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.
497 For platforms without linker support of ELF TLS model, the -femulated-tls
498 flag can be used to generate GCC compatible emulated TLS code.
505 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
506 types <t_struct>`. Literal types are uniqued structurally, but identified types
507 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
508 to forward declare a type that is not yet available.
510 An example of a identified structure specification is:
514 %mytype = type { %mytype*, i32 }
516 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
517 literal types are uniqued in recent versions of LLVM.
524 Global variables define regions of memory allocated at compilation time
527 Global variable definitions must be initialized.
529 Global variables in other translation units can also be declared, in which
530 case they don't have an initializer.
532 Either global variable definitions or declarations may have an explicit section
533 to be placed in and may have an optional explicit alignment specified.
535 A variable may be defined as a global ``constant``, which indicates that
536 the contents of the variable will **never** be modified (enabling better
537 optimization, allowing the global data to be placed in the read-only
538 section of an executable, etc). Note that variables that need runtime
539 initialization cannot be marked ``constant`` as there is a store to the
542 LLVM explicitly allows *declarations* of global variables to be marked
543 constant, even if the final definition of the global is not. This
544 capability can be used to enable slightly better optimization of the
545 program, but requires the language definition to guarantee that
546 optimizations based on the 'constantness' are valid for the translation
547 units that do not include the definition.
549 As SSA values, global variables define pointer values that are in scope
550 (i.e. they dominate) all basic blocks in the program. Global variables
551 always define a pointer to their "content" type because they describe a
552 region of memory, and all memory objects in LLVM are accessed through
555 Global variables can be marked with ``unnamed_addr`` which indicates
556 that the address is not significant, only the content. Constants marked
557 like this can be merged with other constants if they have the same
558 initializer. Note that a constant with significant address *can* be
559 merged with a ``unnamed_addr`` constant, the result being a constant
560 whose address is significant.
562 A global variable may be declared to reside in a target-specific
563 numbered address space. For targets that support them, address spaces
564 may affect how optimizations are performed and/or what target
565 instructions are used to access the variable. The default address space
566 is zero. The address space qualifier must precede any other attributes.
568 LLVM allows an explicit section to be specified for globals. If the
569 target supports it, it will emit globals to the section specified.
570 Additionally, the global can placed in a comdat if the target has the necessary
573 By default, global initializers are optimized by assuming that global
574 variables defined within the module are not modified from their
575 initial values before the start of the global initializer. This is
576 true even for variables potentially accessible from outside the
577 module, including those with external linkage or appearing in
578 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
579 by marking the variable with ``externally_initialized``.
581 An explicit alignment may be specified for a global, which must be a
582 power of 2. If not present, or if the alignment is set to zero, the
583 alignment of the global is set by the target to whatever it feels
584 convenient. If an explicit alignment is specified, the global is forced
585 to have exactly that alignment. Targets and optimizers are not allowed
586 to over-align the global if the global has an assigned section. In this
587 case, the extra alignment could be observable: for example, code could
588 assume that the globals are densely packed in their section and try to
589 iterate over them as an array, alignment padding would break this
590 iteration. The maximum alignment is ``1 << 29``.
592 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
594 Variables and aliases can have a
595 :ref:`Thread Local Storage Model <tls_model>`.
599 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
600 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
601 <global | constant> <Type> [<InitializerConstant>]
602 [, section "name"] [, comdat [($name)]]
603 [, align <Alignment>]
605 For example, the following defines a global in a numbered address space
606 with an initializer, section, and alignment:
610 @G = addrspace(5) constant float 1.0, section "foo", align 4
612 The following example just declares a global variable
616 @G = external global i32
618 The following example defines a thread-local global with the
619 ``initialexec`` TLS model:
623 @G = thread_local(initialexec) global i32 0, align 4
625 .. _functionstructure:
630 LLVM function definitions consist of the "``define``" keyword, an
631 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
632 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
633 an optional :ref:`calling convention <callingconv>`,
634 an optional ``unnamed_addr`` attribute, a return type, an optional
635 :ref:`parameter attribute <paramattrs>` for the return type, a function
636 name, a (possibly empty) argument list (each with optional :ref:`parameter
637 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
638 an optional section, an optional alignment,
639 an optional :ref:`comdat <langref_comdats>`,
640 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
641 an optional :ref:`prologue <prologuedata>`,
642 an optional :ref:`personality <personalityfn>`,
643 an opening curly brace, a list of basic blocks, and a closing curly brace.
645 LLVM function declarations consist of the "``declare``" keyword, an
646 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
647 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
648 an optional :ref:`calling convention <callingconv>`,
649 an optional ``unnamed_addr`` attribute, a return type, an optional
650 :ref:`parameter attribute <paramattrs>` for the return type, a function
651 name, a possibly empty list of arguments, an optional alignment, an optional
652 :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
653 and an optional :ref:`prologue <prologuedata>`.
655 A function definition contains a list of basic blocks, forming the CFG (Control
656 Flow Graph) for the function. Each basic block may optionally start with a label
657 (giving the basic block a symbol table entry), contains a list of instructions,
658 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
659 function return). If an explicit label is not provided, a block is assigned an
660 implicit numbered label, using the next value from the same counter as used for
661 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
662 entry block does not have an explicit label, it will be assigned label "%0",
663 then the first unnamed temporary in that block will be "%1", etc.
665 The first basic block in a function is special in two ways: it is
666 immediately executed on entrance to the function, and it is not allowed
667 to have predecessor basic blocks (i.e. there can not be any branches to
668 the entry block of a function). Because the block can have no
669 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
671 LLVM allows an explicit section to be specified for functions. If the
672 target supports it, it will emit functions to the section specified.
673 Additionally, the function can be placed in a COMDAT.
675 An explicit alignment may be specified for a function. If not present,
676 or if the alignment is set to zero, the alignment of the function is set
677 by the target to whatever it feels convenient. If an explicit alignment
678 is specified, the function is forced to have at least that much
679 alignment. All alignments must be a power of 2.
681 If the ``unnamed_addr`` attribute is given, the address is known to not
682 be significant and two identical functions can be merged.
686 define [linkage] [visibility] [DLLStorageClass]
688 <ResultType> @<FunctionName> ([argument list])
689 [unnamed_addr] [fn Attrs] [section "name"] [comdat [($name)]]
690 [align N] [gc] [prefix Constant] [prologue Constant]
691 [personality Constant] { ... }
693 The argument list is a comma seperated sequence of arguments where each
694 argument is of the following form
698 <type> [parameter Attrs] [name]
706 Aliases, unlike function or variables, don't create any new data. They
707 are just a new symbol and metadata for an existing position.
709 Aliases have a name and an aliasee that is either a global value or a
712 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
713 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
714 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
718 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy> @<Aliasee>
720 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
721 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
722 might not correctly handle dropping a weak symbol that is aliased.
724 Aliases that are not ``unnamed_addr`` are guaranteed to have the same address as
725 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
728 Since aliases are only a second name, some restrictions apply, of which
729 some can only be checked when producing an object file:
731 * The expression defining the aliasee must be computable at assembly
732 time. Since it is just a name, no relocations can be used.
734 * No alias in the expression can be weak as the possibility of the
735 intermediate alias being overridden cannot be represented in an
738 * No global value in the expression can be a declaration, since that
739 would require a relocation, which is not possible.
746 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
748 Comdats have a name which represents the COMDAT key. All global objects that
749 specify this key will only end up in the final object file if the linker chooses
750 that key over some other key. Aliases are placed in the same COMDAT that their
751 aliasee computes to, if any.
753 Comdats have a selection kind to provide input on how the linker should
754 choose between keys in two different object files.
758 $<Name> = comdat SelectionKind
760 The selection kind must be one of the following:
763 The linker may choose any COMDAT key, the choice is arbitrary.
765 The linker may choose any COMDAT key but the sections must contain the
768 The linker will choose the section containing the largest COMDAT key.
770 The linker requires that only section with this COMDAT key exist.
772 The linker may choose any COMDAT key but the sections must contain the
775 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
776 ``any`` as a selection kind.
778 Here is an example of a COMDAT group where a function will only be selected if
779 the COMDAT key's section is the largest:
783 $foo = comdat largest
784 @foo = global i32 2, comdat($foo)
786 define void @bar() comdat($foo) {
790 As a syntactic sugar the ``$name`` can be omitted if the name is the same as
796 @foo = global i32 2, comdat
799 In a COFF object file, this will create a COMDAT section with selection kind
800 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
801 and another COMDAT section with selection kind
802 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
803 section and contains the contents of the ``@bar`` symbol.
805 There are some restrictions on the properties of the global object.
806 It, or an alias to it, must have the same name as the COMDAT group when
808 The contents and size of this object may be used during link-time to determine
809 which COMDAT groups get selected depending on the selection kind.
810 Because the name of the object must match the name of the COMDAT group, the
811 linkage of the global object must not be local; local symbols can get renamed
812 if a collision occurs in the symbol table.
814 The combined use of COMDATS and section attributes may yield surprising results.
821 @g1 = global i32 42, section "sec", comdat($foo)
822 @g2 = global i32 42, section "sec", comdat($bar)
824 From the object file perspective, this requires the creation of two sections
825 with the same name. This is necessary because both globals belong to different
826 COMDAT groups and COMDATs, at the object file level, are represented by
829 Note that certain IR constructs like global variables and functions may
830 create COMDATs in the object file in addition to any which are specified using
831 COMDAT IR. This arises when the code generator is configured to emit globals
832 in individual sections (e.g. when `-data-sections` or `-function-sections`
833 is supplied to `llc`).
835 .. _namedmetadatastructure:
840 Named metadata is a collection of metadata. :ref:`Metadata
841 nodes <metadata>` (but not metadata strings) are the only valid
842 operands for a named metadata.
844 #. Named metadata are represented as a string of characters with the
845 metadata prefix. The rules for metadata names are the same as for
846 identifiers, but quoted names are not allowed. ``"\xx"`` type escapes
847 are still valid, which allows any character to be part of a name.
851 ; Some unnamed metadata nodes, which are referenced by the named metadata.
856 !name = !{!0, !1, !2}
863 The return type and each parameter of a function type may have a set of
864 *parameter attributes* associated with them. Parameter attributes are
865 used to communicate additional information about the result or
866 parameters of a function. Parameter attributes are considered to be part
867 of the function, not of the function type, so functions with different
868 parameter attributes can have the same function type.
870 Parameter attributes are simple keywords that follow the type specified.
871 If multiple parameter attributes are needed, they are space separated.
876 declare i32 @printf(i8* noalias nocapture, ...)
877 declare i32 @atoi(i8 zeroext)
878 declare signext i8 @returns_signed_char()
880 Note that any attributes for the function result (``nounwind``,
881 ``readonly``) come immediately after the argument list.
883 Currently, only the following parameter attributes are defined:
886 This indicates to the code generator that the parameter or return
887 value should be zero-extended to the extent required by the target's
888 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
889 the caller (for a parameter) or the callee (for a return value).
891 This indicates to the code generator that the parameter or return
892 value should be sign-extended to the extent required by the target's
893 ABI (which is usually 32-bits) by the caller (for a parameter) or
894 the callee (for a return value).
896 This indicates that this parameter or return value should be treated
897 in a special target-dependent fashion during while emitting code for
898 a function call or return (usually, by putting it in a register as
899 opposed to memory, though some targets use it to distinguish between
900 two different kinds of registers). Use of this attribute is
903 This indicates that the pointer parameter should really be passed by
904 value to the function. The attribute implies that a hidden copy of
905 the pointee is made between the caller and the callee, so the callee
906 is unable to modify the value in the caller. This attribute is only
907 valid on LLVM pointer arguments. It is generally used to pass
908 structs and arrays by value, but is also valid on pointers to
909 scalars. The copy is considered to belong to the caller not the
910 callee (for example, ``readonly`` functions should not write to
911 ``byval`` parameters). This is not a valid attribute for return
914 The byval attribute also supports specifying an alignment with the
915 align attribute. It indicates the alignment of the stack slot to
916 form and the known alignment of the pointer specified to the call
917 site. If the alignment is not specified, then the code generator
918 makes a target-specific assumption.
924 The ``inalloca`` argument attribute allows the caller to take the
925 address of outgoing stack arguments. An ``inalloca`` argument must
926 be a pointer to stack memory produced by an ``alloca`` instruction.
927 The alloca, or argument allocation, must also be tagged with the
928 inalloca keyword. Only the last argument may have the ``inalloca``
929 attribute, and that argument is guaranteed to be passed in memory.
931 An argument allocation may be used by a call at most once because
932 the call may deallocate it. The ``inalloca`` attribute cannot be
933 used in conjunction with other attributes that affect argument
934 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
935 ``inalloca`` attribute also disables LLVM's implicit lowering of
936 large aggregate return values, which means that frontend authors
937 must lower them with ``sret`` pointers.
939 When the call site is reached, the argument allocation must have
940 been the most recent stack allocation that is still live, or the
941 results are undefined. It is possible to allocate additional stack
942 space after an argument allocation and before its call site, but it
943 must be cleared off with :ref:`llvm.stackrestore
946 See :doc:`InAlloca` for more information on how to use this
950 This indicates that the pointer parameter specifies the address of a
951 structure that is the return value of the function in the source
952 program. This pointer must be guaranteed by the caller to be valid:
953 loads and stores to the structure may be assumed by the callee
954 not to trap and to be properly aligned. This may only be applied to
955 the first parameter. This is not a valid attribute for return
959 This indicates that the pointer value may be assumed by the optimizer to
960 have the specified alignment.
962 Note that this attribute has additional semantics when combined with the
968 This indicates that objects accessed via pointer values
969 :ref:`based <pointeraliasing>` on the argument or return value are not also
970 accessed, during the execution of the function, via pointer values not
971 *based* on the argument or return value. The attribute on a return value
972 also has additional semantics described below. The caller shares the
973 responsibility with the callee for ensuring that these requirements are met.
974 For further details, please see the discussion of the NoAlias response in
975 :ref:`alias analysis <Must, May, or No>`.
977 Note that this definition of ``noalias`` is intentionally similar
978 to the definition of ``restrict`` in C99 for function arguments.
980 For function return values, C99's ``restrict`` is not meaningful,
981 while LLVM's ``noalias`` is. Furthermore, the semantics of the ``noalias``
982 attribute on return values are stronger than the semantics of the attribute
983 when used on function arguments. On function return values, the ``noalias``
984 attribute indicates that the function acts like a system memory allocation
985 function, returning a pointer to allocated storage disjoint from the
986 storage for any other object accessible to the caller.
989 This indicates that the callee does not make any copies of the
990 pointer that outlive the callee itself. This is not a valid
991 attribute for return values.
996 This indicates that the pointer parameter can be excised using the
997 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
998 attribute for return values and can only be applied to one parameter.
1001 This indicates that the function always returns the argument as its return
1002 value. This is an optimization hint to the code generator when generating
1003 the caller, allowing tail call optimization and omission of register saves
1004 and restores in some cases; it is not checked or enforced when generating
1005 the callee. The parameter and the function return type must be valid
1006 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
1007 valid attribute for return values and can only be applied to one parameter.
1010 This indicates that the parameter or return pointer is not null. This
1011 attribute may only be applied to pointer typed parameters. This is not
1012 checked or enforced by LLVM, the caller must ensure that the pointer
1013 passed in is non-null, or the callee must ensure that the returned pointer
1016 ``dereferenceable(<n>)``
1017 This indicates that the parameter or return pointer is dereferenceable. This
1018 attribute may only be applied to pointer typed parameters. A pointer that
1019 is dereferenceable can be loaded from speculatively without a risk of
1020 trapping. The number of bytes known to be dereferenceable must be provided
1021 in parentheses. It is legal for the number of bytes to be less than the
1022 size of the pointee type. The ``nonnull`` attribute does not imply
1023 dereferenceability (consider a pointer to one element past the end of an
1024 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
1025 ``addrspace(0)`` (which is the default address space).
1027 ``dereferenceable_or_null(<n>)``
1028 This indicates that the parameter or return value isn't both
1029 non-null and non-dereferenceable (up to ``<n>`` bytes) at the same
1030 time. All non-null pointers tagged with
1031 ``dereferenceable_or_null(<n>)`` are ``dereferenceable(<n>)``.
1032 For address space 0 ``dereferenceable_or_null(<n>)`` implies that
1033 a pointer is exactly one of ``dereferenceable(<n>)`` or ``null``,
1034 and in other address spaces ``dereferenceable_or_null(<n>)``
1035 implies that a pointer is at least one of ``dereferenceable(<n>)``
1036 or ``null`` (i.e. it may be both ``null`` and
1037 ``dereferenceable(<n>)``). This attribute may only be applied to
1038 pointer typed parameters.
1042 Garbage Collector Strategy Names
1043 --------------------------------
1045 Each function may specify a garbage collector strategy name, which is simply a
1048 .. code-block:: llvm
1050 define void @f() gc "name" { ... }
1052 The supported values of *name* includes those :ref:`built in to LLVM
1053 <builtin-gc-strategies>` and any provided by loaded plugins. Specifying a GC
1054 strategy will cause the compiler to alter its output in order to support the
1055 named garbage collection algorithm. Note that LLVM itself does not contain a
1056 garbage collector, this functionality is restricted to generating machine code
1057 which can interoperate with a collector provided externally.
1064 Prefix data is data associated with a function which the code
1065 generator will emit immediately before the function's entrypoint.
1066 The purpose of this feature is to allow frontends to associate
1067 language-specific runtime metadata with specific functions and make it
1068 available through the function pointer while still allowing the
1069 function pointer to be called.
1071 To access the data for a given function, a program may bitcast the
1072 function pointer to a pointer to the constant's type and dereference
1073 index -1. This implies that the IR symbol points just past the end of
1074 the prefix data. For instance, take the example of a function annotated
1075 with a single ``i32``,
1077 .. code-block:: llvm
1079 define void @f() prefix i32 123 { ... }
1081 The prefix data can be referenced as,
1083 .. code-block:: llvm
1085 %0 = bitcast void* () @f to i32*
1086 %a = getelementptr inbounds i32, i32* %0, i32 -1
1087 %b = load i32, i32* %a
1089 Prefix data is laid out as if it were an initializer for a global variable
1090 of the prefix data's type. The function will be placed such that the
1091 beginning of the prefix data is aligned. This means that if the size
1092 of the prefix data is not a multiple of the alignment size, the
1093 function's entrypoint will not be aligned. If alignment of the
1094 function's entrypoint is desired, padding must be added to the prefix
1097 A function may have prefix data but no body. This has similar semantics
1098 to the ``available_externally`` linkage in that the data may be used by the
1099 optimizers but will not be emitted in the object file.
1106 The ``prologue`` attribute allows arbitrary code (encoded as bytes) to
1107 be inserted prior to the function body. This can be used for enabling
1108 function hot-patching and instrumentation.
1110 To maintain the semantics of ordinary function calls, the prologue data must
1111 have a particular format. Specifically, it must begin with a sequence of
1112 bytes which decode to a sequence of machine instructions, valid for the
1113 module's target, which transfer control to the point immediately succeeding
1114 the prologue data, without performing any other visible action. This allows
1115 the inliner and other passes to reason about the semantics of the function
1116 definition without needing to reason about the prologue data. Obviously this
1117 makes the format of the prologue data highly target dependent.
1119 A trivial example of valid prologue data for the x86 architecture is ``i8 144``,
1120 which encodes the ``nop`` instruction:
1122 .. code-block:: llvm
1124 define void @f() prologue i8 144 { ... }
1126 Generally prologue data can be formed by encoding a relative branch instruction
1127 which skips the metadata, as in this example of valid prologue data for the
1128 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1130 .. code-block:: llvm
1132 %0 = type <{ i8, i8, i8* }>
1134 define void @f() prologue %0 <{ i8 235, i8 8, i8* @md}> { ... }
1136 A function may have prologue data but no body. This has similar semantics
1137 to the ``available_externally`` linkage in that the data may be used by the
1138 optimizers but will not be emitted in the object file.
1142 Personality Function
1143 --------------------
1145 The ``personality`` attribute permits functions to specify what function
1146 to use for exception handling.
1153 Attribute groups are groups of attributes that are referenced by objects within
1154 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1155 functions will use the same set of attributes. In the degenerative case of a
1156 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1157 group will capture the important command line flags used to build that file.
1159 An attribute group is a module-level object. To use an attribute group, an
1160 object references the attribute group's ID (e.g. ``#37``). An object may refer
1161 to more than one attribute group. In that situation, the attributes from the
1162 different groups are merged.
1164 Here is an example of attribute groups for a function that should always be
1165 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1167 .. code-block:: llvm
1169 ; Target-independent attributes:
1170 attributes #0 = { alwaysinline alignstack=4 }
1172 ; Target-dependent attributes:
1173 attributes #1 = { "no-sse" }
1175 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1176 define void @f() #0 #1 { ... }
1183 Function attributes are set to communicate additional information about
1184 a function. Function attributes are considered to be part of the
1185 function, not of the function type, so functions with different function
1186 attributes can have the same function type.
1188 Function attributes are simple keywords that follow the type specified.
1189 If multiple attributes are needed, they are space separated. For
1192 .. code-block:: llvm
1194 define void @f() noinline { ... }
1195 define void @f() alwaysinline { ... }
1196 define void @f() alwaysinline optsize { ... }
1197 define void @f() optsize { ... }
1200 This attribute indicates that, when emitting the prologue and
1201 epilogue, the backend should forcibly align the stack pointer.
1202 Specify the desired alignment, which must be a power of two, in
1205 This attribute indicates that the inliner should attempt to inline
1206 this function into callers whenever possible, ignoring any active
1207 inlining size threshold for this caller.
1209 This indicates that the callee function at a call site should be
1210 recognized as a built-in function, even though the function's declaration
1211 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1212 direct calls to functions that are declared with the ``nobuiltin``
1215 This attribute indicates that this function is rarely called. When
1216 computing edge weights, basic blocks post-dominated by a cold
1217 function call are also considered to be cold; and, thus, given low
1220 This attribute indicates that the callee is dependent on a convergent
1221 thread execution pattern under certain parallel execution models.
1222 Transformations that are execution model agnostic may only move or
1223 tranform this call if the final location is control equivalent to its
1224 original position in the program, where control equivalence is defined as
1225 A dominates B and B post-dominates A, or vice versa.
1227 This attribute indicates that the source code contained a hint that
1228 inlining this function is desirable (such as the "inline" keyword in
1229 C/C++). It is just a hint; it imposes no requirements on the
1232 This attribute indicates that the function should be added to a
1233 jump-instruction table at code-generation time, and that all address-taken
1234 references to this function should be replaced with a reference to the
1235 appropriate jump-instruction-table function pointer. Note that this creates
1236 a new pointer for the original function, which means that code that depends
1237 on function-pointer identity can break. So, any function annotated with
1238 ``jumptable`` must also be ``unnamed_addr``.
1240 This attribute suggests that optimization passes and code generator
1241 passes make choices that keep the code size of this function as small
1242 as possible and perform optimizations that may sacrifice runtime
1243 performance in order to minimize the size of the generated code.
1245 This attribute disables prologue / epilogue emission for the
1246 function. This can have very system-specific consequences.
1248 This indicates that the callee function at a call site is not recognized as
1249 a built-in function. LLVM will retain the original call and not replace it
1250 with equivalent code based on the semantics of the built-in function, unless
1251 the call site uses the ``builtin`` attribute. This is valid at call sites
1252 and on function declarations and definitions.
1254 This attribute indicates that calls to the function cannot be
1255 duplicated. A call to a ``noduplicate`` function may be moved
1256 within its parent function, but may not be duplicated within
1257 its parent function.
1259 A function containing a ``noduplicate`` call may still
1260 be an inlining candidate, provided that the call is not
1261 duplicated by inlining. That implies that the function has
1262 internal linkage and only has one call site, so the original
1263 call is dead after inlining.
1265 This attributes disables implicit floating point instructions.
1267 This attribute indicates that the inliner should never inline this
1268 function in any situation. This attribute may not be used together
1269 with the ``alwaysinline`` attribute.
1271 This attribute suppresses lazy symbol binding for the function. This
1272 may make calls to the function faster, at the cost of extra program
1273 startup time if the function is not called during program startup.
1275 This attribute indicates that the code generator should not use a
1276 red zone, even if the target-specific ABI normally permits it.
1278 This function attribute indicates that the function never returns
1279 normally. This produces undefined behavior at runtime if the
1280 function ever does dynamically return.
1282 This function attribute indicates that the function never raises an
1283 exception. If the function does raise an exception, its runtime
1284 behavior is undefined. However, functions marked nounwind may still
1285 trap or generate asynchronous exceptions. Exception handling schemes
1286 that are recognized by LLVM to handle asynchronous exceptions, such
1287 as SEH, will still provide their implementation defined semantics.
1289 This function attribute indicates that the function is not optimized
1290 by any optimization or code generator passes with the
1291 exception of interprocedural optimization passes.
1292 This attribute cannot be used together with the ``alwaysinline``
1293 attribute; this attribute is also incompatible
1294 with the ``minsize`` attribute and the ``optsize`` attribute.
1296 This attribute requires the ``noinline`` attribute to be specified on
1297 the function as well, so the function is never inlined into any caller.
1298 Only functions with the ``alwaysinline`` attribute are valid
1299 candidates for inlining into the body of this function.
1301 This attribute suggests that optimization passes and code generator
1302 passes make choices that keep the code size of this function low,
1303 and otherwise do optimizations specifically to reduce code size as
1304 long as they do not significantly impact runtime performance.
1306 On a function, this attribute indicates that the function computes its
1307 result (or decides to unwind an exception) based strictly on its arguments,
1308 without dereferencing any pointer arguments or otherwise accessing
1309 any mutable state (e.g. memory, control registers, etc) visible to
1310 caller functions. It does not write through any pointer arguments
1311 (including ``byval`` arguments) and never changes any state visible
1312 to callers. This means that it cannot unwind exceptions by calling
1313 the ``C++`` exception throwing methods.
1315 On an argument, this attribute indicates that the function does not
1316 dereference that pointer argument, even though it may read or write the
1317 memory that the pointer points to if accessed through other pointers.
1319 On a function, this attribute indicates that the function does not write
1320 through any pointer arguments (including ``byval`` arguments) or otherwise
1321 modify any state (e.g. memory, control registers, etc) visible to
1322 caller functions. It may dereference pointer arguments and read
1323 state that may be set in the caller. A readonly function always
1324 returns the same value (or unwinds an exception identically) when
1325 called with the same set of arguments and global state. It cannot
1326 unwind an exception by calling the ``C++`` exception throwing
1329 On an argument, this attribute indicates that the function does not write
1330 through this pointer argument, even though it may write to the memory that
1331 the pointer points to.
1333 This attribute indicates that the only memory accesses inside function are
1334 loads and stores from objects pointed to by its pointer-typed arguments,
1335 with arbitrary offsets. Or in other words, all memory operations in the
1336 function can refer to memory only using pointers based on its function
1338 Note that ``argmemonly`` can be used together with ``readonly`` attribute
1339 in order to specify that function reads only from its arguments.
1341 This attribute indicates that this function can return twice. The C
1342 ``setjmp`` is an example of such a function. The compiler disables
1343 some optimizations (like tail calls) in the caller of these
1346 This attribute indicates that
1347 `SafeStack <http://clang.llvm.org/docs/SafeStack.html>`_
1348 protection is enabled for this function.
1350 If a function that has a ``safestack`` attribute is inlined into a
1351 function that doesn't have a ``safestack`` attribute or which has an
1352 ``ssp``, ``sspstrong`` or ``sspreq`` attribute, then the resulting
1353 function will have a ``safestack`` attribute.
1354 ``sanitize_address``
1355 This attribute indicates that AddressSanitizer checks
1356 (dynamic address safety analysis) are enabled for this function.
1358 This attribute indicates that MemorySanitizer checks (dynamic detection
1359 of accesses to uninitialized memory) are enabled for this function.
1361 This attribute indicates that ThreadSanitizer checks
1362 (dynamic thread safety analysis) are enabled for this function.
1364 This attribute indicates that the function should emit a stack
1365 smashing protector. It is in the form of a "canary" --- a random value
1366 placed on the stack before the local variables that's checked upon
1367 return from the function to see if it has been overwritten. A
1368 heuristic is used to determine if a function needs stack protectors
1369 or not. The heuristic used will enable protectors for functions with:
1371 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1372 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1373 - Calls to alloca() with variable sizes or constant sizes greater than
1374 ``ssp-buffer-size``.
1376 Variables that are identified as requiring a protector will be arranged
1377 on the stack such that they are adjacent to the stack protector guard.
1379 If a function that has an ``ssp`` attribute is inlined into a
1380 function that doesn't have an ``ssp`` attribute, then the resulting
1381 function will have an ``ssp`` attribute.
1383 This attribute indicates that the function should *always* emit a
1384 stack smashing protector. This overrides the ``ssp`` function
1387 Variables that are identified as requiring a protector will be arranged
1388 on the stack such that they are adjacent to the stack protector guard.
1389 The specific layout rules are:
1391 #. Large arrays and structures containing large arrays
1392 (``>= ssp-buffer-size``) are closest to the stack protector.
1393 #. Small arrays and structures containing small arrays
1394 (``< ssp-buffer-size``) are 2nd closest to the protector.
1395 #. Variables that have had their address taken are 3rd closest to the
1398 If a function that has an ``sspreq`` attribute is inlined into a
1399 function that doesn't have an ``sspreq`` attribute or which has an
1400 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1401 an ``sspreq`` attribute.
1403 This attribute indicates that the function should emit a stack smashing
1404 protector. This attribute causes a strong heuristic to be used when
1405 determining if a function needs stack protectors. The strong heuristic
1406 will enable protectors for functions with:
1408 - Arrays of any size and type
1409 - Aggregates containing an array of any size and type.
1410 - Calls to alloca().
1411 - Local variables that have had their address taken.
1413 Variables that are identified as requiring a protector will be arranged
1414 on the stack such that they are adjacent to the stack protector guard.
1415 The specific layout rules are:
1417 #. Large arrays and structures containing large arrays
1418 (``>= ssp-buffer-size``) are closest to the stack protector.
1419 #. Small arrays and structures containing small arrays
1420 (``< ssp-buffer-size``) are 2nd closest to the protector.
1421 #. Variables that have had their address taken are 3rd closest to the
1424 This overrides the ``ssp`` function attribute.
1426 If a function that has an ``sspstrong`` attribute is inlined into a
1427 function that doesn't have an ``sspstrong`` attribute, then the
1428 resulting function will have an ``sspstrong`` attribute.
1430 This attribute indicates that the function will delegate to some other
1431 function with a tail call. The prototype of a thunk should not be used for
1432 optimization purposes. The caller is expected to cast the thunk prototype to
1433 match the thunk target prototype.
1435 This attribute indicates that the ABI being targeted requires that
1436 an unwind table entry be produce for this function even if we can
1437 show that no exceptions passes by it. This is normally the case for
1438 the ELF x86-64 abi, but it can be disabled for some compilation
1443 Module-Level Inline Assembly
1444 ----------------------------
1446 Modules may contain "module-level inline asm" blocks, which corresponds
1447 to the GCC "file scope inline asm" blocks. These blocks are internally
1448 concatenated by LLVM and treated as a single unit, but may be separated
1449 in the ``.ll`` file if desired. The syntax is very simple:
1451 .. code-block:: llvm
1453 module asm "inline asm code goes here"
1454 module asm "more can go here"
1456 The strings can contain any character by escaping non-printable
1457 characters. The escape sequence used is simply "\\xx" where "xx" is the
1458 two digit hex code for the number.
1460 Note that the assembly string *must* be parseable by LLVM's integrated assembler
1461 (unless it is disabled), even when emitting a ``.s`` file.
1463 .. _langref_datalayout:
1468 A module may specify a target specific data layout string that specifies
1469 how data is to be laid out in memory. The syntax for the data layout is
1472 .. code-block:: llvm
1474 target datalayout = "layout specification"
1476 The *layout specification* consists of a list of specifications
1477 separated by the minus sign character ('-'). Each specification starts
1478 with a letter and may include other information after the letter to
1479 define some aspect of the data layout. The specifications accepted are
1483 Specifies that the target lays out data in big-endian form. That is,
1484 the bits with the most significance have the lowest address
1487 Specifies that the target lays out data in little-endian form. That
1488 is, the bits with the least significance have the lowest address
1491 Specifies the natural alignment of the stack in bits. Alignment
1492 promotion of stack variables is limited to the natural stack
1493 alignment to avoid dynamic stack realignment. The stack alignment
1494 must be a multiple of 8-bits. If omitted, the natural stack
1495 alignment defaults to "unspecified", which does not prevent any
1496 alignment promotions.
1497 ``p[n]:<size>:<abi>:<pref>``
1498 This specifies the *size* of a pointer and its ``<abi>`` and
1499 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1500 bits. The address space, ``n`` is optional, and if not specified,
1501 denotes the default address space 0. The value of ``n`` must be
1502 in the range [1,2^23).
1503 ``i<size>:<abi>:<pref>``
1504 This specifies the alignment for an integer type of a given bit
1505 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1506 ``v<size>:<abi>:<pref>``
1507 This specifies the alignment for a vector type of a given bit
1509 ``f<size>:<abi>:<pref>``
1510 This specifies the alignment for a floating point type of a given bit
1511 ``<size>``. Only values of ``<size>`` that are supported by the target
1512 will work. 32 (float) and 64 (double) are supported on all targets; 80
1513 or 128 (different flavors of long double) are also supported on some
1516 This specifies the alignment for an object of aggregate type.
1518 If present, specifies that llvm names are mangled in the output. The
1521 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1522 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1523 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1524 symbols get a ``_`` prefix.
1525 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1526 functions also get a suffix based on the frame size.
1527 ``n<size1>:<size2>:<size3>...``
1528 This specifies a set of native integer widths for the target CPU in
1529 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1530 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1531 this set are considered to support most general arithmetic operations
1534 On every specification that takes a ``<abi>:<pref>``, specifying the
1535 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1536 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1538 When constructing the data layout for a given target, LLVM starts with a
1539 default set of specifications which are then (possibly) overridden by
1540 the specifications in the ``datalayout`` keyword. The default
1541 specifications are given in this list:
1543 - ``E`` - big endian
1544 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1545 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1546 same as the default address space.
1547 - ``S0`` - natural stack alignment is unspecified
1548 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1549 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1550 - ``i16:16:16`` - i16 is 16-bit aligned
1551 - ``i32:32:32`` - i32 is 32-bit aligned
1552 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1553 alignment of 64-bits
1554 - ``f16:16:16`` - half is 16-bit aligned
1555 - ``f32:32:32`` - float is 32-bit aligned
1556 - ``f64:64:64`` - double is 64-bit aligned
1557 - ``f128:128:128`` - quad is 128-bit aligned
1558 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1559 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1560 - ``a:0:64`` - aggregates are 64-bit aligned
1562 When LLVM is determining the alignment for a given type, it uses the
1565 #. If the type sought is an exact match for one of the specifications,
1566 that specification is used.
1567 #. If no match is found, and the type sought is an integer type, then
1568 the smallest integer type that is larger than the bitwidth of the
1569 sought type is used. If none of the specifications are larger than
1570 the bitwidth then the largest integer type is used. For example,
1571 given the default specifications above, the i7 type will use the
1572 alignment of i8 (next largest) while both i65 and i256 will use the
1573 alignment of i64 (largest specified).
1574 #. If no match is found, and the type sought is a vector type, then the
1575 largest vector type that is smaller than the sought vector type will
1576 be used as a fall back. This happens because <128 x double> can be
1577 implemented in terms of 64 <2 x double>, for example.
1579 The function of the data layout string may not be what you expect.
1580 Notably, this is not a specification from the frontend of what alignment
1581 the code generator should use.
1583 Instead, if specified, the target data layout is required to match what
1584 the ultimate *code generator* expects. This string is used by the
1585 mid-level optimizers to improve code, and this only works if it matches
1586 what the ultimate code generator uses. There is no way to generate IR
1587 that does not embed this target-specific detail into the IR. If you
1588 don't specify the string, the default specifications will be used to
1589 generate a Data Layout and the optimization phases will operate
1590 accordingly and introduce target specificity into the IR with respect to
1591 these default specifications.
1598 A module may specify a target triple string that describes the target
1599 host. The syntax for the target triple is simply:
1601 .. code-block:: llvm
1603 target triple = "x86_64-apple-macosx10.7.0"
1605 The *target triple* string consists of a series of identifiers delimited
1606 by the minus sign character ('-'). The canonical forms are:
1610 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1611 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1613 This information is passed along to the backend so that it generates
1614 code for the proper architecture. It's possible to override this on the
1615 command line with the ``-mtriple`` command line option.
1617 .. _pointeraliasing:
1619 Pointer Aliasing Rules
1620 ----------------------
1622 Any memory access must be done through a pointer value associated with
1623 an address range of the memory access, otherwise the behavior is
1624 undefined. Pointer values are associated with address ranges according
1625 to the following rules:
1627 - A pointer value is associated with the addresses associated with any
1628 value it is *based* on.
1629 - An address of a global variable is associated with the address range
1630 of the variable's storage.
1631 - The result value of an allocation instruction is associated with the
1632 address range of the allocated storage.
1633 - A null pointer in the default address-space is associated with no
1635 - An integer constant other than zero or a pointer value returned from
1636 a function not defined within LLVM may be associated with address
1637 ranges allocated through mechanisms other than those provided by
1638 LLVM. Such ranges shall not overlap with any ranges of addresses
1639 allocated by mechanisms provided by LLVM.
1641 A pointer value is *based* on another pointer value according to the
1644 - A pointer value formed from a ``getelementptr`` operation is *based*
1645 on the first value operand of the ``getelementptr``.
1646 - The result value of a ``bitcast`` is *based* on the operand of the
1648 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1649 values that contribute (directly or indirectly) to the computation of
1650 the pointer's value.
1651 - The "*based* on" relationship is transitive.
1653 Note that this definition of *"based"* is intentionally similar to the
1654 definition of *"based"* in C99, though it is slightly weaker.
1656 LLVM IR does not associate types with memory. The result type of a
1657 ``load`` merely indicates the size and alignment of the memory from
1658 which to load, as well as the interpretation of the value. The first
1659 operand type of a ``store`` similarly only indicates the size and
1660 alignment of the store.
1662 Consequently, type-based alias analysis, aka TBAA, aka
1663 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1664 :ref:`Metadata <metadata>` may be used to encode additional information
1665 which specialized optimization passes may use to implement type-based
1670 Volatile Memory Accesses
1671 ------------------------
1673 Certain memory accesses, such as :ref:`load <i_load>`'s,
1674 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1675 marked ``volatile``. The optimizers must not change the number of
1676 volatile operations or change their order of execution relative to other
1677 volatile operations. The optimizers *may* change the order of volatile
1678 operations relative to non-volatile operations. This is not Java's
1679 "volatile" and has no cross-thread synchronization behavior.
1681 IR-level volatile loads and stores cannot safely be optimized into
1682 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1683 flagged volatile. Likewise, the backend should never split or merge
1684 target-legal volatile load/store instructions.
1686 .. admonition:: Rationale
1688 Platforms may rely on volatile loads and stores of natively supported
1689 data width to be executed as single instruction. For example, in C
1690 this holds for an l-value of volatile primitive type with native
1691 hardware support, but not necessarily for aggregate types. The
1692 frontend upholds these expectations, which are intentionally
1693 unspecified in the IR. The rules above ensure that IR transformation
1694 do not violate the frontend's contract with the language.
1698 Memory Model for Concurrent Operations
1699 --------------------------------------
1701 The LLVM IR does not define any way to start parallel threads of
1702 execution or to register signal handlers. Nonetheless, there are
1703 platform-specific ways to create them, and we define LLVM IR's behavior
1704 in their presence. This model is inspired by the C++0x memory model.
1706 For a more informal introduction to this model, see the :doc:`Atomics`.
1708 We define a *happens-before* partial order as the least partial order
1711 - Is a superset of single-thread program order, and
1712 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1713 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1714 techniques, like pthread locks, thread creation, thread joining,
1715 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1716 Constraints <ordering>`).
1718 Note that program order does not introduce *happens-before* edges
1719 between a thread and signals executing inside that thread.
1721 Every (defined) read operation (load instructions, memcpy, atomic
1722 loads/read-modify-writes, etc.) R reads a series of bytes written by
1723 (defined) write operations (store instructions, atomic
1724 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1725 section, initialized globals are considered to have a write of the
1726 initializer which is atomic and happens before any other read or write
1727 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1728 may see any write to the same byte, except:
1730 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1731 write\ :sub:`2` happens before R\ :sub:`byte`, then
1732 R\ :sub:`byte` does not see write\ :sub:`1`.
1733 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1734 R\ :sub:`byte` does not see write\ :sub:`3`.
1736 Given that definition, R\ :sub:`byte` is defined as follows:
1738 - If R is volatile, the result is target-dependent. (Volatile is
1739 supposed to give guarantees which can support ``sig_atomic_t`` in
1740 C/C++, and may be used for accesses to addresses that do not behave
1741 like normal memory. It does not generally provide cross-thread
1743 - Otherwise, if there is no write to the same byte that happens before
1744 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1745 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1746 R\ :sub:`byte` returns the value written by that write.
1747 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1748 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1749 Memory Ordering Constraints <ordering>` section for additional
1750 constraints on how the choice is made.
1751 - Otherwise R\ :sub:`byte` returns ``undef``.
1753 R returns the value composed of the series of bytes it read. This
1754 implies that some bytes within the value may be ``undef`` **without**
1755 the entire value being ``undef``. Note that this only defines the
1756 semantics of the operation; it doesn't mean that targets will emit more
1757 than one instruction to read the series of bytes.
1759 Note that in cases where none of the atomic intrinsics are used, this
1760 model places only one restriction on IR transformations on top of what
1761 is required for single-threaded execution: introducing a store to a byte
1762 which might not otherwise be stored is not allowed in general.
1763 (Specifically, in the case where another thread might write to and read
1764 from an address, introducing a store can change a load that may see
1765 exactly one write into a load that may see multiple writes.)
1769 Atomic Memory Ordering Constraints
1770 ----------------------------------
1772 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1773 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1774 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1775 ordering parameters that determine which other atomic instructions on
1776 the same address they *synchronize with*. These semantics are borrowed
1777 from Java and C++0x, but are somewhat more colloquial. If these
1778 descriptions aren't precise enough, check those specs (see spec
1779 references in the :doc:`atomics guide <Atomics>`).
1780 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1781 differently since they don't take an address. See that instruction's
1782 documentation for details.
1784 For a simpler introduction to the ordering constraints, see the
1788 The set of values that can be read is governed by the happens-before
1789 partial order. A value cannot be read unless some operation wrote
1790 it. This is intended to provide a guarantee strong enough to model
1791 Java's non-volatile shared variables. This ordering cannot be
1792 specified for read-modify-write operations; it is not strong enough
1793 to make them atomic in any interesting way.
1795 In addition to the guarantees of ``unordered``, there is a single
1796 total order for modifications by ``monotonic`` operations on each
1797 address. All modification orders must be compatible with the
1798 happens-before order. There is no guarantee that the modification
1799 orders can be combined to a global total order for the whole program
1800 (and this often will not be possible). The read in an atomic
1801 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1802 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1803 order immediately before the value it writes. If one atomic read
1804 happens before another atomic read of the same address, the later
1805 read must see the same value or a later value in the address's
1806 modification order. This disallows reordering of ``monotonic`` (or
1807 stronger) operations on the same address. If an address is written
1808 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1809 read that address repeatedly, the other threads must eventually see
1810 the write. This corresponds to the C++0x/C1x
1811 ``memory_order_relaxed``.
1813 In addition to the guarantees of ``monotonic``, a
1814 *synchronizes-with* edge may be formed with a ``release`` operation.
1815 This is intended to model C++'s ``memory_order_acquire``.
1817 In addition to the guarantees of ``monotonic``, if this operation
1818 writes a value which is subsequently read by an ``acquire``
1819 operation, it *synchronizes-with* that operation. (This isn't a
1820 complete description; see the C++0x definition of a release
1821 sequence.) This corresponds to the C++0x/C1x
1822 ``memory_order_release``.
1823 ``acq_rel`` (acquire+release)
1824 Acts as both an ``acquire`` and ``release`` operation on its
1825 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1826 ``seq_cst`` (sequentially consistent)
1827 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1828 operation that only reads, ``release`` for an operation that only
1829 writes), there is a global total order on all
1830 sequentially-consistent operations on all addresses, which is
1831 consistent with the *happens-before* partial order and with the
1832 modification orders of all the affected addresses. Each
1833 sequentially-consistent read sees the last preceding write to the
1834 same address in this global order. This corresponds to the C++0x/C1x
1835 ``memory_order_seq_cst`` and Java volatile.
1839 If an atomic operation is marked ``singlethread``, it only *synchronizes
1840 with* or participates in modification and seq\_cst total orderings with
1841 other operations running in the same thread (for example, in signal
1849 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1850 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1851 :ref:`frem <i_frem>`, :ref:`fcmp <i_fcmp>`) have the following flags that can
1852 be set to enable otherwise unsafe floating point operations
1855 No NaNs - Allow optimizations to assume the arguments and result are not
1856 NaN. Such optimizations are required to retain defined behavior over
1857 NaNs, but the value of the result is undefined.
1860 No Infs - Allow optimizations to assume the arguments and result are not
1861 +/-Inf. Such optimizations are required to retain defined behavior over
1862 +/-Inf, but the value of the result is undefined.
1865 No Signed Zeros - Allow optimizations to treat the sign of a zero
1866 argument or result as insignificant.
1869 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1870 argument rather than perform division.
1873 Fast - Allow algebraically equivalent transformations that may
1874 dramatically change results in floating point (e.g. reassociate). This
1875 flag implies all the others.
1879 Use-list Order Directives
1880 -------------------------
1882 Use-list directives encode the in-memory order of each use-list, allowing the
1883 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1884 indexes that are assigned to the referenced value's uses. The referenced
1885 value's use-list is immediately sorted by these indexes.
1887 Use-list directives may appear at function scope or global scope. They are not
1888 instructions, and have no effect on the semantics of the IR. When they're at
1889 function scope, they must appear after the terminator of the final basic block.
1891 If basic blocks have their address taken via ``blockaddress()`` expressions,
1892 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1899 uselistorder <ty> <value>, { <order-indexes> }
1900 uselistorder_bb @function, %block { <order-indexes> }
1906 define void @foo(i32 %arg1, i32 %arg2) {
1908 ; ... instructions ...
1910 ; ... instructions ...
1912 ; At function scope.
1913 uselistorder i32 %arg1, { 1, 0, 2 }
1914 uselistorder label %bb, { 1, 0 }
1918 uselistorder i32* @global, { 1, 2, 0 }
1919 uselistorder i32 7, { 1, 0 }
1920 uselistorder i32 (i32) @bar, { 1, 0 }
1921 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1928 The LLVM type system is one of the most important features of the
1929 intermediate representation. Being typed enables a number of
1930 optimizations to be performed on the intermediate representation
1931 directly, without having to do extra analyses on the side before the
1932 transformation. A strong type system makes it easier to read the
1933 generated code and enables novel analyses and transformations that are
1934 not feasible to perform on normal three address code representations.
1944 The void type does not represent any value and has no size.
1962 The function type can be thought of as a function signature. It consists of a
1963 return type and a list of formal parameter types. The return type of a function
1964 type is a void type or first class type --- except for :ref:`label <t_label>`
1965 and :ref:`metadata <t_metadata>` types.
1971 <returntype> (<parameter list>)
1973 ...where '``<parameter list>``' is a comma-separated list of type
1974 specifiers. Optionally, the parameter list may include a type ``...``, which
1975 indicates that the function takes a variable number of arguments. Variable
1976 argument functions can access their arguments with the :ref:`variable argument
1977 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1978 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1982 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1983 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1984 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1985 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1986 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1987 | ``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. |
1988 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1989 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1990 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1997 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1998 Values of these types are the only ones which can be produced by
2006 These are the types that are valid in registers from CodeGen's perspective.
2015 The integer type is a very simple type that simply specifies an
2016 arbitrary bit width for the integer type desired. Any bit width from 1
2017 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
2025 The number of bits the integer will occupy is specified by the ``N``
2031 +----------------+------------------------------------------------+
2032 | ``i1`` | a single-bit integer. |
2033 +----------------+------------------------------------------------+
2034 | ``i32`` | a 32-bit integer. |
2035 +----------------+------------------------------------------------+
2036 | ``i1942652`` | a really big integer of over 1 million bits. |
2037 +----------------+------------------------------------------------+
2041 Floating Point Types
2042 """"""""""""""""""""
2051 - 16-bit floating point value
2054 - 32-bit floating point value
2057 - 64-bit floating point value
2060 - 128-bit floating point value (112-bit mantissa)
2063 - 80-bit floating point value (X87)
2066 - 128-bit floating point value (two 64-bits)
2073 The x86_mmx type represents a value held in an MMX register on an x86
2074 machine. The operations allowed on it are quite limited: parameters and
2075 return values, load and store, and bitcast. User-specified MMX
2076 instructions are represented as intrinsic or asm calls with arguments
2077 and/or results of this type. There are no arrays, vectors or constants
2094 The pointer type is used to specify memory locations. Pointers are
2095 commonly used to reference objects in memory.
2097 Pointer types may have an optional address space attribute defining the
2098 numbered address space where the pointed-to object resides. The default
2099 address space is number zero. The semantics of non-zero address spaces
2100 are target-specific.
2102 Note that LLVM does not permit pointers to void (``void*``) nor does it
2103 permit pointers to labels (``label*``). Use ``i8*`` instead.
2113 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2114 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2115 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2116 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2117 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2118 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2119 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2128 A vector type is a simple derived type that represents a vector of
2129 elements. Vector types are used when multiple primitive data are
2130 operated in parallel using a single instruction (SIMD). A vector type
2131 requires a size (number of elements) and an underlying primitive data
2132 type. Vector types are considered :ref:`first class <t_firstclass>`.
2138 < <# elements> x <elementtype> >
2140 The number of elements is a constant integer value larger than 0;
2141 elementtype may be any integer, floating point or pointer type. Vectors
2142 of size zero are not allowed.
2146 +-------------------+--------------------------------------------------+
2147 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2148 +-------------------+--------------------------------------------------+
2149 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2150 +-------------------+--------------------------------------------------+
2151 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2152 +-------------------+--------------------------------------------------+
2153 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2154 +-------------------+--------------------------------------------------+
2163 The label type represents code labels.
2178 The metadata type represents embedded metadata. No derived types may be
2179 created from metadata except for :ref:`function <t_function>` arguments.
2192 Aggregate Types are a subset of derived types that can contain multiple
2193 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2194 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2204 The array type is a very simple derived type that arranges elements
2205 sequentially in memory. The array type requires a size (number of
2206 elements) and an underlying data type.
2212 [<# elements> x <elementtype>]
2214 The number of elements is a constant integer value; ``elementtype`` may
2215 be any type with a size.
2219 +------------------+--------------------------------------+
2220 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2221 +------------------+--------------------------------------+
2222 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2223 +------------------+--------------------------------------+
2224 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2225 +------------------+--------------------------------------+
2227 Here are some examples of multidimensional arrays:
2229 +-----------------------------+----------------------------------------------------------+
2230 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2231 +-----------------------------+----------------------------------------------------------+
2232 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2233 +-----------------------------+----------------------------------------------------------+
2234 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2235 +-----------------------------+----------------------------------------------------------+
2237 There is no restriction on indexing beyond the end of the array implied
2238 by a static type (though there are restrictions on indexing beyond the
2239 bounds of an allocated object in some cases). This means that
2240 single-dimension 'variable sized array' addressing can be implemented in
2241 LLVM with a zero length array type. An implementation of 'pascal style
2242 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2252 The structure type is used to represent a collection of data members
2253 together in memory. The elements of a structure may be any type that has
2256 Structures in memory are accessed using '``load``' and '``store``' by
2257 getting a pointer to a field with the '``getelementptr``' instruction.
2258 Structures in registers are accessed using the '``extractvalue``' and
2259 '``insertvalue``' instructions.
2261 Structures may optionally be "packed" structures, which indicate that
2262 the alignment of the struct is one byte, and that there is no padding
2263 between the elements. In non-packed structs, padding between field types
2264 is inserted as defined by the DataLayout string in the module, which is
2265 required to match what the underlying code generator expects.
2267 Structures can either be "literal" or "identified". A literal structure
2268 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2269 identified types are always defined at the top level with a name.
2270 Literal types are uniqued by their contents and can never be recursive
2271 or opaque since there is no way to write one. Identified types can be
2272 recursive, can be opaqued, and are never uniqued.
2278 %T1 = type { <type list> } ; Identified normal struct type
2279 %T2 = type <{ <type list> }> ; Identified packed struct type
2283 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2284 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2285 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2286 | ``{ 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``. |
2287 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2288 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2289 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2293 Opaque Structure Types
2294 """"""""""""""""""""""
2298 Opaque structure types are used to represent named structure types that
2299 do not have a body specified. This corresponds (for example) to the C
2300 notion of a forward declared structure.
2311 +--------------+-------------------+
2312 | ``opaque`` | An opaque type. |
2313 +--------------+-------------------+
2320 LLVM has several different basic types of constants. This section
2321 describes them all and their syntax.
2326 **Boolean constants**
2327 The two strings '``true``' and '``false``' are both valid constants
2329 **Integer constants**
2330 Standard integers (such as '4') are constants of the
2331 :ref:`integer <t_integer>` type. Negative numbers may be used with
2333 **Floating point constants**
2334 Floating point constants use standard decimal notation (e.g.
2335 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2336 hexadecimal notation (see below). The assembler requires the exact
2337 decimal value of a floating-point constant. For example, the
2338 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2339 decimal in binary. Floating point constants must have a :ref:`floating
2340 point <t_floating>` type.
2341 **Null pointer constants**
2342 The identifier '``null``' is recognized as a null pointer constant
2343 and must be of :ref:`pointer type <t_pointer>`.
2345 The one non-intuitive notation for constants is the hexadecimal form of
2346 floating point constants. For example, the form
2347 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2348 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2349 constants are required (and the only time that they are generated by the
2350 disassembler) is when a floating point constant must be emitted but it
2351 cannot be represented as a decimal floating point number in a reasonable
2352 number of digits. For example, NaN's, infinities, and other special
2353 values are represented in their IEEE hexadecimal format so that assembly
2354 and disassembly do not cause any bits to change in the constants.
2356 When using the hexadecimal form, constants of types half, float, and
2357 double are represented using the 16-digit form shown above (which
2358 matches the IEEE754 representation for double); half and float values
2359 must, however, be exactly representable as IEEE 754 half and single
2360 precision, respectively. Hexadecimal format is always used for long
2361 double, and there are three forms of long double. The 80-bit format used
2362 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2363 128-bit format used by PowerPC (two adjacent doubles) is represented by
2364 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2365 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2366 will only work if they match the long double format on your target.
2367 The IEEE 16-bit format (half precision) is represented by ``0xH``
2368 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2369 (sign bit at the left).
2371 There are no constants of type x86_mmx.
2373 .. _complexconstants:
2378 Complex constants are a (potentially recursive) combination of simple
2379 constants and smaller complex constants.
2381 **Structure constants**
2382 Structure constants are represented with notation similar to
2383 structure type definitions (a comma separated list of elements,
2384 surrounded by braces (``{}``)). For example:
2385 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2386 "``@G = external global i32``". Structure constants must have
2387 :ref:`structure type <t_struct>`, and the number and types of elements
2388 must match those specified by the type.
2390 Array constants are represented with notation similar to array type
2391 definitions (a comma separated list of elements, surrounded by
2392 square brackets (``[]``)). For example:
2393 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2394 :ref:`array type <t_array>`, and the number and types of elements must
2395 match those specified by the type. As a special case, character array
2396 constants may also be represented as a double-quoted string using the ``c``
2397 prefix. For example: "``c"Hello World\0A\00"``".
2398 **Vector constants**
2399 Vector constants are represented with notation similar to vector
2400 type definitions (a comma separated list of elements, surrounded by
2401 less-than/greater-than's (``<>``)). For example:
2402 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2403 must have :ref:`vector type <t_vector>`, and the number and types of
2404 elements must match those specified by the type.
2405 **Zero initialization**
2406 The string '``zeroinitializer``' can be used to zero initialize a
2407 value to zero of *any* type, including scalar and
2408 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2409 having to print large zero initializers (e.g. for large arrays) and
2410 is always exactly equivalent to using explicit zero initializers.
2412 A metadata node is a constant tuple without types. For example:
2413 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2414 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2415 Unlike other typed constants that are meant to be interpreted as part of
2416 the instruction stream, metadata is a place to attach additional
2417 information such as debug info.
2419 Global Variable and Function Addresses
2420 --------------------------------------
2422 The addresses of :ref:`global variables <globalvars>` and
2423 :ref:`functions <functionstructure>` are always implicitly valid
2424 (link-time) constants. These constants are explicitly referenced when
2425 the :ref:`identifier for the global <identifiers>` is used and always have
2426 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2429 .. code-block:: llvm
2433 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2440 The string '``undef``' can be used anywhere a constant is expected, and
2441 indicates that the user of the value may receive an unspecified
2442 bit-pattern. Undefined values may be of any type (other than '``label``'
2443 or '``void``') and be used anywhere a constant is permitted.
2445 Undefined values are useful because they indicate to the compiler that
2446 the program is well defined no matter what value is used. This gives the
2447 compiler more freedom to optimize. Here are some examples of
2448 (potentially surprising) transformations that are valid (in pseudo IR):
2450 .. code-block:: llvm
2460 This is safe because all of the output bits are affected by the undef
2461 bits. Any output bit can have a zero or one depending on the input bits.
2463 .. code-block:: llvm
2474 These logical operations have bits that are not always affected by the
2475 input. For example, if ``%X`` has a zero bit, then the output of the
2476 '``and``' operation will always be a zero for that bit, no matter what
2477 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2478 optimize or assume that the result of the '``and``' is '``undef``'.
2479 However, it is safe to assume that all bits of the '``undef``' could be
2480 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2481 all the bits of the '``undef``' operand to the '``or``' could be set,
2482 allowing the '``or``' to be folded to -1.
2484 .. code-block:: llvm
2486 %A = select undef, %X, %Y
2487 %B = select undef, 42, %Y
2488 %C = select %X, %Y, undef
2498 This set of examples shows that undefined '``select``' (and conditional
2499 branch) conditions can go *either way*, but they have to come from one
2500 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2501 both known to have a clear low bit, then ``%A`` would have to have a
2502 cleared low bit. However, in the ``%C`` example, the optimizer is
2503 allowed to assume that the '``undef``' operand could be the same as
2504 ``%Y``, allowing the whole '``select``' to be eliminated.
2506 .. code-block:: llvm
2508 %A = xor undef, undef
2525 This example points out that two '``undef``' operands are not
2526 necessarily the same. This can be surprising to people (and also matches
2527 C semantics) where they assume that "``X^X``" is always zero, even if
2528 ``X`` is undefined. This isn't true for a number of reasons, but the
2529 short answer is that an '``undef``' "variable" can arbitrarily change
2530 its value over its "live range". This is true because the variable
2531 doesn't actually *have a live range*. Instead, the value is logically
2532 read from arbitrary registers that happen to be around when needed, so
2533 the value is not necessarily consistent over time. In fact, ``%A`` and
2534 ``%C`` need to have the same semantics or the core LLVM "replace all
2535 uses with" concept would not hold.
2537 .. code-block:: llvm
2545 These examples show the crucial difference between an *undefined value*
2546 and *undefined behavior*. An undefined value (like '``undef``') is
2547 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2548 operation can be constant folded to '``undef``', because the '``undef``'
2549 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2550 However, in the second example, we can make a more aggressive
2551 assumption: because the ``undef`` is allowed to be an arbitrary value,
2552 we are allowed to assume that it could be zero. Since a divide by zero
2553 has *undefined behavior*, we are allowed to assume that the operation
2554 does not execute at all. This allows us to delete the divide and all
2555 code after it. Because the undefined operation "can't happen", the
2556 optimizer can assume that it occurs in dead code.
2558 .. code-block:: llvm
2560 a: store undef -> %X
2561 b: store %X -> undef
2566 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2567 value can be assumed to not have any effect; we can assume that the
2568 value is overwritten with bits that happen to match what was already
2569 there. However, a store *to* an undefined location could clobber
2570 arbitrary memory, therefore, it has undefined behavior.
2577 Poison values are similar to :ref:`undef values <undefvalues>`, however
2578 they also represent the fact that an instruction or constant expression
2579 that cannot evoke side effects has nevertheless detected a condition
2580 that results in undefined behavior.
2582 There is currently no way of representing a poison value in the IR; they
2583 only exist when produced by operations such as :ref:`add <i_add>` with
2586 Poison value behavior is defined in terms of value *dependence*:
2588 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2589 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2590 their dynamic predecessor basic block.
2591 - Function arguments depend on the corresponding actual argument values
2592 in the dynamic callers of their functions.
2593 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2594 instructions that dynamically transfer control back to them.
2595 - :ref:`Invoke <i_invoke>` instructions depend on the
2596 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2597 call instructions that dynamically transfer control back to them.
2598 - Non-volatile loads and stores depend on the most recent stores to all
2599 of the referenced memory addresses, following the order in the IR
2600 (including loads and stores implied by intrinsics such as
2601 :ref:`@llvm.memcpy <int_memcpy>`.)
2602 - An instruction with externally visible side effects depends on the
2603 most recent preceding instruction with externally visible side
2604 effects, following the order in the IR. (This includes :ref:`volatile
2605 operations <volatile>`.)
2606 - An instruction *control-depends* on a :ref:`terminator
2607 instruction <terminators>` if the terminator instruction has
2608 multiple successors and the instruction is always executed when
2609 control transfers to one of the successors, and may not be executed
2610 when control is transferred to another.
2611 - Additionally, an instruction also *control-depends* on a terminator
2612 instruction if the set of instructions it otherwise depends on would
2613 be different if the terminator had transferred control to a different
2615 - Dependence is transitive.
2617 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2618 with the additional effect that any instruction that has a *dependence*
2619 on a poison value has undefined behavior.
2621 Here are some examples:
2623 .. code-block:: llvm
2626 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2627 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2628 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
2629 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2631 store i32 %poison, i32* @g ; Poison value stored to memory.
2632 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
2634 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2636 %narrowaddr = bitcast i32* @g to i16*
2637 %wideaddr = bitcast i32* @g to i64*
2638 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2639 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
2641 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2642 br i1 %cmp, label %true, label %end ; Branch to either destination.
2645 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2646 ; it has undefined behavior.
2650 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2651 ; Both edges into this PHI are
2652 ; control-dependent on %cmp, so this
2653 ; always results in a poison value.
2655 store volatile i32 0, i32* @g ; This would depend on the store in %true
2656 ; if %cmp is true, or the store in %entry
2657 ; otherwise, so this is undefined behavior.
2659 br i1 %cmp, label %second_true, label %second_end
2660 ; The same branch again, but this time the
2661 ; true block doesn't have side effects.
2668 store volatile i32 0, i32* @g ; This time, the instruction always depends
2669 ; on the store in %end. Also, it is
2670 ; control-equivalent to %end, so this is
2671 ; well-defined (ignoring earlier undefined
2672 ; behavior in this example).
2676 Addresses of Basic Blocks
2677 -------------------------
2679 ``blockaddress(@function, %block)``
2681 The '``blockaddress``' constant computes the address of the specified
2682 basic block in the specified function, and always has an ``i8*`` type.
2683 Taking the address of the entry block is illegal.
2685 This value only has defined behavior when used as an operand to the
2686 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2687 against null. Pointer equality tests between labels addresses results in
2688 undefined behavior --- though, again, comparison against null is ok, and
2689 no label is equal to the null pointer. This may be passed around as an
2690 opaque pointer sized value as long as the bits are not inspected. This
2691 allows ``ptrtoint`` and arithmetic to be performed on these values so
2692 long as the original value is reconstituted before the ``indirectbr``
2695 Finally, some targets may provide defined semantics when using the value
2696 as the operand to an inline assembly, but that is target specific.
2700 Constant Expressions
2701 --------------------
2703 Constant expressions are used to allow expressions involving other
2704 constants to be used as constants. Constant expressions may be of any
2705 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2706 that does not have side effects (e.g. load and call are not supported).
2707 The following is the syntax for constant expressions:
2709 ``trunc (CST to TYPE)``
2710 Truncate a constant to another type. The bit size of CST must be
2711 larger than the bit size of TYPE. Both types must be integers.
2712 ``zext (CST to TYPE)``
2713 Zero 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 ``sext (CST to TYPE)``
2716 Sign extend a constant to another type. The bit size of CST must be
2717 smaller than the bit size of TYPE. Both types must be integers.
2718 ``fptrunc (CST to TYPE)``
2719 Truncate a floating point constant to another floating point type.
2720 The size of CST must be larger than the size of TYPE. Both types
2721 must be floating point.
2722 ``fpext (CST to TYPE)``
2723 Floating point extend a constant to another type. The size of CST
2724 must be smaller or equal to the size of TYPE. Both types must be
2726 ``fptoui (CST to TYPE)``
2727 Convert a floating point constant to the corresponding unsigned
2728 integer constant. TYPE must be a scalar or vector integer type. CST
2729 must be of scalar or vector floating point type. Both CST and TYPE
2730 must be scalars, or vectors of the same number of elements. If the
2731 value won't fit in the integer type, the results are undefined.
2732 ``fptosi (CST to TYPE)``
2733 Convert a floating point constant to the corresponding signed
2734 integer constant. TYPE must be a scalar or vector integer type. CST
2735 must be of scalar or vector floating point type. Both CST and TYPE
2736 must be scalars, or vectors of the same number of elements. If the
2737 value won't fit in the integer type, the results are undefined.
2738 ``uitofp (CST to TYPE)``
2739 Convert an unsigned integer constant to the corresponding floating
2740 point constant. TYPE must be a scalar or vector floating point type.
2741 CST must be of scalar or vector integer type. Both CST and TYPE must
2742 be scalars, or vectors of the same number of elements. If the value
2743 won't fit in the floating point type, the results are undefined.
2744 ``sitofp (CST to TYPE)``
2745 Convert a signed integer constant to the corresponding floating
2746 point constant. TYPE must be a scalar or vector floating point type.
2747 CST must be of scalar or vector integer type. Both CST and TYPE must
2748 be scalars, or vectors of the same number of elements. If the value
2749 won't fit in the floating point type, the results are undefined.
2750 ``ptrtoint (CST to TYPE)``
2751 Convert a pointer typed constant to the corresponding integer
2752 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2753 pointer type. The ``CST`` value is zero extended, truncated, or
2754 unchanged to make it fit in ``TYPE``.
2755 ``inttoptr (CST to TYPE)``
2756 Convert an integer constant to a pointer constant. TYPE must be a
2757 pointer type. CST must be of integer type. The CST value is zero
2758 extended, truncated, or unchanged to make it fit in a pointer size.
2759 This one is *really* dangerous!
2760 ``bitcast (CST to TYPE)``
2761 Convert a constant, CST, to another TYPE. The constraints of the
2762 operands are the same as those for the :ref:`bitcast
2763 instruction <i_bitcast>`.
2764 ``addrspacecast (CST to TYPE)``
2765 Convert a constant pointer or constant vector of pointer, CST, to another
2766 TYPE in a different address space. The constraints of the operands are the
2767 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2768 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
2769 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2770 constants. As with the :ref:`getelementptr <i_getelementptr>`
2771 instruction, the index list may have zero or more indexes, which are
2772 required to make sense for the type of "pointer to TY".
2773 ``select (COND, VAL1, VAL2)``
2774 Perform the :ref:`select operation <i_select>` on constants.
2775 ``icmp COND (VAL1, VAL2)``
2776 Performs the :ref:`icmp operation <i_icmp>` on constants.
2777 ``fcmp COND (VAL1, VAL2)``
2778 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2779 ``extractelement (VAL, IDX)``
2780 Perform the :ref:`extractelement operation <i_extractelement>` on
2782 ``insertelement (VAL, ELT, IDX)``
2783 Perform the :ref:`insertelement operation <i_insertelement>` on
2785 ``shufflevector (VEC1, VEC2, IDXMASK)``
2786 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2788 ``extractvalue (VAL, IDX0, IDX1, ...)``
2789 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2790 constants. The index list is interpreted in a similar manner as
2791 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2792 least one index value must be specified.
2793 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2794 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2795 The index list is interpreted in a similar manner as indices in a
2796 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2797 value must be specified.
2798 ``OPCODE (LHS, RHS)``
2799 Perform the specified operation of the LHS and RHS constants. OPCODE
2800 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2801 binary <bitwiseops>` operations. The constraints on operands are
2802 the same as those for the corresponding instruction (e.g. no bitwise
2803 operations on floating point values are allowed).
2810 Inline Assembler Expressions
2811 ----------------------------
2813 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2814 Inline Assembly <moduleasm>`) through the use of a special value. This value
2815 represents the inline assembler as a template string (containing the
2816 instructions to emit), a list of operand constraints (stored as a string), a
2817 flag that indicates whether or not the inline asm expression has side effects,
2818 and a flag indicating whether the function containing the asm needs to align its
2819 stack conservatively.
2821 The template string supports argument substitution of the operands using "``$``"
2822 followed by a number, to indicate substitution of the given register/memory
2823 location, as specified by the constraint string. "``${NUM:MODIFIER}``" may also
2824 be used, where ``MODIFIER`` is a target-specific annotation for how to print the
2825 operand (See :ref:`inline-asm-modifiers`).
2827 A literal "``$``" may be included by using "``$$``" in the template. To include
2828 other special characters into the output, the usual "``\XX``" escapes may be
2829 used, just as in other strings. Note that after template substitution, the
2830 resulting assembly string is parsed by LLVM's integrated assembler unless it is
2831 disabled -- even when emitting a ``.s`` file -- and thus must contain assembly
2832 syntax known to LLVM.
2834 LLVM's support for inline asm is modeled closely on the requirements of Clang's
2835 GCC-compatible inline-asm support. Thus, the feature-set and the constraint and
2836 modifier codes listed here are similar or identical to those in GCC's inline asm
2837 support. However, to be clear, the syntax of the template and constraint strings
2838 described here is *not* the same as the syntax accepted by GCC and Clang, and,
2839 while most constraint letters are passed through as-is by Clang, some get
2840 translated to other codes when converting from the C source to the LLVM
2843 An example inline assembler expression is:
2845 .. code-block:: llvm
2847 i32 (i32) asm "bswap $0", "=r,r"
2849 Inline assembler expressions may **only** be used as the callee operand
2850 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2851 Thus, typically we have:
2853 .. code-block:: llvm
2855 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2857 Inline asms with side effects not visible in the constraint list must be
2858 marked as having side effects. This is done through the use of the
2859 '``sideeffect``' keyword, like so:
2861 .. code-block:: llvm
2863 call void asm sideeffect "eieio", ""()
2865 In some cases inline asms will contain code that will not work unless
2866 the stack is aligned in some way, such as calls or SSE instructions on
2867 x86, yet will not contain code that does that alignment within the asm.
2868 The compiler should make conservative assumptions about what the asm
2869 might contain and should generate its usual stack alignment code in the
2870 prologue if the '``alignstack``' keyword is present:
2872 .. code-block:: llvm
2874 call void asm alignstack "eieio", ""()
2876 Inline asms also support using non-standard assembly dialects. The
2877 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2878 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2879 the only supported dialects. An example is:
2881 .. code-block:: llvm
2883 call void asm inteldialect "eieio", ""()
2885 If multiple keywords appear the '``sideeffect``' keyword must come
2886 first, the '``alignstack``' keyword second and the '``inteldialect``'
2889 Inline Asm Constraint String
2890 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2892 The constraint list is a comma-separated string, each element containing one or
2893 more constraint codes.
2895 For each element in the constraint list an appropriate register or memory
2896 operand will be chosen, and it will be made available to assembly template
2897 string expansion as ``$0`` for the first constraint in the list, ``$1`` for the
2900 There are three different types of constraints, which are distinguished by a
2901 prefix symbol in front of the constraint code: Output, Input, and Clobber. The
2902 constraints must always be given in that order: outputs first, then inputs, then
2903 clobbers. They cannot be intermingled.
2905 There are also three different categories of constraint codes:
2907 - Register constraint. This is either a register class, or a fixed physical
2908 register. This kind of constraint will allocate a register, and if necessary,
2909 bitcast the argument or result to the appropriate type.
2910 - Memory constraint. This kind of constraint is for use with an instruction
2911 taking a memory operand. Different constraints allow for different addressing
2912 modes used by the target.
2913 - Immediate value constraint. This kind of constraint is for an integer or other
2914 immediate value which can be rendered directly into an instruction. The
2915 various target-specific constraints allow the selection of a value in the
2916 proper range for the instruction you wish to use it with.
2921 Output constraints are specified by an "``=``" prefix (e.g. "``=r``"). This
2922 indicates that the assembly will write to this operand, and the operand will
2923 then be made available as a return value of the ``asm`` expression. Output
2924 constraints do not consume an argument from the call instruction. (Except, see
2925 below about indirect outputs).
2927 Normally, it is expected that no output locations are written to by the assembly
2928 expression until *all* of the inputs have been read. As such, LLVM may assign
2929 the same register to an output and an input. If this is not safe (e.g. if the
2930 assembly contains two instructions, where the first writes to one output, and
2931 the second reads an input and writes to a second output), then the "``&``"
2932 modifier must be used (e.g. "``=&r``") to specify that the output is an
2933 "early-clobber" output. Marking an ouput as "early-clobber" ensures that LLVM
2934 will not use the same register for any inputs (other than an input tied to this
2940 Input constraints do not have a prefix -- just the constraint codes. Each input
2941 constraint will consume one argument from the call instruction. It is not
2942 permitted for the asm to write to any input register or memory location (unless
2943 that input is tied to an output). Note also that multiple inputs may all be
2944 assigned to the same register, if LLVM can determine that they necessarily all
2945 contain the same value.
2947 Instead of providing a Constraint Code, input constraints may also "tie"
2948 themselves to an output constraint, by providing an integer as the constraint
2949 string. Tied inputs still consume an argument from the call instruction, and
2950 take up a position in the asm template numbering as is usual -- they will simply
2951 be constrained to always use the same register as the output they've been tied
2952 to. For example, a constraint string of "``=r,0``" says to assign a register for
2953 output, and use that register as an input as well (it being the 0'th
2956 It is permitted to tie an input to an "early-clobber" output. In that case, no
2957 *other* input may share the same register as the input tied to the early-clobber
2958 (even when the other input has the same value).
2960 You may only tie an input to an output which has a register constraint, not a
2961 memory constraint. Only a single input may be tied to an output.
2963 There is also an "interesting" feature which deserves a bit of explanation: if a
2964 register class constraint allocates a register which is too small for the value
2965 type operand provided as input, the input value will be split into multiple
2966 registers, and all of them passed to the inline asm.
2968 However, this feature is often not as useful as you might think.
2970 Firstly, the registers are *not* guaranteed to be consecutive. So, on those
2971 architectures that have instructions which operate on multiple consecutive
2972 instructions, this is not an appropriate way to support them. (e.g. the 32-bit
2973 SparcV8 has a 64-bit load, which instruction takes a single 32-bit register. The
2974 hardware then loads into both the named register, and the next register. This
2975 feature of inline asm would not be useful to support that.)
2977 A few of the targets provide a template string modifier allowing explicit access
2978 to the second register of a two-register operand (e.g. MIPS ``L``, ``M``, and
2979 ``D``). On such an architecture, you can actually access the second allocated
2980 register (yet, still, not any subsequent ones). But, in that case, you're still
2981 probably better off simply splitting the value into two separate operands, for
2982 clarity. (e.g. see the description of the ``A`` constraint on X86, which,
2983 despite existing only for use with this feature, is not really a good idea to
2986 Indirect inputs and outputs
2987 """""""""""""""""""""""""""
2989 Indirect output or input constraints can be specified by the "``*``" modifier
2990 (which goes after the "``=``" in case of an output). This indicates that the asm
2991 will write to or read from the contents of an *address* provided as an input
2992 argument. (Note that in this way, indirect outputs act more like an *input* than
2993 an output: just like an input, they consume an argument of the call expression,
2994 rather than producing a return value. An indirect output constraint is an
2995 "output" only in that the asm is expected to write to the contents of the input
2996 memory location, instead of just read from it).
2998 This is most typically used for memory constraint, e.g. "``=*m``", to pass the
2999 address of a variable as a value.
3001 It is also possible to use an indirect *register* constraint, but only on output
3002 (e.g. "``=*r``"). This will cause LLVM to allocate a register for an output
3003 value normally, and then, separately emit a store to the address provided as
3004 input, after the provided inline asm. (It's not clear what value this
3005 functionality provides, compared to writing the store explicitly after the asm
3006 statement, and it can only produce worse code, since it bypasses many
3007 optimization passes. I would recommend not using it.)
3013 A clobber constraint is indicated by a "``~``" prefix. A clobber does not
3014 consume an input operand, nor generate an output. Clobbers cannot use any of the
3015 general constraint code letters -- they may use only explicit register
3016 constraints, e.g. "``~{eax}``". The one exception is that a clobber string of
3017 "``~{memory}``" indicates that the assembly writes to arbitrary undeclared
3018 memory locations -- not only the memory pointed to by a declared indirect
3024 After a potential prefix comes constraint code, or codes.
3026 A Constraint Code is either a single letter (e.g. "``r``"), a "``^``" character
3027 followed by two letters (e.g. "``^wc``"), or "``{``" register-name "``}``"
3030 The one and two letter constraint codes are typically chosen to be the same as
3031 GCC's constraint codes.
3033 A single constraint may include one or more than constraint code in it, leaving
3034 it up to LLVM to choose which one to use. This is included mainly for
3035 compatibility with the translation of GCC inline asm coming from clang.
3037 There are two ways to specify alternatives, and either or both may be used in an
3038 inline asm constraint list:
3040 1) Append the codes to each other, making a constraint code set. E.g. "``im``"
3041 or "``{eax}m``". This means "choose any of the options in the set". The
3042 choice of constraint is made independently for each constraint in the
3045 2) Use "``|``" between constraint code sets, creating alternatives. Every
3046 constraint in the constraint list must have the same number of alternative
3047 sets. With this syntax, the same alternative in *all* of the items in the
3048 constraint list will be chosen together.
3050 Putting those together, you might have a two operand constraint string like
3051 ``"rm|r,ri|rm"``. This indicates that if operand 0 is ``r`` or ``m``, then
3052 operand 1 may be one of ``r`` or ``i``. If operand 0 is ``r``, then operand 1
3053 may be one of ``r`` or ``m``. But, operand 0 and 1 cannot both be of type m.
3055 However, the use of either of the alternatives features is *NOT* recommended, as
3056 LLVM is not able to make an intelligent choice about which one to use. (At the
3057 point it currently needs to choose, not enough information is available to do so
3058 in a smart way.) Thus, it simply tries to make a choice that's most likely to
3059 compile, not one that will be optimal performance. (e.g., given "``rm``", it'll
3060 always choose to use memory, not registers). And, if given multiple registers,
3061 or multiple register classes, it will simply choose the first one. (In fact, it
3062 doesn't currently even ensure explicitly specified physical registers are
3063 unique, so specifying multiple physical registers as alternatives, like
3064 ``{r11}{r12},{r11}{r12}``, will assign r11 to both operands, not at all what was
3067 Supported Constraint Code List
3068 """"""""""""""""""""""""""""""
3070 The constraint codes are, in general, expected to behave the same way they do in
3071 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3072 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3073 and GCC likely indicates a bug in LLVM.
3075 Some constraint codes are typically supported by all targets:
3077 - ``r``: A register in the target's general purpose register class.
3078 - ``m``: A memory address operand. It is target-specific what addressing modes
3079 are supported, typical examples are register, or register + register offset,
3080 or register + immediate offset (of some target-specific size).
3081 - ``i``: An integer constant (of target-specific width). Allows either a simple
3082 immediate, or a relocatable value.
3083 - ``n``: An integer constant -- *not* including relocatable values.
3084 - ``s``: An integer constant, but allowing *only* relocatable values.
3085 - ``X``: Allows an operand of any kind, no constraint whatsoever. Typically
3086 useful to pass a label for an asm branch or call.
3088 .. FIXME: but that surely isn't actually okay to jump out of an asm
3089 block without telling llvm about the control transfer???)
3091 - ``{register-name}``: Requires exactly the named physical register.
3093 Other constraints are target-specific:
3097 - ``z``: An immediate integer 0. Outputs ``WZR`` or ``XZR``, as appropriate.
3098 - ``I``: An immediate integer valid for an ``ADD`` or ``SUB`` instruction,
3099 i.e. 0 to 4095 with optional shift by 12.
3100 - ``J``: An immediate integer that, when negated, is valid for an ``ADD`` or
3101 ``SUB`` instruction, i.e. -1 to -4095 with optional left shift by 12.
3102 - ``K``: An immediate integer that is valid for the 'bitmask immediate 32' of a
3103 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 32-bit register.
3104 - ``L``: An immediate integer that is valid for the 'bitmask immediate 64' of a
3105 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 64-bit register.
3106 - ``M``: An immediate integer for use with the ``MOV`` assembly alias on a
3107 32-bit register. This is a superset of ``K``: in addition to the bitmask
3108 immediate, also allows immediate integers which can be loaded with a single
3109 ``MOVZ`` or ``MOVL`` instruction.
3110 - ``N``: An immediate integer for use with the ``MOV`` assembly alias on a
3111 64-bit register. This is a superset of ``L``.
3112 - ``Q``: Memory address operand must be in a single register (no
3113 offsets). (However, LLVM currently does this for the ``m`` constraint as
3115 - ``r``: A 32 or 64-bit integer register (W* or X*).
3116 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register.
3117 - ``x``: A lower 128-bit floating-point/SIMD register (``V0`` to ``V15``).
3121 - ``r``: A 32 or 64-bit integer register.
3122 - ``[0-9]v``: The 32-bit VGPR register, number 0-9.
3123 - ``[0-9]s``: The 32-bit SGPR register, number 0-9.
3128 - ``Q``, ``Um``, ``Un``, ``Uq``, ``Us``, ``Ut``, ``Uv``, ``Uy``: Memory address
3129 operand. Treated the same as operand ``m``, at the moment.
3131 ARM and ARM's Thumb2 mode:
3133 - ``j``: An immediate integer between 0 and 65535 (valid for ``MOVW``)
3134 - ``I``: An immediate integer valid for a data-processing instruction.
3135 - ``J``: An immediate integer between -4095 and 4095.
3136 - ``K``: An immediate integer whose bitwise inverse is valid for a
3137 data-processing instruction. (Can be used with template modifier "``B``" to
3138 print the inverted value).
3139 - ``L``: An immediate integer whose negation is valid for a data-processing
3140 instruction. (Can be used with template modifier "``n``" to print the negated
3142 - ``M``: A power of two or a integer between 0 and 32.
3143 - ``N``: Invalid immediate constraint.
3144 - ``O``: Invalid immediate constraint.
3145 - ``r``: A general-purpose 32-bit integer register (``r0-r15``).
3146 - ``l``: In Thumb2 mode, low 32-bit GPR registers (``r0-r7``). In ARM mode, same
3148 - ``h``: In Thumb2 mode, a high 32-bit GPR register (``r8-r15``). In ARM mode,
3150 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3151 ``d0-d31``, or ``q0-q15``.
3152 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3153 ``d0-d7``, or ``q0-q3``.
3154 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3159 - ``I``: An immediate integer between 0 and 255.
3160 - ``J``: An immediate integer between -255 and -1.
3161 - ``K``: An immediate integer between 0 and 255, with optional left-shift by
3163 - ``L``: An immediate integer between -7 and 7.
3164 - ``M``: An immediate integer which is a multiple of 4 between 0 and 1020.
3165 - ``N``: An immediate integer between 0 and 31.
3166 - ``O``: An immediate integer which is a multiple of 4 between -508 and 508.
3167 - ``r``: A low 32-bit GPR register (``r0-r7``).
3168 - ``l``: A low 32-bit GPR register (``r0-r7``).
3169 - ``h``: A high GPR register (``r0-r7``).
3170 - ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3171 ``d0-d31``, or ``q0-q15``.
3172 - ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3173 ``d0-d7``, or ``q0-q3``.
3174 - ``t``: A floating-point/SIMD register, only supports 32-bit values:
3180 - ``o``, ``v``: A memory address operand, treated the same as constraint ``m``,
3182 - ``r``: A 32 or 64-bit register.
3186 - ``r``: An 8 or 16-bit register.
3190 - ``I``: An immediate signed 16-bit integer.
3191 - ``J``: An immediate integer zero.
3192 - ``K``: An immediate unsigned 16-bit integer.
3193 - ``L``: An immediate 32-bit integer, where the lower 16 bits are 0.
3194 - ``N``: An immediate integer between -65535 and -1.
3195 - ``O``: An immediate signed 15-bit integer.
3196 - ``P``: An immediate integer between 1 and 65535.
3197 - ``m``: A memory address operand. In MIPS-SE mode, allows a base address
3198 register plus 16-bit immediate offset. In MIPS mode, just a base register.
3199 - ``R``: A memory address operand. In MIPS-SE mode, allows a base address
3200 register plus a 9-bit signed offset. In MIPS mode, the same as constraint
3202 - ``ZC``: A memory address operand, suitable for use in a ``pref``, ``ll``, or
3203 ``sc`` instruction on the given subtarget (details vary).
3204 - ``r``, ``d``, ``y``: A 32 or 64-bit GPR register.
3205 - ``f``: A 32 or 64-bit FPU register (``F0-F31``), or a 128-bit MSA register
3206 (``W0-W31``). In the case of MSA registers, it is recommended to use the ``w``
3207 argument modifier for compatibility with GCC.
3208 - ``c``: A 32-bit or 64-bit GPR register suitable for indirect jump (always
3210 - ``l``: The ``lo`` register, 32 or 64-bit.
3215 - ``b``: A 1-bit integer register.
3216 - ``c`` or ``h``: A 16-bit integer register.
3217 - ``r``: A 32-bit integer register.
3218 - ``l`` or ``N``: A 64-bit integer register.
3219 - ``f``: A 32-bit float register.
3220 - ``d``: A 64-bit float register.
3225 - ``I``: An immediate signed 16-bit integer.
3226 - ``J``: An immediate unsigned 16-bit integer, shifted left 16 bits.
3227 - ``K``: An immediate unsigned 16-bit integer.
3228 - ``L``: An immediate signed 16-bit integer, shifted left 16 bits.
3229 - ``M``: An immediate integer greater than 31.
3230 - ``N``: An immediate integer that is an exact power of 2.
3231 - ``O``: The immediate integer constant 0.
3232 - ``P``: An immediate integer constant whose negation is a signed 16-bit
3234 - ``es``, ``o``, ``Q``, ``Z``, ``Zy``: A memory address operand, currently
3235 treated the same as ``m``.
3236 - ``r``: A 32 or 64-bit integer register.
3237 - ``b``: A 32 or 64-bit integer register, excluding ``R0`` (that is:
3239 - ``f``: A 32 or 64-bit float register (``F0-F31``), or when QPX is enabled, a
3240 128 or 256-bit QPX register (``Q0-Q31``; aliases the ``F`` registers).
3241 - ``v``: For ``4 x f32`` or ``4 x f64`` types, when QPX is enabled, a
3242 128 or 256-bit QPX register (``Q0-Q31``), otherwise a 128-bit
3243 altivec vector register (``V0-V31``).
3245 .. FIXME: is this a bug that v accepts QPX registers? I think this
3246 is supposed to only use the altivec vector registers?
3248 - ``y``: Condition register (``CR0-CR7``).
3249 - ``wc``: An individual CR bit in a CR register.
3250 - ``wa``, ``wd``, ``wf``: Any 128-bit VSX vector register, from the full VSX
3251 register set (overlapping both the floating-point and vector register files).
3252 - ``ws``: A 32 or 64-bit floating point register, from the full VSX register
3257 - ``I``: An immediate 13-bit signed integer.
3258 - ``r``: A 32-bit integer register.
3262 - ``I``: An immediate unsigned 8-bit integer.
3263 - ``J``: An immediate unsigned 12-bit integer.
3264 - ``K``: An immediate signed 16-bit integer.
3265 - ``L``: An immediate signed 20-bit integer.
3266 - ``M``: An immediate integer 0x7fffffff.
3267 - ``Q``, ``R``, ``S``, ``T``: A memory address operand, treated the same as
3268 ``m``, at the moment.
3269 - ``r`` or ``d``: A 32, 64, or 128-bit integer register.
3270 - ``a``: A 32, 64, or 128-bit integer address register (excludes R0, which in an
3271 address context evaluates as zero).
3272 - ``h``: A 32-bit value in the high part of a 64bit data register
3274 - ``f``: A 32, 64, or 128-bit floating point register.
3278 - ``I``: An immediate integer between 0 and 31.
3279 - ``J``: An immediate integer between 0 and 64.
3280 - ``K``: An immediate signed 8-bit integer.
3281 - ``L``: An immediate integer, 0xff or 0xffff or (in 64-bit mode only)
3283 - ``M``: An immediate integer between 0 and 3.
3284 - ``N``: An immediate unsigned 8-bit integer.
3285 - ``O``: An immediate integer between 0 and 127.
3286 - ``e``: An immediate 32-bit signed integer.
3287 - ``Z``: An immediate 32-bit unsigned integer.
3288 - ``o``, ``v``: Treated the same as ``m``, at the moment.
3289 - ``q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3290 ``l`` integer register. On X86-32, this is the ``a``, ``b``, ``c``, and ``d``
3291 registers, and on X86-64, it is all of the integer registers.
3292 - ``Q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3293 ``h`` integer register. This is the ``a``, ``b``, ``c``, and ``d`` registers.
3294 - ``r`` or ``l``: An 8, 16, 32, or 64-bit integer register.
3295 - ``R``: An 8, 16, 32, or 64-bit "legacy" integer register -- one which has
3296 existed since i386, and can be accessed without the REX prefix.
3297 - ``f``: A 32, 64, or 80-bit '387 FPU stack pseudo-register.
3298 - ``y``: A 64-bit MMX register, if MMX is enabled.
3299 - ``x``: If SSE is enabled: a 32 or 64-bit scalar operand, or 128-bit vector
3300 operand in a SSE register. If AVX is also enabled, can also be a 256-bit
3301 vector operand in an AVX register. If AVX-512 is also enabled, can also be a
3302 512-bit vector operand in an AVX512 register, Otherwise, an error.
3303 - ``Y``: The same as ``x``, if *SSE2* is enabled, otherwise an error.
3304 - ``A``: Special case: allocates EAX first, then EDX, for a single operand (in
3305 32-bit mode, a 64-bit integer operand will get split into two registers). It
3306 is not recommended to use this constraint, as in 64-bit mode, the 64-bit
3307 operand will get allocated only to RAX -- if two 32-bit operands are needed,
3308 you're better off splitting it yourself, before passing it to the asm
3313 - ``r``: A 32-bit integer register.
3316 .. _inline-asm-modifiers:
3318 Asm template argument modifiers
3319 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3321 In the asm template string, modifiers can be used on the operand reference, like
3324 The modifiers are, in general, expected to behave the same way they do in
3325 GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3326 inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3327 and GCC likely indicates a bug in LLVM.
3331 - ``c``: Print an immediate integer constant unadorned, without
3332 the target-specific immediate punctuation (e.g. no ``$`` prefix).
3333 - ``n``: Negate and print immediate integer constant unadorned, without the
3334 target-specific immediate punctuation (e.g. no ``$`` prefix).
3335 - ``l``: Print as an unadorned label, without the target-specific label
3336 punctuation (e.g. no ``$`` prefix).
3340 - ``w``: Print a GPR register with a ``w*`` name instead of ``x*`` name. E.g.,
3341 instead of ``x30``, print ``w30``.
3342 - ``x``: Print a GPR register with a ``x*`` name. (this is the default, anyhow).
3343 - ``b``, ``h``, ``s``, ``d``, ``q``: Print a floating-point/SIMD register with a
3344 ``b*``, ``h*``, ``s*``, ``d*``, or ``q*`` name, rather than the default of
3353 - ``a``: Print an operand as an address (with ``[`` and ``]`` surrounding a
3357 - ``y``: Print a VFP single-precision register as an indexed double (e.g. print
3358 as ``d4[1]`` instead of ``s9``)
3359 - ``B``: Bitwise invert and print an immediate integer constant without ``#``
3361 - ``L``: Print the low 16-bits of an immediate integer constant.
3362 - ``M``: Print as a register set suitable for ldm/stm. Also prints *all*
3363 register operands subsequent to the specified one (!), so use carefully.
3364 - ``Q``: Print the low-order register of a register-pair, or the low-order
3365 register of a two-register operand.
3366 - ``R``: Print the high-order register of a register-pair, or the high-order
3367 register of a two-register operand.
3368 - ``H``: Print the second register of a register-pair. (On a big-endian system,
3369 ``H`` is equivalent to ``Q``, and on little-endian system, ``H`` is equivalent
3372 .. FIXME: H doesn't currently support printing the second register
3373 of a two-register operand.
3375 - ``e``: Print the low doubleword register of a NEON quad register.
3376 - ``f``: Print the high doubleword register of a NEON quad register.
3377 - ``m``: Print the base register of a memory operand without the ``[`` and ``]``
3382 - ``L``: Print the second register of a two-register operand. Requires that it
3383 has been allocated consecutively to the first.
3385 .. FIXME: why is it restricted to consecutive ones? And there's
3386 nothing that ensures that happens, is there?
3388 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3389 nothing. Used to print 'addi' vs 'add' instructions.
3393 No additional modifiers.
3397 - ``X``: Print an immediate integer as hexadecimal
3398 - ``x``: Print the low 16 bits of an immediate integer as hexadecimal.
3399 - ``d``: Print an immediate integer as decimal.
3400 - ``m``: Subtract one and print an immediate integer as decimal.
3401 - ``z``: Print $0 if an immediate zero, otherwise print normally.
3402 - ``L``: Print the low-order register of a two-register operand, or prints the
3403 address of the low-order word of a double-word memory operand.
3405 .. FIXME: L seems to be missing memory operand support.
3407 - ``M``: Print the high-order register of a two-register operand, or prints the
3408 address of the high-order word of a double-word memory operand.
3410 .. FIXME: M seems to be missing memory operand support.
3412 - ``D``: Print the second register of a two-register operand, or prints the
3413 second word of a double-word memory operand. (On a big-endian system, ``D`` is
3414 equivalent to ``L``, and on little-endian system, ``D`` is equivalent to
3416 - ``w``: No effect. Provided for compatibility with GCC which requires this
3417 modifier in order to print MSA registers (``W0-W31``) with the ``f``
3426 - ``L``: Print the second register of a two-register operand. Requires that it
3427 has been allocated consecutively to the first.
3429 .. FIXME: why is it restricted to consecutive ones? And there's
3430 nothing that ensures that happens, is there?
3432 - ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3433 nothing. Used to print 'addi' vs 'add' instructions.
3434 - ``y``: For a memory operand, prints formatter for a two-register X-form
3435 instruction. (Currently always prints ``r0,OPERAND``).
3436 - ``U``: Prints 'u' if the memory operand is an update form, and nothing
3437 otherwise. (NOTE: LLVM does not support update form, so this will currently
3438 always print nothing)
3439 - ``X``: Prints 'x' if the memory operand is an indexed form. (NOTE: LLVM does
3440 not support indexed form, so this will currently always print nothing)
3448 SystemZ implements only ``n``, and does *not* support any of the other
3449 target-independent modifiers.
3453 - ``c``: Print an unadorned integer or symbol name. (The latter is
3454 target-specific behavior for this typically target-independent modifier).
3455 - ``A``: Print a register name with a '``*``' before it.
3456 - ``b``: Print an 8-bit register name (e.g. ``al``); do nothing on a memory
3458 - ``h``: Print the upper 8-bit register name (e.g. ``ah``); do nothing on a
3460 - ``w``: Print the 16-bit register name (e.g. ``ax``); do nothing on a memory
3462 - ``k``: Print the 32-bit register name (e.g. ``eax``); do nothing on a memory
3464 - ``q``: Print the 64-bit register name (e.g. ``rax``), if 64-bit registers are
3465 available, otherwise the 32-bit register name; do nothing on a memory operand.
3466 - ``n``: Negate and print an unadorned integer, or, for operands other than an
3467 immediate integer (e.g. a relocatable symbol expression), print a '-' before
3468 the operand. (The behavior for relocatable symbol expressions is a
3469 target-specific behavior for this typically target-independent modifier)
3470 - ``H``: Print a memory reference with additional offset +8.
3471 - ``P``: Print a memory reference or operand for use as the argument of a call
3472 instruction. (E.g. omit ``(rip)``, even though it's PC-relative.)
3476 No additional modifiers.
3482 The call instructions that wrap inline asm nodes may have a
3483 "``!srcloc``" MDNode attached to it that contains a list of constant
3484 integers. If present, the code generator will use the integer as the
3485 location cookie value when report errors through the ``LLVMContext``
3486 error reporting mechanisms. This allows a front-end to correlate backend
3487 errors that occur with inline asm back to the source code that produced
3490 .. code-block:: llvm
3492 call void asm sideeffect "something bad", ""(), !srcloc !42
3494 !42 = !{ i32 1234567 }
3496 It is up to the front-end to make sense of the magic numbers it places
3497 in the IR. If the MDNode contains multiple constants, the code generator
3498 will use the one that corresponds to the line of the asm that the error
3506 LLVM IR allows metadata to be attached to instructions in the program
3507 that can convey extra information about the code to the optimizers and
3508 code generator. One example application of metadata is source-level
3509 debug information. There are two metadata primitives: strings and nodes.
3511 Metadata does not have a type, and is not a value. If referenced from a
3512 ``call`` instruction, it uses the ``metadata`` type.
3514 All metadata are identified in syntax by a exclamation point ('``!``').
3516 .. _metadata-string:
3518 Metadata Nodes and Metadata Strings
3519 -----------------------------------
3521 A metadata string is a string surrounded by double quotes. It can
3522 contain any character by escaping non-printable characters with
3523 "``\xx``" where "``xx``" is the two digit hex code. For example:
3526 Metadata nodes are represented with notation similar to structure
3527 constants (a comma separated list of elements, surrounded by braces and
3528 preceded by an exclamation point). Metadata nodes can have any values as
3529 their operand. For example:
3531 .. code-block:: llvm
3533 !{ !"test\00", i32 10}
3535 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
3537 .. code-block:: llvm
3539 !0 = distinct !{!"test\00", i32 10}
3541 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
3542 content. They can also occur when transformations cause uniquing collisions
3543 when metadata operands change.
3545 A :ref:`named metadata <namedmetadatastructure>` is a collection of
3546 metadata nodes, which can be looked up in the module symbol table. For
3549 .. code-block:: llvm
3553 Metadata can be used as function arguments. Here ``llvm.dbg.value``
3554 function is using two metadata arguments:
3556 .. code-block:: llvm
3558 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
3560 Metadata can be attached with an instruction. Here metadata ``!21`` is
3561 attached to the ``add`` instruction using the ``!dbg`` identifier:
3563 .. code-block:: llvm
3565 %indvar.next = add i64 %indvar, 1, !dbg !21
3567 More information about specific metadata nodes recognized by the
3568 optimizers and code generator is found below.
3570 .. _specialized-metadata:
3572 Specialized Metadata Nodes
3573 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3575 Specialized metadata nodes are custom data structures in metadata (as opposed
3576 to generic tuples). Their fields are labelled, and can be specified in any
3579 These aren't inherently debug info centric, but currently all the specialized
3580 metadata nodes are related to debug info.
3587 ``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
3588 ``retainedTypes:``, ``subprograms:``, ``globals:`` and ``imports:`` fields are
3589 tuples containing the debug info to be emitted along with the compile unit,
3590 regardless of code optimizations (some nodes are only emitted if there are
3591 references to them from instructions).
3593 .. code-block:: llvm
3595 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
3596 isOptimized: true, flags: "-O2", runtimeVersion: 2,
3597 splitDebugFilename: "abc.debug", emissionKind: 1,
3598 enums: !2, retainedTypes: !3, subprograms: !4,
3599 globals: !5, imports: !6)
3601 Compile unit descriptors provide the root scope for objects declared in a
3602 specific compilation unit. File descriptors are defined using this scope.
3603 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
3604 keep track of subprograms, global variables, type information, and imported
3605 entities (declarations and namespaces).
3612 ``DIFile`` nodes represent files. The ``filename:`` can include slashes.
3614 .. code-block:: llvm
3616 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
3618 Files are sometimes used in ``scope:`` fields, and are the only valid target
3619 for ``file:`` fields.
3626 ``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
3627 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
3629 .. code-block:: llvm
3631 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3632 encoding: DW_ATE_unsigned_char)
3633 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
3635 The ``encoding:`` describes the details of the type. Usually it's one of the
3638 .. code-block:: llvm
3644 DW_ATE_signed_char = 6
3646 DW_ATE_unsigned_char = 8
3648 .. _DISubroutineType:
3653 ``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
3654 refers to a tuple; the first operand is the return type, while the rest are the
3655 types of the formal arguments in order. If the first operand is ``null``, that
3656 represents a function with no return value (such as ``void foo() {}`` in C++).
3658 .. code-block:: llvm
3660 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
3661 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
3662 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
3669 ``DIDerivedType`` nodes represent types derived from other types, such as
3672 .. code-block:: llvm
3674 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3675 encoding: DW_ATE_unsigned_char)
3676 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
3679 The following ``tag:`` values are valid:
3681 .. code-block:: llvm
3683 DW_TAG_formal_parameter = 5
3685 DW_TAG_pointer_type = 15
3686 DW_TAG_reference_type = 16
3688 DW_TAG_ptr_to_member_type = 31
3689 DW_TAG_const_type = 38
3690 DW_TAG_volatile_type = 53
3691 DW_TAG_restrict_type = 55
3693 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
3694 <DICompositeType>` or :ref:`subprogram <DISubprogram>`. The type of the member
3695 is the ``baseType:``. The ``offset:`` is the member's bit offset.
3696 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
3697 argument of a subprogram.
3699 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3701 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3702 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3705 Note that the ``void *`` type is expressed as a type derived from NULL.
3707 .. _DICompositeType:
3712 ``DICompositeType`` nodes represent types composed of other types, like
3713 structures and unions. ``elements:`` points to a tuple of the composed types.
3715 If the source language supports ODR, the ``identifier:`` field gives the unique
3716 identifier used for type merging between modules. When specified, other types
3717 can refer to composite types indirectly via a :ref:`metadata string
3718 <metadata-string>` that matches their identifier.
3720 .. code-block:: llvm
3722 !0 = !DIEnumerator(name: "SixKind", value: 7)
3723 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3724 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3725 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
3726 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3727 elements: !{!0, !1, !2})
3729 The following ``tag:`` values are valid:
3731 .. code-block:: llvm
3733 DW_TAG_array_type = 1
3734 DW_TAG_class_type = 2
3735 DW_TAG_enumeration_type = 4
3736 DW_TAG_structure_type = 19
3737 DW_TAG_union_type = 23
3738 DW_TAG_subroutine_type = 21
3739 DW_TAG_inheritance = 28
3742 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
3743 descriptors <DISubrange>`, each representing the range of subscripts at that
3744 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3745 array type is a native packed vector.
3747 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
3748 descriptors <DIEnumerator>`, each representing the definition of an enumeration
3749 value for the set. All enumeration type descriptors are collected in the
3750 ``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
3752 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3753 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
3754 <DIDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
3761 ``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3762 :ref:`DICompositeType`. ``count: -1`` indicates an empty array.
3764 .. code-block:: llvm
3766 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
3767 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
3768 !2 = !DISubrange(count: -1) ; empty array.
3775 ``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3776 variants of :ref:`DICompositeType`.
3778 .. code-block:: llvm
3780 !0 = !DIEnumerator(name: "SixKind", value: 7)
3781 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3782 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3784 DITemplateTypeParameter
3785 """""""""""""""""""""""
3787 ``DITemplateTypeParameter`` nodes represent type parameters to generic source
3788 language constructs. They are used (optionally) in :ref:`DICompositeType` and
3789 :ref:`DISubprogram` ``templateParams:`` fields.
3791 .. code-block:: llvm
3793 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
3795 DITemplateValueParameter
3796 """"""""""""""""""""""""
3798 ``DITemplateValueParameter`` nodes represent value parameters to generic source
3799 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3800 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3801 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
3802 :ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
3804 .. code-block:: llvm
3806 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
3811 ``DINamespace`` nodes represent namespaces in the source language.
3813 .. code-block:: llvm
3815 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
3820 ``DIGlobalVariable`` nodes represent global variables in the source language.
3822 .. code-block:: llvm
3824 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
3825 file: !2, line: 7, type: !3, isLocal: true,
3826 isDefinition: false, variable: i32* @foo,
3829 All global variables should be referenced by the `globals:` field of a
3830 :ref:`compile unit <DICompileUnit>`.
3837 ``DISubprogram`` nodes represent functions from the source language. The
3838 ``variables:`` field points at :ref:`variables <DILocalVariable>` that must be
3839 retained, even if their IR counterparts are optimized out of the IR. The
3840 ``type:`` field must point at an :ref:`DISubroutineType`.
3842 .. code-block:: llvm
3844 !0 = !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
3845 file: !2, line: 7, type: !3, isLocal: true,
3846 isDefinition: false, scopeLine: 8, containingType: !4,
3847 virtuality: DW_VIRTUALITY_pure_virtual, virtualIndex: 10,
3848 flags: DIFlagPrototyped, isOptimized: true,
3849 function: void ()* @_Z3foov,
3850 templateParams: !5, declaration: !6, variables: !7)
3857 ``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
3858 <DISubprogram>`. The line number and column numbers are used to dinstinguish
3859 two lexical blocks at same depth. They are valid targets for ``scope:``
3862 .. code-block:: llvm
3864 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
3866 Usually lexical blocks are ``distinct`` to prevent node merging based on
3869 .. _DILexicalBlockFile:
3874 ``DILexicalBlockFile`` nodes are used to discriminate between sections of a
3875 :ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
3876 indicate textual inclusion, or the ``discriminator:`` field can be used to
3877 discriminate between control flow within a single block in the source language.
3879 .. code-block:: llvm
3881 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
3882 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
3883 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
3890 ``DILocation`` nodes represent source debug locations. The ``scope:`` field is
3891 mandatory, and points at an :ref:`DILexicalBlockFile`, an
3892 :ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
3894 .. code-block:: llvm
3896 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
3898 .. _DILocalVariable:
3903 ``DILocalVariable`` nodes represent local variables in the source language.
3904 Instead of ``DW_TAG_variable``, they use LLVM-specific fake tags to
3905 discriminate between local variables (``DW_TAG_auto_variable``) and subprogram
3906 arguments (``DW_TAG_arg_variable``). In the latter case, the ``arg:`` field
3907 specifies the argument position, and this variable will be included in the
3908 ``variables:`` field of its :ref:`DISubprogram`.
3910 .. code-block:: llvm
3912 !0 = !DILocalVariable(tag: DW_TAG_arg_variable, name: "this", arg: 1,
3913 scope: !3, file: !2, line: 7, type: !3,
3914 flags: DIFlagArtificial)
3915 !1 = !DILocalVariable(tag: DW_TAG_arg_variable, name: "x", arg: 2,
3916 scope: !4, file: !2, line: 7, type: !3)
3917 !2 = !DILocalVariable(tag: DW_TAG_auto_variable, name: "y",
3918 scope: !5, file: !2, line: 7, type: !3)
3923 ``DIExpression`` nodes represent DWARF expression sequences. They are used in
3924 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
3925 describe how the referenced LLVM variable relates to the source language
3928 The current supported vocabulary is limited:
3930 - ``DW_OP_deref`` dereferences the working expression.
3931 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
3932 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
3933 here, respectively) of the variable piece from the working expression.
3935 .. code-block:: llvm
3937 !0 = !DIExpression(DW_OP_deref)
3938 !1 = !DIExpression(DW_OP_plus, 3)
3939 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
3940 !3 = !DIExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
3945 ``DIObjCProperty`` nodes represent Objective-C property nodes.
3947 .. code-block:: llvm
3949 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
3950 getter: "getFoo", attributes: 7, type: !2)
3955 ``DIImportedEntity`` nodes represent entities (such as modules) imported into a
3958 .. code-block:: llvm
3960 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
3961 entity: !1, line: 7)
3966 In LLVM IR, memory does not have types, so LLVM's own type system is not
3967 suitable for doing TBAA. Instead, metadata is added to the IR to
3968 describe a type system of a higher level language. This can be used to
3969 implement typical C/C++ TBAA, but it can also be used to implement
3970 custom alias analysis behavior for other languages.
3972 The current metadata format is very simple. TBAA metadata nodes have up
3973 to three fields, e.g.:
3975 .. code-block:: llvm
3977 !0 = !{ !"an example type tree" }
3978 !1 = !{ !"int", !0 }
3979 !2 = !{ !"float", !0 }
3980 !3 = !{ !"const float", !2, i64 1 }
3982 The first field is an identity field. It can be any value, usually a
3983 metadata string, which uniquely identifies the type. The most important
3984 name in the tree is the name of the root node. Two trees with different
3985 root node names are entirely disjoint, even if they have leaves with
3988 The second field identifies the type's parent node in the tree, or is
3989 null or omitted for a root node. A type is considered to alias all of
3990 its descendants and all of its ancestors in the tree. Also, a type is
3991 considered to alias all types in other trees, so that bitcode produced
3992 from multiple front-ends is handled conservatively.
3994 If the third field is present, it's an integer which if equal to 1
3995 indicates that the type is "constant" (meaning
3996 ``pointsToConstantMemory`` should return true; see `other useful
3997 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
3999 '``tbaa.struct``' Metadata
4000 ^^^^^^^^^^^^^^^^^^^^^^^^^^
4002 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
4003 aggregate assignment operations in C and similar languages, however it
4004 is defined to copy a contiguous region of memory, which is more than
4005 strictly necessary for aggregate types which contain holes due to
4006 padding. Also, it doesn't contain any TBAA information about the fields
4009 ``!tbaa.struct`` metadata can describe which memory subregions in a
4010 memcpy are padding and what the TBAA tags of the struct are.
4012 The current metadata format is very simple. ``!tbaa.struct`` metadata
4013 nodes are a list of operands which are in conceptual groups of three.
4014 For each group of three, the first operand gives the byte offset of a
4015 field in bytes, the second gives its size in bytes, and the third gives
4018 .. code-block:: llvm
4020 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
4022 This describes a struct with two fields. The first is at offset 0 bytes
4023 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
4024 and has size 4 bytes and has tbaa tag !2.
4026 Note that the fields need not be contiguous. In this example, there is a
4027 4 byte gap between the two fields. This gap represents padding which
4028 does not carry useful data and need not be preserved.
4030 '``noalias``' and '``alias.scope``' Metadata
4031 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4033 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
4034 noalias memory-access sets. This means that some collection of memory access
4035 instructions (loads, stores, memory-accessing calls, etc.) that carry
4036 ``noalias`` metadata can specifically be specified not to alias with some other
4037 collection of memory access instructions that carry ``alias.scope`` metadata.
4038 Each type of metadata specifies a list of scopes where each scope has an id and
4039 a domain. When evaluating an aliasing query, if for some domain, the set
4040 of scopes with that domain in one instruction's ``alias.scope`` list is a
4041 subset of (or equal to) the set of scopes for that domain in another
4042 instruction's ``noalias`` list, then the two memory accesses are assumed not to
4045 The metadata identifying each domain is itself a list containing one or two
4046 entries. The first entry is the name of the domain. Note that if the name is a
4047 string then it can be combined accross functions and translation units. A
4048 self-reference can be used to create globally unique domain names. A
4049 descriptive string may optionally be provided as a second list entry.
4051 The metadata identifying each scope is also itself a list containing two or
4052 three entries. The first entry is the name of the scope. Note that if the name
4053 is a string then it can be combined accross functions and translation units. A
4054 self-reference can be used to create globally unique scope names. A metadata
4055 reference to the scope's domain is the second entry. A descriptive string may
4056 optionally be provided as a third list entry.
4060 .. code-block:: llvm
4062 ; Two scope domains:
4066 ; Some scopes in these domains:
4072 !5 = !{!4} ; A list containing only scope !4
4076 ; These two instructions don't alias:
4077 %0 = load float, float* %c, align 4, !alias.scope !5
4078 store float %0, float* %arrayidx.i, align 4, !noalias !5
4080 ; These two instructions also don't alias (for domain !1, the set of scopes
4081 ; in the !alias.scope equals that in the !noalias list):
4082 %2 = load float, float* %c, align 4, !alias.scope !5
4083 store float %2, float* %arrayidx.i2, align 4, !noalias !6
4085 ; These two instructions may alias (for domain !0, the set of scopes in
4086 ; the !noalias list is not a superset of, or equal to, the scopes in the
4087 ; !alias.scope list):
4088 %2 = load float, float* %c, align 4, !alias.scope !6
4089 store float %0, float* %arrayidx.i, align 4, !noalias !7
4091 '``fpmath``' Metadata
4092 ^^^^^^^^^^^^^^^^^^^^^
4094 ``fpmath`` metadata may be attached to any instruction of floating point
4095 type. It can be used to express the maximum acceptable error in the
4096 result of that instruction, in ULPs, thus potentially allowing the
4097 compiler to use a more efficient but less accurate method of computing
4098 it. ULP is defined as follows:
4100 If ``x`` is a real number that lies between two finite consecutive
4101 floating-point numbers ``a`` and ``b``, without being equal to one
4102 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
4103 distance between the two non-equal finite floating-point numbers
4104 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
4106 The metadata node shall consist of a single positive floating point
4107 number representing the maximum relative error, for example:
4109 .. code-block:: llvm
4111 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
4115 '``range``' Metadata
4116 ^^^^^^^^^^^^^^^^^^^^
4118 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
4119 integer types. It expresses the possible ranges the loaded value or the value
4120 returned by the called function at this call site is in. The ranges are
4121 represented with a flattened list of integers. The loaded value or the value
4122 returned is known to be in the union of the ranges defined by each consecutive
4123 pair. Each pair has the following properties:
4125 - The type must match the type loaded by the instruction.
4126 - The pair ``a,b`` represents the range ``[a,b)``.
4127 - Both ``a`` and ``b`` are constants.
4128 - The range is allowed to wrap.
4129 - The range should not represent the full or empty set. That is,
4132 In addition, the pairs must be in signed order of the lower bound and
4133 they must be non-contiguous.
4137 .. code-block:: llvm
4139 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
4140 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
4141 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
4142 %d = invoke i8 @bar() to label %cont
4143 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
4145 !0 = !{ i8 0, i8 2 }
4146 !1 = !{ i8 255, i8 2 }
4147 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
4148 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
4153 It is sometimes useful to attach information to loop constructs. Currently,
4154 loop metadata is implemented as metadata attached to the branch instruction
4155 in the loop latch block. This type of metadata refer to a metadata node that is
4156 guaranteed to be separate for each loop. The loop identifier metadata is
4157 specified with the name ``llvm.loop``.
4159 The loop identifier metadata is implemented using a metadata that refers to
4160 itself to avoid merging it with any other identifier metadata, e.g.,
4161 during module linkage or function inlining. That is, each loop should refer
4162 to their own identification metadata even if they reside in separate functions.
4163 The following example contains loop identifier metadata for two separate loop
4166 .. code-block:: llvm
4171 The loop identifier metadata can be used to specify additional
4172 per-loop metadata. Any operands after the first operand can be treated
4173 as user-defined metadata. For example the ``llvm.loop.unroll.count``
4174 suggests an unroll factor to the loop unroller:
4176 .. code-block:: llvm
4178 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
4181 !1 = !{!"llvm.loop.unroll.count", i32 4}
4183 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
4184 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4186 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
4187 used to control per-loop vectorization and interleaving parameters such as
4188 vectorization width and interleave count. These metadata should be used in
4189 conjunction with ``llvm.loop`` loop identification metadata. The
4190 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
4191 optimization hints and the optimizer will only interleave and vectorize loops if
4192 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
4193 which contains information about loop-carried memory dependencies can be helpful
4194 in determining the safety of these transformations.
4196 '``llvm.loop.interleave.count``' Metadata
4197 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4199 This metadata suggests an interleave count to the loop interleaver.
4200 The first operand is the string ``llvm.loop.interleave.count`` and the
4201 second operand is an integer specifying the interleave count. For
4204 .. code-block:: llvm
4206 !0 = !{!"llvm.loop.interleave.count", i32 4}
4208 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
4209 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
4210 then the interleave count will be determined automatically.
4212 '``llvm.loop.vectorize.enable``' Metadata
4213 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4215 This metadata selectively enables or disables vectorization for the loop. The
4216 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
4217 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
4218 0 disables vectorization:
4220 .. code-block:: llvm
4222 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
4223 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
4225 '``llvm.loop.vectorize.width``' Metadata
4226 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4228 This metadata sets the target width of the vectorizer. The first
4229 operand is the string ``llvm.loop.vectorize.width`` and the second
4230 operand is an integer specifying the width. For example:
4232 .. code-block:: llvm
4234 !0 = !{!"llvm.loop.vectorize.width", i32 4}
4236 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
4237 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
4238 0 or if the loop does not have this metadata the width will be
4239 determined automatically.
4241 '``llvm.loop.unroll``'
4242 ^^^^^^^^^^^^^^^^^^^^^^
4244 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
4245 optimization hints such as the unroll factor. ``llvm.loop.unroll``
4246 metadata should be used in conjunction with ``llvm.loop`` loop
4247 identification metadata. The ``llvm.loop.unroll`` metadata are only
4248 optimization hints and the unrolling will only be performed if the
4249 optimizer believes it is safe to do so.
4251 '``llvm.loop.unroll.count``' Metadata
4252 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4254 This metadata suggests an unroll factor to the loop unroller. The
4255 first operand is the string ``llvm.loop.unroll.count`` and the second
4256 operand is a positive integer specifying the unroll factor. For
4259 .. code-block:: llvm
4261 !0 = !{!"llvm.loop.unroll.count", i32 4}
4263 If the trip count of the loop is less than the unroll count the loop
4264 will be partially unrolled.
4266 '``llvm.loop.unroll.disable``' Metadata
4267 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4269 This metadata disables loop unrolling. The metadata has a single operand
4270 which is the string ``llvm.loop.unroll.disable``. For example:
4272 .. code-block:: llvm
4274 !0 = !{!"llvm.loop.unroll.disable"}
4276 '``llvm.loop.unroll.runtime.disable``' Metadata
4277 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4279 This metadata disables runtime loop unrolling. The metadata has a single
4280 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
4282 .. code-block:: llvm
4284 !0 = !{!"llvm.loop.unroll.runtime.disable"}
4286 '``llvm.loop.unroll.full``' Metadata
4287 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4289 This metadata suggests that the loop should be unrolled fully. The
4290 metadata has a single operand which is the string ``llvm.loop.unroll.full``.
4293 .. code-block:: llvm
4295 !0 = !{!"llvm.loop.unroll.full"}
4300 Metadata types used to annotate memory accesses with information helpful
4301 for optimizations are prefixed with ``llvm.mem``.
4303 '``llvm.mem.parallel_loop_access``' Metadata
4304 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4306 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
4307 or metadata containing a list of loop identifiers for nested loops.
4308 The metadata is attached to memory accessing instructions and denotes that
4309 no loop carried memory dependence exist between it and other instructions denoted
4310 with the same loop identifier.
4312 Precisely, given two instructions ``m1`` and ``m2`` that both have the
4313 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
4314 set of loops associated with that metadata, respectively, then there is no loop
4315 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
4318 As a special case, if all memory accessing instructions in a loop have
4319 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
4320 loop has no loop carried memory dependences and is considered to be a parallel
4323 Note that if not all memory access instructions have such metadata referring to
4324 the loop, then the loop is considered not being trivially parallel. Additional
4325 memory dependence analysis is required to make that determination. As a fail
4326 safe mechanism, this causes loops that were originally parallel to be considered
4327 sequential (if optimization passes that are unaware of the parallel semantics
4328 insert new memory instructions into the loop body).
4330 Example of a loop that is considered parallel due to its correct use of
4331 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
4332 metadata types that refer to the same loop identifier metadata.
4334 .. code-block:: llvm
4338 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
4340 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4342 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
4348 It is also possible to have nested parallel loops. In that case the
4349 memory accesses refer to a list of loop identifier metadata nodes instead of
4350 the loop identifier metadata node directly:
4352 .. code-block:: llvm
4356 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
4358 br label %inner.for.body
4362 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
4364 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
4366 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
4370 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
4372 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
4374 outer.for.end: ; preds = %for.body
4376 !0 = !{!1, !2} ; a list of loop identifiers
4377 !1 = !{!1} ; an identifier for the inner loop
4378 !2 = !{!2} ; an identifier for the outer loop
4383 The ``llvm.bitsets`` global metadata is used to implement
4384 :doc:`bitsets <BitSets>`.
4386 Module Flags Metadata
4387 =====================
4389 Information about the module as a whole is difficult to convey to LLVM's
4390 subsystems. The LLVM IR isn't sufficient to transmit this information.
4391 The ``llvm.module.flags`` named metadata exists in order to facilitate
4392 this. These flags are in the form of key / value pairs --- much like a
4393 dictionary --- making it easy for any subsystem who cares about a flag to
4396 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
4397 Each triplet has the following form:
4399 - The first element is a *behavior* flag, which specifies the behavior
4400 when two (or more) modules are merged together, and it encounters two
4401 (or more) metadata with the same ID. The supported behaviors are
4403 - The second element is a metadata string that is a unique ID for the
4404 metadata. Each module may only have one flag entry for each unique ID (not
4405 including entries with the **Require** behavior).
4406 - The third element is the value of the flag.
4408 When two (or more) modules are merged together, the resulting
4409 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
4410 each unique metadata ID string, there will be exactly one entry in the merged
4411 modules ``llvm.module.flags`` metadata table, and the value for that entry will
4412 be determined by the merge behavior flag, as described below. The only exception
4413 is that entries with the *Require* behavior are always preserved.
4415 The following behaviors are supported:
4426 Emits an error if two values disagree, otherwise the resulting value
4427 is that of the operands.
4431 Emits a warning if two values disagree. The result value will be the
4432 operand for the flag from the first module being linked.
4436 Adds a requirement that another module flag be present and have a
4437 specified value after linking is performed. The value must be a
4438 metadata pair, where the first element of the pair is the ID of the
4439 module flag to be restricted, and the second element of the pair is
4440 the value the module flag should be restricted to. This behavior can
4441 be used to restrict the allowable results (via triggering of an
4442 error) of linking IDs with the **Override** behavior.
4446 Uses the specified value, regardless of the behavior or value of the
4447 other module. If both modules specify **Override**, but the values
4448 differ, an error will be emitted.
4452 Appends the two values, which are required to be metadata nodes.
4456 Appends the two values, which are required to be metadata
4457 nodes. However, duplicate entries in the second list are dropped
4458 during the append operation.
4460 It is an error for a particular unique flag ID to have multiple behaviors,
4461 except in the case of **Require** (which adds restrictions on another metadata
4462 value) or **Override**.
4464 An example of module flags:
4466 .. code-block:: llvm
4468 !0 = !{ i32 1, !"foo", i32 1 }
4469 !1 = !{ i32 4, !"bar", i32 37 }
4470 !2 = !{ i32 2, !"qux", i32 42 }
4471 !3 = !{ i32 3, !"qux",
4476 !llvm.module.flags = !{ !0, !1, !2, !3 }
4478 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
4479 if two or more ``!"foo"`` flags are seen is to emit an error if their
4480 values are not equal.
4482 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
4483 behavior if two or more ``!"bar"`` flags are seen is to use the value
4486 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
4487 behavior if two or more ``!"qux"`` flags are seen is to emit a
4488 warning if their values are not equal.
4490 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
4496 The behavior is to emit an error if the ``llvm.module.flags`` does not
4497 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
4500 Objective-C Garbage Collection Module Flags Metadata
4501 ----------------------------------------------------
4503 On the Mach-O platform, Objective-C stores metadata about garbage
4504 collection in a special section called "image info". The metadata
4505 consists of a version number and a bitmask specifying what types of
4506 garbage collection are supported (if any) by the file. If two or more
4507 modules are linked together their garbage collection metadata needs to
4508 be merged rather than appended together.
4510 The Objective-C garbage collection module flags metadata consists of the
4511 following key-value pairs:
4520 * - ``Objective-C Version``
4521 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
4523 * - ``Objective-C Image Info Version``
4524 - **[Required]** --- The version of the image info section. Currently
4527 * - ``Objective-C Image Info Section``
4528 - **[Required]** --- The section to place the metadata. Valid values are
4529 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
4530 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
4531 Objective-C ABI version 2.
4533 * - ``Objective-C Garbage Collection``
4534 - **[Required]** --- Specifies whether garbage collection is supported or
4535 not. Valid values are 0, for no garbage collection, and 2, for garbage
4536 collection supported.
4538 * - ``Objective-C GC Only``
4539 - **[Optional]** --- Specifies that only garbage collection is supported.
4540 If present, its value must be 6. This flag requires that the
4541 ``Objective-C Garbage Collection`` flag have the value 2.
4543 Some important flag interactions:
4545 - If a module with ``Objective-C Garbage Collection`` set to 0 is
4546 merged with a module with ``Objective-C Garbage Collection`` set to
4547 2, then the resulting module has the
4548 ``Objective-C Garbage Collection`` flag set to 0.
4549 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
4550 merged with a module with ``Objective-C GC Only`` set to 6.
4552 Automatic Linker Flags Module Flags Metadata
4553 --------------------------------------------
4555 Some targets support embedding flags to the linker inside individual object
4556 files. Typically this is used in conjunction with language extensions which
4557 allow source files to explicitly declare the libraries they depend on, and have
4558 these automatically be transmitted to the linker via object files.
4560 These flags are encoded in the IR using metadata in the module flags section,
4561 using the ``Linker Options`` key. The merge behavior for this flag is required
4562 to be ``AppendUnique``, and the value for the key is expected to be a metadata
4563 node which should be a list of other metadata nodes, each of which should be a
4564 list of metadata strings defining linker options.
4566 For example, the following metadata section specifies two separate sets of
4567 linker options, presumably to link against ``libz`` and the ``Cocoa``
4570 !0 = !{ i32 6, !"Linker Options",
4573 !{ !"-framework", !"Cocoa" } } }
4574 !llvm.module.flags = !{ !0 }
4576 The metadata encoding as lists of lists of options, as opposed to a collapsed
4577 list of options, is chosen so that the IR encoding can use multiple option
4578 strings to specify e.g., a single library, while still having that specifier be
4579 preserved as an atomic element that can be recognized by a target specific
4580 assembly writer or object file emitter.
4582 Each individual option is required to be either a valid option for the target's
4583 linker, or an option that is reserved by the target specific assembly writer or
4584 object file emitter. No other aspect of these options is defined by the IR.
4586 C type width Module Flags Metadata
4587 ----------------------------------
4589 The ARM backend emits a section into each generated object file describing the
4590 options that it was compiled with (in a compiler-independent way) to prevent
4591 linking incompatible objects, and to allow automatic library selection. Some
4592 of these options are not visible at the IR level, namely wchar_t width and enum
4595 To pass this information to the backend, these options are encoded in module
4596 flags metadata, using the following key-value pairs:
4606 - * 0 --- sizeof(wchar_t) == 4
4607 * 1 --- sizeof(wchar_t) == 2
4610 - * 0 --- Enums are at least as large as an ``int``.
4611 * 1 --- Enums are stored in the smallest integer type which can
4612 represent all of its values.
4614 For example, the following metadata section specifies that the module was
4615 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
4616 enum is the smallest type which can represent all of its values::
4618 !llvm.module.flags = !{!0, !1}
4619 !0 = !{i32 1, !"short_wchar", i32 1}
4620 !1 = !{i32 1, !"short_enum", i32 0}
4622 .. _intrinsicglobalvariables:
4624 Intrinsic Global Variables
4625 ==========================
4627 LLVM has a number of "magic" global variables that contain data that
4628 affect code generation or other IR semantics. These are documented here.
4629 All globals of this sort should have a section specified as
4630 "``llvm.metadata``". This section and all globals that start with
4631 "``llvm.``" are reserved for use by LLVM.
4635 The '``llvm.used``' Global Variable
4636 -----------------------------------
4638 The ``@llvm.used`` global is an array which has
4639 :ref:`appending linkage <linkage_appending>`. This array contains a list of
4640 pointers to named global variables, functions and aliases which may optionally
4641 have a pointer cast formed of bitcast or getelementptr. For example, a legal
4644 .. code-block:: llvm
4649 @llvm.used = appending global [2 x i8*] [
4651 i8* bitcast (i32* @Y to i8*)
4652 ], section "llvm.metadata"
4654 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
4655 and linker are required to treat the symbol as if there is a reference to the
4656 symbol that it cannot see (which is why they have to be named). For example, if
4657 a variable has internal linkage and no references other than that from the
4658 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
4659 references from inline asms and other things the compiler cannot "see", and
4660 corresponds to "``attribute((used))``" in GNU C.
4662 On some targets, the code generator must emit a directive to the
4663 assembler or object file to prevent the assembler and linker from
4664 molesting the symbol.
4666 .. _gv_llvmcompilerused:
4668 The '``llvm.compiler.used``' Global Variable
4669 --------------------------------------------
4671 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4672 directive, except that it only prevents the compiler from touching the
4673 symbol. On targets that support it, this allows an intelligent linker to
4674 optimize references to the symbol without being impeded as it would be
4677 This is a rare construct that should only be used in rare circumstances,
4678 and should not be exposed to source languages.
4680 .. _gv_llvmglobalctors:
4682 The '``llvm.global_ctors``' Global Variable
4683 -------------------------------------------
4685 .. code-block:: llvm
4687 %0 = type { i32, void ()*, i8* }
4688 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
4690 The ``@llvm.global_ctors`` array contains a list of constructor
4691 functions, priorities, and an optional associated global or function.
4692 The functions referenced by this array will be called in ascending order
4693 of priority (i.e. lowest first) when the module is loaded. The order of
4694 functions with the same priority is not defined.
4696 If the third field is present, non-null, and points to a global variable
4697 or function, the initializer function will only run if the associated
4698 data from the current module is not discarded.
4700 .. _llvmglobaldtors:
4702 The '``llvm.global_dtors``' Global Variable
4703 -------------------------------------------
4705 .. code-block:: llvm
4707 %0 = type { i32, void ()*, i8* }
4708 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
4710 The ``@llvm.global_dtors`` array contains a list of destructor
4711 functions, priorities, and an optional associated global or function.
4712 The functions referenced by this array will be called in descending
4713 order of priority (i.e. highest first) when the module is unloaded. The
4714 order of functions with the same priority is not defined.
4716 If the third field is present, non-null, and points to a global variable
4717 or function, the destructor function will only run if the associated
4718 data from the current module is not discarded.
4720 Instruction Reference
4721 =====================
4723 The LLVM instruction set consists of several different classifications
4724 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
4725 instructions <binaryops>`, :ref:`bitwise binary
4726 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
4727 :ref:`other instructions <otherops>`.
4731 Terminator Instructions
4732 -----------------------
4734 As mentioned :ref:`previously <functionstructure>`, every basic block in a
4735 program ends with a "Terminator" instruction, which indicates which
4736 block should be executed after the current block is finished. These
4737 terminator instructions typically yield a '``void``' value: they produce
4738 control flow, not values (the one exception being the
4739 ':ref:`invoke <i_invoke>`' instruction).
4741 The terminator instructions are: ':ref:`ret <i_ret>`',
4742 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
4743 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
4744 ':ref:`resume <i_resume>`', ':ref:`catchpad <i_catchpad>`',
4745 ':ref:`catchendpad <i_catchendpad>`',
4746 ':ref:`catchret <i_catchret>`',
4747 ':ref:`cleanupret <i_cleanupret>`',
4748 ':ref:`terminatepad <i_terminatepad>`',
4749 and ':ref:`unreachable <i_unreachable>`'.
4753 '``ret``' Instruction
4754 ^^^^^^^^^^^^^^^^^^^^^
4761 ret <type> <value> ; Return a value from a non-void function
4762 ret void ; Return from void function
4767 The '``ret``' instruction is used to return control flow (and optionally
4768 a value) from a function back to the caller.
4770 There are two forms of the '``ret``' instruction: one that returns a
4771 value and then causes control flow, and one that just causes control
4777 The '``ret``' instruction optionally accepts a single argument, the
4778 return value. The type of the return value must be a ':ref:`first
4779 class <t_firstclass>`' type.
4781 A function is not :ref:`well formed <wellformed>` if it it has a non-void
4782 return type and contains a '``ret``' instruction with no return value or
4783 a return value with a type that does not match its type, or if it has a
4784 void return type and contains a '``ret``' instruction with a return
4790 When the '``ret``' instruction is executed, control flow returns back to
4791 the calling function's context. If the caller is a
4792 ":ref:`call <i_call>`" instruction, execution continues at the
4793 instruction after the call. If the caller was an
4794 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
4795 beginning of the "normal" destination block. If the instruction returns
4796 a value, that value shall set the call or invoke instruction's return
4802 .. code-block:: llvm
4804 ret i32 5 ; Return an integer value of 5
4805 ret void ; Return from a void function
4806 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
4810 '``br``' Instruction
4811 ^^^^^^^^^^^^^^^^^^^^
4818 br i1 <cond>, label <iftrue>, label <iffalse>
4819 br label <dest> ; Unconditional branch
4824 The '``br``' instruction is used to cause control flow to transfer to a
4825 different basic block in the current function. There are two forms of
4826 this instruction, corresponding to a conditional branch and an
4827 unconditional branch.
4832 The conditional branch form of the '``br``' instruction takes a single
4833 '``i1``' value and two '``label``' values. The unconditional form of the
4834 '``br``' instruction takes a single '``label``' value as a target.
4839 Upon execution of a conditional '``br``' instruction, the '``i1``'
4840 argument is evaluated. If the value is ``true``, control flows to the
4841 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
4842 to the '``iffalse``' ``label`` argument.
4847 .. code-block:: llvm
4850 %cond = icmp eq i32 %a, %b
4851 br i1 %cond, label %IfEqual, label %IfUnequal
4859 '``switch``' Instruction
4860 ^^^^^^^^^^^^^^^^^^^^^^^^
4867 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
4872 The '``switch``' instruction is used to transfer control flow to one of
4873 several different places. It is a generalization of the '``br``'
4874 instruction, allowing a branch to occur to one of many possible
4880 The '``switch``' instruction uses three parameters: an integer
4881 comparison value '``value``', a default '``label``' destination, and an
4882 array of pairs of comparison value constants and '``label``'s. The table
4883 is not allowed to contain duplicate constant entries.
4888 The ``switch`` instruction specifies a table of values and destinations.
4889 When the '``switch``' instruction is executed, this table is searched
4890 for the given value. If the value is found, control flow is transferred
4891 to the corresponding destination; otherwise, control flow is transferred
4892 to the default destination.
4897 Depending on properties of the target machine and the particular
4898 ``switch`` instruction, this instruction may be code generated in
4899 different ways. For example, it could be generated as a series of
4900 chained conditional branches or with a lookup table.
4905 .. code-block:: llvm
4907 ; Emulate a conditional br instruction
4908 %Val = zext i1 %value to i32
4909 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
4911 ; Emulate an unconditional br instruction
4912 switch i32 0, label %dest [ ]
4914 ; Implement a jump table:
4915 switch i32 %val, label %otherwise [ i32 0, label %onzero
4917 i32 2, label %ontwo ]
4921 '``indirectbr``' Instruction
4922 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4929 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
4934 The '``indirectbr``' instruction implements an indirect branch to a
4935 label within the current function, whose address is specified by
4936 "``address``". Address must be derived from a
4937 :ref:`blockaddress <blockaddress>` constant.
4942 The '``address``' argument is the address of the label to jump to. The
4943 rest of the arguments indicate the full set of possible destinations
4944 that the address may point to. Blocks are allowed to occur multiple
4945 times in the destination list, though this isn't particularly useful.
4947 This destination list is required so that dataflow analysis has an
4948 accurate understanding of the CFG.
4953 Control transfers to the block specified in the address argument. All
4954 possible destination blocks must be listed in the label list, otherwise
4955 this instruction has undefined behavior. This implies that jumps to
4956 labels defined in other functions have undefined behavior as well.
4961 This is typically implemented with a jump through a register.
4966 .. code-block:: llvm
4968 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
4972 '``invoke``' Instruction
4973 ^^^^^^^^^^^^^^^^^^^^^^^^
4980 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
4981 to label <normal label> unwind label <exception label>
4986 The '``invoke``' instruction causes control to transfer to a specified
4987 function, with the possibility of control flow transfer to either the
4988 '``normal``' label or the '``exception``' label. If the callee function
4989 returns with the "``ret``" instruction, control flow will return to the
4990 "normal" label. If the callee (or any indirect callees) returns via the
4991 ":ref:`resume <i_resume>`" instruction or other exception handling
4992 mechanism, control is interrupted and continued at the dynamically
4993 nearest "exception" label.
4995 The '``exception``' label is a `landing
4996 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
4997 '``exception``' label is required to have the
4998 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
4999 information about the behavior of the program after unwinding happens,
5000 as its first non-PHI instruction. The restrictions on the
5001 "``landingpad``" instruction's tightly couples it to the "``invoke``"
5002 instruction, so that the important information contained within the
5003 "``landingpad``" instruction can't be lost through normal code motion.
5008 This instruction requires several arguments:
5010 #. The optional "cconv" marker indicates which :ref:`calling
5011 convention <callingconv>` the call should use. If none is
5012 specified, the call defaults to using C calling conventions.
5013 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5014 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5016 #. '``ptr to function ty``': shall be the signature of the pointer to
5017 function value being invoked. In most cases, this is a direct
5018 function invocation, but indirect ``invoke``'s are just as possible,
5019 branching off an arbitrary pointer to function value.
5020 #. '``function ptr val``': An LLVM value containing a pointer to a
5021 function to be invoked.
5022 #. '``function args``': argument list whose types match the function
5023 signature argument types and parameter attributes. All arguments must
5024 be of :ref:`first class <t_firstclass>` type. If the function signature
5025 indicates the function accepts a variable number of arguments, the
5026 extra arguments can be specified.
5027 #. '``normal label``': the label reached when the called function
5028 executes a '``ret``' instruction.
5029 #. '``exception label``': the label reached when a callee returns via
5030 the :ref:`resume <i_resume>` instruction or other exception handling
5032 #. The optional :ref:`function attributes <fnattrs>` list. Only
5033 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5034 attributes are valid here.
5039 This instruction is designed to operate as a standard '``call``'
5040 instruction in most regards. The primary difference is that it
5041 establishes an association with a label, which is used by the runtime
5042 library to unwind the stack.
5044 This instruction is used in languages with destructors to ensure that
5045 proper cleanup is performed in the case of either a ``longjmp`` or a
5046 thrown exception. Additionally, this is important for implementation of
5047 '``catch``' clauses in high-level languages that support them.
5049 For the purposes of the SSA form, the definition of the value returned
5050 by the '``invoke``' instruction is deemed to occur on the edge from the
5051 current block to the "normal" label. If the callee unwinds then no
5052 return value is available.
5057 .. code-block:: llvm
5059 %retval = invoke i32 @Test(i32 15) to label %Continue
5060 unwind label %TestCleanup ; i32:retval set
5061 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
5062 unwind label %TestCleanup ; i32:retval set
5066 '``resume``' Instruction
5067 ^^^^^^^^^^^^^^^^^^^^^^^^
5074 resume <type> <value>
5079 The '``resume``' instruction is a terminator instruction that has no
5085 The '``resume``' instruction requires one argument, which must have the
5086 same type as the result of any '``landingpad``' instruction in the same
5092 The '``resume``' instruction resumes propagation of an existing
5093 (in-flight) exception whose unwinding was interrupted with a
5094 :ref:`landingpad <i_landingpad>` instruction.
5099 .. code-block:: llvm
5101 resume { i8*, i32 } %exn
5105 '``catchpad``' Instruction
5106 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5113 <resultval> = catchpad <resultty> [<args>*]
5114 to label <normal label> unwind label <exception label>
5119 The '``catchpad``' instruction is used by `LLVM's exception handling
5120 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5121 is a catch block --- one where a personality routine attempts to transfer
5122 control to catch an exception.
5123 The ``args`` correspond to whatever information the personality
5124 routine requires to know if this is an appropriate place to catch the
5125 exception. Control is tranfered to the ``exception`` label if the
5126 ``catchpad`` is not an appropriate handler for the in-flight exception.
5127 The ``normal`` label should contain the code found in the ``catch``
5128 portion of a ``try``/``catch`` sequence. It defines values supplied by
5129 the :ref:`personality function <personalityfn>` upon re-entry to the
5130 function. The ``resultval`` has the type ``resultty``.
5135 The instruction takes a list of arbitrary values which are interpreted
5136 by the :ref:`personality function <personalityfn>`.
5138 The ``catchpad`` must be provided a ``normal`` label to transfer control
5139 to if the ``catchpad`` matches the exception and an ``exception``
5140 label to transfer control to if it doesn't.
5145 The '``catchpad``' instruction defines the values which are set by the
5146 :ref:`personality function <personalityfn>` upon re-entry to the function, and
5147 therefore the "result type" of the ``catchpad`` instruction. As with
5148 calling conventions, how the personality function results are
5149 represented in LLVM IR is target specific.
5151 When the call stack is being unwound due to an exception being thrown,
5152 the exception is compared against the ``args``. If it doesn't match,
5153 then control is transfered to the ``exception`` basic block.
5155 The ``catchpad`` instruction has several restrictions:
5157 - A catch block is a basic block which is the unwind destination of
5158 an exceptional instruction.
5159 - A catch block must have a '``catchpad``' instruction as its
5160 first non-PHI instruction.
5161 - A catch block's ``exception`` edge must refer to a catch block or a
5163 - There can be only one '``catchpad``' instruction within the
5165 - A basic block that is not a catch block may not include a
5166 '``catchpad``' instruction.
5167 - It is undefined behavior for control to transfer from a ``catchpad`` to a
5168 ``cleanupret`` without first executing a ``catchret`` and a subsequent
5170 - It is undefined behavior for control to transfer from a ``catchpad`` to a
5171 ``ret`` without first executing a ``catchret``.
5176 .. code-block:: llvm
5178 ;; A catch block which can catch an integer.
5179 %res = catchpad { i8*, i32 } [i8** @_ZTIi]
5180 to label %int.handler unwind label %terminate
5184 '``catchendpad``' Instruction
5185 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5192 catchendpad unwind label <nextaction>
5193 catchendpad unwind to caller
5198 The '``catchendpad``' instruction is used by `LLVM's exception handling
5199 system <ExceptionHandling.html#overview>`_ to communicate to the
5200 :ref:`personality function <personalityfn>` which invokes are associated
5201 with a chain of :ref:`catchpad <i_catchpad>` instructions.
5203 The ``nextaction`` label indicates where control should transfer to if
5204 none of the ``catchpad`` instructions are suitable for catching the
5205 in-flight exception.
5207 If a ``nextaction`` label is not present, the instruction unwinds out of
5208 its parent function. The
5209 :ref:`personality function <personalityfn>` will continue processing
5210 exception handling actions in the caller.
5215 The instruction optionally takes a label, ``nextaction``, indicating
5216 where control should transfer to if none of the preceding
5217 ``catchpad`` instructions are suitable for the in-flight exception.
5222 When the call stack is being unwound due to an exception being thrown
5223 and none of the constituent ``catchpad`` instructions match, then
5224 control is transfered to ``nextaction`` if it is present. If it is not
5225 present, control is transfered to the caller.
5227 The ``catchendpad`` instruction has several restrictions:
5229 - A catch-end block is a basic block which is the unwind destination of
5230 an exceptional instruction.
5231 - A catch-end block must have a '``catchendpad``' instruction as its
5232 first non-PHI instruction.
5233 - There can be only one '``catchendpad``' instruction within the
5235 - A basic block that is not a catch-end block may not include a
5236 '``catchendpad``' instruction.
5237 - Exactly one catch block may unwind to a ``catchendpad``.
5238 - The unwind target of invokes between a ``catchpad`` and a
5239 corresponding ``catchret`` must be its ``catchendpad``.
5244 .. code-block:: llvm
5246 catchendpad unwind label %terminate
5247 catchendpad unwind to caller
5251 '``catchret``' Instruction
5252 ^^^^^^^^^^^^^^^^^^^^^^^^^^
5259 catchret label <normal>
5264 The '``catchret``' instruction is a terminator instruction that has a
5271 The '``catchret``' instruction requires one argument which specifies
5272 where control will transfer to next.
5277 The '``catchret``' instruction ends the existing (in-flight) exception
5278 whose unwinding was interrupted with a
5279 :ref:`catchpad <i_catchpad>` instruction.
5280 The :ref:`personality function <personalityfn>` gets a chance to execute
5281 arbitrary code to, for example, run a C++ destructor.
5282 Control then transfers to ``normal``.
5287 .. code-block:: llvm
5289 catchret label %continue
5293 '``cleanupret``' Instruction
5294 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5301 cleanupret <type> <value> unwind label <continue>
5302 cleanupret <type> <value> unwind to caller
5307 The '``cleanupret``' instruction is a terminator instruction that has
5308 an optional successor.
5314 The '``cleanupret``' instruction requires one argument, which must have the
5315 same type as the result of any '``cleanuppad``' instruction in the same
5316 function. It also has an optional successor, ``continue``.
5321 The '``cleanupret``' instruction indicates to the
5322 :ref:`personality function <personalityfn>` that one
5323 :ref:`cleanuppad <i_cleanuppad>` it transferred control to has ended.
5324 It transfers control to ``continue`` or unwinds out of the function.
5329 .. code-block:: llvm
5331 cleanupret void unwind to caller
5332 cleanupret { i8*, i32 } %exn unwind label %continue
5336 '``terminatepad``' Instruction
5337 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5344 terminatepad [<args>*] unwind label <exception label>
5345 terminatepad [<args>*] unwind to caller
5350 The '``terminatepad``' instruction is used by `LLVM's exception handling
5351 system <ExceptionHandling.html#overview>`_ to specify that a basic block
5352 is a terminate block --- one where a personality routine may decide to
5353 terminate the program.
5354 The ``args`` correspond to whatever information the personality
5355 routine requires to know if this is an appropriate place to terminate the
5356 program. Control is transferred to the ``exception`` label if the
5357 personality routine decides not to terminate the program for the
5358 in-flight exception.
5363 The instruction takes a list of arbitrary values which are interpreted
5364 by the :ref:`personality function <personalityfn>`.
5366 The ``terminatepad`` may be given an ``exception`` label to
5367 transfer control to if the in-flight exception matches the ``args``.
5372 When the call stack is being unwound due to an exception being thrown,
5373 the exception is compared against the ``args``. If it matches,
5374 then control is transfered to the ``exception`` basic block. Otherwise,
5375 the program is terminated via personality-specific means. Typically,
5376 the first argument to ``terminatepad`` specifies what function the
5377 personality should defer to in order to terminate the program.
5379 The ``terminatepad`` instruction has several restrictions:
5381 - A terminate block is a basic block which is the unwind destination of
5382 an exceptional instruction.
5383 - A terminate block must have a '``terminatepad``' instruction as its
5384 first non-PHI instruction.
5385 - There can be only one '``terminatepad``' instruction within the
5387 - A basic block that is not a terminate block may not include a
5388 '``terminatepad``' instruction.
5393 .. code-block:: llvm
5395 ;; A terminate block which only permits integers.
5396 terminatepad [i8** @_ZTIi] unwind label %continue
5400 '``unreachable``' Instruction
5401 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5413 The '``unreachable``' instruction has no defined semantics. This
5414 instruction is used to inform the optimizer that a particular portion of
5415 the code is not reachable. This can be used to indicate that the code
5416 after a no-return function cannot be reached, and other facts.
5421 The '``unreachable``' instruction has no defined semantics.
5428 Binary operators are used to do most of the computation in a program.
5429 They require two operands of the same type, execute an operation on
5430 them, and produce a single value. The operands might represent multiple
5431 data, as is the case with the :ref:`vector <t_vector>` data type. The
5432 result value has the same type as its operands.
5434 There are several different binary operators:
5438 '``add``' Instruction
5439 ^^^^^^^^^^^^^^^^^^^^^
5446 <result> = add <ty> <op1>, <op2> ; yields ty:result
5447 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
5448 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
5449 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
5454 The '``add``' instruction returns the sum of its two operands.
5459 The two arguments to the '``add``' instruction must be
5460 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5461 arguments must have identical types.
5466 The value produced is the integer sum of the two operands.
5468 If the sum has unsigned overflow, the result returned is the
5469 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5472 Because LLVM integers use a two's complement representation, this
5473 instruction is appropriate for both signed and unsigned integers.
5475 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5476 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5477 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
5478 unsigned and/or signed overflow, respectively, occurs.
5483 .. code-block:: llvm
5485 <result> = add i32 4, %var ; yields i32:result = 4 + %var
5489 '``fadd``' Instruction
5490 ^^^^^^^^^^^^^^^^^^^^^^
5497 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5502 The '``fadd``' instruction returns the sum of its two operands.
5507 The two arguments to the '``fadd``' instruction must be :ref:`floating
5508 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5509 Both arguments must have identical types.
5514 The value produced is the floating point sum of the two operands. This
5515 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
5516 which are optimization hints to enable otherwise unsafe floating point
5522 .. code-block:: llvm
5524 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
5526 '``sub``' Instruction
5527 ^^^^^^^^^^^^^^^^^^^^^
5534 <result> = sub <ty> <op1>, <op2> ; yields ty:result
5535 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
5536 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
5537 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
5542 The '``sub``' instruction returns the difference of its two operands.
5544 Note that the '``sub``' instruction is used to represent the '``neg``'
5545 instruction present in most other intermediate representations.
5550 The two arguments to the '``sub``' instruction must be
5551 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5552 arguments must have identical types.
5557 The value produced is the integer difference of the two operands.
5559 If the difference has unsigned overflow, the result returned is the
5560 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5563 Because LLVM integers use a two's complement representation, this
5564 instruction is appropriate for both signed and unsigned integers.
5566 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5567 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5568 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
5569 unsigned and/or signed overflow, respectively, occurs.
5574 .. code-block:: llvm
5576 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
5577 <result> = sub i32 0, %val ; yields i32:result = -%var
5581 '``fsub``' Instruction
5582 ^^^^^^^^^^^^^^^^^^^^^^
5589 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5594 The '``fsub``' instruction returns the difference of its two operands.
5596 Note that the '``fsub``' instruction is used to represent the '``fneg``'
5597 instruction present in most other intermediate representations.
5602 The two arguments to the '``fsub``' instruction must be :ref:`floating
5603 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5604 Both arguments must have identical types.
5609 The value produced is the floating point difference of the two operands.
5610 This instruction can also take any number of :ref:`fast-math
5611 flags <fastmath>`, which are optimization hints to enable otherwise
5612 unsafe floating point optimizations:
5617 .. code-block:: llvm
5619 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
5620 <result> = fsub float -0.0, %val ; yields float:result = -%var
5622 '``mul``' Instruction
5623 ^^^^^^^^^^^^^^^^^^^^^
5630 <result> = mul <ty> <op1>, <op2> ; yields ty:result
5631 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
5632 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
5633 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
5638 The '``mul``' instruction returns the product of its two operands.
5643 The two arguments to the '``mul``' instruction must be
5644 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5645 arguments must have identical types.
5650 The value produced is the integer product of the two operands.
5652 If the result of the multiplication has unsigned overflow, the result
5653 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
5654 bit width of the result.
5656 Because LLVM integers use a two's complement representation, and the
5657 result is the same width as the operands, this instruction returns the
5658 correct result for both signed and unsigned integers. If a full product
5659 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
5660 sign-extended or zero-extended as appropriate to the width of the full
5663 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5664 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5665 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
5666 unsigned and/or signed overflow, respectively, occurs.
5671 .. code-block:: llvm
5673 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
5677 '``fmul``' Instruction
5678 ^^^^^^^^^^^^^^^^^^^^^^
5685 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5690 The '``fmul``' instruction returns the product of its two operands.
5695 The two arguments to the '``fmul``' instruction must be :ref:`floating
5696 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5697 Both arguments must have identical types.
5702 The value produced is the floating point product of the two operands.
5703 This instruction can also take any number of :ref:`fast-math
5704 flags <fastmath>`, which are optimization hints to enable otherwise
5705 unsafe floating point optimizations:
5710 .. code-block:: llvm
5712 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
5714 '``udiv``' Instruction
5715 ^^^^^^^^^^^^^^^^^^^^^^
5722 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
5723 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
5728 The '``udiv``' instruction returns the quotient of its two operands.
5733 The two arguments to the '``udiv``' instruction must be
5734 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5735 arguments must have identical types.
5740 The value produced is the unsigned integer quotient of the two operands.
5742 Note that unsigned integer division and signed integer division are
5743 distinct operations; for signed integer division, use '``sdiv``'.
5745 Division by zero leads to undefined behavior.
5747 If the ``exact`` keyword is present, the result value of the ``udiv`` is
5748 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
5749 such, "((a udiv exact b) mul b) == a").
5754 .. code-block:: llvm
5756 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
5758 '``sdiv``' Instruction
5759 ^^^^^^^^^^^^^^^^^^^^^^
5766 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
5767 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
5772 The '``sdiv``' instruction returns the quotient of its two operands.
5777 The two arguments to the '``sdiv``' instruction must be
5778 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5779 arguments must have identical types.
5784 The value produced is the signed integer quotient of the two operands
5785 rounded towards zero.
5787 Note that signed integer division and unsigned integer division are
5788 distinct operations; for unsigned integer division, use '``udiv``'.
5790 Division by zero leads to undefined behavior. Overflow also leads to
5791 undefined behavior; this is a rare case, but can occur, for example, by
5792 doing a 32-bit division of -2147483648 by -1.
5794 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
5795 a :ref:`poison value <poisonvalues>` if the result would be rounded.
5800 .. code-block:: llvm
5802 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
5806 '``fdiv``' Instruction
5807 ^^^^^^^^^^^^^^^^^^^^^^
5814 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5819 The '``fdiv``' instruction returns the quotient of its two operands.
5824 The two arguments to the '``fdiv``' instruction must be :ref:`floating
5825 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5826 Both arguments must have identical types.
5831 The value produced is the floating point quotient of the two operands.
5832 This instruction can also take any number of :ref:`fast-math
5833 flags <fastmath>`, which are optimization hints to enable otherwise
5834 unsafe floating point optimizations:
5839 .. code-block:: llvm
5841 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
5843 '``urem``' Instruction
5844 ^^^^^^^^^^^^^^^^^^^^^^
5851 <result> = urem <ty> <op1>, <op2> ; yields ty:result
5856 The '``urem``' instruction returns the remainder from the unsigned
5857 division of its two arguments.
5862 The two arguments to the '``urem``' instruction must be
5863 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5864 arguments must have identical types.
5869 This instruction returns the unsigned integer *remainder* of a division.
5870 This instruction always performs an unsigned division to get the
5873 Note that unsigned integer remainder and signed integer remainder are
5874 distinct operations; for signed integer remainder, use '``srem``'.
5876 Taking the remainder of a division by zero leads to undefined behavior.
5881 .. code-block:: llvm
5883 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
5885 '``srem``' Instruction
5886 ^^^^^^^^^^^^^^^^^^^^^^
5893 <result> = srem <ty> <op1>, <op2> ; yields ty:result
5898 The '``srem``' instruction returns the remainder from the signed
5899 division of its two operands. This instruction can also take
5900 :ref:`vector <t_vector>` versions of the values in which case the elements
5906 The two arguments to the '``srem``' instruction must be
5907 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5908 arguments must have identical types.
5913 This instruction returns the *remainder* of a division (where the result
5914 is either zero or has the same sign as the dividend, ``op1``), not the
5915 *modulo* operator (where the result is either zero or has the same sign
5916 as the divisor, ``op2``) of a value. For more information about the
5917 difference, see `The Math
5918 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
5919 table of how this is implemented in various languages, please see
5921 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
5923 Note that signed integer remainder and unsigned integer remainder are
5924 distinct operations; for unsigned integer remainder, use '``urem``'.
5926 Taking the remainder of a division by zero leads to undefined behavior.
5927 Overflow also leads to undefined behavior; this is a rare case, but can
5928 occur, for example, by taking the remainder of a 32-bit division of
5929 -2147483648 by -1. (The remainder doesn't actually overflow, but this
5930 rule lets srem be implemented using instructions that return both the
5931 result of the division and the remainder.)
5936 .. code-block:: llvm
5938 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
5942 '``frem``' Instruction
5943 ^^^^^^^^^^^^^^^^^^^^^^
5950 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5955 The '``frem``' instruction returns the remainder from the division of
5961 The two arguments to the '``frem``' instruction must be :ref:`floating
5962 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5963 Both arguments must have identical types.
5968 This instruction returns the *remainder* of a division. The remainder
5969 has the same sign as the dividend. This instruction can also take any
5970 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
5971 to enable otherwise unsafe floating point optimizations:
5976 .. code-block:: llvm
5978 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
5982 Bitwise Binary Operations
5983 -------------------------
5985 Bitwise binary operators are used to do various forms of bit-twiddling
5986 in a program. They are generally very efficient instructions and can
5987 commonly be strength reduced from other instructions. They require two
5988 operands of the same type, execute an operation on them, and produce a
5989 single value. The resulting value is the same type as its operands.
5991 '``shl``' Instruction
5992 ^^^^^^^^^^^^^^^^^^^^^
5999 <result> = shl <ty> <op1>, <op2> ; yields ty:result
6000 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
6001 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
6002 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
6007 The '``shl``' instruction returns the first operand shifted to the left
6008 a specified number of bits.
6013 Both arguments to the '``shl``' instruction must be the same
6014 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6015 '``op2``' is treated as an unsigned value.
6020 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
6021 where ``n`` is the width of the result. If ``op2`` is (statically or
6022 dynamically) equal to or larger than the number of bits in
6023 ``op1``, the result is undefined. If the arguments are vectors, each
6024 vector element of ``op1`` is shifted by the corresponding shift amount
6027 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
6028 value <poisonvalues>` if it shifts out any non-zero bits. If the
6029 ``nsw`` keyword is present, then the shift produces a :ref:`poison
6030 value <poisonvalues>` if it shifts out any bits that disagree with the
6031 resultant sign bit. As such, NUW/NSW have the same semantics as they
6032 would if the shift were expressed as a mul instruction with the same
6033 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
6038 .. code-block:: llvm
6040 <result> = shl i32 4, %var ; yields i32: 4 << %var
6041 <result> = shl i32 4, 2 ; yields i32: 16
6042 <result> = shl i32 1, 10 ; yields i32: 1024
6043 <result> = shl i32 1, 32 ; undefined
6044 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
6046 '``lshr``' Instruction
6047 ^^^^^^^^^^^^^^^^^^^^^^
6054 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
6055 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
6060 The '``lshr``' instruction (logical shift right) returns the first
6061 operand shifted to the right a specified number of bits with zero fill.
6066 Both arguments to the '``lshr``' instruction must be the same
6067 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6068 '``op2``' is treated as an unsigned value.
6073 This instruction always performs a logical shift right operation. The
6074 most significant bits of the result will be filled with zero bits after
6075 the shift. If ``op2`` is (statically or dynamically) equal to or larger
6076 than the number of bits in ``op1``, the result is undefined. If the
6077 arguments are vectors, each vector element of ``op1`` is shifted by the
6078 corresponding shift amount in ``op2``.
6080 If the ``exact`` keyword is present, the result value of the ``lshr`` is
6081 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6087 .. code-block:: llvm
6089 <result> = lshr i32 4, 1 ; yields i32:result = 2
6090 <result> = lshr i32 4, 2 ; yields i32:result = 1
6091 <result> = lshr i8 4, 3 ; yields i8:result = 0
6092 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
6093 <result> = lshr i32 1, 32 ; undefined
6094 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
6096 '``ashr``' Instruction
6097 ^^^^^^^^^^^^^^^^^^^^^^
6104 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
6105 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
6110 The '``ashr``' instruction (arithmetic shift right) returns the first
6111 operand shifted to the right a specified number of bits with sign
6117 Both arguments to the '``ashr``' instruction must be the same
6118 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6119 '``op2``' is treated as an unsigned value.
6124 This instruction always performs an arithmetic shift right operation,
6125 The most significant bits of the result will be filled with the sign bit
6126 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
6127 than the number of bits in ``op1``, the result is undefined. If the
6128 arguments are vectors, each vector element of ``op1`` is shifted by the
6129 corresponding shift amount in ``op2``.
6131 If the ``exact`` keyword is present, the result value of the ``ashr`` is
6132 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6138 .. code-block:: llvm
6140 <result> = ashr i32 4, 1 ; yields i32:result = 2
6141 <result> = ashr i32 4, 2 ; yields i32:result = 1
6142 <result> = ashr i8 4, 3 ; yields i8:result = 0
6143 <result> = ashr i8 -2, 1 ; yields i8:result = -1
6144 <result> = ashr i32 1, 32 ; undefined
6145 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
6147 '``and``' Instruction
6148 ^^^^^^^^^^^^^^^^^^^^^
6155 <result> = and <ty> <op1>, <op2> ; yields ty:result
6160 The '``and``' instruction returns the bitwise logical and of its two
6166 The two arguments to the '``and``' instruction must be
6167 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6168 arguments must have identical types.
6173 The truth table used for the '``and``' instruction is:
6190 .. code-block:: llvm
6192 <result> = and i32 4, %var ; yields i32:result = 4 & %var
6193 <result> = and i32 15, 40 ; yields i32:result = 8
6194 <result> = and i32 4, 8 ; yields i32:result = 0
6196 '``or``' Instruction
6197 ^^^^^^^^^^^^^^^^^^^^
6204 <result> = or <ty> <op1>, <op2> ; yields ty:result
6209 The '``or``' instruction returns the bitwise logical inclusive or of its
6215 The two arguments to the '``or``' instruction must be
6216 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6217 arguments must have identical types.
6222 The truth table used for the '``or``' instruction is:
6241 <result> = or i32 4, %var ; yields i32:result = 4 | %var
6242 <result> = or i32 15, 40 ; yields i32:result = 47
6243 <result> = or i32 4, 8 ; yields i32:result = 12
6245 '``xor``' Instruction
6246 ^^^^^^^^^^^^^^^^^^^^^
6253 <result> = xor <ty> <op1>, <op2> ; yields ty:result
6258 The '``xor``' instruction returns the bitwise logical exclusive or of
6259 its two operands. The ``xor`` is used to implement the "one's
6260 complement" operation, which is the "~" operator in C.
6265 The two arguments to the '``xor``' instruction must be
6266 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6267 arguments must have identical types.
6272 The truth table used for the '``xor``' instruction is:
6289 .. code-block:: llvm
6291 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
6292 <result> = xor i32 15, 40 ; yields i32:result = 39
6293 <result> = xor i32 4, 8 ; yields i32:result = 12
6294 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
6299 LLVM supports several instructions to represent vector operations in a
6300 target-independent manner. These instructions cover the element-access
6301 and vector-specific operations needed to process vectors effectively.
6302 While LLVM does directly support these vector operations, many
6303 sophisticated algorithms will want to use target-specific intrinsics to
6304 take full advantage of a specific target.
6306 .. _i_extractelement:
6308 '``extractelement``' Instruction
6309 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6316 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
6321 The '``extractelement``' instruction extracts a single scalar element
6322 from a vector at a specified index.
6327 The first operand of an '``extractelement``' instruction is a value of
6328 :ref:`vector <t_vector>` type. The second operand is an index indicating
6329 the position from which to extract the element. The index may be a
6330 variable of any integer type.
6335 The result is a scalar of the same type as the element type of ``val``.
6336 Its value is the value at position ``idx`` of ``val``. If ``idx``
6337 exceeds the length of ``val``, the results are undefined.
6342 .. code-block:: llvm
6344 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
6346 .. _i_insertelement:
6348 '``insertelement``' Instruction
6349 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6356 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
6361 The '``insertelement``' instruction inserts a scalar element into a
6362 vector at a specified index.
6367 The first operand of an '``insertelement``' instruction is a value of
6368 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
6369 type must equal the element type of the first operand. The third operand
6370 is an index indicating the position at which to insert the value. The
6371 index may be a variable of any integer type.
6376 The result is a vector of the same type as ``val``. Its element values
6377 are those of ``val`` except at position ``idx``, where it gets the value
6378 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
6384 .. code-block:: llvm
6386 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
6388 .. _i_shufflevector:
6390 '``shufflevector``' Instruction
6391 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6398 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
6403 The '``shufflevector``' instruction constructs a permutation of elements
6404 from two input vectors, returning a vector with the same element type as
6405 the input and length that is the same as the shuffle mask.
6410 The first two operands of a '``shufflevector``' instruction are vectors
6411 with the same type. The third argument is a shuffle mask whose element
6412 type is always 'i32'. The result of the instruction is a vector whose
6413 length is the same as the shuffle mask and whose element type is the
6414 same as the element type of the first two operands.
6416 The shuffle mask operand is required to be a constant vector with either
6417 constant integer or undef values.
6422 The elements of the two input vectors are numbered from left to right
6423 across both of the vectors. The shuffle mask operand specifies, for each
6424 element of the result vector, which element of the two input vectors the
6425 result element gets. The element selector may be undef (meaning "don't
6426 care") and the second operand may be undef if performing a shuffle from
6432 .. code-block:: llvm
6434 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6435 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
6436 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
6437 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
6438 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
6439 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
6440 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6441 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
6443 Aggregate Operations
6444 --------------------
6446 LLVM supports several instructions for working with
6447 :ref:`aggregate <t_aggregate>` values.
6451 '``extractvalue``' Instruction
6452 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6459 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
6464 The '``extractvalue``' instruction extracts the value of a member field
6465 from an :ref:`aggregate <t_aggregate>` value.
6470 The first operand of an '``extractvalue``' instruction is a value of
6471 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
6472 constant indices to specify which value to extract in a similar manner
6473 as indices in a '``getelementptr``' instruction.
6475 The major differences to ``getelementptr`` indexing are:
6477 - Since the value being indexed is not a pointer, the first index is
6478 omitted and assumed to be zero.
6479 - At least one index must be specified.
6480 - Not only struct indices but also array indices must be in bounds.
6485 The result is the value at the position in the aggregate specified by
6491 .. code-block:: llvm
6493 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
6497 '``insertvalue``' Instruction
6498 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6505 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
6510 The '``insertvalue``' instruction inserts a value into a member field in
6511 an :ref:`aggregate <t_aggregate>` value.
6516 The first operand of an '``insertvalue``' instruction is a value of
6517 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
6518 a first-class value to insert. The following operands are constant
6519 indices indicating the position at which to insert the value in a
6520 similar manner as indices in a '``extractvalue``' instruction. The value
6521 to insert must have the same type as the value identified by the
6527 The result is an aggregate of the same type as ``val``. Its value is
6528 that of ``val`` except that the value at the position specified by the
6529 indices is that of ``elt``.
6534 .. code-block:: llvm
6536 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
6537 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
6538 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
6542 Memory Access and Addressing Operations
6543 ---------------------------------------
6545 A key design point of an SSA-based representation is how it represents
6546 memory. In LLVM, no memory locations are in SSA form, which makes things
6547 very simple. This section describes how to read, write, and allocate
6552 '``alloca``' Instruction
6553 ^^^^^^^^^^^^^^^^^^^^^^^^
6560 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
6565 The '``alloca``' instruction allocates memory on the stack frame of the
6566 currently executing function, to be automatically released when this
6567 function returns to its caller. The object is always allocated in the
6568 generic address space (address space zero).
6573 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
6574 bytes of memory on the runtime stack, returning a pointer of the
6575 appropriate type to the program. If "NumElements" is specified, it is
6576 the number of elements allocated, otherwise "NumElements" is defaulted
6577 to be one. If a constant alignment is specified, the value result of the
6578 allocation is guaranteed to be aligned to at least that boundary. The
6579 alignment may not be greater than ``1 << 29``. If not specified, or if
6580 zero, the target can choose to align the allocation on any convenient
6581 boundary compatible with the type.
6583 '``type``' may be any sized type.
6588 Memory is allocated; a pointer is returned. The operation is undefined
6589 if there is insufficient stack space for the allocation. '``alloca``'d
6590 memory is automatically released when the function returns. The
6591 '``alloca``' instruction is commonly used to represent automatic
6592 variables that must have an address available. When the function returns
6593 (either with the ``ret`` or ``resume`` instructions), the memory is
6594 reclaimed. Allocating zero bytes is legal, but the result is undefined.
6595 The order in which memory is allocated (ie., which way the stack grows)
6601 .. code-block:: llvm
6603 %ptr = alloca i32 ; yields i32*:ptr
6604 %ptr = alloca i32, i32 4 ; yields i32*:ptr
6605 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
6606 %ptr = alloca i32, align 1024 ; yields i32*:ptr
6610 '``load``' Instruction
6611 ^^^^^^^^^^^^^^^^^^^^^^
6618 <result> = load [volatile] <ty>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>][, !nonnull !<index>][, !dereferenceable !<index>][, !dereferenceable_or_null !<index>]
6619 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
6620 !<index> = !{ i32 1 }
6625 The '``load``' instruction is used to read from memory.
6630 The argument to the ``load`` instruction specifies the memory address
6631 from which to load. The type specified must be a :ref:`first
6632 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
6633 then the optimizer is not allowed to modify the number or order of
6634 execution of this ``load`` with other :ref:`volatile
6635 operations <volatile>`.
6637 If the ``load`` is marked as ``atomic``, it takes an extra
6638 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
6639 ``release`` and ``acq_rel`` orderings are not valid on ``load``
6640 instructions. Atomic loads produce :ref:`defined <memmodel>` results
6641 when they may see multiple atomic stores. The type of the pointee must
6642 be an integer type whose bit width is a power of two greater than or
6643 equal to eight and less than or equal to a target-specific size limit.
6644 ``align`` must be explicitly specified on atomic loads, and the load has
6645 undefined behavior if the alignment is not set to a value which is at
6646 least the size in bytes of the pointee. ``!nontemporal`` does not have
6647 any defined semantics for atomic loads.
6649 The optional constant ``align`` argument specifies the alignment of the
6650 operation (that is, the alignment of the memory address). A value of 0
6651 or an omitted ``align`` argument means that the operation has the ABI
6652 alignment for the target. It is the responsibility of the code emitter
6653 to ensure that the alignment information is correct. Overestimating the
6654 alignment results in undefined behavior. Underestimating the alignment
6655 may produce less efficient code. An alignment of 1 is always safe. The
6656 maximum possible alignment is ``1 << 29``.
6658 The optional ``!nontemporal`` metadata must reference a single
6659 metadata name ``<index>`` corresponding to a metadata node with one
6660 ``i32`` entry of value 1. The existence of the ``!nontemporal``
6661 metadata on the instruction tells the optimizer and code generator
6662 that this load is not expected to be reused in the cache. The code
6663 generator may select special instructions to save cache bandwidth, such
6664 as the ``MOVNT`` instruction on x86.
6666 The optional ``!invariant.load`` metadata must reference a single
6667 metadata name ``<index>`` corresponding to a metadata node with no
6668 entries. The existence of the ``!invariant.load`` metadata on the
6669 instruction tells the optimizer and code generator that the address
6670 operand to this load points to memory which can be assumed unchanged.
6671 Being invariant does not imply that a location is dereferenceable,
6672 but it does imply that once the location is known dereferenceable
6673 its value is henceforth unchanging.
6675 The optional ``!nonnull`` metadata must reference a single
6676 metadata name ``<index>`` corresponding to a metadata node with no
6677 entries. The existence of the ``!nonnull`` metadata on the
6678 instruction tells the optimizer that the value loaded is known to
6679 never be null. This is analogous to the ''nonnull'' attribute
6680 on parameters and return values. This metadata can only be applied
6681 to loads of a pointer type.
6683 The optional ``!dereferenceable`` metadata must reference a single
6684 metadata name ``<index>`` corresponding to a metadata node with one ``i64``
6685 entry. The existence of the ``!dereferenceable`` metadata on the instruction
6686 tells the optimizer that the value loaded is known to be dereferenceable.
6687 The number of bytes known to be dereferenceable is specified by the integer
6688 value in the metadata node. This is analogous to the ''dereferenceable''
6689 attribute on parameters and return values. This metadata can only be applied
6690 to loads of a pointer type.
6692 The optional ``!dereferenceable_or_null`` metadata must reference a single
6693 metadata name ``<index>`` corresponding to a metadata node with one ``i64``
6694 entry. The existence of the ``!dereferenceable_or_null`` metadata on the
6695 instruction tells the optimizer that the value loaded is known to be either
6696 dereferenceable or null.
6697 The number of bytes known to be dereferenceable is specified by the integer
6698 value in the metadata node. This is analogous to the ''dereferenceable_or_null''
6699 attribute on parameters and return values. This metadata can only be applied
6700 to loads of a pointer type.
6705 The location of memory pointed to is loaded. If the value being loaded
6706 is of scalar type then the number of bytes read does not exceed the
6707 minimum number of bytes needed to hold all bits of the type. For
6708 example, loading an ``i24`` reads at most three bytes. When loading a
6709 value of a type like ``i20`` with a size that is not an integral number
6710 of bytes, the result is undefined if the value was not originally
6711 written using a store of the same type.
6716 .. code-block:: llvm
6718 %ptr = alloca i32 ; yields i32*:ptr
6719 store i32 3, i32* %ptr ; yields void
6720 %val = load i32, i32* %ptr ; yields i32:val = i32 3
6724 '``store``' Instruction
6725 ^^^^^^^^^^^^^^^^^^^^^^^
6732 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields void
6733 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields void
6738 The '``store``' instruction is used to write to memory.
6743 There are two arguments to the ``store`` instruction: a value to store
6744 and an address at which to store it. The type of the ``<pointer>``
6745 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
6746 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
6747 then the optimizer is not allowed to modify the number or order of
6748 execution of this ``store`` with other :ref:`volatile
6749 operations <volatile>`.
6751 If the ``store`` is marked as ``atomic``, it takes an extra
6752 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
6753 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
6754 instructions. Atomic loads produce :ref:`defined <memmodel>` results
6755 when they may see multiple atomic stores. The type of the pointee must
6756 be an integer type whose bit width is a power of two greater than or
6757 equal to eight and less than or equal to a target-specific size limit.
6758 ``align`` must be explicitly specified on atomic stores, and the store
6759 has undefined behavior if the alignment is not set to a value which is
6760 at least the size in bytes of the pointee. ``!nontemporal`` does not
6761 have any defined semantics for atomic stores.
6763 The optional constant ``align`` argument specifies the alignment of the
6764 operation (that is, the alignment of the memory address). A value of 0
6765 or an omitted ``align`` argument means that the operation has the ABI
6766 alignment for the target. It is the responsibility of the code emitter
6767 to ensure that the alignment information is correct. Overestimating the
6768 alignment results in undefined behavior. Underestimating the
6769 alignment may produce less efficient code. An alignment of 1 is always
6770 safe. The maximum possible alignment is ``1 << 29``.
6772 The optional ``!nontemporal`` metadata must reference a single metadata
6773 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
6774 value 1. The existence of the ``!nontemporal`` metadata on the instruction
6775 tells the optimizer and code generator that this load is not expected to
6776 be reused in the cache. The code generator may select special
6777 instructions to save cache bandwidth, such as the MOVNT instruction on
6783 The contents of memory are updated to contain ``<value>`` at the
6784 location specified by the ``<pointer>`` operand. If ``<value>`` is
6785 of scalar type then the number of bytes written does not exceed the
6786 minimum number of bytes needed to hold all bits of the type. For
6787 example, storing an ``i24`` writes at most three bytes. When writing a
6788 value of a type like ``i20`` with a size that is not an integral number
6789 of bytes, it is unspecified what happens to the extra bits that do not
6790 belong to the type, but they will typically be overwritten.
6795 .. code-block:: llvm
6797 %ptr = alloca i32 ; yields i32*:ptr
6798 store i32 3, i32* %ptr ; yields void
6799 %val = load i32* %ptr ; yields i32:val = i32 3
6803 '``fence``' Instruction
6804 ^^^^^^^^^^^^^^^^^^^^^^^
6811 fence [singlethread] <ordering> ; yields void
6816 The '``fence``' instruction is used to introduce happens-before edges
6822 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
6823 defines what *synchronizes-with* edges they add. They can only be given
6824 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
6829 A fence A which has (at least) ``release`` ordering semantics
6830 *synchronizes with* a fence B with (at least) ``acquire`` ordering
6831 semantics if and only if there exist atomic operations X and Y, both
6832 operating on some atomic object M, such that A is sequenced before X, X
6833 modifies M (either directly or through some side effect of a sequence
6834 headed by X), Y is sequenced before B, and Y observes M. This provides a
6835 *happens-before* dependency between A and B. Rather than an explicit
6836 ``fence``, one (but not both) of the atomic operations X or Y might
6837 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
6838 still *synchronize-with* the explicit ``fence`` and establish the
6839 *happens-before* edge.
6841 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
6842 ``acquire`` and ``release`` semantics specified above, participates in
6843 the global program order of other ``seq_cst`` operations and/or fences.
6845 The optional ":ref:`singlethread <singlethread>`" argument specifies
6846 that the fence only synchronizes with other fences in the same thread.
6847 (This is useful for interacting with signal handlers.)
6852 .. code-block:: llvm
6854 fence acquire ; yields void
6855 fence singlethread seq_cst ; yields void
6859 '``cmpxchg``' Instruction
6860 ^^^^^^^^^^^^^^^^^^^^^^^^^
6867 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
6872 The '``cmpxchg``' instruction is used to atomically modify memory. It
6873 loads a value in memory and compares it to a given value. If they are
6874 equal, it tries to store a new value into the memory.
6879 There are three arguments to the '``cmpxchg``' instruction: an address
6880 to operate on, a value to compare to the value currently be at that
6881 address, and a new value to place at that address if the compared values
6882 are equal. The type of '<cmp>' must be an integer type whose bit width
6883 is a power of two greater than or equal to eight and less than or equal
6884 to a target-specific size limit. '<cmp>' and '<new>' must have the same
6885 type, and the type of '<pointer>' must be a pointer to that type. If the
6886 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
6887 to modify the number or order of execution of this ``cmpxchg`` with
6888 other :ref:`volatile operations <volatile>`.
6890 The success and failure :ref:`ordering <ordering>` arguments specify how this
6891 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
6892 must be at least ``monotonic``, the ordering constraint on failure must be no
6893 stronger than that on success, and the failure ordering cannot be either
6894 ``release`` or ``acq_rel``.
6896 The optional "``singlethread``" argument declares that the ``cmpxchg``
6897 is only atomic with respect to code (usually signal handlers) running in
6898 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
6899 respect to all other code in the system.
6901 The pointer passed into cmpxchg must have alignment greater than or
6902 equal to the size in memory of the operand.
6907 The contents of memory at the location specified by the '``<pointer>``' operand
6908 is read and compared to '``<cmp>``'; if the read value is the equal, the
6909 '``<new>``' is written. The original value at the location is returned, together
6910 with a flag indicating success (true) or failure (false).
6912 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
6913 permitted: the operation may not write ``<new>`` even if the comparison
6916 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
6917 if the value loaded equals ``cmp``.
6919 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
6920 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
6921 load with an ordering parameter determined the second ordering parameter.
6926 .. code-block:: llvm
6929 %orig = atomic load i32, i32* %ptr unordered ; yields i32
6933 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
6934 %squared = mul i32 %cmp, %cmp
6935 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
6936 %value_loaded = extractvalue { i32, i1 } %val_success, 0
6937 %success = extractvalue { i32, i1 } %val_success, 1
6938 br i1 %success, label %done, label %loop
6945 '``atomicrmw``' Instruction
6946 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6953 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
6958 The '``atomicrmw``' instruction is used to atomically modify memory.
6963 There are three arguments to the '``atomicrmw``' instruction: an
6964 operation to apply, an address whose value to modify, an argument to the
6965 operation. The operation must be one of the following keywords:
6979 The type of '<value>' must be an integer type whose bit width is a power
6980 of two greater than or equal to eight and less than or equal to a
6981 target-specific size limit. The type of the '``<pointer>``' operand must
6982 be a pointer to that type. If the ``atomicrmw`` is marked as
6983 ``volatile``, then the optimizer is not allowed to modify the number or
6984 order of execution of this ``atomicrmw`` with other :ref:`volatile
6985 operations <volatile>`.
6990 The contents of memory at the location specified by the '``<pointer>``'
6991 operand are atomically read, modified, and written back. The original
6992 value at the location is returned. The modification is specified by the
6995 - xchg: ``*ptr = val``
6996 - add: ``*ptr = *ptr + val``
6997 - sub: ``*ptr = *ptr - val``
6998 - and: ``*ptr = *ptr & val``
6999 - nand: ``*ptr = ~(*ptr & val)``
7000 - or: ``*ptr = *ptr | val``
7001 - xor: ``*ptr = *ptr ^ val``
7002 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
7003 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
7004 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
7006 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
7012 .. code-block:: llvm
7014 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
7016 .. _i_getelementptr:
7018 '``getelementptr``' Instruction
7019 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7026 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7027 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7028 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
7033 The '``getelementptr``' instruction is used to get the address of a
7034 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
7035 address calculation only and does not access memory. The instruction can also
7036 be used to calculate a vector of such addresses.
7041 The first argument is always a type used as the basis for the calculations.
7042 The second argument is always a pointer or a vector of pointers, and is the
7043 base address to start from. The remaining arguments are indices
7044 that indicate which of the elements of the aggregate object are indexed.
7045 The interpretation of each index is dependent on the type being indexed
7046 into. The first index always indexes the pointer value given as the
7047 first argument, the second index indexes a value of the type pointed to
7048 (not necessarily the value directly pointed to, since the first index
7049 can be non-zero), etc. The first type indexed into must be a pointer
7050 value, subsequent types can be arrays, vectors, and structs. Note that
7051 subsequent types being indexed into can never be pointers, since that
7052 would require loading the pointer before continuing calculation.
7054 The type of each index argument depends on the type it is indexing into.
7055 When indexing into a (optionally packed) structure, only ``i32`` integer
7056 **constants** are allowed (when using a vector of indices they must all
7057 be the **same** ``i32`` integer constant). When indexing into an array,
7058 pointer or vector, integers of any width are allowed, and they are not
7059 required to be constant. These integers are treated as signed values
7062 For example, let's consider a C code fragment and how it gets compiled
7078 int *foo(struct ST *s) {
7079 return &s[1].Z.B[5][13];
7082 The LLVM code generated by Clang is:
7084 .. code-block:: llvm
7086 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
7087 %struct.ST = type { i32, double, %struct.RT }
7089 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
7091 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
7098 In the example above, the first index is indexing into the
7099 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
7100 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
7101 indexes into the third element of the structure, yielding a
7102 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
7103 structure. The third index indexes into the second element of the
7104 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
7105 dimensions of the array are subscripted into, yielding an '``i32``'
7106 type. The '``getelementptr``' instruction returns a pointer to this
7107 element, thus computing a value of '``i32*``' type.
7109 Note that it is perfectly legal to index partially through a structure,
7110 returning a pointer to an inner element. Because of this, the LLVM code
7111 for the given testcase is equivalent to:
7113 .. code-block:: llvm
7115 define i32* @foo(%struct.ST* %s) {
7116 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
7117 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
7118 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
7119 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
7120 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
7124 If the ``inbounds`` keyword is present, the result value of the
7125 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
7126 pointer is not an *in bounds* address of an allocated object, or if any
7127 of the addresses that would be formed by successive addition of the
7128 offsets implied by the indices to the base address with infinitely
7129 precise signed arithmetic are not an *in bounds* address of that
7130 allocated object. The *in bounds* addresses for an allocated object are
7131 all the addresses that point into the object, plus the address one byte
7132 past the end. In cases where the base is a vector of pointers the
7133 ``inbounds`` keyword applies to each of the computations element-wise.
7135 If the ``inbounds`` keyword is not present, the offsets are added to the
7136 base address with silently-wrapping two's complement arithmetic. If the
7137 offsets have a different width from the pointer, they are sign-extended
7138 or truncated to the width of the pointer. The result value of the
7139 ``getelementptr`` may be outside the object pointed to by the base
7140 pointer. The result value may not necessarily be used to access memory
7141 though, even if it happens to point into allocated storage. See the
7142 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
7145 The getelementptr instruction is often confusing. For some more insight
7146 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
7151 .. code-block:: llvm
7153 ; yields [12 x i8]*:aptr
7154 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
7156 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
7158 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
7160 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
7165 The ``getelementptr`` returns a vector of pointers, instead of a single address,
7166 when one or more of its arguments is a vector. In such cases, all vector
7167 arguments should have the same number of elements, and every scalar argument
7168 will be effectively broadcast into a vector during address calculation.
7170 .. code-block:: llvm
7172 ; All arguments are vectors:
7173 ; A[i] = ptrs[i] + offsets[i]*sizeof(i8)
7174 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets
7176 ; Add the same scalar offset to each pointer of a vector:
7177 ; A[i] = ptrs[i] + offset*sizeof(i8)
7178 %A = getelementptr i8, <4 x i8*> %ptrs, i64 %offset
7180 ; Add distinct offsets to the same pointer:
7181 ; A[i] = ptr + offsets[i]*sizeof(i8)
7182 %A = getelementptr i8, i8* %ptr, <4 x i64> %offsets
7184 ; In all cases described above the type of the result is <4 x i8*>
7186 The two following instructions are equivalent:
7188 .. code-block:: llvm
7190 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7191 <4 x i32> <i32 2, i32 2, i32 2, i32 2>,
7192 <4 x i32> <i32 1, i32 1, i32 1, i32 1>,
7194 <4 x i64> <i64 13, i64 13, i64 13, i64 13>
7196 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7197 i32 2, i32 1, <4 x i32> %ind4, i64 13
7199 Let's look at the C code, where the vector version of ``getelementptr``
7204 // Let's assume that we vectorize the following loop:
7205 double *A, B; int *C;
7206 for (int i = 0; i < size; ++i) {
7210 .. code-block:: llvm
7212 ; get pointers for 8 elements from array B
7213 %ptrs = getelementptr double, double* %B, <8 x i32> %C
7214 ; load 8 elements from array B into A
7215 %A = call <8 x double> @llvm.masked.gather.v8f64(<8 x double*> %ptrs,
7216 i32 8, <8 x i1> %mask, <8 x double> %passthru)
7218 Conversion Operations
7219 ---------------------
7221 The instructions in this category are the conversion instructions
7222 (casting) which all take a single operand and a type. They perform
7223 various bit conversions on the operand.
7225 '``trunc .. to``' Instruction
7226 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7233 <result> = trunc <ty> <value> to <ty2> ; yields ty2
7238 The '``trunc``' instruction truncates its operand to the type ``ty2``.
7243 The '``trunc``' instruction takes a value to trunc, and a type to trunc
7244 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
7245 of the same number of integers. The bit size of the ``value`` must be
7246 larger than the bit size of the destination type, ``ty2``. Equal sized
7247 types are not allowed.
7252 The '``trunc``' instruction truncates the high order bits in ``value``
7253 and converts the remaining bits to ``ty2``. Since the source size must
7254 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
7255 It will always truncate bits.
7260 .. code-block:: llvm
7262 %X = trunc i32 257 to i8 ; yields i8:1
7263 %Y = trunc i32 123 to i1 ; yields i1:true
7264 %Z = trunc i32 122 to i1 ; yields i1:false
7265 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
7267 '``zext .. to``' Instruction
7268 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7275 <result> = zext <ty> <value> to <ty2> ; yields ty2
7280 The '``zext``' instruction zero extends its operand to type ``ty2``.
7285 The '``zext``' instruction takes a value to cast, and a type to cast it
7286 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7287 the same number of integers. The bit size of the ``value`` must be
7288 smaller than the bit size of the destination type, ``ty2``.
7293 The ``zext`` fills the high order bits of the ``value`` with zero bits
7294 until it reaches the size of the destination type, ``ty2``.
7296 When zero extending from i1, the result will always be either 0 or 1.
7301 .. code-block:: llvm
7303 %X = zext i32 257 to i64 ; yields i64:257
7304 %Y = zext i1 true to i32 ; yields i32:1
7305 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7307 '``sext .. to``' Instruction
7308 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7315 <result> = sext <ty> <value> to <ty2> ; yields ty2
7320 The '``sext``' sign extends ``value`` to the type ``ty2``.
7325 The '``sext``' instruction takes a value to cast, and a type to cast it
7326 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7327 the same number of integers. The bit size of the ``value`` must be
7328 smaller than the bit size of the destination type, ``ty2``.
7333 The '``sext``' instruction performs a sign extension by copying the sign
7334 bit (highest order bit) of the ``value`` until it reaches the bit size
7335 of the type ``ty2``.
7337 When sign extending from i1, the extension always results in -1 or 0.
7342 .. code-block:: llvm
7344 %X = sext i8 -1 to i16 ; yields i16 :65535
7345 %Y = sext i1 true to i32 ; yields i32:-1
7346 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7348 '``fptrunc .. to``' Instruction
7349 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7356 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
7361 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
7366 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
7367 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
7368 The size of ``value`` must be larger than the size of ``ty2``. This
7369 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
7374 The '``fptrunc``' instruction truncates a ``value`` from a larger
7375 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
7376 point <t_floating>` type. If the value cannot fit within the
7377 destination type, ``ty2``, then the results are undefined.
7382 .. code-block:: llvm
7384 %X = fptrunc double 123.0 to float ; yields float:123.0
7385 %Y = fptrunc double 1.0E+300 to float ; yields undefined
7387 '``fpext .. to``' Instruction
7388 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7395 <result> = fpext <ty> <value> to <ty2> ; yields ty2
7400 The '``fpext``' extends a floating point ``value`` to a larger floating
7406 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
7407 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
7408 to. The source type must be smaller than the destination type.
7413 The '``fpext``' instruction extends the ``value`` from a smaller
7414 :ref:`floating point <t_floating>` type to a larger :ref:`floating
7415 point <t_floating>` type. The ``fpext`` cannot be used to make a
7416 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
7417 *no-op cast* for a floating point cast.
7422 .. code-block:: llvm
7424 %X = fpext float 3.125 to double ; yields double:3.125000e+00
7425 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
7427 '``fptoui .. to``' Instruction
7428 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7435 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
7440 The '``fptoui``' converts a floating point ``value`` to its unsigned
7441 integer equivalent of type ``ty2``.
7446 The '``fptoui``' instruction takes a value to cast, which must be a
7447 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7448 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7449 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7450 type with the same number of elements as ``ty``
7455 The '``fptoui``' instruction converts its :ref:`floating
7456 point <t_floating>` operand into the nearest (rounding towards zero)
7457 unsigned integer value. If the value cannot fit in ``ty2``, the results
7463 .. code-block:: llvm
7465 %X = fptoui double 123.0 to i32 ; yields i32:123
7466 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
7467 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
7469 '``fptosi .. to``' Instruction
7470 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7477 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
7482 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
7483 ``value`` to type ``ty2``.
7488 The '``fptosi``' instruction takes a value to cast, which must be a
7489 scalar or vector :ref:`floating point <t_floating>` value, and a type to
7490 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7491 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7492 type with the same number of elements as ``ty``
7497 The '``fptosi``' instruction converts its :ref:`floating
7498 point <t_floating>` operand into the nearest (rounding towards zero)
7499 signed integer value. If the value cannot fit in ``ty2``, the results
7505 .. code-block:: llvm
7507 %X = fptosi double -123.0 to i32 ; yields i32:-123
7508 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
7509 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
7511 '``uitofp .. to``' Instruction
7512 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7519 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
7524 The '``uitofp``' instruction regards ``value`` as an unsigned integer
7525 and converts that value to the ``ty2`` type.
7530 The '``uitofp``' instruction takes a value to cast, which must be a
7531 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7532 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7533 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7534 type with the same number of elements as ``ty``
7539 The '``uitofp``' instruction interprets its operand as an unsigned
7540 integer quantity and converts it to the corresponding floating point
7541 value. If the value cannot fit in the floating point value, the results
7547 .. code-block:: llvm
7549 %X = uitofp i32 257 to float ; yields float:257.0
7550 %Y = uitofp i8 -1 to double ; yields double:255.0
7552 '``sitofp .. to``' Instruction
7553 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7560 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
7565 The '``sitofp``' instruction regards ``value`` as a signed integer and
7566 converts that value to the ``ty2`` type.
7571 The '``sitofp``' instruction takes a value to cast, which must be a
7572 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7573 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7574 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7575 type with the same number of elements as ``ty``
7580 The '``sitofp``' instruction interprets its operand as a signed integer
7581 quantity and converts it to the corresponding floating point value. If
7582 the value cannot fit in the floating point value, the results are
7588 .. code-block:: llvm
7590 %X = sitofp i32 257 to float ; yields float:257.0
7591 %Y = sitofp i8 -1 to double ; yields double:-1.0
7595 '``ptrtoint .. to``' Instruction
7596 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7603 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
7608 The '``ptrtoint``' instruction converts the pointer or a vector of
7609 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
7614 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
7615 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
7616 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
7617 a vector of integers type.
7622 The '``ptrtoint``' instruction converts ``value`` to integer type
7623 ``ty2`` by interpreting the pointer value as an integer and either
7624 truncating or zero extending that value to the size of the integer type.
7625 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
7626 ``value`` is larger than ``ty2`` then a truncation is done. If they are
7627 the same size, then nothing is done (*no-op cast*) other than a type
7633 .. code-block:: llvm
7635 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
7636 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
7637 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
7641 '``inttoptr .. to``' Instruction
7642 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7649 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
7654 The '``inttoptr``' instruction converts an integer ``value`` to a
7655 pointer type, ``ty2``.
7660 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
7661 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
7667 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
7668 applying either a zero extension or a truncation depending on the size
7669 of the integer ``value``. If ``value`` is larger than the size of a
7670 pointer then a truncation is done. If ``value`` is smaller than the size
7671 of a pointer then a zero extension is done. If they are the same size,
7672 nothing is done (*no-op cast*).
7677 .. code-block:: llvm
7679 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
7680 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
7681 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
7682 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
7686 '``bitcast .. to``' Instruction
7687 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7694 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
7699 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
7705 The '``bitcast``' instruction takes a value to cast, which must be a
7706 non-aggregate first class value, and a type to cast it to, which must
7707 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
7708 bit sizes of ``value`` and the destination type, ``ty2``, must be
7709 identical. If the source type is a pointer, the destination type must
7710 also be a pointer of the same size. This instruction supports bitwise
7711 conversion of vectors to integers and to vectors of other types (as
7712 long as they have the same size).
7717 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
7718 is always a *no-op cast* because no bits change with this
7719 conversion. The conversion is done as if the ``value`` had been stored
7720 to memory and read back as type ``ty2``. Pointer (or vector of
7721 pointers) types may only be converted to other pointer (or vector of
7722 pointers) types with the same address space through this instruction.
7723 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
7724 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
7729 .. code-block:: llvm
7731 %X = bitcast i8 255 to i8 ; yields i8 :-1
7732 %Y = bitcast i32* %x to sint* ; yields sint*:%x
7733 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
7734 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
7736 .. _i_addrspacecast:
7738 '``addrspacecast .. to``' Instruction
7739 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7746 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
7751 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
7752 address space ``n`` to type ``pty2`` in address space ``m``.
7757 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
7758 to cast and a pointer type to cast it to, which must have a different
7764 The '``addrspacecast``' instruction converts the pointer value
7765 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
7766 value modification, depending on the target and the address space
7767 pair. Pointer conversions within the same address space must be
7768 performed with the ``bitcast`` instruction. Note that if the address space
7769 conversion is legal then both result and operand refer to the same memory
7775 .. code-block:: llvm
7777 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
7778 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
7779 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
7786 The instructions in this category are the "miscellaneous" instructions,
7787 which defy better classification.
7791 '``icmp``' Instruction
7792 ^^^^^^^^^^^^^^^^^^^^^^
7799 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
7804 The '``icmp``' instruction returns a boolean value or a vector of
7805 boolean values based on comparison of its two integer, integer vector,
7806 pointer, or pointer vector operands.
7811 The '``icmp``' instruction takes three operands. The first operand is
7812 the condition code indicating the kind of comparison to perform. It is
7813 not a value, just a keyword. The possible condition code are:
7816 #. ``ne``: not equal
7817 #. ``ugt``: unsigned greater than
7818 #. ``uge``: unsigned greater or equal
7819 #. ``ult``: unsigned less than
7820 #. ``ule``: unsigned less or equal
7821 #. ``sgt``: signed greater than
7822 #. ``sge``: signed greater or equal
7823 #. ``slt``: signed less than
7824 #. ``sle``: signed less or equal
7826 The remaining two arguments must be :ref:`integer <t_integer>` or
7827 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
7828 must also be identical types.
7833 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
7834 code given as ``cond``. The comparison performed always yields either an
7835 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
7837 #. ``eq``: yields ``true`` if the operands are equal, ``false``
7838 otherwise. No sign interpretation is necessary or performed.
7839 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
7840 otherwise. No sign interpretation is necessary or performed.
7841 #. ``ugt``: interprets the operands as unsigned values and yields
7842 ``true`` if ``op1`` is greater than ``op2``.
7843 #. ``uge``: interprets the operands as unsigned values and yields
7844 ``true`` if ``op1`` is greater than or equal to ``op2``.
7845 #. ``ult``: interprets the operands as unsigned values and yields
7846 ``true`` if ``op1`` is less than ``op2``.
7847 #. ``ule``: interprets the operands as unsigned values and yields
7848 ``true`` if ``op1`` is less than or equal to ``op2``.
7849 #. ``sgt``: interprets the operands as signed values and yields ``true``
7850 if ``op1`` is greater than ``op2``.
7851 #. ``sge``: interprets the operands as signed values and yields ``true``
7852 if ``op1`` is greater than or equal to ``op2``.
7853 #. ``slt``: interprets the operands as signed values and yields ``true``
7854 if ``op1`` is less than ``op2``.
7855 #. ``sle``: interprets the operands as signed values and yields ``true``
7856 if ``op1`` is less than or equal to ``op2``.
7858 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
7859 are compared as if they were integers.
7861 If the operands are integer vectors, then they are compared element by
7862 element. The result is an ``i1`` vector with the same number of elements
7863 as the values being compared. Otherwise, the result is an ``i1``.
7868 .. code-block:: llvm
7870 <result> = icmp eq i32 4, 5 ; yields: result=false
7871 <result> = icmp ne float* %X, %X ; yields: result=false
7872 <result> = icmp ult i16 4, 5 ; yields: result=true
7873 <result> = icmp sgt i16 4, 5 ; yields: result=false
7874 <result> = icmp ule i16 -4, 5 ; yields: result=false
7875 <result> = icmp sge i16 4, 5 ; yields: result=false
7877 Note that the code generator does not yet support vector types with the
7878 ``icmp`` instruction.
7882 '``fcmp``' Instruction
7883 ^^^^^^^^^^^^^^^^^^^^^^
7890 <result> = fcmp [fast-math flags]* <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
7895 The '``fcmp``' instruction returns a boolean value or vector of boolean
7896 values based on comparison of its operands.
7898 If the operands are floating point scalars, then the result type is a
7899 boolean (:ref:`i1 <t_integer>`).
7901 If the operands are floating point vectors, then the result type is a
7902 vector of boolean with the same number of elements as the operands being
7908 The '``fcmp``' instruction takes three operands. The first operand is
7909 the condition code indicating the kind of comparison to perform. It is
7910 not a value, just a keyword. The possible condition code are:
7912 #. ``false``: no comparison, always returns false
7913 #. ``oeq``: ordered and equal
7914 #. ``ogt``: ordered and greater than
7915 #. ``oge``: ordered and greater than or equal
7916 #. ``olt``: ordered and less than
7917 #. ``ole``: ordered and less than or equal
7918 #. ``one``: ordered and not equal
7919 #. ``ord``: ordered (no nans)
7920 #. ``ueq``: unordered or equal
7921 #. ``ugt``: unordered or greater than
7922 #. ``uge``: unordered or greater than or equal
7923 #. ``ult``: unordered or less than
7924 #. ``ule``: unordered or less than or equal
7925 #. ``une``: unordered or not equal
7926 #. ``uno``: unordered (either nans)
7927 #. ``true``: no comparison, always returns true
7929 *Ordered* means that neither operand is a QNAN while *unordered* means
7930 that either operand may be a QNAN.
7932 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
7933 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
7934 type. They must have identical types.
7939 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
7940 condition code given as ``cond``. If the operands are vectors, then the
7941 vectors are compared element by element. Each comparison performed
7942 always yields an :ref:`i1 <t_integer>` result, as follows:
7944 #. ``false``: always yields ``false``, regardless of operands.
7945 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
7946 is equal to ``op2``.
7947 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
7948 is greater than ``op2``.
7949 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
7950 is greater than or equal to ``op2``.
7951 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
7952 is less than ``op2``.
7953 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
7954 is less than or equal to ``op2``.
7955 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
7956 is not equal to ``op2``.
7957 #. ``ord``: yields ``true`` if both operands are not a QNAN.
7958 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
7960 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
7961 greater than ``op2``.
7962 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
7963 greater than or equal to ``op2``.
7964 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
7966 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
7967 less than or equal to ``op2``.
7968 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
7969 not equal to ``op2``.
7970 #. ``uno``: yields ``true`` if either operand is a QNAN.
7971 #. ``true``: always yields ``true``, regardless of operands.
7973 The ``fcmp`` instruction can also optionally take any number of
7974 :ref:`fast-math flags <fastmath>`, which are optimization hints to enable
7975 otherwise unsafe floating point optimizations.
7977 Any set of fast-math flags are legal on an ``fcmp`` instruction, but the
7978 only flags that have any effect on its semantics are those that allow
7979 assumptions to be made about the values of input arguments; namely
7980 ``nnan``, ``ninf``, and ``nsz``. See :ref:`fastmath` for more information.
7985 .. code-block:: llvm
7987 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
7988 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
7989 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
7990 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
7992 Note that the code generator does not yet support vector types with the
7993 ``fcmp`` instruction.
7997 '``phi``' Instruction
7998 ^^^^^^^^^^^^^^^^^^^^^
8005 <result> = phi <ty> [ <val0>, <label0>], ...
8010 The '``phi``' instruction is used to implement the φ node in the SSA
8011 graph representing the function.
8016 The type of the incoming values is specified with the first type field.
8017 After this, the '``phi``' instruction takes a list of pairs as
8018 arguments, with one pair for each predecessor basic block of the current
8019 block. Only values of :ref:`first class <t_firstclass>` type may be used as
8020 the value arguments to the PHI node. Only labels may be used as the
8023 There must be no non-phi instructions between the start of a basic block
8024 and the PHI instructions: i.e. PHI instructions must be first in a basic
8027 For the purposes of the SSA form, the use of each incoming value is
8028 deemed to occur on the edge from the corresponding predecessor block to
8029 the current block (but after any definition of an '``invoke``'
8030 instruction's return value on the same edge).
8035 At runtime, the '``phi``' instruction logically takes on the value
8036 specified by the pair corresponding to the predecessor basic block that
8037 executed just prior to the current block.
8042 .. code-block:: llvm
8044 Loop: ; Infinite loop that counts from 0 on up...
8045 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
8046 %nextindvar = add i32 %indvar, 1
8051 '``select``' Instruction
8052 ^^^^^^^^^^^^^^^^^^^^^^^^
8059 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
8061 selty is either i1 or {<N x i1>}
8066 The '``select``' instruction is used to choose one value based on a
8067 condition, without IR-level branching.
8072 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
8073 values indicating the condition, and two values of the same :ref:`first
8074 class <t_firstclass>` type.
8079 If the condition is an i1 and it evaluates to 1, the instruction returns
8080 the first value argument; otherwise, it returns the second value
8083 If the condition is a vector of i1, then the value arguments must be
8084 vectors of the same size, and the selection is done element by element.
8086 If the condition is an i1 and the value arguments are vectors of the
8087 same size, then an entire vector is selected.
8092 .. code-block:: llvm
8094 %X = select i1 true, i8 17, i8 42 ; yields i8:17
8098 '``call``' Instruction
8099 ^^^^^^^^^^^^^^^^^^^^^^
8106 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
8111 The '``call``' instruction represents a simple function call.
8116 This instruction requires several arguments:
8118 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
8119 should perform tail call optimization. The ``tail`` marker is a hint that
8120 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
8121 means that the call must be tail call optimized in order for the program to
8122 be correct. The ``musttail`` marker provides these guarantees:
8124 #. The call will not cause unbounded stack growth if it is part of a
8125 recursive cycle in the call graph.
8126 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
8129 Both markers imply that the callee does not access allocas or varargs from
8130 the caller. Calls marked ``musttail`` must obey the following additional
8133 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
8134 or a pointer bitcast followed by a ret instruction.
8135 - The ret instruction must return the (possibly bitcasted) value
8136 produced by the call or void.
8137 - The caller and callee prototypes must match. Pointer types of
8138 parameters or return types may differ in pointee type, but not
8140 - The calling conventions of the caller and callee must match.
8141 - All ABI-impacting function attributes, such as sret, byval, inreg,
8142 returned, and inalloca, must match.
8143 - The callee must be varargs iff the caller is varargs. Bitcasting a
8144 non-varargs function to the appropriate varargs type is legal so
8145 long as the non-varargs prefixes obey the other rules.
8147 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
8148 the following conditions are met:
8150 - Caller and callee both have the calling convention ``fastcc``.
8151 - The call is in tail position (ret immediately follows call and ret
8152 uses value of call or is void).
8153 - Option ``-tailcallopt`` is enabled, or
8154 ``llvm::GuaranteedTailCallOpt`` is ``true``.
8155 - `Platform-specific constraints are
8156 met. <CodeGenerator.html#tailcallopt>`_
8158 #. The optional "cconv" marker indicates which :ref:`calling
8159 convention <callingconv>` the call should use. If none is
8160 specified, the call defaults to using C calling conventions. The
8161 calling convention of the call must match the calling convention of
8162 the target function, or else the behavior is undefined.
8163 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
8164 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
8166 #. '``ty``': the type of the call instruction itself which is also the
8167 type of the return value. Functions that return no value are marked
8169 #. '``fnty``': shall be the signature of the pointer to function value
8170 being invoked. The argument types must match the types implied by
8171 this signature. This type can be omitted if the function is not
8172 varargs and if the function type does not return a pointer to a
8174 #. '``fnptrval``': An LLVM value containing a pointer to a function to
8175 be invoked. In most cases, this is a direct function invocation, but
8176 indirect ``call``'s are just as possible, calling an arbitrary pointer
8178 #. '``function args``': argument list whose types match the function
8179 signature argument types and parameter attributes. All arguments must
8180 be of :ref:`first class <t_firstclass>` type. If the function signature
8181 indicates the function accepts a variable number of arguments, the
8182 extra arguments can be specified.
8183 #. The optional :ref:`function attributes <fnattrs>` list. Only
8184 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
8185 attributes are valid here.
8190 The '``call``' instruction is used to cause control flow to transfer to
8191 a specified function, with its incoming arguments bound to the specified
8192 values. Upon a '``ret``' instruction in the called function, control
8193 flow continues with the instruction after the function call, and the
8194 return value of the function is bound to the result argument.
8199 .. code-block:: llvm
8201 %retval = call i32 @test(i32 %argc)
8202 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
8203 %X = tail call i32 @foo() ; yields i32
8204 %Y = tail call fastcc i32 @foo() ; yields i32
8205 call void %foo(i8 97 signext)
8207 %struct.A = type { i32, i8 }
8208 %r = call %struct.A @foo() ; yields { i32, i8 }
8209 %gr = extractvalue %struct.A %r, 0 ; yields i32
8210 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
8211 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
8212 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
8214 llvm treats calls to some functions with names and arguments that match
8215 the standard C99 library as being the C99 library functions, and may
8216 perform optimizations or generate code for them under that assumption.
8217 This is something we'd like to change in the future to provide better
8218 support for freestanding environments and non-C-based languages.
8222 '``va_arg``' Instruction
8223 ^^^^^^^^^^^^^^^^^^^^^^^^
8230 <resultval> = va_arg <va_list*> <arglist>, <argty>
8235 The '``va_arg``' instruction is used to access arguments passed through
8236 the "variable argument" area of a function call. It is used to implement
8237 the ``va_arg`` macro in C.
8242 This instruction takes a ``va_list*`` value and the type of the
8243 argument. It returns a value of the specified argument type and
8244 increments the ``va_list`` to point to the next argument. The actual
8245 type of ``va_list`` is target specific.
8250 The '``va_arg``' instruction loads an argument of the specified type
8251 from the specified ``va_list`` and causes the ``va_list`` to point to
8252 the next argument. For more information, see the variable argument
8253 handling :ref:`Intrinsic Functions <int_varargs>`.
8255 It is legal for this instruction to be called in a function which does
8256 not take a variable number of arguments, for example, the ``vfprintf``
8259 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
8260 function <intrinsics>` because it takes a type as an argument.
8265 See the :ref:`variable argument processing <int_varargs>` section.
8267 Note that the code generator does not yet fully support va\_arg on many
8268 targets. Also, it does not currently support va\_arg with aggregate
8269 types on any target.
8273 '``landingpad``' Instruction
8274 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8281 <resultval> = landingpad <resultty> <clause>+
8282 <resultval> = landingpad <resultty> cleanup <clause>*
8284 <clause> := catch <type> <value>
8285 <clause> := filter <array constant type> <array constant>
8290 The '``landingpad``' instruction is used by `LLVM's exception handling
8291 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8292 is a landing pad --- one where the exception lands, and corresponds to the
8293 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
8294 defines values supplied by the :ref:`personality function <personalityfn>` upon
8295 re-entry to the function. The ``resultval`` has the type ``resultty``.
8301 ``cleanup`` flag indicates that the landing pad block is a cleanup.
8303 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
8304 contains the global variable representing the "type" that may be caught
8305 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
8306 clause takes an array constant as its argument. Use
8307 "``[0 x i8**] undef``" for a filter which cannot throw. The
8308 '``landingpad``' instruction must contain *at least* one ``clause`` or
8309 the ``cleanup`` flag.
8314 The '``landingpad``' instruction defines the values which are set by the
8315 :ref:`personality function <personalityfn>` upon re-entry to the function, and
8316 therefore the "result type" of the ``landingpad`` instruction. As with
8317 calling conventions, how the personality function results are
8318 represented in LLVM IR is target specific.
8320 The clauses are applied in order from top to bottom. If two
8321 ``landingpad`` instructions are merged together through inlining, the
8322 clauses from the calling function are appended to the list of clauses.
8323 When the call stack is being unwound due to an exception being thrown,
8324 the exception is compared against each ``clause`` in turn. If it doesn't
8325 match any of the clauses, and the ``cleanup`` flag is not set, then
8326 unwinding continues further up the call stack.
8328 The ``landingpad`` instruction has several restrictions:
8330 - A landing pad block is a basic block which is the unwind destination
8331 of an '``invoke``' instruction.
8332 - A landing pad block must have a '``landingpad``' instruction as its
8333 first non-PHI instruction.
8334 - There can be only one '``landingpad``' instruction within the landing
8336 - A basic block that is not a landing pad block may not include a
8337 '``landingpad``' instruction.
8342 .. code-block:: llvm
8344 ;; A landing pad which can catch an integer.
8345 %res = landingpad { i8*, i32 }
8347 ;; A landing pad that is a cleanup.
8348 %res = landingpad { i8*, i32 }
8350 ;; A landing pad which can catch an integer and can only throw a double.
8351 %res = landingpad { i8*, i32 }
8353 filter [1 x i8**] [@_ZTId]
8357 '``cleanuppad``' Instruction
8358 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8365 <resultval> = cleanuppad <resultty> [<args>*]
8370 The '``cleanuppad``' instruction is used by `LLVM's exception handling
8371 system <ExceptionHandling.html#overview>`_ to specify that a basic block
8372 is a cleanup block --- one where a personality routine attempts to
8373 transfer control to run cleanup actions.
8374 The ``args`` correspond to whatever additional
8375 information the :ref:`personality function <personalityfn>` requires to
8376 execute the cleanup.
8377 The ``resultval`` has the type ``resultty``.
8382 The instruction takes a list of arbitrary values which are interpreted
8383 by the :ref:`personality function <personalityfn>`.
8388 The '``cleanuppad``' instruction defines the values which are set by the
8389 :ref:`personality function <personalityfn>` upon re-entry to the function, and
8390 therefore the "result type" of the ``cleanuppad`` instruction. As with
8391 calling conventions, how the personality function results are
8392 represented in LLVM IR is target specific.
8394 When the call stack is being unwound due to an exception being thrown,
8395 the :ref:`personality function <personalityfn>` transfers control to the
8396 ``cleanuppad`` with the aid of the personality-specific arguments.
8398 The ``cleanuppad`` instruction has several restrictions:
8400 - A cleanup block is a basic block which is the unwind destination of
8401 an exceptional instruction.
8402 - A cleanup block must have a '``cleanuppad``' instruction as its
8403 first non-PHI instruction.
8404 - There can be only one '``cleanuppad``' instruction within the
8406 - A basic block that is not a cleanup block may not include a
8407 '``cleanuppad``' instruction.
8408 - It is undefined behavior for control to transfer from a ``cleanuppad`` to a
8409 ``catchret`` without first executing a ``cleanupret`` and a subsequent
8411 - It is undefined behavior for control to transfer from a ``cleanuppad`` to a
8412 ``ret`` without first executing a ``cleanupret``.
8417 .. code-block:: llvm
8419 %res = cleanuppad { i8*, i32 } [label %nextaction]
8426 LLVM supports the notion of an "intrinsic function". These functions
8427 have well known names and semantics and are required to follow certain
8428 restrictions. Overall, these intrinsics represent an extension mechanism
8429 for the LLVM language that does not require changing all of the
8430 transformations in LLVM when adding to the language (or the bitcode
8431 reader/writer, the parser, etc...).
8433 Intrinsic function names must all start with an "``llvm.``" prefix. This
8434 prefix is reserved in LLVM for intrinsic names; thus, function names may
8435 not begin with this prefix. Intrinsic functions must always be external
8436 functions: you cannot define the body of intrinsic functions. Intrinsic
8437 functions may only be used in call or invoke instructions: it is illegal
8438 to take the address of an intrinsic function. Additionally, because
8439 intrinsic functions are part of the LLVM language, it is required if any
8440 are added that they be documented here.
8442 Some intrinsic functions can be overloaded, i.e., the intrinsic
8443 represents a family of functions that perform the same operation but on
8444 different data types. Because LLVM can represent over 8 million
8445 different integer types, overloading is used commonly to allow an
8446 intrinsic function to operate on any integer type. One or more of the
8447 argument types or the result type can be overloaded to accept any
8448 integer type. Argument types may also be defined as exactly matching a
8449 previous argument's type or the result type. This allows an intrinsic
8450 function which accepts multiple arguments, but needs all of them to be
8451 of the same type, to only be overloaded with respect to a single
8452 argument or the result.
8454 Overloaded intrinsics will have the names of its overloaded argument
8455 types encoded into its function name, each preceded by a period. Only
8456 those types which are overloaded result in a name suffix. Arguments
8457 whose type is matched against another type do not. For example, the
8458 ``llvm.ctpop`` function can take an integer of any width and returns an
8459 integer of exactly the same integer width. This leads to a family of
8460 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
8461 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
8462 overloaded, and only one type suffix is required. Because the argument's
8463 type is matched against the return type, it does not require its own
8466 To learn how to add an intrinsic function, please see the `Extending
8467 LLVM Guide <ExtendingLLVM.html>`_.
8471 Variable Argument Handling Intrinsics
8472 -------------------------------------
8474 Variable argument support is defined in LLVM with the
8475 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
8476 functions. These functions are related to the similarly named macros
8477 defined in the ``<stdarg.h>`` header file.
8479 All of these functions operate on arguments that use a target-specific
8480 value type "``va_list``". The LLVM assembly language reference manual
8481 does not define what this type is, so all transformations should be
8482 prepared to handle these functions regardless of the type used.
8484 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
8485 variable argument handling intrinsic functions are used.
8487 .. code-block:: llvm
8489 ; This struct is different for every platform. For most platforms,
8490 ; it is merely an i8*.
8491 %struct.va_list = type { i8* }
8493 ; For Unix x86_64 platforms, va_list is the following struct:
8494 ; %struct.va_list = type { i32, i32, i8*, i8* }
8496 define i32 @test(i32 %X, ...) {
8497 ; Initialize variable argument processing
8498 %ap = alloca %struct.va_list
8499 %ap2 = bitcast %struct.va_list* %ap to i8*
8500 call void @llvm.va_start(i8* %ap2)
8502 ; Read a single integer argument
8503 %tmp = va_arg i8* %ap2, i32
8505 ; Demonstrate usage of llvm.va_copy and llvm.va_end
8507 %aq2 = bitcast i8** %aq to i8*
8508 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
8509 call void @llvm.va_end(i8* %aq2)
8511 ; Stop processing of arguments.
8512 call void @llvm.va_end(i8* %ap2)
8516 declare void @llvm.va_start(i8*)
8517 declare void @llvm.va_copy(i8*, i8*)
8518 declare void @llvm.va_end(i8*)
8522 '``llvm.va_start``' Intrinsic
8523 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8530 declare void @llvm.va_start(i8* <arglist>)
8535 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
8536 subsequent use by ``va_arg``.
8541 The argument is a pointer to a ``va_list`` element to initialize.
8546 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
8547 available in C. In a target-dependent way, it initializes the
8548 ``va_list`` element to which the argument points, so that the next call
8549 to ``va_arg`` will produce the first variable argument passed to the
8550 function. Unlike the C ``va_start`` macro, this intrinsic does not need
8551 to know the last argument of the function as the compiler can figure
8554 '``llvm.va_end``' Intrinsic
8555 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8562 declare void @llvm.va_end(i8* <arglist>)
8567 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
8568 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
8573 The argument is a pointer to a ``va_list`` to destroy.
8578 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
8579 available in C. In a target-dependent way, it destroys the ``va_list``
8580 element to which the argument points. Calls to
8581 :ref:`llvm.va_start <int_va_start>` and
8582 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
8587 '``llvm.va_copy``' Intrinsic
8588 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8595 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
8600 The '``llvm.va_copy``' intrinsic copies the current argument position
8601 from the source argument list to the destination argument list.
8606 The first argument is a pointer to a ``va_list`` element to initialize.
8607 The second argument is a pointer to a ``va_list`` element to copy from.
8612 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
8613 available in C. In a target-dependent way, it copies the source
8614 ``va_list`` element into the destination ``va_list`` element. This
8615 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
8616 arbitrarily complex and require, for example, memory allocation.
8618 Accurate Garbage Collection Intrinsics
8619 --------------------------------------
8621 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
8622 (GC) requires the frontend to generate code containing appropriate intrinsic
8623 calls and select an appropriate GC strategy which knows how to lower these
8624 intrinsics in a manner which is appropriate for the target collector.
8626 These intrinsics allow identification of :ref:`GC roots on the
8627 stack <int_gcroot>`, as well as garbage collector implementations that
8628 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
8629 Frontends for type-safe garbage collected languages should generate
8630 these intrinsics to make use of the LLVM garbage collectors. For more
8631 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
8633 Experimental Statepoint Intrinsics
8634 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8636 LLVM provides an second experimental set of intrinsics for describing garbage
8637 collection safepoints in compiled code. These intrinsics are an alternative
8638 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
8639 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
8640 differences in approach are covered in the `Garbage Collection with LLVM
8641 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
8642 described in :doc:`Statepoints`.
8646 '``llvm.gcroot``' Intrinsic
8647 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8654 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
8659 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
8660 the code generator, and allows some metadata to be associated with it.
8665 The first argument specifies the address of a stack object that contains
8666 the root pointer. The second pointer (which must be either a constant or
8667 a global value address) contains the meta-data to be associated with the
8673 At runtime, a call to this intrinsic stores a null pointer into the
8674 "ptrloc" location. At compile-time, the code generator generates
8675 information to allow the runtime to find the pointer at GC safe points.
8676 The '``llvm.gcroot``' intrinsic may only be used in a function which
8677 :ref:`specifies a GC algorithm <gc>`.
8681 '``llvm.gcread``' Intrinsic
8682 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8689 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
8694 The '``llvm.gcread``' intrinsic identifies reads of references from heap
8695 locations, allowing garbage collector implementations that require read
8701 The second argument is the address to read from, which should be an
8702 address allocated from the garbage collector. The first object is a
8703 pointer to the start of the referenced object, if needed by the language
8704 runtime (otherwise null).
8709 The '``llvm.gcread``' intrinsic has the same semantics as a load
8710 instruction, but may be replaced with substantially more complex code by
8711 the garbage collector runtime, as needed. The '``llvm.gcread``'
8712 intrinsic may only be used in a function which :ref:`specifies a GC
8717 '``llvm.gcwrite``' Intrinsic
8718 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8725 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
8730 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
8731 locations, allowing garbage collector implementations that require write
8732 barriers (such as generational or reference counting collectors).
8737 The first argument is the reference to store, the second is the start of
8738 the object to store it to, and the third is the address of the field of
8739 Obj to store to. If the runtime does not require a pointer to the
8740 object, Obj may be null.
8745 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
8746 instruction, but may be replaced with substantially more complex code by
8747 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
8748 intrinsic may only be used in a function which :ref:`specifies a GC
8751 Code Generator Intrinsics
8752 -------------------------
8754 These intrinsics are provided by LLVM to expose special features that
8755 may only be implemented with code generator support.
8757 '``llvm.returnaddress``' Intrinsic
8758 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8765 declare i8 *@llvm.returnaddress(i32 <level>)
8770 The '``llvm.returnaddress``' intrinsic attempts to compute a
8771 target-specific value indicating the return address of the current
8772 function or one of its callers.
8777 The argument to this intrinsic indicates which function to return the
8778 address for. Zero indicates the calling function, one indicates its
8779 caller, etc. The argument is **required** to be a constant integer
8785 The '``llvm.returnaddress``' intrinsic either returns a pointer
8786 indicating the return address of the specified call frame, or zero if it
8787 cannot be identified. The value returned by this intrinsic is likely to
8788 be incorrect or 0 for arguments other than zero, so it should only be
8789 used for debugging purposes.
8791 Note that calling this intrinsic does not prevent function inlining or
8792 other aggressive transformations, so the value returned may not be that
8793 of the obvious source-language caller.
8795 '``llvm.frameaddress``' Intrinsic
8796 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8803 declare i8* @llvm.frameaddress(i32 <level>)
8808 The '``llvm.frameaddress``' intrinsic attempts to return the
8809 target-specific frame pointer value for the specified stack frame.
8814 The argument to this intrinsic indicates which function to return the
8815 frame pointer for. Zero indicates the calling function, one indicates
8816 its caller, etc. The argument is **required** to be a constant integer
8822 The '``llvm.frameaddress``' intrinsic either returns a pointer
8823 indicating the frame address of the specified call frame, or zero if it
8824 cannot be identified. The value returned by this intrinsic is likely to
8825 be incorrect or 0 for arguments other than zero, so it should only be
8826 used for debugging purposes.
8828 Note that calling this intrinsic does not prevent function inlining or
8829 other aggressive transformations, so the value returned may not be that
8830 of the obvious source-language caller.
8832 '``llvm.localescape``' and '``llvm.localrecover``' Intrinsics
8833 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8840 declare void @llvm.localescape(...)
8841 declare i8* @llvm.localrecover(i8* %func, i8* %fp, i32 %idx)
8846 The '``llvm.localescape``' intrinsic escapes offsets of a collection of static
8847 allocas, and the '``llvm.localrecover``' intrinsic applies those offsets to a
8848 live frame pointer to recover the address of the allocation. The offset is
8849 computed during frame layout of the caller of ``llvm.localescape``.
8854 All arguments to '``llvm.localescape``' must be pointers to static allocas or
8855 casts of static allocas. Each function can only call '``llvm.localescape``'
8856 once, and it can only do so from the entry block.
8858 The ``func`` argument to '``llvm.localrecover``' must be a constant
8859 bitcasted pointer to a function defined in the current module. The code
8860 generator cannot determine the frame allocation offset of functions defined in
8863 The ``fp`` argument to '``llvm.localrecover``' must be a frame pointer of a
8864 call frame that is currently live. The return value of '``llvm.localaddress``'
8865 is one way to produce such a value, but various runtimes also expose a suitable
8866 pointer in platform-specific ways.
8868 The ``idx`` argument to '``llvm.localrecover``' indicates which alloca passed to
8869 '``llvm.localescape``' to recover. It is zero-indexed.
8874 These intrinsics allow a group of functions to share access to a set of local
8875 stack allocations of a one parent function. The parent function may call the
8876 '``llvm.localescape``' intrinsic once from the function entry block, and the
8877 child functions can use '``llvm.localrecover``' to access the escaped allocas.
8878 The '``llvm.localescape``' intrinsic blocks inlining, as inlining changes where
8879 the escaped allocas are allocated, which would break attempts to use
8880 '``llvm.localrecover``'.
8882 .. _int_read_register:
8883 .. _int_write_register:
8885 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
8886 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8893 declare i32 @llvm.read_register.i32(metadata)
8894 declare i64 @llvm.read_register.i64(metadata)
8895 declare void @llvm.write_register.i32(metadata, i32 @value)
8896 declare void @llvm.write_register.i64(metadata, i64 @value)
8902 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
8903 provides access to the named register. The register must be valid on
8904 the architecture being compiled to. The type needs to be compatible
8905 with the register being read.
8910 The '``llvm.read_register``' intrinsic returns the current value of the
8911 register, where possible. The '``llvm.write_register``' intrinsic sets
8912 the current value of the register, where possible.
8914 This is useful to implement named register global variables that need
8915 to always be mapped to a specific register, as is common practice on
8916 bare-metal programs including OS kernels.
8918 The compiler doesn't check for register availability or use of the used
8919 register in surrounding code, including inline assembly. Because of that,
8920 allocatable registers are not supported.
8922 Warning: So far it only works with the stack pointer on selected
8923 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
8924 work is needed to support other registers and even more so, allocatable
8929 '``llvm.stacksave``' Intrinsic
8930 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8937 declare i8* @llvm.stacksave()
8942 The '``llvm.stacksave``' intrinsic is used to remember the current state
8943 of the function stack, for use with
8944 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
8945 implementing language features like scoped automatic variable sized
8951 This intrinsic returns a opaque pointer value that can be passed to
8952 :ref:`llvm.stackrestore <int_stackrestore>`. When an
8953 ``llvm.stackrestore`` intrinsic is executed with a value saved from
8954 ``llvm.stacksave``, it effectively restores the state of the stack to
8955 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
8956 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
8957 were allocated after the ``llvm.stacksave`` was executed.
8959 .. _int_stackrestore:
8961 '``llvm.stackrestore``' Intrinsic
8962 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8969 declare void @llvm.stackrestore(i8* %ptr)
8974 The '``llvm.stackrestore``' intrinsic is used to restore the state of
8975 the function stack to the state it was in when the corresponding
8976 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
8977 useful for implementing language features like scoped automatic variable
8978 sized arrays in C99.
8983 See the description for :ref:`llvm.stacksave <int_stacksave>`.
8985 '``llvm.prefetch``' Intrinsic
8986 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8993 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
8998 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
8999 insert a prefetch instruction if supported; otherwise, it is a noop.
9000 Prefetches have no effect on the behavior of the program but can change
9001 its performance characteristics.
9006 ``address`` is the address to be prefetched, ``rw`` is the specifier
9007 determining if the fetch should be for a read (0) or write (1), and
9008 ``locality`` is a temporal locality specifier ranging from (0) - no
9009 locality, to (3) - extremely local keep in cache. The ``cache type``
9010 specifies whether the prefetch is performed on the data (1) or
9011 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
9012 arguments must be constant integers.
9017 This intrinsic does not modify the behavior of the program. In
9018 particular, prefetches cannot trap and do not produce a value. On
9019 targets that support this intrinsic, the prefetch can provide hints to
9020 the processor cache for better performance.
9022 '``llvm.pcmarker``' Intrinsic
9023 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9030 declare void @llvm.pcmarker(i32 <id>)
9035 The '``llvm.pcmarker``' intrinsic is a method to export a Program
9036 Counter (PC) in a region of code to simulators and other tools. The
9037 method is target specific, but it is expected that the marker will use
9038 exported symbols to transmit the PC of the marker. The marker makes no
9039 guarantees that it will remain with any specific instruction after
9040 optimizations. It is possible that the presence of a marker will inhibit
9041 optimizations. The intended use is to be inserted after optimizations to
9042 allow correlations of simulation runs.
9047 ``id`` is a numerical id identifying the marker.
9052 This intrinsic does not modify the behavior of the program. Backends
9053 that do not support this intrinsic may ignore it.
9055 '``llvm.readcyclecounter``' Intrinsic
9056 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9063 declare i64 @llvm.readcyclecounter()
9068 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
9069 counter register (or similar low latency, high accuracy clocks) on those
9070 targets that support it. On X86, it should map to RDTSC. On Alpha, it
9071 should map to RPCC. As the backing counters overflow quickly (on the
9072 order of 9 seconds on alpha), this should only be used for small
9078 When directly supported, reading the cycle counter should not modify any
9079 memory. Implementations are allowed to either return a application
9080 specific value or a system wide value. On backends without support, this
9081 is lowered to a constant 0.
9083 Note that runtime support may be conditional on the privilege-level code is
9084 running at and the host platform.
9086 '``llvm.clear_cache``' Intrinsic
9087 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9094 declare void @llvm.clear_cache(i8*, i8*)
9099 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
9100 in the specified range to the execution unit of the processor. On
9101 targets with non-unified instruction and data cache, the implementation
9102 flushes the instruction cache.
9107 On platforms with coherent instruction and data caches (e.g. x86), this
9108 intrinsic is a nop. On platforms with non-coherent instruction and data
9109 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
9110 instructions or a system call, if cache flushing requires special
9113 The default behavior is to emit a call to ``__clear_cache`` from the run
9116 This instrinsic does *not* empty the instruction pipeline. Modifications
9117 of the current function are outside the scope of the intrinsic.
9119 '``llvm.instrprof_increment``' Intrinsic
9120 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9127 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
9128 i32 <num-counters>, i32 <index>)
9133 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
9134 frontend for use with instrumentation based profiling. These will be
9135 lowered by the ``-instrprof`` pass to generate execution counts of a
9141 The first argument is a pointer to a global variable containing the
9142 name of the entity being instrumented. This should generally be the
9143 (mangled) function name for a set of counters.
9145 The second argument is a hash value that can be used by the consumer
9146 of the profile data to detect changes to the instrumented source, and
9147 the third is the number of counters associated with ``name``. It is an
9148 error if ``hash`` or ``num-counters`` differ between two instances of
9149 ``instrprof_increment`` that refer to the same name.
9151 The last argument refers to which of the counters for ``name`` should
9152 be incremented. It should be a value between 0 and ``num-counters``.
9157 This intrinsic represents an increment of a profiling counter. It will
9158 cause the ``-instrprof`` pass to generate the appropriate data
9159 structures and the code to increment the appropriate value, in a
9160 format that can be written out by a compiler runtime and consumed via
9161 the ``llvm-profdata`` tool.
9163 Standard C Library Intrinsics
9164 -----------------------------
9166 LLVM provides intrinsics for a few important standard C library
9167 functions. These intrinsics allow source-language front-ends to pass
9168 information about the alignment of the pointer arguments to the code
9169 generator, providing opportunity for more efficient code generation.
9173 '``llvm.memcpy``' Intrinsic
9174 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9179 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
9180 integer bit width and for different address spaces. Not all targets
9181 support all bit widths however.
9185 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9186 i32 <len>, i32 <align>, i1 <isvolatile>)
9187 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9188 i64 <len>, i32 <align>, i1 <isvolatile>)
9193 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9194 source location to the destination location.
9196 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
9197 intrinsics do not return a value, takes extra alignment/isvolatile
9198 arguments and the pointers can be in specified address spaces.
9203 The first argument is a pointer to the destination, the second is a
9204 pointer to the source. The third argument is an integer argument
9205 specifying the number of bytes to copy, the fourth argument is the
9206 alignment of the source and destination locations, and the fifth is a
9207 boolean indicating a volatile access.
9209 If the call to this intrinsic has an alignment value that is not 0 or 1,
9210 then the caller guarantees that both the source and destination pointers
9211 are aligned to that boundary.
9213 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
9214 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9215 very cleanly specified and it is unwise to depend on it.
9220 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9221 source location to the destination location, which are not allowed to
9222 overlap. It copies "len" bytes of memory over. If the argument is known
9223 to be aligned to some boundary, this can be specified as the fourth
9224 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
9226 '``llvm.memmove``' Intrinsic
9227 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9232 This is an overloaded intrinsic. You can use llvm.memmove on any integer
9233 bit width and for different address space. Not all targets support all
9238 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9239 i32 <len>, i32 <align>, i1 <isvolatile>)
9240 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9241 i64 <len>, i32 <align>, i1 <isvolatile>)
9246 The '``llvm.memmove.*``' intrinsics move a block of memory from the
9247 source location to the destination location. It is similar to the
9248 '``llvm.memcpy``' intrinsic but allows the two memory locations to
9251 Note that, unlike the standard libc function, the ``llvm.memmove.*``
9252 intrinsics do not return a value, takes extra alignment/isvolatile
9253 arguments and the pointers can be in specified address spaces.
9258 The first argument is a pointer to the destination, the second is a
9259 pointer to the source. The third argument is an integer argument
9260 specifying the number of bytes to copy, the fourth argument is the
9261 alignment of the source and destination locations, and the fifth is a
9262 boolean indicating a volatile access.
9264 If the call to this intrinsic has an alignment value that is not 0 or 1,
9265 then the caller guarantees that the source and destination pointers are
9266 aligned to that boundary.
9268 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
9269 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
9270 not very cleanly specified and it is unwise to depend on it.
9275 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
9276 source location to the destination location, which may overlap. It
9277 copies "len" bytes of memory over. If the argument is known to be
9278 aligned to some boundary, this can be specified as the fourth argument,
9279 otherwise it should be set to 0 or 1 (both meaning no alignment).
9281 '``llvm.memset.*``' Intrinsics
9282 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9287 This is an overloaded intrinsic. You can use llvm.memset on any integer
9288 bit width and for different address spaces. However, not all targets
9289 support all bit widths.
9293 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
9294 i32 <len>, i32 <align>, i1 <isvolatile>)
9295 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
9296 i64 <len>, i32 <align>, i1 <isvolatile>)
9301 The '``llvm.memset.*``' intrinsics fill a block of memory with a
9302 particular byte value.
9304 Note that, unlike the standard libc function, the ``llvm.memset``
9305 intrinsic does not return a value and takes extra alignment/volatile
9306 arguments. Also, the destination can be in an arbitrary address space.
9311 The first argument is a pointer to the destination to fill, the second
9312 is the byte value with which to fill it, the third argument is an
9313 integer argument specifying the number of bytes to fill, and the fourth
9314 argument is the known alignment of the destination location.
9316 If the call to this intrinsic has an alignment value that is not 0 or 1,
9317 then the caller guarantees that the destination pointer is aligned to
9320 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
9321 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9322 very cleanly specified and it is unwise to depend on it.
9327 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
9328 at the destination location. If the argument is known to be aligned to
9329 some boundary, this can be specified as the fourth argument, otherwise
9330 it should be set to 0 or 1 (both meaning no alignment).
9332 '``llvm.sqrt.*``' Intrinsic
9333 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9338 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
9339 floating point or vector of floating point type. Not all targets support
9344 declare float @llvm.sqrt.f32(float %Val)
9345 declare double @llvm.sqrt.f64(double %Val)
9346 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
9347 declare fp128 @llvm.sqrt.f128(fp128 %Val)
9348 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
9353 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
9354 returning the same value as the libm '``sqrt``' functions would. Unlike
9355 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
9356 negative numbers other than -0.0 (which allows for better optimization,
9357 because there is no need to worry about errno being set).
9358 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
9363 The argument and return value are floating point numbers of the same
9369 This function returns the sqrt of the specified operand if it is a
9370 nonnegative floating point number.
9372 '``llvm.powi.*``' Intrinsic
9373 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9378 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
9379 floating point or vector of floating point type. Not all targets support
9384 declare float @llvm.powi.f32(float %Val, i32 %power)
9385 declare double @llvm.powi.f64(double %Val, i32 %power)
9386 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
9387 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
9388 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
9393 The '``llvm.powi.*``' intrinsics return the first operand raised to the
9394 specified (positive or negative) power. The order of evaluation of
9395 multiplications is not defined. When a vector of floating point type is
9396 used, the second argument remains a scalar integer value.
9401 The second argument is an integer power, and the first is a value to
9402 raise to that power.
9407 This function returns the first value raised to the second power with an
9408 unspecified sequence of rounding operations.
9410 '``llvm.sin.*``' Intrinsic
9411 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9416 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
9417 floating point or vector of floating point type. Not all targets support
9422 declare float @llvm.sin.f32(float %Val)
9423 declare double @llvm.sin.f64(double %Val)
9424 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
9425 declare fp128 @llvm.sin.f128(fp128 %Val)
9426 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
9431 The '``llvm.sin.*``' intrinsics return the sine of the operand.
9436 The argument and return value are floating point numbers of the same
9442 This function returns the sine of the specified operand, returning the
9443 same values as the libm ``sin`` functions would, and handles error
9444 conditions in the same way.
9446 '``llvm.cos.*``' Intrinsic
9447 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9452 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
9453 floating point or vector of floating point type. Not all targets support
9458 declare float @llvm.cos.f32(float %Val)
9459 declare double @llvm.cos.f64(double %Val)
9460 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
9461 declare fp128 @llvm.cos.f128(fp128 %Val)
9462 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
9467 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
9472 The argument and return value are floating point numbers of the same
9478 This function returns the cosine of the specified operand, returning the
9479 same values as the libm ``cos`` functions would, and handles error
9480 conditions in the same way.
9482 '``llvm.pow.*``' Intrinsic
9483 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9488 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
9489 floating point or vector of floating point type. Not all targets support
9494 declare float @llvm.pow.f32(float %Val, float %Power)
9495 declare double @llvm.pow.f64(double %Val, double %Power)
9496 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
9497 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
9498 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
9503 The '``llvm.pow.*``' intrinsics return the first operand raised to the
9504 specified (positive or negative) power.
9509 The second argument is a floating point power, and the first is a value
9510 to raise to that power.
9515 This function returns the first value raised to the second power,
9516 returning the same values as the libm ``pow`` functions would, and
9517 handles error conditions in the same way.
9519 '``llvm.exp.*``' Intrinsic
9520 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9525 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
9526 floating point or vector of floating point type. Not all targets support
9531 declare float @llvm.exp.f32(float %Val)
9532 declare double @llvm.exp.f64(double %Val)
9533 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
9534 declare fp128 @llvm.exp.f128(fp128 %Val)
9535 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
9540 The '``llvm.exp.*``' intrinsics perform the exp function.
9545 The argument and return value are floating point numbers of the same
9551 This function returns the same values as the libm ``exp`` functions
9552 would, and handles error conditions in the same way.
9554 '``llvm.exp2.*``' Intrinsic
9555 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9560 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
9561 floating point or vector of floating point type. Not all targets support
9566 declare float @llvm.exp2.f32(float %Val)
9567 declare double @llvm.exp2.f64(double %Val)
9568 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
9569 declare fp128 @llvm.exp2.f128(fp128 %Val)
9570 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
9575 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
9580 The argument and return value are floating point numbers of the same
9586 This function returns the same values as the libm ``exp2`` functions
9587 would, and handles error conditions in the same way.
9589 '``llvm.log.*``' Intrinsic
9590 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9595 This is an overloaded intrinsic. You can use ``llvm.log`` on any
9596 floating point or vector of floating point type. Not all targets support
9601 declare float @llvm.log.f32(float %Val)
9602 declare double @llvm.log.f64(double %Val)
9603 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
9604 declare fp128 @llvm.log.f128(fp128 %Val)
9605 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
9610 The '``llvm.log.*``' intrinsics perform the log function.
9615 The argument and return value are floating point numbers of the same
9621 This function returns the same values as the libm ``log`` functions
9622 would, and handles error conditions in the same way.
9624 '``llvm.log10.*``' Intrinsic
9625 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9630 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
9631 floating point or vector of floating point type. Not all targets support
9636 declare float @llvm.log10.f32(float %Val)
9637 declare double @llvm.log10.f64(double %Val)
9638 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
9639 declare fp128 @llvm.log10.f128(fp128 %Val)
9640 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
9645 The '``llvm.log10.*``' intrinsics perform the log10 function.
9650 The argument and return value are floating point numbers of the same
9656 This function returns the same values as the libm ``log10`` functions
9657 would, and handles error conditions in the same way.
9659 '``llvm.log2.*``' Intrinsic
9660 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9665 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
9666 floating point or vector of floating point type. Not all targets support
9671 declare float @llvm.log2.f32(float %Val)
9672 declare double @llvm.log2.f64(double %Val)
9673 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
9674 declare fp128 @llvm.log2.f128(fp128 %Val)
9675 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
9680 The '``llvm.log2.*``' intrinsics perform the log2 function.
9685 The argument and return value are floating point numbers of the same
9691 This function returns the same values as the libm ``log2`` functions
9692 would, and handles error conditions in the same way.
9694 '``llvm.fma.*``' Intrinsic
9695 ^^^^^^^^^^^^^^^^^^^^^^^^^^
9700 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
9701 floating point or vector of floating point type. Not all targets support
9706 declare float @llvm.fma.f32(float %a, float %b, float %c)
9707 declare double @llvm.fma.f64(double %a, double %b, double %c)
9708 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
9709 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
9710 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
9715 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
9721 The argument and return value are floating point numbers of the same
9727 This function returns the same values as the libm ``fma`` functions
9728 would, and does not set errno.
9730 '``llvm.fabs.*``' Intrinsic
9731 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9736 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
9737 floating point or vector of floating point type. Not all targets support
9742 declare float @llvm.fabs.f32(float %Val)
9743 declare double @llvm.fabs.f64(double %Val)
9744 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
9745 declare fp128 @llvm.fabs.f128(fp128 %Val)
9746 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
9751 The '``llvm.fabs.*``' intrinsics return the absolute value of the
9757 The argument and return value are floating point numbers of the same
9763 This function returns the same values as the libm ``fabs`` functions
9764 would, and handles error conditions in the same way.
9766 '``llvm.minnum.*``' Intrinsic
9767 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9772 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
9773 floating point or vector of floating point type. Not all targets support
9778 declare float @llvm.minnum.f32(float %Val0, float %Val1)
9779 declare double @llvm.minnum.f64(double %Val0, double %Val1)
9780 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
9781 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
9782 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
9787 The '``llvm.minnum.*``' intrinsics return the minimum of the two
9794 The arguments and return value are floating point numbers of the same
9800 Follows the IEEE-754 semantics for minNum, which also match for libm's
9803 If either operand is a NaN, returns the other non-NaN operand. Returns
9804 NaN only if both operands are NaN. If the operands compare equal,
9805 returns a value that compares equal to both operands. This means that
9806 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
9808 '``llvm.maxnum.*``' Intrinsic
9809 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9814 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
9815 floating point or vector of floating point type. Not all targets support
9820 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
9821 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
9822 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
9823 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
9824 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
9829 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
9836 The arguments and return value are floating point numbers of the same
9841 Follows the IEEE-754 semantics for maxNum, which also match for libm's
9844 If either operand is a NaN, returns the other non-NaN operand. Returns
9845 NaN only if both operands are NaN. If the operands compare equal,
9846 returns a value that compares equal to both operands. This means that
9847 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
9849 '``llvm.copysign.*``' Intrinsic
9850 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9855 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
9856 floating point or vector of floating point type. Not all targets support
9861 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
9862 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
9863 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
9864 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
9865 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
9870 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
9871 first operand and the sign of the second operand.
9876 The arguments and return value are floating point numbers of the same
9882 This function returns the same values as the libm ``copysign``
9883 functions would, and handles error conditions in the same way.
9885 '``llvm.floor.*``' Intrinsic
9886 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9891 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
9892 floating point or vector of floating point type. Not all targets support
9897 declare float @llvm.floor.f32(float %Val)
9898 declare double @llvm.floor.f64(double %Val)
9899 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
9900 declare fp128 @llvm.floor.f128(fp128 %Val)
9901 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
9906 The '``llvm.floor.*``' intrinsics return the floor of the operand.
9911 The argument and return value are floating point numbers of the same
9917 This function returns the same values as the libm ``floor`` functions
9918 would, and handles error conditions in the same way.
9920 '``llvm.ceil.*``' Intrinsic
9921 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9926 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
9927 floating point or vector of floating point type. Not all targets support
9932 declare float @llvm.ceil.f32(float %Val)
9933 declare double @llvm.ceil.f64(double %Val)
9934 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
9935 declare fp128 @llvm.ceil.f128(fp128 %Val)
9936 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
9941 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
9946 The argument and return value are floating point numbers of the same
9952 This function returns the same values as the libm ``ceil`` functions
9953 would, and handles error conditions in the same way.
9955 '``llvm.trunc.*``' Intrinsic
9956 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9961 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
9962 floating point or vector of floating point type. Not all targets support
9967 declare float @llvm.trunc.f32(float %Val)
9968 declare double @llvm.trunc.f64(double %Val)
9969 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
9970 declare fp128 @llvm.trunc.f128(fp128 %Val)
9971 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
9976 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
9977 nearest integer not larger in magnitude than the operand.
9982 The argument and return value are floating point numbers of the same
9988 This function returns the same values as the libm ``trunc`` functions
9989 would, and handles error conditions in the same way.
9991 '``llvm.rint.*``' Intrinsic
9992 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9997 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
9998 floating point or vector of floating point type. Not all targets support
10003 declare float @llvm.rint.f32(float %Val)
10004 declare double @llvm.rint.f64(double %Val)
10005 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
10006 declare fp128 @llvm.rint.f128(fp128 %Val)
10007 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
10012 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
10013 nearest integer. It may raise an inexact floating-point exception if the
10014 operand isn't an integer.
10019 The argument and return value are floating point numbers of the same
10025 This function returns the same values as the libm ``rint`` functions
10026 would, and handles error conditions in the same way.
10028 '``llvm.nearbyint.*``' Intrinsic
10029 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10034 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
10035 floating point or vector of floating point type. Not all targets support
10040 declare float @llvm.nearbyint.f32(float %Val)
10041 declare double @llvm.nearbyint.f64(double %Val)
10042 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
10043 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
10044 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
10049 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
10055 The argument and return value are floating point numbers of the same
10061 This function returns the same values as the libm ``nearbyint``
10062 functions would, and handles error conditions in the same way.
10064 '``llvm.round.*``' Intrinsic
10065 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10070 This is an overloaded intrinsic. You can use ``llvm.round`` on any
10071 floating point or vector of floating point type. Not all targets support
10076 declare float @llvm.round.f32(float %Val)
10077 declare double @llvm.round.f64(double %Val)
10078 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
10079 declare fp128 @llvm.round.f128(fp128 %Val)
10080 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
10085 The '``llvm.round.*``' intrinsics returns the operand rounded to the
10091 The argument and return value are floating point numbers of the same
10097 This function returns the same values as the libm ``round``
10098 functions would, and handles error conditions in the same way.
10100 Bit Manipulation Intrinsics
10101 ---------------------------
10103 LLVM provides intrinsics for a few important bit manipulation
10104 operations. These allow efficient code generation for some algorithms.
10106 '``llvm.bswap.*``' Intrinsics
10107 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10112 This is an overloaded intrinsic function. You can use bswap on any
10113 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
10117 declare i16 @llvm.bswap.i16(i16 <id>)
10118 declare i32 @llvm.bswap.i32(i32 <id>)
10119 declare i64 @llvm.bswap.i64(i64 <id>)
10124 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
10125 values with an even number of bytes (positive multiple of 16 bits).
10126 These are useful for performing operations on data that is not in the
10127 target's native byte order.
10132 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
10133 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
10134 intrinsic returns an i32 value that has the four bytes of the input i32
10135 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
10136 returned i32 will have its bytes in 3, 2, 1, 0 order. The
10137 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
10138 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
10141 '``llvm.ctpop.*``' Intrinsic
10142 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10147 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
10148 bit width, or on any vector with integer elements. Not all targets
10149 support all bit widths or vector types, however.
10153 declare i8 @llvm.ctpop.i8(i8 <src>)
10154 declare i16 @llvm.ctpop.i16(i16 <src>)
10155 declare i32 @llvm.ctpop.i32(i32 <src>)
10156 declare i64 @llvm.ctpop.i64(i64 <src>)
10157 declare i256 @llvm.ctpop.i256(i256 <src>)
10158 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
10163 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
10169 The only argument is the value to be counted. The argument may be of any
10170 integer type, or a vector with integer elements. The return type must
10171 match the argument type.
10176 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
10177 each element of a vector.
10179 '``llvm.ctlz.*``' Intrinsic
10180 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10185 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
10186 integer bit width, or any vector whose elements are integers. Not all
10187 targets support all bit widths or vector types, however.
10191 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
10192 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
10193 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
10194 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
10195 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
10196 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10201 The '``llvm.ctlz``' family of intrinsic functions counts the number of
10202 leading zeros in a variable.
10207 The first argument is the value to be counted. This argument may be of
10208 any integer type, or a vector with integer element type. The return
10209 type must match the first argument type.
10211 The second argument must be a constant and is a flag to indicate whether
10212 the intrinsic should ensure that a zero as the first argument produces a
10213 defined result. Historically some architectures did not provide a
10214 defined result for zero values as efficiently, and many algorithms are
10215 now predicated on avoiding zero-value inputs.
10220 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
10221 zeros in a variable, or within each element of the vector. If
10222 ``src == 0`` then the result is the size in bits of the type of ``src``
10223 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10224 ``llvm.ctlz(i32 2) = 30``.
10226 '``llvm.cttz.*``' Intrinsic
10227 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10232 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
10233 integer bit width, or any vector of integer elements. Not all targets
10234 support all bit widths or vector types, however.
10238 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
10239 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
10240 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
10241 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
10242 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
10243 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10248 The '``llvm.cttz``' family of intrinsic functions counts the number of
10254 The first argument is the value to be counted. This argument may be of
10255 any integer type, or a vector with integer element type. The return
10256 type must match the first argument type.
10258 The second argument must be a constant and is a flag to indicate whether
10259 the intrinsic should ensure that a zero as the first argument produces a
10260 defined result. Historically some architectures did not provide a
10261 defined result for zero values as efficiently, and many algorithms are
10262 now predicated on avoiding zero-value inputs.
10267 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
10268 zeros in a variable, or within each element of a vector. If ``src == 0``
10269 then the result is the size in bits of the type of ``src`` if
10270 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10271 ``llvm.cttz(2) = 1``.
10275 Arithmetic with Overflow Intrinsics
10276 -----------------------------------
10278 LLVM provides intrinsics for some arithmetic with overflow operations.
10280 '``llvm.sadd.with.overflow.*``' Intrinsics
10281 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10286 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
10287 on any integer bit width.
10291 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
10292 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10293 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
10298 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10299 a signed addition of the two arguments, and indicate whether an overflow
10300 occurred during the signed summation.
10305 The arguments (%a and %b) and the first element of the result structure
10306 may be of integer types of any bit width, but they must have the same
10307 bit width. The second element of the result structure must be of type
10308 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10314 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10315 a signed addition of the two variables. They return a structure --- the
10316 first element of which is the signed summation, and the second element
10317 of which is a bit specifying if the signed summation resulted in an
10323 .. code-block:: llvm
10325 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10326 %sum = extractvalue {i32, i1} %res, 0
10327 %obit = extractvalue {i32, i1} %res, 1
10328 br i1 %obit, label %overflow, label %normal
10330 '``llvm.uadd.with.overflow.*``' Intrinsics
10331 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10336 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
10337 on any integer bit width.
10341 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
10342 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10343 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
10348 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10349 an unsigned addition of the two arguments, and indicate whether a carry
10350 occurred during the unsigned summation.
10355 The arguments (%a and %b) and the first element of the result structure
10356 may be of integer types of any bit width, but they must have the same
10357 bit width. The second element of the result structure must be of type
10358 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10364 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10365 an unsigned addition of the two arguments. They return a structure --- the
10366 first element of which is the sum, and the second element of which is a
10367 bit specifying if the unsigned summation resulted in a carry.
10372 .. code-block:: llvm
10374 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10375 %sum = extractvalue {i32, i1} %res, 0
10376 %obit = extractvalue {i32, i1} %res, 1
10377 br i1 %obit, label %carry, label %normal
10379 '``llvm.ssub.with.overflow.*``' Intrinsics
10380 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10385 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
10386 on any integer bit width.
10390 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
10391 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10392 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
10397 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10398 a signed subtraction of the two arguments, and indicate whether an
10399 overflow occurred during the signed subtraction.
10404 The arguments (%a and %b) and the first element of the result structure
10405 may be of integer types of any bit width, but they must have the same
10406 bit width. The second element of the result structure must be of type
10407 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10413 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10414 a signed subtraction of the two arguments. They return a structure --- the
10415 first element of which is the subtraction, and the second element of
10416 which is a bit specifying if the signed subtraction resulted in an
10422 .. code-block:: llvm
10424 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10425 %sum = extractvalue {i32, i1} %res, 0
10426 %obit = extractvalue {i32, i1} %res, 1
10427 br i1 %obit, label %overflow, label %normal
10429 '``llvm.usub.with.overflow.*``' Intrinsics
10430 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10435 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
10436 on any integer bit width.
10440 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
10441 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10442 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
10447 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10448 an unsigned subtraction of the two arguments, and indicate whether an
10449 overflow occurred during the unsigned subtraction.
10454 The arguments (%a and %b) and the first element of the result structure
10455 may be of integer types of any bit width, but they must have the same
10456 bit width. The second element of the result structure must be of type
10457 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10463 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10464 an unsigned subtraction of the two arguments. They return a structure ---
10465 the first element of which is the subtraction, and the second element of
10466 which is a bit specifying if the unsigned subtraction resulted in an
10472 .. code-block:: llvm
10474 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10475 %sum = extractvalue {i32, i1} %res, 0
10476 %obit = extractvalue {i32, i1} %res, 1
10477 br i1 %obit, label %overflow, label %normal
10479 '``llvm.smul.with.overflow.*``' Intrinsics
10480 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10485 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
10486 on any integer bit width.
10490 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
10491 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10492 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
10497 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10498 a signed multiplication of the two arguments, and indicate whether an
10499 overflow occurred during the signed multiplication.
10504 The arguments (%a and %b) and the first element of the result structure
10505 may be of integer types of any bit width, but they must have the same
10506 bit width. The second element of the result structure must be of type
10507 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10513 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10514 a signed multiplication of the two arguments. They return a structure ---
10515 the first element of which is the multiplication, and the second element
10516 of which is a bit specifying if the signed multiplication resulted in an
10522 .. code-block:: llvm
10524 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10525 %sum = extractvalue {i32, i1} %res, 0
10526 %obit = extractvalue {i32, i1} %res, 1
10527 br i1 %obit, label %overflow, label %normal
10529 '``llvm.umul.with.overflow.*``' Intrinsics
10530 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10535 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
10536 on any integer bit width.
10540 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
10541 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10542 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
10547 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10548 a unsigned multiplication of the two arguments, and indicate whether an
10549 overflow occurred during the unsigned multiplication.
10554 The arguments (%a and %b) and the first element of the result structure
10555 may be of integer types of any bit width, but they must have the same
10556 bit width. The second element of the result structure must be of type
10557 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10563 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10564 an unsigned multiplication of the two arguments. They return a structure ---
10565 the first element of which is the multiplication, and the second
10566 element of which is a bit specifying if the unsigned multiplication
10567 resulted in an overflow.
10572 .. code-block:: llvm
10574 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10575 %sum = extractvalue {i32, i1} %res, 0
10576 %obit = extractvalue {i32, i1} %res, 1
10577 br i1 %obit, label %overflow, label %normal
10579 Specialised Arithmetic Intrinsics
10580 ---------------------------------
10582 '``llvm.canonicalize.*``' Intrinsic
10583 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10590 declare float @llvm.canonicalize.f32(float %a)
10591 declare double @llvm.canonicalize.f64(double %b)
10596 The '``llvm.canonicalize.*``' intrinsic returns the platform specific canonical
10597 encoding of a floating point number. This canonicalization is useful for
10598 implementing certain numeric primitives such as frexp. The canonical encoding is
10599 defined by IEEE-754-2008 to be:
10603 2.1.8 canonical encoding: The preferred encoding of a floating-point
10604 representation in a format. Applied to declets, significands of finite
10605 numbers, infinities, and NaNs, especially in decimal formats.
10607 This operation can also be considered equivalent to the IEEE-754-2008
10608 conversion of a floating-point value to the same format. NaNs are handled
10609 according to section 6.2.
10611 Examples of non-canonical encodings:
10613 - x87 pseudo denormals, pseudo NaNs, pseudo Infinity, Unnormals. These are
10614 converted to a canonical representation per hardware-specific protocol.
10615 - Many normal decimal floating point numbers have non-canonical alternative
10617 - Some machines, like GPUs or ARMv7 NEON, do not support subnormal values.
10618 These are treated as non-canonical encodings of zero and with be flushed to
10619 a zero of the same sign by this operation.
10621 Note that per IEEE-754-2008 6.2, systems that support signaling NaNs with
10622 default exception handling must signal an invalid exception, and produce a
10625 This function should always be implementable as multiplication by 1.0, provided
10626 that the compiler does not constant fold the operation. Likewise, division by
10627 1.0 and ``llvm.minnum(x, x)`` are possible implementations. Addition with
10628 -0.0 is also sufficient provided that the rounding mode is not -Infinity.
10630 ``@llvm.canonicalize`` must preserve the equality relation. That is:
10632 - ``(@llvm.canonicalize(x) == x)`` is equivalent to ``(x == x)``
10633 - ``(@llvm.canonicalize(x) == @llvm.canonicalize(y))`` is equivalent to
10636 Additionally, the sign of zero must be conserved:
10637 ``@llvm.canonicalize(-0.0) = -0.0`` and ``@llvm.canonicalize(+0.0) = +0.0``
10639 The payload bits of a NaN must be conserved, with two exceptions.
10640 First, environments which use only a single canonical representation of NaN
10641 must perform said canonicalization. Second, SNaNs must be quieted per the
10644 The canonicalization operation may be optimized away if:
10646 - The input is known to be canonical. For example, it was produced by a
10647 floating-point operation that is required by the standard to be canonical.
10648 - The result is consumed only by (or fused with) other floating-point
10649 operations. That is, the bits of the floating point value are not examined.
10651 '``llvm.fmuladd.*``' Intrinsic
10652 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10659 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
10660 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
10665 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
10666 expressions that can be fused if the code generator determines that (a) the
10667 target instruction set has support for a fused operation, and (b) that the
10668 fused operation is more efficient than the equivalent, separate pair of mul
10669 and add instructions.
10674 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
10675 multiplicands, a and b, and an addend c.
10684 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
10686 is equivalent to the expression a \* b + c, except that rounding will
10687 not be performed between the multiplication and addition steps if the
10688 code generator fuses the operations. Fusion is not guaranteed, even if
10689 the target platform supports it. If a fused multiply-add is required the
10690 corresponding llvm.fma.\* intrinsic function should be used
10691 instead. This never sets errno, just as '``llvm.fma.*``'.
10696 .. code-block:: llvm
10698 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
10701 '``llvm.uabsdiff.*``' and '``llvm.sabsdiff.*``' Intrinsics
10702 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10706 This is an overloaded intrinsic. The loaded data is a vector of any integer bit width.
10708 .. code-block:: llvm
10710 declare <4 x integer> @llvm.uabsdiff.v4i32(<4 x integer> %a, <4 x integer> %b)
10716 The ``llvm.uabsdiff`` intrinsic returns a vector result of the absolute difference of the two operands,
10717 treating them both as unsigned integers.
10719 The ``llvm.sabsdiff`` intrinsic returns a vector result of the absolute difference of the two operands,
10720 treating them both as signed integers.
10724 These intrinsics are primarily used during the code generation stage of compilation.
10725 They are generated by compiler passes such as the Loop and SLP vectorizers.it is not
10726 recommended for users to create them manually.
10731 Both intrinsics take two integer of the same bitwidth.
10738 call <4 x i32> @llvm.uabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
10742 %sub = sub <4 x i32> %a, %b
10743 %ispos = icmp ugt <4 x i32> %sub, <i32 -1, i32 -1, i32 -1, i32 -1>
10744 %neg = sub <4 x i32> zeroinitializer, %sub
10745 %1 = select <4 x i1> %ispos, <4 x i32> %sub, <4 x i32> %neg
10747 Similarly the expression::
10749 call <4 x i32> @llvm.sabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
10753 %sub = sub nsw <4 x i32> %a, %b
10754 %ispos = icmp sgt <4 x i32> %sub, <i32 -1, i32 -1, i32 -1, i32 -1>
10755 %neg = sub nsw <4 x i32> zeroinitializer, %sub
10756 %1 = select <4 x i1> %ispos, <4 x i32> %sub, <4 x i32> %neg
10759 Half Precision Floating Point Intrinsics
10760 ----------------------------------------
10762 For most target platforms, half precision floating point is a
10763 storage-only format. This means that it is a dense encoding (in memory)
10764 but does not support computation in the format.
10766 This means that code must first load the half-precision floating point
10767 value as an i16, then convert it to float with
10768 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
10769 then be performed on the float value (including extending to double
10770 etc). To store the value back to memory, it is first converted to float
10771 if needed, then converted to i16 with
10772 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
10775 .. _int_convert_to_fp16:
10777 '``llvm.convert.to.fp16``' Intrinsic
10778 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10785 declare i16 @llvm.convert.to.fp16.f32(float %a)
10786 declare i16 @llvm.convert.to.fp16.f64(double %a)
10791 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
10792 conventional floating point type to half precision floating point format.
10797 The intrinsic function contains single argument - the value to be
10803 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
10804 conventional floating point format to half precision floating point format. The
10805 return value is an ``i16`` which contains the converted number.
10810 .. code-block:: llvm
10812 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
10813 store i16 %res, i16* @x, align 2
10815 .. _int_convert_from_fp16:
10817 '``llvm.convert.from.fp16``' Intrinsic
10818 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10825 declare float @llvm.convert.from.fp16.f32(i16 %a)
10826 declare double @llvm.convert.from.fp16.f64(i16 %a)
10831 The '``llvm.convert.from.fp16``' intrinsic function performs a
10832 conversion from half precision floating point format to single precision
10833 floating point format.
10838 The intrinsic function contains single argument - the value to be
10844 The '``llvm.convert.from.fp16``' intrinsic function performs a
10845 conversion from half single precision floating point format to single
10846 precision floating point format. The input half-float value is
10847 represented by an ``i16`` value.
10852 .. code-block:: llvm
10854 %a = load i16, i16* @x, align 2
10855 %res = call float @llvm.convert.from.fp16(i16 %a)
10857 .. _dbg_intrinsics:
10859 Debugger Intrinsics
10860 -------------------
10862 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
10863 prefix), are described in the `LLVM Source Level
10864 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
10867 Exception Handling Intrinsics
10868 -----------------------------
10870 The LLVM exception handling intrinsics (which all start with
10871 ``llvm.eh.`` prefix), are described in the `LLVM Exception
10872 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
10874 .. _int_trampoline:
10876 Trampoline Intrinsics
10877 ---------------------
10879 These intrinsics make it possible to excise one parameter, marked with
10880 the :ref:`nest <nest>` attribute, from a function. The result is a
10881 callable function pointer lacking the nest parameter - the caller does
10882 not need to provide a value for it. Instead, the value to use is stored
10883 in advance in a "trampoline", a block of memory usually allocated on the
10884 stack, which also contains code to splice the nest value into the
10885 argument list. This is used to implement the GCC nested function address
10888 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
10889 then the resulting function pointer has signature ``i32 (i32, i32)*``.
10890 It can be created as follows:
10892 .. code-block:: llvm
10894 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
10895 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
10896 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
10897 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
10898 %fp = bitcast i8* %p to i32 (i32, i32)*
10900 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
10901 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
10905 '``llvm.init.trampoline``' Intrinsic
10906 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10913 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
10918 This fills the memory pointed to by ``tramp`` with executable code,
10919 turning it into a trampoline.
10924 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
10925 pointers. The ``tramp`` argument must point to a sufficiently large and
10926 sufficiently aligned block of memory; this memory is written to by the
10927 intrinsic. Note that the size and the alignment are target-specific -
10928 LLVM currently provides no portable way of determining them, so a
10929 front-end that generates this intrinsic needs to have some
10930 target-specific knowledge. The ``func`` argument must hold a function
10931 bitcast to an ``i8*``.
10936 The block of memory pointed to by ``tramp`` is filled with target
10937 dependent code, turning it into a function. Then ``tramp`` needs to be
10938 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
10939 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
10940 function's signature is the same as that of ``func`` with any arguments
10941 marked with the ``nest`` attribute removed. At most one such ``nest``
10942 argument is allowed, and it must be of pointer type. Calling the new
10943 function is equivalent to calling ``func`` with the same argument list,
10944 but with ``nval`` used for the missing ``nest`` argument. If, after
10945 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
10946 modified, then the effect of any later call to the returned function
10947 pointer is undefined.
10951 '``llvm.adjust.trampoline``' Intrinsic
10952 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10959 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
10964 This performs any required machine-specific adjustment to the address of
10965 a trampoline (passed as ``tramp``).
10970 ``tramp`` must point to a block of memory which already has trampoline
10971 code filled in by a previous call to
10972 :ref:`llvm.init.trampoline <int_it>`.
10977 On some architectures the address of the code to be executed needs to be
10978 different than the address where the trampoline is actually stored. This
10979 intrinsic returns the executable address corresponding to ``tramp``
10980 after performing the required machine specific adjustments. The pointer
10981 returned can then be :ref:`bitcast and executed <int_trampoline>`.
10983 .. _int_mload_mstore:
10985 Masked Vector Load and Store Intrinsics
10986 ---------------------------------------
10988 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.
10992 '``llvm.masked.load.*``' Intrinsics
10993 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10997 This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
11001 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11002 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11007 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.
11013 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.
11019 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.
11020 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.
11025 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
11027 ;; The result of the two following instructions is identical aside from potential memory access exception
11028 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
11029 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
11033 '``llvm.masked.store.*``' Intrinsics
11034 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11038 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
11042 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
11043 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
11048 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.
11053 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.
11059 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.
11060 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.
11064 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
11066 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
11067 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
11068 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
11069 store <16 x float> %res, <16 x float>* %ptr, align 4
11072 Masked Vector Gather and Scatter Intrinsics
11073 -------------------------------------------
11075 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.
11079 '``llvm.masked.gather.*``' Intrinsics
11080 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11084 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.
11088 declare <16 x float> @llvm.masked.gather.v16f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11089 declare <2 x double> @llvm.masked.gather.v2f64 (<2 x double*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11094 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.
11100 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.
11106 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.
11107 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.
11112 %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>)
11114 ;; The gather with all-true mask is equivalent to the following instruction sequence
11115 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
11116 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
11117 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
11118 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
11120 %val0 = load double, double* %ptr0, align 8
11121 %val1 = load double, double* %ptr1, align 8
11122 %val2 = load double, double* %ptr2, align 8
11123 %val3 = load double, double* %ptr3, align 8
11125 %vec0 = insertelement <4 x double>undef, %val0, 0
11126 %vec01 = insertelement <4 x double>%vec0, %val1, 1
11127 %vec012 = insertelement <4 x double>%vec01, %val2, 2
11128 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
11132 '``llvm.masked.scatter.*``' Intrinsics
11133 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11137 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.
11141 declare void @llvm.masked.scatter.v8i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
11142 declare void @llvm.masked.scatter.v16f32(<16 x i32> <value>, <16 x i32*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
11147 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.
11152 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.
11158 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.
11162 ;; This instruction unconditionaly stores data vector in multiple addresses
11163 call @llvm.masked.scatter.v8i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
11165 ;; It is equivalent to a list of scalar stores
11166 %val0 = extractelement <8 x i32> %value, i32 0
11167 %val1 = extractelement <8 x i32> %value, i32 1
11169 %val7 = extractelement <8 x i32> %value, i32 7
11170 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
11171 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
11173 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
11174 ;; Note: the order of the following stores is important when they overlap:
11175 store i32 %val0, i32* %ptr0, align 4
11176 store i32 %val1, i32* %ptr1, align 4
11178 store i32 %val7, i32* %ptr7, align 4
11184 This class of intrinsics provides information about the lifetime of
11185 memory objects and ranges where variables are immutable.
11189 '``llvm.lifetime.start``' Intrinsic
11190 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11197 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
11202 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
11208 The first argument is a constant integer representing the size of the
11209 object, or -1 if it is variable sized. The second argument is a pointer
11215 This intrinsic indicates that before this point in the code, the value
11216 of the memory pointed to by ``ptr`` is dead. This means that it is known
11217 to never be used and has an undefined value. A load from the pointer
11218 that precedes this intrinsic can be replaced with ``'undef'``.
11222 '``llvm.lifetime.end``' Intrinsic
11223 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11230 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
11235 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
11241 The first argument is a constant integer representing the size of the
11242 object, or -1 if it is variable sized. The second argument is a pointer
11248 This intrinsic indicates that after this point in the code, the value of
11249 the memory pointed to by ``ptr`` is dead. This means that it is known to
11250 never be used and has an undefined value. Any stores into the memory
11251 object following this intrinsic may be removed as dead.
11253 '``llvm.invariant.start``' Intrinsic
11254 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11261 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
11266 The '``llvm.invariant.start``' intrinsic specifies that the contents of
11267 a memory object will not change.
11272 The first argument is a constant integer representing the size of the
11273 object, or -1 if it is variable sized. The second argument is a pointer
11279 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
11280 the return value, the referenced memory location is constant and
11283 '``llvm.invariant.end``' Intrinsic
11284 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11291 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
11296 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
11297 memory object are mutable.
11302 The first argument is the matching ``llvm.invariant.start`` intrinsic.
11303 The second argument is a constant integer representing the size of the
11304 object, or -1 if it is variable sized and the third argument is a
11305 pointer to the object.
11310 This intrinsic indicates that the memory is mutable again.
11315 This class of intrinsics is designed to be generic and has no specific
11318 '``llvm.var.annotation``' Intrinsic
11319 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11326 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11331 The '``llvm.var.annotation``' intrinsic.
11336 The first argument is a pointer to a value, the second is a pointer to a
11337 global string, the third is a pointer to a global string which is the
11338 source file name, and the last argument is the line number.
11343 This intrinsic allows annotation of local variables with arbitrary
11344 strings. This can be useful for special purpose optimizations that want
11345 to look for these annotations. These have no other defined use; they are
11346 ignored by code generation and optimization.
11348 '``llvm.ptr.annotation.*``' Intrinsic
11349 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11354 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
11355 pointer to an integer of any width. *NOTE* you must specify an address space for
11356 the pointer. The identifier for the default address space is the integer
11361 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11362 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
11363 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
11364 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
11365 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
11370 The '``llvm.ptr.annotation``' intrinsic.
11375 The first argument is a pointer to an integer value of arbitrary bitwidth
11376 (result of some expression), the second is a pointer to a global string, the
11377 third is a pointer to a global string which is the source file name, and the
11378 last argument is the line number. It returns the value of the first argument.
11383 This intrinsic allows annotation of a pointer to an integer with arbitrary
11384 strings. This can be useful for special purpose optimizations that want to look
11385 for these annotations. These have no other defined use; they are ignored by code
11386 generation and optimization.
11388 '``llvm.annotation.*``' Intrinsic
11389 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11394 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
11395 any integer bit width.
11399 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
11400 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
11401 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
11402 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
11403 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
11408 The '``llvm.annotation``' intrinsic.
11413 The first argument is an integer value (result of some expression), the
11414 second is a pointer to a global string, the third is a pointer to a
11415 global string which is the source file name, and the last argument is
11416 the line number. It returns the value of the first argument.
11421 This intrinsic allows annotations to be put on arbitrary expressions
11422 with arbitrary strings. This can be useful for special purpose
11423 optimizations that want to look for these annotations. These have no
11424 other defined use; they are ignored by code generation and optimization.
11426 '``llvm.trap``' Intrinsic
11427 ^^^^^^^^^^^^^^^^^^^^^^^^^
11434 declare void @llvm.trap() noreturn nounwind
11439 The '``llvm.trap``' intrinsic.
11449 This intrinsic is lowered to the target dependent trap instruction. If
11450 the target does not have a trap instruction, this intrinsic will be
11451 lowered to a call of the ``abort()`` function.
11453 '``llvm.debugtrap``' Intrinsic
11454 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11461 declare void @llvm.debugtrap() nounwind
11466 The '``llvm.debugtrap``' intrinsic.
11476 This intrinsic is lowered to code which is intended to cause an
11477 execution trap with the intention of requesting the attention of a
11480 '``llvm.stackprotector``' Intrinsic
11481 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11488 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
11493 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
11494 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
11495 is placed on the stack before local variables.
11500 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
11501 The first argument is the value loaded from the stack guard
11502 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
11503 enough space to hold the value of the guard.
11508 This intrinsic causes the prologue/epilogue inserter to force the position of
11509 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
11510 to ensure that if a local variable on the stack is overwritten, it will destroy
11511 the value of the guard. When the function exits, the guard on the stack is
11512 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
11513 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
11514 calling the ``__stack_chk_fail()`` function.
11516 '``llvm.stackprotectorcheck``' Intrinsic
11517 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11524 declare void @llvm.stackprotectorcheck(i8** <guard>)
11529 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
11530 created stack protector and if they are not equal calls the
11531 ``__stack_chk_fail()`` function.
11536 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
11537 the variable ``@__stack_chk_guard``.
11542 This intrinsic is provided to perform the stack protector check by comparing
11543 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
11544 values do not match call the ``__stack_chk_fail()`` function.
11546 The reason to provide this as an IR level intrinsic instead of implementing it
11547 via other IR operations is that in order to perform this operation at the IR
11548 level without an intrinsic, one would need to create additional basic blocks to
11549 handle the success/failure cases. This makes it difficult to stop the stack
11550 protector check from disrupting sibling tail calls in Codegen. With this
11551 intrinsic, we are able to generate the stack protector basic blocks late in
11552 codegen after the tail call decision has occurred.
11554 '``llvm.objectsize``' Intrinsic
11555 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11562 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
11563 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
11568 The ``llvm.objectsize`` intrinsic is designed to provide information to
11569 the optimizers to determine at compile time whether a) an operation
11570 (like memcpy) will overflow a buffer that corresponds to an object, or
11571 b) that a runtime check for overflow isn't necessary. An object in this
11572 context means an allocation of a specific class, structure, array, or
11578 The ``llvm.objectsize`` intrinsic takes two arguments. The first
11579 argument is a pointer to or into the ``object``. The second argument is
11580 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
11581 or -1 (if false) when the object size is unknown. The second argument
11582 only accepts constants.
11587 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
11588 the size of the object concerned. If the size cannot be determined at
11589 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
11590 on the ``min`` argument).
11592 '``llvm.expect``' Intrinsic
11593 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
11598 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
11603 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
11604 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
11605 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
11610 The ``llvm.expect`` intrinsic provides information about expected (the
11611 most probable) value of ``val``, which can be used by optimizers.
11616 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
11617 a value. The second argument is an expected value, this needs to be a
11618 constant value, variables are not allowed.
11623 This intrinsic is lowered to the ``val``.
11627 '``llvm.assume``' Intrinsic
11628 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11635 declare void @llvm.assume(i1 %cond)
11640 The ``llvm.assume`` allows the optimizer to assume that the provided
11641 condition is true. This information can then be used in simplifying other parts
11647 The condition which the optimizer may assume is always true.
11652 The intrinsic allows the optimizer to assume that the provided condition is
11653 always true whenever the control flow reaches the intrinsic call. No code is
11654 generated for this intrinsic, and instructions that contribute only to the
11655 provided condition are not used for code generation. If the condition is
11656 violated during execution, the behavior is undefined.
11658 Note that the optimizer might limit the transformations performed on values
11659 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
11660 only used to form the intrinsic's input argument. This might prove undesirable
11661 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
11662 sufficient overall improvement in code quality. For this reason,
11663 ``llvm.assume`` should not be used to document basic mathematical invariants
11664 that the optimizer can otherwise deduce or facts that are of little use to the
11669 '``llvm.bitset.test``' Intrinsic
11670 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11677 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
11683 The first argument is a pointer to be tested. The second argument is a
11684 metadata string containing the name of a :doc:`bitset <BitSets>`.
11689 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
11690 member of the given bitset.
11692 '``llvm.donothing``' Intrinsic
11693 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11700 declare void @llvm.donothing() nounwind readnone
11705 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
11706 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
11707 with an invoke instruction.
11717 This intrinsic does nothing, and it's removed by optimizers and ignored
11720 Stack Map Intrinsics
11721 --------------------
11723 LLVM provides experimental intrinsics to support runtime patching
11724 mechanisms commonly desired in dynamic language JITs. These intrinsics
11725 are described in :doc:`StackMaps`.