1 ========================
2 LLVM Programmer's Manual
3 ========================
9 This is always a work in progress.
16 This document is meant to highlight some of the important classes and interfaces
17 available in the LLVM source-base. This manual is not intended to explain what
18 LLVM is, how it works, and what LLVM code looks like. It assumes that you know
19 the basics of LLVM and are interested in writing transformations or otherwise
20 analyzing or manipulating the code.
22 This document should get you oriented so that you can find your way in the
23 continuously growing source code that makes up the LLVM infrastructure. Note
24 that this manual is not intended to serve as a replacement for reading the
25 source code, so if you think there should be a method in one of these classes to
26 do something, but it's not listed, check the source. Links to the `doxygen
27 <http://llvm.org/doxygen/>`__ sources are provided to make this as easy as
30 The first section of this document describes general information that is useful
31 to know when working in the LLVM infrastructure, and the second describes the
32 Core LLVM classes. In the future this manual will be extended with information
33 describing how to use extension libraries, such as dominator information, CFG
34 traversal routines, and useful utilities like the ``InstVisitor`` (`doxygen
35 <http://llvm.org/doxygen/InstVisitor_8h-source.html>`__) template.
42 This section contains general information that is useful if you are working in
43 the LLVM source-base, but that isn't specific to any particular API.
47 The C++ Standard Template Library
48 ---------------------------------
50 LLVM makes heavy use of the C++ Standard Template Library (STL), perhaps much
51 more than you are used to, or have seen before. Because of this, you might want
52 to do a little background reading in the techniques used and capabilities of the
53 library. There are many good pages that discuss the STL, and several books on
54 the subject that you can get, so it will not be discussed in this document.
56 Here are some useful links:
59 <http://en.cppreference.com/w/>`_ - an excellent
60 reference for the STL and other parts of the standard C++ library.
62 #. `C++ In a Nutshell <http://www.tempest-sw.com/cpp/>`_ - This is an O'Reilly
63 book in the making. It has a decent Standard Library Reference that rivals
64 Dinkumware's, and is unfortunately no longer free since the book has been
67 #. `C++ Frequently Asked Questions <http://www.parashift.com/c++-faq-lite/>`_.
69 #. `SGI's STL Programmer's Guide <http://www.sgi.com/tech/stl/>`_ - Contains a
70 useful `Introduction to the STL
71 <http://www.sgi.com/tech/stl/stl_introduction.html>`_.
73 #. `Bjarne Stroustrup's C++ Page
74 <http://www.research.att.com/%7Ebs/C++.html>`_.
76 #. `Bruce Eckel's Thinking in C++, 2nd ed. Volume 2 Revision 4.0
77 (even better, get the book)
78 <http://www.mindview.net/Books/TICPP/ThinkingInCPP2e.html>`_.
80 You are also encouraged to take a look at the :doc:`LLVM Coding Standards
81 <CodingStandards>` guide which focuses on how to write maintainable code more
82 than where to put your curly braces.
86 Other useful references
87 -----------------------
89 #. `Using static and shared libraries across platforms
90 <http://www.fortran-2000.com/ArnaudRecipes/sharedlib.html>`_
94 Important and useful LLVM APIs
95 ==============================
97 Here we highlight some LLVM APIs that are generally useful and good to know
98 about when writing transformations.
102 The ``isa<>``, ``cast<>`` and ``dyn_cast<>`` templates
103 ------------------------------------------------------
105 The LLVM source-base makes extensive use of a custom form of RTTI. These
106 templates have many similarities to the C++ ``dynamic_cast<>`` operator, but
107 they don't have some drawbacks (primarily stemming from the fact that
108 ``dynamic_cast<>`` only works on classes that have a v-table). Because they are
109 used so often, you must know what they do and how they work. All of these
110 templates are defined in the ``llvm/Support/Casting.h`` (`doxygen
111 <http://llvm.org/doxygen/Casting_8h-source.html>`__) file (note that you very
112 rarely have to include this file directly).
115 The ``isa<>`` operator works exactly like the Java "``instanceof``" operator.
116 It returns true or false depending on whether a reference or pointer points to
117 an instance of the specified class. This can be very useful for constraint
118 checking of various sorts (example below).
121 The ``cast<>`` operator is a "checked cast" operation. It converts a pointer
122 or reference from a base class to a derived class, causing an assertion
123 failure if it is not really an instance of the right type. This should be
124 used in cases where you have some information that makes you believe that
125 something is of the right type. An example of the ``isa<>`` and ``cast<>``
130 static bool isLoopInvariant(const Value *V, const Loop *L) {
131 if (isa<Constant>(V) || isa<Argument>(V) || isa<GlobalValue>(V))
134 // Otherwise, it must be an instruction...
135 return !L->contains(cast<Instruction>(V)->getParent());
138 Note that you should **not** use an ``isa<>`` test followed by a ``cast<>``,
139 for that use the ``dyn_cast<>`` operator.
142 The ``dyn_cast<>`` operator is a "checking cast" operation. It checks to see
143 if the operand is of the specified type, and if so, returns a pointer to it
144 (this operator does not work with references). If the operand is not of the
145 correct type, a null pointer is returned. Thus, this works very much like
146 the ``dynamic_cast<>`` operator in C++, and should be used in the same
147 circumstances. Typically, the ``dyn_cast<>`` operator is used in an ``if``
148 statement or some other flow control statement like this:
152 if (AllocationInst *AI = dyn_cast<AllocationInst>(Val)) {
156 This form of the ``if`` statement effectively combines together a call to
157 ``isa<>`` and a call to ``cast<>`` into one statement, which is very
160 Note that the ``dyn_cast<>`` operator, like C++'s ``dynamic_cast<>`` or Java's
161 ``instanceof`` operator, can be abused. In particular, you should not use big
162 chained ``if/then/else`` blocks to check for lots of different variants of
163 classes. If you find yourself wanting to do this, it is much cleaner and more
164 efficient to use the ``InstVisitor`` class to dispatch over the instruction
168 The ``cast_or_null<>`` operator works just like the ``cast<>`` operator,
169 except that it allows for a null pointer as an argument (which it then
170 propagates). This can sometimes be useful, allowing you to combine several
171 null checks into one.
173 ``dyn_cast_or_null<>``:
174 The ``dyn_cast_or_null<>`` operator works just like the ``dyn_cast<>``
175 operator, except that it allows for a null pointer as an argument (which it
176 then propagates). This can sometimes be useful, allowing you to combine
177 several null checks into one.
179 These five templates can be used with any classes, whether they have a v-table
180 or not. If you want to add support for these templates, see the document
181 :doc:`How to set up LLVM-style RTTI for your class hierarchy
182 <HowToSetUpLLVMStyleRTTI>`
186 Passing strings (the ``StringRef`` and ``Twine`` classes)
187 ---------------------------------------------------------
189 Although LLVM generally does not do much string manipulation, we do have several
190 important APIs which take strings. Two important examples are the Value class
191 -- which has names for instructions, functions, etc. -- and the ``StringMap``
192 class which is used extensively in LLVM and Clang.
194 These are generic classes, and they need to be able to accept strings which may
195 have embedded null characters. Therefore, they cannot simply take a ``const
196 char *``, and taking a ``const std::string&`` requires clients to perform a heap
197 allocation which is usually unnecessary. Instead, many LLVM APIs use a
198 ``StringRef`` or a ``const Twine&`` for passing strings efficiently.
202 The ``StringRef`` class
203 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
205 The ``StringRef`` data type represents a reference to a constant string (a
206 character array and a length) and supports the common operations available on
207 ``std::string``, but does not require heap allocation.
209 It can be implicitly constructed using a C style null-terminated string, an
210 ``std::string``, or explicitly with a character pointer and length. For
211 example, the ``StringRef`` find function is declared as:
215 iterator find(StringRef Key);
217 and clients can call it using any one of:
221 Map.find("foo"); // Lookup "foo"
222 Map.find(std::string("bar")); // Lookup "bar"
223 Map.find(StringRef("\0baz", 4)); // Lookup "\0baz"
225 Similarly, APIs which need to return a string may return a ``StringRef``
226 instance, which can be used directly or converted to an ``std::string`` using
227 the ``str`` member function. See ``llvm/ADT/StringRef.h`` (`doxygen
228 <http://llvm.org/doxygen/classllvm_1_1StringRef_8h-source.html>`__) for more
231 You should rarely use the ``StringRef`` class directly, because it contains
232 pointers to external memory it is not generally safe to store an instance of the
233 class (unless you know that the external storage will not be freed).
234 ``StringRef`` is small and pervasive enough in LLVM that it should always be
240 The ``Twine`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1Twine.html>`__)
241 class is an efficient way for APIs to accept concatenated strings. For example,
242 a common LLVM paradigm is to name one instruction based on the name of another
243 instruction with a suffix, for example:
247 New = CmpInst::Create(..., SO->getName() + ".cmp");
249 The ``Twine`` class is effectively a lightweight `rope
250 <http://en.wikipedia.org/wiki/Rope_(computer_science)>`_ which points to
251 temporary (stack allocated) objects. Twines can be implicitly constructed as
252 the result of the plus operator applied to strings (i.e., a C strings, an
253 ``std::string``, or a ``StringRef``). The twine delays the actual concatenation
254 of strings until it is actually required, at which point it can be efficiently
255 rendered directly into a character array. This avoids unnecessary heap
256 allocation involved in constructing the temporary results of string
257 concatenation. See ``llvm/ADT/Twine.h`` (`doxygen
258 <http://llvm.org/doxygen/Twine_8h_source.html>`__) and :ref:`here <dss_twine>`
259 for more information.
261 As with a ``StringRef``, ``Twine`` objects point to external memory and should
262 almost never be stored or mentioned directly. They are intended solely for use
263 when defining a function which should be able to efficiently accept concatenated
268 Passing functions and other callable objects
269 --------------------------------------------
271 Sometimes you may want a function to be passed a callback object. In order to
272 support lambda expressions and other function objects, you should not use the
273 traditional C approach of taking a function pointer and an opaque cookie:
277 void takeCallback(bool (*Callback)(Function *, void *), void *Cookie);
279 Instead, use one of the following approaches:
284 If you don't mind putting the definition of your function into a header file,
285 make it a function template that is templated on the callable type.
289 template<typename Callable>
290 void takeCallback(Callable Callback) {
294 The ``function_ref`` class template
295 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
298 (`doxygen <http://llvm.org/doxygen/classllvm_1_1function_ref.html>`__) class
299 template represents a reference to a callable object, templated over the type
300 of the callable. This is a good choice for passing a callback to a function,
301 if you don't need to hold onto the callback after the function returns. In this
302 way, ``function_ref`` is to ``std::function`` as ``StringRef`` is to
305 ``function_ref<Ret(Param1, Param2, ...)>`` can be implicitly constructed from
306 any callable object that can be called with arguments of type ``Param1``,
307 ``Param2``, ..., and returns a value that can be converted to type ``Ret``.
312 void visitBasicBlocks(Function *F, function_ref<bool (BasicBlock*)> Callback) {
313 for (BasicBlock &BB : *F)
322 visitBasicBlocks(F, [&](BasicBlock *BB) {
328 Note that a ``function_ref`` object contains pointers to external memory, so it
329 is not generally safe to store an instance of the class (unless you know that
330 the external storage will not be freed). If you need this ability, consider
331 using ``std::function``. ``function_ref`` is small enough that it should always
336 The ``DEBUG()`` macro and ``-debug`` option
337 -------------------------------------------
339 Often when working on your pass you will put a bunch of debugging printouts and
340 other code into your pass. After you get it working, you want to remove it, but
341 you may need it again in the future (to work out new bugs that you run across).
343 Naturally, because of this, you don't want to delete the debug printouts, but
344 you don't want them to always be noisy. A standard compromise is to comment
345 them out, allowing you to enable them if you need them in the future.
347 The ``llvm/Support/Debug.h`` (`doxygen
348 <http://llvm.org/doxygen/Debug_8h-source.html>`__) file provides a macro named
349 ``DEBUG()`` that is a much nicer solution to this problem. Basically, you can
350 put arbitrary code into the argument of the ``DEBUG`` macro, and it is only
351 executed if '``opt``' (or any other tool) is run with the '``-debug``' command
356 DEBUG(errs() << "I am here!\n");
358 Then you can run your pass like this:
362 $ opt < a.bc > /dev/null -mypass
364 $ opt < a.bc > /dev/null -mypass -debug
367 Using the ``DEBUG()`` macro instead of a home-brewed solution allows you to not
368 have to create "yet another" command line option for the debug output for your
369 pass. Note that ``DEBUG()`` macros are disabled for non-asserts builds, so they
370 do not cause a performance impact at all (for the same reason, they should also
371 not contain side-effects!).
373 One additional nice thing about the ``DEBUG()`` macro is that you can enable or
374 disable it directly in gdb. Just use "``set DebugFlag=0``" or "``set
375 DebugFlag=1``" from the gdb if the program is running. If the program hasn't
376 been started yet, you can always just run it with ``-debug``.
380 Fine grained debug info with ``DEBUG_TYPE`` and the ``-debug-only`` option
381 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
383 Sometimes you may find yourself in a situation where enabling ``-debug`` just
384 turns on **too much** information (such as when working on the code generator).
385 If you want to enable debug information with more fine-grained control, you
386 should define the ``DEBUG_TYPE`` macro and use the ``-debug-only`` option as
391 #define DEBUG_TYPE "foo"
392 DEBUG(errs() << "'foo' debug type\n");
394 #define DEBUG_TYPE "bar"
395 DEBUG(errs() << "'bar' debug type\n"));
398 Then you can run your pass like this:
402 $ opt < a.bc > /dev/null -mypass
404 $ opt < a.bc > /dev/null -mypass -debug
407 $ opt < a.bc > /dev/null -mypass -debug-only=foo
409 $ opt < a.bc > /dev/null -mypass -debug-only=bar
411 $ opt < a.bc > /dev/null -mypass -debug-only=foo,bar
415 Of course, in practice, you should only set ``DEBUG_TYPE`` at the top of a file,
416 to specify the debug type for the entire module. Be careful that you only do
417 this after including Debug.h and not around any #include of headers. Also, you
418 should use names more meaningful than "foo" and "bar", because there is no
419 system in place to ensure that names do not conflict. If two different modules
420 use the same string, they will all be turned on when the name is specified.
421 This allows, for example, all debug information for instruction scheduling to be
422 enabled with ``-debug-only=InstrSched``, even if the source lives in multiple
423 files. The name must not include a comma (,) as that is used to seperate the
424 arguments of the ``-debug-only`` option.
426 For performance reasons, -debug-only is not available in optimized build
427 (``--enable-optimized``) of LLVM.
429 The ``DEBUG_WITH_TYPE`` macro is also available for situations where you would
430 like to set ``DEBUG_TYPE``, but only for one specific ``DEBUG`` statement. It
431 takes an additional first parameter, which is the type to use. For example, the
432 preceding example could be written as:
436 DEBUG_WITH_TYPE("foo", errs() << "'foo' debug type\n");
437 DEBUG_WITH_TYPE("bar", errs() << "'bar' debug type\n"));
441 The ``Statistic`` class & ``-stats`` option
442 -------------------------------------------
444 The ``llvm/ADT/Statistic.h`` (`doxygen
445 <http://llvm.org/doxygen/Statistic_8h-source.html>`__) file provides a class
446 named ``Statistic`` that is used as a unified way to keep track of what the LLVM
447 compiler is doing and how effective various optimizations are. It is useful to
448 see what optimizations are contributing to making a particular program run
451 Often you may run your pass on some big program, and you're interested to see
452 how many times it makes a certain transformation. Although you can do this with
453 hand inspection, or some ad-hoc method, this is a real pain and not very useful
454 for big programs. Using the ``Statistic`` class makes it very easy to keep
455 track of this information, and the calculated information is presented in a
456 uniform manner with the rest of the passes being executed.
458 There are many examples of ``Statistic`` uses, but the basics of using it are as
461 #. Define your statistic like this:
465 #define DEBUG_TYPE "mypassname" // This goes before any #includes.
466 STATISTIC(NumXForms, "The # of times I did stuff");
468 The ``STATISTIC`` macro defines a static variable, whose name is specified by
469 the first argument. The pass name is taken from the ``DEBUG_TYPE`` macro, and
470 the description is taken from the second argument. The variable defined
471 ("NumXForms" in this case) acts like an unsigned integer.
473 #. Whenever you make a transformation, bump the counter:
477 ++NumXForms; // I did stuff!
479 That's all you have to do. To get '``opt``' to print out the statistics
480 gathered, use the '``-stats``' option:
484 $ opt -stats -mypassname < program.bc > /dev/null
485 ... statistics output ...
487 Note that in order to use the '``-stats``' option, LLVM must be
488 compiled with assertions enabled.
490 When running ``opt`` on a C file from the SPEC benchmark suite, it gives a
491 report that looks like this:
495 7646 bitcodewriter - Number of normal instructions
496 725 bitcodewriter - Number of oversized instructions
497 129996 bitcodewriter - Number of bitcode bytes written
498 2817 raise - Number of insts DCEd or constprop'd
499 3213 raise - Number of cast-of-self removed
500 5046 raise - Number of expression trees converted
501 75 raise - Number of other getelementptr's formed
502 138 raise - Number of load/store peepholes
503 42 deadtypeelim - Number of unused typenames removed from symtab
504 392 funcresolve - Number of varargs functions resolved
505 27 globaldce - Number of global variables removed
506 2 adce - Number of basic blocks removed
507 134 cee - Number of branches revectored
508 49 cee - Number of setcc instruction eliminated
509 532 gcse - Number of loads removed
510 2919 gcse - Number of instructions removed
511 86 indvars - Number of canonical indvars added
512 87 indvars - Number of aux indvars removed
513 25 instcombine - Number of dead inst eliminate
514 434 instcombine - Number of insts combined
515 248 licm - Number of load insts hoisted
516 1298 licm - Number of insts hoisted to a loop pre-header
517 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
518 75 mem2reg - Number of alloca's promoted
519 1444 cfgsimplify - Number of blocks simplified
521 Obviously, with so many optimizations, having a unified framework for this stuff
522 is very nice. Making your pass fit well into the framework makes it more
523 maintainable and useful.
527 Viewing graphs while debugging code
528 -----------------------------------
530 Several of the important data structures in LLVM are graphs: for example CFGs
531 made out of LLVM :ref:`BasicBlocks <BasicBlock>`, CFGs made out of LLVM
532 :ref:`MachineBasicBlocks <MachineBasicBlock>`, and :ref:`Instruction Selection
533 DAGs <SelectionDAG>`. In many cases, while debugging various parts of the
534 compiler, it is nice to instantly visualize these graphs.
536 LLVM provides several callbacks that are available in a debug build to do
537 exactly that. If you call the ``Function::viewCFG()`` method, for example, the
538 current LLVM tool will pop up a window containing the CFG for the function where
539 each basic block is a node in the graph, and each node contains the instructions
540 in the block. Similarly, there also exists ``Function::viewCFGOnly()`` (does
541 not include the instructions), the ``MachineFunction::viewCFG()`` and
542 ``MachineFunction::viewCFGOnly()``, and the ``SelectionDAG::viewGraph()``
543 methods. Within GDB, for example, you can usually use something like ``call
544 DAG.viewGraph()`` to pop up a window. Alternatively, you can sprinkle calls to
545 these functions in your code in places you want to debug.
547 Getting this to work requires a small amount of setup. On Unix systems
548 with X11, install the `graphviz <http://www.graphviz.org>`_ toolkit, and make
549 sure 'dot' and 'gv' are in your path. If you are running on Mac OS X, download
550 and install the Mac OS X `Graphviz program
551 <http://www.pixelglow.com/graphviz/>`_ and add
552 ``/Applications/Graphviz.app/Contents/MacOS/`` (or wherever you install it) to
553 your path. The programs need not be present when configuring, building or
554 running LLVM and can simply be installed when needed during an active debug
557 ``SelectionDAG`` has been extended to make it easier to locate *interesting*
558 nodes in large complex graphs. From gdb, if you ``call DAG.setGraphColor(node,
559 "color")``, then the next ``call DAG.viewGraph()`` would highlight the node in
560 the specified color (choices of colors can be found at `colors
561 <http://www.graphviz.org/doc/info/colors.html>`_.) More complex node attributes
562 can be provided with ``call DAG.setGraphAttrs(node, "attributes")`` (choices can
563 be found at `Graph attributes <http://www.graphviz.org/doc/info/attrs.html>`_.)
564 If you want to restart and clear all the current graph attributes, then you can
565 ``call DAG.clearGraphAttrs()``.
567 Note that graph visualization features are compiled out of Release builds to
568 reduce file size. This means that you need a Debug+Asserts or Release+Asserts
569 build to use these features.
573 Picking the Right Data Structure for a Task
574 ===========================================
576 LLVM has a plethora of data structures in the ``llvm/ADT/`` directory, and we
577 commonly use STL data structures. This section describes the trade-offs you
578 should consider when you pick one.
580 The first step is a choose your own adventure: do you want a sequential
581 container, a set-like container, or a map-like container? The most important
582 thing when choosing a container is the algorithmic properties of how you plan to
583 access the container. Based on that, you should use:
586 * a :ref:`map-like <ds_map>` container if you need efficient look-up of a
587 value based on another value. Map-like containers also support efficient
588 queries for containment (whether a key is in the map). Map-like containers
589 generally do not support efficient reverse mapping (values to keys). If you
590 need that, use two maps. Some map-like containers also support efficient
591 iteration through the keys in sorted order. Map-like containers are the most
592 expensive sort, only use them if you need one of these capabilities.
594 * a :ref:`set-like <ds_set>` container if you need to put a bunch of stuff into
595 a container that automatically eliminates duplicates. Some set-like
596 containers support efficient iteration through the elements in sorted order.
597 Set-like containers are more expensive than sequential containers.
599 * a :ref:`sequential <ds_sequential>` container provides the most efficient way
600 to add elements and keeps track of the order they are added to the collection.
601 They permit duplicates and support efficient iteration, but do not support
602 efficient look-up based on a key.
604 * a :ref:`string <ds_string>` container is a specialized sequential container or
605 reference structure that is used for character or byte arrays.
607 * a :ref:`bit <ds_bit>` container provides an efficient way to store and
608 perform set operations on sets of numeric id's, while automatically
609 eliminating duplicates. Bit containers require a maximum of 1 bit for each
610 identifier you want to store.
612 Once the proper category of container is determined, you can fine tune the
613 memory use, constant factors, and cache behaviors of access by intelligently
614 picking a member of the category. Note that constant factors and cache behavior
615 can be a big deal. If you have a vector that usually only contains a few
616 elements (but could contain many), for example, it's much better to use
617 :ref:`SmallVector <dss_smallvector>` than :ref:`vector <dss_vector>`. Doing so
618 avoids (relatively) expensive malloc/free calls, which dwarf the cost of adding
619 the elements to the container.
623 Sequential Containers (std::vector, std::list, etc)
624 ---------------------------------------------------
626 There are a variety of sequential containers available for you, based on your
627 needs. Pick the first in this section that will do what you want.
634 The ``llvm::ArrayRef`` class is the preferred class to use in an interface that
635 accepts a sequential list of elements in memory and just reads from them. By
636 taking an ``ArrayRef``, the API can be passed a fixed size array, an
637 ``std::vector``, an ``llvm::SmallVector`` and anything else that is contiguous
645 Fixed size arrays are very simple and very fast. They are good if you know
646 exactly how many elements you have, or you have a (low) upper bound on how many
651 Heap Allocated Arrays
652 ^^^^^^^^^^^^^^^^^^^^^
654 Heap allocated arrays (``new[]`` + ``delete[]``) are also simple. They are good
655 if the number of elements is variable, if you know how many elements you will
656 need before the array is allocated, and if the array is usually large (if not,
657 consider a :ref:`SmallVector <dss_smallvector>`). The cost of a heap allocated
658 array is the cost of the new/delete (aka malloc/free). Also note that if you
659 are allocating an array of a type with a constructor, the constructor and
660 destructors will be run for every element in the array (re-sizable vectors only
661 construct those elements actually used).
663 .. _dss_tinyptrvector:
665 llvm/ADT/TinyPtrVector.h
666 ^^^^^^^^^^^^^^^^^^^^^^^^
668 ``TinyPtrVector<Type>`` is a highly specialized collection class that is
669 optimized to avoid allocation in the case when a vector has zero or one
670 elements. It has two major restrictions: 1) it can only hold values of pointer
671 type, and 2) it cannot hold a null pointer.
673 Since this container is highly specialized, it is rarely used.
677 llvm/ADT/SmallVector.h
678 ^^^^^^^^^^^^^^^^^^^^^^
680 ``SmallVector<Type, N>`` is a simple class that looks and smells just like
681 ``vector<Type>``: it supports efficient iteration, lays out elements in memory
682 order (so you can do pointer arithmetic between elements), supports efficient
683 push_back/pop_back operations, supports efficient random access to its elements,
686 The advantage of SmallVector is that it allocates space for some number of
687 elements (N) **in the object itself**. Because of this, if the SmallVector is
688 dynamically smaller than N, no malloc is performed. This can be a big win in
689 cases where the malloc/free call is far more expensive than the code that
690 fiddles around with the elements.
692 This is good for vectors that are "usually small" (e.g. the number of
693 predecessors/successors of a block is usually less than 8). On the other hand,
694 this makes the size of the SmallVector itself large, so you don't want to
695 allocate lots of them (doing so will waste a lot of space). As such,
696 SmallVectors are most useful when on the stack.
698 SmallVector also provides a nice portable and efficient replacement for
703 Prefer to use ``SmallVectorImpl<T>`` as a parameter type.
705 In APIs that don't care about the "small size" (most?), prefer to use
706 the ``SmallVectorImpl<T>`` class, which is basically just the "vector
707 header" (and methods) without the elements allocated after it. Note that
708 ``SmallVector<T, N>`` inherits from ``SmallVectorImpl<T>`` so the
709 conversion is implicit and costs nothing. E.g.
713 // BAD: Clients cannot pass e.g. SmallVector<Foo, 4>.
714 hardcodedSmallSize(SmallVector<Foo, 2> &Out);
715 // GOOD: Clients can pass any SmallVector<Foo, N>.
716 allowsAnySmallSize(SmallVectorImpl<Foo> &Out);
719 SmallVector<Foo, 8> Vec;
720 hardcodedSmallSize(Vec); // Error.
721 allowsAnySmallSize(Vec); // Works.
724 Even though it has "``Impl``" in the name, this is so widely used that
725 it really isn't "private to the implementation" anymore. A name like
726 ``SmallVectorHeader`` would be more appropriate.
733 ``std::vector`` is well loved and respected. It is useful when SmallVector
734 isn't: when the size of the vector is often large (thus the small optimization
735 will rarely be a benefit) or if you will be allocating many instances of the
736 vector itself (which would waste space for elements that aren't in the
737 container). vector is also useful when interfacing with code that expects
740 One worthwhile note about std::vector: avoid code like this:
749 Instead, write this as:
759 Doing so will save (at least) one heap allocation and free per iteration of the
767 ``std::deque`` is, in some senses, a generalized version of ``std::vector``.
768 Like ``std::vector``, it provides constant time random access and other similar
769 properties, but it also provides efficient access to the front of the list. It
770 does not guarantee continuity of elements within memory.
772 In exchange for this extra flexibility, ``std::deque`` has significantly higher
773 constant factor costs than ``std::vector``. If possible, use ``std::vector`` or
781 ``std::list`` is an extremely inefficient class that is rarely useful. It
782 performs a heap allocation for every element inserted into it, thus having an
783 extremely high constant factor, particularly for small data types.
784 ``std::list`` also only supports bidirectional iteration, not random access
787 In exchange for this high cost, std::list supports efficient access to both ends
788 of the list (like ``std::deque``, but unlike ``std::vector`` or
789 ``SmallVector``). In addition, the iterator invalidation characteristics of
790 std::list are stronger than that of a vector class: inserting or removing an
791 element into the list does not invalidate iterator or pointers to other elements
799 ``ilist<T>`` implements an 'intrusive' doubly-linked list. It is intrusive,
800 because it requires the element to store and provide access to the prev/next
801 pointers for the list.
803 ``ilist`` has the same drawbacks as ``std::list``, and additionally requires an
804 ``ilist_traits`` implementation for the element type, but it provides some novel
805 characteristics. In particular, it can efficiently store polymorphic objects,
806 the traits class is informed when an element is inserted or removed from the
807 list, and ``ilist``\ s are guaranteed to support a constant-time splice
810 These properties are exactly what we want for things like ``Instruction``\ s and
811 basic blocks, which is why these are implemented with ``ilist``\ s.
813 Related classes of interest are explained in the following subsections:
815 * :ref:`ilist_traits <dss_ilist_traits>`
817 * :ref:`iplist <dss_iplist>`
819 * :ref:`llvm/ADT/ilist_node.h <dss_ilist_node>`
821 * :ref:`Sentinels <dss_ilist_sentinel>`
823 .. _dss_packedvector:
825 llvm/ADT/PackedVector.h
826 ^^^^^^^^^^^^^^^^^^^^^^^
828 Useful for storing a vector of values using only a few number of bits for each
829 value. Apart from the standard operations of a vector-like container, it can
830 also perform an 'or' set operation.
838 FirstCondition = 0x1,
839 SecondCondition = 0x2,
844 PackedVector<State, 2> Vec1;
845 Vec1.push_back(FirstCondition);
847 PackedVector<State, 2> Vec2;
848 Vec2.push_back(SecondCondition);
851 return Vec1[0]; // returns 'Both'.
854 .. _dss_ilist_traits:
859 ``ilist_traits<T>`` is ``ilist<T>``'s customization mechanism. ``iplist<T>``
860 (and consequently ``ilist<T>``) publicly derive from this traits class.
867 ``iplist<T>`` is ``ilist<T>``'s base and as such supports a slightly narrower
868 interface. Notably, inserters from ``T&`` are absent.
870 ``ilist_traits<T>`` is a public base of this class and can be used for a wide
871 variety of customizations.
875 llvm/ADT/ilist_node.h
876 ^^^^^^^^^^^^^^^^^^^^^
878 ``ilist_node<T>`` implements the forward and backward links that are expected
879 by the ``ilist<T>`` (and analogous containers) in the default manner.
881 ``ilist_node<T>``\ s are meant to be embedded in the node type ``T``, usually
882 ``T`` publicly derives from ``ilist_node<T>``.
884 .. _dss_ilist_sentinel:
889 ``ilist``\ s have another specialty that must be considered. To be a good
890 citizen in the C++ ecosystem, it needs to support the standard container
891 operations, such as ``begin`` and ``end`` iterators, etc. Also, the
892 ``operator--`` must work correctly on the ``end`` iterator in the case of
893 non-empty ``ilist``\ s.
895 The only sensible solution to this problem is to allocate a so-called *sentinel*
896 along with the intrusive list, which serves as the ``end`` iterator, providing
897 the back-link to the last element. However conforming to the C++ convention it
898 is illegal to ``operator++`` beyond the sentinel and it also must not be
901 These constraints allow for some implementation freedom to the ``ilist`` how to
902 allocate and store the sentinel. The corresponding policy is dictated by
903 ``ilist_traits<T>``. By default a ``T`` gets heap-allocated whenever the need
904 for a sentinel arises.
906 While the default policy is sufficient in most cases, it may break down when
907 ``T`` does not provide a default constructor. Also, in the case of many
908 instances of ``ilist``\ s, the memory overhead of the associated sentinels is
909 wasted. To alleviate the situation with numerous and voluminous
910 ``T``-sentinels, sometimes a trick is employed, leading to *ghostly sentinels*.
912 Ghostly sentinels are obtained by specially-crafted ``ilist_traits<T>`` which
913 superpose the sentinel with the ``ilist`` instance in memory. Pointer
914 arithmetic is used to obtain the sentinel, which is relative to the ``ilist``'s
915 ``this`` pointer. The ``ilist`` is augmented by an extra pointer, which serves
916 as the back-link of the sentinel. This is the only field in the ghostly
917 sentinel which can be legally accessed.
921 Other Sequential Container options
922 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
924 Other STL containers are available, such as ``std::string``.
926 There are also various STL adapter classes such as ``std::queue``,
927 ``std::priority_queue``, ``std::stack``, etc. These provide simplified access
928 to an underlying container but don't affect the cost of the container itself.
932 String-like containers
933 ----------------------
935 There are a variety of ways to pass around and use strings in C and C++, and
936 LLVM adds a few new options to choose from. Pick the first option on this list
937 that will do what you need, they are ordered according to their relative cost.
939 Note that it is generally preferred to *not* pass strings around as ``const
940 char*``'s. These have a number of problems, including the fact that they
941 cannot represent embedded nul ("\0") characters, and do not have a length
942 available efficiently. The general replacement for '``const char*``' is
945 For more information on choosing string containers for APIs, please see
946 :ref:`Passing Strings <string_apis>`.
953 The StringRef class is a simple value class that contains a pointer to a
954 character and a length, and is quite related to the :ref:`ArrayRef
955 <dss_arrayref>` class (but specialized for arrays of characters). Because
956 StringRef carries a length with it, it safely handles strings with embedded nul
957 characters in it, getting the length does not require a strlen call, and it even
958 has very convenient APIs for slicing and dicing the character range that it
961 StringRef is ideal for passing simple strings around that are known to be live,
962 either because they are C string literals, std::string, a C array, or a
963 SmallVector. Each of these cases has an efficient implicit conversion to
964 StringRef, which doesn't result in a dynamic strlen being executed.
966 StringRef has a few major limitations which make more powerful string containers
969 #. You cannot directly convert a StringRef to a 'const char*' because there is
970 no way to add a trailing nul (unlike the .c_str() method on various stronger
973 #. StringRef doesn't own or keep alive the underlying string bytes.
974 As such it can easily lead to dangling pointers, and is not suitable for
975 embedding in datastructures in most cases (instead, use an std::string or
976 something like that).
978 #. For the same reason, StringRef cannot be used as the return value of a
979 method if the method "computes" the result string. Instead, use std::string.
981 #. StringRef's do not allow you to mutate the pointed-to string bytes and it
982 doesn't allow you to insert or remove bytes from the range. For editing
983 operations like this, it interoperates with the :ref:`Twine <dss_twine>`
986 Because of its strengths and limitations, it is very common for a function to
987 take a StringRef and for a method on an object to return a StringRef that points
988 into some string that it owns.
995 The Twine class is used as an intermediary datatype for APIs that want to take a
996 string that can be constructed inline with a series of concatenations. Twine
997 works by forming recursive instances of the Twine datatype (a simple value
998 object) on the stack as temporary objects, linking them together into a tree
999 which is then linearized when the Twine is consumed. Twine is only safe to use
1000 as the argument to a function, and should always be a const reference, e.g.:
1004 void foo(const Twine &T);
1008 foo(X + "." + Twine(i));
1010 This example forms a string like "blarg.42" by concatenating the values
1011 together, and does not form intermediate strings containing "blarg" or "blarg.".
1013 Because Twine is constructed with temporary objects on the stack, and because
1014 these instances are destroyed at the end of the current statement, it is an
1015 inherently dangerous API. For example, this simple variant contains undefined
1016 behavior and will probably crash:
1020 void foo(const Twine &T);
1024 const Twine &Tmp = X + "." + Twine(i);
1027 ... because the temporaries are destroyed before the call. That said, Twine's
1028 are much more efficient than intermediate std::string temporaries, and they work
1029 really well with StringRef. Just be aware of their limitations.
1031 .. _dss_smallstring:
1033 llvm/ADT/SmallString.h
1034 ^^^^^^^^^^^^^^^^^^^^^^
1036 SmallString is a subclass of :ref:`SmallVector <dss_smallvector>` that adds some
1037 convenience APIs like += that takes StringRef's. SmallString avoids allocating
1038 memory in the case when the preallocated space is enough to hold its data, and
1039 it calls back to general heap allocation when required. Since it owns its data,
1040 it is very safe to use and supports full mutation of the string.
1042 Like SmallVector's, the big downside to SmallString is their sizeof. While they
1043 are optimized for small strings, they themselves are not particularly small.
1044 This means that they work great for temporary scratch buffers on the stack, but
1045 should not generally be put into the heap: it is very rare to see a SmallString
1046 as the member of a frequently-allocated heap data structure or returned
1054 The standard C++ std::string class is a very general class that (like
1055 SmallString) owns its underlying data. sizeof(std::string) is very reasonable
1056 so it can be embedded into heap data structures and returned by-value. On the
1057 other hand, std::string is highly inefficient for inline editing (e.g.
1058 concatenating a bunch of stuff together) and because it is provided by the
1059 standard library, its performance characteristics depend a lot of the host
1060 standard library (e.g. libc++ and MSVC provide a highly optimized string class,
1061 GCC contains a really slow implementation).
1063 The major disadvantage of std::string is that almost every operation that makes
1064 them larger can allocate memory, which is slow. As such, it is better to use
1065 SmallVector or Twine as a scratch buffer, but then use std::string to persist
1070 Set-Like Containers (std::set, SmallSet, SetVector, etc)
1071 --------------------------------------------------------
1073 Set-like containers are useful when you need to canonicalize multiple values
1074 into a single representation. There are several different choices for how to do
1075 this, providing various trade-offs.
1077 .. _dss_sortedvectorset:
1082 If you intend to insert a lot of elements, then do a lot of queries, a great
1083 approach is to use a vector (or other sequential container) with
1084 std::sort+std::unique to remove duplicates. This approach works really well if
1085 your usage pattern has these two distinct phases (insert then query), and can be
1086 coupled with a good choice of :ref:`sequential container <ds_sequential>`.
1088 This combination provides the several nice properties: the result data is
1089 contiguous in memory (good for cache locality), has few allocations, is easy to
1090 address (iterators in the final vector are just indices or pointers), and can be
1091 efficiently queried with a standard binary search (e.g.
1092 ``std::lower_bound``; if you want the whole range of elements comparing
1093 equal, use ``std::equal_range``).
1100 If you have a set-like data structure that is usually small and whose elements
1101 are reasonably small, a ``SmallSet<Type, N>`` is a good choice. This set has
1102 space for N elements in place (thus, if the set is dynamically smaller than N,
1103 no malloc traffic is required) and accesses them with a simple linear search.
1104 When the set grows beyond N elements, it allocates a more expensive
1105 representation that guarantees efficient access (for most types, it falls back
1106 to :ref:`std::set <dss_set>`, but for pointers it uses something far better,
1107 :ref:`SmallPtrSet <dss_smallptrset>`.
1109 The magic of this class is that it handles small sets extremely efficiently, but
1110 gracefully handles extremely large sets without loss of efficiency. The
1111 drawback is that the interface is quite small: it supports insertion, queries
1112 and erasing, but does not support iteration.
1114 .. _dss_smallptrset:
1116 llvm/ADT/SmallPtrSet.h
1117 ^^^^^^^^^^^^^^^^^^^^^^
1119 ``SmallPtrSet`` has all the advantages of ``SmallSet`` (and a ``SmallSet`` of
1120 pointers is transparently implemented with a ``SmallPtrSet``), but also supports
1121 iterators. If more than N insertions are performed, a single quadratically
1122 probed hash table is allocated and grows as needed, providing extremely
1123 efficient access (constant time insertion/deleting/queries with low constant
1124 factors) and is very stingy with malloc traffic.
1126 Note that, unlike :ref:`std::set <dss_set>`, the iterators of ``SmallPtrSet``
1127 are invalidated whenever an insertion occurs. Also, the values visited by the
1128 iterators are not visited in sorted order.
1132 llvm/ADT/StringSet.h
1133 ^^^^^^^^^^^^^^^^^^^^
1135 ``StringSet`` is a thin wrapper around :ref:`StringMap\<char\> <dss_stringmap>`,
1136 and it allows efficient storage and retrieval of unique strings.
1138 Functionally analogous to ``SmallSet<StringRef>``, ``StringSet`` also suports
1139 iteration. (The iterator dereferences to a ``StringMapEntry<char>``, so you
1140 need to call ``i->getKey()`` to access the item of the StringSet.) On the
1141 other hand, ``StringSet`` doesn't support range-insertion and
1142 copy-construction, which :ref:`SmallSet <dss_smallset>` and :ref:`SmallPtrSet
1143 <dss_smallptrset>` do support.
1150 DenseSet is a simple quadratically probed hash table. It excels at supporting
1151 small values: it uses a single allocation to hold all of the pairs that are
1152 currently inserted in the set. DenseSet is a great way to unique small values
1153 that are not simple pointers (use :ref:`SmallPtrSet <dss_smallptrset>` for
1154 pointers). Note that DenseSet has the same requirements for the value type that
1155 :ref:`DenseMap <dss_densemap>` has.
1159 llvm/ADT/SparseSet.h
1160 ^^^^^^^^^^^^^^^^^^^^
1162 SparseSet holds a small number of objects identified by unsigned keys of
1163 moderate size. It uses a lot of memory, but provides operations that are almost
1164 as fast as a vector. Typical keys are physical registers, virtual registers, or
1165 numbered basic blocks.
1167 SparseSet is useful for algorithms that need very fast clear/find/insert/erase
1168 and fast iteration over small sets. It is not intended for building composite
1171 .. _dss_sparsemultiset:
1173 llvm/ADT/SparseMultiSet.h
1174 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1176 SparseMultiSet adds multiset behavior to SparseSet, while retaining SparseSet's
1177 desirable attributes. Like SparseSet, it typically uses a lot of memory, but
1178 provides operations that are almost as fast as a vector. Typical keys are
1179 physical registers, virtual registers, or numbered basic blocks.
1181 SparseMultiSet is useful for algorithms that need very fast
1182 clear/find/insert/erase of the entire collection, and iteration over sets of
1183 elements sharing a key. It is often a more efficient choice than using composite
1184 data structures (e.g. vector-of-vectors, map-of-vectors). It is not intended for
1185 building composite data structures.
1189 llvm/ADT/FoldingSet.h
1190 ^^^^^^^^^^^^^^^^^^^^^
1192 FoldingSet is an aggregate class that is really good at uniquing
1193 expensive-to-create or polymorphic objects. It is a combination of a chained
1194 hash table with intrusive links (uniqued objects are required to inherit from
1195 FoldingSetNode) that uses :ref:`SmallVector <dss_smallvector>` as part of its ID
1198 Consider a case where you want to implement a "getOrCreateFoo" method for a
1199 complex object (for example, a node in the code generator). The client has a
1200 description of **what** it wants to generate (it knows the opcode and all the
1201 operands), but we don't want to 'new' a node, then try inserting it into a set
1202 only to find out it already exists, at which point we would have to delete it
1203 and return the node that already exists.
1205 To support this style of client, FoldingSet perform a query with a
1206 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1207 element that we want to query for. The query either returns the element
1208 matching the ID or it returns an opaque ID that indicates where insertion should
1209 take place. Construction of the ID usually does not require heap traffic.
1211 Because FoldingSet uses intrusive links, it can support polymorphic objects in
1212 the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1213 Because the elements are individually allocated, pointers to the elements are
1214 stable: inserting or removing elements does not invalidate any pointers to other
1222 ``std::set`` is a reasonable all-around set class, which is decent at many
1223 things but great at nothing. std::set allocates memory for each element
1224 inserted (thus it is very malloc intensive) and typically stores three pointers
1225 per element in the set (thus adding a large amount of per-element space
1226 overhead). It offers guaranteed log(n) performance, which is not particularly
1227 fast from a complexity standpoint (particularly if the elements of the set are
1228 expensive to compare, like strings), and has extremely high constant factors for
1229 lookup, insertion and removal.
1231 The advantages of std::set are that its iterators are stable (deleting or
1232 inserting an element from the set does not affect iterators or pointers to other
1233 elements) and that iteration over the set is guaranteed to be in sorted order.
1234 If the elements in the set are large, then the relative overhead of the pointers
1235 and malloc traffic is not a big deal, but if the elements of the set are small,
1236 std::set is almost never a good choice.
1240 llvm/ADT/SetVector.h
1241 ^^^^^^^^^^^^^^^^^^^^
1243 LLVM's ``SetVector<Type>`` is an adapter class that combines your choice of a
1244 set-like container along with a :ref:`Sequential Container <ds_sequential>` The
1245 important property that this provides is efficient insertion with uniquing
1246 (duplicate elements are ignored) with iteration support. It implements this by
1247 inserting elements into both a set-like container and the sequential container,
1248 using the set-like container for uniquing and the sequential container for
1251 The difference between SetVector and other sets is that the order of iteration
1252 is guaranteed to match the order of insertion into the SetVector. This property
1253 is really important for things like sets of pointers. Because pointer values
1254 are non-deterministic (e.g. vary across runs of the program on different
1255 machines), iterating over the pointers in the set will not be in a well-defined
1258 The drawback of SetVector is that it requires twice as much space as a normal
1259 set and has the sum of constant factors from the set-like container and the
1260 sequential container that it uses. Use it **only** if you need to iterate over
1261 the elements in a deterministic order. SetVector is also expensive to delete
1262 elements out of (linear time), unless you use its "pop_back" method, which is
1265 ``SetVector`` is an adapter class that defaults to using ``std::vector`` and a
1266 size 16 ``SmallSet`` for the underlying containers, so it is quite expensive.
1267 However, ``"llvm/ADT/SetVector.h"`` also provides a ``SmallSetVector`` class,
1268 which defaults to using a ``SmallVector`` and ``SmallSet`` of a specified size.
1269 If you use this, and if your sets are dynamically smaller than ``N``, you will
1270 save a lot of heap traffic.
1272 .. _dss_uniquevector:
1274 llvm/ADT/UniqueVector.h
1275 ^^^^^^^^^^^^^^^^^^^^^^^
1277 UniqueVector is similar to :ref:`SetVector <dss_setvector>` but it retains a
1278 unique ID for each element inserted into the set. It internally contains a map
1279 and a vector, and it assigns a unique ID for each value inserted into the set.
1281 UniqueVector is very expensive: its cost is the sum of the cost of maintaining
1282 both the map and vector, it has high complexity, high constant factors, and
1283 produces a lot of malloc traffic. It should be avoided.
1285 .. _dss_immutableset:
1287 llvm/ADT/ImmutableSet.h
1288 ^^^^^^^^^^^^^^^^^^^^^^^
1290 ImmutableSet is an immutable (functional) set implementation based on an AVL
1291 tree. Adding or removing elements is done through a Factory object and results
1292 in the creation of a new ImmutableSet object. If an ImmutableSet already exists
1293 with the given contents, then the existing one is returned; equality is compared
1294 with a FoldingSetNodeID. The time and space complexity of add or remove
1295 operations is logarithmic in the size of the original set.
1297 There is no method for returning an element of the set, you can only check for
1302 Other Set-Like Container Options
1303 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1305 The STL provides several other options, such as std::multiset and the various
1306 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1307 never use hash_set and unordered_set because they are generally very expensive
1308 (each insertion requires a malloc) and very non-portable.
1310 std::multiset is useful if you're not interested in elimination of duplicates,
1311 but has all the drawbacks of :ref:`std::set <dss_set>`. A sorted vector
1312 (where you don't delete duplicate entries) or some other approach is almost
1317 Map-Like Containers (std::map, DenseMap, etc)
1318 ---------------------------------------------
1320 Map-like containers are useful when you want to associate data to a key. As
1321 usual, there are a lot of different ways to do this. :)
1323 .. _dss_sortedvectormap:
1328 If your usage pattern follows a strict insert-then-query approach, you can
1329 trivially use the same approach as :ref:`sorted vectors for set-like containers
1330 <dss_sortedvectorset>`. The only difference is that your query function (which
1331 uses std::lower_bound to get efficient log(n) lookup) should only compare the
1332 key, not both the key and value. This yields the same advantages as sorted
1337 llvm/ADT/StringMap.h
1338 ^^^^^^^^^^^^^^^^^^^^
1340 Strings are commonly used as keys in maps, and they are difficult to support
1341 efficiently: they are variable length, inefficient to hash and compare when
1342 long, expensive to copy, etc. StringMap is a specialized container designed to
1343 cope with these issues. It supports mapping an arbitrary range of bytes to an
1344 arbitrary other object.
1346 The StringMap implementation uses a quadratically-probed hash table, where the
1347 buckets store a pointer to the heap allocated entries (and some other stuff).
1348 The entries in the map must be heap allocated because the strings are variable
1349 length. The string data (key) and the element object (value) are stored in the
1350 same allocation with the string data immediately after the element object.
1351 This container guarantees the "``(char*)(&Value+1)``" points to the key string
1354 The StringMap is very fast for several reasons: quadratic probing is very cache
1355 efficient for lookups, the hash value of strings in buckets is not recomputed
1356 when looking up an element, StringMap rarely has to touch the memory for
1357 unrelated objects when looking up a value (even when hash collisions happen),
1358 hash table growth does not recompute the hash values for strings already in the
1359 table, and each pair in the map is store in a single allocation (the string data
1360 is stored in the same allocation as the Value of a pair).
1362 StringMap also provides query methods that take byte ranges, so it only ever
1363 copies a string if a value is inserted into the table.
1365 StringMap iteratation order, however, is not guaranteed to be deterministic, so
1366 any uses which require that should instead use a std::map.
1370 llvm/ADT/IndexedMap.h
1371 ^^^^^^^^^^^^^^^^^^^^^
1373 IndexedMap is a specialized container for mapping small dense integers (or
1374 values that can be mapped to small dense integers) to some other type. It is
1375 internally implemented as a vector with a mapping function that maps the keys
1376 to the dense integer range.
1378 This is useful for cases like virtual registers in the LLVM code generator: they
1379 have a dense mapping that is offset by a compile-time constant (the first
1380 virtual register ID).
1387 DenseMap is a simple quadratically probed hash table. It excels at supporting
1388 small keys and values: it uses a single allocation to hold all of the pairs
1389 that are currently inserted in the map. DenseMap is a great way to map
1390 pointers to pointers, or map other small types to each other.
1392 There are several aspects of DenseMap that you should be aware of, however.
1393 The iterators in a DenseMap are invalidated whenever an insertion occurs,
1394 unlike map. Also, because DenseMap allocates space for a large number of
1395 key/value pairs (it starts with 64 by default), it will waste a lot of space if
1396 your keys or values are large. Finally, you must implement a partial
1397 specialization of DenseMapInfo for the key that you want, if it isn't already
1398 supported. This is required to tell DenseMap about two special marker values
1399 (which can never be inserted into the map) that it needs internally.
1401 DenseMap's find_as() method supports lookup operations using an alternate key
1402 type. This is useful in cases where the normal key type is expensive to
1403 construct, but cheap to compare against. The DenseMapInfo is responsible for
1404 defining the appropriate comparison and hashing methods for each alternate key
1412 ValueMap is a wrapper around a :ref:`DenseMap <dss_densemap>` mapping
1413 ``Value*``\ s (or subclasses) to another type. When a Value is deleted or
1414 RAUW'ed, ValueMap will update itself so the new version of the key is mapped to
1415 the same value, just as if the key were a WeakVH. You can configure exactly how
1416 this happens, and what else happens on these two events, by passing a ``Config``
1417 parameter to the ValueMap template.
1419 .. _dss_intervalmap:
1421 llvm/ADT/IntervalMap.h
1422 ^^^^^^^^^^^^^^^^^^^^^^
1424 IntervalMap is a compact map for small keys and values. It maps key intervals
1425 instead of single keys, and it will automatically coalesce adjacent intervals.
1426 When the map only contains a few intervals, they are stored in the map object
1427 itself to avoid allocations.
1429 The IntervalMap iterators are quite big, so they should not be passed around as
1430 STL iterators. The heavyweight iterators allow a smaller data structure.
1437 std::map has similar characteristics to :ref:`std::set <dss_set>`: it uses a
1438 single allocation per pair inserted into the map, it offers log(n) lookup with
1439 an extremely large constant factor, imposes a space penalty of 3 pointers per
1440 pair in the map, etc.
1442 std::map is most useful when your keys or values are very large, if you need to
1443 iterate over the collection in sorted order, or if you need stable iterators
1444 into the map (i.e. they don't get invalidated if an insertion or deletion of
1445 another element takes place).
1449 llvm/ADT/MapVector.h
1450 ^^^^^^^^^^^^^^^^^^^^
1452 ``MapVector<KeyT,ValueT>`` provides a subset of the DenseMap interface. The
1453 main difference is that the iteration order is guaranteed to be the insertion
1454 order, making it an easy (but somewhat expensive) solution for non-deterministic
1455 iteration over maps of pointers.
1457 It is implemented by mapping from key to an index in a vector of key,value
1458 pairs. This provides fast lookup and iteration, but has two main drawbacks:
1459 the key is stored twice and removing elements takes linear time. If it is
1460 necessary to remove elements, it's best to remove them in bulk using
1463 .. _dss_inteqclasses:
1465 llvm/ADT/IntEqClasses.h
1466 ^^^^^^^^^^^^^^^^^^^^^^^
1468 IntEqClasses provides a compact representation of equivalence classes of small
1469 integers. Initially, each integer in the range 0..n-1 has its own equivalence
1470 class. Classes can be joined by passing two class representatives to the
1471 join(a, b) method. Two integers are in the same class when findLeader() returns
1472 the same representative.
1474 Once all equivalence classes are formed, the map can be compressed so each
1475 integer 0..n-1 maps to an equivalence class number in the range 0..m-1, where m
1476 is the total number of equivalence classes. The map must be uncompressed before
1477 it can be edited again.
1479 .. _dss_immutablemap:
1481 llvm/ADT/ImmutableMap.h
1482 ^^^^^^^^^^^^^^^^^^^^^^^
1484 ImmutableMap is an immutable (functional) map implementation based on an AVL
1485 tree. Adding or removing elements is done through a Factory object and results
1486 in the creation of a new ImmutableMap object. If an ImmutableMap already exists
1487 with the given key set, then the existing one is returned; equality is compared
1488 with a FoldingSetNodeID. The time and space complexity of add or remove
1489 operations is logarithmic in the size of the original map.
1493 Other Map-Like Container Options
1494 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1496 The STL provides several other options, such as std::multimap and the various
1497 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1498 never use hash_set and unordered_set because they are generally very expensive
1499 (each insertion requires a malloc) and very non-portable.
1501 std::multimap is useful if you want to map a key to multiple values, but has all
1502 the drawbacks of std::map. A sorted vector or some other approach is almost
1507 Bit storage containers (BitVector, SparseBitVector)
1508 ---------------------------------------------------
1510 Unlike the other containers, there are only two bit storage containers, and
1511 choosing when to use each is relatively straightforward.
1513 One additional option is ``std::vector<bool>``: we discourage its use for two
1514 reasons 1) the implementation in many common compilers (e.g. commonly
1515 available versions of GCC) is extremely inefficient and 2) the C++ standards
1516 committee is likely to deprecate this container and/or change it significantly
1517 somehow. In any case, please don't use it.
1524 The BitVector container provides a dynamic size set of bits for manipulation.
1525 It supports individual bit setting/testing, as well as set operations. The set
1526 operations take time O(size of bitvector), but operations are performed one word
1527 at a time, instead of one bit at a time. This makes the BitVector very fast for
1528 set operations compared to other containers. Use the BitVector when you expect
1529 the number of set bits to be high (i.e. a dense set).
1531 .. _dss_smallbitvector:
1536 The SmallBitVector container provides the same interface as BitVector, but it is
1537 optimized for the case where only a small number of bits, less than 25 or so,
1538 are needed. It also transparently supports larger bit counts, but slightly less
1539 efficiently than a plain BitVector, so SmallBitVector should only be used when
1540 larger counts are rare.
1542 At this time, SmallBitVector does not support set operations (and, or, xor), and
1543 its operator[] does not provide an assignable lvalue.
1545 .. _dss_sparsebitvector:
1550 The SparseBitVector container is much like BitVector, with one major difference:
1551 Only the bits that are set, are stored. This makes the SparseBitVector much
1552 more space efficient than BitVector when the set is sparse, as well as making
1553 set operations O(number of set bits) instead of O(size of universe). The
1554 downside to the SparseBitVector is that setting and testing of random bits is
1555 O(N), and on large SparseBitVectors, this can be slower than BitVector. In our
1556 implementation, setting or testing bits in sorted order (either forwards or
1557 reverse) is O(1) worst case. Testing and setting bits within 128 bits (depends
1558 on size) of the current bit is also O(1). As a general statement,
1559 testing/setting bits in a SparseBitVector is O(distance away from last set bit).
1563 Helpful Hints for Common Operations
1564 ===================================
1566 This section describes how to perform some very simple transformations of LLVM
1567 code. This is meant to give examples of common idioms used, showing the
1568 practical side of LLVM transformations.
1570 Because this is a "how-to" section, you should also read about the main classes
1571 that you will be working with. The :ref:`Core LLVM Class Hierarchy Reference
1572 <coreclasses>` contains details and descriptions of the main classes that you
1577 Basic Inspection and Traversal Routines
1578 ---------------------------------------
1580 The LLVM compiler infrastructure have many different data structures that may be
1581 traversed. Following the example of the C++ standard template library, the
1582 techniques used to traverse these various data structures are all basically the
1583 same. For a enumerable sequence of values, the ``XXXbegin()`` function (or
1584 method) returns an iterator to the start of the sequence, the ``XXXend()``
1585 function returns an iterator pointing to one past the last valid element of the
1586 sequence, and there is some ``XXXiterator`` data type that is common between the
1589 Because the pattern for iteration is common across many different aspects of the
1590 program representation, the standard template library algorithms may be used on
1591 them, and it is easier to remember how to iterate. First we show a few common
1592 examples of the data structures that need to be traversed. Other data
1593 structures are traversed in very similar ways.
1595 .. _iterate_function:
1597 Iterating over the ``BasicBlock`` in a ``Function``
1598 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1600 It's quite common to have a ``Function`` instance that you'd like to transform
1601 in some way; in particular, you'd like to manipulate its ``BasicBlock``\ s. To
1602 facilitate this, you'll need to iterate over all of the ``BasicBlock``\ s that
1603 constitute the ``Function``. The following is an example that prints the name
1604 of a ``BasicBlock`` and the number of ``Instruction``\ s it contains:
1608 // func is a pointer to a Function instance
1609 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
1610 // Print out the name of the basic block if it has one, and then the
1611 // number of instructions that it contains
1612 errs() << "Basic block (name=" << i->getName() << ") has "
1613 << i->size() << " instructions.\n";
1615 Note that i can be used as if it were a pointer for the purposes of invoking
1616 member functions of the ``Instruction`` class. This is because the indirection
1617 operator is overloaded for the iterator classes. In the above code, the
1618 expression ``i->size()`` is exactly equivalent to ``(*i).size()`` just like
1621 .. _iterate_basicblock:
1623 Iterating over the ``Instruction`` in a ``BasicBlock``
1624 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1626 Just like when dealing with ``BasicBlock``\ s in ``Function``\ s, it's easy to
1627 iterate over the individual instructions that make up ``BasicBlock``\ s. Here's
1628 a code snippet that prints out each instruction in a ``BasicBlock``:
1632 // blk is a pointer to a BasicBlock instance
1633 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
1634 // The next statement works since operator<<(ostream&,...)
1635 // is overloaded for Instruction&
1636 errs() << *i << "\n";
1639 However, this isn't really the best way to print out the contents of a
1640 ``BasicBlock``! Since the ostream operators are overloaded for virtually
1641 anything you'll care about, you could have just invoked the print routine on the
1642 basic block itself: ``errs() << *blk << "\n";``.
1644 .. _iterate_insiter:
1646 Iterating over the ``Instruction`` in a ``Function``
1647 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1649 If you're finding that you commonly iterate over a ``Function``'s
1650 ``BasicBlock``\ s and then that ``BasicBlock``'s ``Instruction``\ s,
1651 ``InstIterator`` should be used instead. You'll need to include
1652 ``llvm/IR/InstIterator.h`` (`doxygen
1653 <http://llvm.org/doxygen/InstIterator_8h.html>`__) and then instantiate
1654 ``InstIterator``\ s explicitly in your code. Here's a small example that shows
1655 how to dump all instructions in a function to the standard error stream:
1659 #include "llvm/IR/InstIterator.h"
1661 // F is a pointer to a Function instance
1662 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1663 errs() << *I << "\n";
1665 Easy, isn't it? You can also use ``InstIterator``\ s to fill a work list with
1666 its initial contents. For example, if you wanted to initialize a work list to
1667 contain all instructions in a ``Function`` F, all you would need to do is
1672 std::set<Instruction*> worklist;
1673 // or better yet, SmallPtrSet<Instruction*, 64> worklist;
1675 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1676 worklist.insert(&*I);
1678 The STL set ``worklist`` would now contain all instructions in the ``Function``
1681 .. _iterate_convert:
1683 Turning an iterator into a class pointer (and vice-versa)
1684 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1686 Sometimes, it'll be useful to grab a reference (or pointer) to a class instance
1687 when all you've got at hand is an iterator. Well, extracting a reference or a
1688 pointer from an iterator is very straight-forward. Assuming that ``i`` is a
1689 ``BasicBlock::iterator`` and ``j`` is a ``BasicBlock::const_iterator``:
1693 Instruction& inst = *i; // Grab reference to instruction reference
1694 Instruction* pinst = &*i; // Grab pointer to instruction reference
1695 const Instruction& inst = *j;
1697 However, the iterators you'll be working with in the LLVM framework are special:
1698 they will automatically convert to a ptr-to-instance type whenever they need to.
1699 Instead of derferencing the iterator and then taking the address of the result,
1700 you can simply assign the iterator to the proper pointer type and you get the
1701 dereference and address-of operation as a result of the assignment (behind the
1702 scenes, this is a result of overloading casting mechanisms). Thus the second
1703 line of the last example,
1707 Instruction *pinst = &*i;
1709 is semantically equivalent to
1713 Instruction *pinst = i;
1715 It's also possible to turn a class pointer into the corresponding iterator, and
1716 this is a constant time operation (very efficient). The following code snippet
1717 illustrates use of the conversion constructors provided by LLVM iterators. By
1718 using these, you can explicitly grab the iterator of something without actually
1719 obtaining it via iteration over some structure:
1723 void printNextInstruction(Instruction* inst) {
1724 BasicBlock::iterator it(inst);
1725 ++it; // After this line, it refers to the instruction after *inst
1726 if (it != inst->getParent()->end()) errs() << *it << "\n";
1729 Unfortunately, these implicit conversions come at a cost; they prevent these
1730 iterators from conforming to standard iterator conventions, and thus from being
1731 usable with standard algorithms and containers. For example, they prevent the
1732 following code, where ``B`` is a ``BasicBlock``, from compiling:
1736 llvm::SmallVector<llvm::Instruction *, 16>(B->begin(), B->end());
1738 Because of this, these implicit conversions may be removed some day, and
1739 ``operator*`` changed to return a pointer instead of a reference.
1741 .. _iterate_complex:
1743 Finding call sites: a slightly more complex example
1744 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1746 Say that you're writing a FunctionPass and would like to count all the locations
1747 in the entire module (that is, across every ``Function``) where a certain
1748 function (i.e., some ``Function *``) is already in scope. As you'll learn
1749 later, you may want to use an ``InstVisitor`` to accomplish this in a much more
1750 straight-forward manner, but this example will allow us to explore how you'd do
1751 it if you didn't have ``InstVisitor`` around. In pseudo-code, this is what we
1754 .. code-block:: none
1756 initialize callCounter to zero
1757 for each Function f in the Module
1758 for each BasicBlock b in f
1759 for each Instruction i in b
1760 if (i is a CallInst and calls the given function)
1761 increment callCounter
1763 And the actual code is (remember, because we're writing a ``FunctionPass``, our
1764 ``FunctionPass``-derived class simply has to override the ``runOnFunction``
1769 Function* targetFunc = ...;
1771 class OurFunctionPass : public FunctionPass {
1773 OurFunctionPass(): callCounter(0) { }
1775 virtual runOnFunction(Function& F) {
1776 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1777 for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) {
1778 if (CallInst* callInst = dyn_cast<CallInst>(&*i)) {
1779 // We know we've encountered a call instruction, so we
1780 // need to determine if it's a call to the
1781 // function pointed to by m_func or not.
1782 if (callInst->getCalledFunction() == targetFunc)
1790 unsigned callCounter;
1793 .. _calls_and_invokes:
1795 Treating calls and invokes the same way
1796 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1798 You may have noticed that the previous example was a bit oversimplified in that
1799 it did not deal with call sites generated by 'invoke' instructions. In this,
1800 and in other situations, you may find that you want to treat ``CallInst``\ s and
1801 ``InvokeInst``\ s the same way, even though their most-specific common base
1802 class is ``Instruction``, which includes lots of less closely-related things.
1803 For these cases, LLVM provides a handy wrapper class called ``CallSite``
1804 (`doxygen <http://llvm.org/doxygen/classllvm_1_1CallSite.html>`__) It is
1805 essentially a wrapper around an ``Instruction`` pointer, with some methods that
1806 provide functionality common to ``CallInst``\ s and ``InvokeInst``\ s.
1808 This class has "value semantics": it should be passed by value, not by reference
1809 and it should not be dynamically allocated or deallocated using ``operator new``
1810 or ``operator delete``. It is efficiently copyable, assignable and
1811 constructable, with costs equivalents to that of a bare pointer. If you look at
1812 its definition, it has only a single pointer member.
1816 Iterating over def-use & use-def chains
1817 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1819 Frequently, we might have an instance of the ``Value`` class (`doxygen
1820 <http://llvm.org/doxygen/classllvm_1_1Value.html>`__) and we want to determine
1821 which ``User`` s use the ``Value``. The list of all ``User``\ s of a particular
1822 ``Value`` is called a *def-use* chain. For example, let's say we have a
1823 ``Function*`` named ``F`` to a particular function ``foo``. Finding all of the
1824 instructions that *use* ``foo`` is as simple as iterating over the *def-use*
1831 for (User *U : F->users()) {
1832 if (Instruction *Inst = dyn_cast<Instruction>(U)) {
1833 errs() << "F is used in instruction:\n";
1834 errs() << *Inst << "\n";
1837 Alternatively, it's common to have an instance of the ``User`` Class (`doxygen
1838 <http://llvm.org/doxygen/classllvm_1_1User.html>`__) and need to know what
1839 ``Value``\ s are used by it. The list of all ``Value``\ s used by a ``User`` is
1840 known as a *use-def* chain. Instances of class ``Instruction`` are common
1841 ``User`` s, so we might want to iterate over all of the values that a particular
1842 instruction uses (that is, the operands of the particular ``Instruction``):
1846 Instruction *pi = ...;
1848 for (Use &U : pi->operands()) {
1853 Declaring objects as ``const`` is an important tool of enforcing mutation free
1854 algorithms (such as analyses, etc.). For this purpose above iterators come in
1855 constant flavors as ``Value::const_use_iterator`` and
1856 ``Value::const_op_iterator``. They automatically arise when calling
1857 ``use/op_begin()`` on ``const Value*``\ s or ``const User*``\ s respectively.
1858 Upon dereferencing, they return ``const Use*``\ s. Otherwise the above patterns
1863 Iterating over predecessors & successors of blocks
1864 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1866 Iterating over the predecessors and successors of a block is quite easy with the
1867 routines defined in ``"llvm/IR/CFG.h"``. Just use code like this to
1868 iterate over all predecessors of BB:
1872 #include "llvm/Support/CFG.h"
1873 BasicBlock *BB = ...;
1875 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
1876 BasicBlock *Pred = *PI;
1880 Similarly, to iterate over successors use ``succ_iterator/succ_begin/succ_end``.
1884 Making simple changes
1885 ---------------------
1887 There are some primitive transformation operations present in the LLVM
1888 infrastructure that are worth knowing about. When performing transformations,
1889 it's fairly common to manipulate the contents of basic blocks. This section
1890 describes some of the common methods for doing so and gives example code.
1892 .. _schanges_creating:
1894 Creating and inserting new ``Instruction``\ s
1895 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1897 *Instantiating Instructions*
1899 Creation of ``Instruction``\ s is straight-forward: simply call the constructor
1900 for the kind of instruction to instantiate and provide the necessary parameters.
1901 For example, an ``AllocaInst`` only *requires* a (const-ptr-to) ``Type``. Thus:
1905 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
1907 will create an ``AllocaInst`` instance that represents the allocation of one
1908 integer in the current stack frame, at run time. Each ``Instruction`` subclass
1909 is likely to have varying default parameters which change the semantics of the
1910 instruction, so refer to the `doxygen documentation for the subclass of
1911 Instruction <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_ that
1912 you're interested in instantiating.
1916 It is very useful to name the values of instructions when you're able to, as
1917 this facilitates the debugging of your transformations. If you end up looking
1918 at generated LLVM machine code, you definitely want to have logical names
1919 associated with the results of instructions! By supplying a value for the
1920 ``Name`` (default) parameter of the ``Instruction`` constructor, you associate a
1921 logical name with the result of the instruction's execution at run time. For
1922 example, say that I'm writing a transformation that dynamically allocates space
1923 for an integer on the stack, and that integer is going to be used as some kind
1924 of index by some other code. To accomplish this, I place an ``AllocaInst`` at
1925 the first point in the first ``BasicBlock`` of some ``Function``, and I'm
1926 intending to use it within the same ``Function``. I might do:
1930 AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
1932 where ``indexLoc`` is now the logical name of the instruction's execution value,
1933 which is a pointer to an integer on the run time stack.
1935 *Inserting instructions*
1937 There are essentially three ways to insert an ``Instruction`` into an existing
1938 sequence of instructions that form a ``BasicBlock``:
1940 * Insertion into an explicit instruction list
1942 Given a ``BasicBlock* pb``, an ``Instruction* pi`` within that ``BasicBlock``,
1943 and a newly-created instruction we wish to insert before ``*pi``, we do the
1948 BasicBlock *pb = ...;
1949 Instruction *pi = ...;
1950 Instruction *newInst = new Instruction(...);
1952 pb->getInstList().insert(pi, newInst); // Inserts newInst before pi in pb
1954 Appending to the end of a ``BasicBlock`` is so common that the ``Instruction``
1955 class and ``Instruction``-derived classes provide constructors which take a
1956 pointer to a ``BasicBlock`` to be appended to. For example code that looked
1961 BasicBlock *pb = ...;
1962 Instruction *newInst = new Instruction(...);
1964 pb->getInstList().push_back(newInst); // Appends newInst to pb
1970 BasicBlock *pb = ...;
1971 Instruction *newInst = new Instruction(..., pb);
1973 which is much cleaner, especially if you are creating long instruction
1976 * Insertion into an implicit instruction list
1978 ``Instruction`` instances that are already in ``BasicBlock``\ s are implicitly
1979 associated with an existing instruction list: the instruction list of the
1980 enclosing basic block. Thus, we could have accomplished the same thing as the
1981 above code without being given a ``BasicBlock`` by doing:
1985 Instruction *pi = ...;
1986 Instruction *newInst = new Instruction(...);
1988 pi->getParent()->getInstList().insert(pi, newInst);
1990 In fact, this sequence of steps occurs so frequently that the ``Instruction``
1991 class and ``Instruction``-derived classes provide constructors which take (as
1992 a default parameter) a pointer to an ``Instruction`` which the newly-created
1993 ``Instruction`` should precede. That is, ``Instruction`` constructors are
1994 capable of inserting the newly-created instance into the ``BasicBlock`` of a
1995 provided instruction, immediately before that instruction. Using an
1996 ``Instruction`` constructor with a ``insertBefore`` (default) parameter, the
2001 Instruction* pi = ...;
2002 Instruction* newInst = new Instruction(..., pi);
2004 which is much cleaner, especially if you're creating a lot of instructions and
2005 adding them to ``BasicBlock``\ s.
2007 * Insertion using an instance of ``IRBuilder``
2009 Inserting several ``Instruction``\ s can be quite laborious using the previous
2010 methods. The ``IRBuilder`` is a convenience class that can be used to add
2011 several instructions to the end of a ``BasicBlock`` or before a particular
2012 ``Instruction``. It also supports constant folding and renaming named
2013 registers (see ``IRBuilder``'s template arguments).
2015 The example below demonstrates a very simple use of the ``IRBuilder`` where
2016 three instructions are inserted before the instruction ``pi``. The first two
2017 instructions are Call instructions and third instruction multiplies the return
2018 value of the two calls.
2022 Instruction *pi = ...;
2023 IRBuilder<> Builder(pi);
2024 CallInst* callOne = Builder.CreateCall(...);
2025 CallInst* callTwo = Builder.CreateCall(...);
2026 Value* result = Builder.CreateMul(callOne, callTwo);
2028 The example below is similar to the above example except that the created
2029 ``IRBuilder`` inserts instructions at the end of the ``BasicBlock`` ``pb``.
2033 BasicBlock *pb = ...;
2034 IRBuilder<> Builder(pb);
2035 CallInst* callOne = Builder.CreateCall(...);
2036 CallInst* callTwo = Builder.CreateCall(...);
2037 Value* result = Builder.CreateMul(callOne, callTwo);
2039 See :doc:`tutorial/LangImpl3` for a practical use of the ``IRBuilder``.
2042 .. _schanges_deleting:
2044 Deleting Instructions
2045 ^^^^^^^^^^^^^^^^^^^^^
2047 Deleting an instruction from an existing sequence of instructions that form a
2048 BasicBlock_ is very straight-forward: just call the instruction's
2049 ``eraseFromParent()`` method. For example:
2053 Instruction *I = .. ;
2054 I->eraseFromParent();
2056 This unlinks the instruction from its containing basic block and deletes it. If
2057 you'd just like to unlink the instruction from its containing basic block but
2058 not delete it, you can use the ``removeFromParent()`` method.
2060 .. _schanges_replacing:
2062 Replacing an Instruction with another Value
2063 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2065 Replacing individual instructions
2066 """""""""""""""""""""""""""""""""
2068 Including "`llvm/Transforms/Utils/BasicBlockUtils.h
2069 <http://llvm.org/doxygen/BasicBlockUtils_8h-source.html>`_" permits use of two
2070 very useful replace functions: ``ReplaceInstWithValue`` and
2071 ``ReplaceInstWithInst``.
2073 .. _schanges_deleting_sub:
2075 Deleting Instructions
2076 """""""""""""""""""""
2078 * ``ReplaceInstWithValue``
2080 This function replaces all uses of a given instruction with a value, and then
2081 removes the original instruction. The following example illustrates the
2082 replacement of the result of a particular ``AllocaInst`` that allocates memory
2083 for a single integer with a null pointer to an integer.
2087 AllocaInst* instToReplace = ...;
2088 BasicBlock::iterator ii(instToReplace);
2090 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
2091 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2093 * ``ReplaceInstWithInst``
2095 This function replaces a particular instruction with another instruction,
2096 inserting the new instruction into the basic block at the location where the
2097 old instruction was, and replacing any uses of the old instruction with the
2098 new instruction. The following example illustrates the replacement of one
2099 ``AllocaInst`` with another.
2103 AllocaInst* instToReplace = ...;
2104 BasicBlock::iterator ii(instToReplace);
2106 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
2107 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
2110 Replacing multiple uses of Users and Values
2111 """""""""""""""""""""""""""""""""""""""""""
2113 You can use ``Value::replaceAllUsesWith`` and ``User::replaceUsesOfWith`` to
2114 change more than one use at a time. See the doxygen documentation for the
2115 `Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_ and `User Class
2116 <http://llvm.org/doxygen/classllvm_1_1User.html>`_, respectively, for more
2119 .. _schanges_deletingGV:
2121 Deleting GlobalVariables
2122 ^^^^^^^^^^^^^^^^^^^^^^^^
2124 Deleting a global variable from a module is just as easy as deleting an
2125 Instruction. First, you must have a pointer to the global variable that you
2126 wish to delete. You use this pointer to erase it from its parent, the module.
2131 GlobalVariable *GV = .. ;
2133 GV->eraseFromParent();
2141 In generating IR, you may need some complex types. If you know these types
2142 statically, you can use ``TypeBuilder<...>::get()``, defined in
2143 ``llvm/Support/TypeBuilder.h``, to retrieve them. ``TypeBuilder`` has two forms
2144 depending on whether you're building types for cross-compilation or native
2145 library use. ``TypeBuilder<T, true>`` requires that ``T`` be independent of the
2146 host environment, meaning that it's built out of types from the ``llvm::types``
2147 (`doxygen <http://llvm.org/doxygen/namespacellvm_1_1types.html>`__) namespace
2148 and pointers, functions, arrays, etc. built of those. ``TypeBuilder<T, false>``
2149 additionally allows native C types whose size may depend on the host compiler.
2154 FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get();
2156 is easier to read and write than the equivalent
2160 std::vector<const Type*> params;
2161 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2162 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2164 See the `class comment
2165 <http://llvm.org/doxygen/TypeBuilder_8h-source.html#l00001>`_ for more details.
2172 This section describes the interaction of the LLVM APIs with multithreading,
2173 both on the part of client applications, and in the JIT, in the hosted
2176 Note that LLVM's support for multithreading is still relatively young. Up
2177 through version 2.5, the execution of threaded hosted applications was
2178 supported, but not threaded client access to the APIs. While this use case is
2179 now supported, clients *must* adhere to the guidelines specified below to ensure
2180 proper operation in multithreaded mode.
2182 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2183 intrinsics in order to support threaded operation. If you need a
2184 multhreading-capable LLVM on a platform without a suitably modern system
2185 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2186 using the resultant compiler to build a copy of LLVM with multithreading
2191 Ending Execution with ``llvm_shutdown()``
2192 -----------------------------------------
2194 When you are done using the LLVM APIs, you should call ``llvm_shutdown()`` to
2195 deallocate memory used for internal structures.
2199 Lazy Initialization with ``ManagedStatic``
2200 ------------------------------------------
2202 ``ManagedStatic`` is a utility class in LLVM used to implement static
2203 initialization of static resources, such as the global type tables. In a
2204 single-threaded environment, it implements a simple lazy initialization scheme.
2205 When LLVM is compiled with support for multi-threading, however, it uses
2206 double-checked locking to implement thread-safe lazy initialization.
2210 Achieving Isolation with ``LLVMContext``
2211 ----------------------------------------
2213 ``LLVMContext`` is an opaque class in the LLVM API which clients can use to
2214 operate multiple, isolated instances of LLVM concurrently within the same
2215 address space. For instance, in a hypothetical compile-server, the compilation
2216 of an individual translation unit is conceptually independent from all the
2217 others, and it would be desirable to be able to compile incoming translation
2218 units concurrently on independent server threads. Fortunately, ``LLVMContext``
2219 exists to enable just this kind of scenario!
2221 Conceptually, ``LLVMContext`` provides isolation. Every LLVM entity
2222 (``Module``\ s, ``Value``\ s, ``Type``\ s, ``Constant``\ s, etc.) in LLVM's
2223 in-memory IR belongs to an ``LLVMContext``. Entities in different contexts
2224 *cannot* interact with each other: ``Module``\ s in different contexts cannot be
2225 linked together, ``Function``\ s cannot be added to ``Module``\ s in different
2226 contexts, etc. What this means is that is is safe to compile on multiple
2227 threads simultaneously, as long as no two threads operate on entities within the
2230 In practice, very few places in the API require the explicit specification of a
2231 ``LLVMContext``, other than the ``Type`` creation/lookup APIs. Because every
2232 ``Type`` carries a reference to its owning context, most other entities can
2233 determine what context they belong to by looking at their own ``Type``. If you
2234 are adding new entities to LLVM IR, please try to maintain this interface
2237 For clients that do *not* require the benefits of isolation, LLVM provides a
2238 convenience API ``getGlobalContext()``. This returns a global, lazily
2239 initialized ``LLVMContext`` that may be used in situations where isolation is
2247 LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple
2248 threads can call ``ExecutionEngine::getPointerToFunction()`` or
2249 ``ExecutionEngine::runFunction()`` concurrently, and multiple threads can run
2250 code output by the JIT concurrently. The user must still ensure that only one
2251 thread accesses IR in a given ``LLVMContext`` while another thread might be
2252 modifying it. One way to do that is to always hold the JIT lock while accessing
2253 IR outside the JIT (the JIT *modifies* the IR by adding ``CallbackVH``\ s).
2254 Another way is to only call ``getPointerToFunction()`` from the
2255 ``LLVMContext``'s thread.
2257 When the JIT is configured to compile lazily (using
2258 ``ExecutionEngine::DisableLazyCompilation(false)``), there is currently a `race
2259 condition <http://llvm.org/bugs/show_bug.cgi?id=5184>`_ in updating call sites
2260 after a function is lazily-jitted. It's still possible to use the lazy JIT in a
2261 threaded program if you ensure that only one thread at a time can call any
2262 particular lazy stub and that the JIT lock guards any IR access, but we suggest
2263 using only the eager JIT in threaded programs.
2270 This section describes some of the advanced or obscure API's that most clients
2271 do not need to be aware of. These API's tend manage the inner workings of the
2272 LLVM system, and only need to be accessed in unusual circumstances.
2276 The ``ValueSymbolTable`` class
2277 ------------------------------
2279 The ``ValueSymbolTable`` (`doxygen
2280 <http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html>`__) class provides
2281 a symbol table that the :ref:`Function <c_Function>` and Module_ classes use for
2282 naming value definitions. The symbol table can provide a name for any Value_.
2284 Note that the ``SymbolTable`` class should not be directly accessed by most
2285 clients. It should only be used when iteration over the symbol table names
2286 themselves are required, which is very special purpose. Note that not all LLVM
2287 Value_\ s have names, and those without names (i.e. they have an empty name) do
2288 not exist in the symbol table.
2290 Symbol tables support iteration over the values in the symbol table with
2291 ``begin/end/iterator`` and supports querying to see if a specific name is in the
2292 symbol table (with ``lookup``). The ``ValueSymbolTable`` class exposes no
2293 public mutator methods, instead, simply call ``setName`` on a value, which will
2294 autoinsert it into the appropriate symbol table.
2298 The ``User`` and owned ``Use`` classes' memory layout
2299 -----------------------------------------------------
2301 The ``User`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1User.html>`__)
2302 class provides a basis for expressing the ownership of ``User`` towards other
2303 `Value instance <http://llvm.org/doxygen/classllvm_1_1Value.html>`_\ s. The
2304 ``Use`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1Use.html>`__) helper
2305 class is employed to do the bookkeeping and to facilitate *O(1)* addition and
2310 Interaction and relationship between ``User`` and ``Use`` objects
2311 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2313 A subclass of ``User`` can choose between incorporating its ``Use`` objects or
2314 refer to them out-of-line by means of a pointer. A mixed variant (some ``Use``
2315 s inline others hung off) is impractical and breaks the invariant that the
2316 ``Use`` objects belonging to the same ``User`` form a contiguous array.
2318 We have 2 different layouts in the ``User`` (sub)classes:
2322 The ``Use`` object(s) are inside (resp. at fixed offset) of the ``User``
2323 object and there are a fixed number of them.
2327 The ``Use`` object(s) are referenced by a pointer to an array from the
2328 ``User`` object and there may be a variable number of them.
2330 As of v2.4 each layout still possesses a direct pointer to the start of the
2331 array of ``Use``\ s. Though not mandatory for layout a), we stick to this
2332 redundancy for the sake of simplicity. The ``User`` object also stores the
2333 number of ``Use`` objects it has. (Theoretically this information can also be
2334 calculated given the scheme presented below.)
2336 Special forms of allocation operators (``operator new``) enforce the following
2339 * Layout a) is modelled by prepending the ``User`` object by the ``Use[]``
2342 .. code-block:: none
2344 ...---.---.---.---.-------...
2345 | P | P | P | P | User
2346 '''---'---'---'---'-------'''
2348 * Layout b) is modelled by pointing at the ``Use[]`` array.
2350 .. code-block:: none
2361 *(In the above figures* '``P``' *stands for the* ``Use**`` *that is stored in
2362 each* ``Use`` *object in the member* ``Use::Prev`` *)*
2366 The waymarking algorithm
2367 ^^^^^^^^^^^^^^^^^^^^^^^^
2369 Since the ``Use`` objects are deprived of the direct (back)pointer to their
2370 ``User`` objects, there must be a fast and exact method to recover it. This is
2371 accomplished by the following scheme:
2373 A bit-encoding in the 2 LSBits (least significant bits) of the ``Use::Prev``
2374 allows to find the start of the ``User`` object:
2376 * ``00`` --- binary digit 0
2378 * ``01`` --- binary digit 1
2380 * ``10`` --- stop and calculate (``s``)
2382 * ``11`` --- full stop (``S``)
2384 Given a ``Use*``, all we have to do is to walk till we get a stop and we either
2385 have a ``User`` immediately behind or we have to walk to the next stop picking
2386 up digits and calculating the offset:
2388 .. code-block:: none
2390 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2391 | 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
2392 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2393 |+15 |+10 |+6 |+3 |+1
2396 | | | ______________________>
2397 | | ______________________________________>
2398 | __________________________________________________________>
2400 Only the significant number of bits need to be stored between the stops, so that
2401 the *worst case is 20 memory accesses* when there are 1000 ``Use`` objects
2402 associated with a ``User``.
2406 Reference implementation
2407 ^^^^^^^^^^^^^^^^^^^^^^^^
2409 The following literate Haskell fragment demonstrates the concept:
2411 .. code-block:: haskell
2413 > import Test.QuickCheck
2415 > digits :: Int -> [Char] -> [Char]
2416 > digits 0 acc = '0' : acc
2417 > digits 1 acc = '1' : acc
2418 > digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
2420 > dist :: Int -> [Char] -> [Char]
2423 > dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
2424 > dist n acc = dist (n - 1) $ dist 1 acc
2426 > takeLast n ss = reverse $ take n $ reverse ss
2428 > test = takeLast 40 $ dist 20 []
2431 Printing <test> gives: ``"1s100000s11010s10100s1111s1010s110s11s1S"``
2433 The reverse algorithm computes the length of the string just by examining a
2436 .. code-block:: haskell
2438 > pref :: [Char] -> Int
2440 > pref ('s':'1':rest) = decode 2 1 rest
2441 > pref (_:rest) = 1 + pref rest
2443 > decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
2444 > decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
2445 > decode walk acc _ = walk + acc
2448 Now, as expected, printing <pref test> gives ``40``.
2450 We can *quickCheck* this with following property:
2452 .. code-block:: haskell
2454 > testcase = dist 2000 []
2455 > testcaseLength = length testcase
2457 > identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
2458 > where arr = takeLast n testcase
2461 As expected <quickCheck identityProp> gives:
2465 *Main> quickCheck identityProp
2466 OK, passed 100 tests.
2468 Let's be a bit more exhaustive:
2470 .. code-block:: haskell
2473 > deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
2476 And here is the result of <deepCheck identityProp>:
2480 *Main> deepCheck identityProp
2481 OK, passed 500 tests.
2485 Tagging considerations
2486 ^^^^^^^^^^^^^^^^^^^^^^
2488 To maintain the invariant that the 2 LSBits of each ``Use**`` in ``Use`` never
2489 change after being set up, setters of ``Use::Prev`` must re-tag the new
2490 ``Use**`` on every modification. Accordingly getters must strip the tag bits.
2492 For layout b) instead of the ``User`` we find a pointer (``User*`` with LSBit
2493 set). Following this pointer brings us to the ``User``. A portable trick
2494 ensures that the first bytes of ``User`` (if interpreted as a pointer) never has
2495 the LSBit set. (Portability is relying on the fact that all known compilers
2496 place the ``vptr`` in the first word of the instances.)
2500 Designing Type Hiercharies and Polymorphic Interfaces
2501 -----------------------------------------------------
2503 There are two different design patterns that tend to result in the use of
2504 virtual dispatch for methods in a type hierarchy in C++ programs. The first is
2505 a genuine type hierarchy where different types in the hierarchy model
2506 a specific subset of the functionality and semantics, and these types nest
2507 strictly within each other. Good examples of this can be seen in the ``Value``
2508 or ``Type`` type hierarchies.
2510 A second is the desire to dispatch dynamically across a collection of
2511 polymorphic interface implementations. This latter use case can be modeled with
2512 virtual dispatch and inheritance by defining an abstract interface base class
2513 which all implementations derive from and override. However, this
2514 implementation strategy forces an **"is-a"** relationship to exist that is not
2515 actually meaningful. There is often not some nested hierarchy of useful
2516 generalizations which code might interact with and move up and down. Instead,
2517 there is a singular interface which is dispatched across a range of
2520 The preferred implementation strategy for the second use case is that of
2521 generic programming (sometimes called "compile-time duck typing" or "static
2522 polymorphism"). For example, a template over some type parameter ``T`` can be
2523 instantiated across any particular implementation that conforms to the
2524 interface or *concept*. A good example here is the highly generic properties of
2525 any type which models a node in a directed graph. LLVM models these primarily
2526 through templates and generic programming. Such templates include the
2527 ``LoopInfoBase`` and ``DominatorTreeBase``. When this type of polymorphism
2528 truly needs **dynamic** dispatch you can generalize it using a technique
2529 called *concept-based polymorphism*. This pattern emulates the interfaces and
2530 behaviors of templates using a very limited form of virtual dispatch for type
2531 erasure inside its implementation. You can find examples of this technique in
2532 the ``PassManager.h`` system, and there is a more detailed introduction to it
2533 by Sean Parent in several of his talks and papers:
2535 #. `Inheritance Is The Base Class of Evil
2536 <http://channel9.msdn.com/Events/GoingNative/2013/Inheritance-Is-The-Base-Class-of-Evil>`_
2537 - The GoingNative 2013 talk describing this technique, and probably the best
2539 #. `Value Semantics and Concepts-based Polymorphism
2540 <http://www.youtube.com/watch?v=_BpMYeUFXv8>`_ - The C++Now! 2012 talk
2541 describing this technique in more detail.
2542 #. `Sean Parent's Papers and Presentations
2543 <http://github.com/sean-parent/sean-parent.github.com/wiki/Papers-and-Presentations>`_
2544 - A Github project full of links to slides, video, and sometimes code.
2546 When deciding between creating a type hierarchy (with either tagged or virtual
2547 dispatch) and using templates or concepts-based polymorphism, consider whether
2548 there is some refinement of an abstract base class which is a semantically
2549 meaningful type on an interface boundary. If anything more refined than the
2550 root abstract interface is meaningless to talk about as a partial extension of
2551 the semantic model, then your use case likely fits better with polymorphism and
2552 you should avoid using virtual dispatch. However, there may be some exigent
2553 circumstances that require one technique or the other to be used.
2555 If you do need to introduce a type hierarchy, we prefer to use explicitly
2556 closed type hierarchies with manual tagged dispatch and/or RTTI rather than the
2557 open inheritance model and virtual dispatch that is more common in C++ code.
2558 This is because LLVM rarely encourages library consumers to extend its core
2559 types, and leverages the closed and tag-dispatched nature of its hierarchies to
2560 generate significantly more efficient code. We have also found that a large
2561 amount of our usage of type hierarchies fits better with tag-based pattern
2562 matching rather than dynamic dispatch across a common interface. Within LLVM we
2563 have built custom helpers to facilitate this design. See this document's
2564 section on :ref:`isa and dyn_cast <isa>` and our :doc:`detailed document
2565 <HowToSetUpLLVMStyleRTTI>` which describes how you can implement this
2566 pattern for use with the LLVM helpers.
2568 .. _abi_breaking_checks:
2573 Checks and asserts that alter the LLVM C++ ABI are predicated on the
2574 preprocessor symbol `LLVM_ENABLE_ABI_BREAKING_CHECKS` -- LLVM
2575 libraries built with `LLVM_ENABLE_ABI_BREAKING_CHECKS` are not ABI
2576 compatible LLVM libraries built without it defined. By default,
2577 turning on assertions also turns on `LLVM_ENABLE_ABI_BREAKING_CHECKS`
2578 so a default +Asserts build is not ABI compatible with a
2579 default -Asserts build. Clients that want ABI compatibility
2580 between +Asserts and -Asserts builds should use the CMake or autoconf
2581 build systems to set `LLVM_ENABLE_ABI_BREAKING_CHECKS` independently
2582 of `LLVM_ENABLE_ASSERTIONS`.
2586 The Core LLVM Class Hierarchy Reference
2587 =======================================
2589 ``#include "llvm/IR/Type.h"``
2591 header source: `Type.h <http://llvm.org/doxygen/Type_8h-source.html>`_
2593 doxygen info: `Type Clases <http://llvm.org/doxygen/classllvm_1_1Type.html>`_
2595 The Core LLVM classes are the primary means of representing the program being
2596 inspected or transformed. The core LLVM classes are defined in header files in
2597 the ``include/llvm/IR`` directory, and implemented in the ``lib/IR``
2598 directory. It's worth noting that, for historical reasons, this library is
2599 called ``libLLVMCore.so``, not ``libLLVMIR.so`` as you might expect.
2603 The Type class and Derived Types
2604 --------------------------------
2606 ``Type`` is a superclass of all type classes. Every ``Value`` has a ``Type``.
2607 ``Type`` cannot be instantiated directly but only through its subclasses.
2608 Certain primitive types (``VoidType``, ``LabelType``, ``FloatType`` and
2609 ``DoubleType``) have hidden subclasses. They are hidden because they offer no
2610 useful functionality beyond what the ``Type`` class offers except to distinguish
2611 themselves from other subclasses of ``Type``.
2613 All other types are subclasses of ``DerivedType``. Types can be named, but this
2614 is not a requirement. There exists exactly one instance of a given shape at any
2615 one time. This allows type equality to be performed with address equality of
2616 the Type Instance. That is, given two ``Type*`` values, the types are identical
2617 if the pointers are identical.
2621 Important Public Methods
2622 ^^^^^^^^^^^^^^^^^^^^^^^^
2624 * ``bool isIntegerTy() const``: Returns true for any integer type.
2626 * ``bool isFloatingPointTy()``: Return true if this is one of the five
2627 floating point types.
2629 * ``bool isSized()``: Return true if the type has known size. Things
2630 that don't have a size are abstract types, labels and void.
2634 Important Derived Types
2635 ^^^^^^^^^^^^^^^^^^^^^^^
2638 Subclass of DerivedType that represents integer types of any bit width. Any
2639 bit width between ``IntegerType::MIN_INT_BITS`` (1) and
2640 ``IntegerType::MAX_INT_BITS`` (~8 million) can be represented.
2642 * ``static const IntegerType* get(unsigned NumBits)``: get an integer
2643 type of a specific bit width.
2645 * ``unsigned getBitWidth() const``: Get the bit width of an integer type.
2648 This is subclassed by ArrayType, PointerType and VectorType.
2650 * ``const Type * getElementType() const``: Returns the type of each
2651 of the elements in the sequential type.
2654 This is a subclass of SequentialType and defines the interface for array
2657 * ``unsigned getNumElements() const``: Returns the number of elements
2661 Subclass of SequentialType for pointer types.
2664 Subclass of SequentialType for vector types. A vector type is similar to an
2665 ArrayType but is distinguished because it is a first class type whereas
2666 ArrayType is not. Vector types are used for vector operations and are usually
2667 small vectors of an integer or floating point type.
2670 Subclass of DerivedTypes for struct types.
2675 Subclass of DerivedTypes for function types.
2677 * ``bool isVarArg() const``: Returns true if it's a vararg function.
2679 * ``const Type * getReturnType() const``: Returns the return type of the
2682 * ``const Type * getParamType (unsigned i)``: Returns the type of the ith
2685 * ``const unsigned getNumParams() const``: Returns the number of formal
2690 The ``Module`` class
2691 --------------------
2693 ``#include "llvm/IR/Module.h"``
2695 header source: `Module.h <http://llvm.org/doxygen/Module_8h-source.html>`_
2697 doxygen info: `Module Class <http://llvm.org/doxygen/classllvm_1_1Module.html>`_
2699 The ``Module`` class represents the top level structure present in LLVM
2700 programs. An LLVM module is effectively either a translation unit of the
2701 original program or a combination of several translation units merged by the
2702 linker. The ``Module`` class keeps track of a list of :ref:`Function
2703 <c_Function>`\ s, a list of GlobalVariable_\ s, and a SymbolTable_.
2704 Additionally, it contains a few helpful member functions that try to make common
2709 Important Public Members of the ``Module`` class
2710 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2712 * ``Module::Module(std::string name = "")``
2714 Constructing a Module_ is easy. You can optionally provide a name for it
2715 (probably based on the name of the translation unit).
2717 * | ``Module::iterator`` - Typedef for function list iterator
2718 | ``Module::const_iterator`` - Typedef for const_iterator.
2719 | ``begin()``, ``end()``, ``size()``, ``empty()``
2721 These are forwarding methods that make it easy to access the contents of a
2722 ``Module`` object's :ref:`Function <c_Function>` list.
2724 * ``Module::FunctionListType &getFunctionList()``
2726 Returns the list of :ref:`Function <c_Function>`\ s. This is necessary to use
2727 when you need to update the list or perform a complex action that doesn't have
2728 a forwarding method.
2732 * | ``Module::global_iterator`` - Typedef for global variable list iterator
2733 | ``Module::const_global_iterator`` - Typedef for const_iterator.
2734 | ``global_begin()``, ``global_end()``, ``global_size()``, ``global_empty()``
2736 These are forwarding methods that make it easy to access the contents of a
2737 ``Module`` object's GlobalVariable_ list.
2739 * ``Module::GlobalListType &getGlobalList()``
2741 Returns the list of GlobalVariable_\ s. This is necessary to use when you
2742 need to update the list or perform a complex action that doesn't have a
2747 * ``SymbolTable *getSymbolTable()``
2749 Return a reference to the SymbolTable_ for this ``Module``.
2753 * ``Function *getFunction(StringRef Name) const``
2755 Look up the specified function in the ``Module`` SymbolTable_. If it does not
2756 exist, return ``null``.
2758 * ``Function *getOrInsertFunction(const std::string &Name, const FunctionType
2761 Look up the specified function in the ``Module`` SymbolTable_. If it does not
2762 exist, add an external declaration for the function and return it.
2764 * ``std::string getTypeName(const Type *Ty)``
2766 If there is at least one entry in the SymbolTable_ for the specified Type_,
2767 return it. Otherwise return the empty string.
2769 * ``bool addTypeName(const std::string &Name, const Type *Ty)``
2771 Insert an entry in the SymbolTable_ mapping ``Name`` to ``Ty``. If there is
2772 already an entry for this name, true is returned and the SymbolTable_ is not
2780 ``#include "llvm/IR/Value.h"``
2782 header source: `Value.h <http://llvm.org/doxygen/Value_8h-source.html>`_
2784 doxygen info: `Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_
2786 The ``Value`` class is the most important class in the LLVM Source base. It
2787 represents a typed value that may be used (among other things) as an operand to
2788 an instruction. There are many different types of ``Value``\ s, such as
2789 Constant_\ s, Argument_\ s. Even Instruction_\ s and :ref:`Function
2790 <c_Function>`\ s are ``Value``\ s.
2792 A particular ``Value`` may be used many times in the LLVM representation for a
2793 program. For example, an incoming argument to a function (represented with an
2794 instance of the Argument_ class) is "used" by every instruction in the function
2795 that references the argument. To keep track of this relationship, the ``Value``
2796 class keeps a list of all of the ``User``\ s that is using it (the User_ class
2797 is a base class for all nodes in the LLVM graph that can refer to ``Value``\ s).
2798 This use list is how LLVM represents def-use information in the program, and is
2799 accessible through the ``use_*`` methods, shown below.
2801 Because LLVM is a typed representation, every LLVM ``Value`` is typed, and this
2802 Type_ is available through the ``getType()`` method. In addition, all LLVM
2803 values can be named. The "name" of the ``Value`` is a symbolic string printed
2806 .. code-block:: llvm
2812 The name of this instruction is "foo". **NOTE** that the name of any value may
2813 be missing (an empty string), so names should **ONLY** be used for debugging
2814 (making the source code easier to read, debugging printouts), they should not be
2815 used to keep track of values or map between them. For this purpose, use a
2816 ``std::map`` of pointers to the ``Value`` itself instead.
2818 One important aspect of LLVM is that there is no distinction between an SSA
2819 variable and the operation that produces it. Because of this, any reference to
2820 the value produced by an instruction (or the value available as an incoming
2821 argument, for example) is represented as a direct pointer to the instance of the
2822 class that represents this value. Although this may take some getting used to,
2823 it simplifies the representation and makes it easier to manipulate.
2827 Important Public Members of the ``Value`` class
2828 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2830 * | ``Value::use_iterator`` - Typedef for iterator over the use-list
2831 | ``Value::const_use_iterator`` - Typedef for const_iterator over the
2833 | ``unsigned use_size()`` - Returns the number of users of the value.
2834 | ``bool use_empty()`` - Returns true if there are no users.
2835 | ``use_iterator use_begin()`` - Get an iterator to the start of the
2837 | ``use_iterator use_end()`` - Get an iterator to the end of the use-list.
2838 | ``User *use_back()`` - Returns the last element in the list.
2840 These methods are the interface to access the def-use information in LLVM.
2841 As with all other iterators in LLVM, the naming conventions follow the
2842 conventions defined by the STL_.
2844 * ``Type *getType() const``
2845 This method returns the Type of the Value.
2847 * | ``bool hasName() const``
2848 | ``std::string getName() const``
2849 | ``void setName(const std::string &Name)``
2851 This family of methods is used to access and assign a name to a ``Value``, be
2852 aware of the :ref:`precaution above <nameWarning>`.
2854 * ``void replaceAllUsesWith(Value *V)``
2856 This method traverses the use list of a ``Value`` changing all User_\ s of the
2857 current value to refer to "``V``" instead. For example, if you detect that an
2858 instruction always produces a constant value (for example through constant
2859 folding), you can replace all uses of the instruction with the constant like
2864 Inst->replaceAllUsesWith(ConstVal);
2871 ``#include "llvm/IR/User.h"``
2873 header source: `User.h <http://llvm.org/doxygen/User_8h-source.html>`_
2875 doxygen info: `User Class <http://llvm.org/doxygen/classllvm_1_1User.html>`_
2879 The ``User`` class is the common base class of all LLVM nodes that may refer to
2880 ``Value``\ s. It exposes a list of "Operands" that are all of the ``Value``\ s
2881 that the User is referring to. The ``User`` class itself is a subclass of
2884 The operands of a ``User`` point directly to the LLVM ``Value`` that it refers
2885 to. Because LLVM uses Static Single Assignment (SSA) form, there can only be
2886 one definition referred to, allowing this direct connection. This connection
2887 provides the use-def information in LLVM.
2891 Important Public Members of the ``User`` class
2892 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2894 The ``User`` class exposes the operand list in two ways: through an index access
2895 interface and through an iterator based interface.
2897 * | ``Value *getOperand(unsigned i)``
2898 | ``unsigned getNumOperands()``
2900 These two methods expose the operands of the ``User`` in a convenient form for
2903 * | ``User::op_iterator`` - Typedef for iterator over the operand list
2904 | ``op_iterator op_begin()`` - Get an iterator to the start of the operand
2906 | ``op_iterator op_end()`` - Get an iterator to the end of the operand list.
2908 Together, these methods make up the iterator based interface to the operands
2914 The ``Instruction`` class
2915 -------------------------
2917 ``#include "llvm/IR/Instruction.h"``
2919 header source: `Instruction.h
2920 <http://llvm.org/doxygen/Instruction_8h-source.html>`_
2922 doxygen info: `Instruction Class
2923 <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_
2925 Superclasses: User_, Value_
2927 The ``Instruction`` class is the common base class for all LLVM instructions.
2928 It provides only a few methods, but is a very commonly used class. The primary
2929 data tracked by the ``Instruction`` class itself is the opcode (instruction
2930 type) and the parent BasicBlock_ the ``Instruction`` is embedded into. To
2931 represent a specific type of instruction, one of many subclasses of
2932 ``Instruction`` are used.
2934 Because the ``Instruction`` class subclasses the User_ class, its operands can
2935 be accessed in the same way as for other ``User``\ s (with the
2936 ``getOperand()``/``getNumOperands()`` and ``op_begin()``/``op_end()`` methods).
2937 An important file for the ``Instruction`` class is the ``llvm/Instruction.def``
2938 file. This file contains some meta-data about the various different types of
2939 instructions in LLVM. It describes the enum values that are used as opcodes
2940 (for example ``Instruction::Add`` and ``Instruction::ICmp``), as well as the
2941 concrete sub-classes of ``Instruction`` that implement the instruction (for
2942 example BinaryOperator_ and CmpInst_). Unfortunately, the use of macros in this
2943 file confuses doxygen, so these enum values don't show up correctly in the
2944 `doxygen output <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_.
2948 Important Subclasses of the ``Instruction`` class
2949 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2953 * ``BinaryOperator``
2955 This subclasses represents all two operand instructions whose operands must be
2956 the same type, except for the comparison instructions.
2961 This subclass is the parent of the 12 casting instructions. It provides
2962 common operations on cast instructions.
2968 This subclass respresents the two comparison instructions,
2969 `ICmpInst <LangRef.html#i_icmp>`_ (integer opreands), and
2970 `FCmpInst <LangRef.html#i_fcmp>`_ (floating point operands).
2974 * ``TerminatorInst``
2976 This subclass is the parent of all terminator instructions (those which can
2981 Important Public Members of the ``Instruction`` class
2982 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2984 * ``BasicBlock *getParent()``
2986 Returns the BasicBlock_ that this
2987 ``Instruction`` is embedded into.
2989 * ``bool mayWriteToMemory()``
2991 Returns true if the instruction writes to memory, i.e. it is a ``call``,
2992 ``free``, ``invoke``, or ``store``.
2994 * ``unsigned getOpcode()``
2996 Returns the opcode for the ``Instruction``.
2998 * ``Instruction *clone() const``
3000 Returns another instance of the specified instruction, identical in all ways
3001 to the original except that the instruction has no parent (i.e. it's not
3002 embedded into a BasicBlock_), and it has no name.
3006 The ``Constant`` class and subclasses
3007 -------------------------------------
3009 Constant represents a base class for different types of constants. It is
3010 subclassed by ConstantInt, ConstantArray, etc. for representing the various
3011 types of Constants. GlobalValue_ is also a subclass, which represents the
3012 address of a global variable or function.
3016 Important Subclasses of Constant
3017 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3019 * ConstantInt : This subclass of Constant represents an integer constant of
3022 * ``const APInt& getValue() const``: Returns the underlying
3023 value of this constant, an APInt value.
3025 * ``int64_t getSExtValue() const``: Converts the underlying APInt value to an
3026 int64_t via sign extension. If the value (not the bit width) of the APInt
3027 is too large to fit in an int64_t, an assertion will result. For this
3028 reason, use of this method is discouraged.
3030 * ``uint64_t getZExtValue() const``: Converts the underlying APInt value
3031 to a uint64_t via zero extension. IF the value (not the bit width) of the
3032 APInt is too large to fit in a uint64_t, an assertion will result. For this
3033 reason, use of this method is discouraged.
3035 * ``static ConstantInt* get(const APInt& Val)``: Returns the ConstantInt
3036 object that represents the value provided by ``Val``. The type is implied
3037 as the IntegerType that corresponds to the bit width of ``Val``.
3039 * ``static ConstantInt* get(const Type *Ty, uint64_t Val)``: Returns the
3040 ConstantInt object that represents the value provided by ``Val`` for integer
3043 * ConstantFP : This class represents a floating point constant.
3045 * ``double getValue() const``: Returns the underlying value of this constant.
3047 * ConstantArray : This represents a constant array.
3049 * ``const std::vector<Use> &getValues() const``: Returns a vector of
3050 component constants that makeup this array.
3052 * ConstantStruct : This represents a constant struct.
3054 * ``const std::vector<Use> &getValues() const``: Returns a vector of
3055 component constants that makeup this array.
3057 * GlobalValue : This represents either a global variable or a function. In
3058 either case, the value is a constant fixed address (after linking).
3062 The ``GlobalValue`` class
3063 -------------------------
3065 ``#include "llvm/IR/GlobalValue.h"``
3067 header source: `GlobalValue.h
3068 <http://llvm.org/doxygen/GlobalValue_8h-source.html>`_
3070 doxygen info: `GlobalValue Class
3071 <http://llvm.org/doxygen/classllvm_1_1GlobalValue.html>`_
3073 Superclasses: Constant_, User_, Value_
3075 Global values ( GlobalVariable_\ s or :ref:`Function <c_Function>`\ s) are the
3076 only LLVM values that are visible in the bodies of all :ref:`Function
3077 <c_Function>`\ s. Because they are visible at global scope, they are also
3078 subject to linking with other globals defined in different translation units.
3079 To control the linking process, ``GlobalValue``\ s know their linkage rules.
3080 Specifically, ``GlobalValue``\ s know whether they have internal or external
3081 linkage, as defined by the ``LinkageTypes`` enumeration.
3083 If a ``GlobalValue`` has internal linkage (equivalent to being ``static`` in C),
3084 it is not visible to code outside the current translation unit, and does not
3085 participate in linking. If it has external linkage, it is visible to external
3086 code, and does participate in linking. In addition to linkage information,
3087 ``GlobalValue``\ s keep track of which Module_ they are currently part of.
3089 Because ``GlobalValue``\ s are memory objects, they are always referred to by
3090 their **address**. As such, the Type_ of a global is always a pointer to its
3091 contents. It is important to remember this when using the ``GetElementPtrInst``
3092 instruction because this pointer must be dereferenced first. For example, if
3093 you have a ``GlobalVariable`` (a subclass of ``GlobalValue)`` that is an array
3094 of 24 ints, type ``[24 x i32]``, then the ``GlobalVariable`` is a pointer to
3095 that array. Although the address of the first element of this array and the
3096 value of the ``GlobalVariable`` are the same, they have different types. The
3097 ``GlobalVariable``'s type is ``[24 x i32]``. The first element's type is
3098 ``i32.`` Because of this, accessing a global value requires you to dereference
3099 the pointer with ``GetElementPtrInst`` first, then its elements can be accessed.
3100 This is explained in the `LLVM Language Reference Manual
3101 <LangRef.html#globalvars>`_.
3105 Important Public Members of the ``GlobalValue`` class
3106 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3108 * | ``bool hasInternalLinkage() const``
3109 | ``bool hasExternalLinkage() const``
3110 | ``void setInternalLinkage(bool HasInternalLinkage)``
3112 These methods manipulate the linkage characteristics of the ``GlobalValue``.
3114 * ``Module *getParent()``
3116 This returns the Module_ that the
3117 GlobalValue is currently embedded into.
3121 The ``Function`` class
3122 ----------------------
3124 ``#include "llvm/IR/Function.h"``
3126 header source: `Function.h <http://llvm.org/doxygen/Function_8h-source.html>`_
3128 doxygen info: `Function Class
3129 <http://llvm.org/doxygen/classllvm_1_1Function.html>`_
3131 Superclasses: GlobalValue_, Constant_, User_, Value_
3133 The ``Function`` class represents a single procedure in LLVM. It is actually
3134 one of the more complex classes in the LLVM hierarchy because it must keep track
3135 of a large amount of data. The ``Function`` class keeps track of a list of
3136 BasicBlock_\ s, a list of formal Argument_\ s, and a SymbolTable_.
3138 The list of BasicBlock_\ s is the most commonly used part of ``Function``
3139 objects. The list imposes an implicit ordering of the blocks in the function,
3140 which indicate how the code will be laid out by the backend. Additionally, the
3141 first BasicBlock_ is the implicit entry node for the ``Function``. It is not
3142 legal in LLVM to explicitly branch to this initial block. There are no implicit
3143 exit nodes, and in fact there may be multiple exit nodes from a single
3144 ``Function``. If the BasicBlock_ list is empty, this indicates that the
3145 ``Function`` is actually a function declaration: the actual body of the function
3146 hasn't been linked in yet.
3148 In addition to a list of BasicBlock_\ s, the ``Function`` class also keeps track
3149 of the list of formal Argument_\ s that the function receives. This container
3150 manages the lifetime of the Argument_ nodes, just like the BasicBlock_ list does
3151 for the BasicBlock_\ s.
3153 The SymbolTable_ is a very rarely used LLVM feature that is only used when you
3154 have to look up a value by name. Aside from that, the SymbolTable_ is used
3155 internally to make sure that there are not conflicts between the names of
3156 Instruction_\ s, BasicBlock_\ s, or Argument_\ s in the function body.
3158 Note that ``Function`` is a GlobalValue_ and therefore also a Constant_. The
3159 value of the function is its address (after linking) which is guaranteed to be
3164 Important Public Members of the ``Function``
3165 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3167 * ``Function(const FunctionType *Ty, LinkageTypes Linkage,
3168 const std::string &N = "", Module* Parent = 0)``
3170 Constructor used when you need to create new ``Function``\ s to add the
3171 program. The constructor must specify the type of the function to create and
3172 what type of linkage the function should have. The FunctionType_ argument
3173 specifies the formal arguments and return value for the function. The same
3174 FunctionType_ value can be used to create multiple functions. The ``Parent``
3175 argument specifies the Module in which the function is defined. If this
3176 argument is provided, the function will automatically be inserted into that
3177 module's list of functions.
3179 * ``bool isDeclaration()``
3181 Return whether or not the ``Function`` has a body defined. If the function is
3182 "external", it does not have a body, and thus must be resolved by linking with
3183 a function defined in a different translation unit.
3185 * | ``Function::iterator`` - Typedef for basic block list iterator
3186 | ``Function::const_iterator`` - Typedef for const_iterator.
3187 | ``begin()``, ``end()``, ``size()``, ``empty()``
3189 These are forwarding methods that make it easy to access the contents of a
3190 ``Function`` object's BasicBlock_ list.
3192 * ``Function::BasicBlockListType &getBasicBlockList()``
3194 Returns the list of BasicBlock_\ s. This is necessary to use when you need to
3195 update the list or perform a complex action that doesn't have a forwarding
3198 * | ``Function::arg_iterator`` - Typedef for the argument list iterator
3199 | ``Function::const_arg_iterator`` - Typedef for const_iterator.
3200 | ``arg_begin()``, ``arg_end()``, ``arg_size()``, ``arg_empty()``
3202 These are forwarding methods that make it easy to access the contents of a
3203 ``Function`` object's Argument_ list.
3205 * ``Function::ArgumentListType &getArgumentList()``
3207 Returns the list of Argument_. This is necessary to use when you need to
3208 update the list or perform a complex action that doesn't have a forwarding
3211 * ``BasicBlock &getEntryBlock()``
3213 Returns the entry ``BasicBlock`` for the function. Because the entry block
3214 for the function is always the first block, this returns the first block of
3217 * | ``Type *getReturnType()``
3218 | ``FunctionType *getFunctionType()``
3220 This traverses the Type_ of the ``Function`` and returns the return type of
3221 the function, or the FunctionType_ of the actual function.
3223 * ``SymbolTable *getSymbolTable()``
3225 Return a pointer to the SymbolTable_ for this ``Function``.
3229 The ``GlobalVariable`` class
3230 ----------------------------
3232 ``#include "llvm/IR/GlobalVariable.h"``
3234 header source: `GlobalVariable.h
3235 <http://llvm.org/doxygen/GlobalVariable_8h-source.html>`_
3237 doxygen info: `GlobalVariable Class
3238 <http://llvm.org/doxygen/classllvm_1_1GlobalVariable.html>`_
3240 Superclasses: GlobalValue_, Constant_, User_, Value_
3242 Global variables are represented with the (surprise surprise) ``GlobalVariable``
3243 class. Like functions, ``GlobalVariable``\ s are also subclasses of
3244 GlobalValue_, and as such are always referenced by their address (global values
3245 must live in memory, so their "name" refers to their constant address). See
3246 GlobalValue_ for more on this. Global variables may have an initial value
3247 (which must be a Constant_), and if they have an initializer, they may be marked
3248 as "constant" themselves (indicating that their contents never change at
3251 .. _m_GlobalVariable:
3253 Important Public Members of the ``GlobalVariable`` class
3254 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3256 * ``GlobalVariable(const Type *Ty, bool isConstant, LinkageTypes &Linkage,
3257 Constant *Initializer = 0, const std::string &Name = "", Module* Parent = 0)``
3259 Create a new global variable of the specified type. If ``isConstant`` is true
3260 then the global variable will be marked as unchanging for the program. The
3261 Linkage parameter specifies the type of linkage (internal, external, weak,
3262 linkonce, appending) for the variable. If the linkage is InternalLinkage,
3263 WeakAnyLinkage, WeakODRLinkage, LinkOnceAnyLinkage or LinkOnceODRLinkage, then
3264 the resultant global variable will have internal linkage. AppendingLinkage
3265 concatenates together all instances (in different translation units) of the
3266 variable into a single variable but is only applicable to arrays. See the
3267 `LLVM Language Reference <LangRef.html#modulestructure>`_ for further details
3268 on linkage types. Optionally an initializer, a name, and the module to put
3269 the variable into may be specified for the global variable as well.
3271 * ``bool isConstant() const``
3273 Returns true if this is a global variable that is known not to be modified at
3276 * ``bool hasInitializer()``
3278 Returns true if this ``GlobalVariable`` has an intializer.
3280 * ``Constant *getInitializer()``
3282 Returns the initial value for a ``GlobalVariable``. It is not legal to call
3283 this method if there is no initializer.
3287 The ``BasicBlock`` class
3288 ------------------------
3290 ``#include "llvm/IR/BasicBlock.h"``
3292 header source: `BasicBlock.h
3293 <http://llvm.org/doxygen/BasicBlock_8h-source.html>`_
3295 doxygen info: `BasicBlock Class
3296 <http://llvm.org/doxygen/classllvm_1_1BasicBlock.html>`_
3300 This class represents a single entry single exit section of the code, commonly
3301 known as a basic block by the compiler community. The ``BasicBlock`` class
3302 maintains a list of Instruction_\ s, which form the body of the block. Matching
3303 the language definition, the last element of this list of instructions is always
3304 a terminator instruction (a subclass of the TerminatorInst_ class).
3306 In addition to tracking the list of instructions that make up the block, the
3307 ``BasicBlock`` class also keeps track of the :ref:`Function <c_Function>` that
3308 it is embedded into.
3310 Note that ``BasicBlock``\ s themselves are Value_\ s, because they are
3311 referenced by instructions like branches and can go in the switch tables.
3312 ``BasicBlock``\ s have type ``label``.
3316 Important Public Members of the ``BasicBlock`` class
3317 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3319 * ``BasicBlock(const std::string &Name = "", Function *Parent = 0)``
3321 The ``BasicBlock`` constructor is used to create new basic blocks for
3322 insertion into a function. The constructor optionally takes a name for the
3323 new block, and a :ref:`Function <c_Function>` to insert it into. If the
3324 ``Parent`` parameter is specified, the new ``BasicBlock`` is automatically
3325 inserted at the end of the specified :ref:`Function <c_Function>`, if not
3326 specified, the BasicBlock must be manually inserted into the :ref:`Function
3329 * | ``BasicBlock::iterator`` - Typedef for instruction list iterator
3330 | ``BasicBlock::const_iterator`` - Typedef for const_iterator.
3331 | ``begin()``, ``end()``, ``front()``, ``back()``,
3332 ``size()``, ``empty()``
3333 STL-style functions for accessing the instruction list.
3335 These methods and typedefs are forwarding functions that have the same
3336 semantics as the standard library methods of the same names. These methods
3337 expose the underlying instruction list of a basic block in a way that is easy
3338 to manipulate. To get the full complement of container operations (including
3339 operations to update the list), you must use the ``getInstList()`` method.
3341 * ``BasicBlock::InstListType &getInstList()``
3343 This method is used to get access to the underlying container that actually
3344 holds the Instructions. This method must be used when there isn't a
3345 forwarding function in the ``BasicBlock`` class for the operation that you
3346 would like to perform. Because there are no forwarding functions for
3347 "updating" operations, you need to use this if you want to update the contents
3348 of a ``BasicBlock``.
3350 * ``Function *getParent()``
3352 Returns a pointer to :ref:`Function <c_Function>` the block is embedded into,
3353 or a null pointer if it is homeless.
3355 * ``TerminatorInst *getTerminator()``
3357 Returns a pointer to the terminator instruction that appears at the end of the
3358 ``BasicBlock``. If there is no terminator instruction, or if the last
3359 instruction in the block is not a terminator, then a null pointer is returned.
3363 The ``Argument`` class
3364 ----------------------
3366 This subclass of Value defines the interface for incoming formal arguments to a
3367 function. A Function maintains a list of its formal arguments. An argument has
3368 a pointer to the parent Function.