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 a alias, but then it only governs how
495 the alias is accessed. It will not have any effect in the aliasee.
502 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
503 types <t_struct>`. Literal types are uniqued structurally, but identified types
504 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
505 to forward declare a type that is not yet available.
507 An example of a identified structure specification is:
511 %mytype = type { %mytype*, i32 }
513 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
514 literal types are uniqued in recent versions of LLVM.
521 Global variables define regions of memory allocated at compilation time
524 Global variable definitions must be initialized.
526 Global variables in other translation units can also be declared, in which
527 case they don't have an initializer.
529 Either global variable definitions or declarations may have an explicit section
530 to be placed in and may have an optional explicit alignment specified.
532 A variable may be defined as a global ``constant``, which indicates that
533 the contents of the variable will **never** be modified (enabling better
534 optimization, allowing the global data to be placed in the read-only
535 section of an executable, etc). Note that variables that need runtime
536 initialization cannot be marked ``constant`` as there is a store to the
539 LLVM explicitly allows *declarations* of global variables to be marked
540 constant, even if the final definition of the global is not. This
541 capability can be used to enable slightly better optimization of the
542 program, but requires the language definition to guarantee that
543 optimizations based on the 'constantness' are valid for the translation
544 units that do not include the definition.
546 As SSA values, global variables define pointer values that are in scope
547 (i.e. they dominate) all basic blocks in the program. Global variables
548 always define a pointer to their "content" type because they describe a
549 region of memory, and all memory objects in LLVM are accessed through
552 Global variables can be marked with ``unnamed_addr`` which indicates
553 that the address is not significant, only the content. Constants marked
554 like this can be merged with other constants if they have the same
555 initializer. Note that a constant with significant address *can* be
556 merged with a ``unnamed_addr`` constant, the result being a constant
557 whose address is significant.
559 A global variable may be declared to reside in a target-specific
560 numbered address space. For targets that support them, address spaces
561 may affect how optimizations are performed and/or what target
562 instructions are used to access the variable. The default address space
563 is zero. The address space qualifier must precede any other attributes.
565 LLVM allows an explicit section to be specified for globals. If the
566 target supports it, it will emit globals to the section specified.
567 Additionally, the global can placed in a comdat if the target has the necessary
570 By default, global initializers are optimized by assuming that global
571 variables defined within the module are not modified from their
572 initial values before the start of the global initializer. This is
573 true even for variables potentially accessible from outside the
574 module, including those with external linkage or appearing in
575 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
576 by marking the variable with ``externally_initialized``.
578 An explicit alignment may be specified for a global, which must be a
579 power of 2. If not present, or if the alignment is set to zero, the
580 alignment of the global is set by the target to whatever it feels
581 convenient. If an explicit alignment is specified, the global is forced
582 to have exactly that alignment. Targets and optimizers are not allowed
583 to over-align the global if the global has an assigned section. In this
584 case, the extra alignment could be observable: for example, code could
585 assume that the globals are densely packed in their section and try to
586 iterate over them as an array, alignment padding would break this
587 iteration. The maximum alignment is ``1 << 29``.
589 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
591 Variables and aliases can have a
592 :ref:`Thread Local Storage Model <tls_model>`.
596 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
597 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
598 <global | constant> <Type> [<InitializerConstant>]
599 [, section "name"] [, comdat [($name)]]
600 [, align <Alignment>]
602 For example, the following defines a global in a numbered address space
603 with an initializer, section, and alignment:
607 @G = addrspace(5) constant float 1.0, section "foo", align 4
609 The following example just declares a global variable
613 @G = external global i32
615 The following example defines a thread-local global with the
616 ``initialexec`` TLS model:
620 @G = thread_local(initialexec) global i32 0, align 4
622 .. _functionstructure:
627 LLVM function definitions consist of the "``define``" keyword, an
628 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
629 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
630 an optional :ref:`calling convention <callingconv>`,
631 an optional ``unnamed_addr`` attribute, a return type, an optional
632 :ref:`parameter attribute <paramattrs>` for the return type, a function
633 name, a (possibly empty) argument list (each with optional :ref:`parameter
634 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
635 an optional section, an optional alignment,
636 an optional :ref:`comdat <langref_comdats>`,
637 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
638 an optional :ref:`prologue <prologuedata>`,
639 an optional :ref:`personality <personalityfn>`,
640 an opening curly brace, a list of basic blocks, and a closing curly brace.
642 LLVM function declarations consist of the "``declare``" keyword, an
643 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
644 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
645 an optional :ref:`calling convention <callingconv>`,
646 an optional ``unnamed_addr`` attribute, a return type, an optional
647 :ref:`parameter attribute <paramattrs>` for the return type, a function
648 name, a possibly empty list of arguments, an optional alignment, an optional
649 :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
650 and an optional :ref:`prologue <prologuedata>`.
652 A function definition contains a list of basic blocks, forming the CFG (Control
653 Flow Graph) for the function. Each basic block may optionally start with a label
654 (giving the basic block a symbol table entry), contains a list of instructions,
655 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
656 function return). If an explicit label is not provided, a block is assigned an
657 implicit numbered label, using the next value from the same counter as used for
658 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
659 entry block does not have an explicit label, it will be assigned label "%0",
660 then the first unnamed temporary in that block will be "%1", etc.
662 The first basic block in a function is special in two ways: it is
663 immediately executed on entrance to the function, and it is not allowed
664 to have predecessor basic blocks (i.e. there can not be any branches to
665 the entry block of a function). Because the block can have no
666 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
668 LLVM allows an explicit section to be specified for functions. If the
669 target supports it, it will emit functions to the section specified.
670 Additionally, the function can be placed in a COMDAT.
672 An explicit alignment may be specified for a function. If not present,
673 or if the alignment is set to zero, the alignment of the function is set
674 by the target to whatever it feels convenient. If an explicit alignment
675 is specified, the function is forced to have at least that much
676 alignment. All alignments must be a power of 2.
678 If the ``unnamed_addr`` attribute is given, the address is known to not
679 be significant and two identical functions can be merged.
683 define [linkage] [visibility] [DLLStorageClass]
685 <ResultType> @<FunctionName> ([argument list])
686 [unnamed_addr] [fn Attrs] [section "name"] [comdat [($name)]]
687 [align N] [gc] [prefix Constant] [prologue Constant]
688 [personality Constant] { ... }
690 The argument list is a comma seperated sequence of arguments where each
691 argument is of the following form
695 <type> [parameter Attrs] [name]
703 Aliases, unlike function or variables, don't create any new data. They
704 are just a new symbol and metadata for an existing position.
706 Aliases have a name and an aliasee that is either a global value or a
709 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
710 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
711 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
715 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy> @<Aliasee>
717 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
718 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
719 might not correctly handle dropping a weak symbol that is aliased.
721 Aliases that are not ``unnamed_addr`` are guaranteed to have the same address as
722 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
725 Since aliases are only a second name, some restrictions apply, of which
726 some can only be checked when producing an object file:
728 * The expression defining the aliasee must be computable at assembly
729 time. Since it is just a name, no relocations can be used.
731 * No alias in the expression can be weak as the possibility of the
732 intermediate alias being overridden cannot be represented in an
735 * No global value in the expression can be a declaration, since that
736 would require a relocation, which is not possible.
743 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
745 Comdats have a name which represents the COMDAT key. All global objects that
746 specify this key will only end up in the final object file if the linker chooses
747 that key over some other key. Aliases are placed in the same COMDAT that their
748 aliasee computes to, if any.
750 Comdats have a selection kind to provide input on how the linker should
751 choose between keys in two different object files.
755 $<Name> = comdat SelectionKind
757 The selection kind must be one of the following:
760 The linker may choose any COMDAT key, the choice is arbitrary.
762 The linker may choose any COMDAT key but the sections must contain the
765 The linker will choose the section containing the largest COMDAT key.
767 The linker requires that only section with this COMDAT key exist.
769 The linker may choose any COMDAT key but the sections must contain the
772 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
773 ``any`` as a selection kind.
775 Here is an example of a COMDAT group where a function will only be selected if
776 the COMDAT key's section is the largest:
780 $foo = comdat largest
781 @foo = global i32 2, comdat($foo)
783 define void @bar() comdat($foo) {
787 As a syntactic sugar the ``$name`` can be omitted if the name is the same as
793 @foo = global i32 2, comdat
796 In a COFF object file, this will create a COMDAT section with selection kind
797 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
798 and another COMDAT section with selection kind
799 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
800 section and contains the contents of the ``@bar`` symbol.
802 There are some restrictions on the properties of the global object.
803 It, or an alias to it, must have the same name as the COMDAT group when
805 The contents and size of this object may be used during link-time to determine
806 which COMDAT groups get selected depending on the selection kind.
807 Because the name of the object must match the name of the COMDAT group, the
808 linkage of the global object must not be local; local symbols can get renamed
809 if a collision occurs in the symbol table.
811 The combined use of COMDATS and section attributes may yield surprising results.
818 @g1 = global i32 42, section "sec", comdat($foo)
819 @g2 = global i32 42, section "sec", comdat($bar)
821 From the object file perspective, this requires the creation of two sections
822 with the same name. This is necessary because both globals belong to different
823 COMDAT groups and COMDATs, at the object file level, are represented by
826 Note that certain IR constructs like global variables and functions may
827 create COMDATs in the object file in addition to any which are specified using
828 COMDAT IR. This arises when the code generator is configured to emit globals
829 in individual sections (e.g. when `-data-sections` or `-function-sections`
830 is supplied to `llc`).
832 .. _namedmetadatastructure:
837 Named metadata is a collection of metadata. :ref:`Metadata
838 nodes <metadata>` (but not metadata strings) are the only valid
839 operands for a named metadata.
841 #. Named metadata are represented as a string of characters with the
842 metadata prefix. The rules for metadata names are the same as for
843 identifiers, but quoted names are not allowed. ``"\xx"`` type escapes
844 are still valid, which allows any character to be part of a name.
848 ; Some unnamed metadata nodes, which are referenced by the named metadata.
853 !name = !{!0, !1, !2}
860 The return type and each parameter of a function type may have a set of
861 *parameter attributes* associated with them. Parameter attributes are
862 used to communicate additional information about the result or
863 parameters of a function. Parameter attributes are considered to be part
864 of the function, not of the function type, so functions with different
865 parameter attributes can have the same function type.
867 Parameter attributes are simple keywords that follow the type specified.
868 If multiple parameter attributes are needed, they are space separated.
873 declare i32 @printf(i8* noalias nocapture, ...)
874 declare i32 @atoi(i8 zeroext)
875 declare signext i8 @returns_signed_char()
877 Note that any attributes for the function result (``nounwind``,
878 ``readonly``) come immediately after the argument list.
880 Currently, only the following parameter attributes are defined:
883 This indicates to the code generator that the parameter or return
884 value should be zero-extended to the extent required by the target's
885 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
886 the caller (for a parameter) or the callee (for a return value).
888 This indicates to the code generator that the parameter or return
889 value should be sign-extended to the extent required by the target's
890 ABI (which is usually 32-bits) by the caller (for a parameter) or
891 the callee (for a return value).
893 This indicates that this parameter or return value should be treated
894 in a special target-dependent fashion during while emitting code for
895 a function call or return (usually, by putting it in a register as
896 opposed to memory, though some targets use it to distinguish between
897 two different kinds of registers). Use of this attribute is
900 This indicates that the pointer parameter should really be passed by
901 value to the function. The attribute implies that a hidden copy of
902 the pointee is made between the caller and the callee, so the callee
903 is unable to modify the value in the caller. This attribute is only
904 valid on LLVM pointer arguments. It is generally used to pass
905 structs and arrays by value, but is also valid on pointers to
906 scalars. The copy is considered to belong to the caller not the
907 callee (for example, ``readonly`` functions should not write to
908 ``byval`` parameters). This is not a valid attribute for return
911 The byval attribute also supports specifying an alignment with the
912 align attribute. It indicates the alignment of the stack slot to
913 form and the known alignment of the pointer specified to the call
914 site. If the alignment is not specified, then the code generator
915 makes a target-specific assumption.
921 The ``inalloca`` argument attribute allows the caller to take the
922 address of outgoing stack arguments. An ``inalloca`` argument must
923 be a pointer to stack memory produced by an ``alloca`` instruction.
924 The alloca, or argument allocation, must also be tagged with the
925 inalloca keyword. Only the last argument may have the ``inalloca``
926 attribute, and that argument is guaranteed to be passed in memory.
928 An argument allocation may be used by a call at most once because
929 the call may deallocate it. The ``inalloca`` attribute cannot be
930 used in conjunction with other attributes that affect argument
931 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
932 ``inalloca`` attribute also disables LLVM's implicit lowering of
933 large aggregate return values, which means that frontend authors
934 must lower them with ``sret`` pointers.
936 When the call site is reached, the argument allocation must have
937 been the most recent stack allocation that is still live, or the
938 results are undefined. It is possible to allocate additional stack
939 space after an argument allocation and before its call site, but it
940 must be cleared off with :ref:`llvm.stackrestore
943 See :doc:`InAlloca` for more information on how to use this
947 This indicates that the pointer parameter specifies the address of a
948 structure that is the return value of the function in the source
949 program. This pointer must be guaranteed by the caller to be valid:
950 loads and stores to the structure may be assumed by the callee
951 not to trap and to be properly aligned. This may only be applied to
952 the first parameter. This is not a valid attribute for return
956 This indicates that the pointer value may be assumed by the optimizer to
957 have the specified alignment.
959 Note that this attribute has additional semantics when combined with the
965 This indicates that objects accessed via pointer values
966 :ref:`based <pointeraliasing>` on the argument or return value are not also
967 accessed, during the execution of the function, via pointer values not
968 *based* on the argument or return value. The attribute on a return value
969 also has additional semantics described below. The caller shares the
970 responsibility with the callee for ensuring that these requirements are met.
971 For further details, please see the discussion of the NoAlias response in
972 :ref:`alias analysis <Must, May, or No>`.
974 Note that this definition of ``noalias`` is intentionally similar
975 to the definition of ``restrict`` in C99 for function arguments.
977 For function return values, C99's ``restrict`` is not meaningful,
978 while LLVM's ``noalias`` is. Furthermore, the semantics of the ``noalias``
979 attribute on return values are stronger than the semantics of the attribute
980 when used on function arguments. On function return values, the ``noalias``
981 attribute indicates that the function acts like a system memory allocation
982 function, returning a pointer to allocated storage disjoint from the
983 storage for any other object accessible to the caller.
986 This indicates that the callee does not make any copies of the
987 pointer that outlive the callee itself. This is not a valid
988 attribute for return values.
993 This indicates that the pointer parameter can be excised using the
994 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
995 attribute for return values and can only be applied to one parameter.
998 This indicates that the function always returns the argument as its return
999 value. This is an optimization hint to the code generator when generating
1000 the caller, allowing tail call optimization and omission of register saves
1001 and restores in some cases; it is not checked or enforced when generating
1002 the callee. The parameter and the function return type must be valid
1003 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
1004 valid attribute for return values and can only be applied to one parameter.
1007 This indicates that the parameter or return pointer is not null. This
1008 attribute may only be applied to pointer typed parameters. This is not
1009 checked or enforced by LLVM, the caller must ensure that the pointer
1010 passed in is non-null, or the callee must ensure that the returned pointer
1013 ``dereferenceable(<n>)``
1014 This indicates that the parameter or return pointer is dereferenceable. This
1015 attribute may only be applied to pointer typed parameters. A pointer that
1016 is dereferenceable can be loaded from speculatively without a risk of
1017 trapping. The number of bytes known to be dereferenceable must be provided
1018 in parentheses. It is legal for the number of bytes to be less than the
1019 size of the pointee type. The ``nonnull`` attribute does not imply
1020 dereferenceability (consider a pointer to one element past the end of an
1021 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
1022 ``addrspace(0)`` (which is the default address space).
1024 ``dereferenceable_or_null(<n>)``
1025 This indicates that the parameter or return value isn't both
1026 non-null and non-dereferenceable (up to ``<n>`` bytes) at the same
1027 time. All non-null pointers tagged with
1028 ``dereferenceable_or_null(<n>)`` are ``dereferenceable(<n>)``.
1029 For address space 0 ``dereferenceable_or_null(<n>)`` implies that
1030 a pointer is exactly one of ``dereferenceable(<n>)`` or ``null``,
1031 and in other address spaces ``dereferenceable_or_null(<n>)``
1032 implies that a pointer is at least one of ``dereferenceable(<n>)``
1033 or ``null`` (i.e. it may be both ``null`` and
1034 ``dereferenceable(<n>)``). This attribute may only be applied to
1035 pointer typed parameters.
1039 Garbage Collector Strategy Names
1040 --------------------------------
1042 Each function may specify a garbage collector strategy name, which is simply a
1045 .. code-block:: llvm
1047 define void @f() gc "name" { ... }
1049 The supported values of *name* includes those :ref:`built in to LLVM
1050 <builtin-gc-strategies>` and any provided by loaded plugins. Specifying a GC
1051 strategy will cause the compiler to alter its output in order to support the
1052 named garbage collection algorithm. Note that LLVM itself does not contain a
1053 garbage collector, this functionality is restricted to generating machine code
1054 which can interoperate with a collector provided externally.
1061 Prefix data is data associated with a function which the code
1062 generator will emit immediately before the function's entrypoint.
1063 The purpose of this feature is to allow frontends to associate
1064 language-specific runtime metadata with specific functions and make it
1065 available through the function pointer while still allowing the
1066 function pointer to be called.
1068 To access the data for a given function, a program may bitcast the
1069 function pointer to a pointer to the constant's type and dereference
1070 index -1. This implies that the IR symbol points just past the end of
1071 the prefix data. For instance, take the example of a function annotated
1072 with a single ``i32``,
1074 .. code-block:: llvm
1076 define void @f() prefix i32 123 { ... }
1078 The prefix data can be referenced as,
1080 .. code-block:: llvm
1082 %0 = bitcast void* () @f to i32*
1083 %a = getelementptr inbounds i32, i32* %0, i32 -1
1084 %b = load i32, i32* %a
1086 Prefix data is laid out as if it were an initializer for a global variable
1087 of the prefix data's type. The function will be placed such that the
1088 beginning of the prefix data is aligned. This means that if the size
1089 of the prefix data is not a multiple of the alignment size, the
1090 function's entrypoint will not be aligned. If alignment of the
1091 function's entrypoint is desired, padding must be added to the prefix
1094 A function may have prefix data but no body. This has similar semantics
1095 to the ``available_externally`` linkage in that the data may be used by the
1096 optimizers but will not be emitted in the object file.
1103 The ``prologue`` attribute allows arbitrary code (encoded as bytes) to
1104 be inserted prior to the function body. This can be used for enabling
1105 function hot-patching and instrumentation.
1107 To maintain the semantics of ordinary function calls, the prologue data must
1108 have a particular format. Specifically, it must begin with a sequence of
1109 bytes which decode to a sequence of machine instructions, valid for the
1110 module's target, which transfer control to the point immediately succeeding
1111 the prologue data, without performing any other visible action. This allows
1112 the inliner and other passes to reason about the semantics of the function
1113 definition without needing to reason about the prologue data. Obviously this
1114 makes the format of the prologue data highly target dependent.
1116 A trivial example of valid prologue data for the x86 architecture is ``i8 144``,
1117 which encodes the ``nop`` instruction:
1119 .. code-block:: llvm
1121 define void @f() prologue i8 144 { ... }
1123 Generally prologue data can be formed by encoding a relative branch instruction
1124 which skips the metadata, as in this example of valid prologue data for the
1125 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1127 .. code-block:: llvm
1129 %0 = type <{ i8, i8, i8* }>
1131 define void @f() prologue %0 <{ i8 235, i8 8, i8* @md}> { ... }
1133 A function may have prologue data but no body. This has similar semantics
1134 to the ``available_externally`` linkage in that the data may be used by the
1135 optimizers but will not be emitted in the object file.
1139 Personality Function
1140 --------------------
1142 The ``personality`` attribute permits functions to specify what function
1143 to use for exception handling.
1150 Attribute groups are groups of attributes that are referenced by objects within
1151 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1152 functions will use the same set of attributes. In the degenerative case of a
1153 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1154 group will capture the important command line flags used to build that file.
1156 An attribute group is a module-level object. To use an attribute group, an
1157 object references the attribute group's ID (e.g. ``#37``). An object may refer
1158 to more than one attribute group. In that situation, the attributes from the
1159 different groups are merged.
1161 Here is an example of attribute groups for a function that should always be
1162 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1164 .. code-block:: llvm
1166 ; Target-independent attributes:
1167 attributes #0 = { alwaysinline alignstack=4 }
1169 ; Target-dependent attributes:
1170 attributes #1 = { "no-sse" }
1172 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1173 define void @f() #0 #1 { ... }
1180 Function attributes are set to communicate additional information about
1181 a function. Function attributes are considered to be part of the
1182 function, not of the function type, so functions with different function
1183 attributes can have the same function type.
1185 Function attributes are simple keywords that follow the type specified.
1186 If multiple attributes are needed, they are space separated. For
1189 .. code-block:: llvm
1191 define void @f() noinline { ... }
1192 define void @f() alwaysinline { ... }
1193 define void @f() alwaysinline optsize { ... }
1194 define void @f() optsize { ... }
1197 This attribute indicates that, when emitting the prologue and
1198 epilogue, the backend should forcibly align the stack pointer.
1199 Specify the desired alignment, which must be a power of two, in
1202 This attribute indicates that the inliner should attempt to inline
1203 this function into callers whenever possible, ignoring any active
1204 inlining size threshold for this caller.
1206 This indicates that the callee function at a call site should be
1207 recognized as a built-in function, even though the function's declaration
1208 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1209 direct calls to functions that are declared with the ``nobuiltin``
1212 This attribute indicates that this function is rarely called. When
1213 computing edge weights, basic blocks post-dominated by a cold
1214 function call are also considered to be cold; and, thus, given low
1217 This attribute indicates that the callee is dependent on a convergent
1218 thread execution pattern under certain parallel execution models.
1219 Transformations that are execution model agnostic may only move or
1220 tranform this call if the final location is control equivalent to its
1221 original position in the program, where control equivalence is defined as
1222 A dominates B and B post-dominates A, or vice versa.
1224 This attribute indicates that the source code contained a hint that
1225 inlining this function is desirable (such as the "inline" keyword in
1226 C/C++). It is just a hint; it imposes no requirements on the
1229 This attribute indicates that the function should be added to a
1230 jump-instruction table at code-generation time, and that all address-taken
1231 references to this function should be replaced with a reference to the
1232 appropriate jump-instruction-table function pointer. Note that this creates
1233 a new pointer for the original function, which means that code that depends
1234 on function-pointer identity can break. So, any function annotated with
1235 ``jumptable`` must also be ``unnamed_addr``.
1237 This attribute suggests that optimization passes and code generator
1238 passes make choices that keep the code size of this function as small
1239 as possible and perform optimizations that may sacrifice runtime
1240 performance in order to minimize the size of the generated code.
1242 This attribute disables prologue / epilogue emission for the
1243 function. This can have very system-specific consequences.
1245 This indicates that the callee function at a call site is not recognized as
1246 a built-in function. LLVM will retain the original call and not replace it
1247 with equivalent code based on the semantics of the built-in function, unless
1248 the call site uses the ``builtin`` attribute. This is valid at call sites
1249 and on function declarations and definitions.
1251 This attribute indicates that calls to the function cannot be
1252 duplicated. A call to a ``noduplicate`` function may be moved
1253 within its parent function, but may not be duplicated within
1254 its parent function.
1256 A function containing a ``noduplicate`` call may still
1257 be an inlining candidate, provided that the call is not
1258 duplicated by inlining. That implies that the function has
1259 internal linkage and only has one call site, so the original
1260 call is dead after inlining.
1262 This attributes disables implicit floating point instructions.
1264 This attribute indicates that the inliner should never inline this
1265 function in any situation. This attribute may not be used together
1266 with the ``alwaysinline`` attribute.
1268 This attribute suppresses lazy symbol binding for the function. This
1269 may make calls to the function faster, at the cost of extra program
1270 startup time if the function is not called during program startup.
1272 This attribute indicates that the code generator should not use a
1273 red zone, even if the target-specific ABI normally permits it.
1275 This function attribute indicates that the function never returns
1276 normally. This produces undefined behavior at runtime if the
1277 function ever does dynamically return.
1279 This function attribute indicates that the function never raises an
1280 exception. If the function does raise an exception, its runtime
1281 behavior is undefined. However, functions marked nounwind may still
1282 trap or generate asynchronous exceptions. Exception handling schemes
1283 that are recognized by LLVM to handle asynchronous exceptions, such
1284 as SEH, will still provide their implementation defined semantics.
1286 This function attribute indicates that the function is not optimized
1287 by any optimization or code generator passes with the
1288 exception of interprocedural optimization passes.
1289 This attribute cannot be used together with the ``alwaysinline``
1290 attribute; this attribute is also incompatible
1291 with the ``minsize`` attribute and the ``optsize`` attribute.
1293 This attribute requires the ``noinline`` attribute to be specified on
1294 the function as well, so the function is never inlined into any caller.
1295 Only functions with the ``alwaysinline`` attribute are valid
1296 candidates for inlining into the body of this function.
1298 This attribute suggests that optimization passes and code generator
1299 passes make choices that keep the code size of this function low,
1300 and otherwise do optimizations specifically to reduce code size as
1301 long as they do not significantly impact runtime performance.
1303 On a function, this attribute indicates that the function computes its
1304 result (or decides to unwind an exception) based strictly on its arguments,
1305 without dereferencing any pointer arguments or otherwise accessing
1306 any mutable state (e.g. memory, control registers, etc) visible to
1307 caller functions. It does not write through any pointer arguments
1308 (including ``byval`` arguments) and never changes any state visible
1309 to callers. This means that it cannot unwind exceptions by calling
1310 the ``C++`` exception throwing methods.
1312 On an argument, this attribute indicates that the function does not
1313 dereference that pointer argument, even though it may read or write the
1314 memory that the pointer points to if accessed through other pointers.
1316 On a function, this attribute indicates that the function does not write
1317 through any pointer arguments (including ``byval`` arguments) or otherwise
1318 modify any state (e.g. memory, control registers, etc) visible to
1319 caller functions. It may dereference pointer arguments and read
1320 state that may be set in the caller. A readonly function always
1321 returns the same value (or unwinds an exception identically) when
1322 called with the same set of arguments and global state. It cannot
1323 unwind an exception by calling the ``C++`` exception throwing
1326 On an argument, this attribute indicates that the function does not write
1327 through this pointer argument, even though it may write to the memory that
1328 the pointer points to.
1330 This attribute indicates that the only memory accesses inside function are
1331 loads and stores from objects pointed to by its pointer-typed arguments,
1332 with arbitrary offsets. Or in other words, all memory operations in the
1333 function can refer to memory only using pointers based on its function
1335 Note that ``argmemonly`` can be used together with ``readonly`` attribute
1336 in order to specify that function reads only from its arguments.
1338 This attribute indicates that this function can return twice. The C
1339 ``setjmp`` is an example of such a function. The compiler disables
1340 some optimizations (like tail calls) in the caller of these
1343 This attribute indicates that
1344 `SafeStack <http://clang.llvm.org/docs/SafeStack.html>`_
1345 protection is enabled for this function.
1347 If a function that has a ``safestack`` attribute is inlined into a
1348 function that doesn't have a ``safestack`` attribute or which has an
1349 ``ssp``, ``sspstrong`` or ``sspreq`` attribute, then the resulting
1350 function will have a ``safestack`` attribute.
1351 ``sanitize_address``
1352 This attribute indicates that AddressSanitizer checks
1353 (dynamic address safety analysis) are enabled for this function.
1355 This attribute indicates that MemorySanitizer checks (dynamic detection
1356 of accesses to uninitialized memory) are enabled for this function.
1358 This attribute indicates that ThreadSanitizer checks
1359 (dynamic thread safety analysis) are enabled for this function.
1361 This attribute indicates that the function should emit a stack
1362 smashing protector. It is in the form of a "canary" --- a random value
1363 placed on the stack before the local variables that's checked upon
1364 return from the function to see if it has been overwritten. A
1365 heuristic is used to determine if a function needs stack protectors
1366 or not. The heuristic used will enable protectors for functions with:
1368 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1369 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1370 - Calls to alloca() with variable sizes or constant sizes greater than
1371 ``ssp-buffer-size``.
1373 Variables that are identified as requiring a protector will be arranged
1374 on the stack such that they are adjacent to the stack protector guard.
1376 If a function that has an ``ssp`` attribute is inlined into a
1377 function that doesn't have an ``ssp`` attribute, then the resulting
1378 function will have an ``ssp`` attribute.
1380 This attribute indicates that the function should *always* emit a
1381 stack smashing protector. This overrides the ``ssp`` function
1384 Variables that are identified as requiring a protector will be arranged
1385 on the stack such that they are adjacent to the stack protector guard.
1386 The specific layout rules are:
1388 #. Large arrays and structures containing large arrays
1389 (``>= ssp-buffer-size``) are closest to the stack protector.
1390 #. Small arrays and structures containing small arrays
1391 (``< ssp-buffer-size``) are 2nd closest to the protector.
1392 #. Variables that have had their address taken are 3rd closest to the
1395 If a function that has an ``sspreq`` attribute is inlined into a
1396 function that doesn't have an ``sspreq`` attribute or which has an
1397 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1398 an ``sspreq`` attribute.
1400 This attribute indicates that the function should emit a stack smashing
1401 protector. This attribute causes a strong heuristic to be used when
1402 determining if a function needs stack protectors. The strong heuristic
1403 will enable protectors for functions with:
1405 - Arrays of any size and type
1406 - Aggregates containing an array of any size and type.
1407 - Calls to alloca().
1408 - Local variables that have had their address taken.
1410 Variables that are identified as requiring a protector will be arranged
1411 on the stack such that they are adjacent to the stack protector guard.
1412 The specific layout rules are:
1414 #. Large arrays and structures containing large arrays
1415 (``>= ssp-buffer-size``) are closest to the stack protector.
1416 #. Small arrays and structures containing small arrays
1417 (``< ssp-buffer-size``) are 2nd closest to the protector.
1418 #. Variables that have had their address taken are 3rd closest to the
1421 This overrides the ``ssp`` function attribute.
1423 If a function that has an ``sspstrong`` attribute is inlined into a
1424 function that doesn't have an ``sspstrong`` attribute, then the
1425 resulting function will have an ``sspstrong`` attribute.
1427 This attribute indicates that the function will delegate to some other
1428 function with a tail call. The prototype of a thunk should not be used for
1429 optimization purposes. The caller is expected to cast the thunk prototype to
1430 match the thunk target prototype.
1432 This attribute indicates that the ABI being targeted requires that
1433 an unwind table entry be produce for this function even if we can
1434 show that no exceptions passes by it. This is normally the case for
1435 the ELF x86-64 abi, but it can be disabled for some compilation
1440 Module-Level Inline Assembly
1441 ----------------------------
1443 Modules may contain "module-level inline asm" blocks, which corresponds
1444 to the GCC "file scope inline asm" blocks. These blocks are internally
1445 concatenated by LLVM and treated as a single unit, but may be separated
1446 in the ``.ll`` file if desired. The syntax is very simple:
1448 .. code-block:: llvm
1450 module asm "inline asm code goes here"
1451 module asm "more can go here"
1453 The strings can contain any character by escaping non-printable
1454 characters. The escape sequence used is simply "\\xx" where "xx" is the
1455 two digit hex code for the number.
1457 Note that the assembly string *must* be parseable by LLVM's integrated assembler
1458 (unless it is disabled), even when emitting a ``.s`` file.
1460 .. _langref_datalayout:
1465 A module may specify a target specific data layout string that specifies
1466 how data is to be laid out in memory. The syntax for the data layout is
1469 .. code-block:: llvm
1471 target datalayout = "layout specification"
1473 The *layout specification* consists of a list of specifications
1474 separated by the minus sign character ('-'). Each specification starts
1475 with a letter and may include other information after the letter to
1476 define some aspect of the data layout. The specifications accepted are
1480 Specifies that the target lays out data in big-endian form. That is,
1481 the bits with the most significance have the lowest address
1484 Specifies that the target lays out data in little-endian form. That
1485 is, the bits with the least significance have the lowest address
1488 Specifies the natural alignment of the stack in bits. Alignment
1489 promotion of stack variables is limited to the natural stack
1490 alignment to avoid dynamic stack realignment. The stack alignment
1491 must be a multiple of 8-bits. If omitted, the natural stack
1492 alignment defaults to "unspecified", which does not prevent any
1493 alignment promotions.
1494 ``p[n]:<size>:<abi>:<pref>``
1495 This specifies the *size* of a pointer and its ``<abi>`` and
1496 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1497 bits. The address space, ``n`` is optional, and if not specified,
1498 denotes the default address space 0. The value of ``n`` must be
1499 in the range [1,2^23).
1500 ``i<size>:<abi>:<pref>``
1501 This specifies the alignment for an integer type of a given bit
1502 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1503 ``v<size>:<abi>:<pref>``
1504 This specifies the alignment for a vector type of a given bit
1506 ``f<size>:<abi>:<pref>``
1507 This specifies the alignment for a floating point type of a given bit
1508 ``<size>``. Only values of ``<size>`` that are supported by the target
1509 will work. 32 (float) and 64 (double) are supported on all targets; 80
1510 or 128 (different flavors of long double) are also supported on some
1513 This specifies the alignment for an object of aggregate type.
1515 If present, specifies that llvm names are mangled in the output. The
1518 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1519 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1520 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1521 symbols get a ``_`` prefix.
1522 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1523 functions also get a suffix based on the frame size.
1524 ``n<size1>:<size2>:<size3>...``
1525 This specifies a set of native integer widths for the target CPU in
1526 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1527 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1528 this set are considered to support most general arithmetic operations
1531 On every specification that takes a ``<abi>:<pref>``, specifying the
1532 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1533 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1535 When constructing the data layout for a given target, LLVM starts with a
1536 default set of specifications which are then (possibly) overridden by
1537 the specifications in the ``datalayout`` keyword. The default
1538 specifications are given in this list:
1540 - ``E`` - big endian
1541 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1542 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1543 same as the default address space.
1544 - ``S0`` - natural stack alignment is unspecified
1545 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1546 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1547 - ``i16:16:16`` - i16 is 16-bit aligned
1548 - ``i32:32:32`` - i32 is 32-bit aligned
1549 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1550 alignment of 64-bits
1551 - ``f16:16:16`` - half is 16-bit aligned
1552 - ``f32:32:32`` - float is 32-bit aligned
1553 - ``f64:64:64`` - double is 64-bit aligned
1554 - ``f128:128:128`` - quad is 128-bit aligned
1555 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1556 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1557 - ``a:0:64`` - aggregates are 64-bit aligned
1559 When LLVM is determining the alignment for a given type, it uses the
1562 #. If the type sought is an exact match for one of the specifications,
1563 that specification is used.
1564 #. If no match is found, and the type sought is an integer type, then
1565 the smallest integer type that is larger than the bitwidth of the
1566 sought type is used. If none of the specifications are larger than
1567 the bitwidth then the largest integer type is used. For example,
1568 given the default specifications above, the i7 type will use the
1569 alignment of i8 (next largest) while both i65 and i256 will use the
1570 alignment of i64 (largest specified).
1571 #. If no match is found, and the type sought is a vector type, then the
1572 largest vector type that is smaller than the sought vector type will
1573 be used as a fall back. This happens because <128 x double> can be
1574 implemented in terms of 64 <2 x double>, for example.
1576 The function of the data layout string may not be what you expect.
1577 Notably, this is not a specification from the frontend of what alignment
1578 the code generator should use.
1580 Instead, if specified, the target data layout is required to match what
1581 the ultimate *code generator* expects. This string is used by the
1582 mid-level optimizers to improve code, and this only works if it matches
1583 what the ultimate code generator uses. There is no way to generate IR
1584 that does not embed this target-specific detail into the IR. If you
1585 don't specify the string, the default specifications will be used to
1586 generate a Data Layout and the optimization phases will operate
1587 accordingly and introduce target specificity into the IR with respect to
1588 these default specifications.
1595 A module may specify a target triple string that describes the target
1596 host. The syntax for the target triple is simply:
1598 .. code-block:: llvm
1600 target triple = "x86_64-apple-macosx10.7.0"
1602 The *target triple* string consists of a series of identifiers delimited
1603 by the minus sign character ('-'). The canonical forms are:
1607 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1608 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1610 This information is passed along to the backend so that it generates
1611 code for the proper architecture. It's possible to override this on the
1612 command line with the ``-mtriple`` command line option.
1614 .. _pointeraliasing:
1616 Pointer Aliasing Rules
1617 ----------------------
1619 Any memory access must be done through a pointer value associated with
1620 an address range of the memory access, otherwise the behavior is
1621 undefined. Pointer values are associated with address ranges according
1622 to the following rules:
1624 - A pointer value is associated with the addresses associated with any
1625 value it is *based* on.
1626 - An address of a global variable is associated with the address range
1627 of the variable's storage.
1628 - The result value of an allocation instruction is associated with the
1629 address range of the allocated storage.
1630 - A null pointer in the default address-space is associated with no
1632 - An integer constant other than zero or a pointer value returned from
1633 a function not defined within LLVM may be associated with address
1634 ranges allocated through mechanisms other than those provided by
1635 LLVM. Such ranges shall not overlap with any ranges of addresses
1636 allocated by mechanisms provided by LLVM.
1638 A pointer value is *based* on another pointer value according to the
1641 - A pointer value formed from a ``getelementptr`` operation is *based*
1642 on the first value operand of the ``getelementptr``.
1643 - The result value of a ``bitcast`` is *based* on the operand of the
1645 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1646 values that contribute (directly or indirectly) to the computation of
1647 the pointer's value.
1648 - The "*based* on" relationship is transitive.
1650 Note that this definition of *"based"* is intentionally similar to the
1651 definition of *"based"* in C99, though it is slightly weaker.
1653 LLVM IR does not associate types with memory. The result type of a
1654 ``load`` merely indicates the size and alignment of the memory from
1655 which to load, as well as the interpretation of the value. The first
1656 operand type of a ``store`` similarly only indicates the size and
1657 alignment of the store.
1659 Consequently, type-based alias analysis, aka TBAA, aka
1660 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1661 :ref:`Metadata <metadata>` may be used to encode additional information
1662 which specialized optimization passes may use to implement type-based
1667 Volatile Memory Accesses
1668 ------------------------
1670 Certain memory accesses, such as :ref:`load <i_load>`'s,
1671 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1672 marked ``volatile``. The optimizers must not change the number of
1673 volatile operations or change their order of execution relative to other
1674 volatile operations. The optimizers *may* change the order of volatile
1675 operations relative to non-volatile operations. This is not Java's
1676 "volatile" and has no cross-thread synchronization behavior.
1678 IR-level volatile loads and stores cannot safely be optimized into
1679 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1680 flagged volatile. Likewise, the backend should never split or merge
1681 target-legal volatile load/store instructions.
1683 .. admonition:: Rationale
1685 Platforms may rely on volatile loads and stores of natively supported
1686 data width to be executed as single instruction. For example, in C
1687 this holds for an l-value of volatile primitive type with native
1688 hardware support, but not necessarily for aggregate types. The
1689 frontend upholds these expectations, which are intentionally
1690 unspecified in the IR. The rules above ensure that IR transformation
1691 do not violate the frontend's contract with the language.
1695 Memory Model for Concurrent Operations
1696 --------------------------------------
1698 The LLVM IR does not define any way to start parallel threads of
1699 execution or to register signal handlers. Nonetheless, there are
1700 platform-specific ways to create them, and we define LLVM IR's behavior
1701 in their presence. This model is inspired by the C++0x memory model.
1703 For a more informal introduction to this model, see the :doc:`Atomics`.
1705 We define a *happens-before* partial order as the least partial order
1708 - Is a superset of single-thread program order, and
1709 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1710 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1711 techniques, like pthread locks, thread creation, thread joining,
1712 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1713 Constraints <ordering>`).
1715 Note that program order does not introduce *happens-before* edges
1716 between a thread and signals executing inside that thread.
1718 Every (defined) read operation (load instructions, memcpy, atomic
1719 loads/read-modify-writes, etc.) R reads a series of bytes written by
1720 (defined) write operations (store instructions, atomic
1721 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1722 section, initialized globals are considered to have a write of the
1723 initializer which is atomic and happens before any other read or write
1724 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1725 may see any write to the same byte, except:
1727 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1728 write\ :sub:`2` happens before R\ :sub:`byte`, then
1729 R\ :sub:`byte` does not see write\ :sub:`1`.
1730 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1731 R\ :sub:`byte` does not see write\ :sub:`3`.
1733 Given that definition, R\ :sub:`byte` is defined as follows:
1735 - If R is volatile, the result is target-dependent. (Volatile is
1736 supposed to give guarantees which can support ``sig_atomic_t`` in
1737 C/C++, and may be used for accesses to addresses that do not behave
1738 like normal memory. It does not generally provide cross-thread
1740 - Otherwise, if there is no write to the same byte that happens before
1741 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1742 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1743 R\ :sub:`byte` returns the value written by that write.
1744 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1745 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1746 Memory Ordering Constraints <ordering>` section for additional
1747 constraints on how the choice is made.
1748 - Otherwise R\ :sub:`byte` returns ``undef``.
1750 R returns the value composed of the series of bytes it read. This
1751 implies that some bytes within the value may be ``undef`` **without**
1752 the entire value being ``undef``. Note that this only defines the
1753 semantics of the operation; it doesn't mean that targets will emit more
1754 than one instruction to read the series of bytes.
1756 Note that in cases where none of the atomic intrinsics are used, this
1757 model places only one restriction on IR transformations on top of what
1758 is required for single-threaded execution: introducing a store to a byte
1759 which might not otherwise be stored is not allowed in general.
1760 (Specifically, in the case where another thread might write to and read
1761 from an address, introducing a store can change a load that may see
1762 exactly one write into a load that may see multiple writes.)
1766 Atomic Memory Ordering Constraints
1767 ----------------------------------
1769 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1770 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1771 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1772 ordering parameters that determine which other atomic instructions on
1773 the same address they *synchronize with*. These semantics are borrowed
1774 from Java and C++0x, but are somewhat more colloquial. If these
1775 descriptions aren't precise enough, check those specs (see spec
1776 references in the :doc:`atomics guide <Atomics>`).
1777 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1778 differently since they don't take an address. See that instruction's
1779 documentation for details.
1781 For a simpler introduction to the ordering constraints, see the
1785 The set of values that can be read is governed by the happens-before
1786 partial order. A value cannot be read unless some operation wrote
1787 it. This is intended to provide a guarantee strong enough to model
1788 Java's non-volatile shared variables. This ordering cannot be
1789 specified for read-modify-write operations; it is not strong enough
1790 to make them atomic in any interesting way.
1792 In addition to the guarantees of ``unordered``, there is a single
1793 total order for modifications by ``monotonic`` operations on each
1794 address. All modification orders must be compatible with the
1795 happens-before order. There is no guarantee that the modification
1796 orders can be combined to a global total order for the whole program
1797 (and this often will not be possible). The read in an atomic
1798 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1799 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1800 order immediately before the value it writes. If one atomic read
1801 happens before another atomic read of the same address, the later
1802 read must see the same value or a later value in the address's
1803 modification order. This disallows reordering of ``monotonic`` (or
1804 stronger) operations on the same address. If an address is written
1805 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1806 read that address repeatedly, the other threads must eventually see
1807 the write. This corresponds to the C++0x/C1x
1808 ``memory_order_relaxed``.
1810 In addition to the guarantees of ``monotonic``, a
1811 *synchronizes-with* edge may be formed with a ``release`` operation.
1812 This is intended to model C++'s ``memory_order_acquire``.
1814 In addition to the guarantees of ``monotonic``, if this operation
1815 writes a value which is subsequently read by an ``acquire``
1816 operation, it *synchronizes-with* that operation. (This isn't a
1817 complete description; see the C++0x definition of a release
1818 sequence.) This corresponds to the C++0x/C1x
1819 ``memory_order_release``.
1820 ``acq_rel`` (acquire+release)
1821 Acts as both an ``acquire`` and ``release`` operation on its
1822 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1823 ``seq_cst`` (sequentially consistent)
1824 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1825 operation that only reads, ``release`` for an operation that only
1826 writes), there is a global total order on all
1827 sequentially-consistent operations on all addresses, which is
1828 consistent with the *happens-before* partial order and with the
1829 modification orders of all the affected addresses. Each
1830 sequentially-consistent read sees the last preceding write to the
1831 same address in this global order. This corresponds to the C++0x/C1x
1832 ``memory_order_seq_cst`` and Java volatile.
1836 If an atomic operation is marked ``singlethread``, it only *synchronizes
1837 with* or participates in modification and seq\_cst total orderings with
1838 other operations running in the same thread (for example, in signal
1846 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1847 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1848 :ref:`frem <i_frem>`, :ref:`fcmp <i_fcmp>`) have the following flags that can
1849 be set to enable otherwise unsafe floating point operations
1852 No NaNs - Allow optimizations to assume the arguments and result are not
1853 NaN. Such optimizations are required to retain defined behavior over
1854 NaNs, but the value of the result is undefined.
1857 No Infs - Allow optimizations to assume the arguments and result are not
1858 +/-Inf. Such optimizations are required to retain defined behavior over
1859 +/-Inf, but the value of the result is undefined.
1862 No Signed Zeros - Allow optimizations to treat the sign of a zero
1863 argument or result as insignificant.
1866 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1867 argument rather than perform division.
1870 Fast - Allow algebraically equivalent transformations that may
1871 dramatically change results in floating point (e.g. reassociate). This
1872 flag implies all the others.
1876 Use-list Order Directives
1877 -------------------------
1879 Use-list directives encode the in-memory order of each use-list, allowing the
1880 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1881 indexes that are assigned to the referenced value's uses. The referenced
1882 value's use-list is immediately sorted by these indexes.
1884 Use-list directives may appear at function scope or global scope. They are not
1885 instructions, and have no effect on the semantics of the IR. When they're at
1886 function scope, they must appear after the terminator of the final basic block.
1888 If basic blocks have their address taken via ``blockaddress()`` expressions,
1889 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1896 uselistorder <ty> <value>, { <order-indexes> }
1897 uselistorder_bb @function, %block { <order-indexes> }
1903 define void @foo(i32 %arg1, i32 %arg2) {
1905 ; ... instructions ...
1907 ; ... instructions ...
1909 ; At function scope.
1910 uselistorder i32 %arg1, { 1, 0, 2 }
1911 uselistorder label %bb, { 1, 0 }
1915 uselistorder i32* @global, { 1, 2, 0 }
1916 uselistorder i32 7, { 1, 0 }
1917 uselistorder i32 (i32) @bar, { 1, 0 }
1918 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1925 The LLVM type system is one of the most important features of the
1926 intermediate representation. Being typed enables a number of
1927 optimizations to be performed on the intermediate representation
1928 directly, without having to do extra analyses on the side before the
1929 transformation. A strong type system makes it easier to read the
1930 generated code and enables novel analyses and transformations that are
1931 not feasible to perform on normal three address code representations.
1941 The void type does not represent any value and has no size.
1959 The function type can be thought of as a function signature. It consists of a
1960 return type and a list of formal parameter types. The return type of a function
1961 type is a void type or first class type --- except for :ref:`label <t_label>`
1962 and :ref:`metadata <t_metadata>` types.
1968 <returntype> (<parameter list>)
1970 ...where '``<parameter list>``' is a comma-separated list of type
1971 specifiers. Optionally, the parameter list may include a type ``...``, which
1972 indicates that the function takes a variable number of arguments. Variable
1973 argument functions can access their arguments with the :ref:`variable argument
1974 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1975 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1979 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1980 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1981 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1982 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1983 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1984 | ``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. |
1985 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1986 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1987 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1994 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1995 Values of these types are the only ones which can be produced by
2003 These are the types that are valid in registers from CodeGen's perspective.
2012 The integer type is a very simple type that simply specifies an
2013 arbitrary bit width for the integer type desired. Any bit width from 1
2014 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
2022 The number of bits the integer will occupy is specified by the ``N``
2028 +----------------+------------------------------------------------+
2029 | ``i1`` | a single-bit integer. |
2030 +----------------+------------------------------------------------+
2031 | ``i32`` | a 32-bit integer. |
2032 +----------------+------------------------------------------------+
2033 | ``i1942652`` | a really big integer of over 1 million bits. |
2034 +----------------+------------------------------------------------+
2038 Floating Point Types
2039 """"""""""""""""""""
2048 - 16-bit floating point value
2051 - 32-bit floating point value
2054 - 64-bit floating point value
2057 - 128-bit floating point value (112-bit mantissa)
2060 - 80-bit floating point value (X87)
2063 - 128-bit floating point value (two 64-bits)
2070 The x86_mmx type represents a value held in an MMX register on an x86
2071 machine. The operations allowed on it are quite limited: parameters and
2072 return values, load and store, and bitcast. User-specified MMX
2073 instructions are represented as intrinsic or asm calls with arguments
2074 and/or results of this type. There are no arrays, vectors or constants
2091 The pointer type is used to specify memory locations. Pointers are
2092 commonly used to reference objects in memory.
2094 Pointer types may have an optional address space attribute defining the
2095 numbered address space where the pointed-to object resides. The default
2096 address space is number zero. The semantics of non-zero address spaces
2097 are target-specific.
2099 Note that LLVM does not permit pointers to void (``void*``) nor does it
2100 permit pointers to labels (``label*``). Use ``i8*`` instead.
2110 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2111 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2112 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2113 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2114 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2115 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2116 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2125 A vector type is a simple derived type that represents a vector of
2126 elements. Vector types are used when multiple primitive data are
2127 operated in parallel using a single instruction (SIMD). A vector type
2128 requires a size (number of elements) and an underlying primitive data
2129 type. Vector types are considered :ref:`first class <t_firstclass>`.
2135 < <# elements> x <elementtype> >
2137 The number of elements is a constant integer value larger than 0;
2138 elementtype may be any integer, floating point or pointer type. Vectors
2139 of size zero are not allowed.
2143 +-------------------+--------------------------------------------------+
2144 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2145 +-------------------+--------------------------------------------------+
2146 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2147 +-------------------+--------------------------------------------------+
2148 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2149 +-------------------+--------------------------------------------------+
2150 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2151 +-------------------+--------------------------------------------------+
2160 The label type represents code labels.
2175 The metadata type represents embedded metadata. No derived types may be
2176 created from metadata except for :ref:`function <t_function>` arguments.
2189 Aggregate Types are a subset of derived types that can contain multiple
2190 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2191 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2201 The array type is a very simple derived type that arranges elements
2202 sequentially in memory. The array type requires a size (number of
2203 elements) and an underlying data type.
2209 [<# elements> x <elementtype>]
2211 The number of elements is a constant integer value; ``elementtype`` may
2212 be any type with a size.
2216 +------------------+--------------------------------------+
2217 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2218 +------------------+--------------------------------------+
2219 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2220 +------------------+--------------------------------------+
2221 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2222 +------------------+--------------------------------------+
2224 Here are some examples of multidimensional arrays:
2226 +-----------------------------+----------------------------------------------------------+
2227 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2228 +-----------------------------+----------------------------------------------------------+
2229 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2230 +-----------------------------+----------------------------------------------------------+
2231 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2232 +-----------------------------+----------------------------------------------------------+
2234 There is no restriction on indexing beyond the end of the array implied
2235 by a static type (though there are restrictions on indexing beyond the
2236 bounds of an allocated object in some cases). This means that
2237 single-dimension 'variable sized array' addressing can be implemented in
2238 LLVM with a zero length array type. An implementation of 'pascal style
2239 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2249 The structure type is used to represent a collection of data members
2250 together in memory. The elements of a structure may be any type that has
2253 Structures in memory are accessed using '``load``' and '``store``' by
2254 getting a pointer to a field with the '``getelementptr``' instruction.
2255 Structures in registers are accessed using the '``extractvalue``' and
2256 '``insertvalue``' instructions.
2258 Structures may optionally be "packed" structures, which indicate that
2259 the alignment of the struct is one byte, and that there is no padding
2260 between the elements. In non-packed structs, padding between field types
2261 is inserted as defined by the DataLayout string in the module, which is
2262 required to match what the underlying code generator expects.
2264 Structures can either be "literal" or "identified". A literal structure
2265 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2266 identified types are always defined at the top level with a name.
2267 Literal types are uniqued by their contents and can never be recursive
2268 or opaque since there is no way to write one. Identified types can be
2269 recursive, can be opaqued, and are never uniqued.
2275 %T1 = type { <type list> } ; Identified normal struct type
2276 %T2 = type <{ <type list> }> ; Identified packed struct type
2280 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2281 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2282 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2283 | ``{ 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``. |
2284 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2285 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2286 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2290 Opaque Structure Types
2291 """"""""""""""""""""""
2295 Opaque structure types are used to represent named structure types that
2296 do not have a body specified. This corresponds (for example) to the C
2297 notion of a forward declared structure.
2308 +--------------+-------------------+
2309 | ``opaque`` | An opaque type. |
2310 +--------------+-------------------+
2317 LLVM has several different basic types of constants. This section
2318 describes them all and their syntax.
2323 **Boolean constants**
2324 The two strings '``true``' and '``false``' are both valid constants
2326 **Integer constants**
2327 Standard integers (such as '4') are constants of the
2328 :ref:`integer <t_integer>` type. Negative numbers may be used with
2330 **Floating point constants**
2331 Floating point constants use standard decimal notation (e.g.
2332 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2333 hexadecimal notation (see below). The assembler requires the exact
2334 decimal value of a floating-point constant. For example, the
2335 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2336 decimal in binary. Floating point constants must have a :ref:`floating
2337 point <t_floating>` type.
2338 **Null pointer constants**
2339 The identifier '``null``' is recognized as a null pointer constant
2340 and must be of :ref:`pointer type <t_pointer>`.
2342 The one non-intuitive notation for constants is the hexadecimal form of
2343 floating point constants. For example, the form
2344 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2345 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2346 constants are required (and the only time that they are generated by the
2347 disassembler) is when a floating point constant must be emitted but it
2348 cannot be represented as a decimal floating point number in a reasonable
2349 number of digits. For example, NaN's, infinities, and other special
2350 values are represented in their IEEE hexadecimal format so that assembly
2351 and disassembly do not cause any bits to change in the constants.
2353 When using the hexadecimal form, constants of types half, float, and
2354 double are represented using the 16-digit form shown above (which
2355 matches the IEEE754 representation for double); half and float values
2356 must, however, be exactly representable as IEEE 754 half and single
2357 precision, respectively. Hexadecimal format is always used for long
2358 double, and there are three forms of long double. The 80-bit format used
2359 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2360 128-bit format used by PowerPC (two adjacent doubles) is represented by
2361 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2362 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2363 will only work if they match the long double format on your target.
2364 The IEEE 16-bit format (half precision) is represented by ``0xH``
2365 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2366 (sign bit at the left).
2368 There are no constants of type x86_mmx.
2370 .. _complexconstants:
2375 Complex constants are a (potentially recursive) combination of simple
2376 constants and smaller complex constants.
2378 **Structure constants**
2379 Structure constants are represented with notation similar to
2380 structure type definitions (a comma separated list of elements,
2381 surrounded by braces (``{}``)). For example:
2382 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2383 "``@G = external global i32``". Structure constants must have
2384 :ref:`structure type <t_struct>`, and the number and types of elements
2385 must match those specified by the type.
2387 Array constants are represented with notation similar to array type
2388 definitions (a comma separated list of elements, surrounded by
2389 square brackets (``[]``)). For example:
2390 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2391 :ref:`array type <t_array>`, and the number and types of elements must
2392 match those specified by the type. As a special case, character array
2393 constants may also be represented as a double-quoted string using the ``c``
2394 prefix. For example: "``c"Hello World\0A\00"``".
2395 **Vector constants**
2396 Vector constants are represented with notation similar to vector
2397 type definitions (a comma separated list of elements, surrounded by
2398 less-than/greater-than's (``<>``)). For example:
2399 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2400 must have :ref:`vector type <t_vector>`, and the number and types of
2401 elements must match those specified by the type.
2402 **Zero initialization**
2403 The string '``zeroinitializer``' can be used to zero initialize a
2404 value to zero of *any* type, including scalar and
2405 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2406 having to print large zero initializers (e.g. for large arrays) and
2407 is always exactly equivalent to using explicit zero initializers.
2409 A metadata node is a constant tuple without types. For example:
2410 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2411 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2412 Unlike other typed constants that are meant to be interpreted as part of
2413 the instruction stream, metadata is a place to attach additional
2414 information such as debug info.
2416 Global Variable and Function Addresses
2417 --------------------------------------
2419 The addresses of :ref:`global variables <globalvars>` and
2420 :ref:`functions <functionstructure>` are always implicitly valid
2421 (link-time) constants. These constants are explicitly referenced when
2422 the :ref:`identifier for the global <identifiers>` is used and always have
2423 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2426 .. code-block:: llvm
2430 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2437 The string '``undef``' can be used anywhere a constant is expected, and
2438 indicates that the user of the value may receive an unspecified
2439 bit-pattern. Undefined values may be of any type (other than '``label``'
2440 or '``void``') and be used anywhere a constant is permitted.
2442 Undefined values are useful because they indicate to the compiler that
2443 the program is well defined no matter what value is used. This gives the
2444 compiler more freedom to optimize. Here are some examples of
2445 (potentially surprising) transformations that are valid (in pseudo IR):
2447 .. code-block:: llvm
2457 This is safe because all of the output bits are affected by the undef
2458 bits. Any output bit can have a zero or one depending on the input bits.
2460 .. code-block:: llvm
2471 These logical operations have bits that are not always affected by the
2472 input. For example, if ``%X`` has a zero bit, then the output of the
2473 '``and``' operation will always be a zero for that bit, no matter what
2474 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2475 optimize or assume that the result of the '``and``' is '``undef``'.
2476 However, it is safe to assume that all bits of the '``undef``' could be
2477 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2478 all the bits of the '``undef``' operand to the '``or``' could be set,
2479 allowing the '``or``' to be folded to -1.
2481 .. code-block:: llvm
2483 %A = select undef, %X, %Y
2484 %B = select undef, 42, %Y
2485 %C = select %X, %Y, undef
2495 This set of examples shows that undefined '``select``' (and conditional
2496 branch) conditions can go *either way*, but they have to come from one
2497 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2498 both known to have a clear low bit, then ``%A`` would have to have a
2499 cleared low bit. However, in the ``%C`` example, the optimizer is
2500 allowed to assume that the '``undef``' operand could be the same as
2501 ``%Y``, allowing the whole '``select``' to be eliminated.
2503 .. code-block:: llvm
2505 %A = xor undef, undef
2522 This example points out that two '``undef``' operands are not
2523 necessarily the same. This can be surprising to people (and also matches
2524 C semantics) where they assume that "``X^X``" is always zero, even if
2525 ``X`` is undefined. This isn't true for a number of reasons, but the
2526 short answer is that an '``undef``' "variable" can arbitrarily change
2527 its value over its "live range". This is true because the variable
2528 doesn't actually *have a live range*. Instead, the value is logically
2529 read from arbitrary registers that happen to be around when needed, so
2530 the value is not necessarily consistent over time. In fact, ``%A`` and
2531 ``%C`` need to have the same semantics or the core LLVM "replace all
2532 uses with" concept would not hold.
2534 .. code-block:: llvm
2542 These examples show the crucial difference between an *undefined value*
2543 and *undefined behavior*. An undefined value (like '``undef``') is
2544 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2545 operation can be constant folded to '``undef``', because the '``undef``'
2546 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2547 However, in the second example, we can make a more aggressive
2548 assumption: because the ``undef`` is allowed to be an arbitrary value,
2549 we are allowed to assume that it could be zero. Since a divide by zero
2550 has *undefined behavior*, we are allowed to assume that the operation
2551 does not execute at all. This allows us to delete the divide and all
2552 code after it. Because the undefined operation "can't happen", the
2553 optimizer can assume that it occurs in dead code.
2555 .. code-block:: llvm
2557 a: store undef -> %X
2558 b: store %X -> undef
2563 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2564 value can be assumed to not have any effect; we can assume that the
2565 value is overwritten with bits that happen to match what was already
2566 there. However, a store *to* an undefined location could clobber
2567 arbitrary memory, therefore, it has undefined behavior.
2574 Poison values are similar to :ref:`undef values <undefvalues>`, however
2575 they also represent the fact that an instruction or constant expression
2576 that cannot evoke side effects has nevertheless detected a condition
2577 that results in undefined behavior.
2579 There is currently no way of representing a poison value in the IR; they
2580 only exist when produced by operations such as :ref:`add <i_add>` with
2583 Poison value behavior is defined in terms of value *dependence*:
2585 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2586 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2587 their dynamic predecessor basic block.
2588 - Function arguments depend on the corresponding actual argument values
2589 in the dynamic callers of their functions.
2590 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2591 instructions that dynamically transfer control back to them.
2592 - :ref:`Invoke <i_invoke>` instructions depend on the
2593 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2594 call instructions that dynamically transfer control back to them.
2595 - Non-volatile loads and stores depend on the most recent stores to all
2596 of the referenced memory addresses, following the order in the IR
2597 (including loads and stores implied by intrinsics such as
2598 :ref:`@llvm.memcpy <int_memcpy>`.)
2599 - An instruction with externally visible side effects depends on the
2600 most recent preceding instruction with externally visible side
2601 effects, following the order in the IR. (This includes :ref:`volatile
2602 operations <volatile>`.)
2603 - An instruction *control-depends* on a :ref:`terminator
2604 instruction <terminators>` if the terminator instruction has
2605 multiple successors and the instruction is always executed when
2606 control transfers to one of the successors, and may not be executed
2607 when control is transferred to another.
2608 - Additionally, an instruction also *control-depends* on a terminator
2609 instruction if the set of instructions it otherwise depends on would
2610 be different if the terminator had transferred control to a different
2612 - Dependence is transitive.
2614 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2615 with the additional effect that any instruction that has a *dependence*
2616 on a poison value has undefined behavior.
2618 Here are some examples:
2620 .. code-block:: llvm
2623 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2624 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2625 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
2626 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2628 store i32 %poison, i32* @g ; Poison value stored to memory.
2629 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
2631 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2633 %narrowaddr = bitcast i32* @g to i16*
2634 %wideaddr = bitcast i32* @g to i64*
2635 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2636 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
2638 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2639 br i1 %cmp, label %true, label %end ; Branch to either destination.
2642 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2643 ; it has undefined behavior.
2647 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2648 ; Both edges into this PHI are
2649 ; control-dependent on %cmp, so this
2650 ; always results in a poison value.
2652 store volatile i32 0, i32* @g ; This would depend on the store in %true
2653 ; if %cmp is true, or the store in %entry
2654 ; otherwise, so this is undefined behavior.
2656 br i1 %cmp, label %second_true, label %second_end
2657 ; The same branch again, but this time the
2658 ; true block doesn't have side effects.
2665 store volatile i32 0, i32* @g ; This time, the instruction always depends
2666 ; on the store in %end. Also, it is
2667 ; control-equivalent to %end, so this is
2668 ; well-defined (ignoring earlier undefined
2669 ; behavior in this example).
2673 Addresses of Basic Blocks
2674 -------------------------
2676 ``blockaddress(@function, %block)``
2678 The '``blockaddress``' constant computes the address of the specified
2679 basic block in the specified function, and always has an ``i8*`` type.
2680 Taking the address of the entry block is illegal.
2682 This value only has defined behavior when used as an operand to the
2683 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2684 against null. Pointer equality tests between labels addresses results in
2685 undefined behavior --- though, again, comparison against null is ok, and
2686 no label is equal to the null pointer. This may be passed around as an
2687 opaque pointer sized value as long as the bits are not inspected. This
2688 allows ``ptrtoint`` and arithmetic to be performed on these values so
2689 long as the original value is reconstituted before the ``indirectbr``
2692 Finally, some targets may provide defined semantics when using the value
2693 as the operand to an inline assembly, but that is target specific.
2697 Constant Expressions
2698 --------------------
2700 Constant expressions are used to allow expressions involving other
2701 constants to be used as constants. Constant expressions may be of any
2702 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2703 that does not have side effects (e.g. load and call are not supported).
2704 The following is the syntax for constant expressions:
2706 ``trunc (CST to TYPE)``
2707 Truncate a constant to another type. The bit size of CST must be
2708 larger than the bit size of TYPE. Both types must be integers.
2709 ``zext (CST to TYPE)``
2710 Zero extend a constant to another type. The bit size of CST must be
2711 smaller than the bit size of TYPE. Both types must be integers.
2712 ``sext (CST to TYPE)``
2713 Sign extend a constant to another type. The bit size of CST must be
2714 smaller than the bit size of TYPE. Both types must be integers.
2715 ``fptrunc (CST to TYPE)``
2716 Truncate a floating point constant to another floating point type.
2717 The size of CST must be larger than the size of TYPE. Both types
2718 must be floating point.
2719 ``fpext (CST to TYPE)``
2720 Floating point extend a constant to another type. The size of CST
2721 must be smaller or equal to the size of TYPE. Both types must be
2723 ``fptoui (CST to TYPE)``
2724 Convert a floating point constant to the corresponding unsigned
2725 integer constant. TYPE must be a scalar or vector integer type. CST
2726 must be of scalar or vector floating point type. Both CST and TYPE
2727 must be scalars, or vectors of the same number of elements. If the
2728 value won't fit in the integer type, the results are undefined.
2729 ``fptosi (CST to TYPE)``
2730 Convert a floating point constant to the corresponding signed
2731 integer constant. TYPE must be a scalar or vector integer type. CST
2732 must be of scalar or vector floating point type. Both CST and TYPE
2733 must be scalars, or vectors of the same number of elements. If the
2734 value won't fit in the integer type, the results are undefined.
2735 ``uitofp (CST to TYPE)``
2736 Convert an unsigned integer constant to the corresponding floating
2737 point constant. TYPE must be a scalar or vector floating point type.
2738 CST must be of scalar or vector integer type. Both CST and TYPE must
2739 be scalars, or vectors of the same number of elements. If the value
2740 won't fit in the floating point type, the results are undefined.
2741 ``sitofp (CST to TYPE)``
2742 Convert a signed integer constant to the corresponding floating
2743 point constant. TYPE must be a scalar or vector floating point type.
2744 CST must be of scalar or vector integer type. Both CST and TYPE must
2745 be scalars, or vectors of the same number of elements. If the value
2746 won't fit in the floating point type, the results are undefined.
2747 ``ptrtoint (CST to TYPE)``
2748 Convert a pointer typed constant to the corresponding integer
2749 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2750 pointer type. The ``CST`` value is zero extended, truncated, or
2751 unchanged to make it fit in ``TYPE``.
2752 ``inttoptr (CST to TYPE)``
2753 Convert an integer constant to a pointer constant. TYPE must be a
2754 pointer type. CST must be of integer type. The CST value is zero
2755 extended, truncated, or unchanged to make it fit in a pointer size.
2756 This one is *really* dangerous!
2757 ``bitcast (CST to TYPE)``
2758 Convert a constant, CST, to another TYPE. The constraints of the
2759 operands are the same as those for the :ref:`bitcast
2760 instruction <i_bitcast>`.
2761 ``addrspacecast (CST to TYPE)``
2762 Convert a constant pointer or constant vector of pointer, CST, to another
2763 TYPE in a different address space. The constraints of the operands are the
2764 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2765 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
2766 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2767 constants. As with the :ref:`getelementptr <i_getelementptr>`
2768 instruction, the index list may have zero or more indexes, which are
2769 required to make sense for the type of "pointer to TY".
2770 ``select (COND, VAL1, VAL2)``
2771 Perform the :ref:`select operation <i_select>` on constants.
2772 ``icmp COND (VAL1, VAL2)``
2773 Performs the :ref:`icmp operation <i_icmp>` on constants.
2774 ``fcmp COND (VAL1, VAL2)``
2775 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2776 ``extractelement (VAL, IDX)``
2777 Perform the :ref:`extractelement operation <i_extractelement>` on
2779 ``insertelement (VAL, ELT, IDX)``
2780 Perform the :ref:`insertelement operation <i_insertelement>` on
2782 ``shufflevector (VEC1, VEC2, IDXMASK)``
2783 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2785 ``extractvalue (VAL, IDX0, IDX1, ...)``
2786 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2787 constants. The index list is interpreted in a similar manner as
2788 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2789 least one index value must be specified.
2790 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2791 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2792 The index list is interpreted in a similar manner as indices in a
2793 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2794 value must be specified.
2795 ``OPCODE (LHS, RHS)``
2796 Perform the specified operation of the LHS and RHS constants. OPCODE
2797 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2798 binary <bitwiseops>` operations. The constraints on operands are
2799 the same as those for the corresponding instruction (e.g. no bitwise
2800 operations on floating point values are allowed).
2807 Inline Assembler Expressions
2808 ----------------------------
2810 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2811 Inline Assembly <moduleasm>`) through the use of a special value. This value
2812 represents the inline assembler as a template string (containing the
2813 instructions to emit), a list of operand constraints (stored as a string), a
2814 flag that indicates whether or not the inline asm expression has side effects,
2815 and a flag indicating whether the function containing the asm needs to align its
2816 stack conservatively.
2818 The template string supports argument substitution of the operands using "``$``"
2819 followed by a number, to indicate substitution of the given register/memory
2820 location, as specified by the constraint string. "``${NUM:MODIFIER}``" may also
2821 be used, where ``MODIFIER`` is a target-specific annotation for how to print the
2822 operand (See :ref:`inline-asm-modifiers`).
2824 A literal "``$``" may be included by using "``$$``" in the template. To include
2825 other special characters into the output, the usual "``\XX``" escapes may be
2826 used, just as in other strings. Note that after template substitution, the
2827 resulting assembly string is parsed by LLVM's integrated assembler unless it is
2828 disabled -- even when emitting a ``.s`` file -- and thus must contain assembly
2829 syntax known to LLVM.
2831 LLVM's support for inline asm is modeled closely on the requirements of Clang's
2832 GCC-compatible inline-asm support. Thus, the feature-set and the constraint and
2833 modifier codes listed here are similar or identical to those in GCC's inline asm
2834 support. However, to be clear, the syntax of the template and constraint strings
2835 described here is *not* the same as the syntax accepted by GCC and Clang, and,
2836 while most constraint letters are passed through as-is by Clang, some get
2837 translated to other codes when converting from the C source to the LLVM
2840 An example inline assembler expression is:
2842 .. code-block:: llvm
2844 i32 (i32) asm "bswap $0", "=r,r"
2846 Inline assembler expressions may **only** be used as the callee operand
2847 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2848 Thus, typically we have:
2850 .. code-block:: llvm
2852 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2854 Inline asms with side effects not visible in the constraint list must be
2855 marked as having side effects. This is done through the use of the
2856 '``sideeffect``' keyword, like so:
2858 .. code-block:: llvm
2860 call void asm sideeffect "eieio", ""()
2862 In some cases inline asms will contain code that will not work unless
2863 the stack is aligned in some way, such as calls or SSE instructions on
2864 x86, yet will not contain code that does that alignment within the asm.
2865 The compiler should make conservative assumptions about what the asm
2866 might contain and should generate its usual stack alignment code in the
2867 prologue if the '``alignstack``' keyword is present:
2869 .. code-block:: llvm
2871 call void asm alignstack "eieio", ""()
2873 Inline asms also support using non-standard assembly dialects. The
2874 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2875 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2876 the only supported dialects. An example is:
2878 .. code-block:: llvm
2880 call void asm inteldialect "eieio", ""()
2882 If multiple keywords appear the '``sideeffect``' keyword must come
2883 first, the '``alignstack``' keyword second and the '``inteldialect``'
2886 Inline Asm Constraint String
2887 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2889 The constraint list is a comma-separated string, each element containing one or
2890 more constraint codes.
2892 For each element in the constraint list an appropriate register or memory
2893 operand will be chosen, and it will be made available to assembly template
2894 string expansion as ``$0`` for the first constraint in the list, ``$1`` for the
2897 There are three different types of constraints, which are distinguished by a
2898 prefix symbol in front of the constraint code: Output, Input, and Clobber. The
2899 constraints must always be given in that order: outputs first, then inputs, then
2900 clobbers. They cannot be intermingled.
2902 There are also three different categories of constraint codes:
2904 - Register constraint. This is either a register class, or a fixed physical
2905 register. This kind of constraint will allocate a register, and if necessary,
2906 bitcast the argument or result to the appropriate type.
2907 - Memory constraint. This kind of constraint is for use with an instruction
2908 taking a memory operand. Different constraints allow for different addressing
2909 modes used by the target.
2910 - Immediate value constraint. This kind of constraint is for an integer or other
2911 immediate value which can be rendered directly into an instruction. The
2912 various target-specific constraints allow the selection of a value in the
2913 proper range for the instruction you wish to use it with.
2918 Output constraints are specified by an "``=``" prefix (e.g. "``=r``"). This
2919 indicates that the assembly will write to this operand, and the operand will
2920 then be made available as a return value of the ``asm`` expression. Output
2921 constraints do not consume an argument from the call instruction. (Except, see
2922 below about indirect outputs).
2924 Normally, it is expected that no output locations are written to by the assembly
2925 expression until *all* of the inputs have been read. As such, LLVM may assign
2926 the same register to an output and an input. If this is not safe (e.g. if the
2927 assembly contains two instructions, where the first writes to one output, and
2928 the second reads an input and writes to a second output), then the "``&``"
2929 modifier must be used (e.g. "``=&r``") to specify that the output is an
2930 "early-clobber" output. Marking an ouput as "early-clobber" ensures that LLVM
2931 will not use the same register for any inputs (other than an input tied to this
2937 Input constraints do not have a prefix -- just the constraint codes. Each input
2938 constraint will consume one argument from the call instruction. It is not
2939 permitted for the asm to write to any input register or memory location (unless
2940 that input is tied to an output). Note also that multiple inputs may all be
2941 assigned to the same register, if LLVM can determine that they necessarily all
2942 contain the same value.
2944 Instead of providing a Constraint Code, input constraints may also "tie"
2945 themselves to an output constraint, by providing an integer as the constraint
2946 string. Tied inputs still consume an argument from the call instruction, and
2947 take up a position in the asm template numbering as is usual -- they will simply
2948 be constrained to always use the same register as the output they've been tied
2949 to. For example, a constraint string of "``=r,0``" says to assign a register for
2950 output, and use that register as an input as well (it being the 0'th
2953 It is permitted to tie an input to an "early-clobber" output. In that case, no
2954 *other* input may share the same register as the input tied to the early-clobber
2955 (even when the other input has the same value).
2957 You may only tie an input to an output which has a register constraint, not a
2958 memory constraint. Only a single input may be tied to an output.
2960 There is also an "interesting" feature which deserves a bit of explanation: if a
2961 register class constraint allocates a register which is too small for the value
2962 type operand provided as input, the input value will be split into multiple
2963 registers, and all of them passed to the inline asm.
2965 However, this feature is often not as useful as you might think.
2967 Firstly, the registers are *not* guaranteed to be consecutive. So, on those
2968 architectures that have instructions which operate on multiple consecutive
2969 instructions, this is not an appropriate way to support them. (e.g. the 32-bit
2970 SparcV8 has a 64-bit load, which instruction takes a single 32-bit register. The
2971 hardware then loads into both the named register, and the next register. This
2972 feature of inline asm would not be useful to support that.)
2974 A few of the targets provide a template string modifier allowing explicit access
2975 to the second register of a two-register operand (e.g. MIPS ``L``, ``M``, and
2976 ``D``). On such an architecture, you can actually access the second allocated
2977 register (yet, still, not any subsequent ones). But, in that case, you're still
2978 probably better off simply splitting the value into two separate operands, for
2979 clarity. (e.g. see the description of the ``A`` constraint on X86, which,
2980 despite existing only for use with this feature, is not really a good idea to
2983 Indirect inputs and outputs
2984 """""""""""""""""""""""""""
2986 Indirect output or input constraints can be specified by the "``*``" modifier
2987 (which goes after the "``=``" in case of an output). This indicates that the asm
2988 will write to or read from the contents of an *address* provided as an input
2989 argument. (Note that in this way, indirect outputs act more like an *input* than
2990 an output: just like an input, they consume an argument of the call expression,
2991 rather than producing a return value. An indirect output constraint is an
2992 "output" only in that the asm is expected to write to the contents of the input
2993 memory location, instead of just read from it).
2995 This is most typically used for memory constraint, e.g. "``=*m``", to pass the
2996 address of a variable as a value.
2998 It is also possible to use an indirect *register* constraint, but only on output
2999 (e.g. "``=*r``"). This will cause LLVM to allocate a register for an output
3000 value normally, and then, separately emit a store to the address provided as
3001 input, after the provided inline asm. (It's not clear what value this
3002 functionality provides, compared to writing the store explicitly after the asm
3003 statement, and it can only produce worse code, since it bypasses many
3004 optimization passes. I would recommend not using it.)
3010 A clobber constraint is indicated by a "``~``" prefix. A clobber does not
3011 consume an input operand, nor generate an output. Clobbers cannot use any of the
3012 general constraint code letters -- they may use only explicit register
3013 constraints, e.g. "``~{eax}``". The one exception is that a clobber string of
3014 "``~{memory}``" indicates that the assembly writes to arbitrary undeclared
3015 memory locations -- not only the memory pointed to by a declared indirect
3021 After a potential prefix comes constraint code, or codes.
3023 A Constraint Code is either a single letter (e.g. "``r``"), a "``^``" character
3024 followed by two letters (e.g. "``^wc``"), or "``{``" register-name "``}``"
3027 The one and two letter constraint codes are typically chosen to be the same as
3028 GCC's constraint codes.
3030 A single constraint may include one or more than constraint code in it, leaving
3031 it up to LLVM to choose which one to use. This is included mainly for
3032 compatibility with the translation of GCC inline asm coming from clang.
3034 There are two ways to specify alternatives, and either or both may be used in an
3035 inline asm constraint list:
3037 1) Append the codes to each other, making a constraint code set. E.g. "``im``"
3038 or "``{eax}m``". This means "choose any of the options in the set". The
3039 choice of constraint is made independently for each constraint in the
3042 2) Use "``|``" between constraint code sets, creating alternatives. Every
3043 constraint in the constraint list must have the same number of alternative
3044 sets. With this syntax, the same alternative in *all* of the items in the
3045 constraint list will be chosen together.
3047 Putting those together, you might have a two operand constraint string like
3048 ``"rm|r,ri|rm"``. This indicates that if operand 0 is ``r`` or ``m``, then
3049 operand 1 may be one of ``r`` or ``i``. If operand 0 is ``r``, then operand 1
3050 may be one of ``r`` or ``m``. But, operand 0 and 1 cannot both be of type m.
3052 However, the use of either of the alternatives features is *NOT* recommended, as
3053 LLVM is not able to make an intelligent choice about which one to use. (At the
3054 point it currently needs to choose, not enough information is available to do so
3055 in a smart way.) Thus, it simply tries to make a choice that's most likely to
3056 compile, not one that will be optimal performance. (e.g., given "``rm``", it'll
3057 always choose to use memory, not registers). And, if given multiple registers,
3058 or multiple register classes, it will simply choose the first one. (In fact, it
3059 doesn't currently even ensure explicitly specified physical registers are
3060 unique, so specifying multiple physical registers as alternatives, like
3061 ``{r11}{r12},{r11}{r12}``, will assign r11 to both operands, not at all what was
3064 Supported Constraint Code List
3065 """"""""""""""""""""""""""""""
3067 The constraint codes are, in general, expected to behave the same way they do in
3068 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3069 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3070 and GCC likely indicates a bug in LLVM.
3072 Some constraint codes are typically supported by all targets:
3074 - ``r``: A register in the target's general purpose register class.
3075 - ``m``: A memory address operand. It is target-specific what addressing modes
3076 are supported, typical examples are register, or register + register offset,
3077 or register + immediate offset (of some target-specific size).
3078 - ``i``: An integer constant (of target-specific width). Allows either a simple
3079 immediate, or a relocatable value.
3080 - ``n``: An integer constant -- *not* including relocatable values.
3081 - ``s``: An integer constant, but allowing *only* relocatable values.
3082 - ``X``: Allows an operand of any kind, no constraint whatsoever. Typically
3083 useful to pass a label for an asm branch or call.
3085 .. FIXME: but that surely isn't actually okay to jump out of an asm
3086 block without telling llvm about the control transfer???)
3088 - ``{register-name}``: Requires exactly the named physical register.
3090 Other constraints are target-specific:
3094 - ``z``: An immediate integer 0. Outputs ``WZR`` or ``XZR``, as appropriate.
3095 - ``I``: An immediate integer valid for an ``ADD`` or ``SUB`` instruction,
3096 i.e. 0 to 4095 with optional shift by 12.
3097 - ``J``: An immediate integer that, when negated, is valid for an ``ADD`` or
3098 ``SUB`` instruction, i.e. -1 to -4095 with optional left shift by 12.
3099 - ``K``: An immediate integer that is valid for the 'bitmask immediate 32' of a
3100 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 32-bit register.
3101 - ``L``: An immediate integer that is valid for the 'bitmask immediate 64' of a
3102 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 64-bit register.
3103 - ``M``: An immediate integer for use with the ``MOV`` assembly alias on a
3104 32-bit register. This is a superset of ``K``: in addition to the bitmask
3105 immediate, also allows immediate integers which can be loaded with a single
3106 ``MOVZ`` or ``MOVL`` instruction.
3107 - ``N``: An immediate integer for use with the ``MOV`` assembly alias on a
3108 64-bit register. This is a superset of ``L``.
3109 - ``Q``: Memory address operand must be in a single register (no
3110 offsets). (However, LLVM currently does this for the ``m`` constraint as
3112 - ``r``: A 32 or 64-bit integer register (W* or X*).
3113 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register.
3114 - ``x``: A lower 128-bit floating-point/SIMD register (``V0`` to ``V15``).
3118 - ``r``: A 32 or 64-bit integer register.
3119 - ``[0-9]v``: The 32-bit VGPR register, number 0-9.
3120 - ``[0-9]s``: The 32-bit SGPR register, number 0-9.
3125 - ``Q``, ``Um``, ``Un``, ``Uq``, ``Us``, ``Ut``, ``Uv``, ``Uy``: Memory address
3126 operand. Treated the same as operand ``m``, at the moment.
3128 ARM and ARM's Thumb2 mode:
3130 - ``j``: An immediate integer between 0 and 65535 (valid for ``MOVW``)
3131 - ``I``: An immediate integer valid for a data-processing instruction.
3132 - ``J``: An immediate integer between -4095 and 4095.
3133 - ``K``: An immediate integer whose bitwise inverse is valid for a
3134 data-processing instruction. (Can be used with template modifier "``B``" to
3135 print the inverted value).
3136 - ``L``: An immediate integer whose negation is valid for a data-processing
3137 instruction. (Can be used with template modifier "``n``" to print the negated
3139 - ``M``: A power of two or a integer between 0 and 32.
3140 - ``N``: Invalid immediate constraint.
3141 - ``O``: Invalid immediate constraint.
3142 - ``r``: A general-purpose 32-bit integer register (``r0-r15``).
3143 - ``l``: In Thumb2 mode, low 32-bit GPR registers (``r0-r7``). In ARM mode, same
3145 - ``h``: In Thumb2 mode, a high 32-bit GPR register (``r8-r15``). In ARM mode,
3147 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3148 ``d0-d31``, or ``q0-q15``.
3149 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3150 ``d0-d7``, or ``q0-q3``.
3151 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3156 - ``I``: An immediate integer between 0 and 255.
3157 - ``J``: An immediate integer between -255 and -1.
3158 - ``K``: An immediate integer between 0 and 255, with optional left-shift by
3160 - ``L``: An immediate integer between -7 and 7.
3161 - ``M``: An immediate integer which is a multiple of 4 between 0 and 1020.
3162 - ``N``: An immediate integer between 0 and 31.
3163 - ``O``: An immediate integer which is a multiple of 4 between -508 and 508.
3164 - ``r``: A low 32-bit GPR register (``r0-r7``).
3165 - ``l``: A low 32-bit GPR register (``r0-r7``).
3166 - ``h``: A high GPR register (``r0-r7``).
3167 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3168 ``d0-d31``, or ``q0-q15``.
3169 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3170 ``d0-d7``, or ``q0-q3``.
3171 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3177 - ``o``, ``v``: A memory address operand, treated the same as constraint ``m``,
3179 - ``r``: A 32 or 64-bit register.
3183 - ``r``: An 8 or 16-bit register.
3187 - ``I``: An immediate signed 16-bit integer.
3188 - ``J``: An immediate integer zero.
3189 - ``K``: An immediate unsigned 16-bit integer.
3190 - ``L``: An immediate 32-bit integer, where the lower 16 bits are 0.
3191 - ``N``: An immediate integer between -65535 and -1.
3192 - ``O``: An immediate signed 15-bit integer.
3193 - ``P``: An immediate integer between 1 and 65535.
3194 - ``m``: A memory address operand. In MIPS-SE mode, allows a base address
3195 register plus 16-bit immediate offset. In MIPS mode, just a base register.
3196 - ``R``: A memory address operand. In MIPS-SE mode, allows a base address
3197 register plus a 9-bit signed offset. In MIPS mode, the same as constraint
3199 - ``ZC``: A memory address operand, suitable for use in a ``pref``, ``ll``, or
3200 ``sc`` instruction on the given subtarget (details vary).
3201 - ``r``, ``d``, ``y``: A 32 or 64-bit GPR register.
3202 - ``f``: A 32 or 64-bit FPU register (``F0-F31``), or a 128-bit MSA register
3204 - ``c``: A 32-bit or 64-bit GPR register suitable for indirect jump (always
3206 - ``l``: The ``lo`` register, 32 or 64-bit.
3211 - ``b``: A 1-bit integer register.
3212 - ``c`` or ``h``: A 16-bit integer register.
3213 - ``r``: A 32-bit integer register.
3214 - ``l`` or ``N``: A 64-bit integer register.
3215 - ``f``: A 32-bit float register.
3216 - ``d``: A 64-bit float register.
3221 - ``I``: An immediate signed 16-bit integer.
3222 - ``J``: An immediate unsigned 16-bit integer, shifted left 16 bits.
3223 - ``K``: An immediate unsigned 16-bit integer.
3224 - ``L``: An immediate signed 16-bit integer, shifted left 16 bits.
3225 - ``M``: An immediate integer greater than 31.
3226 - ``N``: An immediate integer that is an exact power of 2.
3227 - ``O``: The immediate integer constant 0.
3228 - ``P``: An immediate integer constant whose negation is a signed 16-bit
3230 - ``es``, ``o``, ``Q``, ``Z``, ``Zy``: A memory address operand, currently
3231 treated the same as ``m``.
3232 - ``r``: A 32 or 64-bit integer register.
3233 - ``b``: A 32 or 64-bit integer register, excluding ``R0`` (that is:
3235 - ``f``: A 32 or 64-bit float register (``F0-F31``), or when QPX is enabled, a
3236 128 or 256-bit QPX register (``Q0-Q31``; aliases the ``F`` registers).
3237 - ``v``: For ``4 x f32`` or ``4 x f64`` types, when QPX is enabled, a
3238 128 or 256-bit QPX register (``Q0-Q31``), otherwise a 128-bit
3239 altivec vector register (``V0-V31``).
3241 .. FIXME: is this a bug that v accepts QPX registers? I think this
3242 is supposed to only use the altivec vector registers?
3244 - ``y``: Condition register (``CR0-CR7``).
3245 - ``wc``: An individual CR bit in a CR register.
3246 - ``wa``, ``wd``, ``wf``: Any 128-bit VSX vector register, from the full VSX
3247 register set (overlapping both the floating-point and vector register files).
3248 - ``ws``: A 32 or 64-bit floating point register, from the full VSX register
3253 - ``I``: An immediate 13-bit signed integer.
3254 - ``r``: A 32-bit integer register.
3258 - ``I``: An immediate unsigned 8-bit integer.
3259 - ``J``: An immediate unsigned 12-bit integer.
3260 - ``K``: An immediate signed 16-bit integer.
3261 - ``L``: An immediate signed 20-bit integer.
3262 - ``M``: An immediate integer 0x7fffffff.
3263 - ``Q``, ``R``, ``S``, ``T``: A memory address operand, treated the same as
3264 ``m``, at the moment.
3265 - ``r`` or ``d``: A 32, 64, or 128-bit integer register.
3266 - ``a``: A 32, 64, or 128-bit integer address register (excludes R0, which in an
3267 address context evaluates as zero).
3268 - ``h``: A 32-bit value in the high part of a 64bit data register
3270 - ``f``: A 32, 64, or 128-bit floating point register.
3274 - ``I``: An immediate integer between 0 and 31.
3275 - ``J``: An immediate integer between 0 and 64.
3276 - ``K``: An immediate signed 8-bit integer.
3277 - ``L``: An immediate integer, 0xff or 0xffff or (in 64-bit mode only)
3279 - ``M``: An immediate integer between 0 and 3.
3280 - ``N``: An immediate unsigned 8-bit integer.
3281 - ``O``: An immediate integer between 0 and 127.
3282 - ``e``: An immediate 32-bit signed integer.
3283 - ``Z``: An immediate 32-bit unsigned integer.
3284 - ``o``, ``v``: Treated the same as ``m``, at the moment.
3285 - ``q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3286 ``l`` integer register. On X86-32, this is the ``a``, ``b``, ``c``, and ``d``
3287 registers, and on X86-64, it is all of the integer registers.
3288 - ``Q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3289 ``h`` integer register. This is the ``a``, ``b``, ``c``, and ``d`` registers.
3290 - ``r`` or ``l``: An 8, 16, 32, or 64-bit integer register.
3291 - ``R``: An 8, 16, 32, or 64-bit "legacy" integer register -- one which has
3292 existed since i386, and can be accessed without the REX prefix.
3293 - ``f``: A 32, 64, or 80-bit '387 FPU stack pseudo-register.
3294 - ``y``: A 64-bit MMX register, if MMX is enabled.
3295 - ``x``: If SSE is enabled: a 32 or 64-bit scalar operand, or 128-bit vector
3296 operand in a SSE register. If AVX is also enabled, can also be a 256-bit
3297 vector operand in an AVX register. If AVX-512 is also enabled, can also be a
3298 512-bit vector operand in an AVX512 register, Otherwise, an error.
3299 - ``Y``: The same as ``x``, if *SSE2* is enabled, otherwise an error.
3300 - ``A``: Special case: allocates EAX first, then EDX, for a single operand (in
3301 32-bit mode, a 64-bit integer operand will get split into two registers). It
3302 is not recommended to use this constraint, as in 64-bit mode, the 64-bit
3303 operand will get allocated only to RAX -- if two 32-bit operands are needed,
3304 you're better off splitting it yourself, before passing it to the asm
3309 - ``r``: A 32-bit integer register.
3312 .. _inline-asm-modifiers:
3314 Asm template argument modifiers
3315 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3317 In the asm template string, modifiers can be used on the operand reference, like
3320 The modifiers are, in general, expected to behave the same way they do in
3321 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3322 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3323 and GCC likely indicates a bug in LLVM.
3327 - ``c``: Print an immediate integer constant unadorned, without
3328 the target-specific immediate punctuation (e.g. no ``$`` prefix).
3329 - ``n``: Negate and print immediate integer constant unadorned, without the
3330 target-specific immediate punctuation (e.g. no ``$`` prefix).
3331 - ``l``: Print as an unadorned label, without the target-specific label
3332 punctuation (e.g. no ``$`` prefix).
3336 - ``w``: Print a GPR register with a ``w*`` name instead of ``x*`` name. E.g.,
3337 instead of ``x30``, print ``w30``.
3338 - ``x``: Print a GPR register with a ``x*`` name. (this is the default, anyhow).
3339 - ``b``, ``h``, ``s``, ``d``, ``q``: Print a floating-point/SIMD register with a
3340 ``b*``, ``h*``, ``s*``, ``d*``, or ``q*`` name, rather than the default of
3349 - ``a``: Print an operand as an address (with ``[`` and ``]`` surrounding a
3353 - ``y``: Print a VFP single-precision register as an indexed double (e.g. print
3354 as ``d4[1]`` instead of ``s9``)
3355 - ``B``: Bitwise invert and print an immediate integer constant without ``#``
3357 - ``L``: Print the low 16-bits of an immediate integer constant.
3358 - ``M``: Print as a register set suitable for ldm/stm. Also prints *all*
3359 register operands subsequent to the specified one (!), so use carefully.
3360 - ``Q``: Print the low-order register of a register-pair, or the low-order
3361 register of a two-register operand.
3362 - ``R``: Print the high-order register of a register-pair, or the high-order
3363 register of a two-register operand.
3364 - ``H``: Print the second register of a register-pair. (On a big-endian system,
3365 ``H`` is equivalent to ``Q``, and on little-endian system, ``H`` is equivalent
3368 .. FIXME: H doesn't currently support printing the second register
3369 of a two-register operand.
3371 - ``e``: Print the low doubleword register of a NEON quad register.
3372 - ``f``: Print the high doubleword register of a NEON quad register.
3373 - ``m``: Print the base register of a memory operand without the ``[`` and ``]``
3378 - ``L``: Print the second register of a two-register operand. Requires that it
3379 has been allocated consecutively to the first.
3381 .. FIXME: why is it restricted to consecutive ones? And there's
3382 nothing that ensures that happens, is there?
3384 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3385 nothing. Used to print 'addi' vs 'add' instructions.
3389 No additional modifiers.
3393 - ``X``: Print an immediate integer as hexadecimal
3394 - ``x``: Print the low 16 bits of an immediate integer as hexadecimal.
3395 - ``d``: Print an immediate integer as decimal.
3396 - ``m``: Subtract one and print an immediate integer as decimal.
3397 - ``z``: Print $0 if an immediate zero, otherwise print normally.
3398 - ``L``: Print the low-order register of a two-register operand, or prints the
3399 address of the low-order word of a double-word memory operand.
3401 .. FIXME: L seems to be missing memory operand support.
3403 - ``M``: Print the high-order register of a two-register operand, or prints the
3404 address of the high-order word of a double-word memory operand.
3406 .. FIXME: M seems to be missing memory operand support.
3408 - ``D``: Print the second register of a two-register operand, or prints the
3409 second word of a double-word memory operand. (On a big-endian system, ``D`` is
3410 equivalent to ``L``, and on little-endian system, ``D`` is equivalent to
3420 - ``L``: Print the second register of a two-register operand. Requires that it
3421 has been allocated consecutively to the first.
3423 .. FIXME: why is it restricted to consecutive ones? And there's
3424 nothing that ensures that happens, is there?
3426 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3427 nothing. Used to print 'addi' vs 'add' instructions.
3428 - ``y``: For a memory operand, prints formatter for a two-register X-form
3429 instruction. (Currently always prints ``r0,OPERAND``).
3430 - ``U``: Prints 'u' if the memory operand is an update form, and nothing
3431 otherwise. (NOTE: LLVM does not support update form, so this will currently
3432 always print nothing)
3433 - ``X``: Prints 'x' if the memory operand is an indexed form. (NOTE: LLVM does
3434 not support indexed form, so this will currently always print nothing)
3442 SystemZ implements only ``n``, and does *not* support any of the other
3443 target-independent modifiers.
3447 - ``c``: Print an unadorned integer or symbol name. (The latter is
3448 target-specific behavior for this typically target-independent modifier).
3449 - ``A``: Print a register name with a '``*``' before it.
3450 - ``b``: Print an 8-bit register name (e.g. ``al``); do nothing on a memory
3452 - ``h``: Print the upper 8-bit register name (e.g. ``ah``); do nothing on a
3454 - ``w``: Print the 16-bit register name (e.g. ``ax``); do nothing on a memory
3456 - ``k``: Print the 32-bit register name (e.g. ``eax``); do nothing on a memory
3458 - ``q``: Print the 64-bit register name (e.g. ``rax``), if 64-bit registers are
3459 available, otherwise the 32-bit register name; do nothing on a memory operand.
3460 - ``n``: Negate and print an unadorned integer, or, for operands other than an
3461 immediate integer (e.g. a relocatable symbol expression), print a '-' before
3462 the operand. (The behavior for relocatable symbol expressions is a
3463 target-specific behavior for this typically target-independent modifier)
3464 - ``H``: Print a memory reference with additional offset +8.
3465 - ``P``: Print a memory reference or operand for use as the argument of a call
3466 instruction. (E.g. omit ``(rip)``, even though it's PC-relative.)
3470 No additional modifiers.
3476 The call instructions that wrap inline asm nodes may have a
3477 "``!srcloc``" MDNode attached to it that contains a list of constant
3478 integers. If present, the code generator will use the integer as the
3479 location cookie value when report errors through the ``LLVMContext``
3480 error reporting mechanisms. This allows a front-end to correlate backend
3481 errors that occur with inline asm back to the source code that produced
3484 .. code-block:: llvm
3486 call void asm sideeffect "something bad", ""(), !srcloc !42
3488 !42 = !{ i32 1234567 }
3490 It is up to the front-end to make sense of the magic numbers it places
3491 in the IR. If the MDNode contains multiple constants, the code generator
3492 will use the one that corresponds to the line of the asm that the error
3500 LLVM IR allows metadata to be attached to instructions in the program
3501 that can convey extra information about the code to the optimizers and
3502 code generator. One example application of metadata is source-level
3503 debug information. There are two metadata primitives: strings and nodes.
3505 Metadata does not have a type, and is not a value. If referenced from a
3506 ``call`` instruction, it uses the ``metadata`` type.
3508 All metadata are identified in syntax by a exclamation point ('``!``').
3510 .. _metadata-string:
3512 Metadata Nodes and Metadata Strings
3513 -----------------------------------
3515 A metadata string is a string surrounded by double quotes. It can
3516 contain any character by escaping non-printable characters with
3517 "``\xx``" where "``xx``" is the two digit hex code. For example:
3520 Metadata nodes are represented with notation similar to structure
3521 constants (a comma separated list of elements, surrounded by braces and
3522 preceded by an exclamation point). Metadata nodes can have any values as
3523 their operand. For example:
3525 .. code-block:: llvm
3527 !{ !"test\00", i32 10}
3529 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
3531 .. code-block:: llvm
3533 !0 = distinct !{!"test\00", i32 10}
3535 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
3536 content. They can also occur when transformations cause uniquing collisions
3537 when metadata operands change.
3539 A :ref:`named metadata <namedmetadatastructure>` is a collection of
3540 metadata nodes, which can be looked up in the module symbol table. For
3543 .. code-block:: llvm
3547 Metadata can be used as function arguments. Here ``llvm.dbg.value``
3548 function is using two metadata arguments:
3550 .. code-block:: llvm
3552 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
3554 Metadata can be attached with an instruction. Here metadata ``!21`` is
3555 attached to the ``add`` instruction using the ``!dbg`` identifier:
3557 .. code-block:: llvm
3559 %indvar.next = add i64 %indvar, 1, !dbg !21
3561 More information about specific metadata nodes recognized by the
3562 optimizers and code generator is found below.
3564 .. _specialized-metadata:
3566 Specialized Metadata Nodes
3567 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3569 Specialized metadata nodes are custom data structures in metadata (as opposed
3570 to generic tuples). Their fields are labelled, and can be specified in any
3573 These aren't inherently debug info centric, but currently all the specialized
3574 metadata nodes are related to debug info.
3581 ``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
3582 ``retainedTypes:``, ``subprograms:``, ``globals:`` and ``imports:`` fields are
3583 tuples containing the debug info to be emitted along with the compile unit,
3584 regardless of code optimizations (some nodes are only emitted if there are
3585 references to them from instructions).
3587 .. code-block:: llvm
3589 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
3590 isOptimized: true, flags: "-O2", runtimeVersion: 2,
3591 splitDebugFilename: "abc.debug", emissionKind: 1,
3592 enums: !2, retainedTypes: !3, subprograms: !4,
3593 globals: !5, imports: !6)
3595 Compile unit descriptors provide the root scope for objects declared in a
3596 specific compilation unit. File descriptors are defined using this scope.
3597 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
3598 keep track of subprograms, global variables, type information, and imported
3599 entities (declarations and namespaces).
3606 ``DIFile`` nodes represent files. The ``filename:`` can include slashes.
3608 .. code-block:: llvm
3610 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
3612 Files are sometimes used in ``scope:`` fields, and are the only valid target
3613 for ``file:`` fields.
3620 ``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
3621 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
3623 .. code-block:: llvm
3625 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3626 encoding: DW_ATE_unsigned_char)
3627 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
3629 The ``encoding:`` describes the details of the type. Usually it's one of the
3632 .. code-block:: llvm
3638 DW_ATE_signed_char = 6
3640 DW_ATE_unsigned_char = 8
3642 .. _DISubroutineType:
3647 ``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
3648 refers to a tuple; the first operand is the return type, while the rest are the
3649 types of the formal arguments in order. If the first operand is ``null``, that
3650 represents a function with no return value (such as ``void foo() {}`` in C++).
3652 .. code-block:: llvm
3654 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
3655 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
3656 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
3663 ``DIDerivedType`` nodes represent types derived from other types, such as
3666 .. code-block:: llvm
3668 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3669 encoding: DW_ATE_unsigned_char)
3670 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
3673 The following ``tag:`` values are valid:
3675 .. code-block:: llvm
3677 DW_TAG_formal_parameter = 5
3679 DW_TAG_pointer_type = 15
3680 DW_TAG_reference_type = 16
3682 DW_TAG_ptr_to_member_type = 31
3683 DW_TAG_const_type = 38
3684 DW_TAG_volatile_type = 53
3685 DW_TAG_restrict_type = 55
3687 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
3688 <DICompositeType>` or :ref:`subprogram <DISubprogram>`. The type of the member
3689 is the ``baseType:``. The ``offset:`` is the member's bit offset.
3690 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
3691 argument of a subprogram.
3693 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3695 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3696 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3699 Note that the ``void *`` type is expressed as a type derived from NULL.
3701 .. _DICompositeType:
3706 ``DICompositeType`` nodes represent types composed of other types, like
3707 structures and unions. ``elements:`` points to a tuple of the composed types.
3709 If the source language supports ODR, the ``identifier:`` field gives the unique
3710 identifier used for type merging between modules. When specified, other types
3711 can refer to composite types indirectly via a :ref:`metadata string
3712 <metadata-string>` that matches their identifier.
3714 .. code-block:: llvm
3716 !0 = !DIEnumerator(name: "SixKind", value: 7)
3717 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3718 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3719 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
3720 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3721 elements: !{!0, !1, !2})
3723 The following ``tag:`` values are valid:
3725 .. code-block:: llvm
3727 DW_TAG_array_type = 1
3728 DW_TAG_class_type = 2
3729 DW_TAG_enumeration_type = 4
3730 DW_TAG_structure_type = 19
3731 DW_TAG_union_type = 23
3732 DW_TAG_subroutine_type = 21
3733 DW_TAG_inheritance = 28
3736 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
3737 descriptors <DISubrange>`, each representing the range of subscripts at that
3738 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3739 array type is a native packed vector.
3741 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
3742 descriptors <DIEnumerator>`, each representing the definition of an enumeration
3743 value for the set. All enumeration type descriptors are collected in the
3744 ``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
3746 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3747 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
3748 <DIDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
3755 ``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3756 :ref:`DICompositeType`. ``count: -1`` indicates an empty array.
3758 .. code-block:: llvm
3760 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
3761 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
3762 !2 = !DISubrange(count: -1) ; empty array.
3769 ``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3770 variants of :ref:`DICompositeType`.
3772 .. code-block:: llvm
3774 !0 = !DIEnumerator(name: "SixKind", value: 7)
3775 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3776 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3778 DITemplateTypeParameter
3779 """""""""""""""""""""""
3781 ``DITemplateTypeParameter`` nodes represent type parameters to generic source
3782 language constructs. They are used (optionally) in :ref:`DICompositeType` and
3783 :ref:`DISubprogram` ``templateParams:`` fields.
3785 .. code-block:: llvm
3787 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
3789 DITemplateValueParameter
3790 """"""""""""""""""""""""
3792 ``DITemplateValueParameter`` nodes represent value parameters to generic source
3793 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3794 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3795 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
3796 :ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
3798 .. code-block:: llvm
3800 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
3805 ``DINamespace`` nodes represent namespaces in the source language.
3807 .. code-block:: llvm
3809 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
3814 ``DIGlobalVariable`` nodes represent global variables in the source language.
3816 .. code-block:: llvm
3818 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
3819 file: !2, line: 7, type: !3, isLocal: true,
3820 isDefinition: false, variable: i32* @foo,
3823 All global variables should be referenced by the `globals:` field of a
3824 :ref:`compile unit <DICompileUnit>`.
3831 ``DISubprogram`` nodes represent functions from the source language. The
3832 ``variables:`` field points at :ref:`variables <DILocalVariable>` that must be
3833 retained, even if their IR counterparts are optimized out of the IR. The
3834 ``type:`` field must point at an :ref:`DISubroutineType`.
3836 .. code-block:: llvm
3838 !0 = !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
3839 file: !2, line: 7, type: !3, isLocal: true,
3840 isDefinition: false, scopeLine: 8, containingType: !4,
3841 virtuality: DW_VIRTUALITY_pure_virtual, virtualIndex: 10,
3842 flags: DIFlagPrototyped, isOptimized: true,
3843 function: void ()* @_Z3foov,
3844 templateParams: !5, declaration: !6, variables: !7)
3851 ``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
3852 <DISubprogram>`. The line number and column numbers are used to dinstinguish
3853 two lexical blocks at same depth. They are valid targets for ``scope:``
3856 .. code-block:: llvm
3858 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
3860 Usually lexical blocks are ``distinct`` to prevent node merging based on
3863 .. _DILexicalBlockFile:
3868 ``DILexicalBlockFile`` nodes are used to discriminate between sections of a
3869 :ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
3870 indicate textual inclusion, or the ``discriminator:`` field can be used to
3871 discriminate between control flow within a single block in the source language.
3873 .. code-block:: llvm
3875 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
3876 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
3877 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
3884 ``DILocation`` nodes represent source debug locations. The ``scope:`` field is
3885 mandatory, and points at an :ref:`DILexicalBlockFile`, an
3886 :ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
3888 .. code-block:: llvm
3890 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
3892 .. _DILocalVariable:
3897 ``DILocalVariable`` nodes represent local variables in the source language.
3898 Instead of ``DW_TAG_variable``, they use LLVM-specific fake tags to
3899 discriminate between local variables (``DW_TAG_auto_variable``) and subprogram
3900 arguments (``DW_TAG_arg_variable``). In the latter case, the ``arg:`` field
3901 specifies the argument position, and this variable will be included in the
3902 ``variables:`` field of its :ref:`DISubprogram`.
3904 .. code-block:: llvm
3906 !0 = !DILocalVariable(tag: DW_TAG_arg_variable, name: "this", arg: 0,
3907 scope: !3, file: !2, line: 7, type: !3,
3908 flags: DIFlagArtificial)
3909 !1 = !DILocalVariable(tag: DW_TAG_arg_variable, name: "x", arg: 1,
3910 scope: !4, file: !2, line: 7, type: !3)
3911 !1 = !DILocalVariable(tag: DW_TAG_auto_variable, name: "y",
3912 scope: !5, file: !2, line: 7, type: !3)
3917 ``DIExpression`` nodes represent DWARF expression sequences. They are used in
3918 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
3919 describe how the referenced LLVM variable relates to the source language
3922 The current supported vocabulary is limited:
3924 - ``DW_OP_deref`` dereferences the working expression.
3925 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
3926 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
3927 here, respectively) of the variable piece from the working expression.
3929 .. code-block:: llvm
3931 !0 = !DIExpression(DW_OP_deref)
3932 !1 = !DIExpression(DW_OP_plus, 3)
3933 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
3934 !3 = !DIExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
3939 ``DIObjCProperty`` nodes represent Objective-C property nodes.
3941 .. code-block:: llvm
3943 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
3944 getter: "getFoo", attributes: 7, type: !2)
3949 ``DIImportedEntity`` nodes represent entities (such as modules) imported into a
3952 .. code-block:: llvm
3954 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
3955 entity: !1, line: 7)
3960 In LLVM IR, memory does not have types, so LLVM's own type system is not
3961 suitable for doing TBAA. Instead, metadata is added to the IR to
3962 describe a type system of a higher level language. This can be used to
3963 implement typical C/C++ TBAA, but it can also be used to implement
3964 custom alias analysis behavior for other languages.
3966 The current metadata format is very simple. TBAA metadata nodes have up
3967 to three fields, e.g.:
3969 .. code-block:: llvm
3971 !0 = !{ !"an example type tree" }
3972 !1 = !{ !"int", !0 }
3973 !2 = !{ !"float", !0 }
3974 !3 = !{ !"const float", !2, i64 1 }
3976 The first field is an identity field. It can be any value, usually a
3977 metadata string, which uniquely identifies the type. The most important
3978 name in the tree is the name of the root node. Two trees with different
3979 root node names are entirely disjoint, even if they have leaves with
3982 The second field identifies the type's parent node in the tree, or is
3983 null or omitted for a root node. A type is considered to alias all of
3984 its descendants and all of its ancestors in the tree. Also, a type is
3985 considered to alias all types in other trees, so that bitcode produced
3986 from multiple front-ends is handled conservatively.
3988 If the third field is present, it's an integer which if equal to 1
3989 indicates that the type is "constant" (meaning
3990 ``pointsToConstantMemory`` should return true; see `other useful
3991 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
3993 '``tbaa.struct``' Metadata
3994 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3996 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
3997 aggregate assignment operations in C and similar languages, however it
3998 is defined to copy a contiguous region of memory, which is more than
3999 strictly necessary for aggregate types which contain holes due to
4000 padding. Also, it doesn't contain any TBAA information about the fields
4003 ``!tbaa.struct`` metadata can describe which memory subregions in a
4004 memcpy are padding and what the TBAA tags of the struct are.
4006 The current metadata format is very simple. ``!tbaa.struct`` metadata
4007 nodes are a list of operands which are in conceptual groups of three.
4008 For each group of three, the first operand gives the byte offset of a
4009 field in bytes, the second gives its size in bytes, and the third gives
4012 .. code-block:: llvm
4014 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
4016 This describes a struct with two fields. The first is at offset 0 bytes
4017 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
4018 and has size 4 bytes and has tbaa tag !2.
4020 Note that the fields need not be contiguous. In this example, there is a
4021 4 byte gap between the two fields. This gap represents padding which
4022 does not carry useful data and need not be preserved.
4024 '``noalias``' and '``alias.scope``' Metadata
4025 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4027 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
4028 noalias memory-access sets. This means that some collection of memory access
4029 instructions (loads, stores, memory-accessing calls, etc.) that carry
4030 ``noalias`` metadata can specifically be specified not to alias with some other
4031 collection of memory access instructions that carry ``alias.scope`` metadata.
4032 Each type of metadata specifies a list of scopes where each scope has an id and
4033 a domain. When evaluating an aliasing query, if for some domain, the set
4034 of scopes with that domain in one instruction's ``alias.scope`` list is a
4035 subset of (or equal to) the set of scopes for that domain in another
4036 instruction's ``noalias`` list, then the two memory accesses are assumed not to
4039 The metadata identifying each domain is itself a list containing one or two
4040 entries. The first entry is the name of the domain. Note that if the name is a
4041 string then it can be combined accross functions and translation units. A
4042 self-reference can be used to create globally unique domain names. A
4043 descriptive string may optionally be provided as a second list entry.
4045 The metadata identifying each scope is also itself a list containing two or
4046 three entries. The first entry is the name of the scope. Note that if the name
4047 is a string then it can be combined accross functions and translation units. A
4048 self-reference can be used to create globally unique scope names. A metadata
4049 reference to the scope's domain is the second entry. A descriptive string may
4050 optionally be provided as a third list entry.
4054 .. code-block:: llvm
4056 ; Two scope domains:
4060 ; Some scopes in these domains:
4066 !5 = !{!4} ; A list containing only scope !4
4070 ; These two instructions don't alias:
4071 %0 = load float, float* %c, align 4, !alias.scope !5
4072 store float %0, float* %arrayidx.i, align 4, !noalias !5
4074 ; These two instructions also don't alias (for domain !1, the set of scopes
4075 ; in the !alias.scope equals that in the !noalias list):
4076 %2 = load float, float* %c, align 4, !alias.scope !5
4077 store float %2, float* %arrayidx.i2, align 4, !noalias !6
4079 ; These two instructions may alias (for domain !0, the set of scopes in
4080 ; the !noalias list is not a superset of, or equal to, the scopes in the
4081 ; !alias.scope list):
4082 %2 = load float, float* %c, align 4, !alias.scope !6
4083 store float %0, float* %arrayidx.i, align 4, !noalias !7
4085 '``fpmath``' Metadata
4086 ^^^^^^^^^^^^^^^^^^^^^
4088 ``fpmath`` metadata may be attached to any instruction of floating point
4089 type. It can be used to express the maximum acceptable error in the
4090 result of that instruction, in ULPs, thus potentially allowing the
4091 compiler to use a more efficient but less accurate method of computing
4092 it. ULP is defined as follows:
4094 If ``x`` is a real number that lies between two finite consecutive
4095 floating-point numbers ``a`` and ``b``, without being equal to one
4096 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
4097 distance between the two non-equal finite floating-point numbers
4098 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
4100 The metadata node shall consist of a single positive floating point
4101 number representing the maximum relative error, for example:
4103 .. code-block:: llvm
4105 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
4109 '``range``' Metadata
4110 ^^^^^^^^^^^^^^^^^^^^
4112 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
4113 integer types. It expresses the possible ranges the loaded value or the value
4114 returned by the called function at this call site is in. The ranges are
4115 represented with a flattened list of integers. The loaded value or the value
4116 returned is known to be in the union of the ranges defined by each consecutive
4117 pair. Each pair has the following properties:
4119 - The type must match the type loaded by the instruction.
4120 - The pair ``a,b`` represents the range ``[a,b)``.
4121 - Both ``a`` and ``b`` are constants.
4122 - The range is allowed to wrap.
4123 - The range should not represent the full or empty set. That is,
4126 In addition, the pairs must be in signed order of the lower bound and
4127 they must be non-contiguous.
4131 .. code-block:: llvm
4133 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
4134 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
4135 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
4136 %d = invoke i8 @bar() to label %cont
4137 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
4139 !0 = !{ i8 0, i8 2 }
4140 !1 = !{ i8 255, i8 2 }
4141 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
4142 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
4147 It is sometimes useful to attach information to loop constructs. Currently,
4148 loop metadata is implemented as metadata attached to the branch instruction
4149 in the loop latch block. This type of metadata refer to a metadata node that is
4150 guaranteed to be separate for each loop. The loop identifier metadata is
4151 specified with the name ``llvm.loop``.
4153 The loop identifier metadata is implemented using a metadata that refers to
4154 itself to avoid merging it with any other identifier metadata, e.g.,
4155 during module linkage or function inlining. That is, each loop should refer
4156 to their own identification metadata even if they reside in separate functions.
4157 The following example contains loop identifier metadata for two separate loop
4160 .. code-block:: llvm
4165 The loop identifier metadata can be used to specify additional
4166 per-loop metadata. Any operands after the first operand can be treated
4167 as user-defined metadata. For example the ``llvm.loop.unroll.count``
4168 suggests an unroll factor to the loop unroller:
4170 .. code-block:: llvm
4172 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
4175 !1 = !{!"llvm.loop.unroll.count", i32 4}
4177 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
4178 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4180 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
4181 used to control per-loop vectorization and interleaving parameters such as
4182 vectorization width and interleave count. These metadata should be used in
4183 conjunction with ``llvm.loop`` loop identification metadata. The
4184 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
4185 optimization hints and the optimizer will only interleave and vectorize loops if
4186 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
4187 which contains information about loop-carried memory dependencies can be helpful
4188 in determining the safety of these transformations.
4190 '``llvm.loop.interleave.count``' Metadata
4191 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4193 This metadata suggests an interleave count to the loop interleaver.
4194 The first operand is the string ``llvm.loop.interleave.count`` and the
4195 second operand is an integer specifying the interleave count. For
4198 .. code-block:: llvm
4200 !0 = !{!"llvm.loop.interleave.count", i32 4}
4202 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
4203 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
4204 then the interleave count will be determined automatically.
4206 '``llvm.loop.vectorize.enable``' Metadata
4207 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4209 This metadata selectively enables or disables vectorization for the loop. The
4210 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
4211 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
4212 0 disables vectorization:
4214 .. code-block:: llvm
4216 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
4217 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
4219 '``llvm.loop.vectorize.width``' Metadata
4220 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4222 This metadata sets the target width of the vectorizer. The first
4223 operand is the string ``llvm.loop.vectorize.width`` and the second
4224 operand is an integer specifying the width. For example:
4226 .. code-block:: llvm
4228 !0 = !{!"llvm.loop.vectorize.width", i32 4}
4230 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
4231 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
4232 0 or if the loop does not have this metadata the width will be
4233 determined automatically.
4235 '``llvm.loop.unroll``'
4236 ^^^^^^^^^^^^^^^^^^^^^^
4238 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
4239 optimization hints such as the unroll factor. ``llvm.loop.unroll``
4240 metadata should be used in conjunction with ``llvm.loop`` loop
4241 identification metadata. The ``llvm.loop.unroll`` metadata are only
4242 optimization hints and the unrolling will only be performed if the
4243 optimizer believes it is safe to do so.
4245 '``llvm.loop.unroll.count``' Metadata
4246 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4248 This metadata suggests an unroll factor to the loop unroller. The
4249 first operand is the string ``llvm.loop.unroll.count`` and the second
4250 operand is a positive integer specifying the unroll factor. For
4253 .. code-block:: llvm
4255 !0 = !{!"llvm.loop.unroll.count", i32 4}
4257 If the trip count of the loop is less than the unroll count the loop
4258 will be partially unrolled.
4260 '``llvm.loop.unroll.disable``' Metadata
4261 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4263 This metadata disables loop unrolling. The metadata has a single operand
4264 which is the string ``llvm.loop.unroll.disable``. For example:
4266 .. code-block:: llvm
4268 !0 = !{!"llvm.loop.unroll.disable"}
4270 '``llvm.loop.unroll.runtime.disable``' Metadata
4271 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4273 This metadata disables runtime loop unrolling. The metadata has a single
4274 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
4276 .. code-block:: llvm
4278 !0 = !{!"llvm.loop.unroll.runtime.disable"}
4280 '``llvm.loop.unroll.full``' Metadata
4281 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4283 This metadata suggests that the loop should be unrolled fully. The
4284 metadata has a single operand which is the string ``llvm.loop.unroll.full``.
4287 .. code-block:: llvm
4289 !0 = !{!"llvm.loop.unroll.full"}
4294 Metadata types used to annotate memory accesses with information helpful
4295 for optimizations are prefixed with ``llvm.mem``.
4297 '``llvm.mem.parallel_loop_access``' Metadata
4298 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4300 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
4301 or metadata containing a list of loop identifiers for nested loops.
4302 The metadata is attached to memory accessing instructions and denotes that
4303 no loop carried memory dependence exist between it and other instructions denoted
4304 with the same loop identifier.
4306 Precisely, given two instructions ``m1`` and ``m2`` that both have the
4307 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
4308 set of loops associated with that metadata, respectively, then there is no loop
4309 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
4312 As a special case, if all memory accessing instructions in a loop have
4313 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
4314 loop has no loop carried memory dependences and is considered to be a parallel
4317 Note that if not all memory access instructions have such metadata referring to
4318 the loop, then the loop is considered not being trivially parallel. Additional
4319 memory dependence analysis is required to make that determination. As a fail
4320 safe mechanism, this causes loops that were originally parallel to be considered
4321 sequential (if optimization passes that are unaware of the parallel semantics
4322 insert new memory instructions into the loop body).
4324 Example of a loop that is considered parallel due to its correct use of
4325 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
4326 metadata types that refer to the same loop identifier metadata.
4328 .. code-block:: llvm
4332 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
4334 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4336 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
4342 It is also possible to have nested parallel loops. In that case the
4343 memory accesses refer to a list of loop identifier metadata nodes instead of
4344 the loop identifier metadata node directly:
4346 .. code-block:: llvm
4350 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
4352 br label %inner.for.body
4356 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4358 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
4360 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
4364 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
4366 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
4368 outer.for.end: ; preds = %for.body
4370 !0 = !{!1, !2} ; a list of loop identifiers
4371 !1 = !{!1} ; an identifier for the inner loop
4372 !2 = !{!2} ; an identifier for the outer loop
4377 The ``llvm.bitsets`` global metadata is used to implement
4378 :doc:`bitsets <BitSets>`.
4380 Module Flags Metadata
4381 =====================
4383 Information about the module as a whole is difficult to convey to LLVM's
4384 subsystems. The LLVM IR isn't sufficient to transmit this information.
4385 The ``llvm.module.flags`` named metadata exists in order to facilitate
4386 this. These flags are in the form of key / value pairs --- much like a
4387 dictionary --- making it easy for any subsystem who cares about a flag to
4390 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
4391 Each triplet has the following form:
4393 - The first element is a *behavior* flag, which specifies the behavior
4394 when two (or more) modules are merged together, and it encounters two
4395 (or more) metadata with the same ID. The supported behaviors are
4397 - The second element is a metadata string that is a unique ID for the
4398 metadata. Each module may only have one flag entry for each unique ID (not
4399 including entries with the **Require** behavior).
4400 - The third element is the value of the flag.
4402 When two (or more) modules are merged together, the resulting
4403 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
4404 each unique metadata ID string, there will be exactly one entry in the merged
4405 modules ``llvm.module.flags`` metadata table, and the value for that entry will
4406 be determined by the merge behavior flag, as described below. The only exception
4407 is that entries with the *Require* behavior are always preserved.
4409 The following behaviors are supported:
4420 Emits an error if two values disagree, otherwise the resulting value
4421 is that of the operands.
4425 Emits a warning if two values disagree. The result value will be the
4426 operand for the flag from the first module being linked.
4430 Adds a requirement that another module flag be present and have a
4431 specified value after linking is performed. The value must be a
4432 metadata pair, where the first element of the pair is the ID of the
4433 module flag to be restricted, and the second element of the pair is
4434 the value the module flag should be restricted to. This behavior can
4435 be used to restrict the allowable results (via triggering of an
4436 error) of linking IDs with the **Override** behavior.
4440 Uses the specified value, regardless of the behavior or value of the
4441 other module. If both modules specify **Override**, but the values
4442 differ, an error will be emitted.
4446 Appends the two values, which are required to be metadata nodes.
4450 Appends the two values, which are required to be metadata
4451 nodes. However, duplicate entries in the second list are dropped
4452 during the append operation.
4454 It is an error for a particular unique flag ID to have multiple behaviors,
4455 except in the case of **Require** (which adds restrictions on another metadata
4456 value) or **Override**.
4458 An example of module flags:
4460 .. code-block:: llvm
4462 !0 = !{ i32 1, !"foo", i32 1 }
4463 !1 = !{ i32 4, !"bar", i32 37 }
4464 !2 = !{ i32 2, !"qux", i32 42 }
4465 !3 = !{ i32 3, !"qux",
4470 !llvm.module.flags = !{ !0, !1, !2, !3 }
4472 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
4473 if two or more ``!"foo"`` flags are seen is to emit an error if their
4474 values are not equal.
4476 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
4477 behavior if two or more ``!"bar"`` flags are seen is to use the value
4480 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
4481 behavior if two or more ``!"qux"`` flags are seen is to emit a
4482 warning if their values are not equal.
4484 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
4490 The behavior is to emit an error if the ``llvm.module.flags`` does not
4491 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
4494 Objective-C Garbage Collection Module Flags Metadata
4495 ----------------------------------------------------
4497 On the Mach-O platform, Objective-C stores metadata about garbage
4498 collection in a special section called "image info". The metadata
4499 consists of a version number and a bitmask specifying what types of
4500 garbage collection are supported (if any) by the file. If two or more
4501 modules are linked together their garbage collection metadata needs to
4502 be merged rather than appended together.
4504 The Objective-C garbage collection module flags metadata consists of the
4505 following key-value pairs:
4514 * - ``Objective-C Version``
4515 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
4517 * - ``Objective-C Image Info Version``
4518 - **[Required]** --- The version of the image info section. Currently
4521 * - ``Objective-C Image Info Section``
4522 - **[Required]** --- The section to place the metadata. Valid values are
4523 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
4524 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
4525 Objective-C ABI version 2.
4527 * - ``Objective-C Garbage Collection``
4528 - **[Required]** --- Specifies whether garbage collection is supported or
4529 not. Valid values are 0, for no garbage collection, and 2, for garbage
4530 collection supported.
4532 * - ``Objective-C GC Only``
4533 - **[Optional]** --- Specifies that only garbage collection is supported.
4534 If present, its value must be 6. This flag requires that the
4535 ``Objective-C Garbage Collection`` flag have the value 2.
4537 Some important flag interactions:
4539 - If a module with ``Objective-C Garbage Collection`` set to 0 is
4540 merged with a module with ``Objective-C Garbage Collection`` set to
4541 2, then the resulting module has the
4542 ``Objective-C Garbage Collection`` flag set to 0.
4543 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
4544 merged with a module with ``Objective-C GC Only`` set to 6.
4546 Automatic Linker Flags Module Flags Metadata
4547 --------------------------------------------
4549 Some targets support embedding flags to the linker inside individual object
4550 files. Typically this is used in conjunction with language extensions which
4551 allow source files to explicitly declare the libraries they depend on, and have
4552 these automatically be transmitted to the linker via object files.
4554 These flags are encoded in the IR using metadata in the module flags section,
4555 using the ``Linker Options`` key. The merge behavior for this flag is required
4556 to be ``AppendUnique``, and the value for the key is expected to be a metadata
4557 node which should be a list of other metadata nodes, each of which should be a
4558 list of metadata strings defining linker options.
4560 For example, the following metadata section specifies two separate sets of
4561 linker options, presumably to link against ``libz`` and the ``Cocoa``
4564 !0 = !{ i32 6, !"Linker Options",
4567 !{ !"-framework", !"Cocoa" } } }
4568 !llvm.module.flags = !{ !0 }
4570 The metadata encoding as lists of lists of options, as opposed to a collapsed
4571 list of options, is chosen so that the IR encoding can use multiple option
4572 strings to specify e.g., a single library, while still having that specifier be
4573 preserved as an atomic element that can be recognized by a target specific
4574 assembly writer or object file emitter.
4576 Each individual option is required to be either a valid option for the target's
4577 linker, or an option that is reserved by the target specific assembly writer or
4578 object file emitter. No other aspect of these options is defined by the IR.
4580 C type width Module Flags Metadata
4581 ----------------------------------
4583 The ARM backend emits a section into each generated object file describing the
4584 options that it was compiled with (in a compiler-independent way) to prevent
4585 linking incompatible objects, and to allow automatic library selection. Some
4586 of these options are not visible at the IR level, namely wchar_t width and enum
4589 To pass this information to the backend, these options are encoded in module
4590 flags metadata, using the following key-value pairs:
4600 - * 0 --- sizeof(wchar_t) == 4
4601 * 1 --- sizeof(wchar_t) == 2
4604 - * 0 --- Enums are at least as large as an ``int``.
4605 * 1 --- Enums are stored in the smallest integer type which can
4606 represent all of its values.
4608 For example, the following metadata section specifies that the module was
4609 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
4610 enum is the smallest type which can represent all of its values::
4612 !llvm.module.flags = !{!0, !1}
4613 !0 = !{i32 1, !"short_wchar", i32 1}
4614 !1 = !{i32 1, !"short_enum", i32 0}
4616 .. _intrinsicglobalvariables:
4618 Intrinsic Global Variables
4619 ==========================
4621 LLVM has a number of "magic" global variables that contain data that
4622 affect code generation or other IR semantics. These are documented here.
4623 All globals of this sort should have a section specified as
4624 "``llvm.metadata``". This section and all globals that start with
4625 "``llvm.``" are reserved for use by LLVM.
4629 The '``llvm.used``' Global Variable
4630 -----------------------------------
4632 The ``@llvm.used`` global is an array which has
4633 :ref:`appending linkage <linkage_appending>`. This array contains a list of
4634 pointers to named global variables, functions and aliases which may optionally
4635 have a pointer cast formed of bitcast or getelementptr. For example, a legal
4638 .. code-block:: llvm
4643 @llvm.used = appending global [2 x i8*] [
4645 i8* bitcast (i32* @Y to i8*)
4646 ], section "llvm.metadata"
4648 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
4649 and linker are required to treat the symbol as if there is a reference to the
4650 symbol that it cannot see (which is why they have to be named). For example, if
4651 a variable has internal linkage and no references other than that from the
4652 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
4653 references from inline asms and other things the compiler cannot "see", and
4654 corresponds to "``attribute((used))``" in GNU C.
4656 On some targets, the code generator must emit a directive to the
4657 assembler or object file to prevent the assembler and linker from
4658 molesting the symbol.
4660 .. _gv_llvmcompilerused:
4662 The '``llvm.compiler.used``' Global Variable
4663 --------------------------------------------
4665 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4666 directive, except that it only prevents the compiler from touching the
4667 symbol. On targets that support it, this allows an intelligent linker to
4668 optimize references to the symbol without being impeded as it would be
4671 This is a rare construct that should only be used in rare circumstances,
4672 and should not be exposed to source languages.
4674 .. _gv_llvmglobalctors:
4676 The '``llvm.global_ctors``' Global Variable
4677 -------------------------------------------
4679 .. code-block:: llvm
4681 %0 = type { i32, void ()*, i8* }
4682 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
4684 The ``@llvm.global_ctors`` array contains a list of constructor
4685 functions, priorities, and an optional associated global or function.
4686 The functions referenced by this array will be called in ascending order
4687 of priority (i.e. lowest first) when the module is loaded. The order of
4688 functions with the same priority is not defined.
4690 If the third field is present, non-null, and points to a global variable
4691 or function, the initializer function will only run if the associated
4692 data from the current module is not discarded.
4694 .. _llvmglobaldtors:
4696 The '``llvm.global_dtors``' Global Variable
4697 -------------------------------------------
4699 .. code-block:: llvm
4701 %0 = type { i32, void ()*, i8* }
4702 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
4704 The ``@llvm.global_dtors`` array contains a list of destructor
4705 functions, priorities, and an optional associated global or function.
4706 The functions referenced by this array will be called in descending
4707 order of priority (i.e. highest first) when the module is unloaded. The
4708 order of functions with the same priority is not defined.
4710 If the third field is present, non-null, and points to a global variable
4711 or function, the destructor function will only run if the associated
4712 data from the current module is not discarded.
4714 Instruction Reference
4715 =====================
4717 The LLVM instruction set consists of several different classifications
4718 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
4719 instructions <binaryops>`, :ref:`bitwise binary
4720 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
4721 :ref:`other instructions <otherops>`.
4725 Terminator Instructions
4726 -----------------------
4728 As mentioned :ref:`previously <functionstructure>`, every basic block in a
4729 program ends with a "Terminator" instruction, which indicates which
4730 block should be executed after the current block is finished. These
4731 terminator instructions typically yield a '``void``' value: they produce
4732 control flow, not values (the one exception being the
4733 ':ref:`invoke <i_invoke>`' instruction).
4735 The terminator instructions are: ':ref:`ret <i_ret>`',
4736 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
4737 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
4738 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
4742 '``ret``' Instruction
4743 ^^^^^^^^^^^^^^^^^^^^^
4750 ret <type> <value> ; Return a value from a non-void function
4751 ret void ; Return from void function
4756 The '``ret``' instruction is used to return control flow (and optionally
4757 a value) from a function back to the caller.
4759 There are two forms of the '``ret``' instruction: one that returns a
4760 value and then causes control flow, and one that just causes control
4766 The '``ret``' instruction optionally accepts a single argument, the
4767 return value. The type of the return value must be a ':ref:`first
4768 class <t_firstclass>`' type.
4770 A function is not :ref:`well formed <wellformed>` if it it has a non-void
4771 return type and contains a '``ret``' instruction with no return value or
4772 a return value with a type that does not match its type, or if it has a
4773 void return type and contains a '``ret``' instruction with a return
4779 When the '``ret``' instruction is executed, control flow returns back to
4780 the calling function's context. If the caller is a
4781 ":ref:`call <i_call>`" instruction, execution continues at the
4782 instruction after the call. If the caller was an
4783 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
4784 beginning of the "normal" destination block. If the instruction returns
4785 a value, that value shall set the call or invoke instruction's return
4791 .. code-block:: llvm
4793 ret i32 5 ; Return an integer value of 5
4794 ret void ; Return from a void function
4795 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
4799 '``br``' Instruction
4800 ^^^^^^^^^^^^^^^^^^^^
4807 br i1 <cond>, label <iftrue>, label <iffalse>
4808 br label <dest> ; Unconditional branch
4813 The '``br``' instruction is used to cause control flow to transfer to a
4814 different basic block in the current function. There are two forms of
4815 this instruction, corresponding to a conditional branch and an
4816 unconditional branch.
4821 The conditional branch form of the '``br``' instruction takes a single
4822 '``i1``' value and two '``label``' values. The unconditional form of the
4823 '``br``' instruction takes a single '``label``' value as a target.
4828 Upon execution of a conditional '``br``' instruction, the '``i1``'
4829 argument is evaluated. If the value is ``true``, control flows to the
4830 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
4831 to the '``iffalse``' ``label`` argument.
4836 .. code-block:: llvm
4839 %cond = icmp eq i32 %a, %b
4840 br i1 %cond, label %IfEqual, label %IfUnequal
4848 '``switch``' Instruction
4849 ^^^^^^^^^^^^^^^^^^^^^^^^
4856 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
4861 The '``switch``' instruction is used to transfer control flow to one of
4862 several different places. It is a generalization of the '``br``'
4863 instruction, allowing a branch to occur to one of many possible
4869 The '``switch``' instruction uses three parameters: an integer
4870 comparison value '``value``', a default '``label``' destination, and an
4871 array of pairs of comparison value constants and '``label``'s. The table
4872 is not allowed to contain duplicate constant entries.
4877 The ``switch`` instruction specifies a table of values and destinations.
4878 When the '``switch``' instruction is executed, this table is searched
4879 for the given value. If the value is found, control flow is transferred
4880 to the corresponding destination; otherwise, control flow is transferred
4881 to the default destination.
4886 Depending on properties of the target machine and the particular
4887 ``switch`` instruction, this instruction may be code generated in
4888 different ways. For example, it could be generated as a series of
4889 chained conditional branches or with a lookup table.
4894 .. code-block:: llvm
4896 ; Emulate a conditional br instruction
4897 %Val = zext i1 %value to i32
4898 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
4900 ; Emulate an unconditional br instruction
4901 switch i32 0, label %dest [ ]
4903 ; Implement a jump table:
4904 switch i32 %val, label %otherwise [ i32 0, label %onzero
4906 i32 2, label %ontwo ]
4910 '``indirectbr``' Instruction
4911 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4918 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
4923 The '``indirectbr``' instruction implements an indirect branch to a
4924 label within the current function, whose address is specified by
4925 "``address``". Address must be derived from a
4926 :ref:`blockaddress <blockaddress>` constant.
4931 The '``address``' argument is the address of the label to jump to. The
4932 rest of the arguments indicate the full set of possible destinations
4933 that the address may point to. Blocks are allowed to occur multiple
4934 times in the destination list, though this isn't particularly useful.
4936 This destination list is required so that dataflow analysis has an
4937 accurate understanding of the CFG.
4942 Control transfers to the block specified in the address argument. All
4943 possible destination blocks must be listed in the label list, otherwise
4944 this instruction has undefined behavior. This implies that jumps to
4945 labels defined in other functions have undefined behavior as well.
4950 This is typically implemented with a jump through a register.
4955 .. code-block:: llvm
4957 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
4961 '``invoke``' Instruction
4962 ^^^^^^^^^^^^^^^^^^^^^^^^
4969 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
4970 to label <normal label> unwind label <exception label>
4975 The '``invoke``' instruction causes control to transfer to a specified
4976 function, with the possibility of control flow transfer to either the
4977 '``normal``' label or the '``exception``' label. If the callee function
4978 returns with the "``ret``" instruction, control flow will return to the
4979 "normal" label. If the callee (or any indirect callees) returns via the
4980 ":ref:`resume <i_resume>`" instruction or other exception handling
4981 mechanism, control is interrupted and continued at the dynamically
4982 nearest "exception" label.
4984 The '``exception``' label is a `landing
4985 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
4986 '``exception``' label is required to have the
4987 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
4988 information about the behavior of the program after unwinding happens,
4989 as its first non-PHI instruction. The restrictions on the
4990 "``landingpad``" instruction's tightly couples it to the "``invoke``"
4991 instruction, so that the important information contained within the
4992 "``landingpad``" instruction can't be lost through normal code motion.
4997 This instruction requires several arguments:
4999 #. The optional "cconv" marker indicates which :ref:`calling
5000 convention <callingconv>` the call should use. If none is
5001 specified, the call defaults to using C calling conventions.
5002 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5003 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5005 #. '``ptr to function ty``': shall be the signature of the pointer to
5006 function value being invoked. In most cases, this is a direct
5007 function invocation, but indirect ``invoke``'s are just as possible,
5008 branching off an arbitrary pointer to function value.
5009 #. '``function ptr val``': An LLVM value containing a pointer to a
5010 function to be invoked.
5011 #. '``function args``': argument list whose types match the function
5012 signature argument types and parameter attributes. All arguments must
5013 be of :ref:`first class <t_firstclass>` type. If the function signature
5014 indicates the function accepts a variable number of arguments, the
5015 extra arguments can be specified.
5016 #. '``normal label``': the label reached when the called function
5017 executes a '``ret``' instruction.
5018 #. '``exception label``': the label reached when a callee returns via
5019 the :ref:`resume <i_resume>` instruction or other exception handling
5021 #. The optional :ref:`function attributes <fnattrs>` list. Only
5022 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5023 attributes are valid here.
5028 This instruction is designed to operate as a standard '``call``'
5029 instruction in most regards. The primary difference is that it
5030 establishes an association with a label, which is used by the runtime
5031 library to unwind the stack.
5033 This instruction is used in languages with destructors to ensure that
5034 proper cleanup is performed in the case of either a ``longjmp`` or a
5035 thrown exception. Additionally, this is important for implementation of
5036 '``catch``' clauses in high-level languages that support them.
5038 For the purposes of the SSA form, the definition of the value returned
5039 by the '``invoke``' instruction is deemed to occur on the edge from the
5040 current block to the "normal" label. If the callee unwinds then no
5041 return value is available.
5046 .. code-block:: llvm
5048 %retval = invoke i32 @Test(i32 15) to label %Continue
5049 unwind label %TestCleanup ; i32:retval set
5050 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
5051 unwind label %TestCleanup ; i32:retval set
5055 '``resume``' Instruction
5056 ^^^^^^^^^^^^^^^^^^^^^^^^
5063 resume <type> <value>
5068 The '``resume``' instruction is a terminator instruction that has no
5074 The '``resume``' instruction requires one argument, which must have the
5075 same type as the result of any '``landingpad``' instruction in the same
5081 The '``resume``' instruction resumes propagation of an existing
5082 (in-flight) exception whose unwinding was interrupted with a
5083 :ref:`landingpad <i_landingpad>` instruction.
5088 .. code-block:: llvm
5090 resume { i8*, i32 } %exn
5094 '``unreachable``' Instruction
5095 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5107 The '``unreachable``' instruction has no defined semantics. This
5108 instruction is used to inform the optimizer that a particular portion of
5109 the code is not reachable. This can be used to indicate that the code
5110 after a no-return function cannot be reached, and other facts.
5115 The '``unreachable``' instruction has no defined semantics.
5122 Binary operators are used to do most of the computation in a program.
5123 They require two operands of the same type, execute an operation on
5124 them, and produce a single value. The operands might represent multiple
5125 data, as is the case with the :ref:`vector <t_vector>` data type. The
5126 result value has the same type as its operands.
5128 There are several different binary operators:
5132 '``add``' Instruction
5133 ^^^^^^^^^^^^^^^^^^^^^
5140 <result> = add <ty> <op1>, <op2> ; yields ty:result
5141 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
5142 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
5143 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
5148 The '``add``' instruction returns the sum of its two operands.
5153 The two arguments to the '``add``' instruction must be
5154 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5155 arguments must have identical types.
5160 The value produced is the integer sum of the two operands.
5162 If the sum has unsigned overflow, the result returned is the
5163 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5166 Because LLVM integers use a two's complement representation, this
5167 instruction is appropriate for both signed and unsigned integers.
5169 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5170 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5171 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
5172 unsigned and/or signed overflow, respectively, occurs.
5177 .. code-block:: llvm
5179 <result> = add i32 4, %var ; yields i32:result = 4 + %var
5183 '``fadd``' Instruction
5184 ^^^^^^^^^^^^^^^^^^^^^^
5191 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5196 The '``fadd``' instruction returns the sum of its two operands.
5201 The two arguments to the '``fadd``' instruction must be :ref:`floating
5202 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5203 Both arguments must have identical types.
5208 The value produced is the floating point sum of the two operands. This
5209 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
5210 which are optimization hints to enable otherwise unsafe floating point
5216 .. code-block:: llvm
5218 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
5220 '``sub``' Instruction
5221 ^^^^^^^^^^^^^^^^^^^^^
5228 <result> = sub <ty> <op1>, <op2> ; yields ty:result
5229 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
5230 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
5231 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
5236 The '``sub``' instruction returns the difference of its two operands.
5238 Note that the '``sub``' instruction is used to represent the '``neg``'
5239 instruction present in most other intermediate representations.
5244 The two arguments to the '``sub``' instruction must be
5245 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5246 arguments must have identical types.
5251 The value produced is the integer difference of the two operands.
5253 If the difference has unsigned overflow, the result returned is the
5254 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5257 Because LLVM integers use a two's complement representation, this
5258 instruction is appropriate for both signed and unsigned integers.
5260 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5261 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5262 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
5263 unsigned and/or signed overflow, respectively, occurs.
5268 .. code-block:: llvm
5270 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
5271 <result> = sub i32 0, %val ; yields i32:result = -%var
5275 '``fsub``' Instruction
5276 ^^^^^^^^^^^^^^^^^^^^^^
5283 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5288 The '``fsub``' instruction returns the difference of its two operands.
5290 Note that the '``fsub``' instruction is used to represent the '``fneg``'
5291 instruction present in most other intermediate representations.
5296 The two arguments to the '``fsub``' instruction must be :ref:`floating
5297 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5298 Both arguments must have identical types.
5303 The value produced is the floating point difference of the two operands.
5304 This instruction can also take any number of :ref:`fast-math
5305 flags <fastmath>`, which are optimization hints to enable otherwise
5306 unsafe floating point optimizations:
5311 .. code-block:: llvm
5313 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
5314 <result> = fsub float -0.0, %val ; yields float:result = -%var
5316 '``mul``' Instruction
5317 ^^^^^^^^^^^^^^^^^^^^^
5324 <result> = mul <ty> <op1>, <op2> ; yields ty:result
5325 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
5326 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
5327 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
5332 The '``mul``' instruction returns the product of its two operands.
5337 The two arguments to the '``mul``' instruction must be
5338 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5339 arguments must have identical types.
5344 The value produced is the integer product of the two operands.
5346 If the result of the multiplication has unsigned overflow, the result
5347 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
5348 bit width of the result.
5350 Because LLVM integers use a two's complement representation, and the
5351 result is the same width as the operands, this instruction returns the
5352 correct result for both signed and unsigned integers. If a full product
5353 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
5354 sign-extended or zero-extended as appropriate to the width of the full
5357 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5358 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5359 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
5360 unsigned and/or signed overflow, respectively, occurs.
5365 .. code-block:: llvm
5367 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
5371 '``fmul``' Instruction
5372 ^^^^^^^^^^^^^^^^^^^^^^
5379 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5384 The '``fmul``' instruction returns the product of its two operands.
5389 The two arguments to the '``fmul``' instruction must be :ref:`floating
5390 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5391 Both arguments must have identical types.
5396 The value produced is the floating point product of the two operands.
5397 This instruction can also take any number of :ref:`fast-math
5398 flags <fastmath>`, which are optimization hints to enable otherwise
5399 unsafe floating point optimizations:
5404 .. code-block:: llvm
5406 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
5408 '``udiv``' Instruction
5409 ^^^^^^^^^^^^^^^^^^^^^^
5416 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
5417 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
5422 The '``udiv``' instruction returns the quotient of its two operands.
5427 The two arguments to the '``udiv``' instruction must be
5428 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5429 arguments must have identical types.
5434 The value produced is the unsigned integer quotient of the two operands.
5436 Note that unsigned integer division and signed integer division are
5437 distinct operations; for signed integer division, use '``sdiv``'.
5439 Division by zero leads to undefined behavior.
5441 If the ``exact`` keyword is present, the result value of the ``udiv`` is
5442 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
5443 such, "((a udiv exact b) mul b) == a").
5448 .. code-block:: llvm
5450 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
5452 '``sdiv``' Instruction
5453 ^^^^^^^^^^^^^^^^^^^^^^
5460 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
5461 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
5466 The '``sdiv``' instruction returns the quotient of its two operands.
5471 The two arguments to the '``sdiv``' instruction must be
5472 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5473 arguments must have identical types.
5478 The value produced is the signed integer quotient of the two operands
5479 rounded towards zero.
5481 Note that signed integer division and unsigned integer division are
5482 distinct operations; for unsigned integer division, use '``udiv``'.
5484 Division by zero leads to undefined behavior. Overflow also leads to
5485 undefined behavior; this is a rare case, but can occur, for example, by
5486 doing a 32-bit division of -2147483648 by -1.
5488 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
5489 a :ref:`poison value <poisonvalues>` if the result would be rounded.
5494 .. code-block:: llvm
5496 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
5500 '``fdiv``' Instruction
5501 ^^^^^^^^^^^^^^^^^^^^^^
5508 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5513 The '``fdiv``' instruction returns the quotient of its two operands.
5518 The two arguments to the '``fdiv``' instruction must be :ref:`floating
5519 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5520 Both arguments must have identical types.
5525 The value produced is the floating point quotient of the two operands.
5526 This instruction can also take any number of :ref:`fast-math
5527 flags <fastmath>`, which are optimization hints to enable otherwise
5528 unsafe floating point optimizations:
5533 .. code-block:: llvm
5535 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
5537 '``urem``' Instruction
5538 ^^^^^^^^^^^^^^^^^^^^^^
5545 <result> = urem <ty> <op1>, <op2> ; yields ty:result
5550 The '``urem``' instruction returns the remainder from the unsigned
5551 division of its two arguments.
5556 The two arguments to the '``urem``' instruction must be
5557 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5558 arguments must have identical types.
5563 This instruction returns the unsigned integer *remainder* of a division.
5564 This instruction always performs an unsigned division to get the
5567 Note that unsigned integer remainder and signed integer remainder are
5568 distinct operations; for signed integer remainder, use '``srem``'.
5570 Taking the remainder of a division by zero leads to undefined behavior.
5575 .. code-block:: llvm
5577 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
5579 '``srem``' Instruction
5580 ^^^^^^^^^^^^^^^^^^^^^^
5587 <result> = srem <ty> <op1>, <op2> ; yields ty:result
5592 The '``srem``' instruction returns the remainder from the signed
5593 division of its two operands. This instruction can also take
5594 :ref:`vector <t_vector>` versions of the values in which case the elements
5600 The two arguments to the '``srem``' instruction must be
5601 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5602 arguments must have identical types.
5607 This instruction returns the *remainder* of a division (where the result
5608 is either zero or has the same sign as the dividend, ``op1``), not the
5609 *modulo* operator (where the result is either zero or has the same sign
5610 as the divisor, ``op2``) of a value. For more information about the
5611 difference, see `The Math
5612 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
5613 table of how this is implemented in various languages, please see
5615 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
5617 Note that signed integer remainder and unsigned integer remainder are
5618 distinct operations; for unsigned integer remainder, use '``urem``'.
5620 Taking the remainder of a division by zero leads to undefined behavior.
5621 Overflow also leads to undefined behavior; this is a rare case, but can
5622 occur, for example, by taking the remainder of a 32-bit division of
5623 -2147483648 by -1. (The remainder doesn't actually overflow, but this
5624 rule lets srem be implemented using instructions that return both the
5625 result of the division and the remainder.)
5630 .. code-block:: llvm
5632 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
5636 '``frem``' Instruction
5637 ^^^^^^^^^^^^^^^^^^^^^^
5644 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5649 The '``frem``' instruction returns the remainder from the division of
5655 The two arguments to the '``frem``' instruction must be :ref:`floating
5656 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5657 Both arguments must have identical types.
5662 This instruction returns the *remainder* of a division. The remainder
5663 has the same sign as the dividend. This instruction can also take any
5664 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
5665 to enable otherwise unsafe floating point optimizations:
5670 .. code-block:: llvm
5672 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
5676 Bitwise Binary Operations
5677 -------------------------
5679 Bitwise binary operators are used to do various forms of bit-twiddling
5680 in a program. They are generally very efficient instructions and can
5681 commonly be strength reduced from other instructions. They require two
5682 operands of the same type, execute an operation on them, and produce a
5683 single value. The resulting value is the same type as its operands.
5685 '``shl``' Instruction
5686 ^^^^^^^^^^^^^^^^^^^^^
5693 <result> = shl <ty> <op1>, <op2> ; yields ty:result
5694 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
5695 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
5696 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
5701 The '``shl``' instruction returns the first operand shifted to the left
5702 a specified number of bits.
5707 Both arguments to the '``shl``' instruction must be the same
5708 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5709 '``op2``' is treated as an unsigned value.
5714 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
5715 where ``n`` is the width of the result. If ``op2`` is (statically or
5716 dynamically) equal to or larger than the number of bits in
5717 ``op1``, the result is undefined. If the arguments are vectors, each
5718 vector element of ``op1`` is shifted by the corresponding shift amount
5721 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
5722 value <poisonvalues>` if it shifts out any non-zero bits. If the
5723 ``nsw`` keyword is present, then the shift produces a :ref:`poison
5724 value <poisonvalues>` if it shifts out any bits that disagree with the
5725 resultant sign bit. As such, NUW/NSW have the same semantics as they
5726 would if the shift were expressed as a mul instruction with the same
5727 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
5732 .. code-block:: llvm
5734 <result> = shl i32 4, %var ; yields i32: 4 << %var
5735 <result> = shl i32 4, 2 ; yields i32: 16
5736 <result> = shl i32 1, 10 ; yields i32: 1024
5737 <result> = shl i32 1, 32 ; undefined
5738 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
5740 '``lshr``' Instruction
5741 ^^^^^^^^^^^^^^^^^^^^^^
5748 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
5749 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
5754 The '``lshr``' instruction (logical shift right) returns the first
5755 operand shifted to the right a specified number of bits with zero fill.
5760 Both arguments to the '``lshr``' instruction must be the same
5761 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5762 '``op2``' is treated as an unsigned value.
5767 This instruction always performs a logical shift right operation. The
5768 most significant bits of the result will be filled with zero bits after
5769 the shift. If ``op2`` is (statically or dynamically) equal to or larger
5770 than the number of bits in ``op1``, the result is undefined. If the
5771 arguments are vectors, each vector element of ``op1`` is shifted by the
5772 corresponding shift amount in ``op2``.
5774 If the ``exact`` keyword is present, the result value of the ``lshr`` is
5775 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
5781 .. code-block:: llvm
5783 <result> = lshr i32 4, 1 ; yields i32:result = 2
5784 <result> = lshr i32 4, 2 ; yields i32:result = 1
5785 <result> = lshr i8 4, 3 ; yields i8:result = 0
5786 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
5787 <result> = lshr i32 1, 32 ; undefined
5788 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
5790 '``ashr``' Instruction
5791 ^^^^^^^^^^^^^^^^^^^^^^
5798 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
5799 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
5804 The '``ashr``' instruction (arithmetic shift right) returns the first
5805 operand shifted to the right a specified number of bits with sign
5811 Both arguments to the '``ashr``' instruction must be the same
5812 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5813 '``op2``' is treated as an unsigned value.
5818 This instruction always performs an arithmetic shift right operation,
5819 The most significant bits of the result will be filled with the sign bit
5820 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
5821 than the number of bits in ``op1``, the result is undefined. If the
5822 arguments are vectors, each vector element of ``op1`` is shifted by the
5823 corresponding shift amount in ``op2``.
5825 If the ``exact`` keyword is present, the result value of the ``ashr`` is
5826 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
5832 .. code-block:: llvm
5834 <result> = ashr i32 4, 1 ; yields i32:result = 2
5835 <result> = ashr i32 4, 2 ; yields i32:result = 1
5836 <result> = ashr i8 4, 3 ; yields i8:result = 0
5837 <result> = ashr i8 -2, 1 ; yields i8:result = -1
5838 <result> = ashr i32 1, 32 ; undefined
5839 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
5841 '``and``' Instruction
5842 ^^^^^^^^^^^^^^^^^^^^^
5849 <result> = and <ty> <op1>, <op2> ; yields ty:result
5854 The '``and``' instruction returns the bitwise logical and of its two
5860 The two arguments to the '``and``' instruction must be
5861 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5862 arguments must have identical types.
5867 The truth table used for the '``and``' instruction is:
5884 .. code-block:: llvm
5886 <result> = and i32 4, %var ; yields i32:result = 4 & %var
5887 <result> = and i32 15, 40 ; yields i32:result = 8
5888 <result> = and i32 4, 8 ; yields i32:result = 0
5890 '``or``' Instruction
5891 ^^^^^^^^^^^^^^^^^^^^
5898 <result> = or <ty> <op1>, <op2> ; yields ty:result
5903 The '``or``' instruction returns the bitwise logical inclusive or of its
5909 The two arguments to the '``or``' instruction must be
5910 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5911 arguments must have identical types.
5916 The truth table used for the '``or``' instruction is:
5935 <result> = or i32 4, %var ; yields i32:result = 4 | %var
5936 <result> = or i32 15, 40 ; yields i32:result = 47
5937 <result> = or i32 4, 8 ; yields i32:result = 12
5939 '``xor``' Instruction
5940 ^^^^^^^^^^^^^^^^^^^^^
5947 <result> = xor <ty> <op1>, <op2> ; yields ty:result
5952 The '``xor``' instruction returns the bitwise logical exclusive or of
5953 its two operands. The ``xor`` is used to implement the "one's
5954 complement" operation, which is the "~" operator in C.
5959 The two arguments to the '``xor``' instruction must be
5960 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5961 arguments must have identical types.
5966 The truth table used for the '``xor``' instruction is:
5983 .. code-block:: llvm
5985 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
5986 <result> = xor i32 15, 40 ; yields i32:result = 39
5987 <result> = xor i32 4, 8 ; yields i32:result = 12
5988 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
5993 LLVM supports several instructions to represent vector operations in a
5994 target-independent manner. These instructions cover the element-access
5995 and vector-specific operations needed to process vectors effectively.
5996 While LLVM does directly support these vector operations, many
5997 sophisticated algorithms will want to use target-specific intrinsics to
5998 take full advantage of a specific target.
6000 .. _i_extractelement:
6002 '``extractelement``' Instruction
6003 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6010 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
6015 The '``extractelement``' instruction extracts a single scalar element
6016 from a vector at a specified index.
6021 The first operand of an '``extractelement``' instruction is a value of
6022 :ref:`vector <t_vector>` type. The second operand is an index indicating
6023 the position from which to extract the element. The index may be a
6024 variable of any integer type.
6029 The result is a scalar of the same type as the element type of ``val``.
6030 Its value is the value at position ``idx`` of ``val``. If ``idx``
6031 exceeds the length of ``val``, the results are undefined.
6036 .. code-block:: llvm
6038 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
6040 .. _i_insertelement:
6042 '``insertelement``' Instruction
6043 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6050 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
6055 The '``insertelement``' instruction inserts a scalar element into a
6056 vector at a specified index.
6061 The first operand of an '``insertelement``' instruction is a value of
6062 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
6063 type must equal the element type of the first operand. The third operand
6064 is an index indicating the position at which to insert the value. The
6065 index may be a variable of any integer type.
6070 The result is a vector of the same type as ``val``. Its element values
6071 are those of ``val`` except at position ``idx``, where it gets the value
6072 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
6078 .. code-block:: llvm
6080 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
6082 .. _i_shufflevector:
6084 '``shufflevector``' Instruction
6085 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6092 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
6097 The '``shufflevector``' instruction constructs a permutation of elements
6098 from two input vectors, returning a vector with the same element type as
6099 the input and length that is the same as the shuffle mask.
6104 The first two operands of a '``shufflevector``' instruction are vectors
6105 with the same type. The third argument is a shuffle mask whose element
6106 type is always 'i32'. The result of the instruction is a vector whose
6107 length is the same as the shuffle mask and whose element type is the
6108 same as the element type of the first two operands.
6110 The shuffle mask operand is required to be a constant vector with either
6111 constant integer or undef values.
6116 The elements of the two input vectors are numbered from left to right
6117 across both of the vectors. The shuffle mask operand specifies, for each
6118 element of the result vector, which element of the two input vectors the
6119 result element gets. The element selector may be undef (meaning "don't
6120 care") and the second operand may be undef if performing a shuffle from
6126 .. code-block:: llvm
6128 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6129 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
6130 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
6131 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
6132 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
6133 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
6134 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6135 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
6137 Aggregate Operations
6138 --------------------
6140 LLVM supports several instructions for working with
6141 :ref:`aggregate <t_aggregate>` values.
6145 '``extractvalue``' Instruction
6146 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6153 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
6158 The '``extractvalue``' instruction extracts the value of a member field
6159 from an :ref:`aggregate <t_aggregate>` value.
6164 The first operand of an '``extractvalue``' instruction is a value of
6165 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
6166 constant indices to specify which value to extract in a similar manner
6167 as indices in a '``getelementptr``' instruction.
6169 The major differences to ``getelementptr`` indexing are:
6171 - Since the value being indexed is not a pointer, the first index is
6172 omitted and assumed to be zero.
6173 - At least one index must be specified.
6174 - Not only struct indices but also array indices must be in bounds.
6179 The result is the value at the position in the aggregate specified by
6185 .. code-block:: llvm
6187 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
6191 '``insertvalue``' Instruction
6192 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6199 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
6204 The '``insertvalue``' instruction inserts a value into a member field in
6205 an :ref:`aggregate <t_aggregate>` value.
6210 The first operand of an '``insertvalue``' instruction is a value of
6211 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
6212 a first-class value to insert. The following operands are constant
6213 indices indicating the position at which to insert the value in a
6214 similar manner as indices in a '``extractvalue``' instruction. The value
6215 to insert must have the same type as the value identified by the
6221 The result is an aggregate of the same type as ``val``. Its value is
6222 that of ``val`` except that the value at the position specified by the
6223 indices is that of ``elt``.
6228 .. code-block:: llvm
6230 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
6231 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
6232 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
6236 Memory Access and Addressing Operations
6237 ---------------------------------------
6239 A key design point of an SSA-based representation is how it represents
6240 memory. In LLVM, no memory locations are in SSA form, which makes things
6241 very simple. This section describes how to read, write, and allocate
6246 '``alloca``' Instruction
6247 ^^^^^^^^^^^^^^^^^^^^^^^^
6254 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
6259 The '``alloca``' instruction allocates memory on the stack frame of the
6260 currently executing function, to be automatically released when this
6261 function returns to its caller. The object is always allocated in the
6262 generic address space (address space zero).
6267 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
6268 bytes of memory on the runtime stack, returning a pointer of the
6269 appropriate type to the program. If "NumElements" is specified, it is
6270 the number of elements allocated, otherwise "NumElements" is defaulted
6271 to be one. If a constant alignment is specified, the value result of the
6272 allocation is guaranteed to be aligned to at least that boundary. The
6273 alignment may not be greater than ``1 << 29``. If not specified, or if
6274 zero, the target can choose to align the allocation on any convenient
6275 boundary compatible with the type.
6277 '``type``' may be any sized type.
6282 Memory is allocated; a pointer is returned. The operation is undefined
6283 if there is insufficient stack space for the allocation. '``alloca``'d
6284 memory is automatically released when the function returns. The
6285 '``alloca``' instruction is commonly used to represent automatic
6286 variables that must have an address available. When the function returns
6287 (either with the ``ret`` or ``resume`` instructions), the memory is
6288 reclaimed. Allocating zero bytes is legal, but the result is undefined.
6289 The order in which memory is allocated (ie., which way the stack grows)
6295 .. code-block:: llvm
6297 %ptr = alloca i32 ; yields i32*:ptr
6298 %ptr = alloca i32, i32 4 ; yields i32*:ptr
6299 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
6300 %ptr = alloca i32, align 1024 ; yields i32*:ptr
6304 '``load``' Instruction
6305 ^^^^^^^^^^^^^^^^^^^^^^
6312 <result> = load [volatile] <ty>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>][, !nonnull !<index>][, !dereferenceable !<index>][, !dereferenceable_or_null !<index>]
6313 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
6314 !<index> = !{ i32 1 }
6319 The '``load``' instruction is used to read from memory.
6324 The argument to the ``load`` instruction specifies the memory address
6325 from which to load. The type specified must be a :ref:`first
6326 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
6327 then the optimizer is not allowed to modify the number or order of
6328 execution of this ``load`` with other :ref:`volatile
6329 operations <volatile>`.
6331 If the ``load`` is marked as ``atomic``, it takes an extra
6332 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
6333 ``release`` and ``acq_rel`` orderings are not valid on ``load``
6334 instructions. Atomic loads produce :ref:`defined <memmodel>` results
6335 when they may see multiple atomic stores. The type of the pointee must
6336 be an integer type whose bit width is a power of two greater than or
6337 equal to eight and less than or equal to a target-specific size limit.
6338 ``align`` must be explicitly specified on atomic loads, and the load has
6339 undefined behavior if the alignment is not set to a value which is at
6340 least the size in bytes of the pointee. ``!nontemporal`` does not have
6341 any defined semantics for atomic loads.
6343 The optional constant ``align`` argument specifies the alignment of the
6344 operation (that is, the alignment of the memory address). A value of 0
6345 or an omitted ``align`` argument means that the operation has the ABI
6346 alignment for the target. It is the responsibility of the code emitter
6347 to ensure that the alignment information is correct. Overestimating the
6348 alignment results in undefined behavior. Underestimating the alignment
6349 may produce less efficient code. An alignment of 1 is always safe. The
6350 maximum possible alignment is ``1 << 29``.
6352 The optional ``!nontemporal`` metadata must reference a single
6353 metadata name ``<index>`` corresponding to a metadata node with one
6354 ``i32`` entry of value 1. The existence of the ``!nontemporal``
6355 metadata on the instruction tells the optimizer and code generator
6356 that this load is not expected to be reused in the cache. The code
6357 generator may select special instructions to save cache bandwidth, such
6358 as the ``MOVNT`` instruction on x86.
6360 The optional ``!invariant.load`` metadata must reference a single
6361 metadata name ``<index>`` corresponding to a metadata node with no
6362 entries. The existence of the ``!invariant.load`` metadata on the
6363 instruction tells the optimizer and code generator that the address
6364 operand to this load points to memory which can be assumed unchanged.
6365 Being invariant does not imply that a location is dereferenceable,
6366 but it does imply that once the location is known dereferenceable
6367 its value is henceforth unchanging.
6369 The optional ``!nonnull`` metadata must reference a single
6370 metadata name ``<index>`` corresponding to a metadata node with no
6371 entries. The existence of the ``!nonnull`` metadata on the
6372 instruction tells the optimizer that the value loaded is known to
6373 never be null. This is analogous to the ''nonnull'' attribute
6374 on parameters and return values. This metadata can only be applied
6375 to loads of a pointer type.
6377 The optional ``!dereferenceable`` metadata must reference a single
6378 metadata name ``<index>`` corresponding to a metadata node with one ``i64``
6379 entry. The existence of the ``!dereferenceable`` metadata on the instruction
6380 tells the optimizer that the value loaded is known to be dereferenceable.
6381 The number of bytes known to be dereferenceable is specified by the integer
6382 value in the metadata node. This is analogous to the ''dereferenceable''
6383 attribute on parameters and return values. This metadata can only be applied
6384 to loads of a pointer type.
6386 The optional ``!dereferenceable_or_null`` metadata must reference a single
6387 metadata name ``<index>`` corresponding to a metadata node with one ``i64``
6388 entry. The existence of the ``!dereferenceable_or_null`` metadata on the
6389 instruction tells the optimizer that the value loaded is known to be either
6390 dereferenceable or null.
6391 The number of bytes known to be dereferenceable is specified by the integer
6392 value in the metadata node. This is analogous to the ''dereferenceable_or_null''
6393 attribute on parameters and return values. This metadata can only be applied
6394 to loads of a pointer type.
6399 The location of memory pointed to is loaded. If the value being loaded
6400 is of scalar type then the number of bytes read does not exceed the
6401 minimum number of bytes needed to hold all bits of the type. For
6402 example, loading an ``i24`` reads at most three bytes. When loading a
6403 value of a type like ``i20`` with a size that is not an integral number
6404 of bytes, the result is undefined if the value was not originally
6405 written using a store of the same type.
6410 .. code-block:: llvm
6412 %ptr = alloca i32 ; yields i32*:ptr
6413 store i32 3, i32* %ptr ; yields void
6414 %val = load i32, i32* %ptr ; yields i32:val = i32 3
6418 '``store``' Instruction
6419 ^^^^^^^^^^^^^^^^^^^^^^^
6426 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields void
6427 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields void
6432 The '``store``' instruction is used to write to memory.
6437 There are two arguments to the ``store`` instruction: a value to store
6438 and an address at which to store it. The type of the ``<pointer>``
6439 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
6440 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
6441 then the optimizer is not allowed to modify the number or order of
6442 execution of this ``store`` with other :ref:`volatile
6443 operations <volatile>`.
6445 If the ``store`` is marked as ``atomic``, it takes an extra
6446 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
6447 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
6448 instructions. Atomic loads produce :ref:`defined <memmodel>` results
6449 when they may see multiple atomic stores. The type of the pointee must
6450 be an integer type whose bit width is a power of two greater than or
6451 equal to eight and less than or equal to a target-specific size limit.
6452 ``align`` must be explicitly specified on atomic stores, and the store
6453 has undefined behavior if the alignment is not set to a value which is
6454 at least the size in bytes of the pointee. ``!nontemporal`` does not
6455 have any defined semantics for atomic stores.
6457 The optional constant ``align`` argument specifies the alignment of the
6458 operation (that is, the alignment of the memory address). A value of 0
6459 or an omitted ``align`` argument means that the operation has the ABI
6460 alignment for the target. It is the responsibility of the code emitter
6461 to ensure that the alignment information is correct. Overestimating the
6462 alignment results in undefined behavior. Underestimating the
6463 alignment may produce less efficient code. An alignment of 1 is always
6464 safe. The maximum possible alignment is ``1 << 29``.
6466 The optional ``!nontemporal`` metadata must reference a single metadata
6467 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
6468 value 1. The existence of the ``!nontemporal`` metadata on the instruction
6469 tells the optimizer and code generator that this load is not expected to
6470 be reused in the cache. The code generator may select special
6471 instructions to save cache bandwidth, such as the MOVNT instruction on
6477 The contents of memory are updated to contain ``<value>`` at the
6478 location specified by the ``<pointer>`` operand. If ``<value>`` is
6479 of scalar type then the number of bytes written does not exceed the
6480 minimum number of bytes needed to hold all bits of the type. For
6481 example, storing an ``i24`` writes at most three bytes. When writing a
6482 value of a type like ``i20`` with a size that is not an integral number
6483 of bytes, it is unspecified what happens to the extra bits that do not
6484 belong to the type, but they will typically be overwritten.
6489 .. code-block:: llvm
6491 %ptr = alloca i32 ; yields i32*:ptr
6492 store i32 3, i32* %ptr ; yields void
6493 %val = load i32* %ptr ; yields i32:val = i32 3
6497 '``fence``' Instruction
6498 ^^^^^^^^^^^^^^^^^^^^^^^
6505 fence [singlethread] <ordering> ; yields void
6510 The '``fence``' instruction is used to introduce happens-before edges
6516 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
6517 defines what *synchronizes-with* edges they add. They can only be given
6518 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
6523 A fence A which has (at least) ``release`` ordering semantics
6524 *synchronizes with* a fence B with (at least) ``acquire`` ordering
6525 semantics if and only if there exist atomic operations X and Y, both
6526 operating on some atomic object M, such that A is sequenced before X, X
6527 modifies M (either directly or through some side effect of a sequence
6528 headed by X), Y is sequenced before B, and Y observes M. This provides a
6529 *happens-before* dependency between A and B. Rather than an explicit
6530 ``fence``, one (but not both) of the atomic operations X or Y might
6531 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
6532 still *synchronize-with* the explicit ``fence`` and establish the
6533 *happens-before* edge.
6535 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
6536 ``acquire`` and ``release`` semantics specified above, participates in
6537 the global program order of other ``seq_cst`` operations and/or fences.
6539 The optional ":ref:`singlethread <singlethread>`" argument specifies
6540 that the fence only synchronizes with other fences in the same thread.
6541 (This is useful for interacting with signal handlers.)
6546 .. code-block:: llvm
6548 fence acquire ; yields void
6549 fence singlethread seq_cst ; yields void
6553 '``cmpxchg``' Instruction
6554 ^^^^^^^^^^^^^^^^^^^^^^^^^
6561 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
6566 The '``cmpxchg``' instruction is used to atomically modify memory. It
6567 loads a value in memory and compares it to a given value. If they are
6568 equal, it tries to store a new value into the memory.
6573 There are three arguments to the '``cmpxchg``' instruction: an address
6574 to operate on, a value to compare to the value currently be at that
6575 address, and a new value to place at that address if the compared values
6576 are equal. The type of '<cmp>' must be an integer type whose bit width
6577 is a power of two greater than or equal to eight and less than or equal
6578 to a target-specific size limit. '<cmp>' and '<new>' must have the same
6579 type, and the type of '<pointer>' must be a pointer to that type. If the
6580 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
6581 to modify the number or order of execution of this ``cmpxchg`` with
6582 other :ref:`volatile operations <volatile>`.
6584 The success and failure :ref:`ordering <ordering>` arguments specify how this
6585 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
6586 must be at least ``monotonic``, the ordering constraint on failure must be no
6587 stronger than that on success, and the failure ordering cannot be either
6588 ``release`` or ``acq_rel``.
6590 The optional "``singlethread``" argument declares that the ``cmpxchg``
6591 is only atomic with respect to code (usually signal handlers) running in
6592 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
6593 respect to all other code in the system.
6595 The pointer passed into cmpxchg must have alignment greater than or
6596 equal to the size in memory of the operand.
6601 The contents of memory at the location specified by the '``<pointer>``' operand
6602 is read and compared to '``<cmp>``'; if the read value is the equal, the
6603 '``<new>``' is written. The original value at the location is returned, together
6604 with a flag indicating success (true) or failure (false).
6606 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
6607 permitted: the operation may not write ``<new>`` even if the comparison
6610 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
6611 if the value loaded equals ``cmp``.
6613 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
6614 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
6615 load with an ordering parameter determined the second ordering parameter.
6620 .. code-block:: llvm
6623 %orig = atomic load i32, i32* %ptr unordered ; yields i32
6627 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
6628 %squared = mul i32 %cmp, %cmp
6629 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
6630 %value_loaded = extractvalue { i32, i1 } %val_success, 0
6631 %success = extractvalue { i32, i1 } %val_success, 1
6632 br i1 %success, label %done, label %loop
6639 '``atomicrmw``' Instruction
6640 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6647 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
6652 The '``atomicrmw``' instruction is used to atomically modify memory.
6657 There are three arguments to the '``atomicrmw``' instruction: an
6658 operation to apply, an address whose value to modify, an argument to the
6659 operation. The operation must be one of the following keywords:
6673 The type of '<value>' must be an integer type whose bit width is a power
6674 of two greater than or equal to eight and less than or equal to a
6675 target-specific size limit. The type of the '``<pointer>``' operand must
6676 be a pointer to that type. If the ``atomicrmw`` is marked as
6677 ``volatile``, then the optimizer is not allowed to modify the number or
6678 order of execution of this ``atomicrmw`` with other :ref:`volatile
6679 operations <volatile>`.
6684 The contents of memory at the location specified by the '``<pointer>``'
6685 operand are atomically read, modified, and written back. The original
6686 value at the location is returned. The modification is specified by the
6689 - xchg: ``*ptr = val``
6690 - add: ``*ptr = *ptr + val``
6691 - sub: ``*ptr = *ptr - val``
6692 - and: ``*ptr = *ptr & val``
6693 - nand: ``*ptr = ~(*ptr & val)``
6694 - or: ``*ptr = *ptr | val``
6695 - xor: ``*ptr = *ptr ^ val``
6696 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
6697 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
6698 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
6700 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
6706 .. code-block:: llvm
6708 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
6710 .. _i_getelementptr:
6712 '``getelementptr``' Instruction
6713 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6720 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
6721 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
6722 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
6727 The '``getelementptr``' instruction is used to get the address of a
6728 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
6729 address calculation only and does not access memory. The instruction can also
6730 be used to calculate a vector of such addresses.
6735 The first argument is always a type used as the basis for the calculations.
6736 The second argument is always a pointer or a vector of pointers, and is the
6737 base address to start from. The remaining arguments are indices
6738 that indicate which of the elements of the aggregate object are indexed.
6739 The interpretation of each index is dependent on the type being indexed
6740 into. The first index always indexes the pointer value given as the
6741 first argument, the second index indexes a value of the type pointed to
6742 (not necessarily the value directly pointed to, since the first index
6743 can be non-zero), etc. The first type indexed into must be a pointer
6744 value, subsequent types can be arrays, vectors, and structs. Note that
6745 subsequent types being indexed into can never be pointers, since that
6746 would require loading the pointer before continuing calculation.
6748 The type of each index argument depends on the type it is indexing into.
6749 When indexing into a (optionally packed) structure, only ``i32`` integer
6750 **constants** are allowed (when using a vector of indices they must all
6751 be the **same** ``i32`` integer constant). When indexing into an array,
6752 pointer or vector, integers of any width are allowed, and they are not
6753 required to be constant. These integers are treated as signed values
6756 For example, let's consider a C code fragment and how it gets compiled
6772 int *foo(struct ST *s) {
6773 return &s[1].Z.B[5][13];
6776 The LLVM code generated by Clang is:
6778 .. code-block:: llvm
6780 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
6781 %struct.ST = type { i32, double, %struct.RT }
6783 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
6785 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
6792 In the example above, the first index is indexing into the
6793 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
6794 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
6795 indexes into the third element of the structure, yielding a
6796 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
6797 structure. The third index indexes into the second element of the
6798 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
6799 dimensions of the array are subscripted into, yielding an '``i32``'
6800 type. The '``getelementptr``' instruction returns a pointer to this
6801 element, thus computing a value of '``i32*``' type.
6803 Note that it is perfectly legal to index partially through a structure,
6804 returning a pointer to an inner element. Because of this, the LLVM code
6805 for the given testcase is equivalent to:
6807 .. code-block:: llvm
6809 define i32* @foo(%struct.ST* %s) {
6810 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
6811 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
6812 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
6813 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
6814 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
6818 If the ``inbounds`` keyword is present, the result value of the
6819 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
6820 pointer is not an *in bounds* address of an allocated object, or if any
6821 of the addresses that would be formed by successive addition of the
6822 offsets implied by the indices to the base address with infinitely
6823 precise signed arithmetic are not an *in bounds* address of that
6824 allocated object. The *in bounds* addresses for an allocated object are
6825 all the addresses that point into the object, plus the address one byte
6826 past the end. In cases where the base is a vector of pointers the
6827 ``inbounds`` keyword applies to each of the computations element-wise.
6829 If the ``inbounds`` keyword is not present, the offsets are added to the
6830 base address with silently-wrapping two's complement arithmetic. If the
6831 offsets have a different width from the pointer, they are sign-extended
6832 or truncated to the width of the pointer. The result value of the
6833 ``getelementptr`` may be outside the object pointed to by the base
6834 pointer. The result value may not necessarily be used to access memory
6835 though, even if it happens to point into allocated storage. See the
6836 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
6839 The getelementptr instruction is often confusing. For some more insight
6840 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
6845 .. code-block:: llvm
6847 ; yields [12 x i8]*:aptr
6848 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
6850 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
6852 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
6854 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
6859 The ``getelementptr`` returns a vector of pointers, instead of a single address,
6860 when one or more of its arguments is a vector. In such cases, all vector
6861 arguments should have the same number of elements, and every scalar argument
6862 will be effectively broadcast into a vector during address calculation.
6864 .. code-block:: llvm
6866 ; All arguments are vectors:
6867 ; A[i] = ptrs[i] + offsets[i]*sizeof(i8)
6868 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets
6870 ; Add the same scalar offset to each pointer of a vector:
6871 ; A[i] = ptrs[i] + offset*sizeof(i8)
6872 %A = getelementptr i8, <4 x i8*> %ptrs, i64 %offset
6874 ; Add distinct offsets to the same pointer:
6875 ; A[i] = ptr + offsets[i]*sizeof(i8)
6876 %A = getelementptr i8, i8* %ptr, <4 x i64> %offsets
6878 ; In all cases described above the type of the result is <4 x i8*>
6880 The two following instructions are equivalent:
6882 .. code-block:: llvm
6884 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
6885 <4 x i32> <i32 2, i32 2, i32 2, i32 2>,
6886 <4 x i32> <i32 1, i32 1, i32 1, i32 1>,
6888 <4 x i64> <i64 13, i64 13, i64 13, i64 13>
6890 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
6891 i32 2, i32 1, <4 x i32> %ind4, i64 13
6893 Let's look at the C code, where the vector version of ``getelementptr``
6898 // Let's assume that we vectorize the following loop:
6899 double *A, B; int *C;
6900 for (int i = 0; i < size; ++i) {
6904 .. code-block:: llvm
6906 ; get pointers for 8 elements from array B
6907 %ptrs = getelementptr double, double* %B, <8 x i32> %C
6908 ; load 8 elements from array B into A
6909 %A = call <8 x double> @llvm.masked.gather.v8f64(<8 x double*> %ptrs,
6910 i32 8, <8 x i1> %mask, <8 x double> %passthru)
6912 Conversion Operations
6913 ---------------------
6915 The instructions in this category are the conversion instructions
6916 (casting) which all take a single operand and a type. They perform
6917 various bit conversions on the operand.
6919 '``trunc .. to``' Instruction
6920 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6927 <result> = trunc <ty> <value> to <ty2> ; yields ty2
6932 The '``trunc``' instruction truncates its operand to the type ``ty2``.
6937 The '``trunc``' instruction takes a value to trunc, and a type to trunc
6938 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
6939 of the same number of integers. The bit size of the ``value`` must be
6940 larger than the bit size of the destination type, ``ty2``. Equal sized
6941 types are not allowed.
6946 The '``trunc``' instruction truncates the high order bits in ``value``
6947 and converts the remaining bits to ``ty2``. Since the source size must
6948 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
6949 It will always truncate bits.
6954 .. code-block:: llvm
6956 %X = trunc i32 257 to i8 ; yields i8:1
6957 %Y = trunc i32 123 to i1 ; yields i1:true
6958 %Z = trunc i32 122 to i1 ; yields i1:false
6959 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
6961 '``zext .. to``' Instruction
6962 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6969 <result> = zext <ty> <value> to <ty2> ; yields ty2
6974 The '``zext``' instruction zero extends its operand to type ``ty2``.
6979 The '``zext``' instruction takes a value to cast, and a type to cast it
6980 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
6981 the same number of integers. The bit size of the ``value`` must be
6982 smaller than the bit size of the destination type, ``ty2``.
6987 The ``zext`` fills the high order bits of the ``value`` with zero bits
6988 until it reaches the size of the destination type, ``ty2``.
6990 When zero extending from i1, the result will always be either 0 or 1.
6995 .. code-block:: llvm
6997 %X = zext i32 257 to i64 ; yields i64:257
6998 %Y = zext i1 true to i32 ; yields i32:1
6999 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7001 '``sext .. to``' Instruction
7002 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7009 <result> = sext <ty> <value> to <ty2> ; yields ty2
7014 The '``sext``' sign extends ``value`` to the type ``ty2``.
7019 The '``sext``' instruction takes a value to cast, and a type to cast it
7020 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7021 the same number of integers. The bit size of the ``value`` must be
7022 smaller than the bit size of the destination type, ``ty2``.
7027 The '``sext``' instruction performs a sign extension by copying the sign
7028 bit (highest order bit) of the ``value`` until it reaches the bit size
7029 of the type ``ty2``.
7031 When sign extending from i1, the extension always results in -1 or 0.
7036 .. code-block:: llvm
7038 %X = sext i8 -1 to i16 ; yields i16 :65535
7039 %Y = sext i1 true to i32 ; yields i32:-1
7040 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7042 '``fptrunc .. to``' Instruction
7043 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7050 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
7055 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
7060 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
7061 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
7062 The size of ``value`` must be larger than the size of ``ty2``. This
7063 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
7068 The '``fptrunc``' instruction truncates a ``value`` from a larger
7069 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
7070 point <t_floating>` type. If the value cannot fit within the
7071 destination type, ``ty2``, then the results are undefined.
7076 .. code-block:: llvm
7078 %X = fptrunc double 123.0 to float ; yields float:123.0
7079 %Y = fptrunc double 1.0E+300 to float ; yields undefined
7081 '``fpext .. to``' Instruction
7082 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7089 <result> = fpext <ty> <value> to <ty2> ; yields ty2
7094 The '``fpext``' extends a floating point ``value`` to a larger floating
7100 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
7101 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
7102 to. The source type must be smaller than the destination type.
7107 The '``fpext``' instruction extends the ``value`` from a smaller
7108 :ref:`floating point <t_floating>` type to a larger :ref:`floating
7109 point <t_floating>` type. The ``fpext`` cannot be used to make a
7110 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
7111 *no-op cast* for a floating point cast.
7116 .. code-block:: llvm
7118 %X = fpext float 3.125 to double ; yields double:3.125000e+00
7119 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
7121 '``fptoui .. to``' Instruction
7122 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7129 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
7134 The '``fptoui``' converts a floating point ``value`` to its unsigned
7135 integer equivalent of type ``ty2``.
7140 The '``fptoui``' instruction takes a value to cast, which must be a
7141 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7142 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7143 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7144 type with the same number of elements as ``ty``
7149 The '``fptoui``' instruction converts its :ref:`floating
7150 point <t_floating>` operand into the nearest (rounding towards zero)
7151 unsigned integer value. If the value cannot fit in ``ty2``, the results
7157 .. code-block:: llvm
7159 %X = fptoui double 123.0 to i32 ; yields i32:123
7160 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
7161 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
7163 '``fptosi .. to``' Instruction
7164 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7171 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
7176 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
7177 ``value`` to type ``ty2``.
7182 The '``fptosi``' instruction takes a value to cast, which must be a
7183 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7184 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7185 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7186 type with the same number of elements as ``ty``
7191 The '``fptosi``' instruction converts its :ref:`floating
7192 point <t_floating>` operand into the nearest (rounding towards zero)
7193 signed integer value. If the value cannot fit in ``ty2``, the results
7199 .. code-block:: llvm
7201 %X = fptosi double -123.0 to i32 ; yields i32:-123
7202 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
7203 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
7205 '``uitofp .. to``' Instruction
7206 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7213 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
7218 The '``uitofp``' instruction regards ``value`` as an unsigned integer
7219 and converts that value to the ``ty2`` type.
7224 The '``uitofp``' instruction takes a value to cast, which must be a
7225 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7226 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7227 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7228 type with the same number of elements as ``ty``
7233 The '``uitofp``' instruction interprets its operand as an unsigned
7234 integer quantity and converts it to the corresponding floating point
7235 value. If the value cannot fit in the floating point value, the results
7241 .. code-block:: llvm
7243 %X = uitofp i32 257 to float ; yields float:257.0
7244 %Y = uitofp i8 -1 to double ; yields double:255.0
7246 '``sitofp .. to``' Instruction
7247 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7254 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
7259 The '``sitofp``' instruction regards ``value`` as a signed integer and
7260 converts that value to the ``ty2`` type.
7265 The '``sitofp``' instruction takes a value to cast, which must be a
7266 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7267 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7268 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7269 type with the same number of elements as ``ty``
7274 The '``sitofp``' instruction interprets its operand as a signed integer
7275 quantity and converts it to the corresponding floating point value. If
7276 the value cannot fit in the floating point value, the results are
7282 .. code-block:: llvm
7284 %X = sitofp i32 257 to float ; yields float:257.0
7285 %Y = sitofp i8 -1 to double ; yields double:-1.0
7289 '``ptrtoint .. to``' Instruction
7290 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7297 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
7302 The '``ptrtoint``' instruction converts the pointer or a vector of
7303 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
7308 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
7309 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
7310 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
7311 a vector of integers type.
7316 The '``ptrtoint``' instruction converts ``value`` to integer type
7317 ``ty2`` by interpreting the pointer value as an integer and either
7318 truncating or zero extending that value to the size of the integer type.
7319 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
7320 ``value`` is larger than ``ty2`` then a truncation is done. If they are
7321 the same size, then nothing is done (*no-op cast*) other than a type
7327 .. code-block:: llvm
7329 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
7330 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
7331 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
7335 '``inttoptr .. to``' Instruction
7336 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7343 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
7348 The '``inttoptr``' instruction converts an integer ``value`` to a
7349 pointer type, ``ty2``.
7354 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
7355 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
7361 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
7362 applying either a zero extension or a truncation depending on the size
7363 of the integer ``value``. If ``value`` is larger than the size of a
7364 pointer then a truncation is done. If ``value`` is smaller than the size
7365 of a pointer then a zero extension is done. If they are the same size,
7366 nothing is done (*no-op cast*).
7371 .. code-block:: llvm
7373 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
7374 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
7375 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
7376 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
7380 '``bitcast .. to``' Instruction
7381 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7388 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
7393 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
7399 The '``bitcast``' instruction takes a value to cast, which must be a
7400 non-aggregate first class value, and a type to cast it to, which must
7401 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
7402 bit sizes of ``value`` and the destination type, ``ty2``, must be
7403 identical. If the source type is a pointer, the destination type must
7404 also be a pointer of the same size. This instruction supports bitwise
7405 conversion of vectors to integers and to vectors of other types (as
7406 long as they have the same size).
7411 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
7412 is always a *no-op cast* because no bits change with this
7413 conversion. The conversion is done as if the ``value`` had been stored
7414 to memory and read back as type ``ty2``. Pointer (or vector of
7415 pointers) types may only be converted to other pointer (or vector of
7416 pointers) types with the same address space through this instruction.
7417 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
7418 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
7423 .. code-block:: llvm
7425 %X = bitcast i8 255 to i8 ; yields i8 :-1
7426 %Y = bitcast i32* %x to sint* ; yields sint*:%x
7427 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
7428 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
7430 .. _i_addrspacecast:
7432 '``addrspacecast .. to``' Instruction
7433 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7440 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
7445 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
7446 address space ``n`` to type ``pty2`` in address space ``m``.
7451 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
7452 to cast and a pointer type to cast it to, which must have a different
7458 The '``addrspacecast``' instruction converts the pointer value
7459 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
7460 value modification, depending on the target and the address space
7461 pair. Pointer conversions within the same address space must be
7462 performed with the ``bitcast`` instruction. Note that if the address space
7463 conversion is legal then both result and operand refer to the same memory
7469 .. code-block:: llvm
7471 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
7472 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
7473 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
7480 The instructions in this category are the "miscellaneous" instructions,
7481 which defy better classification.
7485 '``icmp``' Instruction
7486 ^^^^^^^^^^^^^^^^^^^^^^
7493 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
7498 The '``icmp``' instruction returns a boolean value or a vector of
7499 boolean values based on comparison of its two integer, integer vector,
7500 pointer, or pointer vector operands.
7505 The '``icmp``' instruction takes three operands. The first operand is
7506 the condition code indicating the kind of comparison to perform. It is
7507 not a value, just a keyword. The possible condition code are:
7510 #. ``ne``: not equal
7511 #. ``ugt``: unsigned greater than
7512 #. ``uge``: unsigned greater or equal
7513 #. ``ult``: unsigned less than
7514 #. ``ule``: unsigned less or equal
7515 #. ``sgt``: signed greater than
7516 #. ``sge``: signed greater or equal
7517 #. ``slt``: signed less than
7518 #. ``sle``: signed less or equal
7520 The remaining two arguments must be :ref:`integer <t_integer>` or
7521 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
7522 must also be identical types.
7527 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
7528 code given as ``cond``. The comparison performed always yields either an
7529 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
7531 #. ``eq``: yields ``true`` if the operands are equal, ``false``
7532 otherwise. No sign interpretation is necessary or performed.
7533 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
7534 otherwise. No sign interpretation is necessary or performed.
7535 #. ``ugt``: interprets the operands as unsigned values and yields
7536 ``true`` if ``op1`` is greater than ``op2``.
7537 #. ``uge``: interprets the operands as unsigned values and yields
7538 ``true`` if ``op1`` is greater than or equal to ``op2``.
7539 #. ``ult``: interprets the operands as unsigned values and yields
7540 ``true`` if ``op1`` is less than ``op2``.
7541 #. ``ule``: interprets the operands as unsigned values and yields
7542 ``true`` if ``op1`` is less than or equal to ``op2``.
7543 #. ``sgt``: interprets the operands as signed values and yields ``true``
7544 if ``op1`` is greater than ``op2``.
7545 #. ``sge``: interprets the operands as signed values and yields ``true``
7546 if ``op1`` is greater than or equal to ``op2``.
7547 #. ``slt``: interprets the operands as signed values and yields ``true``
7548 if ``op1`` is less than ``op2``.
7549 #. ``sle``: interprets the operands as signed values and yields ``true``
7550 if ``op1`` is less than or equal to ``op2``.
7552 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
7553 are compared as if they were integers.
7555 If the operands are integer vectors, then they are compared element by
7556 element. The result is an ``i1`` vector with the same number of elements
7557 as the values being compared. Otherwise, the result is an ``i1``.
7562 .. code-block:: llvm
7564 <result> = icmp eq i32 4, 5 ; yields: result=false
7565 <result> = icmp ne float* %X, %X ; yields: result=false
7566 <result> = icmp ult i16 4, 5 ; yields: result=true
7567 <result> = icmp sgt i16 4, 5 ; yields: result=false
7568 <result> = icmp ule i16 -4, 5 ; yields: result=false
7569 <result> = icmp sge i16 4, 5 ; yields: result=false
7571 Note that the code generator does not yet support vector types with the
7572 ``icmp`` instruction.
7576 '``fcmp``' Instruction
7577 ^^^^^^^^^^^^^^^^^^^^^^
7584 <result> = fcmp [fast-math flags]* <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
7589 The '``fcmp``' instruction returns a boolean value or vector of boolean
7590 values based on comparison of its operands.
7592 If the operands are floating point scalars, then the result type is a
7593 boolean (:ref:`i1 <t_integer>`).
7595 If the operands are floating point vectors, then the result type is a
7596 vector of boolean with the same number of elements as the operands being
7602 The '``fcmp``' instruction takes three operands. The first operand is
7603 the condition code indicating the kind of comparison to perform. It is
7604 not a value, just a keyword. The possible condition code are:
7606 #. ``false``: no comparison, always returns false
7607 #. ``oeq``: ordered and equal
7608 #. ``ogt``: ordered and greater than
7609 #. ``oge``: ordered and greater than or equal
7610 #. ``olt``: ordered and less than
7611 #. ``ole``: ordered and less than or equal
7612 #. ``one``: ordered and not equal
7613 #. ``ord``: ordered (no nans)
7614 #. ``ueq``: unordered or equal
7615 #. ``ugt``: unordered or greater than
7616 #. ``uge``: unordered or greater than or equal
7617 #. ``ult``: unordered or less than
7618 #. ``ule``: unordered or less than or equal
7619 #. ``une``: unordered or not equal
7620 #. ``uno``: unordered (either nans)
7621 #. ``true``: no comparison, always returns true
7623 *Ordered* means that neither operand is a QNAN while *unordered* means
7624 that either operand may be a QNAN.
7626 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
7627 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
7628 type. They must have identical types.
7633 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
7634 condition code given as ``cond``. If the operands are vectors, then the
7635 vectors are compared element by element. Each comparison performed
7636 always yields an :ref:`i1 <t_integer>` result, as follows:
7638 #. ``false``: always yields ``false``, regardless of operands.
7639 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
7640 is equal to ``op2``.
7641 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
7642 is greater than ``op2``.
7643 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
7644 is greater than or equal to ``op2``.
7645 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
7646 is less than ``op2``.
7647 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
7648 is less than or equal to ``op2``.
7649 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
7650 is not equal to ``op2``.
7651 #. ``ord``: yields ``true`` if both operands are not a QNAN.
7652 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
7654 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
7655 greater than ``op2``.
7656 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
7657 greater than or equal to ``op2``.
7658 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
7660 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
7661 less than or equal to ``op2``.
7662 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
7663 not equal to ``op2``.
7664 #. ``uno``: yields ``true`` if either operand is a QNAN.
7665 #. ``true``: always yields ``true``, regardless of operands.
7667 The ``fcmp`` instruction can also optionally take any number of
7668 :ref:`fast-math flags <fastmath>`, which are optimization hints to enable
7669 otherwise unsafe floating point optimizations.
7671 Any set of fast-math flags are legal on an ``fcmp`` instruction, but the
7672 only flags that have any effect on its semantics are those that allow
7673 assumptions to be made about the values of input arguments; namely
7674 ``nnan``, ``ninf``, and ``nsz``. See :ref:`fastmath` for more information.
7679 .. code-block:: llvm
7681 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
7682 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
7683 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
7684 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
7686 Note that the code generator does not yet support vector types with the
7687 ``fcmp`` instruction.
7691 '``phi``' Instruction
7692 ^^^^^^^^^^^^^^^^^^^^^
7699 <result> = phi <ty> [ <val0>, <label0>], ...
7704 The '``phi``' instruction is used to implement the φ node in the SSA
7705 graph representing the function.
7710 The type of the incoming values is specified with the first type field.
7711 After this, the '``phi``' instruction takes a list of pairs as
7712 arguments, with one pair for each predecessor basic block of the current
7713 block. Only values of :ref:`first class <t_firstclass>` type may be used as
7714 the value arguments to the PHI node. Only labels may be used as the
7717 There must be no non-phi instructions between the start of a basic block
7718 and the PHI instructions: i.e. PHI instructions must be first in a basic
7721 For the purposes of the SSA form, the use of each incoming value is
7722 deemed to occur on the edge from the corresponding predecessor block to
7723 the current block (but after any definition of an '``invoke``'
7724 instruction's return value on the same edge).
7729 At runtime, the '``phi``' instruction logically takes on the value
7730 specified by the pair corresponding to the predecessor basic block that
7731 executed just prior to the current block.
7736 .. code-block:: llvm
7738 Loop: ; Infinite loop that counts from 0 on up...
7739 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
7740 %nextindvar = add i32 %indvar, 1
7745 '``select``' Instruction
7746 ^^^^^^^^^^^^^^^^^^^^^^^^
7753 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
7755 selty is either i1 or {<N x i1>}
7760 The '``select``' instruction is used to choose one value based on a
7761 condition, without IR-level branching.
7766 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
7767 values indicating the condition, and two values of the same :ref:`first
7768 class <t_firstclass>` type.
7773 If the condition is an i1 and it evaluates to 1, the instruction returns
7774 the first value argument; otherwise, it returns the second value
7777 If the condition is a vector of i1, then the value arguments must be
7778 vectors of the same size, and the selection is done element by element.
7780 If the condition is an i1 and the value arguments are vectors of the
7781 same size, then an entire vector is selected.
7786 .. code-block:: llvm
7788 %X = select i1 true, i8 17, i8 42 ; yields i8:17
7792 '``call``' Instruction
7793 ^^^^^^^^^^^^^^^^^^^^^^
7800 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
7805 The '``call``' instruction represents a simple function call.
7810 This instruction requires several arguments:
7812 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
7813 should perform tail call optimization. The ``tail`` marker is a hint that
7814 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
7815 means that the call must be tail call optimized in order for the program to
7816 be correct. The ``musttail`` marker provides these guarantees:
7818 #. The call will not cause unbounded stack growth if it is part of a
7819 recursive cycle in the call graph.
7820 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
7823 Both markers imply that the callee does not access allocas or varargs from
7824 the caller. Calls marked ``musttail`` must obey the following additional
7827 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
7828 or a pointer bitcast followed by a ret instruction.
7829 - The ret instruction must return the (possibly bitcasted) value
7830 produced by the call or void.
7831 - The caller and callee prototypes must match. Pointer types of
7832 parameters or return types may differ in pointee type, but not
7834 - The calling conventions of the caller and callee must match.
7835 - All ABI-impacting function attributes, such as sret, byval, inreg,
7836 returned, and inalloca, must match.
7837 - The callee must be varargs iff the caller is varargs. Bitcasting a
7838 non-varargs function to the appropriate varargs type is legal so
7839 long as the non-varargs prefixes obey the other rules.
7841 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
7842 the following conditions are met:
7844 - Caller and callee both have the calling convention ``fastcc``.
7845 - The call is in tail position (ret immediately follows call and ret
7846 uses value of call or is void).
7847 - Option ``-tailcallopt`` is enabled, or
7848 ``llvm::GuaranteedTailCallOpt`` is ``true``.
7849 - `Platform-specific constraints are
7850 met. <CodeGenerator.html#tailcallopt>`_
7852 #. The optional "cconv" marker indicates which :ref:`calling
7853 convention <callingconv>` the call should use. If none is
7854 specified, the call defaults to using C calling conventions. The
7855 calling convention of the call must match the calling convention of
7856 the target function, or else the behavior is undefined.
7857 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
7858 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
7860 #. '``ty``': the type of the call instruction itself which is also the
7861 type of the return value. Functions that return no value are marked
7863 #. '``fnty``': shall be the signature of the pointer to function value
7864 being invoked. The argument types must match the types implied by
7865 this signature. This type can be omitted if the function is not
7866 varargs and if the function type does not return a pointer to a
7868 #. '``fnptrval``': An LLVM value containing a pointer to a function to
7869 be invoked. In most cases, this is a direct function invocation, but
7870 indirect ``call``'s are just as possible, calling an arbitrary pointer
7872 #. '``function args``': argument list whose types match the function
7873 signature argument types and parameter attributes. All arguments must
7874 be of :ref:`first class <t_firstclass>` type. If the function signature
7875 indicates the function accepts a variable number of arguments, the
7876 extra arguments can be specified.
7877 #. The optional :ref:`function attributes <fnattrs>` list. Only
7878 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
7879 attributes are valid here.
7884 The '``call``' instruction is used to cause control flow to transfer to
7885 a specified function, with its incoming arguments bound to the specified
7886 values. Upon a '``ret``' instruction in the called function, control
7887 flow continues with the instruction after the function call, and the
7888 return value of the function is bound to the result argument.
7893 .. code-block:: llvm
7895 %retval = call i32 @test(i32 %argc)
7896 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
7897 %X = tail call i32 @foo() ; yields i32
7898 %Y = tail call fastcc i32 @foo() ; yields i32
7899 call void %foo(i8 97 signext)
7901 %struct.A = type { i32, i8 }
7902 %r = call %struct.A @foo() ; yields { i32, i8 }
7903 %gr = extractvalue %struct.A %r, 0 ; yields i32
7904 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
7905 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
7906 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
7908 llvm treats calls to some functions with names and arguments that match
7909 the standard C99 library as being the C99 library functions, and may
7910 perform optimizations or generate code for them under that assumption.
7911 This is something we'd like to change in the future to provide better
7912 support for freestanding environments and non-C-based languages.
7916 '``va_arg``' Instruction
7917 ^^^^^^^^^^^^^^^^^^^^^^^^
7924 <resultval> = va_arg <va_list*> <arglist>, <argty>
7929 The '``va_arg``' instruction is used to access arguments passed through
7930 the "variable argument" area of a function call. It is used to implement
7931 the ``va_arg`` macro in C.
7936 This instruction takes a ``va_list*`` value and the type of the
7937 argument. It returns a value of the specified argument type and
7938 increments the ``va_list`` to point to the next argument. The actual
7939 type of ``va_list`` is target specific.
7944 The '``va_arg``' instruction loads an argument of the specified type
7945 from the specified ``va_list`` and causes the ``va_list`` to point to
7946 the next argument. For more information, see the variable argument
7947 handling :ref:`Intrinsic Functions <int_varargs>`.
7949 It is legal for this instruction to be called in a function which does
7950 not take a variable number of arguments, for example, the ``vfprintf``
7953 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
7954 function <intrinsics>` because it takes a type as an argument.
7959 See the :ref:`variable argument processing <int_varargs>` section.
7961 Note that the code generator does not yet fully support va\_arg on many
7962 targets. Also, it does not currently support va\_arg with aggregate
7963 types on any target.
7967 '``landingpad``' Instruction
7968 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7975 <resultval> = landingpad <resultty> <clause>+
7976 <resultval> = landingpad <resultty> cleanup <clause>*
7978 <clause> := catch <type> <value>
7979 <clause> := filter <array constant type> <array constant>
7984 The '``landingpad``' instruction is used by `LLVM's exception handling
7985 system <ExceptionHandling.html#overview>`_ to specify that a basic block
7986 is a landing pad --- one where the exception lands, and corresponds to the
7987 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
7988 defines values supplied by the :ref:`personality function <personalityfn>` upon
7989 re-entry to the function. The ``resultval`` has the type ``resultty``.
7995 ``cleanup`` flag indicates that the landing pad block is a cleanup.
7997 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
7998 contains the global variable representing the "type" that may be caught
7999 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
8000 clause takes an array constant as its argument. Use
8001 "``[0 x i8**] undef``" for a filter which cannot throw. The
8002 '``landingpad``' instruction must contain *at least* one ``clause`` or
8003 the ``cleanup`` flag.
8008 The '``landingpad``' instruction defines the values which are set by the
8009 :ref:`personality function <personalityfn>` upon re-entry to the function, and
8010 therefore the "result type" of the ``landingpad`` instruction. As with
8011 calling conventions, how the personality function results are
8012 represented in LLVM IR is target specific.
8014 The clauses are applied in order from top to bottom. If two
8015 ``landingpad`` instructions are merged together through inlining, the
8016 clauses from the calling function are appended to the list of clauses.
8017 When the call stack is being unwound due to an exception being thrown,
8018 the exception is compared against each ``clause`` in turn. If it doesn't
8019 match any of the clauses, and the ``cleanup`` flag is not set, then
8020 unwinding continues further up the call stack.
8022 The ``landingpad`` instruction has several restrictions:
8024 - A landing pad block is a basic block which is the unwind destination
8025 of an '``invoke``' instruction.
8026 - A landing pad block must have a '``landingpad``' instruction as its
8027 first non-PHI instruction.
8028 - There can be only one '``landingpad``' instruction within the landing
8030 - A basic block that is not a landing pad block may not include a
8031 '``landingpad``' instruction.
8036 .. code-block:: llvm
8038 ;; A landing pad which can catch an integer.
8039 %res = landingpad { i8*, i32 }
8041 ;; A landing pad that is a cleanup.
8042 %res = landingpad { i8*, i32 }
8044 ;; A landing pad which can catch an integer and can only throw a double.
8045 %res = landingpad { i8*, i32 }
8047 filter [1 x i8**] [@_ZTId]
8054 LLVM supports the notion of an "intrinsic function". These functions
8055 have well known names and semantics and are required to follow certain
8056 restrictions. Overall, these intrinsics represent an extension mechanism
8057 for the LLVM language that does not require changing all of the
8058 transformations in LLVM when adding to the language (or the bitcode
8059 reader/writer, the parser, etc...).
8061 Intrinsic function names must all start with an "``llvm.``" prefix. This
8062 prefix is reserved in LLVM for intrinsic names; thus, function names may
8063 not begin with this prefix. Intrinsic functions must always be external
8064 functions: you cannot define the body of intrinsic functions. Intrinsic
8065 functions may only be used in call or invoke instructions: it is illegal
8066 to take the address of an intrinsic function. Additionally, because
8067 intrinsic functions are part of the LLVM language, it is required if any
8068 are added that they be documented here.
8070 Some intrinsic functions can be overloaded, i.e., the intrinsic
8071 represents a family of functions that perform the same operation but on
8072 different data types. Because LLVM can represent over 8 million
8073 different integer types, overloading is used commonly to allow an
8074 intrinsic function to operate on any integer type. One or more of the
8075 argument types or the result type can be overloaded to accept any
8076 integer type. Argument types may also be defined as exactly matching a
8077 previous argument's type or the result type. This allows an intrinsic
8078 function which accepts multiple arguments, but needs all of them to be
8079 of the same type, to only be overloaded with respect to a single
8080 argument or the result.
8082 Overloaded intrinsics will have the names of its overloaded argument
8083 types encoded into its function name, each preceded by a period. Only
8084 those types which are overloaded result in a name suffix. Arguments
8085 whose type is matched against another type do not. For example, the
8086 ``llvm.ctpop`` function can take an integer of any width and returns an
8087 integer of exactly the same integer width. This leads to a family of
8088 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
8089 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
8090 overloaded, and only one type suffix is required. Because the argument's
8091 type is matched against the return type, it does not require its own
8094 To learn how to add an intrinsic function, please see the `Extending
8095 LLVM Guide <ExtendingLLVM.html>`_.
8099 Variable Argument Handling Intrinsics
8100 -------------------------------------
8102 Variable argument support is defined in LLVM with the
8103 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
8104 functions. These functions are related to the similarly named macros
8105 defined in the ``<stdarg.h>`` header file.
8107 All of these functions operate on arguments that use a target-specific
8108 value type "``va_list``". The LLVM assembly language reference manual
8109 does not define what this type is, so all transformations should be
8110 prepared to handle these functions regardless of the type used.
8112 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
8113 variable argument handling intrinsic functions are used.
8115 .. code-block:: llvm
8117 ; This struct is different for every platform. For most platforms,
8118 ; it is merely an i8*.
8119 %struct.va_list = type { i8* }
8121 ; For Unix x86_64 platforms, va_list is the following struct:
8122 ; %struct.va_list = type { i32, i32, i8*, i8* }
8124 define i32 @test(i32 %X, ...) {
8125 ; Initialize variable argument processing
8126 %ap = alloca %struct.va_list
8127 %ap2 = bitcast %struct.va_list* %ap to i8*
8128 call void @llvm.va_start(i8* %ap2)
8130 ; Read a single integer argument
8131 %tmp = va_arg i8* %ap2, i32
8133 ; Demonstrate usage of llvm.va_copy and llvm.va_end
8135 %aq2 = bitcast i8** %aq to i8*
8136 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
8137 call void @llvm.va_end(i8* %aq2)
8139 ; Stop processing of arguments.
8140 call void @llvm.va_end(i8* %ap2)
8144 declare void @llvm.va_start(i8*)
8145 declare void @llvm.va_copy(i8*, i8*)
8146 declare void @llvm.va_end(i8*)
8150 '``llvm.va_start``' Intrinsic
8151 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8158 declare void @llvm.va_start(i8* <arglist>)
8163 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
8164 subsequent use by ``va_arg``.
8169 The argument is a pointer to a ``va_list`` element to initialize.
8174 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
8175 available in C. In a target-dependent way, it initializes the
8176 ``va_list`` element to which the argument points, so that the next call
8177 to ``va_arg`` will produce the first variable argument passed to the
8178 function. Unlike the C ``va_start`` macro, this intrinsic does not need
8179 to know the last argument of the function as the compiler can figure
8182 '``llvm.va_end``' Intrinsic
8183 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8190 declare void @llvm.va_end(i8* <arglist>)
8195 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
8196 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
8201 The argument is a pointer to a ``va_list`` to destroy.
8206 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
8207 available in C. In a target-dependent way, it destroys the ``va_list``
8208 element to which the argument points. Calls to
8209 :ref:`llvm.va_start <int_va_start>` and
8210 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
8215 '``llvm.va_copy``' Intrinsic
8216 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8223 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
8228 The '``llvm.va_copy``' intrinsic copies the current argument position
8229 from the source argument list to the destination argument list.
8234 The first argument is a pointer to a ``va_list`` element to initialize.
8235 The second argument is a pointer to a ``va_list`` element to copy from.
8240 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
8241 available in C. In a target-dependent way, it copies the source
8242 ``va_list`` element into the destination ``va_list`` element. This
8243 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
8244 arbitrarily complex and require, for example, memory allocation.
8246 Accurate Garbage Collection Intrinsics
8247 --------------------------------------
8249 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
8250 (GC) requires the frontend to generate code containing appropriate intrinsic
8251 calls and select an appropriate GC strategy which knows how to lower these
8252 intrinsics in a manner which is appropriate for the target collector.
8254 These intrinsics allow identification of :ref:`GC roots on the
8255 stack <int_gcroot>`, as well as garbage collector implementations that
8256 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
8257 Frontends for type-safe garbage collected languages should generate
8258 these intrinsics to make use of the LLVM garbage collectors. For more
8259 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
8261 Experimental Statepoint Intrinsics
8262 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8264 LLVM provides an second experimental set of intrinsics for describing garbage
8265 collection safepoints in compiled code. These intrinsics are an alternative
8266 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
8267 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
8268 differences in approach are covered in the `Garbage Collection with LLVM
8269 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
8270 described in :doc:`Statepoints`.
8274 '``llvm.gcroot``' Intrinsic
8275 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8282 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
8287 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
8288 the code generator, and allows some metadata to be associated with it.
8293 The first argument specifies the address of a stack object that contains
8294 the root pointer. The second pointer (which must be either a constant or
8295 a global value address) contains the meta-data to be associated with the
8301 At runtime, a call to this intrinsic stores a null pointer into the
8302 "ptrloc" location. At compile-time, the code generator generates
8303 information to allow the runtime to find the pointer at GC safe points.
8304 The '``llvm.gcroot``' intrinsic may only be used in a function which
8305 :ref:`specifies a GC algorithm <gc>`.
8309 '``llvm.gcread``' Intrinsic
8310 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8317 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
8322 The '``llvm.gcread``' intrinsic identifies reads of references from heap
8323 locations, allowing garbage collector implementations that require read
8329 The second argument is the address to read from, which should be an
8330 address allocated from the garbage collector. The first object is a
8331 pointer to the start of the referenced object, if needed by the language
8332 runtime (otherwise null).
8337 The '``llvm.gcread``' intrinsic has the same semantics as a load
8338 instruction, but may be replaced with substantially more complex code by
8339 the garbage collector runtime, as needed. The '``llvm.gcread``'
8340 intrinsic may only be used in a function which :ref:`specifies a GC
8345 '``llvm.gcwrite``' Intrinsic
8346 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8353 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
8358 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
8359 locations, allowing garbage collector implementations that require write
8360 barriers (such as generational or reference counting collectors).
8365 The first argument is the reference to store, the second is the start of
8366 the object to store it to, and the third is the address of the field of
8367 Obj to store to. If the runtime does not require a pointer to the
8368 object, Obj may be null.
8373 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
8374 instruction, but may be replaced with substantially more complex code by
8375 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
8376 intrinsic may only be used in a function which :ref:`specifies a GC
8379 Code Generator Intrinsics
8380 -------------------------
8382 These intrinsics are provided by LLVM to expose special features that
8383 may only be implemented with code generator support.
8385 '``llvm.returnaddress``' Intrinsic
8386 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8393 declare i8 *@llvm.returnaddress(i32 <level>)
8398 The '``llvm.returnaddress``' intrinsic attempts to compute a
8399 target-specific value indicating the return address of the current
8400 function or one of its callers.
8405 The argument to this intrinsic indicates which function to return the
8406 address for. Zero indicates the calling function, one indicates its
8407 caller, etc. The argument is **required** to be a constant integer
8413 The '``llvm.returnaddress``' intrinsic either returns a pointer
8414 indicating the return address of the specified call frame, or zero if it
8415 cannot be identified. The value returned by this intrinsic is likely to
8416 be incorrect or 0 for arguments other than zero, so it should only be
8417 used for debugging purposes.
8419 Note that calling this intrinsic does not prevent function inlining or
8420 other aggressive transformations, so the value returned may not be that
8421 of the obvious source-language caller.
8423 '``llvm.frameaddress``' Intrinsic
8424 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8431 declare i8* @llvm.frameaddress(i32 <level>)
8436 The '``llvm.frameaddress``' intrinsic attempts to return the
8437 target-specific frame pointer value for the specified stack frame.
8442 The argument to this intrinsic indicates which function to return the
8443 frame pointer for. Zero indicates the calling function, one indicates
8444 its caller, etc. The argument is **required** to be a constant integer
8450 The '``llvm.frameaddress``' intrinsic either returns a pointer
8451 indicating the frame address of the specified call frame, or zero if it
8452 cannot be identified. The value returned by this intrinsic is likely to
8453 be incorrect or 0 for arguments other than zero, so it should only be
8454 used for debugging purposes.
8456 Note that calling this intrinsic does not prevent function inlining or
8457 other aggressive transformations, so the value returned may not be that
8458 of the obvious source-language caller.
8460 '``llvm.localescape``' and '``llvm.localrecover``' Intrinsics
8461 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8468 declare void @llvm.localescape(...)
8469 declare i8* @llvm.localrecover(i8* %func, i8* %fp, i32 %idx)
8474 The '``llvm.localescape``' intrinsic escapes offsets of a collection of static
8475 allocas, and the '``llvm.localrecover``' intrinsic applies those offsets to a
8476 live frame pointer to recover the address of the allocation. The offset is
8477 computed during frame layout of the caller of ``llvm.localescape``.
8482 All arguments to '``llvm.localescape``' must be pointers to static allocas or
8483 casts of static allocas. Each function can only call '``llvm.localescape``'
8484 once, and it can only do so from the entry block.
8486 The ``func`` argument to '``llvm.localrecover``' must be a constant
8487 bitcasted pointer to a function defined in the current module. The code
8488 generator cannot determine the frame allocation offset of functions defined in
8491 The ``fp`` argument to '``llvm.localrecover``' must be a frame pointer of a
8492 call frame that is currently live. The return value of '``llvm.localaddress``'
8493 is one way to produce such a value, but various runtimes also expose a suitable
8494 pointer in platform-specific ways.
8496 The ``idx`` argument to '``llvm.localrecover``' indicates which alloca passed to
8497 '``llvm.localescape``' to recover. It is zero-indexed.
8502 These intrinsics allow a group of functions to share access to a set of local
8503 stack allocations of a one parent function. The parent function may call the
8504 '``llvm.localescape``' intrinsic once from the function entry block, and the
8505 child functions can use '``llvm.localrecover``' to access the escaped allocas.
8506 The '``llvm.localescape``' intrinsic blocks inlining, as inlining changes where
8507 the escaped allocas are allocated, which would break attempts to use
8508 '``llvm.localrecover``'.
8510 .. _int_read_register:
8511 .. _int_write_register:
8513 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
8514 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8521 declare i32 @llvm.read_register.i32(metadata)
8522 declare i64 @llvm.read_register.i64(metadata)
8523 declare void @llvm.write_register.i32(metadata, i32 @value)
8524 declare void @llvm.write_register.i64(metadata, i64 @value)
8530 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
8531 provides access to the named register. The register must be valid on
8532 the architecture being compiled to. The type needs to be compatible
8533 with the register being read.
8538 The '``llvm.read_register``' intrinsic returns the current value of the
8539 register, where possible. The '``llvm.write_register``' intrinsic sets
8540 the current value of the register, where possible.
8542 This is useful to implement named register global variables that need
8543 to always be mapped to a specific register, as is common practice on
8544 bare-metal programs including OS kernels.
8546 The compiler doesn't check for register availability or use of the used
8547 register in surrounding code, including inline assembly. Because of that,
8548 allocatable registers are not supported.
8550 Warning: So far it only works with the stack pointer on selected
8551 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
8552 work is needed to support other registers and even more so, allocatable
8557 '``llvm.stacksave``' Intrinsic
8558 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8565 declare i8* @llvm.stacksave()
8570 The '``llvm.stacksave``' intrinsic is used to remember the current state
8571 of the function stack, for use with
8572 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
8573 implementing language features like scoped automatic variable sized
8579 This intrinsic returns a opaque pointer value that can be passed to
8580 :ref:`llvm.stackrestore <int_stackrestore>`. When an
8581 ``llvm.stackrestore`` intrinsic is executed with a value saved from
8582 ``llvm.stacksave``, it effectively restores the state of the stack to
8583 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
8584 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
8585 were allocated after the ``llvm.stacksave`` was executed.
8587 .. _int_stackrestore:
8589 '``llvm.stackrestore``' Intrinsic
8590 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8597 declare void @llvm.stackrestore(i8* %ptr)
8602 The '``llvm.stackrestore``' intrinsic is used to restore the state of
8603 the function stack to the state it was in when the corresponding
8604 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
8605 useful for implementing language features like scoped automatic variable
8606 sized arrays in C99.
8611 See the description for :ref:`llvm.stacksave <int_stacksave>`.
8613 '``llvm.prefetch``' Intrinsic
8614 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8621 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
8626 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
8627 insert a prefetch instruction if supported; otherwise, it is a noop.
8628 Prefetches have no effect on the behavior of the program but can change
8629 its performance characteristics.
8634 ``address`` is the address to be prefetched, ``rw`` is the specifier
8635 determining if the fetch should be for a read (0) or write (1), and
8636 ``locality`` is a temporal locality specifier ranging from (0) - no
8637 locality, to (3) - extremely local keep in cache. The ``cache type``
8638 specifies whether the prefetch is performed on the data (1) or
8639 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
8640 arguments must be constant integers.
8645 This intrinsic does not modify the behavior of the program. In
8646 particular, prefetches cannot trap and do not produce a value. On
8647 targets that support this intrinsic, the prefetch can provide hints to
8648 the processor cache for better performance.
8650 '``llvm.pcmarker``' Intrinsic
8651 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8658 declare void @llvm.pcmarker(i32 <id>)
8663 The '``llvm.pcmarker``' intrinsic is a method to export a Program
8664 Counter (PC) in a region of code to simulators and other tools. The
8665 method is target specific, but it is expected that the marker will use
8666 exported symbols to transmit the PC of the marker. The marker makes no
8667 guarantees that it will remain with any specific instruction after
8668 optimizations. It is possible that the presence of a marker will inhibit
8669 optimizations. The intended use is to be inserted after optimizations to
8670 allow correlations of simulation runs.
8675 ``id`` is a numerical id identifying the marker.
8680 This intrinsic does not modify the behavior of the program. Backends
8681 that do not support this intrinsic may ignore it.
8683 '``llvm.readcyclecounter``' Intrinsic
8684 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8691 declare i64 @llvm.readcyclecounter()
8696 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
8697 counter register (or similar low latency, high accuracy clocks) on those
8698 targets that support it. On X86, it should map to RDTSC. On Alpha, it
8699 should map to RPCC. As the backing counters overflow quickly (on the
8700 order of 9 seconds on alpha), this should only be used for small
8706 When directly supported, reading the cycle counter should not modify any
8707 memory. Implementations are allowed to either return a application
8708 specific value or a system wide value. On backends without support, this
8709 is lowered to a constant 0.
8711 Note that runtime support may be conditional on the privilege-level code is
8712 running at and the host platform.
8714 '``llvm.clear_cache``' Intrinsic
8715 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8722 declare void @llvm.clear_cache(i8*, i8*)
8727 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
8728 in the specified range to the execution unit of the processor. On
8729 targets with non-unified instruction and data cache, the implementation
8730 flushes the instruction cache.
8735 On platforms with coherent instruction and data caches (e.g. x86), this
8736 intrinsic is a nop. On platforms with non-coherent instruction and data
8737 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
8738 instructions or a system call, if cache flushing requires special
8741 The default behavior is to emit a call to ``__clear_cache`` from the run
8744 This instrinsic does *not* empty the instruction pipeline. Modifications
8745 of the current function are outside the scope of the intrinsic.
8747 '``llvm.instrprof_increment``' Intrinsic
8748 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8755 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
8756 i32 <num-counters>, i32 <index>)
8761 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
8762 frontend for use with instrumentation based profiling. These will be
8763 lowered by the ``-instrprof`` pass to generate execution counts of a
8769 The first argument is a pointer to a global variable containing the
8770 name of the entity being instrumented. This should generally be the
8771 (mangled) function name for a set of counters.
8773 The second argument is a hash value that can be used by the consumer
8774 of the profile data to detect changes to the instrumented source, and
8775 the third is the number of counters associated with ``name``. It is an
8776 error if ``hash`` or ``num-counters`` differ between two instances of
8777 ``instrprof_increment`` that refer to the same name.
8779 The last argument refers to which of the counters for ``name`` should
8780 be incremented. It should be a value between 0 and ``num-counters``.
8785 This intrinsic represents an increment of a profiling counter. It will
8786 cause the ``-instrprof`` pass to generate the appropriate data
8787 structures and the code to increment the appropriate value, in a
8788 format that can be written out by a compiler runtime and consumed via
8789 the ``llvm-profdata`` tool.
8791 Standard C Library Intrinsics
8792 -----------------------------
8794 LLVM provides intrinsics for a few important standard C library
8795 functions. These intrinsics allow source-language front-ends to pass
8796 information about the alignment of the pointer arguments to the code
8797 generator, providing opportunity for more efficient code generation.
8801 '``llvm.memcpy``' Intrinsic
8802 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8807 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
8808 integer bit width and for different address spaces. Not all targets
8809 support all bit widths however.
8813 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
8814 i32 <len>, i32 <align>, i1 <isvolatile>)
8815 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
8816 i64 <len>, i32 <align>, i1 <isvolatile>)
8821 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
8822 source location to the destination location.
8824 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
8825 intrinsics do not return a value, takes extra alignment/isvolatile
8826 arguments and the pointers can be in specified address spaces.
8831 The first argument is a pointer to the destination, the second is a
8832 pointer to the source. The third argument is an integer argument
8833 specifying the number of bytes to copy, the fourth argument is the
8834 alignment of the source and destination locations, and the fifth is a
8835 boolean indicating a volatile access.
8837 If the call to this intrinsic has an alignment value that is not 0 or 1,
8838 then the caller guarantees that both the source and destination pointers
8839 are aligned to that boundary.
8841 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
8842 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
8843 very cleanly specified and it is unwise to depend on it.
8848 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
8849 source location to the destination location, which are not allowed to
8850 overlap. It copies "len" bytes of memory over. If the argument is known
8851 to be aligned to some boundary, this can be specified as the fourth
8852 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
8854 '``llvm.memmove``' Intrinsic
8855 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8860 This is an overloaded intrinsic. You can use llvm.memmove on any integer
8861 bit width and for different address space. Not all targets support all
8866 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
8867 i32 <len>, i32 <align>, i1 <isvolatile>)
8868 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
8869 i64 <len>, i32 <align>, i1 <isvolatile>)
8874 The '``llvm.memmove.*``' intrinsics move a block of memory from the
8875 source location to the destination location. It is similar to the
8876 '``llvm.memcpy``' intrinsic but allows the two memory locations to
8879 Note that, unlike the standard libc function, the ``llvm.memmove.*``
8880 intrinsics do not return a value, takes extra alignment/isvolatile
8881 arguments and the pointers can be in specified address spaces.
8886 The first argument is a pointer to the destination, the second is a
8887 pointer to the source. The third argument is an integer argument
8888 specifying the number of bytes to copy, the fourth argument is the
8889 alignment of the source and destination locations, and the fifth is a
8890 boolean indicating a volatile access.
8892 If the call to this intrinsic has an alignment value that is not 0 or 1,
8893 then the caller guarantees that the source and destination pointers are
8894 aligned to that boundary.
8896 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
8897 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
8898 not very cleanly specified and it is unwise to depend on it.
8903 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
8904 source location to the destination location, which may overlap. It
8905 copies "len" bytes of memory over. If the argument is known to be
8906 aligned to some boundary, this can be specified as the fourth argument,
8907 otherwise it should be set to 0 or 1 (both meaning no alignment).
8909 '``llvm.memset.*``' Intrinsics
8910 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8915 This is an overloaded intrinsic. You can use llvm.memset on any integer
8916 bit width and for different address spaces. However, not all targets
8917 support all bit widths.
8921 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
8922 i32 <len>, i32 <align>, i1 <isvolatile>)
8923 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
8924 i64 <len>, i32 <align>, i1 <isvolatile>)
8929 The '``llvm.memset.*``' intrinsics fill a block of memory with a
8930 particular byte value.
8932 Note that, unlike the standard libc function, the ``llvm.memset``
8933 intrinsic does not return a value and takes extra alignment/volatile
8934 arguments. Also, the destination can be in an arbitrary address space.
8939 The first argument is a pointer to the destination to fill, the second
8940 is the byte value with which to fill it, the third argument is an
8941 integer argument specifying the number of bytes to fill, and the fourth
8942 argument is the known alignment of the destination location.
8944 If the call to this intrinsic has an alignment value that is not 0 or 1,
8945 then the caller guarantees that the destination pointer is aligned to
8948 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
8949 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
8950 very cleanly specified and it is unwise to depend on it.
8955 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
8956 at the destination location. If the argument is known to be aligned to
8957 some boundary, this can be specified as the fourth argument, otherwise
8958 it should be set to 0 or 1 (both meaning no alignment).
8960 '``llvm.sqrt.*``' Intrinsic
8961 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8966 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
8967 floating point or vector of floating point type. Not all targets support
8972 declare float @llvm.sqrt.f32(float %Val)
8973 declare double @llvm.sqrt.f64(double %Val)
8974 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
8975 declare fp128 @llvm.sqrt.f128(fp128 %Val)
8976 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
8981 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
8982 returning the same value as the libm '``sqrt``' functions would. Unlike
8983 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
8984 negative numbers other than -0.0 (which allows for better optimization,
8985 because there is no need to worry about errno being set).
8986 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
8991 The argument and return value are floating point numbers of the same
8997 This function returns the sqrt of the specified operand if it is a
8998 nonnegative floating point number.
9000 '``llvm.powi.*``' Intrinsic
9001 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9006 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
9007 floating point or vector of floating point type. Not all targets support
9012 declare float @llvm.powi.f32(float %Val, i32 %power)
9013 declare double @llvm.powi.f64(double %Val, i32 %power)
9014 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
9015 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
9016 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
9021 The '``llvm.powi.*``' intrinsics return the first operand raised to the
9022 specified (positive or negative) power. The order of evaluation of
9023 multiplications is not defined. When a vector of floating point type is
9024 used, the second argument remains a scalar integer value.
9029 The second argument is an integer power, and the first is a value to
9030 raise to that power.
9035 This function returns the first value raised to the second power with an
9036 unspecified sequence of rounding operations.
9038 '``llvm.sin.*``' Intrinsic
9039 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9044 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
9045 floating point or vector of floating point type. Not all targets support
9050 declare float @llvm.sin.f32(float %Val)
9051 declare double @llvm.sin.f64(double %Val)
9052 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
9053 declare fp128 @llvm.sin.f128(fp128 %Val)
9054 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
9059 The '``llvm.sin.*``' intrinsics return the sine of the operand.
9064 The argument and return value are floating point numbers of the same
9070 This function returns the sine of the specified operand, returning the
9071 same values as the libm ``sin`` functions would, and handles error
9072 conditions in the same way.
9074 '``llvm.cos.*``' Intrinsic
9075 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9080 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
9081 floating point or vector of floating point type. Not all targets support
9086 declare float @llvm.cos.f32(float %Val)
9087 declare double @llvm.cos.f64(double %Val)
9088 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
9089 declare fp128 @llvm.cos.f128(fp128 %Val)
9090 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
9095 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
9100 The argument and return value are floating point numbers of the same
9106 This function returns the cosine of the specified operand, returning the
9107 same values as the libm ``cos`` functions would, and handles error
9108 conditions in the same way.
9110 '``llvm.pow.*``' Intrinsic
9111 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9116 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
9117 floating point or vector of floating point type. Not all targets support
9122 declare float @llvm.pow.f32(float %Val, float %Power)
9123 declare double @llvm.pow.f64(double %Val, double %Power)
9124 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
9125 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
9126 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
9131 The '``llvm.pow.*``' intrinsics return the first operand raised to the
9132 specified (positive or negative) power.
9137 The second argument is a floating point power, and the first is a value
9138 to raise to that power.
9143 This function returns the first value raised to the second power,
9144 returning the same values as the libm ``pow`` functions would, and
9145 handles error conditions in the same way.
9147 '``llvm.exp.*``' Intrinsic
9148 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9153 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
9154 floating point or vector of floating point type. Not all targets support
9159 declare float @llvm.exp.f32(float %Val)
9160 declare double @llvm.exp.f64(double %Val)
9161 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
9162 declare fp128 @llvm.exp.f128(fp128 %Val)
9163 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
9168 The '``llvm.exp.*``' intrinsics perform the exp function.
9173 The argument and return value are floating point numbers of the same
9179 This function returns the same values as the libm ``exp`` functions
9180 would, and handles error conditions in the same way.
9182 '``llvm.exp2.*``' Intrinsic
9183 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9188 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
9189 floating point or vector of floating point type. Not all targets support
9194 declare float @llvm.exp2.f32(float %Val)
9195 declare double @llvm.exp2.f64(double %Val)
9196 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
9197 declare fp128 @llvm.exp2.f128(fp128 %Val)
9198 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
9203 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
9208 The argument and return value are floating point numbers of the same
9214 This function returns the same values as the libm ``exp2`` functions
9215 would, and handles error conditions in the same way.
9217 '``llvm.log.*``' Intrinsic
9218 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9223 This is an overloaded intrinsic. You can use ``llvm.log`` on any
9224 floating point or vector of floating point type. Not all targets support
9229 declare float @llvm.log.f32(float %Val)
9230 declare double @llvm.log.f64(double %Val)
9231 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
9232 declare fp128 @llvm.log.f128(fp128 %Val)
9233 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
9238 The '``llvm.log.*``' intrinsics perform the log function.
9243 The argument and return value are floating point numbers of the same
9249 This function returns the same values as the libm ``log`` functions
9250 would, and handles error conditions in the same way.
9252 '``llvm.log10.*``' Intrinsic
9253 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9258 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
9259 floating point or vector of floating point type. Not all targets support
9264 declare float @llvm.log10.f32(float %Val)
9265 declare double @llvm.log10.f64(double %Val)
9266 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
9267 declare fp128 @llvm.log10.f128(fp128 %Val)
9268 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
9273 The '``llvm.log10.*``' intrinsics perform the log10 function.
9278 The argument and return value are floating point numbers of the same
9284 This function returns the same values as the libm ``log10`` functions
9285 would, and handles error conditions in the same way.
9287 '``llvm.log2.*``' Intrinsic
9288 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9293 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
9294 floating point or vector of floating point type. Not all targets support
9299 declare float @llvm.log2.f32(float %Val)
9300 declare double @llvm.log2.f64(double %Val)
9301 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
9302 declare fp128 @llvm.log2.f128(fp128 %Val)
9303 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
9308 The '``llvm.log2.*``' intrinsics perform the log2 function.
9313 The argument and return value are floating point numbers of the same
9319 This function returns the same values as the libm ``log2`` functions
9320 would, and handles error conditions in the same way.
9322 '``llvm.fma.*``' Intrinsic
9323 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9328 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
9329 floating point or vector of floating point type. Not all targets support
9334 declare float @llvm.fma.f32(float %a, float %b, float %c)
9335 declare double @llvm.fma.f64(double %a, double %b, double %c)
9336 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
9337 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
9338 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
9343 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
9349 The argument and return value are floating point numbers of the same
9355 This function returns the same values as the libm ``fma`` functions
9356 would, and does not set errno.
9358 '``llvm.fabs.*``' Intrinsic
9359 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9364 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
9365 floating point or vector of floating point type. Not all targets support
9370 declare float @llvm.fabs.f32(float %Val)
9371 declare double @llvm.fabs.f64(double %Val)
9372 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
9373 declare fp128 @llvm.fabs.f128(fp128 %Val)
9374 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
9379 The '``llvm.fabs.*``' intrinsics return the absolute value of the
9385 The argument and return value are floating point numbers of the same
9391 This function returns the same values as the libm ``fabs`` functions
9392 would, and handles error conditions in the same way.
9394 '``llvm.minnum.*``' Intrinsic
9395 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9400 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
9401 floating point or vector of floating point type. Not all targets support
9406 declare float @llvm.minnum.f32(float %Val0, float %Val1)
9407 declare double @llvm.minnum.f64(double %Val0, double %Val1)
9408 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
9409 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
9410 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
9415 The '``llvm.minnum.*``' intrinsics return the minimum of the two
9422 The arguments and return value are floating point numbers of the same
9428 Follows the IEEE-754 semantics for minNum, which also match for libm's
9431 If either operand is a NaN, returns the other non-NaN operand. Returns
9432 NaN only if both operands are NaN. If the operands compare equal,
9433 returns a value that compares equal to both operands. This means that
9434 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
9436 '``llvm.maxnum.*``' Intrinsic
9437 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9442 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
9443 floating point or vector of floating point type. Not all targets support
9448 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
9449 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
9450 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
9451 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
9452 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
9457 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
9464 The arguments and return value are floating point numbers of the same
9469 Follows the IEEE-754 semantics for maxNum, which also match for libm's
9472 If either operand is a NaN, returns the other non-NaN operand. Returns
9473 NaN only if both operands are NaN. If the operands compare equal,
9474 returns a value that compares equal to both operands. This means that
9475 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
9477 '``llvm.copysign.*``' Intrinsic
9478 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9483 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
9484 floating point or vector of floating point type. Not all targets support
9489 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
9490 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
9491 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
9492 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
9493 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
9498 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
9499 first operand and the sign of the second operand.
9504 The arguments and return value are floating point numbers of the same
9510 This function returns the same values as the libm ``copysign``
9511 functions would, and handles error conditions in the same way.
9513 '``llvm.floor.*``' Intrinsic
9514 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9519 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
9520 floating point or vector of floating point type. Not all targets support
9525 declare float @llvm.floor.f32(float %Val)
9526 declare double @llvm.floor.f64(double %Val)
9527 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
9528 declare fp128 @llvm.floor.f128(fp128 %Val)
9529 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
9534 The '``llvm.floor.*``' intrinsics return the floor of the operand.
9539 The argument and return value are floating point numbers of the same
9545 This function returns the same values as the libm ``floor`` functions
9546 would, and handles error conditions in the same way.
9548 '``llvm.ceil.*``' Intrinsic
9549 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9554 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
9555 floating point or vector of floating point type. Not all targets support
9560 declare float @llvm.ceil.f32(float %Val)
9561 declare double @llvm.ceil.f64(double %Val)
9562 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
9563 declare fp128 @llvm.ceil.f128(fp128 %Val)
9564 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
9569 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
9574 The argument and return value are floating point numbers of the same
9580 This function returns the same values as the libm ``ceil`` functions
9581 would, and handles error conditions in the same way.
9583 '``llvm.trunc.*``' Intrinsic
9584 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9589 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
9590 floating point or vector of floating point type. Not all targets support
9595 declare float @llvm.trunc.f32(float %Val)
9596 declare double @llvm.trunc.f64(double %Val)
9597 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
9598 declare fp128 @llvm.trunc.f128(fp128 %Val)
9599 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
9604 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
9605 nearest integer not larger in magnitude than the operand.
9610 The argument and return value are floating point numbers of the same
9616 This function returns the same values as the libm ``trunc`` functions
9617 would, and handles error conditions in the same way.
9619 '``llvm.rint.*``' Intrinsic
9620 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9625 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
9626 floating point or vector of floating point type. Not all targets support
9631 declare float @llvm.rint.f32(float %Val)
9632 declare double @llvm.rint.f64(double %Val)
9633 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
9634 declare fp128 @llvm.rint.f128(fp128 %Val)
9635 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
9640 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
9641 nearest integer. It may raise an inexact floating-point exception if the
9642 operand isn't an integer.
9647 The argument and return value are floating point numbers of the same
9653 This function returns the same values as the libm ``rint`` functions
9654 would, and handles error conditions in the same way.
9656 '``llvm.nearbyint.*``' Intrinsic
9657 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9662 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
9663 floating point or vector of floating point type. Not all targets support
9668 declare float @llvm.nearbyint.f32(float %Val)
9669 declare double @llvm.nearbyint.f64(double %Val)
9670 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
9671 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
9672 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
9677 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
9683 The argument and return value are floating point numbers of the same
9689 This function returns the same values as the libm ``nearbyint``
9690 functions would, and handles error conditions in the same way.
9692 '``llvm.round.*``' Intrinsic
9693 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9698 This is an overloaded intrinsic. You can use ``llvm.round`` on any
9699 floating point or vector of floating point type. Not all targets support
9704 declare float @llvm.round.f32(float %Val)
9705 declare double @llvm.round.f64(double %Val)
9706 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
9707 declare fp128 @llvm.round.f128(fp128 %Val)
9708 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
9713 The '``llvm.round.*``' intrinsics returns the operand rounded to the
9719 The argument and return value are floating point numbers of the same
9725 This function returns the same values as the libm ``round``
9726 functions would, and handles error conditions in the same way.
9728 Bit Manipulation Intrinsics
9729 ---------------------------
9731 LLVM provides intrinsics for a few important bit manipulation
9732 operations. These allow efficient code generation for some algorithms.
9734 '``llvm.bswap.*``' Intrinsics
9735 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9740 This is an overloaded intrinsic function. You can use bswap on any
9741 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
9745 declare i16 @llvm.bswap.i16(i16 <id>)
9746 declare i32 @llvm.bswap.i32(i32 <id>)
9747 declare i64 @llvm.bswap.i64(i64 <id>)
9752 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
9753 values with an even number of bytes (positive multiple of 16 bits).
9754 These are useful for performing operations on data that is not in the
9755 target's native byte order.
9760 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
9761 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
9762 intrinsic returns an i32 value that has the four bytes of the input i32
9763 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
9764 returned i32 will have its bytes in 3, 2, 1, 0 order. The
9765 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
9766 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
9769 '``llvm.ctpop.*``' Intrinsic
9770 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9775 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
9776 bit width, or on any vector with integer elements. Not all targets
9777 support all bit widths or vector types, however.
9781 declare i8 @llvm.ctpop.i8(i8 <src>)
9782 declare i16 @llvm.ctpop.i16(i16 <src>)
9783 declare i32 @llvm.ctpop.i32(i32 <src>)
9784 declare i64 @llvm.ctpop.i64(i64 <src>)
9785 declare i256 @llvm.ctpop.i256(i256 <src>)
9786 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
9791 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
9797 The only argument is the value to be counted. The argument may be of any
9798 integer type, or a vector with integer elements. The return type must
9799 match the argument type.
9804 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
9805 each element of a vector.
9807 '``llvm.ctlz.*``' Intrinsic
9808 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9813 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
9814 integer bit width, or any vector whose elements are integers. Not all
9815 targets support all bit widths or vector types, however.
9819 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
9820 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
9821 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
9822 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
9823 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
9824 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
9829 The '``llvm.ctlz``' family of intrinsic functions counts the number of
9830 leading zeros in a variable.
9835 The first argument is the value to be counted. This argument may be of
9836 any integer type, or a vector with integer element type. The return
9837 type must match the first argument type.
9839 The second argument must be a constant and is a flag to indicate whether
9840 the intrinsic should ensure that a zero as the first argument produces a
9841 defined result. Historically some architectures did not provide a
9842 defined result for zero values as efficiently, and many algorithms are
9843 now predicated on avoiding zero-value inputs.
9848 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
9849 zeros in a variable, or within each element of the vector. If
9850 ``src == 0`` then the result is the size in bits of the type of ``src``
9851 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
9852 ``llvm.ctlz(i32 2) = 30``.
9854 '``llvm.cttz.*``' Intrinsic
9855 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9860 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
9861 integer bit width, or any vector of integer elements. Not all targets
9862 support all bit widths or vector types, however.
9866 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
9867 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
9868 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
9869 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
9870 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
9871 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
9876 The '``llvm.cttz``' family of intrinsic functions counts the number of
9882 The first argument is the value to be counted. This argument may be of
9883 any integer type, or a vector with integer element type. The return
9884 type must match the first argument type.
9886 The second argument must be a constant and is a flag to indicate whether
9887 the intrinsic should ensure that a zero as the first argument produces a
9888 defined result. Historically some architectures did not provide a
9889 defined result for zero values as efficiently, and many algorithms are
9890 now predicated on avoiding zero-value inputs.
9895 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
9896 zeros in a variable, or within each element of a vector. If ``src == 0``
9897 then the result is the size in bits of the type of ``src`` if
9898 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
9899 ``llvm.cttz(2) = 1``.
9903 Arithmetic with Overflow Intrinsics
9904 -----------------------------------
9906 LLVM provides intrinsics for some arithmetic with overflow operations.
9908 '``llvm.sadd.with.overflow.*``' Intrinsics
9909 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9914 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
9915 on any integer bit width.
9919 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
9920 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
9921 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
9926 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
9927 a signed addition of the two arguments, and indicate whether an overflow
9928 occurred during the signed summation.
9933 The arguments (%a and %b) and the first element of the result structure
9934 may be of integer types of any bit width, but they must have the same
9935 bit width. The second element of the result structure must be of type
9936 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9942 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
9943 a signed addition of the two variables. They return a structure --- the
9944 first element of which is the signed summation, and the second element
9945 of which is a bit specifying if the signed summation resulted in an
9951 .. code-block:: llvm
9953 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
9954 %sum = extractvalue {i32, i1} %res, 0
9955 %obit = extractvalue {i32, i1} %res, 1
9956 br i1 %obit, label %overflow, label %normal
9958 '``llvm.uadd.with.overflow.*``' Intrinsics
9959 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9964 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
9965 on any integer bit width.
9969 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
9970 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
9971 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
9976 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
9977 an unsigned addition of the two arguments, and indicate whether a carry
9978 occurred during the unsigned summation.
9983 The arguments (%a and %b) and the first element of the result structure
9984 may be of integer types of any bit width, but they must have the same
9985 bit width. The second element of the result structure must be of type
9986 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9992 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
9993 an unsigned addition of the two arguments. They return a structure --- the
9994 first element of which is the sum, and the second element of which is a
9995 bit specifying if the unsigned summation resulted in a carry.
10000 .. code-block:: llvm
10002 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10003 %sum = extractvalue {i32, i1} %res, 0
10004 %obit = extractvalue {i32, i1} %res, 1
10005 br i1 %obit, label %carry, label %normal
10007 '``llvm.ssub.with.overflow.*``' Intrinsics
10008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10013 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
10014 on any integer bit width.
10018 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
10019 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10020 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
10025 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10026 a signed subtraction of the two arguments, and indicate whether an
10027 overflow occurred during the signed subtraction.
10032 The arguments (%a and %b) and the first element of the result structure
10033 may be of integer types of any bit width, but they must have the same
10034 bit width. The second element of the result structure must be of type
10035 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10041 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10042 a signed subtraction of the two arguments. They return a structure --- the
10043 first element of which is the subtraction, and the second element of
10044 which is a bit specifying if the signed subtraction resulted in an
10050 .. code-block:: llvm
10052 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10053 %sum = extractvalue {i32, i1} %res, 0
10054 %obit = extractvalue {i32, i1} %res, 1
10055 br i1 %obit, label %overflow, label %normal
10057 '``llvm.usub.with.overflow.*``' Intrinsics
10058 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10063 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
10064 on any integer bit width.
10068 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
10069 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10070 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
10075 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10076 an unsigned subtraction of the two arguments, and indicate whether an
10077 overflow occurred during the unsigned subtraction.
10082 The arguments (%a and %b) and the first element of the result structure
10083 may be of integer types of any bit width, but they must have the same
10084 bit width. The second element of the result structure must be of type
10085 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10091 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10092 an unsigned subtraction of the two arguments. They return a structure ---
10093 the first element of which is the subtraction, and the second element of
10094 which is a bit specifying if the unsigned subtraction resulted in an
10100 .. code-block:: llvm
10102 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10103 %sum = extractvalue {i32, i1} %res, 0
10104 %obit = extractvalue {i32, i1} %res, 1
10105 br i1 %obit, label %overflow, label %normal
10107 '``llvm.smul.with.overflow.*``' Intrinsics
10108 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10113 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
10114 on any integer bit width.
10118 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
10119 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10120 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
10125 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10126 a signed multiplication of the two arguments, and indicate whether an
10127 overflow occurred during the signed multiplication.
10132 The arguments (%a and %b) and the first element of the result structure
10133 may be of integer types of any bit width, but they must have the same
10134 bit width. The second element of the result structure must be of type
10135 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10141 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10142 a signed multiplication of the two arguments. They return a structure ---
10143 the first element of which is the multiplication, and the second element
10144 of which is a bit specifying if the signed multiplication resulted in an
10150 .. code-block:: llvm
10152 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10153 %sum = extractvalue {i32, i1} %res, 0
10154 %obit = extractvalue {i32, i1} %res, 1
10155 br i1 %obit, label %overflow, label %normal
10157 '``llvm.umul.with.overflow.*``' Intrinsics
10158 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10163 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
10164 on any integer bit width.
10168 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
10169 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10170 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
10175 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10176 a unsigned multiplication of the two arguments, and indicate whether an
10177 overflow occurred during the unsigned multiplication.
10182 The arguments (%a and %b) and the first element of the result structure
10183 may be of integer types of any bit width, but they must have the same
10184 bit width. The second element of the result structure must be of type
10185 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10191 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10192 an unsigned multiplication of the two arguments. They return a structure ---
10193 the first element of which is the multiplication, and the second
10194 element of which is a bit specifying if the unsigned multiplication
10195 resulted in an overflow.
10200 .. code-block:: llvm
10202 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10203 %sum = extractvalue {i32, i1} %res, 0
10204 %obit = extractvalue {i32, i1} %res, 1
10205 br i1 %obit, label %overflow, label %normal
10207 Specialised Arithmetic Intrinsics
10208 ---------------------------------
10210 '``llvm.canonicalize.*``' Intrinsic
10211 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10218 declare float @llvm.canonicalize.f32(float %a)
10219 declare double @llvm.canonicalize.f64(double %b)
10224 The '``llvm.canonicalize.*``' intrinsic returns the platform specific canonical
10225 encoding of a floating point number. This canonicalization is useful for
10226 implementing certain numeric primitives such as frexp. The canonical encoding is
10227 defined by IEEE-754-2008 to be:
10231 2.1.8 canonical encoding: The preferred encoding of a floating-point
10232 representation in a format. Applied to declets, significands of finite
10233 numbers, infinities, and NaNs, especially in decimal formats.
10235 This operation can also be considered equivalent to the IEEE-754-2008
10236 conversion of a floating-point value to the same format. NaNs are handled
10237 according to section 6.2.
10239 Examples of non-canonical encodings:
10241 - x87 pseudo denormals, pseudo NaNs, pseudo Infinity, Unnormals. These are
10242 converted to a canonical representation per hardware-specific protocol.
10243 - Many normal decimal floating point numbers have non-canonical alternative
10245 - Some machines, like GPUs or ARMv7 NEON, do not support subnormal values.
10246 These are treated as non-canonical encodings of zero and with be flushed to
10247 a zero of the same sign by this operation.
10249 Note that per IEEE-754-2008 6.2, systems that support signaling NaNs with
10250 default exception handling must signal an invalid exception, and produce a
10253 This function should always be implementable as multiplication by 1.0, provided
10254 that the compiler does not constant fold the operation. Likewise, division by
10255 1.0 and ``llvm.minnum(x, x)`` are possible implementations. Addition with
10256 -0.0 is also sufficient provided that the rounding mode is not -Infinity.
10258 ``@llvm.canonicalize`` must preserve the equality relation. That is:
10260 - ``(@llvm.canonicalize(x) == x)`` is equivalent to ``(x == x)``
10261 - ``(@llvm.canonicalize(x) == @llvm.canonicalize(y))`` is equivalent to
10264 Additionally, the sign of zero must be conserved:
10265 ``@llvm.canonicalize(-0.0) = -0.0`` and ``@llvm.canonicalize(+0.0) = +0.0``
10267 The payload bits of a NaN must be conserved, with two exceptions.
10268 First, environments which use only a single canonical representation of NaN
10269 must perform said canonicalization. Second, SNaNs must be quieted per the
10272 The canonicalization operation may be optimized away if:
10274 - The input is known to be canonical. For example, it was produced by a
10275 floating-point operation that is required by the standard to be canonical.
10276 - The result is consumed only by (or fused with) other floating-point
10277 operations. That is, the bits of the floating point value are not examined.
10279 '``llvm.fmuladd.*``' Intrinsic
10280 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10287 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
10288 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
10293 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
10294 expressions that can be fused if the code generator determines that (a) the
10295 target instruction set has support for a fused operation, and (b) that the
10296 fused operation is more efficient than the equivalent, separate pair of mul
10297 and add instructions.
10302 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
10303 multiplicands, a and b, and an addend c.
10312 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
10314 is equivalent to the expression a \* b + c, except that rounding will
10315 not be performed between the multiplication and addition steps if the
10316 code generator fuses the operations. Fusion is not guaranteed, even if
10317 the target platform supports it. If a fused multiply-add is required the
10318 corresponding llvm.fma.\* intrinsic function should be used
10319 instead. This never sets errno, just as '``llvm.fma.*``'.
10324 .. code-block:: llvm
10326 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
10328 Half Precision Floating Point Intrinsics
10329 ----------------------------------------
10331 For most target platforms, half precision floating point is a
10332 storage-only format. This means that it is a dense encoding (in memory)
10333 but does not support computation in the format.
10335 This means that code must first load the half-precision floating point
10336 value as an i16, then convert it to float with
10337 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
10338 then be performed on the float value (including extending to double
10339 etc). To store the value back to memory, it is first converted to float
10340 if needed, then converted to i16 with
10341 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
10344 .. _int_convert_to_fp16:
10346 '``llvm.convert.to.fp16``' Intrinsic
10347 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10354 declare i16 @llvm.convert.to.fp16.f32(float %a)
10355 declare i16 @llvm.convert.to.fp16.f64(double %a)
10360 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
10361 conventional floating point type to half precision floating point format.
10366 The intrinsic function contains single argument - the value to be
10372 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
10373 conventional floating point format to half precision floating point format. The
10374 return value is an ``i16`` which contains the converted number.
10379 .. code-block:: llvm
10381 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
10382 store i16 %res, i16* @x, align 2
10384 .. _int_convert_from_fp16:
10386 '``llvm.convert.from.fp16``' Intrinsic
10387 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10394 declare float @llvm.convert.from.fp16.f32(i16 %a)
10395 declare double @llvm.convert.from.fp16.f64(i16 %a)
10400 The '``llvm.convert.from.fp16``' intrinsic function performs a
10401 conversion from half precision floating point format to single precision
10402 floating point format.
10407 The intrinsic function contains single argument - the value to be
10413 The '``llvm.convert.from.fp16``' intrinsic function performs a
10414 conversion from half single precision floating point format to single
10415 precision floating point format. The input half-float value is
10416 represented by an ``i16`` value.
10421 .. code-block:: llvm
10423 %a = load i16, i16* @x, align 2
10424 %res = call float @llvm.convert.from.fp16(i16 %a)
10426 .. _dbg_intrinsics:
10428 Debugger Intrinsics
10429 -------------------
10431 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
10432 prefix), are described in the `LLVM Source Level
10433 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
10436 Exception Handling Intrinsics
10437 -----------------------------
10439 The LLVM exception handling intrinsics (which all start with
10440 ``llvm.eh.`` prefix), are described in the `LLVM Exception
10441 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
10443 .. _int_trampoline:
10445 Trampoline Intrinsics
10446 ---------------------
10448 These intrinsics make it possible to excise one parameter, marked with
10449 the :ref:`nest <nest>` attribute, from a function. The result is a
10450 callable function pointer lacking the nest parameter - the caller does
10451 not need to provide a value for it. Instead, the value to use is stored
10452 in advance in a "trampoline", a block of memory usually allocated on the
10453 stack, which also contains code to splice the nest value into the
10454 argument list. This is used to implement the GCC nested function address
10457 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
10458 then the resulting function pointer has signature ``i32 (i32, i32)*``.
10459 It can be created as follows:
10461 .. code-block:: llvm
10463 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
10464 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
10465 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
10466 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
10467 %fp = bitcast i8* %p to i32 (i32, i32)*
10469 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
10470 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
10474 '``llvm.init.trampoline``' Intrinsic
10475 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10482 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
10487 This fills the memory pointed to by ``tramp`` with executable code,
10488 turning it into a trampoline.
10493 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
10494 pointers. The ``tramp`` argument must point to a sufficiently large and
10495 sufficiently aligned block of memory; this memory is written to by the
10496 intrinsic. Note that the size and the alignment are target-specific -
10497 LLVM currently provides no portable way of determining them, so a
10498 front-end that generates this intrinsic needs to have some
10499 target-specific knowledge. The ``func`` argument must hold a function
10500 bitcast to an ``i8*``.
10505 The block of memory pointed to by ``tramp`` is filled with target
10506 dependent code, turning it into a function. Then ``tramp`` needs to be
10507 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
10508 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
10509 function's signature is the same as that of ``func`` with any arguments
10510 marked with the ``nest`` attribute removed. At most one such ``nest``
10511 argument is allowed, and it must be of pointer type. Calling the new
10512 function is equivalent to calling ``func`` with the same argument list,
10513 but with ``nval`` used for the missing ``nest`` argument. If, after
10514 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
10515 modified, then the effect of any later call to the returned function
10516 pointer is undefined.
10520 '``llvm.adjust.trampoline``' Intrinsic
10521 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10528 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
10533 This performs any required machine-specific adjustment to the address of
10534 a trampoline (passed as ``tramp``).
10539 ``tramp`` must point to a block of memory which already has trampoline
10540 code filled in by a previous call to
10541 :ref:`llvm.init.trampoline <int_it>`.
10546 On some architectures the address of the code to be executed needs to be
10547 different than the address where the trampoline is actually stored. This
10548 intrinsic returns the executable address corresponding to ``tramp``
10549 after performing the required machine specific adjustments. The pointer
10550 returned can then be :ref:`bitcast and executed <int_trampoline>`.
10552 .. _int_mload_mstore:
10554 Masked Vector Load and Store Intrinsics
10555 ---------------------------------------
10557 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.
10561 '``llvm.masked.load.*``' Intrinsics
10562 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10566 This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
10570 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
10571 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
10576 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.
10582 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.
10588 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.
10589 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.
10594 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
10596 ;; The result of the two following instructions is identical aside from potential memory access exception
10597 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
10598 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
10602 '``llvm.masked.store.*``' Intrinsics
10603 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10607 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
10611 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
10612 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
10617 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.
10622 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.
10628 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.
10629 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.
10633 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
10635 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
10636 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
10637 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
10638 store <16 x float> %res, <16 x float>* %ptr, align 4
10641 Masked Vector Gather and Scatter Intrinsics
10642 -------------------------------------------
10644 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.
10648 '``llvm.masked.gather.*``' Intrinsics
10649 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10653 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.
10657 declare <16 x float> @llvm.masked.gather.v16f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
10658 declare <2 x double> @llvm.masked.gather.v2f64 (<2 x double*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
10663 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.
10669 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.
10675 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.
10676 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.
10681 %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>)
10683 ;; The gather with all-true mask is equivalent to the following instruction sequence
10684 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
10685 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
10686 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
10687 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
10689 %val0 = load double, double* %ptr0, align 8
10690 %val1 = load double, double* %ptr1, align 8
10691 %val2 = load double, double* %ptr2, align 8
10692 %val3 = load double, double* %ptr3, align 8
10694 %vec0 = insertelement <4 x double>undef, %val0, 0
10695 %vec01 = insertelement <4 x double>%vec0, %val1, 1
10696 %vec012 = insertelement <4 x double>%vec01, %val2, 2
10697 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
10701 '``llvm.masked.scatter.*``' Intrinsics
10702 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10706 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.
10710 declare void @llvm.masked.scatter.v8i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
10711 declare void @llvm.masked.scatter.v16f32(<16 x i32> <value>, <16 x i32*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
10716 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.
10721 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.
10727 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 divergency. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar store operations.
10731 ;; This instruction unconditionaly stores data vector in multiple addresses
10732 call @llvm.masked.scatter.v8i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
10734 ;; It is equivalent to a list of scalar stores
10735 %val0 = extractelement <8 x i32> %value, i32 0
10736 %val1 = extractelement <8 x i32> %value, i32 1
10738 %val7 = extractelement <8 x i32> %value, i32 7
10739 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
10740 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
10742 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
10743 ;; Note: the order of the following stores is important when they overlap:
10744 store i32 %val0, i32* %ptr0, align 4
10745 store i32 %val1, i32* %ptr1, align 4
10747 store i32 %val7, i32* %ptr7, align 4
10753 This class of intrinsics provides information about the lifetime of
10754 memory objects and ranges where variables are immutable.
10758 '``llvm.lifetime.start``' Intrinsic
10759 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10766 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
10771 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
10777 The first argument is a constant integer representing the size of the
10778 object, or -1 if it is variable sized. The second argument is a pointer
10784 This intrinsic indicates that before this point in the code, the value
10785 of the memory pointed to by ``ptr`` is dead. This means that it is known
10786 to never be used and has an undefined value. A load from the pointer
10787 that precedes this intrinsic can be replaced with ``'undef'``.
10791 '``llvm.lifetime.end``' Intrinsic
10792 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10799 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
10804 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
10810 The first argument is a constant integer representing the size of the
10811 object, or -1 if it is variable sized. The second argument is a pointer
10817 This intrinsic indicates that after this point in the code, the value of
10818 the memory pointed to by ``ptr`` is dead. This means that it is known to
10819 never be used and has an undefined value. Any stores into the memory
10820 object following this intrinsic may be removed as dead.
10822 '``llvm.invariant.start``' Intrinsic
10823 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10830 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
10835 The '``llvm.invariant.start``' intrinsic specifies that the contents of
10836 a memory object will not change.
10841 The first argument is a constant integer representing the size of the
10842 object, or -1 if it is variable sized. The second argument is a pointer
10848 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
10849 the return value, the referenced memory location is constant and
10852 '``llvm.invariant.end``' Intrinsic
10853 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10860 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
10865 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
10866 memory object are mutable.
10871 The first argument is the matching ``llvm.invariant.start`` intrinsic.
10872 The second argument is a constant integer representing the size of the
10873 object, or -1 if it is variable sized and the third argument is a
10874 pointer to the object.
10879 This intrinsic indicates that the memory is mutable again.
10884 This class of intrinsics is designed to be generic and has no specific
10887 '``llvm.var.annotation``' Intrinsic
10888 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10895 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
10900 The '``llvm.var.annotation``' intrinsic.
10905 The first argument is a pointer to a value, the second is a pointer to a
10906 global string, the third is a pointer to a global string which is the
10907 source file name, and the last argument is the line number.
10912 This intrinsic allows annotation of local variables with arbitrary
10913 strings. This can be useful for special purpose optimizations that want
10914 to look for these annotations. These have no other defined use; they are
10915 ignored by code generation and optimization.
10917 '``llvm.ptr.annotation.*``' Intrinsic
10918 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10923 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
10924 pointer to an integer of any width. *NOTE* you must specify an address space for
10925 the pointer. The identifier for the default address space is the integer
10930 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
10931 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
10932 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
10933 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
10934 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
10939 The '``llvm.ptr.annotation``' intrinsic.
10944 The first argument is a pointer to an integer value of arbitrary bitwidth
10945 (result of some expression), the second is a pointer to a global string, the
10946 third is a pointer to a global string which is the source file name, and the
10947 last argument is the line number. It returns the value of the first argument.
10952 This intrinsic allows annotation of a pointer to an integer with arbitrary
10953 strings. This can be useful for special purpose optimizations that want to look
10954 for these annotations. These have no other defined use; they are ignored by code
10955 generation and optimization.
10957 '``llvm.annotation.*``' Intrinsic
10958 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10963 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
10964 any integer bit width.
10968 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
10969 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
10970 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
10971 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
10972 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
10977 The '``llvm.annotation``' intrinsic.
10982 The first argument is an integer value (result of some expression), the
10983 second is a pointer to a global string, the third is a pointer to a
10984 global string which is the source file name, and the last argument is
10985 the line number. It returns the value of the first argument.
10990 This intrinsic allows annotations to be put on arbitrary expressions
10991 with arbitrary strings. This can be useful for special purpose
10992 optimizations that want to look for these annotations. These have no
10993 other defined use; they are ignored by code generation and optimization.
10995 '``llvm.trap``' Intrinsic
10996 ^^^^^^^^^^^^^^^^^^^^^^^^^
11003 declare void @llvm.trap() noreturn nounwind
11008 The '``llvm.trap``' intrinsic.
11018 This intrinsic is lowered to the target dependent trap instruction. If
11019 the target does not have a trap instruction, this intrinsic will be
11020 lowered to a call of the ``abort()`` function.
11022 '``llvm.debugtrap``' Intrinsic
11023 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11030 declare void @llvm.debugtrap() nounwind
11035 The '``llvm.debugtrap``' intrinsic.
11045 This intrinsic is lowered to code which is intended to cause an
11046 execution trap with the intention of requesting the attention of a
11049 '``llvm.stackprotector``' Intrinsic
11050 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11057 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
11062 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
11063 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
11064 is placed on the stack before local variables.
11069 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
11070 The first argument is the value loaded from the stack guard
11071 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
11072 enough space to hold the value of the guard.
11077 This intrinsic causes the prologue/epilogue inserter to force the position of
11078 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
11079 to ensure that if a local variable on the stack is overwritten, it will destroy
11080 the value of the guard. When the function exits, the guard on the stack is
11081 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
11082 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
11083 calling the ``__stack_chk_fail()`` function.
11085 '``llvm.stackprotectorcheck``' Intrinsic
11086 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11093 declare void @llvm.stackprotectorcheck(i8** <guard>)
11098 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
11099 created stack protector and if they are not equal calls the
11100 ``__stack_chk_fail()`` function.
11105 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
11106 the variable ``@__stack_chk_guard``.
11111 This intrinsic is provided to perform the stack protector check by comparing
11112 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
11113 values do not match call the ``__stack_chk_fail()`` function.
11115 The reason to provide this as an IR level intrinsic instead of implementing it
11116 via other IR operations is that in order to perform this operation at the IR
11117 level without an intrinsic, one would need to create additional basic blocks to
11118 handle the success/failure cases. This makes it difficult to stop the stack
11119 protector check from disrupting sibling tail calls in Codegen. With this
11120 intrinsic, we are able to generate the stack protector basic blocks late in
11121 codegen after the tail call decision has occurred.
11123 '``llvm.objectsize``' Intrinsic
11124 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11131 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
11132 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
11137 The ``llvm.objectsize`` intrinsic is designed to provide information to
11138 the optimizers to determine at compile time whether a) an operation
11139 (like memcpy) will overflow a buffer that corresponds to an object, or
11140 b) that a runtime check for overflow isn't necessary. An object in this
11141 context means an allocation of a specific class, structure, array, or
11147 The ``llvm.objectsize`` intrinsic takes two arguments. The first
11148 argument is a pointer to or into the ``object``. The second argument is
11149 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
11150 or -1 (if false) when the object size is unknown. The second argument
11151 only accepts constants.
11156 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
11157 the size of the object concerned. If the size cannot be determined at
11158 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
11159 on the ``min`` argument).
11161 '``llvm.expect``' Intrinsic
11162 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
11167 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
11172 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
11173 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
11174 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
11179 The ``llvm.expect`` intrinsic provides information about expected (the
11180 most probable) value of ``val``, which can be used by optimizers.
11185 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
11186 a value. The second argument is an expected value, this needs to be a
11187 constant value, variables are not allowed.
11192 This intrinsic is lowered to the ``val``.
11196 '``llvm.assume``' Intrinsic
11197 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11204 declare void @llvm.assume(i1 %cond)
11209 The ``llvm.assume`` allows the optimizer to assume that the provided
11210 condition is true. This information can then be used in simplifying other parts
11216 The condition which the optimizer may assume is always true.
11221 The intrinsic allows the optimizer to assume that the provided condition is
11222 always true whenever the control flow reaches the intrinsic call. No code is
11223 generated for this intrinsic, and instructions that contribute only to the
11224 provided condition are not used for code generation. If the condition is
11225 violated during execution, the behavior is undefined.
11227 Note that the optimizer might limit the transformations performed on values
11228 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
11229 only used to form the intrinsic's input argument. This might prove undesirable
11230 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
11231 sufficient overall improvement in code quality. For this reason,
11232 ``llvm.assume`` should not be used to document basic mathematical invariants
11233 that the optimizer can otherwise deduce or facts that are of little use to the
11238 '``llvm.bitset.test``' Intrinsic
11239 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11246 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
11252 The first argument is a pointer to be tested. The second argument is a
11253 metadata string containing the name of a :doc:`bitset <BitSets>`.
11258 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
11259 member of the given bitset.
11261 '``llvm.donothing``' Intrinsic
11262 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11269 declare void @llvm.donothing() nounwind readnone
11274 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
11275 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
11276 with an invoke instruction.
11286 This intrinsic does nothing, and it's removed by optimizers and ignored
11289 Stack Map Intrinsics
11290 --------------------
11292 LLVM provides experimental intrinsics to support runtime patching
11293 mechanisms commonly desired in dynamic language JITs. These intrinsics
11294 are described in :doc:`StackMaps`.