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 are
133 included in this numbering. For example, if the entry basic block is not
134 given a label name, then it will get number 0.
136 It also shows a convention that we follow in this document. When
137 demonstrating instructions, we will follow an instruction with a comment
138 that defines the type and name of value produced.
146 LLVM programs are composed of ``Module``'s, each of which is a
147 translation unit of the input programs. Each module consists of
148 functions, global variables, and symbol table entries. Modules may be
149 combined together with the LLVM linker, which merges function (and
150 global variable) definitions, resolves forward declarations, and merges
151 symbol table entries. Here is an example of the "hello world" module:
155 ; Declare the string constant as a global constant.
156 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
158 ; External declaration of the puts function
159 declare i32 @puts(i8* nocapture) nounwind
161 ; Definition of main function
162 define i32 @main() { ; i32()*
163 ; Convert [13 x i8]* to i8 *...
164 %cast210 = getelementptr [13 x i8]* @.str, i64 0, i64 0
166 ; Call puts function to write out the string to stdout.
167 call i32 @puts(i8* %cast210)
172 !0 = metadata !{i32 42, null, metadata !"string"}
175 This example is made up of a :ref:`global variable <globalvars>` named
176 "``.str``", an external declaration of the "``puts``" function, a
177 :ref:`function definition <functionstructure>` for "``main``" and
178 :ref:`named metadata <namedmetadatastructure>` "``foo``".
180 In general, a module is made up of a list of global values (where both
181 functions and global variables are global values). Global values are
182 represented by a pointer to a memory location (in this case, a pointer
183 to an array of char, and a pointer to a function), and have one of the
184 following :ref:`linkage types <linkage>`.
191 All Global Variables and Functions have one of the following types of
195 Global values with "``private``" linkage are only directly
196 accessible by objects in the current module. In particular, linking
197 code into a module with an private global value may cause the
198 private to be renamed as necessary to avoid collisions. Because the
199 symbol is private to the module, all references can be updated. This
200 doesn't show up in any symbol table in the object file.
202 Similar to private, but the value shows as a local symbol
203 (``STB_LOCAL`` in the case of ELF) in the object file. This
204 corresponds to the notion of the '``static``' keyword in C.
205 ``available_externally``
206 Globals with "``available_externally``" linkage are never emitted
207 into the object file corresponding to the LLVM module. They exist to
208 allow inlining and other optimizations to take place given knowledge
209 of the definition of the global, which is known to be somewhere
210 outside the module. Globals with ``available_externally`` linkage
211 are allowed to be discarded at will, and are otherwise the same as
212 ``linkonce_odr``. This linkage type is only allowed on definitions,
215 Globals with "``linkonce``" linkage are merged with other globals of
216 the same name when linkage occurs. This can be used to implement
217 some forms of inline functions, templates, or other code which must
218 be generated in each translation unit that uses it, but where the
219 body may be overridden with a more definitive definition later.
220 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
221 that ``linkonce`` linkage does not actually allow the optimizer to
222 inline the body of this function into callers because it doesn't
223 know if this definition of the function is the definitive definition
224 within the program or whether it will be overridden by a stronger
225 definition. To enable inlining and other optimizations, use
226 "``linkonce_odr``" linkage.
228 "``weak``" linkage has the same merging semantics as ``linkonce``
229 linkage, except that unreferenced globals with ``weak`` linkage may
230 not be discarded. This is used for globals that are declared "weak"
233 "``common``" linkage is most similar to "``weak``" linkage, but they
234 are used for tentative definitions in C, such as "``int X;``" at
235 global scope. Symbols with "``common``" linkage are merged in the
236 same way as ``weak symbols``, and they may not be deleted if
237 unreferenced. ``common`` symbols may not have an explicit section,
238 must have a zero initializer, and may not be marked
239 ':ref:`constant <globalvars>`'. Functions and aliases may not have
242 .. _linkage_appending:
245 "``appending``" linkage may only be applied to global variables of
246 pointer to array type. When two global variables with appending
247 linkage are linked together, the two global arrays are appended
248 together. This is the LLVM, typesafe, equivalent of having the
249 system linker append together "sections" with identical names when
252 The semantics of this linkage follow the ELF object file model: the
253 symbol is weak until linked, if not linked, the symbol becomes null
254 instead of being an undefined reference.
255 ``linkonce_odr``, ``weak_odr``
256 Some languages allow differing globals to be merged, such as two
257 functions with different semantics. Other languages, such as
258 ``C++``, ensure that only equivalent globals are ever merged (the
259 "one definition rule" --- "ODR"). Such languages can use the
260 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
261 global will only be merged with equivalent globals. These linkage
262 types are otherwise the same as their non-``odr`` versions.
264 If none of the above identifiers are used, the global is externally
265 visible, meaning that it participates in linkage and can be used to
266 resolve external symbol references.
268 It is illegal for a function *declaration* to have any linkage type
269 other than ``external`` or ``extern_weak``.
276 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
277 :ref:`invokes <i_invoke>` can all have an optional calling convention
278 specified for the call. The calling convention of any pair of dynamic
279 caller/callee must match, or the behavior of the program is undefined.
280 The following calling conventions are supported by LLVM, and more may be
283 "``ccc``" - The C calling convention
284 This calling convention (the default if no other calling convention
285 is specified) matches the target C calling conventions. This calling
286 convention supports varargs function calls and tolerates some
287 mismatch in the declared prototype and implemented declaration of
288 the function (as does normal C).
289 "``fastcc``" - The fast calling convention
290 This calling convention attempts to make calls as fast as possible
291 (e.g. by passing things in registers). This calling convention
292 allows the target to use whatever tricks it wants to produce fast
293 code for the target, without having to conform to an externally
294 specified ABI (Application Binary Interface). `Tail calls can only
295 be optimized when this, the GHC or the HiPE convention is
296 used. <CodeGenerator.html#id80>`_ This calling convention does not
297 support varargs and requires the prototype of all callees to exactly
298 match the prototype of the function definition.
299 "``coldcc``" - The cold calling convention
300 This calling convention attempts to make code in the caller as
301 efficient as possible under the assumption that the call is not
302 commonly executed. As such, these calls often preserve all registers
303 so that the call does not break any live ranges in the caller side.
304 This calling convention does not support varargs and requires the
305 prototype of all callees to exactly match the prototype of the
306 function definition. Furthermore the inliner doesn't consider such function
308 "``cc 10``" - GHC convention
309 This calling convention has been implemented specifically for use by
310 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
311 It passes everything in registers, going to extremes to achieve this
312 by disabling callee save registers. This calling convention should
313 not be used lightly but only for specific situations such as an
314 alternative to the *register pinning* performance technique often
315 used when implementing functional programming languages. At the
316 moment only X86 supports this convention and it has the following
319 - On *X86-32* only supports up to 4 bit type parameters. No
320 floating point types are supported.
321 - On *X86-64* only supports up to 10 bit type parameters and 6
322 floating point parameters.
324 This calling convention supports `tail call
325 optimization <CodeGenerator.html#id80>`_ but requires both the
326 caller and callee are using it.
327 "``cc 11``" - The HiPE calling convention
328 This calling convention has been implemented specifically for use by
329 the `High-Performance Erlang
330 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
331 native code compiler of the `Ericsson's Open Source Erlang/OTP
332 system <http://www.erlang.org/download.shtml>`_. It uses more
333 registers for argument passing than the ordinary C calling
334 convention and defines no callee-saved registers. The calling
335 convention properly supports `tail call
336 optimization <CodeGenerator.html#id80>`_ but requires that both the
337 caller and the callee use it. It uses a *register pinning*
338 mechanism, similar to GHC's convention, for keeping frequently
339 accessed runtime components pinned to specific hardware registers.
340 At the moment only X86 supports this convention (both 32 and 64
342 "``webkit_jscc``" - WebKit's JavaScript calling convention
343 This calling convention has been implemented for `WebKit FTL JIT
344 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
345 stack right to left (as cdecl does), and returns a value in the
346 platform's customary return register.
347 "``anyregcc``" - Dynamic calling convention for code patching
348 This is a special convention that supports patching an arbitrary code
349 sequence in place of a call site. This convention forces the call
350 arguments into registers but allows them to be dynamcially
351 allocated. This can currently only be used with calls to
352 llvm.experimental.patchpoint because only this intrinsic records
353 the location of its arguments in a side table. See :doc:`StackMaps`.
354 "``preserve_mostcc``" - The `PreserveMost` calling convention
355 This calling convention attempts to make the code in the caller as little
356 intrusive as possible. This calling convention behaves identical to the `C`
357 calling convention on how arguments and return values are passed, but it
358 uses a different set of caller/callee-saved registers. This alleviates the
359 burden of saving and recovering a large register set before and after the
360 call in the caller. If the arguments are passed in callee-saved registers,
361 then they will be preserved by the callee across the call. This doesn't
362 apply for values returned in callee-saved registers.
364 - On X86-64 the callee preserves all general purpose registers, except for
365 R11. R11 can be used as a scratch register. Floating-point registers
366 (XMMs/YMMs) are not preserved and need to be saved by the caller.
368 The idea behind this convention is to support calls to runtime functions
369 that have a hot path and a cold path. The hot path is usually a small piece
370 of code that doesn't many registers. The cold path might need to call out to
371 another function and therefore only needs to preserve the caller-saved
372 registers, which haven't already been saved by the caller. The
373 `PreserveMost` calling convention is very similar to the `cold` calling
374 convention in terms of caller/callee-saved registers, but they are used for
375 different types of function calls. `coldcc` is for function calls that are
376 rarely executed, whereas `preserve_mostcc` function calls are intended to be
377 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
378 doesn't prevent the inliner from inlining the function call.
380 This calling convention will be used by a future version of the ObjectiveC
381 runtime and should therefore still be considered experimental at this time.
382 Although this convention was created to optimize certain runtime calls to
383 the ObjectiveC runtime, it is not limited to this runtime and might be used
384 by other runtimes in the future too. The current implementation only
385 supports X86-64, but the intention is to support more architectures in the
387 "``preserve_allcc``" - The `PreserveAll` calling convention
388 This calling convention attempts to make the code in the caller even less
389 intrusive than the `PreserveMost` calling convention. This calling
390 convention also behaves identical to the `C` calling convention on how
391 arguments and return values are passed, but it uses a different set of
392 caller/callee-saved registers. This removes the burden of saving and
393 recovering a large register set before and after the call in the caller. If
394 the arguments are passed in callee-saved registers, then they will be
395 preserved by the callee across the call. This doesn't apply for values
396 returned in callee-saved registers.
398 - On X86-64 the callee preserves all general purpose registers, except for
399 R11. R11 can be used as a scratch register. Furthermore it also preserves
400 all floating-point registers (XMMs/YMMs).
402 The idea behind this convention is to support calls to runtime functions
403 that don't need to call out to any other functions.
405 This calling convention, like the `PreserveMost` calling convention, will be
406 used by a future version of the ObjectiveC runtime and should be considered
407 experimental at this time.
408 "``cc <n>``" - Numbered convention
409 Any calling convention may be specified by number, allowing
410 target-specific calling conventions to be used. Target specific
411 calling conventions start at 64.
413 More calling conventions can be added/defined on an as-needed basis, to
414 support Pascal conventions or any other well-known target-independent
417 .. _visibilitystyles:
422 All Global Variables and Functions have one of the following visibility
425 "``default``" - Default style
426 On targets that use the ELF object file format, default visibility
427 means that the declaration is visible to other modules and, in
428 shared libraries, means that the declared entity may be overridden.
429 On Darwin, default visibility means that the declaration is visible
430 to other modules. Default visibility corresponds to "external
431 linkage" in the language.
432 "``hidden``" - Hidden style
433 Two declarations of an object with hidden visibility refer to the
434 same object if they are in the same shared object. Usually, hidden
435 visibility indicates that the symbol will not be placed into the
436 dynamic symbol table, so no other module (executable or shared
437 library) can reference it directly.
438 "``protected``" - Protected style
439 On ELF, protected visibility indicates that the symbol will be
440 placed in the dynamic symbol table, but that references within the
441 defining module will bind to the local symbol. That is, the symbol
442 cannot be overridden by another module.
444 A symbol with ``internal`` or ``private`` linkage must have ``default``
452 All Global Variables, Functions and Aliases can have one of the following
456 "``dllimport``" causes the compiler to reference a function or variable via
457 a global pointer to a pointer that is set up by the DLL exporting the
458 symbol. On Microsoft Windows targets, the pointer name is formed by
459 combining ``__imp_`` and the function or variable name.
461 "``dllexport``" causes the compiler to provide a global pointer to a pointer
462 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
463 Microsoft Windows targets, the pointer name is formed by combining
464 ``__imp_`` and the function or variable name. Since this storage class
465 exists for defining a dll interface, the compiler, assembler and linker know
466 it is externally referenced and must refrain from deleting the symbol.
470 Thread Local Storage Models
471 ---------------------------
473 A variable may be defined as ``thread_local``, which means that it will
474 not be shared by threads (each thread will have a separated copy of the
475 variable). Not all targets support thread-local variables. Optionally, a
476 TLS model may be specified:
479 For variables that are only used within the current shared library.
481 For variables in modules that will not be loaded dynamically.
483 For variables defined in the executable and only used within it.
485 If no explicit model is given, the "general dynamic" model is used.
487 The models correspond to the ELF TLS models; see `ELF Handling For
488 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
489 more information on under which circumstances the different models may
490 be used. The target may choose a different TLS model if the specified
491 model is not supported, or if a better choice of model can be made.
493 A model can also be specified in a alias, but then it only governs how
494 the alias is accessed. It will not have any effect in the aliasee.
501 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
502 types <t_struct>`. Literal types are uniqued structurally, but identified types
503 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
504 to forward declare a type that is not yet available.
506 An example of a identified structure specification is:
510 %mytype = type { %mytype*, i32 }
512 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
513 literal types are uniqued in recent versions of LLVM.
520 Global variables define regions of memory allocated at compilation time
523 Global variables definitions must be initialized.
525 Global variables in other translation units can also be declared, in which
526 case they don't have an initializer.
528 Either global variable definitions or declarations may have an explicit section
529 to be placed in and may have an optional explicit alignment specified.
531 A variable may be defined as a global ``constant``, which indicates that
532 the contents of the variable will **never** be modified (enabling better
533 optimization, allowing the global data to be placed in the read-only
534 section of an executable, etc). Note that variables that need runtime
535 initialization cannot be marked ``constant`` as there is a store to the
538 LLVM explicitly allows *declarations* of global variables to be marked
539 constant, even if the final definition of the global is not. This
540 capability can be used to enable slightly better optimization of the
541 program, but requires the language definition to guarantee that
542 optimizations based on the 'constantness' are valid for the translation
543 units that do not include the definition.
545 As SSA values, global variables define pointer values that are in scope
546 (i.e. they dominate) all basic blocks in the program. Global variables
547 always define a pointer to their "content" type because they describe a
548 region of memory, and all memory objects in LLVM are accessed through
551 Global variables can be marked with ``unnamed_addr`` which indicates
552 that the address is not significant, only the content. Constants marked
553 like this can be merged with other constants if they have the same
554 initializer. Note that a constant with significant address *can* be
555 merged with a ``unnamed_addr`` constant, the result being a constant
556 whose address is significant.
558 A global variable may be declared to reside in a target-specific
559 numbered address space. For targets that support them, address spaces
560 may affect how optimizations are performed and/or what target
561 instructions are used to access the variable. The default address space
562 is zero. The address space qualifier must precede any other attributes.
564 LLVM allows an explicit section to be specified for globals. If the
565 target supports it, it will emit globals to the section specified.
566 Additionally, the global can placed in a comdat if the target has the necessary
569 By default, global initializers are optimized by assuming that global
570 variables defined within the module are not modified from their
571 initial values before the start of the global initializer. This is
572 true even for variables potentially accessible from outside the
573 module, including those with external linkage or appearing in
574 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
575 by marking the variable with ``externally_initialized``.
577 An explicit alignment may be specified for a global, which must be a
578 power of 2. If not present, or if the alignment is set to zero, the
579 alignment of the global is set by the target to whatever it feels
580 convenient. If an explicit alignment is specified, the global is forced
581 to have exactly that alignment. Targets and optimizers are not allowed
582 to over-align the global if the global has an assigned section. In this
583 case, the extra alignment could be observable: for example, code could
584 assume that the globals are densely packed in their section and try to
585 iterate over them as an array, alignment padding would break this
586 iteration. The maximum alignment is ``1 << 29``.
588 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
590 Variables and aliasaes can have a
591 :ref:`Thread Local Storage Model <tls_model>`.
595 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
596 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
597 <global | constant> <Type> [<InitializerConstant>]
598 [, section "name"] [, align <Alignment>]
600 For example, the following defines a global in a numbered address space
601 with an initializer, section, and alignment:
605 @G = addrspace(5) constant float 1.0, section "foo", align 4
607 The following example just declares a global variable
611 @G = external global i32
613 The following example defines a thread-local global with the
614 ``initialexec`` TLS model:
618 @G = thread_local(initialexec) global i32 0, align 4
620 .. _functionstructure:
625 LLVM function definitions consist of the "``define``" keyword, an
626 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
627 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
628 an optional :ref:`calling convention <callingconv>`,
629 an optional ``unnamed_addr`` attribute, a return type, an optional
630 :ref:`parameter attribute <paramattrs>` for the return type, a function
631 name, a (possibly empty) argument list (each with optional :ref:`parameter
632 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
633 an optional section, an optional alignment,
634 an optional :ref:`comdat <langref_comdats>`,
635 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
636 curly brace, a list of basic blocks, and a closing curly brace.
638 LLVM function declarations consist of the "``declare``" keyword, an
639 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
640 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
641 an optional :ref:`calling convention <callingconv>`,
642 an optional ``unnamed_addr`` attribute, a return type, an optional
643 :ref:`parameter attribute <paramattrs>` for the return type, a function
644 name, a possibly empty list of arguments, an optional alignment, an optional
645 :ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
647 A function definition contains a list of basic blocks, forming the CFG (Control
648 Flow Graph) for the function. Each basic block may optionally start with a label
649 (giving the basic block a symbol table entry), contains a list of instructions,
650 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
651 function return). If an explicit label is not provided, a block is assigned an
652 implicit numbered label, using the next value from the same counter as used for
653 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
654 entry block does not have an explicit label, it will be assigned label "%0",
655 then the first unnamed temporary in that block will be "%1", etc.
657 The first basic block in a function is special in two ways: it is
658 immediately executed on entrance to the function, and it is not allowed
659 to have predecessor basic blocks (i.e. there can not be any branches to
660 the entry block of a function). Because the block can have no
661 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
663 LLVM allows an explicit section to be specified for functions. If the
664 target supports it, it will emit functions to the section specified.
665 Additionally, the function can placed in a COMDAT.
667 An explicit alignment may be specified for a function. If not present,
668 or if the alignment is set to zero, the alignment of the function is set
669 by the target to whatever it feels convenient. If an explicit alignment
670 is specified, the function is forced to have at least that much
671 alignment. All alignments must be a power of 2.
673 If the ``unnamed_addr`` attribute is given, the address is know to not
674 be significant and two identical functions can be merged.
678 define [linkage] [visibility] [DLLStorageClass]
680 <ResultType> @<FunctionName> ([argument list])
681 [unnamed_addr] [fn Attrs] [section "name"] [comdat $<ComdatName>]
682 [align N] [gc] [prefix Constant] { ... }
689 Aliases, unlike function or variables, don't create any new data. They
690 are just a new symbol and metadata for an existing position.
692 Aliases have a name and an aliasee that is either a global value or a
695 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
696 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
697 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
701 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy> @<Aliasee>
703 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
704 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
705 might not correctly handle dropping a weak symbol that is aliased.
707 Alias that are not ``unnamed_addr`` are guaranteed to have the same address as
708 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
711 Since aliases are only a second name, some restrictions apply, of which
712 some can only be checked when producing an object file:
714 * The expression defining the aliasee must be computable at assembly
715 time. Since it is just a name, no relocations can be used.
717 * No alias in the expression can be weak as the possibility of the
718 intermediate alias being overridden cannot be represented in an
721 * No global value in the expression can be a declaration, since that
722 would require a relocation, which is not possible.
729 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
731 Comdats have a name which represents the COMDAT key. All global objects that
732 specify this key will only end up in the final object file if the linker chooses
733 that key over some other key. Aliases are placed in the same COMDAT that their
734 aliasee computes to, if any.
736 Comdats have a selection kind to provide input on how the linker should
737 choose between keys in two different object files.
741 $<Name> = comdat SelectionKind
743 The selection kind must be one of the following:
746 The linker may choose any COMDAT key, the choice is arbitrary.
748 The linker may choose any COMDAT key but the sections must contain the
751 The linker will choose the section containing the largest COMDAT key.
753 The linker requires that only section with this COMDAT key exist.
755 The linker may choose any COMDAT key but the sections must contain the
758 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
759 ``any`` as a selection kind.
761 Here is an example of a COMDAT group where a function will only be selected if
762 the COMDAT key's section is the largest:
766 $foo = comdat largest
767 @foo = global i32 2, comdat $foo
769 define void @bar() comdat $foo {
773 In a COFF object file, this will create a COMDAT section with selection kind
774 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
775 and another COMDAT section with selection kind
776 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
777 section and contains the contents of the ``@baz`` symbol.
779 There are some restrictions on the properties of the global object.
780 It, or an alias to it, must have the same name as the COMDAT group when
782 The contents and size of this object may be used during link-time to determine
783 which COMDAT groups get selected depending on the selection kind.
784 Because the name of the object must match the name of the COMDAT group, the
785 linkage of the global object must not be local; local symbols can get renamed
786 if a collision occurs in the symbol table.
788 The combined use of COMDATS and section attributes may yield surprising results.
795 @g1 = global i32 42, section "sec", comdat $foo
796 @g2 = global i32 42, section "sec", comdat $bar
798 From the object file perspective, this requires the creation of two sections
799 with the same name. This is necessary because both globals belong to different
800 COMDAT groups and COMDATs, at the object file level, are represented by
803 Note that certain IR constructs like global variables and functions may create
804 COMDATs in the object file in addition to any which are specified using COMDAT
805 IR. This arises, for example, when a global variable has linkonce_odr linkage.
807 .. _namedmetadatastructure:
812 Named metadata is a collection of metadata. :ref:`Metadata
813 nodes <metadata>` (but not metadata strings) are the only valid
814 operands for a named metadata.
818 ; Some unnamed metadata nodes, which are referenced by the named metadata.
819 !0 = metadata !{metadata !"zero"}
820 !1 = metadata !{metadata !"one"}
821 !2 = metadata !{metadata !"two"}
823 !name = !{!0, !1, !2}
830 The return type and each parameter of a function type may have a set of
831 *parameter attributes* associated with them. Parameter attributes are
832 used to communicate additional information about the result or
833 parameters of a function. Parameter attributes are considered to be part
834 of the function, not of the function type, so functions with different
835 parameter attributes can have the same function type.
837 Parameter attributes are simple keywords that follow the type specified.
838 If multiple parameter attributes are needed, they are space separated.
843 declare i32 @printf(i8* noalias nocapture, ...)
844 declare i32 @atoi(i8 zeroext)
845 declare signext i8 @returns_signed_char()
847 Note that any attributes for the function result (``nounwind``,
848 ``readonly``) come immediately after the argument list.
850 Currently, only the following parameter attributes are defined:
853 This indicates to the code generator that the parameter or return
854 value should be zero-extended to the extent required by the target's
855 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
856 the caller (for a parameter) or the callee (for a return value).
858 This indicates to the code generator that the parameter or return
859 value should be sign-extended to the extent required by the target's
860 ABI (which is usually 32-bits) by the caller (for a parameter) or
861 the callee (for a return value).
863 This indicates that this parameter or return value should be treated
864 in a special target-dependent fashion during while emitting code for
865 a function call or return (usually, by putting it in a register as
866 opposed to memory, though some targets use it to distinguish between
867 two different kinds of registers). Use of this attribute is
870 This indicates that the pointer parameter should really be passed by
871 value to the function. The attribute implies that a hidden copy of
872 the pointee is made between the caller and the callee, so the callee
873 is unable to modify the value in the caller. This attribute is only
874 valid on LLVM pointer arguments. It is generally used to pass
875 structs and arrays by value, but is also valid on pointers to
876 scalars. The copy is considered to belong to the caller not the
877 callee (for example, ``readonly`` functions should not write to
878 ``byval`` parameters). This is not a valid attribute for return
881 The byval attribute also supports specifying an alignment with the
882 align attribute. It indicates the alignment of the stack slot to
883 form and the known alignment of the pointer specified to the call
884 site. If the alignment is not specified, then the code generator
885 makes a target-specific assumption.
891 The ``inalloca`` argument attribute allows the caller to take the
892 address of outgoing stack arguments. An ``inalloca`` argument must
893 be a pointer to stack memory produced by an ``alloca`` instruction.
894 The alloca, or argument allocation, must also be tagged with the
895 inalloca keyword. Only the last argument may have the ``inalloca``
896 attribute, and that argument is guaranteed to be passed in memory.
898 An argument allocation may be used by a call at most once because
899 the call may deallocate it. The ``inalloca`` attribute cannot be
900 used in conjunction with other attributes that affect argument
901 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
902 ``inalloca`` attribute also disables LLVM's implicit lowering of
903 large aggregate return values, which means that frontend authors
904 must lower them with ``sret`` pointers.
906 When the call site is reached, the argument allocation must have
907 been the most recent stack allocation that is still live, or the
908 results are undefined. It is possible to allocate additional stack
909 space after an argument allocation and before its call site, but it
910 must be cleared off with :ref:`llvm.stackrestore
913 See :doc:`InAlloca` for more information on how to use this
917 This indicates that the pointer parameter specifies the address of a
918 structure that is the return value of the function in the source
919 program. This pointer must be guaranteed by the caller to be valid:
920 loads and stores to the structure may be assumed by the callee
921 not to trap and to be properly aligned. This may only be applied to
922 the first parameter. This is not a valid attribute for return
926 This indicates that the pointer value may be assumed by the optimizer to
927 have the specified alignment.
929 Note that this attribute has additional semantics when combined with the
935 This indicates that pointer values :ref:`based <pointeraliasing>` on
936 the argument or return value do not alias pointer values that are
937 not *based* on it, ignoring certain "irrelevant" dependencies. For a
938 call to the parent function, dependencies between memory references
939 from before or after the call and from those during the call are
940 "irrelevant" to the ``noalias`` keyword for the arguments and return
941 value used in that call. The caller shares the responsibility with
942 the callee for ensuring that these requirements are met. For further
943 details, please see the discussion of the NoAlias response in :ref:`alias
944 analysis <Must, May, or No>`.
946 Note that this definition of ``noalias`` is intentionally similar
947 to the definition of ``restrict`` in C99 for function arguments,
948 though it is slightly weaker.
950 For function return values, C99's ``restrict`` is not meaningful,
951 while LLVM's ``noalias`` is.
953 This indicates that the callee does not make any copies of the
954 pointer that outlive the callee itself. This is not a valid
955 attribute for return values.
960 This indicates that the pointer parameter can be excised using the
961 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
962 attribute for return values and can only be applied to one parameter.
965 This indicates that the function always returns the argument as its return
966 value. This is an optimization hint to the code generator when generating
967 the caller, allowing tail call optimization and omission of register saves
968 and restores in some cases; it is not checked or enforced when generating
969 the callee. The parameter and the function return type must be valid
970 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
971 valid attribute for return values and can only be applied to one parameter.
974 This indicates that the parameter or return pointer is not null. This
975 attribute may only be applied to pointer typed parameters. This is not
976 checked or enforced by LLVM, the caller must ensure that the pointer
977 passed in is non-null, or the callee must ensure that the returned pointer
980 ``dereferenceable(<n>)``
981 This indicates that the parameter or return pointer is dereferenceable. This
982 attribute may only be applied to pointer typed parameters. A pointer that
983 is dereferenceable can be loaded from speculatively without a risk of
984 trapping. The number of bytes known to be dereferenceable must be provided
985 in parentheses. It is legal for the number of bytes to be less than the
986 size of the pointee type. The ``nonnull`` attribute does not imply
987 dereferenceability (consider a pointer to one element past the end of an
988 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
989 ``addrspace(0)`` (which is the default address space).
993 Garbage Collector Names
994 -----------------------
996 Each function may specify a garbage collector name, which is simply a
1001 define void @f() gc "name" { ... }
1003 The compiler declares the supported values of *name*. Specifying a
1004 collector will cause the compiler to alter its output in order to
1005 support the named garbage collection algorithm.
1012 Prefix data is data associated with a function which the code generator
1013 will emit immediately before the function body. The purpose of this feature
1014 is to allow frontends to associate language-specific runtime metadata with
1015 specific functions and make it available through the function pointer while
1016 still allowing the function pointer to be called. To access the data for a
1017 given function, a program may bitcast the function pointer to a pointer to
1018 the constant's type. This implies that the IR symbol points to the start
1021 To maintain the semantics of ordinary function calls, the prefix data must
1022 have a particular format. Specifically, it must begin with a sequence of
1023 bytes which decode to a sequence of machine instructions, valid for the
1024 module's target, which transfer control to the point immediately succeeding
1025 the prefix data, without performing any other visible action. This allows
1026 the inliner and other passes to reason about the semantics of the function
1027 definition without needing to reason about the prefix data. Obviously this
1028 makes the format of the prefix data highly target dependent.
1030 Prefix data is laid out as if it were an initializer for a global variable
1031 of the prefix data's type. No padding is automatically placed between the
1032 prefix data and the function body. If padding is required, it must be part
1035 A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
1036 which encodes the ``nop`` instruction:
1038 .. code-block:: llvm
1040 define void @f() prefix i8 144 { ... }
1042 Generally prefix data can be formed by encoding a relative branch instruction
1043 which skips the metadata, as in this example of valid prefix data for the
1044 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1046 .. code-block:: llvm
1048 %0 = type <{ i8, i8, i8* }>
1050 define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
1052 A function may have prefix data but no body. This has similar semantics
1053 to the ``available_externally`` linkage in that the data may be used by the
1054 optimizers but will not be emitted in the object file.
1061 Attribute groups are groups of attributes that are referenced by objects within
1062 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1063 functions will use the same set of attributes. In the degenerative case of a
1064 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1065 group will capture the important command line flags used to build that file.
1067 An attribute group is a module-level object. To use an attribute group, an
1068 object references the attribute group's ID (e.g. ``#37``). An object may refer
1069 to more than one attribute group. In that situation, the attributes from the
1070 different groups are merged.
1072 Here is an example of attribute groups for a function that should always be
1073 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1075 .. code-block:: llvm
1077 ; Target-independent attributes:
1078 attributes #0 = { alwaysinline alignstack=4 }
1080 ; Target-dependent attributes:
1081 attributes #1 = { "no-sse" }
1083 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1084 define void @f() #0 #1 { ... }
1091 Function attributes are set to communicate additional information about
1092 a function. Function attributes are considered to be part of the
1093 function, not of the function type, so functions with different function
1094 attributes can have the same function type.
1096 Function attributes are simple keywords that follow the type specified.
1097 If multiple attributes are needed, they are space separated. For
1100 .. code-block:: llvm
1102 define void @f() noinline { ... }
1103 define void @f() alwaysinline { ... }
1104 define void @f() alwaysinline optsize { ... }
1105 define void @f() optsize { ... }
1108 This attribute indicates that, when emitting the prologue and
1109 epilogue, the backend should forcibly align the stack pointer.
1110 Specify the desired alignment, which must be a power of two, in
1113 This attribute indicates that the inliner should attempt to inline
1114 this function into callers whenever possible, ignoring any active
1115 inlining size threshold for this caller.
1117 This indicates that the callee function at a call site should be
1118 recognized as a built-in function, even though the function's declaration
1119 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1120 direct calls to functions that are declared with the ``nobuiltin``
1123 This attribute indicates that this function is rarely called. When
1124 computing edge weights, basic blocks post-dominated by a cold
1125 function call are also considered to be cold; and, thus, given low
1128 This attribute indicates that the source code contained a hint that
1129 inlining this function is desirable (such as the "inline" keyword in
1130 C/C++). It is just a hint; it imposes no requirements on the
1133 This attribute indicates that the function should be added to a
1134 jump-instruction table at code-generation time, and that all address-taken
1135 references to this function should be replaced with a reference to the
1136 appropriate jump-instruction-table function pointer. Note that this creates
1137 a new pointer for the original function, which means that code that depends
1138 on function-pointer identity can break. So, any function annotated with
1139 ``jumptable`` must also be ``unnamed_addr``.
1141 This attribute suggests that optimization passes and code generator
1142 passes make choices that keep the code size of this function as small
1143 as possible and perform optimizations that may sacrifice runtime
1144 performance in order to minimize the size of the generated code.
1146 This attribute disables prologue / epilogue emission for the
1147 function. This can have very system-specific consequences.
1149 This indicates that the callee function at a call site is not recognized as
1150 a built-in function. LLVM will retain the original call and not replace it
1151 with equivalent code based on the semantics of the built-in function, unless
1152 the call site uses the ``builtin`` attribute. This is valid at call sites
1153 and on function declarations and definitions.
1155 This attribute indicates that calls to the function cannot be
1156 duplicated. A call to a ``noduplicate`` function may be moved
1157 within its parent function, but may not be duplicated within
1158 its parent function.
1160 A function containing a ``noduplicate`` call may still
1161 be an inlining candidate, provided that the call is not
1162 duplicated by inlining. That implies that the function has
1163 internal linkage and only has one call site, so the original
1164 call is dead after inlining.
1166 This attributes disables implicit floating point instructions.
1168 This attribute indicates that the inliner should never inline this
1169 function in any situation. This attribute may not be used together
1170 with the ``alwaysinline`` attribute.
1172 This attribute suppresses lazy symbol binding for the function. This
1173 may make calls to the function faster, at the cost of extra program
1174 startup time if the function is not called during program startup.
1176 This attribute indicates that the code generator should not use a
1177 red zone, even if the target-specific ABI normally permits it.
1179 This function attribute indicates that the function never returns
1180 normally. This produces undefined behavior at runtime if the
1181 function ever does dynamically return.
1183 This function attribute indicates that the function never returns
1184 with an unwind or exceptional control flow. If the function does
1185 unwind, its runtime behavior is undefined.
1187 This function attribute indicates that the function is not optimized
1188 by any optimization or code generator passes with the
1189 exception of interprocedural optimization passes.
1190 This attribute cannot be used together with the ``alwaysinline``
1191 attribute; this attribute is also incompatible
1192 with the ``minsize`` attribute and the ``optsize`` attribute.
1194 This attribute requires the ``noinline`` attribute to be specified on
1195 the function as well, so the function is never inlined into any caller.
1196 Only functions with the ``alwaysinline`` attribute are valid
1197 candidates for inlining into the body of this function.
1199 This attribute suggests that optimization passes and code generator
1200 passes make choices that keep the code size of this function low,
1201 and otherwise do optimizations specifically to reduce code size as
1202 long as they do not significantly impact runtime performance.
1204 On a function, this attribute indicates that the function computes its
1205 result (or decides to unwind an exception) based strictly on its arguments,
1206 without dereferencing any pointer arguments or otherwise accessing
1207 any mutable state (e.g. memory, control registers, etc) visible to
1208 caller functions. It does not write through any pointer arguments
1209 (including ``byval`` arguments) and never changes any state visible
1210 to callers. This means that it cannot unwind exceptions by calling
1211 the ``C++`` exception throwing methods.
1213 On an argument, this attribute indicates that the function does not
1214 dereference that pointer argument, even though it may read or write the
1215 memory that the pointer points to if accessed through other pointers.
1217 On a function, this attribute indicates that the function does not write
1218 through any pointer arguments (including ``byval`` arguments) or otherwise
1219 modify any state (e.g. memory, control registers, etc) visible to
1220 caller functions. It may dereference pointer arguments and read
1221 state that may be set in the caller. A readonly function always
1222 returns the same value (or unwinds an exception identically) when
1223 called with the same set of arguments and global state. It cannot
1224 unwind an exception by calling the ``C++`` exception throwing
1227 On an argument, this attribute indicates that the function does not write
1228 through this pointer argument, even though it may write to the memory that
1229 the pointer points to.
1231 This attribute indicates that this function can return twice. The C
1232 ``setjmp`` is an example of such a function. The compiler disables
1233 some optimizations (like tail calls) in the caller of these
1235 ``sanitize_address``
1236 This attribute indicates that AddressSanitizer checks
1237 (dynamic address safety analysis) are enabled for this function.
1239 This attribute indicates that MemorySanitizer checks (dynamic detection
1240 of accesses to uninitialized memory) are enabled for this function.
1242 This attribute indicates that ThreadSanitizer checks
1243 (dynamic thread safety analysis) are enabled for this function.
1245 This attribute indicates that the function should emit a stack
1246 smashing protector. It is in the form of a "canary" --- a random value
1247 placed on the stack before the local variables that's checked upon
1248 return from the function to see if it has been overwritten. A
1249 heuristic is used to determine if a function needs stack protectors
1250 or not. The heuristic used will enable protectors for functions with:
1252 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1253 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1254 - Calls to alloca() with variable sizes or constant sizes greater than
1255 ``ssp-buffer-size``.
1257 Variables that are identified as requiring a protector will be arranged
1258 on the stack such that they are adjacent to the stack protector guard.
1260 If a function that has an ``ssp`` attribute is inlined into a
1261 function that doesn't have an ``ssp`` attribute, then the resulting
1262 function will have an ``ssp`` attribute.
1264 This attribute indicates that the function should *always* emit a
1265 stack smashing protector. This overrides the ``ssp`` function
1268 Variables that are identified as requiring a protector will be arranged
1269 on the stack such that they are adjacent to the stack protector guard.
1270 The specific layout rules are:
1272 #. Large arrays and structures containing large arrays
1273 (``>= ssp-buffer-size``) are closest to the stack protector.
1274 #. Small arrays and structures containing small arrays
1275 (``< ssp-buffer-size``) are 2nd closest to the protector.
1276 #. Variables that have had their address taken are 3rd closest to the
1279 If a function that has an ``sspreq`` attribute is inlined into a
1280 function that doesn't have an ``sspreq`` attribute or which has an
1281 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1282 an ``sspreq`` attribute.
1284 This attribute indicates that the function should emit a stack smashing
1285 protector. This attribute causes a strong heuristic to be used when
1286 determining if a function needs stack protectors. The strong heuristic
1287 will enable protectors for functions with:
1289 - Arrays of any size and type
1290 - Aggregates containing an array of any size and type.
1291 - Calls to alloca().
1292 - Local variables that have had their address taken.
1294 Variables that are identified as requiring a protector will be arranged
1295 on the stack such that they are adjacent to the stack protector guard.
1296 The specific layout rules are:
1298 #. Large arrays and structures containing large arrays
1299 (``>= ssp-buffer-size``) are closest to the stack protector.
1300 #. Small arrays and structures containing small arrays
1301 (``< ssp-buffer-size``) are 2nd closest to the protector.
1302 #. Variables that have had their address taken are 3rd closest to the
1305 This overrides the ``ssp`` function attribute.
1307 If a function that has an ``sspstrong`` attribute is inlined into a
1308 function that doesn't have an ``sspstrong`` attribute, then the
1309 resulting function will have an ``sspstrong`` attribute.
1311 This attribute indicates that the ABI being targeted requires that
1312 an unwind table entry be produce for this function even if we can
1313 show that no exceptions passes by it. This is normally the case for
1314 the ELF x86-64 abi, but it can be disabled for some compilation
1319 Module-Level Inline Assembly
1320 ----------------------------
1322 Modules may contain "module-level inline asm" blocks, which corresponds
1323 to the GCC "file scope inline asm" blocks. These blocks are internally
1324 concatenated by LLVM and treated as a single unit, but may be separated
1325 in the ``.ll`` file if desired. The syntax is very simple:
1327 .. code-block:: llvm
1329 module asm "inline asm code goes here"
1330 module asm "more can go here"
1332 The strings can contain any character by escaping non-printable
1333 characters. The escape sequence used is simply "\\xx" where "xx" is the
1334 two digit hex code for the number.
1336 The inline asm code is simply printed to the machine code .s file when
1337 assembly code is generated.
1339 .. _langref_datalayout:
1344 A module may specify a target specific data layout string that specifies
1345 how data is to be laid out in memory. The syntax for the data layout is
1348 .. code-block:: llvm
1350 target datalayout = "layout specification"
1352 The *layout specification* consists of a list of specifications
1353 separated by the minus sign character ('-'). Each specification starts
1354 with a letter and may include other information after the letter to
1355 define some aspect of the data layout. The specifications accepted are
1359 Specifies that the target lays out data in big-endian form. That is,
1360 the bits with the most significance have the lowest address
1363 Specifies that the target lays out data in little-endian form. That
1364 is, the bits with the least significance have the lowest address
1367 Specifies the natural alignment of the stack in bits. Alignment
1368 promotion of stack variables is limited to the natural stack
1369 alignment to avoid dynamic stack realignment. The stack alignment
1370 must be a multiple of 8-bits. If omitted, the natural stack
1371 alignment defaults to "unspecified", which does not prevent any
1372 alignment promotions.
1373 ``p[n]:<size>:<abi>:<pref>``
1374 This specifies the *size* of a pointer and its ``<abi>`` and
1375 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1376 bits. The address space, ``n`` is optional, and if not specified,
1377 denotes the default address space 0. The value of ``n`` must be
1378 in the range [1,2^23).
1379 ``i<size>:<abi>:<pref>``
1380 This specifies the alignment for an integer type of a given bit
1381 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1382 ``v<size>:<abi>:<pref>``
1383 This specifies the alignment for a vector type of a given bit
1385 ``f<size>:<abi>:<pref>``
1386 This specifies the alignment for a floating point type of a given bit
1387 ``<size>``. Only values of ``<size>`` that are supported by the target
1388 will work. 32 (float) and 64 (double) are supported on all targets; 80
1389 or 128 (different flavors of long double) are also supported on some
1392 This specifies the alignment for an object of aggregate type.
1394 If present, specifies that llvm names are mangled in the output. The
1397 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1398 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1399 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1400 symbols get a ``_`` prefix.
1401 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1402 functions also get a suffix based on the frame size.
1403 ``n<size1>:<size2>:<size3>...``
1404 This specifies a set of native integer widths for the target CPU in
1405 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1406 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1407 this set are considered to support most general arithmetic operations
1410 On every specification that takes a ``<abi>:<pref>``, specifying the
1411 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1412 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1414 When constructing the data layout for a given target, LLVM starts with a
1415 default set of specifications which are then (possibly) overridden by
1416 the specifications in the ``datalayout`` keyword. The default
1417 specifications are given in this list:
1419 - ``E`` - big endian
1420 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1421 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1422 same as the default address space.
1423 - ``S0`` - natural stack alignment is unspecified
1424 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1425 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1426 - ``i16:16:16`` - i16 is 16-bit aligned
1427 - ``i32:32:32`` - i32 is 32-bit aligned
1428 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1429 alignment of 64-bits
1430 - ``f16:16:16`` - half is 16-bit aligned
1431 - ``f32:32:32`` - float is 32-bit aligned
1432 - ``f64:64:64`` - double is 64-bit aligned
1433 - ``f128:128:128`` - quad is 128-bit aligned
1434 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1435 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1436 - ``a:0:64`` - aggregates are 64-bit aligned
1438 When LLVM is determining the alignment for a given type, it uses the
1441 #. If the type sought is an exact match for one of the specifications,
1442 that specification is used.
1443 #. If no match is found, and the type sought is an integer type, then
1444 the smallest integer type that is larger than the bitwidth of the
1445 sought type is used. If none of the specifications are larger than
1446 the bitwidth then the largest integer type is used. For example,
1447 given the default specifications above, the i7 type will use the
1448 alignment of i8 (next largest) while both i65 and i256 will use the
1449 alignment of i64 (largest specified).
1450 #. If no match is found, and the type sought is a vector type, then the
1451 largest vector type that is smaller than the sought vector type will
1452 be used as a fall back. This happens because <128 x double> can be
1453 implemented in terms of 64 <2 x double>, for example.
1455 The function of the data layout string may not be what you expect.
1456 Notably, this is not a specification from the frontend of what alignment
1457 the code generator should use.
1459 Instead, if specified, the target data layout is required to match what
1460 the ultimate *code generator* expects. This string is used by the
1461 mid-level optimizers to improve code, and this only works if it matches
1462 what the ultimate code generator uses. If you would like to generate IR
1463 that does not embed this target-specific detail into the IR, then you
1464 don't have to specify the string. This will disable some optimizations
1465 that require precise layout information, but this also prevents those
1466 optimizations from introducing target specificity into the IR.
1473 A module may specify a target triple string that describes the target
1474 host. The syntax for the target triple is simply:
1476 .. code-block:: llvm
1478 target triple = "x86_64-apple-macosx10.7.0"
1480 The *target triple* string consists of a series of identifiers delimited
1481 by the minus sign character ('-'). The canonical forms are:
1485 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1486 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1488 This information is passed along to the backend so that it generates
1489 code for the proper architecture. It's possible to override this on the
1490 command line with the ``-mtriple`` command line option.
1492 .. _pointeraliasing:
1494 Pointer Aliasing Rules
1495 ----------------------
1497 Any memory access must be done through a pointer value associated with
1498 an address range of the memory access, otherwise the behavior is
1499 undefined. Pointer values are associated with address ranges according
1500 to the following rules:
1502 - A pointer value is associated with the addresses associated with any
1503 value it is *based* on.
1504 - An address of a global variable is associated with the address range
1505 of the variable's storage.
1506 - The result value of an allocation instruction is associated with the
1507 address range of the allocated storage.
1508 - A null pointer in the default address-space is associated with no
1510 - An integer constant other than zero or a pointer value returned from
1511 a function not defined within LLVM may be associated with address
1512 ranges allocated through mechanisms other than those provided by
1513 LLVM. Such ranges shall not overlap with any ranges of addresses
1514 allocated by mechanisms provided by LLVM.
1516 A pointer value is *based* on another pointer value according to the
1519 - A pointer value formed from a ``getelementptr`` operation is *based*
1520 on the first operand of the ``getelementptr``.
1521 - The result value of a ``bitcast`` is *based* on the operand of the
1523 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1524 values that contribute (directly or indirectly) to the computation of
1525 the pointer's value.
1526 - The "*based* on" relationship is transitive.
1528 Note that this definition of *"based"* is intentionally similar to the
1529 definition of *"based"* in C99, though it is slightly weaker.
1531 LLVM IR does not associate types with memory. The result type of a
1532 ``load`` merely indicates the size and alignment of the memory from
1533 which to load, as well as the interpretation of the value. The first
1534 operand type of a ``store`` similarly only indicates the size and
1535 alignment of the store.
1537 Consequently, type-based alias analysis, aka TBAA, aka
1538 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1539 :ref:`Metadata <metadata>` may be used to encode additional information
1540 which specialized optimization passes may use to implement type-based
1545 Volatile Memory Accesses
1546 ------------------------
1548 Certain memory accesses, such as :ref:`load <i_load>`'s,
1549 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1550 marked ``volatile``. The optimizers must not change the number of
1551 volatile operations or change their order of execution relative to other
1552 volatile operations. The optimizers *may* change the order of volatile
1553 operations relative to non-volatile operations. This is not Java's
1554 "volatile" and has no cross-thread synchronization behavior.
1556 IR-level volatile loads and stores cannot safely be optimized into
1557 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1558 flagged volatile. Likewise, the backend should never split or merge
1559 target-legal volatile load/store instructions.
1561 .. admonition:: Rationale
1563 Platforms may rely on volatile loads and stores of natively supported
1564 data width to be executed as single instruction. For example, in C
1565 this holds for an l-value of volatile primitive type with native
1566 hardware support, but not necessarily for aggregate types. The
1567 frontend upholds these expectations, which are intentionally
1568 unspecified in the IR. The rules above ensure that IR transformation
1569 do not violate the frontend's contract with the language.
1573 Memory Model for Concurrent Operations
1574 --------------------------------------
1576 The LLVM IR does not define any way to start parallel threads of
1577 execution or to register signal handlers. Nonetheless, there are
1578 platform-specific ways to create them, and we define LLVM IR's behavior
1579 in their presence. This model is inspired by the C++0x memory model.
1581 For a more informal introduction to this model, see the :doc:`Atomics`.
1583 We define a *happens-before* partial order as the least partial order
1586 - Is a superset of single-thread program order, and
1587 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1588 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1589 techniques, like pthread locks, thread creation, thread joining,
1590 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1591 Constraints <ordering>`).
1593 Note that program order does not introduce *happens-before* edges
1594 between a thread and signals executing inside that thread.
1596 Every (defined) read operation (load instructions, memcpy, atomic
1597 loads/read-modify-writes, etc.) R reads a series of bytes written by
1598 (defined) write operations (store instructions, atomic
1599 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1600 section, initialized globals are considered to have a write of the
1601 initializer which is atomic and happens before any other read or write
1602 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1603 may see any write to the same byte, except:
1605 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1606 write\ :sub:`2` happens before R\ :sub:`byte`, then
1607 R\ :sub:`byte` does not see write\ :sub:`1`.
1608 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1609 R\ :sub:`byte` does not see write\ :sub:`3`.
1611 Given that definition, R\ :sub:`byte` is defined as follows:
1613 - If R is volatile, the result is target-dependent. (Volatile is
1614 supposed to give guarantees which can support ``sig_atomic_t`` in
1615 C/C++, and may be used for accesses to addresses that do not behave
1616 like normal memory. It does not generally provide cross-thread
1618 - Otherwise, if there is no write to the same byte that happens before
1619 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1620 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1621 R\ :sub:`byte` returns the value written by that write.
1622 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1623 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1624 Memory Ordering Constraints <ordering>` section for additional
1625 constraints on how the choice is made.
1626 - Otherwise R\ :sub:`byte` returns ``undef``.
1628 R returns the value composed of the series of bytes it read. This
1629 implies that some bytes within the value may be ``undef`` **without**
1630 the entire value being ``undef``. Note that this only defines the
1631 semantics of the operation; it doesn't mean that targets will emit more
1632 than one instruction to read the series of bytes.
1634 Note that in cases where none of the atomic intrinsics are used, this
1635 model places only one restriction on IR transformations on top of what
1636 is required for single-threaded execution: introducing a store to a byte
1637 which might not otherwise be stored is not allowed in general.
1638 (Specifically, in the case where another thread might write to and read
1639 from an address, introducing a store can change a load that may see
1640 exactly one write into a load that may see multiple writes.)
1644 Atomic Memory Ordering Constraints
1645 ----------------------------------
1647 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1648 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1649 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1650 ordering parameters that determine which other atomic instructions on
1651 the same address they *synchronize with*. These semantics are borrowed
1652 from Java and C++0x, but are somewhat more colloquial. If these
1653 descriptions aren't precise enough, check those specs (see spec
1654 references in the :doc:`atomics guide <Atomics>`).
1655 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1656 differently since they don't take an address. See that instruction's
1657 documentation for details.
1659 For a simpler introduction to the ordering constraints, see the
1663 The set of values that can be read is governed by the happens-before
1664 partial order. A value cannot be read unless some operation wrote
1665 it. This is intended to provide a guarantee strong enough to model
1666 Java's non-volatile shared variables. This ordering cannot be
1667 specified for read-modify-write operations; it is not strong enough
1668 to make them atomic in any interesting way.
1670 In addition to the guarantees of ``unordered``, there is a single
1671 total order for modifications by ``monotonic`` operations on each
1672 address. All modification orders must be compatible with the
1673 happens-before order. There is no guarantee that the modification
1674 orders can be combined to a global total order for the whole program
1675 (and this often will not be possible). The read in an atomic
1676 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1677 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1678 order immediately before the value it writes. If one atomic read
1679 happens before another atomic read of the same address, the later
1680 read must see the same value or a later value in the address's
1681 modification order. This disallows reordering of ``monotonic`` (or
1682 stronger) operations on the same address. If an address is written
1683 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1684 read that address repeatedly, the other threads must eventually see
1685 the write. This corresponds to the C++0x/C1x
1686 ``memory_order_relaxed``.
1688 In addition to the guarantees of ``monotonic``, a
1689 *synchronizes-with* edge may be formed with a ``release`` operation.
1690 This is intended to model C++'s ``memory_order_acquire``.
1692 In addition to the guarantees of ``monotonic``, if this operation
1693 writes a value which is subsequently read by an ``acquire``
1694 operation, it *synchronizes-with* that operation. (This isn't a
1695 complete description; see the C++0x definition of a release
1696 sequence.) This corresponds to the C++0x/C1x
1697 ``memory_order_release``.
1698 ``acq_rel`` (acquire+release)
1699 Acts as both an ``acquire`` and ``release`` operation on its
1700 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1701 ``seq_cst`` (sequentially consistent)
1702 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1703 operation that only reads, ``release`` for an operation that only
1704 writes), there is a global total order on all
1705 sequentially-consistent operations on all addresses, which is
1706 consistent with the *happens-before* partial order and with the
1707 modification orders of all the affected addresses. Each
1708 sequentially-consistent read sees the last preceding write to the
1709 same address in this global order. This corresponds to the C++0x/C1x
1710 ``memory_order_seq_cst`` and Java volatile.
1714 If an atomic operation is marked ``singlethread``, it only *synchronizes
1715 with* or participates in modification and seq\_cst total orderings with
1716 other operations running in the same thread (for example, in signal
1724 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1725 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1726 :ref:`frem <i_frem>`) have the following flags that can set to enable
1727 otherwise unsafe floating point operations
1730 No NaNs - Allow optimizations to assume the arguments and result are not
1731 NaN. Such optimizations are required to retain defined behavior over
1732 NaNs, but the value of the result is undefined.
1735 No Infs - Allow optimizations to assume the arguments and result are not
1736 +/-Inf. Such optimizations are required to retain defined behavior over
1737 +/-Inf, but the value of the result is undefined.
1740 No Signed Zeros - Allow optimizations to treat the sign of a zero
1741 argument or result as insignificant.
1744 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1745 argument rather than perform division.
1748 Fast - Allow algebraically equivalent transformations that may
1749 dramatically change results in floating point (e.g. reassociate). This
1750 flag implies all the others.
1754 Use-list Order Directives
1755 -------------------------
1757 Use-list directives encode the in-memory order of each use-list, allowing the
1758 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1759 indexes that are assigned to the referenced value's uses. The referenced
1760 value's use-list is immediately sorted by these indexes.
1762 Use-list directives may appear at function scope or global scope. They are not
1763 instructions, and have no effect on the semantics of the IR. When they're at
1764 function scope, they must appear after the terminator of the final basic block.
1766 If basic blocks have their address taken via ``blockaddress()`` expressions,
1767 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1774 uselistorder <ty> <value>, { <order-indexes> }
1775 uselistorder_bb @function, %block { <order-indexes> }
1781 define void @foo(i32 %arg1, i32 %arg2) {
1783 ; ... instructions ...
1785 ; ... instructions ...
1787 ; At function scope.
1788 uselistorder i32 %arg1, { 1, 0, 2 }
1789 uselistorder label %bb, { 1, 0 }
1793 uselistorder i32* @global, { 1, 2, 0 }
1794 uselistorder i32 7, { 1, 0 }
1795 uselistorder i32 (i32) @bar, { 1, 0 }
1796 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1803 The LLVM type system is one of the most important features of the
1804 intermediate representation. Being typed enables a number of
1805 optimizations to be performed on the intermediate representation
1806 directly, without having to do extra analyses on the side before the
1807 transformation. A strong type system makes it easier to read the
1808 generated code and enables novel analyses and transformations that are
1809 not feasible to perform on normal three address code representations.
1819 The void type does not represent any value and has no size.
1837 The function type can be thought of as a function signature. It consists of a
1838 return type and a list of formal parameter types. The return type of a function
1839 type is a void type or first class type --- except for :ref:`label <t_label>`
1840 and :ref:`metadata <t_metadata>` types.
1846 <returntype> (<parameter list>)
1848 ...where '``<parameter list>``' is a comma-separated list of type
1849 specifiers. Optionally, the parameter list may include a type ``...``, which
1850 indicates that the function takes a variable number of arguments. Variable
1851 argument functions can access their arguments with the :ref:`variable argument
1852 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1853 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1857 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1858 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1859 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1860 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1861 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1862 | ``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. |
1863 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1864 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1865 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1872 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1873 Values of these types are the only ones which can be produced by
1881 These are the types that are valid in registers from CodeGen's perspective.
1890 The integer type is a very simple type that simply specifies an
1891 arbitrary bit width for the integer type desired. Any bit width from 1
1892 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1900 The number of bits the integer will occupy is specified by the ``N``
1906 +----------------+------------------------------------------------+
1907 | ``i1`` | a single-bit integer. |
1908 +----------------+------------------------------------------------+
1909 | ``i32`` | a 32-bit integer. |
1910 +----------------+------------------------------------------------+
1911 | ``i1942652`` | a really big integer of over 1 million bits. |
1912 +----------------+------------------------------------------------+
1916 Floating Point Types
1917 """"""""""""""""""""
1926 - 16-bit floating point value
1929 - 32-bit floating point value
1932 - 64-bit floating point value
1935 - 128-bit floating point value (112-bit mantissa)
1938 - 80-bit floating point value (X87)
1941 - 128-bit floating point value (two 64-bits)
1948 The x86_mmx type represents a value held in an MMX register on an x86
1949 machine. The operations allowed on it are quite limited: parameters and
1950 return values, load and store, and bitcast. User-specified MMX
1951 instructions are represented as intrinsic or asm calls with arguments
1952 and/or results of this type. There are no arrays, vectors or constants
1969 The pointer type is used to specify memory locations. Pointers are
1970 commonly used to reference objects in memory.
1972 Pointer types may have an optional address space attribute defining the
1973 numbered address space where the pointed-to object resides. The default
1974 address space is number zero. The semantics of non-zero address spaces
1975 are target-specific.
1977 Note that LLVM does not permit pointers to void (``void*``) nor does it
1978 permit pointers to labels (``label*``). Use ``i8*`` instead.
1988 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1989 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
1990 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1991 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
1992 +-------------------------+--------------------------------------------------------------------------------------------------------------+
1993 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
1994 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2003 A vector type is a simple derived type that represents a vector of
2004 elements. Vector types are used when multiple primitive data are
2005 operated in parallel using a single instruction (SIMD). A vector type
2006 requires a size (number of elements) and an underlying primitive data
2007 type. Vector types are considered :ref:`first class <t_firstclass>`.
2013 < <# elements> x <elementtype> >
2015 The number of elements is a constant integer value larger than 0;
2016 elementtype may be any integer, floating point or pointer type. Vectors
2017 of size zero are not allowed.
2021 +-------------------+--------------------------------------------------+
2022 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2023 +-------------------+--------------------------------------------------+
2024 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2025 +-------------------+--------------------------------------------------+
2026 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2027 +-------------------+--------------------------------------------------+
2028 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2029 +-------------------+--------------------------------------------------+
2038 The label type represents code labels.
2053 The metadata type represents embedded metadata. No derived types may be
2054 created from metadata except for :ref:`function <t_function>` arguments.
2067 Aggregate Types are a subset of derived types that can contain multiple
2068 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2069 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2079 The array type is a very simple derived type that arranges elements
2080 sequentially in memory. The array type requires a size (number of
2081 elements) and an underlying data type.
2087 [<# elements> x <elementtype>]
2089 The number of elements is a constant integer value; ``elementtype`` may
2090 be any type with a size.
2094 +------------------+--------------------------------------+
2095 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2096 +------------------+--------------------------------------+
2097 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2098 +------------------+--------------------------------------+
2099 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2100 +------------------+--------------------------------------+
2102 Here are some examples of multidimensional arrays:
2104 +-----------------------------+----------------------------------------------------------+
2105 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2106 +-----------------------------+----------------------------------------------------------+
2107 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2108 +-----------------------------+----------------------------------------------------------+
2109 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2110 +-----------------------------+----------------------------------------------------------+
2112 There is no restriction on indexing beyond the end of the array implied
2113 by a static type (though there are restrictions on indexing beyond the
2114 bounds of an allocated object in some cases). This means that
2115 single-dimension 'variable sized array' addressing can be implemented in
2116 LLVM with a zero length array type. An implementation of 'pascal style
2117 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2127 The structure type is used to represent a collection of data members
2128 together in memory. The elements of a structure may be any type that has
2131 Structures in memory are accessed using '``load``' and '``store``' by
2132 getting a pointer to a field with the '``getelementptr``' instruction.
2133 Structures in registers are accessed using the '``extractvalue``' and
2134 '``insertvalue``' instructions.
2136 Structures may optionally be "packed" structures, which indicate that
2137 the alignment of the struct is one byte, and that there is no padding
2138 between the elements. In non-packed structs, padding between field types
2139 is inserted as defined by the DataLayout string in the module, which is
2140 required to match what the underlying code generator expects.
2142 Structures can either be "literal" or "identified". A literal structure
2143 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2144 identified types are always defined at the top level with a name.
2145 Literal types are uniqued by their contents and can never be recursive
2146 or opaque since there is no way to write one. Identified types can be
2147 recursive, can be opaqued, and are never uniqued.
2153 %T1 = type { <type list> } ; Identified normal struct type
2154 %T2 = type <{ <type list> }> ; Identified packed struct type
2158 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2159 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2160 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2161 | ``{ 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``. |
2162 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2163 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2164 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2168 Opaque Structure Types
2169 """"""""""""""""""""""
2173 Opaque structure types are used to represent named structure types that
2174 do not have a body specified. This corresponds (for example) to the C
2175 notion of a forward declared structure.
2186 +--------------+-------------------+
2187 | ``opaque`` | An opaque type. |
2188 +--------------+-------------------+
2195 LLVM has several different basic types of constants. This section
2196 describes them all and their syntax.
2201 **Boolean constants**
2202 The two strings '``true``' and '``false``' are both valid constants
2204 **Integer constants**
2205 Standard integers (such as '4') are constants of the
2206 :ref:`integer <t_integer>` type. Negative numbers may be used with
2208 **Floating point constants**
2209 Floating point constants use standard decimal notation (e.g.
2210 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2211 hexadecimal notation (see below). The assembler requires the exact
2212 decimal value of a floating-point constant. For example, the
2213 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2214 decimal in binary. Floating point constants must have a :ref:`floating
2215 point <t_floating>` type.
2216 **Null pointer constants**
2217 The identifier '``null``' is recognized as a null pointer constant
2218 and must be of :ref:`pointer type <t_pointer>`.
2220 The one non-intuitive notation for constants is the hexadecimal form of
2221 floating point constants. For example, the form
2222 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2223 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2224 constants are required (and the only time that they are generated by the
2225 disassembler) is when a floating point constant must be emitted but it
2226 cannot be represented as a decimal floating point number in a reasonable
2227 number of digits. For example, NaN's, infinities, and other special
2228 values are represented in their IEEE hexadecimal format so that assembly
2229 and disassembly do not cause any bits to change in the constants.
2231 When using the hexadecimal form, constants of types half, float, and
2232 double are represented using the 16-digit form shown above (which
2233 matches the IEEE754 representation for double); half and float values
2234 must, however, be exactly representable as IEEE 754 half and single
2235 precision, respectively. Hexadecimal format is always used for long
2236 double, and there are three forms of long double. The 80-bit format used
2237 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2238 128-bit format used by PowerPC (two adjacent doubles) is represented by
2239 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2240 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2241 will only work if they match the long double format on your target.
2242 The IEEE 16-bit format (half precision) is represented by ``0xH``
2243 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2244 (sign bit at the left).
2246 There are no constants of type x86_mmx.
2248 .. _complexconstants:
2253 Complex constants are a (potentially recursive) combination of simple
2254 constants and smaller complex constants.
2256 **Structure constants**
2257 Structure constants are represented with notation similar to
2258 structure type definitions (a comma separated list of elements,
2259 surrounded by braces (``{}``)). For example:
2260 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2261 "``@G = external global i32``". Structure constants must have
2262 :ref:`structure type <t_struct>`, and the number and types of elements
2263 must match those specified by the type.
2265 Array constants are represented with notation similar to array type
2266 definitions (a comma separated list of elements, surrounded by
2267 square brackets (``[]``)). For example:
2268 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2269 :ref:`array type <t_array>`, and the number and types of elements must
2270 match those specified by the type.
2271 **Vector constants**
2272 Vector constants are represented with notation similar to vector
2273 type definitions (a comma separated list of elements, surrounded by
2274 less-than/greater-than's (``<>``)). For example:
2275 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2276 must have :ref:`vector type <t_vector>`, and the number and types of
2277 elements must match those specified by the type.
2278 **Zero initialization**
2279 The string '``zeroinitializer``' can be used to zero initialize a
2280 value to zero of *any* type, including scalar and
2281 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2282 having to print large zero initializers (e.g. for large arrays) and
2283 is always exactly equivalent to using explicit zero initializers.
2285 A metadata node is a structure-like constant with :ref:`metadata
2286 type <t_metadata>`. For example:
2287 "``metadata !{ i32 0, metadata !"test" }``". Unlike other
2288 constants that are meant to be interpreted as part of the
2289 instruction stream, metadata is a place to attach additional
2290 information such as debug info.
2292 Global Variable and Function Addresses
2293 --------------------------------------
2295 The addresses of :ref:`global variables <globalvars>` and
2296 :ref:`functions <functionstructure>` are always implicitly valid
2297 (link-time) constants. These constants are explicitly referenced when
2298 the :ref:`identifier for the global <identifiers>` is used and always have
2299 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2302 .. code-block:: llvm
2306 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2313 The string '``undef``' can be used anywhere a constant is expected, and
2314 indicates that the user of the value may receive an unspecified
2315 bit-pattern. Undefined values may be of any type (other than '``label``'
2316 or '``void``') and be used anywhere a constant is permitted.
2318 Undefined values are useful because they indicate to the compiler that
2319 the program is well defined no matter what value is used. This gives the
2320 compiler more freedom to optimize. Here are some examples of
2321 (potentially surprising) transformations that are valid (in pseudo IR):
2323 .. code-block:: llvm
2333 This is safe because all of the output bits are affected by the undef
2334 bits. Any output bit can have a zero or one depending on the input bits.
2336 .. code-block:: llvm
2347 These logical operations have bits that are not always affected by the
2348 input. For example, if ``%X`` has a zero bit, then the output of the
2349 '``and``' operation will always be a zero for that bit, no matter what
2350 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2351 optimize or assume that the result of the '``and``' is '``undef``'.
2352 However, it is safe to assume that all bits of the '``undef``' could be
2353 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2354 all the bits of the '``undef``' operand to the '``or``' could be set,
2355 allowing the '``or``' to be folded to -1.
2357 .. code-block:: llvm
2359 %A = select undef, %X, %Y
2360 %B = select undef, 42, %Y
2361 %C = select %X, %Y, undef
2371 This set of examples shows that undefined '``select``' (and conditional
2372 branch) conditions can go *either way*, but they have to come from one
2373 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2374 both known to have a clear low bit, then ``%A`` would have to have a
2375 cleared low bit. However, in the ``%C`` example, the optimizer is
2376 allowed to assume that the '``undef``' operand could be the same as
2377 ``%Y``, allowing the whole '``select``' to be eliminated.
2379 .. code-block:: llvm
2381 %A = xor undef, undef
2398 This example points out that two '``undef``' operands are not
2399 necessarily the same. This can be surprising to people (and also matches
2400 C semantics) where they assume that "``X^X``" is always zero, even if
2401 ``X`` is undefined. This isn't true for a number of reasons, but the
2402 short answer is that an '``undef``' "variable" can arbitrarily change
2403 its value over its "live range". This is true because the variable
2404 doesn't actually *have a live range*. Instead, the value is logically
2405 read from arbitrary registers that happen to be around when needed, so
2406 the value is not necessarily consistent over time. In fact, ``%A`` and
2407 ``%C`` need to have the same semantics or the core LLVM "replace all
2408 uses with" concept would not hold.
2410 .. code-block:: llvm
2418 These examples show the crucial difference between an *undefined value*
2419 and *undefined behavior*. An undefined value (like '``undef``') is
2420 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2421 operation can be constant folded to '``undef``', because the '``undef``'
2422 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2423 However, in the second example, we can make a more aggressive
2424 assumption: because the ``undef`` is allowed to be an arbitrary value,
2425 we are allowed to assume that it could be zero. Since a divide by zero
2426 has *undefined behavior*, we are allowed to assume that the operation
2427 does not execute at all. This allows us to delete the divide and all
2428 code after it. Because the undefined operation "can't happen", the
2429 optimizer can assume that it occurs in dead code.
2431 .. code-block:: llvm
2433 a: store undef -> %X
2434 b: store %X -> undef
2439 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2440 value can be assumed to not have any effect; we can assume that the
2441 value is overwritten with bits that happen to match what was already
2442 there. However, a store *to* an undefined location could clobber
2443 arbitrary memory, therefore, it has undefined behavior.
2450 Poison values are similar to :ref:`undef values <undefvalues>`, however
2451 they also represent the fact that an instruction or constant expression
2452 that cannot evoke side effects has nevertheless detected a condition
2453 that results in undefined behavior.
2455 There is currently no way of representing a poison value in the IR; they
2456 only exist when produced by operations such as :ref:`add <i_add>` with
2459 Poison value behavior is defined in terms of value *dependence*:
2461 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2462 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2463 their dynamic predecessor basic block.
2464 - Function arguments depend on the corresponding actual argument values
2465 in the dynamic callers of their functions.
2466 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2467 instructions that dynamically transfer control back to them.
2468 - :ref:`Invoke <i_invoke>` instructions depend on the
2469 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2470 call instructions that dynamically transfer control back to them.
2471 - Non-volatile loads and stores depend on the most recent stores to all
2472 of the referenced memory addresses, following the order in the IR
2473 (including loads and stores implied by intrinsics such as
2474 :ref:`@llvm.memcpy <int_memcpy>`.)
2475 - An instruction with externally visible side effects depends on the
2476 most recent preceding instruction with externally visible side
2477 effects, following the order in the IR. (This includes :ref:`volatile
2478 operations <volatile>`.)
2479 - An instruction *control-depends* on a :ref:`terminator
2480 instruction <terminators>` if the terminator instruction has
2481 multiple successors and the instruction is always executed when
2482 control transfers to one of the successors, and may not be executed
2483 when control is transferred to another.
2484 - Additionally, an instruction also *control-depends* on a terminator
2485 instruction if the set of instructions it otherwise depends on would
2486 be different if the terminator had transferred control to a different
2488 - Dependence is transitive.
2490 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2491 with the additional effect that any instruction that has a *dependence*
2492 on a poison value has undefined behavior.
2494 Here are some examples:
2496 .. code-block:: llvm
2499 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2500 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2501 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2502 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2504 store i32 %poison, i32* @g ; Poison value stored to memory.
2505 %poison2 = load i32* @g ; Poison value loaded back from memory.
2507 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2509 %narrowaddr = bitcast i32* @g to i16*
2510 %wideaddr = bitcast i32* @g to i64*
2511 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2512 %poison4 = load i64* %wideaddr ; Returns a poison value.
2514 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2515 br i1 %cmp, label %true, label %end ; Branch to either destination.
2518 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2519 ; it has undefined behavior.
2523 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2524 ; Both edges into this PHI are
2525 ; control-dependent on %cmp, so this
2526 ; always results in a poison value.
2528 store volatile i32 0, i32* @g ; This would depend on the store in %true
2529 ; if %cmp is true, or the store in %entry
2530 ; otherwise, so this is undefined behavior.
2532 br i1 %cmp, label %second_true, label %second_end
2533 ; The same branch again, but this time the
2534 ; true block doesn't have side effects.
2541 store volatile i32 0, i32* @g ; This time, the instruction always depends
2542 ; on the store in %end. Also, it is
2543 ; control-equivalent to %end, so this is
2544 ; well-defined (ignoring earlier undefined
2545 ; behavior in this example).
2549 Addresses of Basic Blocks
2550 -------------------------
2552 ``blockaddress(@function, %block)``
2554 The '``blockaddress``' constant computes the address of the specified
2555 basic block in the specified function, and always has an ``i8*`` type.
2556 Taking the address of the entry block is illegal.
2558 This value only has defined behavior when used as an operand to the
2559 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2560 against null. Pointer equality tests between labels addresses results in
2561 undefined behavior --- though, again, comparison against null is ok, and
2562 no label is equal to the null pointer. This may be passed around as an
2563 opaque pointer sized value as long as the bits are not inspected. This
2564 allows ``ptrtoint`` and arithmetic to be performed on these values so
2565 long as the original value is reconstituted before the ``indirectbr``
2568 Finally, some targets may provide defined semantics when using the value
2569 as the operand to an inline assembly, but that is target specific.
2573 Constant Expressions
2574 --------------------
2576 Constant expressions are used to allow expressions involving other
2577 constants to be used as constants. Constant expressions may be of any
2578 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2579 that does not have side effects (e.g. load and call are not supported).
2580 The following is the syntax for constant expressions:
2582 ``trunc (CST to TYPE)``
2583 Truncate a constant to another type. The bit size of CST must be
2584 larger than the bit size of TYPE. Both types must be integers.
2585 ``zext (CST to TYPE)``
2586 Zero extend a constant to another type. The bit size of CST must be
2587 smaller than the bit size of TYPE. Both types must be integers.
2588 ``sext (CST to TYPE)``
2589 Sign extend a constant to another type. The bit size of CST must be
2590 smaller than the bit size of TYPE. Both types must be integers.
2591 ``fptrunc (CST to TYPE)``
2592 Truncate a floating point constant to another floating point type.
2593 The size of CST must be larger than the size of TYPE. Both types
2594 must be floating point.
2595 ``fpext (CST to TYPE)``
2596 Floating point extend a constant to another type. The size of CST
2597 must be smaller or equal to the size of TYPE. Both types must be
2599 ``fptoui (CST to TYPE)``
2600 Convert a floating point constant to the corresponding unsigned
2601 integer constant. TYPE must be a scalar or vector integer type. CST
2602 must be of scalar or vector floating point type. Both CST and TYPE
2603 must be scalars, or vectors of the same number of elements. If the
2604 value won't fit in the integer type, the results are undefined.
2605 ``fptosi (CST to TYPE)``
2606 Convert a floating point constant to the corresponding signed
2607 integer constant. TYPE must be a scalar or vector integer type. CST
2608 must be of scalar or vector floating point type. Both CST and TYPE
2609 must be scalars, or vectors of the same number of elements. If the
2610 value won't fit in the integer type, the results are undefined.
2611 ``uitofp (CST to TYPE)``
2612 Convert an unsigned integer constant to the corresponding floating
2613 point constant. TYPE must be a scalar or vector floating point type.
2614 CST must be of scalar or vector integer type. Both CST and TYPE must
2615 be scalars, or vectors of the same number of elements. If the value
2616 won't fit in the floating point type, the results are undefined.
2617 ``sitofp (CST to TYPE)``
2618 Convert a signed integer constant to the corresponding floating
2619 point constant. TYPE must be a scalar or vector floating point type.
2620 CST must be of scalar or vector integer type. Both CST and TYPE must
2621 be scalars, or vectors of the same number of elements. If the value
2622 won't fit in the floating point type, the results are undefined.
2623 ``ptrtoint (CST to TYPE)``
2624 Convert a pointer typed constant to the corresponding integer
2625 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2626 pointer type. The ``CST`` value is zero extended, truncated, or
2627 unchanged to make it fit in ``TYPE``.
2628 ``inttoptr (CST to TYPE)``
2629 Convert an integer constant to a pointer constant. TYPE must be a
2630 pointer type. CST must be of integer type. The CST value is zero
2631 extended, truncated, or unchanged to make it fit in a pointer size.
2632 This one is *really* dangerous!
2633 ``bitcast (CST to TYPE)``
2634 Convert a constant, CST, to another TYPE. The constraints of the
2635 operands are the same as those for the :ref:`bitcast
2636 instruction <i_bitcast>`.
2637 ``addrspacecast (CST to TYPE)``
2638 Convert a constant pointer or constant vector of pointer, CST, to another
2639 TYPE in a different address space. The constraints of the operands are the
2640 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2641 ``getelementptr (CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)``
2642 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2643 constants. As with the :ref:`getelementptr <i_getelementptr>`
2644 instruction, the index list may have zero or more indexes, which are
2645 required to make sense for the type of "CSTPTR".
2646 ``select (COND, VAL1, VAL2)``
2647 Perform the :ref:`select operation <i_select>` on constants.
2648 ``icmp COND (VAL1, VAL2)``
2649 Performs the :ref:`icmp operation <i_icmp>` on constants.
2650 ``fcmp COND (VAL1, VAL2)``
2651 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2652 ``extractelement (VAL, IDX)``
2653 Perform the :ref:`extractelement operation <i_extractelement>` on
2655 ``insertelement (VAL, ELT, IDX)``
2656 Perform the :ref:`insertelement operation <i_insertelement>` on
2658 ``shufflevector (VEC1, VEC2, IDXMASK)``
2659 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2661 ``extractvalue (VAL, IDX0, IDX1, ...)``
2662 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2663 constants. The index list is interpreted in a similar manner as
2664 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2665 least one index value must be specified.
2666 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2667 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2668 The index list is interpreted in a similar manner as indices in a
2669 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2670 value must be specified.
2671 ``OPCODE (LHS, RHS)``
2672 Perform the specified operation of the LHS and RHS constants. OPCODE
2673 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2674 binary <bitwiseops>` operations. The constraints on operands are
2675 the same as those for the corresponding instruction (e.g. no bitwise
2676 operations on floating point values are allowed).
2683 Inline Assembler Expressions
2684 ----------------------------
2686 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2687 Inline Assembly <moduleasm>`) through the use of a special value. This
2688 value represents the inline assembler as a string (containing the
2689 instructions to emit), a list of operand constraints (stored as a
2690 string), a flag that indicates whether or not the inline asm expression
2691 has side effects, and a flag indicating whether the function containing
2692 the asm needs to align its stack conservatively. An example inline
2693 assembler expression is:
2695 .. code-block:: llvm
2697 i32 (i32) asm "bswap $0", "=r,r"
2699 Inline assembler expressions may **only** be used as the callee operand
2700 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2701 Thus, typically we have:
2703 .. code-block:: llvm
2705 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2707 Inline asms with side effects not visible in the constraint list must be
2708 marked as having side effects. This is done through the use of the
2709 '``sideeffect``' keyword, like so:
2711 .. code-block:: llvm
2713 call void asm sideeffect "eieio", ""()
2715 In some cases inline asms will contain code that will not work unless
2716 the stack is aligned in some way, such as calls or SSE instructions on
2717 x86, yet will not contain code that does that alignment within the asm.
2718 The compiler should make conservative assumptions about what the asm
2719 might contain and should generate its usual stack alignment code in the
2720 prologue if the '``alignstack``' keyword is present:
2722 .. code-block:: llvm
2724 call void asm alignstack "eieio", ""()
2726 Inline asms also support using non-standard assembly dialects. The
2727 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2728 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2729 the only supported dialects. An example is:
2731 .. code-block:: llvm
2733 call void asm inteldialect "eieio", ""()
2735 If multiple keywords appear the '``sideeffect``' keyword must come
2736 first, the '``alignstack``' keyword second and the '``inteldialect``'
2742 The call instructions that wrap inline asm nodes may have a
2743 "``!srcloc``" MDNode attached to it that contains a list of constant
2744 integers. If present, the code generator will use the integer as the
2745 location cookie value when report errors through the ``LLVMContext``
2746 error reporting mechanisms. This allows a front-end to correlate backend
2747 errors that occur with inline asm back to the source code that produced
2750 .. code-block:: llvm
2752 call void asm sideeffect "something bad", ""(), !srcloc !42
2754 !42 = !{ i32 1234567 }
2756 It is up to the front-end to make sense of the magic numbers it places
2757 in the IR. If the MDNode contains multiple constants, the code generator
2758 will use the one that corresponds to the line of the asm that the error
2763 Metadata Nodes and Metadata Strings
2764 -----------------------------------
2766 LLVM IR allows metadata to be attached to instructions in the program
2767 that can convey extra information about the code to the optimizers and
2768 code generator. One example application of metadata is source-level
2769 debug information. There are two metadata primitives: strings and nodes.
2770 All metadata has the ``metadata`` type and is identified in syntax by a
2771 preceding exclamation point ('``!``').
2773 A metadata string is a string surrounded by double quotes. It can
2774 contain any character by escaping non-printable characters with
2775 "``\xx``" where "``xx``" is the two digit hex code. For example:
2778 Metadata nodes are represented with notation similar to structure
2779 constants (a comma separated list of elements, surrounded by braces and
2780 preceded by an exclamation point). Metadata nodes can have any values as
2781 their operand. For example:
2783 .. code-block:: llvm
2785 !{ metadata !"test\00", i32 10}
2787 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2788 metadata nodes, which can be looked up in the module symbol table. For
2791 .. code-block:: llvm
2793 !foo = metadata !{!4, !3}
2795 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2796 function is using two metadata arguments:
2798 .. code-block:: llvm
2800 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2802 Metadata can be attached with an instruction. Here metadata ``!21`` is
2803 attached to the ``add`` instruction using the ``!dbg`` identifier:
2805 .. code-block:: llvm
2807 %indvar.next = add i64 %indvar, 1, !dbg !21
2809 More information about specific metadata nodes recognized by the
2810 optimizers and code generator is found below.
2815 In LLVM IR, memory does not have types, so LLVM's own type system is not
2816 suitable for doing TBAA. Instead, metadata is added to the IR to
2817 describe a type system of a higher level language. This can be used to
2818 implement typical C/C++ TBAA, but it can also be used to implement
2819 custom alias analysis behavior for other languages.
2821 The current metadata format is very simple. TBAA metadata nodes have up
2822 to three fields, e.g.:
2824 .. code-block:: llvm
2826 !0 = metadata !{ metadata !"an example type tree" }
2827 !1 = metadata !{ metadata !"int", metadata !0 }
2828 !2 = metadata !{ metadata !"float", metadata !0 }
2829 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2831 The first field is an identity field. It can be any value, usually a
2832 metadata string, which uniquely identifies the type. The most important
2833 name in the tree is the name of the root node. Two trees with different
2834 root node names are entirely disjoint, even if they have leaves with
2837 The second field identifies the type's parent node in the tree, or is
2838 null or omitted for a root node. A type is considered to alias all of
2839 its descendants and all of its ancestors in the tree. Also, a type is
2840 considered to alias all types in other trees, so that bitcode produced
2841 from multiple front-ends is handled conservatively.
2843 If the third field is present, it's an integer which if equal to 1
2844 indicates that the type is "constant" (meaning
2845 ``pointsToConstantMemory`` should return true; see `other useful
2846 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
2848 '``tbaa.struct``' Metadata
2849 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2851 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
2852 aggregate assignment operations in C and similar languages, however it
2853 is defined to copy a contiguous region of memory, which is more than
2854 strictly necessary for aggregate types which contain holes due to
2855 padding. Also, it doesn't contain any TBAA information about the fields
2858 ``!tbaa.struct`` metadata can describe which memory subregions in a
2859 memcpy are padding and what the TBAA tags of the struct are.
2861 The current metadata format is very simple. ``!tbaa.struct`` metadata
2862 nodes are a list of operands which are in conceptual groups of three.
2863 For each group of three, the first operand gives the byte offset of a
2864 field in bytes, the second gives its size in bytes, and the third gives
2867 .. code-block:: llvm
2869 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
2871 This describes a struct with two fields. The first is at offset 0 bytes
2872 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
2873 and has size 4 bytes and has tbaa tag !2.
2875 Note that the fields need not be contiguous. In this example, there is a
2876 4 byte gap between the two fields. This gap represents padding which
2877 does not carry useful data and need not be preserved.
2879 '``noalias``' and '``alias.scope``' Metadata
2880 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2882 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
2883 noalias memory-access sets. This means that some collection of memory access
2884 instructions (loads, stores, memory-accessing calls, etc.) that carry
2885 ``noalias`` metadata can specifically be specified not to alias with some other
2886 collection of memory access instructions that carry ``alias.scope`` metadata.
2887 Each type of metadata specifies a list of scopes where each scope has an id and
2888 a domain. When evaluating an aliasing query, if for some some domain, the set
2889 of scopes with that domain in one instruction's ``alias.scope`` list is a
2890 subset of (or qual to) the set of scopes for that domain in another
2891 instruction's ``noalias`` list, then the two memory accesses are assumed not to
2894 The metadata identifying each domain is itself a list containing one or two
2895 entries. The first entry is the name of the domain. Note that if the name is a
2896 string then it can be combined accross functions and translation units. A
2897 self-reference can be used to create globally unique domain names. A
2898 descriptive string may optionally be provided as a second list entry.
2900 The metadata identifying each scope is also itself a list containing two or
2901 three entries. The first entry is the name of the scope. Note that if the name
2902 is a string then it can be combined accross functions and translation units. A
2903 self-reference can be used to create globally unique scope names. A metadata
2904 reference to the scope's domain is the second entry. A descriptive string may
2905 optionally be provided as a third list entry.
2909 .. code-block:: llvm
2911 ; Two scope domains:
2912 !0 = metadata !{metadata !0}
2913 !1 = metadata !{metadata !1}
2915 ; Some scopes in these domains:
2916 !2 = metadata !{metadata !2, metadata !0}
2917 !3 = metadata !{metadata !3, metadata !0}
2918 !4 = metadata !{metadata !4, metadata !1}
2921 !5 = metadata !{metadata !4} ; A list containing only scope !4
2922 !6 = metadata !{metadata !4, metadata !3, metadata !2}
2923 !7 = metadata !{metadata !3}
2925 ; These two instructions don't alias:
2926 %0 = load float* %c, align 4, !alias.scope !5
2927 store float %0, float* %arrayidx.i, align 4, !noalias !5
2929 ; These two instructions also don't alias (for domain !1, the set of scopes
2930 ; in the !alias.scope equals that in the !noalias list):
2931 %2 = load float* %c, align 4, !alias.scope !5
2932 store float %2, float* %arrayidx.i2, align 4, !noalias !6
2934 ; These two instructions don't alias (for domain !0, the set of scopes in
2935 ; the !noalias list is not a superset of, or equal to, the scopes in the
2936 ; !alias.scope list):
2937 %2 = load float* %c, align 4, !alias.scope !6
2938 store float %0, float* %arrayidx.i, align 4, !noalias !7
2940 '``fpmath``' Metadata
2941 ^^^^^^^^^^^^^^^^^^^^^
2943 ``fpmath`` metadata may be attached to any instruction of floating point
2944 type. It can be used to express the maximum acceptable error in the
2945 result of that instruction, in ULPs, thus potentially allowing the
2946 compiler to use a more efficient but less accurate method of computing
2947 it. ULP is defined as follows:
2949 If ``x`` is a real number that lies between two finite consecutive
2950 floating-point numbers ``a`` and ``b``, without being equal to one
2951 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
2952 distance between the two non-equal finite floating-point numbers
2953 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
2955 The metadata node shall consist of a single positive floating point
2956 number representing the maximum relative error, for example:
2958 .. code-block:: llvm
2960 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
2962 '``range``' Metadata
2963 ^^^^^^^^^^^^^^^^^^^^
2965 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
2966 integer types. It expresses the possible ranges the loaded value or the value
2967 returned by the called function at this call site is in. The ranges are
2968 represented with a flattened list of integers. The loaded value or the value
2969 returned is known to be in the union of the ranges defined by each consecutive
2970 pair. Each pair has the following properties:
2972 - The type must match the type loaded by the instruction.
2973 - The pair ``a,b`` represents the range ``[a,b)``.
2974 - Both ``a`` and ``b`` are constants.
2975 - The range is allowed to wrap.
2976 - The range should not represent the full or empty set. That is,
2979 In addition, the pairs must be in signed order of the lower bound and
2980 they must be non-contiguous.
2984 .. code-block:: llvm
2986 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
2987 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
2988 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
2989 %d = invoke i8 @bar() to label %cont
2990 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
2992 !0 = metadata !{ i8 0, i8 2 }
2993 !1 = metadata !{ i8 255, i8 2 }
2994 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
2995 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
3000 It is sometimes useful to attach information to loop constructs. Currently,
3001 loop metadata is implemented as metadata attached to the branch instruction
3002 in the loop latch block. This type of metadata refer to a metadata node that is
3003 guaranteed to be separate for each loop. The loop identifier metadata is
3004 specified with the name ``llvm.loop``.
3006 The loop identifier metadata is implemented using a metadata that refers to
3007 itself to avoid merging it with any other identifier metadata, e.g.,
3008 during module linkage or function inlining. That is, each loop should refer
3009 to their own identification metadata even if they reside in separate functions.
3010 The following example contains loop identifier metadata for two separate loop
3013 .. code-block:: llvm
3015 !0 = metadata !{ metadata !0 }
3016 !1 = metadata !{ metadata !1 }
3018 The loop identifier metadata can be used to specify additional
3019 per-loop metadata. Any operands after the first operand can be treated
3020 as user-defined metadata. For example the ``llvm.loop.unroll.count``
3021 suggests an unroll factor to the loop unroller:
3023 .. code-block:: llvm
3025 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
3027 !0 = metadata !{ metadata !0, metadata !1 }
3028 !1 = metadata !{ metadata !"llvm.loop.unroll.count", i32 4 }
3030 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
3031 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3033 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
3034 used to control per-loop vectorization and interleaving parameters such as
3035 vectorization width and interleave count. These metadata should be used in
3036 conjunction with ``llvm.loop`` loop identification metadata. The
3037 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
3038 optimization hints and the optimizer will only interleave and vectorize loops if
3039 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
3040 which contains information about loop-carried memory dependencies can be helpful
3041 in determining the safety of these transformations.
3043 '``llvm.loop.interleave.count``' Metadata
3044 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3046 This metadata suggests an interleave count to the loop interleaver.
3047 The first operand is the string ``llvm.loop.interleave.count`` and the
3048 second operand is an integer specifying the interleave count. For
3051 .. code-block:: llvm
3053 !0 = metadata !{ metadata !"llvm.loop.interleave.count", i32 4 }
3055 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
3056 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
3057 then the interleave count will be determined automatically.
3059 '``llvm.loop.vectorize.enable``' Metadata
3060 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3062 This metadata selectively enables or disables vectorization for the loop. The
3063 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
3064 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
3065 0 disables vectorization:
3067 .. code-block:: llvm
3069 !0 = metadata !{ metadata !"llvm.loop.vectorize.enable", i1 0 }
3070 !1 = metadata !{ metadata !"llvm.loop.vectorize.enable", i1 1 }
3072 '``llvm.loop.vectorize.width``' Metadata
3073 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3075 This metadata sets the target width of the vectorizer. The first
3076 operand is the string ``llvm.loop.vectorize.width`` and the second
3077 operand is an integer specifying the width. For example:
3079 .. code-block:: llvm
3081 !0 = metadata !{ metadata !"llvm.loop.vectorize.width", i32 4 }
3083 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
3084 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
3085 0 or if the loop does not have this metadata the width will be
3086 determined automatically.
3088 '``llvm.loop.unroll``'
3089 ^^^^^^^^^^^^^^^^^^^^^^
3091 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
3092 optimization hints such as the unroll factor. ``llvm.loop.unroll``
3093 metadata should be used in conjunction with ``llvm.loop`` loop
3094 identification metadata. The ``llvm.loop.unroll`` metadata are only
3095 optimization hints and the unrolling will only be performed if the
3096 optimizer believes it is safe to do so.
3098 '``llvm.loop.unroll.count``' Metadata
3099 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3101 This metadata suggests an unroll factor to the loop unroller. The
3102 first operand is the string ``llvm.loop.unroll.count`` and the second
3103 operand is a positive integer specifying the unroll factor. For
3106 .. code-block:: llvm
3108 !0 = metadata !{ metadata !"llvm.loop.unroll.count", i32 4 }
3110 If the trip count of the loop is less than the unroll count the loop
3111 will be partially unrolled.
3113 '``llvm.loop.unroll.disable``' Metadata
3114 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3116 This metadata either disables loop unrolling. The metadata has a single operand
3117 which is the string ``llvm.loop.unroll.disable``. For example:
3119 .. code-block:: llvm
3121 !0 = metadata !{ metadata !"llvm.loop.unroll.disable" }
3123 '``llvm.loop.unroll.full``' Metadata
3124 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3126 This metadata either suggests that the loop should be unrolled fully. The
3127 metadata has a single operand which is the string ``llvm.loop.unroll.disable``.
3130 .. code-block:: llvm
3132 !0 = metadata !{ metadata !"llvm.loop.unroll.full" }
3137 Metadata types used to annotate memory accesses with information helpful
3138 for optimizations are prefixed with ``llvm.mem``.
3140 '``llvm.mem.parallel_loop_access``' Metadata
3141 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3143 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
3144 or metadata containing a list of loop identifiers for nested loops.
3145 The metadata is attached to memory accessing instructions and denotes that
3146 no loop carried memory dependence exist between it and other instructions denoted
3147 with the same loop identifier.
3149 Precisely, given two instructions ``m1`` and ``m2`` that both have the
3150 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
3151 set of loops associated with that metadata, respectively, then there is no loop
3152 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
3155 As a special case, if all memory accessing instructions in a loop have
3156 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
3157 loop has no loop carried memory dependences and is considered to be a parallel
3160 Note that if not all memory access instructions have such metadata referring to
3161 the loop, then the loop is considered not being trivially parallel. Additional
3162 memory dependence analysis is required to make that determination. As a fail
3163 safe mechanism, this causes loops that were originally parallel to be considered
3164 sequential (if optimization passes that are unaware of the parallel semantics
3165 insert new memory instructions into the loop body).
3167 Example of a loop that is considered parallel due to its correct use of
3168 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
3169 metadata types that refer to the same loop identifier metadata.
3171 .. code-block:: llvm
3175 %val0 = load i32* %arrayidx, !llvm.mem.parallel_loop_access !0
3177 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3179 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
3183 !0 = metadata !{ metadata !0 }
3185 It is also possible to have nested parallel loops. In that case the
3186 memory accesses refer to a list of loop identifier metadata nodes instead of
3187 the loop identifier metadata node directly:
3189 .. code-block:: llvm
3193 %val1 = load i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
3195 br label %inner.for.body
3199 %val0 = load i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3201 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
3203 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
3207 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
3209 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
3211 outer.for.end: ; preds = %for.body
3213 !0 = metadata !{ metadata !1, metadata !2 } ; a list of loop identifiers
3214 !1 = metadata !{ metadata !1 } ; an identifier for the inner loop
3215 !2 = metadata !{ metadata !2 } ; an identifier for the outer loop
3217 Module Flags Metadata
3218 =====================
3220 Information about the module as a whole is difficult to convey to LLVM's
3221 subsystems. The LLVM IR isn't sufficient to transmit this information.
3222 The ``llvm.module.flags`` named metadata exists in order to facilitate
3223 this. These flags are in the form of key / value pairs --- much like a
3224 dictionary --- making it easy for any subsystem who cares about a flag to
3227 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
3228 Each triplet has the following form:
3230 - The first element is a *behavior* flag, which specifies the behavior
3231 when two (or more) modules are merged together, and it encounters two
3232 (or more) metadata with the same ID. The supported behaviors are
3234 - The second element is a metadata string that is a unique ID for the
3235 metadata. Each module may only have one flag entry for each unique ID (not
3236 including entries with the **Require** behavior).
3237 - The third element is the value of the flag.
3239 When two (or more) modules are merged together, the resulting
3240 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
3241 each unique metadata ID string, there will be exactly one entry in the merged
3242 modules ``llvm.module.flags`` metadata table, and the value for that entry will
3243 be determined by the merge behavior flag, as described below. The only exception
3244 is that entries with the *Require* behavior are always preserved.
3246 The following behaviors are supported:
3257 Emits an error if two values disagree, otherwise the resulting value
3258 is that of the operands.
3262 Emits a warning if two values disagree. The result value will be the
3263 operand for the flag from the first module being linked.
3267 Adds a requirement that another module flag be present and have a
3268 specified value after linking is performed. The value must be a
3269 metadata pair, where the first element of the pair is the ID of the
3270 module flag to be restricted, and the second element of the pair is
3271 the value the module flag should be restricted to. This behavior can
3272 be used to restrict the allowable results (via triggering of an
3273 error) of linking IDs with the **Override** behavior.
3277 Uses the specified value, regardless of the behavior or value of the
3278 other module. If both modules specify **Override**, but the values
3279 differ, an error will be emitted.
3283 Appends the two values, which are required to be metadata nodes.
3287 Appends the two values, which are required to be metadata
3288 nodes. However, duplicate entries in the second list are dropped
3289 during the append operation.
3291 It is an error for a particular unique flag ID to have multiple behaviors,
3292 except in the case of **Require** (which adds restrictions on another metadata
3293 value) or **Override**.
3295 An example of module flags:
3297 .. code-block:: llvm
3299 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
3300 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
3301 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
3302 !3 = metadata !{ i32 3, metadata !"qux",
3304 metadata !"foo", i32 1
3307 !llvm.module.flags = !{ !0, !1, !2, !3 }
3309 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
3310 if two or more ``!"foo"`` flags are seen is to emit an error if their
3311 values are not equal.
3313 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
3314 behavior if two or more ``!"bar"`` flags are seen is to use the value
3317 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
3318 behavior if two or more ``!"qux"`` flags are seen is to emit a
3319 warning if their values are not equal.
3321 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
3325 metadata !{ metadata !"foo", i32 1 }
3327 The behavior is to emit an error if the ``llvm.module.flags`` does not
3328 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
3331 Objective-C Garbage Collection Module Flags Metadata
3332 ----------------------------------------------------
3334 On the Mach-O platform, Objective-C stores metadata about garbage
3335 collection in a special section called "image info". The metadata
3336 consists of a version number and a bitmask specifying what types of
3337 garbage collection are supported (if any) by the file. If two or more
3338 modules are linked together their garbage collection metadata needs to
3339 be merged rather than appended together.
3341 The Objective-C garbage collection module flags metadata consists of the
3342 following key-value pairs:
3351 * - ``Objective-C Version``
3352 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3354 * - ``Objective-C Image Info Version``
3355 - **[Required]** --- The version of the image info section. Currently
3358 * - ``Objective-C Image Info Section``
3359 - **[Required]** --- The section to place the metadata. Valid values are
3360 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3361 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3362 Objective-C ABI version 2.
3364 * - ``Objective-C Garbage Collection``
3365 - **[Required]** --- Specifies whether garbage collection is supported or
3366 not. Valid values are 0, for no garbage collection, and 2, for garbage
3367 collection supported.
3369 * - ``Objective-C GC Only``
3370 - **[Optional]** --- Specifies that only garbage collection is supported.
3371 If present, its value must be 6. This flag requires that the
3372 ``Objective-C Garbage Collection`` flag have the value 2.
3374 Some important flag interactions:
3376 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3377 merged with a module with ``Objective-C Garbage Collection`` set to
3378 2, then the resulting module has the
3379 ``Objective-C Garbage Collection`` flag set to 0.
3380 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3381 merged with a module with ``Objective-C GC Only`` set to 6.
3383 Automatic Linker Flags Module Flags Metadata
3384 --------------------------------------------
3386 Some targets support embedding flags to the linker inside individual object
3387 files. Typically this is used in conjunction with language extensions which
3388 allow source files to explicitly declare the libraries they depend on, and have
3389 these automatically be transmitted to the linker via object files.
3391 These flags are encoded in the IR using metadata in the module flags section,
3392 using the ``Linker Options`` key. The merge behavior for this flag is required
3393 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3394 node which should be a list of other metadata nodes, each of which should be a
3395 list of metadata strings defining linker options.
3397 For example, the following metadata section specifies two separate sets of
3398 linker options, presumably to link against ``libz`` and the ``Cocoa``
3401 !0 = metadata !{ i32 6, metadata !"Linker Options",
3403 metadata !{ metadata !"-lz" },
3404 metadata !{ metadata !"-framework", metadata !"Cocoa" } } }
3405 !llvm.module.flags = !{ !0 }
3407 The metadata encoding as lists of lists of options, as opposed to a collapsed
3408 list of options, is chosen so that the IR encoding can use multiple option
3409 strings to specify e.g., a single library, while still having that specifier be
3410 preserved as an atomic element that can be recognized by a target specific
3411 assembly writer or object file emitter.
3413 Each individual option is required to be either a valid option for the target's
3414 linker, or an option that is reserved by the target specific assembly writer or
3415 object file emitter. No other aspect of these options is defined by the IR.
3417 C type width Module Flags Metadata
3418 ----------------------------------
3420 The ARM backend emits a section into each generated object file describing the
3421 options that it was compiled with (in a compiler-independent way) to prevent
3422 linking incompatible objects, and to allow automatic library selection. Some
3423 of these options are not visible at the IR level, namely wchar_t width and enum
3426 To pass this information to the backend, these options are encoded in module
3427 flags metadata, using the following key-value pairs:
3437 - * 0 --- sizeof(wchar_t) == 4
3438 * 1 --- sizeof(wchar_t) == 2
3441 - * 0 --- Enums are at least as large as an ``int``.
3442 * 1 --- Enums are stored in the smallest integer type which can
3443 represent all of its values.
3445 For example, the following metadata section specifies that the module was
3446 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
3447 enum is the smallest type which can represent all of its values::
3449 !llvm.module.flags = !{!0, !1}
3450 !0 = metadata !{i32 1, metadata !"short_wchar", i32 1}
3451 !1 = metadata !{i32 1, metadata !"short_enum", i32 0}
3453 .. _intrinsicglobalvariables:
3455 Intrinsic Global Variables
3456 ==========================
3458 LLVM has a number of "magic" global variables that contain data that
3459 affect code generation or other IR semantics. These are documented here.
3460 All globals of this sort should have a section specified as
3461 "``llvm.metadata``". This section and all globals that start with
3462 "``llvm.``" are reserved for use by LLVM.
3466 The '``llvm.used``' Global Variable
3467 -----------------------------------
3469 The ``@llvm.used`` global is an array which has
3470 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3471 pointers to named global variables, functions and aliases which may optionally
3472 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3475 .. code-block:: llvm
3480 @llvm.used = appending global [2 x i8*] [
3482 i8* bitcast (i32* @Y to i8*)
3483 ], section "llvm.metadata"
3485 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
3486 and linker are required to treat the symbol as if there is a reference to the
3487 symbol that it cannot see (which is why they have to be named). For example, if
3488 a variable has internal linkage and no references other than that from the
3489 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
3490 references from inline asms and other things the compiler cannot "see", and
3491 corresponds to "``attribute((used))``" in GNU C.
3493 On some targets, the code generator must emit a directive to the
3494 assembler or object file to prevent the assembler and linker from
3495 molesting the symbol.
3497 .. _gv_llvmcompilerused:
3499 The '``llvm.compiler.used``' Global Variable
3500 --------------------------------------------
3502 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
3503 directive, except that it only prevents the compiler from touching the
3504 symbol. On targets that support it, this allows an intelligent linker to
3505 optimize references to the symbol without being impeded as it would be
3508 This is a rare construct that should only be used in rare circumstances,
3509 and should not be exposed to source languages.
3511 .. _gv_llvmglobalctors:
3513 The '``llvm.global_ctors``' Global Variable
3514 -------------------------------------------
3516 .. code-block:: llvm
3518 %0 = type { i32, void ()*, i8* }
3519 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
3521 The ``@llvm.global_ctors`` array contains a list of constructor
3522 functions, priorities, and an optional associated global or function.
3523 The functions referenced by this array will be called in ascending order
3524 of priority (i.e. lowest first) when the module is loaded. The order of
3525 functions with the same priority is not defined.
3527 If the third field is present, non-null, and points to a global variable
3528 or function, the initializer function will only run if the associated
3529 data from the current module is not discarded.
3531 .. _llvmglobaldtors:
3533 The '``llvm.global_dtors``' Global Variable
3534 -------------------------------------------
3536 .. code-block:: llvm
3538 %0 = type { i32, void ()*, i8* }
3539 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
3541 The ``@llvm.global_dtors`` array contains a list of destructor
3542 functions, priorities, and an optional associated global or function.
3543 The functions referenced by this array will be called in descending
3544 order of priority (i.e. highest first) when the module is unloaded. The
3545 order of functions with the same priority is not defined.
3547 If the third field is present, non-null, and points to a global variable
3548 or function, the destructor function will only run if the associated
3549 data from the current module is not discarded.
3551 Instruction Reference
3552 =====================
3554 The LLVM instruction set consists of several different classifications
3555 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
3556 instructions <binaryops>`, :ref:`bitwise binary
3557 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
3558 :ref:`other instructions <otherops>`.
3562 Terminator Instructions
3563 -----------------------
3565 As mentioned :ref:`previously <functionstructure>`, every basic block in a
3566 program ends with a "Terminator" instruction, which indicates which
3567 block should be executed after the current block is finished. These
3568 terminator instructions typically yield a '``void``' value: they produce
3569 control flow, not values (the one exception being the
3570 ':ref:`invoke <i_invoke>`' instruction).
3572 The terminator instructions are: ':ref:`ret <i_ret>`',
3573 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
3574 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
3575 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
3579 '``ret``' Instruction
3580 ^^^^^^^^^^^^^^^^^^^^^
3587 ret <type> <value> ; Return a value from a non-void function
3588 ret void ; Return from void function
3593 The '``ret``' instruction is used to return control flow (and optionally
3594 a value) from a function back to the caller.
3596 There are two forms of the '``ret``' instruction: one that returns a
3597 value and then causes control flow, and one that just causes control
3603 The '``ret``' instruction optionally accepts a single argument, the
3604 return value. The type of the return value must be a ':ref:`first
3605 class <t_firstclass>`' type.
3607 A function is not :ref:`well formed <wellformed>` if it it has a non-void
3608 return type and contains a '``ret``' instruction with no return value or
3609 a return value with a type that does not match its type, or if it has a
3610 void return type and contains a '``ret``' instruction with a return
3616 When the '``ret``' instruction is executed, control flow returns back to
3617 the calling function's context. If the caller is a
3618 ":ref:`call <i_call>`" instruction, execution continues at the
3619 instruction after the call. If the caller was an
3620 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
3621 beginning of the "normal" destination block. If the instruction returns
3622 a value, that value shall set the call or invoke instruction's return
3628 .. code-block:: llvm
3630 ret i32 5 ; Return an integer value of 5
3631 ret void ; Return from a void function
3632 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
3636 '``br``' Instruction
3637 ^^^^^^^^^^^^^^^^^^^^
3644 br i1 <cond>, label <iftrue>, label <iffalse>
3645 br label <dest> ; Unconditional branch
3650 The '``br``' instruction is used to cause control flow to transfer to a
3651 different basic block in the current function. There are two forms of
3652 this instruction, corresponding to a conditional branch and an
3653 unconditional branch.
3658 The conditional branch form of the '``br``' instruction takes a single
3659 '``i1``' value and two '``label``' values. The unconditional form of the
3660 '``br``' instruction takes a single '``label``' value as a target.
3665 Upon execution of a conditional '``br``' instruction, the '``i1``'
3666 argument is evaluated. If the value is ``true``, control flows to the
3667 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
3668 to the '``iffalse``' ``label`` argument.
3673 .. code-block:: llvm
3676 %cond = icmp eq i32 %a, %b
3677 br i1 %cond, label %IfEqual, label %IfUnequal
3685 '``switch``' Instruction
3686 ^^^^^^^^^^^^^^^^^^^^^^^^
3693 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3698 The '``switch``' instruction is used to transfer control flow to one of
3699 several different places. It is a generalization of the '``br``'
3700 instruction, allowing a branch to occur to one of many possible
3706 The '``switch``' instruction uses three parameters: an integer
3707 comparison value '``value``', a default '``label``' destination, and an
3708 array of pairs of comparison value constants and '``label``'s. The table
3709 is not allowed to contain duplicate constant entries.
3714 The ``switch`` instruction specifies a table of values and destinations.
3715 When the '``switch``' instruction is executed, this table is searched
3716 for the given value. If the value is found, control flow is transferred
3717 to the corresponding destination; otherwise, control flow is transferred
3718 to the default destination.
3723 Depending on properties of the target machine and the particular
3724 ``switch`` instruction, this instruction may be code generated in
3725 different ways. For example, it could be generated as a series of
3726 chained conditional branches or with a lookup table.
3731 .. code-block:: llvm
3733 ; Emulate a conditional br instruction
3734 %Val = zext i1 %value to i32
3735 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3737 ; Emulate an unconditional br instruction
3738 switch i32 0, label %dest [ ]
3740 ; Implement a jump table:
3741 switch i32 %val, label %otherwise [ i32 0, label %onzero
3743 i32 2, label %ontwo ]
3747 '``indirectbr``' Instruction
3748 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3755 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3760 The '``indirectbr``' instruction implements an indirect branch to a
3761 label within the current function, whose address is specified by
3762 "``address``". Address must be derived from a
3763 :ref:`blockaddress <blockaddress>` constant.
3768 The '``address``' argument is the address of the label to jump to. The
3769 rest of the arguments indicate the full set of possible destinations
3770 that the address may point to. Blocks are allowed to occur multiple
3771 times in the destination list, though this isn't particularly useful.
3773 This destination list is required so that dataflow analysis has an
3774 accurate understanding of the CFG.
3779 Control transfers to the block specified in the address argument. All
3780 possible destination blocks must be listed in the label list, otherwise
3781 this instruction has undefined behavior. This implies that jumps to
3782 labels defined in other functions have undefined behavior as well.
3787 This is typically implemented with a jump through a register.
3792 .. code-block:: llvm
3794 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3798 '``invoke``' Instruction
3799 ^^^^^^^^^^^^^^^^^^^^^^^^
3806 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
3807 to label <normal label> unwind label <exception label>
3812 The '``invoke``' instruction causes control to transfer to a specified
3813 function, with the possibility of control flow transfer to either the
3814 '``normal``' label or the '``exception``' label. If the callee function
3815 returns with the "``ret``" instruction, control flow will return to the
3816 "normal" label. If the callee (or any indirect callees) returns via the
3817 ":ref:`resume <i_resume>`" instruction or other exception handling
3818 mechanism, control is interrupted and continued at the dynamically
3819 nearest "exception" label.
3821 The '``exception``' label is a `landing
3822 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
3823 '``exception``' label is required to have the
3824 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
3825 information about the behavior of the program after unwinding happens,
3826 as its first non-PHI instruction. The restrictions on the
3827 "``landingpad``" instruction's tightly couples it to the "``invoke``"
3828 instruction, so that the important information contained within the
3829 "``landingpad``" instruction can't be lost through normal code motion.
3834 This instruction requires several arguments:
3836 #. The optional "cconv" marker indicates which :ref:`calling
3837 convention <callingconv>` the call should use. If none is
3838 specified, the call defaults to using C calling conventions.
3839 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
3840 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
3842 #. '``ptr to function ty``': shall be the signature of the pointer to
3843 function value being invoked. In most cases, this is a direct
3844 function invocation, but indirect ``invoke``'s are just as possible,
3845 branching off an arbitrary pointer to function value.
3846 #. '``function ptr val``': An LLVM value containing a pointer to a
3847 function to be invoked.
3848 #. '``function args``': argument list whose types match the function
3849 signature argument types and parameter attributes. All arguments must
3850 be of :ref:`first class <t_firstclass>` type. If the function signature
3851 indicates the function accepts a variable number of arguments, the
3852 extra arguments can be specified.
3853 #. '``normal label``': the label reached when the called function
3854 executes a '``ret``' instruction.
3855 #. '``exception label``': the label reached when a callee returns via
3856 the :ref:`resume <i_resume>` instruction or other exception handling
3858 #. The optional :ref:`function attributes <fnattrs>` list. Only
3859 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
3860 attributes are valid here.
3865 This instruction is designed to operate as a standard '``call``'
3866 instruction in most regards. The primary difference is that it
3867 establishes an association with a label, which is used by the runtime
3868 library to unwind the stack.
3870 This instruction is used in languages with destructors to ensure that
3871 proper cleanup is performed in the case of either a ``longjmp`` or a
3872 thrown exception. Additionally, this is important for implementation of
3873 '``catch``' clauses in high-level languages that support them.
3875 For the purposes of the SSA form, the definition of the value returned
3876 by the '``invoke``' instruction is deemed to occur on the edge from the
3877 current block to the "normal" label. If the callee unwinds then no
3878 return value is available.
3883 .. code-block:: llvm
3885 %retval = invoke i32 @Test(i32 15) to label %Continue
3886 unwind label %TestCleanup ; i32:retval set
3887 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
3888 unwind label %TestCleanup ; i32:retval set
3892 '``resume``' Instruction
3893 ^^^^^^^^^^^^^^^^^^^^^^^^
3900 resume <type> <value>
3905 The '``resume``' instruction is a terminator instruction that has no
3911 The '``resume``' instruction requires one argument, which must have the
3912 same type as the result of any '``landingpad``' instruction in the same
3918 The '``resume``' instruction resumes propagation of an existing
3919 (in-flight) exception whose unwinding was interrupted with a
3920 :ref:`landingpad <i_landingpad>` instruction.
3925 .. code-block:: llvm
3927 resume { i8*, i32 } %exn
3931 '``unreachable``' Instruction
3932 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3944 The '``unreachable``' instruction has no defined semantics. This
3945 instruction is used to inform the optimizer that a particular portion of
3946 the code is not reachable. This can be used to indicate that the code
3947 after a no-return function cannot be reached, and other facts.
3952 The '``unreachable``' instruction has no defined semantics.
3959 Binary operators are used to do most of the computation in a program.
3960 They require two operands of the same type, execute an operation on
3961 them, and produce a single value. The operands might represent multiple
3962 data, as is the case with the :ref:`vector <t_vector>` data type. The
3963 result value has the same type as its operands.
3965 There are several different binary operators:
3969 '``add``' Instruction
3970 ^^^^^^^^^^^^^^^^^^^^^
3977 <result> = add <ty> <op1>, <op2> ; yields ty:result
3978 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
3979 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
3980 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
3985 The '``add``' instruction returns the sum of its two operands.
3990 The two arguments to the '``add``' instruction must be
3991 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
3992 arguments must have identical types.
3997 The value produced is the integer sum of the two operands.
3999 If the sum has unsigned overflow, the result returned is the
4000 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4003 Because LLVM integers use a two's complement representation, this
4004 instruction is appropriate for both signed and unsigned integers.
4006 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4007 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4008 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
4009 unsigned and/or signed overflow, respectively, occurs.
4014 .. code-block:: llvm
4016 <result> = add i32 4, %var ; yields i32:result = 4 + %var
4020 '``fadd``' Instruction
4021 ^^^^^^^^^^^^^^^^^^^^^^
4028 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4033 The '``fadd``' instruction returns the sum of its two operands.
4038 The two arguments to the '``fadd``' instruction must be :ref:`floating
4039 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4040 Both arguments must have identical types.
4045 The value produced is the floating point sum of the two operands. This
4046 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
4047 which are optimization hints to enable otherwise unsafe floating point
4053 .. code-block:: llvm
4055 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
4057 '``sub``' Instruction
4058 ^^^^^^^^^^^^^^^^^^^^^
4065 <result> = sub <ty> <op1>, <op2> ; yields ty:result
4066 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
4067 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
4068 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
4073 The '``sub``' instruction returns the difference of its two operands.
4075 Note that the '``sub``' instruction is used to represent the '``neg``'
4076 instruction present in most other intermediate representations.
4081 The two arguments to the '``sub``' instruction must be
4082 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4083 arguments must have identical types.
4088 The value produced is the integer difference of the two operands.
4090 If the difference has unsigned overflow, the result returned is the
4091 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4094 Because LLVM integers use a two's complement representation, this
4095 instruction is appropriate for both signed and unsigned integers.
4097 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4098 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4099 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
4100 unsigned and/or signed overflow, respectively, occurs.
4105 .. code-block:: llvm
4107 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
4108 <result> = sub i32 0, %val ; yields i32:result = -%var
4112 '``fsub``' Instruction
4113 ^^^^^^^^^^^^^^^^^^^^^^
4120 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4125 The '``fsub``' instruction returns the difference of its two operands.
4127 Note that the '``fsub``' instruction is used to represent the '``fneg``'
4128 instruction present in most other intermediate representations.
4133 The two arguments to the '``fsub``' instruction must be :ref:`floating
4134 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4135 Both arguments must have identical types.
4140 The value produced is the floating point difference of the two operands.
4141 This instruction can also take any number of :ref:`fast-math
4142 flags <fastmath>`, which are optimization hints to enable otherwise
4143 unsafe floating point optimizations:
4148 .. code-block:: llvm
4150 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
4151 <result> = fsub float -0.0, %val ; yields float:result = -%var
4153 '``mul``' Instruction
4154 ^^^^^^^^^^^^^^^^^^^^^
4161 <result> = mul <ty> <op1>, <op2> ; yields ty:result
4162 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
4163 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
4164 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
4169 The '``mul``' instruction returns the product of its two operands.
4174 The two arguments to the '``mul``' instruction must be
4175 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4176 arguments must have identical types.
4181 The value produced is the integer product of the two operands.
4183 If the result of the multiplication has unsigned overflow, the result
4184 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
4185 bit width of the result.
4187 Because LLVM integers use a two's complement representation, and the
4188 result is the same width as the operands, this instruction returns the
4189 correct result for both signed and unsigned integers. If a full product
4190 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
4191 sign-extended or zero-extended as appropriate to the width of the full
4194 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4195 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4196 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
4197 unsigned and/or signed overflow, respectively, occurs.
4202 .. code-block:: llvm
4204 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
4208 '``fmul``' Instruction
4209 ^^^^^^^^^^^^^^^^^^^^^^
4216 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4221 The '``fmul``' instruction returns the product of its two operands.
4226 The two arguments to the '``fmul``' instruction must be :ref:`floating
4227 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4228 Both arguments must have identical types.
4233 The value produced is the floating point product of the two operands.
4234 This instruction can also take any number of :ref:`fast-math
4235 flags <fastmath>`, which are optimization hints to enable otherwise
4236 unsafe floating point optimizations:
4241 .. code-block:: llvm
4243 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
4245 '``udiv``' Instruction
4246 ^^^^^^^^^^^^^^^^^^^^^^
4253 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
4254 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
4259 The '``udiv``' instruction returns the quotient of its two operands.
4264 The two arguments to the '``udiv``' instruction must be
4265 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4266 arguments must have identical types.
4271 The value produced is the unsigned integer quotient of the two operands.
4273 Note that unsigned integer division and signed integer division are
4274 distinct operations; for signed integer division, use '``sdiv``'.
4276 Division by zero leads to undefined behavior.
4278 If the ``exact`` keyword is present, the result value of the ``udiv`` is
4279 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
4280 such, "((a udiv exact b) mul b) == a").
4285 .. code-block:: llvm
4287 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
4289 '``sdiv``' Instruction
4290 ^^^^^^^^^^^^^^^^^^^^^^
4297 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
4298 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
4303 The '``sdiv``' instruction returns the quotient of its two operands.
4308 The two arguments to the '``sdiv``' instruction must be
4309 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4310 arguments must have identical types.
4315 The value produced is the signed integer quotient of the two operands
4316 rounded towards zero.
4318 Note that signed integer division and unsigned integer division are
4319 distinct operations; for unsigned integer division, use '``udiv``'.
4321 Division by zero leads to undefined behavior. Overflow also leads to
4322 undefined behavior; this is a rare case, but can occur, for example, by
4323 doing a 32-bit division of -2147483648 by -1.
4325 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
4326 a :ref:`poison value <poisonvalues>` if the result would be rounded.
4331 .. code-block:: llvm
4333 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
4337 '``fdiv``' Instruction
4338 ^^^^^^^^^^^^^^^^^^^^^^
4345 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4350 The '``fdiv``' instruction returns the quotient of its two operands.
4355 The two arguments to the '``fdiv``' instruction must be :ref:`floating
4356 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4357 Both arguments must have identical types.
4362 The value produced is the floating point quotient of the two operands.
4363 This instruction can also take any number of :ref:`fast-math
4364 flags <fastmath>`, which are optimization hints to enable otherwise
4365 unsafe floating point optimizations:
4370 .. code-block:: llvm
4372 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
4374 '``urem``' Instruction
4375 ^^^^^^^^^^^^^^^^^^^^^^
4382 <result> = urem <ty> <op1>, <op2> ; yields ty:result
4387 The '``urem``' instruction returns the remainder from the unsigned
4388 division of its two arguments.
4393 The two arguments to the '``urem``' instruction must be
4394 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4395 arguments must have identical types.
4400 This instruction returns the unsigned integer *remainder* of a division.
4401 This instruction always performs an unsigned division to get the
4404 Note that unsigned integer remainder and signed integer remainder are
4405 distinct operations; for signed integer remainder, use '``srem``'.
4407 Taking the remainder of a division by zero leads to undefined behavior.
4412 .. code-block:: llvm
4414 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
4416 '``srem``' Instruction
4417 ^^^^^^^^^^^^^^^^^^^^^^
4424 <result> = srem <ty> <op1>, <op2> ; yields ty:result
4429 The '``srem``' instruction returns the remainder from the signed
4430 division of its two operands. This instruction can also take
4431 :ref:`vector <t_vector>` versions of the values in which case the elements
4437 The two arguments to the '``srem``' instruction must be
4438 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4439 arguments must have identical types.
4444 This instruction returns the *remainder* of a division (where the result
4445 is either zero or has the same sign as the dividend, ``op1``), not the
4446 *modulo* operator (where the result is either zero or has the same sign
4447 as the divisor, ``op2``) of a value. For more information about the
4448 difference, see `The Math
4449 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4450 table of how this is implemented in various languages, please see
4452 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4454 Note that signed integer remainder and unsigned integer remainder are
4455 distinct operations; for unsigned integer remainder, use '``urem``'.
4457 Taking the remainder of a division by zero leads to undefined behavior.
4458 Overflow also leads to undefined behavior; this is a rare case, but can
4459 occur, for example, by taking the remainder of a 32-bit division of
4460 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4461 rule lets srem be implemented using instructions that return both the
4462 result of the division and the remainder.)
4467 .. code-block:: llvm
4469 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
4473 '``frem``' Instruction
4474 ^^^^^^^^^^^^^^^^^^^^^^
4481 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4486 The '``frem``' instruction returns the remainder from the division of
4492 The two arguments to the '``frem``' instruction must be :ref:`floating
4493 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4494 Both arguments must have identical types.
4499 This instruction returns the *remainder* of a division. The remainder
4500 has the same sign as the dividend. This instruction can also take any
4501 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
4502 to enable otherwise unsafe floating point optimizations:
4507 .. code-block:: llvm
4509 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
4513 Bitwise Binary Operations
4514 -------------------------
4516 Bitwise binary operators are used to do various forms of bit-twiddling
4517 in a program. They are generally very efficient instructions and can
4518 commonly be strength reduced from other instructions. They require two
4519 operands of the same type, execute an operation on them, and produce a
4520 single value. The resulting value is the same type as its operands.
4522 '``shl``' Instruction
4523 ^^^^^^^^^^^^^^^^^^^^^
4530 <result> = shl <ty> <op1>, <op2> ; yields ty:result
4531 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
4532 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
4533 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
4538 The '``shl``' instruction returns the first operand shifted to the left
4539 a specified number of bits.
4544 Both arguments to the '``shl``' instruction must be the same
4545 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4546 '``op2``' is treated as an unsigned value.
4551 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
4552 where ``n`` is the width of the result. If ``op2`` is (statically or
4553 dynamically) negative or equal to or larger than the number of bits in
4554 ``op1``, the result is undefined. If the arguments are vectors, each
4555 vector element of ``op1`` is shifted by the corresponding shift amount
4558 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
4559 value <poisonvalues>` if it shifts out any non-zero bits. If the
4560 ``nsw`` keyword is present, then the shift produces a :ref:`poison
4561 value <poisonvalues>` if it shifts out any bits that disagree with the
4562 resultant sign bit. As such, NUW/NSW have the same semantics as they
4563 would if the shift were expressed as a mul instruction with the same
4564 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
4569 .. code-block:: llvm
4571 <result> = shl i32 4, %var ; yields i32: 4 << %var
4572 <result> = shl i32 4, 2 ; yields i32: 16
4573 <result> = shl i32 1, 10 ; yields i32: 1024
4574 <result> = shl i32 1, 32 ; undefined
4575 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
4577 '``lshr``' Instruction
4578 ^^^^^^^^^^^^^^^^^^^^^^
4585 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
4586 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
4591 The '``lshr``' instruction (logical shift right) returns the first
4592 operand shifted to the right a specified number of bits with zero fill.
4597 Both arguments to the '``lshr``' instruction must be the same
4598 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4599 '``op2``' is treated as an unsigned value.
4604 This instruction always performs a logical shift right operation. The
4605 most significant bits of the result will be filled with zero bits after
4606 the shift. If ``op2`` is (statically or dynamically) equal to or larger
4607 than the number of bits in ``op1``, the result is undefined. If the
4608 arguments are vectors, each vector element of ``op1`` is shifted by the
4609 corresponding shift amount in ``op2``.
4611 If the ``exact`` keyword is present, the result value of the ``lshr`` is
4612 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4618 .. code-block:: llvm
4620 <result> = lshr i32 4, 1 ; yields i32:result = 2
4621 <result> = lshr i32 4, 2 ; yields i32:result = 1
4622 <result> = lshr i8 4, 3 ; yields i8:result = 0
4623 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
4624 <result> = lshr i32 1, 32 ; undefined
4625 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
4627 '``ashr``' Instruction
4628 ^^^^^^^^^^^^^^^^^^^^^^
4635 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
4636 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
4641 The '``ashr``' instruction (arithmetic shift right) returns the first
4642 operand shifted to the right a specified number of bits with sign
4648 Both arguments to the '``ashr``' instruction must be the same
4649 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
4650 '``op2``' is treated as an unsigned value.
4655 This instruction always performs an arithmetic shift right operation,
4656 The most significant bits of the result will be filled with the sign bit
4657 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
4658 than the number of bits in ``op1``, the result is undefined. If the
4659 arguments are vectors, each vector element of ``op1`` is shifted by the
4660 corresponding shift amount in ``op2``.
4662 If the ``exact`` keyword is present, the result value of the ``ashr`` is
4663 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
4669 .. code-block:: llvm
4671 <result> = ashr i32 4, 1 ; yields i32:result = 2
4672 <result> = ashr i32 4, 2 ; yields i32:result = 1
4673 <result> = ashr i8 4, 3 ; yields i8:result = 0
4674 <result> = ashr i8 -2, 1 ; yields i8:result = -1
4675 <result> = ashr i32 1, 32 ; undefined
4676 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
4678 '``and``' Instruction
4679 ^^^^^^^^^^^^^^^^^^^^^
4686 <result> = and <ty> <op1>, <op2> ; yields ty:result
4691 The '``and``' instruction returns the bitwise logical and of its two
4697 The two arguments to the '``and``' instruction must be
4698 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4699 arguments must have identical types.
4704 The truth table used for the '``and``' instruction is:
4721 .. code-block:: llvm
4723 <result> = and i32 4, %var ; yields i32:result = 4 & %var
4724 <result> = and i32 15, 40 ; yields i32:result = 8
4725 <result> = and i32 4, 8 ; yields i32:result = 0
4727 '``or``' Instruction
4728 ^^^^^^^^^^^^^^^^^^^^
4735 <result> = or <ty> <op1>, <op2> ; yields ty:result
4740 The '``or``' instruction returns the bitwise logical inclusive or of its
4746 The two arguments to the '``or``' instruction must be
4747 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4748 arguments must have identical types.
4753 The truth table used for the '``or``' instruction is:
4772 <result> = or i32 4, %var ; yields i32:result = 4 | %var
4773 <result> = or i32 15, 40 ; yields i32:result = 47
4774 <result> = or i32 4, 8 ; yields i32:result = 12
4776 '``xor``' Instruction
4777 ^^^^^^^^^^^^^^^^^^^^^
4784 <result> = xor <ty> <op1>, <op2> ; yields ty:result
4789 The '``xor``' instruction returns the bitwise logical exclusive or of
4790 its two operands. The ``xor`` is used to implement the "one's
4791 complement" operation, which is the "~" operator in C.
4796 The two arguments to the '``xor``' instruction must be
4797 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4798 arguments must have identical types.
4803 The truth table used for the '``xor``' instruction is:
4820 .. code-block:: llvm
4822 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
4823 <result> = xor i32 15, 40 ; yields i32:result = 39
4824 <result> = xor i32 4, 8 ; yields i32:result = 12
4825 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
4830 LLVM supports several instructions to represent vector operations in a
4831 target-independent manner. These instructions cover the element-access
4832 and vector-specific operations needed to process vectors effectively.
4833 While LLVM does directly support these vector operations, many
4834 sophisticated algorithms will want to use target-specific intrinsics to
4835 take full advantage of a specific target.
4837 .. _i_extractelement:
4839 '``extractelement``' Instruction
4840 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4847 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
4852 The '``extractelement``' instruction extracts a single scalar element
4853 from a vector at a specified index.
4858 The first operand of an '``extractelement``' instruction is a value of
4859 :ref:`vector <t_vector>` type. The second operand is an index indicating
4860 the position from which to extract the element. The index may be a
4861 variable of any integer type.
4866 The result is a scalar of the same type as the element type of ``val``.
4867 Its value is the value at position ``idx`` of ``val``. If ``idx``
4868 exceeds the length of ``val``, the results are undefined.
4873 .. code-block:: llvm
4875 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
4877 .. _i_insertelement:
4879 '``insertelement``' Instruction
4880 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4887 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
4892 The '``insertelement``' instruction inserts a scalar element into a
4893 vector at a specified index.
4898 The first operand of an '``insertelement``' instruction is a value of
4899 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
4900 type must equal the element type of the first operand. The third operand
4901 is an index indicating the position at which to insert the value. The
4902 index may be a variable of any integer type.
4907 The result is a vector of the same type as ``val``. Its element values
4908 are those of ``val`` except at position ``idx``, where it gets the value
4909 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
4915 .. code-block:: llvm
4917 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
4919 .. _i_shufflevector:
4921 '``shufflevector``' Instruction
4922 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4929 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
4934 The '``shufflevector``' instruction constructs a permutation of elements
4935 from two input vectors, returning a vector with the same element type as
4936 the input and length that is the same as the shuffle mask.
4941 The first two operands of a '``shufflevector``' instruction are vectors
4942 with the same type. The third argument is a shuffle mask whose element
4943 type is always 'i32'. The result of the instruction is a vector whose
4944 length is the same as the shuffle mask and whose element type is the
4945 same as the element type of the first two operands.
4947 The shuffle mask operand is required to be a constant vector with either
4948 constant integer or undef values.
4953 The elements of the two input vectors are numbered from left to right
4954 across both of the vectors. The shuffle mask operand specifies, for each
4955 element of the result vector, which element of the two input vectors the
4956 result element gets. The element selector may be undef (meaning "don't
4957 care") and the second operand may be undef if performing a shuffle from
4963 .. code-block:: llvm
4965 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4966 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
4967 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4968 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
4969 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4970 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
4971 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4972 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
4974 Aggregate Operations
4975 --------------------
4977 LLVM supports several instructions for working with
4978 :ref:`aggregate <t_aggregate>` values.
4982 '``extractvalue``' Instruction
4983 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4990 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4995 The '``extractvalue``' instruction extracts the value of a member field
4996 from an :ref:`aggregate <t_aggregate>` value.
5001 The first operand of an '``extractvalue``' instruction is a value of
5002 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
5003 constant indices to specify which value to extract in a similar manner
5004 as indices in a '``getelementptr``' instruction.
5006 The major differences to ``getelementptr`` indexing are:
5008 - Since the value being indexed is not a pointer, the first index is
5009 omitted and assumed to be zero.
5010 - At least one index must be specified.
5011 - Not only struct indices but also array indices must be in bounds.
5016 The result is the value at the position in the aggregate specified by
5022 .. code-block:: llvm
5024 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
5028 '``insertvalue``' Instruction
5029 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5036 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
5041 The '``insertvalue``' instruction inserts a value into a member field in
5042 an :ref:`aggregate <t_aggregate>` value.
5047 The first operand of an '``insertvalue``' instruction is a value of
5048 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
5049 a first-class value to insert. The following operands are constant
5050 indices indicating the position at which to insert the value in a
5051 similar manner as indices in a '``extractvalue``' instruction. The value
5052 to insert must have the same type as the value identified by the
5058 The result is an aggregate of the same type as ``val``. Its value is
5059 that of ``val`` except that the value at the position specified by the
5060 indices is that of ``elt``.
5065 .. code-block:: llvm
5067 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
5068 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
5069 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 ; yields {i32 1, float %val}
5073 Memory Access and Addressing Operations
5074 ---------------------------------------
5076 A key design point of an SSA-based representation is how it represents
5077 memory. In LLVM, no memory locations are in SSA form, which makes things
5078 very simple. This section describes how to read, write, and allocate
5083 '``alloca``' Instruction
5084 ^^^^^^^^^^^^^^^^^^^^^^^^
5091 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
5096 The '``alloca``' instruction allocates memory on the stack frame of the
5097 currently executing function, to be automatically released when this
5098 function returns to its caller. The object is always allocated in the
5099 generic address space (address space zero).
5104 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
5105 bytes of memory on the runtime stack, returning a pointer of the
5106 appropriate type to the program. If "NumElements" is specified, it is
5107 the number of elements allocated, otherwise "NumElements" is defaulted
5108 to be one. If a constant alignment is specified, the value result of the
5109 allocation is guaranteed to be aligned to at least that boundary. The
5110 alignment may not be greater than ``1 << 29``. If not specified, or if
5111 zero, the target can choose to align the allocation on any convenient
5112 boundary compatible with the type.
5114 '``type``' may be any sized type.
5119 Memory is allocated; a pointer is returned. The operation is undefined
5120 if there is insufficient stack space for the allocation. '``alloca``'d
5121 memory is automatically released when the function returns. The
5122 '``alloca``' instruction is commonly used to represent automatic
5123 variables that must have an address available. When the function returns
5124 (either with the ``ret`` or ``resume`` instructions), the memory is
5125 reclaimed. Allocating zero bytes is legal, but the result is undefined.
5126 The order in which memory is allocated (ie., which way the stack grows)
5132 .. code-block:: llvm
5134 %ptr = alloca i32 ; yields i32*:ptr
5135 %ptr = alloca i32, i32 4 ; yields i32*:ptr
5136 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
5137 %ptr = alloca i32, align 1024 ; yields i32*:ptr
5141 '``load``' Instruction
5142 ^^^^^^^^^^^^^^^^^^^^^^
5149 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
5150 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
5151 !<index> = !{ i32 1 }
5156 The '``load``' instruction is used to read from memory.
5161 The argument to the ``load`` instruction specifies the memory address
5162 from which to load. The pointer must point to a :ref:`first
5163 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
5164 then the optimizer is not allowed to modify the number or order of
5165 execution of this ``load`` with other :ref:`volatile
5166 operations <volatile>`.
5168 If the ``load`` is marked as ``atomic``, it takes an extra
5169 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5170 ``release`` and ``acq_rel`` orderings are not valid on ``load``
5171 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5172 when they may see multiple atomic stores. The type of the pointee must
5173 be an integer type whose bit width is a power of two greater than or
5174 equal to eight and less than or equal to a target-specific size limit.
5175 ``align`` must be explicitly specified on atomic loads, and the load has
5176 undefined behavior if the alignment is not set to a value which is at
5177 least the size in bytes of the pointee. ``!nontemporal`` does not have
5178 any defined semantics for atomic loads.
5180 The optional constant ``align`` argument specifies the alignment of the
5181 operation (that is, the alignment of the memory address). A value of 0
5182 or an omitted ``align`` argument means that the operation has the ABI
5183 alignment for the target. It is the responsibility of the code emitter
5184 to ensure that the alignment information is correct. Overestimating the
5185 alignment results in undefined behavior. Underestimating the alignment
5186 may produce less efficient code. An alignment of 1 is always safe. The
5187 maximum possible alignment is ``1 << 29``.
5189 The optional ``!nontemporal`` metadata must reference a single
5190 metadata name ``<index>`` corresponding to a metadata node with one
5191 ``i32`` entry of value 1. The existence of the ``!nontemporal``
5192 metadata on the instruction tells the optimizer and code generator
5193 that this load is not expected to be reused in the cache. The code
5194 generator may select special instructions to save cache bandwidth, such
5195 as the ``MOVNT`` instruction on x86.
5197 The optional ``!invariant.load`` metadata must reference a single
5198 metadata name ``<index>`` corresponding to a metadata node with no
5199 entries. The existence of the ``!invariant.load`` metadata on the
5200 instruction tells the optimizer and code generator that this load
5201 address points to memory which does not change value during program
5202 execution. The optimizer may then move this load around, for example, by
5203 hoisting it out of loops using loop invariant code motion.
5208 The location of memory pointed to is loaded. If the value being loaded
5209 is of scalar type then the number of bytes read does not exceed the
5210 minimum number of bytes needed to hold all bits of the type. For
5211 example, loading an ``i24`` reads at most three bytes. When loading a
5212 value of a type like ``i20`` with a size that is not an integral number
5213 of bytes, the result is undefined if the value was not originally
5214 written using a store of the same type.
5219 .. code-block:: llvm
5221 %ptr = alloca i32 ; yields i32*:ptr
5222 store i32 3, i32* %ptr ; yields void
5223 %val = load i32* %ptr ; yields i32:val = i32 3
5227 '``store``' Instruction
5228 ^^^^^^^^^^^^^^^^^^^^^^^
5235 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields void
5236 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields void
5241 The '``store``' instruction is used to write to memory.
5246 There are two arguments to the ``store`` instruction: a value to store
5247 and an address at which to store it. The type of the ``<pointer>``
5248 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
5249 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
5250 then the optimizer is not allowed to modify the number or order of
5251 execution of this ``store`` with other :ref:`volatile
5252 operations <volatile>`.
5254 If the ``store`` is marked as ``atomic``, it takes an extra
5255 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5256 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
5257 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5258 when they may see multiple atomic stores. The type of the pointee must
5259 be an integer type whose bit width is a power of two greater than or
5260 equal to eight and less than or equal to a target-specific size limit.
5261 ``align`` must be explicitly specified on atomic stores, and the store
5262 has undefined behavior if the alignment is not set to a value which is
5263 at least the size in bytes of the pointee. ``!nontemporal`` does not
5264 have any defined semantics for atomic stores.
5266 The optional constant ``align`` argument specifies the alignment of the
5267 operation (that is, the alignment of the memory address). A value of 0
5268 or an omitted ``align`` argument means that the operation has the ABI
5269 alignment for the target. It is the responsibility of the code emitter
5270 to ensure that the alignment information is correct. Overestimating the
5271 alignment results in undefined behavior. Underestimating the
5272 alignment may produce less efficient code. An alignment of 1 is always
5273 safe. The maximum possible alignment is ``1 << 29``.
5275 The optional ``!nontemporal`` metadata must reference a single metadata
5276 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
5277 value 1. The existence of the ``!nontemporal`` metadata on the instruction
5278 tells the optimizer and code generator that this load is not expected to
5279 be reused in the cache. The code generator may select special
5280 instructions to save cache bandwidth, such as the MOVNT instruction on
5286 The contents of memory are updated to contain ``<value>`` at the
5287 location specified by the ``<pointer>`` operand. If ``<value>`` is
5288 of scalar type then the number of bytes written does not exceed the
5289 minimum number of bytes needed to hold all bits of the type. For
5290 example, storing an ``i24`` writes at most three bytes. When writing a
5291 value of a type like ``i20`` with a size that is not an integral number
5292 of bytes, it is unspecified what happens to the extra bits that do not
5293 belong to the type, but they will typically be overwritten.
5298 .. code-block:: llvm
5300 %ptr = alloca i32 ; yields i32*:ptr
5301 store i32 3, i32* %ptr ; yields void
5302 %val = load i32* %ptr ; yields i32:val = i32 3
5306 '``fence``' Instruction
5307 ^^^^^^^^^^^^^^^^^^^^^^^
5314 fence [singlethread] <ordering> ; yields void
5319 The '``fence``' instruction is used to introduce happens-before edges
5325 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
5326 defines what *synchronizes-with* edges they add. They can only be given
5327 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
5332 A fence A which has (at least) ``release`` ordering semantics
5333 *synchronizes with* a fence B with (at least) ``acquire`` ordering
5334 semantics if and only if there exist atomic operations X and Y, both
5335 operating on some atomic object M, such that A is sequenced before X, X
5336 modifies M (either directly or through some side effect of a sequence
5337 headed by X), Y is sequenced before B, and Y observes M. This provides a
5338 *happens-before* dependency between A and B. Rather than an explicit
5339 ``fence``, one (but not both) of the atomic operations X or Y might
5340 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
5341 still *synchronize-with* the explicit ``fence`` and establish the
5342 *happens-before* edge.
5344 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
5345 ``acquire`` and ``release`` semantics specified above, participates in
5346 the global program order of other ``seq_cst`` operations and/or fences.
5348 The optional ":ref:`singlethread <singlethread>`" argument specifies
5349 that the fence only synchronizes with other fences in the same thread.
5350 (This is useful for interacting with signal handlers.)
5355 .. code-block:: llvm
5357 fence acquire ; yields void
5358 fence singlethread seq_cst ; yields void
5362 '``cmpxchg``' Instruction
5363 ^^^^^^^^^^^^^^^^^^^^^^^^^
5370 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
5375 The '``cmpxchg``' instruction is used to atomically modify memory. It
5376 loads a value in memory and compares it to a given value. If they are
5377 equal, it tries to store a new value into the memory.
5382 There are three arguments to the '``cmpxchg``' instruction: an address
5383 to operate on, a value to compare to the value currently be at that
5384 address, and a new value to place at that address if the compared values
5385 are equal. The type of '<cmp>' must be an integer type whose bit width
5386 is a power of two greater than or equal to eight and less than or equal
5387 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5388 type, and the type of '<pointer>' must be a pointer to that type. If the
5389 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5390 to modify the number or order of execution of this ``cmpxchg`` with
5391 other :ref:`volatile operations <volatile>`.
5393 The success and failure :ref:`ordering <ordering>` arguments specify how this
5394 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
5395 must be at least ``monotonic``, the ordering constraint on failure must be no
5396 stronger than that on success, and the failure ordering cannot be either
5397 ``release`` or ``acq_rel``.
5399 The optional "``singlethread``" argument declares that the ``cmpxchg``
5400 is only atomic with respect to code (usually signal handlers) running in
5401 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5402 respect to all other code in the system.
5404 The pointer passed into cmpxchg must have alignment greater than or
5405 equal to the size in memory of the operand.
5410 The contents of memory at the location specified by the '``<pointer>``' operand
5411 is read and compared to '``<cmp>``'; if the read value is the equal, the
5412 '``<new>``' is written. The original value at the location is returned, together
5413 with a flag indicating success (true) or failure (false).
5415 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
5416 permitted: the operation may not write ``<new>`` even if the comparison
5419 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
5420 if the value loaded equals ``cmp``.
5422 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5423 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5424 load with an ordering parameter determined the second ordering parameter.
5429 .. code-block:: llvm
5432 %orig = atomic load i32* %ptr unordered ; yields i32
5436 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5437 %squared = mul i32 %cmp, %cmp
5438 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
5439 %value_loaded = extractvalue { i32, i1 } %val_success, 0
5440 %success = extractvalue { i32, i1 } %val_success, 1
5441 br i1 %success, label %done, label %loop
5448 '``atomicrmw``' Instruction
5449 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
5456 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
5461 The '``atomicrmw``' instruction is used to atomically modify memory.
5466 There are three arguments to the '``atomicrmw``' instruction: an
5467 operation to apply, an address whose value to modify, an argument to the
5468 operation. The operation must be one of the following keywords:
5482 The type of '<value>' must be an integer type whose bit width is a power
5483 of two greater than or equal to eight and less than or equal to a
5484 target-specific size limit. The type of the '``<pointer>``' operand must
5485 be a pointer to that type. If the ``atomicrmw`` is marked as
5486 ``volatile``, then the optimizer is not allowed to modify the number or
5487 order of execution of this ``atomicrmw`` with other :ref:`volatile
5488 operations <volatile>`.
5493 The contents of memory at the location specified by the '``<pointer>``'
5494 operand are atomically read, modified, and written back. The original
5495 value at the location is returned. The modification is specified by the
5498 - xchg: ``*ptr = val``
5499 - add: ``*ptr = *ptr + val``
5500 - sub: ``*ptr = *ptr - val``
5501 - and: ``*ptr = *ptr & val``
5502 - nand: ``*ptr = ~(*ptr & val)``
5503 - or: ``*ptr = *ptr | val``
5504 - xor: ``*ptr = *ptr ^ val``
5505 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
5506 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
5507 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
5509 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
5515 .. code-block:: llvm
5517 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
5519 .. _i_getelementptr:
5521 '``getelementptr``' Instruction
5522 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5529 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5530 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5531 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5536 The '``getelementptr``' instruction is used to get the address of a
5537 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
5538 address calculation only and does not access memory.
5543 The first argument is always a pointer or a vector of pointers, and
5544 forms the basis of the calculation. The remaining arguments are indices
5545 that indicate which of the elements of the aggregate object are indexed.
5546 The interpretation of each index is dependent on the type being indexed
5547 into. The first index always indexes the pointer value given as the
5548 first argument, the second index indexes a value of the type pointed to
5549 (not necessarily the value directly pointed to, since the first index
5550 can be non-zero), etc. The first type indexed into must be a pointer
5551 value, subsequent types can be arrays, vectors, and structs. Note that
5552 subsequent types being indexed into can never be pointers, since that
5553 would require loading the pointer before continuing calculation.
5555 The type of each index argument depends on the type it is indexing into.
5556 When indexing into a (optionally packed) structure, only ``i32`` integer
5557 **constants** are allowed (when using a vector of indices they must all
5558 be the **same** ``i32`` integer constant). When indexing into an array,
5559 pointer or vector, integers of any width are allowed, and they are not
5560 required to be constant. These integers are treated as signed values
5563 For example, let's consider a C code fragment and how it gets compiled
5579 int *foo(struct ST *s) {
5580 return &s[1].Z.B[5][13];
5583 The LLVM code generated by Clang is:
5585 .. code-block:: llvm
5587 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
5588 %struct.ST = type { i32, double, %struct.RT }
5590 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5592 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5599 In the example above, the first index is indexing into the
5600 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
5601 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
5602 indexes into the third element of the structure, yielding a
5603 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
5604 structure. The third index indexes into the second element of the
5605 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
5606 dimensions of the array are subscripted into, yielding an '``i32``'
5607 type. The '``getelementptr``' instruction returns a pointer to this
5608 element, thus computing a value of '``i32*``' type.
5610 Note that it is perfectly legal to index partially through a structure,
5611 returning a pointer to an inner element. Because of this, the LLVM code
5612 for the given testcase is equivalent to:
5614 .. code-block:: llvm
5616 define i32* @foo(%struct.ST* %s) {
5617 %t1 = getelementptr %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
5618 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
5619 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
5620 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
5621 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
5625 If the ``inbounds`` keyword is present, the result value of the
5626 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
5627 pointer is not an *in bounds* address of an allocated object, or if any
5628 of the addresses that would be formed by successive addition of the
5629 offsets implied by the indices to the base address with infinitely
5630 precise signed arithmetic are not an *in bounds* address of that
5631 allocated object. The *in bounds* addresses for an allocated object are
5632 all the addresses that point into the object, plus the address one byte
5633 past the end. In cases where the base is a vector of pointers the
5634 ``inbounds`` keyword applies to each of the computations element-wise.
5636 If the ``inbounds`` keyword is not present, the offsets are added to the
5637 base address with silently-wrapping two's complement arithmetic. If the
5638 offsets have a different width from the pointer, they are sign-extended
5639 or truncated to the width of the pointer. The result value of the
5640 ``getelementptr`` may be outside the object pointed to by the base
5641 pointer. The result value may not necessarily be used to access memory
5642 though, even if it happens to point into allocated storage. See the
5643 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
5646 The getelementptr instruction is often confusing. For some more insight
5647 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
5652 .. code-block:: llvm
5654 ; yields [12 x i8]*:aptr
5655 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5657 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5659 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5661 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5663 In cases where the pointer argument is a vector of pointers, each index
5664 must be a vector with the same number of elements. For example:
5666 .. code-block:: llvm
5668 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5670 Conversion Operations
5671 ---------------------
5673 The instructions in this category are the conversion instructions
5674 (casting) which all take a single operand and a type. They perform
5675 various bit conversions on the operand.
5677 '``trunc .. to``' Instruction
5678 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5685 <result> = trunc <ty> <value> to <ty2> ; yields ty2
5690 The '``trunc``' instruction truncates its operand to the type ``ty2``.
5695 The '``trunc``' instruction takes a value to trunc, and a type to trunc
5696 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
5697 of the same number of integers. The bit size of the ``value`` must be
5698 larger than the bit size of the destination type, ``ty2``. Equal sized
5699 types are not allowed.
5704 The '``trunc``' instruction truncates the high order bits in ``value``
5705 and converts the remaining bits to ``ty2``. Since the source size must
5706 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
5707 It will always truncate bits.
5712 .. code-block:: llvm
5714 %X = trunc i32 257 to i8 ; yields i8:1
5715 %Y = trunc i32 123 to i1 ; yields i1:true
5716 %Z = trunc i32 122 to i1 ; yields i1:false
5717 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
5719 '``zext .. to``' Instruction
5720 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5727 <result> = zext <ty> <value> to <ty2> ; yields ty2
5732 The '``zext``' instruction zero extends its operand to type ``ty2``.
5737 The '``zext``' instruction takes a value to cast, and a type to cast it
5738 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5739 the same number of integers. The bit size of the ``value`` must be
5740 smaller than the bit size of the destination type, ``ty2``.
5745 The ``zext`` fills the high order bits of the ``value`` with zero bits
5746 until it reaches the size of the destination type, ``ty2``.
5748 When zero extending from i1, the result will always be either 0 or 1.
5753 .. code-block:: llvm
5755 %X = zext i32 257 to i64 ; yields i64:257
5756 %Y = zext i1 true to i32 ; yields i32:1
5757 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5759 '``sext .. to``' Instruction
5760 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5767 <result> = sext <ty> <value> to <ty2> ; yields ty2
5772 The '``sext``' sign extends ``value`` to the type ``ty2``.
5777 The '``sext``' instruction takes a value to cast, and a type to cast it
5778 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
5779 the same number of integers. The bit size of the ``value`` must be
5780 smaller than the bit size of the destination type, ``ty2``.
5785 The '``sext``' instruction performs a sign extension by copying the sign
5786 bit (highest order bit) of the ``value`` until it reaches the bit size
5787 of the type ``ty2``.
5789 When sign extending from i1, the extension always results in -1 or 0.
5794 .. code-block:: llvm
5796 %X = sext i8 -1 to i16 ; yields i16 :65535
5797 %Y = sext i1 true to i32 ; yields i32:-1
5798 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
5800 '``fptrunc .. to``' Instruction
5801 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5808 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
5813 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
5818 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
5819 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
5820 The size of ``value`` must be larger than the size of ``ty2``. This
5821 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
5826 The '``fptrunc``' instruction truncates a ``value`` from a larger
5827 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
5828 point <t_floating>` type. If the value cannot fit within the
5829 destination type, ``ty2``, then the results are undefined.
5834 .. code-block:: llvm
5836 %X = fptrunc double 123.0 to float ; yields float:123.0
5837 %Y = fptrunc double 1.0E+300 to float ; yields undefined
5839 '``fpext .. to``' Instruction
5840 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5847 <result> = fpext <ty> <value> to <ty2> ; yields ty2
5852 The '``fpext``' extends a floating point ``value`` to a larger floating
5858 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
5859 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
5860 to. The source type must be smaller than the destination type.
5865 The '``fpext``' instruction extends the ``value`` from a smaller
5866 :ref:`floating point <t_floating>` type to a larger :ref:`floating
5867 point <t_floating>` type. The ``fpext`` cannot be used to make a
5868 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
5869 *no-op cast* for a floating point cast.
5874 .. code-block:: llvm
5876 %X = fpext float 3.125 to double ; yields double:3.125000e+00
5877 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
5879 '``fptoui .. to``' Instruction
5880 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5887 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
5892 The '``fptoui``' converts a floating point ``value`` to its unsigned
5893 integer equivalent of type ``ty2``.
5898 The '``fptoui``' instruction takes a value to cast, which must be a
5899 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5900 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5901 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5902 type with the same number of elements as ``ty``
5907 The '``fptoui``' instruction converts its :ref:`floating
5908 point <t_floating>` operand into the nearest (rounding towards zero)
5909 unsigned integer value. If the value cannot fit in ``ty2``, the results
5915 .. code-block:: llvm
5917 %X = fptoui double 123.0 to i32 ; yields i32:123
5918 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
5919 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
5921 '``fptosi .. to``' Instruction
5922 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5929 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
5934 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
5935 ``value`` to type ``ty2``.
5940 The '``fptosi``' instruction takes a value to cast, which must be a
5941 scalar or vector :ref:`floating point <t_floating>` value, and a type to
5942 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
5943 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
5944 type with the same number of elements as ``ty``
5949 The '``fptosi``' instruction converts its :ref:`floating
5950 point <t_floating>` operand into the nearest (rounding towards zero)
5951 signed integer value. If the value cannot fit in ``ty2``, the results
5957 .. code-block:: llvm
5959 %X = fptosi double -123.0 to i32 ; yields i32:-123
5960 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
5961 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
5963 '``uitofp .. to``' Instruction
5964 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5971 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
5976 The '``uitofp``' instruction regards ``value`` as an unsigned integer
5977 and converts that value to the ``ty2`` type.
5982 The '``uitofp``' instruction takes a value to cast, which must be a
5983 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
5984 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
5985 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
5986 type with the same number of elements as ``ty``
5991 The '``uitofp``' instruction interprets its operand as an unsigned
5992 integer quantity and converts it to the corresponding floating point
5993 value. If the value cannot fit in the floating point value, the results
5999 .. code-block:: llvm
6001 %X = uitofp i32 257 to float ; yields float:257.0
6002 %Y = uitofp i8 -1 to double ; yields double:255.0
6004 '``sitofp .. to``' Instruction
6005 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6012 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
6017 The '``sitofp``' instruction regards ``value`` as a signed integer and
6018 converts that value to the ``ty2`` type.
6023 The '``sitofp``' instruction takes a value to cast, which must be a
6024 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6025 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6026 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6027 type with the same number of elements as ``ty``
6032 The '``sitofp``' instruction interprets its operand as a signed integer
6033 quantity and converts it to the corresponding floating point value. If
6034 the value cannot fit in the floating point value, the results are
6040 .. code-block:: llvm
6042 %X = sitofp i32 257 to float ; yields float:257.0
6043 %Y = sitofp i8 -1 to double ; yields double:-1.0
6047 '``ptrtoint .. to``' Instruction
6048 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6055 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
6060 The '``ptrtoint``' instruction converts the pointer or a vector of
6061 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
6066 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
6067 a a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
6068 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
6069 a vector of integers type.
6074 The '``ptrtoint``' instruction converts ``value`` to integer type
6075 ``ty2`` by interpreting the pointer value as an integer and either
6076 truncating or zero extending that value to the size of the integer type.
6077 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
6078 ``value`` is larger than ``ty2`` then a truncation is done. If they are
6079 the same size, then nothing is done (*no-op cast*) other than a type
6085 .. code-block:: llvm
6087 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
6088 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
6089 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
6093 '``inttoptr .. to``' Instruction
6094 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6101 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
6106 The '``inttoptr``' instruction converts an integer ``value`` to a
6107 pointer type, ``ty2``.
6112 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
6113 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
6119 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
6120 applying either a zero extension or a truncation depending on the size
6121 of the integer ``value``. If ``value`` is larger than the size of a
6122 pointer then a truncation is done. If ``value`` is smaller than the size
6123 of a pointer then a zero extension is done. If they are the same size,
6124 nothing is done (*no-op cast*).
6129 .. code-block:: llvm
6131 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
6132 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
6133 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
6134 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
6138 '``bitcast .. to``' Instruction
6139 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6146 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
6151 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
6157 The '``bitcast``' instruction takes a value to cast, which must be a
6158 non-aggregate first class value, and a type to cast it to, which must
6159 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
6160 bit sizes of ``value`` and the destination type, ``ty2``, must be
6161 identical. If the source type is a pointer, the destination type must
6162 also be a pointer of the same size. This instruction supports bitwise
6163 conversion of vectors to integers and to vectors of other types (as
6164 long as they have the same size).
6169 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
6170 is always a *no-op cast* because no bits change with this
6171 conversion. The conversion is done as if the ``value`` had been stored
6172 to memory and read back as type ``ty2``. Pointer (or vector of
6173 pointers) types may only be converted to other pointer (or vector of
6174 pointers) types with the same address space through this instruction.
6175 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
6176 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
6181 .. code-block:: llvm
6183 %X = bitcast i8 255 to i8 ; yields i8 :-1
6184 %Y = bitcast i32* %x to sint* ; yields sint*:%x
6185 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
6186 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
6188 .. _i_addrspacecast:
6190 '``addrspacecast .. to``' Instruction
6191 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6198 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
6203 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
6204 address space ``n`` to type ``pty2`` in address space ``m``.
6209 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
6210 to cast and a pointer type to cast it to, which must have a different
6216 The '``addrspacecast``' instruction converts the pointer value
6217 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
6218 value modification, depending on the target and the address space
6219 pair. Pointer conversions within the same address space must be
6220 performed with the ``bitcast`` instruction. Note that if the address space
6221 conversion is legal then both result and operand refer to the same memory
6227 .. code-block:: llvm
6229 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
6230 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
6231 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
6238 The instructions in this category are the "miscellaneous" instructions,
6239 which defy better classification.
6243 '``icmp``' Instruction
6244 ^^^^^^^^^^^^^^^^^^^^^^
6251 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6256 The '``icmp``' instruction returns a boolean value or a vector of
6257 boolean values based on comparison of its two integer, integer vector,
6258 pointer, or pointer vector operands.
6263 The '``icmp``' instruction takes three operands. The first operand is
6264 the condition code indicating the kind of comparison to perform. It is
6265 not a value, just a keyword. The possible condition code are:
6268 #. ``ne``: not equal
6269 #. ``ugt``: unsigned greater than
6270 #. ``uge``: unsigned greater or equal
6271 #. ``ult``: unsigned less than
6272 #. ``ule``: unsigned less or equal
6273 #. ``sgt``: signed greater than
6274 #. ``sge``: signed greater or equal
6275 #. ``slt``: signed less than
6276 #. ``sle``: signed less or equal
6278 The remaining two arguments must be :ref:`integer <t_integer>` or
6279 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
6280 must also be identical types.
6285 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
6286 code given as ``cond``. The comparison performed always yields either an
6287 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
6289 #. ``eq``: yields ``true`` if the operands are equal, ``false``
6290 otherwise. No sign interpretation is necessary or performed.
6291 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
6292 otherwise. No sign interpretation is necessary or performed.
6293 #. ``ugt``: interprets the operands as unsigned values and yields
6294 ``true`` if ``op1`` is greater than ``op2``.
6295 #. ``uge``: interprets the operands as unsigned values and yields
6296 ``true`` if ``op1`` is greater than or equal to ``op2``.
6297 #. ``ult``: interprets the operands as unsigned values and yields
6298 ``true`` if ``op1`` is less than ``op2``.
6299 #. ``ule``: interprets the operands as unsigned values and yields
6300 ``true`` if ``op1`` is less than or equal to ``op2``.
6301 #. ``sgt``: interprets the operands as signed values and yields ``true``
6302 if ``op1`` is greater than ``op2``.
6303 #. ``sge``: interprets the operands as signed values and yields ``true``
6304 if ``op1`` is greater than or equal to ``op2``.
6305 #. ``slt``: interprets the operands as signed values and yields ``true``
6306 if ``op1`` is less than ``op2``.
6307 #. ``sle``: interprets the operands as signed values and yields ``true``
6308 if ``op1`` is less than or equal to ``op2``.
6310 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
6311 are compared as if they were integers.
6313 If the operands are integer vectors, then they are compared element by
6314 element. The result is an ``i1`` vector with the same number of elements
6315 as the values being compared. Otherwise, the result is an ``i1``.
6320 .. code-block:: llvm
6322 <result> = icmp eq i32 4, 5 ; yields: result=false
6323 <result> = icmp ne float* %X, %X ; yields: result=false
6324 <result> = icmp ult i16 4, 5 ; yields: result=true
6325 <result> = icmp sgt i16 4, 5 ; yields: result=false
6326 <result> = icmp ule i16 -4, 5 ; yields: result=false
6327 <result> = icmp sge i16 4, 5 ; yields: result=false
6329 Note that the code generator does not yet support vector types with the
6330 ``icmp`` instruction.
6334 '``fcmp``' Instruction
6335 ^^^^^^^^^^^^^^^^^^^^^^
6342 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6347 The '``fcmp``' instruction returns a boolean value or vector of boolean
6348 values based on comparison of its operands.
6350 If the operands are floating point scalars, then the result type is a
6351 boolean (:ref:`i1 <t_integer>`).
6353 If the operands are floating point vectors, then the result type is a
6354 vector of boolean with the same number of elements as the operands being
6360 The '``fcmp``' instruction takes three operands. The first operand is
6361 the condition code indicating the kind of comparison to perform. It is
6362 not a value, just a keyword. The possible condition code are:
6364 #. ``false``: no comparison, always returns false
6365 #. ``oeq``: ordered and equal
6366 #. ``ogt``: ordered and greater than
6367 #. ``oge``: ordered and greater than or equal
6368 #. ``olt``: ordered and less than
6369 #. ``ole``: ordered and less than or equal
6370 #. ``one``: ordered and not equal
6371 #. ``ord``: ordered (no nans)
6372 #. ``ueq``: unordered or equal
6373 #. ``ugt``: unordered or greater than
6374 #. ``uge``: unordered or greater than or equal
6375 #. ``ult``: unordered or less than
6376 #. ``ule``: unordered or less than or equal
6377 #. ``une``: unordered or not equal
6378 #. ``uno``: unordered (either nans)
6379 #. ``true``: no comparison, always returns true
6381 *Ordered* means that neither operand is a QNAN while *unordered* means
6382 that either operand may be a QNAN.
6384 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
6385 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
6386 type. They must have identical types.
6391 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6392 condition code given as ``cond``. If the operands are vectors, then the
6393 vectors are compared element by element. Each comparison performed
6394 always yields an :ref:`i1 <t_integer>` result, as follows:
6396 #. ``false``: always yields ``false``, regardless of operands.
6397 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6398 is equal to ``op2``.
6399 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6400 is greater than ``op2``.
6401 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6402 is greater than or equal to ``op2``.
6403 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6404 is less than ``op2``.
6405 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6406 is less than or equal to ``op2``.
6407 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6408 is not equal to ``op2``.
6409 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6410 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6412 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6413 greater than ``op2``.
6414 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6415 greater than or equal to ``op2``.
6416 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6418 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6419 less than or equal to ``op2``.
6420 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6421 not equal to ``op2``.
6422 #. ``uno``: yields ``true`` if either operand is a QNAN.
6423 #. ``true``: always yields ``true``, regardless of operands.
6428 .. code-block:: llvm
6430 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6431 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6432 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6433 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6435 Note that the code generator does not yet support vector types with the
6436 ``fcmp`` instruction.
6440 '``phi``' Instruction
6441 ^^^^^^^^^^^^^^^^^^^^^
6448 <result> = phi <ty> [ <val0>, <label0>], ...
6453 The '``phi``' instruction is used to implement the φ node in the SSA
6454 graph representing the function.
6459 The type of the incoming values is specified with the first type field.
6460 After this, the '``phi``' instruction takes a list of pairs as
6461 arguments, with one pair for each predecessor basic block of the current
6462 block. Only values of :ref:`first class <t_firstclass>` type may be used as
6463 the value arguments to the PHI node. Only labels may be used as the
6466 There must be no non-phi instructions between the start of a basic block
6467 and the PHI instructions: i.e. PHI instructions must be first in a basic
6470 For the purposes of the SSA form, the use of each incoming value is
6471 deemed to occur on the edge from the corresponding predecessor block to
6472 the current block (but after any definition of an '``invoke``'
6473 instruction's return value on the same edge).
6478 At runtime, the '``phi``' instruction logically takes on the value
6479 specified by the pair corresponding to the predecessor basic block that
6480 executed just prior to the current block.
6485 .. code-block:: llvm
6487 Loop: ; Infinite loop that counts from 0 on up...
6488 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6489 %nextindvar = add i32 %indvar, 1
6494 '``select``' Instruction
6495 ^^^^^^^^^^^^^^^^^^^^^^^^
6502 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
6504 selty is either i1 or {<N x i1>}
6509 The '``select``' instruction is used to choose one value based on a
6510 condition, without IR-level branching.
6515 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
6516 values indicating the condition, and two values of the same :ref:`first
6517 class <t_firstclass>` type. If the val1/val2 are vectors and the
6518 condition is a scalar, then entire vectors are selected, not individual
6524 If the condition is an i1 and it evaluates to 1, the instruction returns
6525 the first value argument; otherwise, it returns the second value
6528 If the condition is a vector of i1, then the value arguments must be
6529 vectors of the same size, and the selection is done element by element.
6534 .. code-block:: llvm
6536 %X = select i1 true, i8 17, i8 42 ; yields i8:17
6540 '``call``' Instruction
6541 ^^^^^^^^^^^^^^^^^^^^^^
6548 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
6553 The '``call``' instruction represents a simple function call.
6558 This instruction requires several arguments:
6560 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
6561 should perform tail call optimization. The ``tail`` marker is a hint that
6562 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
6563 means that the call must be tail call optimized in order for the program to
6564 be correct. The ``musttail`` marker provides these guarantees:
6566 #. The call will not cause unbounded stack growth if it is part of a
6567 recursive cycle in the call graph.
6568 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
6571 Both markers imply that the callee does not access allocas or varargs from
6572 the caller. Calls marked ``musttail`` must obey the following additional
6575 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
6576 or a pointer bitcast followed by a ret instruction.
6577 - The ret instruction must return the (possibly bitcasted) value
6578 produced by the call or void.
6579 - The caller and callee prototypes must match. Pointer types of
6580 parameters or return types may differ in pointee type, but not
6582 - The calling conventions of the caller and callee must match.
6583 - All ABI-impacting function attributes, such as sret, byval, inreg,
6584 returned, and inalloca, must match.
6586 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
6587 the following conditions are met:
6589 - Caller and callee both have the calling convention ``fastcc``.
6590 - The call is in tail position (ret immediately follows call and ret
6591 uses value of call or is void).
6592 - Option ``-tailcallopt`` is enabled, or
6593 ``llvm::GuaranteedTailCallOpt`` is ``true``.
6594 - `Platform-specific constraints are
6595 met. <CodeGenerator.html#tailcallopt>`_
6597 #. The optional "cconv" marker indicates which :ref:`calling
6598 convention <callingconv>` the call should use. If none is
6599 specified, the call defaults to using C calling conventions. The
6600 calling convention of the call must match the calling convention of
6601 the target function, or else the behavior is undefined.
6602 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
6603 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
6605 #. '``ty``': the type of the call instruction itself which is also the
6606 type of the return value. Functions that return no value are marked
6608 #. '``fnty``': shall be the signature of the pointer to function value
6609 being invoked. The argument types must match the types implied by
6610 this signature. This type can be omitted if the function is not
6611 varargs and if the function type does not return a pointer to a
6613 #. '``fnptrval``': An LLVM value containing a pointer to a function to
6614 be invoked. In most cases, this is a direct function invocation, but
6615 indirect ``call``'s are just as possible, calling an arbitrary pointer
6617 #. '``function args``': argument list whose types match the function
6618 signature argument types and parameter attributes. All arguments must
6619 be of :ref:`first class <t_firstclass>` type. If the function signature
6620 indicates the function accepts a variable number of arguments, the
6621 extra arguments can be specified.
6622 #. The optional :ref:`function attributes <fnattrs>` list. Only
6623 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
6624 attributes are valid here.
6629 The '``call``' instruction is used to cause control flow to transfer to
6630 a specified function, with its incoming arguments bound to the specified
6631 values. Upon a '``ret``' instruction in the called function, control
6632 flow continues with the instruction after the function call, and the
6633 return value of the function is bound to the result argument.
6638 .. code-block:: llvm
6640 %retval = call i32 @test(i32 %argc)
6641 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
6642 %X = tail call i32 @foo() ; yields i32
6643 %Y = tail call fastcc i32 @foo() ; yields i32
6644 call void %foo(i8 97 signext)
6646 %struct.A = type { i32, i8 }
6647 %r = call %struct.A @foo() ; yields { i32, i8 }
6648 %gr = extractvalue %struct.A %r, 0 ; yields i32
6649 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
6650 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
6651 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
6653 llvm treats calls to some functions with names and arguments that match
6654 the standard C99 library as being the C99 library functions, and may
6655 perform optimizations or generate code for them under that assumption.
6656 This is something we'd like to change in the future to provide better
6657 support for freestanding environments and non-C-based languages.
6661 '``va_arg``' Instruction
6662 ^^^^^^^^^^^^^^^^^^^^^^^^
6669 <resultval> = va_arg <va_list*> <arglist>, <argty>
6674 The '``va_arg``' instruction is used to access arguments passed through
6675 the "variable argument" area of a function call. It is used to implement
6676 the ``va_arg`` macro in C.
6681 This instruction takes a ``va_list*`` value and the type of the
6682 argument. It returns a value of the specified argument type and
6683 increments the ``va_list`` to point to the next argument. The actual
6684 type of ``va_list`` is target specific.
6689 The '``va_arg``' instruction loads an argument of the specified type
6690 from the specified ``va_list`` and causes the ``va_list`` to point to
6691 the next argument. For more information, see the variable argument
6692 handling :ref:`Intrinsic Functions <int_varargs>`.
6694 It is legal for this instruction to be called in a function which does
6695 not take a variable number of arguments, for example, the ``vfprintf``
6698 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
6699 function <intrinsics>` because it takes a type as an argument.
6704 See the :ref:`variable argument processing <int_varargs>` section.
6706 Note that the code generator does not yet fully support va\_arg on many
6707 targets. Also, it does not currently support va\_arg with aggregate
6708 types on any target.
6712 '``landingpad``' Instruction
6713 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6720 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6721 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6723 <clause> := catch <type> <value>
6724 <clause> := filter <array constant type> <array constant>
6729 The '``landingpad``' instruction is used by `LLVM's exception handling
6730 system <ExceptionHandling.html#overview>`_ to specify that a basic block
6731 is a landing pad --- one where the exception lands, and corresponds to the
6732 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
6733 defines values supplied by the personality function (``pers_fn``) upon
6734 re-entry to the function. The ``resultval`` has the type ``resultty``.
6739 This instruction takes a ``pers_fn`` value. This is the personality
6740 function associated with the unwinding mechanism. The optional
6741 ``cleanup`` flag indicates that the landing pad block is a cleanup.
6743 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
6744 contains the global variable representing the "type" that may be caught
6745 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
6746 clause takes an array constant as its argument. Use
6747 "``[0 x i8**] undef``" for a filter which cannot throw. The
6748 '``landingpad``' instruction must contain *at least* one ``clause`` or
6749 the ``cleanup`` flag.
6754 The '``landingpad``' instruction defines the values which are set by the
6755 personality function (``pers_fn``) upon re-entry to the function, and
6756 therefore the "result type" of the ``landingpad`` instruction. As with
6757 calling conventions, how the personality function results are
6758 represented in LLVM IR is target specific.
6760 The clauses are applied in order from top to bottom. If two
6761 ``landingpad`` instructions are merged together through inlining, the
6762 clauses from the calling function are appended to the list of clauses.
6763 When the call stack is being unwound due to an exception being thrown,
6764 the exception is compared against each ``clause`` in turn. If it doesn't
6765 match any of the clauses, and the ``cleanup`` flag is not set, then
6766 unwinding continues further up the call stack.
6768 The ``landingpad`` instruction has several restrictions:
6770 - A landing pad block is a basic block which is the unwind destination
6771 of an '``invoke``' instruction.
6772 - A landing pad block must have a '``landingpad``' instruction as its
6773 first non-PHI instruction.
6774 - There can be only one '``landingpad``' instruction within the landing
6776 - A basic block that is not a landing pad block may not include a
6777 '``landingpad``' instruction.
6778 - All '``landingpad``' instructions in a function must have the same
6779 personality function.
6784 .. code-block:: llvm
6786 ;; A landing pad which can catch an integer.
6787 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6789 ;; A landing pad that is a cleanup.
6790 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6792 ;; A landing pad which can catch an integer and can only throw a double.
6793 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6795 filter [1 x i8**] [@_ZTId]
6802 LLVM supports the notion of an "intrinsic function". These functions
6803 have well known names and semantics and are required to follow certain
6804 restrictions. Overall, these intrinsics represent an extension mechanism
6805 for the LLVM language that does not require changing all of the
6806 transformations in LLVM when adding to the language (or the bitcode
6807 reader/writer, the parser, etc...).
6809 Intrinsic function names must all start with an "``llvm.``" prefix. This
6810 prefix is reserved in LLVM for intrinsic names; thus, function names may
6811 not begin with this prefix. Intrinsic functions must always be external
6812 functions: you cannot define the body of intrinsic functions. Intrinsic
6813 functions may only be used in call or invoke instructions: it is illegal
6814 to take the address of an intrinsic function. Additionally, because
6815 intrinsic functions are part of the LLVM language, it is required if any
6816 are added that they be documented here.
6818 Some intrinsic functions can be overloaded, i.e., the intrinsic
6819 represents a family of functions that perform the same operation but on
6820 different data types. Because LLVM can represent over 8 million
6821 different integer types, overloading is used commonly to allow an
6822 intrinsic function to operate on any integer type. One or more of the
6823 argument types or the result type can be overloaded to accept any
6824 integer type. Argument types may also be defined as exactly matching a
6825 previous argument's type or the result type. This allows an intrinsic
6826 function which accepts multiple arguments, but needs all of them to be
6827 of the same type, to only be overloaded with respect to a single
6828 argument or the result.
6830 Overloaded intrinsics will have the names of its overloaded argument
6831 types encoded into its function name, each preceded by a period. Only
6832 those types which are overloaded result in a name suffix. Arguments
6833 whose type is matched against another type do not. For example, the
6834 ``llvm.ctpop`` function can take an integer of any width and returns an
6835 integer of exactly the same integer width. This leads to a family of
6836 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
6837 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
6838 overloaded, and only one type suffix is required. Because the argument's
6839 type is matched against the return type, it does not require its own
6842 To learn how to add an intrinsic function, please see the `Extending
6843 LLVM Guide <ExtendingLLVM.html>`_.
6847 Variable Argument Handling Intrinsics
6848 -------------------------------------
6850 Variable argument support is defined in LLVM with the
6851 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
6852 functions. These functions are related to the similarly named macros
6853 defined in the ``<stdarg.h>`` header file.
6855 All of these functions operate on arguments that use a target-specific
6856 value type "``va_list``". The LLVM assembly language reference manual
6857 does not define what this type is, so all transformations should be
6858 prepared to handle these functions regardless of the type used.
6860 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
6861 variable argument handling intrinsic functions are used.
6863 .. code-block:: llvm
6865 define i32 @test(i32 %X, ...) {
6866 ; Initialize variable argument processing
6868 %ap2 = bitcast i8** %ap to i8*
6869 call void @llvm.va_start(i8* %ap2)
6871 ; Read a single integer argument
6872 %tmp = va_arg i8** %ap, i32
6874 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6876 %aq2 = bitcast i8** %aq to i8*
6877 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6878 call void @llvm.va_end(i8* %aq2)
6880 ; Stop processing of arguments.
6881 call void @llvm.va_end(i8* %ap2)
6885 declare void @llvm.va_start(i8*)
6886 declare void @llvm.va_copy(i8*, i8*)
6887 declare void @llvm.va_end(i8*)
6891 '``llvm.va_start``' Intrinsic
6892 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6899 declare void @llvm.va_start(i8* <arglist>)
6904 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
6905 subsequent use by ``va_arg``.
6910 The argument is a pointer to a ``va_list`` element to initialize.
6915 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
6916 available in C. In a target-dependent way, it initializes the
6917 ``va_list`` element to which the argument points, so that the next call
6918 to ``va_arg`` will produce the first variable argument passed to the
6919 function. Unlike the C ``va_start`` macro, this intrinsic does not need
6920 to know the last argument of the function as the compiler can figure
6923 '``llvm.va_end``' Intrinsic
6924 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6931 declare void @llvm.va_end(i8* <arglist>)
6936 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
6937 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
6942 The argument is a pointer to a ``va_list`` to destroy.
6947 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
6948 available in C. In a target-dependent way, it destroys the ``va_list``
6949 element to which the argument points. Calls to
6950 :ref:`llvm.va_start <int_va_start>` and
6951 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
6956 '``llvm.va_copy``' Intrinsic
6957 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6964 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6969 The '``llvm.va_copy``' intrinsic copies the current argument position
6970 from the source argument list to the destination argument list.
6975 The first argument is a pointer to a ``va_list`` element to initialize.
6976 The second argument is a pointer to a ``va_list`` element to copy from.
6981 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
6982 available in C. In a target-dependent way, it copies the source
6983 ``va_list`` element into the destination ``va_list`` element. This
6984 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
6985 arbitrarily complex and require, for example, memory allocation.
6987 Accurate Garbage Collection Intrinsics
6988 --------------------------------------
6990 LLVM support for `Accurate Garbage Collection <GarbageCollection.html>`_
6991 (GC) requires the implementation and generation of these intrinsics.
6992 These intrinsics allow identification of :ref:`GC roots on the
6993 stack <int_gcroot>`, as well as garbage collector implementations that
6994 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
6995 Front-ends for type-safe garbage collected languages should generate
6996 these intrinsics to make use of the LLVM garbage collectors. For more
6997 details, see `Accurate Garbage Collection with
6998 LLVM <GarbageCollection.html>`_.
7000 The garbage collection intrinsics only operate on objects in the generic
7001 address space (address space zero).
7005 '``llvm.gcroot``' Intrinsic
7006 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7013 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
7018 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
7019 the code generator, and allows some metadata to be associated with it.
7024 The first argument specifies the address of a stack object that contains
7025 the root pointer. The second pointer (which must be either a constant or
7026 a global value address) contains the meta-data to be associated with the
7032 At runtime, a call to this intrinsic stores a null pointer into the
7033 "ptrloc" location. At compile-time, the code generator generates
7034 information to allow the runtime to find the pointer at GC safe points.
7035 The '``llvm.gcroot``' intrinsic may only be used in a function which
7036 :ref:`specifies a GC algorithm <gc>`.
7040 '``llvm.gcread``' Intrinsic
7041 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7048 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
7053 The '``llvm.gcread``' intrinsic identifies reads of references from heap
7054 locations, allowing garbage collector implementations that require read
7060 The second argument is the address to read from, which should be an
7061 address allocated from the garbage collector. The first object is a
7062 pointer to the start of the referenced object, if needed by the language
7063 runtime (otherwise null).
7068 The '``llvm.gcread``' intrinsic has the same semantics as a load
7069 instruction, but may be replaced with substantially more complex code by
7070 the garbage collector runtime, as needed. The '``llvm.gcread``'
7071 intrinsic may only be used in a function which :ref:`specifies a GC
7076 '``llvm.gcwrite``' Intrinsic
7077 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7084 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
7089 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
7090 locations, allowing garbage collector implementations that require write
7091 barriers (such as generational or reference counting collectors).
7096 The first argument is the reference to store, the second is the start of
7097 the object to store it to, and the third is the address of the field of
7098 Obj to store to. If the runtime does not require a pointer to the
7099 object, Obj may be null.
7104 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
7105 instruction, but may be replaced with substantially more complex code by
7106 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
7107 intrinsic may only be used in a function which :ref:`specifies a GC
7110 Code Generator Intrinsics
7111 -------------------------
7113 These intrinsics are provided by LLVM to expose special features that
7114 may only be implemented with code generator support.
7116 '``llvm.returnaddress``' Intrinsic
7117 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7124 declare i8 *@llvm.returnaddress(i32 <level>)
7129 The '``llvm.returnaddress``' intrinsic attempts to compute a
7130 target-specific value indicating the return address of the current
7131 function or one of its callers.
7136 The argument to this intrinsic indicates which function to return the
7137 address for. Zero indicates the calling function, one indicates its
7138 caller, etc. The argument is **required** to be a constant integer
7144 The '``llvm.returnaddress``' intrinsic either returns a pointer
7145 indicating the return address of the specified call frame, or zero if it
7146 cannot be identified. The value returned by this intrinsic is likely to
7147 be incorrect or 0 for arguments other than zero, so it should only be
7148 used for debugging purposes.
7150 Note that calling this intrinsic does not prevent function inlining or
7151 other aggressive transformations, so the value returned may not be that
7152 of the obvious source-language caller.
7154 '``llvm.frameaddress``' Intrinsic
7155 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7162 declare i8* @llvm.frameaddress(i32 <level>)
7167 The '``llvm.frameaddress``' intrinsic attempts to return the
7168 target-specific frame pointer value for the specified stack frame.
7173 The argument to this intrinsic indicates which function to return the
7174 frame pointer for. Zero indicates the calling function, one indicates
7175 its caller, etc. The argument is **required** to be a constant integer
7181 The '``llvm.frameaddress``' intrinsic either returns a pointer
7182 indicating the frame address of the specified call frame, or zero if it
7183 cannot be identified. The value returned by this intrinsic is likely to
7184 be incorrect or 0 for arguments other than zero, so it should only be
7185 used for debugging purposes.
7187 Note that calling this intrinsic does not prevent function inlining or
7188 other aggressive transformations, so the value returned may not be that
7189 of the obvious source-language caller.
7191 .. _int_read_register:
7192 .. _int_write_register:
7194 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
7195 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7202 declare i32 @llvm.read_register.i32(metadata)
7203 declare i64 @llvm.read_register.i64(metadata)
7204 declare void @llvm.write_register.i32(metadata, i32 @value)
7205 declare void @llvm.write_register.i64(metadata, i64 @value)
7206 !0 = metadata !{metadata !"sp\00"}
7211 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
7212 provides access to the named register. The register must be valid on
7213 the architecture being compiled to. The type needs to be compatible
7214 with the register being read.
7219 The '``llvm.read_register``' intrinsic returns the current value of the
7220 register, where possible. The '``llvm.write_register``' intrinsic sets
7221 the current value of the register, where possible.
7223 This is useful to implement named register global variables that need
7224 to always be mapped to a specific register, as is common practice on
7225 bare-metal programs including OS kernels.
7227 The compiler doesn't check for register availability or use of the used
7228 register in surrounding code, including inline assembly. Because of that,
7229 allocatable registers are not supported.
7231 Warning: So far it only works with the stack pointer on selected
7232 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
7233 work is needed to support other registers and even more so, allocatable
7238 '``llvm.stacksave``' Intrinsic
7239 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7246 declare i8* @llvm.stacksave()
7251 The '``llvm.stacksave``' intrinsic is used to remember the current state
7252 of the function stack, for use with
7253 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
7254 implementing language features like scoped automatic variable sized
7260 This intrinsic returns a opaque pointer value that can be passed to
7261 :ref:`llvm.stackrestore <int_stackrestore>`. When an
7262 ``llvm.stackrestore`` intrinsic is executed with a value saved from
7263 ``llvm.stacksave``, it effectively restores the state of the stack to
7264 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
7265 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
7266 were allocated after the ``llvm.stacksave`` was executed.
7268 .. _int_stackrestore:
7270 '``llvm.stackrestore``' Intrinsic
7271 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7278 declare void @llvm.stackrestore(i8* %ptr)
7283 The '``llvm.stackrestore``' intrinsic is used to restore the state of
7284 the function stack to the state it was in when the corresponding
7285 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
7286 useful for implementing language features like scoped automatic variable
7287 sized arrays in C99.
7292 See the description for :ref:`llvm.stacksave <int_stacksave>`.
7294 '``llvm.prefetch``' Intrinsic
7295 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7302 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
7307 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
7308 insert a prefetch instruction if supported; otherwise, it is a noop.
7309 Prefetches have no effect on the behavior of the program but can change
7310 its performance characteristics.
7315 ``address`` is the address to be prefetched, ``rw`` is the specifier
7316 determining if the fetch should be for a read (0) or write (1), and
7317 ``locality`` is a temporal locality specifier ranging from (0) - no
7318 locality, to (3) - extremely local keep in cache. The ``cache type``
7319 specifies whether the prefetch is performed on the data (1) or
7320 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
7321 arguments must be constant integers.
7326 This intrinsic does not modify the behavior of the program. In
7327 particular, prefetches cannot trap and do not produce a value. On
7328 targets that support this intrinsic, the prefetch can provide hints to
7329 the processor cache for better performance.
7331 '``llvm.pcmarker``' Intrinsic
7332 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7339 declare void @llvm.pcmarker(i32 <id>)
7344 The '``llvm.pcmarker``' intrinsic is a method to export a Program
7345 Counter (PC) in a region of code to simulators and other tools. The
7346 method is target specific, but it is expected that the marker will use
7347 exported symbols to transmit the PC of the marker. The marker makes no
7348 guarantees that it will remain with any specific instruction after
7349 optimizations. It is possible that the presence of a marker will inhibit
7350 optimizations. The intended use is to be inserted after optimizations to
7351 allow correlations of simulation runs.
7356 ``id`` is a numerical id identifying the marker.
7361 This intrinsic does not modify the behavior of the program. Backends
7362 that do not support this intrinsic may ignore it.
7364 '``llvm.readcyclecounter``' Intrinsic
7365 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7372 declare i64 @llvm.readcyclecounter()
7377 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
7378 counter register (or similar low latency, high accuracy clocks) on those
7379 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7380 should map to RPCC. As the backing counters overflow quickly (on the
7381 order of 9 seconds on alpha), this should only be used for small
7387 When directly supported, reading the cycle counter should not modify any
7388 memory. Implementations are allowed to either return a application
7389 specific value or a system wide value. On backends without support, this
7390 is lowered to a constant 0.
7392 Note that runtime support may be conditional on the privilege-level code is
7393 running at and the host platform.
7395 '``llvm.clear_cache``' Intrinsic
7396 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7403 declare void @llvm.clear_cache(i8*, i8*)
7408 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
7409 in the specified range to the execution unit of the processor. On
7410 targets with non-unified instruction and data cache, the implementation
7411 flushes the instruction cache.
7416 On platforms with coherent instruction and data caches (e.g. x86), this
7417 intrinsic is a nop. On platforms with non-coherent instruction and data
7418 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
7419 instructions or a system call, if cache flushing requires special
7422 The default behavior is to emit a call to ``__clear_cache`` from the run
7425 This instrinsic does *not* empty the instruction pipeline. Modifications
7426 of the current function are outside the scope of the intrinsic.
7428 Standard C Library Intrinsics
7429 -----------------------------
7431 LLVM provides intrinsics for a few important standard C library
7432 functions. These intrinsics allow source-language front-ends to pass
7433 information about the alignment of the pointer arguments to the code
7434 generator, providing opportunity for more efficient code generation.
7438 '``llvm.memcpy``' Intrinsic
7439 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7444 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
7445 integer bit width and for different address spaces. Not all targets
7446 support all bit widths however.
7450 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7451 i32 <len>, i32 <align>, i1 <isvolatile>)
7452 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7453 i64 <len>, i32 <align>, i1 <isvolatile>)
7458 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7459 source location to the destination location.
7461 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
7462 intrinsics do not return a value, takes extra alignment/isvolatile
7463 arguments and the pointers can be in specified address spaces.
7468 The first argument is a pointer to the destination, the second is a
7469 pointer to the source. The third argument is an integer argument
7470 specifying the number of bytes to copy, the fourth argument is the
7471 alignment of the source and destination locations, and the fifth is a
7472 boolean indicating a volatile access.
7474 If the call to this intrinsic has an alignment value that is not 0 or 1,
7475 then the caller guarantees that both the source and destination pointers
7476 are aligned to that boundary.
7478 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
7479 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7480 very cleanly specified and it is unwise to depend on it.
7485 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
7486 source location to the destination location, which are not allowed to
7487 overlap. It copies "len" bytes of memory over. If the argument is known
7488 to be aligned to some boundary, this can be specified as the fourth
7489 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
7491 '``llvm.memmove``' Intrinsic
7492 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7497 This is an overloaded intrinsic. You can use llvm.memmove on any integer
7498 bit width and for different address space. Not all targets support all
7503 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7504 i32 <len>, i32 <align>, i1 <isvolatile>)
7505 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7506 i64 <len>, i32 <align>, i1 <isvolatile>)
7511 The '``llvm.memmove.*``' intrinsics move a block of memory from the
7512 source location to the destination location. It is similar to the
7513 '``llvm.memcpy``' intrinsic but allows the two memory locations to
7516 Note that, unlike the standard libc function, the ``llvm.memmove.*``
7517 intrinsics do not return a value, takes extra alignment/isvolatile
7518 arguments and the pointers can be in specified address spaces.
7523 The first argument is a pointer to the destination, the second is a
7524 pointer to the source. The third argument is an integer argument
7525 specifying the number of bytes to copy, the fourth argument is the
7526 alignment of the source and destination locations, and the fifth is a
7527 boolean indicating a volatile access.
7529 If the call to this intrinsic has an alignment value that is not 0 or 1,
7530 then the caller guarantees that the source and destination pointers are
7531 aligned to that boundary.
7533 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
7534 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
7535 not very cleanly specified and it is unwise to depend on it.
7540 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
7541 source location to the destination location, which may overlap. It
7542 copies "len" bytes of memory over. If the argument is known to be
7543 aligned to some boundary, this can be specified as the fourth argument,
7544 otherwise it should be set to 0 or 1 (both meaning no alignment).
7546 '``llvm.memset.*``' Intrinsics
7547 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7552 This is an overloaded intrinsic. You can use llvm.memset on any integer
7553 bit width and for different address spaces. However, not all targets
7554 support all bit widths.
7558 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7559 i32 <len>, i32 <align>, i1 <isvolatile>)
7560 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7561 i64 <len>, i32 <align>, i1 <isvolatile>)
7566 The '``llvm.memset.*``' intrinsics fill a block of memory with a
7567 particular byte value.
7569 Note that, unlike the standard libc function, the ``llvm.memset``
7570 intrinsic does not return a value and takes extra alignment/volatile
7571 arguments. Also, the destination can be in an arbitrary address space.
7576 The first argument is a pointer to the destination to fill, the second
7577 is the byte value with which to fill it, the third argument is an
7578 integer argument specifying the number of bytes to fill, and the fourth
7579 argument is the known alignment of the destination location.
7581 If the call to this intrinsic has an alignment value that is not 0 or 1,
7582 then the caller guarantees that the destination pointer is aligned to
7585 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
7586 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
7587 very cleanly specified and it is unwise to depend on it.
7592 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
7593 at the destination location. If the argument is known to be aligned to
7594 some boundary, this can be specified as the fourth argument, otherwise
7595 it should be set to 0 or 1 (both meaning no alignment).
7597 '``llvm.sqrt.*``' Intrinsic
7598 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7603 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
7604 floating point or vector of floating point type. Not all targets support
7609 declare float @llvm.sqrt.f32(float %Val)
7610 declare double @llvm.sqrt.f64(double %Val)
7611 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7612 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7613 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7618 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
7619 returning the same value as the libm '``sqrt``' functions would. Unlike
7620 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
7621 negative numbers other than -0.0 (which allows for better optimization,
7622 because there is no need to worry about errno being set).
7623 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
7628 The argument and return value are floating point numbers of the same
7634 This function returns the sqrt of the specified operand if it is a
7635 nonnegative floating point number.
7637 '``llvm.powi.*``' Intrinsic
7638 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7643 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
7644 floating point or vector of floating point type. Not all targets support
7649 declare float @llvm.powi.f32(float %Val, i32 %power)
7650 declare double @llvm.powi.f64(double %Val, i32 %power)
7651 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7652 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7653 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7658 The '``llvm.powi.*``' intrinsics return the first operand raised to the
7659 specified (positive or negative) power. The order of evaluation of
7660 multiplications is not defined. When a vector of floating point type is
7661 used, the second argument remains a scalar integer value.
7666 The second argument is an integer power, and the first is a value to
7667 raise to that power.
7672 This function returns the first value raised to the second power with an
7673 unspecified sequence of rounding operations.
7675 '``llvm.sin.*``' Intrinsic
7676 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7681 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
7682 floating point or vector of floating point type. Not all targets support
7687 declare float @llvm.sin.f32(float %Val)
7688 declare double @llvm.sin.f64(double %Val)
7689 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7690 declare fp128 @llvm.sin.f128(fp128 %Val)
7691 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7696 The '``llvm.sin.*``' intrinsics return the sine of the operand.
7701 The argument and return value are floating point numbers of the same
7707 This function returns the sine of the specified operand, returning the
7708 same values as the libm ``sin`` functions would, and handles error
7709 conditions in the same way.
7711 '``llvm.cos.*``' Intrinsic
7712 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7717 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
7718 floating point or vector of floating point type. Not all targets support
7723 declare float @llvm.cos.f32(float %Val)
7724 declare double @llvm.cos.f64(double %Val)
7725 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7726 declare fp128 @llvm.cos.f128(fp128 %Val)
7727 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7732 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
7737 The argument and return value are floating point numbers of the same
7743 This function returns the cosine of the specified operand, returning the
7744 same values as the libm ``cos`` functions would, and handles error
7745 conditions in the same way.
7747 '``llvm.pow.*``' Intrinsic
7748 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7753 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
7754 floating point or vector of floating point type. Not all targets support
7759 declare float @llvm.pow.f32(float %Val, float %Power)
7760 declare double @llvm.pow.f64(double %Val, double %Power)
7761 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7762 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7763 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7768 The '``llvm.pow.*``' intrinsics return the first operand raised to the
7769 specified (positive or negative) power.
7774 The second argument is a floating point power, and the first is a value
7775 to raise to that power.
7780 This function returns the first value raised to the second power,
7781 returning the same values as the libm ``pow`` functions would, and
7782 handles error conditions in the same way.
7784 '``llvm.exp.*``' Intrinsic
7785 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7790 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
7791 floating point or vector of floating point type. Not all targets support
7796 declare float @llvm.exp.f32(float %Val)
7797 declare double @llvm.exp.f64(double %Val)
7798 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7799 declare fp128 @llvm.exp.f128(fp128 %Val)
7800 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7805 The '``llvm.exp.*``' intrinsics perform the exp function.
7810 The argument and return value are floating point numbers of the same
7816 This function returns the same values as the libm ``exp`` functions
7817 would, and handles error conditions in the same way.
7819 '``llvm.exp2.*``' Intrinsic
7820 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7825 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
7826 floating point or vector of floating point type. Not all targets support
7831 declare float @llvm.exp2.f32(float %Val)
7832 declare double @llvm.exp2.f64(double %Val)
7833 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7834 declare fp128 @llvm.exp2.f128(fp128 %Val)
7835 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7840 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
7845 The argument and return value are floating point numbers of the same
7851 This function returns the same values as the libm ``exp2`` functions
7852 would, and handles error conditions in the same way.
7854 '``llvm.log.*``' Intrinsic
7855 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7860 This is an overloaded intrinsic. You can use ``llvm.log`` on any
7861 floating point or vector of floating point type. Not all targets support
7866 declare float @llvm.log.f32(float %Val)
7867 declare double @llvm.log.f64(double %Val)
7868 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7869 declare fp128 @llvm.log.f128(fp128 %Val)
7870 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7875 The '``llvm.log.*``' intrinsics perform the log function.
7880 The argument and return value are floating point numbers of the same
7886 This function returns the same values as the libm ``log`` functions
7887 would, and handles error conditions in the same way.
7889 '``llvm.log10.*``' Intrinsic
7890 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7895 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
7896 floating point or vector of floating point type. Not all targets support
7901 declare float @llvm.log10.f32(float %Val)
7902 declare double @llvm.log10.f64(double %Val)
7903 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7904 declare fp128 @llvm.log10.f128(fp128 %Val)
7905 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7910 The '``llvm.log10.*``' intrinsics perform the log10 function.
7915 The argument and return value are floating point numbers of the same
7921 This function returns the same values as the libm ``log10`` functions
7922 would, and handles error conditions in the same way.
7924 '``llvm.log2.*``' Intrinsic
7925 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7930 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
7931 floating point or vector of floating point type. Not all targets support
7936 declare float @llvm.log2.f32(float %Val)
7937 declare double @llvm.log2.f64(double %Val)
7938 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7939 declare fp128 @llvm.log2.f128(fp128 %Val)
7940 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7945 The '``llvm.log2.*``' intrinsics perform the log2 function.
7950 The argument and return value are floating point numbers of the same
7956 This function returns the same values as the libm ``log2`` functions
7957 would, and handles error conditions in the same way.
7959 '``llvm.fma.*``' Intrinsic
7960 ^^^^^^^^^^^^^^^^^^^^^^^^^^
7965 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
7966 floating point or vector of floating point type. Not all targets support
7971 declare float @llvm.fma.f32(float %a, float %b, float %c)
7972 declare double @llvm.fma.f64(double %a, double %b, double %c)
7973 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7974 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7975 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7980 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
7986 The argument and return value are floating point numbers of the same
7992 This function returns the same values as the libm ``fma`` functions
7993 would, and does not set errno.
7995 '``llvm.fabs.*``' Intrinsic
7996 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8001 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
8002 floating point or vector of floating point type. Not all targets support
8007 declare float @llvm.fabs.f32(float %Val)
8008 declare double @llvm.fabs.f64(double %Val)
8009 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
8010 declare fp128 @llvm.fabs.f128(fp128 %Val)
8011 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
8016 The '``llvm.fabs.*``' intrinsics return the absolute value of the
8022 The argument and return value are floating point numbers of the same
8028 This function returns the same values as the libm ``fabs`` functions
8029 would, and handles error conditions in the same way.
8031 '``llvm.copysign.*``' Intrinsic
8032 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8037 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
8038 floating point or vector of floating point type. Not all targets support
8043 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
8044 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
8045 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
8046 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
8047 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
8052 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
8053 first operand and the sign of the second operand.
8058 The arguments and return value are floating point numbers of the same
8064 This function returns the same values as the libm ``copysign``
8065 functions would, and handles error conditions in the same way.
8067 '``llvm.floor.*``' Intrinsic
8068 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8073 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
8074 floating point or vector of floating point type. Not all targets support
8079 declare float @llvm.floor.f32(float %Val)
8080 declare double @llvm.floor.f64(double %Val)
8081 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
8082 declare fp128 @llvm.floor.f128(fp128 %Val)
8083 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
8088 The '``llvm.floor.*``' intrinsics return the floor of the operand.
8093 The argument and return value are floating point numbers of the same
8099 This function returns the same values as the libm ``floor`` functions
8100 would, and handles error conditions in the same way.
8102 '``llvm.ceil.*``' Intrinsic
8103 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8108 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
8109 floating point or vector of floating point type. Not all targets support
8114 declare float @llvm.ceil.f32(float %Val)
8115 declare double @llvm.ceil.f64(double %Val)
8116 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
8117 declare fp128 @llvm.ceil.f128(fp128 %Val)
8118 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
8123 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
8128 The argument and return value are floating point numbers of the same
8134 This function returns the same values as the libm ``ceil`` functions
8135 would, and handles error conditions in the same way.
8137 '``llvm.trunc.*``' Intrinsic
8138 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8143 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
8144 floating point or vector of floating point type. Not all targets support
8149 declare float @llvm.trunc.f32(float %Val)
8150 declare double @llvm.trunc.f64(double %Val)
8151 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
8152 declare fp128 @llvm.trunc.f128(fp128 %Val)
8153 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
8158 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
8159 nearest integer not larger in magnitude than the operand.
8164 The argument and return value are floating point numbers of the same
8170 This function returns the same values as the libm ``trunc`` functions
8171 would, and handles error conditions in the same way.
8173 '``llvm.rint.*``' Intrinsic
8174 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8179 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
8180 floating point or vector of floating point type. Not all targets support
8185 declare float @llvm.rint.f32(float %Val)
8186 declare double @llvm.rint.f64(double %Val)
8187 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
8188 declare fp128 @llvm.rint.f128(fp128 %Val)
8189 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
8194 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
8195 nearest integer. It may raise an inexact floating-point exception if the
8196 operand isn't an integer.
8201 The argument and return value are floating point numbers of the same
8207 This function returns the same values as the libm ``rint`` functions
8208 would, and handles error conditions in the same way.
8210 '``llvm.nearbyint.*``' Intrinsic
8211 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8216 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
8217 floating point or vector of floating point type. Not all targets support
8222 declare float @llvm.nearbyint.f32(float %Val)
8223 declare double @llvm.nearbyint.f64(double %Val)
8224 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
8225 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
8226 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
8231 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
8237 The argument and return value are floating point numbers of the same
8243 This function returns the same values as the libm ``nearbyint``
8244 functions would, and handles error conditions in the same way.
8246 '``llvm.round.*``' Intrinsic
8247 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8252 This is an overloaded intrinsic. You can use ``llvm.round`` on any
8253 floating point or vector of floating point type. Not all targets support
8258 declare float @llvm.round.f32(float %Val)
8259 declare double @llvm.round.f64(double %Val)
8260 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
8261 declare fp128 @llvm.round.f128(fp128 %Val)
8262 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
8267 The '``llvm.round.*``' intrinsics returns the operand rounded to the
8273 The argument and return value are floating point numbers of the same
8279 This function returns the same values as the libm ``round``
8280 functions would, and handles error conditions in the same way.
8282 Bit Manipulation Intrinsics
8283 ---------------------------
8285 LLVM provides intrinsics for a few important bit manipulation
8286 operations. These allow efficient code generation for some algorithms.
8288 '``llvm.bswap.*``' Intrinsics
8289 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8294 This is an overloaded intrinsic function. You can use bswap on any
8295 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
8299 declare i16 @llvm.bswap.i16(i16 <id>)
8300 declare i32 @llvm.bswap.i32(i32 <id>)
8301 declare i64 @llvm.bswap.i64(i64 <id>)
8306 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
8307 values with an even number of bytes (positive multiple of 16 bits).
8308 These are useful for performing operations on data that is not in the
8309 target's native byte order.
8314 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
8315 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
8316 intrinsic returns an i32 value that has the four bytes of the input i32
8317 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
8318 returned i32 will have its bytes in 3, 2, 1, 0 order. The
8319 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
8320 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
8323 '``llvm.ctpop.*``' Intrinsic
8324 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8329 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
8330 bit width, or on any vector with integer elements. Not all targets
8331 support all bit widths or vector types, however.
8335 declare i8 @llvm.ctpop.i8(i8 <src>)
8336 declare i16 @llvm.ctpop.i16(i16 <src>)
8337 declare i32 @llvm.ctpop.i32(i32 <src>)
8338 declare i64 @llvm.ctpop.i64(i64 <src>)
8339 declare i256 @llvm.ctpop.i256(i256 <src>)
8340 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
8345 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
8351 The only argument is the value to be counted. The argument may be of any
8352 integer type, or a vector with integer elements. The return type must
8353 match the argument type.
8358 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
8359 each element of a vector.
8361 '``llvm.ctlz.*``' Intrinsic
8362 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8367 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
8368 integer bit width, or any vector whose elements are integers. Not all
8369 targets support all bit widths or vector types, however.
8373 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
8374 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
8375 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
8376 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
8377 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
8378 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8383 The '``llvm.ctlz``' family of intrinsic functions counts the number of
8384 leading zeros in a variable.
8389 The first argument is the value to be counted. This argument may be of
8390 any integer type, or a vectory with integer element type. The return
8391 type must match the first argument type.
8393 The second argument must be a constant and is a flag to indicate whether
8394 the intrinsic should ensure that a zero as the first argument produces a
8395 defined result. Historically some architectures did not provide a
8396 defined result for zero values as efficiently, and many algorithms are
8397 now predicated on avoiding zero-value inputs.
8402 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
8403 zeros in a variable, or within each element of the vector. If
8404 ``src == 0`` then the result is the size in bits of the type of ``src``
8405 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8406 ``llvm.ctlz(i32 2) = 30``.
8408 '``llvm.cttz.*``' Intrinsic
8409 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8414 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
8415 integer bit width, or any vector of integer elements. Not all targets
8416 support all bit widths or vector types, however.
8420 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
8421 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
8422 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
8423 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
8424 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
8425 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8430 The '``llvm.cttz``' family of intrinsic functions counts the number of
8436 The first argument is the value to be counted. This argument may be of
8437 any integer type, or a vectory with integer element type. The return
8438 type must match the first argument type.
8440 The second argument must be a constant and is a flag to indicate whether
8441 the intrinsic should ensure that a zero as the first argument produces a
8442 defined result. Historically some architectures did not provide a
8443 defined result for zero values as efficiently, and many algorithms are
8444 now predicated on avoiding zero-value inputs.
8449 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
8450 zeros in a variable, or within each element of a vector. If ``src == 0``
8451 then the result is the size in bits of the type of ``src`` if
8452 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
8453 ``llvm.cttz(2) = 1``.
8455 Arithmetic with Overflow Intrinsics
8456 -----------------------------------
8458 LLVM provides intrinsics for some arithmetic with overflow operations.
8460 '``llvm.sadd.with.overflow.*``' Intrinsics
8461 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8466 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
8467 on any integer bit width.
8471 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
8472 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8473 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
8478 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8479 a signed addition of the two arguments, and indicate whether an overflow
8480 occurred during the signed summation.
8485 The arguments (%a and %b) and the first element of the result structure
8486 may be of integer types of any bit width, but they must have the same
8487 bit width. The second element of the result structure must be of type
8488 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8494 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
8495 a signed addition of the two variables. They return a structure --- the
8496 first element of which is the signed summation, and the second element
8497 of which is a bit specifying if the signed summation resulted in an
8503 .. code-block:: llvm
8505 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8506 %sum = extractvalue {i32, i1} %res, 0
8507 %obit = extractvalue {i32, i1} %res, 1
8508 br i1 %obit, label %overflow, label %normal
8510 '``llvm.uadd.with.overflow.*``' Intrinsics
8511 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8516 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
8517 on any integer bit width.
8521 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8522 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8523 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8528 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8529 an unsigned addition of the two arguments, and indicate whether a carry
8530 occurred during the unsigned summation.
8535 The arguments (%a and %b) and the first element of the result structure
8536 may be of integer types of any bit width, but they must have the same
8537 bit width. The second element of the result structure must be of type
8538 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8544 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
8545 an unsigned addition of the two arguments. They return a structure --- the
8546 first element of which is the sum, and the second element of which is a
8547 bit specifying if the unsigned summation resulted in a carry.
8552 .. code-block:: llvm
8554 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8555 %sum = extractvalue {i32, i1} %res, 0
8556 %obit = extractvalue {i32, i1} %res, 1
8557 br i1 %obit, label %carry, label %normal
8559 '``llvm.ssub.with.overflow.*``' Intrinsics
8560 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8565 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
8566 on any integer bit width.
8570 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8571 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8572 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8577 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8578 a signed subtraction of the two arguments, and indicate whether an
8579 overflow occurred during the signed subtraction.
8584 The arguments (%a and %b) and the first element of the result structure
8585 may be of integer types of any bit width, but they must have the same
8586 bit width. The second element of the result structure must be of type
8587 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8593 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
8594 a signed subtraction of the two arguments. They return a structure --- the
8595 first element of which is the subtraction, and the second element of
8596 which is a bit specifying if the signed subtraction resulted in an
8602 .. code-block:: llvm
8604 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8605 %sum = extractvalue {i32, i1} %res, 0
8606 %obit = extractvalue {i32, i1} %res, 1
8607 br i1 %obit, label %overflow, label %normal
8609 '``llvm.usub.with.overflow.*``' Intrinsics
8610 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8615 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
8616 on any integer bit width.
8620 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8621 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8622 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8627 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8628 an unsigned subtraction of the two arguments, and indicate whether an
8629 overflow occurred during the unsigned subtraction.
8634 The arguments (%a and %b) and the first element of the result structure
8635 may be of integer types of any bit width, but they must have the same
8636 bit width. The second element of the result structure must be of type
8637 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8643 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
8644 an unsigned subtraction of the two arguments. They return a structure ---
8645 the first element of which is the subtraction, and the second element of
8646 which is a bit specifying if the unsigned subtraction resulted in an
8652 .. code-block:: llvm
8654 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8655 %sum = extractvalue {i32, i1} %res, 0
8656 %obit = extractvalue {i32, i1} %res, 1
8657 br i1 %obit, label %overflow, label %normal
8659 '``llvm.smul.with.overflow.*``' Intrinsics
8660 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8665 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
8666 on any integer bit width.
8670 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8671 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8672 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8677 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8678 a signed multiplication of the two arguments, and indicate whether an
8679 overflow occurred during the signed multiplication.
8684 The arguments (%a and %b) and the first element of the result structure
8685 may be of integer types of any bit width, but they must have the same
8686 bit width. The second element of the result structure must be of type
8687 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
8693 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
8694 a signed multiplication of the two arguments. They return a structure ---
8695 the first element of which is the multiplication, and the second element
8696 of which is a bit specifying if the signed multiplication resulted in an
8702 .. code-block:: llvm
8704 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8705 %sum = extractvalue {i32, i1} %res, 0
8706 %obit = extractvalue {i32, i1} %res, 1
8707 br i1 %obit, label %overflow, label %normal
8709 '``llvm.umul.with.overflow.*``' Intrinsics
8710 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8715 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
8716 on any integer bit width.
8720 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8721 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8722 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8727 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8728 a unsigned multiplication of the two arguments, and indicate whether an
8729 overflow occurred during the unsigned multiplication.
8734 The arguments (%a and %b) and the first element of the result structure
8735 may be of integer types of any bit width, but they must have the same
8736 bit width. The second element of the result structure must be of type
8737 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
8743 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
8744 an unsigned multiplication of the two arguments. They return a structure ---
8745 the first element of which is the multiplication, and the second
8746 element of which is a bit specifying if the unsigned multiplication
8747 resulted in an overflow.
8752 .. code-block:: llvm
8754 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8755 %sum = extractvalue {i32, i1} %res, 0
8756 %obit = extractvalue {i32, i1} %res, 1
8757 br i1 %obit, label %overflow, label %normal
8759 Specialised Arithmetic Intrinsics
8760 ---------------------------------
8762 '``llvm.fmuladd.*``' Intrinsic
8763 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8770 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8771 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8776 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
8777 expressions that can be fused if the code generator determines that (a) the
8778 target instruction set has support for a fused operation, and (b) that the
8779 fused operation is more efficient than the equivalent, separate pair of mul
8780 and add instructions.
8785 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
8786 multiplicands, a and b, and an addend c.
8795 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8797 is equivalent to the expression a \* b + c, except that rounding will
8798 not be performed between the multiplication and addition steps if the
8799 code generator fuses the operations. Fusion is not guaranteed, even if
8800 the target platform supports it. If a fused multiply-add is required the
8801 corresponding llvm.fma.\* intrinsic function should be used
8802 instead. This never sets errno, just as '``llvm.fma.*``'.
8807 .. code-block:: llvm
8809 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
8811 Half Precision Floating Point Intrinsics
8812 ----------------------------------------
8814 For most target platforms, half precision floating point is a
8815 storage-only format. This means that it is a dense encoding (in memory)
8816 but does not support computation in the format.
8818 This means that code must first load the half-precision floating point
8819 value as an i16, then convert it to float with
8820 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
8821 then be performed on the float value (including extending to double
8822 etc). To store the value back to memory, it is first converted to float
8823 if needed, then converted to i16 with
8824 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
8827 .. _int_convert_to_fp16:
8829 '``llvm.convert.to.fp16``' Intrinsic
8830 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8837 declare i16 @llvm.convert.to.fp16.f32(float %a)
8838 declare i16 @llvm.convert.to.fp16.f64(double %a)
8843 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
8844 conventional floating point type to half precision floating point format.
8849 The intrinsic function contains single argument - the value to be
8855 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
8856 conventional floating point format to half precision floating point format. The
8857 return value is an ``i16`` which contains the converted number.
8862 .. code-block:: llvm
8864 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
8865 store i16 %res, i16* @x, align 2
8867 .. _int_convert_from_fp16:
8869 '``llvm.convert.from.fp16``' Intrinsic
8870 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8877 declare float @llvm.convert.from.fp16.f32(i16 %a)
8878 declare double @llvm.convert.from.fp16.f64(i16 %a)
8883 The '``llvm.convert.from.fp16``' intrinsic function performs a
8884 conversion from half precision floating point format to single precision
8885 floating point format.
8890 The intrinsic function contains single argument - the value to be
8896 The '``llvm.convert.from.fp16``' intrinsic function performs a
8897 conversion from half single precision floating point format to single
8898 precision floating point format. The input half-float value is
8899 represented by an ``i16`` value.
8904 .. code-block:: llvm
8906 %a = load i16* @x, align 2
8907 %res = call float @llvm.convert.from.fp16(i16 %a)
8912 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
8913 prefix), are described in the `LLVM Source Level
8914 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
8917 Exception Handling Intrinsics
8918 -----------------------------
8920 The LLVM exception handling intrinsics (which all start with
8921 ``llvm.eh.`` prefix), are described in the `LLVM Exception
8922 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
8926 Trampoline Intrinsics
8927 ---------------------
8929 These intrinsics make it possible to excise one parameter, marked with
8930 the :ref:`nest <nest>` attribute, from a function. The result is a
8931 callable function pointer lacking the nest parameter - the caller does
8932 not need to provide a value for it. Instead, the value to use is stored
8933 in advance in a "trampoline", a block of memory usually allocated on the
8934 stack, which also contains code to splice the nest value into the
8935 argument list. This is used to implement the GCC nested function address
8938 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
8939 then the resulting function pointer has signature ``i32 (i32, i32)*``.
8940 It can be created as follows:
8942 .. code-block:: llvm
8944 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8945 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8946 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8947 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8948 %fp = bitcast i8* %p to i32 (i32, i32)*
8950 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
8951 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
8955 '``llvm.init.trampoline``' Intrinsic
8956 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8963 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8968 This fills the memory pointed to by ``tramp`` with executable code,
8969 turning it into a trampoline.
8974 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
8975 pointers. The ``tramp`` argument must point to a sufficiently large and
8976 sufficiently aligned block of memory; this memory is written to by the
8977 intrinsic. Note that the size and the alignment are target-specific -
8978 LLVM currently provides no portable way of determining them, so a
8979 front-end that generates this intrinsic needs to have some
8980 target-specific knowledge. The ``func`` argument must hold a function
8981 bitcast to an ``i8*``.
8986 The block of memory pointed to by ``tramp`` is filled with target
8987 dependent code, turning it into a function. Then ``tramp`` needs to be
8988 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
8989 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
8990 function's signature is the same as that of ``func`` with any arguments
8991 marked with the ``nest`` attribute removed. At most one such ``nest``
8992 argument is allowed, and it must be of pointer type. Calling the new
8993 function is equivalent to calling ``func`` with the same argument list,
8994 but with ``nval`` used for the missing ``nest`` argument. If, after
8995 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
8996 modified, then the effect of any later call to the returned function
8997 pointer is undefined.
9001 '``llvm.adjust.trampoline``' Intrinsic
9002 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9009 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
9014 This performs any required machine-specific adjustment to the address of
9015 a trampoline (passed as ``tramp``).
9020 ``tramp`` must point to a block of memory which already has trampoline
9021 code filled in by a previous call to
9022 :ref:`llvm.init.trampoline <int_it>`.
9027 On some architectures the address of the code to be executed needs to be
9028 different than the address where the trampoline is actually stored. This
9029 intrinsic returns the executable address corresponding to ``tramp``
9030 after performing the required machine specific adjustments. The pointer
9031 returned can then be :ref:`bitcast and executed <int_trampoline>`.
9036 This class of intrinsics provides information about the lifetime of
9037 memory objects and ranges where variables are immutable.
9041 '``llvm.lifetime.start``' Intrinsic
9042 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9049 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
9054 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
9060 The first argument is a constant integer representing the size of the
9061 object, or -1 if it is variable sized. The second argument is a pointer
9067 This intrinsic indicates that before this point in the code, the value
9068 of the memory pointed to by ``ptr`` is dead. This means that it is known
9069 to never be used and has an undefined value. A load from the pointer
9070 that precedes this intrinsic can be replaced with ``'undef'``.
9074 '``llvm.lifetime.end``' Intrinsic
9075 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9082 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
9087 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
9093 The first argument is a constant integer representing the size of the
9094 object, or -1 if it is variable sized. The second argument is a pointer
9100 This intrinsic indicates that after this point in the code, the value of
9101 the memory pointed to by ``ptr`` is dead. This means that it is known to
9102 never be used and has an undefined value. Any stores into the memory
9103 object following this intrinsic may be removed as dead.
9105 '``llvm.invariant.start``' Intrinsic
9106 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9113 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
9118 The '``llvm.invariant.start``' intrinsic specifies that the contents of
9119 a memory object will not change.
9124 The first argument is a constant integer representing the size of the
9125 object, or -1 if it is variable sized. The second argument is a pointer
9131 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
9132 the return value, the referenced memory location is constant and
9135 '``llvm.invariant.end``' Intrinsic
9136 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9143 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
9148 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
9149 memory object are mutable.
9154 The first argument is the matching ``llvm.invariant.start`` intrinsic.
9155 The second argument is a constant integer representing the size of the
9156 object, or -1 if it is variable sized and the third argument is a
9157 pointer to the object.
9162 This intrinsic indicates that the memory is mutable again.
9167 This class of intrinsics is designed to be generic and has no specific
9170 '``llvm.var.annotation``' Intrinsic
9171 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9178 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
9183 The '``llvm.var.annotation``' intrinsic.
9188 The first argument is a pointer to a value, the second is a pointer to a
9189 global string, the third is a pointer to a global string which is the
9190 source file name, and the last argument is the line number.
9195 This intrinsic allows annotation of local variables with arbitrary
9196 strings. This can be useful for special purpose optimizations that want
9197 to look for these annotations. These have no other defined use; they are
9198 ignored by code generation and optimization.
9200 '``llvm.ptr.annotation.*``' Intrinsic
9201 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9206 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
9207 pointer to an integer of any width. *NOTE* you must specify an address space for
9208 the pointer. The identifier for the default address space is the integer
9213 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
9214 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
9215 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
9216 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
9217 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
9222 The '``llvm.ptr.annotation``' intrinsic.
9227 The first argument is a pointer to an integer value of arbitrary bitwidth
9228 (result of some expression), the second is a pointer to a global string, the
9229 third is a pointer to a global string which is the source file name, and the
9230 last argument is the line number. It returns the value of the first argument.
9235 This intrinsic allows annotation of a pointer to an integer with arbitrary
9236 strings. This can be useful for special purpose optimizations that want to look
9237 for these annotations. These have no other defined use; they are ignored by code
9238 generation and optimization.
9240 '``llvm.annotation.*``' Intrinsic
9241 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9246 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
9247 any integer bit width.
9251 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
9252 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
9253 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
9254 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
9255 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
9260 The '``llvm.annotation``' intrinsic.
9265 The first argument is an integer value (result of some expression), the
9266 second is a pointer to a global string, the third is a pointer to a
9267 global string which is the source file name, and the last argument is
9268 the line number. It returns the value of the first argument.
9273 This intrinsic allows annotations to be put on arbitrary expressions
9274 with arbitrary strings. This can be useful for special purpose
9275 optimizations that want to look for these annotations. These have no
9276 other defined use; they are ignored by code generation and optimization.
9278 '``llvm.trap``' Intrinsic
9279 ^^^^^^^^^^^^^^^^^^^^^^^^^
9286 declare void @llvm.trap() noreturn nounwind
9291 The '``llvm.trap``' intrinsic.
9301 This intrinsic is lowered to the target dependent trap instruction. If
9302 the target does not have a trap instruction, this intrinsic will be
9303 lowered to a call of the ``abort()`` function.
9305 '``llvm.debugtrap``' Intrinsic
9306 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9313 declare void @llvm.debugtrap() nounwind
9318 The '``llvm.debugtrap``' intrinsic.
9328 This intrinsic is lowered to code which is intended to cause an
9329 execution trap with the intention of requesting the attention of a
9332 '``llvm.stackprotector``' Intrinsic
9333 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9340 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
9345 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
9346 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
9347 is placed on the stack before local variables.
9352 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
9353 The first argument is the value loaded from the stack guard
9354 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
9355 enough space to hold the value of the guard.
9360 This intrinsic causes the prologue/epilogue inserter to force the position of
9361 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
9362 to ensure that if a local variable on the stack is overwritten, it will destroy
9363 the value of the guard. When the function exits, the guard on the stack is
9364 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
9365 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
9366 calling the ``__stack_chk_fail()`` function.
9368 '``llvm.stackprotectorcheck``' Intrinsic
9369 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9376 declare void @llvm.stackprotectorcheck(i8** <guard>)
9381 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
9382 created stack protector and if they are not equal calls the
9383 ``__stack_chk_fail()`` function.
9388 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
9389 the variable ``@__stack_chk_guard``.
9394 This intrinsic is provided to perform the stack protector check by comparing
9395 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
9396 values do not match call the ``__stack_chk_fail()`` function.
9398 The reason to provide this as an IR level intrinsic instead of implementing it
9399 via other IR operations is that in order to perform this operation at the IR
9400 level without an intrinsic, one would need to create additional basic blocks to
9401 handle the success/failure cases. This makes it difficult to stop the stack
9402 protector check from disrupting sibling tail calls in Codegen. With this
9403 intrinsic, we are able to generate the stack protector basic blocks late in
9404 codegen after the tail call decision has occurred.
9406 '``llvm.objectsize``' Intrinsic
9407 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9414 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
9415 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
9420 The ``llvm.objectsize`` intrinsic is designed to provide information to
9421 the optimizers to determine at compile time whether a) an operation
9422 (like memcpy) will overflow a buffer that corresponds to an object, or
9423 b) that a runtime check for overflow isn't necessary. An object in this
9424 context means an allocation of a specific class, structure, array, or
9430 The ``llvm.objectsize`` intrinsic takes two arguments. The first
9431 argument is a pointer to or into the ``object``. The second argument is
9432 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
9433 or -1 (if false) when the object size is unknown. The second argument
9434 only accepts constants.
9439 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
9440 the size of the object concerned. If the size cannot be determined at
9441 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
9442 on the ``min`` argument).
9444 '``llvm.expect``' Intrinsic
9445 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9450 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
9455 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
9456 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
9457 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
9462 The ``llvm.expect`` intrinsic provides information about expected (the
9463 most probable) value of ``val``, which can be used by optimizers.
9468 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
9469 a value. The second argument is an expected value, this needs to be a
9470 constant value, variables are not allowed.
9475 This intrinsic is lowered to the ``val``.
9477 '``llvm.assume``' Intrinsic
9478 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9485 declare void @llvm.assume(i1 %cond)
9490 The ``llvm.assume`` allows the optimizer to assume that the provided
9491 condition is true. This information can then be used in simplifying other parts
9497 The condition which the optimizer may assume is always true.
9502 The intrinsic allows the optimizer to assume that the provided condition is
9503 always true whenever the control flow reaches the intrinsic call. No code is
9504 generated for this intrinsic, and instructions that contribute only to the
9505 provided condition are not used for code generation. If the condition is
9506 violated during execution, the behavior is undefined.
9508 Please note that optimizer might limit the transformations performed on values
9509 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
9510 only used to form the intrinsic's input argument. This might prove undesirable
9511 if the extra information provided by the ``llvm.assume`` intrinsic does cause
9512 sufficient overall improvement in code quality. For this reason,
9513 ``llvm.assume`` should not be used to document basic mathematical invariants
9514 that the optimizer can otherwise deduce or facts that are of little use to the
9517 '``llvm.donothing``' Intrinsic
9518 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9525 declare void @llvm.donothing() nounwind readnone
9530 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's the
9531 only intrinsic that can be called with an invoke instruction.
9541 This intrinsic does nothing, and it's removed by optimizers and ignored
9544 Stack Map Intrinsics
9545 --------------------
9547 LLVM provides experimental intrinsics to support runtime patching
9548 mechanisms commonly desired in dynamic language JITs. These intrinsics
9549 are described in :doc:`StackMaps`.