blob: ca5e8cf5e64b25b7cd5891cef31a97f95e5280e2 [file] [log] [blame]
Sean Silvab084af42012-12-07 10:36:55 +00001==============================
2LLVM Language Reference Manual
3==============================
4
5.. contents::
6 :local:
Rafael Espindola08013342013-12-07 19:34:20 +00007 :depth: 4
Sean Silvab084af42012-12-07 10:36:55 +00008
Sean Silvab084af42012-12-07 10:36:55 +00009Abstract
10========
11
12This document is a reference manual for the LLVM assembly language. LLVM
13is a Static Single Assignment (SSA) based representation that provides
14type safety, low-level operations, flexibility, and the capability of
15representing 'all' high-level languages cleanly. It is the common code
16representation used throughout all phases of the LLVM compilation
17strategy.
18
19Introduction
20============
21
22The LLVM code representation is designed to be used in three different
23forms: 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
25readable assembly language representation. This allows LLVM to provide a
26powerful intermediate representation for efficient compiler
27transformations and analysis, while providing a natural means to debug
28and visualize the transformations. The three different forms of LLVM are
29all equivalent. This document describes the human readable
30representation and notation.
31
32The LLVM representation aims to be light-weight and low-level while
33being expressive, typed, and extensible at the same time. It aims to be
34a "universal IR" of sorts, by being at a low enough level that
35high-level ideas may be cleanly mapped to it (similar to how
36microprocessors are "universal IR's", allowing many source languages to
37be mapped to them). By providing type information, LLVM can be used as
38the target of optimizations: for example, through pointer analysis, it
39can be proven that a C automatic variable is never accessed outside of
40the current function, allowing it to be promoted to a simple SSA value
41instead of a memory location.
42
43.. _wellformed:
44
45Well-Formedness
46---------------
47
48It is important to note that this document describes 'well formed' LLVM
49assembly language. There is a difference between what the parser accepts
50and what is considered 'well formed'. For example, the following
51instruction is syntactically okay, but not well formed:
52
53.. code-block:: llvm
54
55 %x = add i32 1, %x
56
57because the definition of ``%x`` does not dominate all of its uses. The
58LLVM infrastructure provides a verification pass that may be used to
59verify that an LLVM module is well formed. This pass is automatically
60run by the parser after parsing input assembly and by the optimizer
61before it outputs bitcode. The violations pointed out by the verifier
62pass indicate bugs in transformation passes or input to the parser.
63
64.. _identifiers:
65
66Identifiers
67===========
68
69LLVM identifiers come in two basic types: global and local. Global
70identifiers (functions, global variables) begin with the ``'@'``
71character. Local identifiers (register names, types) begin with the
72``'%'`` character. Additionally, there are three different formats for
73identifiers, for different purposes:
74
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
Sean Silva9d01a5b2015-01-07 21:35:14 +000078 '``[%@][-a-zA-Z$._][-a-zA-Z$._0-9]*``'. Identifiers that require other
Sean Silvab084af42012-12-07 10:36:55 +000079 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
Hans Wennborg85e06532014-07-30 20:02:08 +000082 be used in a name value, even quotes themselves. The ``"\01"`` prefix
83 can be used on global variables to suppress mangling.
Sean Silvab084af42012-12-07 10:36:55 +000084#. 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.
87
88LLVM requires that values start with a prefix for two reasons: Compilers
89don't need to worry about name clashes with reserved words, and the set
90of reserved words may be expanded in the future without penalty.
91Additionally, unnamed identifiers allow a compiler to quickly come up
92with a temporary variable without having to avoid symbol table
93conflicts.
94
95Reserved words in LLVM are very similar to reserved words in other
96languages. 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
99with variable names, because none of them start with a prefix character
100(``'%'`` or ``'@'``).
101
102Here is an example of LLVM code to multiply the integer variable
103'``%X``' by 8:
104
105The easy way:
106
107.. code-block:: llvm
108
109 %result = mul i32 %X, 8
110
111After strength reduction:
112
113.. code-block:: llvm
114
Dmitri Gribenko675911d2013-01-26 13:30:13 +0000115 %result = shl i32 %X, 3
Sean Silvab084af42012-12-07 10:36:55 +0000116
117And the hard way:
118
119.. code-block:: llvm
120
Tim Northover675a0962014-06-13 14:24:23 +0000121 %0 = add i32 %X, %X ; yields i32:%0
122 %1 = add i32 %0, %0 ; yields i32:%1
Sean Silvab084af42012-12-07 10:36:55 +0000123 %result = add i32 %1, %1
124
125This last way of multiplying ``%X`` by 8 illustrates several important
126lexical features of LLVM:
127
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.
Sean Silva8ca11782013-05-20 23:31:12 +0000131#. Unnamed temporaries are numbered sequentially (using a per-function
Dan Liew2661dfc2014-08-20 15:06:30 +0000132 incrementing counter, starting with 0). Note that basic blocks and unnamed
133 function parameters are included in this numbering. For example, if the
134 entry basic block is not given a label name and all function parameters are
135 named, then it will get number 0.
Sean Silvab084af42012-12-07 10:36:55 +0000136
137It also shows a convention that we follow in this document. When
138demonstrating instructions, we will follow an instruction with a comment
139that defines the type and name of value produced.
140
141High Level Structure
142====================
143
144Module Structure
145----------------
146
147LLVM programs are composed of ``Module``'s, each of which is a
148translation unit of the input programs. Each module consists of
149functions, global variables, and symbol table entries. Modules may be
150combined together with the LLVM linker, which merges function (and
151global variable) definitions, resolves forward declarations, and merges
152symbol table entries. Here is an example of the "hello world" module:
153
154.. code-block:: llvm
155
Michael Liaoa7699082013-03-06 18:24:34 +0000156 ; Declare the string constant as a global constant.
157 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
Sean Silvab084af42012-12-07 10:36:55 +0000158
Michael Liaoa7699082013-03-06 18:24:34 +0000159 ; External declaration of the puts function
160 declare i32 @puts(i8* nocapture) nounwind
Sean Silvab084af42012-12-07 10:36:55 +0000161
162 ; Definition of main function
Michael Liaoa7699082013-03-06 18:24:34 +0000163 define i32 @main() { ; i32()*
164 ; Convert [13 x i8]* to i8 *...
David Blaikie16a97eb2015-03-04 22:02:58 +0000165 %cast210 = getelementptr [13 x i8], [13 x i8]* @.str, i64 0, i64 0
Sean Silvab084af42012-12-07 10:36:55 +0000166
Michael Liaoa7699082013-03-06 18:24:34 +0000167 ; Call puts function to write out the string to stdout.
Sean Silvab084af42012-12-07 10:36:55 +0000168 call i32 @puts(i8* %cast210)
Michael Liaoa7699082013-03-06 18:24:34 +0000169 ret i32 0
Sean Silvab084af42012-12-07 10:36:55 +0000170 }
171
172 ; Named metadata
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +0000173 !0 = !{i32 42, null, !"string"}
Nick Lewyckya0de40a2014-08-13 04:54:05 +0000174 !foo = !{!0}
Sean Silvab084af42012-12-07 10:36:55 +0000175
176This example is made up of a :ref:`global variable <globalvars>` named
177"``.str``", an external declaration of the "``puts``" function, a
178:ref:`function definition <functionstructure>` for "``main``" and
179:ref:`named metadata <namedmetadatastructure>` "``foo``".
180
181In general, a module is made up of a list of global values (where both
182functions and global variables are global values). Global values are
183represented by a pointer to a memory location (in this case, a pointer
184to an array of char, and a pointer to a function), and have one of the
185following :ref:`linkage types <linkage>`.
186
187.. _linkage:
188
189Linkage Types
190-------------
191
192All Global Variables and Functions have one of the following types of
193linkage:
194
195``private``
196 Global values with "``private``" linkage are only directly
197 accessible by objects in the current module. In particular, linking
198 code into a module with an private global value may cause the
199 private to be renamed as necessary to avoid collisions. Because the
200 symbol is private to the module, all references can be updated. This
201 doesn't show up in any symbol table in the object file.
Sean Silvab084af42012-12-07 10:36:55 +0000202``internal``
203 Similar to private, but the value shows as a local symbol
204 (``STB_LOCAL`` in the case of ELF) in the object file. This
205 corresponds to the notion of the '``static``' keyword in C.
206``available_externally``
207 Globals with "``available_externally``" linkage are never emitted
208 into the object file corresponding to the LLVM module. They exist to
209 allow inlining and other optimizations to take place given knowledge
210 of the definition of the global, which is known to be somewhere
211 outside the module. Globals with ``available_externally`` linkage
212 are allowed to be discarded at will, and are otherwise the same as
213 ``linkonce_odr``. This linkage type is only allowed on definitions,
214 not declarations.
215``linkonce``
216 Globals with "``linkonce``" linkage are merged with other globals of
217 the same name when linkage occurs. This can be used to implement
218 some forms of inline functions, templates, or other code which must
219 be generated in each translation unit that uses it, but where the
220 body may be overridden with a more definitive definition later.
221 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
222 that ``linkonce`` linkage does not actually allow the optimizer to
223 inline the body of this function into callers because it doesn't
224 know if this definition of the function is the definitive definition
225 within the program or whether it will be overridden by a stronger
226 definition. To enable inlining and other optimizations, use
227 "``linkonce_odr``" linkage.
228``weak``
229 "``weak``" linkage has the same merging semantics as ``linkonce``
230 linkage, except that unreferenced globals with ``weak`` linkage may
231 not be discarded. This is used for globals that are declared "weak"
232 in C source code.
233``common``
234 "``common``" linkage is most similar to "``weak``" linkage, but they
235 are used for tentative definitions in C, such as "``int X;``" at
236 global scope. Symbols with "``common``" linkage are merged in the
237 same way as ``weak symbols``, and they may not be deleted if
238 unreferenced. ``common`` symbols may not have an explicit section,
239 must have a zero initializer, and may not be marked
240 ':ref:`constant <globalvars>`'. Functions and aliases may not have
241 common linkage.
242
243.. _linkage_appending:
244
245``appending``
246 "``appending``" linkage may only be applied to global variables of
247 pointer to array type. When two global variables with appending
248 linkage are linked together, the two global arrays are appended
249 together. This is the LLVM, typesafe, equivalent of having the
250 system linker append together "sections" with identical names when
251 .o files are linked.
252``extern_weak``
253 The semantics of this linkage follow the ELF object file model: the
254 symbol is weak until linked, if not linked, the symbol becomes null
255 instead of being an undefined reference.
256``linkonce_odr``, ``weak_odr``
257 Some languages allow differing globals to be merged, such as two
258 functions with different semantics. Other languages, such as
259 ``C++``, ensure that only equivalent globals are ever merged (the
Dmitri Gribenkoe8131122013-01-19 20:34:20 +0000260 "one definition rule" --- "ODR"). Such languages can use the
Sean Silvab084af42012-12-07 10:36:55 +0000261 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
262 global will only be merged with equivalent globals. These linkage
263 types are otherwise the same as their non-``odr`` versions.
Sean Silvab084af42012-12-07 10:36:55 +0000264``external``
265 If none of the above identifiers are used, the global is externally
266 visible, meaning that it participates in linkage and can be used to
267 resolve external symbol references.
268
Sean Silvab084af42012-12-07 10:36:55 +0000269It is illegal for a function *declaration* to have any linkage type
Nico Rieck7157bb72014-01-14 15:22:47 +0000270other than ``external`` or ``extern_weak``.
Sean Silvab084af42012-12-07 10:36:55 +0000271
Sean Silvab084af42012-12-07 10:36:55 +0000272.. _callingconv:
273
274Calling Conventions
275-------------------
276
277LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
278:ref:`invokes <i_invoke>` can all have an optional calling convention
279specified for the call. The calling convention of any pair of dynamic
280caller/callee must match, or the behavior of the program is undefined.
281The following calling conventions are supported by LLVM, and more may be
282added in the future:
283
284"``ccc``" - The C calling convention
285 This calling convention (the default if no other calling convention
286 is specified) matches the target C calling conventions. This calling
287 convention supports varargs function calls and tolerates some
288 mismatch in the declared prototype and implemented declaration of
289 the function (as does normal C).
290"``fastcc``" - The fast calling convention
291 This calling convention attempts to make calls as fast as possible
292 (e.g. by passing things in registers). This calling convention
293 allows the target to use whatever tricks it wants to produce fast
294 code for the target, without having to conform to an externally
295 specified ABI (Application Binary Interface). `Tail calls can only
296 be optimized when this, the GHC or the HiPE convention is
297 used. <CodeGenerator.html#id80>`_ This calling convention does not
298 support varargs and requires the prototype of all callees to exactly
299 match the prototype of the function definition.
300"``coldcc``" - The cold calling convention
301 This calling convention attempts to make code in the caller as
302 efficient as possible under the assumption that the call is not
303 commonly executed. As such, these calls often preserve all registers
304 so that the call does not break any live ranges in the caller side.
305 This calling convention does not support varargs and requires the
306 prototype of all callees to exactly match the prototype of the
Juergen Ributzka5d05ed12014-01-17 22:24:35 +0000307 function definition. Furthermore the inliner doesn't consider such function
308 calls for inlining.
Sean Silvab084af42012-12-07 10:36:55 +0000309"``cc 10``" - GHC convention
310 This calling convention has been implemented specifically for use by
311 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
312 It passes everything in registers, going to extremes to achieve this
313 by disabling callee save registers. This calling convention should
314 not be used lightly but only for specific situations such as an
315 alternative to the *register pinning* performance technique often
316 used when implementing functional programming languages. At the
317 moment only X86 supports this convention and it has the following
318 limitations:
319
320 - On *X86-32* only supports up to 4 bit type parameters. No
321 floating point types are supported.
322 - On *X86-64* only supports up to 10 bit type parameters and 6
323 floating point parameters.
324
325 This calling convention supports `tail call
326 optimization <CodeGenerator.html#id80>`_ but requires both the
327 caller and callee are using it.
328"``cc 11``" - The HiPE calling convention
329 This calling convention has been implemented specifically for use by
330 the `High-Performance Erlang
331 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
332 native code compiler of the `Ericsson's Open Source Erlang/OTP
333 system <http://www.erlang.org/download.shtml>`_. It uses more
334 registers for argument passing than the ordinary C calling
335 convention and defines no callee-saved registers. The calling
336 convention properly supports `tail call
337 optimization <CodeGenerator.html#id80>`_ but requires that both the
338 caller and the callee use it. It uses a *register pinning*
339 mechanism, similar to GHC's convention, for keeping frequently
340 accessed runtime components pinned to specific hardware registers.
341 At the moment only X86 supports this convention (both 32 and 64
342 bit).
Andrew Trick5e029ce2013-12-24 02:57:25 +0000343"``webkit_jscc``" - WebKit's JavaScript calling convention
344 This calling convention has been implemented for `WebKit FTL JIT
345 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
346 stack right to left (as cdecl does), and returns a value in the
347 platform's customary return register.
348"``anyregcc``" - Dynamic calling convention for code patching
349 This is a special convention that supports patching an arbitrary code
350 sequence in place of a call site. This convention forces the call
Eli Bendersky45324ce2015-04-02 15:20:04 +0000351 arguments into registers but allows them to be dynamically
Andrew Trick5e029ce2013-12-24 02:57:25 +0000352 allocated. This can currently only be used with calls to
353 llvm.experimental.patchpoint because only this intrinsic records
354 the location of its arguments in a side table. See :doc:`StackMaps`.
Juergen Ributzkae6250132014-01-17 19:47:03 +0000355"``preserve_mostcc``" - The `PreserveMost` calling convention
Eli Bendersky45324ce2015-04-02 15:20:04 +0000356 This calling convention attempts to make the code in the caller as
357 unintrusive as possible. This convention behaves identically to the `C`
Juergen Ributzkae6250132014-01-17 19:47:03 +0000358 calling convention on how arguments and return values are passed, but it
359 uses a different set of caller/callee-saved registers. This alleviates the
360 burden of saving and recovering a large register set before and after the
Juergen Ributzka980f2dc2014-01-30 02:39:00 +0000361 call in the caller. If the arguments are passed in callee-saved registers,
362 then they will be preserved by the callee across the call. This doesn't
363 apply for values returned in callee-saved registers.
Juergen Ributzkae6250132014-01-17 19:47:03 +0000364
365 - On X86-64 the callee preserves all general purpose registers, except for
366 R11. R11 can be used as a scratch register. Floating-point registers
367 (XMMs/YMMs) are not preserved and need to be saved by the caller.
368
369 The idea behind this convention is to support calls to runtime functions
370 that have a hot path and a cold path. The hot path is usually a small piece
Eric Christopher1e61ffd2015-02-19 18:46:25 +0000371 of code that doesn't use many registers. The cold path might need to call out to
Juergen Ributzkae6250132014-01-17 19:47:03 +0000372 another function and therefore only needs to preserve the caller-saved
Juergen Ributzka5d05ed12014-01-17 22:24:35 +0000373 registers, which haven't already been saved by the caller. The
374 `PreserveMost` calling convention is very similar to the `cold` calling
375 convention in terms of caller/callee-saved registers, but they are used for
376 different types of function calls. `coldcc` is for function calls that are
377 rarely executed, whereas `preserve_mostcc` function calls are intended to be
378 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
379 doesn't prevent the inliner from inlining the function call.
Juergen Ributzkae6250132014-01-17 19:47:03 +0000380
381 This calling convention will be used by a future version of the ObjectiveC
382 runtime and should therefore still be considered experimental at this time.
383 Although this convention was created to optimize certain runtime calls to
384 the ObjectiveC runtime, it is not limited to this runtime and might be used
385 by other runtimes in the future too. The current implementation only
386 supports X86-64, but the intention is to support more architectures in the
387 future.
388"``preserve_allcc``" - The `PreserveAll` calling convention
389 This calling convention attempts to make the code in the caller even less
390 intrusive than the `PreserveMost` calling convention. This calling
391 convention also behaves identical to the `C` calling convention on how
392 arguments and return values are passed, but it uses a different set of
393 caller/callee-saved registers. This removes the burden of saving and
Juergen Ributzka980f2dc2014-01-30 02:39:00 +0000394 recovering a large register set before and after the call in the caller. If
395 the arguments are passed in callee-saved registers, then they will be
396 preserved by the callee across the call. This doesn't apply for values
397 returned in callee-saved registers.
Juergen Ributzkae6250132014-01-17 19:47:03 +0000398
399 - On X86-64 the callee preserves all general purpose registers, except for
400 R11. R11 can be used as a scratch register. Furthermore it also preserves
401 all floating-point registers (XMMs/YMMs).
402
403 The idea behind this convention is to support calls to runtime functions
404 that don't need to call out to any other functions.
405
406 This calling convention, like the `PreserveMost` calling convention, will be
407 used by a future version of the ObjectiveC runtime and should be considered
408 experimental at this time.
Sean Silvab084af42012-12-07 10:36:55 +0000409"``cc <n>``" - Numbered convention
410 Any calling convention may be specified by number, allowing
411 target-specific calling conventions to be used. Target specific
412 calling conventions start at 64.
413
414More calling conventions can be added/defined on an as-needed basis, to
415support Pascal conventions or any other well-known target-independent
416convention.
417
Eli Benderskyfdc529a2013-06-07 19:40:08 +0000418.. _visibilitystyles:
419
Sean Silvab084af42012-12-07 10:36:55 +0000420Visibility Styles
421-----------------
422
423All Global Variables and Functions have one of the following visibility
424styles:
425
426"``default``" - Default style
427 On targets that use the ELF object file format, default visibility
428 means that the declaration is visible to other modules and, in
429 shared libraries, means that the declared entity may be overridden.
430 On Darwin, default visibility means that the declaration is visible
431 to other modules. Default visibility corresponds to "external
432 linkage" in the language.
433"``hidden``" - Hidden style
434 Two declarations of an object with hidden visibility refer to the
435 same object if they are in the same shared object. Usually, hidden
436 visibility indicates that the symbol will not be placed into the
437 dynamic symbol table, so no other module (executable or shared
438 library) can reference it directly.
439"``protected``" - Protected style
440 On ELF, protected visibility indicates that the symbol will be
441 placed in the dynamic symbol table, but that references within the
442 defining module will bind to the local symbol. That is, the symbol
443 cannot be overridden by another module.
444
Duncan P. N. Exon Smithb80de102014-05-07 22:57:20 +0000445A symbol with ``internal`` or ``private`` linkage must have ``default``
446visibility.
447
Rafael Espindola3bc64d52014-05-26 21:30:40 +0000448.. _dllstorageclass:
Eli Benderskyfdc529a2013-06-07 19:40:08 +0000449
Nico Rieck7157bb72014-01-14 15:22:47 +0000450DLL Storage Classes
451-------------------
452
453All Global Variables, Functions and Aliases can have one of the following
454DLL storage class:
455
456``dllimport``
457 "``dllimport``" causes the compiler to reference a function or variable via
458 a global pointer to a pointer that is set up by the DLL exporting the
459 symbol. On Microsoft Windows targets, the pointer name is formed by
460 combining ``__imp_`` and the function or variable name.
461``dllexport``
462 "``dllexport``" causes the compiler to provide a global pointer to a pointer
463 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
464 Microsoft Windows targets, the pointer name is formed by combining
465 ``__imp_`` and the function or variable name. Since this storage class
466 exists for defining a dll interface, the compiler, assembler and linker know
467 it is externally referenced and must refrain from deleting the symbol.
468
Rafael Espindola59f7eba2014-05-28 18:15:43 +0000469.. _tls_model:
470
471Thread Local Storage Models
472---------------------------
473
474A variable may be defined as ``thread_local``, which means that it will
475not be shared by threads (each thread will have a separated copy of the
476variable). Not all targets support thread-local variables. Optionally, a
477TLS model may be specified:
478
479``localdynamic``
480 For variables that are only used within the current shared library.
481``initialexec``
482 For variables in modules that will not be loaded dynamically.
483``localexec``
484 For variables defined in the executable and only used within it.
485
486If no explicit model is given, the "general dynamic" model is used.
487
488The models correspond to the ELF TLS models; see `ELF Handling For
489Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
490more information on under which circumstances the different models may
491be used. The target may choose a different TLS model if the specified
492model is not supported, or if a better choice of model can be made.
493
Sean Silva706fba52015-08-06 22:56:24 +0000494A model can also be specified in an alias, but then it only governs how
Rafael Espindola59f7eba2014-05-28 18:15:43 +0000495the alias is accessed. It will not have any effect in the aliasee.
496
Chih-Hung Hsieh1e859582015-07-28 16:24:05 +0000497For platforms without linker support of ELF TLS model, the -femulated-tls
498flag can be used to generate GCC compatible emulated TLS code.
499
Rafael Espindola3bc64d52014-05-26 21:30:40 +0000500.. _namedtypes:
501
Reid Kleckner7c84d1d2014-03-05 02:21:50 +0000502Structure Types
503---------------
Sean Silvab084af42012-12-07 10:36:55 +0000504
Reid Kleckner7c84d1d2014-03-05 02:21:50 +0000505LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
506types <t_struct>`. Literal types are uniqued structurally, but identified types
507are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
Richard Smith32dbdf62014-07-31 04:25:36 +0000508to forward declare a type that is not yet available.
Reid Kleckner7c84d1d2014-03-05 02:21:50 +0000509
Sean Silva706fba52015-08-06 22:56:24 +0000510An example of an identified structure specification is:
Sean Silvab084af42012-12-07 10:36:55 +0000511
512.. code-block:: llvm
513
514 %mytype = type { %mytype*, i32 }
515
Reid Kleckner7c84d1d2014-03-05 02:21:50 +0000516Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
517literal types are uniqued in recent versions of LLVM.
Sean Silvab084af42012-12-07 10:36:55 +0000518
519.. _globalvars:
520
521Global Variables
522----------------
523
524Global variables define regions of memory allocated at compilation time
Rafael Espindola5d1b7452013-10-29 13:44:11 +0000525instead of run-time.
526
Eric Christopher1e61ffd2015-02-19 18:46:25 +0000527Global variable definitions must be initialized.
Rafael Espindola5d1b7452013-10-29 13:44:11 +0000528
529Global variables in other translation units can also be declared, in which
530case they don't have an initializer.
Sean Silvab084af42012-12-07 10:36:55 +0000531
Bob Wilson85b24f22014-06-12 20:40:33 +0000532Either global variable definitions or declarations may have an explicit section
533to be placed in and may have an optional explicit alignment specified.
534
Michael Gottesman006039c2013-01-31 05:48:48 +0000535A variable may be defined as a global ``constant``, which indicates that
Sean Silvab084af42012-12-07 10:36:55 +0000536the contents of the variable will **never** be modified (enabling better
537optimization, allowing the global data to be placed in the read-only
538section of an executable, etc). Note that variables that need runtime
Michael Gottesman1cffcf742013-01-31 05:44:04 +0000539initialization cannot be marked ``constant`` as there is a store to the
Sean Silvab084af42012-12-07 10:36:55 +0000540variable.
541
542LLVM explicitly allows *declarations* of global variables to be marked
543constant, even if the final definition of the global is not. This
544capability can be used to enable slightly better optimization of the
545program, but requires the language definition to guarantee that
546optimizations based on the 'constantness' are valid for the translation
547units that do not include the definition.
548
549As SSA values, global variables define pointer values that are in scope
550(i.e. they dominate) all basic blocks in the program. Global variables
551always define a pointer to their "content" type because they describe a
552region of memory, and all memory objects in LLVM are accessed through
553pointers.
554
555Global variables can be marked with ``unnamed_addr`` which indicates
556that the address is not significant, only the content. Constants marked
557like this can be merged with other constants if they have the same
558initializer. Note that a constant with significant address *can* be
559merged with a ``unnamed_addr`` constant, the result being a constant
560whose address is significant.
561
562A global variable may be declared to reside in a target-specific
563numbered address space. For targets that support them, address spaces
564may affect how optimizations are performed and/or what target
565instructions are used to access the variable. The default address space
566is zero. The address space qualifier must precede any other attributes.
567
568LLVM allows an explicit section to be specified for globals. If the
569target supports it, it will emit globals to the section specified.
David Majnemerdad0a642014-06-27 18:19:56 +0000570Additionally, the global can placed in a comdat if the target has the necessary
571support.
Sean Silvab084af42012-12-07 10:36:55 +0000572
Michael Gottesmane743a302013-02-04 03:22:00 +0000573By default, global initializers are optimized by assuming that global
Michael Gottesmanef2bc772013-02-03 09:57:15 +0000574variables defined within the module are not modified from their
575initial values before the start of the global initializer. This is
576true even for variables potentially accessible from outside the
577module, including those with external linkage or appearing in
Yunzhong Gaof5b769e2013-12-05 18:37:54 +0000578``@llvm.used`` or dllexported variables. This assumption may be suppressed
579by marking the variable with ``externally_initialized``.
Michael Gottesmanef2bc772013-02-03 09:57:15 +0000580
Sean Silvab084af42012-12-07 10:36:55 +0000581An explicit alignment may be specified for a global, which must be a
582power of 2. If not present, or if the alignment is set to zero, the
583alignment of the global is set by the target to whatever it feels
584convenient. If an explicit alignment is specified, the global is forced
585to have exactly that alignment. Targets and optimizers are not allowed
586to over-align the global if the global has an assigned section. In this
587case, the extra alignment could be observable: for example, code could
588assume that the globals are densely packed in their section and try to
589iterate over them as an array, alignment padding would break this
Reid Kleckner15fe7a52014-07-15 01:16:09 +0000590iteration. The maximum alignment is ``1 << 29``.
Sean Silvab084af42012-12-07 10:36:55 +0000591
Nico Rieck7157bb72014-01-14 15:22:47 +0000592Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
593
Peter Collingbourne69ba0162015-02-04 00:42:45 +0000594Variables and aliases can have a
Rafael Espindola59f7eba2014-05-28 18:15:43 +0000595:ref:`Thread Local Storage Model <tls_model>`.
596
Nico Rieck7157bb72014-01-14 15:22:47 +0000597Syntax::
598
599 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
Rafael Espindola28f3ca62014-06-09 21:21:33 +0000600 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
Bob Wilson85b24f22014-06-12 20:40:33 +0000601 <global | constant> <Type> [<InitializerConstant>]
Rafael Espindola83a362c2015-01-06 22:55:16 +0000602 [, section "name"] [, comdat [($name)]]
603 [, align <Alignment>]
Nico Rieck7157bb72014-01-14 15:22:47 +0000604
Sean Silvab084af42012-12-07 10:36:55 +0000605For example, the following defines a global in a numbered address space
606with an initializer, section, and alignment:
607
608.. code-block:: llvm
609
610 @G = addrspace(5) constant float 1.0, section "foo", align 4
611
Rafael Espindola5d1b7452013-10-29 13:44:11 +0000612The following example just declares a global variable
613
614.. code-block:: llvm
615
616 @G = external global i32
617
Sean Silvab084af42012-12-07 10:36:55 +0000618The following example defines a thread-local global with the
619``initialexec`` TLS model:
620
621.. code-block:: llvm
622
623 @G = thread_local(initialexec) global i32 0, align 4
624
625.. _functionstructure:
626
627Functions
628---------
629
630LLVM function definitions consist of the "``define``" keyword, an
631optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
Nico Rieck7157bb72014-01-14 15:22:47 +0000632style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
633an optional :ref:`calling convention <callingconv>`,
Sean Silvab084af42012-12-07 10:36:55 +0000634an optional ``unnamed_addr`` attribute, a return type, an optional
635:ref:`parameter attribute <paramattrs>` for the return type, a function
636name, a (possibly empty) argument list (each with optional :ref:`parameter
637attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
David Majnemerdad0a642014-06-27 18:19:56 +0000638an optional section, an optional alignment,
639an optional :ref:`comdat <langref_comdats>`,
Peter Collingbourne51d2de72014-12-03 02:08:38 +0000640an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
David Majnemer7fddecc2015-06-17 20:52:32 +0000641an optional :ref:`prologue <prologuedata>`,
642an optional :ref:`personality <personalityfn>`,
643an opening curly brace, a list of basic blocks, and a closing curly brace.
Sean Silvab084af42012-12-07 10:36:55 +0000644
645LLVM function declarations consist of the "``declare``" keyword, an
646optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
Nico Rieck7157bb72014-01-14 15:22:47 +0000647style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
648an optional :ref:`calling convention <callingconv>`,
Sean Silvab084af42012-12-07 10:36:55 +0000649an optional ``unnamed_addr`` attribute, a return type, an optional
650:ref:`parameter attribute <paramattrs>` for the return type, a function
Peter Collingbourne3fa50f92013-09-16 01:08:15 +0000651name, a possibly empty list of arguments, an optional alignment, an optional
Peter Collingbourne51d2de72014-12-03 02:08:38 +0000652:ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
653and an optional :ref:`prologue <prologuedata>`.
Sean Silvab084af42012-12-07 10:36:55 +0000654
Bill Wendling6822ecb2013-10-27 05:09:12 +0000655A function definition contains a list of basic blocks, forming the CFG (Control
656Flow Graph) for the function. Each basic block may optionally start with a label
657(giving the basic block a symbol table entry), contains a list of instructions,
658and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
659function return). If an explicit label is not provided, a block is assigned an
660implicit numbered label, using the next value from the same counter as used for
661unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
662entry block does not have an explicit label, it will be assigned label "%0",
663then the first unnamed temporary in that block will be "%1", etc.
Sean Silvab084af42012-12-07 10:36:55 +0000664
665The first basic block in a function is special in two ways: it is
666immediately executed on entrance to the function, and it is not allowed
667to have predecessor basic blocks (i.e. there can not be any branches to
668the entry block of a function). Because the block can have no
669predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
670
671LLVM allows an explicit section to be specified for functions. If the
672target supports it, it will emit functions to the section specified.
Eric Christopher1e61ffd2015-02-19 18:46:25 +0000673Additionally, the function can be placed in a COMDAT.
Sean Silvab084af42012-12-07 10:36:55 +0000674
675An explicit alignment may be specified for a function. If not present,
676or if the alignment is set to zero, the alignment of the function is set
677by the target to whatever it feels convenient. If an explicit alignment
678is specified, the function is forced to have at least that much
679alignment. All alignments must be a power of 2.
680
Eric Christopher1e61ffd2015-02-19 18:46:25 +0000681If the ``unnamed_addr`` attribute is given, the address is known to not
Sean Silvab084af42012-12-07 10:36:55 +0000682be significant and two identical functions can be merged.
683
684Syntax::
685
Nico Rieck7157bb72014-01-14 15:22:47 +0000686 define [linkage] [visibility] [DLLStorageClass]
Sean Silvab084af42012-12-07 10:36:55 +0000687 [cconv] [ret attrs]
688 <ResultType> @<FunctionName> ([argument list])
Rafael Espindola83a362c2015-01-06 22:55:16 +0000689 [unnamed_addr] [fn Attrs] [section "name"] [comdat [($name)]]
David Majnemer7fddecc2015-06-17 20:52:32 +0000690 [align N] [gc] [prefix Constant] [prologue Constant]
691 [personality Constant] { ... }
Sean Silvab084af42012-12-07 10:36:55 +0000692
Sean Silva706fba52015-08-06 22:56:24 +0000693The argument list is a comma separated sequence of arguments where each
694argument is of the following form:
Dan Liew2661dfc2014-08-20 15:06:30 +0000695
696Syntax::
697
698 <type> [parameter Attrs] [name]
699
700
Eli Benderskyfdc529a2013-06-07 19:40:08 +0000701.. _langref_aliases:
702
Sean Silvab084af42012-12-07 10:36:55 +0000703Aliases
704-------
705
Rafael Espindola64c1e182014-06-03 02:41:57 +0000706Aliases, unlike function or variables, don't create any new data. They
707are just a new symbol and metadata for an existing position.
708
709Aliases have a name and an aliasee that is either a global value or a
710constant expression.
711
Nico Rieck7157bb72014-01-14 15:22:47 +0000712Aliases may have an optional :ref:`linkage type <linkage>`, an optional
Rafael Espindola64c1e182014-06-03 02:41:57 +0000713:ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
714<dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
Sean Silvab084af42012-12-07 10:36:55 +0000715
716Syntax::
717
Rafael Espindola464fe022014-07-30 22:51:54 +0000718 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy> @<Aliasee>
Sean Silvab084af42012-12-07 10:36:55 +0000719
Rafael Espindola2fb5bc32014-03-13 23:18:37 +0000720The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
Rafael Espindola716e7402013-11-01 17:09:14 +0000721``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
Rafael Espindola64c1e182014-06-03 02:41:57 +0000722might not correctly handle dropping a weak symbol that is aliased.
Rafael Espindola78527052013-10-06 15:10:43 +0000723
Eric Christopher1e61ffd2015-02-19 18:46:25 +0000724Aliases that are not ``unnamed_addr`` are guaranteed to have the same address as
Rafael Espindola42a4c9f2014-06-06 01:20:28 +0000725the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
726to the same content.
Rafael Espindolaf3336bc2014-03-12 20:15:49 +0000727
Rafael Espindola64c1e182014-06-03 02:41:57 +0000728Since aliases are only a second name, some restrictions apply, of which
729some can only be checked when producing an object file:
Rafael Espindolaf3336bc2014-03-12 20:15:49 +0000730
Rafael Espindola64c1e182014-06-03 02:41:57 +0000731* The expression defining the aliasee must be computable at assembly
732 time. Since it is just a name, no relocations can be used.
733
734* No alias in the expression can be weak as the possibility of the
735 intermediate alias being overridden cannot be represented in an
736 object file.
737
738* No global value in the expression can be a declaration, since that
739 would require a relocation, which is not possible.
Rafael Espindola24a669d2014-03-27 15:26:56 +0000740
David Majnemerdad0a642014-06-27 18:19:56 +0000741.. _langref_comdats:
742
743Comdats
744-------
745
746Comdat IR provides access to COFF and ELF object file COMDAT functionality.
747
Richard Smith32dbdf62014-07-31 04:25:36 +0000748Comdats have a name which represents the COMDAT key. All global objects that
David Majnemerdad0a642014-06-27 18:19:56 +0000749specify this key will only end up in the final object file if the linker chooses
750that key over some other key. Aliases are placed in the same COMDAT that their
751aliasee computes to, if any.
752
753Comdats have a selection kind to provide input on how the linker should
754choose between keys in two different object files.
755
756Syntax::
757
758 $<Name> = comdat SelectionKind
759
760The selection kind must be one of the following:
761
762``any``
763 The linker may choose any COMDAT key, the choice is arbitrary.
764``exactmatch``
765 The linker may choose any COMDAT key but the sections must contain the
766 same data.
767``largest``
768 The linker will choose the section containing the largest COMDAT key.
769``noduplicates``
770 The linker requires that only section with this COMDAT key exist.
771``samesize``
772 The linker may choose any COMDAT key but the sections must contain the
773 same amount of data.
774
775Note that the Mach-O platform doesn't support COMDATs and ELF only supports
776``any`` as a selection kind.
777
778Here is an example of a COMDAT group where a function will only be selected if
779the COMDAT key's section is the largest:
780
781.. code-block:: llvm
782
783 $foo = comdat largest
Rafael Espindola83a362c2015-01-06 22:55:16 +0000784 @foo = global i32 2, comdat($foo)
David Majnemerdad0a642014-06-27 18:19:56 +0000785
Rafael Espindola83a362c2015-01-06 22:55:16 +0000786 define void @bar() comdat($foo) {
David Majnemerdad0a642014-06-27 18:19:56 +0000787 ret void
788 }
789
Rafael Espindola83a362c2015-01-06 22:55:16 +0000790As a syntactic sugar the ``$name`` can be omitted if the name is the same as
791the global name:
792
793.. code-block:: llvm
794
795 $foo = comdat any
796 @foo = global i32 2, comdat
797
798
David Majnemerdad0a642014-06-27 18:19:56 +0000799In a COFF object file, this will create a COMDAT section with selection kind
800``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
801and another COMDAT section with selection kind
802``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
Hans Wennborg0def0662014-09-10 17:05:08 +0000803section and contains the contents of the ``@bar`` symbol.
David Majnemerdad0a642014-06-27 18:19:56 +0000804
805There are some restrictions on the properties of the global object.
806It, or an alias to it, must have the same name as the COMDAT group when
807targeting COFF.
808The contents and size of this object may be used during link-time to determine
809which COMDAT groups get selected depending on the selection kind.
810Because the name of the object must match the name of the COMDAT group, the
811linkage of the global object must not be local; local symbols can get renamed
812if a collision occurs in the symbol table.
813
814The combined use of COMDATS and section attributes may yield surprising results.
815For example:
816
817.. code-block:: llvm
818
819 $foo = comdat any
820 $bar = comdat any
Rafael Espindola83a362c2015-01-06 22:55:16 +0000821 @g1 = global i32 42, section "sec", comdat($foo)
822 @g2 = global i32 42, section "sec", comdat($bar)
David Majnemerdad0a642014-06-27 18:19:56 +0000823
824From the object file perspective, this requires the creation of two sections
825with the same name. This is necessary because both globals belong to different
826COMDAT groups and COMDATs, at the object file level, are represented by
827sections.
828
Peter Collingbourne1feef2e2015-06-30 19:10:31 +0000829Note that certain IR constructs like global variables and functions may
830create COMDATs in the object file in addition to any which are specified using
831COMDAT IR. This arises when the code generator is configured to emit globals
832in individual sections (e.g. when `-data-sections` or `-function-sections`
833is supplied to `llc`).
David Majnemerdad0a642014-06-27 18:19:56 +0000834
Sean Silvab084af42012-12-07 10:36:55 +0000835.. _namedmetadatastructure:
836
837Named Metadata
838--------------
839
840Named metadata is a collection of metadata. :ref:`Metadata
841nodes <metadata>` (but not metadata strings) are the only valid
842operands for a named metadata.
843
Filipe Cabecinhas62431b12015-06-02 21:25:08 +0000844#. Named metadata are represented as a string of characters with the
845 metadata prefix. The rules for metadata names are the same as for
846 identifiers, but quoted names are not allowed. ``"\xx"`` type escapes
847 are still valid, which allows any character to be part of a name.
848
Sean Silvab084af42012-12-07 10:36:55 +0000849Syntax::
850
851 ; Some unnamed metadata nodes, which are referenced by the named metadata.
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +0000852 !0 = !{!"zero"}
853 !1 = !{!"one"}
854 !2 = !{!"two"}
Sean Silvab084af42012-12-07 10:36:55 +0000855 ; A named metadata.
856 !name = !{!0, !1, !2}
857
858.. _paramattrs:
859
860Parameter Attributes
861--------------------
862
863The return type and each parameter of a function type may have a set of
864*parameter attributes* associated with them. Parameter attributes are
865used to communicate additional information about the result or
866parameters of a function. Parameter attributes are considered to be part
867of the function, not of the function type, so functions with different
868parameter attributes can have the same function type.
869
870Parameter attributes are simple keywords that follow the type specified.
871If multiple parameter attributes are needed, they are space separated.
872For example:
873
874.. code-block:: llvm
875
876 declare i32 @printf(i8* noalias nocapture, ...)
877 declare i32 @atoi(i8 zeroext)
878 declare signext i8 @returns_signed_char()
879
880Note that any attributes for the function result (``nounwind``,
881``readonly``) come immediately after the argument list.
882
883Currently, only the following parameter attributes are defined:
884
885``zeroext``
886 This indicates to the code generator that the parameter or return
887 value should be zero-extended to the extent required by the target's
888 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
889 the caller (for a parameter) or the callee (for a return value).
890``signext``
891 This indicates to the code generator that the parameter or return
892 value should be sign-extended to the extent required by the target's
893 ABI (which is usually 32-bits) by the caller (for a parameter) or
894 the callee (for a return value).
895``inreg``
896 This indicates that this parameter or return value should be treated
Sean Silva706fba52015-08-06 22:56:24 +0000897 in a special target-dependent fashion while emitting code for
Sean Silvab084af42012-12-07 10:36:55 +0000898 a function call or return (usually, by putting it in a register as
899 opposed to memory, though some targets use it to distinguish between
900 two different kinds of registers). Use of this attribute is
901 target-specific.
902``byval``
903 This indicates that the pointer parameter should really be passed by
904 value to the function. The attribute implies that a hidden copy of
905 the pointee is made between the caller and the callee, so the callee
906 is unable to modify the value in the caller. This attribute is only
907 valid on LLVM pointer arguments. It is generally used to pass
908 structs and arrays by value, but is also valid on pointers to
909 scalars. The copy is considered to belong to the caller not the
910 callee (for example, ``readonly`` functions should not write to
911 ``byval`` parameters). This is not a valid attribute for return
912 values.
913
914 The byval attribute also supports specifying an alignment with the
915 align attribute. It indicates the alignment of the stack slot to
916 form and the known alignment of the pointer specified to the call
917 site. If the alignment is not specified, then the code generator
918 makes a target-specific assumption.
919
Reid Klecknera534a382013-12-19 02:14:12 +0000920.. _attr_inalloca:
921
922``inalloca``
923
Reid Kleckner60d3a832014-01-16 22:59:24 +0000924 The ``inalloca`` argument attribute allows the caller to take the
Reid Kleckner436c42e2014-01-17 23:58:17 +0000925 address of outgoing stack arguments. An ``inalloca`` argument must
926 be a pointer to stack memory produced by an ``alloca`` instruction.
927 The alloca, or argument allocation, must also be tagged with the
Hal Finkelc8491d32014-07-16 21:22:46 +0000928 inalloca keyword. Only the last argument may have the ``inalloca``
Reid Kleckner436c42e2014-01-17 23:58:17 +0000929 attribute, and that argument is guaranteed to be passed in memory.
Reid Klecknera534a382013-12-19 02:14:12 +0000930
Reid Kleckner436c42e2014-01-17 23:58:17 +0000931 An argument allocation may be used by a call at most once because
932 the call may deallocate it. The ``inalloca`` attribute cannot be
933 used in conjunction with other attributes that affect argument
Reid Klecknerf5b76512014-01-31 23:50:57 +0000934 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
935 ``inalloca`` attribute also disables LLVM's implicit lowering of
936 large aggregate return values, which means that frontend authors
937 must lower them with ``sret`` pointers.
Reid Klecknera534a382013-12-19 02:14:12 +0000938
Reid Kleckner60d3a832014-01-16 22:59:24 +0000939 When the call site is reached, the argument allocation must have
940 been the most recent stack allocation that is still live, or the
941 results are undefined. It is possible to allocate additional stack
942 space after an argument allocation and before its call site, but it
943 must be cleared off with :ref:`llvm.stackrestore
944 <int_stackrestore>`.
Reid Klecknera534a382013-12-19 02:14:12 +0000945
946 See :doc:`InAlloca` for more information on how to use this
947 attribute.
948
Sean Silvab084af42012-12-07 10:36:55 +0000949``sret``
950 This indicates that the pointer parameter specifies the address of a
951 structure that is the return value of the function in the source
952 program. This pointer must be guaranteed by the caller to be valid:
Eli Bendersky4f2162f2013-01-23 22:05:19 +0000953 loads and stores to the structure may be assumed by the callee
Sean Silvab084af42012-12-07 10:36:55 +0000954 not to trap and to be properly aligned. This may only be applied to
955 the first parameter. This is not a valid attribute for return
956 values.
Sean Silva1703e702014-04-08 21:06:22 +0000957
Hal Finkelccc70902014-07-22 16:58:55 +0000958``align <n>``
959 This indicates that the pointer value may be assumed by the optimizer to
960 have the specified alignment.
961
962 Note that this attribute has additional semantics when combined with the
963 ``byval`` attribute.
964
Sean Silva1703e702014-04-08 21:06:22 +0000965.. _noalias:
966
Sean Silvab084af42012-12-07 10:36:55 +0000967``noalias``
Hal Finkel12d36302014-11-21 02:22:46 +0000968 This indicates that objects accessed via pointer values
969 :ref:`based <pointeraliasing>` on the argument or return value are not also
970 accessed, during the execution of the function, via pointer values not
971 *based* on the argument or return value. The attribute on a return value
972 also has additional semantics described below. The caller shares the
973 responsibility with the callee for ensuring that these requirements are met.
974 For further details, please see the discussion of the NoAlias response in
975 :ref:`alias analysis <Must, May, or No>`.
Sean Silvab084af42012-12-07 10:36:55 +0000976
977 Note that this definition of ``noalias`` is intentionally similar
Hal Finkel12d36302014-11-21 02:22:46 +0000978 to the definition of ``restrict`` in C99 for function arguments.
Sean Silvab084af42012-12-07 10:36:55 +0000979
980 For function return values, C99's ``restrict`` is not meaningful,
Hal Finkel12d36302014-11-21 02:22:46 +0000981 while LLVM's ``noalias`` is. Furthermore, the semantics of the ``noalias``
982 attribute on return values are stronger than the semantics of the attribute
983 when used on function arguments. On function return values, the ``noalias``
984 attribute indicates that the function acts like a system memory allocation
985 function, returning a pointer to allocated storage disjoint from the
986 storage for any other object accessible to the caller.
987
Sean Silvab084af42012-12-07 10:36:55 +0000988``nocapture``
989 This indicates that the callee does not make any copies of the
990 pointer that outlive the callee itself. This is not a valid
991 attribute for return values.
992
993.. _nest:
994
995``nest``
996 This indicates that the pointer parameter can be excised using the
997 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
Stephen Linb8bd2322013-04-20 05:14:40 +0000998 attribute for return values and can only be applied to one parameter.
999
1000``returned``
Stephen Linfec5b0b2013-06-20 21:55:10 +00001001 This indicates that the function always returns the argument as its return
1002 value. This is an optimization hint to the code generator when generating
1003 the caller, allowing tail call optimization and omission of register saves
1004 and restores in some cases; it is not checked or enforced when generating
1005 the callee. The parameter and the function return type must be valid
1006 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
1007 valid attribute for return values and can only be applied to one parameter.
Sean Silvab084af42012-12-07 10:36:55 +00001008
Nick Lewyckyd52b1522014-05-20 01:23:40 +00001009``nonnull``
1010 This indicates that the parameter or return pointer is not null. This
1011 attribute may only be applied to pointer typed parameters. This is not
1012 checked or enforced by LLVM, the caller must ensure that the pointer
Mehdi Amini4a121fa2015-03-14 22:04:06 +00001013 passed in is non-null, or the callee must ensure that the returned pointer
Nick Lewyckyd52b1522014-05-20 01:23:40 +00001014 is non-null.
1015
Hal Finkelb0407ba2014-07-18 15:51:28 +00001016``dereferenceable(<n>)``
1017 This indicates that the parameter or return pointer is dereferenceable. This
1018 attribute may only be applied to pointer typed parameters. A pointer that
1019 is dereferenceable can be loaded from speculatively without a risk of
1020 trapping. The number of bytes known to be dereferenceable must be provided
1021 in parentheses. It is legal for the number of bytes to be less than the
1022 size of the pointee type. The ``nonnull`` attribute does not imply
1023 dereferenceability (consider a pointer to one element past the end of an
1024 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
1025 ``addrspace(0)`` (which is the default address space).
1026
Sanjoy Das31ea6d12015-04-16 20:29:50 +00001027``dereferenceable_or_null(<n>)``
1028 This indicates that the parameter or return value isn't both
1029 non-null and non-dereferenceable (up to ``<n>`` bytes) at the same
1030 time. All non-null pointers tagged with
1031 ``dereferenceable_or_null(<n>)`` are ``dereferenceable(<n>)``.
1032 For address space 0 ``dereferenceable_or_null(<n>)`` implies that
1033 a pointer is exactly one of ``dereferenceable(<n>)`` or ``null``,
1034 and in other address spaces ``dereferenceable_or_null(<n>)``
1035 implies that a pointer is at least one of ``dereferenceable(<n>)``
1036 or ``null`` (i.e. it may be both ``null`` and
1037 ``dereferenceable(<n>)``). This attribute may only be applied to
1038 pointer typed parameters.
1039
Sean Silvab084af42012-12-07 10:36:55 +00001040.. _gc:
1041
Philip Reamesf80bbff2015-02-25 23:45:20 +00001042Garbage Collector Strategy Names
1043--------------------------------
Sean Silvab084af42012-12-07 10:36:55 +00001044
Philip Reamesf80bbff2015-02-25 23:45:20 +00001045Each function may specify a garbage collector strategy name, which is simply a
Sean Silvab084af42012-12-07 10:36:55 +00001046string:
1047
1048.. code-block:: llvm
1049
1050 define void @f() gc "name" { ... }
1051
Mehdi Amini4a121fa2015-03-14 22:04:06 +00001052The supported values of *name* includes those :ref:`built in to LLVM
Philip Reamesf80bbff2015-02-25 23:45:20 +00001053<builtin-gc-strategies>` and any provided by loaded plugins. Specifying a GC
Mehdi Amini4a121fa2015-03-14 22:04:06 +00001054strategy will cause the compiler to alter its output in order to support the
1055named garbage collection algorithm. Note that LLVM itself does not contain a
Philip Reamesf80bbff2015-02-25 23:45:20 +00001056garbage collector, this functionality is restricted to generating machine code
Mehdi Amini4a121fa2015-03-14 22:04:06 +00001057which can interoperate with a collector provided externally.
Sean Silvab084af42012-12-07 10:36:55 +00001058
Peter Collingbourne3fa50f92013-09-16 01:08:15 +00001059.. _prefixdata:
1060
1061Prefix Data
1062-----------
1063
Peter Collingbourne51d2de72014-12-03 02:08:38 +00001064Prefix data is data associated with a function which the code
1065generator will emit immediately before the function's entrypoint.
1066The purpose of this feature is to allow frontends to associate
1067language-specific runtime metadata with specific functions and make it
1068available through the function pointer while still allowing the
1069function pointer to be called.
Peter Collingbourne3fa50f92013-09-16 01:08:15 +00001070
Peter Collingbourne51d2de72014-12-03 02:08:38 +00001071To access the data for a given function, a program may bitcast the
1072function pointer to a pointer to the constant's type and dereference
1073index -1. This implies that the IR symbol points just past the end of
1074the prefix data. For instance, take the example of a function annotated
1075with a single ``i32``,
1076
1077.. code-block:: llvm
1078
1079 define void @f() prefix i32 123 { ... }
1080
1081The prefix data can be referenced as,
1082
1083.. code-block:: llvm
1084
David Blaikie16a97eb2015-03-04 22:02:58 +00001085 %0 = bitcast void* () @f to i32*
1086 %a = getelementptr inbounds i32, i32* %0, i32 -1
David Blaikiec7aabbb2015-03-04 22:06:14 +00001087 %b = load i32, i32* %a
Peter Collingbourne51d2de72014-12-03 02:08:38 +00001088
1089Prefix data is laid out as if it were an initializer for a global variable
1090of the prefix data's type. The function will be placed such that the
1091beginning of the prefix data is aligned. This means that if the size
1092of the prefix data is not a multiple of the alignment size, the
1093function's entrypoint will not be aligned. If alignment of the
1094function's entrypoint is desired, padding must be added to the prefix
1095data.
1096
1097A function may have prefix data but no body. This has similar semantics
1098to the ``available_externally`` linkage in that the data may be used by the
1099optimizers but will not be emitted in the object file.
1100
1101.. _prologuedata:
1102
1103Prologue Data
1104-------------
1105
1106The ``prologue`` attribute allows arbitrary code (encoded as bytes) to
1107be inserted prior to the function body. This can be used for enabling
1108function hot-patching and instrumentation.
1109
1110To maintain the semantics of ordinary function calls, the prologue data must
Peter Collingbourne3fa50f92013-09-16 01:08:15 +00001111have a particular format. Specifically, it must begin with a sequence of
1112bytes which decode to a sequence of machine instructions, valid for the
1113module's target, which transfer control to the point immediately succeeding
Peter Collingbourne51d2de72014-12-03 02:08:38 +00001114the prologue data, without performing any other visible action. This allows
Peter Collingbourne3fa50f92013-09-16 01:08:15 +00001115the inliner and other passes to reason about the semantics of the function
Peter Collingbourne51d2de72014-12-03 02:08:38 +00001116definition without needing to reason about the prologue data. Obviously this
1117makes the format of the prologue data highly target dependent.
Peter Collingbourne3fa50f92013-09-16 01:08:15 +00001118
Peter Collingbourne51d2de72014-12-03 02:08:38 +00001119A trivial example of valid prologue data for the x86 architecture is ``i8 144``,
Peter Collingbourne3fa50f92013-09-16 01:08:15 +00001120which encodes the ``nop`` instruction:
1121
1122.. code-block:: llvm
1123
Peter Collingbourne51d2de72014-12-03 02:08:38 +00001124 define void @f() prologue i8 144 { ... }
Peter Collingbourne3fa50f92013-09-16 01:08:15 +00001125
Peter Collingbourne51d2de72014-12-03 02:08:38 +00001126Generally prologue data can be formed by encoding a relative branch instruction
1127which skips the metadata, as in this example of valid prologue data for the
Peter Collingbourne3fa50f92013-09-16 01:08:15 +00001128x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1129
1130.. code-block:: llvm
1131
1132 %0 = type <{ i8, i8, i8* }>
1133
Peter Collingbourne51d2de72014-12-03 02:08:38 +00001134 define void @f() prologue %0 <{ i8 235, i8 8, i8* @md}> { ... }
Peter Collingbourne3fa50f92013-09-16 01:08:15 +00001135
Peter Collingbourne51d2de72014-12-03 02:08:38 +00001136A function may have prologue data but no body. This has similar semantics
Peter Collingbourne3fa50f92013-09-16 01:08:15 +00001137to the ``available_externally`` linkage in that the data may be used by the
1138optimizers but will not be emitted in the object file.
1139
David Majnemer7fddecc2015-06-17 20:52:32 +00001140.. _personalityfn:
1141
1142Personality Function
David Majnemerc5ad8a92015-06-17 21:21:16 +00001143--------------------
David Majnemer7fddecc2015-06-17 20:52:32 +00001144
1145The ``personality`` attribute permits functions to specify what function
1146to use for exception handling.
1147
Bill Wendling63b88192013-02-06 06:52:58 +00001148.. _attrgrp:
1149
1150Attribute Groups
1151----------------
1152
1153Attribute groups are groups of attributes that are referenced by objects within
1154the IR. They are important for keeping ``.ll`` files readable, because a lot of
1155functions will use the same set of attributes. In the degenerative case of a
1156``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1157group will capture the important command line flags used to build that file.
1158
1159An attribute group is a module-level object. To use an attribute group, an
1160object references the attribute group's ID (e.g. ``#37``). An object may refer
1161to more than one attribute group. In that situation, the attributes from the
1162different groups are merged.
1163
1164Here is an example of attribute groups for a function that should always be
1165inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1166
1167.. code-block:: llvm
1168
1169 ; Target-independent attributes:
Eli Bendersky97ad9242013-04-18 16:11:44 +00001170 attributes #0 = { alwaysinline alignstack=4 }
Bill Wendling63b88192013-02-06 06:52:58 +00001171
1172 ; Target-dependent attributes:
Eli Bendersky97ad9242013-04-18 16:11:44 +00001173 attributes #1 = { "no-sse" }
Bill Wendling63b88192013-02-06 06:52:58 +00001174
1175 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1176 define void @f() #0 #1 { ... }
1177
Sean Silvab084af42012-12-07 10:36:55 +00001178.. _fnattrs:
1179
1180Function Attributes
1181-------------------
1182
1183Function attributes are set to communicate additional information about
1184a function. Function attributes are considered to be part of the
1185function, not of the function type, so functions with different function
1186attributes can have the same function type.
1187
1188Function attributes are simple keywords that follow the type specified.
1189If multiple attributes are needed, they are space separated. For
1190example:
1191
1192.. code-block:: llvm
1193
1194 define void @f() noinline { ... }
1195 define void @f() alwaysinline { ... }
1196 define void @f() alwaysinline optsize { ... }
1197 define void @f() optsize { ... }
1198
Sean Silvab084af42012-12-07 10:36:55 +00001199``alignstack(<n>)``
1200 This attribute indicates that, when emitting the prologue and
1201 epilogue, the backend should forcibly align the stack pointer.
1202 Specify the desired alignment, which must be a power of two, in
1203 parentheses.
1204``alwaysinline``
1205 This attribute indicates that the inliner should attempt to inline
1206 this function into callers whenever possible, ignoring any active
1207 inlining size threshold for this caller.
Michael Gottesman41748d72013-06-27 00:25:01 +00001208``builtin``
1209 This indicates that the callee function at a call site should be
1210 recognized as a built-in function, even though the function's declaration
Michael Gottesman3a6a9672013-07-02 21:32:56 +00001211 uses the ``nobuiltin`` attribute. This is only valid at call sites for
Richard Smith32dbdf62014-07-31 04:25:36 +00001212 direct calls to functions that are declared with the ``nobuiltin``
Michael Gottesman41748d72013-06-27 00:25:01 +00001213 attribute.
Michael Gottesman296adb82013-06-27 22:48:08 +00001214``cold``
1215 This attribute indicates that this function is rarely called. When
1216 computing edge weights, basic blocks post-dominated by a cold
1217 function call are also considered to be cold; and, thus, given low
1218 weight.
Owen Anderson85fa7d52015-05-26 23:48:40 +00001219``convergent``
1220 This attribute indicates that the callee is dependent on a convergent
1221 thread execution pattern under certain parallel execution models.
1222 Transformations that are execution model agnostic may only move or
1223 tranform this call if the final location is control equivalent to its
1224 original position in the program, where control equivalence is defined as
1225 A dominates B and B post-dominates A, or vice versa.
Sean Silvab084af42012-12-07 10:36:55 +00001226``inlinehint``
1227 This attribute indicates that the source code contained a hint that
1228 inlining this function is desirable (such as the "inline" keyword in
1229 C/C++). It is just a hint; it imposes no requirements on the
1230 inliner.
Tom Roeder44cb65f2014-06-05 19:29:43 +00001231``jumptable``
1232 This attribute indicates that the function should be added to a
1233 jump-instruction table at code-generation time, and that all address-taken
1234 references to this function should be replaced with a reference to the
1235 appropriate jump-instruction-table function pointer. Note that this creates
1236 a new pointer for the original function, which means that code that depends
1237 on function-pointer identity can break. So, any function annotated with
1238 ``jumptable`` must also be ``unnamed_addr``.
Andrea Di Biagio9b5d23b2013-08-09 18:42:18 +00001239``minsize``
1240 This attribute suggests that optimization passes and code generator
1241 passes make choices that keep the code size of this function as small
Andrew Trickd4d1d9c2013-10-31 17:18:07 +00001242 as possible and perform optimizations that may sacrifice runtime
Andrea Di Biagio9b5d23b2013-08-09 18:42:18 +00001243 performance in order to minimize the size of the generated code.
Sean Silvab084af42012-12-07 10:36:55 +00001244``naked``
1245 This attribute disables prologue / epilogue emission for the
1246 function. This can have very system-specific consequences.
Eli Bendersky97ad9242013-04-18 16:11:44 +00001247``nobuiltin``
Michael Gottesman41748d72013-06-27 00:25:01 +00001248 This indicates that the callee function at a call site is not recognized as
1249 a built-in function. LLVM will retain the original call and not replace it
1250 with equivalent code based on the semantics of the built-in function, unless
1251 the call site uses the ``builtin`` attribute. This is valid at call sites
1252 and on function declarations and definitions.
Bill Wendlingbf902f12013-02-06 06:22:58 +00001253``noduplicate``
1254 This attribute indicates that calls to the function cannot be
1255 duplicated. A call to a ``noduplicate`` function may be moved
1256 within its parent function, but may not be duplicated within
1257 its parent function.
1258
1259 A function containing a ``noduplicate`` call may still
1260 be an inlining candidate, provided that the call is not
1261 duplicated by inlining. That implies that the function has
1262 internal linkage and only has one call site, so the original
1263 call is dead after inlining.
Sean Silvab084af42012-12-07 10:36:55 +00001264``noimplicitfloat``
1265 This attributes disables implicit floating point instructions.
1266``noinline``
1267 This attribute indicates that the inliner should never inline this
1268 function in any situation. This attribute may not be used together
1269 with the ``alwaysinline`` attribute.
Sean Silva1cbbcf12013-08-06 19:34:37 +00001270``nonlazybind``
1271 This attribute suppresses lazy symbol binding for the function. This
1272 may make calls to the function faster, at the cost of extra program
1273 startup time if the function is not called during program startup.
Sean Silvab084af42012-12-07 10:36:55 +00001274``noredzone``
1275 This attribute indicates that the code generator should not use a
1276 red zone, even if the target-specific ABI normally permits it.
1277``noreturn``
1278 This function attribute indicates that the function never returns
1279 normally. This produces undefined behavior at runtime if the
1280 function ever does dynamically return.
1281``nounwind``
Reid Kleckner96d01132015-02-11 01:23:16 +00001282 This function attribute indicates that the function never raises an
1283 exception. If the function does raise an exception, its runtime
1284 behavior is undefined. However, functions marked nounwind may still
1285 trap or generate asynchronous exceptions. Exception handling schemes
1286 that are recognized by LLVM to handle asynchronous exceptions, such
1287 as SEH, will still provide their implementation defined semantics.
Andrea Di Biagio377496b2013-08-23 11:53:55 +00001288``optnone``
1289 This function attribute indicates that the function is not optimized
Andrew Trickd4d1d9c2013-10-31 17:18:07 +00001290 by any optimization or code generator passes with the
Andrea Di Biagio377496b2013-08-23 11:53:55 +00001291 exception of interprocedural optimization passes.
1292 This attribute cannot be used together with the ``alwaysinline``
1293 attribute; this attribute is also incompatible
1294 with the ``minsize`` attribute and the ``optsize`` attribute.
Andrew Trickd4d1d9c2013-10-31 17:18:07 +00001295
Paul Robinsondcbe35b2013-11-18 21:44:03 +00001296 This attribute requires the ``noinline`` attribute to be specified on
1297 the function as well, so the function is never inlined into any caller.
Andrea Di Biagio377496b2013-08-23 11:53:55 +00001298 Only functions with the ``alwaysinline`` attribute are valid
Paul Robinsondcbe35b2013-11-18 21:44:03 +00001299 candidates for inlining into the body of this function.
Sean Silvab084af42012-12-07 10:36:55 +00001300``optsize``
1301 This attribute suggests that optimization passes and code generator
1302 passes make choices that keep the code size of this function low,
Andrea Di Biagio9b5d23b2013-08-09 18:42:18 +00001303 and otherwise do optimizations specifically to reduce code size as
1304 long as they do not significantly impact runtime performance.
Sean Silvab084af42012-12-07 10:36:55 +00001305``readnone``
Nick Lewyckyc2ec0722013-07-06 00:29:58 +00001306 On a function, this attribute indicates that the function computes its
1307 result (or decides to unwind an exception) based strictly on its arguments,
Sean Silvab084af42012-12-07 10:36:55 +00001308 without dereferencing any pointer arguments or otherwise accessing
1309 any mutable state (e.g. memory, control registers, etc) visible to
1310 caller functions. It does not write through any pointer arguments
1311 (including ``byval`` arguments) and never changes any state visible
1312 to callers. This means that it cannot unwind exceptions by calling
1313 the ``C++`` exception throwing methods.
Andrew Trickd4d1d9c2013-10-31 17:18:07 +00001314
Nick Lewyckyc2ec0722013-07-06 00:29:58 +00001315 On an argument, this attribute indicates that the function does not
1316 dereference that pointer argument, even though it may read or write the
Nick Lewyckyefe31f22013-07-06 01:04:47 +00001317 memory that the pointer points to if accessed through other pointers.
Sean Silvab084af42012-12-07 10:36:55 +00001318``readonly``
Nick Lewyckyc2ec0722013-07-06 00:29:58 +00001319 On a function, this attribute indicates that the function does not write
1320 through any pointer arguments (including ``byval`` arguments) or otherwise
Sean Silvab084af42012-12-07 10:36:55 +00001321 modify any state (e.g. memory, control registers, etc) visible to
1322 caller functions. It may dereference pointer arguments and read
1323 state that may be set in the caller. A readonly function always
1324 returns the same value (or unwinds an exception identically) when
1325 called with the same set of arguments and global state. It cannot
1326 unwind an exception by calling the ``C++`` exception throwing
1327 methods.
Andrew Trickd4d1d9c2013-10-31 17:18:07 +00001328
Nick Lewyckyc2ec0722013-07-06 00:29:58 +00001329 On an argument, this attribute indicates that the function does not write
1330 through this pointer argument, even though it may write to the memory that
1331 the pointer points to.
Igor Laevsky39d662f2015-07-11 10:30:36 +00001332``argmemonly``
1333 This attribute indicates that the only memory accesses inside function are
1334 loads and stores from objects pointed to by its pointer-typed arguments,
1335 with arbitrary offsets. Or in other words, all memory operations in the
1336 function can refer to memory only using pointers based on its function
1337 arguments.
1338 Note that ``argmemonly`` can be used together with ``readonly`` attribute
1339 in order to specify that function reads only from its arguments.
Sean Silvab084af42012-12-07 10:36:55 +00001340``returns_twice``
1341 This attribute indicates that this function can return twice. The C
1342 ``setjmp`` is an example of such a function. The compiler disables
1343 some optimizations (like tail calls) in the caller of these
1344 functions.
Peter Collingbourne82437bf2015-06-15 21:07:11 +00001345``safestack``
1346 This attribute indicates that
1347 `SafeStack <http://clang.llvm.org/docs/SafeStack.html>`_
1348 protection is enabled for this function.
1349
1350 If a function that has a ``safestack`` attribute is inlined into a
1351 function that doesn't have a ``safestack`` attribute or which has an
1352 ``ssp``, ``sspstrong`` or ``sspreq`` attribute, then the resulting
1353 function will have a ``safestack`` attribute.
Kostya Serebryanycf880b92013-02-26 06:58:09 +00001354``sanitize_address``
1355 This attribute indicates that AddressSanitizer checks
1356 (dynamic address safety analysis) are enabled for this function.
1357``sanitize_memory``
1358 This attribute indicates that MemorySanitizer checks (dynamic detection
1359 of accesses to uninitialized memory) are enabled for this function.
1360``sanitize_thread``
1361 This attribute indicates that ThreadSanitizer checks
1362 (dynamic thread safety analysis) are enabled for this function.
Sean Silvab084af42012-12-07 10:36:55 +00001363``ssp``
1364 This attribute indicates that the function should emit a stack
Dmitri Gribenkoe8131122013-01-19 20:34:20 +00001365 smashing protector. It is in the form of a "canary" --- a random value
Sean Silvab084af42012-12-07 10:36:55 +00001366 placed on the stack before the local variables that's checked upon
1367 return from the function to see if it has been overwritten. A
1368 heuristic is used to determine if a function needs stack protectors
Bill Wendling7c8f96a2013-01-23 06:43:53 +00001369 or not. The heuristic used will enable protectors for functions with:
Dmitri Gribenko69b56472013-01-29 23:14:41 +00001370
Bill Wendling7c8f96a2013-01-23 06:43:53 +00001371 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1372 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1373 - Calls to alloca() with variable sizes or constant sizes greater than
1374 ``ssp-buffer-size``.
Sean Silvab084af42012-12-07 10:36:55 +00001375
Josh Magee24c7f062014-02-01 01:36:16 +00001376 Variables that are identified as requiring a protector will be arranged
1377 on the stack such that they are adjacent to the stack protector guard.
1378
Sean Silvab084af42012-12-07 10:36:55 +00001379 If a function that has an ``ssp`` attribute is inlined into a
1380 function that doesn't have an ``ssp`` attribute, then the resulting
1381 function will have an ``ssp`` attribute.
1382``sspreq``
1383 This attribute indicates that the function should *always* emit a
1384 stack smashing protector. This overrides the ``ssp`` function
1385 attribute.
1386
Josh Magee24c7f062014-02-01 01:36:16 +00001387 Variables that are identified as requiring a protector will be arranged
1388 on the stack such that they are adjacent to the stack protector guard.
1389 The specific layout rules are:
1390
1391 #. Large arrays and structures containing large arrays
1392 (``>= ssp-buffer-size``) are closest to the stack protector.
1393 #. Small arrays and structures containing small arrays
1394 (``< ssp-buffer-size``) are 2nd closest to the protector.
1395 #. Variables that have had their address taken are 3rd closest to the
1396 protector.
1397
Sean Silvab084af42012-12-07 10:36:55 +00001398 If a function that has an ``sspreq`` attribute is inlined into a
1399 function that doesn't have an ``sspreq`` attribute or which has an
Bill Wendlingd154e2832013-01-23 06:41:41 +00001400 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1401 an ``sspreq`` attribute.
1402``sspstrong``
1403 This attribute indicates that the function should emit a stack smashing
Bill Wendling7c8f96a2013-01-23 06:43:53 +00001404 protector. This attribute causes a strong heuristic to be used when
1405 determining if a function needs stack protectors. The strong heuristic
1406 will enable protectors for functions with:
Dmitri Gribenko69b56472013-01-29 23:14:41 +00001407
Bill Wendling7c8f96a2013-01-23 06:43:53 +00001408 - Arrays of any size and type
1409 - Aggregates containing an array of any size and type.
1410 - Calls to alloca().
1411 - Local variables that have had their address taken.
1412
Josh Magee24c7f062014-02-01 01:36:16 +00001413 Variables that are identified as requiring a protector will be arranged
1414 on the stack such that they are adjacent to the stack protector guard.
1415 The specific layout rules are:
1416
1417 #. Large arrays and structures containing large arrays
1418 (``>= ssp-buffer-size``) are closest to the stack protector.
1419 #. Small arrays and structures containing small arrays
1420 (``< ssp-buffer-size``) are 2nd closest to the protector.
1421 #. Variables that have had their address taken are 3rd closest to the
1422 protector.
1423
Bill Wendling7c8f96a2013-01-23 06:43:53 +00001424 This overrides the ``ssp`` function attribute.
Bill Wendlingd154e2832013-01-23 06:41:41 +00001425
1426 If a function that has an ``sspstrong`` attribute is inlined into a
1427 function that doesn't have an ``sspstrong`` attribute, then the
1428 resulting function will have an ``sspstrong`` attribute.
Reid Kleckner5a2ab2b2015-03-04 00:08:56 +00001429``"thunk"``
1430 This attribute indicates that the function will delegate to some other
1431 function with a tail call. The prototype of a thunk should not be used for
1432 optimization purposes. The caller is expected to cast the thunk prototype to
1433 match the thunk target prototype.
Sean Silvab084af42012-12-07 10:36:55 +00001434``uwtable``
1435 This attribute indicates that the ABI being targeted requires that
Sean Silva706fba52015-08-06 22:56:24 +00001436 an unwind table entry be produced for this function even if we can
Sean Silvab084af42012-12-07 10:36:55 +00001437 show that no exceptions passes by it. This is normally the case for
1438 the ELF x86-64 abi, but it can be disabled for some compilation
1439 units.
Sean Silvab084af42012-12-07 10:36:55 +00001440
1441.. _moduleasm:
1442
1443Module-Level Inline Assembly
1444----------------------------
1445
1446Modules may contain "module-level inline asm" blocks, which corresponds
1447to the GCC "file scope inline asm" blocks. These blocks are internally
1448concatenated by LLVM and treated as a single unit, but may be separated
1449in the ``.ll`` file if desired. The syntax is very simple:
1450
1451.. code-block:: llvm
1452
1453 module asm "inline asm code goes here"
1454 module asm "more can go here"
1455
1456The strings can contain any character by escaping non-printable
1457characters. The escape sequence used is simply "\\xx" where "xx" is the
1458two digit hex code for the number.
1459
James Y Knightbc832ed2015-07-08 18:08:36 +00001460Note that the assembly string *must* be parseable by LLVM's integrated assembler
1461(unless it is disabled), even when emitting a ``.s`` file.
Sean Silvab084af42012-12-07 10:36:55 +00001462
Eli Benderskyfdc529a2013-06-07 19:40:08 +00001463.. _langref_datalayout:
1464
Sean Silvab084af42012-12-07 10:36:55 +00001465Data Layout
1466-----------
1467
1468A module may specify a target specific data layout string that specifies
1469how data is to be laid out in memory. The syntax for the data layout is
1470simply:
1471
1472.. code-block:: llvm
1473
1474 target datalayout = "layout specification"
1475
1476The *layout specification* consists of a list of specifications
1477separated by the minus sign character ('-'). Each specification starts
1478with a letter and may include other information after the letter to
1479define some aspect of the data layout. The specifications accepted are
1480as follows:
1481
1482``E``
1483 Specifies that the target lays out data in big-endian form. That is,
1484 the bits with the most significance have the lowest address
1485 location.
1486``e``
1487 Specifies that the target lays out data in little-endian form. That
1488 is, the bits with the least significance have the lowest address
1489 location.
1490``S<size>``
1491 Specifies the natural alignment of the stack in bits. Alignment
1492 promotion of stack variables is limited to the natural stack
1493 alignment to avoid dynamic stack realignment. The stack alignment
1494 must be a multiple of 8-bits. If omitted, the natural stack
1495 alignment defaults to "unspecified", which does not prevent any
1496 alignment promotions.
1497``p[n]:<size>:<abi>:<pref>``
1498 This specifies the *size* of a pointer and its ``<abi>`` and
1499 ``<pref>``\erred alignments for address space ``n``. All sizes are in
Sean Silva706fba52015-08-06 22:56:24 +00001500 bits. The address space, ``n``, is optional, and if not specified,
Rafael Espindolaabdd7262014-01-06 21:40:24 +00001501 denotes the default address space 0. The value of ``n`` must be
1502 in the range [1,2^23).
Sean Silvab084af42012-12-07 10:36:55 +00001503``i<size>:<abi>:<pref>``
1504 This specifies the alignment for an integer type of a given bit
1505 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1506``v<size>:<abi>:<pref>``
1507 This specifies the alignment for a vector type of a given bit
1508 ``<size>``.
1509``f<size>:<abi>:<pref>``
1510 This specifies the alignment for a floating point type of a given bit
1511 ``<size>``. Only values of ``<size>`` that are supported by the target
1512 will work. 32 (float) and 64 (double) are supported on all targets; 80
1513 or 128 (different flavors of long double) are also supported on some
1514 targets.
Rafael Espindolaabdd7262014-01-06 21:40:24 +00001515``a:<abi>:<pref>``
1516 This specifies the alignment for an object of aggregate type.
Rafael Espindola58873562014-01-03 19:21:54 +00001517``m:<mangling>``
Hans Wennborgd4245ac2014-01-15 02:49:17 +00001518 If present, specifies that llvm names are mangled in the output. The
1519 options are
1520
1521 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1522 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1523 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1524 symbols get a ``_`` prefix.
1525 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1526 functions also get a suffix based on the frame size.
Sean Silvab084af42012-12-07 10:36:55 +00001527``n<size1>:<size2>:<size3>...``
1528 This specifies a set of native integer widths for the target CPU in
1529 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1530 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1531 this set are considered to support most general arithmetic operations
1532 efficiently.
1533
Rafael Espindolaabdd7262014-01-06 21:40:24 +00001534On every specification that takes a ``<abi>:<pref>``, specifying the
1535``<pref>`` alignment is optional. If omitted, the preceding ``:``
1536should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1537
Sean Silvab084af42012-12-07 10:36:55 +00001538When constructing the data layout for a given target, LLVM starts with a
1539default set of specifications which are then (possibly) overridden by
1540the specifications in the ``datalayout`` keyword. The default
1541specifications are given in this list:
1542
1543- ``E`` - big endian
Matt Arsenault24b49c42013-07-31 17:49:08 +00001544- ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1545- ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1546 same as the default address space.
Patrik Hagglunda832ab12013-01-30 09:02:06 +00001547- ``S0`` - natural stack alignment is unspecified
Sean Silvab084af42012-12-07 10:36:55 +00001548- ``i1:8:8`` - i1 is 8-bit (byte) aligned
1549- ``i8:8:8`` - i8 is 8-bit (byte) aligned
1550- ``i16:16:16`` - i16 is 16-bit aligned
1551- ``i32:32:32`` - i32 is 32-bit aligned
1552- ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1553 alignment of 64-bits
Patrik Hagglunda832ab12013-01-30 09:02:06 +00001554- ``f16:16:16`` - half is 16-bit aligned
Sean Silvab084af42012-12-07 10:36:55 +00001555- ``f32:32:32`` - float is 32-bit aligned
1556- ``f64:64:64`` - double is 64-bit aligned
Patrik Hagglunda832ab12013-01-30 09:02:06 +00001557- ``f128:128:128`` - quad is 128-bit aligned
Sean Silvab084af42012-12-07 10:36:55 +00001558- ``v64:64:64`` - 64-bit vector is 64-bit aligned
1559- ``v128:128:128`` - 128-bit vector is 128-bit aligned
Rafael Espindolae8f4d582013-12-12 17:21:51 +00001560- ``a:0:64`` - aggregates are 64-bit aligned
Sean Silvab084af42012-12-07 10:36:55 +00001561
1562When LLVM is determining the alignment for a given type, it uses the
1563following rules:
1564
1565#. If the type sought is an exact match for one of the specifications,
1566 that specification is used.
1567#. If no match is found, and the type sought is an integer type, then
1568 the smallest integer type that is larger than the bitwidth of the
1569 sought type is used. If none of the specifications are larger than
1570 the bitwidth then the largest integer type is used. For example,
1571 given the default specifications above, the i7 type will use the
1572 alignment of i8 (next largest) while both i65 and i256 will use the
1573 alignment of i64 (largest specified).
1574#. If no match is found, and the type sought is a vector type, then the
1575 largest vector type that is smaller than the sought vector type will
1576 be used as a fall back. This happens because <128 x double> can be
1577 implemented in terms of 64 <2 x double>, for example.
1578
1579The function of the data layout string may not be what you expect.
1580Notably, this is not a specification from the frontend of what alignment
1581the code generator should use.
1582
1583Instead, if specified, the target data layout is required to match what
1584the ultimate *code generator* expects. This string is used by the
1585mid-level optimizers to improve code, and this only works if it matches
Mehdi Amini4a121fa2015-03-14 22:04:06 +00001586what the ultimate code generator uses. There is no way to generate IR
1587that does not embed this target-specific detail into the IR. If you
1588don't specify the string, the default specifications will be used to
1589generate a Data Layout and the optimization phases will operate
1590accordingly and introduce target specificity into the IR with respect to
1591these default specifications.
Sean Silvab084af42012-12-07 10:36:55 +00001592
Bill Wendling5cc90842013-10-18 23:41:25 +00001593.. _langref_triple:
1594
1595Target Triple
1596-------------
1597
1598A module may specify a target triple string that describes the target
1599host. The syntax for the target triple is simply:
1600
1601.. code-block:: llvm
1602
1603 target triple = "x86_64-apple-macosx10.7.0"
1604
1605The *target triple* string consists of a series of identifiers delimited
1606by the minus sign character ('-'). The canonical forms are:
1607
1608::
1609
1610 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1611 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1612
1613This information is passed along to the backend so that it generates
1614code for the proper architecture. It's possible to override this on the
1615command line with the ``-mtriple`` command line option.
1616
Sean Silvab084af42012-12-07 10:36:55 +00001617.. _pointeraliasing:
1618
1619Pointer Aliasing Rules
1620----------------------
1621
1622Any memory access must be done through a pointer value associated with
1623an address range of the memory access, otherwise the behavior is
1624undefined. Pointer values are associated with address ranges according
1625to the following rules:
1626
1627- A pointer value is associated with the addresses associated with any
1628 value it is *based* on.
1629- An address of a global variable is associated with the address range
1630 of the variable's storage.
1631- The result value of an allocation instruction is associated with the
1632 address range of the allocated storage.
1633- A null pointer in the default address-space is associated with no
1634 address.
1635- An integer constant other than zero or a pointer value returned from
1636 a function not defined within LLVM may be associated with address
1637 ranges allocated through mechanisms other than those provided by
1638 LLVM. Such ranges shall not overlap with any ranges of addresses
1639 allocated by mechanisms provided by LLVM.
1640
1641A pointer value is *based* on another pointer value according to the
1642following rules:
1643
1644- A pointer value formed from a ``getelementptr`` operation is *based*
David Blaikie16a97eb2015-03-04 22:02:58 +00001645 on the first value operand of the ``getelementptr``.
Sean Silvab084af42012-12-07 10:36:55 +00001646- The result value of a ``bitcast`` is *based* on the operand of the
1647 ``bitcast``.
1648- A pointer value formed by an ``inttoptr`` is *based* on all pointer
1649 values that contribute (directly or indirectly) to the computation of
1650 the pointer's value.
1651- The "*based* on" relationship is transitive.
1652
1653Note that this definition of *"based"* is intentionally similar to the
1654definition of *"based"* in C99, though it is slightly weaker.
1655
1656LLVM IR does not associate types with memory. The result type of a
1657``load`` merely indicates the size and alignment of the memory from
1658which to load, as well as the interpretation of the value. The first
1659operand type of a ``store`` similarly only indicates the size and
1660alignment of the store.
1661
1662Consequently, type-based alias analysis, aka TBAA, aka
1663``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1664:ref:`Metadata <metadata>` may be used to encode additional information
1665which specialized optimization passes may use to implement type-based
1666alias analysis.
1667
1668.. _volatile:
1669
1670Volatile Memory Accesses
1671------------------------
1672
1673Certain memory accesses, such as :ref:`load <i_load>`'s,
1674:ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1675marked ``volatile``. The optimizers must not change the number of
1676volatile operations or change their order of execution relative to other
1677volatile operations. The optimizers *may* change the order of volatile
1678operations relative to non-volatile operations. This is not Java's
1679"volatile" and has no cross-thread synchronization behavior.
1680
Andrew Trick89fc5a62013-01-30 21:19:35 +00001681IR-level volatile loads and stores cannot safely be optimized into
1682llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1683flagged volatile. Likewise, the backend should never split or merge
1684target-legal volatile load/store instructions.
1685
Andrew Trick7e6f9282013-01-31 00:49:39 +00001686.. admonition:: Rationale
1687
1688 Platforms may rely on volatile loads and stores of natively supported
1689 data width to be executed as single instruction. For example, in C
1690 this holds for an l-value of volatile primitive type with native
1691 hardware support, but not necessarily for aggregate types. The
1692 frontend upholds these expectations, which are intentionally
Sean Silva706fba52015-08-06 22:56:24 +00001693 unspecified in the IR. The rules above ensure that IR transformations
Andrew Trick7e6f9282013-01-31 00:49:39 +00001694 do not violate the frontend's contract with the language.
1695
Sean Silvab084af42012-12-07 10:36:55 +00001696.. _memmodel:
1697
1698Memory Model for Concurrent Operations
1699--------------------------------------
1700
1701The LLVM IR does not define any way to start parallel threads of
1702execution or to register signal handlers. Nonetheless, there are
1703platform-specific ways to create them, and we define LLVM IR's behavior
1704in their presence. This model is inspired by the C++0x memory model.
1705
1706For a more informal introduction to this model, see the :doc:`Atomics`.
1707
1708We define a *happens-before* partial order as the least partial order
1709that
1710
1711- Is a superset of single-thread program order, and
1712- When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1713 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1714 techniques, like pthread locks, thread creation, thread joining,
1715 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1716 Constraints <ordering>`).
1717
1718Note that program order does not introduce *happens-before* edges
1719between a thread and signals executing inside that thread.
1720
1721Every (defined) read operation (load instructions, memcpy, atomic
1722loads/read-modify-writes, etc.) R reads a series of bytes written by
1723(defined) write operations (store instructions, atomic
1724stores/read-modify-writes, memcpy, etc.). For the purposes of this
1725section, initialized globals are considered to have a write of the
1726initializer which is atomic and happens before any other read or write
1727of the memory in question. For each byte of a read R, R\ :sub:`byte`
1728may see any write to the same byte, except:
1729
1730- If write\ :sub:`1` happens before write\ :sub:`2`, and
1731 write\ :sub:`2` happens before R\ :sub:`byte`, then
1732 R\ :sub:`byte` does not see write\ :sub:`1`.
1733- If R\ :sub:`byte` happens before write\ :sub:`3`, then
1734 R\ :sub:`byte` does not see write\ :sub:`3`.
1735
1736Given that definition, R\ :sub:`byte` is defined as follows:
1737
1738- If R is volatile, the result is target-dependent. (Volatile is
1739 supposed to give guarantees which can support ``sig_atomic_t`` in
Richard Smith32dbdf62014-07-31 04:25:36 +00001740 C/C++, and may be used for accesses to addresses that do not behave
Sean Silvab084af42012-12-07 10:36:55 +00001741 like normal memory. It does not generally provide cross-thread
1742 synchronization.)
1743- Otherwise, if there is no write to the same byte that happens before
1744 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1745- Otherwise, if R\ :sub:`byte` may see exactly one write,
1746 R\ :sub:`byte` returns the value written by that write.
1747- Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1748 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1749 Memory Ordering Constraints <ordering>` section for additional
1750 constraints on how the choice is made.
1751- Otherwise R\ :sub:`byte` returns ``undef``.
1752
1753R returns the value composed of the series of bytes it read. This
1754implies that some bytes within the value may be ``undef`` **without**
1755the entire value being ``undef``. Note that this only defines the
1756semantics of the operation; it doesn't mean that targets will emit more
1757than one instruction to read the series of bytes.
1758
1759Note that in cases where none of the atomic intrinsics are used, this
1760model places only one restriction on IR transformations on top of what
1761is required for single-threaded execution: introducing a store to a byte
1762which might not otherwise be stored is not allowed in general.
1763(Specifically, in the case where another thread might write to and read
1764from an address, introducing a store can change a load that may see
1765exactly one write into a load that may see multiple writes.)
1766
1767.. _ordering:
1768
1769Atomic Memory Ordering Constraints
1770----------------------------------
1771
1772Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1773:ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1774:ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
Tim Northovere94a5182014-03-11 10:48:52 +00001775ordering parameters that determine which other atomic instructions on
Sean Silvab084af42012-12-07 10:36:55 +00001776the same address they *synchronize with*. These semantics are borrowed
1777from Java and C++0x, but are somewhat more colloquial. If these
1778descriptions aren't precise enough, check those specs (see spec
1779references in the :doc:`atomics guide <Atomics>`).
1780:ref:`fence <i_fence>` instructions treat these orderings somewhat
1781differently since they don't take an address. See that instruction's
1782documentation for details.
1783
1784For a simpler introduction to the ordering constraints, see the
1785:doc:`Atomics`.
1786
1787``unordered``
1788 The set of values that can be read is governed by the happens-before
1789 partial order. A value cannot be read unless some operation wrote
1790 it. This is intended to provide a guarantee strong enough to model
1791 Java's non-volatile shared variables. This ordering cannot be
1792 specified for read-modify-write operations; it is not strong enough
1793 to make them atomic in any interesting way.
1794``monotonic``
1795 In addition to the guarantees of ``unordered``, there is a single
1796 total order for modifications by ``monotonic`` operations on each
1797 address. All modification orders must be compatible with the
1798 happens-before order. There is no guarantee that the modification
1799 orders can be combined to a global total order for the whole program
1800 (and this often will not be possible). The read in an atomic
1801 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1802 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1803 order immediately before the value it writes. If one atomic read
1804 happens before another atomic read of the same address, the later
1805 read must see the same value or a later value in the address's
1806 modification order. This disallows reordering of ``monotonic`` (or
1807 stronger) operations on the same address. If an address is written
1808 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1809 read that address repeatedly, the other threads must eventually see
1810 the write. This corresponds to the C++0x/C1x
1811 ``memory_order_relaxed``.
1812``acquire``
1813 In addition to the guarantees of ``monotonic``, a
1814 *synchronizes-with* edge may be formed with a ``release`` operation.
1815 This is intended to model C++'s ``memory_order_acquire``.
1816``release``
1817 In addition to the guarantees of ``monotonic``, if this operation
1818 writes a value which is subsequently read by an ``acquire``
1819 operation, it *synchronizes-with* that operation. (This isn't a
1820 complete description; see the C++0x definition of a release
1821 sequence.) This corresponds to the C++0x/C1x
1822 ``memory_order_release``.
1823``acq_rel`` (acquire+release)
1824 Acts as both an ``acquire`` and ``release`` operation on its
1825 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1826``seq_cst`` (sequentially consistent)
1827 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
Richard Smith32dbdf62014-07-31 04:25:36 +00001828 operation that only reads, ``release`` for an operation that only
Sean Silvab084af42012-12-07 10:36:55 +00001829 writes), there is a global total order on all
1830 sequentially-consistent operations on all addresses, which is
1831 consistent with the *happens-before* partial order and with the
1832 modification orders of all the affected addresses. Each
1833 sequentially-consistent read sees the last preceding write to the
1834 same address in this global order. This corresponds to the C++0x/C1x
1835 ``memory_order_seq_cst`` and Java volatile.
1836
1837.. _singlethread:
1838
1839If an atomic operation is marked ``singlethread``, it only *synchronizes
1840with* or participates in modification and seq\_cst total orderings with
1841other operations running in the same thread (for example, in signal
1842handlers).
1843
1844.. _fastmath:
1845
1846Fast-Math Flags
1847---------------
1848
1849LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1850:ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
James Molloy88eb5352015-07-10 12:52:00 +00001851:ref:`frem <i_frem>`, :ref:`fcmp <i_fcmp>`) have the following flags that can
1852be set to enable otherwise unsafe floating point operations
Sean Silvab084af42012-12-07 10:36:55 +00001853
1854``nnan``
1855 No NaNs - Allow optimizations to assume the arguments and result are not
1856 NaN. Such optimizations are required to retain defined behavior over
1857 NaNs, but the value of the result is undefined.
1858
1859``ninf``
1860 No Infs - Allow optimizations to assume the arguments and result are not
1861 +/-Inf. Such optimizations are required to retain defined behavior over
1862 +/-Inf, but the value of the result is undefined.
1863
1864``nsz``
1865 No Signed Zeros - Allow optimizations to treat the sign of a zero
1866 argument or result as insignificant.
1867
1868``arcp``
1869 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1870 argument rather than perform division.
1871
1872``fast``
1873 Fast - Allow algebraically equivalent transformations that may
1874 dramatically change results in floating point (e.g. reassociate). This
1875 flag implies all the others.
1876
Duncan P. N. Exon Smith0a448fb2014-08-19 21:30:15 +00001877.. _uselistorder:
1878
1879Use-list Order Directives
1880-------------------------
1881
1882Use-list directives encode the in-memory order of each use-list, allowing the
1883order to be recreated. ``<order-indexes>`` is a comma-separated list of
1884indexes that are assigned to the referenced value's uses. The referenced
1885value's use-list is immediately sorted by these indexes.
1886
1887Use-list directives may appear at function scope or global scope. They are not
1888instructions, and have no effect on the semantics of the IR. When they're at
1889function scope, they must appear after the terminator of the final basic block.
1890
1891If basic blocks have their address taken via ``blockaddress()`` expressions,
1892``uselistorder_bb`` can be used to reorder their use-lists from outside their
1893function's scope.
1894
1895:Syntax:
1896
1897::
1898
1899 uselistorder <ty> <value>, { <order-indexes> }
1900 uselistorder_bb @function, %block { <order-indexes> }
1901
1902:Examples:
1903
1904::
1905
Duncan P. N. Exon Smith23046652014-08-19 21:48:04 +00001906 define void @foo(i32 %arg1, i32 %arg2) {
1907 entry:
1908 ; ... instructions ...
1909 bb:
1910 ; ... instructions ...
1911
1912 ; At function scope.
1913 uselistorder i32 %arg1, { 1, 0, 2 }
1914 uselistorder label %bb, { 1, 0 }
1915 }
Duncan P. N. Exon Smith0a448fb2014-08-19 21:30:15 +00001916
1917 ; At global scope.
1918 uselistorder i32* @global, { 1, 2, 0 }
1919 uselistorder i32 7, { 1, 0 }
1920 uselistorder i32 (i32) @bar, { 1, 0 }
1921 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1922
Sean Silvab084af42012-12-07 10:36:55 +00001923.. _typesystem:
1924
1925Type System
1926===========
1927
1928The LLVM type system is one of the most important features of the
1929intermediate representation. Being typed enables a number of
1930optimizations to be performed on the intermediate representation
1931directly, without having to do extra analyses on the side before the
1932transformation. A strong type system makes it easier to read the
1933generated code and enables novel analyses and transformations that are
1934not feasible to perform on normal three address code representations.
1935
Rafael Espindola08013342013-12-07 19:34:20 +00001936.. _t_void:
Eli Bendersky0220e6b2013-06-07 20:24:43 +00001937
Rafael Espindola08013342013-12-07 19:34:20 +00001938Void Type
1939---------
Sean Silvab084af42012-12-07 10:36:55 +00001940
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00001941:Overview:
1942
Rafael Espindola08013342013-12-07 19:34:20 +00001943
1944The void type does not represent any value and has no size.
1945
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00001946:Syntax:
1947
Rafael Espindola08013342013-12-07 19:34:20 +00001948
1949::
1950
1951 void
Sean Silvab084af42012-12-07 10:36:55 +00001952
1953
Rafael Espindola08013342013-12-07 19:34:20 +00001954.. _t_function:
Sean Silvab084af42012-12-07 10:36:55 +00001955
Rafael Espindola08013342013-12-07 19:34:20 +00001956Function Type
1957-------------
Sean Silvab084af42012-12-07 10:36:55 +00001958
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00001959:Overview:
1960
Sean Silvab084af42012-12-07 10:36:55 +00001961
Rafael Espindola08013342013-12-07 19:34:20 +00001962The function type can be thought of as a function signature. It consists of a
1963return type and a list of formal parameter types. The return type of a function
1964type is a void type or first class type --- except for :ref:`label <t_label>`
1965and :ref:`metadata <t_metadata>` types.
Sean Silvab084af42012-12-07 10:36:55 +00001966
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00001967:Syntax:
Sean Silvab084af42012-12-07 10:36:55 +00001968
Rafael Espindola08013342013-12-07 19:34:20 +00001969::
Sean Silvab084af42012-12-07 10:36:55 +00001970
Rafael Espindola08013342013-12-07 19:34:20 +00001971 <returntype> (<parameter list>)
Sean Silvab084af42012-12-07 10:36:55 +00001972
Rafael Espindola08013342013-12-07 19:34:20 +00001973...where '``<parameter list>``' is a comma-separated list of type
1974specifiers. Optionally, the parameter list may include a type ``...``, which
1975indicates that the function takes a variable number of arguments. Variable
1976argument functions can access their arguments with the :ref:`variable argument
1977handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1978except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
Sean Silvab084af42012-12-07 10:36:55 +00001979
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00001980:Examples:
Sean Silvab084af42012-12-07 10:36:55 +00001981
Rafael Espindola08013342013-12-07 19:34:20 +00001982+---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1983| ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1984+---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1985| ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1986+---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1987| ``i32 (i8*, ...)`` | A vararg function that takes at least one :ref:`pointer <t_pointer>` to ``i8`` (char in C), which returns an integer. This is the signature for ``printf`` in LLVM. |
1988+---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1989| ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1990+---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1991
1992.. _t_firstclass:
1993
1994First Class Types
1995-----------------
Sean Silvab084af42012-12-07 10:36:55 +00001996
1997The :ref:`first class <t_firstclass>` types are perhaps the most important.
1998Values of these types are the only ones which can be produced by
1999instructions.
2000
Rafael Espindola08013342013-12-07 19:34:20 +00002001.. _t_single_value:
Sean Silvab084af42012-12-07 10:36:55 +00002002
Rafael Espindola08013342013-12-07 19:34:20 +00002003Single Value Types
2004^^^^^^^^^^^^^^^^^^
Sean Silvab084af42012-12-07 10:36:55 +00002005
Rafael Espindola08013342013-12-07 19:34:20 +00002006These are the types that are valid in registers from CodeGen's perspective.
Sean Silvab084af42012-12-07 10:36:55 +00002007
2008.. _t_integer:
2009
2010Integer Type
Rafael Espindola08013342013-12-07 19:34:20 +00002011""""""""""""
Sean Silvab084af42012-12-07 10:36:55 +00002012
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002013:Overview:
Sean Silvab084af42012-12-07 10:36:55 +00002014
2015The integer type is a very simple type that simply specifies an
2016arbitrary bit width for the integer type desired. Any bit width from 1
2017bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
2018
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002019:Syntax:
Sean Silvab084af42012-12-07 10:36:55 +00002020
2021::
2022
2023 iN
2024
2025The number of bits the integer will occupy is specified by the ``N``
2026value.
2027
2028Examples:
Rafael Espindola08013342013-12-07 19:34:20 +00002029*********
Sean Silvab084af42012-12-07 10:36:55 +00002030
2031+----------------+------------------------------------------------+
2032| ``i1`` | a single-bit integer. |
2033+----------------+------------------------------------------------+
2034| ``i32`` | a 32-bit integer. |
2035+----------------+------------------------------------------------+
2036| ``i1942652`` | a really big integer of over 1 million bits. |
2037+----------------+------------------------------------------------+
2038
2039.. _t_floating:
2040
2041Floating Point Types
Rafael Espindola08013342013-12-07 19:34:20 +00002042""""""""""""""""""""
Sean Silvab084af42012-12-07 10:36:55 +00002043
2044.. list-table::
2045 :header-rows: 1
2046
2047 * - Type
2048 - Description
2049
2050 * - ``half``
2051 - 16-bit floating point value
2052
2053 * - ``float``
2054 - 32-bit floating point value
2055
2056 * - ``double``
2057 - 64-bit floating point value
2058
2059 * - ``fp128``
2060 - 128-bit floating point value (112-bit mantissa)
2061
2062 * - ``x86_fp80``
2063 - 80-bit floating point value (X87)
2064
2065 * - ``ppc_fp128``
2066 - 128-bit floating point value (two 64-bits)
2067
Reid Kleckner9a16d082014-03-05 02:41:37 +00002068X86_mmx Type
2069""""""""""""
Sean Silvab084af42012-12-07 10:36:55 +00002070
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002071:Overview:
Sean Silvab084af42012-12-07 10:36:55 +00002072
Reid Kleckner9a16d082014-03-05 02:41:37 +00002073The x86_mmx type represents a value held in an MMX register on an x86
Sean Silvab084af42012-12-07 10:36:55 +00002074machine. The operations allowed on it are quite limited: parameters and
2075return values, load and store, and bitcast. User-specified MMX
2076instructions are represented as intrinsic or asm calls with arguments
2077and/or results of this type. There are no arrays, vectors or constants
2078of this type.
2079
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002080:Syntax:
Sean Silvab084af42012-12-07 10:36:55 +00002081
2082::
2083
Reid Kleckner9a16d082014-03-05 02:41:37 +00002084 x86_mmx
Sean Silvab084af42012-12-07 10:36:55 +00002085
Sean Silvab084af42012-12-07 10:36:55 +00002086
Rafael Espindola08013342013-12-07 19:34:20 +00002087.. _t_pointer:
2088
2089Pointer Type
2090""""""""""""
Sean Silvab084af42012-12-07 10:36:55 +00002091
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002092:Overview:
Sean Silvab084af42012-12-07 10:36:55 +00002093
Rafael Espindola08013342013-12-07 19:34:20 +00002094The pointer type is used to specify memory locations. Pointers are
2095commonly used to reference objects in memory.
2096
2097Pointer types may have an optional address space attribute defining the
2098numbered address space where the pointed-to object resides. The default
2099address space is number zero. The semantics of non-zero address spaces
2100are target-specific.
2101
2102Note that LLVM does not permit pointers to void (``void*``) nor does it
2103permit pointers to labels (``label*``). Use ``i8*`` instead.
Sean Silvab084af42012-12-07 10:36:55 +00002104
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002105:Syntax:
Sean Silvab084af42012-12-07 10:36:55 +00002106
2107::
2108
Rafael Espindola08013342013-12-07 19:34:20 +00002109 <type> *
2110
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002111:Examples:
Rafael Espindola08013342013-12-07 19:34:20 +00002112
2113+-------------------------+--------------------------------------------------------------------------------------------------------------+
2114| ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2115+-------------------------+--------------------------------------------------------------------------------------------------------------+
2116| ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2117+-------------------------+--------------------------------------------------------------------------------------------------------------+
2118| ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2119+-------------------------+--------------------------------------------------------------------------------------------------------------+
2120
2121.. _t_vector:
2122
2123Vector Type
2124"""""""""""
2125
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002126:Overview:
Rafael Espindola08013342013-12-07 19:34:20 +00002127
2128A vector type is a simple derived type that represents a vector of
2129elements. Vector types are used when multiple primitive data are
2130operated in parallel using a single instruction (SIMD). A vector type
2131requires a size (number of elements) and an underlying primitive data
2132type. Vector types are considered :ref:`first class <t_firstclass>`.
2133
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002134:Syntax:
Rafael Espindola08013342013-12-07 19:34:20 +00002135
2136::
2137
2138 < <# elements> x <elementtype> >
2139
2140The number of elements is a constant integer value larger than 0;
Manuel Jacob961f7872014-07-30 12:30:06 +00002141elementtype may be any integer, floating point or pointer type. Vectors
2142of size zero are not allowed.
Rafael Espindola08013342013-12-07 19:34:20 +00002143
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002144:Examples:
Rafael Espindola08013342013-12-07 19:34:20 +00002145
2146+-------------------+--------------------------------------------------+
2147| ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2148+-------------------+--------------------------------------------------+
2149| ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2150+-------------------+--------------------------------------------------+
2151| ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2152+-------------------+--------------------------------------------------+
2153| ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2154+-------------------+--------------------------------------------------+
Sean Silvab084af42012-12-07 10:36:55 +00002155
2156.. _t_label:
2157
2158Label Type
2159^^^^^^^^^^
2160
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002161:Overview:
Sean Silvab084af42012-12-07 10:36:55 +00002162
2163The label type represents code labels.
2164
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002165:Syntax:
Sean Silvab084af42012-12-07 10:36:55 +00002166
2167::
2168
2169 label
2170
2171.. _t_metadata:
2172
2173Metadata Type
2174^^^^^^^^^^^^^
2175
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002176:Overview:
Sean Silvab084af42012-12-07 10:36:55 +00002177
2178The metadata type represents embedded metadata. No derived types may be
2179created from metadata except for :ref:`function <t_function>` arguments.
2180
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002181:Syntax:
Sean Silvab084af42012-12-07 10:36:55 +00002182
2183::
2184
2185 metadata
2186
Sean Silvab084af42012-12-07 10:36:55 +00002187.. _t_aggregate:
2188
2189Aggregate Types
2190^^^^^^^^^^^^^^^
2191
2192Aggregate Types are a subset of derived types that can contain multiple
2193member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2194aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2195aggregate types.
2196
2197.. _t_array:
2198
2199Array Type
Rafael Espindola08013342013-12-07 19:34:20 +00002200""""""""""
Sean Silvab084af42012-12-07 10:36:55 +00002201
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002202:Overview:
Sean Silvab084af42012-12-07 10:36:55 +00002203
2204The array type is a very simple derived type that arranges elements
2205sequentially in memory. The array type requires a size (number of
2206elements) and an underlying data type.
2207
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002208:Syntax:
Sean Silvab084af42012-12-07 10:36:55 +00002209
2210::
2211
2212 [<# elements> x <elementtype>]
2213
2214The number of elements is a constant integer value; ``elementtype`` may
2215be any type with a size.
2216
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002217:Examples:
Sean Silvab084af42012-12-07 10:36:55 +00002218
2219+------------------+--------------------------------------+
2220| ``[40 x i32]`` | Array of 40 32-bit integer values. |
2221+------------------+--------------------------------------+
2222| ``[41 x i32]`` | Array of 41 32-bit integer values. |
2223+------------------+--------------------------------------+
2224| ``[4 x i8]`` | Array of 4 8-bit integer values. |
2225+------------------+--------------------------------------+
2226
2227Here are some examples of multidimensional arrays:
2228
2229+-----------------------------+----------------------------------------------------------+
2230| ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2231+-----------------------------+----------------------------------------------------------+
2232| ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2233+-----------------------------+----------------------------------------------------------+
2234| ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2235+-----------------------------+----------------------------------------------------------+
2236
2237There is no restriction on indexing beyond the end of the array implied
2238by a static type (though there are restrictions on indexing beyond the
2239bounds of an allocated object in some cases). This means that
2240single-dimension 'variable sized array' addressing can be implemented in
2241LLVM with a zero length array type. An implementation of 'pascal style
2242arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2243example.
2244
Sean Silvab084af42012-12-07 10:36:55 +00002245.. _t_struct:
2246
2247Structure Type
Rafael Espindola08013342013-12-07 19:34:20 +00002248""""""""""""""
Sean Silvab084af42012-12-07 10:36:55 +00002249
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002250:Overview:
Sean Silvab084af42012-12-07 10:36:55 +00002251
2252The structure type is used to represent a collection of data members
2253together in memory. The elements of a structure may be any type that has
2254a size.
2255
2256Structures in memory are accessed using '``load``' and '``store``' by
2257getting a pointer to a field with the '``getelementptr``' instruction.
2258Structures in registers are accessed using the '``extractvalue``' and
2259'``insertvalue``' instructions.
2260
2261Structures may optionally be "packed" structures, which indicate that
2262the alignment of the struct is one byte, and that there is no padding
2263between the elements. In non-packed structs, padding between field types
2264is inserted as defined by the DataLayout string in the module, which is
2265required to match what the underlying code generator expects.
2266
2267Structures can either be "literal" or "identified". A literal structure
2268is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2269identified types are always defined at the top level with a name.
2270Literal types are uniqued by their contents and can never be recursive
2271or opaque since there is no way to write one. Identified types can be
2272recursive, can be opaqued, and are never uniqued.
2273
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002274:Syntax:
Sean Silvab084af42012-12-07 10:36:55 +00002275
2276::
2277
2278 %T1 = type { <type list> } ; Identified normal struct type
2279 %T2 = type <{ <type list> }> ; Identified packed struct type
2280
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002281:Examples:
Sean Silvab084af42012-12-07 10:36:55 +00002282
2283+------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2284| ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2285+------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
Daniel Dunbar1dc66ca2013-01-17 18:57:32 +00002286| ``{ 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``. |
Sean Silvab084af42012-12-07 10:36:55 +00002287+------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2288| ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2289+------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2290
2291.. _t_opaque:
2292
2293Opaque Structure Types
Rafael Espindola08013342013-12-07 19:34:20 +00002294""""""""""""""""""""""
Sean Silvab084af42012-12-07 10:36:55 +00002295
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002296:Overview:
Sean Silvab084af42012-12-07 10:36:55 +00002297
2298Opaque structure types are used to represent named structure types that
2299do not have a body specified. This corresponds (for example) to the C
2300notion of a forward declared structure.
2301
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002302:Syntax:
Sean Silvab084af42012-12-07 10:36:55 +00002303
2304::
2305
2306 %X = type opaque
2307 %52 = type opaque
2308
Rafael Espindola2f6d7b92013-12-10 14:53:22 +00002309:Examples:
Sean Silvab084af42012-12-07 10:36:55 +00002310
2311+--------------+-------------------+
2312| ``opaque`` | An opaque type. |
2313+--------------+-------------------+
2314
Sean Silva1703e702014-04-08 21:06:22 +00002315.. _constants:
2316
Sean Silvab084af42012-12-07 10:36:55 +00002317Constants
2318=========
2319
2320LLVM has several different basic types of constants. This section
2321describes them all and their syntax.
2322
2323Simple Constants
2324----------------
2325
2326**Boolean constants**
2327 The two strings '``true``' and '``false``' are both valid constants
2328 of the ``i1`` type.
2329**Integer constants**
2330 Standard integers (such as '4') are constants of the
2331 :ref:`integer <t_integer>` type. Negative numbers may be used with
2332 integer types.
2333**Floating point constants**
2334 Floating point constants use standard decimal notation (e.g.
2335 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2336 hexadecimal notation (see below). The assembler requires the exact
2337 decimal value of a floating-point constant. For example, the
2338 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2339 decimal in binary. Floating point constants must have a :ref:`floating
2340 point <t_floating>` type.
2341**Null pointer constants**
2342 The identifier '``null``' is recognized as a null pointer constant
2343 and must be of :ref:`pointer type <t_pointer>`.
2344
2345The one non-intuitive notation for constants is the hexadecimal form of
2346floating point constants. For example, the form
2347'``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2348than) '``double 4.5e+15``'. The only time hexadecimal floating point
2349constants are required (and the only time that they are generated by the
2350disassembler) is when a floating point constant must be emitted but it
2351cannot be represented as a decimal floating point number in a reasonable
2352number of digits. For example, NaN's, infinities, and other special
2353values are represented in their IEEE hexadecimal format so that assembly
2354and disassembly do not cause any bits to change in the constants.
2355
2356When using the hexadecimal form, constants of types half, float, and
2357double are represented using the 16-digit form shown above (which
2358matches the IEEE754 representation for double); half and float values
Dmitri Gribenko4dc2ba12013-01-16 23:40:37 +00002359must, however, be exactly representable as IEEE 754 half and single
Sean Silvab084af42012-12-07 10:36:55 +00002360precision, respectively. Hexadecimal format is always used for long
2361double, and there are three forms of long double. The 80-bit format used
2362by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2363128-bit format used by PowerPC (two adjacent doubles) is represented by
2364``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
Richard Sandifordae426b42013-05-03 14:32:27 +00002365represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2366will only work if they match the long double format on your target.
2367The IEEE 16-bit format (half precision) is represented by ``0xH``
2368followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2369(sign bit at the left).
Sean Silvab084af42012-12-07 10:36:55 +00002370
Reid Kleckner9a16d082014-03-05 02:41:37 +00002371There are no constants of type x86_mmx.
Sean Silvab084af42012-12-07 10:36:55 +00002372
Eli Bendersky0220e6b2013-06-07 20:24:43 +00002373.. _complexconstants:
2374
Sean Silvab084af42012-12-07 10:36:55 +00002375Complex Constants
2376-----------------
2377
2378Complex constants are a (potentially recursive) combination of simple
2379constants and smaller complex constants.
2380
2381**Structure constants**
2382 Structure constants are represented with notation similar to
2383 structure type definitions (a comma separated list of elements,
2384 surrounded by braces (``{}``)). For example:
2385 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2386 "``@G = external global i32``". Structure constants must have
2387 :ref:`structure type <t_struct>`, and the number and types of elements
2388 must match those specified by the type.
2389**Array constants**
2390 Array constants are represented with notation similar to array type
2391 definitions (a comma separated list of elements, surrounded by
2392 square brackets (``[]``)). For example:
2393 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2394 :ref:`array type <t_array>`, and the number and types of elements must
Daniel Sandersf6051842014-09-11 12:02:59 +00002395 match those specified by the type. As a special case, character array
2396 constants may also be represented as a double-quoted string using the ``c``
2397 prefix. For example: "``c"Hello World\0A\00"``".
Sean Silvab084af42012-12-07 10:36:55 +00002398**Vector constants**
2399 Vector constants are represented with notation similar to vector
2400 type definitions (a comma separated list of elements, surrounded by
2401 less-than/greater-than's (``<>``)). For example:
2402 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2403 must have :ref:`vector type <t_vector>`, and the number and types of
2404 elements must match those specified by the type.
2405**Zero initialization**
2406 The string '``zeroinitializer``' can be used to zero initialize a
2407 value to zero of *any* type, including scalar and
2408 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2409 having to print large zero initializers (e.g. for large arrays) and
2410 is always exactly equivalent to using explicit zero initializers.
2411**Metadata node**
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00002412 A metadata node is a constant tuple without types. For example:
2413 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2414 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2415 Unlike other typed constants that are meant to be interpreted as part of
2416 the instruction stream, metadata is a place to attach additional
Sean Silvab084af42012-12-07 10:36:55 +00002417 information such as debug info.
2418
2419Global Variable and Function Addresses
2420--------------------------------------
2421
2422The addresses of :ref:`global variables <globalvars>` and
2423:ref:`functions <functionstructure>` are always implicitly valid
2424(link-time) constants. These constants are explicitly referenced when
2425the :ref:`identifier for the global <identifiers>` is used and always have
2426:ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2427file:
2428
2429.. code-block:: llvm
2430
2431 @X = global i32 17
2432 @Y = global i32 42
2433 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2434
2435.. _undefvalues:
2436
2437Undefined Values
2438----------------
2439
2440The string '``undef``' can be used anywhere a constant is expected, and
2441indicates that the user of the value may receive an unspecified
2442bit-pattern. Undefined values may be of any type (other than '``label``'
2443or '``void``') and be used anywhere a constant is permitted.
2444
2445Undefined values are useful because they indicate to the compiler that
2446the program is well defined no matter what value is used. This gives the
2447compiler more freedom to optimize. Here are some examples of
2448(potentially surprising) transformations that are valid (in pseudo IR):
2449
2450.. code-block:: llvm
2451
2452 %A = add %X, undef
2453 %B = sub %X, undef
2454 %C = xor %X, undef
2455 Safe:
2456 %A = undef
2457 %B = undef
2458 %C = undef
2459
2460This is safe because all of the output bits are affected by the undef
2461bits. Any output bit can have a zero or one depending on the input bits.
2462
2463.. code-block:: llvm
2464
2465 %A = or %X, undef
2466 %B = and %X, undef
2467 Safe:
2468 %A = -1
2469 %B = 0
2470 Unsafe:
2471 %A = undef
2472 %B = undef
2473
2474These logical operations have bits that are not always affected by the
2475input. For example, if ``%X`` has a zero bit, then the output of the
2476'``and``' operation will always be a zero for that bit, no matter what
2477the corresponding bit from the '``undef``' is. As such, it is unsafe to
2478optimize or assume that the result of the '``and``' is '``undef``'.
2479However, it is safe to assume that all bits of the '``undef``' could be
24800, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2481all the bits of the '``undef``' operand to the '``or``' could be set,
2482allowing the '``or``' to be folded to -1.
2483
2484.. code-block:: llvm
2485
2486 %A = select undef, %X, %Y
2487 %B = select undef, 42, %Y
2488 %C = select %X, %Y, undef
2489 Safe:
2490 %A = %X (or %Y)
2491 %B = 42 (or %Y)
2492 %C = %Y
2493 Unsafe:
2494 %A = undef
2495 %B = undef
2496 %C = undef
2497
2498This set of examples shows that undefined '``select``' (and conditional
2499branch) conditions can go *either way*, but they have to come from one
2500of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2501both known to have a clear low bit, then ``%A`` would have to have a
2502cleared low bit. However, in the ``%C`` example, the optimizer is
2503allowed to assume that the '``undef``' operand could be the same as
2504``%Y``, allowing the whole '``select``' to be eliminated.
2505
2506.. code-block:: llvm
2507
2508 %A = xor undef, undef
2509
2510 %B = undef
2511 %C = xor %B, %B
2512
2513 %D = undef
Jonathan Roelofsec81c0b2014-10-16 19:28:10 +00002514 %E = icmp slt %D, 4
Sean Silvab084af42012-12-07 10:36:55 +00002515 %F = icmp gte %D, 4
2516
2517 Safe:
2518 %A = undef
2519 %B = undef
2520 %C = undef
2521 %D = undef
2522 %E = undef
2523 %F = undef
2524
2525This example points out that two '``undef``' operands are not
2526necessarily the same. This can be surprising to people (and also matches
2527C semantics) where they assume that "``X^X``" is always zero, even if
2528``X`` is undefined. This isn't true for a number of reasons, but the
2529short answer is that an '``undef``' "variable" can arbitrarily change
2530its value over its "live range". This is true because the variable
2531doesn't actually *have a live range*. Instead, the value is logically
2532read from arbitrary registers that happen to be around when needed, so
2533the value is not necessarily consistent over time. In fact, ``%A`` and
2534``%C`` need to have the same semantics or the core LLVM "replace all
2535uses with" concept would not hold.
2536
2537.. code-block:: llvm
2538
2539 %A = fdiv undef, %X
2540 %B = fdiv %X, undef
2541 Safe:
2542 %A = undef
2543 b: unreachable
2544
2545These examples show the crucial difference between an *undefined value*
2546and *undefined behavior*. An undefined value (like '``undef``') is
2547allowed to have an arbitrary bit-pattern. This means that the ``%A``
2548operation can be constant folded to '``undef``', because the '``undef``'
2549could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2550However, in the second example, we can make a more aggressive
2551assumption: because the ``undef`` is allowed to be an arbitrary value,
2552we are allowed to assume that it could be zero. Since a divide by zero
2553has *undefined behavior*, we are allowed to assume that the operation
2554does not execute at all. This allows us to delete the divide and all
2555code after it. Because the undefined operation "can't happen", the
2556optimizer can assume that it occurs in dead code.
2557
2558.. code-block:: llvm
2559
2560 a: store undef -> %X
2561 b: store %X -> undef
2562 Safe:
2563 a: <deleted>
2564 b: unreachable
2565
2566These examples reiterate the ``fdiv`` example: a store *of* an undefined
2567value can be assumed to not have any effect; we can assume that the
2568value is overwritten with bits that happen to match what was already
2569there. However, a store *to* an undefined location could clobber
2570arbitrary memory, therefore, it has undefined behavior.
2571
2572.. _poisonvalues:
2573
2574Poison Values
2575-------------
2576
2577Poison values are similar to :ref:`undef values <undefvalues>`, however
2578they also represent the fact that an instruction or constant expression
Richard Smith32dbdf62014-07-31 04:25:36 +00002579that cannot evoke side effects has nevertheless detected a condition
2580that results in undefined behavior.
Sean Silvab084af42012-12-07 10:36:55 +00002581
2582There is currently no way of representing a poison value in the IR; they
2583only exist when produced by operations such as :ref:`add <i_add>` with
2584the ``nsw`` flag.
2585
2586Poison value behavior is defined in terms of value *dependence*:
2587
2588- Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2589- :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2590 their dynamic predecessor basic block.
2591- Function arguments depend on the corresponding actual argument values
2592 in the dynamic callers of their functions.
2593- :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2594 instructions that dynamically transfer control back to them.
2595- :ref:`Invoke <i_invoke>` instructions depend on the
2596 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2597 call instructions that dynamically transfer control back to them.
2598- Non-volatile loads and stores depend on the most recent stores to all
2599 of the referenced memory addresses, following the order in the IR
2600 (including loads and stores implied by intrinsics such as
2601 :ref:`@llvm.memcpy <int_memcpy>`.)
2602- An instruction with externally visible side effects depends on the
2603 most recent preceding instruction with externally visible side
2604 effects, following the order in the IR. (This includes :ref:`volatile
2605 operations <volatile>`.)
2606- An instruction *control-depends* on a :ref:`terminator
2607 instruction <terminators>` if the terminator instruction has
2608 multiple successors and the instruction is always executed when
2609 control transfers to one of the successors, and may not be executed
2610 when control is transferred to another.
2611- Additionally, an instruction also *control-depends* on a terminator
2612 instruction if the set of instructions it otherwise depends on would
2613 be different if the terminator had transferred control to a different
2614 successor.
2615- Dependence is transitive.
2616
Richard Smith32dbdf62014-07-31 04:25:36 +00002617Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2618with the additional effect that any instruction that has a *dependence*
Sean Silvab084af42012-12-07 10:36:55 +00002619on a poison value has undefined behavior.
2620
2621Here are some examples:
2622
2623.. code-block:: llvm
2624
2625 entry:
2626 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2627 %still_poison = and i32 %poison, 0 ; 0, but also poison.
David Blaikie16a97eb2015-03-04 22:02:58 +00002628 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
Sean Silvab084af42012-12-07 10:36:55 +00002629 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2630
2631 store i32 %poison, i32* @g ; Poison value stored to memory.
David Blaikiec7aabbb2015-03-04 22:06:14 +00002632 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
Sean Silvab084af42012-12-07 10:36:55 +00002633
2634 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2635
2636 %narrowaddr = bitcast i32* @g to i16*
2637 %wideaddr = bitcast i32* @g to i64*
David Blaikiec7aabbb2015-03-04 22:06:14 +00002638 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2639 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
Sean Silvab084af42012-12-07 10:36:55 +00002640
2641 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2642 br i1 %cmp, label %true, label %end ; Branch to either destination.
2643
2644 true:
2645 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2646 ; it has undefined behavior.
2647 br label %end
2648
2649 end:
2650 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2651 ; Both edges into this PHI are
2652 ; control-dependent on %cmp, so this
2653 ; always results in a poison value.
2654
2655 store volatile i32 0, i32* @g ; This would depend on the store in %true
2656 ; if %cmp is true, or the store in %entry
2657 ; otherwise, so this is undefined behavior.
2658
2659 br i1 %cmp, label %second_true, label %second_end
2660 ; The same branch again, but this time the
2661 ; true block doesn't have side effects.
2662
2663 second_true:
2664 ; No side effects!
2665 ret void
2666
2667 second_end:
2668 store volatile i32 0, i32* @g ; This time, the instruction always depends
2669 ; on the store in %end. Also, it is
2670 ; control-equivalent to %end, so this is
2671 ; well-defined (ignoring earlier undefined
2672 ; behavior in this example).
2673
2674.. _blockaddress:
2675
2676Addresses of Basic Blocks
2677-------------------------
2678
2679``blockaddress(@function, %block)``
2680
2681The '``blockaddress``' constant computes the address of the specified
2682basic block in the specified function, and always has an ``i8*`` type.
2683Taking the address of the entry block is illegal.
2684
2685This value only has defined behavior when used as an operand to the
2686':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2687against null. Pointer equality tests between labels addresses results in
Dmitri Gribenkoe8131122013-01-19 20:34:20 +00002688undefined behavior --- though, again, comparison against null is ok, and
Sean Silvab084af42012-12-07 10:36:55 +00002689no label is equal to the null pointer. This may be passed around as an
2690opaque pointer sized value as long as the bits are not inspected. This
2691allows ``ptrtoint`` and arithmetic to be performed on these values so
2692long as the original value is reconstituted before the ``indirectbr``
2693instruction.
2694
2695Finally, some targets may provide defined semantics when using the value
2696as the operand to an inline assembly, but that is target specific.
2697
Eli Bendersky0220e6b2013-06-07 20:24:43 +00002698.. _constantexprs:
2699
Sean Silvab084af42012-12-07 10:36:55 +00002700Constant Expressions
2701--------------------
2702
2703Constant expressions are used to allow expressions involving other
2704constants to be used as constants. Constant expressions may be of any
2705:ref:`first class <t_firstclass>` type and may involve any LLVM operation
2706that does not have side effects (e.g. load and call are not supported).
2707The following is the syntax for constant expressions:
2708
2709``trunc (CST to TYPE)``
2710 Truncate a constant to another type. The bit size of CST must be
2711 larger than the bit size of TYPE. Both types must be integers.
2712``zext (CST to TYPE)``
2713 Zero extend a constant to another type. The bit size of CST must be
2714 smaller than the bit size of TYPE. Both types must be integers.
2715``sext (CST to TYPE)``
2716 Sign extend a constant to another type. The bit size of CST must be
2717 smaller than the bit size of TYPE. Both types must be integers.
2718``fptrunc (CST to TYPE)``
2719 Truncate a floating point constant to another floating point type.
2720 The size of CST must be larger than the size of TYPE. Both types
2721 must be floating point.
2722``fpext (CST to TYPE)``
2723 Floating point extend a constant to another type. The size of CST
2724 must be smaller or equal to the size of TYPE. Both types must be
2725 floating point.
2726``fptoui (CST to TYPE)``
2727 Convert a floating point constant to the corresponding unsigned
2728 integer constant. TYPE must be a scalar or vector integer type. CST
2729 must be of scalar or vector floating point type. Both CST and TYPE
2730 must be scalars, or vectors of the same number of elements. If the
2731 value won't fit in the integer type, the results are undefined.
2732``fptosi (CST to TYPE)``
2733 Convert a floating point constant to the corresponding signed
2734 integer constant. TYPE must be a scalar or vector integer type. CST
2735 must be of scalar or vector floating point type. Both CST and TYPE
2736 must be scalars, or vectors of the same number of elements. If the
2737 value won't fit in the integer type, the results are undefined.
2738``uitofp (CST to TYPE)``
2739 Convert an unsigned integer constant to the corresponding floating
2740 point constant. TYPE must be a scalar or vector floating point type.
2741 CST must be of scalar or vector integer type. Both CST and TYPE must
2742 be scalars, or vectors of the same number of elements. If the value
2743 won't fit in the floating point type, the results are undefined.
2744``sitofp (CST to TYPE)``
2745 Convert a signed integer constant to the corresponding floating
2746 point constant. TYPE must be a scalar or vector floating point type.
2747 CST must be of scalar or vector integer type. Both CST and TYPE must
2748 be scalars, or vectors of the same number of elements. If the value
2749 won't fit in the floating point type, the results are undefined.
2750``ptrtoint (CST to TYPE)``
2751 Convert a pointer typed constant to the corresponding integer
Eli Bendersky9c0d4932013-03-11 16:51:15 +00002752 constant. ``TYPE`` must be an integer type. ``CST`` must be of
Sean Silvab084af42012-12-07 10:36:55 +00002753 pointer type. The ``CST`` value is zero extended, truncated, or
2754 unchanged to make it fit in ``TYPE``.
2755``inttoptr (CST to TYPE)``
2756 Convert an integer constant to a pointer constant. TYPE must be a
2757 pointer type. CST must be of integer type. The CST value is zero
2758 extended, truncated, or unchanged to make it fit in a pointer size.
2759 This one is *really* dangerous!
2760``bitcast (CST to TYPE)``
2761 Convert a constant, CST, to another TYPE. The constraints of the
2762 operands are the same as those for the :ref:`bitcast
2763 instruction <i_bitcast>`.
Matt Arsenaultb03bd4d2013-11-15 01:34:59 +00002764``addrspacecast (CST to TYPE)``
2765 Convert a constant pointer or constant vector of pointer, CST, to another
2766 TYPE in a different address space. The constraints of the operands are the
2767 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
David Blaikief72d05b2015-03-13 18:20:45 +00002768``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
Sean Silvab084af42012-12-07 10:36:55 +00002769 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2770 constants. As with the :ref:`getelementptr <i_getelementptr>`
2771 instruction, the index list may have zero or more indexes, which are
David Blaikief72d05b2015-03-13 18:20:45 +00002772 required to make sense for the type of "pointer to TY".
Sean Silvab084af42012-12-07 10:36:55 +00002773``select (COND, VAL1, VAL2)``
2774 Perform the :ref:`select operation <i_select>` on constants.
2775``icmp COND (VAL1, VAL2)``
2776 Performs the :ref:`icmp operation <i_icmp>` on constants.
2777``fcmp COND (VAL1, VAL2)``
2778 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2779``extractelement (VAL, IDX)``
2780 Perform the :ref:`extractelement operation <i_extractelement>` on
2781 constants.
2782``insertelement (VAL, ELT, IDX)``
2783 Perform the :ref:`insertelement operation <i_insertelement>` on
2784 constants.
2785``shufflevector (VEC1, VEC2, IDXMASK)``
2786 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2787 constants.
2788``extractvalue (VAL, IDX0, IDX1, ...)``
2789 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2790 constants. The index list is interpreted in a similar manner as
2791 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2792 least one index value must be specified.
2793``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2794 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2795 The index list is interpreted in a similar manner as indices in a
2796 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2797 value must be specified.
2798``OPCODE (LHS, RHS)``
2799 Perform the specified operation of the LHS and RHS constants. OPCODE
2800 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2801 binary <bitwiseops>` operations. The constraints on operands are
2802 the same as those for the corresponding instruction (e.g. no bitwise
2803 operations on floating point values are allowed).
2804
2805Other Values
2806============
2807
Eli Bendersky0220e6b2013-06-07 20:24:43 +00002808.. _inlineasmexprs:
2809
Sean Silvab084af42012-12-07 10:36:55 +00002810Inline Assembler Expressions
2811----------------------------
2812
2813LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
James Y Knightbc832ed2015-07-08 18:08:36 +00002814Inline Assembly <moduleasm>`) through the use of a special value. This value
2815represents the inline assembler as a template string (containing the
2816instructions to emit), a list of operand constraints (stored as a string), a
2817flag that indicates whether or not the inline asm expression has side effects,
2818and a flag indicating whether the function containing the asm needs to align its
2819stack conservatively.
2820
2821The template string supports argument substitution of the operands using "``$``"
2822followed by a number, to indicate substitution of the given register/memory
2823location, as specified by the constraint string. "``${NUM:MODIFIER}``" may also
2824be used, where ``MODIFIER`` is a target-specific annotation for how to print the
2825operand (See :ref:`inline-asm-modifiers`).
2826
2827A literal "``$``" may be included by using "``$$``" in the template. To include
2828other special characters into the output, the usual "``\XX``" escapes may be
2829used, just as in other strings. Note that after template substitution, the
2830resulting assembly string is parsed by LLVM's integrated assembler unless it is
2831disabled -- even when emitting a ``.s`` file -- and thus must contain assembly
2832syntax known to LLVM.
2833
2834LLVM's support for inline asm is modeled closely on the requirements of Clang's
2835GCC-compatible inline-asm support. Thus, the feature-set and the constraint and
2836modifier codes listed here are similar or identical to those in GCC's inline asm
2837support. However, to be clear, the syntax of the template and constraint strings
2838described here is *not* the same as the syntax accepted by GCC and Clang, and,
2839while most constraint letters are passed through as-is by Clang, some get
2840translated to other codes when converting from the C source to the LLVM
2841assembly.
2842
2843An example inline assembler expression is:
Sean Silvab084af42012-12-07 10:36:55 +00002844
2845.. code-block:: llvm
2846
2847 i32 (i32) asm "bswap $0", "=r,r"
2848
2849Inline assembler expressions may **only** be used as the callee operand
2850of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2851Thus, typically we have:
2852
2853.. code-block:: llvm
2854
2855 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2856
2857Inline asms with side effects not visible in the constraint list must be
2858marked as having side effects. This is done through the use of the
2859'``sideeffect``' keyword, like so:
2860
2861.. code-block:: llvm
2862
2863 call void asm sideeffect "eieio", ""()
2864
2865In some cases inline asms will contain code that will not work unless
2866the stack is aligned in some way, such as calls or SSE instructions on
2867x86, yet will not contain code that does that alignment within the asm.
2868The compiler should make conservative assumptions about what the asm
2869might contain and should generate its usual stack alignment code in the
2870prologue if the '``alignstack``' keyword is present:
2871
2872.. code-block:: llvm
2873
2874 call void asm alignstack "eieio", ""()
2875
2876Inline asms also support using non-standard assembly dialects. The
2877assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2878the inline asm is using the Intel dialect. Currently, ATT and Intel are
2879the only supported dialects. An example is:
2880
2881.. code-block:: llvm
2882
2883 call void asm inteldialect "eieio", ""()
2884
2885If multiple keywords appear the '``sideeffect``' keyword must come
2886first, the '``alignstack``' keyword second and the '``inteldialect``'
2887keyword last.
2888
James Y Knightbc832ed2015-07-08 18:08:36 +00002889Inline Asm Constraint String
2890^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2891
2892The constraint list is a comma-separated string, each element containing one or
2893more constraint codes.
2894
2895For each element in the constraint list an appropriate register or memory
2896operand will be chosen, and it will be made available to assembly template
2897string expansion as ``$0`` for the first constraint in the list, ``$1`` for the
2898second, etc.
2899
2900There are three different types of constraints, which are distinguished by a
2901prefix symbol in front of the constraint code: Output, Input, and Clobber. The
2902constraints must always be given in that order: outputs first, then inputs, then
2903clobbers. They cannot be intermingled.
2904
2905There are also three different categories of constraint codes:
2906
2907- Register constraint. This is either a register class, or a fixed physical
2908 register. This kind of constraint will allocate a register, and if necessary,
2909 bitcast the argument or result to the appropriate type.
2910- Memory constraint. This kind of constraint is for use with an instruction
2911 taking a memory operand. Different constraints allow for different addressing
2912 modes used by the target.
2913- Immediate value constraint. This kind of constraint is for an integer or other
2914 immediate value which can be rendered directly into an instruction. The
2915 various target-specific constraints allow the selection of a value in the
2916 proper range for the instruction you wish to use it with.
2917
2918Output constraints
2919""""""""""""""""""
2920
2921Output constraints are specified by an "``=``" prefix (e.g. "``=r``"). This
2922indicates that the assembly will write to this operand, and the operand will
2923then be made available as a return value of the ``asm`` expression. Output
2924constraints do not consume an argument from the call instruction. (Except, see
2925below about indirect outputs).
2926
2927Normally, it is expected that no output locations are written to by the assembly
2928expression until *all* of the inputs have been read. As such, LLVM may assign
2929the same register to an output and an input. If this is not safe (e.g. if the
2930assembly contains two instructions, where the first writes to one output, and
2931the second reads an input and writes to a second output), then the "``&``"
2932modifier must be used (e.g. "``=&r``") to specify that the output is an
2933"early-clobber" output. Marking an ouput as "early-clobber" ensures that LLVM
2934will not use the same register for any inputs (other than an input tied to this
2935output).
2936
2937Input constraints
2938"""""""""""""""""
2939
2940Input constraints do not have a prefix -- just the constraint codes. Each input
2941constraint will consume one argument from the call instruction. It is not
2942permitted for the asm to write to any input register or memory location (unless
2943that input is tied to an output). Note also that multiple inputs may all be
2944assigned to the same register, if LLVM can determine that they necessarily all
2945contain the same value.
2946
2947Instead of providing a Constraint Code, input constraints may also "tie"
2948themselves to an output constraint, by providing an integer as the constraint
2949string. Tied inputs still consume an argument from the call instruction, and
2950take up a position in the asm template numbering as is usual -- they will simply
2951be constrained to always use the same register as the output they've been tied
2952to. For example, a constraint string of "``=r,0``" says to assign a register for
2953output, and use that register as an input as well (it being the 0'th
2954constraint).
2955
2956It is permitted to tie an input to an "early-clobber" output. In that case, no
2957*other* input may share the same register as the input tied to the early-clobber
2958(even when the other input has the same value).
2959
2960You may only tie an input to an output which has a register constraint, not a
2961memory constraint. Only a single input may be tied to an output.
2962
2963There is also an "interesting" feature which deserves a bit of explanation: if a
2964register class constraint allocates a register which is too small for the value
2965type operand provided as input, the input value will be split into multiple
2966registers, and all of them passed to the inline asm.
2967
2968However, this feature is often not as useful as you might think.
2969
2970Firstly, the registers are *not* guaranteed to be consecutive. So, on those
2971architectures that have instructions which operate on multiple consecutive
2972instructions, this is not an appropriate way to support them. (e.g. the 32-bit
2973SparcV8 has a 64-bit load, which instruction takes a single 32-bit register. The
2974hardware then loads into both the named register, and the next register. This
2975feature of inline asm would not be useful to support that.)
2976
2977A few of the targets provide a template string modifier allowing explicit access
2978to the second register of a two-register operand (e.g. MIPS ``L``, ``M``, and
2979``D``). On such an architecture, you can actually access the second allocated
2980register (yet, still, not any subsequent ones). But, in that case, you're still
2981probably better off simply splitting the value into two separate operands, for
2982clarity. (e.g. see the description of the ``A`` constraint on X86, which,
2983despite existing only for use with this feature, is not really a good idea to
2984use)
2985
2986Indirect inputs and outputs
2987"""""""""""""""""""""""""""
2988
2989Indirect output or input constraints can be specified by the "``*``" modifier
2990(which goes after the "``=``" in case of an output). This indicates that the asm
2991will write to or read from the contents of an *address* provided as an input
2992argument. (Note that in this way, indirect outputs act more like an *input* than
2993an output: just like an input, they consume an argument of the call expression,
2994rather than producing a return value. An indirect output constraint is an
2995"output" only in that the asm is expected to write to the contents of the input
2996memory location, instead of just read from it).
2997
2998This is most typically used for memory constraint, e.g. "``=*m``", to pass the
2999address of a variable as a value.
3000
3001It is also possible to use an indirect *register* constraint, but only on output
3002(e.g. "``=*r``"). This will cause LLVM to allocate a register for an output
3003value normally, and then, separately emit a store to the address provided as
3004input, after the provided inline asm. (It's not clear what value this
3005functionality provides, compared to writing the store explicitly after the asm
3006statement, and it can only produce worse code, since it bypasses many
3007optimization passes. I would recommend not using it.)
3008
3009
3010Clobber constraints
3011"""""""""""""""""""
3012
3013A clobber constraint is indicated by a "``~``" prefix. A clobber does not
3014consume an input operand, nor generate an output. Clobbers cannot use any of the
3015general constraint code letters -- they may use only explicit register
3016constraints, e.g. "``~{eax}``". The one exception is that a clobber string of
3017"``~{memory}``" indicates that the assembly writes to arbitrary undeclared
3018memory locations -- not only the memory pointed to by a declared indirect
3019output.
3020
3021
3022Constraint Codes
3023""""""""""""""""
3024After a potential prefix comes constraint code, or codes.
3025
3026A Constraint Code is either a single letter (e.g. "``r``"), a "``^``" character
3027followed by two letters (e.g. "``^wc``"), or "``{``" register-name "``}``"
3028(e.g. "``{eax}``").
3029
3030The one and two letter constraint codes are typically chosen to be the same as
3031GCC's constraint codes.
3032
3033A single constraint may include one or more than constraint code in it, leaving
3034it up to LLVM to choose which one to use. This is included mainly for
3035compatibility with the translation of GCC inline asm coming from clang.
3036
3037There are two ways to specify alternatives, and either or both may be used in an
3038inline asm constraint list:
3039
30401) Append the codes to each other, making a constraint code set. E.g. "``im``"
3041 or "``{eax}m``". This means "choose any of the options in the set". The
3042 choice of constraint is made independently for each constraint in the
3043 constraint list.
3044
30452) Use "``|``" between constraint code sets, creating alternatives. Every
3046 constraint in the constraint list must have the same number of alternative
3047 sets. With this syntax, the same alternative in *all* of the items in the
3048 constraint list will be chosen together.
3049
3050Putting those together, you might have a two operand constraint string like
3051``"rm|r,ri|rm"``. This indicates that if operand 0 is ``r`` or ``m``, then
3052operand 1 may be one of ``r`` or ``i``. If operand 0 is ``r``, then operand 1
3053may be one of ``r`` or ``m``. But, operand 0 and 1 cannot both be of type m.
3054
3055However, the use of either of the alternatives features is *NOT* recommended, as
3056LLVM is not able to make an intelligent choice about which one to use. (At the
3057point it currently needs to choose, not enough information is available to do so
3058in a smart way.) Thus, it simply tries to make a choice that's most likely to
3059compile, not one that will be optimal performance. (e.g., given "``rm``", it'll
3060always choose to use memory, not registers). And, if given multiple registers,
3061or multiple register classes, it will simply choose the first one. (In fact, it
3062doesn't currently even ensure explicitly specified physical registers are
3063unique, so specifying multiple physical registers as alternatives, like
3064``{r11}{r12},{r11}{r12}``, will assign r11 to both operands, not at all what was
3065intended.)
3066
3067Supported Constraint Code List
3068""""""""""""""""""""""""""""""
3069
3070The constraint codes are, in general, expected to behave the same way they do in
3071GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3072inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3073and GCC likely indicates a bug in LLVM.
3074
3075Some constraint codes are typically supported by all targets:
3076
3077- ``r``: A register in the target's general purpose register class.
3078- ``m``: A memory address operand. It is target-specific what addressing modes
3079 are supported, typical examples are register, or register + register offset,
3080 or register + immediate offset (of some target-specific size).
3081- ``i``: An integer constant (of target-specific width). Allows either a simple
3082 immediate, or a relocatable value.
3083- ``n``: An integer constant -- *not* including relocatable values.
3084- ``s``: An integer constant, but allowing *only* relocatable values.
3085- ``X``: Allows an operand of any kind, no constraint whatsoever. Typically
3086 useful to pass a label for an asm branch or call.
3087
3088 .. FIXME: but that surely isn't actually okay to jump out of an asm
3089 block without telling llvm about the control transfer???)
3090
3091- ``{register-name}``: Requires exactly the named physical register.
3092
3093Other constraints are target-specific:
3094
3095AArch64:
3096
3097- ``z``: An immediate integer 0. Outputs ``WZR`` or ``XZR``, as appropriate.
3098- ``I``: An immediate integer valid for an ``ADD`` or ``SUB`` instruction,
3099 i.e. 0 to 4095 with optional shift by 12.
3100- ``J``: An immediate integer that, when negated, is valid for an ``ADD`` or
3101 ``SUB`` instruction, i.e. -1 to -4095 with optional left shift by 12.
3102- ``K``: An immediate integer that is valid for the 'bitmask immediate 32' of a
3103 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 32-bit register.
3104- ``L``: An immediate integer that is valid for the 'bitmask immediate 64' of a
3105 logical instruction like ``AND``, ``EOR``, or ``ORR`` with a 64-bit register.
3106- ``M``: An immediate integer for use with the ``MOV`` assembly alias on a
3107 32-bit register. This is a superset of ``K``: in addition to the bitmask
3108 immediate, also allows immediate integers which can be loaded with a single
3109 ``MOVZ`` or ``MOVL`` instruction.
3110- ``N``: An immediate integer for use with the ``MOV`` assembly alias on a
3111 64-bit register. This is a superset of ``L``.
3112- ``Q``: Memory address operand must be in a single register (no
3113 offsets). (However, LLVM currently does this for the ``m`` constraint as
3114 well.)
3115- ``r``: A 32 or 64-bit integer register (W* or X*).
3116- ``w``: A 32, 64, or 128-bit floating-point/SIMD register.
3117- ``x``: A lower 128-bit floating-point/SIMD register (``V0`` to ``V15``).
3118
3119AMDGPU:
3120
3121- ``r``: A 32 or 64-bit integer register.
3122- ``[0-9]v``: The 32-bit VGPR register, number 0-9.
3123- ``[0-9]s``: The 32-bit SGPR register, number 0-9.
3124
3125
3126All ARM modes:
3127
3128- ``Q``, ``Um``, ``Un``, ``Uq``, ``Us``, ``Ut``, ``Uv``, ``Uy``: Memory address
3129 operand. Treated the same as operand ``m``, at the moment.
3130
3131ARM and ARM's Thumb2 mode:
3132
3133- ``j``: An immediate integer between 0 and 65535 (valid for ``MOVW``)
3134- ``I``: An immediate integer valid for a data-processing instruction.
3135- ``J``: An immediate integer between -4095 and 4095.
3136- ``K``: An immediate integer whose bitwise inverse is valid for a
3137 data-processing instruction. (Can be used with template modifier "``B``" to
3138 print the inverted value).
3139- ``L``: An immediate integer whose negation is valid for a data-processing
3140 instruction. (Can be used with template modifier "``n``" to print the negated
3141 value).
3142- ``M``: A power of two or a integer between 0 and 32.
3143- ``N``: Invalid immediate constraint.
3144- ``O``: Invalid immediate constraint.
3145- ``r``: A general-purpose 32-bit integer register (``r0-r15``).
3146- ``l``: In Thumb2 mode, low 32-bit GPR registers (``r0-r7``). In ARM mode, same
3147 as ``r``.
3148- ``h``: In Thumb2 mode, a high 32-bit GPR register (``r8-r15``). In ARM mode,
3149 invalid.
3150- ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3151 ``d0-d31``, or ``q0-q15``.
3152- ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3153 ``d0-d7``, or ``q0-q3``.
3154- ``t``: A floating-point/SIMD register, only supports 32-bit values:
3155 ``s0-s31``.
3156
3157ARM's Thumb1 mode:
3158
3159- ``I``: An immediate integer between 0 and 255.
3160- ``J``: An immediate integer between -255 and -1.
3161- ``K``: An immediate integer between 0 and 255, with optional left-shift by
3162 some amount.
3163- ``L``: An immediate integer between -7 and 7.
3164- ``M``: An immediate integer which is a multiple of 4 between 0 and 1020.
3165- ``N``: An immediate integer between 0 and 31.
3166- ``O``: An immediate integer which is a multiple of 4 between -508 and 508.
3167- ``r``: A low 32-bit GPR register (``r0-r7``).
3168- ``l``: A low 32-bit GPR register (``r0-r7``).
3169- ``h``: A high GPR register (``r0-r7``).
3170- ``w``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s31``,
3171 ``d0-d31``, or ``q0-q15``.
3172- ``x``: A 32, 64, or 128-bit floating-point/SIMD register: ``s0-s15``,
3173 ``d0-d7``, or ``q0-q3``.
3174- ``t``: A floating-point/SIMD register, only supports 32-bit values:
3175 ``s0-s31``.
3176
3177
3178Hexagon:
3179
3180- ``o``, ``v``: A memory address operand, treated the same as constraint ``m``,
3181 at the moment.
3182- ``r``: A 32 or 64-bit register.
3183
3184MSP430:
3185
3186- ``r``: An 8 or 16-bit register.
3187
3188MIPS:
3189
3190- ``I``: An immediate signed 16-bit integer.
3191- ``J``: An immediate integer zero.
3192- ``K``: An immediate unsigned 16-bit integer.
3193- ``L``: An immediate 32-bit integer, where the lower 16 bits are 0.
3194- ``N``: An immediate integer between -65535 and -1.
3195- ``O``: An immediate signed 15-bit integer.
3196- ``P``: An immediate integer between 1 and 65535.
3197- ``m``: A memory address operand. In MIPS-SE mode, allows a base address
3198 register plus 16-bit immediate offset. In MIPS mode, just a base register.
3199- ``R``: A memory address operand. In MIPS-SE mode, allows a base address
3200 register plus a 9-bit signed offset. In MIPS mode, the same as constraint
3201 ``m``.
3202- ``ZC``: A memory address operand, suitable for use in a ``pref``, ``ll``, or
3203 ``sc`` instruction on the given subtarget (details vary).
3204- ``r``, ``d``, ``y``: A 32 or 64-bit GPR register.
3205- ``f``: A 32 or 64-bit FPU register (``F0-F31``), or a 128-bit MSA register
Daniel Sanders3745e022015-07-13 09:24:21 +00003206 (``W0-W31``). In the case of MSA registers, it is recommended to use the ``w``
3207 argument modifier for compatibility with GCC.
James Y Knightbc832ed2015-07-08 18:08:36 +00003208- ``c``: A 32-bit or 64-bit GPR register suitable for indirect jump (always
3209 ``25``).
3210- ``l``: The ``lo`` register, 32 or 64-bit.
3211- ``x``: Invalid.
3212
3213NVPTX:
3214
3215- ``b``: A 1-bit integer register.
3216- ``c`` or ``h``: A 16-bit integer register.
3217- ``r``: A 32-bit integer register.
3218- ``l`` or ``N``: A 64-bit integer register.
3219- ``f``: A 32-bit float register.
3220- ``d``: A 64-bit float register.
3221
3222
3223PowerPC:
3224
3225- ``I``: An immediate signed 16-bit integer.
3226- ``J``: An immediate unsigned 16-bit integer, shifted left 16 bits.
3227- ``K``: An immediate unsigned 16-bit integer.
3228- ``L``: An immediate signed 16-bit integer, shifted left 16 bits.
3229- ``M``: An immediate integer greater than 31.
3230- ``N``: An immediate integer that is an exact power of 2.
3231- ``O``: The immediate integer constant 0.
3232- ``P``: An immediate integer constant whose negation is a signed 16-bit
3233 constant.
3234- ``es``, ``o``, ``Q``, ``Z``, ``Zy``: A memory address operand, currently
3235 treated the same as ``m``.
3236- ``r``: A 32 or 64-bit integer register.
3237- ``b``: A 32 or 64-bit integer register, excluding ``R0`` (that is:
3238 ``R1-R31``).
3239- ``f``: A 32 or 64-bit float register (``F0-F31``), or when QPX is enabled, a
3240 128 or 256-bit QPX register (``Q0-Q31``; aliases the ``F`` registers).
3241- ``v``: For ``4 x f32`` or ``4 x f64`` types, when QPX is enabled, a
3242 128 or 256-bit QPX register (``Q0-Q31``), otherwise a 128-bit
3243 altivec vector register (``V0-V31``).
3244
3245 .. FIXME: is this a bug that v accepts QPX registers? I think this
3246 is supposed to only use the altivec vector registers?
3247
3248- ``y``: Condition register (``CR0-CR7``).
3249- ``wc``: An individual CR bit in a CR register.
3250- ``wa``, ``wd``, ``wf``: Any 128-bit VSX vector register, from the full VSX
3251 register set (overlapping both the floating-point and vector register files).
3252- ``ws``: A 32 or 64-bit floating point register, from the full VSX register
3253 set.
3254
3255Sparc:
3256
3257- ``I``: An immediate 13-bit signed integer.
3258- ``r``: A 32-bit integer register.
3259
3260SystemZ:
3261
3262- ``I``: An immediate unsigned 8-bit integer.
3263- ``J``: An immediate unsigned 12-bit integer.
3264- ``K``: An immediate signed 16-bit integer.
3265- ``L``: An immediate signed 20-bit integer.
3266- ``M``: An immediate integer 0x7fffffff.
3267- ``Q``, ``R``, ``S``, ``T``: A memory address operand, treated the same as
3268 ``m``, at the moment.
3269- ``r`` or ``d``: A 32, 64, or 128-bit integer register.
3270- ``a``: A 32, 64, or 128-bit integer address register (excludes R0, which in an
3271 address context evaluates as zero).
3272- ``h``: A 32-bit value in the high part of a 64bit data register
3273 (LLVM-specific)
3274- ``f``: A 32, 64, or 128-bit floating point register.
3275
3276X86:
3277
3278- ``I``: An immediate integer between 0 and 31.
3279- ``J``: An immediate integer between 0 and 64.
3280- ``K``: An immediate signed 8-bit integer.
3281- ``L``: An immediate integer, 0xff or 0xffff or (in 64-bit mode only)
3282 0xffffffff.
3283- ``M``: An immediate integer between 0 and 3.
3284- ``N``: An immediate unsigned 8-bit integer.
3285- ``O``: An immediate integer between 0 and 127.
3286- ``e``: An immediate 32-bit signed integer.
3287- ``Z``: An immediate 32-bit unsigned integer.
3288- ``o``, ``v``: Treated the same as ``m``, at the moment.
3289- ``q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3290 ``l`` integer register. On X86-32, this is the ``a``, ``b``, ``c``, and ``d``
3291 registers, and on X86-64, it is all of the integer registers.
3292- ``Q``: An 8, 16, 32, or 64-bit register which can be accessed as an 8-bit
3293 ``h`` integer register. This is the ``a``, ``b``, ``c``, and ``d`` registers.
3294- ``r`` or ``l``: An 8, 16, 32, or 64-bit integer register.
3295- ``R``: An 8, 16, 32, or 64-bit "legacy" integer register -- one which has
3296 existed since i386, and can be accessed without the REX prefix.
3297- ``f``: A 32, 64, or 80-bit '387 FPU stack pseudo-register.
3298- ``y``: A 64-bit MMX register, if MMX is enabled.
3299- ``x``: If SSE is enabled: a 32 or 64-bit scalar operand, or 128-bit vector
3300 operand in a SSE register. If AVX is also enabled, can also be a 256-bit
3301 vector operand in an AVX register. If AVX-512 is also enabled, can also be a
3302 512-bit vector operand in an AVX512 register, Otherwise, an error.
3303- ``Y``: The same as ``x``, if *SSE2* is enabled, otherwise an error.
3304- ``A``: Special case: allocates EAX first, then EDX, for a single operand (in
3305 32-bit mode, a 64-bit integer operand will get split into two registers). It
3306 is not recommended to use this constraint, as in 64-bit mode, the 64-bit
3307 operand will get allocated only to RAX -- if two 32-bit operands are needed,
3308 you're better off splitting it yourself, before passing it to the asm
3309 statement.
3310
3311XCore:
3312
3313- ``r``: A 32-bit integer register.
3314
3315
3316.. _inline-asm-modifiers:
3317
3318Asm template argument modifiers
3319^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3320
3321In the asm template string, modifiers can be used on the operand reference, like
3322"``${0:n}``".
3323
3324The modifiers are, in general, expected to behave the same way they do in
3325GCC. LLVM's support is often implemented on an 'as-needed' basis, to support C
3326inline asm code which was supported by GCC. A mismatch in behavior between LLVM
3327and GCC likely indicates a bug in LLVM.
3328
3329Target-independent:
3330
3331- ``c``: Print an immediate integer constant unadorned, without
3332 the target-specific immediate punctuation (e.g. no ``$`` prefix).
3333- ``n``: Negate and print immediate integer constant unadorned, without the
3334 target-specific immediate punctuation (e.g. no ``$`` prefix).
3335- ``l``: Print as an unadorned label, without the target-specific label
3336 punctuation (e.g. no ``$`` prefix).
3337
3338AArch64:
3339
3340- ``w``: Print a GPR register with a ``w*`` name instead of ``x*`` name. E.g.,
3341 instead of ``x30``, print ``w30``.
3342- ``x``: Print a GPR register with a ``x*`` name. (this is the default, anyhow).
3343- ``b``, ``h``, ``s``, ``d``, ``q``: Print a floating-point/SIMD register with a
3344 ``b*``, ``h*``, ``s*``, ``d*``, or ``q*`` name, rather than the default of
3345 ``v*``.
3346
3347AMDGPU:
3348
3349- ``r``: No effect.
3350
3351ARM:
3352
3353- ``a``: Print an operand as an address (with ``[`` and ``]`` surrounding a
3354 register).
3355- ``P``: No effect.
3356- ``q``: No effect.
3357- ``y``: Print a VFP single-precision register as an indexed double (e.g. print
3358 as ``d4[1]`` instead of ``s9``)
3359- ``B``: Bitwise invert and print an immediate integer constant without ``#``
3360 prefix.
3361- ``L``: Print the low 16-bits of an immediate integer constant.
3362- ``M``: Print as a register set suitable for ldm/stm. Also prints *all*
3363 register operands subsequent to the specified one (!), so use carefully.
3364- ``Q``: Print the low-order register of a register-pair, or the low-order
3365 register of a two-register operand.
3366- ``R``: Print the high-order register of a register-pair, or the high-order
3367 register of a two-register operand.
3368- ``H``: Print the second register of a register-pair. (On a big-endian system,
3369 ``H`` is equivalent to ``Q``, and on little-endian system, ``H`` is equivalent
3370 to ``R``.)
3371
3372 .. FIXME: H doesn't currently support printing the second register
3373 of a two-register operand.
3374
3375- ``e``: Print the low doubleword register of a NEON quad register.
3376- ``f``: Print the high doubleword register of a NEON quad register.
3377- ``m``: Print the base register of a memory operand without the ``[`` and ``]``
3378 adornment.
3379
3380Hexagon:
3381
3382- ``L``: Print the second register of a two-register operand. Requires that it
3383 has been allocated consecutively to the first.
3384
3385 .. FIXME: why is it restricted to consecutive ones? And there's
3386 nothing that ensures that happens, is there?
3387
3388- ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3389 nothing. Used to print 'addi' vs 'add' instructions.
3390
3391MSP430:
3392
3393No additional modifiers.
3394
3395MIPS:
3396
3397- ``X``: Print an immediate integer as hexadecimal
3398- ``x``: Print the low 16 bits of an immediate integer as hexadecimal.
3399- ``d``: Print an immediate integer as decimal.
3400- ``m``: Subtract one and print an immediate integer as decimal.
3401- ``z``: Print $0 if an immediate zero, otherwise print normally.
3402- ``L``: Print the low-order register of a two-register operand, or prints the
3403 address of the low-order word of a double-word memory operand.
3404
3405 .. FIXME: L seems to be missing memory operand support.
3406
3407- ``M``: Print the high-order register of a two-register operand, or prints the
3408 address of the high-order word of a double-word memory operand.
3409
3410 .. FIXME: M seems to be missing memory operand support.
3411
3412- ``D``: Print the second register of a two-register operand, or prints the
3413 second word of a double-word memory operand. (On a big-endian system, ``D`` is
3414 equivalent to ``L``, and on little-endian system, ``D`` is equivalent to
3415 ``M``.)
Daniel Sanders3745e022015-07-13 09:24:21 +00003416- ``w``: No effect. Provided for compatibility with GCC which requires this
3417 modifier in order to print MSA registers (``W0-W31``) with the ``f``
3418 constraint.
James Y Knightbc832ed2015-07-08 18:08:36 +00003419
3420NVPTX:
3421
3422- ``r``: No effect.
3423
3424PowerPC:
3425
3426- ``L``: Print the second register of a two-register operand. Requires that it
3427 has been allocated consecutively to the first.
3428
3429 .. FIXME: why is it restricted to consecutive ones? And there's
3430 nothing that ensures that happens, is there?
3431
3432- ``I``: Print the letter 'i' if the operand is an integer constant, otherwise
3433 nothing. Used to print 'addi' vs 'add' instructions.
3434- ``y``: For a memory operand, prints formatter for a two-register X-form
3435 instruction. (Currently always prints ``r0,OPERAND``).
3436- ``U``: Prints 'u' if the memory operand is an update form, and nothing
3437 otherwise. (NOTE: LLVM does not support update form, so this will currently
3438 always print nothing)
3439- ``X``: Prints 'x' if the memory operand is an indexed form. (NOTE: LLVM does
3440 not support indexed form, so this will currently always print nothing)
3441
3442Sparc:
3443
3444- ``r``: No effect.
3445
3446SystemZ:
3447
3448SystemZ implements only ``n``, and does *not* support any of the other
3449target-independent modifiers.
3450
3451X86:
3452
3453- ``c``: Print an unadorned integer or symbol name. (The latter is
3454 target-specific behavior for this typically target-independent modifier).
3455- ``A``: Print a register name with a '``*``' before it.
3456- ``b``: Print an 8-bit register name (e.g. ``al``); do nothing on a memory
3457 operand.
3458- ``h``: Print the upper 8-bit register name (e.g. ``ah``); do nothing on a
3459 memory operand.
3460- ``w``: Print the 16-bit register name (e.g. ``ax``); do nothing on a memory
3461 operand.
3462- ``k``: Print the 32-bit register name (e.g. ``eax``); do nothing on a memory
3463 operand.
3464- ``q``: Print the 64-bit register name (e.g. ``rax``), if 64-bit registers are
3465 available, otherwise the 32-bit register name; do nothing on a memory operand.
3466- ``n``: Negate and print an unadorned integer, or, for operands other than an
3467 immediate integer (e.g. a relocatable symbol expression), print a '-' before
3468 the operand. (The behavior for relocatable symbol expressions is a
3469 target-specific behavior for this typically target-independent modifier)
3470- ``H``: Print a memory reference with additional offset +8.
3471- ``P``: Print a memory reference or operand for use as the argument of a call
3472 instruction. (E.g. omit ``(rip)``, even though it's PC-relative.)
3473
3474XCore:
3475
3476No additional modifiers.
3477
3478
Sean Silvab084af42012-12-07 10:36:55 +00003479Inline Asm Metadata
3480^^^^^^^^^^^^^^^^^^^
3481
3482The call instructions that wrap inline asm nodes may have a
3483"``!srcloc``" MDNode attached to it that contains a list of constant
3484integers. If present, the code generator will use the integer as the
3485location cookie value when report errors through the ``LLVMContext``
3486error reporting mechanisms. This allows a front-end to correlate backend
3487errors that occur with inline asm back to the source code that produced
3488it. For example:
3489
3490.. code-block:: llvm
3491
3492 call void asm sideeffect "something bad", ""(), !srcloc !42
3493 ...
3494 !42 = !{ i32 1234567 }
3495
3496It is up to the front-end to make sense of the magic numbers it places
3497in the IR. If the MDNode contains multiple constants, the code generator
3498will use the one that corresponds to the line of the asm that the error
3499occurs on.
3500
3501.. _metadata:
3502
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00003503Metadata
3504========
Sean Silvab084af42012-12-07 10:36:55 +00003505
3506LLVM IR allows metadata to be attached to instructions in the program
3507that can convey extra information about the code to the optimizers and
3508code generator. One example application of metadata is source-level
3509debug information. There are two metadata primitives: strings and nodes.
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00003510
3511Metadata does not have a type, and is not a value. If referenced from a
3512``call`` instruction, it uses the ``metadata`` type.
3513
3514All metadata are identified in syntax by a exclamation point ('``!``').
3515
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003516.. _metadata-string:
3517
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00003518Metadata Nodes and Metadata Strings
3519-----------------------------------
Sean Silvab084af42012-12-07 10:36:55 +00003520
3521A metadata string is a string surrounded by double quotes. It can
3522contain any character by escaping non-printable characters with
3523"``\xx``" where "``xx``" is the two digit hex code. For example:
3524"``!"test\00"``".
3525
3526Metadata nodes are represented with notation similar to structure
3527constants (a comma separated list of elements, surrounded by braces and
3528preceded by an exclamation point). Metadata nodes can have any values as
3529their operand. For example:
3530
3531.. code-block:: llvm
3532
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00003533 !{ !"test\00", i32 10}
Sean Silvab084af42012-12-07 10:36:55 +00003534
Duncan P. N. Exon Smith090a19b2015-01-08 22:38:29 +00003535Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
3536
3537.. code-block:: llvm
3538
3539 !0 = distinct !{!"test\00", i32 10}
3540
Duncan P. N. Exon Smith99010342015-01-08 23:50:26 +00003541``distinct`` nodes are useful when nodes shouldn't be merged based on their
3542content. They can also occur when transformations cause uniquing collisions
3543when metadata operands change.
3544
Sean Silvab084af42012-12-07 10:36:55 +00003545A :ref:`named metadata <namedmetadatastructure>` is a collection of
3546metadata nodes, which can be looked up in the module symbol table. For
3547example:
3548
3549.. code-block:: llvm
3550
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00003551 !foo = !{!4, !3}
Sean Silvab084af42012-12-07 10:36:55 +00003552
3553Metadata can be used as function arguments. Here ``llvm.dbg.value``
3554function is using two metadata arguments:
3555
3556.. code-block:: llvm
3557
3558 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
3559
3560Metadata can be attached with an instruction. Here metadata ``!21`` is
3561attached to the ``add`` instruction using the ``!dbg`` identifier:
3562
3563.. code-block:: llvm
3564
3565 %indvar.next = add i64 %indvar, 1, !dbg !21
3566
3567More information about specific metadata nodes recognized by the
3568optimizers and code generator is found below.
3569
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003570.. _specialized-metadata:
3571
Duncan P. N. Exon Smith6a484832015-01-13 21:10:44 +00003572Specialized Metadata Nodes
3573^^^^^^^^^^^^^^^^^^^^^^^^^^
3574
3575Specialized metadata nodes are custom data structures in metadata (as opposed
3576to generic tuples). Their fields are labelled, and can be specified in any
3577order.
3578
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003579These aren't inherently debug info centric, but currently all the specialized
3580metadata nodes are related to debug info.
3581
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003582.. _DICompileUnit:
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003583
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003584DICompileUnit
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003585"""""""""""""
3586
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003587``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003588``retainedTypes:``, ``subprograms:``, ``globals:`` and ``imports:`` fields are
3589tuples containing the debug info to be emitted along with the compile unit,
3590regardless of code optimizations (some nodes are only emitted if there are
3591references to them from instructions).
3592
3593.. code-block:: llvm
3594
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003595 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003596 isOptimized: true, flags: "-O2", runtimeVersion: 2,
3597 splitDebugFilename: "abc.debug", emissionKind: 1,
3598 enums: !2, retainedTypes: !3, subprograms: !4,
3599 globals: !5, imports: !6)
3600
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003601Compile unit descriptors provide the root scope for objects declared in a
3602specific compilation unit. File descriptors are defined using this scope.
3603These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
3604keep track of subprograms, global variables, type information, and imported
3605entities (declarations and namespaces).
3606
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003607.. _DIFile:
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003608
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003609DIFile
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003610""""""
3611
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003612``DIFile`` nodes represent files. The ``filename:`` can include slashes.
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003613
3614.. code-block:: llvm
3615
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003616 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003617
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003618Files are sometimes used in ``scope:`` fields, and are the only valid target
3619for ``file:`` fields.
3620
Michael Kuperstein605308a2015-05-14 10:58:59 +00003621.. _DIBasicType:
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003622
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003623DIBasicType
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003624"""""""""""
3625
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003626``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003627``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003628
3629.. code-block:: llvm
3630
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003631 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003632 encoding: DW_ATE_unsigned_char)
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003633 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003634
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003635The ``encoding:`` describes the details of the type. Usually it's one of the
3636following:
3637
3638.. code-block:: llvm
3639
3640 DW_ATE_address = 1
3641 DW_ATE_boolean = 2
3642 DW_ATE_float = 4
3643 DW_ATE_signed = 5
3644 DW_ATE_signed_char = 6
3645 DW_ATE_unsigned = 7
3646 DW_ATE_unsigned_char = 8
3647
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003648.. _DISubroutineType:
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003649
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003650DISubroutineType
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003651""""""""""""""""
3652
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003653``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003654refers to a tuple; the first operand is the return type, while the rest are the
3655types of the formal arguments in order. If the first operand is ``null``, that
3656represents a function with no return value (such as ``void foo() {}`` in C++).
3657
3658.. code-block:: llvm
3659
3660 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
3661 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003662 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003663
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003664.. _DIDerivedType:
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003665
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003666DIDerivedType
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003667"""""""""""""
3668
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003669``DIDerivedType`` nodes represent types derived from other types, such as
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003670qualified types.
3671
3672.. code-block:: llvm
3673
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003674 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003675 encoding: DW_ATE_unsigned_char)
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003676 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003677 align: 32)
3678
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003679The following ``tag:`` values are valid:
3680
3681.. code-block:: llvm
3682
3683 DW_TAG_formal_parameter = 5
3684 DW_TAG_member = 13
3685 DW_TAG_pointer_type = 15
3686 DW_TAG_reference_type = 16
3687 DW_TAG_typedef = 22
3688 DW_TAG_ptr_to_member_type = 31
3689 DW_TAG_const_type = 38
3690 DW_TAG_volatile_type = 53
3691 DW_TAG_restrict_type = 55
3692
3693``DW_TAG_member`` is used to define a member of a :ref:`composite type
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003694<DICompositeType>` or :ref:`subprogram <DISubprogram>`. The type of the member
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003695is the ``baseType:``. The ``offset:`` is the member's bit offset.
3696``DW_TAG_formal_parameter`` is used to define a member which is a formal
3697argument of a subprogram.
3698
3699``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3700
3701``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3702``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3703``baseType:``.
3704
3705Note that the ``void *`` type is expressed as a type derived from NULL.
3706
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003707.. _DICompositeType:
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003708
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003709DICompositeType
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003710"""""""""""""""
3711
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003712``DICompositeType`` nodes represent types composed of other types, like
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003713structures and unions. ``elements:`` points to a tuple of the composed types.
3714
3715If the source language supports ODR, the ``identifier:`` field gives the unique
3716identifier used for type merging between modules. When specified, other types
3717can refer to composite types indirectly via a :ref:`metadata string
3718<metadata-string>` that matches their identifier.
3719
3720.. code-block:: llvm
3721
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003722 !0 = !DIEnumerator(name: "SixKind", value: 7)
3723 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3724 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3725 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003726 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3727 elements: !{!0, !1, !2})
3728
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003729The following ``tag:`` values are valid:
3730
3731.. code-block:: llvm
3732
3733 DW_TAG_array_type = 1
3734 DW_TAG_class_type = 2
3735 DW_TAG_enumeration_type = 4
3736 DW_TAG_structure_type = 19
3737 DW_TAG_union_type = 23
3738 DW_TAG_subroutine_type = 21
3739 DW_TAG_inheritance = 28
3740
3741
3742For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003743descriptors <DISubrange>`, each representing the range of subscripts at that
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003744level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3745array type is a native packed vector.
3746
3747For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003748descriptors <DIEnumerator>`, each representing the definition of an enumeration
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003749value for the set. All enumeration type descriptors are collected in the
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003750``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003751
3752For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3753``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003754<DIDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003755
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003756.. _DISubrange:
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003757
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003758DISubrange
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003759""""""""""
3760
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003761``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3762:ref:`DICompositeType`. ``count: -1`` indicates an empty array.
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003763
3764.. code-block:: llvm
3765
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003766 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
3767 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
3768 !2 = !DISubrange(count: -1) ; empty array.
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003769
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003770.. _DIEnumerator:
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003771
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003772DIEnumerator
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003773""""""""""""
3774
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003775``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3776variants of :ref:`DICompositeType`.
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003777
3778.. code-block:: llvm
3779
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003780 !0 = !DIEnumerator(name: "SixKind", value: 7)
3781 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3782 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003783
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003784DITemplateTypeParameter
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003785"""""""""""""""""""""""
3786
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003787``DITemplateTypeParameter`` nodes represent type parameters to generic source
3788language constructs. They are used (optionally) in :ref:`DICompositeType` and
3789:ref:`DISubprogram` ``templateParams:`` fields.
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003790
3791.. code-block:: llvm
3792
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003793 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003794
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003795DITemplateValueParameter
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003796""""""""""""""""""""""""
3797
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003798``DITemplateValueParameter`` nodes represent value parameters to generic source
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003799language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3800but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3801``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003802:ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003803
3804.. code-block:: llvm
3805
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003806 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003807
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003808DINamespace
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003809"""""""""""
3810
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003811``DINamespace`` nodes represent namespaces in the source language.
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003812
3813.. code-block:: llvm
3814
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003815 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003816
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003817DIGlobalVariable
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003818""""""""""""""""
3819
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003820``DIGlobalVariable`` nodes represent global variables in the source language.
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003821
3822.. code-block:: llvm
3823
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003824 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003825 file: !2, line: 7, type: !3, isLocal: true,
3826 isDefinition: false, variable: i32* @foo,
3827 declaration: !4)
3828
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003829All global variables should be referenced by the `globals:` field of a
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003830:ref:`compile unit <DICompileUnit>`.
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003831
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003832.. _DISubprogram:
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003833
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003834DISubprogram
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003835""""""""""""
3836
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003837``DISubprogram`` nodes represent functions from the source language. The
3838``variables:`` field points at :ref:`variables <DILocalVariable>` that must be
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003839retained, even if their IR counterparts are optimized out of the IR. The
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003840``type:`` field must point at an :ref:`DISubroutineType`.
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003841
3842.. code-block:: llvm
3843
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003844 !0 = !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003845 file: !2, line: 7, type: !3, isLocal: true,
3846 isDefinition: false, scopeLine: 8, containingType: !4,
3847 virtuality: DW_VIRTUALITY_pure_virtual, virtualIndex: 10,
3848 flags: DIFlagPrototyped, isOptimized: true,
3849 function: void ()* @_Z3foov,
3850 templateParams: !5, declaration: !6, variables: !7)
3851
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003852.. _DILexicalBlock:
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003853
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003854DILexicalBlock
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003855""""""""""""""
3856
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003857``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
3858<DISubprogram>`. The line number and column numbers are used to dinstinguish
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003859two lexical blocks at same depth. They are valid targets for ``scope:``
3860fields.
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003861
3862.. code-block:: llvm
3863
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003864 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
Duncan P. N. Exon Smithd937cd92015-03-17 23:41:05 +00003865
3866Usually lexical blocks are ``distinct`` to prevent node merging based on
3867operands.
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003868
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003869.. _DILexicalBlockFile:
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003870
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003871DILexicalBlockFile
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003872""""""""""""""""""
3873
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003874``DILexicalBlockFile`` nodes are used to discriminate between sections of a
3875:ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003876indicate textual inclusion, or the ``discriminator:`` field can be used to
3877discriminate between control flow within a single block in the source language.
3878
3879.. code-block:: llvm
3880
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003881 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
3882 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
3883 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003884
Michael Kuperstein605308a2015-05-14 10:58:59 +00003885.. _DILocation:
3886
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003887DILocation
Duncan P. N. Exon Smith6a484832015-01-13 21:10:44 +00003888""""""""""
3889
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003890``DILocation`` nodes represent source debug locations. The ``scope:`` field is
3891mandatory, and points at an :ref:`DILexicalBlockFile`, an
3892:ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
Duncan P. N. Exon Smith6a484832015-01-13 21:10:44 +00003893
3894.. code-block:: llvm
3895
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003896 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
Duncan P. N. Exon Smith6a484832015-01-13 21:10:44 +00003897
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003898.. _DILocalVariable:
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003899
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003900DILocalVariable
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003901"""""""""""""""
3902
Duncan P. N. Exon Smithed013cd2015-07-31 18:58:39 +00003903``DILocalVariable`` nodes represent local variables in the source language. If
3904the ``arg:`` field is set to non-zero, then this variable is a subprogram
3905parameter, and it will be included in the ``variables:`` field of its
3906:ref:`DISubprogram`.
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003907
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003908.. code-block:: llvm
3909
Duncan P. N. Exon Smithed013cd2015-07-31 18:58:39 +00003910 !0 = !DILocalVariable(name: "this", arg: 1, scope: !3, file: !2, line: 7,
3911 type: !3, flags: DIFlagArtificial)
3912 !1 = !DILocalVariable(name: "x", arg: 2, scope: !4, file: !2, line: 7,
3913 type: !3)
3914 !2 = !DILocalVariable(name: "y", scope: !5, file: !2, line: 7, type: !3)
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003915
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003916DIExpression
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003917""""""""""""
3918
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003919``DIExpression`` nodes represent DWARF expression sequences. They are used in
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003920:ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
3921describe how the referenced LLVM variable relates to the source language
3922variable.
3923
3924The current supported vocabulary is limited:
3925
3926- ``DW_OP_deref`` dereferences the working expression.
3927- ``DW_OP_plus, 93`` adds ``93`` to the working expression.
3928- ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
3929 here, respectively) of the variable piece from the working expression.
3930
3931.. code-block:: llvm
3932
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003933 !0 = !DIExpression(DW_OP_deref)
3934 !1 = !DIExpression(DW_OP_plus, 3)
3935 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
3936 !3 = !DIExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003937
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003938DIObjCProperty
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003939""""""""""""""
3940
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003941``DIObjCProperty`` nodes represent Objective-C property nodes.
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003942
3943.. code-block:: llvm
3944
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003945 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003946 getter: "getFoo", attributes: 7, type: !2)
3947
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003948DIImportedEntity
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003949""""""""""""""""
3950
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003951``DIImportedEntity`` nodes represent entities (such as modules) imported into a
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003952compile unit.
3953
3954.. code-block:: llvm
3955
Duncan P. N. Exon Smitha9308c42015-04-29 16:38:44 +00003956 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +00003957 entity: !1, line: 7)
3958
Sean Silvab084af42012-12-07 10:36:55 +00003959'``tbaa``' Metadata
3960^^^^^^^^^^^^^^^^^^^
3961
3962In LLVM IR, memory does not have types, so LLVM's own type system is not
3963suitable for doing TBAA. Instead, metadata is added to the IR to
3964describe a type system of a higher level language. This can be used to
3965implement typical C/C++ TBAA, but it can also be used to implement
3966custom alias analysis behavior for other languages.
3967
3968The current metadata format is very simple. TBAA metadata nodes have up
3969to three fields, e.g.:
3970
3971.. code-block:: llvm
3972
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00003973 !0 = !{ !"an example type tree" }
3974 !1 = !{ !"int", !0 }
3975 !2 = !{ !"float", !0 }
3976 !3 = !{ !"const float", !2, i64 1 }
Sean Silvab084af42012-12-07 10:36:55 +00003977
3978The first field is an identity field. It can be any value, usually a
3979metadata string, which uniquely identifies the type. The most important
3980name in the tree is the name of the root node. Two trees with different
3981root node names are entirely disjoint, even if they have leaves with
3982common names.
3983
3984The second field identifies the type's parent node in the tree, or is
3985null or omitted for a root node. A type is considered to alias all of
3986its descendants and all of its ancestors in the tree. Also, a type is
3987considered to alias all types in other trees, so that bitcode produced
3988from multiple front-ends is handled conservatively.
3989
3990If the third field is present, it's an integer which if equal to 1
3991indicates that the type is "constant" (meaning
3992``pointsToConstantMemory`` should return true; see `other useful
3993AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
3994
3995'``tbaa.struct``' Metadata
3996^^^^^^^^^^^^^^^^^^^^^^^^^^
3997
3998The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
3999aggregate assignment operations in C and similar languages, however it
4000is defined to copy a contiguous region of memory, which is more than
4001strictly necessary for aggregate types which contain holes due to
4002padding. Also, it doesn't contain any TBAA information about the fields
4003of the aggregate.
4004
4005``!tbaa.struct`` metadata can describe which memory subregions in a
4006memcpy are padding and what the TBAA tags of the struct are.
4007
4008The current metadata format is very simple. ``!tbaa.struct`` metadata
4009nodes are a list of operands which are in conceptual groups of three.
4010For each group of three, the first operand gives the byte offset of a
4011field in bytes, the second gives its size in bytes, and the third gives
4012its tbaa tag. e.g.:
4013
4014.. code-block:: llvm
4015
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004016 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
Sean Silvab084af42012-12-07 10:36:55 +00004017
4018This describes a struct with two fields. The first is at offset 0 bytes
4019with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
4020and has size 4 bytes and has tbaa tag !2.
4021
4022Note that the fields need not be contiguous. In this example, there is a
40234 byte gap between the two fields. This gap represents padding which
4024does not carry useful data and need not be preserved.
4025
Hal Finkel94146652014-07-24 14:25:39 +00004026'``noalias``' and '``alias.scope``' Metadata
Dan Liewbafdcba2014-07-28 13:33:51 +00004027^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Hal Finkel94146652014-07-24 14:25:39 +00004028
4029``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
4030noalias memory-access sets. This means that some collection of memory access
4031instructions (loads, stores, memory-accessing calls, etc.) that carry
4032``noalias`` metadata can specifically be specified not to alias with some other
4033collection of memory access instructions that carry ``alias.scope`` metadata.
Hal Finkel029cde62014-07-25 15:50:02 +00004034Each type of metadata specifies a list of scopes where each scope has an id and
Ed Maste8ed40ce2015-04-14 20:52:58 +00004035a domain. When evaluating an aliasing query, if for some domain, the set
Hal Finkel029cde62014-07-25 15:50:02 +00004036of scopes with that domain in one instruction's ``alias.scope`` list is a
Arch D. Robison96cf7ab2015-02-24 20:11:49 +00004037subset of (or equal to) the set of scopes for that domain in another
Hal Finkel029cde62014-07-25 15:50:02 +00004038instruction's ``noalias`` list, then the two memory accesses are assumed not to
4039alias.
Hal Finkel94146652014-07-24 14:25:39 +00004040
Hal Finkel029cde62014-07-25 15:50:02 +00004041The metadata identifying each domain is itself a list containing one or two
4042entries. The first entry is the name of the domain. Note that if the name is a
Hal Finkel94146652014-07-24 14:25:39 +00004043string then it can be combined accross functions and translation units. A
Hal Finkel029cde62014-07-25 15:50:02 +00004044self-reference can be used to create globally unique domain names. A
4045descriptive string may optionally be provided as a second list entry.
4046
4047The metadata identifying each scope is also itself a list containing two or
4048three entries. The first entry is the name of the scope. Note that if the name
4049is a string then it can be combined accross functions and translation units. A
4050self-reference can be used to create globally unique scope names. A metadata
4051reference to the scope's domain is the second entry. A descriptive string may
4052optionally be provided as a third list entry.
Hal Finkel94146652014-07-24 14:25:39 +00004053
4054For example,
4055
4056.. code-block:: llvm
4057
Hal Finkel029cde62014-07-25 15:50:02 +00004058 ; Two scope domains:
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004059 !0 = !{!0}
4060 !1 = !{!1}
Hal Finkel94146652014-07-24 14:25:39 +00004061
Hal Finkel029cde62014-07-25 15:50:02 +00004062 ; Some scopes in these domains:
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004063 !2 = !{!2, !0}
4064 !3 = !{!3, !0}
4065 !4 = !{!4, !1}
Hal Finkel94146652014-07-24 14:25:39 +00004066
Hal Finkel029cde62014-07-25 15:50:02 +00004067 ; Some scope lists:
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004068 !5 = !{!4} ; A list containing only scope !4
4069 !6 = !{!4, !3, !2}
4070 !7 = !{!3}
Hal Finkel94146652014-07-24 14:25:39 +00004071
4072 ; These two instructions don't alias:
David Blaikiec7aabbb2015-03-04 22:06:14 +00004073 %0 = load float, float* %c, align 4, !alias.scope !5
Hal Finkel029cde62014-07-25 15:50:02 +00004074 store float %0, float* %arrayidx.i, align 4, !noalias !5
Hal Finkel94146652014-07-24 14:25:39 +00004075
Hal Finkel029cde62014-07-25 15:50:02 +00004076 ; These two instructions also don't alias (for domain !1, the set of scopes
4077 ; in the !alias.scope equals that in the !noalias list):
David Blaikiec7aabbb2015-03-04 22:06:14 +00004078 %2 = load float, float* %c, align 4, !alias.scope !5
Hal Finkel029cde62014-07-25 15:50:02 +00004079 store float %2, float* %arrayidx.i2, align 4, !noalias !6
Hal Finkel94146652014-07-24 14:25:39 +00004080
Adam Nemet0a8416f2015-05-11 08:30:28 +00004081 ; These two instructions may alias (for domain !0, the set of scopes in
Hal Finkel029cde62014-07-25 15:50:02 +00004082 ; the !noalias list is not a superset of, or equal to, the scopes in the
4083 ; !alias.scope list):
David Blaikiec7aabbb2015-03-04 22:06:14 +00004084 %2 = load float, float* %c, align 4, !alias.scope !6
Hal Finkel029cde62014-07-25 15:50:02 +00004085 store float %0, float* %arrayidx.i, align 4, !noalias !7
Hal Finkel94146652014-07-24 14:25:39 +00004086
Sean Silvab084af42012-12-07 10:36:55 +00004087'``fpmath``' Metadata
4088^^^^^^^^^^^^^^^^^^^^^
4089
4090``fpmath`` metadata may be attached to any instruction of floating point
4091type. It can be used to express the maximum acceptable error in the
4092result of that instruction, in ULPs, thus potentially allowing the
4093compiler to use a more efficient but less accurate method of computing
4094it. ULP is defined as follows:
4095
4096 If ``x`` is a real number that lies between two finite consecutive
4097 floating-point numbers ``a`` and ``b``, without being equal to one
4098 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
4099 distance between the two non-equal finite floating-point numbers
4100 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
4101
4102The metadata node shall consist of a single positive floating point
4103number representing the maximum relative error, for example:
4104
4105.. code-block:: llvm
4106
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004107 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
Sean Silvab084af42012-12-07 10:36:55 +00004108
Philip Reamesf8bf9dd2015-02-27 23:14:50 +00004109.. _range-metadata:
4110
Sean Silvab084af42012-12-07 10:36:55 +00004111'``range``' Metadata
4112^^^^^^^^^^^^^^^^^^^^
4113
Jingyue Wu37fcb592014-06-19 16:50:16 +00004114``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
4115integer types. It expresses the possible ranges the loaded value or the value
4116returned by the called function at this call site is in. The ranges are
4117represented with a flattened list of integers. The loaded value or the value
4118returned is known to be in the union of the ranges defined by each consecutive
4119pair. Each pair has the following properties:
Sean Silvab084af42012-12-07 10:36:55 +00004120
4121- The type must match the type loaded by the instruction.
4122- The pair ``a,b`` represents the range ``[a,b)``.
4123- Both ``a`` and ``b`` are constants.
4124- The range is allowed to wrap.
4125- The range should not represent the full or empty set. That is,
4126 ``a!=b``.
4127
4128In addition, the pairs must be in signed order of the lower bound and
4129they must be non-contiguous.
4130
4131Examples:
4132
4133.. code-block:: llvm
4134
David Blaikiec7aabbb2015-03-04 22:06:14 +00004135 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
4136 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
Jingyue Wu37fcb592014-06-19 16:50:16 +00004137 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
4138 %d = invoke i8 @bar() to label %cont
4139 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
Sean Silvab084af42012-12-07 10:36:55 +00004140 ...
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004141 !0 = !{ i8 0, i8 2 }
4142 !1 = !{ i8 255, i8 2 }
4143 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
4144 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
Sean Silvab084af42012-12-07 10:36:55 +00004145
Pekka Jaaskelainen0d237252013-02-13 18:08:57 +00004146'``llvm.loop``'
4147^^^^^^^^^^^^^^^
4148
4149It is sometimes useful to attach information to loop constructs. Currently,
4150loop metadata is implemented as metadata attached to the branch instruction
4151in the loop latch block. This type of metadata refer to a metadata node that is
Matt Arsenault24b49c42013-07-31 17:49:08 +00004152guaranteed to be separate for each loop. The loop identifier metadata is
Paul Redmond5fdf8362013-05-28 20:00:34 +00004153specified with the name ``llvm.loop``.
Pekka Jaaskelainen0d237252013-02-13 18:08:57 +00004154
4155The loop identifier metadata is implemented using a metadata that refers to
Michael Liaoa7699082013-03-06 18:24:34 +00004156itself to avoid merging it with any other identifier metadata, e.g.,
4157during module linkage or function inlining. That is, each loop should refer
4158to their own identification metadata even if they reside in separate functions.
4159The following example contains loop identifier metadata for two separate loop
Pekka Jaaskelainen119a2b62013-02-22 12:03:07 +00004160constructs:
Pekka Jaaskelainen0d237252013-02-13 18:08:57 +00004161
4162.. code-block:: llvm
Paul Redmondeaaed3b2013-02-21 17:20:45 +00004163
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004164 !0 = !{!0}
4165 !1 = !{!1}
Pekka Jaaskelainen119a2b62013-02-22 12:03:07 +00004166
Mark Heffernan893752a2014-07-18 19:24:51 +00004167The loop identifier metadata can be used to specify additional
4168per-loop metadata. Any operands after the first operand can be treated
4169as user-defined metadata. For example the ``llvm.loop.unroll.count``
4170suggests an unroll factor to the loop unroller:
Pekka Jaaskelainen0d237252013-02-13 18:08:57 +00004171
Paul Redmond5fdf8362013-05-28 20:00:34 +00004172.. code-block:: llvm
Pekka Jaaskelainen0d237252013-02-13 18:08:57 +00004173
Paul Redmond5fdf8362013-05-28 20:00:34 +00004174 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
4175 ...
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004176 !0 = !{!0, !1}
4177 !1 = !{!"llvm.loop.unroll.count", i32 4}
Mark Heffernan893752a2014-07-18 19:24:51 +00004178
Mark Heffernan9d20e422014-07-21 23:11:03 +00004179'``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
4180^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Mark Heffernan893752a2014-07-18 19:24:51 +00004181
Mark Heffernan9d20e422014-07-21 23:11:03 +00004182Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
4183used to control per-loop vectorization and interleaving parameters such as
4184vectorization width and interleave count. These metadata should be used in
Mark Heffernan893752a2014-07-18 19:24:51 +00004185conjunction with ``llvm.loop`` loop identification metadata. The
Mark Heffernan9d20e422014-07-21 23:11:03 +00004186``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
4187optimization hints and the optimizer will only interleave and vectorize loops if
4188it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
4189which contains information about loop-carried memory dependencies can be helpful
4190in determining the safety of these transformations.
Mark Heffernan893752a2014-07-18 19:24:51 +00004191
Mark Heffernan9d20e422014-07-21 23:11:03 +00004192'``llvm.loop.interleave.count``' Metadata
Mark Heffernan893752a2014-07-18 19:24:51 +00004193^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4194
Mark Heffernan9d20e422014-07-21 23:11:03 +00004195This metadata suggests an interleave count to the loop interleaver.
4196The first operand is the string ``llvm.loop.interleave.count`` and the
Mark Heffernan893752a2014-07-18 19:24:51 +00004197second operand is an integer specifying the interleave count. For
4198example:
4199
4200.. code-block:: llvm
4201
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004202 !0 = !{!"llvm.loop.interleave.count", i32 4}
Mark Heffernan893752a2014-07-18 19:24:51 +00004203
Mark Heffernan9d20e422014-07-21 23:11:03 +00004204Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
4205multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
4206then the interleave count will be determined automatically.
4207
4208'``llvm.loop.vectorize.enable``' Metadata
Dan Liew9a1829d2014-07-22 14:59:38 +00004209^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Mark Heffernan9d20e422014-07-21 23:11:03 +00004210
4211This metadata selectively enables or disables vectorization for the loop. The
4212first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
4213is a bit. If the bit operand value is 1 vectorization is enabled. A value of
42140 disables vectorization:
4215
4216.. code-block:: llvm
4217
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004218 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
4219 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
Mark Heffernan893752a2014-07-18 19:24:51 +00004220
4221'``llvm.loop.vectorize.width``' Metadata
4222^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4223
4224This metadata sets the target width of the vectorizer. The first
4225operand is the string ``llvm.loop.vectorize.width`` and the second
4226operand is an integer specifying the width. For example:
4227
4228.. code-block:: llvm
4229
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004230 !0 = !{!"llvm.loop.vectorize.width", i32 4}
Mark Heffernan893752a2014-07-18 19:24:51 +00004231
4232Note that setting ``llvm.loop.vectorize.width`` to 1 disables
4233vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
42340 or if the loop does not have this metadata the width will be
4235determined automatically.
4236
4237'``llvm.loop.unroll``'
4238^^^^^^^^^^^^^^^^^^^^^^
4239
4240Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
4241optimization hints such as the unroll factor. ``llvm.loop.unroll``
4242metadata should be used in conjunction with ``llvm.loop`` loop
4243identification metadata. The ``llvm.loop.unroll`` metadata are only
4244optimization hints and the unrolling will only be performed if the
4245optimizer believes it is safe to do so.
4246
Mark Heffernan893752a2014-07-18 19:24:51 +00004247'``llvm.loop.unroll.count``' Metadata
4248^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4249
4250This metadata suggests an unroll factor to the loop unroller. The
4251first operand is the string ``llvm.loop.unroll.count`` and the second
4252operand is a positive integer specifying the unroll factor. For
4253example:
4254
4255.. code-block:: llvm
4256
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004257 !0 = !{!"llvm.loop.unroll.count", i32 4}
Mark Heffernan893752a2014-07-18 19:24:51 +00004258
4259If the trip count of the loop is less than the unroll count the loop
4260will be partially unrolled.
4261
Mark Heffernane6b4ba12014-07-23 17:31:37 +00004262'``llvm.loop.unroll.disable``' Metadata
4263^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4264
Mark Heffernan3e32a4e2015-06-30 22:48:51 +00004265This metadata disables loop unrolling. The metadata has a single operand
Mark Heffernane6b4ba12014-07-23 17:31:37 +00004266which is the string ``llvm.loop.unroll.disable``. For example:
4267
4268.. code-block:: llvm
4269
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004270 !0 = !{!"llvm.loop.unroll.disable"}
Mark Heffernane6b4ba12014-07-23 17:31:37 +00004271
Kevin Qin715b01e2015-03-09 06:14:18 +00004272'``llvm.loop.unroll.runtime.disable``' Metadata
Dan Liew868b0742015-03-11 13:34:49 +00004273^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Kevin Qin715b01e2015-03-09 06:14:18 +00004274
Mark Heffernan3e32a4e2015-06-30 22:48:51 +00004275This metadata disables runtime loop unrolling. The metadata has a single
Kevin Qin715b01e2015-03-09 06:14:18 +00004276operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
4277
4278.. code-block:: llvm
4279
4280 !0 = !{!"llvm.loop.unroll.runtime.disable"}
4281
Mark Heffernane6b4ba12014-07-23 17:31:37 +00004282'``llvm.loop.unroll.full``' Metadata
4283^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4284
Mark Heffernan3e32a4e2015-06-30 22:48:51 +00004285This metadata suggests that the loop should be unrolled fully. The
4286metadata has a single operand which is the string ``llvm.loop.unroll.full``.
Mark Heffernane6b4ba12014-07-23 17:31:37 +00004287For example:
4288
4289.. code-block:: llvm
4290
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004291 !0 = !{!"llvm.loop.unroll.full"}
Pekka Jaaskelainen0d237252013-02-13 18:08:57 +00004292
4293'``llvm.mem``'
4294^^^^^^^^^^^^^^^
4295
4296Metadata types used to annotate memory accesses with information helpful
4297for optimizations are prefixed with ``llvm.mem``.
4298
4299'``llvm.mem.parallel_loop_access``' Metadata
4300^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4301
Mehdi Amini4a121fa2015-03-14 22:04:06 +00004302The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
4303or metadata containing a list of loop identifiers for nested loops.
4304The metadata is attached to memory accessing instructions and denotes that
4305no loop carried memory dependence exist between it and other instructions denoted
Pekka Jaaskelainen23b222cc2014-05-23 11:35:46 +00004306with the same loop identifier.
4307
Mehdi Amini4a121fa2015-03-14 22:04:06 +00004308Precisely, given two instructions ``m1`` and ``m2`` that both have the
4309``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
4310set of loops associated with that metadata, respectively, then there is no loop
4311carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
Pekka Jaaskelainen23b222cc2014-05-23 11:35:46 +00004312``L2``.
4313
Mehdi Amini4a121fa2015-03-14 22:04:06 +00004314As a special case, if all memory accessing instructions in a loop have
4315``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
4316loop has no loop carried memory dependences and is considered to be a parallel
4317loop.
Pekka Jaaskelainen23b222cc2014-05-23 11:35:46 +00004318
Mehdi Amini4a121fa2015-03-14 22:04:06 +00004319Note that if not all memory access instructions have such metadata referring to
4320the loop, then the loop is considered not being trivially parallel. Additional
4321memory dependence analysis is required to make that determination. As a fail
4322safe mechanism, this causes loops that were originally parallel to be considered
4323sequential (if optimization passes that are unaware of the parallel semantics
Pekka Jaaskelainen23b222cc2014-05-23 11:35:46 +00004324insert new memory instructions into the loop body).
Pekka Jaaskelainen0d237252013-02-13 18:08:57 +00004325
4326Example of a loop that is considered parallel due to its correct use of
Paul Redmond5fdf8362013-05-28 20:00:34 +00004327both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
Pekka Jaaskelainen0d237252013-02-13 18:08:57 +00004328metadata types that refer to the same loop identifier metadata.
4329
4330.. code-block:: llvm
4331
4332 for.body:
Paul Redmond5fdf8362013-05-28 20:00:34 +00004333 ...
David Blaikiec7aabbb2015-03-04 22:06:14 +00004334 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
Paul Redmond5fdf8362013-05-28 20:00:34 +00004335 ...
Tobias Grosserfbe95dc2014-03-05 13:36:04 +00004336 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
Paul Redmond5fdf8362013-05-28 20:00:34 +00004337 ...
4338 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
Pekka Jaaskelainen0d237252013-02-13 18:08:57 +00004339
4340 for.end:
4341 ...
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004342 !0 = !{!0}
Pekka Jaaskelainen0d237252013-02-13 18:08:57 +00004343
4344It is also possible to have nested parallel loops. In that case the
4345memory accesses refer to a list of loop identifier metadata nodes instead of
4346the loop identifier metadata node directly:
4347
4348.. code-block:: llvm
4349
4350 outer.for.body:
Tobias Grosserfbe95dc2014-03-05 13:36:04 +00004351 ...
David Blaikiec7aabbb2015-03-04 22:06:14 +00004352 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
Tobias Grosserfbe95dc2014-03-05 13:36:04 +00004353 ...
4354 br label %inner.for.body
Pekka Jaaskelainen0d237252013-02-13 18:08:57 +00004355
4356 inner.for.body:
Paul Redmond5fdf8362013-05-28 20:00:34 +00004357 ...
David Blaikiec7aabbb2015-03-04 22:06:14 +00004358 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
Paul Redmond5fdf8362013-05-28 20:00:34 +00004359 ...
Tobias Grosserfbe95dc2014-03-05 13:36:04 +00004360 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
Paul Redmond5fdf8362013-05-28 20:00:34 +00004361 ...
4362 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
Pekka Jaaskelainen0d237252013-02-13 18:08:57 +00004363
4364 inner.for.end:
Paul Redmond5fdf8362013-05-28 20:00:34 +00004365 ...
Tobias Grosserfbe95dc2014-03-05 13:36:04 +00004366 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
Paul Redmond5fdf8362013-05-28 20:00:34 +00004367 ...
4368 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
Pekka Jaaskelainen0d237252013-02-13 18:08:57 +00004369
4370 outer.for.end: ; preds = %for.body
4371 ...
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004372 !0 = !{!1, !2} ; a list of loop identifiers
4373 !1 = !{!1} ; an identifier for the inner loop
4374 !2 = !{!2} ; an identifier for the outer loop
Pekka Jaaskelainen0d237252013-02-13 18:08:57 +00004375
Peter Collingbournee6909c82015-02-20 20:30:47 +00004376'``llvm.bitsets``'
4377^^^^^^^^^^^^^^^^^^
4378
4379The ``llvm.bitsets`` global metadata is used to implement
4380:doc:`bitsets <BitSets>`.
4381
Sean Silvab084af42012-12-07 10:36:55 +00004382Module Flags Metadata
4383=====================
4384
4385Information about the module as a whole is difficult to convey to LLVM's
4386subsystems. The LLVM IR isn't sufficient to transmit this information.
4387The ``llvm.module.flags`` named metadata exists in order to facilitate
Dmitri Gribenkoe8131122013-01-19 20:34:20 +00004388this. These flags are in the form of key / value pairs --- much like a
4389dictionary --- making it easy for any subsystem who cares about a flag to
Sean Silvab084af42012-12-07 10:36:55 +00004390look it up.
4391
4392The ``llvm.module.flags`` metadata contains a list of metadata triplets.
4393Each triplet has the following form:
4394
4395- The first element is a *behavior* flag, which specifies the behavior
4396 when two (or more) modules are merged together, and it encounters two
4397 (or more) metadata with the same ID. The supported behaviors are
4398 described below.
4399- The second element is a metadata string that is a unique ID for the
Daniel Dunbar25c4b572013-01-15 01:22:53 +00004400 metadata. Each module may only have one flag entry for each unique ID (not
4401 including entries with the **Require** behavior).
Sean Silvab084af42012-12-07 10:36:55 +00004402- The third element is the value of the flag.
4403
4404When two (or more) modules are merged together, the resulting
Daniel Dunbar25c4b572013-01-15 01:22:53 +00004405``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
4406each unique metadata ID string, there will be exactly one entry in the merged
4407modules ``llvm.module.flags`` metadata table, and the value for that entry will
4408be determined by the merge behavior flag, as described below. The only exception
4409is that entries with the *Require* behavior are always preserved.
Sean Silvab084af42012-12-07 10:36:55 +00004410
4411The following behaviors are supported:
4412
4413.. list-table::
4414 :header-rows: 1
4415 :widths: 10 90
4416
4417 * - Value
4418 - Behavior
4419
4420 * - 1
4421 - **Error**
Daniel Dunbar25c4b572013-01-15 01:22:53 +00004422 Emits an error if two values disagree, otherwise the resulting value
4423 is that of the operands.
Sean Silvab084af42012-12-07 10:36:55 +00004424
4425 * - 2
4426 - **Warning**
Daniel Dunbar25c4b572013-01-15 01:22:53 +00004427 Emits a warning if two values disagree. The result value will be the
4428 operand for the flag from the first module being linked.
Sean Silvab084af42012-12-07 10:36:55 +00004429
4430 * - 3
4431 - **Require**
Daniel Dunbar25c4b572013-01-15 01:22:53 +00004432 Adds a requirement that another module flag be present and have a
4433 specified value after linking is performed. The value must be a
4434 metadata pair, where the first element of the pair is the ID of the
4435 module flag to be restricted, and the second element of the pair is
4436 the value the module flag should be restricted to. This behavior can
4437 be used to restrict the allowable results (via triggering of an
4438 error) of linking IDs with the **Override** behavior.
Sean Silvab084af42012-12-07 10:36:55 +00004439
4440 * - 4
4441 - **Override**
Daniel Dunbar25c4b572013-01-15 01:22:53 +00004442 Uses the specified value, regardless of the behavior or value of the
4443 other module. If both modules specify **Override**, but the values
4444 differ, an error will be emitted.
4445
Daniel Dunbard77d9fb2013-01-16 21:38:56 +00004446 * - 5
4447 - **Append**
4448 Appends the two values, which are required to be metadata nodes.
4449
4450 * - 6
4451 - **AppendUnique**
4452 Appends the two values, which are required to be metadata
4453 nodes. However, duplicate entries in the second list are dropped
4454 during the append operation.
4455
Daniel Dunbar25c4b572013-01-15 01:22:53 +00004456It is an error for a particular unique flag ID to have multiple behaviors,
4457except in the case of **Require** (which adds restrictions on another metadata
4458value) or **Override**.
Sean Silvab084af42012-12-07 10:36:55 +00004459
4460An example of module flags:
4461
4462.. code-block:: llvm
4463
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004464 !0 = !{ i32 1, !"foo", i32 1 }
4465 !1 = !{ i32 4, !"bar", i32 37 }
4466 !2 = !{ i32 2, !"qux", i32 42 }
4467 !3 = !{ i32 3, !"qux",
4468 !{
4469 !"foo", i32 1
Sean Silvab084af42012-12-07 10:36:55 +00004470 }
4471 }
4472 !llvm.module.flags = !{ !0, !1, !2, !3 }
4473
4474- Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
4475 if two or more ``!"foo"`` flags are seen is to emit an error if their
4476 values are not equal.
4477
4478- Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
4479 behavior if two or more ``!"bar"`` flags are seen is to use the value
Daniel Dunbar25c4b572013-01-15 01:22:53 +00004480 '37'.
Sean Silvab084af42012-12-07 10:36:55 +00004481
4482- Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
4483 behavior if two or more ``!"qux"`` flags are seen is to emit a
4484 warning if their values are not equal.
4485
4486- Metadata ``!3`` has the ID ``!"qux"`` and the value:
4487
4488 ::
4489
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004490 !{ !"foo", i32 1 }
Sean Silvab084af42012-12-07 10:36:55 +00004491
Daniel Dunbar25c4b572013-01-15 01:22:53 +00004492 The behavior is to emit an error if the ``llvm.module.flags`` does not
4493 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
4494 performed.
Sean Silvab084af42012-12-07 10:36:55 +00004495
4496Objective-C Garbage Collection Module Flags Metadata
4497----------------------------------------------------
4498
4499On the Mach-O platform, Objective-C stores metadata about garbage
4500collection in a special section called "image info". The metadata
4501consists of a version number and a bitmask specifying what types of
4502garbage collection are supported (if any) by the file. If two or more
4503modules are linked together their garbage collection metadata needs to
4504be merged rather than appended together.
4505
4506The Objective-C garbage collection module flags metadata consists of the
4507following key-value pairs:
4508
4509.. list-table::
4510 :header-rows: 1
4511 :widths: 30 70
4512
4513 * - Key
4514 - Value
4515
Daniel Dunbar1dc66ca2013-01-17 18:57:32 +00004516 * - ``Objective-C Version``
Dmitri Gribenkoe8131122013-01-19 20:34:20 +00004517 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
Sean Silvab084af42012-12-07 10:36:55 +00004518
Daniel Dunbar1dc66ca2013-01-17 18:57:32 +00004519 * - ``Objective-C Image Info Version``
Dmitri Gribenkoe8131122013-01-19 20:34:20 +00004520 - **[Required]** --- The version of the image info section. Currently
Sean Silvab084af42012-12-07 10:36:55 +00004521 always 0.
4522
Daniel Dunbar1dc66ca2013-01-17 18:57:32 +00004523 * - ``Objective-C Image Info Section``
Dmitri Gribenkoe8131122013-01-19 20:34:20 +00004524 - **[Required]** --- The section to place the metadata. Valid values are
Sean Silvab084af42012-12-07 10:36:55 +00004525 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
4526 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
4527 Objective-C ABI version 2.
4528
Daniel Dunbar1dc66ca2013-01-17 18:57:32 +00004529 * - ``Objective-C Garbage Collection``
Dmitri Gribenkoe8131122013-01-19 20:34:20 +00004530 - **[Required]** --- Specifies whether garbage collection is supported or
Sean Silvab084af42012-12-07 10:36:55 +00004531 not. Valid values are 0, for no garbage collection, and 2, for garbage
4532 collection supported.
4533
Daniel Dunbar1dc66ca2013-01-17 18:57:32 +00004534 * - ``Objective-C GC Only``
Dmitri Gribenkoe8131122013-01-19 20:34:20 +00004535 - **[Optional]** --- Specifies that only garbage collection is supported.
Sean Silvab084af42012-12-07 10:36:55 +00004536 If present, its value must be 6. This flag requires that the
4537 ``Objective-C Garbage Collection`` flag have the value 2.
4538
4539Some important flag interactions:
4540
4541- If a module with ``Objective-C Garbage Collection`` set to 0 is
4542 merged with a module with ``Objective-C Garbage Collection`` set to
4543 2, then the resulting module has the
4544 ``Objective-C Garbage Collection`` flag set to 0.
4545- A module with ``Objective-C Garbage Collection`` set to 0 cannot be
4546 merged with a module with ``Objective-C GC Only`` set to 6.
4547
Daniel Dunbar252bedc2013-01-17 00:16:27 +00004548Automatic Linker Flags Module Flags Metadata
4549--------------------------------------------
4550
4551Some targets support embedding flags to the linker inside individual object
4552files. Typically this is used in conjunction with language extensions which
4553allow source files to explicitly declare the libraries they depend on, and have
4554these automatically be transmitted to the linker via object files.
4555
4556These flags are encoded in the IR using metadata in the module flags section,
Daniel Dunbar1dc66ca2013-01-17 18:57:32 +00004557using the ``Linker Options`` key. The merge behavior for this flag is required
Daniel Dunbar252bedc2013-01-17 00:16:27 +00004558to be ``AppendUnique``, and the value for the key is expected to be a metadata
4559node which should be a list of other metadata nodes, each of which should be a
4560list of metadata strings defining linker options.
4561
4562For example, the following metadata section specifies two separate sets of
4563linker options, presumably to link against ``libz`` and the ``Cocoa``
4564framework::
4565
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004566 !0 = !{ i32 6, !"Linker Options",
4567 !{
4568 !{ !"-lz" },
4569 !{ !"-framework", !"Cocoa" } } }
Daniel Dunbar252bedc2013-01-17 00:16:27 +00004570 !llvm.module.flags = !{ !0 }
4571
4572The metadata encoding as lists of lists of options, as opposed to a collapsed
4573list of options, is chosen so that the IR encoding can use multiple option
4574strings to specify e.g., a single library, while still having that specifier be
4575preserved as an atomic element that can be recognized by a target specific
4576assembly writer or object file emitter.
4577
4578Each individual option is required to be either a valid option for the target's
4579linker, or an option that is reserved by the target specific assembly writer or
4580object file emitter. No other aspect of these options is defined by the IR.
4581
Oliver Stannard5dc29342014-06-20 10:08:11 +00004582C type width Module Flags Metadata
4583----------------------------------
4584
4585The ARM backend emits a section into each generated object file describing the
4586options that it was compiled with (in a compiler-independent way) to prevent
4587linking incompatible objects, and to allow automatic library selection. Some
4588of these options are not visible at the IR level, namely wchar_t width and enum
4589width.
4590
4591To pass this information to the backend, these options are encoded in module
4592flags metadata, using the following key-value pairs:
4593
4594.. list-table::
4595 :header-rows: 1
4596 :widths: 30 70
4597
4598 * - Key
4599 - Value
4600
4601 * - short_wchar
4602 - * 0 --- sizeof(wchar_t) == 4
4603 * 1 --- sizeof(wchar_t) == 2
4604
4605 * - short_enum
4606 - * 0 --- Enums are at least as large as an ``int``.
4607 * 1 --- Enums are stored in the smallest integer type which can
4608 represent all of its values.
4609
4610For example, the following metadata section specifies that the module was
4611compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
4612enum is the smallest type which can represent all of its values::
4613
4614 !llvm.module.flags = !{!0, !1}
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00004615 !0 = !{i32 1, !"short_wchar", i32 1}
4616 !1 = !{i32 1, !"short_enum", i32 0}
Oliver Stannard5dc29342014-06-20 10:08:11 +00004617
Eli Bendersky0220e6b2013-06-07 20:24:43 +00004618.. _intrinsicglobalvariables:
4619
Sean Silvab084af42012-12-07 10:36:55 +00004620Intrinsic Global Variables
4621==========================
4622
4623LLVM has a number of "magic" global variables that contain data that
4624affect code generation or other IR semantics. These are documented here.
4625All globals of this sort should have a section specified as
4626"``llvm.metadata``". This section and all globals that start with
4627"``llvm.``" are reserved for use by LLVM.
4628
Eli Bendersky0220e6b2013-06-07 20:24:43 +00004629.. _gv_llvmused:
4630
Sean Silvab084af42012-12-07 10:36:55 +00004631The '``llvm.used``' Global Variable
4632-----------------------------------
4633
Rafael Espindola74f2e462013-04-22 14:58:02 +00004634The ``@llvm.used`` global is an array which has
Paul Redmond219ef812013-05-30 17:24:32 +00004635:ref:`appending linkage <linkage_appending>`. This array contains a list of
Rafael Espindola70a729d2013-06-11 13:18:13 +00004636pointers to named global variables, functions and aliases which may optionally
4637have a pointer cast formed of bitcast or getelementptr. For example, a legal
Sean Silvab084af42012-12-07 10:36:55 +00004638use of it is:
4639
4640.. code-block:: llvm
4641
4642 @X = global i8 4
4643 @Y = global i32 123
4644
4645 @llvm.used = appending global [2 x i8*] [
4646 i8* @X,
4647 i8* bitcast (i32* @Y to i8*)
4648 ], section "llvm.metadata"
4649
Rafael Espindola74f2e462013-04-22 14:58:02 +00004650If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
4651and linker are required to treat the symbol as if there is a reference to the
Rafael Espindola70a729d2013-06-11 13:18:13 +00004652symbol that it cannot see (which is why they have to be named). For example, if
4653a variable has internal linkage and no references other than that from the
4654``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
4655references from inline asms and other things the compiler cannot "see", and
4656corresponds to "``attribute((used))``" in GNU C.
Sean Silvab084af42012-12-07 10:36:55 +00004657
4658On some targets, the code generator must emit a directive to the
4659assembler or object file to prevent the assembler and linker from
4660molesting the symbol.
4661
Eli Bendersky0220e6b2013-06-07 20:24:43 +00004662.. _gv_llvmcompilerused:
4663
Sean Silvab084af42012-12-07 10:36:55 +00004664The '``llvm.compiler.used``' Global Variable
4665--------------------------------------------
4666
4667The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4668directive, except that it only prevents the compiler from touching the
4669symbol. On targets that support it, this allows an intelligent linker to
4670optimize references to the symbol without being impeded as it would be
4671by ``@llvm.used``.
4672
4673This is a rare construct that should only be used in rare circumstances,
4674and should not be exposed to source languages.
4675
Eli Bendersky0220e6b2013-06-07 20:24:43 +00004676.. _gv_llvmglobalctors:
4677
Sean Silvab084af42012-12-07 10:36:55 +00004678The '``llvm.global_ctors``' Global Variable
4679-------------------------------------------
4680
4681.. code-block:: llvm
4682
Reid Klecknerfceb76f2014-05-16 20:39:27 +00004683 %0 = type { i32, void ()*, i8* }
4684 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
Sean Silvab084af42012-12-07 10:36:55 +00004685
4686The ``@llvm.global_ctors`` array contains a list of constructor
Reid Klecknerfceb76f2014-05-16 20:39:27 +00004687functions, priorities, and an optional associated global or function.
4688The functions referenced by this array will be called in ascending order
4689of priority (i.e. lowest first) when the module is loaded. The order of
4690functions with the same priority is not defined.
4691
4692If the third field is present, non-null, and points to a global variable
4693or function, the initializer function will only run if the associated
4694data from the current module is not discarded.
Sean Silvab084af42012-12-07 10:36:55 +00004695
Eli Bendersky0220e6b2013-06-07 20:24:43 +00004696.. _llvmglobaldtors:
4697
Sean Silvab084af42012-12-07 10:36:55 +00004698The '``llvm.global_dtors``' Global Variable
4699-------------------------------------------
4700
4701.. code-block:: llvm
4702
Reid Klecknerfceb76f2014-05-16 20:39:27 +00004703 %0 = type { i32, void ()*, i8* }
4704 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
Sean Silvab084af42012-12-07 10:36:55 +00004705
Reid Klecknerfceb76f2014-05-16 20:39:27 +00004706The ``@llvm.global_dtors`` array contains a list of destructor
4707functions, priorities, and an optional associated global or function.
4708The functions referenced by this array will be called in descending
Reid Klecknerbffbcc52014-05-27 21:35:17 +00004709order of priority (i.e. highest first) when the module is unloaded. The
Reid Klecknerfceb76f2014-05-16 20:39:27 +00004710order of functions with the same priority is not defined.
4711
4712If the third field is present, non-null, and points to a global variable
4713or function, the destructor function will only run if the associated
4714data from the current module is not discarded.
Sean Silvab084af42012-12-07 10:36:55 +00004715
4716Instruction Reference
4717=====================
4718
4719The LLVM instruction set consists of several different classifications
4720of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
4721instructions <binaryops>`, :ref:`bitwise binary
4722instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
4723:ref:`other instructions <otherops>`.
4724
4725.. _terminators:
4726
4727Terminator Instructions
4728-----------------------
4729
4730As mentioned :ref:`previously <functionstructure>`, every basic block in a
4731program ends with a "Terminator" instruction, which indicates which
4732block should be executed after the current block is finished. These
4733terminator instructions typically yield a '``void``' value: they produce
4734control flow, not values (the one exception being the
4735':ref:`invoke <i_invoke>`' instruction).
4736
4737The terminator instructions are: ':ref:`ret <i_ret>`',
4738':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
4739':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
David Majnemer654e1302015-07-31 17:58:14 +00004740':ref:`resume <i_resume>`', ':ref:`catchpad <i_catchpad>`',
4741':ref:`catchendpad <i_catchendpad>`',
4742':ref:`catchret <i_catchret>`',
4743':ref:`cleanupret <i_cleanupret>`',
4744':ref:`terminatepad <i_terminatepad>`',
4745and ':ref:`unreachable <i_unreachable>`'.
Sean Silvab084af42012-12-07 10:36:55 +00004746
4747.. _i_ret:
4748
4749'``ret``' Instruction
4750^^^^^^^^^^^^^^^^^^^^^
4751
4752Syntax:
4753"""""""
4754
4755::
4756
4757 ret <type> <value> ; Return a value from a non-void function
4758 ret void ; Return from void function
4759
4760Overview:
4761"""""""""
4762
4763The '``ret``' instruction is used to return control flow (and optionally
4764a value) from a function back to the caller.
4765
4766There are two forms of the '``ret``' instruction: one that returns a
4767value and then causes control flow, and one that just causes control
4768flow to occur.
4769
4770Arguments:
4771""""""""""
4772
4773The '``ret``' instruction optionally accepts a single argument, the
4774return value. The type of the return value must be a ':ref:`first
4775class <t_firstclass>`' type.
4776
4777A function is not :ref:`well formed <wellformed>` if it it has a non-void
4778return type and contains a '``ret``' instruction with no return value or
4779a return value with a type that does not match its type, or if it has a
4780void return type and contains a '``ret``' instruction with a return
4781value.
4782
4783Semantics:
4784""""""""""
4785
4786When the '``ret``' instruction is executed, control flow returns back to
4787the calling function's context. If the caller is a
4788":ref:`call <i_call>`" instruction, execution continues at the
4789instruction after the call. If the caller was an
4790":ref:`invoke <i_invoke>`" instruction, execution continues at the
4791beginning of the "normal" destination block. If the instruction returns
4792a value, that value shall set the call or invoke instruction's return
4793value.
4794
4795Example:
4796""""""""
4797
4798.. code-block:: llvm
4799
4800 ret i32 5 ; Return an integer value of 5
4801 ret void ; Return from a void function
4802 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
4803
4804.. _i_br:
4805
4806'``br``' Instruction
4807^^^^^^^^^^^^^^^^^^^^
4808
4809Syntax:
4810"""""""
4811
4812::
4813
4814 br i1 <cond>, label <iftrue>, label <iffalse>
4815 br label <dest> ; Unconditional branch
4816
4817Overview:
4818"""""""""
4819
4820The '``br``' instruction is used to cause control flow to transfer to a
4821different basic block in the current function. There are two forms of
4822this instruction, corresponding to a conditional branch and an
4823unconditional branch.
4824
4825Arguments:
4826""""""""""
4827
4828The conditional branch form of the '``br``' instruction takes a single
4829'``i1``' value and two '``label``' values. The unconditional form of the
4830'``br``' instruction takes a single '``label``' value as a target.
4831
4832Semantics:
4833""""""""""
4834
4835Upon execution of a conditional '``br``' instruction, the '``i1``'
4836argument is evaluated. If the value is ``true``, control flows to the
4837'``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
4838to the '``iffalse``' ``label`` argument.
4839
4840Example:
4841""""""""
4842
4843.. code-block:: llvm
4844
4845 Test:
4846 %cond = icmp eq i32 %a, %b
4847 br i1 %cond, label %IfEqual, label %IfUnequal
4848 IfEqual:
4849 ret i32 1
4850 IfUnequal:
4851 ret i32 0
4852
4853.. _i_switch:
4854
4855'``switch``' Instruction
4856^^^^^^^^^^^^^^^^^^^^^^^^
4857
4858Syntax:
4859"""""""
4860
4861::
4862
4863 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
4864
4865Overview:
4866"""""""""
4867
4868The '``switch``' instruction is used to transfer control flow to one of
4869several different places. It is a generalization of the '``br``'
4870instruction, allowing a branch to occur to one of many possible
4871destinations.
4872
4873Arguments:
4874""""""""""
4875
4876The '``switch``' instruction uses three parameters: an integer
4877comparison value '``value``', a default '``label``' destination, and an
4878array of pairs of comparison value constants and '``label``'s. The table
4879is not allowed to contain duplicate constant entries.
4880
4881Semantics:
4882""""""""""
4883
4884The ``switch`` instruction specifies a table of values and destinations.
4885When the '``switch``' instruction is executed, this table is searched
4886for the given value. If the value is found, control flow is transferred
4887to the corresponding destination; otherwise, control flow is transferred
4888to the default destination.
4889
4890Implementation:
4891"""""""""""""""
4892
4893Depending on properties of the target machine and the particular
4894``switch`` instruction, this instruction may be code generated in
4895different ways. For example, it could be generated as a series of
4896chained conditional branches or with a lookup table.
4897
4898Example:
4899""""""""
4900
4901.. code-block:: llvm
4902
4903 ; Emulate a conditional br instruction
4904 %Val = zext i1 %value to i32
4905 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
4906
4907 ; Emulate an unconditional br instruction
4908 switch i32 0, label %dest [ ]
4909
4910 ; Implement a jump table:
4911 switch i32 %val, label %otherwise [ i32 0, label %onzero
4912 i32 1, label %onone
4913 i32 2, label %ontwo ]
4914
4915.. _i_indirectbr:
4916
4917'``indirectbr``' Instruction
4918^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4919
4920Syntax:
4921"""""""
4922
4923::
4924
4925 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
4926
4927Overview:
4928"""""""""
4929
4930The '``indirectbr``' instruction implements an indirect branch to a
4931label within the current function, whose address is specified by
4932"``address``". Address must be derived from a
4933:ref:`blockaddress <blockaddress>` constant.
4934
4935Arguments:
4936""""""""""
4937
4938The '``address``' argument is the address of the label to jump to. The
4939rest of the arguments indicate the full set of possible destinations
4940that the address may point to. Blocks are allowed to occur multiple
4941times in the destination list, though this isn't particularly useful.
4942
4943This destination list is required so that dataflow analysis has an
4944accurate understanding of the CFG.
4945
4946Semantics:
4947""""""""""
4948
4949Control transfers to the block specified in the address argument. All
4950possible destination blocks must be listed in the label list, otherwise
4951this instruction has undefined behavior. This implies that jumps to
4952labels defined in other functions have undefined behavior as well.
4953
4954Implementation:
4955"""""""""""""""
4956
4957This is typically implemented with a jump through a register.
4958
4959Example:
4960""""""""
4961
4962.. code-block:: llvm
4963
4964 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
4965
4966.. _i_invoke:
4967
4968'``invoke``' Instruction
4969^^^^^^^^^^^^^^^^^^^^^^^^
4970
4971Syntax:
4972"""""""
4973
4974::
4975
4976 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
4977 to label <normal label> unwind label <exception label>
4978
4979Overview:
4980"""""""""
4981
4982The '``invoke``' instruction causes control to transfer to a specified
4983function, with the possibility of control flow transfer to either the
4984'``normal``' label or the '``exception``' label. If the callee function
4985returns with the "``ret``" instruction, control flow will return to the
4986"normal" label. If the callee (or any indirect callees) returns via the
4987":ref:`resume <i_resume>`" instruction or other exception handling
4988mechanism, control is interrupted and continued at the dynamically
4989nearest "exception" label.
4990
4991The '``exception``' label is a `landing
4992pad <ExceptionHandling.html#overview>`_ for the exception. As such,
4993'``exception``' label is required to have the
4994":ref:`landingpad <i_landingpad>`" instruction, which contains the
4995information about the behavior of the program after unwinding happens,
4996as its first non-PHI instruction. The restrictions on the
4997"``landingpad``" instruction's tightly couples it to the "``invoke``"
4998instruction, so that the important information contained within the
4999"``landingpad``" instruction can't be lost through normal code motion.
5000
5001Arguments:
5002""""""""""
5003
5004This instruction requires several arguments:
5005
5006#. The optional "cconv" marker indicates which :ref:`calling
5007 convention <callingconv>` the call should use. If none is
5008 specified, the call defaults to using C calling conventions.
5009#. The optional :ref:`Parameter Attributes <paramattrs>` list for return
5010 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
5011 are valid here.
5012#. '``ptr to function ty``': shall be the signature of the pointer to
5013 function value being invoked. In most cases, this is a direct
5014 function invocation, but indirect ``invoke``'s are just as possible,
5015 branching off an arbitrary pointer to function value.
5016#. '``function ptr val``': An LLVM value containing a pointer to a
5017 function to be invoked.
5018#. '``function args``': argument list whose types match the function
5019 signature argument types and parameter attributes. All arguments must
5020 be of :ref:`first class <t_firstclass>` type. If the function signature
5021 indicates the function accepts a variable number of arguments, the
5022 extra arguments can be specified.
5023#. '``normal label``': the label reached when the called function
5024 executes a '``ret``' instruction.
5025#. '``exception label``': the label reached when a callee returns via
5026 the :ref:`resume <i_resume>` instruction or other exception handling
5027 mechanism.
5028#. The optional :ref:`function attributes <fnattrs>` list. Only
5029 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
5030 attributes are valid here.
5031
5032Semantics:
5033""""""""""
5034
5035This instruction is designed to operate as a standard '``call``'
5036instruction in most regards. The primary difference is that it
5037establishes an association with a label, which is used by the runtime
5038library to unwind the stack.
5039
5040This instruction is used in languages with destructors to ensure that
5041proper cleanup is performed in the case of either a ``longjmp`` or a
5042thrown exception. Additionally, this is important for implementation of
5043'``catch``' clauses in high-level languages that support them.
5044
5045For the purposes of the SSA form, the definition of the value returned
5046by the '``invoke``' instruction is deemed to occur on the edge from the
5047current block to the "normal" label. If the callee unwinds then no
5048return value is available.
5049
5050Example:
5051""""""""
5052
5053.. code-block:: llvm
5054
5055 %retval = invoke i32 @Test(i32 15) to label %Continue
Tim Northover675a0962014-06-13 14:24:23 +00005056 unwind label %TestCleanup ; i32:retval set
Sean Silvab084af42012-12-07 10:36:55 +00005057 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
Tim Northover675a0962014-06-13 14:24:23 +00005058 unwind label %TestCleanup ; i32:retval set
Sean Silvab084af42012-12-07 10:36:55 +00005059
5060.. _i_resume:
5061
5062'``resume``' Instruction
5063^^^^^^^^^^^^^^^^^^^^^^^^
5064
5065Syntax:
5066"""""""
5067
5068::
5069
5070 resume <type> <value>
5071
5072Overview:
5073"""""""""
5074
5075The '``resume``' instruction is a terminator instruction that has no
5076successors.
5077
5078Arguments:
5079""""""""""
5080
5081The '``resume``' instruction requires one argument, which must have the
5082same type as the result of any '``landingpad``' instruction in the same
5083function.
5084
5085Semantics:
5086""""""""""
5087
5088The '``resume``' instruction resumes propagation of an existing
5089(in-flight) exception whose unwinding was interrupted with a
5090:ref:`landingpad <i_landingpad>` instruction.
5091
5092Example:
5093""""""""
5094
5095.. code-block:: llvm
5096
5097 resume { i8*, i32 } %exn
5098
David Majnemer654e1302015-07-31 17:58:14 +00005099.. _i_catchpad:
5100
5101'``catchpad``' Instruction
5102^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5103
5104Syntax:
5105"""""""
5106
5107::
5108
5109 <resultval> = catchpad <resultty> [<args>*]
5110 to label <normal label> unwind label <exception label>
5111
5112Overview:
5113"""""""""
5114
5115The '``catchpad``' instruction is used by `LLVM's exception handling
5116system <ExceptionHandling.html#overview>`_ to specify that a basic block
5117is a catch block --- one where a personality routine attempts to transfer
5118control to catch an exception.
5119The ``args`` correspond to whatever information the personality
5120routine requires to know if this is an appropriate place to catch the
5121exception. Control is tranfered to the ``exception`` label if the
5122``catchpad`` is not an appropriate handler for the in-flight exception.
5123The ``normal`` label should contain the code found in the ``catch``
5124portion of a ``try``/``catch`` sequence. It defines values supplied by
5125the :ref:`personality function <personalityfn>` upon re-entry to the
5126function. The ``resultval`` has the type ``resultty``.
5127
5128Arguments:
5129""""""""""
5130
5131The instruction takes a list of arbitrary values which are interpreted
5132by the :ref:`personality function <personalityfn>`.
5133
5134The ``catchpad`` must be provided a ``normal`` label to transfer control
5135to if the ``catchpad`` matches the exception and an ``exception``
5136label to transfer control to if it doesn't.
5137
5138Semantics:
5139""""""""""
5140
5141The '``catchpad``' instruction defines the values which are set by the
5142:ref:`personality function <personalityfn>` upon re-entry to the function, and
5143therefore the "result type" of the ``catchpad`` instruction. As with
5144calling conventions, how the personality function results are
5145represented in LLVM IR is target specific.
5146
5147When the call stack is being unwound due to an exception being thrown,
5148the exception is compared against the ``args``. If it doesn't match,
5149then control is transfered to the ``exception`` basic block.
5150
5151The ``catchpad`` instruction has several restrictions:
5152
5153- A catch block is a basic block which is the unwind destination of
5154 an exceptional instruction.
5155- A catch block must have a '``catchpad``' instruction as its
5156 first non-PHI instruction.
5157- A catch block's ``exception`` edge must refer to a catch block or a
5158 catch-end block.
5159- There can be only one '``catchpad``' instruction within the
5160 catch block.
5161- A basic block that is not a catch block may not include a
5162 '``catchpad``' instruction.
5163- It is undefined behavior for control to transfer from a ``catchpad`` to a
5164 ``cleanupret`` without first executing a ``catchret`` and a subsequent
5165 ``cleanuppad``.
5166- It is undefined behavior for control to transfer from a ``catchpad`` to a
5167 ``ret`` without first executing a ``catchret``.
5168
5169Example:
5170""""""""
5171
5172.. code-block:: llvm
5173
5174 ;; A catch block which can catch an integer.
5175 %res = catchpad { i8*, i32 } [i8** @_ZTIi]
5176 to label %int.handler unwind label %terminate
5177
5178.. _i_catchendpad:
5179
5180'``catchendpad``' Instruction
5181^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5182
5183Syntax:
5184"""""""
5185
5186::
5187
5188 catchendpad unwind label <nextaction>
5189 catchendpad unwind to caller
5190
5191Overview:
5192"""""""""
5193
5194The '``catchendpad``' instruction is used by `LLVM's exception handling
5195system <ExceptionHandling.html#overview>`_ to communicate to the
5196:ref:`personality function <personalityfn>` which invokes are associated
5197with a chain of :ref:`catchpad <i_catchpad>` instructions.
5198
5199The ``nextaction`` label indicates where control should transfer to if
5200none of the ``catchpad`` instructions are suitable for catching the
5201in-flight exception.
5202
5203If a ``nextaction`` label is not present, the instruction unwinds out of
5204its parent function. The
5205:ref:`personality function <personalityfn>` will continue processing
5206exception handling actions in the caller.
5207
5208Arguments:
5209""""""""""
5210
5211The instruction optionally takes a label, ``nextaction``, indicating
5212where control should transfer to if none of the preceding
5213``catchpad`` instructions are suitable for the in-flight exception.
5214
5215Semantics:
5216""""""""""
5217
5218When the call stack is being unwound due to an exception being thrown
5219and none of the constituent ``catchpad`` instructions match, then
5220control is transfered to ``nextaction`` if it is present. If it is not
5221present, control is transfered to the caller.
5222
5223The ``catchendpad`` instruction has several restrictions:
5224
5225- A catch-end block is a basic block which is the unwind destination of
5226 an exceptional instruction.
5227- A catch-end block must have a '``catchendpad``' instruction as its
5228 first non-PHI instruction.
5229- There can be only one '``catchendpad``' instruction within the
5230 catch block.
5231- A basic block that is not a catch-end block may not include a
5232 '``catchendpad``' instruction.
5233- Exactly one catch block may unwind to a ``catchendpad``.
5234- The unwind target of invokes between a ``catchpad`` and a
5235 corresponding ``catchret`` must be its ``catchendpad``.
5236
5237Example:
5238""""""""
5239
5240.. code-block:: llvm
5241
5242 catchendpad unwind label %terminate
5243 catchendpad unwind to caller
5244
5245.. _i_catchret:
5246
5247'``catchret``' Instruction
5248^^^^^^^^^^^^^^^^^^^^^^^^^^
5249
5250Syntax:
5251"""""""
5252
5253::
5254
5255 catchret label <normal>
5256
5257Overview:
5258"""""""""
5259
5260The '``catchret``' instruction is a terminator instruction that has a
5261single successor.
5262
5263
5264Arguments:
5265""""""""""
5266
5267The '``catchret``' instruction requires one argument which specifies
5268where control will transfer to next.
5269
5270Semantics:
5271""""""""""
5272
5273The '``catchret``' instruction ends the existing (in-flight) exception
5274whose unwinding was interrupted with a
5275:ref:`catchpad <i_catchpad>` instruction.
5276The :ref:`personality function <personalityfn>` gets a chance to execute
5277arbitrary code to, for example, run a C++ destructor.
5278Control then transfers to ``normal``.
5279
5280Example:
5281""""""""
5282
5283.. code-block:: llvm
5284
5285 catchret label %continue
5286
5287.. _i_cleanupret:
5288
5289'``cleanupret``' Instruction
5290^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5291
5292Syntax:
5293"""""""
5294
5295::
5296
5297 cleanupret <type> <value> unwind label <continue>
5298 cleanupret <type> <value> unwind to caller
5299
5300Overview:
5301"""""""""
5302
5303The '``cleanupret``' instruction is a terminator instruction that has
5304an optional successor.
5305
5306
5307Arguments:
5308""""""""""
5309
5310The '``cleanupret``' instruction requires one argument, which must have the
5311same type as the result of any '``cleanuppad``' instruction in the same
5312function. It also has an optional successor, ``continue``.
5313
5314Semantics:
5315""""""""""
5316
5317The '``cleanupret``' instruction indicates to the
5318:ref:`personality function <personalityfn>` that one
5319:ref:`cleanuppad <i_cleanuppad>` it transferred control to has ended.
5320It transfers control to ``continue`` or unwinds out of the function.
5321
5322Example:
5323""""""""
5324
5325.. code-block:: llvm
5326
5327 cleanupret void unwind to caller
5328 cleanupret { i8*, i32 } %exn unwind label %continue
5329
5330.. _i_terminatepad:
5331
5332'``terminatepad``' Instruction
5333^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5334
5335Syntax:
5336"""""""
5337
5338::
5339
5340 terminatepad [<args>*] unwind label <exception label>
5341 terminatepad [<args>*] unwind to caller
5342
5343Overview:
5344"""""""""
5345
5346The '``terminatepad``' instruction is used by `LLVM's exception handling
5347system <ExceptionHandling.html#overview>`_ to specify that a basic block
5348is a terminate block --- one where a personality routine may decide to
5349terminate the program.
5350The ``args`` correspond to whatever information the personality
5351routine requires to know if this is an appropriate place to terminate the
5352program. Control is transferred to the ``exception`` label if the
5353personality routine decides not to terminate the program for the
5354in-flight exception.
5355
5356Arguments:
5357""""""""""
5358
5359The instruction takes a list of arbitrary values which are interpreted
5360by the :ref:`personality function <personalityfn>`.
5361
5362The ``terminatepad`` may be given an ``exception`` label to
5363transfer control to if the in-flight exception matches the ``args``.
5364
5365Semantics:
5366""""""""""
5367
5368When the call stack is being unwound due to an exception being thrown,
5369the exception is compared against the ``args``. If it matches,
5370then control is transfered to the ``exception`` basic block. Otherwise,
5371the program is terminated via personality-specific means. Typically,
5372the first argument to ``terminatepad`` specifies what function the
5373personality should defer to in order to terminate the program.
5374
5375The ``terminatepad`` instruction has several restrictions:
5376
5377- A terminate block is a basic block which is the unwind destination of
5378 an exceptional instruction.
5379- A terminate block must have a '``terminatepad``' instruction as its
5380 first non-PHI instruction.
5381- There can be only one '``terminatepad``' instruction within the
5382 terminate block.
5383- A basic block that is not a terminate block may not include a
5384 '``terminatepad``' instruction.
5385
5386Example:
5387""""""""
5388
5389.. code-block:: llvm
5390
5391 ;; A terminate block which only permits integers.
5392 terminatepad [i8** @_ZTIi] unwind label %continue
5393
Sean Silvab084af42012-12-07 10:36:55 +00005394.. _i_unreachable:
5395
5396'``unreachable``' Instruction
5397^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5398
5399Syntax:
5400"""""""
5401
5402::
5403
5404 unreachable
5405
5406Overview:
5407"""""""""
5408
5409The '``unreachable``' instruction has no defined semantics. This
5410instruction is used to inform the optimizer that a particular portion of
5411the code is not reachable. This can be used to indicate that the code
5412after a no-return function cannot be reached, and other facts.
5413
5414Semantics:
5415""""""""""
5416
5417The '``unreachable``' instruction has no defined semantics.
5418
5419.. _binaryops:
5420
5421Binary Operations
5422-----------------
5423
5424Binary operators are used to do most of the computation in a program.
5425They require two operands of the same type, execute an operation on
5426them, and produce a single value. The operands might represent multiple
5427data, as is the case with the :ref:`vector <t_vector>` data type. The
5428result value has the same type as its operands.
5429
5430There are several different binary operators:
5431
5432.. _i_add:
5433
5434'``add``' Instruction
5435^^^^^^^^^^^^^^^^^^^^^
5436
5437Syntax:
5438"""""""
5439
5440::
5441
Tim Northover675a0962014-06-13 14:24:23 +00005442 <result> = add <ty> <op1>, <op2> ; yields ty:result
5443 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
5444 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
5445 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00005446
5447Overview:
5448"""""""""
5449
5450The '``add``' instruction returns the sum of its two operands.
5451
5452Arguments:
5453""""""""""
5454
5455The two arguments to the '``add``' instruction must be
5456:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5457arguments must have identical types.
5458
5459Semantics:
5460""""""""""
5461
5462The value produced is the integer sum of the two operands.
5463
5464If the sum has unsigned overflow, the result returned is the
5465mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5466the result.
5467
5468Because LLVM integers use a two's complement representation, this
5469instruction is appropriate for both signed and unsigned integers.
5470
5471``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5472respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5473result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
5474unsigned and/or signed overflow, respectively, occurs.
5475
5476Example:
5477""""""""
5478
5479.. code-block:: llvm
5480
Tim Northover675a0962014-06-13 14:24:23 +00005481 <result> = add i32 4, %var ; yields i32:result = 4 + %var
Sean Silvab084af42012-12-07 10:36:55 +00005482
5483.. _i_fadd:
5484
5485'``fadd``' Instruction
5486^^^^^^^^^^^^^^^^^^^^^^
5487
5488Syntax:
5489"""""""
5490
5491::
5492
Tim Northover675a0962014-06-13 14:24:23 +00005493 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00005494
5495Overview:
5496"""""""""
5497
5498The '``fadd``' instruction returns the sum of its two operands.
5499
5500Arguments:
5501""""""""""
5502
5503The two arguments to the '``fadd``' instruction must be :ref:`floating
5504point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5505Both arguments must have identical types.
5506
5507Semantics:
5508""""""""""
5509
5510The value produced is the floating point sum of the two operands. This
5511instruction can also take any number of :ref:`fast-math flags <fastmath>`,
5512which are optimization hints to enable otherwise unsafe floating point
5513optimizations:
5514
5515Example:
5516""""""""
5517
5518.. code-block:: llvm
5519
Tim Northover675a0962014-06-13 14:24:23 +00005520 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
Sean Silvab084af42012-12-07 10:36:55 +00005521
5522'``sub``' Instruction
5523^^^^^^^^^^^^^^^^^^^^^
5524
5525Syntax:
5526"""""""
5527
5528::
5529
Tim Northover675a0962014-06-13 14:24:23 +00005530 <result> = sub <ty> <op1>, <op2> ; yields ty:result
5531 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
5532 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
5533 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00005534
5535Overview:
5536"""""""""
5537
5538The '``sub``' instruction returns the difference of its two operands.
5539
5540Note that the '``sub``' instruction is used to represent the '``neg``'
5541instruction present in most other intermediate representations.
5542
5543Arguments:
5544""""""""""
5545
5546The two arguments to the '``sub``' instruction must be
5547:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5548arguments must have identical types.
5549
5550Semantics:
5551""""""""""
5552
5553The value produced is the integer difference of the two operands.
5554
5555If the difference has unsigned overflow, the result returned is the
5556mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
5557the result.
5558
5559Because LLVM integers use a two's complement representation, this
5560instruction is appropriate for both signed and unsigned integers.
5561
5562``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5563respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5564result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
5565unsigned and/or signed overflow, respectively, occurs.
5566
5567Example:
5568""""""""
5569
5570.. code-block:: llvm
5571
Tim Northover675a0962014-06-13 14:24:23 +00005572 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
5573 <result> = sub i32 0, %val ; yields i32:result = -%var
Sean Silvab084af42012-12-07 10:36:55 +00005574
5575.. _i_fsub:
5576
5577'``fsub``' Instruction
5578^^^^^^^^^^^^^^^^^^^^^^
5579
5580Syntax:
5581"""""""
5582
5583::
5584
Tim Northover675a0962014-06-13 14:24:23 +00005585 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00005586
5587Overview:
5588"""""""""
5589
5590The '``fsub``' instruction returns the difference of its two operands.
5591
5592Note that the '``fsub``' instruction is used to represent the '``fneg``'
5593instruction present in most other intermediate representations.
5594
5595Arguments:
5596""""""""""
5597
5598The two arguments to the '``fsub``' instruction must be :ref:`floating
5599point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5600Both arguments must have identical types.
5601
5602Semantics:
5603""""""""""
5604
5605The value produced is the floating point difference of the two operands.
5606This instruction can also take any number of :ref:`fast-math
5607flags <fastmath>`, which are optimization hints to enable otherwise
5608unsafe floating point optimizations:
5609
5610Example:
5611""""""""
5612
5613.. code-block:: llvm
5614
Tim Northover675a0962014-06-13 14:24:23 +00005615 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
5616 <result> = fsub float -0.0, %val ; yields float:result = -%var
Sean Silvab084af42012-12-07 10:36:55 +00005617
5618'``mul``' Instruction
5619^^^^^^^^^^^^^^^^^^^^^
5620
5621Syntax:
5622"""""""
5623
5624::
5625
Tim Northover675a0962014-06-13 14:24:23 +00005626 <result> = mul <ty> <op1>, <op2> ; yields ty:result
5627 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
5628 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
5629 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00005630
5631Overview:
5632"""""""""
5633
5634The '``mul``' instruction returns the product of its two operands.
5635
5636Arguments:
5637""""""""""
5638
5639The two arguments to the '``mul``' instruction must be
5640:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5641arguments must have identical types.
5642
5643Semantics:
5644""""""""""
5645
5646The value produced is the integer product of the two operands.
5647
5648If the result of the multiplication has unsigned overflow, the result
5649returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
5650bit width of the result.
5651
5652Because LLVM integers use a two's complement representation, and the
5653result is the same width as the operands, this instruction returns the
5654correct result for both signed and unsigned integers. If a full product
5655(e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
5656sign-extended or zero-extended as appropriate to the width of the full
5657product.
5658
5659``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
5660respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
5661result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
5662unsigned and/or signed overflow, respectively, occurs.
5663
5664Example:
5665""""""""
5666
5667.. code-block:: llvm
5668
Tim Northover675a0962014-06-13 14:24:23 +00005669 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
Sean Silvab084af42012-12-07 10:36:55 +00005670
5671.. _i_fmul:
5672
5673'``fmul``' Instruction
5674^^^^^^^^^^^^^^^^^^^^^^
5675
5676Syntax:
5677"""""""
5678
5679::
5680
Tim Northover675a0962014-06-13 14:24:23 +00005681 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00005682
5683Overview:
5684"""""""""
5685
5686The '``fmul``' instruction returns the product of its two operands.
5687
5688Arguments:
5689""""""""""
5690
5691The two arguments to the '``fmul``' instruction must be :ref:`floating
5692point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5693Both arguments must have identical types.
5694
5695Semantics:
5696""""""""""
5697
5698The value produced is the floating point product of the two operands.
5699This instruction can also take any number of :ref:`fast-math
5700flags <fastmath>`, which are optimization hints to enable otherwise
5701unsafe floating point optimizations:
5702
5703Example:
5704""""""""
5705
5706.. code-block:: llvm
5707
Tim Northover675a0962014-06-13 14:24:23 +00005708 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
Sean Silvab084af42012-12-07 10:36:55 +00005709
5710'``udiv``' Instruction
5711^^^^^^^^^^^^^^^^^^^^^^
5712
5713Syntax:
5714"""""""
5715
5716::
5717
Tim Northover675a0962014-06-13 14:24:23 +00005718 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
5719 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00005720
5721Overview:
5722"""""""""
5723
5724The '``udiv``' instruction returns the quotient of its two operands.
5725
5726Arguments:
5727""""""""""
5728
5729The two arguments to the '``udiv``' instruction must be
5730:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5731arguments must have identical types.
5732
5733Semantics:
5734""""""""""
5735
5736The value produced is the unsigned integer quotient of the two operands.
5737
5738Note that unsigned integer division and signed integer division are
5739distinct operations; for signed integer division, use '``sdiv``'.
5740
5741Division by zero leads to undefined behavior.
5742
5743If the ``exact`` keyword is present, the result value of the ``udiv`` is
5744a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
5745such, "((a udiv exact b) mul b) == a").
5746
5747Example:
5748""""""""
5749
5750.. code-block:: llvm
5751
Tim Northover675a0962014-06-13 14:24:23 +00005752 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
Sean Silvab084af42012-12-07 10:36:55 +00005753
5754'``sdiv``' Instruction
5755^^^^^^^^^^^^^^^^^^^^^^
5756
5757Syntax:
5758"""""""
5759
5760::
5761
Tim Northover675a0962014-06-13 14:24:23 +00005762 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
5763 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00005764
5765Overview:
5766"""""""""
5767
5768The '``sdiv``' instruction returns the quotient of its two operands.
5769
5770Arguments:
5771""""""""""
5772
5773The two arguments to the '``sdiv``' instruction must be
5774:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5775arguments must have identical types.
5776
5777Semantics:
5778""""""""""
5779
5780The value produced is the signed integer quotient of the two operands
5781rounded towards zero.
5782
5783Note that signed integer division and unsigned integer division are
5784distinct operations; for unsigned integer division, use '``udiv``'.
5785
5786Division by zero leads to undefined behavior. Overflow also leads to
5787undefined behavior; this is a rare case, but can occur, for example, by
5788doing a 32-bit division of -2147483648 by -1.
5789
5790If the ``exact`` keyword is present, the result value of the ``sdiv`` is
5791a :ref:`poison value <poisonvalues>` if the result would be rounded.
5792
5793Example:
5794""""""""
5795
5796.. code-block:: llvm
5797
Tim Northover675a0962014-06-13 14:24:23 +00005798 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
Sean Silvab084af42012-12-07 10:36:55 +00005799
5800.. _i_fdiv:
5801
5802'``fdiv``' Instruction
5803^^^^^^^^^^^^^^^^^^^^^^
5804
5805Syntax:
5806"""""""
5807
5808::
5809
Tim Northover675a0962014-06-13 14:24:23 +00005810 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00005811
5812Overview:
5813"""""""""
5814
5815The '``fdiv``' instruction returns the quotient of its two operands.
5816
5817Arguments:
5818""""""""""
5819
5820The two arguments to the '``fdiv``' instruction must be :ref:`floating
5821point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5822Both arguments must have identical types.
5823
5824Semantics:
5825""""""""""
5826
5827The value produced is the floating point quotient of the two operands.
5828This instruction can also take any number of :ref:`fast-math
5829flags <fastmath>`, which are optimization hints to enable otherwise
5830unsafe floating point optimizations:
5831
5832Example:
5833""""""""
5834
5835.. code-block:: llvm
5836
Tim Northover675a0962014-06-13 14:24:23 +00005837 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
Sean Silvab084af42012-12-07 10:36:55 +00005838
5839'``urem``' Instruction
5840^^^^^^^^^^^^^^^^^^^^^^
5841
5842Syntax:
5843"""""""
5844
5845::
5846
Tim Northover675a0962014-06-13 14:24:23 +00005847 <result> = urem <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00005848
5849Overview:
5850"""""""""
5851
5852The '``urem``' instruction returns the remainder from the unsigned
5853division of its two arguments.
5854
5855Arguments:
5856""""""""""
5857
5858The two arguments to the '``urem``' instruction must be
5859:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5860arguments must have identical types.
5861
5862Semantics:
5863""""""""""
5864
5865This instruction returns the unsigned integer *remainder* of a division.
5866This instruction always performs an unsigned division to get the
5867remainder.
5868
5869Note that unsigned integer remainder and signed integer remainder are
5870distinct operations; for signed integer remainder, use '``srem``'.
5871
5872Taking the remainder of a division by zero leads to undefined behavior.
5873
5874Example:
5875""""""""
5876
5877.. code-block:: llvm
5878
Tim Northover675a0962014-06-13 14:24:23 +00005879 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
Sean Silvab084af42012-12-07 10:36:55 +00005880
5881'``srem``' Instruction
5882^^^^^^^^^^^^^^^^^^^^^^
5883
5884Syntax:
5885"""""""
5886
5887::
5888
Tim Northover675a0962014-06-13 14:24:23 +00005889 <result> = srem <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00005890
5891Overview:
5892"""""""""
5893
5894The '``srem``' instruction returns the remainder from the signed
5895division of its two operands. This instruction can also take
5896:ref:`vector <t_vector>` versions of the values in which case the elements
5897must be integers.
5898
5899Arguments:
5900""""""""""
5901
5902The two arguments to the '``srem``' instruction must be
5903:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5904arguments must have identical types.
5905
5906Semantics:
5907""""""""""
5908
5909This instruction returns the *remainder* of a division (where the result
5910is either zero or has the same sign as the dividend, ``op1``), not the
5911*modulo* operator (where the result is either zero or has the same sign
5912as the divisor, ``op2``) of a value. For more information about the
5913difference, see `The Math
5914Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
5915table of how this is implemented in various languages, please see
5916`Wikipedia: modulo
5917operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
5918
5919Note that signed integer remainder and unsigned integer remainder are
5920distinct operations; for unsigned integer remainder, use '``urem``'.
5921
5922Taking the remainder of a division by zero leads to undefined behavior.
5923Overflow also leads to undefined behavior; this is a rare case, but can
5924occur, for example, by taking the remainder of a 32-bit division of
5925-2147483648 by -1. (The remainder doesn't actually overflow, but this
5926rule lets srem be implemented using instructions that return both the
5927result of the division and the remainder.)
5928
5929Example:
5930""""""""
5931
5932.. code-block:: llvm
5933
Tim Northover675a0962014-06-13 14:24:23 +00005934 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
Sean Silvab084af42012-12-07 10:36:55 +00005935
5936.. _i_frem:
5937
5938'``frem``' Instruction
5939^^^^^^^^^^^^^^^^^^^^^^
5940
5941Syntax:
5942"""""""
5943
5944::
5945
Tim Northover675a0962014-06-13 14:24:23 +00005946 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00005947
5948Overview:
5949"""""""""
5950
5951The '``frem``' instruction returns the remainder from the division of
5952its two operands.
5953
5954Arguments:
5955""""""""""
5956
5957The two arguments to the '``frem``' instruction must be :ref:`floating
5958point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5959Both arguments must have identical types.
5960
5961Semantics:
5962""""""""""
5963
5964This instruction returns the *remainder* of a division. The remainder
5965has the same sign as the dividend. This instruction can also take any
5966number of :ref:`fast-math flags <fastmath>`, which are optimization hints
5967to enable otherwise unsafe floating point optimizations:
5968
5969Example:
5970""""""""
5971
5972.. code-block:: llvm
5973
Tim Northover675a0962014-06-13 14:24:23 +00005974 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
Sean Silvab084af42012-12-07 10:36:55 +00005975
5976.. _bitwiseops:
5977
5978Bitwise Binary Operations
5979-------------------------
5980
5981Bitwise binary operators are used to do various forms of bit-twiddling
5982in a program. They are generally very efficient instructions and can
5983commonly be strength reduced from other instructions. They require two
5984operands of the same type, execute an operation on them, and produce a
5985single value. The resulting value is the same type as its operands.
5986
5987'``shl``' Instruction
5988^^^^^^^^^^^^^^^^^^^^^
5989
5990Syntax:
5991"""""""
5992
5993::
5994
Tim Northover675a0962014-06-13 14:24:23 +00005995 <result> = shl <ty> <op1>, <op2> ; yields ty:result
5996 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
5997 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
5998 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00005999
6000Overview:
6001"""""""""
6002
6003The '``shl``' instruction returns the first operand shifted to the left
6004a specified number of bits.
6005
6006Arguments:
6007""""""""""
6008
6009Both arguments to the '``shl``' instruction must be the same
6010:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6011'``op2``' is treated as an unsigned value.
6012
6013Semantics:
6014""""""""""
6015
6016The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
6017where ``n`` is the width of the result. If ``op2`` is (statically or
Sean Silvab8a108c2015-04-17 21:58:55 +00006018dynamically) equal to or larger than the number of bits in
Sean Silvab084af42012-12-07 10:36:55 +00006019``op1``, the result is undefined. If the arguments are vectors, each
6020vector element of ``op1`` is shifted by the corresponding shift amount
6021in ``op2``.
6022
6023If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
6024value <poisonvalues>` if it shifts out any non-zero bits. If the
6025``nsw`` keyword is present, then the shift produces a :ref:`poison
6026value <poisonvalues>` if it shifts out any bits that disagree with the
6027resultant sign bit. As such, NUW/NSW have the same semantics as they
6028would if the shift were expressed as a mul instruction with the same
6029nsw/nuw bits in (mul %op1, (shl 1, %op2)).
6030
6031Example:
6032""""""""
6033
6034.. code-block:: llvm
6035
Tim Northover675a0962014-06-13 14:24:23 +00006036 <result> = shl i32 4, %var ; yields i32: 4 << %var
6037 <result> = shl i32 4, 2 ; yields i32: 16
6038 <result> = shl i32 1, 10 ; yields i32: 1024
Sean Silvab084af42012-12-07 10:36:55 +00006039 <result> = shl i32 1, 32 ; undefined
6040 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
6041
6042'``lshr``' Instruction
6043^^^^^^^^^^^^^^^^^^^^^^
6044
6045Syntax:
6046"""""""
6047
6048::
6049
Tim Northover675a0962014-06-13 14:24:23 +00006050 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
6051 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00006052
6053Overview:
6054"""""""""
6055
6056The '``lshr``' instruction (logical shift right) returns the first
6057operand shifted to the right a specified number of bits with zero fill.
6058
6059Arguments:
6060""""""""""
6061
6062Both arguments to the '``lshr``' instruction must be the same
6063:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6064'``op2``' is treated as an unsigned value.
6065
6066Semantics:
6067""""""""""
6068
6069This instruction always performs a logical shift right operation. The
6070most significant bits of the result will be filled with zero bits after
6071the shift. If ``op2`` is (statically or dynamically) equal to or larger
6072than the number of bits in ``op1``, the result is undefined. If the
6073arguments are vectors, each vector element of ``op1`` is shifted by the
6074corresponding shift amount in ``op2``.
6075
6076If the ``exact`` keyword is present, the result value of the ``lshr`` is
6077a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6078non-zero.
6079
6080Example:
6081""""""""
6082
6083.. code-block:: llvm
6084
Tim Northover675a0962014-06-13 14:24:23 +00006085 <result> = lshr i32 4, 1 ; yields i32:result = 2
6086 <result> = lshr i32 4, 2 ; yields i32:result = 1
6087 <result> = lshr i8 4, 3 ; yields i8:result = 0
6088 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
Sean Silvab084af42012-12-07 10:36:55 +00006089 <result> = lshr i32 1, 32 ; undefined
6090 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
6091
6092'``ashr``' Instruction
6093^^^^^^^^^^^^^^^^^^^^^^
6094
6095Syntax:
6096"""""""
6097
6098::
6099
Tim Northover675a0962014-06-13 14:24:23 +00006100 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
6101 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00006102
6103Overview:
6104"""""""""
6105
6106The '``ashr``' instruction (arithmetic shift right) returns the first
6107operand shifted to the right a specified number of bits with sign
6108extension.
6109
6110Arguments:
6111""""""""""
6112
6113Both arguments to the '``ashr``' instruction must be the same
6114:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
6115'``op2``' is treated as an unsigned value.
6116
6117Semantics:
6118""""""""""
6119
6120This instruction always performs an arithmetic shift right operation,
6121The most significant bits of the result will be filled with the sign bit
6122of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
6123than the number of bits in ``op1``, the result is undefined. If the
6124arguments are vectors, each vector element of ``op1`` is shifted by the
6125corresponding shift amount in ``op2``.
6126
6127If the ``exact`` keyword is present, the result value of the ``ashr`` is
6128a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
6129non-zero.
6130
6131Example:
6132""""""""
6133
6134.. code-block:: llvm
6135
Tim Northover675a0962014-06-13 14:24:23 +00006136 <result> = ashr i32 4, 1 ; yields i32:result = 2
6137 <result> = ashr i32 4, 2 ; yields i32:result = 1
6138 <result> = ashr i8 4, 3 ; yields i8:result = 0
6139 <result> = ashr i8 -2, 1 ; yields i8:result = -1
Sean Silvab084af42012-12-07 10:36:55 +00006140 <result> = ashr i32 1, 32 ; undefined
6141 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
6142
6143'``and``' Instruction
6144^^^^^^^^^^^^^^^^^^^^^
6145
6146Syntax:
6147"""""""
6148
6149::
6150
Tim Northover675a0962014-06-13 14:24:23 +00006151 <result> = and <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00006152
6153Overview:
6154"""""""""
6155
6156The '``and``' instruction returns the bitwise logical and of its two
6157operands.
6158
6159Arguments:
6160""""""""""
6161
6162The two arguments to the '``and``' instruction must be
6163:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6164arguments must have identical types.
6165
6166Semantics:
6167""""""""""
6168
6169The truth table used for the '``and``' instruction is:
6170
6171+-----+-----+-----+
6172| In0 | In1 | Out |
6173+-----+-----+-----+
6174| 0 | 0 | 0 |
6175+-----+-----+-----+
6176| 0 | 1 | 0 |
6177+-----+-----+-----+
6178| 1 | 0 | 0 |
6179+-----+-----+-----+
6180| 1 | 1 | 1 |
6181+-----+-----+-----+
6182
6183Example:
6184""""""""
6185
6186.. code-block:: llvm
6187
Tim Northover675a0962014-06-13 14:24:23 +00006188 <result> = and i32 4, %var ; yields i32:result = 4 & %var
6189 <result> = and i32 15, 40 ; yields i32:result = 8
6190 <result> = and i32 4, 8 ; yields i32:result = 0
Sean Silvab084af42012-12-07 10:36:55 +00006191
6192'``or``' Instruction
6193^^^^^^^^^^^^^^^^^^^^
6194
6195Syntax:
6196"""""""
6197
6198::
6199
Tim Northover675a0962014-06-13 14:24:23 +00006200 <result> = or <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00006201
6202Overview:
6203"""""""""
6204
6205The '``or``' instruction returns the bitwise logical inclusive or of its
6206two operands.
6207
6208Arguments:
6209""""""""""
6210
6211The two arguments to the '``or``' instruction must be
6212:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6213arguments must have identical types.
6214
6215Semantics:
6216""""""""""
6217
6218The truth table used for the '``or``' instruction is:
6219
6220+-----+-----+-----+
6221| In0 | In1 | Out |
6222+-----+-----+-----+
6223| 0 | 0 | 0 |
6224+-----+-----+-----+
6225| 0 | 1 | 1 |
6226+-----+-----+-----+
6227| 1 | 0 | 1 |
6228+-----+-----+-----+
6229| 1 | 1 | 1 |
6230+-----+-----+-----+
6231
6232Example:
6233""""""""
6234
6235::
6236
Tim Northover675a0962014-06-13 14:24:23 +00006237 <result> = or i32 4, %var ; yields i32:result = 4 | %var
6238 <result> = or i32 15, 40 ; yields i32:result = 47
6239 <result> = or i32 4, 8 ; yields i32:result = 12
Sean Silvab084af42012-12-07 10:36:55 +00006240
6241'``xor``' Instruction
6242^^^^^^^^^^^^^^^^^^^^^
6243
6244Syntax:
6245"""""""
6246
6247::
6248
Tim Northover675a0962014-06-13 14:24:23 +00006249 <result> = xor <ty> <op1>, <op2> ; yields ty:result
Sean Silvab084af42012-12-07 10:36:55 +00006250
6251Overview:
6252"""""""""
6253
6254The '``xor``' instruction returns the bitwise logical exclusive or of
6255its two operands. The ``xor`` is used to implement the "one's
6256complement" operation, which is the "~" operator in C.
6257
6258Arguments:
6259""""""""""
6260
6261The two arguments to the '``xor``' instruction must be
6262:ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
6263arguments must have identical types.
6264
6265Semantics:
6266""""""""""
6267
6268The truth table used for the '``xor``' instruction is:
6269
6270+-----+-----+-----+
6271| In0 | In1 | Out |
6272+-----+-----+-----+
6273| 0 | 0 | 0 |
6274+-----+-----+-----+
6275| 0 | 1 | 1 |
6276+-----+-----+-----+
6277| 1 | 0 | 1 |
6278+-----+-----+-----+
6279| 1 | 1 | 0 |
6280+-----+-----+-----+
6281
6282Example:
6283""""""""
6284
6285.. code-block:: llvm
6286
Tim Northover675a0962014-06-13 14:24:23 +00006287 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
6288 <result> = xor i32 15, 40 ; yields i32:result = 39
6289 <result> = xor i32 4, 8 ; yields i32:result = 12
6290 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
Sean Silvab084af42012-12-07 10:36:55 +00006291
6292Vector Operations
6293-----------------
6294
6295LLVM supports several instructions to represent vector operations in a
6296target-independent manner. These instructions cover the element-access
6297and vector-specific operations needed to process vectors effectively.
6298While LLVM does directly support these vector operations, many
6299sophisticated algorithms will want to use target-specific intrinsics to
6300take full advantage of a specific target.
6301
6302.. _i_extractelement:
6303
6304'``extractelement``' Instruction
6305^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6306
6307Syntax:
6308"""""""
6309
6310::
6311
Michael J. Spencer1f10c5ea2014-05-01 22:12:39 +00006312 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
Sean Silvab084af42012-12-07 10:36:55 +00006313
6314Overview:
6315"""""""""
6316
6317The '``extractelement``' instruction extracts a single scalar element
6318from a vector at a specified index.
6319
6320Arguments:
6321""""""""""
6322
6323The first operand of an '``extractelement``' instruction is a value of
6324:ref:`vector <t_vector>` type. The second operand is an index indicating
6325the position from which to extract the element. The index may be a
Michael J. Spencer1f10c5ea2014-05-01 22:12:39 +00006326variable of any integer type.
Sean Silvab084af42012-12-07 10:36:55 +00006327
6328Semantics:
6329""""""""""
6330
6331The result is a scalar of the same type as the element type of ``val``.
6332Its value is the value at position ``idx`` of ``val``. If ``idx``
6333exceeds the length of ``val``, the results are undefined.
6334
6335Example:
6336""""""""
6337
6338.. code-block:: llvm
6339
6340 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
6341
6342.. _i_insertelement:
6343
6344'``insertelement``' Instruction
6345^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6346
6347Syntax:
6348"""""""
6349
6350::
6351
Michael J. Spencer1f10c5ea2014-05-01 22:12:39 +00006352 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
Sean Silvab084af42012-12-07 10:36:55 +00006353
6354Overview:
6355"""""""""
6356
6357The '``insertelement``' instruction inserts a scalar element into a
6358vector at a specified index.
6359
6360Arguments:
6361""""""""""
6362
6363The first operand of an '``insertelement``' instruction is a value of
6364:ref:`vector <t_vector>` type. The second operand is a scalar value whose
6365type must equal the element type of the first operand. The third operand
6366is an index indicating the position at which to insert the value. The
Michael J. Spencer1f10c5ea2014-05-01 22:12:39 +00006367index may be a variable of any integer type.
Sean Silvab084af42012-12-07 10:36:55 +00006368
6369Semantics:
6370""""""""""
6371
6372The result is a vector of the same type as ``val``. Its element values
6373are those of ``val`` except at position ``idx``, where it gets the value
6374``elt``. If ``idx`` exceeds the length of ``val``, the results are
6375undefined.
6376
6377Example:
6378""""""""
6379
6380.. code-block:: llvm
6381
6382 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
6383
6384.. _i_shufflevector:
6385
6386'``shufflevector``' Instruction
6387^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6388
6389Syntax:
6390"""""""
6391
6392::
6393
6394 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
6395
6396Overview:
6397"""""""""
6398
6399The '``shufflevector``' instruction constructs a permutation of elements
6400from two input vectors, returning a vector with the same element type as
6401the input and length that is the same as the shuffle mask.
6402
6403Arguments:
6404""""""""""
6405
6406The first two operands of a '``shufflevector``' instruction are vectors
6407with the same type. The third argument is a shuffle mask whose element
6408type is always 'i32'. The result of the instruction is a vector whose
6409length is the same as the shuffle mask and whose element type is the
6410same as the element type of the first two operands.
6411
6412The shuffle mask operand is required to be a constant vector with either
6413constant integer or undef values.
6414
6415Semantics:
6416""""""""""
6417
6418The elements of the two input vectors are numbered from left to right
6419across both of the vectors. The shuffle mask operand specifies, for each
6420element of the result vector, which element of the two input vectors the
6421result element gets. The element selector may be undef (meaning "don't
6422care") and the second operand may be undef if performing a shuffle from
6423only one vector.
6424
6425Example:
6426""""""""
6427
6428.. code-block:: llvm
6429
6430 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6431 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
6432 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
6433 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
6434 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
6435 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
6436 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
6437 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
6438
6439Aggregate Operations
6440--------------------
6441
6442LLVM supports several instructions for working with
6443:ref:`aggregate <t_aggregate>` values.
6444
6445.. _i_extractvalue:
6446
6447'``extractvalue``' Instruction
6448^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6449
6450Syntax:
6451"""""""
6452
6453::
6454
6455 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
6456
6457Overview:
6458"""""""""
6459
6460The '``extractvalue``' instruction extracts the value of a member field
6461from an :ref:`aggregate <t_aggregate>` value.
6462
6463Arguments:
6464""""""""""
6465
6466The first operand of an '``extractvalue``' instruction is a value of
6467:ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
6468constant indices to specify which value to extract in a similar manner
6469as indices in a '``getelementptr``' instruction.
6470
6471The major differences to ``getelementptr`` indexing are:
6472
6473- Since the value being indexed is not a pointer, the first index is
6474 omitted and assumed to be zero.
6475- At least one index must be specified.
6476- Not only struct indices but also array indices must be in bounds.
6477
6478Semantics:
6479""""""""""
6480
6481The result is the value at the position in the aggregate specified by
6482the index operands.
6483
6484Example:
6485""""""""
6486
6487.. code-block:: llvm
6488
6489 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
6490
6491.. _i_insertvalue:
6492
6493'``insertvalue``' Instruction
6494^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6495
6496Syntax:
6497"""""""
6498
6499::
6500
6501 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
6502
6503Overview:
6504"""""""""
6505
6506The '``insertvalue``' instruction inserts a value into a member field in
6507an :ref:`aggregate <t_aggregate>` value.
6508
6509Arguments:
6510""""""""""
6511
6512The first operand of an '``insertvalue``' instruction is a value of
6513:ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
6514a first-class value to insert. The following operands are constant
6515indices indicating the position at which to insert the value in a
6516similar manner as indices in a '``extractvalue``' instruction. The value
6517to insert must have the same type as the value identified by the
6518indices.
6519
6520Semantics:
6521""""""""""
6522
6523The result is an aggregate of the same type as ``val``. Its value is
6524that of ``val`` except that the value at the position specified by the
6525indices is that of ``elt``.
6526
6527Example:
6528""""""""
6529
6530.. code-block:: llvm
6531
6532 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
6533 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
Dan Liewffcfe7f2014-09-08 21:19:46 +00006534 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
Sean Silvab084af42012-12-07 10:36:55 +00006535
6536.. _memoryops:
6537
6538Memory Access and Addressing Operations
6539---------------------------------------
6540
6541A key design point of an SSA-based representation is how it represents
6542memory. In LLVM, no memory locations are in SSA form, which makes things
6543very simple. This section describes how to read, write, and allocate
6544memory in LLVM.
6545
6546.. _i_alloca:
6547
6548'``alloca``' Instruction
6549^^^^^^^^^^^^^^^^^^^^^^^^
6550
6551Syntax:
6552"""""""
6553
6554::
6555
Tim Northover675a0962014-06-13 14:24:23 +00006556 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
Sean Silvab084af42012-12-07 10:36:55 +00006557
6558Overview:
6559"""""""""
6560
6561The '``alloca``' instruction allocates memory on the stack frame of the
6562currently executing function, to be automatically released when this
6563function returns to its caller. The object is always allocated in the
6564generic address space (address space zero).
6565
6566Arguments:
6567""""""""""
6568
6569The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
6570bytes of memory on the runtime stack, returning a pointer of the
6571appropriate type to the program. If "NumElements" is specified, it is
6572the number of elements allocated, otherwise "NumElements" is defaulted
6573to be one. If a constant alignment is specified, the value result of the
Reid Kleckner15fe7a52014-07-15 01:16:09 +00006574allocation is guaranteed to be aligned to at least that boundary. The
6575alignment may not be greater than ``1 << 29``. If not specified, or if
6576zero, the target can choose to align the allocation on any convenient
6577boundary compatible with the type.
Sean Silvab084af42012-12-07 10:36:55 +00006578
6579'``type``' may be any sized type.
6580
6581Semantics:
6582""""""""""
6583
6584Memory is allocated; a pointer is returned. The operation is undefined
6585if there is insufficient stack space for the allocation. '``alloca``'d
6586memory is automatically released when the function returns. The
6587'``alloca``' instruction is commonly used to represent automatic
6588variables that must have an address available. When the function returns
6589(either with the ``ret`` or ``resume`` instructions), the memory is
6590reclaimed. Allocating zero bytes is legal, but the result is undefined.
6591The order in which memory is allocated (ie., which way the stack grows)
6592is not specified.
6593
6594Example:
6595""""""""
6596
6597.. code-block:: llvm
6598
Tim Northover675a0962014-06-13 14:24:23 +00006599 %ptr = alloca i32 ; yields i32*:ptr
6600 %ptr = alloca i32, i32 4 ; yields i32*:ptr
6601 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
6602 %ptr = alloca i32, align 1024 ; yields i32*:ptr
Sean Silvab084af42012-12-07 10:36:55 +00006603
6604.. _i_load:
6605
6606'``load``' Instruction
6607^^^^^^^^^^^^^^^^^^^^^^
6608
6609Syntax:
6610"""""""
6611
6612::
6613
Sanjoy Dasf9995472015-05-19 20:10:19 +00006614 <result> = load [volatile] <ty>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>][, !nonnull !<index>][, !dereferenceable !<index>][, !dereferenceable_or_null !<index>]
Sean Silvab084af42012-12-07 10:36:55 +00006615 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
6616 !<index> = !{ i32 1 }
6617
6618Overview:
6619"""""""""
6620
6621The '``load``' instruction is used to read from memory.
6622
6623Arguments:
6624""""""""""
6625
Eli Bendersky239a78b2013-04-17 20:17:08 +00006626The argument to the ``load`` instruction specifies the memory address
David Blaikiec7aabbb2015-03-04 22:06:14 +00006627from which to load. The type specified must be a :ref:`first
Sean Silvab084af42012-12-07 10:36:55 +00006628class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
6629then the optimizer is not allowed to modify the number or order of
6630execution of this ``load`` with other :ref:`volatile
6631operations <volatile>`.
6632
6633If the ``load`` is marked as ``atomic``, it takes an extra
6634:ref:`ordering <ordering>` and optional ``singlethread`` argument. The
6635``release`` and ``acq_rel`` orderings are not valid on ``load``
6636instructions. Atomic loads produce :ref:`defined <memmodel>` results
6637when they may see multiple atomic stores. The type of the pointee must
6638be an integer type whose bit width is a power of two greater than or
6639equal to eight and less than or equal to a target-specific size limit.
6640``align`` must be explicitly specified on atomic loads, and the load has
6641undefined behavior if the alignment is not set to a value which is at
6642least the size in bytes of the pointee. ``!nontemporal`` does not have
6643any defined semantics for atomic loads.
6644
6645The optional constant ``align`` argument specifies the alignment of the
6646operation (that is, the alignment of the memory address). A value of 0
Eli Bendersky239a78b2013-04-17 20:17:08 +00006647or an omitted ``align`` argument means that the operation has the ABI
Sean Silvab084af42012-12-07 10:36:55 +00006648alignment for the target. It is the responsibility of the code emitter
6649to ensure that the alignment information is correct. Overestimating the
6650alignment results in undefined behavior. Underestimating the alignment
Reid Kleckner15fe7a52014-07-15 01:16:09 +00006651may produce less efficient code. An alignment of 1 is always safe. The
6652maximum possible alignment is ``1 << 29``.
Sean Silvab084af42012-12-07 10:36:55 +00006653
6654The optional ``!nontemporal`` metadata must reference a single
Stefanus Du Toit736e2e22013-06-20 14:02:44 +00006655metadata name ``<index>`` corresponding to a metadata node with one
Sean Silvab084af42012-12-07 10:36:55 +00006656``i32`` entry of value 1. The existence of the ``!nontemporal``
Stefanus Du Toit736e2e22013-06-20 14:02:44 +00006657metadata on the instruction tells the optimizer and code generator
Sean Silvab084af42012-12-07 10:36:55 +00006658that this load is not expected to be reused in the cache. The code
6659generator may select special instructions to save cache bandwidth, such
6660as the ``MOVNT`` instruction on x86.
6661
6662The optional ``!invariant.load`` metadata must reference a single
Stefanus Du Toit736e2e22013-06-20 14:02:44 +00006663metadata name ``<index>`` corresponding to a metadata node with no
6664entries. The existence of the ``!invariant.load`` metadata on the
Philip Reamese1526fc2014-11-24 22:32:43 +00006665instruction tells the optimizer and code generator that the address
6666operand to this load points to memory which can be assumed unchanged.
Mehdi Amini4a121fa2015-03-14 22:04:06 +00006667Being invariant does not imply that a location is dereferenceable,
6668but it does imply that once the location is known dereferenceable
6669its value is henceforth unchanging.
Sean Silvab084af42012-12-07 10:36:55 +00006670
Philip Reamescdb72f32014-10-20 22:40:55 +00006671The optional ``!nonnull`` metadata must reference a single
6672metadata name ``<index>`` corresponding to a metadata node with no
6673entries. The existence of the ``!nonnull`` metadata on the
6674instruction tells the optimizer that the value loaded is known to
6675never be null. This is analogous to the ''nonnull'' attribute
Mehdi Amini4a121fa2015-03-14 22:04:06 +00006676on parameters and return values. This metadata can only be applied
6677to loads of a pointer type.
Philip Reamescdb72f32014-10-20 22:40:55 +00006678
Sanjoy Dasf9995472015-05-19 20:10:19 +00006679The optional ``!dereferenceable`` metadata must reference a single
6680metadata name ``<index>`` corresponding to a metadata node with one ``i64``
Sean Silva706fba52015-08-06 22:56:24 +00006681entry. The existence of the ``!dereferenceable`` metadata on the instruction
Sanjoy Dasf9995472015-05-19 20:10:19 +00006682tells the optimizer that the value loaded is known to be dereferenceable.
Sean Silva706fba52015-08-06 22:56:24 +00006683The number of bytes known to be dereferenceable is specified by the integer
6684value in the metadata node. This is analogous to the ''dereferenceable''
6685attribute on parameters and return values. This metadata can only be applied
Sanjoy Dasf9995472015-05-19 20:10:19 +00006686to loads of a pointer type.
6687
6688The optional ``!dereferenceable_or_null`` metadata must reference a single
6689metadata name ``<index>`` corresponding to a metadata node with one ``i64``
Sean Silva706fba52015-08-06 22:56:24 +00006690entry. The existence of the ``!dereferenceable_or_null`` metadata on the
Sanjoy Dasf9995472015-05-19 20:10:19 +00006691instruction tells the optimizer that the value loaded is known to be either
6692dereferenceable or null.
Sean Silva706fba52015-08-06 22:56:24 +00006693The number of bytes known to be dereferenceable is specified by the integer
6694value in the metadata node. This is analogous to the ''dereferenceable_or_null''
6695attribute on parameters and return values. This metadata can only be applied
Sanjoy Dasf9995472015-05-19 20:10:19 +00006696to loads of a pointer type.
6697
Sean Silvab084af42012-12-07 10:36:55 +00006698Semantics:
6699""""""""""
6700
6701The location of memory pointed to is loaded. If the value being loaded
6702is of scalar type then the number of bytes read does not exceed the
6703minimum number of bytes needed to hold all bits of the type. For
6704example, loading an ``i24`` reads at most three bytes. When loading a
6705value of a type like ``i20`` with a size that is not an integral number
6706of bytes, the result is undefined if the value was not originally
6707written using a store of the same type.
6708
6709Examples:
6710"""""""""
6711
6712.. code-block:: llvm
6713
Tim Northover675a0962014-06-13 14:24:23 +00006714 %ptr = alloca i32 ; yields i32*:ptr
6715 store i32 3, i32* %ptr ; yields void
David Blaikiec7aabbb2015-03-04 22:06:14 +00006716 %val = load i32, i32* %ptr ; yields i32:val = i32 3
Sean Silvab084af42012-12-07 10:36:55 +00006717
6718.. _i_store:
6719
6720'``store``' Instruction
6721^^^^^^^^^^^^^^^^^^^^^^^
6722
6723Syntax:
6724"""""""
6725
6726::
6727
Tim Northover675a0962014-06-13 14:24:23 +00006728 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields void
6729 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields void
Sean Silvab084af42012-12-07 10:36:55 +00006730
6731Overview:
6732"""""""""
6733
6734The '``store``' instruction is used to write to memory.
6735
6736Arguments:
6737""""""""""
6738
Eli Benderskyca380842013-04-17 17:17:20 +00006739There are two arguments to the ``store`` instruction: a value to store
6740and an address at which to store it. The type of the ``<pointer>``
Sean Silvab084af42012-12-07 10:36:55 +00006741operand must be a pointer to the :ref:`first class <t_firstclass>` type of
Eli Benderskyca380842013-04-17 17:17:20 +00006742the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
Sean Silvab084af42012-12-07 10:36:55 +00006743then the optimizer is not allowed to modify the number or order of
6744execution of this ``store`` with other :ref:`volatile
6745operations <volatile>`.
6746
6747If the ``store`` is marked as ``atomic``, it takes an extra
6748:ref:`ordering <ordering>` and optional ``singlethread`` argument. The
6749``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
6750instructions. Atomic loads produce :ref:`defined <memmodel>` results
6751when they may see multiple atomic stores. The type of the pointee must
6752be an integer type whose bit width is a power of two greater than or
6753equal to eight and less than or equal to a target-specific size limit.
6754``align`` must be explicitly specified on atomic stores, and the store
6755has undefined behavior if the alignment is not set to a value which is
6756at least the size in bytes of the pointee. ``!nontemporal`` does not
6757have any defined semantics for atomic stores.
6758
Eli Benderskyca380842013-04-17 17:17:20 +00006759The optional constant ``align`` argument specifies the alignment of the
Sean Silvab084af42012-12-07 10:36:55 +00006760operation (that is, the alignment of the memory address). A value of 0
Eli Benderskyca380842013-04-17 17:17:20 +00006761or an omitted ``align`` argument means that the operation has the ABI
Sean Silvab084af42012-12-07 10:36:55 +00006762alignment for the target. It is the responsibility of the code emitter
6763to ensure that the alignment information is correct. Overestimating the
Eli Benderskyca380842013-04-17 17:17:20 +00006764alignment results in undefined behavior. Underestimating the
Sean Silvab084af42012-12-07 10:36:55 +00006765alignment may produce less efficient code. An alignment of 1 is always
Reid Kleckner15fe7a52014-07-15 01:16:09 +00006766safe. The maximum possible alignment is ``1 << 29``.
Sean Silvab084af42012-12-07 10:36:55 +00006767
Stefanus Du Toit736e2e22013-06-20 14:02:44 +00006768The optional ``!nontemporal`` metadata must reference a single metadata
Eli Benderskyca380842013-04-17 17:17:20 +00006769name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
Stefanus Du Toit736e2e22013-06-20 14:02:44 +00006770value 1. The existence of the ``!nontemporal`` metadata on the instruction
Sean Silvab084af42012-12-07 10:36:55 +00006771tells the optimizer and code generator that this load is not expected to
6772be reused in the cache. The code generator may select special
6773instructions to save cache bandwidth, such as the MOVNT instruction on
6774x86.
6775
6776Semantics:
6777""""""""""
6778
Eli Benderskyca380842013-04-17 17:17:20 +00006779The contents of memory are updated to contain ``<value>`` at the
6780location specified by the ``<pointer>`` operand. If ``<value>`` is
Sean Silvab084af42012-12-07 10:36:55 +00006781of scalar type then the number of bytes written does not exceed the
6782minimum number of bytes needed to hold all bits of the type. For
6783example, storing an ``i24`` writes at most three bytes. When writing a
6784value of a type like ``i20`` with a size that is not an integral number
6785of bytes, it is unspecified what happens to the extra bits that do not
6786belong to the type, but they will typically be overwritten.
6787
6788Example:
6789""""""""
6790
6791.. code-block:: llvm
6792
Tim Northover675a0962014-06-13 14:24:23 +00006793 %ptr = alloca i32 ; yields i32*:ptr
6794 store i32 3, i32* %ptr ; yields void
6795 %val = load i32* %ptr ; yields i32:val = i32 3
Sean Silvab084af42012-12-07 10:36:55 +00006796
6797.. _i_fence:
6798
6799'``fence``' Instruction
6800^^^^^^^^^^^^^^^^^^^^^^^
6801
6802Syntax:
6803"""""""
6804
6805::
6806
Tim Northover675a0962014-06-13 14:24:23 +00006807 fence [singlethread] <ordering> ; yields void
Sean Silvab084af42012-12-07 10:36:55 +00006808
6809Overview:
6810"""""""""
6811
6812The '``fence``' instruction is used to introduce happens-before edges
6813between operations.
6814
6815Arguments:
6816""""""""""
6817
6818'``fence``' instructions take an :ref:`ordering <ordering>` argument which
6819defines what *synchronizes-with* edges they add. They can only be given
6820``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
6821
6822Semantics:
6823""""""""""
6824
6825A fence A which has (at least) ``release`` ordering semantics
6826*synchronizes with* a fence B with (at least) ``acquire`` ordering
6827semantics if and only if there exist atomic operations X and Y, both
6828operating on some atomic object M, such that A is sequenced before X, X
6829modifies M (either directly or through some side effect of a sequence
6830headed by X), Y is sequenced before B, and Y observes M. This provides a
6831*happens-before* dependency between A and B. Rather than an explicit
6832``fence``, one (but not both) of the atomic operations X or Y might
6833provide a ``release`` or ``acquire`` (resp.) ordering constraint and
6834still *synchronize-with* the explicit ``fence`` and establish the
6835*happens-before* edge.
6836
6837A ``fence`` which has ``seq_cst`` ordering, in addition to having both
6838``acquire`` and ``release`` semantics specified above, participates in
6839the global program order of other ``seq_cst`` operations and/or fences.
6840
6841The optional ":ref:`singlethread <singlethread>`" argument specifies
6842that the fence only synchronizes with other fences in the same thread.
6843(This is useful for interacting with signal handlers.)
6844
6845Example:
6846""""""""
6847
6848.. code-block:: llvm
6849
Tim Northover675a0962014-06-13 14:24:23 +00006850 fence acquire ; yields void
6851 fence singlethread seq_cst ; yields void
Sean Silvab084af42012-12-07 10:36:55 +00006852
6853.. _i_cmpxchg:
6854
6855'``cmpxchg``' Instruction
6856^^^^^^^^^^^^^^^^^^^^^^^^^
6857
6858Syntax:
6859"""""""
6860
6861::
6862
Tim Northover675a0962014-06-13 14:24:23 +00006863 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
Sean Silvab084af42012-12-07 10:36:55 +00006864
6865Overview:
6866"""""""""
6867
6868The '``cmpxchg``' instruction is used to atomically modify memory. It
6869loads a value in memory and compares it to a given value. If they are
Tim Northover420a2162014-06-13 14:24:07 +00006870equal, it tries to store a new value into the memory.
Sean Silvab084af42012-12-07 10:36:55 +00006871
6872Arguments:
6873""""""""""
6874
6875There are three arguments to the '``cmpxchg``' instruction: an address
6876to operate on, a value to compare to the value currently be at that
6877address, and a new value to place at that address if the compared values
6878are equal. The type of '<cmp>' must be an integer type whose bit width
6879is a power of two greater than or equal to eight and less than or equal
6880to a target-specific size limit. '<cmp>' and '<new>' must have the same
6881type, and the type of '<pointer>' must be a pointer to that type. If the
6882``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
6883to modify the number or order of execution of this ``cmpxchg`` with
6884other :ref:`volatile operations <volatile>`.
6885
Tim Northovere94a5182014-03-11 10:48:52 +00006886The success and failure :ref:`ordering <ordering>` arguments specify how this
Tim Northover1dcc9f92014-06-13 14:24:16 +00006887``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
6888must be at least ``monotonic``, the ordering constraint on failure must be no
6889stronger than that on success, and the failure ordering cannot be either
6890``release`` or ``acq_rel``.
Sean Silvab084af42012-12-07 10:36:55 +00006891
6892The optional "``singlethread``" argument declares that the ``cmpxchg``
6893is only atomic with respect to code (usually signal handlers) running in
6894the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
6895respect to all other code in the system.
6896
6897The pointer passed into cmpxchg must have alignment greater than or
6898equal to the size in memory of the operand.
6899
6900Semantics:
6901""""""""""
6902
Tim Northover420a2162014-06-13 14:24:07 +00006903The contents of memory at the location specified by the '``<pointer>``' operand
6904is read and compared to '``<cmp>``'; if the read value is the equal, the
6905'``<new>``' is written. The original value at the location is returned, together
6906with a flag indicating success (true) or failure (false).
6907
6908If the cmpxchg operation is marked as ``weak`` then a spurious failure is
6909permitted: the operation may not write ``<new>`` even if the comparison
6910matched.
6911
6912If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
6913if the value loaded equals ``cmp``.
Sean Silvab084af42012-12-07 10:36:55 +00006914
Tim Northovere94a5182014-03-11 10:48:52 +00006915A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
6916identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
6917load with an ordering parameter determined the second ordering parameter.
Sean Silvab084af42012-12-07 10:36:55 +00006918
6919Example:
6920""""""""
6921
6922.. code-block:: llvm
6923
6924 entry:
David Blaikiec7aabbb2015-03-04 22:06:14 +00006925 %orig = atomic load i32, i32* %ptr unordered ; yields i32
Sean Silvab084af42012-12-07 10:36:55 +00006926 br label %loop
6927
6928 loop:
6929 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
6930 %squared = mul i32 %cmp, %cmp
Tim Northover675a0962014-06-13 14:24:23 +00006931 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
Tim Northover420a2162014-06-13 14:24:07 +00006932 %value_loaded = extractvalue { i32, i1 } %val_success, 0
6933 %success = extractvalue { i32, i1 } %val_success, 1
Sean Silvab084af42012-12-07 10:36:55 +00006934 br i1 %success, label %done, label %loop
6935
6936 done:
6937 ...
6938
6939.. _i_atomicrmw:
6940
6941'``atomicrmw``' Instruction
6942^^^^^^^^^^^^^^^^^^^^^^^^^^^
6943
6944Syntax:
6945"""""""
6946
6947::
6948
Tim Northover675a0962014-06-13 14:24:23 +00006949 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
Sean Silvab084af42012-12-07 10:36:55 +00006950
6951Overview:
6952"""""""""
6953
6954The '``atomicrmw``' instruction is used to atomically modify memory.
6955
6956Arguments:
6957""""""""""
6958
6959There are three arguments to the '``atomicrmw``' instruction: an
6960operation to apply, an address whose value to modify, an argument to the
6961operation. The operation must be one of the following keywords:
6962
6963- xchg
6964- add
6965- sub
6966- and
6967- nand
6968- or
6969- xor
6970- max
6971- min
6972- umax
6973- umin
6974
6975The type of '<value>' must be an integer type whose bit width is a power
6976of two greater than or equal to eight and less than or equal to a
6977target-specific size limit. The type of the '``<pointer>``' operand must
6978be a pointer to that type. If the ``atomicrmw`` is marked as
6979``volatile``, then the optimizer is not allowed to modify the number or
6980order of execution of this ``atomicrmw`` with other :ref:`volatile
6981operations <volatile>`.
6982
6983Semantics:
6984""""""""""
6985
6986The contents of memory at the location specified by the '``<pointer>``'
6987operand are atomically read, modified, and written back. The original
6988value at the location is returned. The modification is specified by the
6989operation argument:
6990
6991- xchg: ``*ptr = val``
6992- add: ``*ptr = *ptr + val``
6993- sub: ``*ptr = *ptr - val``
6994- and: ``*ptr = *ptr & val``
6995- nand: ``*ptr = ~(*ptr & val)``
6996- or: ``*ptr = *ptr | val``
6997- xor: ``*ptr = *ptr ^ val``
6998- max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
6999- min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
7000- umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
7001 comparison)
7002- umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
7003 comparison)
7004
7005Example:
7006""""""""
7007
7008.. code-block:: llvm
7009
Tim Northover675a0962014-06-13 14:24:23 +00007010 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
Sean Silvab084af42012-12-07 10:36:55 +00007011
7012.. _i_getelementptr:
7013
7014'``getelementptr``' Instruction
7015^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7016
7017Syntax:
7018"""""""
7019
7020::
7021
David Blaikie16a97eb2015-03-04 22:02:58 +00007022 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7023 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
7024 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
Sean Silvab084af42012-12-07 10:36:55 +00007025
7026Overview:
7027"""""""""
7028
7029The '``getelementptr``' instruction is used to get the address of a
7030subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
Elena Demikhovsky37a4da82015-07-09 07:42:48 +00007031address calculation only and does not access memory. The instruction can also
7032be used to calculate a vector of such addresses.
Sean Silvab084af42012-12-07 10:36:55 +00007033
7034Arguments:
7035""""""""""
7036
David Blaikie16a97eb2015-03-04 22:02:58 +00007037The first argument is always a type used as the basis for the calculations.
7038The second argument is always a pointer or a vector of pointers, and is the
7039base address to start from. The remaining arguments are indices
Sean Silvab084af42012-12-07 10:36:55 +00007040that indicate which of the elements of the aggregate object are indexed.
7041The interpretation of each index is dependent on the type being indexed
7042into. The first index always indexes the pointer value given as the
7043first argument, the second index indexes a value of the type pointed to
7044(not necessarily the value directly pointed to, since the first index
7045can be non-zero), etc. The first type indexed into must be a pointer
7046value, subsequent types can be arrays, vectors, and structs. Note that
7047subsequent types being indexed into can never be pointers, since that
7048would require loading the pointer before continuing calculation.
7049
7050The type of each index argument depends on the type it is indexing into.
7051When indexing into a (optionally packed) structure, only ``i32`` integer
7052**constants** are allowed (when using a vector of indices they must all
7053be the **same** ``i32`` integer constant). When indexing into an array,
7054pointer or vector, integers of any width are allowed, and they are not
7055required to be constant. These integers are treated as signed values
7056where relevant.
7057
7058For example, let's consider a C code fragment and how it gets compiled
7059to LLVM:
7060
7061.. code-block:: c
7062
7063 struct RT {
7064 char A;
7065 int B[10][20];
7066 char C;
7067 };
7068 struct ST {
7069 int X;
7070 double Y;
7071 struct RT Z;
7072 };
7073
7074 int *foo(struct ST *s) {
7075 return &s[1].Z.B[5][13];
7076 }
7077
7078The LLVM code generated by Clang is:
7079
7080.. code-block:: llvm
7081
7082 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
7083 %struct.ST = type { i32, double, %struct.RT }
7084
7085 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
7086 entry:
David Blaikie16a97eb2015-03-04 22:02:58 +00007087 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
Sean Silvab084af42012-12-07 10:36:55 +00007088 ret i32* %arrayidx
7089 }
7090
7091Semantics:
7092""""""""""
7093
7094In the example above, the first index is indexing into the
7095'``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
7096= '``{ i32, double, %struct.RT }``' type, a structure. The second index
7097indexes into the third element of the structure, yielding a
7098'``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
7099structure. The third index indexes into the second element of the
7100structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
7101dimensions of the array are subscripted into, yielding an '``i32``'
7102type. The '``getelementptr``' instruction returns a pointer to this
7103element, thus computing a value of '``i32*``' type.
7104
7105Note that it is perfectly legal to index partially through a structure,
7106returning a pointer to an inner element. Because of this, the LLVM code
7107for the given testcase is equivalent to:
7108
7109.. code-block:: llvm
7110
7111 define i32* @foo(%struct.ST* %s) {
David Blaikie16a97eb2015-03-04 22:02:58 +00007112 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
7113 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
7114 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
7115 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
7116 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
Sean Silvab084af42012-12-07 10:36:55 +00007117 ret i32* %t5
7118 }
7119
7120If the ``inbounds`` keyword is present, the result value of the
7121``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
7122pointer is not an *in bounds* address of an allocated object, or if any
7123of the addresses that would be formed by successive addition of the
7124offsets implied by the indices to the base address with infinitely
7125precise signed arithmetic are not an *in bounds* address of that
7126allocated object. The *in bounds* addresses for an allocated object are
7127all the addresses that point into the object, plus the address one byte
7128past the end. In cases where the base is a vector of pointers the
7129``inbounds`` keyword applies to each of the computations element-wise.
7130
7131If the ``inbounds`` keyword is not present, the offsets are added to the
7132base address with silently-wrapping two's complement arithmetic. If the
7133offsets have a different width from the pointer, they are sign-extended
7134or truncated to the width of the pointer. The result value of the
7135``getelementptr`` may be outside the object pointed to by the base
7136pointer. The result value may not necessarily be used to access memory
7137though, even if it happens to point into allocated storage. See the
7138:ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
7139information.
7140
7141The getelementptr instruction is often confusing. For some more insight
7142into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
7143
7144Example:
7145""""""""
7146
7147.. code-block:: llvm
7148
7149 ; yields [12 x i8]*:aptr
David Blaikie16a97eb2015-03-04 22:02:58 +00007150 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
Sean Silvab084af42012-12-07 10:36:55 +00007151 ; yields i8*:vptr
David Blaikie16a97eb2015-03-04 22:02:58 +00007152 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
Sean Silvab084af42012-12-07 10:36:55 +00007153 ; yields i8*:eptr
David Blaikie16a97eb2015-03-04 22:02:58 +00007154 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
Sean Silvab084af42012-12-07 10:36:55 +00007155 ; yields i32*:iptr
David Blaikie16a97eb2015-03-04 22:02:58 +00007156 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
Sean Silvab084af42012-12-07 10:36:55 +00007157
Elena Demikhovsky37a4da82015-07-09 07:42:48 +00007158Vector of pointers:
7159"""""""""""""""""""
7160
7161The ``getelementptr`` returns a vector of pointers, instead of a single address,
7162when one or more of its arguments is a vector. In such cases, all vector
7163arguments should have the same number of elements, and every scalar argument
7164will be effectively broadcast into a vector during address calculation.
Sean Silvab084af42012-12-07 10:36:55 +00007165
7166.. code-block:: llvm
7167
Elena Demikhovsky37a4da82015-07-09 07:42:48 +00007168 ; All arguments are vectors:
7169 ; A[i] = ptrs[i] + offsets[i]*sizeof(i8)
7170 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets
Sean Silva706fba52015-08-06 22:56:24 +00007171
Elena Demikhovsky37a4da82015-07-09 07:42:48 +00007172 ; Add the same scalar offset to each pointer of a vector:
7173 ; A[i] = ptrs[i] + offset*sizeof(i8)
7174 %A = getelementptr i8, <4 x i8*> %ptrs, i64 %offset
Sean Silva706fba52015-08-06 22:56:24 +00007175
Elena Demikhovsky37a4da82015-07-09 07:42:48 +00007176 ; Add distinct offsets to the same pointer:
7177 ; A[i] = ptr + offsets[i]*sizeof(i8)
7178 %A = getelementptr i8, i8* %ptr, <4 x i64> %offsets
Sean Silva706fba52015-08-06 22:56:24 +00007179
Elena Demikhovsky37a4da82015-07-09 07:42:48 +00007180 ; In all cases described above the type of the result is <4 x i8*>
7181
7182The two following instructions are equivalent:
7183
7184.. code-block:: llvm
7185
7186 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7187 <4 x i32> <i32 2, i32 2, i32 2, i32 2>,
7188 <4 x i32> <i32 1, i32 1, i32 1, i32 1>,
7189 <4 x i32> %ind4,
7190 <4 x i64> <i64 13, i64 13, i64 13, i64 13>
Sean Silva706fba52015-08-06 22:56:24 +00007191
Elena Demikhovsky37a4da82015-07-09 07:42:48 +00007192 getelementptr %struct.ST, <4 x %struct.ST*> %s, <4 x i64> %ind1,
7193 i32 2, i32 1, <4 x i32> %ind4, i64 13
7194
7195Let's look at the C code, where the vector version of ``getelementptr``
7196makes sense:
7197
7198.. code-block:: c
7199
7200 // Let's assume that we vectorize the following loop:
7201 double *A, B; int *C;
7202 for (int i = 0; i < size; ++i) {
7203 A[i] = B[C[i]];
7204 }
7205
7206.. code-block:: llvm
7207
7208 ; get pointers for 8 elements from array B
7209 %ptrs = getelementptr double, double* %B, <8 x i32> %C
7210 ; load 8 elements from array B into A
7211 %A = call <8 x double> @llvm.masked.gather.v8f64(<8 x double*> %ptrs,
7212 i32 8, <8 x i1> %mask, <8 x double> %passthru)
Sean Silvab084af42012-12-07 10:36:55 +00007213
7214Conversion Operations
7215---------------------
7216
7217The instructions in this category are the conversion instructions
7218(casting) which all take a single operand and a type. They perform
7219various bit conversions on the operand.
7220
7221'``trunc .. to``' Instruction
7222^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7223
7224Syntax:
7225"""""""
7226
7227::
7228
7229 <result> = trunc <ty> <value> to <ty2> ; yields ty2
7230
7231Overview:
7232"""""""""
7233
7234The '``trunc``' instruction truncates its operand to the type ``ty2``.
7235
7236Arguments:
7237""""""""""
7238
7239The '``trunc``' instruction takes a value to trunc, and a type to trunc
7240it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
7241of the same number of integers. The bit size of the ``value`` must be
7242larger than the bit size of the destination type, ``ty2``. Equal sized
7243types are not allowed.
7244
7245Semantics:
7246""""""""""
7247
7248The '``trunc``' instruction truncates the high order bits in ``value``
7249and converts the remaining bits to ``ty2``. Since the source size must
7250be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
7251It will always truncate bits.
7252
7253Example:
7254""""""""
7255
7256.. code-block:: llvm
7257
7258 %X = trunc i32 257 to i8 ; yields i8:1
7259 %Y = trunc i32 123 to i1 ; yields i1:true
7260 %Z = trunc i32 122 to i1 ; yields i1:false
7261 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
7262
7263'``zext .. to``' Instruction
7264^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7265
7266Syntax:
7267"""""""
7268
7269::
7270
7271 <result> = zext <ty> <value> to <ty2> ; yields ty2
7272
7273Overview:
7274"""""""""
7275
7276The '``zext``' instruction zero extends its operand to type ``ty2``.
7277
7278Arguments:
7279""""""""""
7280
7281The '``zext``' instruction takes a value to cast, and a type to cast it
7282to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7283the same number of integers. The bit size of the ``value`` must be
7284smaller than the bit size of the destination type, ``ty2``.
7285
7286Semantics:
7287""""""""""
7288
7289The ``zext`` fills the high order bits of the ``value`` with zero bits
7290until it reaches the size of the destination type, ``ty2``.
7291
7292When zero extending from i1, the result will always be either 0 or 1.
7293
7294Example:
7295""""""""
7296
7297.. code-block:: llvm
7298
7299 %X = zext i32 257 to i64 ; yields i64:257
7300 %Y = zext i1 true to i32 ; yields i32:1
7301 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7302
7303'``sext .. to``' Instruction
7304^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7305
7306Syntax:
7307"""""""
7308
7309::
7310
7311 <result> = sext <ty> <value> to <ty2> ; yields ty2
7312
7313Overview:
7314"""""""""
7315
7316The '``sext``' sign extends ``value`` to the type ``ty2``.
7317
7318Arguments:
7319""""""""""
7320
7321The '``sext``' instruction takes a value to cast, and a type to cast it
7322to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
7323the same number of integers. The bit size of the ``value`` must be
7324smaller than the bit size of the destination type, ``ty2``.
7325
7326Semantics:
7327""""""""""
7328
7329The '``sext``' instruction performs a sign extension by copying the sign
7330bit (highest order bit) of the ``value`` until it reaches the bit size
7331of the type ``ty2``.
7332
7333When sign extending from i1, the extension always results in -1 or 0.
7334
7335Example:
7336""""""""
7337
7338.. code-block:: llvm
7339
7340 %X = sext i8 -1 to i16 ; yields i16 :65535
7341 %Y = sext i1 true to i32 ; yields i32:-1
7342 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
7343
7344'``fptrunc .. to``' Instruction
7345^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7346
7347Syntax:
7348"""""""
7349
7350::
7351
7352 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
7353
7354Overview:
7355"""""""""
7356
7357The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
7358
7359Arguments:
7360""""""""""
7361
7362The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
7363value to cast and a :ref:`floating point <t_floating>` type to cast it to.
7364The size of ``value`` must be larger than the size of ``ty2``. This
7365implies that ``fptrunc`` cannot be used to make a *no-op cast*.
7366
7367Semantics:
7368""""""""""
7369
7370The '``fptrunc``' instruction truncates a ``value`` from a larger
7371:ref:`floating point <t_floating>` type to a smaller :ref:`floating
7372point <t_floating>` type. If the value cannot fit within the
7373destination type, ``ty2``, then the results are undefined.
7374
7375Example:
7376""""""""
7377
7378.. code-block:: llvm
7379
7380 %X = fptrunc double 123.0 to float ; yields float:123.0
7381 %Y = fptrunc double 1.0E+300 to float ; yields undefined
7382
7383'``fpext .. to``' Instruction
7384^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7385
7386Syntax:
7387"""""""
7388
7389::
7390
7391 <result> = fpext <ty> <value> to <ty2> ; yields ty2
7392
7393Overview:
7394"""""""""
7395
7396The '``fpext``' extends a floating point ``value`` to a larger floating
7397point value.
7398
7399Arguments:
7400""""""""""
7401
7402The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
7403``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
7404to. The source type must be smaller than the destination type.
7405
7406Semantics:
7407""""""""""
7408
7409The '``fpext``' instruction extends the ``value`` from a smaller
7410:ref:`floating point <t_floating>` type to a larger :ref:`floating
7411point <t_floating>` type. The ``fpext`` cannot be used to make a
7412*no-op cast* because it always changes bits. Use ``bitcast`` to make a
7413*no-op cast* for a floating point cast.
7414
7415Example:
7416""""""""
7417
7418.. code-block:: llvm
7419
7420 %X = fpext float 3.125 to double ; yields double:3.125000e+00
7421 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
7422
7423'``fptoui .. to``' Instruction
7424^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7425
7426Syntax:
7427"""""""
7428
7429::
7430
7431 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
7432
7433Overview:
7434"""""""""
7435
7436The '``fptoui``' converts a floating point ``value`` to its unsigned
7437integer equivalent of type ``ty2``.
7438
7439Arguments:
7440""""""""""
7441
7442The '``fptoui``' instruction takes a value to cast, which must be a
7443scalar or vector :ref:`floating point <t_floating>` value, and a type to
7444cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7445``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7446type with the same number of elements as ``ty``
7447
7448Semantics:
7449""""""""""
7450
7451The '``fptoui``' instruction converts its :ref:`floating
7452point <t_floating>` operand into the nearest (rounding towards zero)
7453unsigned integer value. If the value cannot fit in ``ty2``, the results
7454are undefined.
7455
7456Example:
7457""""""""
7458
7459.. code-block:: llvm
7460
7461 %X = fptoui double 123.0 to i32 ; yields i32:123
7462 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
7463 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
7464
7465'``fptosi .. to``' Instruction
7466^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7467
7468Syntax:
7469"""""""
7470
7471::
7472
7473 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
7474
7475Overview:
7476"""""""""
7477
7478The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
7479``value`` to type ``ty2``.
7480
7481Arguments:
7482""""""""""
7483
7484The '``fptosi``' instruction takes a value to cast, which must be a
7485scalar or vector :ref:`floating point <t_floating>` value, and a type to
7486cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
7487``ty`` is a vector floating point type, ``ty2`` must be a vector integer
7488type with the same number of elements as ``ty``
7489
7490Semantics:
7491""""""""""
7492
7493The '``fptosi``' instruction converts its :ref:`floating
7494point <t_floating>` operand into the nearest (rounding towards zero)
7495signed integer value. If the value cannot fit in ``ty2``, the results
7496are undefined.
7497
7498Example:
7499""""""""
7500
7501.. code-block:: llvm
7502
7503 %X = fptosi double -123.0 to i32 ; yields i32:-123
7504 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
7505 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
7506
7507'``uitofp .. to``' Instruction
7508^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7509
7510Syntax:
7511"""""""
7512
7513::
7514
7515 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
7516
7517Overview:
7518"""""""""
7519
7520The '``uitofp``' instruction regards ``value`` as an unsigned integer
7521and converts that value to the ``ty2`` type.
7522
7523Arguments:
7524""""""""""
7525
7526The '``uitofp``' instruction takes a value to cast, which must be a
7527scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7528``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7529``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7530type with the same number of elements as ``ty``
7531
7532Semantics:
7533""""""""""
7534
7535The '``uitofp``' instruction interprets its operand as an unsigned
7536integer quantity and converts it to the corresponding floating point
7537value. If the value cannot fit in the floating point value, the results
7538are undefined.
7539
7540Example:
7541""""""""
7542
7543.. code-block:: llvm
7544
7545 %X = uitofp i32 257 to float ; yields float:257.0
7546 %Y = uitofp i8 -1 to double ; yields double:255.0
7547
7548'``sitofp .. to``' Instruction
7549^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7550
7551Syntax:
7552"""""""
7553
7554::
7555
7556 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
7557
7558Overview:
7559"""""""""
7560
7561The '``sitofp``' instruction regards ``value`` as a signed integer and
7562converts that value to the ``ty2`` type.
7563
7564Arguments:
7565""""""""""
7566
7567The '``sitofp``' instruction takes a value to cast, which must be a
7568scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
7569``ty2``, which must be an :ref:`floating point <t_floating>` type. If
7570``ty`` is a vector integer type, ``ty2`` must be a vector floating point
7571type with the same number of elements as ``ty``
7572
7573Semantics:
7574""""""""""
7575
7576The '``sitofp``' instruction interprets its operand as a signed integer
7577quantity and converts it to the corresponding floating point value. If
7578the value cannot fit in the floating point value, the results are
7579undefined.
7580
7581Example:
7582""""""""
7583
7584.. code-block:: llvm
7585
7586 %X = sitofp i32 257 to float ; yields float:257.0
7587 %Y = sitofp i8 -1 to double ; yields double:-1.0
7588
7589.. _i_ptrtoint:
7590
7591'``ptrtoint .. to``' Instruction
7592^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7593
7594Syntax:
7595"""""""
7596
7597::
7598
7599 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
7600
7601Overview:
7602"""""""""
7603
7604The '``ptrtoint``' instruction converts the pointer or a vector of
7605pointers ``value`` to the integer (or vector of integers) type ``ty2``.
7606
7607Arguments:
7608""""""""""
7609
7610The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
Ed Maste8ed40ce2015-04-14 20:52:58 +00007611a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
Sean Silvab084af42012-12-07 10:36:55 +00007612type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
7613a vector of integers type.
7614
7615Semantics:
7616""""""""""
7617
7618The '``ptrtoint``' instruction converts ``value`` to integer type
7619``ty2`` by interpreting the pointer value as an integer and either
7620truncating or zero extending that value to the size of the integer type.
7621If ``value`` is smaller than ``ty2`` then a zero extension is done. If
7622``value`` is larger than ``ty2`` then a truncation is done. If they are
7623the same size, then nothing is done (*no-op cast*) other than a type
7624change.
7625
7626Example:
7627""""""""
7628
7629.. code-block:: llvm
7630
7631 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
7632 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
7633 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
7634
7635.. _i_inttoptr:
7636
7637'``inttoptr .. to``' Instruction
7638^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7639
7640Syntax:
7641"""""""
7642
7643::
7644
7645 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
7646
7647Overview:
7648"""""""""
7649
7650The '``inttoptr``' instruction converts an integer ``value`` to a
7651pointer type, ``ty2``.
7652
7653Arguments:
7654""""""""""
7655
7656The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
7657cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
7658type.
7659
7660Semantics:
7661""""""""""
7662
7663The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
7664applying either a zero extension or a truncation depending on the size
7665of the integer ``value``. If ``value`` is larger than the size of a
7666pointer then a truncation is done. If ``value`` is smaller than the size
7667of a pointer then a zero extension is done. If they are the same size,
7668nothing is done (*no-op cast*).
7669
7670Example:
7671""""""""
7672
7673.. code-block:: llvm
7674
7675 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
7676 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
7677 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
7678 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
7679
7680.. _i_bitcast:
7681
7682'``bitcast .. to``' Instruction
7683^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7684
7685Syntax:
7686"""""""
7687
7688::
7689
7690 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
7691
7692Overview:
7693"""""""""
7694
7695The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
7696changing any bits.
7697
7698Arguments:
7699""""""""""
7700
7701The '``bitcast``' instruction takes a value to cast, which must be a
7702non-aggregate first class value, and a type to cast it to, which must
Matt Arsenault24b49c42013-07-31 17:49:08 +00007703also be a non-aggregate :ref:`first class <t_firstclass>` type. The
7704bit sizes of ``value`` and the destination type, ``ty2``, must be
7705identical. If the source type is a pointer, the destination type must
7706also be a pointer of the same size. This instruction supports bitwise
7707conversion of vectors to integers and to vectors of other types (as
7708long as they have the same size).
Sean Silvab084af42012-12-07 10:36:55 +00007709
7710Semantics:
7711""""""""""
7712
Matt Arsenault24b49c42013-07-31 17:49:08 +00007713The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
7714is always a *no-op cast* because no bits change with this
7715conversion. The conversion is done as if the ``value`` had been stored
7716to memory and read back as type ``ty2``. Pointer (or vector of
7717pointers) types may only be converted to other pointer (or vector of
Matt Arsenaultb03bd4d2013-11-15 01:34:59 +00007718pointers) types with the same address space through this instruction.
7719To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
7720or :ref:`ptrtoint <i_ptrtoint>` instructions first.
Sean Silvab084af42012-12-07 10:36:55 +00007721
7722Example:
7723""""""""
7724
7725.. code-block:: llvm
7726
7727 %X = bitcast i8 255 to i8 ; yields i8 :-1
7728 %Y = bitcast i32* %x to sint* ; yields sint*:%x
7729 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
7730 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
7731
Matt Arsenaultb03bd4d2013-11-15 01:34:59 +00007732.. _i_addrspacecast:
7733
7734'``addrspacecast .. to``' Instruction
7735^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7736
7737Syntax:
7738"""""""
7739
7740::
7741
7742 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
7743
7744Overview:
7745"""""""""
7746
7747The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
7748address space ``n`` to type ``pty2`` in address space ``m``.
7749
7750Arguments:
7751""""""""""
7752
7753The '``addrspacecast``' instruction takes a pointer or vector of pointer value
7754to cast and a pointer type to cast it to, which must have a different
7755address space.
7756
7757Semantics:
7758""""""""""
7759
7760The '``addrspacecast``' instruction converts the pointer value
7761``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
Matt Arsenault54a2a172013-11-15 05:44:56 +00007762value modification, depending on the target and the address space
7763pair. Pointer conversions within the same address space must be
7764performed with the ``bitcast`` instruction. Note that if the address space
Matt Arsenaultb03bd4d2013-11-15 01:34:59 +00007765conversion is legal then both result and operand refer to the same memory
7766location.
7767
7768Example:
7769""""""""
7770
7771.. code-block:: llvm
7772
Matt Arsenault9c13dd02013-11-15 22:43:50 +00007773 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
7774 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
7775 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
Matt Arsenaultb03bd4d2013-11-15 01:34:59 +00007776
Sean Silvab084af42012-12-07 10:36:55 +00007777.. _otherops:
7778
7779Other Operations
7780----------------
7781
7782The instructions in this category are the "miscellaneous" instructions,
7783which defy better classification.
7784
7785.. _i_icmp:
7786
7787'``icmp``' Instruction
7788^^^^^^^^^^^^^^^^^^^^^^
7789
7790Syntax:
7791"""""""
7792
7793::
7794
Tim Northover675a0962014-06-13 14:24:23 +00007795 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
Sean Silvab084af42012-12-07 10:36:55 +00007796
7797Overview:
7798"""""""""
7799
7800The '``icmp``' instruction returns a boolean value or a vector of
7801boolean values based on comparison of its two integer, integer vector,
7802pointer, or pointer vector operands.
7803
7804Arguments:
7805""""""""""
7806
7807The '``icmp``' instruction takes three operands. The first operand is
7808the condition code indicating the kind of comparison to perform. It is
7809not a value, just a keyword. The possible condition code are:
7810
7811#. ``eq``: equal
7812#. ``ne``: not equal
7813#. ``ugt``: unsigned greater than
7814#. ``uge``: unsigned greater or equal
7815#. ``ult``: unsigned less than
7816#. ``ule``: unsigned less or equal
7817#. ``sgt``: signed greater than
7818#. ``sge``: signed greater or equal
7819#. ``slt``: signed less than
7820#. ``sle``: signed less or equal
7821
7822The remaining two arguments must be :ref:`integer <t_integer>` or
7823:ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
7824must also be identical types.
7825
7826Semantics:
7827""""""""""
7828
7829The '``icmp``' compares ``op1`` and ``op2`` according to the condition
7830code given as ``cond``. The comparison performed always yields either an
7831:ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
7832
7833#. ``eq``: yields ``true`` if the operands are equal, ``false``
7834 otherwise. No sign interpretation is necessary or performed.
7835#. ``ne``: yields ``true`` if the operands are unequal, ``false``
7836 otherwise. No sign interpretation is necessary or performed.
7837#. ``ugt``: interprets the operands as unsigned values and yields
7838 ``true`` if ``op1`` is greater than ``op2``.
7839#. ``uge``: interprets the operands as unsigned values and yields
7840 ``true`` if ``op1`` is greater than or equal to ``op2``.
7841#. ``ult``: interprets the operands as unsigned values and yields
7842 ``true`` if ``op1`` is less than ``op2``.
7843#. ``ule``: interprets the operands as unsigned values and yields
7844 ``true`` if ``op1`` is less than or equal to ``op2``.
7845#. ``sgt``: interprets the operands as signed values and yields ``true``
7846 if ``op1`` is greater than ``op2``.
7847#. ``sge``: interprets the operands as signed values and yields ``true``
7848 if ``op1`` is greater than or equal to ``op2``.
7849#. ``slt``: interprets the operands as signed values and yields ``true``
7850 if ``op1`` is less than ``op2``.
7851#. ``sle``: interprets the operands as signed values and yields ``true``
7852 if ``op1`` is less than or equal to ``op2``.
7853
7854If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
7855are compared as if they were integers.
7856
7857If the operands are integer vectors, then they are compared element by
7858element. The result is an ``i1`` vector with the same number of elements
7859as the values being compared. Otherwise, the result is an ``i1``.
7860
7861Example:
7862""""""""
7863
7864.. code-block:: llvm
7865
7866 <result> = icmp eq i32 4, 5 ; yields: result=false
7867 <result> = icmp ne float* %X, %X ; yields: result=false
7868 <result> = icmp ult i16 4, 5 ; yields: result=true
7869 <result> = icmp sgt i16 4, 5 ; yields: result=false
7870 <result> = icmp ule i16 -4, 5 ; yields: result=false
7871 <result> = icmp sge i16 4, 5 ; yields: result=false
7872
7873Note that the code generator does not yet support vector types with the
7874``icmp`` instruction.
7875
7876.. _i_fcmp:
7877
7878'``fcmp``' Instruction
7879^^^^^^^^^^^^^^^^^^^^^^
7880
7881Syntax:
7882"""""""
7883
7884::
7885
James Molloy88eb5352015-07-10 12:52:00 +00007886 <result> = fcmp [fast-math flags]* <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
Sean Silvab084af42012-12-07 10:36:55 +00007887
7888Overview:
7889"""""""""
7890
7891The '``fcmp``' instruction returns a boolean value or vector of boolean
7892values based on comparison of its operands.
7893
7894If the operands are floating point scalars, then the result type is a
7895boolean (:ref:`i1 <t_integer>`).
7896
7897If the operands are floating point vectors, then the result type is a
7898vector of boolean with the same number of elements as the operands being
7899compared.
7900
7901Arguments:
7902""""""""""
7903
7904The '``fcmp``' instruction takes three operands. The first operand is
7905the condition code indicating the kind of comparison to perform. It is
7906not a value, just a keyword. The possible condition code are:
7907
7908#. ``false``: no comparison, always returns false
7909#. ``oeq``: ordered and equal
7910#. ``ogt``: ordered and greater than
7911#. ``oge``: ordered and greater than or equal
7912#. ``olt``: ordered and less than
7913#. ``ole``: ordered and less than or equal
7914#. ``one``: ordered and not equal
7915#. ``ord``: ordered (no nans)
7916#. ``ueq``: unordered or equal
7917#. ``ugt``: unordered or greater than
7918#. ``uge``: unordered or greater than or equal
7919#. ``ult``: unordered or less than
7920#. ``ule``: unordered or less than or equal
7921#. ``une``: unordered or not equal
7922#. ``uno``: unordered (either nans)
7923#. ``true``: no comparison, always returns true
7924
7925*Ordered* means that neither operand is a QNAN while *unordered* means
7926that either operand may be a QNAN.
7927
7928Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
7929point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
7930type. They must have identical types.
7931
7932Semantics:
7933""""""""""
7934
7935The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
7936condition code given as ``cond``. If the operands are vectors, then the
7937vectors are compared element by element. Each comparison performed
7938always yields an :ref:`i1 <t_integer>` result, as follows:
7939
7940#. ``false``: always yields ``false``, regardless of operands.
7941#. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
7942 is equal to ``op2``.
7943#. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
7944 is greater than ``op2``.
7945#. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
7946 is greater than or equal to ``op2``.
7947#. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
7948 is less than ``op2``.
7949#. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
7950 is less than or equal to ``op2``.
7951#. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
7952 is not equal to ``op2``.
7953#. ``ord``: yields ``true`` if both operands are not a QNAN.
7954#. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
7955 equal to ``op2``.
7956#. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
7957 greater than ``op2``.
7958#. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
7959 greater than or equal to ``op2``.
7960#. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
7961 less than ``op2``.
7962#. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
7963 less than or equal to ``op2``.
7964#. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
7965 not equal to ``op2``.
7966#. ``uno``: yields ``true`` if either operand is a QNAN.
7967#. ``true``: always yields ``true``, regardless of operands.
7968
James Molloy88eb5352015-07-10 12:52:00 +00007969The ``fcmp`` instruction can also optionally take any number of
7970:ref:`fast-math flags <fastmath>`, which are optimization hints to enable
7971otherwise unsafe floating point optimizations.
7972
7973Any set of fast-math flags are legal on an ``fcmp`` instruction, but the
7974only flags that have any effect on its semantics are those that allow
7975assumptions to be made about the values of input arguments; namely
7976``nnan``, ``ninf``, and ``nsz``. See :ref:`fastmath` for more information.
7977
Sean Silvab084af42012-12-07 10:36:55 +00007978Example:
7979""""""""
7980
7981.. code-block:: llvm
7982
7983 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
7984 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
7985 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
7986 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
7987
7988Note that the code generator does not yet support vector types with the
7989``fcmp`` instruction.
7990
7991.. _i_phi:
7992
7993'``phi``' Instruction
7994^^^^^^^^^^^^^^^^^^^^^
7995
7996Syntax:
7997"""""""
7998
7999::
8000
8001 <result> = phi <ty> [ <val0>, <label0>], ...
8002
8003Overview:
8004"""""""""
8005
8006The '``phi``' instruction is used to implement the φ node in the SSA
8007graph representing the function.
8008
8009Arguments:
8010""""""""""
8011
8012The type of the incoming values is specified with the first type field.
8013After this, the '``phi``' instruction takes a list of pairs as
8014arguments, with one pair for each predecessor basic block of the current
8015block. Only values of :ref:`first class <t_firstclass>` type may be used as
8016the value arguments to the PHI node. Only labels may be used as the
8017label arguments.
8018
8019There must be no non-phi instructions between the start of a basic block
8020and the PHI instructions: i.e. PHI instructions must be first in a basic
8021block.
8022
8023For the purposes of the SSA form, the use of each incoming value is
8024deemed to occur on the edge from the corresponding predecessor block to
8025the current block (but after any definition of an '``invoke``'
8026instruction's return value on the same edge).
8027
8028Semantics:
8029""""""""""
8030
8031At runtime, the '``phi``' instruction logically takes on the value
8032specified by the pair corresponding to the predecessor basic block that
8033executed just prior to the current block.
8034
8035Example:
8036""""""""
8037
8038.. code-block:: llvm
8039
8040 Loop: ; Infinite loop that counts from 0 on up...
8041 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
8042 %nextindvar = add i32 %indvar, 1
8043 br label %Loop
8044
8045.. _i_select:
8046
8047'``select``' Instruction
8048^^^^^^^^^^^^^^^^^^^^^^^^
8049
8050Syntax:
8051"""""""
8052
8053::
8054
8055 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
8056
8057 selty is either i1 or {<N x i1>}
8058
8059Overview:
8060"""""""""
8061
8062The '``select``' instruction is used to choose one value based on a
Joerg Sonnenberger94321ec2014-03-26 15:30:21 +00008063condition, without IR-level branching.
Sean Silvab084af42012-12-07 10:36:55 +00008064
8065Arguments:
8066""""""""""
8067
8068The '``select``' instruction requires an 'i1' value or a vector of 'i1'
8069values indicating the condition, and two values of the same :ref:`first
David Majnemer40a0b592015-03-03 22:45:47 +00008070class <t_firstclass>` type.
Sean Silvab084af42012-12-07 10:36:55 +00008071
8072Semantics:
8073""""""""""
8074
8075If the condition is an i1 and it evaluates to 1, the instruction returns
8076the first value argument; otherwise, it returns the second value
8077argument.
8078
8079If the condition is a vector of i1, then the value arguments must be
8080vectors of the same size, and the selection is done element by element.
8081
David Majnemer40a0b592015-03-03 22:45:47 +00008082If the condition is an i1 and the value arguments are vectors of the
8083same size, then an entire vector is selected.
8084
Sean Silvab084af42012-12-07 10:36:55 +00008085Example:
8086""""""""
8087
8088.. code-block:: llvm
8089
8090 %X = select i1 true, i8 17, i8 42 ; yields i8:17
8091
8092.. _i_call:
8093
8094'``call``' Instruction
8095^^^^^^^^^^^^^^^^^^^^^^
8096
8097Syntax:
8098"""""""
8099
8100::
8101
Reid Kleckner5772b772014-04-24 20:14:34 +00008102 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
Sean Silvab084af42012-12-07 10:36:55 +00008103
8104Overview:
8105"""""""""
8106
8107The '``call``' instruction represents a simple function call.
8108
8109Arguments:
8110""""""""""
8111
8112This instruction requires several arguments:
8113
Reid Kleckner5772b772014-04-24 20:14:34 +00008114#. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
8115 should perform tail call optimization. The ``tail`` marker is a hint that
8116 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
8117 means that the call must be tail call optimized in order for the program to
8118 be correct. The ``musttail`` marker provides these guarantees:
8119
8120 #. The call will not cause unbounded stack growth if it is part of a
8121 recursive cycle in the call graph.
8122 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
8123 forwarded in place.
8124
8125 Both markers imply that the callee does not access allocas or varargs from
8126 the caller. Calls marked ``musttail`` must obey the following additional
8127 rules:
8128
8129 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
8130 or a pointer bitcast followed by a ret instruction.
8131 - The ret instruction must return the (possibly bitcasted) value
8132 produced by the call or void.
8133 - The caller and callee prototypes must match. Pointer types of
8134 parameters or return types may differ in pointee type, but not
8135 in address space.
8136 - The calling conventions of the caller and callee must match.
8137 - All ABI-impacting function attributes, such as sret, byval, inreg,
8138 returned, and inalloca, must match.
Reid Kleckner83498642014-08-26 00:33:28 +00008139 - The callee must be varargs iff the caller is varargs. Bitcasting a
8140 non-varargs function to the appropriate varargs type is legal so
8141 long as the non-varargs prefixes obey the other rules.
Reid Kleckner5772b772014-04-24 20:14:34 +00008142
8143 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
8144 the following conditions are met:
Sean Silvab084af42012-12-07 10:36:55 +00008145
8146 - Caller and callee both have the calling convention ``fastcc``.
8147 - The call is in tail position (ret immediately follows call and ret
8148 uses value of call or is void).
8149 - Option ``-tailcallopt`` is enabled, or
8150 ``llvm::GuaranteedTailCallOpt`` is ``true``.
Alp Tokercf218752014-06-30 18:57:16 +00008151 - `Platform-specific constraints are
Sean Silvab084af42012-12-07 10:36:55 +00008152 met. <CodeGenerator.html#tailcallopt>`_
8153
8154#. The optional "cconv" marker indicates which :ref:`calling
8155 convention <callingconv>` the call should use. If none is
8156 specified, the call defaults to using C calling conventions. The
8157 calling convention of the call must match the calling convention of
8158 the target function, or else the behavior is undefined.
8159#. The optional :ref:`Parameter Attributes <paramattrs>` list for return
8160 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
8161 are valid here.
8162#. '``ty``': the type of the call instruction itself which is also the
8163 type of the return value. Functions that return no value are marked
8164 ``void``.
8165#. '``fnty``': shall be the signature of the pointer to function value
8166 being invoked. The argument types must match the types implied by
8167 this signature. This type can be omitted if the function is not
8168 varargs and if the function type does not return a pointer to a
8169 function.
8170#. '``fnptrval``': An LLVM value containing a pointer to a function to
8171 be invoked. In most cases, this is a direct function invocation, but
8172 indirect ``call``'s are just as possible, calling an arbitrary pointer
8173 to function value.
8174#. '``function args``': argument list whose types match the function
8175 signature argument types and parameter attributes. All arguments must
8176 be of :ref:`first class <t_firstclass>` type. If the function signature
8177 indicates the function accepts a variable number of arguments, the
8178 extra arguments can be specified.
8179#. The optional :ref:`function attributes <fnattrs>` list. Only
8180 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
8181 attributes are valid here.
8182
8183Semantics:
8184""""""""""
8185
8186The '``call``' instruction is used to cause control flow to transfer to
8187a specified function, with its incoming arguments bound to the specified
8188values. Upon a '``ret``' instruction in the called function, control
8189flow continues with the instruction after the function call, and the
8190return value of the function is bound to the result argument.
8191
8192Example:
8193""""""""
8194
8195.. code-block:: llvm
8196
8197 %retval = call i32 @test(i32 %argc)
8198 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
8199 %X = tail call i32 @foo() ; yields i32
8200 %Y = tail call fastcc i32 @foo() ; yields i32
8201 call void %foo(i8 97 signext)
8202
8203 %struct.A = type { i32, i8 }
Tim Northover675a0962014-06-13 14:24:23 +00008204 %r = call %struct.A @foo() ; yields { i32, i8 }
Sean Silvab084af42012-12-07 10:36:55 +00008205 %gr = extractvalue %struct.A %r, 0 ; yields i32
8206 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
8207 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
8208 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
8209
8210llvm treats calls to some functions with names and arguments that match
8211the standard C99 library as being the C99 library functions, and may
8212perform optimizations or generate code for them under that assumption.
8213This is something we'd like to change in the future to provide better
8214support for freestanding environments and non-C-based languages.
8215
8216.. _i_va_arg:
8217
8218'``va_arg``' Instruction
8219^^^^^^^^^^^^^^^^^^^^^^^^
8220
8221Syntax:
8222"""""""
8223
8224::
8225
8226 <resultval> = va_arg <va_list*> <arglist>, <argty>
8227
8228Overview:
8229"""""""""
8230
8231The '``va_arg``' instruction is used to access arguments passed through
8232the "variable argument" area of a function call. It is used to implement
8233the ``va_arg`` macro in C.
8234
8235Arguments:
8236""""""""""
8237
8238This instruction takes a ``va_list*`` value and the type of the
8239argument. It returns a value of the specified argument type and
8240increments the ``va_list`` to point to the next argument. The actual
8241type of ``va_list`` is target specific.
8242
8243Semantics:
8244""""""""""
8245
8246The '``va_arg``' instruction loads an argument of the specified type
8247from the specified ``va_list`` and causes the ``va_list`` to point to
8248the next argument. For more information, see the variable argument
8249handling :ref:`Intrinsic Functions <int_varargs>`.
8250
8251It is legal for this instruction to be called in a function which does
8252not take a variable number of arguments, for example, the ``vfprintf``
8253function.
8254
8255``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
8256function <intrinsics>` because it takes a type as an argument.
8257
8258Example:
8259""""""""
8260
8261See the :ref:`variable argument processing <int_varargs>` section.
8262
8263Note that the code generator does not yet fully support va\_arg on many
8264targets. Also, it does not currently support va\_arg with aggregate
8265types on any target.
8266
8267.. _i_landingpad:
8268
8269'``landingpad``' Instruction
8270^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8271
8272Syntax:
8273"""""""
8274
8275::
8276
David Majnemer7fddecc2015-06-17 20:52:32 +00008277 <resultval> = landingpad <resultty> <clause>+
8278 <resultval> = landingpad <resultty> cleanup <clause>*
Sean Silvab084af42012-12-07 10:36:55 +00008279
8280 <clause> := catch <type> <value>
8281 <clause> := filter <array constant type> <array constant>
8282
8283Overview:
8284"""""""""
8285
8286The '``landingpad``' instruction is used by `LLVM's exception handling
8287system <ExceptionHandling.html#overview>`_ to specify that a basic block
Dmitri Gribenkoe8131122013-01-19 20:34:20 +00008288is a landing pad --- one where the exception lands, and corresponds to the
Sean Silvab084af42012-12-07 10:36:55 +00008289code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
David Majnemer7fddecc2015-06-17 20:52:32 +00008290defines values supplied by the :ref:`personality function <personalityfn>` upon
Sean Silvab084af42012-12-07 10:36:55 +00008291re-entry to the function. The ``resultval`` has the type ``resultty``.
8292
8293Arguments:
8294""""""""""
8295
David Majnemer7fddecc2015-06-17 20:52:32 +00008296The optional
Sean Silvab084af42012-12-07 10:36:55 +00008297``cleanup`` flag indicates that the landing pad block is a cleanup.
8298
Dmitri Gribenkoe8131122013-01-19 20:34:20 +00008299A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
Sean Silvab084af42012-12-07 10:36:55 +00008300contains the global variable representing the "type" that may be caught
8301or filtered respectively. Unlike the ``catch`` clause, the ``filter``
8302clause takes an array constant as its argument. Use
8303"``[0 x i8**] undef``" for a filter which cannot throw. The
8304'``landingpad``' instruction must contain *at least* one ``clause`` or
8305the ``cleanup`` flag.
8306
8307Semantics:
8308""""""""""
8309
8310The '``landingpad``' instruction defines the values which are set by the
David Majnemer7fddecc2015-06-17 20:52:32 +00008311:ref:`personality function <personalityfn>` upon re-entry to the function, and
Sean Silvab084af42012-12-07 10:36:55 +00008312therefore the "result type" of the ``landingpad`` instruction. As with
8313calling conventions, how the personality function results are
8314represented in LLVM IR is target specific.
8315
8316The clauses are applied in order from top to bottom. If two
8317``landingpad`` instructions are merged together through inlining, the
8318clauses from the calling function are appended to the list of clauses.
8319When the call stack is being unwound due to an exception being thrown,
8320the exception is compared against each ``clause`` in turn. If it doesn't
8321match any of the clauses, and the ``cleanup`` flag is not set, then
8322unwinding continues further up the call stack.
8323
8324The ``landingpad`` instruction has several restrictions:
8325
8326- A landing pad block is a basic block which is the unwind destination
8327 of an '``invoke``' instruction.
8328- A landing pad block must have a '``landingpad``' instruction as its
8329 first non-PHI instruction.
8330- There can be only one '``landingpad``' instruction within the landing
8331 pad block.
8332- A basic block that is not a landing pad block may not include a
8333 '``landingpad``' instruction.
Sean Silvab084af42012-12-07 10:36:55 +00008334
8335Example:
8336""""""""
8337
8338.. code-block:: llvm
8339
8340 ;; A landing pad which can catch an integer.
David Majnemer7fddecc2015-06-17 20:52:32 +00008341 %res = landingpad { i8*, i32 }
Sean Silvab084af42012-12-07 10:36:55 +00008342 catch i8** @_ZTIi
8343 ;; A landing pad that is a cleanup.
David Majnemer7fddecc2015-06-17 20:52:32 +00008344 %res = landingpad { i8*, i32 }
Sean Silvab084af42012-12-07 10:36:55 +00008345 cleanup
8346 ;; A landing pad which can catch an integer and can only throw a double.
David Majnemer7fddecc2015-06-17 20:52:32 +00008347 %res = landingpad { i8*, i32 }
Sean Silvab084af42012-12-07 10:36:55 +00008348 catch i8** @_ZTIi
8349 filter [1 x i8**] [@_ZTId]
8350
David Majnemer654e1302015-07-31 17:58:14 +00008351.. _i_cleanuppad:
8352
8353'``cleanuppad``' Instruction
8354^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8355
8356Syntax:
8357"""""""
8358
8359::
8360
8361 <resultval> = cleanuppad <resultty> [<args>*]
8362
8363Overview:
8364"""""""""
8365
8366The '``cleanuppad``' instruction is used by `LLVM's exception handling
8367system <ExceptionHandling.html#overview>`_ to specify that a basic block
8368is a cleanup block --- one where a personality routine attempts to
8369transfer control to run cleanup actions.
8370The ``args`` correspond to whatever additional
8371information the :ref:`personality function <personalityfn>` requires to
8372execute the cleanup.
8373The ``resultval`` has the type ``resultty``.
8374
8375Arguments:
8376""""""""""
8377
8378The instruction takes a list of arbitrary values which are interpreted
8379by the :ref:`personality function <personalityfn>`.
8380
8381Semantics:
8382""""""""""
8383
8384The '``cleanuppad``' instruction defines the values which are set by the
8385:ref:`personality function <personalityfn>` upon re-entry to the function, and
8386therefore the "result type" of the ``cleanuppad`` instruction. As with
8387calling conventions, how the personality function results are
8388represented in LLVM IR is target specific.
8389
8390When the call stack is being unwound due to an exception being thrown,
8391the :ref:`personality function <personalityfn>` transfers control to the
8392``cleanuppad`` with the aid of the personality-specific arguments.
8393
8394The ``cleanuppad`` instruction has several restrictions:
8395
8396- A cleanup block is a basic block which is the unwind destination of
8397 an exceptional instruction.
8398- A cleanup block must have a '``cleanuppad``' instruction as its
8399 first non-PHI instruction.
8400- There can be only one '``cleanuppad``' instruction within the
8401 cleanup block.
8402- A basic block that is not a cleanup block may not include a
8403 '``cleanuppad``' instruction.
8404- It is undefined behavior for control to transfer from a ``cleanuppad`` to a
8405 ``catchret`` without first executing a ``cleanupret`` and a subsequent
8406 ``catchpad``.
8407- It is undefined behavior for control to transfer from a ``cleanuppad`` to a
8408 ``ret`` without first executing a ``cleanupret``.
8409
8410Example:
8411""""""""
8412
8413.. code-block:: llvm
8414
8415 %res = cleanuppad { i8*, i32 } [label %nextaction]
8416
Sean Silvab084af42012-12-07 10:36:55 +00008417.. _intrinsics:
8418
8419Intrinsic Functions
8420===================
8421
8422LLVM supports the notion of an "intrinsic function". These functions
8423have well known names and semantics and are required to follow certain
8424restrictions. Overall, these intrinsics represent an extension mechanism
8425for the LLVM language that does not require changing all of the
8426transformations in LLVM when adding to the language (or the bitcode
8427reader/writer, the parser, etc...).
8428
8429Intrinsic function names must all start with an "``llvm.``" prefix. This
8430prefix is reserved in LLVM for intrinsic names; thus, function names may
8431not begin with this prefix. Intrinsic functions must always be external
8432functions: you cannot define the body of intrinsic functions. Intrinsic
8433functions may only be used in call or invoke instructions: it is illegal
8434to take the address of an intrinsic function. Additionally, because
8435intrinsic functions are part of the LLVM language, it is required if any
8436are added that they be documented here.
8437
8438Some intrinsic functions can be overloaded, i.e., the intrinsic
8439represents a family of functions that perform the same operation but on
8440different data types. Because LLVM can represent over 8 million
8441different integer types, overloading is used commonly to allow an
8442intrinsic function to operate on any integer type. One or more of the
8443argument types or the result type can be overloaded to accept any
8444integer type. Argument types may also be defined as exactly matching a
8445previous argument's type or the result type. This allows an intrinsic
8446function which accepts multiple arguments, but needs all of them to be
8447of the same type, to only be overloaded with respect to a single
8448argument or the result.
8449
8450Overloaded intrinsics will have the names of its overloaded argument
8451types encoded into its function name, each preceded by a period. Only
8452those types which are overloaded result in a name suffix. Arguments
8453whose type is matched against another type do not. For example, the
8454``llvm.ctpop`` function can take an integer of any width and returns an
8455integer of exactly the same integer width. This leads to a family of
8456functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
8457``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
8458overloaded, and only one type suffix is required. Because the argument's
8459type is matched against the return type, it does not require its own
8460name suffix.
8461
8462To learn how to add an intrinsic function, please see the `Extending
8463LLVM Guide <ExtendingLLVM.html>`_.
8464
8465.. _int_varargs:
8466
8467Variable Argument Handling Intrinsics
8468-------------------------------------
8469
8470Variable argument support is defined in LLVM with the
8471:ref:`va_arg <i_va_arg>` instruction and these three intrinsic
8472functions. These functions are related to the similarly named macros
8473defined in the ``<stdarg.h>`` header file.
8474
8475All of these functions operate on arguments that use a target-specific
8476value type "``va_list``". The LLVM assembly language reference manual
8477does not define what this type is, so all transformations should be
8478prepared to handle these functions regardless of the type used.
8479
8480This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
8481variable argument handling intrinsic functions are used.
8482
8483.. code-block:: llvm
8484
Tim Northoverab60bb92014-11-02 01:21:51 +00008485 ; This struct is different for every platform. For most platforms,
8486 ; it is merely an i8*.
8487 %struct.va_list = type { i8* }
8488
8489 ; For Unix x86_64 platforms, va_list is the following struct:
8490 ; %struct.va_list = type { i32, i32, i8*, i8* }
8491
Sean Silvab084af42012-12-07 10:36:55 +00008492 define i32 @test(i32 %X, ...) {
8493 ; Initialize variable argument processing
Tim Northoverab60bb92014-11-02 01:21:51 +00008494 %ap = alloca %struct.va_list
8495 %ap2 = bitcast %struct.va_list* %ap to i8*
Sean Silvab084af42012-12-07 10:36:55 +00008496 call void @llvm.va_start(i8* %ap2)
8497
8498 ; Read a single integer argument
Tim Northoverab60bb92014-11-02 01:21:51 +00008499 %tmp = va_arg i8* %ap2, i32
Sean Silvab084af42012-12-07 10:36:55 +00008500
8501 ; Demonstrate usage of llvm.va_copy and llvm.va_end
8502 %aq = alloca i8*
8503 %aq2 = bitcast i8** %aq to i8*
8504 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
8505 call void @llvm.va_end(i8* %aq2)
8506
8507 ; Stop processing of arguments.
8508 call void @llvm.va_end(i8* %ap2)
8509 ret i32 %tmp
8510 }
8511
8512 declare void @llvm.va_start(i8*)
8513 declare void @llvm.va_copy(i8*, i8*)
8514 declare void @llvm.va_end(i8*)
8515
8516.. _int_va_start:
8517
8518'``llvm.va_start``' Intrinsic
8519^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8520
8521Syntax:
8522"""""""
8523
8524::
8525
Nick Lewycky04f6de02013-09-11 22:04:52 +00008526 declare void @llvm.va_start(i8* <arglist>)
Sean Silvab084af42012-12-07 10:36:55 +00008527
8528Overview:
8529"""""""""
8530
8531The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
8532subsequent use by ``va_arg``.
8533
8534Arguments:
8535""""""""""
8536
8537The argument is a pointer to a ``va_list`` element to initialize.
8538
8539Semantics:
8540""""""""""
8541
8542The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
8543available in C. In a target-dependent way, it initializes the
8544``va_list`` element to which the argument points, so that the next call
8545to ``va_arg`` will produce the first variable argument passed to the
8546function. Unlike the C ``va_start`` macro, this intrinsic does not need
8547to know the last argument of the function as the compiler can figure
8548that out.
8549
8550'``llvm.va_end``' Intrinsic
8551^^^^^^^^^^^^^^^^^^^^^^^^^^^
8552
8553Syntax:
8554"""""""
8555
8556::
8557
8558 declare void @llvm.va_end(i8* <arglist>)
8559
8560Overview:
8561"""""""""
8562
8563The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
8564initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
8565
8566Arguments:
8567""""""""""
8568
8569The argument is a pointer to a ``va_list`` to destroy.
8570
8571Semantics:
8572""""""""""
8573
8574The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
8575available in C. In a target-dependent way, it destroys the ``va_list``
8576element to which the argument points. Calls to
8577:ref:`llvm.va_start <int_va_start>` and
8578:ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
8579``llvm.va_end``.
8580
8581.. _int_va_copy:
8582
8583'``llvm.va_copy``' Intrinsic
8584^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8585
8586Syntax:
8587"""""""
8588
8589::
8590
8591 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
8592
8593Overview:
8594"""""""""
8595
8596The '``llvm.va_copy``' intrinsic copies the current argument position
8597from the source argument list to the destination argument list.
8598
8599Arguments:
8600""""""""""
8601
8602The first argument is a pointer to a ``va_list`` element to initialize.
8603The second argument is a pointer to a ``va_list`` element to copy from.
8604
8605Semantics:
8606""""""""""
8607
8608The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
8609available in C. In a target-dependent way, it copies the source
8610``va_list`` element into the destination ``va_list`` element. This
8611intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
8612arbitrarily complex and require, for example, memory allocation.
8613
8614Accurate Garbage Collection Intrinsics
8615--------------------------------------
8616
Philip Reamesc5b0f562015-02-25 23:52:06 +00008617LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
Mehdi Amini4a121fa2015-03-14 22:04:06 +00008618(GC) requires the frontend to generate code containing appropriate intrinsic
8619calls and select an appropriate GC strategy which knows how to lower these
Philip Reamesc5b0f562015-02-25 23:52:06 +00008620intrinsics in a manner which is appropriate for the target collector.
8621
Sean Silvab084af42012-12-07 10:36:55 +00008622These intrinsics allow identification of :ref:`GC roots on the
8623stack <int_gcroot>`, as well as garbage collector implementations that
8624require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
Philip Reamesc5b0f562015-02-25 23:52:06 +00008625Frontends for type-safe garbage collected languages should generate
Sean Silvab084af42012-12-07 10:36:55 +00008626these intrinsics to make use of the LLVM garbage collectors. For more
Philip Reamesf80bbff2015-02-25 23:45:20 +00008627details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
Sean Silvab084af42012-12-07 10:36:55 +00008628
Philip Reamesf80bbff2015-02-25 23:45:20 +00008629Experimental Statepoint Intrinsics
8630^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8631
8632LLVM provides an second experimental set of intrinsics for describing garbage
Mehdi Amini4a121fa2015-03-14 22:04:06 +00008633collection safepoints in compiled code. These intrinsics are an alternative
8634to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
8635:ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
8636differences in approach are covered in the `Garbage Collection with LLVM
8637<GarbageCollection.html>`_ documentation. The intrinsics themselves are
Philip Reamesf80bbff2015-02-25 23:45:20 +00008638described in :doc:`Statepoints`.
Sean Silvab084af42012-12-07 10:36:55 +00008639
8640.. _int_gcroot:
8641
8642'``llvm.gcroot``' Intrinsic
8643^^^^^^^^^^^^^^^^^^^^^^^^^^^
8644
8645Syntax:
8646"""""""
8647
8648::
8649
8650 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
8651
8652Overview:
8653"""""""""
8654
8655The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
8656the code generator, and allows some metadata to be associated with it.
8657
8658Arguments:
8659""""""""""
8660
8661The first argument specifies the address of a stack object that contains
8662the root pointer. The second pointer (which must be either a constant or
8663a global value address) contains the meta-data to be associated with the
8664root.
8665
8666Semantics:
8667""""""""""
8668
8669At runtime, a call to this intrinsic stores a null pointer into the
8670"ptrloc" location. At compile-time, the code generator generates
8671information to allow the runtime to find the pointer at GC safe points.
8672The '``llvm.gcroot``' intrinsic may only be used in a function which
8673:ref:`specifies a GC algorithm <gc>`.
8674
8675.. _int_gcread:
8676
8677'``llvm.gcread``' Intrinsic
8678^^^^^^^^^^^^^^^^^^^^^^^^^^^
8679
8680Syntax:
8681"""""""
8682
8683::
8684
8685 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
8686
8687Overview:
8688"""""""""
8689
8690The '``llvm.gcread``' intrinsic identifies reads of references from heap
8691locations, allowing garbage collector implementations that require read
8692barriers.
8693
8694Arguments:
8695""""""""""
8696
8697The second argument is the address to read from, which should be an
8698address allocated from the garbage collector. The first object is a
8699pointer to the start of the referenced object, if needed by the language
8700runtime (otherwise null).
8701
8702Semantics:
8703""""""""""
8704
8705The '``llvm.gcread``' intrinsic has the same semantics as a load
8706instruction, but may be replaced with substantially more complex code by
8707the garbage collector runtime, as needed. The '``llvm.gcread``'
8708intrinsic may only be used in a function which :ref:`specifies a GC
8709algorithm <gc>`.
8710
8711.. _int_gcwrite:
8712
8713'``llvm.gcwrite``' Intrinsic
8714^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8715
8716Syntax:
8717"""""""
8718
8719::
8720
8721 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
8722
8723Overview:
8724"""""""""
8725
8726The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
8727locations, allowing garbage collector implementations that require write
8728barriers (such as generational or reference counting collectors).
8729
8730Arguments:
8731""""""""""
8732
8733The first argument is the reference to store, the second is the start of
8734the object to store it to, and the third is the address of the field of
8735Obj to store to. If the runtime does not require a pointer to the
8736object, Obj may be null.
8737
8738Semantics:
8739""""""""""
8740
8741The '``llvm.gcwrite``' intrinsic has the same semantics as a store
8742instruction, but may be replaced with substantially more complex code by
8743the garbage collector runtime, as needed. The '``llvm.gcwrite``'
8744intrinsic may only be used in a function which :ref:`specifies a GC
8745algorithm <gc>`.
8746
8747Code Generator Intrinsics
8748-------------------------
8749
8750These intrinsics are provided by LLVM to expose special features that
8751may only be implemented with code generator support.
8752
8753'``llvm.returnaddress``' Intrinsic
8754^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8755
8756Syntax:
8757"""""""
8758
8759::
8760
8761 declare i8 *@llvm.returnaddress(i32 <level>)
8762
8763Overview:
8764"""""""""
8765
8766The '``llvm.returnaddress``' intrinsic attempts to compute a
8767target-specific value indicating the return address of the current
8768function or one of its callers.
8769
8770Arguments:
8771""""""""""
8772
8773The argument to this intrinsic indicates which function to return the
8774address for. Zero indicates the calling function, one indicates its
8775caller, etc. The argument is **required** to be a constant integer
8776value.
8777
8778Semantics:
8779""""""""""
8780
8781The '``llvm.returnaddress``' intrinsic either returns a pointer
8782indicating the return address of the specified call frame, or zero if it
8783cannot be identified. The value returned by this intrinsic is likely to
8784be incorrect or 0 for arguments other than zero, so it should only be
8785used for debugging purposes.
8786
8787Note that calling this intrinsic does not prevent function inlining or
8788other aggressive transformations, so the value returned may not be that
8789of the obvious source-language caller.
8790
8791'``llvm.frameaddress``' Intrinsic
8792^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8793
8794Syntax:
8795"""""""
8796
8797::
8798
8799 declare i8* @llvm.frameaddress(i32 <level>)
8800
8801Overview:
8802"""""""""
8803
8804The '``llvm.frameaddress``' intrinsic attempts to return the
8805target-specific frame pointer value for the specified stack frame.
8806
8807Arguments:
8808""""""""""
8809
8810The argument to this intrinsic indicates which function to return the
8811frame pointer for. Zero indicates the calling function, one indicates
8812its caller, etc. The argument is **required** to be a constant integer
8813value.
8814
8815Semantics:
8816""""""""""
8817
8818The '``llvm.frameaddress``' intrinsic either returns a pointer
8819indicating the frame address of the specified call frame, or zero if it
8820cannot be identified. The value returned by this intrinsic is likely to
8821be incorrect or 0 for arguments other than zero, so it should only be
8822used for debugging purposes.
8823
8824Note that calling this intrinsic does not prevent function inlining or
8825other aggressive transformations, so the value returned may not be that
8826of the obvious source-language caller.
8827
Reid Kleckner60381792015-07-07 22:25:32 +00008828'``llvm.localescape``' and '``llvm.localrecover``' Intrinsics
Reid Klecknere9b89312015-01-13 00:48:10 +00008829^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8830
8831Syntax:
8832"""""""
8833
8834::
8835
Reid Kleckner60381792015-07-07 22:25:32 +00008836 declare void @llvm.localescape(...)
8837 declare i8* @llvm.localrecover(i8* %func, i8* %fp, i32 %idx)
Reid Klecknere9b89312015-01-13 00:48:10 +00008838
8839Overview:
8840"""""""""
8841
Reid Kleckner60381792015-07-07 22:25:32 +00008842The '``llvm.localescape``' intrinsic escapes offsets of a collection of static
8843allocas, and the '``llvm.localrecover``' intrinsic applies those offsets to a
Reid Klecknercfb9ce52015-03-05 18:26:34 +00008844live frame pointer to recover the address of the allocation. The offset is
Reid Kleckner60381792015-07-07 22:25:32 +00008845computed during frame layout of the caller of ``llvm.localescape``.
Reid Klecknere9b89312015-01-13 00:48:10 +00008846
8847Arguments:
8848""""""""""
8849
Reid Kleckner60381792015-07-07 22:25:32 +00008850All arguments to '``llvm.localescape``' must be pointers to static allocas or
8851casts of static allocas. Each function can only call '``llvm.localescape``'
Reid Klecknercfb9ce52015-03-05 18:26:34 +00008852once, and it can only do so from the entry block.
Reid Klecknere9b89312015-01-13 00:48:10 +00008853
Reid Kleckner60381792015-07-07 22:25:32 +00008854The ``func`` argument to '``llvm.localrecover``' must be a constant
Reid Klecknere9b89312015-01-13 00:48:10 +00008855bitcasted pointer to a function defined in the current module. The code
8856generator cannot determine the frame allocation offset of functions defined in
8857other modules.
8858
Reid Klecknerd5afc62f2015-07-07 23:23:03 +00008859The ``fp`` argument to '``llvm.localrecover``' must be a frame pointer of a
8860call frame that is currently live. The return value of '``llvm.localaddress``'
8861is one way to produce such a value, but various runtimes also expose a suitable
8862pointer in platform-specific ways.
Reid Klecknere9b89312015-01-13 00:48:10 +00008863
Reid Kleckner60381792015-07-07 22:25:32 +00008864The ``idx`` argument to '``llvm.localrecover``' indicates which alloca passed to
8865'``llvm.localescape``' to recover. It is zero-indexed.
Reid Klecknercfb9ce52015-03-05 18:26:34 +00008866
Reid Klecknere9b89312015-01-13 00:48:10 +00008867Semantics:
8868""""""""""
8869
Reid Kleckner60381792015-07-07 22:25:32 +00008870These intrinsics allow a group of functions to share access to a set of local
8871stack allocations of a one parent function. The parent function may call the
8872'``llvm.localescape``' intrinsic once from the function entry block, and the
8873child functions can use '``llvm.localrecover``' to access the escaped allocas.
8874The '``llvm.localescape``' intrinsic blocks inlining, as inlining changes where
8875the escaped allocas are allocated, which would break attempts to use
8876'``llvm.localrecover``'.
Reid Klecknere9b89312015-01-13 00:48:10 +00008877
Renato Golinc7aea402014-05-06 16:51:25 +00008878.. _int_read_register:
8879.. _int_write_register:
8880
8881'``llvm.read_register``' and '``llvm.write_register``' Intrinsics
8882^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8883
8884Syntax:
8885"""""""
8886
8887::
8888
8889 declare i32 @llvm.read_register.i32(metadata)
8890 declare i64 @llvm.read_register.i64(metadata)
8891 declare void @llvm.write_register.i32(metadata, i32 @value)
8892 declare void @llvm.write_register.i64(metadata, i64 @value)
Duncan P. N. Exon Smithbe7ea192014-12-15 19:07:53 +00008893 !0 = !{!"sp\00"}
Renato Golinc7aea402014-05-06 16:51:25 +00008894
8895Overview:
8896"""""""""
8897
8898The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
8899provides access to the named register. The register must be valid on
8900the architecture being compiled to. The type needs to be compatible
8901with the register being read.
8902
8903Semantics:
8904""""""""""
8905
8906The '``llvm.read_register``' intrinsic returns the current value of the
8907register, where possible. The '``llvm.write_register``' intrinsic sets
8908the current value of the register, where possible.
8909
8910This is useful to implement named register global variables that need
8911to always be mapped to a specific register, as is common practice on
8912bare-metal programs including OS kernels.
8913
8914The compiler doesn't check for register availability or use of the used
8915register in surrounding code, including inline assembly. Because of that,
8916allocatable registers are not supported.
8917
8918Warning: So far it only works with the stack pointer on selected
Tim Northover3b0846e2014-05-24 12:50:23 +00008919architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
Renato Golinc7aea402014-05-06 16:51:25 +00008920work is needed to support other registers and even more so, allocatable
8921registers.
8922
Sean Silvab084af42012-12-07 10:36:55 +00008923.. _int_stacksave:
8924
8925'``llvm.stacksave``' Intrinsic
8926^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8927
8928Syntax:
8929"""""""
8930
8931::
8932
8933 declare i8* @llvm.stacksave()
8934
8935Overview:
8936"""""""""
8937
8938The '``llvm.stacksave``' intrinsic is used to remember the current state
8939of the function stack, for use with
8940:ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
8941implementing language features like scoped automatic variable sized
8942arrays in C99.
8943
8944Semantics:
8945""""""""""
8946
8947This intrinsic returns a opaque pointer value that can be passed to
8948:ref:`llvm.stackrestore <int_stackrestore>`. When an
8949``llvm.stackrestore`` intrinsic is executed with a value saved from
8950``llvm.stacksave``, it effectively restores the state of the stack to
8951the state it was in when the ``llvm.stacksave`` intrinsic executed. In
8952practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
8953were allocated after the ``llvm.stacksave`` was executed.
8954
8955.. _int_stackrestore:
8956
8957'``llvm.stackrestore``' Intrinsic
8958^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8959
8960Syntax:
8961"""""""
8962
8963::
8964
8965 declare void @llvm.stackrestore(i8* %ptr)
8966
8967Overview:
8968"""""""""
8969
8970The '``llvm.stackrestore``' intrinsic is used to restore the state of
8971the function stack to the state it was in when the corresponding
8972:ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
8973useful for implementing language features like scoped automatic variable
8974sized arrays in C99.
8975
8976Semantics:
8977""""""""""
8978
8979See the description for :ref:`llvm.stacksave <int_stacksave>`.
8980
8981'``llvm.prefetch``' Intrinsic
8982^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8983
8984Syntax:
8985"""""""
8986
8987::
8988
8989 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
8990
8991Overview:
8992"""""""""
8993
8994The '``llvm.prefetch``' intrinsic is a hint to the code generator to
8995insert a prefetch instruction if supported; otherwise, it is a noop.
8996Prefetches have no effect on the behavior of the program but can change
8997its performance characteristics.
8998
8999Arguments:
9000""""""""""
9001
9002``address`` is the address to be prefetched, ``rw`` is the specifier
9003determining if the fetch should be for a read (0) or write (1), and
9004``locality`` is a temporal locality specifier ranging from (0) - no
9005locality, to (3) - extremely local keep in cache. The ``cache type``
9006specifies whether the prefetch is performed on the data (1) or
9007instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
9008arguments must be constant integers.
9009
9010Semantics:
9011""""""""""
9012
9013This intrinsic does not modify the behavior of the program. In
9014particular, prefetches cannot trap and do not produce a value. On
9015targets that support this intrinsic, the prefetch can provide hints to
9016the processor cache for better performance.
9017
9018'``llvm.pcmarker``' Intrinsic
9019^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9020
9021Syntax:
9022"""""""
9023
9024::
9025
9026 declare void @llvm.pcmarker(i32 <id>)
9027
9028Overview:
9029"""""""""
9030
9031The '``llvm.pcmarker``' intrinsic is a method to export a Program
9032Counter (PC) in a region of code to simulators and other tools. The
9033method is target specific, but it is expected that the marker will use
9034exported symbols to transmit the PC of the marker. The marker makes no
9035guarantees that it will remain with any specific instruction after
9036optimizations. It is possible that the presence of a marker will inhibit
9037optimizations. The intended use is to be inserted after optimizations to
9038allow correlations of simulation runs.
9039
9040Arguments:
9041""""""""""
9042
9043``id`` is a numerical id identifying the marker.
9044
9045Semantics:
9046""""""""""
9047
9048This intrinsic does not modify the behavior of the program. Backends
9049that do not support this intrinsic may ignore it.
9050
9051'``llvm.readcyclecounter``' Intrinsic
9052^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9053
9054Syntax:
9055"""""""
9056
9057::
9058
9059 declare i64 @llvm.readcyclecounter()
9060
9061Overview:
9062"""""""""
9063
9064The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
9065counter register (or similar low latency, high accuracy clocks) on those
9066targets that support it. On X86, it should map to RDTSC. On Alpha, it
9067should map to RPCC. As the backing counters overflow quickly (on the
9068order of 9 seconds on alpha), this should only be used for small
9069timings.
9070
9071Semantics:
9072""""""""""
9073
9074When directly supported, reading the cycle counter should not modify any
9075memory. Implementations are allowed to either return a application
9076specific value or a system wide value. On backends without support, this
9077is lowered to a constant 0.
9078
Tim Northoverbc933082013-05-23 19:11:20 +00009079Note that runtime support may be conditional on the privilege-level code is
9080running at and the host platform.
9081
Renato Golinc0a3c1d2014-03-26 12:52:28 +00009082'``llvm.clear_cache``' Intrinsic
9083^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9084
9085Syntax:
9086"""""""
9087
9088::
9089
9090 declare void @llvm.clear_cache(i8*, i8*)
9091
9092Overview:
9093"""""""""
9094
Joerg Sonnenberger03014d62014-03-26 14:35:21 +00009095The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
9096in the specified range to the execution unit of the processor. On
9097targets with non-unified instruction and data cache, the implementation
9098flushes the instruction cache.
Renato Golinc0a3c1d2014-03-26 12:52:28 +00009099
9100Semantics:
9101""""""""""
9102
Joerg Sonnenberger03014d62014-03-26 14:35:21 +00009103On platforms with coherent instruction and data caches (e.g. x86), this
9104intrinsic is a nop. On platforms with non-coherent instruction and data
Alp Toker16f98b22014-04-09 14:47:27 +00009105cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
Joerg Sonnenberger03014d62014-03-26 14:35:21 +00009106instructions or a system call, if cache flushing requires special
9107privileges.
Renato Golinc0a3c1d2014-03-26 12:52:28 +00009108
Sean Silvad02bf3e2014-04-07 22:29:53 +00009109The default behavior is to emit a call to ``__clear_cache`` from the run
Joerg Sonnenberger03014d62014-03-26 14:35:21 +00009110time library.
Renato Golin93010e62014-03-26 14:01:32 +00009111
Joerg Sonnenberger03014d62014-03-26 14:35:21 +00009112This instrinsic does *not* empty the instruction pipeline. Modifications
9113of the current function are outside the scope of the intrinsic.
Renato Golinc0a3c1d2014-03-26 12:52:28 +00009114
Justin Bogner61ba2e32014-12-08 18:02:35 +00009115'``llvm.instrprof_increment``' Intrinsic
9116^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9117
9118Syntax:
9119"""""""
9120
9121::
9122
9123 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
9124 i32 <num-counters>, i32 <index>)
9125
9126Overview:
9127"""""""""
9128
9129The '``llvm.instrprof_increment``' intrinsic can be emitted by a
9130frontend for use with instrumentation based profiling. These will be
9131lowered by the ``-instrprof`` pass to generate execution counts of a
9132program at runtime.
9133
9134Arguments:
9135""""""""""
9136
9137The first argument is a pointer to a global variable containing the
9138name of the entity being instrumented. This should generally be the
9139(mangled) function name for a set of counters.
9140
9141The second argument is a hash value that can be used by the consumer
9142of the profile data to detect changes to the instrumented source, and
9143the third is the number of counters associated with ``name``. It is an
9144error if ``hash`` or ``num-counters`` differ between two instances of
9145``instrprof_increment`` that refer to the same name.
9146
9147The last argument refers to which of the counters for ``name`` should
9148be incremented. It should be a value between 0 and ``num-counters``.
9149
9150Semantics:
9151""""""""""
9152
9153This intrinsic represents an increment of a profiling counter. It will
9154cause the ``-instrprof`` pass to generate the appropriate data
9155structures and the code to increment the appropriate value, in a
9156format that can be written out by a compiler runtime and consumed via
9157the ``llvm-profdata`` tool.
9158
Sean Silvab084af42012-12-07 10:36:55 +00009159Standard C Library Intrinsics
9160-----------------------------
9161
9162LLVM provides intrinsics for a few important standard C library
9163functions. These intrinsics allow source-language front-ends to pass
9164information about the alignment of the pointer arguments to the code
9165generator, providing opportunity for more efficient code generation.
9166
9167.. _int_memcpy:
9168
9169'``llvm.memcpy``' Intrinsic
9170^^^^^^^^^^^^^^^^^^^^^^^^^^^
9171
9172Syntax:
9173"""""""
9174
9175This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
9176integer bit width and for different address spaces. Not all targets
9177support all bit widths however.
9178
9179::
9180
9181 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9182 i32 <len>, i32 <align>, i1 <isvolatile>)
9183 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9184 i64 <len>, i32 <align>, i1 <isvolatile>)
9185
9186Overview:
9187"""""""""
9188
9189The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9190source location to the destination location.
9191
9192Note that, unlike the standard libc function, the ``llvm.memcpy.*``
9193intrinsics do not return a value, takes extra alignment/isvolatile
9194arguments and the pointers can be in specified address spaces.
9195
9196Arguments:
9197""""""""""
9198
9199The first argument is a pointer to the destination, the second is a
9200pointer to the source. The third argument is an integer argument
9201specifying the number of bytes to copy, the fourth argument is the
9202alignment of the source and destination locations, and the fifth is a
9203boolean indicating a volatile access.
9204
9205If the call to this intrinsic has an alignment value that is not 0 or 1,
9206then the caller guarantees that both the source and destination pointers
9207are aligned to that boundary.
9208
9209If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
9210a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9211very cleanly specified and it is unwise to depend on it.
9212
9213Semantics:
9214""""""""""
9215
9216The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
9217source location to the destination location, which are not allowed to
9218overlap. It copies "len" bytes of memory over. If the argument is known
9219to be aligned to some boundary, this can be specified as the fourth
Bill Wendling61163152013-10-18 23:26:55 +00009220argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
Sean Silvab084af42012-12-07 10:36:55 +00009221
9222'``llvm.memmove``' Intrinsic
9223^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9224
9225Syntax:
9226"""""""
9227
9228This is an overloaded intrinsic. You can use llvm.memmove on any integer
9229bit width and for different address space. Not all targets support all
9230bit widths however.
9231
9232::
9233
9234 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
9235 i32 <len>, i32 <align>, i1 <isvolatile>)
9236 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
9237 i64 <len>, i32 <align>, i1 <isvolatile>)
9238
9239Overview:
9240"""""""""
9241
9242The '``llvm.memmove.*``' intrinsics move a block of memory from the
9243source location to the destination location. It is similar to the
9244'``llvm.memcpy``' intrinsic but allows the two memory locations to
9245overlap.
9246
9247Note that, unlike the standard libc function, the ``llvm.memmove.*``
9248intrinsics do not return a value, takes extra alignment/isvolatile
9249arguments and the pointers can be in specified address spaces.
9250
9251Arguments:
9252""""""""""
9253
9254The first argument is a pointer to the destination, the second is a
9255pointer to the source. The third argument is an integer argument
9256specifying the number of bytes to copy, the fourth argument is the
9257alignment of the source and destination locations, and the fifth is a
9258boolean indicating a volatile access.
9259
9260If the call to this intrinsic has an alignment value that is not 0 or 1,
9261then the caller guarantees that the source and destination pointers are
9262aligned to that boundary.
9263
9264If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
9265is a :ref:`volatile operation <volatile>`. The detailed access behavior is
9266not very cleanly specified and it is unwise to depend on it.
9267
9268Semantics:
9269""""""""""
9270
9271The '``llvm.memmove.*``' intrinsics copy a block of memory from the
9272source location to the destination location, which may overlap. It
9273copies "len" bytes of memory over. If the argument is known to be
9274aligned to some boundary, this can be specified as the fourth argument,
Bill Wendling61163152013-10-18 23:26:55 +00009275otherwise it should be set to 0 or 1 (both meaning no alignment).
Sean Silvab084af42012-12-07 10:36:55 +00009276
9277'``llvm.memset.*``' Intrinsics
9278^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9279
9280Syntax:
9281"""""""
9282
9283This is an overloaded intrinsic. You can use llvm.memset on any integer
9284bit width and for different address spaces. However, not all targets
9285support all bit widths.
9286
9287::
9288
9289 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
9290 i32 <len>, i32 <align>, i1 <isvolatile>)
9291 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
9292 i64 <len>, i32 <align>, i1 <isvolatile>)
9293
9294Overview:
9295"""""""""
9296
9297The '``llvm.memset.*``' intrinsics fill a block of memory with a
9298particular byte value.
9299
9300Note that, unlike the standard libc function, the ``llvm.memset``
9301intrinsic does not return a value and takes extra alignment/volatile
9302arguments. Also, the destination can be in an arbitrary address space.
9303
9304Arguments:
9305""""""""""
9306
9307The first argument is a pointer to the destination to fill, the second
9308is the byte value with which to fill it, the third argument is an
9309integer argument specifying the number of bytes to fill, and the fourth
9310argument is the known alignment of the destination location.
9311
9312If the call to this intrinsic has an alignment value that is not 0 or 1,
9313then the caller guarantees that the destination pointer is aligned to
9314that boundary.
9315
9316If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
9317a :ref:`volatile operation <volatile>`. The detailed access behavior is not
9318very cleanly specified and it is unwise to depend on it.
9319
9320Semantics:
9321""""""""""
9322
9323The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
9324at the destination location. If the argument is known to be aligned to
9325some boundary, this can be specified as the fourth argument, otherwise
Bill Wendling61163152013-10-18 23:26:55 +00009326it should be set to 0 or 1 (both meaning no alignment).
Sean Silvab084af42012-12-07 10:36:55 +00009327
9328'``llvm.sqrt.*``' Intrinsic
9329^^^^^^^^^^^^^^^^^^^^^^^^^^^
9330
9331Syntax:
9332"""""""
9333
9334This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
9335floating point or vector of floating point type. Not all targets support
9336all types however.
9337
9338::
9339
9340 declare float @llvm.sqrt.f32(float %Val)
9341 declare double @llvm.sqrt.f64(double %Val)
9342 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
9343 declare fp128 @llvm.sqrt.f128(fp128 %Val)
9344 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
9345
9346Overview:
9347"""""""""
9348
9349The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
9350returning the same value as the libm '``sqrt``' functions would. Unlike
9351``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
9352negative numbers other than -0.0 (which allows for better optimization,
9353because there is no need to worry about errno being set).
9354``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
9355
9356Arguments:
9357""""""""""
9358
9359The argument and return value are floating point numbers of the same
9360type.
9361
9362Semantics:
9363""""""""""
9364
9365This function returns the sqrt of the specified operand if it is a
9366nonnegative floating point number.
9367
9368'``llvm.powi.*``' Intrinsic
9369^^^^^^^^^^^^^^^^^^^^^^^^^^^
9370
9371Syntax:
9372"""""""
9373
9374This is an overloaded intrinsic. You can use ``llvm.powi`` on any
9375floating point or vector of floating point type. Not all targets support
9376all types however.
9377
9378::
9379
9380 declare float @llvm.powi.f32(float %Val, i32 %power)
9381 declare double @llvm.powi.f64(double %Val, i32 %power)
9382 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
9383 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
9384 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
9385
9386Overview:
9387"""""""""
9388
9389The '``llvm.powi.*``' intrinsics return the first operand raised to the
9390specified (positive or negative) power. The order of evaluation of
9391multiplications is not defined. When a vector of floating point type is
9392used, the second argument remains a scalar integer value.
9393
9394Arguments:
9395""""""""""
9396
9397The second argument is an integer power, and the first is a value to
9398raise to that power.
9399
9400Semantics:
9401""""""""""
9402
9403This function returns the first value raised to the second power with an
9404unspecified sequence of rounding operations.
9405
9406'``llvm.sin.*``' Intrinsic
9407^^^^^^^^^^^^^^^^^^^^^^^^^^
9408
9409Syntax:
9410"""""""
9411
9412This is an overloaded intrinsic. You can use ``llvm.sin`` on any
9413floating point or vector of floating point type. Not all targets support
9414all types however.
9415
9416::
9417
9418 declare float @llvm.sin.f32(float %Val)
9419 declare double @llvm.sin.f64(double %Val)
9420 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
9421 declare fp128 @llvm.sin.f128(fp128 %Val)
9422 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
9423
9424Overview:
9425"""""""""
9426
9427The '``llvm.sin.*``' intrinsics return the sine of the operand.
9428
9429Arguments:
9430""""""""""
9431
9432The argument and return value are floating point numbers of the same
9433type.
9434
9435Semantics:
9436""""""""""
9437
9438This function returns the sine of the specified operand, returning the
9439same values as the libm ``sin`` functions would, and handles error
9440conditions in the same way.
9441
9442'``llvm.cos.*``' Intrinsic
9443^^^^^^^^^^^^^^^^^^^^^^^^^^
9444
9445Syntax:
9446"""""""
9447
9448This is an overloaded intrinsic. You can use ``llvm.cos`` on any
9449floating point or vector of floating point type. Not all targets support
9450all types however.
9451
9452::
9453
9454 declare float @llvm.cos.f32(float %Val)
9455 declare double @llvm.cos.f64(double %Val)
9456 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
9457 declare fp128 @llvm.cos.f128(fp128 %Val)
9458 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
9459
9460Overview:
9461"""""""""
9462
9463The '``llvm.cos.*``' intrinsics return the cosine of the operand.
9464
9465Arguments:
9466""""""""""
9467
9468The argument and return value are floating point numbers of the same
9469type.
9470
9471Semantics:
9472""""""""""
9473
9474This function returns the cosine of the specified operand, returning the
9475same values as the libm ``cos`` functions would, and handles error
9476conditions in the same way.
9477
9478'``llvm.pow.*``' Intrinsic
9479^^^^^^^^^^^^^^^^^^^^^^^^^^
9480
9481Syntax:
9482"""""""
9483
9484This is an overloaded intrinsic. You can use ``llvm.pow`` on any
9485floating point or vector of floating point type. Not all targets support
9486all types however.
9487
9488::
9489
9490 declare float @llvm.pow.f32(float %Val, float %Power)
9491 declare double @llvm.pow.f64(double %Val, double %Power)
9492 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
9493 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
9494 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
9495
9496Overview:
9497"""""""""
9498
9499The '``llvm.pow.*``' intrinsics return the first operand raised to the
9500specified (positive or negative) power.
9501
9502Arguments:
9503""""""""""
9504
9505The second argument is a floating point power, and the first is a value
9506to raise to that power.
9507
9508Semantics:
9509""""""""""
9510
9511This function returns the first value raised to the second power,
9512returning the same values as the libm ``pow`` functions would, and
9513handles error conditions in the same way.
9514
9515'``llvm.exp.*``' Intrinsic
9516^^^^^^^^^^^^^^^^^^^^^^^^^^
9517
9518Syntax:
9519"""""""
9520
9521This is an overloaded intrinsic. You can use ``llvm.exp`` on any
9522floating point or vector of floating point type. Not all targets support
9523all types however.
9524
9525::
9526
9527 declare float @llvm.exp.f32(float %Val)
9528 declare double @llvm.exp.f64(double %Val)
9529 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
9530 declare fp128 @llvm.exp.f128(fp128 %Val)
9531 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
9532
9533Overview:
9534"""""""""
9535
9536The '``llvm.exp.*``' intrinsics perform the exp function.
9537
9538Arguments:
9539""""""""""
9540
9541The argument and return value are floating point numbers of the same
9542type.
9543
9544Semantics:
9545""""""""""
9546
9547This function returns the same values as the libm ``exp`` functions
9548would, and handles error conditions in the same way.
9549
9550'``llvm.exp2.*``' Intrinsic
9551^^^^^^^^^^^^^^^^^^^^^^^^^^^
9552
9553Syntax:
9554"""""""
9555
9556This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
9557floating point or vector of floating point type. Not all targets support
9558all types however.
9559
9560::
9561
9562 declare float @llvm.exp2.f32(float %Val)
9563 declare double @llvm.exp2.f64(double %Val)
9564 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
9565 declare fp128 @llvm.exp2.f128(fp128 %Val)
9566 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
9567
9568Overview:
9569"""""""""
9570
9571The '``llvm.exp2.*``' intrinsics perform the exp2 function.
9572
9573Arguments:
9574""""""""""
9575
9576The argument and return value are floating point numbers of the same
9577type.
9578
9579Semantics:
9580""""""""""
9581
9582This function returns the same values as the libm ``exp2`` functions
9583would, and handles error conditions in the same way.
9584
9585'``llvm.log.*``' Intrinsic
9586^^^^^^^^^^^^^^^^^^^^^^^^^^
9587
9588Syntax:
9589"""""""
9590
9591This is an overloaded intrinsic. You can use ``llvm.log`` on any
9592floating point or vector of floating point type. Not all targets support
9593all types however.
9594
9595::
9596
9597 declare float @llvm.log.f32(float %Val)
9598 declare double @llvm.log.f64(double %Val)
9599 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
9600 declare fp128 @llvm.log.f128(fp128 %Val)
9601 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
9602
9603Overview:
9604"""""""""
9605
9606The '``llvm.log.*``' intrinsics perform the log function.
9607
9608Arguments:
9609""""""""""
9610
9611The argument and return value are floating point numbers of the same
9612type.
9613
9614Semantics:
9615""""""""""
9616
9617This function returns the same values as the libm ``log`` functions
9618would, and handles error conditions in the same way.
9619
9620'``llvm.log10.*``' Intrinsic
9621^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9622
9623Syntax:
9624"""""""
9625
9626This is an overloaded intrinsic. You can use ``llvm.log10`` on any
9627floating point or vector of floating point type. Not all targets support
9628all types however.
9629
9630::
9631
9632 declare float @llvm.log10.f32(float %Val)
9633 declare double @llvm.log10.f64(double %Val)
9634 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
9635 declare fp128 @llvm.log10.f128(fp128 %Val)
9636 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
9637
9638Overview:
9639"""""""""
9640
9641The '``llvm.log10.*``' intrinsics perform the log10 function.
9642
9643Arguments:
9644""""""""""
9645
9646The argument and return value are floating point numbers of the same
9647type.
9648
9649Semantics:
9650""""""""""
9651
9652This function returns the same values as the libm ``log10`` functions
9653would, and handles error conditions in the same way.
9654
9655'``llvm.log2.*``' Intrinsic
9656^^^^^^^^^^^^^^^^^^^^^^^^^^^
9657
9658Syntax:
9659"""""""
9660
9661This is an overloaded intrinsic. You can use ``llvm.log2`` on any
9662floating point or vector of floating point type. Not all targets support
9663all types however.
9664
9665::
9666
9667 declare float @llvm.log2.f32(float %Val)
9668 declare double @llvm.log2.f64(double %Val)
9669 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
9670 declare fp128 @llvm.log2.f128(fp128 %Val)
9671 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
9672
9673Overview:
9674"""""""""
9675
9676The '``llvm.log2.*``' intrinsics perform the log2 function.
9677
9678Arguments:
9679""""""""""
9680
9681The argument and return value are floating point numbers of the same
9682type.
9683
9684Semantics:
9685""""""""""
9686
9687This function returns the same values as the libm ``log2`` functions
9688would, and handles error conditions in the same way.
9689
9690'``llvm.fma.*``' Intrinsic
9691^^^^^^^^^^^^^^^^^^^^^^^^^^
9692
9693Syntax:
9694"""""""
9695
9696This is an overloaded intrinsic. You can use ``llvm.fma`` on any
9697floating point or vector of floating point type. Not all targets support
9698all types however.
9699
9700::
9701
9702 declare float @llvm.fma.f32(float %a, float %b, float %c)
9703 declare double @llvm.fma.f64(double %a, double %b, double %c)
9704 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
9705 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
9706 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
9707
9708Overview:
9709"""""""""
9710
9711The '``llvm.fma.*``' intrinsics perform the fused multiply-add
9712operation.
9713
9714Arguments:
9715""""""""""
9716
9717The argument and return value are floating point numbers of the same
9718type.
9719
9720Semantics:
9721""""""""""
9722
9723This function returns the same values as the libm ``fma`` functions
Matt Arsenaultee364ee2014-01-31 00:09:00 +00009724would, and does not set errno.
Sean Silvab084af42012-12-07 10:36:55 +00009725
9726'``llvm.fabs.*``' Intrinsic
9727^^^^^^^^^^^^^^^^^^^^^^^^^^^
9728
9729Syntax:
9730"""""""
9731
9732This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
9733floating point or vector of floating point type. Not all targets support
9734all types however.
9735
9736::
9737
9738 declare float @llvm.fabs.f32(float %Val)
9739 declare double @llvm.fabs.f64(double %Val)
Matt Arsenaultd6511b42014-10-21 23:00:20 +00009740 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
Sean Silvab084af42012-12-07 10:36:55 +00009741 declare fp128 @llvm.fabs.f128(fp128 %Val)
Matt Arsenaultd6511b42014-10-21 23:00:20 +00009742 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
Sean Silvab084af42012-12-07 10:36:55 +00009743
9744Overview:
9745"""""""""
9746
9747The '``llvm.fabs.*``' intrinsics return the absolute value of the
9748operand.
9749
9750Arguments:
9751""""""""""
9752
9753The argument and return value are floating point numbers of the same
9754type.
9755
9756Semantics:
9757""""""""""
9758
9759This function returns the same values as the libm ``fabs`` functions
9760would, and handles error conditions in the same way.
9761
Matt Arsenaultd6511b42014-10-21 23:00:20 +00009762'``llvm.minnum.*``' Intrinsic
Matt Arsenault9886b0d2014-10-22 00:15:53 +00009763^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Matt Arsenaultd6511b42014-10-21 23:00:20 +00009764
9765Syntax:
9766"""""""
9767
9768This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
9769floating point or vector of floating point type. Not all targets support
9770all types however.
9771
9772::
9773
Matt Arsenault64313c92014-10-22 18:25:02 +00009774 declare float @llvm.minnum.f32(float %Val0, float %Val1)
9775 declare double @llvm.minnum.f64(double %Val0, double %Val1)
9776 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
9777 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
9778 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
Matt Arsenaultd6511b42014-10-21 23:00:20 +00009779
9780Overview:
9781"""""""""
9782
9783The '``llvm.minnum.*``' intrinsics return the minimum of the two
9784arguments.
9785
9786
9787Arguments:
9788""""""""""
9789
9790The arguments and return value are floating point numbers of the same
9791type.
9792
9793Semantics:
9794""""""""""
9795
9796Follows the IEEE-754 semantics for minNum, which also match for libm's
9797fmin.
9798
9799If either operand is a NaN, returns the other non-NaN operand. Returns
9800NaN only if both operands are NaN. If the operands compare equal,
9801returns a value that compares equal to both operands. This means that
9802fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
9803
9804'``llvm.maxnum.*``' Intrinsic
Matt Arsenault9886b0d2014-10-22 00:15:53 +00009805^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Matt Arsenaultd6511b42014-10-21 23:00:20 +00009806
9807Syntax:
9808"""""""
9809
9810This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
9811floating point or vector of floating point type. Not all targets support
9812all types however.
9813
9814::
9815
Matt Arsenault64313c92014-10-22 18:25:02 +00009816 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
9817 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
9818 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
9819 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
9820 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
Matt Arsenaultd6511b42014-10-21 23:00:20 +00009821
9822Overview:
9823"""""""""
9824
9825The '``llvm.maxnum.*``' intrinsics return the maximum of the two
9826arguments.
9827
9828
9829Arguments:
9830""""""""""
9831
9832The arguments and return value are floating point numbers of the same
9833type.
9834
9835Semantics:
9836""""""""""
9837Follows the IEEE-754 semantics for maxNum, which also match for libm's
9838fmax.
9839
9840If either operand is a NaN, returns the other non-NaN operand. Returns
9841NaN only if both operands are NaN. If the operands compare equal,
9842returns a value that compares equal to both operands. This means that
9843fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
9844
Hal Finkel0c5c01aa2013-08-19 23:35:46 +00009845'``llvm.copysign.*``' Intrinsic
9846^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9847
9848Syntax:
9849"""""""
9850
9851This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
9852floating point or vector of floating point type. Not all targets support
9853all types however.
9854
9855::
9856
9857 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
9858 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
9859 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
9860 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
9861 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
9862
9863Overview:
9864"""""""""
9865
9866The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
9867first operand and the sign of the second operand.
9868
9869Arguments:
9870""""""""""
9871
9872The arguments and return value are floating point numbers of the same
9873type.
9874
9875Semantics:
9876""""""""""
9877
9878This function returns the same values as the libm ``copysign``
9879functions would, and handles error conditions in the same way.
9880
Sean Silvab084af42012-12-07 10:36:55 +00009881'``llvm.floor.*``' Intrinsic
9882^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9883
9884Syntax:
9885"""""""
9886
9887This is an overloaded intrinsic. You can use ``llvm.floor`` on any
9888floating point or vector of floating point type. Not all targets support
9889all types however.
9890
9891::
9892
9893 declare float @llvm.floor.f32(float %Val)
9894 declare double @llvm.floor.f64(double %Val)
9895 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
9896 declare fp128 @llvm.floor.f128(fp128 %Val)
9897 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
9898
9899Overview:
9900"""""""""
9901
9902The '``llvm.floor.*``' intrinsics return the floor of the operand.
9903
9904Arguments:
9905""""""""""
9906
9907The argument and return value are floating point numbers of the same
9908type.
9909
9910Semantics:
9911""""""""""
9912
9913This function returns the same values as the libm ``floor`` functions
9914would, and handles error conditions in the same way.
9915
9916'``llvm.ceil.*``' Intrinsic
9917^^^^^^^^^^^^^^^^^^^^^^^^^^^
9918
9919Syntax:
9920"""""""
9921
9922This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
9923floating point or vector of floating point type. Not all targets support
9924all types however.
9925
9926::
9927
9928 declare float @llvm.ceil.f32(float %Val)
9929 declare double @llvm.ceil.f64(double %Val)
9930 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
9931 declare fp128 @llvm.ceil.f128(fp128 %Val)
9932 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
9933
9934Overview:
9935"""""""""
9936
9937The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
9938
9939Arguments:
9940""""""""""
9941
9942The argument and return value are floating point numbers of the same
9943type.
9944
9945Semantics:
9946""""""""""
9947
9948This function returns the same values as the libm ``ceil`` functions
9949would, and handles error conditions in the same way.
9950
9951'``llvm.trunc.*``' Intrinsic
9952^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9953
9954Syntax:
9955"""""""
9956
9957This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
9958floating point or vector of floating point type. Not all targets support
9959all types however.
9960
9961::
9962
9963 declare float @llvm.trunc.f32(float %Val)
9964 declare double @llvm.trunc.f64(double %Val)
9965 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
9966 declare fp128 @llvm.trunc.f128(fp128 %Val)
9967 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
9968
9969Overview:
9970"""""""""
9971
9972The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
9973nearest integer not larger in magnitude than the operand.
9974
9975Arguments:
9976""""""""""
9977
9978The argument and return value are floating point numbers of the same
9979type.
9980
9981Semantics:
9982""""""""""
9983
9984This function returns the same values as the libm ``trunc`` functions
9985would, and handles error conditions in the same way.
9986
9987'``llvm.rint.*``' Intrinsic
9988^^^^^^^^^^^^^^^^^^^^^^^^^^^
9989
9990Syntax:
9991"""""""
9992
9993This is an overloaded intrinsic. You can use ``llvm.rint`` on any
9994floating point or vector of floating point type. Not all targets support
9995all types however.
9996
9997::
9998
9999 declare float @llvm.rint.f32(float %Val)
10000 declare double @llvm.rint.f64(double %Val)
10001 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
10002 declare fp128 @llvm.rint.f128(fp128 %Val)
10003 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
10004
10005Overview:
10006"""""""""
10007
10008The '``llvm.rint.*``' intrinsics returns the operand rounded to the
10009nearest integer. It may raise an inexact floating-point exception if the
10010operand isn't an integer.
10011
10012Arguments:
10013""""""""""
10014
10015The argument and return value are floating point numbers of the same
10016type.
10017
10018Semantics:
10019""""""""""
10020
10021This function returns the same values as the libm ``rint`` functions
10022would, and handles error conditions in the same way.
10023
10024'``llvm.nearbyint.*``' Intrinsic
10025^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10026
10027Syntax:
10028"""""""
10029
10030This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
10031floating point or vector of floating point type. Not all targets support
10032all types however.
10033
10034::
10035
10036 declare float @llvm.nearbyint.f32(float %Val)
10037 declare double @llvm.nearbyint.f64(double %Val)
10038 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
10039 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
10040 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
10041
10042Overview:
10043"""""""""
10044
10045The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
10046nearest integer.
10047
10048Arguments:
10049""""""""""
10050
10051The argument and return value are floating point numbers of the same
10052type.
10053
10054Semantics:
10055""""""""""
10056
10057This function returns the same values as the libm ``nearbyint``
10058functions would, and handles error conditions in the same way.
10059
Hal Finkel171817e2013-08-07 22:49:12 +000010060'``llvm.round.*``' Intrinsic
10061^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10062
10063Syntax:
10064"""""""
10065
10066This is an overloaded intrinsic. You can use ``llvm.round`` on any
10067floating point or vector of floating point type. Not all targets support
10068all types however.
10069
10070::
10071
10072 declare float @llvm.round.f32(float %Val)
10073 declare double @llvm.round.f64(double %Val)
10074 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
10075 declare fp128 @llvm.round.f128(fp128 %Val)
10076 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
10077
10078Overview:
10079"""""""""
10080
10081The '``llvm.round.*``' intrinsics returns the operand rounded to the
10082nearest integer.
10083
10084Arguments:
10085""""""""""
10086
10087The argument and return value are floating point numbers of the same
10088type.
10089
10090Semantics:
10091""""""""""
10092
10093This function returns the same values as the libm ``round``
10094functions would, and handles error conditions in the same way.
10095
Sean Silvab084af42012-12-07 10:36:55 +000010096Bit Manipulation Intrinsics
10097---------------------------
10098
10099LLVM provides intrinsics for a few important bit manipulation
10100operations. These allow efficient code generation for some algorithms.
10101
10102'``llvm.bswap.*``' Intrinsics
10103^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10104
10105Syntax:
10106"""""""
10107
10108This is an overloaded intrinsic function. You can use bswap on any
10109integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
10110
10111::
10112
10113 declare i16 @llvm.bswap.i16(i16 <id>)
10114 declare i32 @llvm.bswap.i32(i32 <id>)
10115 declare i64 @llvm.bswap.i64(i64 <id>)
10116
10117Overview:
10118"""""""""
10119
10120The '``llvm.bswap``' family of intrinsics is used to byte swap integer
10121values with an even number of bytes (positive multiple of 16 bits).
10122These are useful for performing operations on data that is not in the
10123target's native byte order.
10124
10125Semantics:
10126""""""""""
10127
10128The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
10129and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
10130intrinsic returns an i32 value that has the four bytes of the input i32
10131swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
10132returned i32 will have its bytes in 3, 2, 1, 0 order. The
10133``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
10134concept to additional even-byte lengths (6 bytes, 8 bytes and more,
10135respectively).
10136
10137'``llvm.ctpop.*``' Intrinsic
10138^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10139
10140Syntax:
10141"""""""
10142
10143This is an overloaded intrinsic. You can use llvm.ctpop on any integer
10144bit width, or on any vector with integer elements. Not all targets
10145support all bit widths or vector types, however.
10146
10147::
10148
10149 declare i8 @llvm.ctpop.i8(i8 <src>)
10150 declare i16 @llvm.ctpop.i16(i16 <src>)
10151 declare i32 @llvm.ctpop.i32(i32 <src>)
10152 declare i64 @llvm.ctpop.i64(i64 <src>)
10153 declare i256 @llvm.ctpop.i256(i256 <src>)
10154 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
10155
10156Overview:
10157"""""""""
10158
10159The '``llvm.ctpop``' family of intrinsics counts the number of bits set
10160in a value.
10161
10162Arguments:
10163""""""""""
10164
10165The only argument is the value to be counted. The argument may be of any
10166integer type, or a vector with integer elements. The return type must
10167match the argument type.
10168
10169Semantics:
10170""""""""""
10171
10172The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
10173each element of a vector.
10174
10175'``llvm.ctlz.*``' Intrinsic
10176^^^^^^^^^^^^^^^^^^^^^^^^^^^
10177
10178Syntax:
10179"""""""
10180
10181This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
10182integer bit width, or any vector whose elements are integers. Not all
10183targets support all bit widths or vector types, however.
10184
10185::
10186
10187 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
10188 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
10189 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
10190 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
10191 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
10192 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10193
10194Overview:
10195"""""""""
10196
10197The '``llvm.ctlz``' family of intrinsic functions counts the number of
10198leading zeros in a variable.
10199
10200Arguments:
10201""""""""""
10202
10203The first argument is the value to be counted. This argument may be of
Hal Finkel5dd82782015-01-05 04:05:21 +000010204any integer type, or a vector with integer element type. The return
Sean Silvab084af42012-12-07 10:36:55 +000010205type must match the first argument type.
10206
10207The second argument must be a constant and is a flag to indicate whether
10208the intrinsic should ensure that a zero as the first argument produces a
10209defined result. Historically some architectures did not provide a
10210defined result for zero values as efficiently, and many algorithms are
10211now predicated on avoiding zero-value inputs.
10212
10213Semantics:
10214""""""""""
10215
10216The '``llvm.ctlz``' intrinsic counts the leading (most significant)
10217zeros in a variable, or within each element of the vector. If
10218``src == 0`` then the result is the size in bits of the type of ``src``
10219if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10220``llvm.ctlz(i32 2) = 30``.
10221
10222'``llvm.cttz.*``' Intrinsic
10223^^^^^^^^^^^^^^^^^^^^^^^^^^^
10224
10225Syntax:
10226"""""""
10227
10228This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
10229integer bit width, or any vector of integer elements. Not all targets
10230support all bit widths or vector types, however.
10231
10232::
10233
10234 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
10235 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
10236 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
10237 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
10238 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
10239 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
10240
10241Overview:
10242"""""""""
10243
10244The '``llvm.cttz``' family of intrinsic functions counts the number of
10245trailing zeros.
10246
10247Arguments:
10248""""""""""
10249
10250The first argument is the value to be counted. This argument may be of
Hal Finkel5dd82782015-01-05 04:05:21 +000010251any integer type, or a vector with integer element type. The return
Sean Silvab084af42012-12-07 10:36:55 +000010252type must match the first argument type.
10253
10254The second argument must be a constant and is a flag to indicate whether
10255the intrinsic should ensure that a zero as the first argument produces a
10256defined result. Historically some architectures did not provide a
10257defined result for zero values as efficiently, and many algorithms are
10258now predicated on avoiding zero-value inputs.
10259
10260Semantics:
10261""""""""""
10262
10263The '``llvm.cttz``' intrinsic counts the trailing (least significant)
10264zeros in a variable, or within each element of a vector. If ``src == 0``
10265then the result is the size in bits of the type of ``src`` if
10266``is_zero_undef == 0`` and ``undef`` otherwise. For example,
10267``llvm.cttz(2) = 1``.
10268
Philip Reames34843ae2015-03-05 05:55:55 +000010269.. _int_overflow:
10270
Sean Silvab084af42012-12-07 10:36:55 +000010271Arithmetic with Overflow Intrinsics
10272-----------------------------------
10273
10274LLVM provides intrinsics for some arithmetic with overflow operations.
10275
10276'``llvm.sadd.with.overflow.*``' Intrinsics
10277^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10278
10279Syntax:
10280"""""""
10281
10282This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
10283on any integer bit width.
10284
10285::
10286
10287 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
10288 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10289 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
10290
10291Overview:
10292"""""""""
10293
10294The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
10295a signed addition of the two arguments, and indicate whether an overflow
10296occurred during the signed summation.
10297
10298Arguments:
10299""""""""""
10300
10301The arguments (%a and %b) and the first element of the result structure
10302may be of integer types of any bit width, but they must have the same
10303bit width. The second element of the result structure must be of type
10304``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10305addition.
10306
10307Semantics:
10308""""""""""
10309
10310The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
Dmitri Gribenkoe8131122013-01-19 20:34:20 +000010311a signed addition of the two variables. They return a structure --- the
Sean Silvab084af42012-12-07 10:36:55 +000010312first element of which is the signed summation, and the second element
10313of which is a bit specifying if the signed summation resulted in an
10314overflow.
10315
10316Examples:
10317"""""""""
10318
10319.. code-block:: llvm
10320
10321 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
10322 %sum = extractvalue {i32, i1} %res, 0
10323 %obit = extractvalue {i32, i1} %res, 1
10324 br i1 %obit, label %overflow, label %normal
10325
10326'``llvm.uadd.with.overflow.*``' Intrinsics
10327^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10328
10329Syntax:
10330"""""""
10331
10332This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
10333on any integer bit width.
10334
10335::
10336
10337 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
10338 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10339 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
10340
10341Overview:
10342"""""""""
10343
10344The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
10345an unsigned addition of the two arguments, and indicate whether a carry
10346occurred during the unsigned summation.
10347
10348Arguments:
10349""""""""""
10350
10351The arguments (%a and %b) and the first element of the result structure
10352may be of integer types of any bit width, but they must have the same
10353bit width. The second element of the result structure must be of type
10354``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10355addition.
10356
10357Semantics:
10358""""""""""
10359
10360The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
Dmitri Gribenkoe8131122013-01-19 20:34:20 +000010361an unsigned addition of the two arguments. They return a structure --- the
Sean Silvab084af42012-12-07 10:36:55 +000010362first element of which is the sum, and the second element of which is a
10363bit specifying if the unsigned summation resulted in a carry.
10364
10365Examples:
10366"""""""""
10367
10368.. code-block:: llvm
10369
10370 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
10371 %sum = extractvalue {i32, i1} %res, 0
10372 %obit = extractvalue {i32, i1} %res, 1
10373 br i1 %obit, label %carry, label %normal
10374
10375'``llvm.ssub.with.overflow.*``' Intrinsics
10376^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10377
10378Syntax:
10379"""""""
10380
10381This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
10382on any integer bit width.
10383
10384::
10385
10386 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
10387 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10388 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
10389
10390Overview:
10391"""""""""
10392
10393The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
10394a signed subtraction of the two arguments, and indicate whether an
10395overflow occurred during the signed subtraction.
10396
10397Arguments:
10398""""""""""
10399
10400The arguments (%a and %b) and the first element of the result structure
10401may be of integer types of any bit width, but they must have the same
10402bit width. The second element of the result structure must be of type
10403``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10404subtraction.
10405
10406Semantics:
10407""""""""""
10408
10409The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
Dmitri Gribenkoe8131122013-01-19 20:34:20 +000010410a signed subtraction of the two arguments. They return a structure --- the
Sean Silvab084af42012-12-07 10:36:55 +000010411first element of which is the subtraction, and the second element of
10412which is a bit specifying if the signed subtraction resulted in an
10413overflow.
10414
10415Examples:
10416"""""""""
10417
10418.. code-block:: llvm
10419
10420 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
10421 %sum = extractvalue {i32, i1} %res, 0
10422 %obit = extractvalue {i32, i1} %res, 1
10423 br i1 %obit, label %overflow, label %normal
10424
10425'``llvm.usub.with.overflow.*``' Intrinsics
10426^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10427
10428Syntax:
10429"""""""
10430
10431This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
10432on any integer bit width.
10433
10434::
10435
10436 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
10437 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10438 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
10439
10440Overview:
10441"""""""""
10442
10443The '``llvm.usub.with.overflow``' family of intrinsic functions perform
10444an unsigned subtraction of the two arguments, and indicate whether an
10445overflow occurred during the unsigned subtraction.
10446
10447Arguments:
10448""""""""""
10449
10450The arguments (%a and %b) and the first element of the result structure
10451may be of integer types of any bit width, but they must have the same
10452bit width. The second element of the result structure must be of type
10453``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10454subtraction.
10455
10456Semantics:
10457""""""""""
10458
10459The '``llvm.usub.with.overflow``' family of intrinsic functions perform
Dmitri Gribenkoe8131122013-01-19 20:34:20 +000010460an unsigned subtraction of the two arguments. They return a structure ---
Sean Silvab084af42012-12-07 10:36:55 +000010461the first element of which is the subtraction, and the second element of
10462which is a bit specifying if the unsigned subtraction resulted in an
10463overflow.
10464
10465Examples:
10466"""""""""
10467
10468.. code-block:: llvm
10469
10470 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
10471 %sum = extractvalue {i32, i1} %res, 0
10472 %obit = extractvalue {i32, i1} %res, 1
10473 br i1 %obit, label %overflow, label %normal
10474
10475'``llvm.smul.with.overflow.*``' Intrinsics
10476^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10477
10478Syntax:
10479"""""""
10480
10481This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
10482on any integer bit width.
10483
10484::
10485
10486 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
10487 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10488 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
10489
10490Overview:
10491"""""""""
10492
10493The '``llvm.smul.with.overflow``' family of intrinsic functions perform
10494a signed multiplication of the two arguments, and indicate whether an
10495overflow occurred during the signed multiplication.
10496
10497Arguments:
10498""""""""""
10499
10500The arguments (%a and %b) and the first element of the result structure
10501may be of integer types of any bit width, but they must have the same
10502bit width. The second element of the result structure must be of type
10503``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
10504multiplication.
10505
10506Semantics:
10507""""""""""
10508
10509The '``llvm.smul.with.overflow``' family of intrinsic functions perform
Dmitri Gribenkoe8131122013-01-19 20:34:20 +000010510a signed multiplication of the two arguments. They return a structure ---
Sean Silvab084af42012-12-07 10:36:55 +000010511the first element of which is the multiplication, and the second element
10512of which is a bit specifying if the signed multiplication resulted in an
10513overflow.
10514
10515Examples:
10516"""""""""
10517
10518.. code-block:: llvm
10519
10520 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
10521 %sum = extractvalue {i32, i1} %res, 0
10522 %obit = extractvalue {i32, i1} %res, 1
10523 br i1 %obit, label %overflow, label %normal
10524
10525'``llvm.umul.with.overflow.*``' Intrinsics
10526^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10527
10528Syntax:
10529"""""""
10530
10531This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
10532on any integer bit width.
10533
10534::
10535
10536 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
10537 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10538 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
10539
10540Overview:
10541"""""""""
10542
10543The '``llvm.umul.with.overflow``' family of intrinsic functions perform
10544a unsigned multiplication of the two arguments, and indicate whether an
10545overflow occurred during the unsigned multiplication.
10546
10547Arguments:
10548""""""""""
10549
10550The arguments (%a and %b) and the first element of the result structure
10551may be of integer types of any bit width, but they must have the same
10552bit width. The second element of the result structure must be of type
10553``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
10554multiplication.
10555
10556Semantics:
10557""""""""""
10558
10559The '``llvm.umul.with.overflow``' family of intrinsic functions perform
Dmitri Gribenkoe8131122013-01-19 20:34:20 +000010560an unsigned multiplication of the two arguments. They return a structure ---
10561the first element of which is the multiplication, and the second
Sean Silvab084af42012-12-07 10:36:55 +000010562element of which is a bit specifying if the unsigned multiplication
10563resulted in an overflow.
10564
10565Examples:
10566"""""""""
10567
10568.. code-block:: llvm
10569
10570 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
10571 %sum = extractvalue {i32, i1} %res, 0
10572 %obit = extractvalue {i32, i1} %res, 1
10573 br i1 %obit, label %overflow, label %normal
10574
10575Specialised Arithmetic Intrinsics
10576---------------------------------
10577
Owen Anderson1056a922015-07-11 07:01:27 +000010578'``llvm.canonicalize.*``' Intrinsic
10579^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10580
10581Syntax:
10582"""""""
10583
10584::
10585
10586 declare float @llvm.canonicalize.f32(float %a)
10587 declare double @llvm.canonicalize.f64(double %b)
10588
10589Overview:
10590"""""""""
10591
10592The '``llvm.canonicalize.*``' intrinsic returns the platform specific canonical
10593encoding of a floating point number. This canonicalization is useful for
10594implementing certain numeric primitives such as frexp. The canonical encoding is
10595defined by IEEE-754-2008 to be:
10596
10597::
10598
10599 2.1.8 canonical encoding: The preferred encoding of a floating-point
10600 representation in a format. Applied to declets, significands of finite
10601 numbers, infinities, and NaNs, especially in decimal formats.
10602
10603This operation can also be considered equivalent to the IEEE-754-2008
10604conversion of a floating-point value to the same format. NaNs are handled
10605according to section 6.2.
10606
10607Examples of non-canonical encodings:
10608
10609- x87 pseudo denormals, pseudo NaNs, pseudo Infinity, Unnormals. These are
10610 converted to a canonical representation per hardware-specific protocol.
10611- Many normal decimal floating point numbers have non-canonical alternative
10612 encodings.
10613- Some machines, like GPUs or ARMv7 NEON, do not support subnormal values.
10614 These are treated as non-canonical encodings of zero and with be flushed to
10615 a zero of the same sign by this operation.
10616
10617Note that per IEEE-754-2008 6.2, systems that support signaling NaNs with
10618default exception handling must signal an invalid exception, and produce a
10619quiet NaN result.
10620
10621This function should always be implementable as multiplication by 1.0, provided
10622that the compiler does not constant fold the operation. Likewise, division by
106231.0 and ``llvm.minnum(x, x)`` are possible implementations. Addition with
10624-0.0 is also sufficient provided that the rounding mode is not -Infinity.
10625
10626``@llvm.canonicalize`` must preserve the equality relation. That is:
10627
10628- ``(@llvm.canonicalize(x) == x)`` is equivalent to ``(x == x)``
10629- ``(@llvm.canonicalize(x) == @llvm.canonicalize(y))`` is equivalent to
10630 to ``(x == y)``
10631
10632Additionally, the sign of zero must be conserved:
10633``@llvm.canonicalize(-0.0) = -0.0`` and ``@llvm.canonicalize(+0.0) = +0.0``
10634
10635The payload bits of a NaN must be conserved, with two exceptions.
10636First, environments which use only a single canonical representation of NaN
10637must perform said canonicalization. Second, SNaNs must be quieted per the
10638usual methods.
10639
10640The canonicalization operation may be optimized away if:
10641
10642- The input is known to be canonical. For example, it was produced by a
10643 floating-point operation that is required by the standard to be canonical.
10644- The result is consumed only by (or fused with) other floating-point
10645 operations. That is, the bits of the floating point value are not examined.
10646
Sean Silvab084af42012-12-07 10:36:55 +000010647'``llvm.fmuladd.*``' Intrinsic
10648^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10649
10650Syntax:
10651"""""""
10652
10653::
10654
10655 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
10656 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
10657
10658Overview:
10659"""""""""
10660
10661The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
Lang Hames045f4392013-01-17 00:00:49 +000010662expressions that can be fused if the code generator determines that (a) the
10663target instruction set has support for a fused operation, and (b) that the
10664fused operation is more efficient than the equivalent, separate pair of mul
10665and add instructions.
Sean Silvab084af42012-12-07 10:36:55 +000010666
10667Arguments:
10668""""""""""
10669
10670The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
10671multiplicands, a and b, and an addend c.
10672
10673Semantics:
10674""""""""""
10675
10676The expression:
10677
10678::
10679
10680 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
10681
10682is equivalent to the expression a \* b + c, except that rounding will
10683not be performed between the multiplication and addition steps if the
10684code generator fuses the operations. Fusion is not guaranteed, even if
10685the target platform supports it. If a fused multiply-add is required the
Matt Arsenaultee364ee2014-01-31 00:09:00 +000010686corresponding llvm.fma.\* intrinsic function should be used
10687instead. This never sets errno, just as '``llvm.fma.*``'.
Sean Silvab084af42012-12-07 10:36:55 +000010688
10689Examples:
10690"""""""""
10691
10692.. code-block:: llvm
10693
Tim Northover675a0962014-06-13 14:24:23 +000010694 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
Sean Silvab084af42012-12-07 10:36:55 +000010695
James Molloy7395a812015-07-16 15:22:46 +000010696
10697'``llvm.uabsdiff.*``' and '``llvm.sabsdiff.*``' Intrinsics
10698^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10699
10700Syntax:
10701"""""""
10702This is an overloaded intrinsic. The loaded data is a vector of any integer bit width.
10703
10704.. code-block:: llvm
10705
10706 declare <4 x integer> @llvm.uabsdiff.v4i32(<4 x integer> %a, <4 x integer> %b)
10707
10708
10709Overview:
10710"""""""""
10711
10712The ``llvm.uabsdiff`` intrinsic returns a vector result of the absolute difference of the two operands,
10713treating them both as unsigned integers.
10714
10715The ``llvm.sabsdiff`` intrinsic returns a vector result of the absolute difference of the two operands,
10716treating them both as signed integers.
10717
10718.. note::
10719
10720 These intrinsics are primarily used during the code generation stage of compilation.
10721 They are generated by compiler passes such as the Loop and SLP vectorizers.it is not
10722 recommended for users to create them manually.
10723
10724Arguments:
10725""""""""""
10726
10727Both intrinsics take two integer of the same bitwidth.
10728
10729Semantics:
10730""""""""""
10731
10732The expression::
10733
10734 call <4 x i32> @llvm.uabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
10735
10736is equivalent to::
10737
10738 %sub = sub <4 x i32> %a, %b
10739 %ispos = icmp ugt <4 x i32> %sub, <i32 -1, i32 -1, i32 -1, i32 -1>
10740 %neg = sub <4 x i32> zeroinitializer, %sub
10741 %1 = select <4 x i1> %ispos, <4 x i32> %sub, <4 x i32> %neg
10742
10743Similarly the expression::
10744
10745 call <4 x i32> @llvm.sabsdiff.v4i32(<4 x i32> %a, <4 x i32> %b)
10746
10747is equivalent to::
10748
10749 %sub = sub nsw <4 x i32> %a, %b
10750 %ispos = icmp sgt <4 x i32> %sub, <i32 -1, i32 -1, i32 -1, i32 -1>
10751 %neg = sub nsw <4 x i32> zeroinitializer, %sub
10752 %1 = select <4 x i1> %ispos, <4 x i32> %sub, <4 x i32> %neg
10753
10754
Sean Silvab084af42012-12-07 10:36:55 +000010755Half Precision Floating Point Intrinsics
10756----------------------------------------
10757
10758For most target platforms, half precision floating point is a
10759storage-only format. This means that it is a dense encoding (in memory)
10760but does not support computation in the format.
10761
10762This means that code must first load the half-precision floating point
10763value as an i16, then convert it to float with
10764:ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
10765then be performed on the float value (including extending to double
10766etc). To store the value back to memory, it is first converted to float
10767if needed, then converted to i16 with
10768:ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
10769i16 value.
10770
10771.. _int_convert_to_fp16:
10772
10773'``llvm.convert.to.fp16``' Intrinsic
10774^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10775
10776Syntax:
10777"""""""
10778
10779::
10780
Tim Northoverfd7e4242014-07-17 10:51:23 +000010781 declare i16 @llvm.convert.to.fp16.f32(float %a)
10782 declare i16 @llvm.convert.to.fp16.f64(double %a)
Sean Silvab084af42012-12-07 10:36:55 +000010783
10784Overview:
10785"""""""""
10786
Tim Northoverfd7e4242014-07-17 10:51:23 +000010787The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
10788conventional floating point type to half precision floating point format.
Sean Silvab084af42012-12-07 10:36:55 +000010789
10790Arguments:
10791""""""""""
10792
10793The intrinsic function contains single argument - the value to be
10794converted.
10795
10796Semantics:
10797""""""""""
10798
Tim Northoverfd7e4242014-07-17 10:51:23 +000010799The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
10800conventional floating point format to half precision floating point format. The
10801return value is an ``i16`` which contains the converted number.
Sean Silvab084af42012-12-07 10:36:55 +000010802
10803Examples:
10804"""""""""
10805
10806.. code-block:: llvm
10807
Tim Northoverfd7e4242014-07-17 10:51:23 +000010808 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
Sean Silvab084af42012-12-07 10:36:55 +000010809 store i16 %res, i16* @x, align 2
10810
10811.. _int_convert_from_fp16:
10812
10813'``llvm.convert.from.fp16``' Intrinsic
10814^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10815
10816Syntax:
10817"""""""
10818
10819::
10820
Tim Northoverfd7e4242014-07-17 10:51:23 +000010821 declare float @llvm.convert.from.fp16.f32(i16 %a)
10822 declare double @llvm.convert.from.fp16.f64(i16 %a)
Sean Silvab084af42012-12-07 10:36:55 +000010823
10824Overview:
10825"""""""""
10826
10827The '``llvm.convert.from.fp16``' intrinsic function performs a
10828conversion from half precision floating point format to single precision
10829floating point format.
10830
10831Arguments:
10832""""""""""
10833
10834The intrinsic function contains single argument - the value to be
10835converted.
10836
10837Semantics:
10838""""""""""
10839
10840The '``llvm.convert.from.fp16``' intrinsic function performs a
10841conversion from half single precision floating point format to single
10842precision floating point format. The input half-float value is
10843represented by an ``i16`` value.
10844
10845Examples:
10846"""""""""
10847
10848.. code-block:: llvm
10849
David Blaikiec7aabbb2015-03-04 22:06:14 +000010850 %a = load i16, i16* @x, align 2
Matt Arsenault3e3ddda2014-07-10 03:22:16 +000010851 %res = call float @llvm.convert.from.fp16(i16 %a)
Sean Silvab084af42012-12-07 10:36:55 +000010852
Duncan P. N. Exon Smithe2741802015-03-03 17:24:31 +000010853.. _dbg_intrinsics:
10854
Sean Silvab084af42012-12-07 10:36:55 +000010855Debugger Intrinsics
10856-------------------
10857
10858The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
10859prefix), are described in the `LLVM Source Level
10860Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
10861document.
10862
10863Exception Handling Intrinsics
10864-----------------------------
10865
10866The LLVM exception handling intrinsics (which all start with
10867``llvm.eh.`` prefix), are described in the `LLVM Exception
10868Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
10869
10870.. _int_trampoline:
10871
10872Trampoline Intrinsics
10873---------------------
10874
10875These intrinsics make it possible to excise one parameter, marked with
10876the :ref:`nest <nest>` attribute, from a function. The result is a
10877callable function pointer lacking the nest parameter - the caller does
10878not need to provide a value for it. Instead, the value to use is stored
10879in advance in a "trampoline", a block of memory usually allocated on the
10880stack, which also contains code to splice the nest value into the
10881argument list. This is used to implement the GCC nested function address
10882extension.
10883
10884For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
10885then the resulting function pointer has signature ``i32 (i32, i32)*``.
10886It can be created as follows:
10887
10888.. code-block:: llvm
10889
10890 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
David Blaikie16a97eb2015-03-04 22:02:58 +000010891 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
Sean Silvab084af42012-12-07 10:36:55 +000010892 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
10893 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
10894 %fp = bitcast i8* %p to i32 (i32, i32)*
10895
10896The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
10897``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
10898
10899.. _int_it:
10900
10901'``llvm.init.trampoline``' Intrinsic
10902^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10903
10904Syntax:
10905"""""""
10906
10907::
10908
10909 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
10910
10911Overview:
10912"""""""""
10913
10914This fills the memory pointed to by ``tramp`` with executable code,
10915turning it into a trampoline.
10916
10917Arguments:
10918""""""""""
10919
10920The ``llvm.init.trampoline`` intrinsic takes three arguments, all
10921pointers. The ``tramp`` argument must point to a sufficiently large and
10922sufficiently aligned block of memory; this memory is written to by the
10923intrinsic. Note that the size and the alignment are target-specific -
10924LLVM currently provides no portable way of determining them, so a
10925front-end that generates this intrinsic needs to have some
10926target-specific knowledge. The ``func`` argument must hold a function
10927bitcast to an ``i8*``.
10928
10929Semantics:
10930""""""""""
10931
10932The block of memory pointed to by ``tramp`` is filled with target
10933dependent code, turning it into a function. Then ``tramp`` needs to be
10934passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
10935be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
10936function's signature is the same as that of ``func`` with any arguments
10937marked with the ``nest`` attribute removed. At most one such ``nest``
10938argument is allowed, and it must be of pointer type. Calling the new
10939function is equivalent to calling ``func`` with the same argument list,
10940but with ``nval`` used for the missing ``nest`` argument. If, after
10941calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
10942modified, then the effect of any later call to the returned function
10943pointer is undefined.
10944
10945.. _int_at:
10946
10947'``llvm.adjust.trampoline``' Intrinsic
10948^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10949
10950Syntax:
10951"""""""
10952
10953::
10954
10955 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
10956
10957Overview:
10958"""""""""
10959
10960This performs any required machine-specific adjustment to the address of
10961a trampoline (passed as ``tramp``).
10962
10963Arguments:
10964""""""""""
10965
10966``tramp`` must point to a block of memory which already has trampoline
10967code filled in by a previous call to
10968:ref:`llvm.init.trampoline <int_it>`.
10969
10970Semantics:
10971""""""""""
10972
10973On some architectures the address of the code to be executed needs to be
Sanjay Patel69bf48e2014-07-04 19:40:43 +000010974different than the address where the trampoline is actually stored. This
Sean Silvab084af42012-12-07 10:36:55 +000010975intrinsic returns the executable address corresponding to ``tramp``
10976after performing the required machine specific adjustments. The pointer
10977returned can then be :ref:`bitcast and executed <int_trampoline>`.
10978
Elena Demikhovsky82cdd652015-05-07 12:25:11 +000010979.. _int_mload_mstore:
10980
Elena Demikhovsky3d13f1c2014-12-25 09:29:13 +000010981Masked Vector Load and Store Intrinsics
10982---------------------------------------
10983
10984LLVM provides intrinsics for predicated vector load and store operations. The predicate is specified by a mask operand, which holds one bit per vector element, switching the associated vector lane on or off. The memory addresses corresponding to the "off" lanes are not accessed. When all bits of the mask are on, the intrinsic is identical to a regular vector load or store. When all bits are off, no memory is accessed.
10985
10986.. _int_mload:
10987
10988'``llvm.masked.load.*``' Intrinsics
10989^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10990
10991Syntax:
10992"""""""
10993This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
10994
10995::
10996
10997 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
10998 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
10999
11000Overview:
11001"""""""""
11002
Elena Demikhovsky82cdd652015-05-07 12:25:11 +000011003Reads a vector from memory according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes. The masked-off lanes in the result vector are taken from the corresponding lanes of the '``passthru``' operand.
Elena Demikhovsky3d13f1c2014-12-25 09:29:13 +000011004
11005
11006Arguments:
11007""""""""""
11008
Elena Demikhovsky82cdd652015-05-07 12:25:11 +000011009The first operand is the base pointer for the load. The second operand is the alignment of the source location. It must be a constant integer value. The third operand, mask, is a vector of boolean values with the same number of elements as the return type. The fourth is a pass-through value that is used to fill the masked-off lanes of the result. The return type, underlying type of the base pointer and the type of the '``passthru``' operand are the same vector types.
Elena Demikhovsky3d13f1c2014-12-25 09:29:13 +000011010
11011
11012Semantics:
11013""""""""""
11014
11015The '``llvm.masked.load``' intrinsic is designed for conditional reading of selected vector elements in a single IR operation. It is useful for targets that support vector masked loads and allows vectorizing predicated basic blocks on these targets. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar load operations.
11016The result of this operation is equivalent to a regular vector load instruction followed by a 'select' between the loaded and the passthru values, predicated on the same mask. However, using this intrinsic prevents exceptions on memory access to masked-off lanes.
11017
11018
11019::
11020
11021 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
Mehdi Amini4a121fa2015-03-14 22:04:06 +000011022
Elena Demikhovsky3d13f1c2014-12-25 09:29:13 +000011023 ;; The result of the two following instructions is identical aside from potential memory access exception
David Blaikiec7aabbb2015-03-04 22:06:14 +000011024 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
Elena Demikhovskye86c8c82014-12-29 09:47:51 +000011025 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
Elena Demikhovsky3d13f1c2014-12-25 09:29:13 +000011026
11027.. _int_mstore:
11028
11029'``llvm.masked.store.*``' Intrinsics
11030^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11031
11032Syntax:
11033"""""""
11034This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
11035
11036::
11037
11038 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
11039 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
11040
11041Overview:
11042"""""""""
11043
11044Writes a vector to memory according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes.
11045
11046Arguments:
11047""""""""""
11048
11049The first operand is the vector value to be written to memory. The second operand is the base pointer for the store, it has the same underlying type as the value operand. The third operand is the alignment of the destination location. The fourth operand, mask, is a vector of boolean values. The types of the mask and the value operand must have the same number of vector elements.
11050
11051
11052Semantics:
11053""""""""""
11054
11055The '``llvm.masked.store``' intrinsics is designed for conditional writing of selected vector elements in a single IR operation. It is useful for targets that support vector masked store and allows vectorizing predicated basic blocks on these targets. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar store operations.
11056The result of this operation is equivalent to a load-modify-store sequence. However, using this intrinsic prevents exceptions and data races on memory access to masked-off lanes.
11057
11058::
11059
11060 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
Mehdi Amini4a121fa2015-03-14 22:04:06 +000011061
Elena Demikhovskye86c8c82014-12-29 09:47:51 +000011062 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
David Blaikiec7aabbb2015-03-04 22:06:14 +000011063 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
Elena Demikhovsky3d13f1c2014-12-25 09:29:13 +000011064 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
11065 store <16 x float> %res, <16 x float>* %ptr, align 4
11066
11067
Elena Demikhovsky82cdd652015-05-07 12:25:11 +000011068Masked Vector Gather and Scatter Intrinsics
11069-------------------------------------------
11070
11071LLVM provides intrinsics for vector gather and scatter operations. They are similar to :ref:`Masked Vector Load and Store <int_mload_mstore>`, except they are designed for arbitrary memory accesses, rather than sequential memory accesses. Gather and scatter also employ a mask operand, which holds one bit per vector element, switching the associated vector lane on or off. The memory addresses corresponding to the "off" lanes are not accessed. When all bits are off, no memory is accessed.
11072
11073.. _int_mgather:
11074
11075'``llvm.masked.gather.*``' Intrinsics
11076^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11077
11078Syntax:
11079"""""""
11080This is an overloaded intrinsic. The loaded data are multiple scalar values of any integer or floating point data type gathered together into one vector.
11081
11082::
11083
11084 declare <16 x float> @llvm.masked.gather.v16f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
11085 declare <2 x double> @llvm.masked.gather.v2f64 (<2 x double*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
11086
11087Overview:
11088"""""""""
11089
11090Reads scalar values from arbitrary memory locations and gathers them into one vector. The memory locations are provided in the vector of pointers '``ptrs``'. The memory is accessed according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes. The masked-off lanes in the result vector are taken from the corresponding lanes of the '``passthru``' operand.
11091
11092
11093Arguments:
11094""""""""""
11095
11096The first operand is a vector of pointers which holds all memory addresses to read. The second operand is an alignment of the source addresses. It must be a constant integer value. The third operand, mask, is a vector of boolean values with the same number of elements as the return type. The fourth is a pass-through value that is used to fill the masked-off lanes of the result. The return type, underlying type of the vector of pointers and the type of the '``passthru``' operand are the same vector types.
11097
11098
11099Semantics:
11100""""""""""
11101
11102The '``llvm.masked.gather``' intrinsic is designed for conditional reading of multiple scalar values from arbitrary memory locations in a single IR operation. It is useful for targets that support vector masked gathers and allows vectorizing basic blocks with data and control divergence. Other targets may support this intrinsic differently, for example by lowering it into a sequence of scalar load operations.
11103The semantics of this operation are equivalent to a sequence of conditional scalar loads with subsequent gathering all loaded values into a single vector. The mask restricts memory access to certain lanes and facilitates vectorization of predicated basic blocks.
11104
11105
11106::
11107
11108 %res = call <4 x double> @llvm.masked.gather.v4f64 (<4 x double*> %ptrs, i32 8, <4 x i1>%mask, <4 x double> <true, true, true, true>)
11109
11110 ;; The gather with all-true mask is equivalent to the following instruction sequence
11111 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
11112 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
11113 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
11114 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
11115
11116 %val0 = load double, double* %ptr0, align 8
11117 %val1 = load double, double* %ptr1, align 8
11118 %val2 = load double, double* %ptr2, align 8
11119 %val3 = load double, double* %ptr3, align 8
11120
11121 %vec0 = insertelement <4 x double>undef, %val0, 0
11122 %vec01 = insertelement <4 x double>%vec0, %val1, 1
11123 %vec012 = insertelement <4 x double>%vec01, %val2, 2
11124 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
11125
11126.. _int_mscatter:
11127
11128'``llvm.masked.scatter.*``' Intrinsics
11129^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11130
11131Syntax:
11132"""""""
11133This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type. Each vector element is stored in an arbitrary memory addresses. Scatter with overlapping addresses is guaranteed to be ordered from least-significant to most-significant element.
11134
11135::
11136
11137 declare void @llvm.masked.scatter.v8i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
11138 declare void @llvm.masked.scatter.v16f32(<16 x i32> <value>, <16 x i32*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
11139
11140Overview:
11141"""""""""
11142
11143Writes each element from the value vector to the corresponding memory address. The memory addresses are represented as a vector of pointers. Writing is done according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes.
11144
11145Arguments:
11146""""""""""
11147
11148The first operand is a vector value to be written to memory. The second operand is a vector of pointers, pointing to where the value elements should be stored. It has the same underlying type as the value operand. The third operand is an alignment of the destination addresses. The fourth operand, mask, is a vector of boolean values. The types of the mask and the value operand must have the same number of vector elements.
11149
11150
11151Semantics:
11152""""""""""
11153
11154The '``llvm.masked.scatter``' intrinsics is designed for writing selected vector elements to arbitrary memory addresses in a single IR operation. The operation may be conditional, when not all bits in the mask are switched on. It is useful for targets that support vector masked scatter and allows vectorizing basic blocks with data and control divergency. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar store operations.
11155
11156::
11157
11158 ;; This instruction unconditionaly stores data vector in multiple addresses
11159 call @llvm.masked.scatter.v8i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
11160
11161 ;; It is equivalent to a list of scalar stores
11162 %val0 = extractelement <8 x i32> %value, i32 0
11163 %val1 = extractelement <8 x i32> %value, i32 1
11164 ..
11165 %val7 = extractelement <8 x i32> %value, i32 7
11166 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
11167 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
11168 ..
11169 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
11170 ;; Note: the order of the following stores is important when they overlap:
11171 store i32 %val0, i32* %ptr0, align 4
11172 store i32 %val1, i32* %ptr1, align 4
11173 ..
11174 store i32 %val7, i32* %ptr7, align 4
11175
11176
Sean Silvab084af42012-12-07 10:36:55 +000011177Memory Use Markers
11178------------------
11179
Sanjay Patel69bf48e2014-07-04 19:40:43 +000011180This class of intrinsics provides information about the lifetime of
Sean Silvab084af42012-12-07 10:36:55 +000011181memory objects and ranges where variables are immutable.
11182
Reid Klecknera534a382013-12-19 02:14:12 +000011183.. _int_lifestart:
11184
Sean Silvab084af42012-12-07 10:36:55 +000011185'``llvm.lifetime.start``' Intrinsic
11186^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11187
11188Syntax:
11189"""""""
11190
11191::
11192
11193 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
11194
11195Overview:
11196"""""""""
11197
11198The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
11199object's lifetime.
11200
11201Arguments:
11202""""""""""
11203
11204The first argument is a constant integer representing the size of the
11205object, or -1 if it is variable sized. The second argument is a pointer
11206to the object.
11207
11208Semantics:
11209""""""""""
11210
11211This intrinsic indicates that before this point in the code, the value
11212of the memory pointed to by ``ptr`` is dead. This means that it is known
11213to never be used and has an undefined value. A load from the pointer
11214that precedes this intrinsic can be replaced with ``'undef'``.
11215
Reid Klecknera534a382013-12-19 02:14:12 +000011216.. _int_lifeend:
11217
Sean Silvab084af42012-12-07 10:36:55 +000011218'``llvm.lifetime.end``' Intrinsic
11219^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11220
11221Syntax:
11222"""""""
11223
11224::
11225
11226 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
11227
11228Overview:
11229"""""""""
11230
11231The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
11232object's lifetime.
11233
11234Arguments:
11235""""""""""
11236
11237The first argument is a constant integer representing the size of the
11238object, or -1 if it is variable sized. The second argument is a pointer
11239to the object.
11240
11241Semantics:
11242""""""""""
11243
11244This intrinsic indicates that after this point in the code, the value of
11245the memory pointed to by ``ptr`` is dead. This means that it is known to
11246never be used and has an undefined value. Any stores into the memory
11247object following this intrinsic may be removed as dead.
11248
11249'``llvm.invariant.start``' Intrinsic
11250^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11251
11252Syntax:
11253"""""""
11254
11255::
11256
11257 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
11258
11259Overview:
11260"""""""""
11261
11262The '``llvm.invariant.start``' intrinsic specifies that the contents of
11263a memory object will not change.
11264
11265Arguments:
11266""""""""""
11267
11268The first argument is a constant integer representing the size of the
11269object, or -1 if it is variable sized. The second argument is a pointer
11270to the object.
11271
11272Semantics:
11273""""""""""
11274
11275This intrinsic indicates that until an ``llvm.invariant.end`` that uses
11276the return value, the referenced memory location is constant and
11277unchanging.
11278
11279'``llvm.invariant.end``' Intrinsic
11280^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11281
11282Syntax:
11283"""""""
11284
11285::
11286
11287 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
11288
11289Overview:
11290"""""""""
11291
11292The '``llvm.invariant.end``' intrinsic specifies that the contents of a
11293memory object are mutable.
11294
11295Arguments:
11296""""""""""
11297
11298The first argument is the matching ``llvm.invariant.start`` intrinsic.
11299The second argument is a constant integer representing the size of the
11300object, or -1 if it is variable sized and the third argument is a
11301pointer to the object.
11302
11303Semantics:
11304""""""""""
11305
11306This intrinsic indicates that the memory is mutable again.
11307
11308General Intrinsics
11309------------------
11310
11311This class of intrinsics is designed to be generic and has no specific
11312purpose.
11313
11314'``llvm.var.annotation``' Intrinsic
11315^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11316
11317Syntax:
11318"""""""
11319
11320::
11321
11322 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11323
11324Overview:
11325"""""""""
11326
11327The '``llvm.var.annotation``' intrinsic.
11328
11329Arguments:
11330""""""""""
11331
11332The first argument is a pointer to a value, the second is a pointer to a
11333global string, the third is a pointer to a global string which is the
11334source file name, and the last argument is the line number.
11335
11336Semantics:
11337""""""""""
11338
11339This intrinsic allows annotation of local variables with arbitrary
11340strings. This can be useful for special purpose optimizations that want
11341to look for these annotations. These have no other defined use; they are
11342ignored by code generation and optimization.
11343
Michael Gottesman88d18832013-03-26 00:34:27 +000011344'``llvm.ptr.annotation.*``' Intrinsic
11345^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11346
11347Syntax:
11348"""""""
11349
11350This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
11351pointer to an integer of any width. *NOTE* you must specify an address space for
11352the pointer. The identifier for the default address space is the integer
11353'``0``'.
11354
11355::
11356
11357 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
11358 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
11359 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
11360 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
11361 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
11362
11363Overview:
11364"""""""""
11365
11366The '``llvm.ptr.annotation``' intrinsic.
11367
11368Arguments:
11369""""""""""
11370
11371The first argument is a pointer to an integer value of arbitrary bitwidth
11372(result of some expression), the second is a pointer to a global string, the
11373third is a pointer to a global string which is the source file name, and the
11374last argument is the line number. It returns the value of the first argument.
11375
11376Semantics:
11377""""""""""
11378
11379This intrinsic allows annotation of a pointer to an integer with arbitrary
11380strings. This can be useful for special purpose optimizations that want to look
11381for these annotations. These have no other defined use; they are ignored by code
11382generation and optimization.
11383
Sean Silvab084af42012-12-07 10:36:55 +000011384'``llvm.annotation.*``' Intrinsic
11385^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11386
11387Syntax:
11388"""""""
11389
11390This is an overloaded intrinsic. You can use '``llvm.annotation``' on
11391any integer bit width.
11392
11393::
11394
11395 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
11396 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
11397 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
11398 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
11399 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
11400
11401Overview:
11402"""""""""
11403
11404The '``llvm.annotation``' intrinsic.
11405
11406Arguments:
11407""""""""""
11408
11409The first argument is an integer value (result of some expression), the
11410second is a pointer to a global string, the third is a pointer to a
11411global string which is the source file name, and the last argument is
11412the line number. It returns the value of the first argument.
11413
11414Semantics:
11415""""""""""
11416
11417This intrinsic allows annotations to be put on arbitrary expressions
11418with arbitrary strings. This can be useful for special purpose
11419optimizations that want to look for these annotations. These have no
11420other defined use; they are ignored by code generation and optimization.
11421
11422'``llvm.trap``' Intrinsic
11423^^^^^^^^^^^^^^^^^^^^^^^^^
11424
11425Syntax:
11426"""""""
11427
11428::
11429
11430 declare void @llvm.trap() noreturn nounwind
11431
11432Overview:
11433"""""""""
11434
11435The '``llvm.trap``' intrinsic.
11436
11437Arguments:
11438""""""""""
11439
11440None.
11441
11442Semantics:
11443""""""""""
11444
11445This intrinsic is lowered to the target dependent trap instruction. If
11446the target does not have a trap instruction, this intrinsic will be
11447lowered to a call of the ``abort()`` function.
11448
11449'``llvm.debugtrap``' Intrinsic
11450^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11451
11452Syntax:
11453"""""""
11454
11455::
11456
11457 declare void @llvm.debugtrap() nounwind
11458
11459Overview:
11460"""""""""
11461
11462The '``llvm.debugtrap``' intrinsic.
11463
11464Arguments:
11465""""""""""
11466
11467None.
11468
11469Semantics:
11470""""""""""
11471
11472This intrinsic is lowered to code which is intended to cause an
11473execution trap with the intention of requesting the attention of a
11474debugger.
11475
11476'``llvm.stackprotector``' Intrinsic
11477^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11478
11479Syntax:
11480"""""""
11481
11482::
11483
11484 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
11485
11486Overview:
11487"""""""""
11488
11489The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
11490onto the stack at ``slot``. The stack slot is adjusted to ensure that it
11491is placed on the stack before local variables.
11492
11493Arguments:
11494""""""""""
11495
11496The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
11497The first argument is the value loaded from the stack guard
11498``@__stack_chk_guard``. The second variable is an ``alloca`` that has
11499enough space to hold the value of the guard.
11500
11501Semantics:
11502""""""""""
11503
Michael Gottesmandafc7d92013-08-12 18:35:32 +000011504This intrinsic causes the prologue/epilogue inserter to force the position of
11505the ``AllocaInst`` stack slot to be before local variables on the stack. This is
11506to ensure that if a local variable on the stack is overwritten, it will destroy
11507the value of the guard. When the function exits, the guard on the stack is
11508checked against the original guard by ``llvm.stackprotectorcheck``. If they are
11509different, then ``llvm.stackprotectorcheck`` causes the program to abort by
11510calling the ``__stack_chk_fail()`` function.
11511
11512'``llvm.stackprotectorcheck``' Intrinsic
Sean Silva9d1e1a32013-09-09 19:13:28 +000011513^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Michael Gottesmandafc7d92013-08-12 18:35:32 +000011514
11515Syntax:
11516"""""""
11517
11518::
11519
11520 declare void @llvm.stackprotectorcheck(i8** <guard>)
11521
11522Overview:
11523"""""""""
11524
11525The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
Michael Gottesman98850bd2013-08-12 19:44:09 +000011526created stack protector and if they are not equal calls the
Sean Silvab084af42012-12-07 10:36:55 +000011527``__stack_chk_fail()`` function.
11528
Michael Gottesmandafc7d92013-08-12 18:35:32 +000011529Arguments:
11530""""""""""
11531
11532The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
11533the variable ``@__stack_chk_guard``.
11534
11535Semantics:
11536""""""""""
11537
11538This intrinsic is provided to perform the stack protector check by comparing
11539``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
11540values do not match call the ``__stack_chk_fail()`` function.
11541
11542The reason to provide this as an IR level intrinsic instead of implementing it
11543via other IR operations is that in order to perform this operation at the IR
11544level without an intrinsic, one would need to create additional basic blocks to
11545handle the success/failure cases. This makes it difficult to stop the stack
11546protector check from disrupting sibling tail calls in Codegen. With this
11547intrinsic, we are able to generate the stack protector basic blocks late in
Benjamin Kramer3b32b2f2013-10-29 17:53:27 +000011548codegen after the tail call decision has occurred.
Michael Gottesmandafc7d92013-08-12 18:35:32 +000011549
Sean Silvab084af42012-12-07 10:36:55 +000011550'``llvm.objectsize``' Intrinsic
11551^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11552
11553Syntax:
11554"""""""
11555
11556::
11557
11558 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
11559 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
11560
11561Overview:
11562"""""""""
11563
11564The ``llvm.objectsize`` intrinsic is designed to provide information to
11565the optimizers to determine at compile time whether a) an operation
11566(like memcpy) will overflow a buffer that corresponds to an object, or
11567b) that a runtime check for overflow isn't necessary. An object in this
11568context means an allocation of a specific class, structure, array, or
11569other object.
11570
11571Arguments:
11572""""""""""
11573
11574The ``llvm.objectsize`` intrinsic takes two arguments. The first
11575argument is a pointer to or into the ``object``. The second argument is
11576a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
11577or -1 (if false) when the object size is unknown. The second argument
11578only accepts constants.
11579
11580Semantics:
11581""""""""""
11582
11583The ``llvm.objectsize`` intrinsic is lowered to a constant representing
11584the size of the object concerned. If the size cannot be determined at
11585compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
11586on the ``min`` argument).
11587
11588'``llvm.expect``' Intrinsic
11589^^^^^^^^^^^^^^^^^^^^^^^^^^^
11590
11591Syntax:
11592"""""""
11593
Duncan P. N. Exon Smith1ff08e32014-02-02 22:43:55 +000011594This is an overloaded intrinsic. You can use ``llvm.expect`` on any
11595integer bit width.
11596
Sean Silvab084af42012-12-07 10:36:55 +000011597::
11598
Duncan P. N. Exon Smith1ff08e32014-02-02 22:43:55 +000011599 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
Sean Silvab084af42012-12-07 10:36:55 +000011600 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
11601 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
11602
11603Overview:
11604"""""""""
11605
11606The ``llvm.expect`` intrinsic provides information about expected (the
11607most probable) value of ``val``, which can be used by optimizers.
11608
11609Arguments:
11610""""""""""
11611
11612The ``llvm.expect`` intrinsic takes two arguments. The first argument is
11613a value. The second argument is an expected value, this needs to be a
11614constant value, variables are not allowed.
11615
11616Semantics:
11617""""""""""
11618
11619This intrinsic is lowered to the ``val``.
11620
Philip Reamese0e90832015-04-26 22:23:12 +000011621.. _int_assume:
11622
Hal Finkel93046912014-07-25 21:13:35 +000011623'``llvm.assume``' Intrinsic
11624^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11625
11626Syntax:
11627"""""""
11628
11629::
11630
11631 declare void @llvm.assume(i1 %cond)
11632
11633Overview:
11634"""""""""
11635
11636The ``llvm.assume`` allows the optimizer to assume that the provided
11637condition is true. This information can then be used in simplifying other parts
11638of the code.
11639
11640Arguments:
11641""""""""""
11642
11643The condition which the optimizer may assume is always true.
11644
11645Semantics:
11646""""""""""
11647
11648The intrinsic allows the optimizer to assume that the provided condition is
11649always true whenever the control flow reaches the intrinsic call. No code is
11650generated for this intrinsic, and instructions that contribute only to the
11651provided condition are not used for code generation. If the condition is
11652violated during execution, the behavior is undefined.
11653
Sanjay Patel1ed2bb52015-01-14 16:03:58 +000011654Note that the optimizer might limit the transformations performed on values
Hal Finkel93046912014-07-25 21:13:35 +000011655used by the ``llvm.assume`` intrinsic in order to preserve the instructions
11656only used to form the intrinsic's input argument. This might prove undesirable
Sanjay Patel1ed2bb52015-01-14 16:03:58 +000011657if the extra information provided by the ``llvm.assume`` intrinsic does not cause
Hal Finkel93046912014-07-25 21:13:35 +000011658sufficient overall improvement in code quality. For this reason,
11659``llvm.assume`` should not be used to document basic mathematical invariants
11660that the optimizer can otherwise deduce or facts that are of little use to the
11661optimizer.
11662
Peter Collingbournee6909c82015-02-20 20:30:47 +000011663.. _bitset.test:
11664
11665'``llvm.bitset.test``' Intrinsic
11666^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11667
11668Syntax:
11669"""""""
11670
11671::
11672
11673 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
11674
11675
11676Arguments:
11677""""""""""
11678
11679The first argument is a pointer to be tested. The second argument is a
11680metadata string containing the name of a :doc:`bitset <BitSets>`.
11681
11682Overview:
11683"""""""""
11684
11685The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
11686member of the given bitset.
11687
Sean Silvab084af42012-12-07 10:36:55 +000011688'``llvm.donothing``' Intrinsic
11689^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
11690
11691Syntax:
11692"""""""
11693
11694::
11695
11696 declare void @llvm.donothing() nounwind readnone
11697
11698Overview:
11699"""""""""
11700
Juergen Ributzkac9161192014-10-23 22:36:13 +000011701The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
11702two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
11703with an invoke instruction.
Sean Silvab084af42012-12-07 10:36:55 +000011704
11705Arguments:
11706""""""""""
11707
11708None.
11709
11710Semantics:
11711""""""""""
11712
11713This intrinsic does nothing, and it's removed by optimizers and ignored
11714by codegen.
Andrew Trick5e029ce2013-12-24 02:57:25 +000011715
11716Stack Map Intrinsics
11717--------------------
11718
11719LLVM provides experimental intrinsics to support runtime patching
11720mechanisms commonly desired in dynamic language JITs. These intrinsics
11721are described in :doc:`StackMaps`.