense encoding (in memory)
11068 but does not support computation in the format.
11070 This means that code must first load the half-precision floating point
11071 value as an i16, then convert it to float with
11072 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
11073 then be performed on the float value (including extending to double
11074 etc). To store the value back to memory, it is first converted to float
11075 if needed, then converted to i16 with
11076 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
11079 .. _int_convert_to_fp16:
11081 '``llvm.convert.to.fp16``' Intrinsic
11082 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11089 declare i16 @llvm.convert.to.fp16.f32(float %a)
11090 declare i16 @llvm.convert.to.fp16.f64(double %a)
11095 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11096 conventional floating point type to half precision floating point format.
11101 The intrinsic function contains single argument - the value to be
11107 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11108 conventional floating point format to half precision floating point format. The
11109 return value is an ``i16`` which contains the converted number.
11114 .. code-block:: llvm
11116 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
11117 store i16 %res, i16* @x, align 2
11119 .. _int_convert_from_fp16:
11121 '``llvm.convert.from.fp16``' Intrinsic
11122 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11129 declare float @llvm.convert.from.fp16.f32(i16 %a)
11130 declare double @llvm.convert.from.fp16.f64(i16 %a)
11135 The '``llvm.convert.from.fp16``' intrinsic function performs a
11136 conversion from half precision floating point format to single precision
11137 floating point format.
11142 The intrinsic function contains single argument - the value to be
11148 The '``llvm.convert.from.fp16``' intrinsic function performs a
11149 conversion from half single precision floating point format to single
11150 precision floating point format. The input half-float value is
11151 represented by an ``i16`` value.
11156 .. code-block:: llvm
11158 %a = load i16, i16* @x, align 2
11159 %res = call float @llvm.convert.from.fp16(i16 %a)
11161 .. _dbg_intrinsics:
11163 Debugger Intrinsics
11164 -------------------
11166 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
11167 prefix), are described in the `LLVM Source Level
11168 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
11171 Exception Handling Intrinsics
11172 -----------------------------
11174 The LLVM exception handling intrinsics (which all start with
11175 ``llvm.eh.`` prefix), are described in the `LLVM Exception
11176 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
11178 .. _int_trampoline:
11180 Trampoline Intrinsics
11181 ---------------------
11183 These intrinsics make it possible to excise one parameter, marked with
11184 the :ref:`nest <nest>` attribute, from a function. The result is a
11185 callable function pointer lacking the nest parameter - the caller does
11186 not need to provide a value for it. Instead, the value to use is stored
11187 in advance in a "trampoline", a block of memory usually allocated on the
11188 stack, which also contains code to splice the nest value into the
11189 argument list. This is used to implement the GCC nested function address
11192 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
11193 then the resulting function pointer has signature ``i32 (i32, i32)*``.
11194 It can be created as follows:
11196 .. code-block:: llvm
11198 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
11199 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
11200 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
11201 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
11202 %fp = bitcast i8* %p to i32 (i32, i32)*
11204 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
11205 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
11209 '``llvm.init.trampoline``' Intrinsic
11210 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11217 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
11222 This fills the memory pointed to by ``tramp`` with executable code,
11223 turning it into a trampoline.
11228 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
11229 pointers. The ``tramp`` argument must point to a sufficiently large and
11230 sufficiently aligned block of memory; this memory is written to by the
11231 intrinsic. Note that the size and the alignment are target-specific -
11232 LLVM currently provides no portable way of determining them, so a
11233 front-end that generates this intrinsic needs to have some
11234 target-specific knowledge. The ``func`` argument must hold a function
11235 bitcast to an ``i8*``.
11240 The block of memory pointed to by ``tramp`` is filled with target
11241 dependent code, turning it into a function. Then ``tramp`` needs to be
11242 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
11243 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
11244 function's signature is the same as that of ``func`` with any arguments
11245 marked with the ``nest`` attribute removed. At most one such ``nest``
11246 argument is allowed, and it must be of pointer type. Calling the new
11247 function is equivalent to calling ``func`` with the same argument list,
11248 but with ``nval`` used for the missing ``nest`` argument. If, after
11249 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
11250 modified, then the effect of any later call to the returned function
11251 pointer is undefined.
11255 '``llvm.adjust.trampoline``' Intrinsic
11256 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11263 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
11268 This performs any required machine-specific adjustment to the address of
11269 a trampoline (passed as ``tramp``).
11274 ``tramp`` must point to a block of memory which already has trampoline
11275 code filled in by a previous call to
11276 :ref:`llvm.init.trampoline <int_it>`.
11281 On some architectures the address of the code to be executed needs to be
11282 different than the address where the trampoline is actually stored. This
11283 intrinsic returns the executable address corresponding to ``tramp``
11284 after performing the required machine specific adjustments. The pointer
11285 returned can then be :ref:`bitcast and executed <int_trampoline>`.
11287 .. _int_mload_mstore:
11289 Masked Vector Load and Store Intrinsics
11290 ---------------------------------------
11292 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.
11296 '``llvm.masked.load.*``' Intrinsics
11297 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11301 This is an overloaded intrinsic. The loaded data is a vector of any integer, floating point or pointer data type.
11305 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11306 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11307 ;; The data is a vector of pointers to double
11308 declare <8 x double*> @llvm.masked.load.v8p0f64 (<8 x double*>* <ptr>, i32 <alignment>, <8 x i1> <mask>, <8 x double*> <passthru>)
11309 ;; The data is a vector of function pointers
11310 declare <8 x i32 ()*> @llvm.masked.load.v8p0f_i32f (<8 x i32 ()*>* <ptr>, i32 <alignment>, <8 x i1> <mask>, <8 x i32 ()*> <passthru>)
11315 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.
11321 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.
11327 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.
11328 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.
11333 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
11335 ;; The result of the two following instructions is identical aside from potential memory access exception
11336 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
11337 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
11341 '``llvm.masked.store.*``' Intrinsics
11342 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11346 This is an overloaded intrinsic. The data stored in memory is a vector of any integer, floating point or pointer data type.
11350 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32>* <ptr>, i32 <alignment>, <8 x i1> <mask>)
11351 declare void @llvm.masked.store.v16f32 (<16 x float> <value>, <16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
11352 ;; The data is a vector of pointers to double
11353 declare void @llvm.masked.store.v8p0f64 (<8 x double*> <value>, <8 x double*>* <ptr>, i32 <alignment>, <8 x i1> <mask>)
11354 ;; The data is a vector of function pointers
11355 declare void @llvm.masked.store.v4p0f_i32f (<4 x i32 ()*> <value>, <4 x i32 ()*>* <ptr>, i32 <alignment>, <4 x i1> <mask>)
11360 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.
11365 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.
11371 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.
11372 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.
11376 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
11378 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
11379 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
11380 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
11381 store <16 x float> %res, <16 x float>* %ptr, align 4
11384 Masked Vector Gather and Scatter Intrinsics
11385 -------------------------------------------
11387 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.
11391 '``llvm.masked.gather.*``' Intrinsics
11392 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11396 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.
11400 declare <16 x float> @llvm.masked.gather.v16f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11401 declare <2 x double> @llvm.masked.gather.v2f64 (<2 x double*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11402 declare <8 x float*> @llvm.masked.gather.v8p0f32 (<8 x float**> <ptrs>, i32 <alignment>, <8 x i1> <mask>, <8 x float*> <passthru>)
11407 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.
11413 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.
11419 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.
11420 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.
11425 %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>)
11427 ;; The gather with all-true mask is equivalent to the following instruction sequence
11428 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
11429 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
11430 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
11431 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
11433 %val0 = load double, double* %ptr0, align 8
11434 %val1 = load double, double* %ptr1, align 8
11435 %val2 = load double, double* %ptr2, align 8
11436 %val3 = load double, double* %ptr3, align 8
11438 %vec0 = insertelement <4 x double>undef, %val0, 0
11439 %vec01 = insertelement <4 x double>%vec0, %val1, 1
11440 %vec012 = insertelement <4 x double>%vec01, %val2, 2
11441 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
11445 '``llvm.masked.scatter.*``' Intrinsics
11446 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11450 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.
11454 declare void @llvm.masked.scatter.v8i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
11455 declare void @llvm.masked.scatter.v16f32 (<16 x float> <value>, <16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
11456 declare void @llvm.masked.scatter.v4p0f64 (<4 x double*> <value>, <4 x double**> <ptrs>, i32 <alignment>, <4 x i1> <mask>)
11461 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.
11466 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.
11472 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.
11476 ;; This instruction unconditionaly stores data vector in multiple addresses
11477 call @llvm.masked.scatter.v8i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
11479 ;; It is equivalent to a list of scalar stores
11480 %val0 = extractelement <8 x i32> %value, i32 0
11481 %val1 = extractelement <8 x i32> %value, i32 1
11483 %val7 = extractelement <8 x i32> %value, i32 7
11484 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
11485 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
11487 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
11488 ;; Note: the order of the following stores is important when they overlap:
11489 store i32 %val0, i32* %ptr0, align 4
11490 store i32 %val1, i32* %ptr1, align 4
11492 store i32 %val7, i32* %ptr7, align 4
11498 This class of intrinsics provides information about the lifetime of
11499 memory objects and ranges where variables are immutable.
11503 '``llvm.lifetime.start``' Intrinsic
11504 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11511 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
11516 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
11522 The first argument is a constant integer representing the size of the
11523 object, or -1 if it is variable sized. The second argument is a pointer
11529 This intrinsic indicates that before this point in the code, the value
11530 of the memory pointed to by ``ptr`` is dead. This means that it is known
11531 to never be used and has an undefined value. A load from the pointer
11532 that precedes this intrinsic can be replaced with ``'undef'``.
11536 '``llvm.lifetime.end``' Intrinsic
11537 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11544 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
11549 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
11555 The first argument is a constant integer representing the size of the
11556 object, or -1 if it is variable sized. The second argument is a pointer
11562 This intrinsic indicates that after this point in the code, the value of
11563 the memory pointed to by ``ptr`` is dead. This means that it is known to
11564 never be used and has an undefined value. Any stores into the memory
11565 object following this intrinsic may be removed as dead.
11567 '``llvm.invariant.start``' Intrinsic
11568 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11575 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
11580 The '``llvm.invariant.start``' intrinsic specifies that the contents of
11581 a memory object will not change.
11586 The first argument is a constant integer representing the size of the
11587 object, or -1 if it is variable sized. The second argument is a pointer
11593 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
11594 the return value, the referenced memory location is constant and
11597 '``llvm.invariant.end``' Intrinsic
11598 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11605 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
11610 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
11611 memory object are mutable.
11616 The first argument is the matching ``llvm.invariant.start`` intrinsic.
11617 The second argument is a constant integer representing the size of the
11618 object, or -1 if it is variable sized and the third argument is a
11619 pointer to the object.
11624 This intrinsic indicates that the memory is mutable again.
11626 '``llvm.invariant.group.barrier``' Intrinsic
11627 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11634 declare i8* @llvm.invariant.group.barrier(i8* <ptr>)
11639 The '``llvm.invariant.group.barrier``' intrinsic can be used when an invariant
11640 established by invariant.group metadata no longer holds, to obtain a new pointer
11641 value that does not carry the invariant information.
11647 The ``llvm.invariant.group.barrier`` takes only one argument, which is
11648 the pointer to the memory for which the ``invariant.group`` no longer holds.
11653 Returns another pointer that aliases its argument but which is considered different
11654 for the purposes of ``load``/``store`` ``invariant.group`` metadata.
11659 This class of intrinsics is designed to be generic and has no specific
11662 '``llvm.var.annotation``' Intrinsic
11663 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11670 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11675 The '``llvm.var.annotation``' intrinsic.
11680 The first argument is a pointer to a value, the second is a pointer to a
11681 global string, the third is a pointer to a global string which is the
11682 source file name, and the last argument is the line number.
11687 This intrinsic allows annotation of local variables with arbitrary
11688 strings. This can be useful for special purpose optimizations that want
11689 to look for these annotations. These have no other defined use; they are
11690 ignored by code generation and optimization.
11692 '``llvm.ptr.annotation.*``' Intrinsic
11693 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11698 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
11699 pointer to an integer of any width. *NOTE* you must specify an address space for
11700 the pointer. The identifier for the default address space is the integer
11705 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11706 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
11707 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
11708 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
11709 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
11714 The '``llvm.ptr.annotation``' intrinsic.
11719 The first argument is a pointer to an integer value of arbitrary bitwidth
11720 (result of some expression), the second is a pointer to a global string, the
11721 third is a pointer to a global string which is the source file name, and the
11722 last argument is the line number. It returns the value of the first argument.
11727 This intrinsic allows annotation of a pointer to an integer with arbitrary
11728 strings. This can be useful for special purpose optimizations that want to look
11729 for these annotations. These have no other defined use; they are ignored by code
11730 generation and optimization.
11732 '``llvm.annotation.*``' Intrinsic
11733 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11738 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
11739 any integer bit width.
11743 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
11744 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
11745 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
11746 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
11747 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
11752 The '``llvm.annotation``' intrinsic.
11757 The first argument is an integer value (result of some expression), the
11758 second is a pointer to a global string, the third is a pointer to a
11759 global string which is the source file name, and the last argument is
11760 the line number. It returns the value of the first argument.
11765 This intrinsic allows annotations to be put on arbitrary expressions
11766 with arbitrary strings. This can be useful for special purpose
11767 optimizations that want to look for these annotations. These have no
11768 other defined use; they are ignored by code generation and optimization.
11770 '``llvm.trap``' Intrinsic
11771 ^^^^^^^^^^^^^^^^^^^^^^^^^
11778 declare void @llvm.trap() noreturn nounwind
11783 The '``llvm.trap``' intrinsic.
11793 This intrinsic is lowered to the target dependent trap instruction. If
11794 the target does not have a trap instruction, this intrinsic will be
11795 lowered to a call of the ``abort()`` function.
11797 '``llvm.debugtrap``' Intrinsic
11798 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11805 declare void @llvm.debugtrap() nounwind
11810 The '``llvm.debugtrap``' intrinsic.
11820 This intrinsic is lowered to code which is intended to cause an
11821 execution trap with the intention of requesting the attention of a
11824 '``llvm.stackprotector``' Intrinsic
11825 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11832 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
11837 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
11838 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
11839 is placed on the stack before local variables.
11844 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
11845 The first argument is the value loaded from the stack guard
11846 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
11847 enough space to hold the value of the guard.
11852 This intrinsic causes the prologue/epilogue inserter to force the position of
11853 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
11854 to ensure that if a local variable on the stack is overwritten, it will destroy
11855 the value of the guard. When the function exits, the guard on the stack is
11856 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
11857 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
11858 calling the ``__stack_chk_fail()`` function.
11860 '``llvm.stackprotectorcheck``' Intrinsic
11861 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11868 declare void @llvm.stackprotectorcheck(i8** <guard>)
11873 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
11874 created stack protector and if they are not equal calls the
11875 ``__stack_chk_fail()`` function.
11880 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
11881 the variable ``@__stack_chk_guard``.
11886 This intrinsic is provided to perform the stack protector check by comparing
11887 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
11888 values do not match call the ``__stack_chk_fail()`` function.
11890 The reason to provide this as an IR level intrinsic instead of implementing it
11891 via other IR operations is that in order to perform this operation at the IR
11892 level without an intrinsic, one would need to create additional basic blocks to
11893 handle the success/failure cases. This makes it difficult to stop the stack
11894 protector check from disrupting sibling tail calls in Codegen. With this
11895 intrinsic, we are able to generate the stack protector basic blocks late in
11896 codegen after the tail call decision has occurred.
11898 '``llvm.objectsize``' Intrinsic
11899 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11906 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
11907 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
11912 The ``llvm.objectsize`` intrinsic is designed to provide information to
11913 the optimizers to determine at compile time whether a) an operation
11914 (like memcpy) will overflow a buffer that corresponds to an object, or
11915 b) that a runtime check for overflow isn't necessary. An object in this
11916 context means an allocation of a specific class, structure, array, or
11922 The ``llvm.objectsize`` intrinsic takes two arguments. The first
11923 argument is a pointer to or into the ``object``. The second argument is
11924 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
11925 or -1 (if false) when the object size is unknown. The second argument
11926 only accepts constants.
11931 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
11932 the size of the object concerned. If the size cannot be determined at
11933 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
11934 on the ``min`` argument).
11936 '``llvm.expect``' Intrinsic
11937 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
11942 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
11947 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
11948 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
11949 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
11954 The ``llvm.expect`` intrinsic provides information about expected (the
11955 most probable) value of ``val``, which can be used by optimizers.
11960 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
11961 a value. The second argument is an expected value, this needs to be a
11962 constant value, variables are not allowed.
11967 This intrinsic is lowered to the ``val``.
11971 '``llvm.assume``' Intrinsic
11972 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11979 declare void @llvm.assume(i1 %cond)
11984 The ``llvm.assume`` allows the optimizer to assume that the provided
11985 condition is true. This information can then be used in simplifying other parts
11991 The condition which the optimizer may assume is always true.
11996 The intrinsic allows the optimizer to assume that the provided condition is
11997 always true whenever the control flow reaches the intrinsic call. No code is
11998 generated for this intrinsic, and instructions that contribute only to the
11999 provided condition are not used for code generation. If the condition is
12000 violated during execution, the behavior is undefined.
12002 Note that the optimizer might limit the transformations performed on values
12003 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
12004 only used to form the intrinsic's input argument. This might prove undesirable
12005 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
12006 sufficient overall improvement in code quality. For this reason,
12007 ``llvm.assume`` should not be used to document basic mathematical invariants
12008 that the optimizer can otherwise deduce or facts that are of little use to the
12013 '``llvm.bitset.test``' Intrinsic
12014 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12021 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
12027 The first argument is a pointer to be tested. The second argument is a
12028 metadata object representing an identifier for a :doc:`bitset <BitSets>`.
12033 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
12034 member of the given bitset.
12036 '``llvm.donothing``' Intrinsic
12037 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12044 declare void @llvm.donothing() nounwind readnone
12049 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
12050 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
12051 with an invoke instruction.
12061 This intrinsic does nothing, and it's removed by optimizers and ignored
12064 Stack Map Intrinsics
12065 --------------------
12067 LLVM provides experimental intrinsics to support runtime patching
12068 mechanisms commonly desired in dynamic language JITs. These intrinsics
12069 are described in :doc:`StackMaps`.