1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][-a-zA-Z$._][-a-zA-Z$._0-9]*``'. Identifiers that require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves. The ``"\01"`` prefix
83 can be used on global variables to suppress mangling.
84 #. Unnamed values are represented as an unsigned numeric value with
85 their prefix. For example, ``%12``, ``@2``, ``%44``.
86 #. Constants, which are described in the section Constants_ below.
88 LLVM requires that values start with a prefix for two reasons: Compilers
89 don't need to worry about name clashes with reserved words, and the set
90 of reserved words may be expanded in the future without penalty.
91 Additionally, unnamed identifiers allow a compiler to quickly come up
92 with a temporary variable without having to avoid symbol table
95 Reserved words in LLVM are very similar to reserved words in other
96 languages. There are keywords for different opcodes ('``add``',
97 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
98 '``i32``', etc...), and others. These reserved words cannot conflict
99 with variable names, because none of them start with a prefix character
100 (``'%'`` or ``'@'``).
102 Here is an example of LLVM code to multiply the integer variable
109 %result = mul i32 %X, 8
111 After strength reduction:
115 %result = shl i32 %X, 3
121 %0 = add i32 %X, %X ; yields i32:%0
122 %1 = add i32 %0, %0 ; yields i32:%1
123 %result = add i32 %1, %1
125 This last way of multiplying ``%X`` by 8 illustrates several important
126 lexical features of LLVM:
128 #. Comments are delimited with a '``;``' and go until the end of line.
129 #. Unnamed temporaries are created when the result of a computation is
130 not assigned to a named value.
131 #. Unnamed temporaries are numbered sequentially (using a per-function
132 incrementing counter, starting with 0). Note that basic blocks and unnamed
133 function parameters are included in this numbering. For example, if the
134 entry basic block is not given a label name and all function parameters are
135 named, then it will get number 0.
137 It also shows a convention that we follow in this document. When
138 demonstrating instructions, we will follow an instruction with a comment
139 that defines the type and name of value produced.
147 LLVM programs are composed of ``Module``'s, each of which is a
148 translation unit of the input programs. Each module consists of
149 functions, global variables, and symbol table entries. Modules may be
150 combined together with the LLVM linker, which merges function (and
151 global variable) definitions, resolves forward declarations, and merges
152 symbol table entries. Here is an example of the "hello world" module:
156 ; Declare the string constant as a global constant.
157 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
159 ; External declaration of the puts function
160 declare i32 @puts(i8* nocapture) nounwind
162 ; Definition of main function
163 define i32 @main() { ; i32()*
164 ; Convert [13 x i8]* to i8 *...
165 %cast210 = getelementptr [13 x i8], [13 x i8]* @.str, i64 0, i64 0
167 ; Call puts function to write out the string to stdout.
168 call i32 @puts(i8* %cast210)
173 !0 = !{i32 42, null, !"string"}
176 This example is made up of a :ref:`global variable <globalvars>` named
177 "``.str``", an external declaration of the "``puts``" function, a
178 :ref:`function definition <functionstructure>` for "``main``" and
179 :ref:`named metadata <namedmetadatastructure>` "``foo``".
181 In general, a module is made up of a list of global values (where both
182 functions and global variables are global values). Global values are
183 represented by a pointer to a memory location (in this case, a pointer
184 to an array of char, and a pointer to a function), and have one of the
185 following :ref:`linkage types <linkage>`.
192 All Global Variables and Functions have one of the following types of
196 Global values with "``private``" linkage are only directly
197 accessible by objects in the current module. In particular, linking
198 code into a module with an private global value may cause the
199 private to be renamed as necessary to avoid collisions. Because the
200 symbol is private to the module, all references can be updated. This
201 doesn't show up in any symbol table in the object file.
203 Similar to private, but the value shows as a local symbol
204 (``STB_LOCAL`` in the case of ELF) in the object file. This
205 corresponds to the notion of the '``static``' keyword in C.
206 ``available_externally``
207 Globals with "``available_externally``" linkage are never emitted
208 into the object file corresponding to the LLVM module. They exist to
209 allow inlining and other optimizations to take place given knowledge
210 of the definition of the global, which is known to be somewhere
211 outside the module. Globals with ``available_externally`` linkage
212 are allowed to be discarded at will, and are otherwise the same as
213 ``linkonce_odr``. This linkage type is only allowed on definitions,
216 Globals with "``linkonce``" linkage are merged with other globals of
217 the same name when linkage occurs. This can be used to implement
218 some forms of inline functions, templates, or other code which must
219 be generated in each translation unit that uses it, but where the
220 body may be overridden with a more definitive definition later.
221 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
222 that ``linkonce`` linkage does not actually allow the optimizer to
223 inline the body of this function into callers because it doesn't
224 know if this definition of the function is the definitive definition
225 within the program or whether it will be overridden by a stronger
226 definition. To enable inlining and other optimizations, use
227 "``linkonce_odr``" linkage.
229 "``weak``" linkage has the same merging semantics as ``linkonce``
230 linkage, except that unreferenced globals with ``weak`` linkage may
231 not be discarded. This is used for globals that are declared "weak"
234 "``common``" linkage is most similar to "``weak``" linkage, but they
235 are used for tentative definitions in C, such as "``int X;``" at
236 global scope. Symbols with "``common``" linkage are merged in the
237 same way as ``weak symbols``, and they may not be deleted if
238 unreferenced. ``common`` symbols may not have an explicit section,
239 must have a zero initializer, and may not be marked
240 ':ref:`constant <globalvars>`'. Functions and aliases may not have
243 .. _linkage_appending:
246 "``appending``" linkage may only be applied to global variables of
247 pointer to array type. When two global variables with appending
248 linkage are linked together, the two global arrays are appended
249 together. This is the LLVM, typesafe, equivalent of having the
250 system linker append together "sections" with identical names when
253 The semantics of this linkage follow the ELF object file model: the
254 symbol is weak until linked, if not linked, the symbol becomes null
255 instead of being an undefined reference.
256 ``linkonce_odr``, ``weak_odr``
257 Some languages allow differing globals to be merged, such as two
258 functions with different semantics. Other languages, such as
259 ``C++``, ensure that only equivalent globals are ever merged (the
260 "one definition rule" --- "ODR"). Such languages can use the
261 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
262 global will only be merged with equivalent globals. These linkage
263 types are otherwise the same as their non-``odr`` versions.
265 If none of the above identifiers are used, the global is externally
266 visible, meaning that it participates in linkage and can be used to
267 resolve external symbol references.
269 It is illegal for a function *declaration* to have any linkage type
270 other than ``external`` or ``extern_weak``.
277 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
278 :ref:`invokes <i_invoke>` can all have an optional calling convention
279 specified for the call. The calling convention of any pair of dynamic
280 caller/callee must match, or the behavior of the program is undefined.
281 The following calling conventions are supported by LLVM, and more may be
284 "``ccc``" - The C calling convention
285 This calling convention (the default if no other calling convention
286 is specified) matches the target C calling conventions. This calling
287 convention supports varargs function calls and tolerates some
288 mismatch in the declared prototype and implemented declaration of
289 the function (as does normal C).
290 "``fastcc``" - The fast calling convention
291 This calling convention attempts to make calls as fast as possible
292 (e.g. by passing things in registers). This calling convention
293 allows the target to use whatever tricks it wants to produce fast
294 code for the target, without having to conform to an externally
295 specified ABI (Application Binary Interface). `Tail calls can only
296 be optimized when this, the GHC or the HiPE convention is
297 used. <CodeGenerator.html#id80>`_ This calling convention does not
298 support varargs and requires the prototype of all callees to exactly
299 match the prototype of the function definition.
300 "``coldcc``" - The cold calling convention
301 This calling convention attempts to make code in the caller as
302 efficient as possible under the assumption that the call is not
303 commonly executed. As such, these calls often preserve all registers
304 so that the call does not break any live ranges in the caller side.
305 This calling convention does not support varargs and requires the
306 prototype of all callees to exactly match the prototype of the
307 function definition. Furthermore the inliner doesn't consider such function
309 "``cc 10``" - GHC convention
310 This calling convention has been implemented specifically for use by
311 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
312 It passes everything in registers, going to extremes to achieve this
313 by disabling callee save registers. This calling convention should
314 not be used lightly but only for specific situations such as an
315 alternative to the *register pinning* performance technique often
316 used when implementing functional programming languages. At the
317 moment only X86 supports this convention and it has the following
320 - On *X86-32* only supports up to 4 bit type parameters. No
321 floating point types are supported.
322 - On *X86-64* only supports up to 10 bit type parameters and 6
323 floating point parameters.
325 This calling convention supports `tail call
326 optimization <CodeGenerator.html#id80>`_ but requires both the
327 caller and callee are using it.
328 "``cc 11``" - The HiPE calling convention
329 This calling convention has been implemented specifically for use by
330 the `High-Performance Erlang
331 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
332 native code compiler of the `Ericsson's Open Source Erlang/OTP
333 system <http://www.erlang.org/download.shtml>`_. It uses more
334 registers for argument passing than the ordinary C calling
335 convention and defines no callee-saved registers. The calling
336 convention properly supports `tail call
337 optimization <CodeGenerator.html#id80>`_ but requires that both the
338 caller and the callee use it. It uses a *register pinning*
339 mechanism, similar to GHC's convention, for keeping frequently
340 accessed runtime components pinned to specific hardware registers.
341 At the moment only X86 supports this convention (both 32 and 64
343 "``webkit_jscc``" - WebKit's JavaScript calling convention
344 This calling convention has been implemented for `WebKit FTL JIT
345 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
346 stack right to left (as cdecl does), and returns a value in the
347 platform's customary return register.
348 "``anyregcc``" - Dynamic calling convention for code patching
349 This is a special convention that supports patching an arbitrary code
350 sequence in place of a call site. This convention forces the call
351 arguments into registers but allows them to be dynamically
352 allocated. This can currently only be used with calls to
353 llvm.experimental.patchpoint because only this intrinsic records
354 the location of its arguments in a side table. See :doc:`StackMaps`.
355 "``preserve_mostcc``" - The `PreserveMost` calling convention
356 This calling convention attempts to make the code in the caller as
357 unintrusive as possible. This convention behaves identically to the `C`
358 calling convention on how arguments and return values are passed, but it
359 uses a different set of caller/callee-saved registers. This alleviates the
360 burden of saving and recovering a large register set before and after the
361 call in the caller. If the arguments are passed in callee-saved registers,
362 then they will be preserved by the callee across the call. This doesn't
363 apply for values returned in callee-saved registers.
365 - On X86-64 the callee preserves all general purpose registers, except for
366 R11. R11 can be used as a scratch register. Floating-point registers
367 (XMMs/YMMs) are not preserved and need to be saved by the caller.
369 The idea behind this convention is to support calls to runtime functions
370 that have a hot path and a cold path. The hot path is usually a small piece
371 of code that doesn't use many registers. The cold path might need to call out to
372 another function and therefore only needs to preserve the caller-saved
373 registers, which haven't already been saved by the caller. The
374 `PreserveMost` calling convention is very similar to the `cold` calling
375 convention in terms of caller/callee-saved registers, but they are used for
376 different types of function calls. `coldcc` is for function calls that are
377 rarely executed, whereas `preserve_mostcc` function calls are intended to be
378 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
379 doesn't prevent the inliner from inlining the function call.
381 This calling convention will be used by a future version of the ObjectiveC
382 runtime and should therefore still be considered experimental at this time.
383 Although this convention was created to optimize certain runtime calls to
384 the ObjectiveC runtime, it is not limited to this runtime and might be used
385 by other runtimes in the future too. The current implementation only
386 supports X86-64, but the intention is to support more architectures in the
388 "``preserve_allcc``" - The `PreserveAll` calling convention
389 This calling convention attempts to make the code in the caller even less
390 intrusive than the `PreserveMost` calling convention. This calling
391 convention also behaves identical to the `C` calling convention on how
392 arguments and return values are passed, but it uses a different set of
393 caller/callee-saved registers. This removes the burden of saving and
394 recovering a large register set before and after the call in the caller. If
395 the arguments are passed in callee-saved registers, then they will be
396 preserved by the callee across the call. This doesn't apply for values
397 returned in callee-saved registers.
399 - On X86-64 the callee preserves all general purpose registers, except for
400 R11. R11 can be used as a scratch register. Furthermore it also preserves
401 all floating-point registers (XMMs/YMMs).
403 The idea behind this convention is to support calls to runtime functions
404 that don't need to call out to any other functions.
406 This calling convention, like the `PreserveMost` calling convention, will be
407 used by a future version of the ObjectiveC runtime and should be considered
408 experimental at this time.
409 "``cc <n>``" - Numbered convention
410 Any calling convention may be specified by number, allowing
411 target-specific calling conventions to be used. Target specific
412 calling conventions start at 64.
414 More calling conventions can be added/defined on an as-needed basis, to
415 support Pascal conventions or any other well-known target-independent
418 .. _visibilitystyles:
423 All Global Variables and Functions have one of the following visibility
426 "``default``" - Default style
427 On targets that use the ELF object file format, default visibility
428 means that the declaration is visible to other modules and, in
429 shared libraries, means that the declared entity may be overridden.
430 On Darwin, default visibility means that the declaration is visible
431 to other modules. Default visibility corresponds to "external
432 linkage" in the language.
433 "``hidden``" - Hidden style
434 Two declarations of an object with hidden visibility refer to the
435 same object if they are in the same shared object. Usually, hidden
436 visibility indicates that the symbol will not be placed into the
437 dynamic symbol table, so no other module (executable or shared
438 library) can reference it directly.
439 "``protected``" - Protected style
440 On ELF, protected visibility indicates that the symbol will be
441 placed in the dynamic symbol table, but that references within the
442 defining module will bind to the local symbol. That is, the symbol
443 cannot be overridden by another module.
445 A symbol with ``internal`` or ``private`` linkage must have ``default``
453 All Global Variables, Functions and Aliases can have one of the following
457 "``dllimport``" causes the compiler to reference a function or variable via
458 a global pointer to a pointer that is set up by the DLL exporting the
459 symbol. On Microsoft Windows targets, the pointer name is formed by
460 combining ``__imp_`` and the function or variable name.
462 "``dllexport``" causes the compiler to provide a global pointer to a pointer
463 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
464 Microsoft Windows targets, the pointer name is formed by combining
465 ``__imp_`` and the function or variable name. Since this storage class
466 exists for defining a dll interface, the compiler, assembler and linker know
467 it is externally referenced and must refrain from deleting the symbol.
471 Thread Local Storage Models
472 ---------------------------
474 A variable may be defined as ``thread_local``, which means that it will
475 not be shared by threads (each thread will have a separated copy of the
476 variable). Not all targets support thread-local variables. Optionally, a
477 TLS model may be specified:
480 For variables that are only used within the current shared library.
482 For variables in modules that will not be loaded dynamically.
484 For variables defined in the executable and only used within it.
486 If no explicit model is given, the "general dynamic" model is used.
488 The models correspond to the ELF TLS models; see `ELF Handling For
489 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
490 more information on under which circumstances the different models may
491 be used. The target may choose a different TLS model if the specified
492 model is not supported, or if a better choice of model can be made.
494 A model can also be specified in an alias, but then it only governs how
495 the alias is accessed. It will not have any effect in the aliasee.
497 For platforms without linker support of ELF TLS model, the -femulated-tls
498 flag can be used to generate GCC compatible emulated TLS code.
505 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
506 types <t_struct>`. Literal types are uniqued structurally, but identified types
507 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
508 to forward declare a type that is not yet available.
510 An example of an identified structure specification is:
514 %mytype = type { %mytype*, i32 }
516 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
517 literal types are uniqued in recent versions of LLVM.
524 Global variables define regions of memory allocated at compilation time
527 Global variable definitions must be initialized.
529 Global variables in other translation units can also be declared, in which
530 case they don't have an initializer.
532 Either global variable definitions or declarations may have an explicit section
533 to be placed in and may have an optional explicit alignment specified.
535 A variable may be defined as a global ``constant``, which indicates that
536 the contents of the variable will **never** be modified (enabling better
537 optimization, allowing the global data to be placed in the read-only
538 section of an executable, etc). Note that variables that need runtime
539 initialization cannot be marked ``constant`` as there is a store to the
542 LLVM explicitly allows *declarations* of global variables to be marked
543 constant, even if the final definition of the global is not. This
544 capability can be used to enable slightly better optimization of the
545 program, but requires the language definition to guarantee that
546 optimizations based on the 'constantness' are valid for the translation
547 units that do not include the definition.
549 As SSA values, global variables define pointer values that are in scope
550 (i.e. they dominate) all basic blocks in the program. Global variables
551 always define a pointer to their "content" type because they describe a
552 region of memory, and all memory objects in LLVM are accessed through
555 Global variables can be marked with ``unnamed_addr`` which indicates
556 that the address is not significant, only the content. Constants marked
557 like this can be merged with other constants if they have the same
558 initializer. Note that a constant with significant address *can* be
559 merged with a ``unnamed_addr`` constant, the result being a constant
560 whose address is significant.
562 A global variable may be declared to reside in a target-specific
563 numbered address space. For targets that support them, address spaces
564 may affect how optimizations are performed and/or what target
565 instructions are used to access the variable. The default address space
566 is zero. The address space qualifier must precede any other attributes.
568 LLVM allows an explicit section to be specified for globals. If the
569 target supports it, it will emit globals to the section specified.
570 Additionally, the global can placed in a comdat if the target has the necessary
573 By default, global initializers are optimized by assuming that global
574 variables defined within the module are not modified from their
575 initial values before the start of the global initializer. This is
576 true even for variables potentially accessible from outside the
577 module, including those with external linkage or appearing in
578 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
579 by marking the variable with ``externally_initialized``.
581 An explicit alignment may be specified for a global, which must be a
582 power of 2. If not present, or if the alignment is set to zero, the
583 alignment of the global is set by the target to whatever it feels
584 convenient. If an explicit alignment is specified, the global is forced
585 to have exactly that alignment. Targets and optimizers are not allowed
586 to over-align the global if the global has an assigned section. In this
587 case, the extra alignment could be observable: for example, code could
588 assume that the globals are densely packed in their section and try to
589 iterate over them as an array, alignment padding would break this
590 iteration. The maximum alignment is ``1 << 29``.
592 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
594 Variables and aliases can have a
595 :ref:`Thread Local Storage Model <tls_model>`.
599 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
600 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
601 <global | constant> <Type> [<InitializerConstant>]
602 [, section "name"] [, comdat [($name)]]
603 [, align <Alignment>]
605 For example, the following defines a global in a numbered address space
606 with an initializer, section, and alignment:
610 @G = addrspace(5) constant float 1.0, section "foo", align 4
612 The following example just declares a global variable
616 @G = external global i32
618 The following example defines a thread-local global with the
619 ``initialexec`` TLS model:
623 @G = thread_local(initialexec) global i32 0, align 4
625 .. _functionstructure:
630 LLVM function definitions consist of the "``define``" keyword, an
631 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
632 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
633 an optional :ref:`calling convention <callingconv>`,
634 an optional ``unnamed_addr`` attribute, a return type, an optional
635 :ref:`parameter attribute <paramattrs>` for the return type, a function
636 name, a (possibly empty) argument list (each with optional :ref:`parameter
637 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
638 an optional section, an optional alignment,
639 an optional :ref:`comdat <langref_comdats>`,
640 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
641 an optional :ref:`prologue <prologuedata>`,
642 an optional :ref:`personality <personalityfn>`,
643 an optional list of attached :ref:`metadata <metadata>`,
644 an opening curly brace, a list of basic blocks, and a closing curly brace.
646 LLVM function declarations consist of the "``declare``" keyword, an
647 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
648 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
649 an optional :ref:`calling convention <callingconv>`,
650 an optional ``unnamed_addr`` attribute, a return type, an optional
651 :ref:`parameter attribute <paramattrs>` for the return type, a function
652 name, a possibly empty list of arguments, an optional alignment, an optional
653 :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
654 and an optional :ref:`prologue <prologuedata>`.
656 A function definition contains a list of basic blocks, forming the CFG (Control
657 Flow Graph) for the function. Each basic block may optionally start with a label
658 (giving the basic block a symbol table entry), contains a list of instructions,
659 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
660 function return). If an explicit label is not provided, a block is assigned an
661 implicit numbered label, using the next value from the same counter as used for
662 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
663 entry block does not have an explicit label, it will be assigned label "%0",
664 then the first unnamed temporary in that block will be "%1", etc.
666 The first basic block in a function is special in two ways: it is
667 immediately executed on entrance to the function, and it is not allowed
668 to have predecessor basic blocks (i.e. there can not be any branches to
669 the entry block of a function). Because the block can have no
670 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
672 LLVM allows an explicit section to be specified for functions. If the
673 target supports it, it will emit functions to the section specified.
674 Additionally, the function can be placed in a COMDAT.
676 An explicit alignment may be specified for a function. If not present,
677 or if the alignment is set to zero, the alignment of the function is set
678 by the target to whatever it feels convenient. If an explicit alignment
679 is specified, the function is forced to have at least that much
680 alignment. All alignments must be a power of 2.
682 If the ``unnamed_addr`` attribute is given, the address is known to not
683 be significant and two identical functions can be merged.
687 define [linkage] [visibility] [DLLStorageClass]
689 <ResultType> @<FunctionName> ([argument list])
690 [unnamed_addr] [fn Attrs] [section "name"] [comdat [($name)]]
691 [align N] [gc] [prefix Constant] [prologue Constant]
692 [personality Constant] (!name !N)* { ... }
694 The argument list is a comma separated sequence of arguments where each
695 argument is of the following form:
699 <type> [parameter Attrs] [name]
707 Aliases, unlike function or variables, don't create any new data. They
708 are just a new symbol and metadata for an existing position.
710 Aliases have a name and an aliasee that is either a global value or a
713 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
714 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
715 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
719 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy>, <AliaseeTy>* @<Aliasee>
721 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
722 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
723 might not correctly handle dropping a weak symbol that is aliased.
725 Aliases that are not ``unnamed_addr`` are guaranteed to have the same address as
726 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
729 Since aliases are only a second name, some restrictions apply, of which
730 some can only be checked when producing an object file:
732 * The expression defining the aliasee must be computable at assembly
733 time. Since it is just a name, no relocations can be used.
735 * No alias in the expression can be weak as the possibility of the
736 intermediate alias being overridden cannot be represented in an
739 * No global value in the expression can be a declaration, since that
740 would require a relocation, which is not possible.
747 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
749 Comdats have a name which represents the COMDAT key. All global objects that
750 specify this key will only end up in the final object file if the linker chooses
751 that key over some other key. Aliases are placed in the same COMDAT that their
752 aliasee computes to, if any.
754 Comdats have a selection kind to provide input on how the linker should
755 choose between keys in two different object files.
759 $<Name> = comdat SelectionKind
761 The selection kind must be one of the following:
764 The linker may choose any COMDAT key, the choice is arbitrary.
766 The linker may choose any COMDAT key but the sections must contain the
769 The linker will choose the section containing the largest COMDAT key.
771 The linker requires that only section with this COMDAT key exist.
773 The linker may choose any COMDAT key but the sections must contain the
776 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
777 ``any`` as a selection kind.
779 Here is an example of a COMDAT group where a function will only be selected if
780 the COMDAT key's section is the largest:
784 $foo = comdat largest
785 @foo = global i32 2, comdat($foo)
787 define void @bar() comdat($foo) {
791 As a syntactic sugar the ``$name`` can be omitted if the name is the same as
797 @foo = global i32 2, comdat
800 In a COFF object file, this will create a COMDAT section with selection kind
801 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
802 and another COMDAT section with selection kind
803 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
804 section and contains the contents of the ``@bar`` symbol.
806 There are some restrictions on the properties of the global object.
807 It, or an alias to it, must have the same name as the COMDAT group when
809 The contents and size of this object may be used during link-time to determine
810 which COMDAT groups get selected depending on the selection kind.
811 Because the name of the object must match the name of the COMDAT group, the
812 linkage of the global object must not be local; local symbols can get renamed
813 if a collision occurs in the symbol table.
815 The combined use of COMDATS and section attributes may yield surprising results.
822 @g1 = global i32 42, section "sec", comdat($foo)
823 @g2 = global i32 42, section "sec", comdat($bar)
825 From the object file perspective, this requires the creation of two sections
826 with the same name. This is necessary because both globals belong to different
827 COMDAT groups and COMDATs, at the object file level, are represented by
830 Note that certain IR constructs like global variables and functions may
831 create COMDATs in the object file in addition to any which are specified using
832 COMDAT IR. This arises when the code generator is configured to emit globals
833 in individual sections (e.g. when `-data-sections` or `-function-sections`
834 is supplied to `llc`).
836 .. _namedmetadatastructure:
841 Named metadata is a collection of metadata. :ref:`Metadata
842 nodes <metadata>` (but not metadata strings) are the only valid
843 operands for a named metadata.
845 #. Named metadata are represented as a string of characters with the
846 metadata prefix. The rules for metadata names are the same as for
847 identifiers, but quoted names are not allowed. ``"\xx"`` type escapes
848 are still valid, which allows any character to be part of a name.
852 ; Some unnamed metadata nodes, which are referenced by the named metadata.
857 !name = !{!0, !1, !2}
864 The return type and each parameter of a function type may have a set of
865 *parameter attributes* associated with them. Parameter attributes are
866 used to communicate additional information about the result or
867 parameters of a function. Parameter attributes are considered to be part
868 of the function, not of the function type, so functions with different
869 parameter attributes can have the same function type.
871 Parameter attributes are simple keywords that follow the type specified.
872 If multiple parameter attributes are needed, they are space separated.
877 declare i32 @printf(i8* noalias nocapture, ...)
878 declare i32 @atoi(i8 zeroext)
879 declare signext i8 @returns_signed_char()
881 Note that any attributes for the function result (``nounwind``,
882 ``readonly``) come immediately after the argument list.
884 Currently, only the following parameter attributes are defined:
887 This indicates to the code generator that the parameter or return
888 value should be zero-extended to the extent required by the target's
889 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
890 the caller (for a parameter) or the callee (for a return value).
892 This indicates to the code generator that the parameter or return
893 value should be sign-extended to the extent required by the target's
894 ABI (which is usually 32-bits) by the caller (for a parameter) or
895 the callee (for a return value).
897 This indicates that this parameter or return value should be treated
898 in a special target-dependent fashion while emitting code for
899 a function call or return (usually, by putting it in a register as
900 opposed to memory, though some targets use it to distinguish between
901 two different kinds of registers). Use of this attribute is
904 This indicates that the pointer parameter should really be passed by
905 value to the function. The attribute implies that a hidden copy of
906 the pointee is made between the caller and the callee, so the callee
907 is unable to modify the value in the caller. This attribute is only
908 valid on LLVM pointer arguments. It is generally used to pass
909 structs and arrays by value, but is also valid on pointers to
910 scalars. The copy is considered to belong to the caller not the
911 callee (for example, ``readonly`` functions should not write to
912 ``byval`` parameters). This is not a valid attribute for return
915 The byval attribute also supports specifying an alignment with the
916 align attribute. It indicates the alignment of the stack slot to
917 form and the known alignment of the pointer specified to the call
918 site. If the alignment is not specified, then the code generator
919 makes a target-specific assumption.
925 The ``inalloca`` argument attribute allows the caller to take the
926 address of outgoing stack arguments. An ``inalloca`` argument must
927 be a pointer to stack memory produced by an ``alloca`` instruction.
928 The alloca, or argument allocation, must also be tagged with the
929 inalloca keyword. Only the last argument may have the ``inalloca``
930 attribute, and that argument is guaranteed to be passed in memory.
932 An argument allocation may be used by a call at most once because
933 the call may deallocate it. The ``inalloca`` attribute cannot be
934 used in conjunction with other attributes that affect argument
935 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
936 ``inalloca`` attribute also disables LLVM's implicit lowering of
937 large aggregate return values, which means that frontend authors
938 must lower them with ``sret`` pointers.
940 When the call site is reached, the argument allocation must have
941 been the most recent stack allocation that is still live, or the
942 results are undefined. It is possible to allocate additional stack
943 space after an argument allocation and before its call site, but it
944 must be cleared off with :ref:`llvm.stackrestore
947 See :doc:`InAlloca` for more information on how to use this
951 This indicates that the pointer parameter specifies the address of a
952 structure that is the return value of the function in the source
953 program. This pointer must be guaranteed by the caller to be valid:
954 loads and stores to the structure may be assumed by the callee
955 not to trap and to be properly aligned. This may only be applied to
956 the first parameter. This is not a valid attribute for return
960 This indicates that the pointer value may be assumed by the optimizer to
961 have the specified alignment.
963 Note that this attribute has additional semantics when combined with the
969 This indicates that objects accessed via pointer values
970 :ref:`based <pointeraliasing>` on the argument or return value are not also
971 accessed, during the execution of the function, via pointer values not
972 *based* on the argument or return value. The attribute on a return value
973 also has additional semantics described below. The caller shares the
974 responsibility with the callee for ensuring that these requirements are met.
975 For further details, please see the discussion of the NoAlias response in
976 :ref:`alias analysis <Must, May, or No>`.
978 Note that this definition of ``noalias`` is intentionally similar
979 to the definition of ``restrict`` in C99 for function arguments.
981 For function return values, C99's ``restrict`` is not meaningful,
982 while LLVM's ``noalias`` is. Furthermore, the semantics of the ``noalias``
983 attribute on return values are stronger than the semantics of the attribute
984 when used on function arguments. On function return values, the ``noalias``
985 attribute indicates that the function acts like a system memory allocation
986 function, returning a pointer to allocated storage disjoint from the
987 storage for any other object accessible to the caller.
990 This indicates that the callee does not make any copies of the
991 pointer that outlive the callee itself. This is not a valid
992 attribute for return values.
997 This indicates that the pointer parameter can be excised using the
998 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
999 attribute for return values and can only be applied to one parameter.
1002 This indicates that the function always returns the argument as its return
1003 value. This is an optimization hint to the code generator when generating
1004 the caller, allowing tail call optimization and omission of register saves
1005 and restores in some cases; it is not checked or enforced when generating
1006 the callee. The parameter and the function return type must be valid
1007 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
1008 valid attribute for return values and can only be applied to one parameter.
1011 This indicates that the parameter or return pointer is not null. This
1012 attribute may only be applied to pointer typed parameters. This is not
1013 checked or enforced by LLVM, the caller must ensure that the pointer
1014 passed in is non-null, or the callee must ensure that the returned pointer
1017 ``dereferenceable(<n>)``
1018 This indicates that the parameter or return pointer is dereferenceable. This
1019 attribute may only be applied to pointer typed parameters. A pointer that
1020 is dereferenceable can be loaded from speculatively without a risk of
1021 trapping. The number of bytes known to be dereferenceable must be provided
1022 in parentheses. It is legal for the number of bytes to be less than the
1023 size of the pointee type. The ``nonnull`` attribute does not imply
1024 dereferenceability (consider a pointer to one element past the end of an
1025 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
1026 ``addrspace(0)`` (which is the default address space).
1028 ``dereferenceable_or_null(<n>)``
1029 This indicates that the parameter or return value isn't both
1030 non-null and non-dereferenceable (up to ``<n>`` bytes) at the same
1031 time. All non-null pointers tagged with
1032 ``dereferenceable_or_null(<n>)`` are ``dereferenceable(<n>)``.
1033 For address space 0 ``dereferenceable_or_null(<n>)`` implies that
1034 a pointer is exactly one of ``dereferenceable(<n>)`` or ``null``,
1035 and in other address spaces ``dereferenceable_or_null(<n>)``
1036 implies that a pointer is at least one of ``dereferenceable(<n>)``
1037 or ``null`` (i.e. it may be both ``null`` and
1038 ``dereferenceable(<n>)``). This attribute may only be applied to
1039 pointer typed parameters.
1043 Garbage Collector Strategy Names
1044 --------------------------------
1046 Each function may specify a garbage collector strategy name, which is simply a
1049 .. code-block:: llvm
1051 define void @f() gc "name" { ... }
1053 The supported values of *name* includes those :ref:`built in to LLVM
1054 <builtin-gc-strategies>` and any provided by loaded plugins. Specifying a GC
1055 strategy will cause the compiler to alter its output in order to support the
1056 named garbage collection algorithm. Note that LLVM itself does not contain a
1057 garbage collector, this functionality is restricted to generating machine code
1058 which can interoperate with a collector provided externally.
1065 Prefix data is data associated with a function which the code
1066 generator will emit immediately before the function's entrypoint.
1067 The purpose of this feature is to allow frontends to associate
1068 language-specific runtime metadata with specific functions and make it
1069 available through the function pointer while still allowing the
1070 function pointer to be called.
1072 To access the data for a given function, a program may bitcast the
1073 function pointer to a pointer to the constant's type and dereference
1074 index -1. This implies that the IR symbol points just past the end of
1075 the prefix data. For instance, take the example of a function annotated
1076 with a single ``i32``,
1078 .. code-block:: llvm
1080 define void @f() prefix i32 123 { ... }
1082 The prefix data can be referenced as,
1084 .. code-block:: llvm
1086 %0 = bitcast void* () @f to i32*
1087 %a = getelementptr inbounds i32, i32* %0, i32 -1
1088 %b = load i32, i32* %a
1090 Prefix data is laid out as if it were an initializer for a global variable
1091 of the prefix data's type. The function will be placed such that the
1092 beginning of the prefix data is aligned. This means that if the size
1093 of the prefix data is not a multiple of the alignment size, the
1094 function's entrypoint will not be aligned. If alignment of the
1095 function's entrypoint is desired, padding must be added to the prefix
1098 A function may have prefix data but no body. This has similar semantics
1099 to the ``available_externally`` linkage in that the data may be used by the
1100 optimizers but will not be emitted in the object file.
1107 The ``prologue`` attribute allows arbitrary code (encoded as bytes) to
1108 be inserted prior to the function body. This can be used for enabling
1109 function hot-patching and instrumentation.
1111 To maintain the semantics of ordinary function calls, the prologue data must
1112 have a particular format. Specifically, it must begin with a sequence of
1113 bytes which decode to a sequence of machine instructions, valid for the
1114 module's target, which transfer control to the point immediately succeeding
1115 the prologue data, without performing any other visible action. This allows
1116 the inliner and other passes to reason about the semantics of the function
1117 definition without needing to reason about the prologue data. Obviously this
1118 makes the format of the prologue data highly target dependent.
1120 A trivial example of valid prologue data for the x86 architecture is ``i8 144``,
1121 which encodes the ``nop`` instruction:
1123 .. code-block:: llvm
1125 define void @f() prologue i8 144 { ... }
1127 Generally prologue data can be formed by encoding a relative branch instruction
1128 which skips the metadata, as in this example of valid prologue data for the
1129 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1131 .. code-block:: llvm
1133 %0 = type <{ i8, i8, i8* }>
1135 define void @f() prologue %0 <{ i8 235, i8 8, i8* @md}> { ... }
1137 A function may have prologue data but no body. This has similar semantics
1138 to the ``available_externally`` linkage in that the data may be used by the
1139 optimizers but will not be emitted in the object file.
1143 Personality Function
1144 --------------------
1146 The ``personality`` attribute permits functions to specify what function
1147 to use for exception handling.
1154 Attribute groups are groups of attributes that are referenced by objects within
1155 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1156 functions will use the same set of attributes. In the degenerative case of a
1157 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1158 group will capture the important command line flags used to build that file.
1160 An attribute group is a module-level object. To use an attribute group, an
1161 object references the attribute group's ID (e.g. ``#37``). An object may refer
1162 to more than one attribute group. In that situation, the attributes from the
1163 different groups are merged.
1165 Here is an example of attribute groups for a function that should always be
1166 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1168 .. code-block:: llvm
1170 ; Target-independent attributes:
1171 attributes #0 = { alwaysinline alignstack=4 }
1173 ; Target-dependent attributes:
1174 attributes #1 = { "no-sse" }
1176 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1177 define void @f() #0 #1 { ... }
1184 Function attributes are set to communicate additional information about
1185 a function. Function attributes are considered to be part of the
1186 function, not of the function type, so functions with different function
1187 attributes can have the same function type.
1189 Function attributes are simple keywords that follow the type specified.
1190 If multiple attributes are needed, they are space separated. For
1193 .. code-block:: llvm
1195 define void @f() noinline { ... }
1196 define void @f() alwaysinline { ... }
1197 define void @f() alwaysinline optsize { ... }
1198 define void @f() optsize { ... }
1201 This attribute indicates that, when emitting the prologue and
1202 epilogue, the backend should forcibly align the stack pointer.
1203 Specify the desired alignment, which must be a power of two, in
1206 This attribute indicates that the inliner should attempt to inline
1207 this function into callers whenever possible, ignoring any active
1208 inlining size threshold for this caller.
1210 This indicates that the callee function at a call site should be
1211 recognized as a built-in function, even though the function's declaration
1212 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1213 direct calls to functions that are declared with the ``nobuiltin``
1216 This attribute indicates that this function is rarely called. When
1217 computing edge weights, basic blocks post-dominated by a cold
1218 function call are also considered to be cold; and, thus, given low
1221 This attribute indicates that the callee is dependent on a convergent
1222 thread execution pattern under certain parallel execution models.
1223 Transformations that are execution model agnostic may not make the execution
1224 of a convergent operation control dependent on any additional values.
1226 This attribute indicates that the source code contained a hint that
1227 inlining this function is desirable (such as the "inline" keyword in
1228 C/C++). It is just a hint; it imposes no requirements on the
1231 This attribute indicates that the function should be added to a
1232 jump-instruction table at code-generation time, and that all address-taken
1233 references to this function should be replaced with a reference to the
1234 appropriate jump-instruction-table function pointer. Note that this creates
1235 a new pointer for the original function, which means that code that depends
1236 on function-pointer identity can break. So, any function annotated with
1237 ``jumptable`` must also be ``unnamed_addr``.
1239 This attribute suggests that optimization passes and code generator
1240 passes make choices that keep the code size of this function as small
1241 as possible and perform optimizations that may sacrifice runtime
1242 performance in order to minimize the size of the generated code.
1244 This attribute disables prologue / epilogue emission for the
1245 function. This can have very system-specific consequences.
1247 This indicates that the callee function at a call site is not recognized as
1248 a built-in function. LLVM will retain the original call and not replace it
1249 with equivalent code based on the semantics of the built-in function, unless
1250 the call site uses the ``builtin`` attribute. This is valid at call sites
1251 and on function declarations and definitions.
1253 This attribute indicates that calls to the function cannot be
1254 duplicated. A call to a ``noduplicate`` function may be moved
1255 within its parent function, but may not be duplicated within
1256 its parent function.
1258 A function containing a ``noduplicate`` call may still
1259 be an inlining candidate, provided that the call is not
1260 duplicated by inlining. That implies that the function has
1261 internal linkage and only has one call site, so the original
1262 call is dead after inlining.
1264 This attributes disables implicit floating point instructions.
1266 This attribute indicates that the inliner should never inline this
1267 function in any situation. This attribute may not be used together
1268 with the ``alwaysinline`` attribute.
1270 This attribute suppresses lazy symbol binding for the function. This
1271 may make calls to the function faster, at the cost of extra program
1272 startup time if the function is not called during program startup.
1274 This attribute indicates that the code generator should not use a
1275 red zone, even if the target-specific ABI normally permits it.
1277 This function attribute indicates that the function never returns
1278 normally. This produces undefined behavior at runtime if the
1279 function ever does dynamically return.
1281 This function attribute indicates that the function does not call itself
1282 either directly or indirectly down any possible call path. This produces
1283 undefined behavior at runtime if the function ever does recurse.
1285 This function attribute indicates that the function never raises an
1286 exception. If the function does raise an exception, its runtime
1287 behavior is undefined. However, functions marked nounwind may still
1288 trap or generate asynchronous exceptions. Exception handling schemes
1289 that are recognized by LLVM to handle asynchronous exceptions, such
1290 as SEH, will still provide their implementation defined semantics.
1292 This function attribute indicates that the function is not optimized
1293 by any optimization or code generator passes with the
1294 exception of interprocedural optimization passes.
1295 This attribute cannot be used together with the ``alwaysinline``
1296 attribute; this attribute is also incompatible
1297 with the ``minsize`` attribute and the ``optsize`` attribute.
1299 This attribute requires the ``noinline`` attribute to be specified on
1300 the function as well, so the function is never inlined into any caller.
1301 Only functions with the ``alwaysinline`` attribute are valid
1302 candidates for inlining into the body of this function.
1304 This attribute suggests that optimization passes and code generator
1305 passes make choices that keep the code size of this function low,
1306 and otherwise do optimizations specifically to reduce code size as
1307 long as they do not significantly impact runtime performance.
1309 On a function, this attribute indicates that the function computes its
1310 result (or decides to unwind an exception) based strictly on its arguments,
1311 without dereferencing any pointer arguments or otherwise accessing
1312 any mutable state (e.g. memory, control registers, etc) visible to
1313 caller functions. It does not write through any pointer arguments
1314 (including ``byval`` arguments) and never changes any state visible
1315 to callers. This means that it cannot unwind exceptions by calling
1316 the ``C++`` exception throwing methods.
1318 On an argument, this attribute indicates that the function does not
1319 dereference that pointer argument, even though it may read or write the
1320 memory that the pointer points to if accessed through other pointers.
1322 On a function, this attribute indicates that the function does not write
1323 through any pointer arguments (including ``byval`` arguments) or otherwise
1324 modify any state (e.g. memory, control registers, etc) visible to
1325 caller functions. It may dereference pointer arguments and read
1326 state that may be set in the caller. A readonly function always
1327 returns the same value (or unwinds an exception identically) when
1328 called with the same set of arguments and global state. It cannot
1329 unwind an exception by calling the ``C++`` exception throwing
1332 On an argument, this attribute indicates that the function does not write
1333 through this pointer argument, even though it may write to the memory that
1334 the pointer points to.
1336 This attribute indicates that the only memory accesses inside function are
1337 loads and stores from objects pointed to by its pointer-typed arguments,
1338 with arbitrary offsets. Or in other words, all memory operations in the
1339 function can refer to memory only using pointers based on its function
1341 Note that ``argmemonly`` can be used together with ``readonly`` attribute
1342 in order to specify that function reads only from its arguments.
1344 This attribute indicates that this function can return twice. The C
1345 ``setjmp`` is an example of such a function. The compiler disables
1346 some optimizations (like tail calls) in the caller of these
1349 This attribute indicates that
1350 `SafeStack <http://clang.llvm.org/docs/SafeStack.html>`_
1351 protection is enabled for this function.
1353 If a function that has a ``safestack`` attribute is inlined into a
1354 function that doesn't have a ``safestack`` attribute or which has an
1355 ``ssp``, ``sspstrong`` or ``sspreq`` attribute, then the resulting
1356 function will have a ``safestack`` attribute.
1357 ``sanitize_address``
1358 This attribute indicates that AddressSanitizer checks
1359 (dynamic address safety analysis) are enabled for this function.
1361 This attribute indicates that MemorySanitizer checks (dynamic detection
1362 of accesses to uninitialized memory) are enabled for this function.
1364 This attribute indicates that ThreadSanitizer checks
1365 (dynamic thread safety analysis) are enabled for this function.
1367 This attribute indicates that the function should emit a stack
1368 smashing protector. It is in the form of a "canary" --- a random value
1369 placed on the stack before the local variables that's checked upon
1370 return from the function to see if it has been overwritten. A
1371 heuristic is used to determine if a function needs stack protectors
1372 or not. The heuristic used will enable protectors for functions with:
1374 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1375 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1376 - Calls to alloca() with variable sizes or constant sizes greater than
1377 ``ssp-buffer-size``.
1379 Variables that are identified as requiring a protector will be arranged
1380 on the stack such that they are adjacent to the stack protector guard.
1382 If a function that has an ``ssp`` attribute is inlined into a
1383 function that doesn't have an ``ssp`` attribute, then the resulting
1384 function will have an ``ssp`` attribute.
1386 This attribute indicates that the function should *always* emit a
1387 stack smashing protector. This overrides the ``ssp`` function
1390 Variables that are identified as requiring a protector will be arranged
1391 on the stack such that they are adjacent to the stack protector guard.
1392 The specific layout rules are:
1394 #. Large arrays and structures containing large arrays
1395 (``>= ssp-buffer-size``) are closest to the stack protector.
1396 #. Small arrays and structures containing small arrays
1397 (``< ssp-buffer-size``) are 2nd closest to the protector.
1398 #. Variables that have had their address taken are 3rd closest to the
1401 If a function that has an ``sspreq`` attribute is inlined into a
1402 function that doesn't have an ``sspreq`` attribute or which has an
1403 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1404 an ``sspreq`` attribute.
1406 This attribute indicates that the function should emit a stack smashing
1407 protector. This attribute causes a strong heuristic to be used when
1408 determining if a function needs stack protectors. The strong heuristic
1409 will enable protectors for functions with:
1411 - Arrays of any size and type
1412 - Aggregates containing an array of any size and type.
1413 - Calls to alloca().
1414 - Local variables that have had their address taken.
1416 Variables that are identified as requiring a protector will be arranged
1417 on the stack such that they are adjacent to the stack protector guard.
1418 The specific layout rules are:
1420 #. Large arrays and structures containing large arrays
1421 (``>= ssp-buffer-size``) are closest to the stack protector.
1422 #. Small arrays and structures containing small arrays
1423 (``< ssp-buffer-size``) are 2nd closest to the protector.
1424 #. Variables that have had their address taken are 3rd closest to the
1427 This overrides the ``ssp`` function attribute.
1429 If a function that has an ``sspstrong`` attribute is inlined into a
1430 function that doesn't have an ``sspstrong`` attribute, then the
1431 resulting function will have an ``sspstrong`` attribute.
1433 This attribute indicates that the function will delegate to some other
1434 function with a tail call. The prototype of a thunk should not be used for
1435 optimization purposes. The caller is expected to cast the thunk prototype to
1436 match the thunk target prototype.
1438 This attribute indicates that the ABI being targeted requires that
1439 an unwind table entry be produced for this function even if we can
1440 show that no exceptions passes by it. This is normally the case for
1441 the ELF x86-64 abi, but it can be disabled for some compilation
1450 Note: operand bundles are a work in progress, and they should be
1451 considered experimental at this time.
1453 Operand bundles are tagged sets of SSA values that can be associated
1454 with certain LLVM instructions (currently only ``call`` s and
1455 ``invoke`` s). In a way they are like metadata, but dropping them is
1456 incorrect and will change program semantics.
1460 operand bundle set ::= '[' operand bundle ']'
1461 operand bundle ::= tag '(' [ bundle operand ] (, bundle operand )* ')'
1462 bundle operand ::= SSA value
1463 tag ::= string constant
1465 Operand bundles are **not** part of a function's signature, and a
1466 given function may be called from multiple places with different kinds
1467 of operand bundles. This reflects the fact that the operand bundles
1468 are conceptually a part of the ``call`` (or ``invoke``), not the
1469 callee being dispatched to.
1471 Operand bundles are a generic mechanism intended to support
1472 runtime-introspection-like functionality for managed languages. While
1473 the exact semantics of an operand bundle depend on the bundle tag,
1474 there are certain limitations to how much the presence of an operand
1475 bundle can influence the semantics of a program. These restrictions
1476 are described as the semantics of an "unknown" operand bundle. As
1477 long as the behavior of an operand bundle is describable within these
1478 restrictions, LLVM does not need to have special knowledge of the
1479 operand bundle to not miscompile programs containing it.
1481 - The bundle operands for an unknown operand bundle escape in unknown
1482 ways before control is transferred to the callee or invokee.
1483 - Calls and invokes with operand bundles have unknown read / write
1484 effect on the heap on entry and exit (even if the call target is
1485 ``readnone`` or ``readonly``), unless they're overriden with
1486 callsite specific attributes.
1487 - An operand bundle at a call site cannot change the implementation
1488 of the called function. Inter-procedural optimizations work as
1489 usual as long as they take into account the first two properties.
1491 More specific types of operand bundles are described below.
1493 Deoptimization Operand Bundles
1494 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1496 Deoptimization operand bundles are characterized by the ``"deopt``
1497 operand bundle tag. These operand bundles represent an alternate
1498 "safe" continuation for the call site they're attached to, and can be
1499 used by a suitable runtime to deoptimize the compiled frame at the
1500 specified call site. Exact details of deoptimization is out of scope
1501 for the language reference, but it usually involves rewriting a
1502 compiled frame into a set of interpreted frames.
1504 From the compiler's perspective, deoptimization operand bundles make
1505 the call sites they're attached to at least ``readonly``. They read
1506 through all of their pointer typed operands (even if they're not
1507 otherwise escaped) and the entire visible heap. Deoptimization
1508 operand bundles do not capture their operands except during
1509 deoptimization, in which case control will not be returned to the
1514 Module-Level Inline Assembly
1515 ----------------------------
1517 Modules may contain "module-level inline asm" blocks, which corresponds
1518 to the GCC "file scope inline asm" blocks. These blocks are internally
1519 concatenated by LLVM and treated as a single unit, but may be separated
1520 in the ``.ll`` file if desired. The syntax is very simple:
1522 .. code-block:: llvm
1524 module asm "inline asm code goes here"
1525 module asm "more can go here"
1527 The strings can contain any character by escaping non-printable
1528 characters. The escape sequence used is simply "\\xx" where "xx" is the
1529 two digit hex code for the number.
1531 Note that the assembly string *must* be parseable by LLVM's integrated assembler
1532 (unless it is disabled), even when emitting a ``.s`` file.
1534 .. _langref_datalayout:
1539 A module may specify a target specific data layout string that specifies
1540 how data is to be laid out in memory. The syntax for the data layout is
1543 .. code-block:: llvm
1545 target datalayout = "layout specification"
1547 The *layout specification* consists of a list of specifications
1548 separated by the minus sign character ('-'). Each specification starts
1549 with a letter and may include other information after the letter to
1550 define some aspect of the data layout. The specifications accepted are
1554 Specifies that the target lays out data in big-endian form. That is,
1555 the bits with the most significance have the lowest address
1558 Specifies that the target lays out data in little-endian form. That
1559 is, the bits with the least significance have the lowest address
1562 Specifies the natural alignment of the stack in bits. Alignment
1563 promotion of stack variables is limited to the natural stack
1564 alignment to avoid dynamic stack realignment. The stack alignment
1565 must be a multiple of 8-bits. If omitted, the natural stack
1566 alignment defaults to "unspecified", which does not prevent any
1567 alignment promotions.
1568 ``p[n]:<size>:<abi>:<pref>``
1569 This specifies the *size* of a pointer and its ``<abi>`` and
1570 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1571 bits. The address space, ``n``, is optional, and if not specified,
1572 denotes the default address space 0. The value of ``n`` must be
1573 in the range [1,2^23).
1574 ``i<size>:<abi>:<pref>``
1575 This specifies the alignment for an integer type of a given bit
1576 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1577 ``v<size>:<abi>:<pref>``
1578 This specifies the alignment for a vector type of a given bit
1580 ``f<size>:<abi>:<pref>``
1581 This specifies the alignment for a floating point type of a given bit
1582 ``<size>``. Only values of ``<size>`` that are supported by the target
1583 will work. 32 (float) and 64 (double) are supported on all targets; 80
1584 or 128 (different flavors of long double) are also supported on some
1587 This specifies the alignment for an object of aggregate type.
1589 If present, specifies that llvm names are mangled in the output. The
1592 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1593 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1594 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1595 symbols get a ``_`` prefix.
1596 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1597 functions also get a suffix based on the frame size.
1598 * ``x``: Windows x86 COFF prefix: Similar to Windows COFF, but use a ``_``
1599 prefix for ``__cdecl`` functions.
1600 ``n<size1>:<size2>:<size3>...``
1601 This specifies a set of native integer widths for the target CPU in
1602 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1603 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1604 this set are considered to support most general arithmetic operations
1607 On every specification that takes a ``<abi>:<pref>``, specifying the
1608 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1609 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1611 When constructing the data layout for a given target, LLVM starts with a
1612 default set of specifications which are then (possibly) overridden by
1613 the specifications in the ``datalayout`` keyword. The default
1614 specifications are given in this list:
1616 - ``E`` - big endian
1617 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1618 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1619 same as the default address space.
1620 - ``S0`` - natural stack alignment is unspecified
1621 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1622 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1623 - ``i16:16:16`` - i16 is 16-bit aligned
1624 - ``i32:32:32`` - i32 is 32-bit aligned
1625 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1626 alignment of 64-bits
1627 - ``f16:16:16`` - half is 16-bit aligned
1628 - ``f32:32:32`` - float is 32-bit aligned
1629 - ``f64:64:64`` - double is 64-bit aligned
1630 - ``f128:128:128`` - quad is 128-bit aligned
1631 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1632 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1633 - ``a:0:64`` - aggregates are 64-bit aligned
1635 When LLVM is determining the alignment for a given type, it uses the
1638 #. If the type sought is an exact match for one of the specifications,
1639 that specification is used.
1640 #. If no match is found, and the type sought is an integer type, then
1641 the smallest integer type that is larger than the bitwidth of the
1642 sought type is used. If none of the specifications are larger than
1643 the bitwidth then the largest integer type is used. For example,
1644 given the default specifications above, the i7 type will use the
1645 alignment of i8 (next largest) while both i65 and i256 will use the
1646 alignment of i64 (largest specified).
1647 #. If no match is found, and the type sought is a vector type, then the
1648 largest vector type that is smaller than the sought vector type will
1649 be used as a fall back. This happens because <128 x double> can be
1650 implemented in terms of 64 <2 x double>, for example.
1652 The function of the data layout string may not be what you expect.
1653 Notably, this is not a specification from the frontend of what alignment
1654 the code generator should use.
1656 Instead, if specified, the target data layout is required to match what
1657 the ultimate *code generator* expects. This string is used by the
1658 mid-level optimizers to improve code, and this only works if it matches
1659 what the ultimate code generator uses. There is no way to generate IR
1660 that does not embed this target-specific detail into the IR. If you
1661 don't specify the string, the default specifications will be used to
1662 generate a Data Layout and the optimization phases will operate
1663 accordingly and introduce target specificity into the IR with respect to
1664 these default specifications.
1671 A module may specify a target triple string that describes the target
1672 host. The syntax for the target triple is simply:
1674 .. code-block:: llvm
1676 target triple = "x86_64-apple-macosx10.7.0"
1678 The *target triple* string consists of a series of identifiers delimited
1679 by the minus sign character ('-'). The canonical forms are:
1683 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1684 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1686 This information is passed along to the backend so that it generates
1687 code for the proper architecture. It's possible to override this on the
1688 command line with the ``-mtriple`` command line option.
1690 .. _pointeraliasing:
1692 Pointer Aliasing Rules
1693 ----------------------
1695 Any memory access must be done through a pointer value associated with
1696 an address range of the memory access, otherwise the behavior is
1697 undefined. Pointer values are associated with address ranges according
1698 to the following rules:
1700 - A pointer value is associated with the addresses associated with any
1701 value it is *based* on.
1702 - An address of a global variable is associated with the address range
1703 of the variable's storage.
1704 - The result value of an allocation instruction is associated with the
1705 address range of the allocated storage.
1706 - A null pointer in the default address-space is associated with no
1708 - An integer constant other than zero or a pointer value returned from
1709 a function not defined within LLVM may be associated with address
1710 ranges allocated through mechanisms other than those provided by
1711 LLVM. Such ranges shall not overlap with any ranges of addresses
1712 allocated by mechanisms provided by LLVM.
1714 A pointer value is *based* on another pointer value according to the
1717 - A pointer value formed from a ``getelementptr`` operation is *based*
1718 on the first value operand of the ``getelementptr``.
1719 - The result value of a ``bitcast`` is *based* on the operand of the
1721 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1722 values that contribute (directly or indirectly) to the computation of
1723 the pointer's value.
1724 - The "*based* on" relationship is transitive.
1726 Note that this definition of *"based"* is intentionally similar to the
1727 definition of *"based"* in C99, though it is slightly weaker.
1729 LLVM IR does not associate types with memory. The result type of a
1730 ``load`` merely indicates the size and alignment of the memory from
1731 which to load, as well as the interpretation of the value. The first
1732 operand type of a ``store`` similarly only indicates the size and
1733 alignment of the store.
1735 Consequently, type-based alias analysis, aka TBAA, aka
1736 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1737 :ref:`Metadata <metadata>` may be used to encode additional information
1738 which specialized optimization passes may use to implement type-based
1743 Volatile Memory Accesses
1744 ------------------------
1746 Certain memory accesses, such as :ref:`load <i_load>`'s,
1747 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1748 marked ``volatile``. The optimizers must not change the number of
1749 volatile operations or change their order of execution relative to other
1750 volatile operations. The optimizers *may* change the order of volatile
1751 operations relative to non-volatile operations. This is not Java's
1752 "volatile" and has no cross-thread synchronization behavior.
1754 IR-level volatile loads and stores cannot safely be optimized into
1755 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1756 flagged volatile. Likewise, the backend should never split or merge
1757 target-legal volatile load/store instructions.
1759 .. admonition:: Rationale
1761 Platforms may rely on volatile loads and stores of natively supported
1762 data width to be executed as single instruction. For example, in C
1763 this holds for an l-value of volatile primitive type with native
1764 hardware support, but not necessarily for aggregate types. The
1765 frontend upholds these expectations, which are intentionally
1766 unspecified in the IR. The rules above ensure that IR transformations
1767 do not violate the frontend's contract with the language.
1771 Memory Model for Concurrent Operations
1772 --------------------------------------
1774 The LLVM IR does not define any way to start parallel threads of
1775 execution or to register signal handlers. Nonetheless, there are
1776 platform-specific ways to create them, and we define LLVM IR's behavior
1777 in their presence. This model is inspired by the C++0x memory model.
1779 For a more informal introduction to this model, see the :doc:`Atomics`.
1781 We define a *happens-before* partial order as the least partial order
1784 - Is a superset of single-thread program order, and
1785 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1786 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1787 techniques, like pthread locks, thread creation, thread joining,
1788 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1789 Constraints <ordering>`).
1791 Note that program order does not introduce *happens-before* edges
1792 between a thread and signals executing inside that thread.
1794 Every (defined) read operation (load instructions, memcpy, atomic
1795 loads/read-modify-writes, etc.) R reads a series of bytes written by
1796 (defined) write operations (store instructions, atomic
1797 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1798 section, initialized globals are considered to have a write of the
1799 initializer which is atomic and happens before any other read or write
1800 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1801 may see any write to the same byte, except:
1803 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1804 write\ :sub:`2` happens before R\ :sub:`byte`, then
1805 R\ :sub:`byte` does not see write\ :sub:`1`.
1806 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1807 R\ :sub:`byte` does not see write\ :sub:`3`.
1809 Given that definition, R\ :sub:`byte` is defined as follows:
1811 - If R is volatile, the result is target-dependent. (Volatile is
1812 supposed to give guarantees which can support ``sig_atomic_t`` in
1813 C/C++, and may be used for accesses to addresses that do not behave
1814 like normal memory. It does not generally provide cross-thread
1816 - Otherwise, if there is no write to the same byte that happens before
1817 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1818 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1819 R\ :sub:`byte` returns the value written by that write.
1820 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1821 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1822 Memory Ordering Constraints <ordering>` section for additional
1823 constraints on how the choice is made.
1824 - Otherwise R\ :sub:`byte` returns ``undef``.
1826 R returns the value composed of the series of bytes it read. This
1827 implies that some bytes within the value may be ``undef`` **without**
1828 the entire value being ``undef``. Note that this only defines the
1829 semantics of the operation; it doesn't mean that targets will emit more
1830 than one instruction to read the series of bytes.
1832 Note that in cases where none of the atomic intrinsics are used, this
1833 model places only one restriction on IR transformations on top of what
1834 is required for single-threaded execution: introducing a store to a byte
1835 which might not otherwise be stored is not allowed in general.
1836 (Specifically, in the case where another thread might write to and read
1837 from an address, introducing a store can change a load that may see
1838 exactly one write into a load that may see multiple writes.)
1842 Atomic Memory Ordering Constraints
1843 ----------------------------------
1845 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1846 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1847 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1848 ordering parameters that determine which other atomic instructions on
1849 the same address they *synchronize with*. These semantics are borrowed
1850 from Java and C++0x, but are somewhat more colloquial. If these
1851 descriptions aren't precise enough, check those specs (see spec
1852 references in the :doc:`atomics guide <Atomics>`).
1853 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1854 differently since they don't take an address. See that instruction's
1855 documentation for details.
1857 For a simpler introduction to the ordering constraints, see the
1861 The set of values that can be read is governed by the happens-before
1862 partial order. A value cannot be read unless some operation wrote
1863 it. This is intended to provide a guarantee strong enough to model
1864 Java's non-volatile shared variables. This ordering cannot be
1865 specified for read-modify-write operations; it is not strong enough
1866 to make them atomic in any interesting way.
1868 In addition to the guarantees of ``unordered``, there is a single
1869 total order for modifications by ``monotonic`` operations on each
1870 address. All modification orders must be compatible with the
1871 happens-before order. There is no guarantee that the modification
1872 orders can be combined to a global total order for the whole program
1873 (and this often will not be possible). The read in an atomic
1874 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1875 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1876 order immediately before the value it writes. If one atomic read
1877 happens before another atomic read of the same address, the later
1878 read must see the same value or a later value in the address's
1879 modification order. This disallows reordering of ``monotonic`` (or
1880 stronger) operations on the same address. If an address is written
1881 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1882 read that address repeatedly, the other threads must eventually see
1883 the write. This corresponds to the C++0x/C1x
1884 ``memory_order_relaxed``.
1886 In addition to the guarantees of ``monotonic``, a
1887 *synchronizes-with* edge may be formed with a ``release`` operation.
1888 This is intended to model C++'s ``memory_order_acquire``.
1890 In addition to the guarantees of ``monotonic``, if this operation
1891 writes a value which is subsequently read by an ``acquire``
1892 operation, it *synchronizes-with* that operation. (This isn't a
1893 complete description; see the C++0x definition of a release
1894 sequence.) This corresponds to the C++0x/C1x
1895 ``memory_order_release``.
1896 ``acq_rel`` (acquire+release)
1897 Acts as both an ``acquire`` and ``release`` operation on its
1898 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1899 ``seq_cst`` (sequentially consistent)
1900 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1901 operation that only reads, ``release`` for an operation that only
1902 writes), there is a global total order on all
1903 sequentially-consistent operations on all addresses, which is
1904 consistent with the *happens-before* partial order and with the
1905 modification orders of all the affected addresses. Each
1906 sequentially-consistent read sees the last preceding write to the
1907 same address in this global order. This corresponds to the C++0x/C1x
1908 ``memory_order_seq_cst`` and Java volatile.
1912 If an atomic operation is marked ``singlethread``, it only *synchronizes
1913 with* or participates in modification and seq\_cst total orderings with
1914 other operations running in the same thread (for example, in signal
1922 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1923 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1924 :ref:`frem <i_frem>`, :ref:`fcmp <i_fcmp>`) have the following flags that can
1925 be set to enable otherwise unsafe floating point operations
1928 No NaNs - Allow optimizations to assume the arguments and result are not
1929 NaN. Such optimizations are required to retain defined behavior over
1930 NaNs, but the value of the result is undefined.
1933 No Infs - Allow optimizations to assume the arguments and result are not
1934 +/-Inf. Such optimizations are required to retain defined behavior over
1935 +/-Inf, but the value of the result is undefined.
1938 No Signed Zeros - Allow optimizations to treat the sign of a zero
1939 argument or result as insignificant.
1942 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1943 argument rather than perform division.
1946 Fast - Allow algebraically equivalent transformations that may
1947 dramatically change results in floating point (e.g. reassociate). This
1948 flag implies all the others.
1952 Use-list Order Directives
1953 -------------------------
1955 Use-list directives encode the in-memory order of each use-list, allowing the
1956 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1957 indexes that are assigned to the referenced value's uses. The referenced
1958 value's use-list is immediately sorted by these indexes.
1960 Use-list directives may appear at function scope or global scope. They are not
1961 instructions, and have no effect on the semantics of the IR. When they're at
1962 function scope, they must appear after the terminator of the final basic block.
1964 If basic blocks have their address taken via ``blockaddress()`` expressions,
1965 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1972 uselistorder <ty> <value>, { <order-indexes> }
1973 uselistorder_bb @function, %block { <order-indexes> }
1979 define void @foo(i32 %arg1, i32 %arg2) {
1981 ; ... instructions ...
1983 ; ... instructions ...
1985 ; At function scope.
1986 uselistorder i32 %arg1, { 1, 0, 2 }
1987 uselistorder label %bb, { 1, 0 }
1991 uselistorder i32* @global, { 1, 2, 0 }
1992 uselistorder i32 7, { 1, 0 }
1993 uselistorder i32 (i32) @bar, { 1, 0 }
1994 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
2001 The LLVM type system is one of the most important features of the
2002 intermediate representation. Being typed enables a number of
2003 optimizations to be performed on the intermediate representation
2004 directly, without having to do extra analyses on the side before the
2005 transformation. A strong type system makes it easier to read the
2006 generated code and enables novel analyses and transformations that are
2007 not feasible to perform on normal three address code representations.
2017 The void type does not represent any value and has no size.
2035 The function type can be thought of as a function signature. It consists of a
2036 return type and a list of formal parameter types. The return type of a function
2037 type is a void type or first class type --- except for :ref:`label <t_label>`
2038 and :ref:`metadata <t_metadata>` types.
2044 <returntype> (<parameter list>)
2046 ...where '``<parameter list>``' is a comma-separated list of type
2047 specifiers. Optionally, the parameter list may include a type ``...``, which
2048 indicates that the function takes a variable number of arguments. Variable
2049 argument functions can access their arguments with the :ref:`variable argument
2050 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
2051 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
2055 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2056 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
2057 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2058 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
2059 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2060 | ``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. |
2061 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2062 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
2063 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2070 The :ref:`first class <t_firstclass>` types are perhaps the most important.
2071 Values of these types are the only ones which can be produced by
2079 These are the types that are valid in registers from CodeGen's perspective.
2088 The integer type is a very simple type that simply specifies an
2089 arbitrary bit width for the integer type desired. Any bit width from 1
2090 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
2098 The number of bits the integer will occupy is specified by the ``N``
2104 +----------------+------------------------------------------------+
2105 | ``i1`` | a single-bit integer. |
2106 +----------------+------------------------------------------------+
2107 | ``i32`` | a 32-bit integer. |
2108 +----------------+------------------------------------------------+
2109 | ``i1942652`` | a really big integer of over 1 million bits. |
2110 +----------------+------------------------------------------------+
2114 Floating Point Types
2115 """"""""""""""""""""
2124 - 16-bit floating point value
2127 - 32-bit floating point value
2130 - 64-bit floating point value
2133 - 128-bit floating point value (112-bit mantissa)
2136 - 80-bit floating point value (X87)
2139 - 128-bit floating point value (two 64-bits)
2146 The x86_mmx type represents a value held in an MMX register on an x86
2147 machine. The operations allowed on it are quite limited: parameters and
2148 return values, load and store, and bitcast. User-specified MMX
2149 instructions are represented as intrinsic or asm calls with arguments
2150 and/or results of this type. There are no arrays, vectors or constants
2167 The pointer type is used to specify memory locations. Pointers are
2168 commonly used to reference objects in memory.
2170 Pointer types may have an optional address space attribute defining the
2171 numbered address space where the pointed-to object resides. The default
2172 address space is number zero. The semantics of non-zero address spaces
2173 are target-specific.
2175 Note that LLVM does not permit pointers to void (``void*``) nor does it
2176 permit pointers to labels (``label*``). Use ``i8*`` instead.
2186 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2187 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2188 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2189 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2190 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2191 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2192 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2201 A vector type is a simple derived type that represents a vector of
2202 elements. Vector types are used when multiple primitive data are
2203 operated in parallel using a single instruction (SIMD). A vector type
2204 requires a size (number of elements) and an underlying primitive data
2205 type. Vector types are considered :ref:`first class <t_firstclass>`.
2211 < <# elements> x <elementtype> >
2213 The number of elements is a constant integer value larger than 0;
2214 elementtype may be any integer, floating point or pointer type. Vectors
2215 of size zero are not allowed.
2219 +-------------------+--------------------------------------------------+
2220 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2221 +-------------------+--------------------------------------------------+
2222 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2223 +-------------------+--------------------------------------------------+
2224 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2225 +-------------------+--------------------------------------------------+
2226 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2227 +-------------------+--------------------------------------------------+
2236 The label type represents code labels.
2251 The token type is used when a value is associated with an instruction
2252 but all uses of the value must not attempt to introspect or obscure it.
2253 As such, it is not appropriate to have a :ref:`phi <i_phi>` or
2254 :ref:`select <i_select>` of type token.
2271 The metadata type represents embedded metadata. No derived types may be
2272 created from metadata except for :ref:`function <t_function>` arguments.
2285 Aggregate Types are a subset of derived types that can contain multiple
2286 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2287 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2297 The array type is a very simple derived type that arranges elements
2298 sequentially in memory. The array type requires a size (number of
2299 elements) and an underlying data type.
2305 [<# elements> x <elementtype>]
2307 The number of elements is a constant integer value; ``elementtype`` may
2308 be any type with a size.
2312 +------------------+--------------------------------------+
2313 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2314 +------------------+--------------------------------------+
2315 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2316 +------------------+--------------------------------------+
2317 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2318 +------------------+--------------------------------------+
2320 Here are some examples of multidimensional arrays:
2322 +-----------------------------+----------------------------------------------------------+
2323 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2324 +-----------------------------+----------------------------------------------------------+
2325 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2326 +-----------------------------+----------------------------------------------------------+
2327 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2328 +-----------------------------+----------------------------------------------------------+
2330 There is no restriction on indexing beyond the end of the array implied
2331 by a static type (though there are restrictions on indexing beyond the
2332 bounds of an allocated object in some cases). This means that
2333 single-dimension 'variable sized array' addressing can be implemented in
2334 LLVM with a zero length array type. An implementation of 'pascal style
2335 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2345 The structure type is used to represent a collection of data members
2346 together in memory. The elements of a structure may be any type that has
2349 Structures in memory are accessed using '``load``' and '``store``' by
2350 getting a pointer to a field with the '``getelementptr``' instruction.
2351 Structures in registers are accessed using the '``extractvalue``' and
2352 '``insertvalue``' instructions.
2354 Structures may optionally be "packed" structures, which indicate that
2355 the alignment of the struct is one byte, and that there is no padding
2356 between the elements. In non-packed structs, padding between field types
2357 is inserted as defined by the DataLayout string in the module, which is
2358 required to match what the underlying code generator expects.
2360 Structures can either be "literal" or "identified". A literal structure
2361 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2362 identified types are always defined at the top level with a name.
2363 Literal types are uniqued by their contents and can never be recursive
2364 or opaque since there is no way to write one. Identified types can be
2365 recursive, can be opaqued, and are never uniqued.
2371 %T1 = type { <type list> } ; Identified normal struct type
2372 %T2 = type <{ <type list> }> ; Identified packed struct type
2376 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2377 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2378 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2379 | ``{ 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``. |
2380 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2381 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2382 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2386 Opaque Structure Types
2387 """"""""""""""""""""""
2391 Opaque structure types are used to represent named structure types that
2392 do not have a body specified. This corresponds (for example) to the C
2393 notion of a forward declared structure.
2404 +--------------+-------------------+
2405 | ``opaque`` | An opaque type. |
2406 +--------------+-------------------+
2413 LLVM has several different basic types of constants. This section
2414 describes them all and their syntax.
2419 **Boolean constants**
2420 The two strings '``true``' and '``false``' are both valid constants
2422 **Integer constants**
2423 Standard integers (such as '4') are constants of the
2424 :ref:`integer <t_integer>` type. Negative numbers may be used with
2426 **Floating point constants**
2427 Floating point constants use standard decimal notation (e.g.
2428 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2429 hexadecimal notation (see below). The assembler requires the exact
2430 decimal value of a floating-point constant. For example, the
2431 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2432 decimal in binary. Floating point constants must have a :ref:`floating
2433 point <t_floating>` type.
2434 **Null pointer constants**
2435 The identifier '``null``' is recognized as a null pointer constant
2436 and must be of :ref:`pointer type <t_pointer>`.
2438 The identifier '``none``' is recognized as an empty token constant
2439 and must be of :ref:`token type <t_token>`.
2441 The one non-intuitive notation for constants is the hexadecimal form of
2442 floating point constants. For example, the form
2443 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2444 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2445 constants are required (and the only time that they are generated by the
2446 disassembler) is when a floating point constant must be emitted but it
2447 cannot be represented as a decimal floating point number in a reasonable
2448 number of digits. For example, NaN's, infinities, and other special
2449 values are represented in their IEEE hexadecimal format so that assembly
2450 and disassembly do not cause any bits to change in the constants.
2452 When using the hexadecimal form, constants of types half, float, and
2453 double are represented using the 16-digit form shown above (which
2454 matches the IEEE754 representation for double); half and float values
2455 must, however, be exactly representable as IEEE 754 half and single
2456 precision, respectively. Hexadecimal format is always used for long
2457 double, and there are three forms of long double. The 80-bit format used
2458 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2459 128-bit format used by PowerPC (two adjacent doubles) is represented by
2460 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2461 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2462 will only work if they match the long double format on your target.
2463 The IEEE 16-bit format (half precision) is represented by ``0xH``
2464 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2465 (sign bit at the left).
2467 There are no constants of type x86_mmx.
2469 .. _complexconstants:
2474 Complex constants are a (potentially recursive) combination of simple
2475 constants and smaller complex constants.
2477 **Structure constants**
2478 Structure constants are represented with notation similar to
2479 structure type definitions (a comma separated list of elements,
2480 surrounded by braces (``{}``)). For example:
2481 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2482 "``@G = external global i32``". Structure constants must have
2483 :ref:`structure type <t_struct>`, and the number and types of elements
2484 must match those specified by the type.
2486 Array constants are represented with notation similar to array type
2487 definitions (a comma separated list of elements, surrounded by
2488 square brackets (``[]``)). For example:
2489 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2490 :ref:`array type <t_array>`, and the number and types of elements must
2491 match those specified by the type. As a special case, character array
2492 constants may also be represented as a double-quoted string using the ``c``
2493 prefix. For example: "``c"Hello World\0A\00"``".
2494 **Vector constants**
2495 Vector constants are represented with notation similar to vector
2496 type definitions (a comma separated list of elements, surrounded by
2497 less-than/greater-than's (``<>``)). For example:
2498 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2499 must have :ref:`vector type <t_vector>`, and the number and types of
2500 elements must match those specified by the type.
2501 **Zero initialization**
2502 The string '``zeroinitializer``' can be used to zero initialize a
2503 value to zero of *any* type, including scalar and
2504 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2505 having to print large zero initializers (e.g. for large arrays) and
2506 is always exactly equivalent to using explicit zero initializers.
2508 A metadata node is a constant tuple without types. For example:
2509 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2510 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2511 Unlike other typed constants that are meant to be interpreted as part of
2512 the instruction stream, metadata is a place to attach additional
2513 information such as debug info.
2515 Global Variable and Function Addresses
2516 --------------------------------------
2518 The addresses of :ref:`global variables <globalvars>` and
2519 :ref:`functions <functionstructure>` are always implicitly valid
2520 (link-time) constants. These constants are explicitly referenced when
2521 the :ref:`identifier for the global <identifiers>` is used and always have
2522 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2525 .. code-block:: llvm
2529 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2536 The string '``undef``' can be used anywhere a constant is expected, and
2537 indicates that the user of the value may receive an unspecified
2538 bit-pattern. Undefined values may be of any type (other than '``label``'
2539 or '``void``') and be used anywhere a constant is permitted.
2541 Undefined values are useful because they indicate to the compiler that
2542 the program is well defined no matter what value is used. This gives the
2543 compiler more freedom to optimize. Here are some examples of
2544 (potentially surprising) transformations that are valid (in pseudo IR):
2546 .. code-block:: llvm
2556 This is safe because all of the output bits are affected by the undef
2557 bits. Any output bit can have a zero or one depending on the input bits.
2559 .. code-block:: llvm
2570 These logical operations have bits that are not always affected by the
2571 input. For example, if ``%X`` has a zero bit, then the output of the
2572 '``and``' operation will always be a zero for that bit, no matter what
2573 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2574 optimize or assume that the result of the '``and``' is '``undef``'.
2575 However, it is safe to assume that all bits of the '``undef``' could be
2576 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2577 all the bits of the '``undef``' operand to the '``or``' could be set,
2578 allowing the '``or``' to be folded to -1.
2580 .. code-block:: llvm
2582 %A = select undef, %X, %Y
2583 %B = select undef, 42, %Y
2584 %C = select %X, %Y, undef
2594 This set of examples shows that undefined '``select``' (and conditional
2595 branch) conditions can go *either way*, but they have to come from one
2596 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2597 both known to have a clear low bit, then ``%A`` would have to have a
2598 cleared low bit. However, in the ``%C`` example, the optimizer is
2599 allowed to assume that the '``undef``' operand could be the same as
2600 ``%Y``, allowing the whole '``select``' to be eliminated.
2602 .. code-block:: llvm
2604 %A = xor undef, undef
2621 This example points out that two '``undef``' operands are not
2622 necessarily the same. This can be surprising to people (and also matches
2623 C semantics) where they assume that "``X^X``" is always zero, even if
2624 ``X`` is undefined. This isn't true for a number of reasons, but the
2625 short answer is that an '``undef``' "variable" can arbitrarily change
2626 its value over its "live range". This is true because the variable
2627 doesn't actually *have a live range*. Instead, the value is logically
2628 read from arbitrary registers that happen to be around when needed, so
2629 the value is not necessarily consistent over time. In fact, ``%A`` and
2630 ``%C`` need to have the same semantics or the core LLVM "replace all
2631 uses with" concept would not hold.
2633 .. code-block:: llvm
2641 These examples show the crucial difference between an *undefined value*
2642 and *undefined behavior*. An undefined value (like '``undef``') is
2643 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2644 operation can be constant folded to '``undef``', because the '``undef``'
2645 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2646 However, in the second example, we can make a more aggressive
2647 assumption: because the ``undef`` is allowed to be an arbitrary value,
2648 we are allowed to assume that it could be zero. Since a divide by zero
2649 has *undefined behavior*, we are allowed to assume that the operation
2650 does not execute at all. This allows us to delete the divide and all
2651 code after it. Because the undefined operation "can't happen", the
2652 optimizer can assume that it occurs in dead code.
2654 .. code-block:: llvm
2656 a: store undef -> %X
2657 b: store %X -> undef
2662 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2663 value can be assumed to not have any effect; we can assume that the
2664 value is overwritten with bits that happen to match what was already
2665 there. However, a store *to* an undefined location could clobber
2666 arbitrary memory, therefore, it has undefined behavior.
2673 Poison values are similar to :ref:`undef values <undefvalues>`, however
2674 they also represent the fact that an instruction or constant expression
2675 that cannot evoke side effects has nevertheless detected a condition
2676 that results in undefined behavior.
2678 There is currently no way of representing a poison value in the IR; they
2679 only exist when produced by operations such as :ref:`add <i_add>` with
2682 Poison value behavior is defined in terms of value *dependence*:
2684 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2685 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2686 their dynamic predecessor basic block.
2687 - Function arguments depend on the corresponding actual argument values
2688 in the dynamic callers of their functions.
2689 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2690 instructions that dynamically transfer control back to them.
2691 - :ref:`Invoke <i_invoke>` instructions depend on the
2692 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2693 call instructions that dynamically transfer control back to them.
2694 - Non-volatile loads and stores depend on the most recent stores to all
2695 of the referenced memory addresses, following the order in the IR
2696 (including loads and stores implied by intrinsics such as
2697 :ref:`@llvm.memcpy <int_memcpy>`.)
2698 - An instruction with externally visible side effects depends on the
2699 most recent preceding instruction with externally visible side
2700 effects, following the order in the IR. (This includes :ref:`volatile
2701 operations <volatile>`.)
2702 - An instruction *control-depends* on a :ref:`terminator
2703 instruction <terminators>` if the terminator instruction has
2704 multiple successors and the instruction is always executed when
2705 control transfers to one of the successors, and may not be executed
2706 when control is transferred to another.
2707 - Additionally, an instruction also *control-depends* on a terminator
2708 instruction if the set of instructions it otherwise depends on would
2709 be different if the terminator had transferred control to a different
2711 - Dependence is transitive.
2713 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2714 with the additional effect that any instruction that has a *dependence*
2715 on a poison value has undefined behavior.
2717 Here are some examples:
2719 .. code-block:: llvm
2722 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2723 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2724 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
2725 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2727 store i32 %poison, i32* @g ; Poison value stored to memory.
2728 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
2730 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2732 %narrowaddr = bitcast i32* @g to i16*
2733 %wideaddr = bitcast i32* @g to i64*
2734 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2735 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
2737 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2738 br i1 %cmp, label %true, label %end ; Branch to either destination.
2741 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2742 ; it has undefined behavior.
2746 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2747 ; Both edges into this PHI are
2748 ; control-dependent on %cmp, so this
2749 ; always results in a poison value.
2751 store volatile i32 0, i32* @g ; This would depend on the store in %true
2752 ; if %cmp is true, or the store in %entry
2753 ; otherwise, so this is undefined behavior.
2755 br i1 %cmp, label %second_true, label %second_end
2756 ; The same branch again, but this time the
2757 ; true block doesn't have side effects.
2764 store volatile i32 0, i32* @g ; This time, the instruction always depends
2765 ; on the store in %end. Also, it is
2766 ; control-equivalent to %end, so this is
2767 ; well-defined (ignoring earlier undefined
2768 ; behavior in this example).
2772 Addresses of Basic Blocks
2773 -------------------------
2775 ``blockaddress(@function, %block)``
2777 The '``blockaddress``' constant computes the address of the specified
2778 basic block in the specified function, and always has an ``i8*`` type.
2779 Taking the address of the entry block is illegal.
2781 This value only has defined behavior when used as an operand to the
2782 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2783 against null. Pointer equality tests between labels addresses results in
2784 undefined behavior --- though, again, comparison against null is ok, and
2785 no label is equal to the null pointer. This may be passed around as an
2786 opaque pointer sized value as long as the bits are not inspected. This
2787 allows ``ptrtoint`` and arithmetic to be performed on these values so
2788 long as the original value is reconstituted before the ``indirectbr``
2791 Finally, some targets may provide defined semantics when using the value
2792 as the operand to an inline assembly, but that is target specific.
2796 Constant Expressions
2797 --------------------
2799 Constant expressions are used to allow expressions involving other
2800 constants to be used as constants. Constant expressions may be of any
2801 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2802 that does not have side effects (e.g. load and call are not supported).
2803 The following is the syntax for constant expressions:
2805 ``trunc (CST to TYPE)``
2806 Truncate a constant to another type. The bit size of CST must be
2807 larger than the bit size of TYPE. Both types must be integers.
2808 ``zext (CST to TYPE)``
2809 Zero extend a constant to another type. The bit size of CST must be
2810 smaller than the bit size of TYPE. Both types must be integers.
2811 ``sext (CST to TYPE)``
2812 Sign extend a constant to another type. The bit size of CST must be
2813 smaller than the bit size of TYPE. Both types must be integers.
2814 ``fptrunc (CST to TYPE)``
2815 Truncate a floating point constant to another floating point type.
2816 The size of CST must be larger than the size of TYPE. Both types
2817 must be floating point.
2818 ``fpext (CST to TYPE)``
2819 Floating point extend a constant to another type. The size of CST
2820 must be smaller or equal to the size of TYPE. Both types must be
2822 ``fptoui (CST to TYPE)``
2823 Convert a floating point constant to the corresponding unsigned
2824 integer constant. TYPE must be a scalar or vector integer type. CST
2825 must be of scalar or vector floating point type. Both CST and TYPE
2826 must be scalars, or vectors of the same number of elements. If the
2827 value won't fit in the integer type, the results are undefined.
2828 ``fptosi (CST to TYPE)``
2829 Convert a floating point constant to the corresponding signed
2830 integer constant. TYPE must be a scalar or vector integer type. CST
2831 must be of scalar or vector floating point type. Both CST and TYPE
2832 must be scalars, or vectors of the same number of elements. If the
2833 value won't fit in the integer type, the results are undefined.
2834 ``uitofp (CST to TYPE)``
2835 Convert an unsigned integer constant to the corresponding floating
2836 point constant. TYPE must be a scalar or vector floating point type.
2837 CST must be of scalar or vector integer type. Both CST and TYPE must
2838 be scalars, or vectors of the same number of elements. If the value
2839 won't fit in the floating point type, the results are undefined.
2840 ``sitofp (CST to TYPE)``
2841 Convert a signed integer constant to the corresponding floating
2842 point constant. TYPE must be a scalar or vector floating point type.
2843 CST must be of scalar or vector integer type. Both CST and TYPE must
2844 be scalars, or vectors of the same number of elements. If the value
2845 won't fit in the floating point type, the results are undefined.
2846 ``ptrtoint (CST to TYPE)``
2847 Convert a pointer typed constant to the corresponding integer
2848 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2849 pointer type. The ``CST`` value is zero extended, truncated, or
2850 unchanged to make it fit in ``TYPE``.
2851 ``inttoptr (CST to TYPE)``
2852 Convert an integer constant to a pointer constant. TYPE must be a
2853 pointer type. CST must be of integer type. The CST value is zero
2854 extended, truncated, or unchanged to make it fit in a pointer size.
2855 This one is *really* dangerous!
2856 ``bitcast (CST to TYPE)``
2857 Convert a constant, CST, to another TYPE. The constraints of the
2858 operands are the same as those for the :ref:`bitcast
2859 instruction <i_bitcast>`.
2860 ``addrspacecast (CST to TYPE)``
2861 Convert a constant pointer or constant vector of pointer, CST, to another
2862 TYPE in a different address space. The constraints of the operands are the
2863 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2864 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
2865 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2866 constants. As with the :ref:`getelementptr <i_getelementptr>`
2867 instruction, the index list may have zero or more indexes, which are
2868 required to make sense for the type of "pointer to TY".
2869 ``select (COND, VAL1, VAL2)``
2870 Perform the :ref:`select operation <i_select>` on constants.
2871 ``icmp COND (VAL1, VAL2)``
2872 Performs the :ref:`icmp operation <i_icmp>` on constants.
2873 ``fcmp COND (VAL1, VAL2)``
2874 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2875 ``extractelement (VAL, IDX)``
2876 Perform the :ref:`extractelement operation <i_extractelement>` on
2878 ``insertelement (VAL, ELT, IDX)``
2879 Perform the :ref:`insertelement operation <i_insertelement>` on
2881 ``shufflevector (VEC1, VEC2, IDXMASK)``
2882 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2884 ``extractvalue (VAL, IDX0, IDX1, ...)``
2885 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2886 constants. The index list is interpreted in a similar manner as
2887 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2888 least one index value must be specified.
2889 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2890 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2891 The index list is interpreted in a similar manner as indices in a
2892 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2893 value must be specified.
2894 ``OPCODE (LHS, RHS)``
2895 Perform the specified operation of the LHS and RHS constants. OPCODE
2896 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2897 binary <bitwiseops>` operations. The constraints on operands are
2898 the same as those for the corresponding instruction (e.g. no bitwise
2899 operations on floating point values are allowed).
2906 Inline Assembler Expressions
2907 ----------------------------
2909 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2910 Inline Assembly <moduleasm>`) through the use of a special value. This value
2911 represents the inline assembler as a template string (containing the
2912 instructions to emit), a list of operand constraints (stored as a string), a
2913 flag that indicates whether or not the inline asm expression has side effects,
2914 and a flag indicating whether the function containing the asm needs to align its
2915 stack conservatively.
2917 The template string supports argument substitution of the operands using "``$``"
2918 followed by a number, to indicate substitution of the given register/memory
2919 location, as specified by the constraint string. "``${NUM:MODIFIER}``" may also
2920 be used, where ``MODIFIER`` is a target-specific annotation for how to print the
2921 operand (See :ref:`inline-asm-modifiers`).
2923 A literal "``$``" may be included by using "``$$``" in the template. To include
2924 other special characters into the output, the usual "``\XX``" escapes may be
2925 used, just as in other strings. Note that after template substitution, the
2926 resulting assembly string is parsed by LLVM's integrated assembler unless it is
2927 disabled -- even when emitting a ``.s`` file -- and thus must contain assembly
2928 syntax known to LLVM.
2930 LLVM's support for inline asm is modeled closely on the requirements of Clang's
2931 GCC-compatible inline-asm support. Thus, the feature-set and the constraint and
2932 modifier codes listed here are similar or identical to those in GCC's inline asm
2933 support. However, to be clear, the syntax of the template and constraint strings
2934 described here is *not* the same as the syntax accepted by GCC and Clang, and,
2935 while most constraint letters are passed through as-is by Clang, some get
2936 translated to other codes when converting from the C source to the LLVM
2939 An example inline assembler expression is:
2941 .. code-block:: llvm
2943 i32 (i32) asm "bswap $0", "=r,r"
2945 Inline assembler expressions may **only** be used as the callee operand
2946 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2947 Thus, typically we have:
2949 .. code-block:: llvm
2951 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2953 Inline asms with side effects not visible in the constraint list must be
2954 marked as having side effects. This is done through the use of the
2955 '``sideeffect``' keyword, like so:
2957 .. code-block:: llvm
2959 call void asm sideeffect "eieio", ""()
2961 In some cases inline asms will contain code that will not work unless
2962 the stack is aligned in some way, such as calls or SSE instructions on
2963 x86, yet will not contain code that does that alignment within the asm.
2964 The compiler should make conservative assumptions about what the asm
2965 might contain and should generate its usual stack alignment code in the
2966 prologue if the '``alignstack``' keyword is present:
2968 .. code-block:: llvm
2970 call void asm alignstack "eieio", ""()
2972 Inline asms also support using non-standard assembly dialects. The
2973 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2974 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2975 the only supported dialects. An example is:
2977 .. code-block:: llvm
2979 call void asm inteldialect "eieio", ""()
2981 If multiple keywords appear the '``sideeffect``' keyword must come
2982 first, the '``alignstack``' keyword second and the '``inteldialect``'
2985 Inline Asm Constraint String
2986 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2988 The constraint list is a comma-separated string, each element containing one or
2989 more constraint codes.
2991 For each element in the constraint list an appropriate register or memory
2992 operand will be chosen, and it will be made available to assembly template
2993 string expansion as ``$0`` for the first constraint in the list, ``$1`` for the
2996 There are three different types of constraints, which are distinguished by a
2997 prefix symbol in front of the constraint code: Output, Input, and Clobber. The
2998 constraints must always be given in that order: outputs first, then inputs, then
2999 clobbers. They cannot be intermingled.
3001 There are also three different categories of constraint codes:
3003 - Register constraint. This is either a register class, or a fixed physical
3004 register. This kind of constraint will allocate a register, and if necessary,
3005 bitcast the argument or result to the appropriate type.
3006 - Memory constraint. This kind of constraint is for use with an instruction
3007 taking a memory operand. Different constraints allow for different addressing
3008 modes used by the target.
3009 - Immediate value constraint. This kind of constraint is for an integer or other
3010 immediate value which can be rendered directly into an instruction. The
3011 various target-specific constraints allow the selection of a value in the
3012 proper range for the instruction you wish to use it with.
3017 Output constraints are specified by an "``=``" prefix (e.g. "``=r``"). This
3018 indicates that the assembly will write to this operand, and the operand will
3019 then be made available as a return value of the ``asm`` expression. Output
3020 constraints do not consume an argument from the call instruction. (Except, see
3021 below about indirect outputs).
3023 Normally, it is expected that no output locations are written to by the assembly
3024 expression until *all* of the inputs have been read. As such, LLVM may assign
3025 the same register to an output and an input. If this is not safe (e.g. if the
3026 assembly contains two instructions, where the first writes to one output, and
3027 the second reads an input and writes to a second output), then the "``&``"
3028 modifier must be used (e.g. "``=&r``") to specify that the output is an
3029 "early-clobber" output. Marking an ouput as "early-clobber" ensures that LLVM
3030 will not use the same register for any inputs (other than an input tied to this
3036 Input constraints do not have a prefix -- just the constraint codes. Each input
3037 constraint will consume one argument from the call instruction. It is not
3038 permitted for the asm to write to any input register or memory location (unless
3039 that input is tied to an output). Note also that multiple inputs may all be
3040 assigned to the same register, if LLVM can determine that they necessarily all
3041 contain the same value.
3043 Instead of providing a Constraint Code, input constraints may also "tie"
3044 themselves to an output constraint, by providing an integer as the constraint
3045 string. Tied inputs still consume an argument from the call instruction, and
3046 take up a position in the asm template numbering as is usual -- they will simply
3047 be constrained to always use the same register as the output they've been tied
3048 to. For example, a constraint string of "``=r,0``" says to assign a register for
3049 output, and use that register as an input as well (it being the 0'th
3052 It is permitted to tie an input to an "early-clobber" output. In that case, no
3053 *other* input may share the same register as the input tied to the early-clobber
3054 (even when the other input has the same value).
3056 You may only tie an input to an output which has a register constraint, not a
3057 memory constraint. Only a single input may be tied to an output.
3059 There is also an "interesting" feature which deserves a bit of explanation: if a
3060 register class constraint allocates a register which is too small for the value
3061 type operand provided as input, the input value will be split into multiple
3062 registers, and all of them passed to the inline asm.
3064 However, this feature is often not as useful as you might think.
3066 Firstly, the registers are *not* guaranteed to be consecutive. So, on those
3067 architectures that have instructions which operate on multiple consecutive
3068 instructions, this is not an appropriate way to support them. (e.g. the 32-bit
3069 SparcV8 has a 64-bit load, which instruction takes a single 32-bit register. The
3070 hardware then loads into both the named register, and the next register. This
3071 feature of inline asm would not be useful to support that.)
3073 A few of the targets provide a template string modifier allowing explicit access
3074 to the second register of a two-register operand (e.g. MIPS ``L``, ``M``, and
3075 ``D``). On such an architecture, you can actually access the second allocated
3076 register (yet, still, not any subsequent ones). But, in that case, you're still
3077 probably better off simply splitting the value into two separate operands, for
3078 clarity. (e.g. see the description of the ``A`` constraint on X86, which,
3079 despite existing only for use with this feature, is not really a good idea to
3082 Indirect inputs and outputs
3083 """""""""""""""""""""""""""
3085 Indirect output or input constraints can be specified by the "``*``" modifier
3086 (which goes after the "``=``" in case of an output). This indicates that the asm
3087 will write to or read from the contents of an *address* provided as an input
3088 argument. (Note that in this way, indirect outputs act more like an *input* than
3089 an output: just like an input, they consume an argument of the call expression,
3090 rather than producing a return value. An indirect output constraint is an
3091 "output" only in that the asm is expected to write to the contents of the input
3092 memory location, instead of just read from it).
3094 This is most typically used for memory constraint, e.g. "``=*m``", to pass the
3095 address of a variable as a value.
3097 It is also possible to use an indirect *register* constraint, but only on output
3098 (e.g. "``=*r``"). This will cause LLVM to allocate a register for an output
3099 value normally, and then, separately emit a store to the address provided as
3100 input, after the provided inline asm. (It's not clear what value this
3101 functionality provides, compared to writing the store explicitly after the asm
3102 statement, and it can only produce worse code, since it bypasses many
3103 optimization passes. I would recommend not using it.)
3109 A clobber constraint is indicated by a "``~``" prefix. A clobber does not
3110 consume an input operand, nor generate an output. Clobbers cannot use any of the
3111 general constraint code letters -- they may use only explicit register
3112 constraints, e.g. "``~{eax}``". The one exception is that a clobber string of
3113 "``~{memory}``" indicates that the assembly writes to arbitrary undeclared
3114 memory locations -- not only the memory pointed to by a declared indirect
3120 After a potential prefix comes constraint code, or codes.
3122 A Constraint Code is either a single letter (e.g. "``r``"), a "``^``" character
3123 followed by two letters (e.g. "``^wc``"), or "``{``" register-name "``}``"
3126 The one and two letter constraint codes are typically chosen to be the same as
3127 GCC's constraint codes.
3129 A single constraint may include one or more than constraint code in it, leaving
3130 it up to LLVM to choose which one to use. This is included mainly for
3131 compatibility with the translation of GCC inline asm coming from clang.
3133 There are two ways to specify alternatives, and either or both may be used in an
3134 inline asm constraint list:
3136 1) Append the codes to each other, making a constraint code set. E.g. "``im``"
3137 or "``{eax}m``". This means "choose any of the options in the set". The
3138 choice of constraint is made independently for each constraint in the
3141 2) Use "``|``" between constraint code sets, creating alternatives. Every
3142 constraint in the constraint list must have the same number of alternative
3143 sets. With this syntax, the same alternative in *all* of the items in the
3144 constraint list will be chosen together.
3146 Putting those together, you might have a two operand constraint string like
3147 ``"rm|r,ri|rm"``. This indicates that if operand 0 is ``r`` or ``m``, then
3148 operand 1 may be one of ``r`` or ``i``. If operand 0 is ``r``, then operand 1
3149 may be one of ``r`` or ``m``. But, operand 0 and 1 cannot both be of type m.
3151 However, the use of either of the alternatives features is *NOT* recommended, as
3152 LLVM is not able to make an intelligent choice about which one to use. (At the
3153 point it currently needs to choose, not enough information is available to do so
3154 in a smart way.) Thus, it simply tries to make a choice that's most likely to
3155 compile, not one that will be optimal performance. (e.g., given "``rm``", it'll
3156 always choose to use memory, not registers). And, if given multiple registers,
3157 or multiple register classes, it will simply choose the first one. (In fact, it
3158 doesn't currently even ensure explicitly specified physical registers are
3159 unique, so specifying multiple physical registers as alternatives, like
3160 ``{r11}{r12},{r11}{r12}``, will assign r11 to both operands, not at all what was
3163 Supported Constraint Code List
3164 """"""""""""""""""""""""""""""
3166 The constraint codes are, in general, expected to behave the same way they do in
3167 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3168 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3169 and GCC likely indicates a bug in LLVM.
3171 Some constraint codes are typically supported by all targets:
3173 - ``r``: A register in the target's general purpose register class.
3174 - ``m``: A memory address operand. It is target-specific what addressing modes
3175 are supported, typical examples are register, or register + register offset,
3176 or register + immediate offset (of some target-specific size).
3177 - ``i``: An integer constant (of target-specific width). Allows either a simple
3178 immediate, or a relocatable value.
3179 - ``n``: An integer constant -- *not* including relocatable values.
3180 - ``s``: An integer constant, but allowing *only* relocatable values.
3181 - ``X``: Allows an operand of any kind, no constraint whatsoever. Typically
3182 useful to pass a label for an asm branch or call.
3184 .. FIXME: but that surely isn't actually okay to jump out of an asm
3185 block without telling llvm about the control transfer???)
3187 - ``{register-name}``: Requires exactly the named physical register.
3189 Other constraints are target-specific:
3193 - ``z``: An immediate integer 0. Outputs ``WZR`` or ``XZR``, as appropriate.
3194 - ``I``: An immediate integer valid for an ``ADD`` or ``SUB`` instruction,
3195 i.e. 0 to 4095 with optional shift by 12.
3196 - ``J``: An immediate integer that, when negated, is valid for an ``ADD`` or
3197 ``SUB`` instruction, i.e. -1 to -4095 with optional left shift by 12.
3198 - ``K``: An immediate integer that is valid for the 'bitmask immediate 32' of a
3199 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 32-bit register.
3200 - ``L``: An immediate integer that is valid for the 'bitmask immediate 64' of a
3201 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 64-bit register.
3202 - ``M``: An immediate integer for use with the ``MOV`` assembly alias on a
3203 32-bit register. This is a superset of ``K``: in addition to the bitmask
3204 immediate, also allows immediate integers which can be loaded with a single
3205 ``MOVZ`` or ``MOVL`` instruction.
3206 - ``N``: An immediate integer for use with the ``MOV`` assembly alias on a
3207 64-bit register. This is a superset of ``L``.
3208 - ``Q``: Memory address operand must be in a single register (no
3209 offsets). (However, LLVM currently does this for the ``m`` constraint as
3211 - ``r``: A 32 or 64-bit integer register (W* or X*).
3212 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register.
3213 - ``x``: A lower 128-bit floating-point/SIMD register (``V0`` to ``V15``).
3217 - ``r``: A 32 or 64-bit integer register.
3218 - ``[0-9]v``: The 32-bit VGPR register, number 0-9.
3219 - ``[0-9]s``: The 32-bit SGPR register, number 0-9.
3224 - ``Q``, ``Um``, ``Un``, ``Uq``, ``Us``, ``Ut``, ``Uv``, ``Uy``: Memory address
3225 operand. Treated the same as operand ``m``, at the moment.
3227 ARM and ARM's Thumb2 mode:
3229 - ``j``: An immediate integer between 0 and 65535 (valid for ``MOVW``)
3230 - ``I``: An immediate integer valid for a data-processing instruction.
3231 - ``J``: An immediate integer between -4095 and 4095.
3232 - ``K``: An immediate integer whose bitwise inverse is valid for a
3233 data-processing instruction. (Can be used with template modifier "``B``" to
3234 print the inverted value).
3235 - ``L``: An immediate integer whose negation is valid for a data-processing
3236 instruction. (Can be used with template modifier "``n``" to print the negated
3238 - ``M``: A power of two or a integer between 0 and 32.
3239 - ``N``: Invalid immediate constraint.
3240 - ``O``: Invalid immediate constraint.
3241 - ``r``: A general-purpose 32-bit integer register (``r0-r15``).
3242 - ``l``: In Thumb2 mode, low 32-bit GPR registers (``r0-r7``). In ARM mode, same
3244 - ``h``: In Thumb2 mode, a high 32-bit GPR register (``r8-r15``). In ARM mode,
3246 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3247 ``d0-d31``, or ``q0-q15``.
3248 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3249 ``d0-d7``, or ``q0-q3``.
3250 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3255 - ``I``: An immediate integer between 0 and 255.
3256 - ``J``: An immediate integer between -255 and -1.
3257 - ``K``: An immediate integer between 0 and 255, with optional left-shift by
3259 - ``L``: An immediate integer between -7 and 7.
3260 - ``M``: An immediate integer which is a multiple of 4 between 0 and 1020.
3261 - ``N``: An immediate integer between 0 and 31.
3262 - ``O``: An immediate integer which is a multiple of 4 between -508 and 508.
3263 - ``r``: A low 32-bit GPR register (``r0-r7``).
3264 - ``l``: A low 32-bit GPR register (``r0-r7``).
3265 - ``h``: A high GPR register (``r0-r7``).
3266 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3267 ``d0-d31``, or ``q0-q15``.
3268 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3269 ``d0-d7``, or ``q0-q3``.
3270 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3276 - ``o``, ``v``: A memory address operand, treated the same as constraint ``m``,
3278 - ``r``: A 32 or 64-bit register.
3282 - ``r``: An 8 or 16-bit register.
3286 - ``I``: An immediate signed 16-bit integer.
3287 - ``J``: An immediate integer zero.
3288 - ``K``: An immediate unsigned 16-bit integer.
3289 - ``L``: An immediate 32-bit integer, where the lower 16 bits are 0.
3290 - ``N``: An immediate integer between -65535 and -1.
3291 - ``O``: An immediate signed 15-bit integer.
3292 - ``P``: An immediate integer between 1 and 65535.
3293 - ``m``: A memory address operand. In MIPS-SE mode, allows a base address
3294 register plus 16-bit immediate offset. In MIPS mode, just a base register.
3295 - ``R``: A memory address operand. In MIPS-SE mode, allows a base address
3296 register plus a 9-bit signed offset. In MIPS mode, the same as constraint
3298 - ``ZC``: A memory address operand, suitable for use in a ``pref``, ``ll``, or
3299 ``sc`` instruction on the given subtarget (details vary).
3300 - ``r``, ``d``, ``y``: A 32 or 64-bit GPR register.
3301 - ``f``: A 32 or 64-bit FPU register (``F0-F31``), or a 128-bit MSA register
3302 (``W0-W31``). In the case of MSA registers, it is recommended to use the ``w``
3303 argument modifier for compatibility with GCC.
3304 - ``c``: A 32-bit or 64-bit GPR register suitable for indirect jump (always
3306 - ``l``: The ``lo`` register, 32 or 64-bit.
3311 - ``b``: A 1-bit integer register.
3312 - ``c`` or ``h``: A 16-bit integer register.
3313 - ``r``: A 32-bit integer register.
3314 - ``l`` or ``N``: A 64-bit integer register.
3315 - ``f``: A 32-bit float register.
3316 - ``d``: A 64-bit float register.
3321 - ``I``: An immediate signed 16-bit integer.
3322 - ``J``: An immediate unsigned 16-bit integer, shifted left 16 bits.
3323 - ``K``: An immediate unsigned 16-bit integer.
3324 - ``L``: An immediate signed 16-bit integer, shifted left 16 bits.
3325 - ``M``: An immediate integer greater than 31.
3326 - ``N``: An immediate integer that is an exact power of 2.
3327 - ``O``: The immediate integer constant 0.
3328 - ``P``: An immediate integer constant whose negation is a signed 16-bit
3330 - ``es``, ``o``, ``Q``, ``Z``, ``Zy``: A memory address operand, currently
3331 treated the same as ``m``.
3332 - ``r``: A 32 or 64-bit integer register.
3333 - ``b``: A 32 or 64-bit integer register, excluding ``R0`` (that is:
3335 - ``f``: A 32 or 64-bit float register (``F0-F31``), or when QPX is enabled, a
3336 128 or 256-bit QPX register (``Q0-Q31``; aliases the ``F`` registers).
3337 - ``v``: For ``4 x f32`` or ``4 x f64`` types, when QPX is enabled, a
3338 128 or 256-bit QPX register (``Q0-Q31``), otherwise a 128-bit
3339 altivec vector register (``V0-V31``).
3341 .. FIXME: is this a bug that v accepts QPX registers? I think this
3342 is supposed to only use the altivec vector registers?
3344 - ``y``: Condition register (``CR0-CR7``).
3345 - ``wc``: An individual CR bit in a CR register.
3346 - ``wa``, ``wd``, ``wf``: Any 128-bit VSX vector register, from the full VSX
3347 register set (overlapping both the floating-point and vector register files).
3348 - ``ws``: A 32 or 64-bit floating point register, from the full VSX register
3353 - ``I``: An immediate 13-bit signed integer.
3354 - ``r``: A 32-bit integer register.
3358 - ``I``: An immediate unsigned 8-bit integer.
3359 - ``J``: An immediate unsigned 12-bit integer.
3360 - ``K``: An immediate signed 16-bit integer.
3361 - ``L``: An immediate signed 20-bit integer.
3362 - ``M``: An immediate integer 0x7fffffff.
3363 - ``Q``, ``R``, ``S``, ``T``: A memory address operand, treated the same as
3364 ``m``, at the moment.
3365 - ``r`` or ``d``: A 32, 64, or 128-bit integer register.
3366 - ``a``: A 32, 64, or 128-bit integer address register (excludes R0, which in an
3367 address context evaluates as zero).
3368 - ``h``: A 32-bit value in the high part of a 64bit data register
3370 - ``f``: A 32, 64, or 128-bit floating point register.
3374 - ``I``: An immediate integer between 0 and 31.
3375 - ``J``: An immediate integer between 0 and 64.
3376 - ``K``: An immediate signed 8-bit integer.
3377 - ``L``: An immediate integer, 0xff or 0xffff or (in 64-bit mode only)
3379 - ``M``: An immediate integer between 0 and 3.
3380 - ``N``: An immediate unsigned 8-bit integer.
3381 - ``O``: An immediate integer between 0 and 127.
3382 - ``e``: An immediate 32-bit signed integer.
3383 - ``Z``: An immediate 32-bit unsigned integer.
3384 - ``o``, ``v``: Treated the same as ``m``, at the moment.
3385 - ``q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3386 ``l`` integer register. On X86-32, this is the ``a``, ``b``, ``c``, and ``d``
3387 registers, and on X86-64, it is all of the integer registers.
3388 - ``Q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3389 ``h`` integer register. This is the ``a``, ``b``, ``c``, and ``d`` registers.
3390 - ``r`` or ``l``: An 8, 16, 32, or 64-bit integer register.
3391 - ``R``: An 8, 16, 32, or 64-bit "legacy" integer register -- one which has
3392 existed since i386, and can be accessed without the REX prefix.
3393 - ``f``: A 32, 64, or 80-bit '387 FPU stack pseudo-register.
3394 - ``y``: A 64-bit MMX register, if MMX is enabled.
3395 - ``x``: If SSE is enabled: a 32 or 64-bit scalar operand, or 128-bit vector
3396 operand in a SSE register. If AVX is also enabled, can also be a 256-bit
3397 vector operand in an AVX register. If AVX-512 is also enabled, can also be a
3398 512-bit vector operand in an AVX512 register, Otherwise, an error.
3399 - ``Y``: The same as ``x``, if *SSE2* is enabled, otherwise an error.
3400 - ``A``: Special case: allocates EAX first, then EDX, for a single operand (in
3401 32-bit mode, a 64-bit integer operand will get split into two registers). It
3402 is not recommended to use this constraint, as in 64-bit mode, the 64-bit
3403 operand will get allocated only to RAX -- if two 32-bit operands are needed,
3404 you're better off splitting it yourself, before passing it to the asm
3409 - ``r``: A 32-bit integer register.
3412 .. _inline-asm-modifiers:
3414 Asm template argument modifiers
3415 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3417 In the asm template string, modifiers can be used on the operand reference, like
3420 The modifiers are, in general, expected to behave the same way they do in
3421 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3422 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3423 and GCC likely indicates a bug in LLVM.
3427 - ``c``: Print an immediate integer constant unadorned, without
3428 the target-specific immediate punctuation (e.g. no ``$`` prefix).
3429 - ``n``: Negate and print immediate integer constant unadorned, without the
3430 target-specific immediate punctuation (e.g. no ``$`` prefix).
3431 - ``l``: Print as an unadorned label, without the target-specific label
3432 punctuation (e.g. no ``$`` prefix).
3436 - ``w``: Print a GPR register with a ``w*`` name instead of ``x*`` name. E.g.,
3437 instead of ``x30``, print ``w30``.
3438 - ``x``: Print a GPR register with a ``x*`` name. (this is the default, anyhow).
3439 - ``b``, ``h``, ``s``, ``d``, ``q``: Print a floating-point/SIMD register with a
3440 ``b*``, ``h*``, ``s*``, ``d*``, or ``q*`` name, rather than the default of
3449 - ``a``: Print an operand as an address (with ``[`` and ``]`` surrounding a
3453 - ``y``: Print a VFP single-precision register as an indexed double (e.g. print
3454 as ``d4[1]`` instead of ``s9``)
3455 - ``B``: Bitwise invert and print an immediate integer constant without ``#``
3457 - ``L``: Print the low 16-bits of an immediate integer constant.
3458 - ``M``: Print as a register set suitable for ldm/stm. Also prints *all*
3459 register operands subsequent to the specified one (!), so use carefully.
3460 - ``Q``: Print the low-order register of a register-pair, or the low-order
3461 register of a two-register operand.
3462 - ``R``: Print the high-order register of a register-pair, or the high-order
3463 register of a two-register operand.
3464 - ``H``: Print the second register of a register-pair. (On a big-endian system,
3465 ``H`` is equivalent to ``Q``, and on little-endian system, ``H`` is equivalent
3468 .. FIXME: H doesn't currently support printing the second register
3469 of a two-register operand.
3471 - ``e``: Print the low doubleword register of a NEON quad register.
3472 - ``f``: Print the high doubleword register of a NEON quad register.
3473 - ``m``: Print the base register of a memory operand without the ``[`` and ``]``
3478 - ``L``: Print the second register of a two-register operand. Requires that it
3479 has been allocated consecutively to the first.
3481 .. FIXME: why is it restricted to consecutive ones? And there's
3482 nothing that ensures that happens, is there?
3484 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3485 nothing. Used to print 'addi' vs 'add' instructions.
3489 No additional modifiers.
3493 - ``X``: Print an immediate integer as hexadecimal
3494 - ``x``: Print the low 16 bits of an immediate integer as hexadecimal.
3495 - ``d``: Print an immediate integer as decimal.
3496 - ``m``: Subtract one and print an immediate integer as decimal.
3497 - ``z``: Print $0 if an immediate zero, otherwise print normally.
3498 - ``L``: Print the low-order register of a two-register operand, or prints the
3499 address of the low-order word of a double-word memory operand.
3501 .. FIXME: L seems to be missing memory operand support.
3503 - ``M``: Print the high-order register of a two-register operand, or prints the
3504 address of the high-order word of a double-word memory operand.
3506 .. FIXME: M seems to be missing memory operand support.
3508 - ``D``: Print the second register of a two-register operand, or prints the
3509 second word of a double-word memory operand. (On a big-endian system, ``D`` is
3510 equivalent to ``L``, and on little-endian system, ``D`` is equivalent to
3512 - ``w``: No effect. Provided for compatibility with GCC which requires this
3513 modifier in order to print MSA registers (``W0-W31``) with the ``f``
3522 - ``L``: Print the second register of a two-register operand. Requires that it
3523 has been allocated consecutively to the first.
3525 .. FIXME: why is it restricted to consecutive ones? And there's
3526 nothing that ensures that happens, is there?
3528 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3529 nothing. Used to print 'addi' vs 'add' instructions.
3530 - ``y``: For a memory operand, prints formatter for a two-register X-form
3531 instruction. (Currently always prints ``r0,OPERAND``).
3532 - ``U``: Prints 'u' if the memory operand is an update form, and nothing
3533 otherwise. (NOTE: LLVM does not support update form, so this will currently
3534 always print nothing)
3535 - ``X``: Prints 'x' if the memory operand is an indexed form. (NOTE: LLVM does
3536 not support indexed form, so this will currently always print nothing)
3544 SystemZ implements only ``n``, and does *not* support any of the other
3545 target-independent modifiers.
3549 - ``c``: Print an unadorned integer or symbol name. (The latter is
3550 target-specific behavior for this typically target-independent modifier).
3551 - ``A``: Print a register name with a '``*``' before it.
3552 - ``b``: Print an 8-bit register name (e.g. ``al``); do nothing on a memory
3554 - ``h``: Print the upper 8-bit register name (e.g. ``ah``); do nothing on a
3556 - ``w``: Print the 16-bit register name (e.g. ``ax``); do nothing on a memory
3558 - ``k``: Print the 32-bit register name (e.g. ``eax``); do nothing on a memory
3560 - ``q``: Print the 64-bit register name (e.g. ``rax``), if 64-bit registers are
3561 available, otherwise the 32-bit register name; do nothing on a memory operand.
3562 - ``n``: Negate and print an unadorned integer, or, for operands other than an
3563 immediate integer (e.g. a relocatable symbol expression), print a '-' before
3564 the operand. (The behavior for relocatable symbol expressions is a
3565 target-specific behavior for this typically target-independent modifier)
3566 - ``H``: Print a memory reference with additional offset +8.
3567 - ``P``: Print a memory reference or operand for use as the argument of a call
3568 instruction. (E.g. omit ``(rip)``, even though it's PC-relative.)
3572 No additional modifiers.
3578 The call instructions that wrap inline asm nodes may have a
3579 "``!srcloc``" MDNode attached to it that contains a list of constant
3580 integers. If present, the code generator will use the integer as the
3581 location cookie value when report errors through the ``LLVMContext``
3582 error reporting mechanisms. This allows a front-end to correlate backend
3583 errors that occur with inline asm back to the source code that produced
3586 .. code-block:: llvm
3588 call void asm sideeffect "something bad", ""(), !srcloc !42
3590 !42 = !{ i32 1234567 }
3592 It is up to the front-end to make sense of the magic numbers it places
3593 in the IR. If the MDNode contains multiple constants, the code generator
3594 will use the one that corresponds to the line of the asm that the error
3602 LLVM IR allows metadata to be attached to instructions in the program
3603 that can convey extra information about the code to the optimizers and
3604 code generator. One example application of metadata is source-level
3605 debug information. There are two metadata primitives: strings and nodes.
3607 Metadata does not have a type, and is not a value. If referenced from a
3608 ``call`` instruction, it uses the ``metadata`` type.
3610 All metadata are identified in syntax by a exclamation point ('``!``').
3612 .. _metadata-string:
3614 Metadata Nodes and Metadata Strings
3615 -----------------------------------
3617 A metadata string is a string surrounded by double quotes. It can
3618 contain any character by escaping non-printable characters with
3619 "``\xx``" where "``xx``" is the two digit hex code. For example:
3622 Metadata nodes are represented with notation similar to structure
3623 constants (a comma separated list of elements, surrounded by braces and
3624 preceded by an exclamation point). Metadata nodes can have any values as
3625 their operand. For example:
3627 .. code-block:: llvm
3629 !{ !"test\00", i32 10}
3631 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
3633 .. code-block:: llvm
3635 !0 = distinct !{!"test\00", i32 10}
3637 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
3638 content. They can also occur when transformations cause uniquing collisions
3639 when metadata operands change.
3641 A :ref:`named metadata <namedmetadatastructure>` is a collection of
3642 metadata nodes, which can be looked up in the module symbol table. For
3645 .. code-block:: llvm
3649 Metadata can be used as function arguments. Here ``llvm.dbg.value``
3650 function is using two metadata arguments:
3652 .. code-block:: llvm
3654 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
3656 Metadata can be attached to an instruction. Here metadata ``!21`` is attached
3657 to the ``add`` instruction using the ``!dbg`` identifier:
3659 .. code-block:: llvm
3661 %indvar.next = add i64 %indvar, 1, !dbg !21
3663 Metadata can also be attached to a function definition. Here metadata ``!22``
3664 is attached to the ``foo`` function using the ``!dbg`` identifier:
3666 .. code-block:: llvm
3668 define void @foo() !dbg !22 {
3672 More information about specific metadata nodes recognized by the
3673 optimizers and code generator is found below.
3675 .. _specialized-metadata:
3677 Specialized Metadata Nodes
3678 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3680 Specialized metadata nodes are custom data structures in metadata (as opposed
3681 to generic tuples). Their fields are labelled, and can be specified in any
3684 These aren't inherently debug info centric, but currently all the specialized
3685 metadata nodes are related to debug info.
3692 ``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
3693 ``retainedTypes:``, ``subprograms:``, ``globals:`` and ``imports:`` fields are
3694 tuples containing the debug info to be emitted along with the compile unit,
3695 regardless of code optimizations (some nodes are only emitted if there are
3696 references to them from instructions).
3698 .. code-block:: llvm
3700 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
3701 isOptimized: true, flags: "-O2", runtimeVersion: 2,
3702 splitDebugFilename: "abc.debug", emissionKind: 1,
3703 enums: !2, retainedTypes: !3, subprograms: !4,
3704 globals: !5, imports: !6)
3706 Compile unit descriptors provide the root scope for objects declared in a
3707 specific compilation unit. File descriptors are defined using this scope.
3708 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
3709 keep track of subprograms, global variables, type information, and imported
3710 entities (declarations and namespaces).
3717 ``DIFile`` nodes represent files. The ``filename:`` can include slashes.
3719 .. code-block:: llvm
3721 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
3723 Files are sometimes used in ``scope:`` fields, and are the only valid target
3724 for ``file:`` fields.
3731 ``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
3732 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
3734 .. code-block:: llvm
3736 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3737 encoding: DW_ATE_unsigned_char)
3738 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
3740 The ``encoding:`` describes the details of the type. Usually it's one of the
3743 .. code-block:: llvm
3749 DW_ATE_signed_char = 6
3751 DW_ATE_unsigned_char = 8
3753 .. _DISubroutineType:
3758 ``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
3759 refers to a tuple; the first operand is the return type, while the rest are the
3760 types of the formal arguments in order. If the first operand is ``null``, that
3761 represents a function with no return value (such as ``void foo() {}`` in C++).
3763 .. code-block:: llvm
3765 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
3766 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
3767 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
3774 ``DIDerivedType`` nodes represent types derived from other types, such as
3777 .. code-block:: llvm
3779 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3780 encoding: DW_ATE_unsigned_char)
3781 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
3784 The following ``tag:`` values are valid:
3786 .. code-block:: llvm
3788 DW_TAG_formal_parameter = 5
3790 DW_TAG_pointer_type = 15
3791 DW_TAG_reference_type = 16
3793 DW_TAG_ptr_to_member_type = 31
3794 DW_TAG_const_type = 38
3795 DW_TAG_volatile_type = 53
3796 DW_TAG_restrict_type = 55
3798 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
3799 <DICompositeType>` or :ref:`subprogram <DISubprogram>`. The type of the member
3800 is the ``baseType:``. The ``offset:`` is the member's bit offset.
3801 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
3802 argument of a subprogram.
3804 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3806 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3807 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3810 Note that the ``void *`` type is expressed as a type derived from NULL.
3812 .. _DICompositeType:
3817 ``DICompositeType`` nodes represent types composed of other types, like
3818 structures and unions. ``elements:`` points to a tuple of the composed types.
3820 If the source language supports ODR, the ``identifier:`` field gives the unique
3821 identifier used for type merging between modules. When specified, other types
3822 can refer to composite types indirectly via a :ref:`metadata string
3823 <metadata-string>` that matches their identifier.
3825 .. code-block:: llvm
3827 !0 = !DIEnumerator(name: "SixKind", value: 7)
3828 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3829 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3830 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
3831 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3832 elements: !{!0, !1, !2})
3834 The following ``tag:`` values are valid:
3836 .. code-block:: llvm
3838 DW_TAG_array_type = 1
3839 DW_TAG_class_type = 2
3840 DW_TAG_enumeration_type = 4
3841 DW_TAG_structure_type = 19
3842 DW_TAG_union_type = 23
3843 DW_TAG_subroutine_type = 21
3844 DW_TAG_inheritance = 28
3847 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
3848 descriptors <DISubrange>`, each representing the range of subscripts at that
3849 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3850 array type is a native packed vector.
3852 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
3853 descriptors <DIEnumerator>`, each representing the definition of an enumeration
3854 value for the set. All enumeration type descriptors are collected in the
3855 ``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
3857 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3858 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
3859 <DIDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
3866 ``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3867 :ref:`DICompositeType`. ``count: -1`` indicates an empty array.
3869 .. code-block:: llvm
3871 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
3872 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
3873 !2 = !DISubrange(count: -1) ; empty array.
3880 ``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3881 variants of :ref:`DICompositeType`.
3883 .. code-block:: llvm
3885 !0 = !DIEnumerator(name: "SixKind", value: 7)
3886 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3887 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3889 DITemplateTypeParameter
3890 """""""""""""""""""""""
3892 ``DITemplateTypeParameter`` nodes represent type parameters to generic source
3893 language constructs. They are used (optionally) in :ref:`DICompositeType` and
3894 :ref:`DISubprogram` ``templateParams:`` fields.
3896 .. code-block:: llvm
3898 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
3900 DITemplateValueParameter
3901 """"""""""""""""""""""""
3903 ``DITemplateValueParameter`` nodes represent value parameters to generic source
3904 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3905 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3906 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
3907 :ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
3909 .. code-block:: llvm
3911 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
3916 ``DINamespace`` nodes represent namespaces in the source language.
3918 .. code-block:: llvm
3920 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
3925 ``DIGlobalVariable`` nodes represent global variables in the source language.
3927 .. code-block:: llvm
3929 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
3930 file: !2, line: 7, type: !3, isLocal: true,
3931 isDefinition: false, variable: i32* @foo,
3934 All global variables should be referenced by the `globals:` field of a
3935 :ref:`compile unit <DICompileUnit>`.
3942 ``DISubprogram`` nodes represent functions from the source language. A
3943 ``DISubprogram`` may be attached to a function definition using ``!dbg``
3944 metadata. The ``variables:`` field points at :ref:`variables <DILocalVariable>`
3945 that must be retained, even if their IR counterparts are optimized out of
3946 the IR. The ``type:`` field must point at an :ref:`DISubroutineType`.
3948 .. code-block:: llvm
3950 define void @_Z3foov() !dbg !0 {
3954 !0 = distinct !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
3955 file: !2, line: 7, type: !3, isLocal: true,
3956 isDefinition: false, scopeLine: 8,
3958 virtuality: DW_VIRTUALITY_pure_virtual,
3959 virtualIndex: 10, flags: DIFlagPrototyped,
3960 isOptimized: true, templateParams: !5,
3961 declaration: !6, variables: !7)
3968 ``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
3969 <DISubprogram>`. The line number and column numbers are used to distinguish
3970 two lexical blocks at same depth. They are valid targets for ``scope:``
3973 .. code-block:: llvm
3975 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
3977 Usually lexical blocks are ``distinct`` to prevent node merging based on
3980 .. _DILexicalBlockFile:
3985 ``DILexicalBlockFile`` nodes are used to discriminate between sections of a
3986 :ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
3987 indicate textual inclusion, or the ``discriminator:`` field can be used to
3988 discriminate between control flow within a single block in the source language.
3990 .. code-block:: llvm
3992 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
3993 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
3994 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
4001 ``DILocation`` nodes represent source debug locations. The ``scope:`` field is
4002 mandatory, and points at an :ref:`DILexicalBlockFile`, an
4003 :ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
4005 .. code-block:: llvm
4007 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
4009 .. _DILocalVariable:
4014 ``DILocalVariable`` nodes represent local variables in the source language. If
4015 the ``arg:`` field is set to non-zero, then this variable is a subprogram
4016 parameter, and it will be included in the ``variables:`` field of its
4017 :ref:`DISubprogram`.
4019 .. code-block:: llvm
4021 !0 = !DILocalVariable(name: "this", arg: 1, scope: !3, file: !2, line: 7,
4022 type: !3, flags: DIFlagArtificial)
4023 !1 = !DILocalVariable(name: "x", arg: 2, scope: !4, file: !2, line: 7,
4025 !2 = !DILocalVariable(name: "y", scope: !5, file: !2, line: 7, type: !3)
4030 ``DIExpression`` nodes represent DWARF expression sequences. They are used in
4031 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
4032 describe how the referenced LLVM variable relates to the source language
4035 The current supported vocabulary is limited:
4037 - ``DW_OP_deref`` dereferences the working expression.
4038 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
4039 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
4040 here, respectively) of the variable piece from the working expression.
4042 .. code-block:: llvm
4044 !0 = !DIExpression(DW_OP_deref)
4045 !1 = !DIExpression(DW_OP_plus, 3)
4046 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
4047 !3 = !DIExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
4052 ``DIObjCProperty`` nodes represent Objective-C property nodes.
4054 .. code-block:: llvm
4056 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
4057 getter: "getFoo", attributes: 7, type: !2)
4062 ``DIImportedEntity`` nodes represent entities (such as modules) imported into a
4065 .. code-block:: llvm
4067 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
4068 entity: !1, line: 7)
4073 In LLVM IR, memory does not have types, so LLVM's own type system is not
4074 suitable for doing TBAA. Instead, metadata is added to the IR to
4075 describe a type system of a higher level language. This can be used to
4076 implement typical C/C++ TBAA, but it can also be used to implement
4077 custom alias analysis behavior for other languages.
4079 The current metadata format is very simple. TBAA metadata nodes have up
4080 to three fields, e.g.:
4082 .. code-block:: llvm
4084 !0 = !{ !"an example type tree" }
4085 !1 = !{ !"int", !0 }
4086 !2 = !{ !"float", !0 }
4087 !3 = !{ !"const float", !2, i64 1 }
4089 The first field is an identity field. It can be any value, usually a
4090 metadata string, which uniquely identifies the type. The most important
4091 name in the tree is the name of the root node. Two trees with different
4092 root node names are entirely disjoint, even if they have leaves with
4095 The second field identifies the type's parent node in the tree, or is
4096 null or omitted for a root node. A type is considered to alias all of
4097 its descendants and all of its ancestors in the tree. Also, a type is
4098 considered to alias all types in other trees, so that bitcode produced
4099 from multiple front-ends is handled conservatively.
4101 If the third field is present, it's an integer which if equal to 1
4102 indicates that the type is "constant" (meaning
4103 ``pointsToConstantMemory`` should return true; see `other useful
4104 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
4106 '``tbaa.struct``' Metadata
4107 ^^^^^^^^^^^^^^^^^^^^^^^^^^
4109 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
4110 aggregate assignment operations in C and similar languages, however it
4111 is defined to copy a contiguous region of memory, which is more than
4112 strictly necessary for aggregate types which contain holes due to
4113 padding. Also, it doesn't contain any TBAA information about the fields
4116 ``!tbaa.struct`` metadata can describe which memory subregions in a
4117 memcpy are padding and what the TBAA tags of the struct are.
4119 The current metadata format is very simple. ``!tbaa.struct`` metadata
4120 nodes are a list of operands which are in conceptual groups of three.
4121 For each group of three, the first operand gives the byte offset of a
4122 field in bytes, the second gives its size in bytes, and the third gives
4125 .. code-block:: llvm
4127 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
4129 This describes a struct with two fields. The first is at offset 0 bytes
4130 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
4131 and has size 4 bytes and has tbaa tag !2.
4133 Note that the fields need not be contiguous. In this example, there is a
4134 4 byte gap between the two fields. This gap represents padding which
4135 does not carry useful data and need not be preserved.
4137 '``noalias``' and '``alias.scope``' Metadata
4138 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4140 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
4141 noalias memory-access sets. This means that some collection of memory access
4142 instructions (loads, stores, memory-accessing calls, etc.) that carry
4143 ``noalias`` metadata can specifically be specified not to alias with some other
4144 collection of memory access instructions that carry ``alias.scope`` metadata.
4145 Each type of metadata specifies a list of scopes where each scope has an id and
4146 a domain. When evaluating an aliasing query, if for some domain, the set
4147 of scopes with that domain in one instruction's ``alias.scope`` list is a
4148 subset of (or equal to) the set of scopes for that domain in another
4149 instruction's ``noalias`` list, then the two memory accesses are assumed not to
4152 The metadata identifying each domain is itself a list containing one or two
4153 entries. The first entry is the name of the domain. Note that if the name is a
4154 string then it can be combined across functions and translation units. A
4155 self-reference can be used to create globally unique domain names. A
4156 descriptive string may optionally be provided as a second list entry.
4158 The metadata identifying each scope is also itself a list containing two or
4159 three entries. The first entry is the name of the scope. Note that if the name
4160 is a string then it can be combined across functions and translation units. A
4161 self-reference can be used to create globally unique scope names. A metadata
4162 reference to the scope's domain is the second entry. A descriptive string may
4163 optionally be provided as a third list entry.
4167 .. code-block:: llvm
4169 ; Two scope domains:
4173 ; Some scopes in these domains:
4179 !5 = !{!4} ; A list containing only scope !4
4183 ; These two instructions don't alias:
4184 %0 = load float, float* %c, align 4, !alias.scope !5
4185 store float %0, float* %arrayidx.i, align 4, !noalias !5
4187 ; These two instructions also don't alias (for domain !1, the set of scopes
4188 ; in the !alias.scope equals that in the !noalias list):
4189 %2 = load float, float* %c, align 4, !alias.scope !5
4190 store float %2, float* %arrayidx.i2, align 4, !noalias !6
4192 ; These two instructions may alias (for domain !0, the set of scopes in
4193 ; the !noalias list is not a superset of, or equal to, the scopes in the
4194 ; !alias.scope list):
4195 %2 = load float, float* %c, align 4, !alias.scope !6
4196 store float %0, float* %arrayidx.i, align 4, !noalias !7
4198 '``fpmath``' Metadata
4199 ^^^^^^^^^^^^^^^^^^^^^
4201 ``fpmath`` metadata may be attached to any instruction of floating point
4202 type. It can be used to express the maximum acceptable error in the
4203 result of that instruction, in ULPs, thus potentially allowing the
4204 compiler to use a more efficient but less accurate method of computing
4205 it. ULP is defined as follows:
4207 If ``x`` is a real number that lies between two finite consecutive
4208 floating-point numbers ``a`` and ``b``, without being equal to one
4209 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
4210 distance between the two non-equal finite floating-point numbers
4211 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
4213 The metadata node shall consist of a single positive floating point
4214 number representing the maximum relative error, for example:
4216 .. code-block:: llvm
4218 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
4222 '``range``' Metadata
4223 ^^^^^^^^^^^^^^^^^^^^
4225 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
4226 integer types. It expresses the possible ranges the loaded value or the value
4227 returned by the called function at this call site is in. The ranges are
4228 represented with a flattened list of integers. The loaded value or the value
4229 returned is known to be in the union of the ranges defined by each consecutive
4230 pair. Each pair has the following properties:
4232 - The type must match the type loaded by the instruction.
4233 - The pair ``a,b`` represents the range ``[a,b)``.
4234 - Both ``a`` and ``b`` are constants.
4235 - The range is allowed to wrap.
4236 - The range should not represent the full or empty set. That is,
4239 In addition, the pairs must be in signed order of the lower bound and
4240 they must be non-contiguous.
4244 .. code-block:: llvm
4246 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
4247 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
4248 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
4249 %d = invoke i8 @bar() to label %cont
4250 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
4252 !0 = !{ i8 0, i8 2 }
4253 !1 = !{ i8 255, i8 2 }
4254 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
4255 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
4257 '``unpredictable``' Metadata
4258 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4260 ``unpredictable`` metadata may be attached to any branch or switch
4261 instruction. It can be used to express the unpredictability of control
4262 flow. Similar to the llvm.expect intrinsic, it may be used to alter
4263 optimizations related to compare and branch instructions. The metadata
4264 is treated as a boolean value; if it exists, it signals that the branch
4265 or switch that it is attached to is completely unpredictable.
4270 It is sometimes useful to attach information to loop constructs. Currently,
4271 loop metadata is implemented as metadata attached to the branch instruction
4272 in the loop latch block. This type of metadata refer to a metadata node that is
4273 guaranteed to be separate for each loop. The loop identifier metadata is
4274 specified with the name ``llvm.loop``.
4276 The loop identifier metadata is implemented using a metadata that refers to
4277 itself to avoid merging it with any other identifier metadata, e.g.,
4278 during module linkage or function inlining. That is, each loop should refer
4279 to their own identification metadata even if they reside in separate functions.
4280 The following example contains loop identifier metadata for two separate loop
4283 .. code-block:: llvm
4288 The loop identifier metadata can be used to specify additional
4289 per-loop metadata. Any operands after the first operand can be treated
4290 as user-defined metadata. For example the ``llvm.loop.unroll.count``
4291 suggests an unroll factor to the loop unroller:
4293 .. code-block:: llvm
4295 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
4298 !1 = !{!"llvm.loop.unroll.count", i32 4}
4300 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
4301 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4303 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
4304 used to control per-loop vectorization and interleaving parameters such as
4305 vectorization width and interleave count. These metadata should be used in
4306 conjunction with ``llvm.loop`` loop identification metadata. The
4307 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
4308 optimization hints and the optimizer will only interleave and vectorize loops if
4309 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
4310 which contains information about loop-carried memory dependencies can be helpful
4311 in determining the safety of these transformations.
4313 '``llvm.loop.interleave.count``' Metadata
4314 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4316 This metadata suggests an interleave count to the loop interleaver.
4317 The first operand is the string ``llvm.loop.interleave.count`` and the
4318 second operand is an integer specifying the interleave count. For
4321 .. code-block:: llvm
4323 !0 = !{!"llvm.loop.interleave.count", i32 4}
4325 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
4326 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
4327 then the interleave count will be determined automatically.
4329 '``llvm.loop.vectorize.enable``' Metadata
4330 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4332 This metadata selectively enables or disables vectorization for the loop. The
4333 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
4334 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
4335 0 disables vectorization:
4337 .. code-block:: llvm
4339 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
4340 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
4342 '``llvm.loop.vectorize.width``' Metadata
4343 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4345 This metadata sets the target width of the vectorizer. The first
4346 operand is the string ``llvm.loop.vectorize.width`` and the second
4347 operand is an integer specifying the width. For example:
4349 .. code-block:: llvm
4351 !0 = !{!"llvm.loop.vectorize.width", i32 4}
4353 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
4354 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
4355 0 or if the loop does not have this metadata the width will be
4356 determined automatically.
4358 '``llvm.loop.unroll``'
4359 ^^^^^^^^^^^^^^^^^^^^^^
4361 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
4362 optimization hints such as the unroll factor. ``llvm.loop.unroll``
4363 metadata should be used in conjunction with ``llvm.loop`` loop
4364 identification metadata. The ``llvm.loop.unroll`` metadata are only
4365 optimization hints and the unrolling will only be performed if the
4366 optimizer believes it is safe to do so.
4368 '``llvm.loop.unroll.count``' Metadata
4369 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4371 This metadata suggests an unroll factor to the loop unroller. The
4372 first operand is the string ``llvm.loop.unroll.count`` and the second
4373 operand is a positive integer specifying the unroll factor. For
4376 .. code-block:: llvm
4378 !0 = !{!"llvm.loop.unroll.count", i32 4}
4380 If the trip count of the loop is less than the unroll count the loop
4381 will be partially unrolled.
4383 '``llvm.loop.unroll.disable``' Metadata
4384 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4386 This metadata disables loop unrolling. The metadata has a single operand
4387 which is the string ``llvm.loop.unroll.disable``. For example:
4389 .. code-block:: llvm
4391 !0 = !{!"llvm.loop.unroll.disable"}
4393 '``llvm.loop.unroll.runtime.disable``' Metadata
4394 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4396 This metadata disables runtime loop unrolling. The metadata has a single
4397 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
4399 .. code-block:: llvm
4401 !0 = !{!"llvm.loop.unroll.runtime.disable"}
4403 '``llvm.loop.unroll.enable``' Metadata
4404 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4406 This metadata suggests that the loop should be fully unrolled if the trip count
4407 is known at compile time and partially unrolled if the trip count is not known
4408 at compile time. The metadata has a single operand which is the string
4409 ``llvm.loop.unroll.enable``. For example:
4411 .. code-block:: llvm
4413 !0 = !{!"llvm.loop.unroll.enable"}
4415 '``llvm.loop.unroll.full``' Metadata
4416 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4418 This metadata suggests that the loop should be unrolled fully. The
4419 metadata has a single operand which is the string ``llvm.loop.unroll.full``.
4422 .. code-block:: llvm
4424 !0 = !{!"llvm.loop.unroll.full"}
4429 Metadata types used to annotate memory accesses with information helpful
4430 for optimizations are prefixed with ``llvm.mem``.
4432 '``llvm.mem.parallel_loop_access``' Metadata
4433 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4435 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
4436 or metadata containing a list of loop identifiers for nested loops.
4437 The metadata is attached to memory accessing instructions and denotes that
4438 no loop carried memory dependence exist between it and other instructions denoted
4439 with the same loop identifier.
4441 Precisely, given two instructions ``m1`` and ``m2`` that both have the
4442 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
4443 set of loops associated with that metadata, respectively, then there is no loop
4444 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
4447 As a special case, if all memory accessing instructions in a loop have
4448 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
4449 loop has no loop carried memory dependences and is considered to be a parallel
4452 Note that if not all memory access instructions have such metadata referring to
4453 the loop, then the loop is considered not being trivially parallel. Additional
4454 memory dependence analysis is required to make that determination. As a fail
4455 safe mechanism, this causes loops that were originally parallel to be considered
4456 sequential (if optimization passes that are unaware of the parallel semantics
4457 insert new memory instructions into the loop body).
4459 Example of a loop that is considered parallel due to its correct use of
4460 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
4461 metadata types that refer to the same loop identifier metadata.
4463 .. code-block:: llvm
4467 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
4469 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4471 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
4477 It is also possible to have nested parallel loops. In that case the
4478 memory accesses refer to a list of loop identifier metadata nodes instead of
4479 the loop identifier metadata node directly:
4481 .. code-block:: llvm
4485 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
4487 br label %inner.for.body
4491 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4493 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
4495 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
4499 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
4501 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
4503 outer.for.end: ; preds = %for.body
4505 !0 = !{!1, !2} ; a list of loop identifiers
4506 !1 = !{!1} ; an identifier for the inner loop
4507 !2 = !{!2} ; an identifier for the outer loop
4512 The ``llvm.bitsets`` global metadata is used to implement
4513 :doc:`bitsets <BitSets>`.
4515 '``invariant.group``' Metadata
4516 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4518 The ``invariant.group`` metadata may be attached to ``load``/``store`` instructions.
4519 The existence of the ``invariant.group`` metadata on the instruction tells
4520 the optimizer that every ``load`` and ``store`` to the same pointer operand
4521 within the same invariant group can be assumed to load or store the same
4522 value (but see the ``llvm.invariant.group.barrier`` intrinsic which affects
4523 when two pointers are considered the same).
4527 .. code-block:: llvm
4529 @unknownPtr = external global i8
4532 store i8 42, i8* %ptr, !invariant.group !0
4533 call void @foo(i8* %ptr)
4535 %a = load i8, i8* %ptr, !invariant.group !0 ; Can assume that value under %ptr didn't change
4536 call void @foo(i8* %ptr)
4537 %b = load i8, i8* %ptr, !invariant.group !1 ; Can't assume anything, because group changed
4539 %newPtr = call i8* @getPointer(i8* %ptr)
4540 %c = load i8, i8* %newPtr, !invariant.group !0 ; Can't assume anything, because we only have information about %ptr
4542 %unknownValue = load i8, i8* @unknownPtr
4543 store i8 %unknownValue, i8* %ptr, !invariant.group !0 ; Can assume that %unknownValue == 42
4545 call void @foo(i8* %ptr)
4546 %newPtr2 = call i8* @llvm.invariant.group.barrier(i8* %ptr)
4547 %d = load i8, i8* %newPtr2, !invariant.group !0 ; Can't step through invariant.group.barrier to get value of %ptr
4550 declare void @foo(i8*)
4551 declare i8* @getPointer(i8*)
4552 declare i8* @llvm.invariant.group.barrier(i8*)
4554 !0 = !{!"magic ptr"}
4555 !1 = !{!"other ptr"}
4559 Module Flags Metadata
4560 =====================
4562 Information about the module as a whole is difficult to convey to LLVM's
4563 subsystems. The LLVM IR isn't sufficient to transmit this information.
4564 The ``llvm.module.flags`` named metadata exists in order to facilitate
4565 this. These flags are in the form of key / value pairs --- much like a
4566 dictionary --- making it easy for any subsystem who cares about a flag to
4569 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
4570 Each triplet has the following form:
4572 - The first element is a *behavior* flag, which specifies the behavior
4573 when two (or more) modules are merged together, and it encounters two
4574 (or more) metadata with the same ID. The supported behaviors are
4576 - The second element is a metadata string that is a unique ID for the
4577 metadata. Each module may only have one flag entry for each unique ID (not
4578 including entries with the **Require** behavior).
4579 - The third element is the value of the flag.
4581 When two (or more) modules are merged together, the resulting
4582 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
4583 each unique metadata ID string, there will be exactly one entry in the merged
4584 modules ``llvm.module.flags`` metadata table, and the value for that entry will
4585 be determined by the merge behavior flag, as described below. The only exception
4586 is that entries with the *Require* behavior are always preserved.
4588 The following behaviors are supported:
4599 Emits an error if two values disagree, otherwise the resulting value
4600 is that of the operands.
4604 Emits a warning if two values disagree. The result value will be the
4605 operand for the flag from the first module being linked.
4609 Adds a requirement that another module flag be present and have a
4610 specified value after linking is performed. The value must be a
4611 metadata pair, where the first element of the pair is the ID of the
4612 module flag to be restricted, and the second element of the pair is
4613 the value the module flag should be restricted to. This behavior can
4614 be used to restrict the allowable results (via triggering of an
4615 error) of linking IDs with the **Override** behavior.
4619 Uses the specified value, regardless of the behavior or value of the
4620 other module. If both modules specify **Override**, but the values
4621 differ, an error will be emitted.
4625 Appends the two values, which are required to be metadata nodes.
4629 Appends the two values, which are required to be metadata
4630 nodes. However, duplicate entries in the second list are dropped
4631 during the append operation.
4633 It is an error for a particular unique flag ID to have multiple behaviors,
4634 except in the case of **Require** (which adds restrictions on another metadata
4635 value) or **Override**.
4637 An example of module flags:
4639 .. code-block:: llvm
4641 !0 = !{ i32 1, !"foo", i32 1 }
4642 !1 = !{ i32 4, !"bar", i32 37 }
4643 !2 = !{ i32 2, !"qux", i32 42 }
4644 !3 = !{ i32 3, !"qux",
4649 !llvm.module.flags = !{ !0, !1, !2, !3 }
4651 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
4652 if two or more ``!"foo"`` flags are seen is to emit an error if their
4653 values are not equal.
4655 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
4656 behavior if two or more ``!"bar"`` flags are seen is to use the value
4659 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
4660 behavior if two or more ``!"qux"`` flags are seen is to emit a
4661 warning if their values are not equal.
4663 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
4669 The behavior is to emit an error if the ``llvm.module.flags`` does not
4670 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
4673 Objective-C Garbage Collection Module Flags Metadata
4674 ----------------------------------------------------
4676 On the Mach-O platform, Objective-C stores metadata about garbage
4677 collection in a special section called "image info". The metadata
4678 consists of a version number and a bitmask specifying what types of
4679 garbage collection are supported (if any) by the file. If two or more
4680 modules are linked together their garbage collection metadata needs to
4681 be merged rather than appended together.
4683 The Objective-C garbage collection module flags metadata consists of the
4684 following key-value pairs:
4693 * - ``Objective-C Version``
4694 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
4696 * - ``Objective-C Image Info Version``
4697 - **[Required]** --- The version of the image info section. Currently
4700 * - ``Objective-C Image Info Section``
4701 - **[Required]** --- The section to place the metadata. Valid values are
4702 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
4703 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
4704 Objective-C ABI version 2.
4706 * - ``Objective-C Garbage Collection``
4707 - **[Required]** --- Specifies whether garbage collection is supported or
4708 not. Valid values are 0, for no garbage collection, and 2, for garbage
4709 collection supported.
4711 * - ``Objective-C GC Only``
4712 - **[Optional]** --- Specifies that only garbage collection is supported.
4713 If present, its value must be 6. This flag requires that the
4714 ``Objective-C Garbage Collection`` flag have the value 2.
4716 Some important flag interactions:
4718 - If a module with ``Objective-C Garbage Collection`` set to 0 is
4719 merged with a module with ``Objective-C Garbage Collection`` set to
4720 2, then the resulting module has the
4721 ``Objective-C Garbage Collection`` flag set to 0.
4722 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
4723 merged with a module with ``Objective-C GC Only`` set to 6.
4725 Automatic Linker Flags Module Flags Metadata
4726 --------------------------------------------
4728 Some targets support embedding flags to the linker inside individual object
4729 files. Typically this is used in conjunction with language extensions which
4730 allow source files to explicitly declare the libraries they depend on, and have
4731 these automatically be transmitted to the linker via object files.
4733 These flags are encoded in the IR using metadata in the module flags section,
4734 using the ``Linker Options`` key. The merge behavior for this flag is required
4735 to be ``AppendUnique``, and the value for the key is expected to be a metadata
4736 node which should be a list of other metadata nodes, each of which should be a
4737 list of metadata strings defining linker options.
4739 For example, the following metadata section specifies two separate sets of
4740 linker options, presumably to link against ``libz`` and the ``Cocoa``
4743 !0 = !{ i32 6, !"Linker Options",
4746 !{ !"-framework", !"Cocoa" } } }
4747 !llvm.module.flags = !{ !0 }
4749 The metadata encoding as lists of lists of options, as opposed to a collapsed
4750 list of options, is chosen so that the IR encoding can use multiple option
4751 strings to specify e.g., a single library, while still having that specifier be
4752 preserved as an atomic element that can be recognized by a target specific
4753 assembly writer or object file emitter.
4755 Each individual option is required to be either a valid option for the target's
4756 linker, or an option that is reserved by the target specific assembly writer or
4757 object file emitter. No other aspect of these options is defined by the IR.
4759 C type width Module Flags Metadata
4760 ----------------------------------
4762 The ARM backend emits a section into each generated object file describing the
4763 options that it was compiled with (in a compiler-independent way) to prevent
4764 linking incompatible objects, and to allow automatic library selection. Some
4765 of these options are not visible at the IR level, namely wchar_t width and enum
4768 To pass this information to the backend, these options are encoded in module
4769 flags metadata, using the following key-value pairs:
4779 - * 0 --- sizeof(wchar_t) == 4
4780 * 1 --- sizeof(wchar_t) == 2
4783 - * 0 --- Enums are at least as large as an ``int``.
4784 * 1 --- Enums are stored in the smallest integer type which can
4785 represent all of its values.
4787 For example, the following metadata section specifies that the module was
4788 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
4789 enum is the smallest type which can represent all of its values::
4791 !llvm.module.flags = !{!0, !1}
4792 !0 = !{i32 1, !"short_wchar", i32 1}
4793 !1 = !{i32 1, !"short_enum", i32 0}
4795 .. _intrinsicglobalvariables:
4797 Intrinsic Global Variables
4798 ==========================
4800 LLVM has a number of "magic" global variables that contain data that
4801 affect code generation or other IR semantics. These are documented here.
4802 All globals of this sort should have a section specified as
4803 "``llvm.metadata``". This section and all globals that start with
4804 "``llvm.``" are reserved for use by LLVM.
4808 The '``llvm.used``' Global Variable
4809 -----------------------------------
4811 The ``@llvm.used`` global is an array which has
4812 :ref:`appending linkage <linkage_appending>`. This array contains a list of
4813 pointers to named global variables, functions and aliases which may optionally
4814 have a pointer cast formed of bitcast or getelementptr. For example, a legal
4817 .. code-block:: llvm
4822 @llvm.used = appending global [2 x i8*] [
4824 i8* bitcast (i32* @Y to i8*)
4825 ], section "llvm.metadata"
4827 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
4828 and linker are required to treat the symbol as if there is a reference to the
4829 symbol that it cannot see (which is why they have to be named). For example, if
4830 a variable has internal linkage and no references other than that from the
4831 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
4832 references from inline asms and other things the compiler cannot "see", and
4833 corresponds to "``attribute((used))``" in GNU C.
4835 On some targets, the code generator must emit a directive to the
4836 assembler or object file to prevent the assembler and linker from
4837 molesting the symbol.
4839 .. _gv_llvmcompilerused:
4841 The '``llvm.compiler.used``' Global Variable
4842 --------------------------------------------
4844 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4845 directive, except that it only prevents the compiler from touching the
4846 symbol. On targets that support it, this allows an intelligent linker to
4847 optimize references to the symbol without being impeded as it would be
4850 This is a rare construct that should only be used in rare circumstances,
4851 and should not be exposed to source languages.
4853 .. _gv_llvmglobalctors:
4855 The '``llvm.global_ctors``' Global Variable
4856 -------------------------------------------
4858 .. code-block:: llvm
4860 %0 = type { i32, void ()*, i8* }
4861 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
4863 The ``@llvm.global_ctors`` array contains a list of constructor
4864 functions, priorities, and an optional associated global or function.
4865 The functions referenced by this array will be called in ascending order
4866 of priority (i.e. lowest first) when the module is loaded. The order of
4867 functions with the same priority is not defined.
4869 If the third field is present, non-null, and points to a global variable
4870 or function, the initializer function will only run if the associated
4871 data from the current module is not discarded.
4873 .. _llvmglobaldtors:
4875 The '``llvm.global_dtors``' Global Variable
4876 -------------------------------------------
4878 .. code-block:: llvm
4880 %0 = type { i32, void ()*, i8* }
4881 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
4883 The ``@llvm.global_dtors`` array contains a list of destructor
4884 functions, priorities, and an optional associated global or function.
4885 The functions referenced by this array will be called in descending
4886 order of priority (i.e. highest first) when the module is unloaded. The
4887 order of functions with the same priority is not defined.
4889 If the third field is present, non-null, and points to a global variable
4890 or function, the destructor function will only run if the associated
4891 data from the current module is not discarded.
4893 Instruction Reference
4894 =====================
4896 The LLVM instruction set consists of several different classifications
4897 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
4898 instructions <binaryops>`, :ref:`bitwise binary
4899 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
4900 :ref:`other instructions <otherops>`.
4904 Terminator Instructions
4905 -----------------------
4907 As mentioned :ref:`previously <functionstructure>`, every basic block in a
4908 program ends with a "Terminator" instruction, which indicates which
4909 block should be executed after the current block is finished. These
4910 terminator instructions typically yield a '``void``' value: they produce
4911 control flow, not values (the one exception being the
4912 ':ref:`invoke <i_invoke>`' instruction).
4914 The terminator instructions are: ':ref:`ret <i_ret>`',
4915 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
4916 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
4917 ':ref:`resume <i_resume>`', ':ref:`catchpad <i_catchpad>`',
4918 ':ref:`catchendpad <i_catchendpad>`',
4919 ':ref:`catchret <i_catchret>`',
4920 ':ref:`cleanupendpad <i_cleanupendpad>`',
4921 ':ref:`cleanupret <i_cleanupret>`',
4922 ':ref:`terminatepad <i_terminatepad>`',
4923 and ':ref:`unreachable <i_unreachable>`'.
4927 '``ret``' Instruction
4928 ^^^^^^^^^^^^^^^^^^^^^
4935 ret <type> <value> ; Return a value from a non-void function
4936 ret void ; Return from void function
4941 The '``ret``' instruction is used to return control flow (and optionally
4942 a value) from a function back to the caller.
4944 There are two forms of the '``ret``' instruction: one that returns a
4945 value and then causes control flow, and one that just causes control
4951 The '``ret``' instruction optionally accepts a single argument, the
4952 return value. The type of the return value must be a ':ref:`first
4953 class <t_firstclass>`' type.
4955 A function is not :ref:`well formed <wellformed>` if it it has a non-void
4956 return type and contains a '``ret``' instruction with no return value or
4957 a return value with a type that does not match its type, or if it has a
4958 void return type and contains a '``ret``' instruction with a return
4964 When the '``ret``' instruction is executed, control flow returns back to
4965 the calling function's context. If the caller is a
4966 ":ref:`call <i_call>`" instruction, execution continues at the
4967 instruction after the call. If the caller was an
4968 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
4969 beginning of the "normal" destination block. If the instruction returns
4970 a value, that value shall set the call or invoke instruction's return
4976 .. code-block:: llvm
4978 ret i32 5 ; Return an integer value of 5
4979 ret void ; Return from a void function
4980 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
4984 '``br``' Instruction
4985 ^^^^^^^^^^^^^^^^^^^^
4992 br i1 <cond>, label <iftrue>, label <iffalse>
4993 br label <dest> ; Unconditional branch
4998 The '``br``' instruction is used to cause control flow to transfer to a
4999 different basic block in the current function. There are two forms of
5000 this instruction, corresponding to a conditional branch and an
5001 unconditional branch.
5006 The conditional branch form of the '``br``' instruction takes a single
5007 '``i1``' value and two '``label``' values. The unconditional form of the
5008 '``br``' instruction takes a single '``label``' value as a target.
5013 Upon execution of a conditional '``br``' instruction, the '``i1``'
5014 argument is evaluated. If the value is ``true``, control flows to the
5015 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
5016 to the '``iffalse``' ``label`` argument.
5021 .. code-block:: llvm
5024 %cond = icmp eq i32 %a, %b
5025 br i1 %cond, label %IfEqual, label %IfUnequal
5033 '``switch``' Instruction
5034 ^^^^^^^^^^^^^^^^^^^^^^^^
5041 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
5046 The '``switch``' instruction is used to transfer control flow to one of
5047 several different places. It is a generalization of the '``br``'
5048 instruction, allowing a branch to occur to one of many possible
5054 The '``switch``' instruction uses three parameters: an integer
5055 comparison value '``value``', a default '``label``' destination, and an
5056 array of pairs of comparison value constants and '``label``'s. The table
5057 is not allowed to contain duplicate constant entries.
5062 The ``switch`` instruction specifies a table of values and destinations.
5063 When the '``switch``' instruction is executed, this table is searched
5064 for the given value. If the value is found, control flow is transferred
5065 to the corresponding destination; otherwise, control flow is transferred
5066 to the default destination.
5071 Depending on properties of the target machine and the particular
5072 ``switch`` instruction, this instruction may be code generated in
5073 different ways. For example, it could be generated as a series of
5074 chained conditional branches or with a lookup table.
5079 .. code-block:: llvm
5081 ; Emulate a conditional br instruction
5082 %Val = zext i1 %value to i32
5083 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
5085 ; Emulate an unconditional br instruction
5086 switch i32 0, label %dest [ ]
5088 ; Implement a jump table:
5089 switch i32 %val, label %otherwise [ i32 0, label %onzero
5091 i32 2, label %ontwo ]
5095 '``indirectbr``' Instruction
5096 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5103 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
5108 The '``indirectbr``' instruction implements an indirect branch to a
5109 label within the current function, whose address is specified by
5110 "``address``". Address must be derived from a
5111 :ref:`blockaddress <blockaddress>` constant.
5116 The '``address``' argument is the address of the label to jump to. The
5117 rest of the arguments indicate the full set of possible destinations
5118 that the address may point to. Blocks are allowed to occur multiple
5119 times in the destination list, though this isn't particularly useful.
5121 This destination list is required so that dataflow analysis has an
5122 accurate understanding of the CFG.
5127 Control transfers to the block specified in the address argument. All
5128 possible destination blocks must be listed in the label list, otherwise
5129 this instruction has undefined behavior. This implies that jumps to
5130 labels defined in other functions have undefined behavior as well.
5135 This is typically implemented with a jump through a register.
5140 .. code-block:: llvm
5142 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
5146 '``invoke``' Instruction
5147 ^^^^^^^^^^^^^^^^^^^^^^^^
5154 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
5155 [operand bundles] to label <normal label> unwind label <exception label>
5160 The '``invoke``' instruction causes control to transfer to a specified
5161 function, with the possibility of control flow transfer to either the
5162 '``normal``' label or the '``exception``' label. If the callee function
5163 returns with the "``ret``" instruction, control flow will return to the
5164 "normal" label. If the callee (or any indirect callees) returns via the
5165 ":ref:`resume <i_resume>`" instruction or other exception handling
5166 mechanism, control is interrupted and continued at the dynamically
5167 nearest "exception" label.
5169 The '``exception``' label is a `landing
5170 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
5171 '``exception``' label is required to have the
5172 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
5173 information about the behavior of the program after unwinding happens,
5174 as its first non-PHI instruction. The restrictions on the
5175 "``landingpad``" instruction's tightly couples it to the "``invoke``"
5176 instruction, so that the important information contained within the
5177 "``landingpad``" instruction can't be lost through normal code motion.
5182 This instruction requires several arguments:
5184 #. The optional "cconv" marker indicates which :ref:`calling
5185 convention <callingconv>` the call should use. If none is
5186 specified, the call defaults to using C calling conventions.
5187 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5188 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5190 #. '``ptr to function ty``': shall be the signature of the pointer to
5191 function value being invoked. In most cases, this is a direct
5192 function invocation, but indirect ``invoke``'s are just as possible,
5193 branching off an arbitrary pointer to function value.
5194 #. '``function ptr val``': An LLVM value containing a pointer to a
5195 function to be invoked.
5196 #. '``function args``': argument list whose types match the function
5197 signature argument types and parameter attributes. All arguments must
5198 be of :ref:`first class <t_firstclass>` type. If the function signature
5199 indicates the function accepts a variable number of arguments, the
5200 extra arguments can be specified.
5201 #. '``normal label``': the label reached when the called function
5202 executes a '``ret``' instruction.
5203 #. '``exception label``': the label reached when a callee returns via
5204 the :ref:`resume <i_resume>` instruction or other exception handling
5206 #. The optional :ref:`function attributes <fnattrs>` list. Only
5207 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5208 attributes are valid here.
5209 #. The optional :ref:`operand bundles <opbundles>` list.
5214 This instruction is designed to operate as a standard '``call``'
5215 instruction in most regards. The primary difference is that it
5216 establishes an association with a label, which is used by the runtime
5217 library to unwind the stack.
5219 This instruction is used in languages with destructors to ensure that
5220 proper cleanup is performed in the case of either a ``longjmp`` or a
5221 thrown exception. Additionally, this is important for implementation of
5222 '``catch``' clauses in high-level languages that support them.
5224 For the purposes of the SSA form, the definition of the value returned
5225 by the '``invoke``' instruction is deemed to occur on the edge from the
5226 current block to the "normal" label. If the callee unwinds then no
5227 return value is available.
5232 .. code-block:: llvm
5234 %retval = invoke i32 @Test(i32 15) to label %Continue
5235 unwind label %TestCleanup ; i32:retval set
5236 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
5237 unwind label %TestCleanup ; i32:retval set
5241 '``resume``' Instruction
5242 ^^^^^^^^^^^^^^^^^^^^^^^^
5249 resume <type> <value>
5254 The '``resume``' instruction is a terminator instruction that has no
5260 The '``resume``' instruction requires one argument, which must have the
5261 same type as the result of any '``landingpad``' instruction in the same
5267 The '``resume``' instruction resumes propagation of an existing
5268 (in-flight) exception whose unwinding was interrupted with a
5269 :ref:`landingpad <i_landingpad>` instruction.
5274 .. code-block:: llvm
5276 resume { i8*, i32 } %exn
5280 '``catchpad``' Instruction
5281 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5288 <resultval> = catchpad [<args>*]
5289 to label <normal label> unwind label <exception label>
5294 The '``catchpad``' instruction is used by `LLVM's exception handling
5295 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5296 is a catch block --- one where a personality routine attempts to transfer
5297 control to catch an exception.
5298 The ``args`` correspond to whatever information the personality
5299 routine requires to know if this is an appropriate place to catch the
5300 exception. Control is transfered to the ``exception`` label if the
5301 ``catchpad`` is not an appropriate handler for the in-flight exception.
5302 The ``normal`` label should contain the code found in the ``catch``
5303 portion of a ``try``/``catch`` sequence. The ``resultval`` has the type
5304 :ref:`token <t_token>` and is used to match the ``catchpad`` to
5305 corresponding :ref:`catchrets <i_catchret>`.
5310 The instruction takes a list of arbitrary values which are interpreted
5311 by the :ref:`personality function <personalityfn>`.
5313 The ``catchpad`` must be provided a ``normal`` label to transfer control
5314 to if the ``catchpad`` matches the exception and an ``exception``
5315 label to transfer control to if it doesn't.
5320 When the call stack is being unwound due to an exception being thrown,
5321 the exception is compared against the ``args``. If it doesn't match,
5322 then control is transfered to the ``exception`` basic block.
5323 As with calling conventions, how the personality function results are
5324 represented in LLVM IR is target specific.
5326 The ``catchpad`` instruction has several restrictions:
5328 - A catch block is a basic block which is the unwind destination of
5329 an exceptional instruction.
5330 - A catch block must have a '``catchpad``' instruction as its
5331 first non-PHI instruction.
5332 - A catch block's ``exception`` edge must refer to a catch block or a
5334 - There can be only one '``catchpad``' instruction within the
5336 - A basic block that is not a catch block may not include a
5337 '``catchpad``' instruction.
5338 - A catch block which has another catch block as a predecessor may not have
5339 any other predecessors.
5340 - It is undefined behavior for control to transfer from a ``catchpad`` to a
5341 ``ret`` without first executing a ``catchret`` that consumes the
5342 ``catchpad`` or unwinding through its ``catchendpad``.
5343 - It is undefined behavior for control to transfer from a ``catchpad`` to
5344 itself without first executing a ``catchret`` that consumes the
5345 ``catchpad`` or unwinding through its ``catchendpad``.
5350 .. code-block:: llvm
5352 ;; A catch block which can catch an integer.
5353 %tok = catchpad [i8** @_ZTIi]
5354 to label %int.handler unwind label %terminate
5358 '``catchendpad``' Instruction
5359 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5366 catchendpad unwind label <nextaction>
5367 catchendpad unwind to caller
5372 The '``catchendpad``' instruction is used by `LLVM's exception handling
5373 system <ExceptionHandling.html#overview>`_ to communicate to the
5374 :ref:`personality function <personalityfn>` which invokes are associated
5375 with a chain of :ref:`catchpad <i_catchpad>` instructions; propagating an
5376 exception out of a catch handler is represented by unwinding through its
5377 ``catchendpad``. Unwinding to the outer scope when a chain of catch handlers
5378 do not handle an exception is also represented by unwinding through their
5381 The ``nextaction`` label indicates where control should transfer to if
5382 none of the ``catchpad`` instructions are suitable for catching the
5383 in-flight exception.
5385 If a ``nextaction`` label is not present, the instruction unwinds out of
5386 its parent function. The
5387 :ref:`personality function <personalityfn>` will continue processing
5388 exception handling actions in the caller.
5393 The instruction optionally takes a label, ``nextaction``, indicating
5394 where control should transfer to if none of the preceding
5395 ``catchpad`` instructions are suitable for the in-flight exception.
5400 When the call stack is being unwound due to an exception being thrown
5401 and none of the constituent ``catchpad`` instructions match, then
5402 control is transfered to ``nextaction`` if it is present. If it is not
5403 present, control is transfered to the caller.
5405 The ``catchendpad`` instruction has several restrictions:
5407 - A catch-end block is a basic block which is the unwind destination of
5408 an exceptional instruction.
5409 - A catch-end block must have a '``catchendpad``' instruction as its
5410 first non-PHI instruction.
5411 - There can be only one '``catchendpad``' instruction within the
5413 - A basic block that is not a catch-end block may not include a
5414 '``catchendpad``' instruction.
5415 - Exactly one catch block may unwind to a ``catchendpad``.
5416 - It is undefined behavior to execute a ``catchendpad`` if none of the
5417 '``catchpad``'s chained to it have been executed.
5418 - It is undefined behavior to execute a ``catchendpad`` twice without an
5419 intervening execution of one or more of the '``catchpad``'s chained to it.
5420 - It is undefined behavior to execute a ``catchendpad`` if, after the most
5421 recent execution of the normal successor edge of any ``catchpad`` chained
5422 to it, some ``catchret`` consuming that ``catchpad`` has already been
5424 - It is undefined behavior to execute a ``catchendpad`` if, after the most
5425 recent execution of the normal successor edge of any ``catchpad`` chained
5426 to it, any other ``catchpad`` or ``cleanuppad`` has been executed but has
5427 not had a corresponding
5428 ``catchret``/``cleanupret``/``catchendpad``/``cleanupendpad`` executed.
5433 .. code-block:: llvm
5435 catchendpad unwind label %terminate
5436 catchendpad unwind to caller
5440 '``catchret``' Instruction
5441 ^^^^^^^^^^^^^^^^^^^^^^^^^^
5448 catchret <value> to label <normal>
5453 The '``catchret``' instruction is a terminator instruction that has a
5460 The first argument to a '``catchret``' indicates which ``catchpad`` it
5461 exits. It must be a :ref:`catchpad <i_catchpad>`.
5462 The second argument to a '``catchret``' specifies where control will
5468 The '``catchret``' instruction ends the existing (in-flight) exception
5469 whose unwinding was interrupted with a
5470 :ref:`catchpad <i_catchpad>` instruction.
5471 The :ref:`personality function <personalityfn>` gets a chance to execute
5472 arbitrary code to, for example, run a C++ destructor.
5473 Control then transfers to ``normal``.
5474 It may be passed an optional, personality specific, value.
5476 It is undefined behavior to execute a ``catchret`` whose ``catchpad`` has
5479 It is undefined behavior to execute a ``catchret`` if, after the most recent
5480 execution of its ``catchpad``, some ``catchret`` or ``catchendpad`` linked
5481 to the same ``catchpad`` has already been executed.
5483 It is undefined behavior to execute a ``catchret`` if, after the most recent
5484 execution of its ``catchpad``, any other ``catchpad`` or ``cleanuppad`` has
5485 been executed but has not had a corresponding
5486 ``catchret``/``cleanupret``/``catchendpad``/``cleanupendpad`` executed.
5491 .. code-block:: llvm
5493 catchret %catch label %continue
5495 .. _i_cleanupendpad:
5497 '``cleanupendpad``' Instruction
5498 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5505 cleanupendpad <value> unwind label <nextaction>
5506 cleanupendpad <value> unwind to caller
5511 The '``cleanupendpad``' instruction is used by `LLVM's exception handling
5512 system <ExceptionHandling.html#overview>`_ to communicate to the
5513 :ref:`personality function <personalityfn>` which invokes are associated
5514 with a :ref:`cleanuppad <i_cleanuppad>` instructions; propagating an exception
5515 out of a cleanup is represented by unwinding through its ``cleanupendpad``.
5517 The ``nextaction`` label indicates where control should unwind to next, in the
5518 event that a cleanup is exited by means of an(other) exception being raised.
5520 If a ``nextaction`` label is not present, the instruction unwinds out of
5521 its parent function. The
5522 :ref:`personality function <personalityfn>` will continue processing
5523 exception handling actions in the caller.
5528 The '``cleanupendpad``' instruction requires one argument, which indicates
5529 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
5530 It also has an optional successor, ``nextaction``, indicating where control
5536 When and exception propagates to a ``cleanupendpad``, control is transfered to
5537 ``nextaction`` if it is present. If it is not present, control is transfered to
5540 The ``cleanupendpad`` instruction has several restrictions:
5542 - A cleanup-end block is a basic block which is the unwind destination of
5543 an exceptional instruction.
5544 - A cleanup-end block must have a '``cleanupendpad``' instruction as its
5545 first non-PHI instruction.
5546 - There can be only one '``cleanupendpad``' instruction within the
5548 - A basic block that is not a cleanup-end block may not include a
5549 '``cleanupendpad``' instruction.
5550 - It is undefined behavior to execute a ``cleanupendpad`` whose ``cleanuppad``
5551 has not been executed.
5552 - It is undefined behavior to execute a ``cleanupendpad`` if, after the most
5553 recent execution of its ``cleanuppad``, some ``cleanupret`` or ``cleanupendpad``
5554 consuming the same ``cleanuppad`` has already been executed.
5555 - It is undefined behavior to execute a ``cleanupendpad`` if, after the most
5556 recent execution of its ``cleanuppad``, any other ``cleanuppad`` or
5557 ``catchpad`` has been executed but has not had a corresponding
5558 ``cleanupret``/``catchret``/``cleanupendpad``/``catchendpad`` executed.
5563 .. code-block:: llvm
5565 cleanupendpad %cleanup unwind label %terminate
5566 cleanupendpad %cleanup unwind to caller
5570 '``cleanupret``' Instruction
5571 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5578 cleanupret <value> unwind label <continue>
5579 cleanupret <value> unwind to caller
5584 The '``cleanupret``' instruction is a terminator instruction that has
5585 an optional successor.
5591 The '``cleanupret``' instruction requires one argument, which indicates
5592 which ``cleanuppad`` it exits, and must be a :ref:`cleanuppad <i_cleanuppad>`.
5593 It also has an optional successor, ``continue``.
5598 The '``cleanupret``' instruction indicates to the
5599 :ref:`personality function <personalityfn>` that one
5600 :ref:`cleanuppad <i_cleanuppad>` it transferred control to has ended.
5601 It transfers control to ``continue`` or unwinds out of the function.
5603 It is undefined behavior to execute a ``cleanupret`` whose ``cleanuppad`` has
5606 It is undefined behavior to execute a ``cleanupret`` if, after the most recent
5607 execution of its ``cleanuppad``, some ``cleanupret`` or ``cleanupendpad``
5608 consuming the same ``cleanuppad`` has already been executed.
5610 It is undefined behavior to execute a ``cleanupret`` if, after the most recent
5611 execution of its ``cleanuppad``, any other ``cleanuppad`` or ``catchpad`` has
5612 been executed but has not had a corresponding
5613 ``cleanupret``/``catchret``/``cleanupendpad``/``catchendpad`` executed.
5618 .. code-block:: llvm
5620 cleanupret %cleanup unwind to caller
5621 cleanupret %cleanup unwind label %continue
5625 '``terminatepad``' Instruction
5626 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5633 terminatepad [<args>*] unwind label <exception label>
5634 terminatepad [<args>*] unwind to caller
5639 The '``terminatepad``' instruction is used by `LLVM's exception handling
5640 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5641 is a terminate block --- one where a personality routine may decide to
5642 terminate the program.
5643 The ``args`` correspond to whatever information the personality
5644 routine requires to know if this is an appropriate place to terminate the
5645 program. Control is transferred to the ``exception`` label if the
5646 personality routine decides not to terminate the program for the
5647 in-flight exception.
5652 The instruction takes a list of arbitrary values which are interpreted
5653 by the :ref:`personality function <personalityfn>`.
5655 The ``terminatepad`` may be given an ``exception`` label to
5656 transfer control to if the in-flight exception matches the ``args``.
5661 When the call stack is being unwound due to an exception being thrown,
5662 the exception is compared against the ``args``. If it matches,
5663 then control is transfered to the ``exception`` basic block. Otherwise,
5664 the program is terminated via personality-specific means. Typically,
5665 the first argument to ``terminatepad`` specifies what function the
5666 personality should defer to in order to terminate the program.
5668 The ``terminatepad`` instruction has several restrictions:
5670 - A terminate block is a basic block which is the unwind destination of
5671 an exceptional instruction.
5672 - A terminate block must have a '``terminatepad``' instruction as its
5673 first non-PHI instruction.
5674 - There can be only one '``terminatepad``' instruction within the
5676 - A basic block that is not a terminate block may not include a
5677 '``terminatepad``' instruction.
5682 .. code-block:: llvm
5684 ;; A terminate block which only permits integers.
5685 terminatepad [i8** @_ZTIi] unwind label %continue
5689 '``unreachable``' Instruction
5690 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5702 The '``unreachable``' instruction has no defined semantics. This
5703 instruction is used to inform the optimizer that a particular portion of
5704 the code is not reachable. This can be used to indicate that the code
5705 after a no-return function cannot be reached, and other facts.
5710 The '``unreachable``' instruction has no defined semantics.
5717 Binary operators are used to do most of the computation in a program.
5718 They require two operands of the same type, execute an operation on
5719 them, and produce a single value. The operands might represent multiple
5720 data, as is the case with the :ref:`vector <t_vector>` data type. The
5721 result value has the same type as its operands.
5723 There are several different binary operators:
5727 '``add``' Instruction
5728 ^^^^^^^^^^^^^^^^^^^^^
5735 <result> = add <ty> <op1>, <op2> ; yields ty:result
5736 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
5737 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
5738 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
5743 The '``add``' instruction returns the sum of its two operands.
5748 The two arguments to the '``add``' instruction must be
5749 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5750 arguments must have identical types.
5755 The value produced is the integer sum of the two operands.
5757 If the sum has unsigned overflow, the result returned is the
5758 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5761 Because LLVM integers use a two's complement representation, this
5762 instruction is appropriate for both signed and unsigned integers.
5764 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5765 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5766 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
5767 unsigned and/or signed overflow, respectively, occurs.
5772 .. code-block:: llvm
5774 <result> = add i32 4, %var ; yields i32:result = 4 + %var
5778 '``fadd``' Instruction
5779 ^^^^^^^^^^^^^^^^^^^^^^
5786 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5791 The '``fadd``' instruction returns the sum of its two operands.
5796 The two arguments to the '``fadd``' instruction must be :ref:`floating
5797 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5798 Both arguments must have identical types.
5803 The value produced is the floating point sum of the two operands. This
5804 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
5805 which are optimization hints to enable otherwise unsafe floating point
5811 .. code-block:: llvm
5813 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
5815 '``sub``' Instruction
5816 ^^^^^^^^^^^^^^^^^^^^^
5823 <result> = sub <ty> <op1>, <op2> ; yields ty:result
5824 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
5825 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
5826 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
5831 The '``sub``' instruction returns the difference of its two operands.
5833 Note that the '``sub``' instruction is used to represent the '``neg``'
5834 instruction present in most other intermediate representations.
5839 The two arguments to the '``sub``' instruction must be
5840 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5841 arguments must have identical types.
5846 The value produced is the integer difference of the two operands.
5848 If the difference has unsigned overflow, the result returned is the
5849 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5852 Because LLVM integers use a two's complement representation, this
5853 instruction is appropriate for both signed and unsigned integers.
5855 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5856 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5857 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
5858 unsigned and/or signed overflow, respectively, occurs.
5863 .. code-block:: llvm
5865 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
5866 <result> = sub i32 0, %val ; yields i32:result = -%var
5870 '``fsub``' Instruction
5871 ^^^^^^^^^^^^^^^^^^^^^^
5878 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5883 The '``fsub``' instruction returns the difference of its two operands.
5885 Note that the '``fsub``' instruction is used to represent the '``fneg``'
5886 instruction present in most other intermediate representations.
5891 The two arguments to the '``fsub``' instruction must be :ref:`floating
5892 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5893 Both arguments must have identical types.
5898 The value produced is the floating point difference of the two operands.
5899 This instruction can also take any number of :ref:`fast-math
5900 flags <fastmath>`, which are optimization hints to enable otherwise
5901 unsafe floating point optimizations:
5906 .. code-block:: llvm
5908 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
5909 <result> = fsub float -0.0, %val ; yields float:result = -%var
5911 '``mul``' Instruction
5912 ^^^^^^^^^^^^^^^^^^^^^
5919 <result> = mul <ty> <op1>, <op2> ; yields ty:result
5920 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
5921 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
5922 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
5927 The '``mul``' instruction returns the product of its two operands.
5932 The two arguments to the '``mul``' instruction must be
5933 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5934 arguments must have identical types.
5939 The value produced is the integer product of the two operands.
5941 If the result of the multiplication has unsigned overflow, the result
5942 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
5943 bit width of the result.
5945 Because LLVM integers use a two's complement representation, and the
5946 result is the same width as the operands, this instruction returns the
5947 correct result for both signed and unsigned integers. If a full product
5948 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
5949 sign-extended or zero-extended as appropriate to the width of the full
5952 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5953 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5954 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
5955 unsigned and/or signed overflow, respectively, occurs.
5960 .. code-block:: llvm
5962 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
5966 '``fmul``' Instruction
5967 ^^^^^^^^^^^^^^^^^^^^^^
5974 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5979 The '``fmul``' instruction returns the product of its two operands.
5984 The two arguments to the '``fmul``' instruction must be :ref:`floating
5985 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5986 Both arguments must have identical types.
5991 The value produced is the floating point product of the two operands.
5992 This instruction can also take any number of :ref:`fast-math
5993 flags <fastmath>`, which are optimization hints to enable otherwise
5994 unsafe floating point optimizations:
5999 .. code-block:: llvm
6001 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
6003 '``udiv``' Instruction
6004 ^^^^^^^^^^^^^^^^^^^^^^
6011 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
6012 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
6017 The '``udiv``' instruction returns the quotient of its two operands.
6022 The two arguments to the '``udiv``' instruction must be
6023 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6024 arguments must have identical types.
6029 The value produced is the unsigned integer quotient of the two operands.
6031 Note that unsigned integer division and signed integer division are
6032 distinct operations; for signed integer division, use '``sdiv``'.
6034 Division by zero leads to undefined behavior.
6036 If the ``exact`` keyword is present, the result value of the ``udiv`` is
6037 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
6038 such, "((a udiv exact b) mul b) == a").
6043 .. code-block:: llvm
6045 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
6047 '``sdiv``' Instruction
6048 ^^^^^^^^^^^^^^^^^^^^^^
6055 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
6056 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
6061 The '``sdiv``' instruction returns the quotient of its two operands.
6066 The two arguments to the '``sdiv``' instruction must be
6067 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6068 arguments must have identical types.
6073 The value produced is the signed integer quotient of the two operands
6074 rounded towards zero.
6076 Note that signed integer division and unsigned integer division are
6077 distinct operations; for unsigned integer division, use '``udiv``'.
6079 Division by zero leads to undefined behavior. Overflow also leads to
6080 undefined behavior; this is a rare case, but can occur, for example, by
6081 doing a 32-bit division of -2147483648 by -1.
6083 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
6084 a :ref:`poison value <poisonvalues>` if the result would be rounded.
6089 .. code-block:: llvm
6091 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
6095 '``fdiv``' Instruction
6096 ^^^^^^^^^^^^^^^^^^^^^^
6103 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6108 The '``fdiv``' instruction returns the quotient of its two operands.
6113 The two arguments to the '``fdiv``' instruction must be :ref:`floating
6114 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6115 Both arguments must have identical types.
6120 The value produced is the floating point quotient of the two operands.
6121 This instruction can also take any number of :ref:`fast-math
6122 flags <fastmath>`, which are optimization hints to enable otherwise
6123 unsafe floating point optimizations:
6128 .. code-block:: llvm
6130 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
6132 '``urem``' Instruction
6133 ^^^^^^^^^^^^^^^^^^^^^^
6140 <result> = urem <ty> <op1>, <op2> ; yields ty:result
6145 The '``urem``' instruction returns the remainder from the unsigned
6146 division of its two arguments.
6151 The two arguments to the '``urem``' instruction must be
6152 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6153 arguments must have identical types.
6158 This instruction returns the unsigned integer *remainder* of a division.
6159 This instruction always performs an unsigned division to get the
6162 Note that unsigned integer remainder and signed integer remainder are
6163 distinct operations; for signed integer remainder, use '``srem``'.
6165 Taking the remainder of a division by zero leads to undefined behavior.
6170 .. code-block:: llvm
6172 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
6174 '``srem``' Instruction
6175 ^^^^^^^^^^^^^^^^^^^^^^
6182 <result> = srem <ty> <op1>, <op2> ; yields ty:result
6187 The '``srem``' instruction returns the remainder from the signed
6188 division of its two operands. This instruction can also take
6189 :ref:`vector <t_vector>` versions of the values in which case the elements
6195 The two arguments to the '``srem``' instruction must be
6196 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6197 arguments must have identical types.
6202 This instruction returns the *remainder* of a division (where the result
6203 is either zero or has the same sign as the dividend, ``op1``), not the
6204 *modulo* operator (where the result is either zero or has the same sign
6205 as the divisor, ``op2``) of a value. For more information about the
6206 difference, see `The Math
6207 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
6208 table of how this is implemented in various languages, please see
6210 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
6212 Note that signed integer remainder and unsigned integer remainder are
6213 distinct operations; for unsigned integer remainder, use '``urem``'.
6215 Taking the remainder of a division by zero leads to undefined behavior.
6216 Overflow also leads to undefined behavior; this is a rare case, but can
6217 occur, for example, by taking the remainder of a 32-bit division of
6218 -2147483648 by -1. (The remainder doesn't actually overflow, but this
6219 rule lets srem be implemented using instructions that return both the
6220 result of the division and the remainder.)
6225 .. code-block:: llvm
6227 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
6231 '``frem``' Instruction
6232 ^^^^^^^^^^^^^^^^^^^^^^
6239 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
6244 The '``frem``' instruction returns the remainder from the division of
6250 The two arguments to the '``frem``' instruction must be :ref:`floating
6251 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
6252 Both arguments must have identical types.
6257 This instruction returns the *remainder* of a division. The remainder
6258 has the same sign as the dividend. This instruction can also take any
6259 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
6260 to enable otherwise unsafe floating point optimizations:
6265 .. code-block:: llvm
6267 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
6271 Bitwise Binary Operations
6272 -------------------------
6274 Bitwise binary operators are used to do various forms of bit-twiddling
6275 in a program. They are generally very efficient instructions and can
6276 commonly be strength reduced from other instructions. They require two
6277 operands of the same type, execute an operation on them, and produce a
6278 single value. The resulting value is the same type as its operands.
6280 '``shl``' Instruction
6281 ^^^^^^^^^^^^^^^^^^^^^
6288 <result> = shl <ty> <op1>, <op2> ; yields ty:result
6289 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
6290 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
6291 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
6296 The '``shl``' instruction returns the first operand shifted to the left
6297 a specified number of bits.
6302 Both arguments to the '``shl``' instruction must be the same
6303 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6304 '``op2``' is treated as an unsigned value.
6309 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
6310 where ``n`` is the width of the result. If ``op2`` is (statically or
6311 dynamically) equal to or larger than the number of bits in
6312 ``op1``, the result is undefined. If the arguments are vectors, each
6313 vector element of ``op1`` is shifted by the corresponding shift amount
6316 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
6317 value <poisonvalues>` if it shifts out any non-zero bits. If the
6318 ``nsw`` keyword is present, then the shift produces a :ref:`poison
6319 value <poisonvalues>` if it shifts out any bits that disagree with the
6320 resultant sign bit. As such, NUW/NSW have the same semantics as they
6321 would if the shift were expressed as a mul instruction with the same
6322 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
6327 .. code-block:: llvm
6329 <result> = shl i32 4, %var ; yields i32: 4 << %var
6330 <result> = shl i32 4, 2 ; yields i32: 16
6331 <result> = shl i32 1, 10 ; yields i32: 1024
6332 <result> = shl i32 1, 32 ; undefined
6333 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
6335 '``lshr``' Instruction
6336 ^^^^^^^^^^^^^^^^^^^^^^
6343 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
6344 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
6349 The '``lshr``' instruction (logical shift right) returns the first
6350 operand shifted to the right a specified number of bits with zero fill.
6355 Both arguments to the '``lshr``' instruction must be the same
6356 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6357 '``op2``' is treated as an unsigned value.
6362 This instruction always performs a logical shift right operation. The
6363 most significant bits of the result will be filled with zero bits after
6364 the shift. If ``op2`` is (statically or dynamically) equal to or larger
6365 than the number of bits in ``op1``, the result is undefined. If the
6366 arguments are vectors, each vector element of ``op1`` is shifted by the
6367 corresponding shift amount in ``op2``.
6369 If the ``exact`` keyword is present, the result value of the ``lshr`` is
6370 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6376 .. code-block:: llvm
6378 <result> = lshr i32 4, 1 ; yields i32:result = 2
6379 <result> = lshr i32 4, 2 ; yields i32:result = 1
6380 <result> = lshr i8 4, 3 ; yields i8:result = 0
6381 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
6382 <result> = lshr i32 1, 32 ; undefined
6383 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
6385 '``ashr``' Instruction
6386 ^^^^^^^^^^^^^^^^^^^^^^
6393 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
6394 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
6399 The '``ashr``' instruction (arithmetic shift right) returns the first
6400 operand shifted to the right a specified number of bits with sign
6406 Both arguments to the '``ashr``' instruction must be the same
6407 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6408 '``op2``' is treated as an unsigned value.
6413 This instruction always performs an arithmetic shift right operation,
6414 The most significant bits of the result will be filled with the sign bit
6415 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
6416 than the number of bits in ``op1``, the result is undefined. If the
6417 arguments are vectors, each vector element of ``op1`` is shifted by the
6418 corresponding shift amount in ``op2``.
6420 If the ``exact`` keyword is present, the result value of the ``ashr`` is
6421 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6427 .. code-block:: llvm
6429 <result> = ashr i32 4, 1 ; yields i32:result = 2
6430 <result> = ashr i32 4, 2 ; yields i32:result = 1
6431 <result> = ashr i8 4, 3 ; yields i8:result = 0
6432 <result> = ashr i8 -2, 1 ; yields i8:result = -1
6433 <result> = ashr i32 1, 32 ; undefined
6434 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
6436 '``and``' Instruction
6437 ^^^^^^^^^^^^^^^^^^^^^
6444 <result> = and <ty> <op1>, <op2> ; yields ty:result
6449 The '``and``' instruction returns the bitwise logical and of its two
6455 The two arguments to the '``and``' instruction must be
6456 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6457 arguments must have identical types.
6462 The truth table used for the '``and``' instruction is:
6479 .. code-block:: llvm
6481 <result> = and i32 4, %var ; yields i32:result = 4 & %var
6482 <result> = and i32 15, 40 ; yields i32:result = 8
6483 <result> = and i32 4, 8 ; yields i32:result = 0
6485 '``or``' Instruction
6486 ^^^^^^^^^^^^^^^^^^^^
6493 <result> = or <ty> <op1>, <op2> ; yields ty:result
6498 The '``or``' instruction returns the bitwise logical inclusive or of its
6504 The two arguments to the '``or``' instruction must be
6505 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6506 arguments must have identical types.
6511 The truth table used for the '``or``' instruction is:
6530 <result> = or i32 4, %var ; yields i32:result = 4 | %var
6531 <result> = or i32 15, 40 ; yields i32:result = 47
6532 <result> = or i32 4, 8 ; yields i32:result = 12
6534 '``xor``' Instruction
6535 ^^^^^^^^^^^^^^^^^^^^^
6542 <result> = xor <ty> <op1>, <op2> ; yields ty:result
6547 The '``xor``' instruction returns the bitwise logical exclusive or of
6548 its two operands. The ``xor`` is used to implement the "one's
6549 complement" operation, which is the "~" operator in C.
6554 The two arguments to the '``xor``' instruction must be
6555 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6556 arguments must have identical types.
6561 The truth table used for the '``xor``' instruction is:
6578 .. code-block:: llvm
6580 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
6581 <result> = xor i32 15, 40 ; yields i32:result = 39
6582 <result> = xor i32 4, 8 ; yields i32:result = 12
6583 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
6588 LLVM supports several instructions to represent vector operations in a
6589 target-independent manner. These instructions cover the element-access
6590 and vector-specific operations needed to process vectors effectively.
6591 While LLVM does directly support these vector operations, many
6592 sophisticated algorithms will want to use target-specific intrinsics to
6593 take full advantage of a specific target.
6595 .. _i_extractelement:
6597 '``extractelement``' Instruction
6598 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6605 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
6610 The '``extractelement``' instruction extracts a single scalar element
6611 from a vector at a specified index.
6616 The first operand of an '``extractelement``' instruction is a value of
6617 :ref:`vector <t_vector>` type. The second operand is an index indicating
6618 the position from which to extract the element. The index may be a
6619 variable of any integer type.
6624 The result is a scalar of the same type as the element type of ``val``.
6625 Its value is the value at position ``idx`` of ``val``. If ``idx``
6626 exceeds the length of ``val``, the results are undefined.
6631 .. code-block:: llvm
6633 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
6635 .. _i_insertelement:
6637 '``insertelement``' Instruction
6638 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6645 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
6650 The '``insertelement``' instruction inserts a scalar element into a
6651 vector at a specified index.
6656 The first operand of an '``insertelement``' instruction is a value of
6657 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
6658 type must equal the element type of the first operand. The third operand
6659 is an index indicating the position at which to insert the value. The
6660 index may be a variable of any integer type.
6665 The result is a vector of the same type as ``val``. Its element values
6666 are those of ``val`` except at position ``idx``, where it gets the value
6667 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
6673 .. code-block:: llvm
6675 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
6677 .. _i_shufflevector:
6679 '``shufflevector``' Instruction
6680 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6687 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
6692 The '``shufflevector``' instruction constructs a permutation of elements
6693 from two input vectors, returning a vector with the same element type as
6694 the input and length that is the same as the shuffle mask.
6699 The first two operands of a '``shufflevector``' instruction are vectors
6700 with the same type. The third argument is a shuffle mask whose element
6701 type is always 'i32'. The result of the instruction is a vector whose
6702 length is the same as the shuffle mask and whose element type is the
6703 same as the element type of the first two operands.
6705 The shuffle mask operand is required to be a constant vector with either
6706 constant integer or undef values.
6711 The elements of the two input vectors are numbered from left to right
6712 across both of the vectors. The shuffle mask operand specifies, for each
6713 element of the result vector, which element of the two input vectors the
6714 result element gets. The element selector may be undef (meaning "don't
6715 care") and the second operand may be undef if performing a shuffle from
6721 .. code-block:: llvm
6723 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6724 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
6725 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
6726 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
6727 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
6728 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
6729 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6730 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
6732 Aggregate Operations
6733 --------------------
6735 LLVM supports several instructions for working with
6736 :ref:`aggregate <t_aggregate>` values.
6740 '``extractvalue``' Instruction
6741 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6748 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
6753 The '``extractvalue``' instruction extracts the value of a member field
6754 from an :ref:`aggregate <t_aggregate>` value.
6759 The first operand of an '``extractvalue``' instruction is a value of
6760 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The other operands are
6761 constant indices to specify which value to extract in a similar manner
6762 as indices in a '``getelementptr``' instruction.
6764 The major differences to ``getelementptr`` indexing are:
6766 - Since the value being indexed is not a pointer, the first index is
6767 omitted and assumed to be zero.
6768 - At least one index must be specified.
6769 - Not only struct indices but also array indices must be in bounds.
6774 The result is the value at the position in the aggregate specified by
6780 .. code-block:: llvm
6782 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
6786 '``insertvalue``' Instruction
6787 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6794 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
6799 The '``insertvalue``' instruction inserts a value into a member field in
6800 an :ref:`aggregate <t_aggregate>` value.
6805 The first operand of an '``insertvalue``' instruction is a value of
6806 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
6807 a first-class value to insert. The following operands are constant
6808 indices indicating the position at which to insert the value in a
6809 similar manner as indices in a '``extractvalue``' instruction. The value
6810 to insert must have the same type as the value identified by the
6816 The result is an aggregate of the same type as ``val``. Its value is
6817 that of ``val`` except that the value at the position specified by the
6818 indices is that of ``elt``.
6823 .. code-block:: llvm
6825 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
6826 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
6827 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
6831 Memory Access and Addressing Operations
6832 ---------------------------------------
6834 A key design point of an SSA-based representation is how it represents
6835 memory. In LLVM, no memory locations are in SSA form, which makes things
6836 very simple. This section describes how to read, write, and allocate
6841 '``alloca``' Instruction
6842 ^^^^^^^^^^^^^^^^^^^^^^^^
6849 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
6854 The '``alloca``' instruction allocates memory on the stack frame of the
6855 currently executing function, to be automatically released when this
6856 function returns to its caller. The object is always allocated in the
6857 generic address space (address space zero).
6862 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
6863 bytes of memory on the runtime stack, returning a pointer of the
6864 appropriate type to the program. If "NumElements" is specified, it is
6865 the number of elements allocated, otherwise "NumElements" is defaulted
6866 to be one. If a constant alignment is specified, the value result of the
6867 allocation is guaranteed to be aligned to at least that boundary. The
6868 alignment may not be greater than ``1 << 29``. If not specified, or if
6869 zero, the target can choose to align the allocation on any convenient
6870 boundary compatible with the type.
6872 '``type``' may be any sized type.
6877 Memory is allocated; a pointer is returned. The operation is undefined
6878 if there is insufficient stack space for the allocation. '``alloca``'d
6879 memory is automatically released when the function returns. The
6880 '``alloca``' instruction is commonly used to represent automatic
6881 variables that must have an address available. When the function returns
6882 (either with the ``ret`` or ``resume`` instructions), the memory is
6883 reclaimed. Allocating zero bytes is legal, but the result is undefined.
6884 The order in which memory is allocated (ie., which way the stack grows)
6890 .. code-block:: llvm
6892 %ptr = alloca i32 ; yields i32*:ptr
6893 %ptr = alloca i32, i32 4 ; yields i32*:ptr
6894 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
6895 %ptr = alloca i32, align 1024 ; yields i32*:ptr
6899 '``load``' Instruction
6900 ^^^^^^^^^^^^^^^^^^^^^^
6907 <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>]
6908 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>]
6909 !<index> = !{ i32 1 }
6910 !<deref_bytes_node> = !{i64 <dereferenceable_bytes>}
6911 !<align_node> = !{ i64 <value_alignment> }
6916 The '``load``' instruction is used to read from memory.
6921 The argument to the ``load`` instruction specifies the memory address
6922 from which to load. The type specified must be a :ref:`first
6923 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
6924 then the optimizer is not allowed to modify the number or order of
6925 execution of this ``load`` with other :ref:`volatile
6926 operations <volatile>`.
6928 If the ``load`` is marked as ``atomic``, it takes an extra
6929 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
6930 ``release`` and ``acq_rel`` orderings are not valid on ``load``
6931 instructions. Atomic loads produce :ref:`defined <memmodel>` results
6932 when they may see multiple atomic stores. The type of the pointee must
6933 be an integer type whose bit width is a power of two greater than or
6934 equal to eight and less than or equal to a target-specific size limit.
6935 ``align`` must be explicitly specified on atomic loads, and the load has
6936 undefined behavior if the alignment is not set to a value which is at
6937 least the size in bytes of the pointee. ``!nontemporal`` does not have
6938 any defined semantics for atomic loads.
6940 The optional constant ``align`` argument specifies the alignment of the
6941 operation (that is, the alignment of the memory address). A value of 0
6942 or an omitted ``align`` argument means that the operation has the ABI
6943 alignment for the target. It is the responsibility of the code emitter
6944 to ensure that the alignment information is correct. Overestimating the
6945 alignment results in undefined behavior. Underestimating the alignment
6946 may produce less efficient code. An alignment of 1 is always safe. The
6947 maximum possible alignment is ``1 << 29``.
6949 The optional ``!nontemporal`` metadata must reference a single
6950 metadata name ``<index>`` corresponding to a metadata node with one
6951 ``i32`` entry of value 1. The existence of the ``!nontemporal``
6952 metadata on the instruction tells the optimizer and code generator
6953 that this load is not expected to be reused in the cache. The code
6954 generator may select special instructions to save cache bandwidth, such
6955 as the ``MOVNT`` instruction on x86.
6957 The optional ``!invariant.load`` metadata must reference a single
6958 metadata name ``<index>`` corresponding to a metadata node with no
6959 entries. The existence of the ``!invariant.load`` metadata on the
6960 instruction tells the optimizer and code generator that the address
6961 operand to this load points to memory which can be assumed unchanged.
6962 Being invariant does not imply that a location is dereferenceable,
6963 but it does imply that once the location is known dereferenceable
6964 its value is henceforth unchanging.
6966 The optional ``!invariant.group`` metadata must reference a single metadata name
6967 ``<index>`` corresponding to a metadata node. See ``invariant.group`` metadata.
6969 The optional ``!nonnull`` metadata must reference a single
6970 metadata name ``<index>`` corresponding to a metadata node with no
6971 entries. The existence of the ``!nonnull`` metadata on the
6972 instruction tells the optimizer that the value loaded is known to
6973 never be null. This is analogous to the ``nonnull`` attribute
6974 on parameters and return values. This metadata can only be applied
6975 to loads of a pointer type.
6977 The optional ``!dereferenceable`` metadata must reference a single metadata
6978 name ``<deref_bytes_node>`` corresponding to a metadata node with one ``i64``
6979 entry. The existence of the ``!dereferenceable`` metadata on the instruction
6980 tells the optimizer that the value loaded is known to be dereferenceable.
6981 The number of bytes known to be dereferenceable is specified by the integer
6982 value in the metadata node. This is analogous to the ''dereferenceable''
6983 attribute on parameters and return values. This metadata can only be applied
6984 to loads of a pointer type.
6986 The optional ``!dereferenceable_or_null`` metadata must reference a single
6987 metadata name ``<deref_bytes_node>`` corresponding to a metadata node with one
6988 ``i64`` entry. The existence of the ``!dereferenceable_or_null`` metadata on the
6989 instruction tells the optimizer that the value loaded is known to be either
6990 dereferenceable or null.
6991 The number of bytes known to be dereferenceable is specified by the integer
6992 value in the metadata node. This is analogous to the ''dereferenceable_or_null''
6993 attribute on parameters and return values. This metadata can only be applied
6994 to loads of a pointer type.
6996 The optional ``!align`` metadata must reference a single metadata name
6997 ``<align_node>`` corresponding to a metadata node with one ``i64`` entry.
6998 The existence of the ``!align`` metadata on the instruction tells the
6999 optimizer that the value loaded is known to be aligned to a boundary specified
7000 by the integer value in the metadata node. The alignment must be a power of 2.
7001 This is analogous to the ''align'' attribute on parameters and return values.
7002 This metadata can only be applied to loads of a pointer type.
7007 The location of memory pointed to is loaded. If the value being loaded
7008 is of scalar type then the number of bytes read does not exceed the
7009 minimum number of bytes needed to hold all bits of the type. For
7010 example, loading an ``i24`` reads at most three bytes. When loading a
7011 value of a type like ``i20`` with a size that is not an integral number
7012 of bytes, the result is undefined if the value was not originally
7013 written using a store of the same type.
7018 .. code-block:: llvm
7020 %ptr = alloca i32 ; yields i32*:ptr
7021 store i32 3, i32* %ptr ; yields void
7022 %val = load i32, i32* %ptr ; yields i32:val = i32 3
7026 '``store``' Instruction
7027 ^^^^^^^^^^^^^^^^^^^^^^^
7034 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.group !<index>] ; yields void
7035 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> [, !invariant.group !<index>] ; yields void
7040 The '``store``' instruction is used to write to memory.
7045 There are two arguments to the ``store`` instruction: a value to store
7046 and an address at which to store it. The type of the ``<pointer>``
7047 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
7048 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
7049 then the optimizer is not allowed to modify the number or order of
7050 execution of this ``store`` with other :ref:`volatile
7051 operations <volatile>`.
7053 If the ``store`` is marked as ``atomic``, it takes an extra
7054 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
7055 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
7056 instructions. Atomic loads produce :ref:`defined <memmodel>` results
7057 when they may see multiple atomic stores. The type of the pointee must
7058 be an integer type whose bit width is a power of two greater than or
7059 equal to eight and less than or equal to a target-specific size limit.
7060 ``align`` must be explicitly specified on atomic stores, and the store
7061 has undefined behavior if the alignment is not set to a value which is
7062 at least the size in bytes of the pointee. ``!nontemporal`` does not
7063 have any defined semantics for atomic stores.
7065 The optional constant ``align`` argument specifies the alignment of the
7066 operation (that is, the alignment of the memory address). A value of 0
7067 or an omitted ``align`` argument means that the operation has the ABI
7068 alignment for the target. It is the responsibility of the code emitter
7069 to ensure that the alignment information is correct. Overestimating the
7070 alignment results in undefined behavior. Underestimating the
7071 alignment may produce less efficient code. An alignment of 1 is always
7072 safe. The maximum possible alignment is ``1 << 29``.
7074 The optional ``!nontemporal`` metadata must reference a single metadata
7075 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
7076 value 1. The existence of the ``!nontemporal`` metadata on the instruction
7077 tells the optimizer and code generator that this load is not expected to
7078 be reused in the cache. The code generator may select special
7079 instructions to save cache bandwidth, such as the MOVNT instruction on
7082 The optional ``!invariant.group`` metadata must reference a
7083 single metadata name ``<index>``. See ``invariant.group`` metadata.
7088 The contents of memory are updated to contain ``<value>`` at the
7089 location specified by the ``<pointer>`` operand. If ``<value>`` is
7090 of scalar type then the number of bytes written does not exceed the
7091 minimum number of bytes needed to hold all bits of the type. For
7092 example, storing an ``i24`` writes at most three bytes. When writing a
7093 value of a type like ``i20`` with a size that is not an integral number
7094 of bytes, it is unspecified what happens to the extra bits that do not
7095 belong to the type, but they will typically be overwritten.
7100 .. code-block:: llvm
7102 %ptr = alloca i32 ; yields i32*:ptr
7103 store i32 3, i32* %ptr ; yields void
7104 %val = load i32, i32* %ptr ; yields i32:val = i32 3
7108 '``fence``' Instruction
7109 ^^^^^^^^^^^^^^^^^^^^^^^
7116 fence [singlethread] <ordering> ; yields void
7121 The '``fence``' instruction is used to introduce happens-before edges
7127 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
7128 defines what *synchronizes-with* edges they add. They can only be given
7129 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
7134 A fence A which has (at least) ``release`` ordering semantics
7135 *synchronizes with* a fence B with (at least) ``acquire`` ordering
7136 semantics if and only if there exist atomic operations X and Y, both
7137 operating on some atomic object M, such that A is sequenced before X, X
7138 modifies M (either directly or through some side effect of a sequence
7139 headed by X), Y is sequenced before B, and Y observes M. This provides a
7140 *happens-before* dependency between A and B. Rather than an explicit
7141 ``fence``, one (but not both) of the atomic operations X or Y might
7142 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
7143 still *synchronize-with* the explicit ``fence`` and establish the
7144 *happens-before* edge.
7146 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
7147 ``acquire`` and ``release`` semantics specified above, participates in
7148 the global program order of other ``seq_cst`` operations and/or fences.
7150 The optional ":ref:`singlethread <singlethread>`" argument specifies
7151 that the fence only synchronizes with other fences in the same thread.
7152 (This is useful for interacting with signal handlers.)
7157 .. code-block:: llvm
7159 fence acquire ; yields void
7160 fence singlethread seq_cst ; yields void
7164 '``cmpxchg``' Instruction
7165 ^^^^^^^^^^^^^^^^^^^^^^^^^
7172 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
7177 The '``cmpxchg``' instruction is used to atomically modify memory. It
7178 loads a value in memory and compares it to a given value. If they are
7179 equal, it tries to store a new value into the memory.
7184 There are three arguments to the '``cmpxchg``' instruction: an address
7185 to operate on, a value to compare to the value currently be at that
7186 address, and a new value to place at that address if the compared values
7187 are equal. The type of '<cmp>' must be an integer type whose bit width
7188 is a power of two greater than or equal to eight and less than or equal
7189 to a target-specific size limit. '<cmp>' and '<new>' must have the same
7190 type, and the type of '<pointer>' must be a pointer to that type. If the
7191 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
7192 to modify the number or order of execution of this ``cmpxchg`` with
7193 other :ref:`volatile operations <volatile>`.
7195 The success and failure :ref:`ordering <ordering>` arguments specify how this
7196 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
7197 must be at least ``monotonic``, the ordering constraint on failure must be no
7198 stronger than that on success, and the failure ordering cannot be either
7199 ``release`` or ``acq_rel``.
7201 The optional "``singlethread``" argument declares that the ``cmpxchg``
7202 is only atomic with respect to code (usually signal handlers) running in
7203 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
7204 respect to all other code in the system.
7206 The pointer passed into cmpxchg must have alignment greater than or
7207 equal to the size in memory of the operand.
7212 The contents of memory at the location specified by the '``<pointer>``' operand
7213 is read and compared to '``<cmp>``'; if the read value is the equal, the
7214 '``<new>``' is written. The original value at the location is returned, together
7215 with a flag indicating success (true) or failure (false).
7217 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
7218 permitted: the operation may not write ``<new>`` even if the comparison
7221 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
7222 if the value loaded equals ``cmp``.
7224 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
7225 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
7226 load with an ordering parameter determined the second ordering parameter.
7231 .. code-block:: llvm
7234 %orig = atomic load i32, i32* %ptr unordered ; yields i32
7238 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
7239 %squared = mul i32 %cmp, %cmp
7240 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
7241 %value_loaded = extractvalue { i32, i1 } %val_success, 0
7242 %success = extractvalue { i32, i1 } %val_success, 1
7243 br i1 %success, label %done, label %loop
7250 '``atomicrmw``' Instruction
7251 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7258 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
7263 The '``atomicrmw``' instruction is used to atomically modify memory.
7268 There are three arguments to the '``atomicrmw``' instruction: an
7269 operation to apply, an address whose value to modify, an argument to the
7270 operation. The operation must be one of the following keywords:
7284 The type of '<value>' must be an integer type whose bit width is a power
7285 of two greater than or equal to eight and less than or equal to a
7286 target-specific size limit. The type of the '``<pointer>``' operand must
7287 be a pointer to that type. If the ``atomicrmw`` is marked as
7288 ``volatile``, then the optimizer is not allowed to modify the number or
7289 order of execution of this ``atomicrmw`` with other :ref:`volatile
7290 operations <volatile>`.
7295 The contents of memory at the location specified by the '``<pointer>``'
7296 operand are atomically read, modified, and written back. The original
7297 value at the location is returned. The modification is specified by the
7300 - xchg: ``*ptr = val``
7301 - add: ``*ptr = *ptr + val``
7302 - sub: ``*ptr = *ptr - val``
7303 - and: ``*ptr = *ptr & val``
7304 - nand: ``*ptr = ~(*ptr & val)``
7305 - or: ``*ptr = *ptr | val``
7306 - xor: ``*ptr = *ptr ^ val``
7307 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
7308 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
7309 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
7311 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
7317 .. code-block:: llvm
7319 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
7321 .. _i_getelementptr:
7323 '``getelementptr``' Instruction
7324 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7331 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7332 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7333 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
7338 The '``getelementptr``' instruction is used to get the address of a
7339 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
7340 address calculation only and does not access memory. The instruction can also
7341 be used to calculate a vector of such addresses.
7346 The first argument is always a type used as the basis for the calculations.
7347 The second argument is always a pointer or a vector of pointers, and is the
7348 base address to start from. The remaining arguments are indices
7349 that indicate which of the elements of the aggregate object are indexed.
7350 The interpretation of each index is dependent on the type being indexed
7351 into. The first index always indexes the pointer value given as the
7352 first argument, the second index indexes a value of the type pointed to
7353 (not necessarily the value directly pointed to, since the first index
7354 can be non-zero), etc. The first type indexed into must be a pointer
7355 value, subsequent types can be arrays, vectors, and structs. Note that
7356 subsequent types being indexed into can never be pointers, since that
7357 would require loading the pointer before continuing calculation.
7359 The type of each index argument depends on the type it is indexing into.
7360 When indexing into a (optionally packed) structure, only ``i32`` integer
7361 **constants** are allowed (when using a vector of indices they must all
7362 be the **same** ``i32`` integer constant). When indexing into an array,
7363 pointer or vector, integers of any width are allowed, and they are not
7364 required to be constant. These integers are treated as signed values
7367 For example, let's consider a C code fragment and how it gets compiled
7383 int *foo(struct ST *s) {
7384 return &s[1].Z.B[5][13];
7387 The LLVM code generated by Clang is:
7389 .. code-block:: llvm
7391 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
7392 %struct.ST = type { i32, double, %struct.RT }
7394 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
7396 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
7403 In the example above, the first index is indexing into the
7404 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
7405 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
7406 indexes into the third element of the structure, yielding a
7407 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
7408 structure. The third index indexes into the second element of the
7409 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
7410 dimensions of the array are subscripted into, yielding an '``i32``'
7411 type. The '``getelementptr``' instruction returns a pointer to this
7412 element, thus computing a value of '``i32*``' type.
7414 Note that it is perfectly legal to index partially through a structure,
7415 returning a pointer to an inner element. Because of this, the LLVM code
7416 for the given testcase is equivalent to:
7418 .. code-block:: llvm
7420 define i32* @foo(%struct.ST* %s) {
7421 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
7422 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
7423 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
7424 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
7425 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
7429 If the ``inbounds`` keyword is present, the result value of the
7430 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
7431 pointer is not an *in bounds* address of an allocated object, or if any
7432 of the addresses that would be formed by successive addition of the
7433 offsets implied by the indices to the base address with infinitely
7434 precise signed arithmetic are not an *in bounds* address of that
7435 allocated object. The *in bounds* addresses for an allocated object are
7436 all the addresses that point into the object, plus the address one byte
7437 past the end. In cases where the base is a vector of pointers the
7438 ``inbounds`` keyword applies to each of the computations element-wise.
7440 If the ``inbounds`` keyword is not present, the offsets are added to the
7441 base address with silently-wrapping two's complement arithmetic. If the
7442 offsets have a different width from the pointer, they are sign-extended
7443 or truncated to the width of the pointer. The result value of the
7444 ``getelementptr`` may be outside the object pointed to by the base
7445 pointer. The result value may not necessarily be used to access memory
7446 though, even if it happens to point into allocated storage. See the
7447 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
7450 The getelementptr instruction is often confusing. For some more insight
7451 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
7456 .. code-block:: llvm
7458 ; yields [12 x i8]*:aptr
7459 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
7461 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
7463 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
7465 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
7470 The ``getelementptr`` returns a vector of pointers, instead of a single address,
7471 when one or more of its arguments is a vector. In such cases, all vector
7472 arguments should have the same number of elements, and every scalar argument
7473 will be effectively broadcast into a vector during address calculation.
7475 .. code-block:: llvm
7477 ; All arguments are vectors:
7478 ; A[i] = ptrs[i] + offsets[i]*sizeof(i8)
7479 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets
7481 ; Add the same scalar offset to each pointer of a vector:
7482 ; A[i] = ptrs[i] + offset*sizeof(i8)
7483 %A = getelementptr i8, <4 x i8*> %ptrs, i64 %offset
7485 ; Add distinct offsets to the same pointer:
7486 ; A[i] = ptr + offsets[i]*sizeof(i8)
7487 %A = getelementptr i8, i8* %ptr, <4 x i64> %offsets
7489 ; In all cases described above the type of the result is <4 x i8*>
7491 The two following instructions are equivalent:
7493 .. code-block:: llvm
7495 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7496 <4 x i32> <i32 2, i32 2, i32 2, i32 2>,
7497 <4 x i32> <i32 1, i32 1, i32 1, i32 1>,
7499 <4 x i64> <i64 13, i64 13, i64 13, i64 13>
7501 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7502 i32 2, i32 1, <4 x i32> %ind4, i64 13
7504 Let's look at the C code, where the vector version of ``getelementptr``
7509 // Let's assume that we vectorize the following loop:
7510 double *A, B; int *C;
7511 for (int i = 0; i < size; ++i) {
7515 .. code-block:: llvm
7517 ; get pointers for 8 elements from array B
7518 %ptrs = getelementptr double, double* %B, <8 x i32> %C
7519 ; load 8 elements from array B into A
7520 %A = call <8 x double> @llvm.masked.gather.v8f64(<8 x double*> %ptrs,
7521 i32 8, <8 x i1> %mask, <8 x double> %passthru)
7523 Conversion Operations
7524 ---------------------
7526 The instructions in this category are the conversion instructions
7527 (casting) which all take a single operand and a type. They perform
7528 various bit conversions on the operand.
7530 '``trunc .. to``' Instruction
7531 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7538 <result> = trunc <ty> <value> to <ty2> ; yields ty2
7543 The '``trunc``' instruction truncates its operand to the type ``ty2``.
7548 The '``trunc``' instruction takes a value to trunc, and a type to trunc
7549 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
7550 of the same number of integers. The bit size of the ``value`` must be
7551 larger than the bit size of the destination type, ``ty2``. Equal sized
7552 types are not allowed.
7557 The '``trunc``' instruction truncates the high order bits in ``value``
7558 and converts the remaining bits to ``ty2``. Since the source size must
7559 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
7560 It will always truncate bits.
7565 .. code-block:: llvm
7567 %X = trunc i32 257 to i8 ; yields i8:1
7568 %Y = trunc i32 123 to i1 ; yields i1:true
7569 %Z = trunc i32 122 to i1 ; yields i1:false
7570 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
7572 '``zext .. to``' Instruction
7573 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7580 <result> = zext <ty> <value> to <ty2> ; yields ty2
7585 The '``zext``' instruction zero extends its operand to type ``ty2``.
7590 The '``zext``' instruction takes a value to cast, and a type to cast it
7591 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7592 the same number of integers. The bit size of the ``value`` must be
7593 smaller than the bit size of the destination type, ``ty2``.
7598 The ``zext`` fills the high order bits of the ``value`` with zero bits
7599 until it reaches the size of the destination type, ``ty2``.
7601 When zero extending from i1, the result will always be either 0 or 1.
7606 .. code-block:: llvm
7608 %X = zext i32 257 to i64 ; yields i64:257
7609 %Y = zext i1 true to i32 ; yields i32:1
7610 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7612 '``sext .. to``' Instruction
7613 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7620 <result> = sext <ty> <value> to <ty2> ; yields ty2
7625 The '``sext``' sign extends ``value`` to the type ``ty2``.
7630 The '``sext``' instruction takes a value to cast, and a type to cast it
7631 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7632 the same number of integers. The bit size of the ``value`` must be
7633 smaller than the bit size of the destination type, ``ty2``.
7638 The '``sext``' instruction performs a sign extension by copying the sign
7639 bit (highest order bit) of the ``value`` until it reaches the bit size
7640 of the type ``ty2``.
7642 When sign extending from i1, the extension always results in -1 or 0.
7647 .. code-block:: llvm
7649 %X = sext i8 -1 to i16 ; yields i16 :65535
7650 %Y = sext i1 true to i32 ; yields i32:-1
7651 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7653 '``fptrunc .. to``' Instruction
7654 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7661 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
7666 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
7671 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
7672 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
7673 The size of ``value`` must be larger than the size of ``ty2``. This
7674 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
7679 The '``fptrunc``' instruction casts a ``value`` from a larger
7680 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
7681 point <t_floating>` type. If the value cannot fit (i.e. overflows) within the
7682 destination type, ``ty2``, then the results are undefined. If the cast produces
7683 an inexact result, how rounding is performed (e.g. truncation, also known as
7684 round to zero) is undefined.
7689 .. code-block:: llvm
7691 %X = fptrunc double 123.0 to float ; yields float:123.0
7692 %Y = fptrunc double 1.0E+300 to float ; yields undefined
7694 '``fpext .. to``' Instruction
7695 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7702 <result> = fpext <ty> <value> to <ty2> ; yields ty2
7707 The '``fpext``' extends a floating point ``value`` to a larger floating
7713 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
7714 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
7715 to. The source type must be smaller than the destination type.
7720 The '``fpext``' instruction extends the ``value`` from a smaller
7721 :ref:`floating point <t_floating>` type to a larger :ref:`floating
7722 point <t_floating>` type. The ``fpext`` cannot be used to make a
7723 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
7724 *no-op cast* for a floating point cast.
7729 .. code-block:: llvm
7731 %X = fpext float 3.125 to double ; yields double:3.125000e+00
7732 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
7734 '``fptoui .. to``' Instruction
7735 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7742 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
7747 The '``fptoui``' converts a floating point ``value`` to its unsigned
7748 integer equivalent of type ``ty2``.
7753 The '``fptoui``' instruction takes a value to cast, which must be a
7754 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7755 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7756 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7757 type with the same number of elements as ``ty``
7762 The '``fptoui``' instruction converts its :ref:`floating
7763 point <t_floating>` operand into the nearest (rounding towards zero)
7764 unsigned integer value. If the value cannot fit in ``ty2``, the results
7770 .. code-block:: llvm
7772 %X = fptoui double 123.0 to i32 ; yields i32:123
7773 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
7774 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
7776 '``fptosi .. to``' Instruction
7777 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7784 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
7789 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
7790 ``value`` to type ``ty2``.
7795 The '``fptosi``' instruction takes a value to cast, which must be a
7796 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7797 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7798 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7799 type with the same number of elements as ``ty``
7804 The '``fptosi``' instruction converts its :ref:`floating
7805 point <t_floating>` operand into the nearest (rounding towards zero)
7806 signed integer value. If the value cannot fit in ``ty2``, the results
7812 .. code-block:: llvm
7814 %X = fptosi double -123.0 to i32 ; yields i32:-123
7815 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
7816 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
7818 '``uitofp .. to``' Instruction
7819 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7826 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
7831 The '``uitofp``' instruction regards ``value`` as an unsigned integer
7832 and converts that value to the ``ty2`` type.
7837 The '``uitofp``' instruction takes a value to cast, which must be a
7838 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7839 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7840 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7841 type with the same number of elements as ``ty``
7846 The '``uitofp``' instruction interprets its operand as an unsigned
7847 integer quantity and converts it to the corresponding floating point
7848 value. If the value cannot fit in the floating point value, the results
7854 .. code-block:: llvm
7856 %X = uitofp i32 257 to float ; yields float:257.0
7857 %Y = uitofp i8 -1 to double ; yields double:255.0
7859 '``sitofp .. to``' Instruction
7860 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7867 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
7872 The '``sitofp``' instruction regards ``value`` as a signed integer and
7873 converts that value to the ``ty2`` type.
7878 The '``sitofp``' instruction takes a value to cast, which must be a
7879 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7880 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7881 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7882 type with the same number of elements as ``ty``
7887 The '``sitofp``' instruction interprets its operand as a signed integer
7888 quantity and converts it to the corresponding floating point value. If
7889 the value cannot fit in the floating point value, the results are
7895 .. code-block:: llvm
7897 %X = sitofp i32 257 to float ; yields float:257.0
7898 %Y = sitofp i8 -1 to double ; yields double:-1.0
7902 '``ptrtoint .. to``' Instruction
7903 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7910 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
7915 The '``ptrtoint``' instruction converts the pointer or a vector of
7916 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
7921 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
7922 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
7923 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
7924 a vector of integers type.
7929 The '``ptrtoint``' instruction converts ``value`` to integer type
7930 ``ty2`` by interpreting the pointer value as an integer and either
7931 truncating or zero extending that value to the size of the integer type.
7932 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
7933 ``value`` is larger than ``ty2`` then a truncation is done. If they are
7934 the same size, then nothing is done (*no-op cast*) other than a type
7940 .. code-block:: llvm
7942 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
7943 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
7944 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
7948 '``inttoptr .. to``' Instruction
7949 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7956 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
7961 The '``inttoptr``' instruction converts an integer ``value`` to a
7962 pointer type, ``ty2``.
7967 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
7968 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
7974 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
7975 applying either a zero extension or a truncation depending on the size
7976 of the integer ``value``. If ``value`` is larger than the size of a
7977 pointer then a truncation is done. If ``value`` is smaller than the size
7978 of a pointer then a zero extension is done. If they are the same size,
7979 nothing is done (*no-op cast*).
7984 .. code-block:: llvm
7986 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
7987 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
7988 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
7989 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
7993 '``bitcast .. to``' Instruction
7994 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8001 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
8006 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
8012 The '``bitcast``' instruction takes a value to cast, which must be a
8013 non-aggregate first class value, and a type to cast it to, which must
8014 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
8015 bit sizes of ``value`` and the destination type, ``ty2``, must be
8016 identical. If the source type is a pointer, the destination type must
8017 also be a pointer of the same size. This instruction supports bitwise
8018 conversion of vectors to integers and to vectors of other types (as
8019 long as they have the same size).
8024 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
8025 is always a *no-op cast* because no bits change with this
8026 conversion. The conversion is done as if the ``value`` had been stored
8027 to memory and read back as type ``ty2``. Pointer (or vector of
8028 pointers) types may only be converted to other pointer (or vector of
8029 pointers) types with the same address space through this instruction.
8030 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
8031 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
8036 .. code-block:: llvm
8038 %X = bitcast i8 255 to i8 ; yields i8 :-1
8039 %Y = bitcast i32* %x to sint* ; yields sint*:%x
8040 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
8041 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
8043 .. _i_addrspacecast:
8045 '``addrspacecast .. to``' Instruction
8046 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8053 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
8058 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
8059 address space ``n`` to type ``pty2`` in address space ``m``.
8064 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
8065 to cast and a pointer type to cast it to, which must have a different
8071 The '``addrspacecast``' instruction converts the pointer value
8072 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
8073 value modification, depending on the target and the address space
8074 pair. Pointer conversions within the same address space must be
8075 performed with the ``bitcast`` instruction. Note that if the address space
8076 conversion is legal then both result and operand refer to the same memory
8082 .. code-block:: llvm
8084 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
8085 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
8086 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
8093 The instructions in this category are the "miscellaneous" instructions,
8094 which defy better classification.
8098 '``icmp``' Instruction
8099 ^^^^^^^^^^^^^^^^^^^^^^
8106 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8111 The '``icmp``' instruction returns a boolean value or a vector of
8112 boolean values based on comparison of its two integer, integer vector,
8113 pointer, or pointer vector operands.
8118 The '``icmp``' instruction takes three operands. The first operand is
8119 the condition code indicating the kind of comparison to perform. It is
8120 not a value, just a keyword. The possible condition code are:
8123 #. ``ne``: not equal
8124 #. ``ugt``: unsigned greater than
8125 #. ``uge``: unsigned greater or equal
8126 #. ``ult``: unsigned less than
8127 #. ``ule``: unsigned less or equal
8128 #. ``sgt``: signed greater than
8129 #. ``sge``: signed greater or equal
8130 #. ``slt``: signed less than
8131 #. ``sle``: signed less or equal
8133 The remaining two arguments must be :ref:`integer <t_integer>` or
8134 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
8135 must also be identical types.
8140 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
8141 code given as ``cond``. The comparison performed always yields either an
8142 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
8144 #. ``eq``: yields ``true`` if the operands are equal, ``false``
8145 otherwise. No sign interpretation is necessary or performed.
8146 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
8147 otherwise. No sign interpretation is necessary or performed.
8148 #. ``ugt``: interprets the operands as unsigned values and yields
8149 ``true`` if ``op1`` is greater than ``op2``.
8150 #. ``uge``: interprets the operands as unsigned values and yields
8151 ``true`` if ``op1`` is greater than or equal to ``op2``.
8152 #. ``ult``: interprets the operands as unsigned values and yields
8153 ``true`` if ``op1`` is less than ``op2``.
8154 #. ``ule``: interprets the operands as unsigned values and yields
8155 ``true`` if ``op1`` is less than or equal to ``op2``.
8156 #. ``sgt``: interprets the operands as signed values and yields ``true``
8157 if ``op1`` is greater than ``op2``.
8158 #. ``sge``: interprets the operands as signed values and yields ``true``
8159 if ``op1`` is greater than or equal to ``op2``.
8160 #. ``slt``: interprets the operands as signed values and yields ``true``
8161 if ``op1`` is less than ``op2``.
8162 #. ``sle``: interprets the operands as signed values and yields ``true``
8163 if ``op1`` is less than or equal to ``op2``.
8165 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
8166 are compared as if they were integers.
8168 If the operands are integer vectors, then they are compared element by
8169 element. The result is an ``i1`` vector with the same number of elements
8170 as the values being compared. Otherwise, the result is an ``i1``.
8175 .. code-block:: llvm
8177 <result> = icmp eq i32 4, 5 ; yields: result=false
8178 <result> = icmp ne float* %X, %X ; yields: result=false
8179 <result> = icmp ult i16 4, 5 ; yields: result=true
8180 <result> = icmp sgt i16 4, 5 ; yields: result=false
8181 <result> = icmp ule i16 -4, 5 ; yields: result=false
8182 <result> = icmp sge i16 4, 5 ; yields: result=false
8184 Note that the code generator does not yet support vector types with the
8185 ``icmp`` instruction.
8189 '``fcmp``' Instruction
8190 ^^^^^^^^^^^^^^^^^^^^^^
8197 <result> = fcmp [fast-math flags]* <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
8202 The '``fcmp``' instruction returns a boolean value or vector of boolean
8203 values based on comparison of its operands.
8205 If the operands are floating point scalars, then the result type is a
8206 boolean (:ref:`i1 <t_integer>`).
8208 If the operands are floating point vectors, then the result type is a
8209 vector of boolean with the same number of elements as the operands being
8215 The '``fcmp``' instruction takes three operands. The first operand is
8216 the condition code indicating the kind of comparison to perform. It is
8217 not a value, just a keyword. The possible condition code are:
8219 #. ``false``: no comparison, always returns false
8220 #. ``oeq``: ordered and equal
8221 #. ``ogt``: ordered and greater than
8222 #. ``oge``: ordered and greater than or equal
8223 #. ``olt``: ordered and less than
8224 #. ``ole``: ordered and less than or equal
8225 #. ``one``: ordered and not equal
8226 #. ``ord``: ordered (no nans)
8227 #. ``ueq``: unordered or equal
8228 #. ``ugt``: unordered or greater than
8229 #. ``uge``: unordered or greater than or equal
8230 #. ``ult``: unordered or less than
8231 #. ``ule``: unordered or less than or equal
8232 #. ``une``: unordered or not equal
8233 #. ``uno``: unordered (either nans)
8234 #. ``true``: no comparison, always returns true
8236 *Ordered* means that neither operand is a QNAN while *unordered* means
8237 that either operand may be a QNAN.
8239 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
8240 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
8241 type. They must have identical types.
8246 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
8247 condition code given as ``cond``. If the operands are vectors, then the
8248 vectors are compared element by element. Each comparison performed
8249 always yields an :ref:`i1 <t_integer>` result, as follows:
8251 #. ``false``: always yields ``false``, regardless of operands.
8252 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
8253 is equal to ``op2``.
8254 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
8255 is greater than ``op2``.
8256 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
8257 is greater than or equal to ``op2``.
8258 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
8259 is less than ``op2``.
8260 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
8261 is less than or equal to ``op2``.
8262 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
8263 is not equal to ``op2``.
8264 #. ``ord``: yields ``true`` if both operands are not a QNAN.
8265 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
8267 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
8268 greater than ``op2``.
8269 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
8270 greater than or equal to ``op2``.
8271 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
8273 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
8274 less than or equal to ``op2``.
8275 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
8276 not equal to ``op2``.
8277 #. ``uno``: yields ``true`` if either operand is a QNAN.
8278 #. ``true``: always yields ``true``, regardless of operands.
8280 The ``fcmp`` instruction can also optionally take any number of
8281 :ref:`fast-math flags <fastmath>`, which are optimization hints to enable
8282 otherwise unsafe floating point optimizations.
8284 Any set of fast-math flags are legal on an ``fcmp`` instruction, but the
8285 only flags that have any effect on its semantics are those that allow
8286 assumptions to be made about the values of input arguments; namely
8287 ``nnan``, ``ninf``, and ``nsz``. See :ref:`fastmath` for more information.
8292 .. code-block:: llvm
8294 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
8295 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
8296 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
8297 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
8299 Note that the code generator does not yet support vector types with the
8300 ``fcmp`` instruction.
8304 '``phi``' Instruction
8305 ^^^^^^^^^^^^^^^^^^^^^
8312 <result> = phi <ty> [ <val0>, <label0>], ...
8317 The '``phi``' instruction is used to implement the φ node in the SSA
8318 graph representing the function.
8323 The type of the incoming values is specified with the first type field.
8324 After this, the '``phi``' instruction takes a list of pairs as
8325 arguments, with one pair for each predecessor basic block of the current
8326 block. Only values of :ref:`first class <t_firstclass>` type may be used as
8327 the value arguments to the PHI node. Only labels may be used as the
8330 There must be no non-phi instructions between the start of a basic block
8331 and the PHI instructions: i.e. PHI instructions must be first in a basic
8334 For the purposes of the SSA form, the use of each incoming value is
8335 deemed to occur on the edge from the corresponding predecessor block to
8336 the current block (but after any definition of an '``invoke``'
8337 instruction's return value on the same edge).
8342 At runtime, the '``phi``' instruction logically takes on the value
8343 specified by the pair corresponding to the predecessor basic block that
8344 executed just prior to the current block.
8349 .. code-block:: llvm
8351 Loop: ; Infinite loop that counts from 0 on up...
8352 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
8353 %nextindvar = add i32 %indvar, 1
8358 '``select``' Instruction
8359 ^^^^^^^^^^^^^^^^^^^^^^^^
8366 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
8368 selty is either i1 or {<N x i1>}
8373 The '``select``' instruction is used to choose one value based on a
8374 condition, without IR-level branching.
8379 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
8380 values indicating the condition, and two values of the same :ref:`first
8381 class <t_firstclass>` type.
8386 If the condition is an i1 and it evaluates to 1, the instruction returns
8387 the first value argument; otherwise, it returns the second value
8390 If the condition is a vector of i1, then the value arguments must be
8391 vectors of the same size, and the selection is done element by element.
8393 If the condition is an i1 and the value arguments are vectors of the
8394 same size, then an entire vector is selected.
8399 .. code-block:: llvm
8401 %X = select i1 true, i8 17, i8 42 ; yields i8:17
8405 '``call``' Instruction
8406 ^^^^^^^^^^^^^^^^^^^^^^
8413 <result> = [tail | musttail | notail ] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
8419 The '``call``' instruction represents a simple function call.
8424 This instruction requires several arguments:
8426 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
8427 should perform tail call optimization. The ``tail`` marker is a hint that
8428 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
8429 means that the call must be tail call optimized in order for the program to
8430 be correct. The ``musttail`` marker provides these guarantees:
8432 #. The call will not cause unbounded stack growth if it is part of a
8433 recursive cycle in the call graph.
8434 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
8437 Both markers imply that the callee does not access allocas or varargs from
8438 the caller. Calls marked ``musttail`` must obey the following additional
8441 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
8442 or a pointer bitcast followed by a ret instruction.
8443 - The ret instruction must return the (possibly bitcasted) value
8444 produced by the call or void.
8445 - The caller and callee prototypes must match. Pointer types of
8446 parameters or return types may differ in pointee type, but not
8448 - The calling conventions of the caller and callee must match.
8449 - All ABI-impacting function attributes, such as sret, byval, inreg,
8450 returned, and inalloca, must match.
8451 - The callee must be varargs iff the caller is varargs. Bitcasting a
8452 non-varargs function to the appropriate varargs type is legal so
8453 long as the non-varargs prefixes obey the other rules.
8455 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
8456 the following conditions are met:
8458 - Caller and callee both have the calling convention ``fastcc``.
8459 - The call is in tail position (ret immediately follows call and ret
8460 uses value of call or is void).
8461 - Option ``-tailcallopt`` is enabled, or
8462 ``llvm::GuaranteedTailCallOpt`` is ``true``.
8463 - `Platform-specific constraints are
8464 met. <CodeGenerator.html#tailcallopt>`_
8466 #. The optional ``notail`` marker indicates that the optimizers should not add
8467 ``tail`` or ``musttail`` markers to the call. It is used to prevent tail
8468 call optimization from being performed on the call.
8470 #. The optional "cconv" marker indicates which :ref:`calling
8471 convention <callingconv>` the call should use. If none is
8472 specified, the call defaults to using C calling conventions. The
8473 calling convention of the call must match the calling convention of
8474 the target function, or else the behavior is undefined.
8475 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
8476 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
8478 #. '``ty``': the type of the call instruction itself which is also the
8479 type of the return value. Functions that return no value are marked
8481 #. '``fnty``': shall be the signature of the pointer to function value
8482 being invoked. The argument types must match the types implied by
8483 this signature. This type can be omitted if the function is not
8484 varargs and if the function type does not return a pointer to a
8486 #. '``fnptrval``': An LLVM value containing a pointer to a function to
8487 be invoked. In most cases, this is a direct function invocation, but
8488 indirect ``call``'s are just as possible, calling an arbitrary pointer
8490 #. '``function args``': argument list whose types match the function
8491 signature argument types and parameter attributes. All arguments must
8492 be of :ref:`first class <t_firstclass>` type. If the function signature
8493 indicates the function accepts a variable number of arguments, the
8494 extra arguments can be specified.
8495 #. The optional :ref:`function attributes <fnattrs>` list. Only
8496 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
8497 attributes are valid here.
8498 #. The optional :ref:`operand bundles <opbundles>` list.
8503 The '``call``' instruction is used to cause control flow to transfer to
8504 a specified function, with its incoming arguments bound to the specified
8505 values. Upon a '``ret``' instruction in the called function, control
8506 flow continues with the instruction after the function call, and the
8507 return value of the function is bound to the result argument.
8512 .. code-block:: llvm
8514 %retval = call i32 @test(i32 %argc)
8515 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
8516 %X = tail call i32 @foo() ; yields i32
8517 %Y = tail call fastcc i32 @foo() ; yields i32
8518 call void %foo(i8 97 signext)
8520 %struct.A = type { i32, i8 }
8521 %r = call %struct.A @foo() ; yields { i32, i8 }
8522 %gr = extractvalue %struct.A %r, 0 ; yields i32
8523 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
8524 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
8525 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
8527 llvm treats calls to some functions with names and arguments that match
8528 the standard C99 library as being the C99 library functions, and may
8529 perform optimizations or generate code for them under that assumption.
8530 This is something we'd like to change in the future to provide better
8531 support for freestanding environments and non-C-based languages.
8535 '``va_arg``' Instruction
8536 ^^^^^^^^^^^^^^^^^^^^^^^^
8543 <resultval> = va_arg <va_list*> <arglist>, <argty>
8548 The '``va_arg``' instruction is used to access arguments passed through
8549 the "variable argument" area of a function call. It is used to implement
8550 the ``va_arg`` macro in C.
8555 This instruction takes a ``va_list*`` value and the type of the
8556 argument. It returns a value of the specified argument type and
8557 increments the ``va_list`` to point to the next argument. The actual
8558 type of ``va_list`` is target specific.
8563 The '``va_arg``' instruction loads an argument of the specified type
8564 from the specified ``va_list`` and causes the ``va_list`` to point to
8565 the next argument. For more information, see the variable argument
8566 handling :ref:`Intrinsic Functions <int_varargs>`.
8568 It is legal for this instruction to be called in a function which does
8569 not take a variable number of arguments, for example, the ``vfprintf``
8572 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
8573 function <intrinsics>` because it takes a type as an argument.
8578 See the :ref:`variable argument processing <int_varargs>` section.
8580 Note that the code generator does not yet fully support va\_arg on many
8581 targets. Also, it does not currently support va\_arg with aggregate
8582 types on any target.
8586 '``landingpad``' Instruction
8587 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8594 <resultval> = landingpad <resultty> <clause>+
8595 <resultval> = landingpad <resultty> cleanup <clause>*
8597 <clause> := catch <type> <value>
8598 <clause> := filter <array constant type> <array constant>
8603 The '``landingpad``' instruction is used by `LLVM's exception handling
8604 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8605 is a landing pad --- one where the exception lands, and corresponds to the
8606 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
8607 defines values supplied by the :ref:`personality function <personalityfn>` upon
8608 re-entry to the function. The ``resultval`` has the type ``resultty``.
8614 ``cleanup`` flag indicates that the landing pad block is a cleanup.
8616 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
8617 contains the global variable representing the "type" that may be caught
8618 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
8619 clause takes an array constant as its argument. Use
8620 "``[0 x i8**] undef``" for a filter which cannot throw. The
8621 '``landingpad``' instruction must contain *at least* one ``clause`` or
8622 the ``cleanup`` flag.
8627 The '``landingpad``' instruction defines the values which are set by the
8628 :ref:`personality function <personalityfn>` upon re-entry to the function, and
8629 therefore the "result type" of the ``landingpad`` instruction. As with
8630 calling conventions, how the personality function results are
8631 represented in LLVM IR is target specific.
8633 The clauses are applied in order from top to bottom. If two
8634 ``landingpad`` instructions are merged together through inlining, the
8635 clauses from the calling function are appended to the list of clauses.
8636 When the call stack is being unwound due to an exception being thrown,
8637 the exception is compared against each ``clause`` in turn. If it doesn't
8638 match any of the clauses, and the ``cleanup`` flag is not set, then
8639 unwinding continues further up the call stack.
8641 The ``landingpad`` instruction has several restrictions:
8643 - A landing pad block is a basic block which is the unwind destination
8644 of an '``invoke``' instruction.
8645 - A landing pad block must have a '``landingpad``' instruction as its
8646 first non-PHI instruction.
8647 - There can be only one '``landingpad``' instruction within the landing
8649 - A basic block that is not a landing pad block may not include a
8650 '``landingpad``' instruction.
8655 .. code-block:: llvm
8657 ;; A landing pad which can catch an integer.
8658 %res = landingpad { i8*, i32 }
8660 ;; A landing pad that is a cleanup.
8661 %res = landingpad { i8*, i32 }
8663 ;; A landing pad which can catch an integer and can only throw a double.
8664 %res = landingpad { i8*, i32 }
8666 filter [1 x i8**] [@_ZTId]
8670 '``cleanuppad``' Instruction
8671 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8678 <resultval> = cleanuppad [<args>*]
8683 The '``cleanuppad``' instruction is used by `LLVM's exception handling
8684 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8685 is a cleanup block --- one where a personality routine attempts to
8686 transfer control to run cleanup actions.
8687 The ``args`` correspond to whatever additional
8688 information the :ref:`personality function <personalityfn>` requires to
8689 execute the cleanup.
8690 The ``resultval`` has the type :ref:`token <t_token>` and is used to
8691 match the ``cleanuppad`` to corresponding :ref:`cleanuprets <i_cleanupret>`
8692 and :ref:`cleanupendpads <i_cleanupendpad>`.
8697 The instruction takes a list of arbitrary values which are interpreted
8698 by the :ref:`personality function <personalityfn>`.
8703 When the call stack is being unwound due to an exception being thrown,
8704 the :ref:`personality function <personalityfn>` transfers control to the
8705 ``cleanuppad`` with the aid of the personality-specific arguments.
8706 As with calling conventions, how the personality function results are
8707 represented in LLVM IR is target specific.
8709 The ``cleanuppad`` instruction has several restrictions:
8711 - A cleanup block is a basic block which is the unwind destination of
8712 an exceptional instruction.
8713 - A cleanup block must have a '``cleanuppad``' instruction as its
8714 first non-PHI instruction.
8715 - There can be only one '``cleanuppad``' instruction within the
8717 - A basic block that is not a cleanup block may not include a
8718 '``cleanuppad``' instruction.
8719 - All '``cleanupret``'s and '``cleanupendpad``'s which consume a ``cleanuppad``
8720 must have the same exceptional successor.
8721 - It is undefined behavior for control to transfer from a ``cleanuppad`` to a
8722 ``ret`` without first executing a ``cleanupret`` or ``cleanupendpad`` that
8723 consumes the ``cleanuppad``.
8724 - It is undefined behavior for control to transfer from a ``cleanuppad`` to
8725 itself without first executing a ``cleanupret`` or ``cleanupendpad`` that
8726 consumes the ``cleanuppad``.
8731 .. code-block:: llvm
8733 %tok = cleanuppad []
8740 LLVM supports the notion of an "intrinsic function". These functions
8741 have well known names and semantics and are required to follow certain
8742 restrictions. Overall, these intrinsics represent an extension mechanism
8743 for the LLVM language that does not require changing all of the
8744 transformations in LLVM when adding to the language (or the bitcode
8745 reader/writer, the parser, etc...).
8747 Intrinsic function names must all start with an "``llvm.``" prefix. This
8748 prefix is reserved in LLVM for intrinsic names; thus, function names may
8749 not begin with this prefix. Intrinsic functions must always be external
8750 functions: you cannot define the body of intrinsic functions. Intrinsic
8751 functions may only be used in call or invoke instructions: it is illegal
8752 to take the address of an intrinsic function. Additionally, because
8753 intrinsic functions are part of the LLVM language, it is required if any
8754 are added that they be documented here.
8756 Some intrinsic functions can be overloaded, i.e., the intrinsic
8757 represents a family of functions that perform the same operation but on
8758 different data types. Because LLVM can represent over 8 million
8759 different integer types, overloading is used commonly to allow an
8760 intrinsic function to operate on any integer type. One or more of the
8761 argument types or the result type can be overloaded to accept any
8762 integer type. Argument types may also be defined as exactly matching a
8763 previous argument's type or the result type. This allows an intrinsic
8764 function which accepts multiple arguments, but needs all of them to be
8765 of the same type, to only be overloaded with respect to a single
8766 argument or the result.
8768 Overloaded intrinsics will have the names of its overloaded argument
8769 types encoded into its function name, each preceded by a period. Only
8770 those types which are overloaded result in a name suffix. Arguments
8771 whose type is matched against another type do not. For example, the
8772 ``llvm.ctpop`` function can take an integer of any width and returns an
8773 integer of exactly the same integer width. This leads to a family of
8774 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
8775 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
8776 overloaded, and only one type suffix is required. Because the argument's
8777 type is matched against the return type, it does not require its own
8780 To learn how to add an intrinsic function, please see the `Extending
8781 LLVM Guide <ExtendingLLVM.html>`_.
8785 Variable Argument Handling Intrinsics
8786 -------------------------------------
8788 Variable argument support is defined in LLVM with the
8789 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
8790 functions. These functions are related to the similarly named macros
8791 defined in the ``<stdarg.h>`` header file.
8793 All of these functions operate on arguments that use a target-specific
8794 value type "``va_list``". The LLVM assembly language reference manual
8795 does not define what this type is, so all transformations should be
8796 prepared to handle these functions regardless of the type used.
8798 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
8799 variable argument handling intrinsic functions are used.
8801 .. code-block:: llvm
8803 ; This struct is different for every platform. For most platforms,
8804 ; it is merely an i8*.
8805 %struct.va_list = type { i8* }
8807 ; For Unix x86_64 platforms, va_list is the following struct:
8808 ; %struct.va_list = type { i32, i32, i8*, i8* }
8810 define i32 @test(i32 %X, ...) {
8811 ; Initialize variable argument processing
8812 %ap = alloca %struct.va_list
8813 %ap2 = bitcast %struct.va_list* %ap to i8*
8814 call void @llvm.va_start(i8* %ap2)
8816 ; Read a single integer argument
8817 %tmp = va_arg i8* %ap2, i32
8819 ; Demonstrate usage of llvm.va_copy and llvm.va_end
8821 %aq2 = bitcast i8** %aq to i8*
8822 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
8823 call void @llvm.va_end(i8* %aq2)
8825 ; Stop processing of arguments.
8826 call void @llvm.va_end(i8* %ap2)
8830 declare void @llvm.va_start(i8*)
8831 declare void @llvm.va_copy(i8*, i8*)
8832 declare void @llvm.va_end(i8*)
8836 '``llvm.va_start``' Intrinsic
8837 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8844 declare void @llvm.va_start(i8* <arglist>)
8849 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
8850 subsequent use by ``va_arg``.
8855 The argument is a pointer to a ``va_list`` element to initialize.
8860 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
8861 available in C. In a target-dependent way, it initializes the
8862 ``va_list`` element to which the argument points, so that the next call
8863 to ``va_arg`` will produce the first variable argument passed to the
8864 function. Unlike the C ``va_start`` macro, this intrinsic does not need
8865 to know the last argument of the function as the compiler can figure
8868 '``llvm.va_end``' Intrinsic
8869 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8876 declare void @llvm.va_end(i8* <arglist>)
8881 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
8882 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
8887 The argument is a pointer to a ``va_list`` to destroy.
8892 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
8893 available in C. In a target-dependent way, it destroys the ``va_list``
8894 element to which the argument points. Calls to
8895 :ref:`llvm.va_start <int_va_start>` and
8896 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
8901 '``llvm.va_copy``' Intrinsic
8902 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8909 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
8914 The '``llvm.va_copy``' intrinsic copies the current argument position
8915 from the source argument list to the destination argument list.
8920 The first argument is a pointer to a ``va_list`` element to initialize.
8921 The second argument is a pointer to a ``va_list`` element to copy from.
8926 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
8927 available in C. In a target-dependent way, it copies the source
8928 ``va_list`` element into the destination ``va_list`` element. This
8929 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
8930 arbitrarily complex and require, for example, memory allocation.
8932 Accurate Garbage Collection Intrinsics
8933 --------------------------------------
8935 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
8936 (GC) requires the frontend to generate code containing appropriate intrinsic
8937 calls and select an appropriate GC strategy which knows how to lower these
8938 intrinsics in a manner which is appropriate for the target collector.
8940 These intrinsics allow identification of :ref:`GC roots on the
8941 stack <int_gcroot>`, as well as garbage collector implementations that
8942 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
8943 Frontends for type-safe garbage collected languages should generate
8944 these intrinsics to make use of the LLVM garbage collectors. For more
8945 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
8947 Experimental Statepoint Intrinsics
8948 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8950 LLVM provides an second experimental set of intrinsics for describing garbage
8951 collection safepoints in compiled code. These intrinsics are an alternative
8952 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
8953 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
8954 differences in approach are covered in the `Garbage Collection with LLVM
8955 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
8956 described in :doc:`Statepoints`.
8960 '``llvm.gcroot``' Intrinsic
8961 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8968 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
8973 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
8974 the code generator, and allows some metadata to be associated with it.
8979 The first argument specifies the address of a stack object that contains
8980 the root pointer. The second pointer (which must be either a constant or
8981 a global value address) contains the meta-data to be associated with the
8987 At runtime, a call to this intrinsic stores a null pointer into the
8988 "ptrloc" location. At compile-time, the code generator generates
8989 information to allow the runtime to find the pointer at GC safe points.
8990 The '``llvm.gcroot``' intrinsic may only be used in a function which
8991 :ref:`specifies a GC algorithm <gc>`.
8995 '``llvm.gcread``' Intrinsic
8996 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9003 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
9008 The '``llvm.gcread``' intrinsic identifies reads of references from heap
9009 locations, allowing garbage collector implementations that require read
9015 The second argument is the address to read from, which should be an
9016 address allocated from the garbage collector. The first object is a
9017 pointer to the start of the referenced object, if needed by the language
9018 runtime (otherwise null).
9023 The '``llvm.gcread``' intrinsic has the same semantics as a load
9024 instruction, but may be replaced with substantially more complex code by
9025 the garbage collector runtime, as needed. The '``llvm.gcread``'
9026 intrinsic may only be used in a function which :ref:`specifies a GC
9031 '``llvm.gcwrite``' Intrinsic
9032 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9039 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
9044 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
9045 locations, allowing garbage collector implementations that require write
9046 barriers (such as generational or reference counting collectors).
9051 The first argument is the reference to store, the second is the start of
9052 the object to store it to, and the third is the address of the field of
9053 Obj to store to. If the runtime does not require a pointer to the
9054 object, Obj may be null.
9059 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
9060 instruction, but may be replaced with substantially more complex code by
9061 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
9062 intrinsic may only be used in a function which :ref:`specifies a GC
9065 Code Generator Intrinsics
9066 -------------------------
9068 These intrinsics are provided by LLVM to expose special features that
9069 may only be implemented with code generator support.
9071 '``llvm.returnaddress``' Intrinsic
9072 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9079 declare i8 *@llvm.returnaddress(i32 <level>)
9084 The '``llvm.returnaddress``' intrinsic attempts to compute a
9085 target-specific value indicating the return address of the current
9086 function or one of its callers.
9091 The argument to this intrinsic indicates which function to return the
9092 address for. Zero indicates the calling function, one indicates its
9093 caller, etc. The argument is **required** to be a constant integer
9099 The '``llvm.returnaddress``' intrinsic either returns a pointer
9100 indicating the return address of the specified call frame, or zero if it
9101 cannot be identified. The value returned by this intrinsic is likely to
9102 be incorrect or 0 for arguments other than zero, so it should only be
9103 used for debugging purposes.
9105 Note that calling this intrinsic does not prevent function inlining or
9106 other aggressive transformations, so the value returned may not be that
9107 of the obvious source-language caller.
9109 '``llvm.frameaddress``' Intrinsic
9110 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9117 declare i8* @llvm.frameaddress(i32 <level>)
9122 The '``llvm.frameaddress``' intrinsic attempts to return the
9123 target-specific frame pointer value for the specified stack frame.
9128 The argument to this intrinsic indicates which function to return the
9129 frame pointer for. Zero indicates the calling function, one indicates
9130 its caller, etc. The argument is **required** to be a constant integer
9136 The '``llvm.frameaddress``' intrinsic either returns a pointer
9137 indicating the frame address of the specified call frame, or zero if it
9138 cannot be identified. The value returned by this intrinsic is likely to
9139 be incorrect or 0 for arguments other than zero, so it should only be
9140 used for debugging purposes.
9142 Note that calling this intrinsic does not prevent function inlining or
9143 other aggressive transformations, so the value returned may not be that
9144 of the obvious source-language caller.
9146 '``llvm.localescape``' and '``llvm.localrecover``' Intrinsics
9147 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9154 declare void @llvm.localescape(...)
9155 declare i8* @llvm.localrecover(i8* %func, i8* %fp, i32 %idx)
9160 The '``llvm.localescape``' intrinsic escapes offsets of a collection of static
9161 allocas, and the '``llvm.localrecover``' intrinsic applies those offsets to a
9162 live frame pointer to recover the address of the allocation. The offset is
9163 computed during frame layout of the caller of ``llvm.localescape``.
9168 All arguments to '``llvm.localescape``' must be pointers to static allocas or
9169 casts of static allocas. Each function can only call '``llvm.localescape``'
9170 once, and it can only do so from the entry block.
9172 The ``func`` argument to '``llvm.localrecover``' must be a constant
9173 bitcasted pointer to a function defined in the current module. The code
9174 generator cannot determine the frame allocation offset of functions defined in
9177 The ``fp`` argument to '``llvm.localrecover``' must be a frame pointer of a
9178 call frame that is currently live. The return value of '``llvm.localaddress``'
9179 is one way to produce such a value, but various runtimes also expose a suitable
9180 pointer in platform-specific ways.
9182 The ``idx`` argument to '``llvm.localrecover``' indicates which alloca passed to
9183 '``llvm.localescape``' to recover. It is zero-indexed.
9188 These intrinsics allow a group of functions to share access to a set of local
9189 stack allocations of a one parent function. The parent function may call the
9190 '``llvm.localescape``' intrinsic once from the function entry block, and the
9191 child functions can use '``llvm.localrecover``' to access the escaped allocas.
9192 The '``llvm.localescape``' intrinsic blocks inlining, as inlining changes where
9193 the escaped allocas are allocated, which would break attempts to use
9194 '``llvm.localrecover``'.
9196 .. _int_read_register:
9197 .. _int_write_register:
9199 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
9200 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9207 declare i32 @llvm.read_register.i32(metadata)
9208 declare i64 @llvm.read_register.i64(metadata)
9209 declare void @llvm.write_register.i32(metadata, i32 @value)
9210 declare void @llvm.write_register.i64(metadata, i64 @value)
9216 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
9217 provides access to the named register. The register must be valid on
9218 the architecture being compiled to. The type needs to be compatible
9219 with the register being read.
9224 The '``llvm.read_register``' intrinsic returns the current value of the
9225 register, where possible. The '``llvm.write_register``' intrinsic sets
9226 the current value of the register, where possible.
9228 This is useful to implement named register global variables that need
9229 to always be mapped to a specific register, as is common practice on
9230 bare-metal programs including OS kernels.
9232 The compiler doesn't check for register availability or use of the used
9233 register in surrounding code, including inline assembly. Because of that,
9234 allocatable registers are not supported.
9236 Warning: So far it only works with the stack pointer on selected
9237 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
9238 work is needed to support other registers and even more so, allocatable
9243 '``llvm.stacksave``' Intrinsic
9244 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9251 declare i8* @llvm.stacksave()
9256 The '``llvm.stacksave``' intrinsic is used to remember the current state
9257 of the function stack, for use with
9258 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
9259 implementing language features like scoped automatic variable sized
9265 This intrinsic returns a opaque pointer value that can be passed to
9266 :ref:`llvm.stackrestore <int_stackrestore>`. When an
9267 ``llvm.stackrestore`` intrinsic is executed with a value saved from
9268 ``llvm.stacksave``, it effectively restores the state of the stack to
9269 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
9270 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
9271 were allocated after the ``llvm.stacksave`` was executed.
9273 .. _int_stackrestore:
9275 '``llvm.stackrestore``' Intrinsic
9276 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9283 declare void @llvm.stackrestore(i8* %ptr)
9288 The '``llvm.stackrestore``' intrinsic is used to restore the state of
9289 the function stack to the state it was in when the corresponding
9290 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
9291 useful for implementing language features like scoped automatic variable
9292 sized arrays in C99.
9297 See the description for :ref:`llvm.stacksave <int_stacksave>`.
9299 '``llvm.prefetch``' Intrinsic
9300 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9307 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
9312 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
9313 insert a prefetch instruction if supported; otherwise, it is a noop.
9314 Prefetches have no effect on the behavior of the program but can change
9315 its performance characteristics.
9320 ``address`` is the address to be prefetched, ``rw`` is the specifier
9321 determining if the fetch should be for a read (0) or write (1), and
9322 ``locality`` is a temporal locality specifier ranging from (0) - no
9323 locality, to (3) - extremely local keep in cache. The ``cache type``
9324 specifies whether the prefetch is performed on the data (1) or
9325 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
9326 arguments must be constant integers.
9331 This intrinsic does not modify the behavior of the program. In
9332 particular, prefetches cannot trap and do not produce a value. On
9333 targets that support this intrinsic, the prefetch can provide hints to
9334 the processor cache for better performance.
9336 '``llvm.pcmarker``' Intrinsic
9337 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9344 declare void @llvm.pcmarker(i32 <id>)
9349 The '``llvm.pcmarker``' intrinsic is a method to export a Program
9350 Counter (PC) in a region of code to simulators and other tools. The
9351 method is target specific, but it is expected that the marker will use
9352 exported symbols to transmit the PC of the marker. The marker makes no
9353 guarantees that it will remain with any specific instruction after
9354 optimizations. It is possible that the presence of a marker will inhibit
9355 optimizations. The intended use is to be inserted after optimizations to
9356 allow correlations of simulation runs.
9361 ``id`` is a numerical id identifying the marker.
9366 This intrinsic does not modify the behavior of the program. Backends
9367 that do not support this intrinsic may ignore it.
9369 '``llvm.readcyclecounter``' Intrinsic
9370 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9377 declare i64 @llvm.readcyclecounter()
9382 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
9383 counter register (or similar low latency, high accuracy clocks) on those
9384 targets that support it. On X86, it should map to RDTSC. On Alpha, it
9385 should map to RPCC. As the backing counters overflow quickly (on the
9386 order of 9 seconds on alpha), this should only be used for small
9392 When directly supported, reading the cycle counter should not modify any
9393 memory. Implementations are allowed to either return a application
9394 specific value or a system wide value. On backends without support, this
9395 is lowered to a constant 0.
9397 Note that runtime support may be conditional on the privilege-level code is
9398 running at and the host platform.
9400 '``llvm.clear_cache``' Intrinsic
9401 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9408 declare void @llvm.clear_cache(i8*, i8*)
9413 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
9414 in the specified range to the execution unit of the processor. On
9415 targets with non-unified instruction and data cache, the implementation
9416 flushes the instruction cache.
9421 On platforms with coherent instruction and data caches (e.g. x86), this
9422 intrinsic is a nop. On platforms with non-coherent instruction and data
9423 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
9424 instructions or a system call, if cache flushing requires special
9427 The default behavior is to emit a call to ``__clear_cache`` from the run
9430 This instrinsic does *not* empty the instruction pipeline. Modifications
9431 of the current function are outside the scope of the intrinsic.
9433 '``llvm.instrprof_increment``' Intrinsic
9434 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9441 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
9442 i32 <num-counters>, i32 <index>)
9447 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
9448 frontend for use with instrumentation based profiling. These will be
9449 lowered by the ``-instrprof`` pass to generate execution counts of a
9455 The first argument is a pointer to a global variable containing the
9456 name of the entity being instrumented. This should generally be the
9457 (mangled) function name for a set of counters.
9459 The second argument is a hash value that can be used by the consumer
9460 of the profile data to detect changes to the instrumented source, and
9461 the third is the number of counters associated with ``name``. It is an
9462 error if ``hash`` or ``num-counters`` differ between two instances of
9463 ``instrprof_increment`` that refer to the same name.
9465 The last argument refers to which of the counters for ``name`` should
9466 be incremented. It should be a value between 0 and ``num-counters``.
9471 This intrinsic represents an increment of a profiling counter. It will
9472 cause the ``-instrprof`` pass to generate the appropriate data
9473 structures and the code to increment the appropriate value, in a
9474 format that can be written out by a compiler runtime and consumed via
9475 the ``llvm-profdata`` tool.
9477 Standard C Library Intrinsics
9478 -----------------------------
9480 LLVM provides intrinsics for a few important standard C library
9481 functions. These intrinsics allow source-language front-ends to pass
9482 information about the alignment of the pointer arguments to the code
9483 generator, providing opportunity for more efficient code generation.
9487 '``llvm.memcpy``' Intrinsic
9488 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9493 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
9494 integer bit width and for different address spaces. Not all targets
9495 support all bit widths however.
9499 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9500 i32 <len>, i32 <align>, i1 <isvolatile>)
9501 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9502 i64 <len>, i32 <align>, i1 <isvolatile>)
9507 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9508 source location to the destination location.
9510 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
9511 intrinsics do not return a value, takes extra alignment/isvolatile
9512 arguments and the pointers can be in specified address spaces.
9517 The first argument is a pointer to the destination, the second is a
9518 pointer to the source. The third argument is an integer argument
9519 specifying the number of bytes to copy, the fourth argument is the
9520 alignment of the source and destination locations, and the fifth is a
9521 boolean indicating a volatile access.
9523 If the call to this intrinsic has an alignment value that is not 0 or 1,
9524 then the caller guarantees that both the source and destination pointers
9525 are aligned to that boundary.
9527 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
9528 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9529 very cleanly specified and it is unwise to depend on it.
9534 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9535 source location to the destination location, which are not allowed to
9536 overlap. It copies "len" bytes of memory over. If the argument is known
9537 to be aligned to some boundary, this can be specified as the fourth
9538 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
9540 '``llvm.memmove``' Intrinsic
9541 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9546 This is an overloaded intrinsic. You can use llvm.memmove on any integer
9547 bit width and for different address space. Not all targets support all
9552 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9553 i32 <len>, i32 <align>, i1 <isvolatile>)
9554 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9555 i64 <len>, i32 <align>, i1 <isvolatile>)
9560 The '``llvm.memmove.*``' intrinsics move a block of memory from the
9561 source location to the destination location. It is similar to the
9562 '``llvm.memcpy``' intrinsic but allows the two memory locations to
9565 Note that, unlike the standard libc function, the ``llvm.memmove.*``
9566 intrinsics do not return a value, takes extra alignment/isvolatile
9567 arguments and the pointers can be in specified address spaces.
9572 The first argument is a pointer to the destination, the second is a
9573 pointer to the source. The third argument is an integer argument
9574 specifying the number of bytes to copy, the fourth argument is the
9575 alignment of the source and destination locations, and the fifth is a
9576 boolean indicating a volatile access.
9578 If the call to this intrinsic has an alignment value that is not 0 or 1,
9579 then the caller guarantees that the source and destination pointers are
9580 aligned to that boundary.
9582 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
9583 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
9584 not very cleanly specified and it is unwise to depend on it.
9589 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
9590 source location to the destination location, which may overlap. It
9591 copies "len" bytes of memory over. If the argument is known to be
9592 aligned to some boundary, this can be specified as the fourth argument,
9593 otherwise it should be set to 0 or 1 (both meaning no alignment).
9595 '``llvm.memset.*``' Intrinsics
9596 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9601 This is an overloaded intrinsic. You can use llvm.memset on any integer
9602 bit width and for different address spaces. However, not all targets
9603 support all bit widths.
9607 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
9608 i32 <len>, i32 <align>, i1 <isvolatile>)
9609 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
9610 i64 <len>, i32 <align>, i1 <isvolatile>)
9615 The '``llvm.memset.*``' intrinsics fill a block of memory with a
9616 particular byte value.
9618 Note that, unlike the standard libc function, the ``llvm.memset``
9619 intrinsic does not return a value and takes extra alignment/volatile
9620 arguments. Also, the destination can be in an arbitrary address space.
9625 The first argument is a pointer to the destination to fill, the second
9626 is the byte value with which to fill it, the third argument is an
9627 integer argument specifying the number of bytes to fill, and the fourth
9628 argument is the known alignment of the destination location.
9630 If the call to this intrinsic has an alignment value that is not 0 or 1,
9631 then the caller guarantees that the destination pointer is aligned to
9634 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
9635 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9636 very cleanly specified and it is unwise to depend on it.
9641 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
9642 at the destination location. If the argument is known to be aligned to
9643 some boundary, this can be specified as the fourth argument, otherwise
9644 it should be set to 0 or 1 (both meaning no alignment).
9646 '``llvm.sqrt.*``' Intrinsic
9647 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9652 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
9653 floating point or vector of floating point type. Not all targets support
9658 declare float @llvm.sqrt.f32(float %Val)
9659 declare double @llvm.sqrt.f64(double %Val)
9660 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
9661 declare fp128 @llvm.sqrt.f128(fp128 %Val)
9662 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
9667 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
9668 returning the same value as the libm '``sqrt``' functions would. Unlike
9669 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
9670 negative numbers other than -0.0 (which allows for better optimization,
9671 because there is no need to worry about errno being set).
9672 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
9677 The argument and return value are floating point numbers of the same
9683 This function returns the sqrt of the specified operand if it is a
9684 nonnegative floating point number.
9686 '``llvm.powi.*``' Intrinsic
9687 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9692 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
9693 floating point or vector of floating point type. Not all targets support
9698 declare float @llvm.powi.f32(float %Val, i32 %power)
9699 declare double @llvm.powi.f64(double %Val, i32 %power)
9700 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
9701 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
9702 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
9707 The '``llvm.powi.*``' intrinsics return the first operand raised to the
9708 specified (positive or negative) power. The order of evaluation of
9709 multiplications is not defined. When a vector of floating point type is
9710 used, the second argument remains a scalar integer value.
9715 The second argument is an integer power, and the first is a value to
9716 raise to that power.
9721 This function returns the first value raised to the second power with an
9722 unspecified sequence of rounding operations.
9724 '``llvm.sin.*``' Intrinsic
9725 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9730 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
9731 floating point or vector of floating point type. Not all targets support
9736 declare float @llvm.sin.f32(float %Val)
9737 declare double @llvm.sin.f64(double %Val)
9738 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
9739 declare fp128 @llvm.sin.f128(fp128 %Val)
9740 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
9745 The '``llvm.sin.*``' intrinsics return the sine of the operand.
9750 The argument and return value are floating point numbers of the same
9756 This function returns the sine of the specified operand, returning the
9757 same values as the libm ``sin`` functions would, and handles error
9758 conditions in the same way.
9760 '``llvm.cos.*``' Intrinsic
9761 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9766 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
9767 floating point or vector of floating point type. Not all targets support
9772 declare float @llvm.cos.f32(float %Val)
9773 declare double @llvm.cos.f64(double %Val)
9774 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
9775 declare fp128 @llvm.cos.f128(fp128 %Val)
9776 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
9781 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
9786 The argument and return value are floating point numbers of the same
9792 This function returns the cosine of the specified operand, returning the
9793 same values as the libm ``cos`` functions would, and handles error
9794 conditions in the same way.
9796 '``llvm.pow.*``' Intrinsic
9797 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9802 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
9803 floating point or vector of floating point type. Not all targets support
9808 declare float @llvm.pow.f32(float %Val, float %Power)
9809 declare double @llvm.pow.f64(double %Val, double %Power)
9810 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
9811 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
9812 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
9817 The '``llvm.pow.*``' intrinsics return the first operand raised to the
9818 specified (positive or negative) power.
9823 The second argument is a floating point power, and the first is a value
9824 to raise to that power.
9829 This function returns the first value raised to the second power,
9830 returning the same values as the libm ``pow`` functions would, and
9831 handles error conditions in the same way.
9833 '``llvm.exp.*``' Intrinsic
9834 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9839 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
9840 floating point or vector of floating point type. Not all targets support
9845 declare float @llvm.exp.f32(float %Val)
9846 declare double @llvm.exp.f64(double %Val)
9847 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
9848 declare fp128 @llvm.exp.f128(fp128 %Val)
9849 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
9854 The '``llvm.exp.*``' intrinsics perform the exp function.
9859 The argument and return value are floating point numbers of the same
9865 This function returns the same values as the libm ``exp`` functions
9866 would, and handles error conditions in the same way.
9868 '``llvm.exp2.*``' Intrinsic
9869 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9874 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
9875 floating point or vector of floating point type. Not all targets support
9880 declare float @llvm.exp2.f32(float %Val)
9881 declare double @llvm.exp2.f64(double %Val)
9882 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
9883 declare fp128 @llvm.exp2.f128(fp128 %Val)
9884 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
9889 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
9894 The argument and return value are floating point numbers of the same
9900 This function returns the same values as the libm ``exp2`` functions
9901 would, and handles error conditions in the same way.
9903 '``llvm.log.*``' Intrinsic
9904 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9909 This is an overloaded intrinsic. You can use ``llvm.log`` on any
9910 floating point or vector of floating point type. Not all targets support
9915 declare float @llvm.log.f32(float %Val)
9916 declare double @llvm.log.f64(double %Val)
9917 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
9918 declare fp128 @llvm.log.f128(fp128 %Val)
9919 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
9924 The '``llvm.log.*``' intrinsics perform the log function.
9929 The argument and return value are floating point numbers of the same
9935 This function returns the same values as the libm ``log`` functions
9936 would, and handles error conditions in the same way.
9938 '``llvm.log10.*``' Intrinsic
9939 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9944 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
9945 floating point or vector of floating point type. Not all targets support
9950 declare float @llvm.log10.f32(float %Val)
9951 declare double @llvm.log10.f64(double %Val)
9952 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
9953 declare fp128 @llvm.log10.f128(fp128 %Val)
9954 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
9959 The '``llvm.log10.*``' intrinsics perform the log10 function.
9964 The argument and return value are floating point numbers of the same
9970 This function returns the same values as the libm ``log10`` functions
9971 would, and handles error conditions in the same way.
9973 '``llvm.log2.*``' Intrinsic
9974 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9979 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
9980 floating point or vector of floating point type. Not all targets support
9985 declare float @llvm.log2.f32(float %Val)
9986 declare double @llvm.log2.f64(double %Val)
9987 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
9988 declare fp128 @llvm.log2.f128(fp128 %Val)
9989 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
9994 The '``llvm.log2.*``' intrinsics perform the log2 function.
9999 The argument and return value are floating point numbers of the same
10005 This function returns the same values as the libm ``log2`` functions
10006 would, and handles error conditions in the same way.
10008 '``llvm.fma.*``' Intrinsic
10009 ^^^^^^^^^^^^^^^^^^^^^^^^^^
10014 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
10015 floating point or vector of floating point type. Not all targets support
10020 declare float @llvm.fma.f32(float %a, float %b, float %c)
10021 declare double @llvm.fma.f64(double %a, double %b, double %c)
10022 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
10023 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
10024 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
10029 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
10035 The argument and return value are floating point numbers of the same
10041 This function returns the same values as the libm ``fma`` functions
10042 would, and does not set errno.
10044 '``llvm.fabs.*``' Intrinsic
10045 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10050 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
10051 floating point or vector of floating point type. Not all targets support
10056 declare float @llvm.fabs.f32(float %Val)
10057 declare double @llvm.fabs.f64(double %Val)
10058 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
10059 declare fp128 @llvm.fabs.f128(fp128 %Val)
10060 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
10065 The '``llvm.fabs.*``' intrinsics return the absolute value of the
10071 The argument and return value are floating point numbers of the same
10077 This function returns the same values as the libm ``fabs`` functions
10078 would, and handles error conditions in the same way.
10080 '``llvm.minnum.*``' Intrinsic
10081 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10086 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
10087 floating point or vector of floating point type. Not all targets support
10092 declare float @llvm.minnum.f32(float %Val0, float %Val1)
10093 declare double @llvm.minnum.f64(double %Val0, double %Val1)
10094 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10095 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
10096 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10101 The '``llvm.minnum.*``' intrinsics return the minimum of the two
10108 The arguments and return value are floating point numbers of the same
10114 Follows the IEEE-754 semantics for minNum, which also match for libm's
10117 If either operand is a NaN, returns the other non-NaN operand. Returns
10118 NaN only if both operands are NaN. If the operands compare equal,
10119 returns a value that compares equal to both operands. This means that
10120 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10122 '``llvm.maxnum.*``' Intrinsic
10123 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10128 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
10129 floating point or vector of floating point type. Not all targets support
10134 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
10135 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
10136 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
10137 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
10138 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
10143 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
10150 The arguments and return value are floating point numbers of the same
10155 Follows the IEEE-754 semantics for maxNum, which also match for libm's
10158 If either operand is a NaN, returns the other non-NaN operand. Returns
10159 NaN only if both operands are NaN. If the operands compare equal,
10160 returns a value that compares equal to both operands. This means that
10161 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
10163 '``llvm.copysign.*``' Intrinsic
10164 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10169 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
10170 floating point or vector of floating point type. Not all targets support
10175 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
10176 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
10177 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
10178 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
10179 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
10184 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
10185 first operand and the sign of the second operand.
10190 The arguments and return value are floating point numbers of the same
10196 This function returns the same values as the libm ``copysign``
10197 functions would, and handles error conditions in the same way.
10199 '``llvm.floor.*``' Intrinsic
10200 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10205 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
10206 floating point or vector of floating point type. Not all targets support
10211 declare float @llvm.floor.f32(float %Val)
10212 declare double @llvm.floor.f64(double %Val)
10213 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
10214 declare fp128 @llvm.floor.f128(fp128 %Val)
10215 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
10220 The '``llvm.floor.*``' intrinsics return the floor of the operand.
10225 The argument and return value are floating point numbers of the same
10231 This function returns the same values as the libm ``floor`` functions
10232 would, and handles error conditions in the same way.
10234 '``llvm.ceil.*``' Intrinsic
10235 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10240 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
10241 floating point or vector of floating point type. Not all targets support
10246 declare float @llvm.ceil.f32(float %Val)
10247 declare double @llvm.ceil.f64(double %Val)
10248 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
10249 declare fp128 @llvm.ceil.f128(fp128 %Val)
10250 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
10255 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
10260 The argument and return value are floating point numbers of the same
10266 This function returns the same values as the libm ``ceil`` functions
10267 would, and handles error conditions in the same way.
10269 '``llvm.trunc.*``' Intrinsic
10270 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10275 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
10276 floating point or vector of floating point type. Not all targets support
10281 declare float @llvm.trunc.f32(float %Val)
10282 declare double @llvm.trunc.f64(double %Val)
10283 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
10284 declare fp128 @llvm.trunc.f128(fp128 %Val)
10285 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
10290 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
10291 nearest integer not larger in magnitude than the operand.
10296 The argument and return value are floating point numbers of the same
10302 This function returns the same values as the libm ``trunc`` functions
10303 would, and handles error conditions in the same way.
10305 '``llvm.rint.*``' Intrinsic
10306 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10311 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
10312 floating point or vector of floating point type. Not all targets support
10317 declare float @llvm.rint.f32(float %Val)
10318 declare double @llvm.rint.f64(double %Val)
10319 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
10320 declare fp128 @llvm.rint.f128(fp128 %Val)
10321 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
10326 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
10327 nearest integer. It may raise an inexact floating-point exception if the
10328 operand isn't an integer.
10333 The argument and return value are floating point numbers of the same
10339 This function returns the same values as the libm ``rint`` functions
10340 would, and handles error conditions in the same way.
10342 '``llvm.nearbyint.*``' Intrinsic
10343 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10348 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
10349 floating point or vector of floating point type. Not all targets support
10354 declare float @llvm.nearbyint.f32(float %Val)
10355 declare double @llvm.nearbyint.f64(double %Val)
10356 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
10357 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
10358 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
10363 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
10369 The argument and return value are floating point numbers of the same
10375 This function returns the same values as the libm ``nearbyint``
10376 functions would, and handles error conditions in the same way.
10378 '``llvm.round.*``' Intrinsic
10379 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10384 This is an overloaded intrinsic. You can use ``llvm.round`` on any
10385 floating point or vector of floating point type. Not all targets support
10390 declare float @llvm.round.f32(float %Val)
10391 declare double @llvm.round.f64(double %Val)
10392 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
10393 declare fp128 @llvm.round.f128(fp128 %Val)
10394 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
10399 The '``llvm.round.*``' intrinsics returns the operand rounded to the
10405 The argument and return value are floating point numbers of the same
10411 This function returns the same values as the libm ``round``
10412 functions would, and handles error conditions in the same way.
10414 Bit Manipulation Intrinsics
10415 ---------------------------
10417 LLVM provides intrinsics for a few important bit manipulation
10418 operations. These allow efficient code generation for some algorithms.
10420 '``llvm.bswap.*``' Intrinsics
10421 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10426 This is an overloaded intrinsic function. You can use bswap on any
10427 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
10431 declare i16 @llvm.bswap.i16(i16 <id>)
10432 declare i32 @llvm.bswap.i32(i32 <id>)
10433 declare i64 @llvm.bswap.i64(i64 <id>)
10438 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
10439 values with an even number of bytes (positive multiple of 16 bits).
10440 These are useful for performing operations on data that is not in the
10441 target's native byte order.
10446 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
10447 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
10448 intrinsic returns an i32 value that has the four bytes of the input i32
10449 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
10450 returned i32 will have its bytes in 3, 2, 1, 0 order. The
10451 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
10452 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
10455 '``llvm.ctpop.*``' Intrinsic
10456 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10461 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
10462 bit width, or on any vector with integer elements. Not all targets
10463 support all bit widths or vector types, however.
10467 declare i8 @llvm.ctpop.i8(i8 <src>)
10468 declare i16 @llvm.ctpop.i16(i16 <src>)
10469 declare i32 @llvm.ctpop.i32(i32 <src>)
10470 declare i64 @llvm.ctpop.i64(i64 <src>)
10471 declare i256 @llvm.ctpop.i256(i256 <src>)
10472 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
10477 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
10483 The only argument is the value to be counted. The argument may be of any
10484 integer type, or a vector with integer elements. The return type must
10485 match the argument type.
10490 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
10491 each element of a vector.
10493 '``llvm.ctlz.*``' Intrinsic
10494 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10499 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
10500 integer bit width, or any vector whose elements are integers. Not all
10501 targets support all bit widths or vector types, however.
10505 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
10506 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
10507 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
10508 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
10509 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
10510 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10515 The '``llvm.ctlz``' family of intrinsic functions counts the number of
10516 leading zeros in a variable.
10521 The first argument is the value to be counted. This argument may be of
10522 any integer type, or a vector with integer element type. The return
10523 type must match the first argument type.
10525 The second argument must be a constant and is a flag to indicate whether
10526 the intrinsic should ensure that a zero as the first argument produces a
10527 defined result. Historically some architectures did not provide a
10528 defined result for zero values as efficiently, and many algorithms are
10529 now predicated on avoiding zero-value inputs.
10534 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
10535 zeros in a variable, or within each element of the vector. If
10536 ``src == 0`` then the result is the size in bits of the type of ``src``
10537 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10538 ``llvm.ctlz(i32 2) = 30``.
10540 '``llvm.cttz.*``' Intrinsic
10541 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10546 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
10547 integer bit width, or any vector of integer elements. Not all targets
10548 support all bit widths or vector types, however.
10552 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
10553 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
10554 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
10555 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
10556 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
10557 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10562 The '``llvm.cttz``' family of intrinsic functions counts the number of
10568 The first argument is the value to be counted. This argument may be of
10569 any integer type, or a vector with integer element type. The return
10570 type must match the first argument type.
10572 The second argument must be a constant and is a flag to indicate whether
10573 the intrinsic should ensure that a zero as the first argument produces a
10574 defined result. Historically some architectures did not provide a
10575 defined result for zero values as efficiently, and many algorithms are
10576 now predicated on avoiding zero-value inputs.
10581 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
10582 zeros in a variable, or within each element of a vector. If ``src == 0``
10583 then the result is the size in bits of the type of ``src`` if
10584 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10585 ``llvm.cttz(2) = 1``.
10589 Arithmetic with Overflow Intrinsics
10590 -----------------------------------
10592 LLVM provides intrinsics for some arithmetic with overflow operations.
10594 '``llvm.sadd.with.overflow.*``' Intrinsics
10595 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10600 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
10601 on any integer bit width.
10605 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
10606 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10607 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
10612 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10613 a signed addition of the two arguments, and indicate whether an overflow
10614 occurred during the signed summation.
10619 The arguments (%a and %b) and the first element of the result structure
10620 may be of integer types of any bit width, but they must have the same
10621 bit width. The second element of the result structure must be of type
10622 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10628 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10629 a signed addition of the two variables. They return a structure --- the
10630 first element of which is the signed summation, and the second element
10631 of which is a bit specifying if the signed summation resulted in an
10637 .. code-block:: llvm
10639 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10640 %sum = extractvalue {i32, i1} %res, 0
10641 %obit = extractvalue {i32, i1} %res, 1
10642 br i1 %obit, label %overflow, label %normal
10644 '``llvm.uadd.with.overflow.*``' Intrinsics
10645 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10650 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
10651 on any integer bit width.
10655 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
10656 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10657 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
10662 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10663 an unsigned addition of the two arguments, and indicate whether a carry
10664 occurred during the unsigned summation.
10669 The arguments (%a and %b) and the first element of the result structure
10670 may be of integer types of any bit width, but they must have the same
10671 bit width. The second element of the result structure must be of type
10672 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10678 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10679 an unsigned addition of the two arguments. They return a structure --- the
10680 first element of which is the sum, and the second element of which is a
10681 bit specifying if the unsigned summation resulted in a carry.
10686 .. code-block:: llvm
10688 %res = call {i32, i1} @llvm.uadd.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 %carry, label %normal
10693 '``llvm.ssub.with.overflow.*``' Intrinsics
10694 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10699 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
10700 on any integer bit width.
10704 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
10705 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10706 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
10711 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10712 a signed subtraction of the two arguments, and indicate whether an
10713 overflow occurred during the signed subtraction.
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 signed
10727 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10728 a signed subtraction of the two arguments. They return a structure --- the
10729 first element of which is the subtraction, and the second element of
10730 which is a bit specifying if the signed subtraction resulted in an
10736 .. code-block:: llvm
10738 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10739 %sum = extractvalue {i32, i1} %res, 0
10740 %obit = extractvalue {i32, i1} %res, 1
10741 br i1 %obit, label %overflow, label %normal
10743 '``llvm.usub.with.overflow.*``' Intrinsics
10744 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10749 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
10750 on any integer bit width.
10754 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
10755 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10756 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
10761 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10762 an unsigned subtraction of the two arguments, and indicate whether an
10763 overflow occurred during the unsigned subtraction.
10768 The arguments (%a and %b) and the first element of the result structure
10769 may be of integer types of any bit width, but they must have the same
10770 bit width. The second element of the result structure must be of type
10771 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10777 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10778 an unsigned subtraction of the two arguments. They return a structure ---
10779 the first element of which is the subtraction, and the second element of
10780 which is a bit specifying if the unsigned subtraction resulted in an
10786 .. code-block:: llvm
10788 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10789 %sum = extractvalue {i32, i1} %res, 0
10790 %obit = extractvalue {i32, i1} %res, 1
10791 br i1 %obit, label %overflow, label %normal
10793 '``llvm.smul.with.overflow.*``' Intrinsics
10794 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10799 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
10800 on any integer bit width.
10804 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
10805 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10806 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
10811 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10812 a signed multiplication of the two arguments, and indicate whether an
10813 overflow occurred during the signed multiplication.
10818 The arguments (%a and %b) and the first element of the result structure
10819 may be of integer types of any bit width, but they must have the same
10820 bit width. The second element of the result structure must be of type
10821 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10827 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10828 a signed multiplication of the two arguments. They return a structure ---
10829 the first element of which is the multiplication, and the second element
10830 of which is a bit specifying if the signed multiplication resulted in an
10836 .. code-block:: llvm
10838 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10839 %sum = extractvalue {i32, i1} %res, 0
10840 %obit = extractvalue {i32, i1} %res, 1
10841 br i1 %obit, label %overflow, label %normal
10843 '``llvm.umul.with.overflow.*``' Intrinsics
10844 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10849 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
10850 on any integer bit width.
10854 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
10855 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10856 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
10861 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10862 a unsigned multiplication of the two arguments, and indicate whether an
10863 overflow occurred during the unsigned multiplication.
10868 The arguments (%a and %b) and the first element of the result structure
10869 may be of integer types of any bit width, but they must have the same
10870 bit width. The second element of the result structure must be of type
10871 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10877 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10878 an unsigned multiplication of the two arguments. They return a structure ---
10879 the first element of which is the multiplication, and the second
10880 element of which is a bit specifying if the unsigned multiplication
10881 resulted in an overflow.
10886 .. code-block:: llvm
10888 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10889 %sum = extractvalue {i32, i1} %res, 0
10890 %obit = extractvalue {i32, i1} %res, 1
10891 br i1 %obit, label %overflow, label %normal
10893 Specialised Arithmetic Intrinsics
10894 ---------------------------------
10896 '``llvm.canonicalize.*``' Intrinsic
10897 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10904 declare float @llvm.canonicalize.f32(float %a)
10905 declare double @llvm.canonicalize.f64(double %b)
10910 The '``llvm.canonicalize.*``' intrinsic returns the platform specific canonical
10911 encoding of a floating point number. This canonicalization is useful for
10912 implementing certain numeric primitives such as frexp. The canonical encoding is
10913 defined by IEEE-754-2008 to be:
10917 2.1.8 canonical encoding: The preferred encoding of a floating-point
10918 representation in a format. Applied to declets, significands of finite
10919 numbers, infinities, and NaNs, especially in decimal formats.
10921 This operation can also be considered equivalent to the IEEE-754-2008
10922 conversion of a floating-point value to the same format. NaNs are handled
10923 according to section 6.2.
10925 Examples of non-canonical encodings:
10927 - x87 pseudo denormals, pseudo NaNs, pseudo Infinity, Unnormals. These are
10928 converted to a canonical representation per hardware-specific protocol.
10929 - Many normal decimal floating point numbers have non-canonical alternative
10931 - Some machines, like GPUs or ARMv7 NEON, do not support subnormal values.
10932 These are treated as non-canonical encodings of zero and with be flushed to
10933 a zero of the same sign by this operation.
10935 Note that per IEEE-754-2008 6.2, systems that support signaling NaNs with
10936 default exception handling must signal an invalid exception, and produce a
10939 This function should always be implementable as multiplication by 1.0, provided
10940 that the compiler does not constant fold the operation. Likewise, division by
10941 1.0 and ``llvm.minnum(x, x)`` are possible implementations. Addition with
10942 -0.0 is also sufficient provided that the rounding mode is not -Infinity.
10944 ``@llvm.canonicalize`` must preserve the equality relation. That is:
10946 - ``(@llvm.canonicalize(x) == x)`` is equivalent to ``(x == x)``
10947 - ``(@llvm.canonicalize(x) == @llvm.canonicalize(y))`` is equivalent to
10950 Additionally, the sign of zero must be conserved:
10951 ``@llvm.canonicalize(-0.0) = -0.0`` and ``@llvm.canonicalize(+0.0) = +0.0``
10953 The payload bits of a NaN must be conserved, with two exceptions.
10954 First, environments which use only a single canonical representation of NaN
10955 must perform said canonicalization. Second, SNaNs must be quieted per the
10958 The canonicalization operation may be optimized away if:
10960 - The input is known to be canonical. For example, it was produced by a
10961 floating-point operation that is required by the standard to be canonical.
10962 - The result is consumed only by (or fused with) other floating-point
10963 operations. That is, the bits of the floating point value are not examined.
10965 '``llvm.fmuladd.*``' Intrinsic
10966 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10973 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
10974 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
10979 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
10980 expressions that can be fused if the code generator determines that (a) the
10981 target instruction set has support for a fused operation, and (b) that the
10982 fused operation is more efficient than the equivalent, separate pair of mul
10983 and add instructions.
10988 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
10989 multiplicands, a and b, and an addend c.
10998 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
11000 is equivalent to the expression a \* b + c, except that rounding will
11001 not be performed between the multiplication and addition steps if the
11002 code generator fuses the operations. Fusion is not guaranteed, even if
11003 the target platform supports it. If a fused multiply-add is required the
11004 corresponding llvm.fma.\* intrinsic function should be used
11005 instead. This never sets errno, just as '``llvm.fma.*``'.
11010 .. code-block:: llvm
11012 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
11015 '``llvm.uabsdiff.*``' and '``llvm.sabsdiff.*``' Intrinsics
11016 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11020 This is an overloaded intrinsic. The loaded data is a vector of any integer bit width.
11022 .. code-block:: llvm
11024 declare <4 x integer> @llvm.uabsdiff.v4i32(<4 x integer> %a, <4 x integer> %b)
11030 The ``llvm.uabsdiff`` intrinsic returns a vector result of the absolute difference
11031 of the two operands, treating them both as unsigned integers. The intermediate
11032 calculations are computed using infinitely precise unsigned arithmetic. The final
11033 result will be truncated to the given type.
11035 The ``llvm.sabsdiff`` intrinsic returns a vector result of the absolute difference of
11036 the two operands, treating them both as signed integers. If the result overflows, the
11037 behavior is undefined.
11041 These intrinsics are primarily used during the code generation stage of compilation.
11042 They are generated by compiler passes such as the Loop and SLP vectorizers. It is not
11043 recommended for users to create them manually.
11048 Both intrinsics take two integer of the same bitwidth.
11055 call <4 x i32> @llvm.uabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
11059 %1 = zext <4 x i32> %a to <4 x i64>
11060 %2 = zext <4 x i32> %b to <4 x i64>
11061 %sub = sub <4 x i64> %1, %2
11062 %trunc = trunc <4 x i64> to <4 x i32>
11064 and the expression::
11066 call <4 x i32> @llvm.sabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
11070 %sub = sub nsw <4 x i32> %a, %b
11071 %ispos = icmp sge <4 x i32> %sub, zeroinitializer
11072 %neg = sub nsw <4 x i32> zeroinitializer, %sub
11073 %1 = select <4 x i1> %ispos, <4 x i32> %sub, <4 x i32> %neg
11076 Half Precision Floating Point Intrinsics
11077 ----------------------------------------
11079 For most target platforms, half precision floating point is a
11080 storage-only format. This means that it is a dense encoding (in memory)
11081 but does not support computation in the format.
11083 This means that code must first load the half-precision floating point
11084 value as an i16, then convert it to float with
11085 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
11086 then be performed on the float value (including extending to double
11087 etc). To store the value back to memory, it is first converted to float
11088 if needed, then converted to i16 with
11089 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
11092 .. _int_convert_to_fp16:
11094 '``llvm.convert.to.fp16``' Intrinsic
11095 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11102 declare i16 @llvm.convert.to.fp16.f32(float %a)
11103 declare i16 @llvm.convert.to.fp16.f64(double %a)
11108 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11109 conventional floating point type to half precision floating point format.
11114 The intrinsic function contains single argument - the value to be
11120 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
11121 conventional floating point format to half precision floating point format. The
11122 return value is an ``i16`` which contains the converted number.
11127 .. code-block:: llvm
11129 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
11130 store i16 %res, i16* @x, align 2
11132 .. _int_convert_from_fp16:
11134 '``llvm.convert.from.fp16``' Intrinsic
11135 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11142 declare float @llvm.convert.from.fp16.f32(i16 %a)
11143 declare double @llvm.convert.from.fp16.f64(i16 %a)
11148 The '``llvm.convert.from.fp16``' intrinsic function performs a
11149 conversion from half precision floating point format to single precision
11150 floating point format.
11155 The intrinsic function contains single argument - the value to be
11161 The '``llvm.convert.from.fp16``' intrinsic function performs a
11162 conversion from half single precision floating point format to single
11163 precision floating point format. The input half-float value is
11164 represented by an ``i16`` value.
11169 .. code-block:: llvm
11171 %a = load i16, i16* @x, align 2
11172 %res = call float @llvm.convert.from.fp16(i16 %a)
11174 .. _dbg_intrinsics:
11176 Debugger Intrinsics
11177 -------------------
11179 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
11180 prefix), are described in the `LLVM Source Level
11181 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
11184 Exception Handling Intrinsics
11185 -----------------------------
11187 The LLVM exception handling intrinsics (which all start with
11188 ``llvm.eh.`` prefix), are described in the `LLVM Exception
11189 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
11191 .. _int_trampoline:
11193 Trampoline Intrinsics
11194 ---------------------
11196 These intrinsics make it possible to excise one parameter, marked with
11197 the :ref:`nest <nest>` attribute, from a function. The result is a
11198 callable function pointer lacking the nest parameter - the caller does
11199 not need to provide a value for it. Instead, the value to use is stored
11200 in advance in a "trampoline", a block of memory usually allocated on the
11201 stack, which also contains code to splice the nest value into the
11202 argument list. This is used to implement the GCC nested function address
11205 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
11206 then the resulting function pointer has signature ``i32 (i32, i32)*``.
11207 It can be created as follows:
11209 .. code-block:: llvm
11211 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
11212 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
11213 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
11214 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
11215 %fp = bitcast i8* %p to i32 (i32, i32)*
11217 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
11218 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
11222 '``llvm.init.trampoline``' Intrinsic
11223 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11230 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
11235 This fills the memory pointed to by ``tramp`` with executable code,
11236 turning it into a trampoline.
11241 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
11242 pointers. The ``tramp`` argument must point to a sufficiently large and
11243 sufficiently aligned block of memory; this memory is written to by the
11244 intrinsic. Note that the size and the alignment are target-specific -
11245 LLVM currently provides no portable way of determining them, so a
11246 front-end that generates this intrinsic needs to have some
11247 target-specific knowledge. The ``func`` argument must hold a function
11248 bitcast to an ``i8*``.
11253 The block of memory pointed to by ``tramp`` is filled with target
11254 dependent code, turning it into a function. Then ``tramp`` needs to be
11255 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
11256 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
11257 function's signature is the same as that of ``func`` with any arguments
11258 marked with the ``nest`` attribute removed. At most one such ``nest``
11259 argument is allowed, and it must be of pointer type. Calling the new
11260 function is equivalent to calling ``func`` with the same argument list,
11261 but with ``nval`` used for the missing ``nest`` argument. If, after
11262 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
11263 modified, then the effect of any later call to the returned function
11264 pointer is undefined.
11268 '``llvm.adjust.trampoline``' Intrinsic
11269 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11276 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
11281 This performs any required machine-specific adjustment to the address of
11282 a trampoline (passed as ``tramp``).
11287 ``tramp`` must point to a block of memory which already has trampoline
11288 code filled in by a previous call to
11289 :ref:`llvm.init.trampoline <int_it>`.
11294 On some architectures the address of the code to be executed needs to be
11295 different than the address where the trampoline is actually stored. This
11296 intrinsic returns the executable address corresponding to ``tramp``
11297 after performing the required machine specific adjustments. The pointer
11298 returned can then be :ref:`bitcast and executed <int_trampoline>`.
11300 .. _int_mload_mstore:
11302 Masked Vector Load and Store Intrinsics
11303 ---------------------------------------
11305 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.
11309 '``llvm.masked.load.*``' Intrinsics
11310 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11314 This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
11318 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11319 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11324 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.
11330 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.
11336 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.
11337 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.
11342 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
11344 ;; The result of the two following instructions is identical aside from potential memory access exception
11345 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
11346 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
11350 '``llvm.masked.store.*``' Intrinsics
11351 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11355 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
11359 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
11360 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
11365 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.
11370 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.
11376 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.
11377 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.
11381 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
11383 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
11384 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
11385 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
11386 store <16 x float> %res, <16 x float>* %ptr, align 4
11389 Masked Vector Gather and Scatter Intrinsics
11390 -------------------------------------------
11392 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.
11396 '``llvm.masked.gather.*``' Intrinsics
11397 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11401 This is an overloaded intrinsic. The loaded data are multiple scalar values of any integer or floating point data type gathered together into one vector.
11405 declare <16 x float> @llvm.masked.gather.v16f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11406 declare <2 x double> @llvm.masked.gather.v2f64 (<2 x double*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11411 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.
11417 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.
11423 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.
11424 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.
11429 %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>)
11431 ;; The gather with all-true mask is equivalent to the following instruction sequence
11432 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
11433 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
11434 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
11435 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
11437 %val0 = load double, double* %ptr0, align 8
11438 %val1 = load double, double* %ptr1, align 8
11439 %val2 = load double, double* %ptr2, align 8
11440 %val3 = load double, double* %ptr3, align 8
11442 %vec0 = insertelement <4 x double>undef, %val0, 0
11443 %vec01 = insertelement <4 x double>%vec0, %val1, 1
11444 %vec012 = insertelement <4 x double>%vec01, %val2, 2
11445 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
11449 '``llvm.masked.scatter.*``' Intrinsics
11450 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11454 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type. Each vector element is stored in an arbitrary memory addresses. Scatter with overlapping addresses is guaranteed to be ordered from least-significant to most-significant element.
11458 declare void @llvm.masked.scatter.v8i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
11459 declare void @llvm.masked.scatter.v16f32(<16 x i32> <value>, <16 x i32*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
11464 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.
11469 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.
11475 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.
11479 ;; This instruction unconditionaly stores data vector in multiple addresses
11480 call @llvm.masked.scatter.v8i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
11482 ;; It is equivalent to a list of scalar stores
11483 %val0 = extractelement <8 x i32> %value, i32 0
11484 %val1 = extractelement <8 x i32> %value, i32 1
11486 %val7 = extractelement <8 x i32> %value, i32 7
11487 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
11488 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
11490 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
11491 ;; Note: the order of the following stores is important when they overlap:
11492 store i32 %val0, i32* %ptr0, align 4
11493 store i32 %val1, i32* %ptr1, align 4
11495 store i32 %val7, i32* %ptr7, align 4
11501 This class of intrinsics provides information about the lifetime of
11502 memory objects and ranges where variables are immutable.
11506 '``llvm.lifetime.start``' Intrinsic
11507 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11514 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
11519 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
11525 The first argument is a constant integer representing the size of the
11526 object, or -1 if it is variable sized. The second argument is a pointer
11532 This intrinsic indicates that before this point in the code, the value
11533 of the memory pointed to by ``ptr`` is dead. This means that it is known
11534 to never be used and has an undefined value. A load from the pointer
11535 that precedes this intrinsic can be replaced with ``'undef'``.
11539 '``llvm.lifetime.end``' Intrinsic
11540 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11547 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
11552 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
11558 The first argument is a constant integer representing the size of the
11559 object, or -1 if it is variable sized. The second argument is a pointer
11565 This intrinsic indicates that after this point in the code, the value of
11566 the memory pointed to by ``ptr`` is dead. This means that it is known to
11567 never be used and has an undefined value. Any stores into the memory
11568 object following this intrinsic may be removed as dead.
11570 '``llvm.invariant.start``' Intrinsic
11571 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11578 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
11583 The '``llvm.invariant.start``' intrinsic specifies that the contents of
11584 a memory object will not change.
11589 The first argument is a constant integer representing the size of the
11590 object, or -1 if it is variable sized. The second argument is a pointer
11596 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
11597 the return value, the referenced memory location is constant and
11600 '``llvm.invariant.end``' Intrinsic
11601 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11608 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
11613 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
11614 memory object are mutable.
11619 The first argument is the matching ``llvm.invariant.start`` intrinsic.
11620 The second argument is a constant integer representing the size of the
11621 object, or -1 if it is variable sized and the third argument is a
11622 pointer to the object.
11627 This intrinsic indicates that the memory is mutable again.
11629 '``llvm.invariant.group.barrier``' Intrinsic
11630 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11637 declare i8* @llvm.invariant.group.barrier(i8* <ptr>)
11642 The '``llvm.invariant.group.barrier``' intrinsic can be used when an invariant
11643 established by invariant.group metadata no longer holds, to obtain a new pointer
11644 value that does not carry the invariant information.
11650 The ``llvm.invariant.group.barrier`` takes only one argument, which is
11651 the pointer to the memory for which the ``invariant.group`` no longer holds.
11656 Returns another pointer that aliases its argument but which is considered different
11657 for the purposes of ``load``/``store`` ``invariant.group`` metadata.
11662 This class of intrinsics is designed to be generic and has no specific
11665 '``llvm.var.annotation``' Intrinsic
11666 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11673 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11678 The '``llvm.var.annotation``' intrinsic.
11683 The first argument is a pointer to a value, the second is a pointer to a
11684 global string, the third is a pointer to a global string which is the
11685 source file name, and the last argument is the line number.
11690 This intrinsic allows annotation of local variables with arbitrary
11691 strings. This can be useful for special purpose optimizations that want
11692 to look for these annotations. These have no other defined use; they are
11693 ignored by code generation and optimization.
11695 '``llvm.ptr.annotation.*``' Intrinsic
11696 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11701 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
11702 pointer to an integer of any width. *NOTE* you must specify an address space for
11703 the pointer. The identifier for the default address space is the integer
11708 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11709 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
11710 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
11711 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
11712 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
11717 The '``llvm.ptr.annotation``' intrinsic.
11722 The first argument is a pointer to an integer value of arbitrary bitwidth
11723 (result of some expression), the second is a pointer to a global string, the
11724 third is a pointer to a global string which is the source file name, and the
11725 last argument is the line number. It returns the value of the first argument.
11730 This intrinsic allows annotation of a pointer to an integer with arbitrary
11731 strings. This can be useful for special purpose optimizations that want to look
11732 for these annotations. These have no other defined use; they are ignored by code
11733 generation and optimization.
11735 '``llvm.annotation.*``' Intrinsic
11736 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11741 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
11742 any integer bit width.
11746 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
11747 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
11748 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
11749 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
11750 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
11755 The '``llvm.annotation``' intrinsic.
11760 The first argument is an integer value (result of some expression), the
11761 second is a pointer to a global string, the third is a pointer to a
11762 global string which is the source file name, and the last argument is
11763 the line number. It returns the value of the first argument.
11768 This intrinsic allows annotations to be put on arbitrary expressions
11769 with arbitrary strings. This can be useful for special purpose
11770 optimizations that want to look for these annotations. These have no
11771 other defined use; they are ignored by code generation and optimization.
11773 '``llvm.trap``' Intrinsic
11774 ^^^^^^^^^^^^^^^^^^^^^^^^^
11781 declare void @llvm.trap() noreturn nounwind
11786 The '``llvm.trap``' intrinsic.
11796 This intrinsic is lowered to the target dependent trap instruction. If
11797 the target does not have a trap instruction, this intrinsic will be
11798 lowered to a call of the ``abort()`` function.
11800 '``llvm.debugtrap``' Intrinsic
11801 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11808 declare void @llvm.debugtrap() nounwind
11813 The '``llvm.debugtrap``' intrinsic.
11823 This intrinsic is lowered to code which is intended to cause an
11824 execution trap with the intention of requesting the attention of a
11827 '``llvm.stackprotector``' Intrinsic
11828 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11835 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
11840 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
11841 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
11842 is placed on the stack before local variables.
11847 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
11848 The first argument is the value loaded from the stack guard
11849 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
11850 enough space to hold the value of the guard.
11855 This intrinsic causes the prologue/epilogue inserter to force the position of
11856 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
11857 to ensure that if a local variable on the stack is overwritten, it will destroy
11858 the value of the guard. When the function exits, the guard on the stack is
11859 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
11860 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
11861 calling the ``__stack_chk_fail()`` function.
11863 '``llvm.stackprotectorcheck``' Intrinsic
11864 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11871 declare void @llvm.stackprotectorcheck(i8** <guard>)
11876 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
11877 created stack protector and if they are not equal calls the
11878 ``__stack_chk_fail()`` function.
11883 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
11884 the variable ``@__stack_chk_guard``.
11889 This intrinsic is provided to perform the stack protector check by comparing
11890 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
11891 values do not match call the ``__stack_chk_fail()`` function.
11893 The reason to provide this as an IR level intrinsic instead of implementing it
11894 via other IR operations is that in order to perform this operation at the IR
11895 level without an intrinsic, one would need to create additional basic blocks to
11896 handle the success/failure cases. This makes it difficult to stop the stack
11897 protector check from disrupting sibling tail calls in Codegen. With this
11898 intrinsic, we are able to generate the stack protector basic blocks late in
11899 codegen after the tail call decision has occurred.
11901 '``llvm.objectsize``' Intrinsic
11902 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11909 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
11910 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
11915 The ``llvm.objectsize`` intrinsic is designed to provide information to
11916 the optimizers to determine at compile time whether a) an operation
11917 (like memcpy) will overflow a buffer that corresponds to an object, or
11918 b) that a runtime check for overflow isn't necessary. An object in this
11919 context means an allocation of a specific class, structure, array, or
11925 The ``llvm.objectsize`` intrinsic takes two arguments. The first
11926 argument is a pointer to or into the ``object``. The second argument is
11927 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
11928 or -1 (if false) when the object size is unknown. The second argument
11929 only accepts constants.
11934 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
11935 the size of the object concerned. If the size cannot be determined at
11936 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
11937 on the ``min`` argument).
11939 '``llvm.expect``' Intrinsic
11940 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
11945 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
11950 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
11951 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
11952 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
11957 The ``llvm.expect`` intrinsic provides information about expected (the
11958 most probable) value of ``val``, which can be used by optimizers.
11963 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
11964 a value. The second argument is an expected value, this needs to be a
11965 constant value, variables are not allowed.
11970 This intrinsic is lowered to the ``val``.
11974 '``llvm.assume``' Intrinsic
11975 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11982 declare void @llvm.assume(i1 %cond)
11987 The ``llvm.assume`` allows the optimizer to assume that the provided
11988 condition is true. This information can then be used in simplifying other parts
11994 The condition which the optimizer may assume is always true.
11999 The intrinsic allows the optimizer to assume that the provided condition is
12000 always true whenever the control flow reaches the intrinsic call. No code is
12001 generated for this intrinsic, and instructions that contribute only to the
12002 provided condition are not used for code generation. If the condition is
12003 violated during execution, the behavior is undefined.
12005 Note that the optimizer might limit the transformations performed on values
12006 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
12007 only used to form the intrinsic's input argument. This might prove undesirable
12008 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
12009 sufficient overall improvement in code quality. For this reason,
12010 ``llvm.assume`` should not be used to document basic mathematical invariants
12011 that the optimizer can otherwise deduce or facts that are of little use to the
12016 '``llvm.bitset.test``' Intrinsic
12017 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12024 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
12030 The first argument is a pointer to be tested. The second argument is a
12031 metadata object representing an identifier for a :doc:`bitset <BitSets>`.
12036 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
12037 member of the given bitset.
12039 '``llvm.donothing``' Intrinsic
12040 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
12047 declare void @llvm.donothing() nounwind readnone
12052 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
12053 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
12054 with an invoke instruction.
12064 This intrinsic does nothing, and it's removed by optimizers and ignored
12067 Stack Map Intrinsics
12068 --------------------
12070 LLVM provides experimental intrinsics to support runtime patching
12071 mechanisms commonly desired in dynamic language JITs. These intrinsics
12072 are described in :doc:`StackMaps`.