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 optimized 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 can define the ``DEBUG_TYPE`` macro and use the ``-debug-only`` option as
392 DEBUG(errs() << "No debug type\n");
393 #define DEBUG_TYPE "foo"
394 DEBUG(errs() << "'foo' debug type\n");
396 #define DEBUG_TYPE "bar"
397 DEBUG(errs() << "'bar' debug type\n"));
399 #define DEBUG_TYPE ""
400 DEBUG(errs() << "No debug type (2)\n");
402 Then you can run your pass like this:
406 $ opt < a.bc > /dev/null -mypass
408 $ opt < a.bc > /dev/null -mypass -debug
413 $ opt < a.bc > /dev/null -mypass -debug-only=foo
415 $ opt < a.bc > /dev/null -mypass -debug-only=bar
418 Of course, in practice, you should only set ``DEBUG_TYPE`` at the top of a file,
419 to specify the debug type for the entire module (if you do this before you
420 ``#include "llvm/Support/Debug.h"``, you don't have to insert the ugly
421 ``#undef``'s). Also, you should use names more meaningful than "foo" and "bar",
422 because there is no system in place to ensure that names do not conflict. If
423 two different modules use the same string, they will all be turned on when the
424 name is specified. This allows, for example, all debug information for
425 instruction scheduling to be enabled with ``-debug-only=InstrSched``, even if
426 the source lives in multiple files.
428 For performance reasons, -debug-only is not available in optimized build
429 (``--enable-optimized``) of LLVM.
431 The ``DEBUG_WITH_TYPE`` macro is also available for situations where you would
432 like to set ``DEBUG_TYPE``, but only for one specific ``DEBUG`` statement. It
433 takes an additional first parameter, which is the type to use. For example, the
434 preceding example could be written as:
438 DEBUG_WITH_TYPE("", errs() << "No debug type\n");
439 DEBUG_WITH_TYPE("foo", errs() << "'foo' debug type\n");
440 DEBUG_WITH_TYPE("bar", errs() << "'bar' debug type\n"));
441 DEBUG_WITH_TYPE("", errs() << "No debug type (2)\n");
445 The ``Statistic`` class & ``-stats`` option
446 -------------------------------------------
448 The ``llvm/ADT/Statistic.h`` (`doxygen
449 <http://llvm.org/doxygen/Statistic_8h-source.html>`__) file provides a class
450 named ``Statistic`` that is used as a unified way to keep track of what the LLVM
451 compiler is doing and how effective various optimizations are. It is useful to
452 see what optimizations are contributing to making a particular program run
455 Often you may run your pass on some big program, and you're interested to see
456 how many times it makes a certain transformation. Although you can do this with
457 hand inspection, or some ad-hoc method, this is a real pain and not very useful
458 for big programs. Using the ``Statistic`` class makes it very easy to keep
459 track of this information, and the calculated information is presented in a
460 uniform manner with the rest of the passes being executed.
462 There are many examples of ``Statistic`` uses, but the basics of using it are as
465 #. Define your statistic like this:
469 #define DEBUG_TYPE "mypassname" // This goes before any #includes.
470 STATISTIC(NumXForms, "The # of times I did stuff");
472 The ``STATISTIC`` macro defines a static variable, whose name is specified by
473 the first argument. The pass name is taken from the ``DEBUG_TYPE`` macro, and
474 the description is taken from the second argument. The variable defined
475 ("NumXForms" in this case) acts like an unsigned integer.
477 #. Whenever you make a transformation, bump the counter:
481 ++NumXForms; // I did stuff!
483 That's all you have to do. To get '``opt``' to print out the statistics
484 gathered, use the '``-stats``' option:
488 $ opt -stats -mypassname < program.bc > /dev/null
489 ... statistics output ...
491 Note that in order to use the '``-stats``' option, LLVM must be
492 compiled with assertions enabled.
494 When running ``opt`` on a C file from the SPEC benchmark suite, it gives a
495 report that looks like this:
499 7646 bitcodewriter - Number of normal instructions
500 725 bitcodewriter - Number of oversized instructions
501 129996 bitcodewriter - Number of bitcode bytes written
502 2817 raise - Number of insts DCEd or constprop'd
503 3213 raise - Number of cast-of-self removed
504 5046 raise - Number of expression trees converted
505 75 raise - Number of other getelementptr's formed
506 138 raise - Number of load/store peepholes
507 42 deadtypeelim - Number of unused typenames removed from symtab
508 392 funcresolve - Number of varargs functions resolved
509 27 globaldce - Number of global variables removed
510 2 adce - Number of basic blocks removed
511 134 cee - Number of branches revectored
512 49 cee - Number of setcc instruction eliminated
513 532 gcse - Number of loads removed
514 2919 gcse - Number of instructions removed
515 86 indvars - Number of canonical indvars added
516 87 indvars - Number of aux indvars removed
517 25 instcombine - Number of dead inst eliminate
518 434 instcombine - Number of insts combined
519 248 licm - Number of load insts hoisted
520 1298 licm - Number of insts hoisted to a loop pre-header
521 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
522 75 mem2reg - Number of alloca's promoted
523 1444 cfgsimplify - Number of blocks simplified
525 Obviously, with so many optimizations, having a unified framework for this stuff
526 is very nice. Making your pass fit well into the framework makes it more
527 maintainable and useful.
531 Viewing graphs while debugging code
532 -----------------------------------
534 Several of the important data structures in LLVM are graphs: for example CFGs
535 made out of LLVM :ref:`BasicBlocks <BasicBlock>`, CFGs made out of LLVM
536 :ref:`MachineBasicBlocks <MachineBasicBlock>`, and :ref:`Instruction Selection
537 DAGs <SelectionDAG>`. In many cases, while debugging various parts of the
538 compiler, it is nice to instantly visualize these graphs.
540 LLVM provides several callbacks that are available in a debug build to do
541 exactly that. If you call the ``Function::viewCFG()`` method, for example, the
542 current LLVM tool will pop up a window containing the CFG for the function where
543 each basic block is a node in the graph, and each node contains the instructions
544 in the block. Similarly, there also exists ``Function::viewCFGOnly()`` (does
545 not include the instructions), the ``MachineFunction::viewCFG()`` and
546 ``MachineFunction::viewCFGOnly()``, and the ``SelectionDAG::viewGraph()``
547 methods. Within GDB, for example, you can usually use something like ``call
548 DAG.viewGraph()`` to pop up a window. Alternatively, you can sprinkle calls to
549 these functions in your code in places you want to debug.
551 Getting this to work requires a small amount of setup. On Unix systems
552 with X11, install the `graphviz <http://www.graphviz.org>`_ toolkit, and make
553 sure 'dot' and 'gv' are in your path. If you are running on Mac OS X, download
554 and install the Mac OS X `Graphviz program
555 <http://www.pixelglow.com/graphviz/>`_ and add
556 ``/Applications/Graphviz.app/Contents/MacOS/`` (or wherever you install it) to
557 your path. The programs need not be present when configuring, building or
558 running LLVM and can simply be installed when needed during an active debug
561 ``SelectionDAG`` has been extended to make it easier to locate *interesting*
562 nodes in large complex graphs. From gdb, if you ``call DAG.setGraphColor(node,
563 "color")``, then the next ``call DAG.viewGraph()`` would highlight the node in
564 the specified color (choices of colors can be found at `colors
565 <http://www.graphviz.org/doc/info/colors.html>`_.) More complex node attributes
566 can be provided with ``call DAG.setGraphAttrs(node, "attributes")`` (choices can
567 be found at `Graph attributes <http://www.graphviz.org/doc/info/attrs.html>`_.)
568 If you want to restart and clear all the current graph attributes, then you can
569 ``call DAG.clearGraphAttrs()``.
571 Note that graph visualization features are compiled out of Release builds to
572 reduce file size. This means that you need a Debug+Asserts or Release+Asserts
573 build to use these features.
577 Picking the Right Data Structure for a Task
578 ===========================================
580 LLVM has a plethora of data structures in the ``llvm/ADT/`` directory, and we
581 commonly use STL data structures. This section describes the trade-offs you
582 should consider when you pick one.
584 The first step is a choose your own adventure: do you want a sequential
585 container, a set-like container, or a map-like container? The most important
586 thing when choosing a container is the algorithmic properties of how you plan to
587 access the container. Based on that, you should use:
590 * a :ref:`map-like <ds_map>` container if you need efficient look-up of a
591 value based on another value. Map-like containers also support efficient
592 queries for containment (whether a key is in the map). Map-like containers
593 generally do not support efficient reverse mapping (values to keys). If you
594 need that, use two maps. Some map-like containers also support efficient
595 iteration through the keys in sorted order. Map-like containers are the most
596 expensive sort, only use them if you need one of these capabilities.
598 * a :ref:`set-like <ds_set>` container if you need to put a bunch of stuff into
599 a container that automatically eliminates duplicates. Some set-like
600 containers support efficient iteration through the elements in sorted order.
601 Set-like containers are more expensive than sequential containers.
603 * a :ref:`sequential <ds_sequential>` container provides the most efficient way
604 to add elements and keeps track of the order they are added to the collection.
605 They permit duplicates and support efficient iteration, but do not support
606 efficient look-up based on a key.
608 * a :ref:`string <ds_string>` container is a specialized sequential container or
609 reference structure that is used for character or byte arrays.
611 * a :ref:`bit <ds_bit>` container provides an efficient way to store and
612 perform set operations on sets of numeric id's, while automatically
613 eliminating duplicates. Bit containers require a maximum of 1 bit for each
614 identifier you want to store.
616 Once the proper category of container is determined, you can fine tune the
617 memory use, constant factors, and cache behaviors of access by intelligently
618 picking a member of the category. Note that constant factors and cache behavior
619 can be a big deal. If you have a vector that usually only contains a few
620 elements (but could contain many), for example, it's much better to use
621 :ref:`SmallVector <dss_smallvector>` than :ref:`vector <dss_vector>`. Doing so
622 avoids (relatively) expensive malloc/free calls, which dwarf the cost of adding
623 the elements to the container.
627 Sequential Containers (std::vector, std::list, etc)
628 ---------------------------------------------------
630 There are a variety of sequential containers available for you, based on your
631 needs. Pick the first in this section that will do what you want.
638 The ``llvm::ArrayRef`` class is the preferred class to use in an interface that
639 accepts a sequential list of elements in memory and just reads from them. By
640 taking an ``ArrayRef``, the API can be passed a fixed size array, an
641 ``std::vector``, an ``llvm::SmallVector`` and anything else that is contiguous
649 Fixed size arrays are very simple and very fast. They are good if you know
650 exactly how many elements you have, or you have a (low) upper bound on how many
655 Heap Allocated Arrays
656 ^^^^^^^^^^^^^^^^^^^^^
658 Heap allocated arrays (``new[]`` + ``delete[]``) are also simple. They are good
659 if the number of elements is variable, if you know how many elements you will
660 need before the array is allocated, and if the array is usually large (if not,
661 consider a :ref:`SmallVector <dss_smallvector>`). The cost of a heap allocated
662 array is the cost of the new/delete (aka malloc/free). Also note that if you
663 are allocating an array of a type with a constructor, the constructor and
664 destructors will be run for every element in the array (re-sizable vectors only
665 construct those elements actually used).
667 .. _dss_tinyptrvector:
669 llvm/ADT/TinyPtrVector.h
670 ^^^^^^^^^^^^^^^^^^^^^^^^
672 ``TinyPtrVector<Type>`` is a highly specialized collection class that is
673 optimized to avoid allocation in the case when a vector has zero or one
674 elements. It has two major restrictions: 1) it can only hold values of pointer
675 type, and 2) it cannot hold a null pointer.
677 Since this container is highly specialized, it is rarely used.
681 llvm/ADT/SmallVector.h
682 ^^^^^^^^^^^^^^^^^^^^^^
684 ``SmallVector<Type, N>`` is a simple class that looks and smells just like
685 ``vector<Type>``: it supports efficient iteration, lays out elements in memory
686 order (so you can do pointer arithmetic between elements), supports efficient
687 push_back/pop_back operations, supports efficient random access to its elements,
690 The advantage of SmallVector is that it allocates space for some number of
691 elements (N) **in the object itself**. Because of this, if the SmallVector is
692 dynamically smaller than N, no malloc is performed. This can be a big win in
693 cases where the malloc/free call is far more expensive than the code that
694 fiddles around with the elements.
696 This is good for vectors that are "usually small" (e.g. the number of
697 predecessors/successors of a block is usually less than 8). On the other hand,
698 this makes the size of the SmallVector itself large, so you don't want to
699 allocate lots of them (doing so will waste a lot of space). As such,
700 SmallVectors are most useful when on the stack.
702 SmallVector also provides a nice portable and efficient replacement for
707 Prefer to use ``SmallVectorImpl<T>`` as a parameter type.
709 In APIs that don't care about the "small size" (most?), prefer to use
710 the ``SmallVectorImpl<T>`` class, which is basically just the "vector
711 header" (and methods) without the elements allocated after it. Note that
712 ``SmallVector<T, N>`` inherits from ``SmallVectorImpl<T>`` so the
713 conversion is implicit and costs nothing. E.g.
717 // BAD: Clients cannot pass e.g. SmallVector<Foo, 4>.
718 hardcodedSmallSize(SmallVector<Foo, 2> &Out);
719 // GOOD: Clients can pass any SmallVector<Foo, N>.
720 allowsAnySmallSize(SmallVectorImpl<Foo> &Out);
723 SmallVector<Foo, 8> Vec;
724 hardcodedSmallSize(Vec); // Error.
725 allowsAnySmallSize(Vec); // Works.
728 Even though it has "``Impl``" in the name, this is so widely used that
729 it really isn't "private to the implementation" anymore. A name like
730 ``SmallVectorHeader`` would be more appropriate.
737 ``std::vector`` is well loved and respected. It is useful when SmallVector
738 isn't: when the size of the vector is often large (thus the small optimization
739 will rarely be a benefit) or if you will be allocating many instances of the
740 vector itself (which would waste space for elements that aren't in the
741 container). vector is also useful when interfacing with code that expects
744 One worthwhile note about std::vector: avoid code like this:
753 Instead, write this as:
763 Doing so will save (at least) one heap allocation and free per iteration of the
771 ``std::deque`` is, in some senses, a generalized version of ``std::vector``.
772 Like ``std::vector``, it provides constant time random access and other similar
773 properties, but it also provides efficient access to the front of the list. It
774 does not guarantee continuity of elements within memory.
776 In exchange for this extra flexibility, ``std::deque`` has significantly higher
777 constant factor costs than ``std::vector``. If possible, use ``std::vector`` or
785 ``std::list`` is an extremely inefficient class that is rarely useful. It
786 performs a heap allocation for every element inserted into it, thus having an
787 extremely high constant factor, particularly for small data types.
788 ``std::list`` also only supports bidirectional iteration, not random access
791 In exchange for this high cost, std::list supports efficient access to both ends
792 of the list (like ``std::deque``, but unlike ``std::vector`` or
793 ``SmallVector``). In addition, the iterator invalidation characteristics of
794 std::list are stronger than that of a vector class: inserting or removing an
795 element into the list does not invalidate iterator or pointers to other elements
803 ``ilist<T>`` implements an 'intrusive' doubly-linked list. It is intrusive,
804 because it requires the element to store and provide access to the prev/next
805 pointers for the list.
807 ``ilist`` has the same drawbacks as ``std::list``, and additionally requires an
808 ``ilist_traits`` implementation for the element type, but it provides some novel
809 characteristics. In particular, it can efficiently store polymorphic objects,
810 the traits class is informed when an element is inserted or removed from the
811 list, and ``ilist``\ s are guaranteed to support a constant-time splice
814 These properties are exactly what we want for things like ``Instruction``\ s and
815 basic blocks, which is why these are implemented with ``ilist``\ s.
817 Related classes of interest are explained in the following subsections:
819 * :ref:`ilist_traits <dss_ilist_traits>`
821 * :ref:`iplist <dss_iplist>`
823 * :ref:`llvm/ADT/ilist_node.h <dss_ilist_node>`
825 * :ref:`Sentinels <dss_ilist_sentinel>`
827 .. _dss_packedvector:
829 llvm/ADT/PackedVector.h
830 ^^^^^^^^^^^^^^^^^^^^^^^
832 Useful for storing a vector of values using only a few number of bits for each
833 value. Apart from the standard operations of a vector-like container, it can
834 also perform an 'or' set operation.
842 FirstCondition = 0x1,
843 SecondCondition = 0x2,
848 PackedVector<State, 2> Vec1;
849 Vec1.push_back(FirstCondition);
851 PackedVector<State, 2> Vec2;
852 Vec2.push_back(SecondCondition);
855 return Vec1[0]; // returns 'Both'.
858 .. _dss_ilist_traits:
863 ``ilist_traits<T>`` is ``ilist<T>``'s customization mechanism. ``iplist<T>``
864 (and consequently ``ilist<T>``) publicly derive from this traits class.
871 ``iplist<T>`` is ``ilist<T>``'s base and as such supports a slightly narrower
872 interface. Notably, inserters from ``T&`` are absent.
874 ``ilist_traits<T>`` is a public base of this class and can be used for a wide
875 variety of customizations.
879 llvm/ADT/ilist_node.h
880 ^^^^^^^^^^^^^^^^^^^^^
882 ``ilist_node<T>`` implements the forward and backward links that are expected
883 by the ``ilist<T>`` (and analogous containers) in the default manner.
885 ``ilist_node<T>``\ s are meant to be embedded in the node type ``T``, usually
886 ``T`` publicly derives from ``ilist_node<T>``.
888 .. _dss_ilist_sentinel:
893 ``ilist``\ s have another specialty that must be considered. To be a good
894 citizen in the C++ ecosystem, it needs to support the standard container
895 operations, such as ``begin`` and ``end`` iterators, etc. Also, the
896 ``operator--`` must work correctly on the ``end`` iterator in the case of
897 non-empty ``ilist``\ s.
899 The only sensible solution to this problem is to allocate a so-called *sentinel*
900 along with the intrusive list, which serves as the ``end`` iterator, providing
901 the back-link to the last element. However conforming to the C++ convention it
902 is illegal to ``operator++`` beyond the sentinel and it also must not be
905 These constraints allow for some implementation freedom to the ``ilist`` how to
906 allocate and store the sentinel. The corresponding policy is dictated by
907 ``ilist_traits<T>``. By default a ``T`` gets heap-allocated whenever the need
908 for a sentinel arises.
910 While the default policy is sufficient in most cases, it may break down when
911 ``T`` does not provide a default constructor. Also, in the case of many
912 instances of ``ilist``\ s, the memory overhead of the associated sentinels is
913 wasted. To alleviate the situation with numerous and voluminous
914 ``T``-sentinels, sometimes a trick is employed, leading to *ghostly sentinels*.
916 Ghostly sentinels are obtained by specially-crafted ``ilist_traits<T>`` which
917 superpose the sentinel with the ``ilist`` instance in memory. Pointer
918 arithmetic is used to obtain the sentinel, which is relative to the ``ilist``'s
919 ``this`` pointer. The ``ilist`` is augmented by an extra pointer, which serves
920 as the back-link of the sentinel. This is the only field in the ghostly
921 sentinel which can be legally accessed.
925 Other Sequential Container options
926 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
928 Other STL containers are available, such as ``std::string``.
930 There are also various STL adapter classes such as ``std::queue``,
931 ``std::priority_queue``, ``std::stack``, etc. These provide simplified access
932 to an underlying container but don't affect the cost of the container itself.
936 String-like containers
937 ----------------------
939 There are a variety of ways to pass around and use strings in C and C++, and
940 LLVM adds a few new options to choose from. Pick the first option on this list
941 that will do what you need, they are ordered according to their relative cost.
943 Note that is is generally preferred to *not* pass strings around as ``const
944 char*``'s. These have a number of problems, including the fact that they
945 cannot represent embedded nul ("\0") characters, and do not have a length
946 available efficiently. The general replacement for '``const char*``' is
949 For more information on choosing string containers for APIs, please see
950 :ref:`Passing Strings <string_apis>`.
957 The StringRef class is a simple value class that contains a pointer to a
958 character and a length, and is quite related to the :ref:`ArrayRef
959 <dss_arrayref>` class (but specialized for arrays of characters). Because
960 StringRef carries a length with it, it safely handles strings with embedded nul
961 characters in it, getting the length does not require a strlen call, and it even
962 has very convenient APIs for slicing and dicing the character range that it
965 StringRef is ideal for passing simple strings around that are known to be live,
966 either because they are C string literals, std::string, a C array, or a
967 SmallVector. Each of these cases has an efficient implicit conversion to
968 StringRef, which doesn't result in a dynamic strlen being executed.
970 StringRef has a few major limitations which make more powerful string containers
973 #. You cannot directly convert a StringRef to a 'const char*' because there is
974 no way to add a trailing nul (unlike the .c_str() method on various stronger
977 #. StringRef doesn't own or keep alive the underlying string bytes.
978 As such it can easily lead to dangling pointers, and is not suitable for
979 embedding in datastructures in most cases (instead, use an std::string or
980 something like that).
982 #. For the same reason, StringRef cannot be used as the return value of a
983 method if the method "computes" the result string. Instead, use std::string.
985 #. StringRef's do not allow you to mutate the pointed-to string bytes and it
986 doesn't allow you to insert or remove bytes from the range. For editing
987 operations like this, it interoperates with the :ref:`Twine <dss_twine>`
990 Because of its strengths and limitations, it is very common for a function to
991 take a StringRef and for a method on an object to return a StringRef that points
992 into some string that it owns.
999 The Twine class is used as an intermediary datatype for APIs that want to take a
1000 string that can be constructed inline with a series of concatenations. Twine
1001 works by forming recursive instances of the Twine datatype (a simple value
1002 object) on the stack as temporary objects, linking them together into a tree
1003 which is then linearized when the Twine is consumed. Twine is only safe to use
1004 as the argument to a function, and should always be a const reference, e.g.:
1008 void foo(const Twine &T);
1012 foo(X + "." + Twine(i));
1014 This example forms a string like "blarg.42" by concatenating the values
1015 together, and does not form intermediate strings containing "blarg" or "blarg.".
1017 Because Twine is constructed with temporary objects on the stack, and because
1018 these instances are destroyed at the end of the current statement, it is an
1019 inherently dangerous API. For example, this simple variant contains undefined
1020 behavior and will probably crash:
1024 void foo(const Twine &T);
1028 const Twine &Tmp = X + "." + Twine(i);
1031 ... because the temporaries are destroyed before the call. That said, Twine's
1032 are much more efficient than intermediate std::string temporaries, and they work
1033 really well with StringRef. Just be aware of their limitations.
1035 .. _dss_smallstring:
1037 llvm/ADT/SmallString.h
1038 ^^^^^^^^^^^^^^^^^^^^^^
1040 SmallString is a subclass of :ref:`SmallVector <dss_smallvector>` that adds some
1041 convenience APIs like += that takes StringRef's. SmallString avoids allocating
1042 memory in the case when the preallocated space is enough to hold its data, and
1043 it calls back to general heap allocation when required. Since it owns its data,
1044 it is very safe to use and supports full mutation of the string.
1046 Like SmallVector's, the big downside to SmallString is their sizeof. While they
1047 are optimized for small strings, they themselves are not particularly small.
1048 This means that they work great for temporary scratch buffers on the stack, but
1049 should not generally be put into the heap: it is very rare to see a SmallString
1050 as the member of a frequently-allocated heap data structure or returned
1058 The standard C++ std::string class is a very general class that (like
1059 SmallString) owns its underlying data. sizeof(std::string) is very reasonable
1060 so it can be embedded into heap data structures and returned by-value. On the
1061 other hand, std::string is highly inefficient for inline editing (e.g.
1062 concatenating a bunch of stuff together) and because it is provided by the
1063 standard library, its performance characteristics depend a lot of the host
1064 standard library (e.g. libc++ and MSVC provide a highly optimized string class,
1065 GCC contains a really slow implementation).
1067 The major disadvantage of std::string is that almost every operation that makes
1068 them larger can allocate memory, which is slow. As such, it is better to use
1069 SmallVector or Twine as a scratch buffer, but then use std::string to persist
1074 Set-Like Containers (std::set, SmallSet, SetVector, etc)
1075 --------------------------------------------------------
1077 Set-like containers are useful when you need to canonicalize multiple values
1078 into a single representation. There are several different choices for how to do
1079 this, providing various trade-offs.
1081 .. _dss_sortedvectorset:
1086 If you intend to insert a lot of elements, then do a lot of queries, a great
1087 approach is to use a vector (or other sequential container) with
1088 std::sort+std::unique to remove duplicates. This approach works really well if
1089 your usage pattern has these two distinct phases (insert then query), and can be
1090 coupled with a good choice of :ref:`sequential container <ds_sequential>`.
1092 This combination provides the several nice properties: the result data is
1093 contiguous in memory (good for cache locality), has few allocations, is easy to
1094 address (iterators in the final vector are just indices or pointers), and can be
1095 efficiently queried with a standard binary search (e.g.
1096 ``std::lower_bound``; if you want the whole range of elements comparing
1097 equal, use ``std::equal_range``).
1104 If you have a set-like data structure that is usually small and whose elements
1105 are reasonably small, a ``SmallSet<Type, N>`` is a good choice. This set has
1106 space for N elements in place (thus, if the set is dynamically smaller than N,
1107 no malloc traffic is required) and accesses them with a simple linear search.
1108 When the set grows beyond 'N' elements, it allocates a more expensive
1109 representation that guarantees efficient access (for most types, it falls back
1110 to std::set, but for pointers it uses something far better, :ref:`SmallPtrSet
1113 The magic of this class is that it handles small sets extremely efficiently, but
1114 gracefully handles extremely large sets without loss of efficiency. The
1115 drawback is that the interface is quite small: it supports insertion, queries
1116 and erasing, but does not support iteration.
1118 .. _dss_smallptrset:
1120 llvm/ADT/SmallPtrSet.h
1121 ^^^^^^^^^^^^^^^^^^^^^^
1123 SmallPtrSet has all the advantages of ``SmallSet`` (and a ``SmallSet`` of
1124 pointers is transparently implemented with a ``SmallPtrSet``), but also supports
1125 iterators. If more than 'N' insertions are performed, a single quadratically
1126 probed hash table is allocated and grows as needed, providing extremely
1127 efficient access (constant time insertion/deleting/queries with low constant
1128 factors) and is very stingy with malloc traffic.
1130 Note that, unlike ``std::set``, the iterators of ``SmallPtrSet`` are invalidated
1131 whenever an insertion occurs. Also, the values visited by the iterators are not
1132 visited in sorted order.
1139 DenseSet is a simple quadratically probed hash table. It excels at supporting
1140 small values: it uses a single allocation to hold all of the pairs that are
1141 currently inserted in the set. DenseSet is a great way to unique small values
1142 that are not simple pointers (use :ref:`SmallPtrSet <dss_smallptrset>` for
1143 pointers). Note that DenseSet has the same requirements for the value type that
1144 :ref:`DenseMap <dss_densemap>` has.
1148 llvm/ADT/SparseSet.h
1149 ^^^^^^^^^^^^^^^^^^^^
1151 SparseSet holds a small number of objects identified by unsigned keys of
1152 moderate size. It uses a lot of memory, but provides operations that are almost
1153 as fast as a vector. Typical keys are physical registers, virtual registers, or
1154 numbered basic blocks.
1156 SparseSet is useful for algorithms that need very fast clear/find/insert/erase
1157 and fast iteration over small sets. It is not intended for building composite
1160 .. _dss_sparsemultiset:
1162 llvm/ADT/SparseMultiSet.h
1163 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1165 SparseMultiSet adds multiset behavior to SparseSet, while retaining SparseSet's
1166 desirable attributes. Like SparseSet, it typically uses a lot of memory, but
1167 provides operations that are almost as fast as a vector. Typical keys are
1168 physical registers, virtual registers, or numbered basic blocks.
1170 SparseMultiSet is useful for algorithms that need very fast
1171 clear/find/insert/erase of the entire collection, and iteration over sets of
1172 elements sharing a key. It is often a more efficient choice than using composite
1173 data structures (e.g. vector-of-vectors, map-of-vectors). It is not intended for
1174 building composite data structures.
1178 llvm/ADT/FoldingSet.h
1179 ^^^^^^^^^^^^^^^^^^^^^
1181 FoldingSet is an aggregate class that is really good at uniquing
1182 expensive-to-create or polymorphic objects. It is a combination of a chained
1183 hash table with intrusive links (uniqued objects are required to inherit from
1184 FoldingSetNode) that uses :ref:`SmallVector <dss_smallvector>` as part of its ID
1187 Consider a case where you want to implement a "getOrCreateFoo" method for a
1188 complex object (for example, a node in the code generator). The client has a
1189 description of **what** it wants to generate (it knows the opcode and all the
1190 operands), but we don't want to 'new' a node, then try inserting it into a set
1191 only to find out it already exists, at which point we would have to delete it
1192 and return the node that already exists.
1194 To support this style of client, FoldingSet perform a query with a
1195 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1196 element that we want to query for. The query either returns the element
1197 matching the ID or it returns an opaque ID that indicates where insertion should
1198 take place. Construction of the ID usually does not require heap traffic.
1200 Because FoldingSet uses intrusive links, it can support polymorphic objects in
1201 the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1202 Because the elements are individually allocated, pointers to the elements are
1203 stable: inserting or removing elements does not invalidate any pointers to other
1211 ``std::set`` is a reasonable all-around set class, which is decent at many
1212 things but great at nothing. std::set allocates memory for each element
1213 inserted (thus it is very malloc intensive) and typically stores three pointers
1214 per element in the set (thus adding a large amount of per-element space
1215 overhead). It offers guaranteed log(n) performance, which is not particularly
1216 fast from a complexity standpoint (particularly if the elements of the set are
1217 expensive to compare, like strings), and has extremely high constant factors for
1218 lookup, insertion and removal.
1220 The advantages of std::set are that its iterators are stable (deleting or
1221 inserting an element from the set does not affect iterators or pointers to other
1222 elements) and that iteration over the set is guaranteed to be in sorted order.
1223 If the elements in the set are large, then the relative overhead of the pointers
1224 and malloc traffic is not a big deal, but if the elements of the set are small,
1225 std::set is almost never a good choice.
1229 llvm/ADT/SetVector.h
1230 ^^^^^^^^^^^^^^^^^^^^
1232 LLVM's ``SetVector<Type>`` is an adapter class that combines your choice of a
1233 set-like container along with a :ref:`Sequential Container <ds_sequential>` The
1234 important property that this provides is efficient insertion with uniquing
1235 (duplicate elements are ignored) with iteration support. It implements this by
1236 inserting elements into both a set-like container and the sequential container,
1237 using the set-like container for uniquing and the sequential container for
1240 The difference between SetVector and other sets is that the order of iteration
1241 is guaranteed to match the order of insertion into the SetVector. This property
1242 is really important for things like sets of pointers. Because pointer values
1243 are non-deterministic (e.g. vary across runs of the program on different
1244 machines), iterating over the pointers in the set will not be in a well-defined
1247 The drawback of SetVector is that it requires twice as much space as a normal
1248 set and has the sum of constant factors from the set-like container and the
1249 sequential container that it uses. Use it **only** if you need to iterate over
1250 the elements in a deterministic order. SetVector is also expensive to delete
1251 elements out of (linear time), unless you use its "pop_back" method, which is
1254 ``SetVector`` is an adapter class that defaults to using ``std::vector`` and a
1255 size 16 ``SmallSet`` for the underlying containers, so it is quite expensive.
1256 However, ``"llvm/ADT/SetVector.h"`` also provides a ``SmallSetVector`` class,
1257 which defaults to using a ``SmallVector`` and ``SmallSet`` of a specified size.
1258 If you use this, and if your sets are dynamically smaller than ``N``, you will
1259 save a lot of heap traffic.
1261 .. _dss_uniquevector:
1263 llvm/ADT/UniqueVector.h
1264 ^^^^^^^^^^^^^^^^^^^^^^^
1266 UniqueVector is similar to :ref:`SetVector <dss_setvector>` but it retains a
1267 unique ID for each element inserted into the set. It internally contains a map
1268 and a vector, and it assigns a unique ID for each value inserted into the set.
1270 UniqueVector is very expensive: its cost is the sum of the cost of maintaining
1271 both the map and vector, it has high complexity, high constant factors, and
1272 produces a lot of malloc traffic. It should be avoided.
1274 .. _dss_immutableset:
1276 llvm/ADT/ImmutableSet.h
1277 ^^^^^^^^^^^^^^^^^^^^^^^
1279 ImmutableSet is an immutable (functional) set implementation based on an AVL
1280 tree. Adding or removing elements is done through a Factory object and results
1281 in the creation of a new ImmutableSet object. If an ImmutableSet already exists
1282 with the given contents, then the existing one is returned; equality is compared
1283 with a FoldingSetNodeID. The time and space complexity of add or remove
1284 operations is logarithmic in the size of the original set.
1286 There is no method for returning an element of the set, you can only check for
1291 Other Set-Like Container Options
1292 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1294 The STL provides several other options, such as std::multiset and the various
1295 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1296 never use hash_set and unordered_set because they are generally very expensive
1297 (each insertion requires a malloc) and very non-portable.
1299 std::multiset is useful if you're not interested in elimination of duplicates,
1300 but has all the drawbacks of std::set. A sorted vector (where you don't delete
1301 duplicate entries) or some other approach is almost always better.
1305 Map-Like Containers (std::map, DenseMap, etc)
1306 ---------------------------------------------
1308 Map-like containers are useful when you want to associate data to a key. As
1309 usual, there are a lot of different ways to do this. :)
1311 .. _dss_sortedvectormap:
1316 If your usage pattern follows a strict insert-then-query approach, you can
1317 trivially use the same approach as :ref:`sorted vectors for set-like containers
1318 <dss_sortedvectorset>`. The only difference is that your query function (which
1319 uses std::lower_bound to get efficient log(n) lookup) should only compare the
1320 key, not both the key and value. This yields the same advantages as sorted
1325 llvm/ADT/StringMap.h
1326 ^^^^^^^^^^^^^^^^^^^^
1328 Strings are commonly used as keys in maps, and they are difficult to support
1329 efficiently: they are variable length, inefficient to hash and compare when
1330 long, expensive to copy, etc. StringMap is a specialized container designed to
1331 cope with these issues. It supports mapping an arbitrary range of bytes to an
1332 arbitrary other object.
1334 The StringMap implementation uses a quadratically-probed hash table, where the
1335 buckets store a pointer to the heap allocated entries (and some other stuff).
1336 The entries in the map must be heap allocated because the strings are variable
1337 length. The string data (key) and the element object (value) are stored in the
1338 same allocation with the string data immediately after the element object.
1339 This container guarantees the "``(char*)(&Value+1)``" points to the key string
1342 The StringMap is very fast for several reasons: quadratic probing is very cache
1343 efficient for lookups, the hash value of strings in buckets is not recomputed
1344 when looking up an element, StringMap rarely has to touch the memory for
1345 unrelated objects when looking up a value (even when hash collisions happen),
1346 hash table growth does not recompute the hash values for strings already in the
1347 table, and each pair in the map is store in a single allocation (the string data
1348 is stored in the same allocation as the Value of a pair).
1350 StringMap also provides query methods that take byte ranges, so it only ever
1351 copies a string if a value is inserted into the table.
1353 StringMap iteratation order, however, is not guaranteed to be deterministic, so
1354 any uses which require that should instead use a std::map.
1358 llvm/ADT/IndexedMap.h
1359 ^^^^^^^^^^^^^^^^^^^^^
1361 IndexedMap is a specialized container for mapping small dense integers (or
1362 values that can be mapped to small dense integers) to some other type. It is
1363 internally implemented as a vector with a mapping function that maps the keys
1364 to the dense integer range.
1366 This is useful for cases like virtual registers in the LLVM code generator: they
1367 have a dense mapping that is offset by a compile-time constant (the first
1368 virtual register ID).
1375 DenseMap is a simple quadratically probed hash table. It excels at supporting
1376 small keys and values: it uses a single allocation to hold all of the pairs
1377 that are currently inserted in the map. DenseMap is a great way to map
1378 pointers to pointers, or map other small types to each other.
1380 There are several aspects of DenseMap that you should be aware of, however.
1381 The iterators in a DenseMap are invalidated whenever an insertion occurs,
1382 unlike map. Also, because DenseMap allocates space for a large number of
1383 key/value pairs (it starts with 64 by default), it will waste a lot of space if
1384 your keys or values are large. Finally, you must implement a partial
1385 specialization of DenseMapInfo for the key that you want, if it isn't already
1386 supported. This is required to tell DenseMap about two special marker values
1387 (which can never be inserted into the map) that it needs internally.
1389 DenseMap's find_as() method supports lookup operations using an alternate key
1390 type. This is useful in cases where the normal key type is expensive to
1391 construct, but cheap to compare against. The DenseMapInfo is responsible for
1392 defining the appropriate comparison and hashing methods for each alternate key
1400 ValueMap is a wrapper around a :ref:`DenseMap <dss_densemap>` mapping
1401 ``Value*``\ s (or subclasses) to another type. When a Value is deleted or
1402 RAUW'ed, ValueMap will update itself so the new version of the key is mapped to
1403 the same value, just as if the key were a WeakVH. You can configure exactly how
1404 this happens, and what else happens on these two events, by passing a ``Config``
1405 parameter to the ValueMap template.
1407 .. _dss_intervalmap:
1409 llvm/ADT/IntervalMap.h
1410 ^^^^^^^^^^^^^^^^^^^^^^
1412 IntervalMap is a compact map for small keys and values. It maps key intervals
1413 instead of single keys, and it will automatically coalesce adjacent intervals.
1414 When the map only contains a few intervals, they are stored in the map object
1415 itself to avoid allocations.
1417 The IntervalMap iterators are quite big, so they should not be passed around as
1418 STL iterators. The heavyweight iterators allow a smaller data structure.
1425 std::map has similar characteristics to :ref:`std::set <dss_set>`: it uses a
1426 single allocation per pair inserted into the map, it offers log(n) lookup with
1427 an extremely large constant factor, imposes a space penalty of 3 pointers per
1428 pair in the map, etc.
1430 std::map is most useful when your keys or values are very large, if you need to
1431 iterate over the collection in sorted order, or if you need stable iterators
1432 into the map (i.e. they don't get invalidated if an insertion or deletion of
1433 another element takes place).
1437 llvm/ADT/MapVector.h
1438 ^^^^^^^^^^^^^^^^^^^^
1440 ``MapVector<KeyT,ValueT>`` provides a subset of the DenseMap interface. The
1441 main difference is that the iteration order is guaranteed to be the insertion
1442 order, making it an easy (but somewhat expensive) solution for non-deterministic
1443 iteration over maps of pointers.
1445 It is implemented by mapping from key to an index in a vector of key,value
1446 pairs. This provides fast lookup and iteration, but has two main drawbacks:
1447 the key is stored twice and removing elements takes linear time. If it is
1448 necessary to remove elements, it's best to remove them in bulk using
1451 .. _dss_inteqclasses:
1453 llvm/ADT/IntEqClasses.h
1454 ^^^^^^^^^^^^^^^^^^^^^^^
1456 IntEqClasses provides a compact representation of equivalence classes of small
1457 integers. Initially, each integer in the range 0..n-1 has its own equivalence
1458 class. Classes can be joined by passing two class representatives to the
1459 join(a, b) method. Two integers are in the same class when findLeader() returns
1460 the same representative.
1462 Once all equivalence classes are formed, the map can be compressed so each
1463 integer 0..n-1 maps to an equivalence class number in the range 0..m-1, where m
1464 is the total number of equivalence classes. The map must be uncompressed before
1465 it can be edited again.
1467 .. _dss_immutablemap:
1469 llvm/ADT/ImmutableMap.h
1470 ^^^^^^^^^^^^^^^^^^^^^^^
1472 ImmutableMap is an immutable (functional) map implementation based on an AVL
1473 tree. Adding or removing elements is done through a Factory object and results
1474 in the creation of a new ImmutableMap object. If an ImmutableMap already exists
1475 with the given key set, then the existing one is returned; equality is compared
1476 with a FoldingSetNodeID. The time and space complexity of add or remove
1477 operations is logarithmic in the size of the original map.
1481 Other Map-Like Container Options
1482 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1484 The STL provides several other options, such as std::multimap and the various
1485 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1486 never use hash_set and unordered_set because they are generally very expensive
1487 (each insertion requires a malloc) and very non-portable.
1489 std::multimap is useful if you want to map a key to multiple values, but has all
1490 the drawbacks of std::map. A sorted vector or some other approach is almost
1495 Bit storage containers (BitVector, SparseBitVector)
1496 ---------------------------------------------------
1498 Unlike the other containers, there are only two bit storage containers, and
1499 choosing when to use each is relatively straightforward.
1501 One additional option is ``std::vector<bool>``: we discourage its use for two
1502 reasons 1) the implementation in many common compilers (e.g. commonly
1503 available versions of GCC) is extremely inefficient and 2) the C++ standards
1504 committee is likely to deprecate this container and/or change it significantly
1505 somehow. In any case, please don't use it.
1512 The BitVector container provides a dynamic size set of bits for manipulation.
1513 It supports individual bit setting/testing, as well as set operations. The set
1514 operations take time O(size of bitvector), but operations are performed one word
1515 at a time, instead of one bit at a time. This makes the BitVector very fast for
1516 set operations compared to other containers. Use the BitVector when you expect
1517 the number of set bits to be high (i.e. a dense set).
1519 .. _dss_smallbitvector:
1524 The SmallBitVector container provides the same interface as BitVector, but it is
1525 optimized for the case where only a small number of bits, less than 25 or so,
1526 are needed. It also transparently supports larger bit counts, but slightly less
1527 efficiently than a plain BitVector, so SmallBitVector should only be used when
1528 larger counts are rare.
1530 At this time, SmallBitVector does not support set operations (and, or, xor), and
1531 its operator[] does not provide an assignable lvalue.
1533 .. _dss_sparsebitvector:
1538 The SparseBitVector container is much like BitVector, with one major difference:
1539 Only the bits that are set, are stored. This makes the SparseBitVector much
1540 more space efficient than BitVector when the set is sparse, as well as making
1541 set operations O(number of set bits) instead of O(size of universe). The
1542 downside to the SparseBitVector is that setting and testing of random bits is
1543 O(N), and on large SparseBitVectors, this can be slower than BitVector. In our
1544 implementation, setting or testing bits in sorted order (either forwards or
1545 reverse) is O(1) worst case. Testing and setting bits within 128 bits (depends
1546 on size) of the current bit is also O(1). As a general statement,
1547 testing/setting bits in a SparseBitVector is O(distance away from last set bit).
1551 Helpful Hints for Common Operations
1552 ===================================
1554 This section describes how to perform some very simple transformations of LLVM
1555 code. This is meant to give examples of common idioms used, showing the
1556 practical side of LLVM transformations.
1558 Because this is a "how-to" section, you should also read about the main classes
1559 that you will be working with. The :ref:`Core LLVM Class Hierarchy Reference
1560 <coreclasses>` contains details and descriptions of the main classes that you
1565 Basic Inspection and Traversal Routines
1566 ---------------------------------------
1568 The LLVM compiler infrastructure have many different data structures that may be
1569 traversed. Following the example of the C++ standard template library, the
1570 techniques used to traverse these various data structures are all basically the
1571 same. For a enumerable sequence of values, the ``XXXbegin()`` function (or
1572 method) returns an iterator to the start of the sequence, the ``XXXend()``
1573 function returns an iterator pointing to one past the last valid element of the
1574 sequence, and there is some ``XXXiterator`` data type that is common between the
1577 Because the pattern for iteration is common across many different aspects of the
1578 program representation, the standard template library algorithms may be used on
1579 them, and it is easier to remember how to iterate. First we show a few common
1580 examples of the data structures that need to be traversed. Other data
1581 structures are traversed in very similar ways.
1583 .. _iterate_function:
1585 Iterating over the ``BasicBlock`` in a ``Function``
1586 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1588 It's quite common to have a ``Function`` instance that you'd like to transform
1589 in some way; in particular, you'd like to manipulate its ``BasicBlock``\ s. To
1590 facilitate this, you'll need to iterate over all of the ``BasicBlock``\ s that
1591 constitute the ``Function``. The following is an example that prints the name
1592 of a ``BasicBlock`` and the number of ``Instruction``\ s it contains:
1596 // func is a pointer to a Function instance
1597 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
1598 // Print out the name of the basic block if it has one, and then the
1599 // number of instructions that it contains
1600 errs() << "Basic block (name=" << i->getName() << ") has "
1601 << i->size() << " instructions.\n";
1603 Note that i can be used as if it were a pointer for the purposes of invoking
1604 member functions of the ``Instruction`` class. This is because the indirection
1605 operator is overloaded for the iterator classes. In the above code, the
1606 expression ``i->size()`` is exactly equivalent to ``(*i).size()`` just like
1609 .. _iterate_basicblock:
1611 Iterating over the ``Instruction`` in a ``BasicBlock``
1612 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1614 Just like when dealing with ``BasicBlock``\ s in ``Function``\ s, it's easy to
1615 iterate over the individual instructions that make up ``BasicBlock``\ s. Here's
1616 a code snippet that prints out each instruction in a ``BasicBlock``:
1620 // blk is a pointer to a BasicBlock instance
1621 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
1622 // The next statement works since operator<<(ostream&,...)
1623 // is overloaded for Instruction&
1624 errs() << *i << "\n";
1627 However, this isn't really the best way to print out the contents of a
1628 ``BasicBlock``! Since the ostream operators are overloaded for virtually
1629 anything you'll care about, you could have just invoked the print routine on the
1630 basic block itself: ``errs() << *blk << "\n";``.
1632 .. _iterate_insiter:
1634 Iterating over the ``Instruction`` in a ``Function``
1635 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1637 If you're finding that you commonly iterate over a ``Function``'s
1638 ``BasicBlock``\ s and then that ``BasicBlock``'s ``Instruction``\ s,
1639 ``InstIterator`` should be used instead. You'll need to include
1640 ``llvm/IR/InstIterator.h`` (`doxygen
1641 <http://llvm.org/doxygen/InstIterator_8h.html>`__) and then instantiate
1642 ``InstIterator``\ s explicitly in your code. Here's a small example that shows
1643 how to dump all instructions in a function to the standard error stream:
1647 #include "llvm/IR/InstIterator.h"
1649 // F is a pointer to a Function instance
1650 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1651 errs() << *I << "\n";
1653 Easy, isn't it? You can also use ``InstIterator``\ s to fill a work list with
1654 its initial contents. For example, if you wanted to initialize a work list to
1655 contain all instructions in a ``Function`` F, all you would need to do is
1660 std::set<Instruction*> worklist;
1661 // or better yet, SmallPtrSet<Instruction*, 64> worklist;
1663 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1664 worklist.insert(&*I);
1666 The STL set ``worklist`` would now contain all instructions in the ``Function``
1669 .. _iterate_convert:
1671 Turning an iterator into a class pointer (and vice-versa)
1672 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1674 Sometimes, it'll be useful to grab a reference (or pointer) to a class instance
1675 when all you've got at hand is an iterator. Well, extracting a reference or a
1676 pointer from an iterator is very straight-forward. Assuming that ``i`` is a
1677 ``BasicBlock::iterator`` and ``j`` is a ``BasicBlock::const_iterator``:
1681 Instruction& inst = *i; // Grab reference to instruction reference
1682 Instruction* pinst = &*i; // Grab pointer to instruction reference
1683 const Instruction& inst = *j;
1685 However, the iterators you'll be working with in the LLVM framework are special:
1686 they will automatically convert to a ptr-to-instance type whenever they need to.
1687 Instead of derferencing the iterator and then taking the address of the result,
1688 you can simply assign the iterator to the proper pointer type and you get the
1689 dereference and address-of operation as a result of the assignment (behind the
1690 scenes, this is a result of overloading casting mechanisms). Thus the last line
1691 of the last example,
1695 Instruction *pinst = &*i;
1697 is semantically equivalent to
1701 Instruction *pinst = i;
1703 It's also possible to turn a class pointer into the corresponding iterator, and
1704 this is a constant time operation (very efficient). The following code snippet
1705 illustrates use of the conversion constructors provided by LLVM iterators. By
1706 using these, you can explicitly grab the iterator of something without actually
1707 obtaining it via iteration over some structure:
1711 void printNextInstruction(Instruction* inst) {
1712 BasicBlock::iterator it(inst);
1713 ++it; // After this line, it refers to the instruction after *inst
1714 if (it != inst->getParent()->end()) errs() << *it << "\n";
1717 Unfortunately, these implicit conversions come at a cost; they prevent these
1718 iterators from conforming to standard iterator conventions, and thus from being
1719 usable with standard algorithms and containers. For example, they prevent the
1720 following code, where ``B`` is a ``BasicBlock``, from compiling:
1724 llvm::SmallVector<llvm::Instruction *, 16>(B->begin(), B->end());
1726 Because of this, these implicit conversions may be removed some day, and
1727 ``operator*`` changed to return a pointer instead of a reference.
1729 .. _iterate_complex:
1731 Finding call sites: a slightly more complex example
1732 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1734 Say that you're writing a FunctionPass and would like to count all the locations
1735 in the entire module (that is, across every ``Function``) where a certain
1736 function (i.e., some ``Function *``) is already in scope. As you'll learn
1737 later, you may want to use an ``InstVisitor`` to accomplish this in a much more
1738 straight-forward manner, but this example will allow us to explore how you'd do
1739 it if you didn't have ``InstVisitor`` around. In pseudo-code, this is what we
1742 .. code-block:: none
1744 initialize callCounter to zero
1745 for each Function f in the Module
1746 for each BasicBlock b in f
1747 for each Instruction i in b
1748 if (i is a CallInst and calls the given function)
1749 increment callCounter
1751 And the actual code is (remember, because we're writing a ``FunctionPass``, our
1752 ``FunctionPass``-derived class simply has to override the ``runOnFunction``
1757 Function* targetFunc = ...;
1759 class OurFunctionPass : public FunctionPass {
1761 OurFunctionPass(): callCounter(0) { }
1763 virtual runOnFunction(Function& F) {
1764 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1765 for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) {
1766 if (CallInst* callInst = dyn_cast<CallInst>(&*i)) {
1767 // We know we've encountered a call instruction, so we
1768 // need to determine if it's a call to the
1769 // function pointed to by m_func or not.
1770 if (callInst->getCalledFunction() == targetFunc)
1778 unsigned callCounter;
1781 .. _calls_and_invokes:
1783 Treating calls and invokes the same way
1784 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1786 You may have noticed that the previous example was a bit oversimplified in that
1787 it did not deal with call sites generated by 'invoke' instructions. In this,
1788 and in other situations, you may find that you want to treat ``CallInst``\ s and
1789 ``InvokeInst``\ s the same way, even though their most-specific common base
1790 class is ``Instruction``, which includes lots of less closely-related things.
1791 For these cases, LLVM provides a handy wrapper class called ``CallSite``
1792 (`doxygen <http://llvm.org/doxygen/classllvm_1_1CallSite.html>`__) It is
1793 essentially a wrapper around an ``Instruction`` pointer, with some methods that
1794 provide functionality common to ``CallInst``\ s and ``InvokeInst``\ s.
1796 This class has "value semantics": it should be passed by value, not by reference
1797 and it should not be dynamically allocated or deallocated using ``operator new``
1798 or ``operator delete``. It is efficiently copyable, assignable and
1799 constructable, with costs equivalents to that of a bare pointer. If you look at
1800 its definition, it has only a single pointer member.
1804 Iterating over def-use & use-def chains
1805 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1807 Frequently, we might have an instance of the ``Value`` class (`doxygen
1808 <http://llvm.org/doxygen/classllvm_1_1Value.html>`__) and we want to determine
1809 which ``User`` s use the ``Value``. The list of all ``User``\ s of a particular
1810 ``Value`` is called a *def-use* chain. For example, let's say we have a
1811 ``Function*`` named ``F`` to a particular function ``foo``. Finding all of the
1812 instructions that *use* ``foo`` is as simple as iterating over the *def-use*
1819 for (User *U : F->users()) {
1820 if (Instruction *Inst = dyn_cast<Instruction>(U)) {
1821 errs() << "F is used in instruction:\n";
1822 errs() << *Inst << "\n";
1825 Alternatively, it's common to have an instance of the ``User`` Class (`doxygen
1826 <http://llvm.org/doxygen/classllvm_1_1User.html>`__) and need to know what
1827 ``Value``\ s are used by it. The list of all ``Value``\ s used by a ``User`` is
1828 known as a *use-def* chain. Instances of class ``Instruction`` are common
1829 ``User`` s, so we might want to iterate over all of the values that a particular
1830 instruction uses (that is, the operands of the particular ``Instruction``):
1834 Instruction *pi = ...;
1836 for (Use &U : pi->operands()) {
1841 Declaring objects as ``const`` is an important tool of enforcing mutation free
1842 algorithms (such as analyses, etc.). For this purpose above iterators come in
1843 constant flavors as ``Value::const_use_iterator`` and
1844 ``Value::const_op_iterator``. They automatically arise when calling
1845 ``use/op_begin()`` on ``const Value*``\ s or ``const User*``\ s respectively.
1846 Upon dereferencing, they return ``const Use*``\ s. Otherwise the above patterns
1851 Iterating over predecessors & successors of blocks
1852 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1854 Iterating over the predecessors and successors of a block is quite easy with the
1855 routines defined in ``"llvm/Support/CFG.h"``. Just use code like this to
1856 iterate over all predecessors of BB:
1860 #include "llvm/Support/CFG.h"
1861 BasicBlock *BB = ...;
1863 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
1864 BasicBlock *Pred = *PI;
1868 Similarly, to iterate over successors use ``succ_iterator/succ_begin/succ_end``.
1872 Making simple changes
1873 ---------------------
1875 There are some primitive transformation operations present in the LLVM
1876 infrastructure that are worth knowing about. When performing transformations,
1877 it's fairly common to manipulate the contents of basic blocks. This section
1878 describes some of the common methods for doing so and gives example code.
1880 .. _schanges_creating:
1882 Creating and inserting new ``Instruction``\ s
1883 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1885 *Instantiating Instructions*
1887 Creation of ``Instruction``\ s is straight-forward: simply call the constructor
1888 for the kind of instruction to instantiate and provide the necessary parameters.
1889 For example, an ``AllocaInst`` only *requires* a (const-ptr-to) ``Type``. Thus:
1893 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
1895 will create an ``AllocaInst`` instance that represents the allocation of one
1896 integer in the current stack frame, at run time. Each ``Instruction`` subclass
1897 is likely to have varying default parameters which change the semantics of the
1898 instruction, so refer to the `doxygen documentation for the subclass of
1899 Instruction <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_ that
1900 you're interested in instantiating.
1904 It is very useful to name the values of instructions when you're able to, as
1905 this facilitates the debugging of your transformations. If you end up looking
1906 at generated LLVM machine code, you definitely want to have logical names
1907 associated with the results of instructions! By supplying a value for the
1908 ``Name`` (default) parameter of the ``Instruction`` constructor, you associate a
1909 logical name with the result of the instruction's execution at run time. For
1910 example, say that I'm writing a transformation that dynamically allocates space
1911 for an integer on the stack, and that integer is going to be used as some kind
1912 of index by some other code. To accomplish this, I place an ``AllocaInst`` at
1913 the first point in the first ``BasicBlock`` of some ``Function``, and I'm
1914 intending to use it within the same ``Function``. I might do:
1918 AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
1920 where ``indexLoc`` is now the logical name of the instruction's execution value,
1921 which is a pointer to an integer on the run time stack.
1923 *Inserting instructions*
1925 There are essentially three ways to insert an ``Instruction`` into an existing
1926 sequence of instructions that form a ``BasicBlock``:
1928 * Insertion into an explicit instruction list
1930 Given a ``BasicBlock* pb``, an ``Instruction* pi`` within that ``BasicBlock``,
1931 and a newly-created instruction we wish to insert before ``*pi``, we do the
1936 BasicBlock *pb = ...;
1937 Instruction *pi = ...;
1938 Instruction *newInst = new Instruction(...);
1940 pb->getInstList().insert(pi, newInst); // Inserts newInst before pi in pb
1942 Appending to the end of a ``BasicBlock`` is so common that the ``Instruction``
1943 class and ``Instruction``-derived classes provide constructors which take a
1944 pointer to a ``BasicBlock`` to be appended to. For example code that looked
1949 BasicBlock *pb = ...;
1950 Instruction *newInst = new Instruction(...);
1952 pb->getInstList().push_back(newInst); // Appends newInst to pb
1958 BasicBlock *pb = ...;
1959 Instruction *newInst = new Instruction(..., pb);
1961 which is much cleaner, especially if you are creating long instruction
1964 * Insertion into an implicit instruction list
1966 ``Instruction`` instances that are already in ``BasicBlock``\ s are implicitly
1967 associated with an existing instruction list: the instruction list of the
1968 enclosing basic block. Thus, we could have accomplished the same thing as the
1969 above code without being given a ``BasicBlock`` by doing:
1973 Instruction *pi = ...;
1974 Instruction *newInst = new Instruction(...);
1976 pi->getParent()->getInstList().insert(pi, newInst);
1978 In fact, this sequence of steps occurs so frequently that the ``Instruction``
1979 class and ``Instruction``-derived classes provide constructors which take (as
1980 a default parameter) a pointer to an ``Instruction`` which the newly-created
1981 ``Instruction`` should precede. That is, ``Instruction`` constructors are
1982 capable of inserting the newly-created instance into the ``BasicBlock`` of a
1983 provided instruction, immediately before that instruction. Using an
1984 ``Instruction`` constructor with a ``insertBefore`` (default) parameter, the
1989 Instruction* pi = ...;
1990 Instruction* newInst = new Instruction(..., pi);
1992 which is much cleaner, especially if you're creating a lot of instructions and
1993 adding them to ``BasicBlock``\ s.
1995 * Insertion using an instance of ``IRBuilder``
1997 Inserting several ``Instruction``\ s can be quite laborious using the previous
1998 methods. The ``IRBuilder`` is a convenience class that can be used to add
1999 several instructions to the end of a ``BasicBlock`` or before a particular
2000 ``Instruction``. It also supports constant folding and renaming named
2001 registers (see ``IRBuilder``'s template arguments).
2003 The example below demonstrates a very simple use of the ``IRBuilder`` where
2004 three instructions are inserted before the instruction ``pi``. The first two
2005 instructions are Call instructions and third instruction multiplies the return
2006 value of the two calls.
2010 Instruction *pi = ...;
2011 IRBuilder<> Builder(pi);
2012 CallInst* callOne = Builder.CreateCall(...);
2013 CallInst* callTwo = Builder.CreateCall(...);
2014 Value* result = Builder.CreateMul(callOne, callTwo);
2016 The example below is similar to the above example except that the created
2017 ``IRBuilder`` inserts instructions at the end of the ``BasicBlock`` ``pb``.
2021 BasicBlock *pb = ...;
2022 IRBuilder<> Builder(pb);
2023 CallInst* callOne = Builder.CreateCall(...);
2024 CallInst* callTwo = Builder.CreateCall(...);
2025 Value* result = Builder.CreateMul(callOne, callTwo);
2027 See :doc:`tutorial/LangImpl3` for a practical use of the ``IRBuilder``.
2030 .. _schanges_deleting:
2032 Deleting Instructions
2033 ^^^^^^^^^^^^^^^^^^^^^
2035 Deleting an instruction from an existing sequence of instructions that form a
2036 BasicBlock_ is very straight-forward: just call the instruction's
2037 ``eraseFromParent()`` method. For example:
2041 Instruction *I = .. ;
2042 I->eraseFromParent();
2044 This unlinks the instruction from its containing basic block and deletes it. If
2045 you'd just like to unlink the instruction from its containing basic block but
2046 not delete it, you can use the ``removeFromParent()`` method.
2048 .. _schanges_replacing:
2050 Replacing an Instruction with another Value
2051 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2053 Replacing individual instructions
2054 """""""""""""""""""""""""""""""""
2056 Including "`llvm/Transforms/Utils/BasicBlockUtils.h
2057 <http://llvm.org/doxygen/BasicBlockUtils_8h-source.html>`_" permits use of two
2058 very useful replace functions: ``ReplaceInstWithValue`` and
2059 ``ReplaceInstWithInst``.
2061 .. _schanges_deleting_sub:
2063 Deleting Instructions
2064 """""""""""""""""""""
2066 * ``ReplaceInstWithValue``
2068 This function replaces all uses of a given instruction with a value, and then
2069 removes the original instruction. The following example illustrates the
2070 replacement of the result of a particular ``AllocaInst`` that allocates memory
2071 for a single integer with a null pointer to an integer.
2075 AllocaInst* instToReplace = ...;
2076 BasicBlock::iterator ii(instToReplace);
2078 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
2079 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2081 * ``ReplaceInstWithInst``
2083 This function replaces a particular instruction with another instruction,
2084 inserting the new instruction into the basic block at the location where the
2085 old instruction was, and replacing any uses of the old instruction with the
2086 new instruction. The following example illustrates the replacement of one
2087 ``AllocaInst`` with another.
2091 AllocaInst* instToReplace = ...;
2092 BasicBlock::iterator ii(instToReplace);
2094 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
2095 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
2098 Replacing multiple uses of Users and Values
2099 """""""""""""""""""""""""""""""""""""""""""
2101 You can use ``Value::replaceAllUsesWith`` and ``User::replaceUsesOfWith`` to
2102 change more than one use at a time. See the doxygen documentation for the
2103 `Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_ and `User Class
2104 <http://llvm.org/doxygen/classllvm_1_1User.html>`_, respectively, for more
2107 .. _schanges_deletingGV:
2109 Deleting GlobalVariables
2110 ^^^^^^^^^^^^^^^^^^^^^^^^
2112 Deleting a global variable from a module is just as easy as deleting an
2113 Instruction. First, you must have a pointer to the global variable that you
2114 wish to delete. You use this pointer to erase it from its parent, the module.
2119 GlobalVariable *GV = .. ;
2121 GV->eraseFromParent();
2129 In generating IR, you may need some complex types. If you know these types
2130 statically, you can use ``TypeBuilder<...>::get()``, defined in
2131 ``llvm/Support/TypeBuilder.h``, to retrieve them. ``TypeBuilder`` has two forms
2132 depending on whether you're building types for cross-compilation or native
2133 library use. ``TypeBuilder<T, true>`` requires that ``T`` be independent of the
2134 host environment, meaning that it's built out of types from the ``llvm::types``
2135 (`doxygen <http://llvm.org/doxygen/namespacellvm_1_1types.html>`__) namespace
2136 and pointers, functions, arrays, etc. built of those. ``TypeBuilder<T, false>``
2137 additionally allows native C types whose size may depend on the host compiler.
2142 FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get();
2144 is easier to read and write than the equivalent
2148 std::vector<const Type*> params;
2149 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2150 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2152 See the `class comment
2153 <http://llvm.org/doxygen/TypeBuilder_8h-source.html#l00001>`_ for more details.
2160 This section describes the interaction of the LLVM APIs with multithreading,
2161 both on the part of client applications, and in the JIT, in the hosted
2164 Note that LLVM's support for multithreading is still relatively young. Up
2165 through version 2.5, the execution of threaded hosted applications was
2166 supported, but not threaded client access to the APIs. While this use case is
2167 now supported, clients *must* adhere to the guidelines specified below to ensure
2168 proper operation in multithreaded mode.
2170 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2171 intrinsics in order to support threaded operation. If you need a
2172 multhreading-capable LLVM on a platform without a suitably modern system
2173 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2174 using the resultant compiler to build a copy of LLVM with multithreading
2179 Ending Execution with ``llvm_shutdown()``
2180 -----------------------------------------
2182 When you are done using the LLVM APIs, you should call ``llvm_shutdown()`` to
2183 deallocate memory used for internal structures.
2187 Lazy Initialization with ``ManagedStatic``
2188 ------------------------------------------
2190 ``ManagedStatic`` is a utility class in LLVM used to implement static
2191 initialization of static resources, such as the global type tables. In a
2192 single-threaded environment, it implements a simple lazy initialization scheme.
2193 When LLVM is compiled with support for multi-threading, however, it uses
2194 double-checked locking to implement thread-safe lazy initialization.
2198 Achieving Isolation with ``LLVMContext``
2199 ----------------------------------------
2201 ``LLVMContext`` is an opaque class in the LLVM API which clients can use to
2202 operate multiple, isolated instances of LLVM concurrently within the same
2203 address space. For instance, in a hypothetical compile-server, the compilation
2204 of an individual translation unit is conceptually independent from all the
2205 others, and it would be desirable to be able to compile incoming translation
2206 units concurrently on independent server threads. Fortunately, ``LLVMContext``
2207 exists to enable just this kind of scenario!
2209 Conceptually, ``LLVMContext`` provides isolation. Every LLVM entity
2210 (``Module``\ s, ``Value``\ s, ``Type``\ s, ``Constant``\ s, etc.) in LLVM's
2211 in-memory IR belongs to an ``LLVMContext``. Entities in different contexts
2212 *cannot* interact with each other: ``Module``\ s in different contexts cannot be
2213 linked together, ``Function``\ s cannot be added to ``Module``\ s in different
2214 contexts, etc. What this means is that is is safe to compile on multiple
2215 threads simultaneously, as long as no two threads operate on entities within the
2218 In practice, very few places in the API require the explicit specification of a
2219 ``LLVMContext``, other than the ``Type`` creation/lookup APIs. Because every
2220 ``Type`` carries a reference to its owning context, most other entities can
2221 determine what context they belong to by looking at their own ``Type``. If you
2222 are adding new entities to LLVM IR, please try to maintain this interface
2225 For clients that do *not* require the benefits of isolation, LLVM provides a
2226 convenience API ``getGlobalContext()``. This returns a global, lazily
2227 initialized ``LLVMContext`` that may be used in situations where isolation is
2235 LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple
2236 threads can call ``ExecutionEngine::getPointerToFunction()`` or
2237 ``ExecutionEngine::runFunction()`` concurrently, and multiple threads can run
2238 code output by the JIT concurrently. The user must still ensure that only one
2239 thread accesses IR in a given ``LLVMContext`` while another thread might be
2240 modifying it. One way to do that is to always hold the JIT lock while accessing
2241 IR outside the JIT (the JIT *modifies* the IR by adding ``CallbackVH``\ s).
2242 Another way is to only call ``getPointerToFunction()`` from the
2243 ``LLVMContext``'s thread.
2245 When the JIT is configured to compile lazily (using
2246 ``ExecutionEngine::DisableLazyCompilation(false)``), there is currently a `race
2247 condition <http://llvm.org/bugs/show_bug.cgi?id=5184>`_ in updating call sites
2248 after a function is lazily-jitted. It's still possible to use the lazy JIT in a
2249 threaded program if you ensure that only one thread at a time can call any
2250 particular lazy stub and that the JIT lock guards any IR access, but we suggest
2251 using only the eager JIT in threaded programs.
2258 This section describes some of the advanced or obscure API's that most clients
2259 do not need to be aware of. These API's tend manage the inner workings of the
2260 LLVM system, and only need to be accessed in unusual circumstances.
2264 The ``ValueSymbolTable`` class
2265 ------------------------------
2267 The ``ValueSymbolTable`` (`doxygen
2268 <http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html>`__) class provides
2269 a symbol table that the :ref:`Function <c_Function>` and Module_ classes use for
2270 naming value definitions. The symbol table can provide a name for any Value_.
2272 Note that the ``SymbolTable`` class should not be directly accessed by most
2273 clients. It should only be used when iteration over the symbol table names
2274 themselves are required, which is very special purpose. Note that not all LLVM
2275 Value_\ s have names, and those without names (i.e. they have an empty name) do
2276 not exist in the symbol table.
2278 Symbol tables support iteration over the values in the symbol table with
2279 ``begin/end/iterator`` and supports querying to see if a specific name is in the
2280 symbol table (with ``lookup``). The ``ValueSymbolTable`` class exposes no
2281 public mutator methods, instead, simply call ``setName`` on a value, which will
2282 autoinsert it into the appropriate symbol table.
2286 The ``User`` and owned ``Use`` classes' memory layout
2287 -----------------------------------------------------
2289 The ``User`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1User.html>`__)
2290 class provides a basis for expressing the ownership of ``User`` towards other
2291 `Value instance <http://llvm.org/doxygen/classllvm_1_1Value.html>`_\ s. The
2292 ``Use`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1Use.html>`__) helper
2293 class is employed to do the bookkeeping and to facilitate *O(1)* addition and
2298 Interaction and relationship between ``User`` and ``Use`` objects
2299 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2301 A subclass of ``User`` can choose between incorporating its ``Use`` objects or
2302 refer to them out-of-line by means of a pointer. A mixed variant (some ``Use``
2303 s inline others hung off) is impractical and breaks the invariant that the
2304 ``Use`` objects belonging to the same ``User`` form a contiguous array.
2306 We have 2 different layouts in the ``User`` (sub)classes:
2310 The ``Use`` object(s) are inside (resp. at fixed offset) of the ``User``
2311 object and there are a fixed number of them.
2315 The ``Use`` object(s) are referenced by a pointer to an array from the
2316 ``User`` object and there may be a variable number of them.
2318 As of v2.4 each layout still possesses a direct pointer to the start of the
2319 array of ``Use``\ s. Though not mandatory for layout a), we stick to this
2320 redundancy for the sake of simplicity. The ``User`` object also stores the
2321 number of ``Use`` objects it has. (Theoretically this information can also be
2322 calculated given the scheme presented below.)
2324 Special forms of allocation operators (``operator new``) enforce the following
2327 * Layout a) is modelled by prepending the ``User`` object by the ``Use[]``
2330 .. code-block:: none
2332 ...---.---.---.---.-------...
2333 | P | P | P | P | User
2334 '''---'---'---'---'-------'''
2336 * Layout b) is modelled by pointing at the ``Use[]`` array.
2338 .. code-block:: none
2349 *(In the above figures* '``P``' *stands for the* ``Use**`` *that is stored in
2350 each* ``Use`` *object in the member* ``Use::Prev`` *)*
2354 The waymarking algorithm
2355 ^^^^^^^^^^^^^^^^^^^^^^^^
2357 Since the ``Use`` objects are deprived of the direct (back)pointer to their
2358 ``User`` objects, there must be a fast and exact method to recover it. This is
2359 accomplished by the following scheme:
2361 A bit-encoding in the 2 LSBits (least significant bits) of the ``Use::Prev``
2362 allows to find the start of the ``User`` object:
2364 * ``00`` --- binary digit 0
2366 * ``01`` --- binary digit 1
2368 * ``10`` --- stop and calculate (``s``)
2370 * ``11`` --- full stop (``S``)
2372 Given a ``Use*``, all we have to do is to walk till we get a stop and we either
2373 have a ``User`` immediately behind or we have to walk to the next stop picking
2374 up digits and calculating the offset:
2376 .. code-block:: none
2378 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2379 | 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
2380 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2381 |+15 |+10 |+6 |+3 |+1
2384 | | | ______________________>
2385 | | ______________________________________>
2386 | __________________________________________________________>
2388 Only the significant number of bits need to be stored between the stops, so that
2389 the *worst case is 20 memory accesses* when there are 1000 ``Use`` objects
2390 associated with a ``User``.
2394 Reference implementation
2395 ^^^^^^^^^^^^^^^^^^^^^^^^
2397 The following literate Haskell fragment demonstrates the concept:
2399 .. code-block:: haskell
2401 > import Test.QuickCheck
2403 > digits :: Int -> [Char] -> [Char]
2404 > digits 0 acc = '0' : acc
2405 > digits 1 acc = '1' : acc
2406 > digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
2408 > dist :: Int -> [Char] -> [Char]
2411 > dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
2412 > dist n acc = dist (n - 1) $ dist 1 acc
2414 > takeLast n ss = reverse $ take n $ reverse ss
2416 > test = takeLast 40 $ dist 20 []
2419 Printing <test> gives: ``"1s100000s11010s10100s1111s1010s110s11s1S"``
2421 The reverse algorithm computes the length of the string just by examining a
2424 .. code-block:: haskell
2426 > pref :: [Char] -> Int
2428 > pref ('s':'1':rest) = decode 2 1 rest
2429 > pref (_:rest) = 1 + pref rest
2431 > decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
2432 > decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
2433 > decode walk acc _ = walk + acc
2436 Now, as expected, printing <pref test> gives ``40``.
2438 We can *quickCheck* this with following property:
2440 .. code-block:: haskell
2442 > testcase = dist 2000 []
2443 > testcaseLength = length testcase
2445 > identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
2446 > where arr = takeLast n testcase
2449 As expected <quickCheck identityProp> gives:
2453 *Main> quickCheck identityProp
2454 OK, passed 100 tests.
2456 Let's be a bit more exhaustive:
2458 .. code-block:: haskell
2461 > deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
2464 And here is the result of <deepCheck identityProp>:
2468 *Main> deepCheck identityProp
2469 OK, passed 500 tests.
2473 Tagging considerations
2474 ^^^^^^^^^^^^^^^^^^^^^^
2476 To maintain the invariant that the 2 LSBits of each ``Use**`` in ``Use`` never
2477 change after being set up, setters of ``Use::Prev`` must re-tag the new
2478 ``Use**`` on every modification. Accordingly getters must strip the tag bits.
2480 For layout b) instead of the ``User`` we find a pointer (``User*`` with LSBit
2481 set). Following this pointer brings us to the ``User``. A portable trick
2482 ensures that the first bytes of ``User`` (if interpreted as a pointer) never has
2483 the LSBit set. (Portability is relying on the fact that all known compilers
2484 place the ``vptr`` in the first word of the instances.)
2488 Designing Type Hiercharies and Polymorphic Interfaces
2489 -----------------------------------------------------
2491 There are two different design patterns that tend to result in the use of
2492 virtual dispatch for methods in a type hierarchy in C++ programs. The first is
2493 a genuine type hierarchy where different types in the hierarchy model
2494 a specific subset of the functionality and semantics, and these types nest
2495 strictly within each other. Good examples of this can be seen in the ``Value``
2496 or ``Type`` type hierarchies.
2498 A second is the desire to dispatch dynamically across a collection of
2499 polymorphic interface implementations. This latter use case can be modeled with
2500 virtual dispatch and inheritance by defining an abstract interface base class
2501 which all implementations derive from and override. However, this
2502 implementation strategy forces an **"is-a"** relationship to exist that is not
2503 actually meaningful. There is often not some nested hierarchy of useful
2504 generalizations which code might interact with and move up and down. Instead,
2505 there is a singular interface which is dispatched across a range of
2508 The preferred implementation strategy for the second use case is that of
2509 generic programming (sometimes called "compile-time duck typing" or "static
2510 polymorphism"). For example, a template over some type parameter ``T`` can be
2511 instantiated across any particular implementation that conforms to the
2512 interface or *concept*. A good example here is the highly generic properties of
2513 any type which models a node in a directed graph. LLVM models these primarily
2514 through templates and generic programming. Such templates include the
2515 ``LoopInfoBase`` and ``DominatorTreeBase``. When this type of polymorphism
2516 truly needs **dynamic** dispatch you can generalize it using a technique
2517 called *concept-based polymorphism*. This pattern emulates the interfaces and
2518 behaviors of templates using a very limited form of virtual dispatch for type
2519 erasure inside its implementation. You can find examples of this technique in
2520 the ``PassManager.h`` system, and there is a more detailed introduction to it
2521 by Sean Parent in several of his talks and papers:
2523 #. `Inheritance Is The Base Class of Evil
2524 <http://channel9.msdn.com/Events/GoingNative/2013/Inheritance-Is-The-Base-Class-of-Evil>`_
2525 - The GoingNative 2013 talk describing this technique, and probably the best
2527 #. `Value Semantics and Concepts-based Polymorphism
2528 <http://www.youtube.com/watch?v=_BpMYeUFXv8>`_ - The C++Now! 2012 talk
2529 describing this technique in more detail.
2530 #. `Sean Parent's Papers and Presentations
2531 <http://github.com/sean-parent/sean-parent.github.com/wiki/Papers-and-Presentations>`_
2532 - A Github project full of links to slides, video, and sometimes code.
2534 When deciding between creating a type hierarchy (with either tagged or virtual
2535 dispatch) and using templates or concepts-based polymorphism, consider whether
2536 there is some refinement of an abstract base class which is a semantically
2537 meaningful type on an interface boundary. If anything more refined than the
2538 root abstract interface is meaningless to talk about as a partial extension of
2539 the semantic model, then your use case likely fits better with polymorphism and
2540 you should avoid using virtual dispatch. However, there may be some exigent
2541 circumstances that require one technique or the other to be used.
2543 If you do need to introduce a type hierarchy, we prefer to use explicitly
2544 closed type hierarchies with manual tagged dispatch and/or RTTI rather than the
2545 open inheritance model and virtual dispatch that is more common in C++ code.
2546 This is because LLVM rarely encourages library consumers to extend its core
2547 types, and leverages the closed and tag-dispatched nature of its hierarchies to
2548 generate significantly more efficient code. We have also found that a large
2549 amount of our usage of type hierarchies fits better with tag-based pattern
2550 matching rather than dynamic dispatch across a common interface. Within LLVM we
2551 have built custom helpers to facilitate this design. See this document's
2552 section on :ref:`isa and dyn_cast <isa>` and our :doc:`detailed document
2553 <HowToSetUpLLVMStyleRTTI>` which describes how you can implement this
2554 pattern for use with the LLVM helpers.
2556 .. _abi_breaking_checks:
2561 Checks and asserts that alter the LLVM C++ ABI are predicated on the
2562 preprocessor symbol `LLVM_ENABLE_ABI_BREAKING_CHECKS` -- LLVM
2563 libraries built with `LLVM_ENABLE_ABI_BREAKING_CHECKS` are not ABI
2564 compatible LLVM libraries built without it defined. By default,
2565 turning on assertions also turns on `LLVM_ENABLE_ABI_BREAKING_CHECKS`
2566 so a default +Asserts build is not ABI compatible with a
2567 default -Asserts build. Clients that want ABI compatibility
2568 between +Asserts and -Asserts builds should use the CMake or autoconf
2569 build systems to set `LLVM_ENABLE_ABI_BREAKING_CHECKS` independently
2570 of `LLVM_ENABLE_ASSERTIONS`.
2574 The Core LLVM Class Hierarchy Reference
2575 =======================================
2577 ``#include "llvm/IR/Type.h"``
2579 header source: `Type.h <http://llvm.org/doxygen/Type_8h-source.html>`_
2581 doxygen info: `Type Clases <http://llvm.org/doxygen/classllvm_1_1Type.html>`_
2583 The Core LLVM classes are the primary means of representing the program being
2584 inspected or transformed. The core LLVM classes are defined in header files in
2585 the ``include/llvm/`` directory, and implemented in the ``lib/VMCore``
2590 The Type class and Derived Types
2591 --------------------------------
2593 ``Type`` is a superclass of all type classes. Every ``Value`` has a ``Type``.
2594 ``Type`` cannot be instantiated directly but only through its subclasses.
2595 Certain primitive types (``VoidType``, ``LabelType``, ``FloatType`` and
2596 ``DoubleType``) have hidden subclasses. They are hidden because they offer no
2597 useful functionality beyond what the ``Type`` class offers except to distinguish
2598 themselves from other subclasses of ``Type``.
2600 All other types are subclasses of ``DerivedType``. Types can be named, but this
2601 is not a requirement. There exists exactly one instance of a given shape at any
2602 one time. This allows type equality to be performed with address equality of
2603 the Type Instance. That is, given two ``Type*`` values, the types are identical
2604 if the pointers are identical.
2608 Important Public Methods
2609 ^^^^^^^^^^^^^^^^^^^^^^^^
2611 * ``bool isIntegerTy() const``: Returns true for any integer type.
2613 * ``bool isFloatingPointTy()``: Return true if this is one of the five
2614 floating point types.
2616 * ``bool isSized()``: Return true if the type has known size. Things
2617 that don't have a size are abstract types, labels and void.
2621 Important Derived Types
2622 ^^^^^^^^^^^^^^^^^^^^^^^
2625 Subclass of DerivedType that represents integer types of any bit width. Any
2626 bit width between ``IntegerType::MIN_INT_BITS`` (1) and
2627 ``IntegerType::MAX_INT_BITS`` (~8 million) can be represented.
2629 * ``static const IntegerType* get(unsigned NumBits)``: get an integer
2630 type of a specific bit width.
2632 * ``unsigned getBitWidth() const``: Get the bit width of an integer type.
2635 This is subclassed by ArrayType, PointerType and VectorType.
2637 * ``const Type * getElementType() const``: Returns the type of each
2638 of the elements in the sequential type.
2641 This is a subclass of SequentialType and defines the interface for array
2644 * ``unsigned getNumElements() const``: Returns the number of elements
2648 Subclass of SequentialType for pointer types.
2651 Subclass of SequentialType for vector types. A vector type is similar to an
2652 ArrayType but is distinguished because it is a first class type whereas
2653 ArrayType is not. Vector types are used for vector operations and are usually
2654 small vectors of of an integer or floating point type.
2657 Subclass of DerivedTypes for struct types.
2662 Subclass of DerivedTypes for function types.
2664 * ``bool isVarArg() const``: Returns true if it's a vararg function.
2666 * ``const Type * getReturnType() const``: Returns the return type of the
2669 * ``const Type * getParamType (unsigned i)``: Returns the type of the ith
2672 * ``const unsigned getNumParams() const``: Returns the number of formal
2677 The ``Module`` class
2678 --------------------
2680 ``#include "llvm/IR/Module.h"``
2682 header source: `Module.h <http://llvm.org/doxygen/Module_8h-source.html>`_
2684 doxygen info: `Module Class <http://llvm.org/doxygen/classllvm_1_1Module.html>`_
2686 The ``Module`` class represents the top level structure present in LLVM
2687 programs. An LLVM module is effectively either a translation unit of the
2688 original program or a combination of several translation units merged by the
2689 linker. The ``Module`` class keeps track of a list of :ref:`Function
2690 <c_Function>`\ s, a list of GlobalVariable_\ s, and a SymbolTable_.
2691 Additionally, it contains a few helpful member functions that try to make common
2696 Important Public Members of the ``Module`` class
2697 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2699 * ``Module::Module(std::string name = "")``
2701 Constructing a Module_ is easy. You can optionally provide a name for it
2702 (probably based on the name of the translation unit).
2704 * | ``Module::iterator`` - Typedef for function list iterator
2705 | ``Module::const_iterator`` - Typedef for const_iterator.
2706 | ``begin()``, ``end()``, ``size()``, ``empty()``
2708 These are forwarding methods that make it easy to access the contents of a
2709 ``Module`` object's :ref:`Function <c_Function>` list.
2711 * ``Module::FunctionListType &getFunctionList()``
2713 Returns the list of :ref:`Function <c_Function>`\ s. This is necessary to use
2714 when you need to update the list or perform a complex action that doesn't have
2715 a forwarding method.
2719 * | ``Module::global_iterator`` - Typedef for global variable list iterator
2720 | ``Module::const_global_iterator`` - Typedef for const_iterator.
2721 | ``global_begin()``, ``global_end()``, ``global_size()``, ``global_empty()``
2723 These are forwarding methods that make it easy to access the contents of a
2724 ``Module`` object's GlobalVariable_ list.
2726 * ``Module::GlobalListType &getGlobalList()``
2728 Returns the list of GlobalVariable_\ s. This is necessary to use when you
2729 need to update the list or perform a complex action that doesn't have a
2734 * ``SymbolTable *getSymbolTable()``
2736 Return a reference to the SymbolTable_ for this ``Module``.
2740 * ``Function *getFunction(StringRef Name) const``
2742 Look up the specified function in the ``Module`` SymbolTable_. If it does not
2743 exist, return ``null``.
2745 * ``Function *getOrInsertFunction(const std::string &Name, const FunctionType
2748 Look up the specified function in the ``Module`` SymbolTable_. If it does not
2749 exist, add an external declaration for the function and return it.
2751 * ``std::string getTypeName(const Type *Ty)``
2753 If there is at least one entry in the SymbolTable_ for the specified Type_,
2754 return it. Otherwise return the empty string.
2756 * ``bool addTypeName(const std::string &Name, const Type *Ty)``
2758 Insert an entry in the SymbolTable_ mapping ``Name`` to ``Ty``. If there is
2759 already an entry for this name, true is returned and the SymbolTable_ is not
2767 ``#include "llvm/IR/Value.h"``
2769 header source: `Value.h <http://llvm.org/doxygen/Value_8h-source.html>`_
2771 doxygen info: `Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_
2773 The ``Value`` class is the most important class in the LLVM Source base. It
2774 represents a typed value that may be used (among other things) as an operand to
2775 an instruction. There are many different types of ``Value``\ s, such as
2776 Constant_\ s, Argument_\ s. Even Instruction_\ s and :ref:`Function
2777 <c_Function>`\ s are ``Value``\ s.
2779 A particular ``Value`` may be used many times in the LLVM representation for a
2780 program. For example, an incoming argument to a function (represented with an
2781 instance of the Argument_ class) is "used" by every instruction in the function
2782 that references the argument. To keep track of this relationship, the ``Value``
2783 class keeps a list of all of the ``User``\ s that is using it (the User_ class
2784 is a base class for all nodes in the LLVM graph that can refer to ``Value``\ s).
2785 This use list is how LLVM represents def-use information in the program, and is
2786 accessible through the ``use_*`` methods, shown below.
2788 Because LLVM is a typed representation, every LLVM ``Value`` is typed, and this
2789 Type_ is available through the ``getType()`` method. In addition, all LLVM
2790 values can be named. The "name" of the ``Value`` is a symbolic string printed
2793 .. code-block:: llvm
2799 The name of this instruction is "foo". **NOTE** that the name of any value may
2800 be missing (an empty string), so names should **ONLY** be used for debugging
2801 (making the source code easier to read, debugging printouts), they should not be
2802 used to keep track of values or map between them. For this purpose, use a
2803 ``std::map`` of pointers to the ``Value`` itself instead.
2805 One important aspect of LLVM is that there is no distinction between an SSA
2806 variable and the operation that produces it. Because of this, any reference to
2807 the value produced by an instruction (or the value available as an incoming
2808 argument, for example) is represented as a direct pointer to the instance of the
2809 class that represents this value. Although this may take some getting used to,
2810 it simplifies the representation and makes it easier to manipulate.
2814 Important Public Members of the ``Value`` class
2815 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2817 * | ``Value::use_iterator`` - Typedef for iterator over the use-list
2818 | ``Value::const_use_iterator`` - Typedef for const_iterator over the
2820 | ``unsigned use_size()`` - Returns the number of users of the value.
2821 | ``bool use_empty()`` - Returns true if there are no users.
2822 | ``use_iterator use_begin()`` - Get an iterator to the start of the
2824 | ``use_iterator use_end()`` - Get an iterator to the end of the use-list.
2825 | ``User *use_back()`` - Returns the last element in the list.
2827 These methods are the interface to access the def-use information in LLVM.
2828 As with all other iterators in LLVM, the naming conventions follow the
2829 conventions defined by the STL_.
2831 * ``Type *getType() const``
2832 This method returns the Type of the Value.
2834 * | ``bool hasName() const``
2835 | ``std::string getName() const``
2836 | ``void setName(const std::string &Name)``
2838 This family of methods is used to access and assign a name to a ``Value``, be
2839 aware of the :ref:`precaution above <nameWarning>`.
2841 * ``void replaceAllUsesWith(Value *V)``
2843 This method traverses the use list of a ``Value`` changing all User_\ s of the
2844 current value to refer to "``V``" instead. For example, if you detect that an
2845 instruction always produces a constant value (for example through constant
2846 folding), you can replace all uses of the instruction with the constant like
2851 Inst->replaceAllUsesWith(ConstVal);
2858 ``#include "llvm/IR/User.h"``
2860 header source: `User.h <http://llvm.org/doxygen/User_8h-source.html>`_
2862 doxygen info: `User Class <http://llvm.org/doxygen/classllvm_1_1User.html>`_
2866 The ``User`` class is the common base class of all LLVM nodes that may refer to
2867 ``Value``\ s. It exposes a list of "Operands" that are all of the ``Value``\ s
2868 that the User is referring to. The ``User`` class itself is a subclass of
2871 The operands of a ``User`` point directly to the LLVM ``Value`` that it refers
2872 to. Because LLVM uses Static Single Assignment (SSA) form, there can only be
2873 one definition referred to, allowing this direct connection. This connection
2874 provides the use-def information in LLVM.
2878 Important Public Members of the ``User`` class
2879 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2881 The ``User`` class exposes the operand list in two ways: through an index access
2882 interface and through an iterator based interface.
2884 * | ``Value *getOperand(unsigned i)``
2885 | ``unsigned getNumOperands()``
2887 These two methods expose the operands of the ``User`` in a convenient form for
2890 * | ``User::op_iterator`` - Typedef for iterator over the operand list
2891 | ``op_iterator op_begin()`` - Get an iterator to the start of the operand
2893 | ``op_iterator op_end()`` - Get an iterator to the end of the operand list.
2895 Together, these methods make up the iterator based interface to the operands
2901 The ``Instruction`` class
2902 -------------------------
2904 ``#include "llvm/IR/Instruction.h"``
2906 header source: `Instruction.h
2907 <http://llvm.org/doxygen/Instruction_8h-source.html>`_
2909 doxygen info: `Instruction Class
2910 <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_
2912 Superclasses: User_, Value_
2914 The ``Instruction`` class is the common base class for all LLVM instructions.
2915 It provides only a few methods, but is a very commonly used class. The primary
2916 data tracked by the ``Instruction`` class itself is the opcode (instruction
2917 type) and the parent BasicBlock_ the ``Instruction`` is embedded into. To
2918 represent a specific type of instruction, one of many subclasses of
2919 ``Instruction`` are used.
2921 Because the ``Instruction`` class subclasses the User_ class, its operands can
2922 be accessed in the same way as for other ``User``\ s (with the
2923 ``getOperand()``/``getNumOperands()`` and ``op_begin()``/``op_end()`` methods).
2924 An important file for the ``Instruction`` class is the ``llvm/Instruction.def``
2925 file. This file contains some meta-data about the various different types of
2926 instructions in LLVM. It describes the enum values that are used as opcodes
2927 (for example ``Instruction::Add`` and ``Instruction::ICmp``), as well as the
2928 concrete sub-classes of ``Instruction`` that implement the instruction (for
2929 example BinaryOperator_ and CmpInst_). Unfortunately, the use of macros in this
2930 file confuses doxygen, so these enum values don't show up correctly in the
2931 `doxygen output <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_.
2935 Important Subclasses of the ``Instruction`` class
2936 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2940 * ``BinaryOperator``
2942 This subclasses represents all two operand instructions whose operands must be
2943 the same type, except for the comparison instructions.
2948 This subclass is the parent of the 12 casting instructions. It provides
2949 common operations on cast instructions.
2955 This subclass respresents the two comparison instructions,
2956 `ICmpInst <LangRef.html#i_icmp>`_ (integer opreands), and
2957 `FCmpInst <LangRef.html#i_fcmp>`_ (floating point operands).
2961 * ``TerminatorInst``
2963 This subclass is the parent of all terminator instructions (those which can
2968 Important Public Members of the ``Instruction`` class
2969 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2971 * ``BasicBlock *getParent()``
2973 Returns the BasicBlock_ that this
2974 ``Instruction`` is embedded into.
2976 * ``bool mayWriteToMemory()``
2978 Returns true if the instruction writes to memory, i.e. it is a ``call``,
2979 ``free``, ``invoke``, or ``store``.
2981 * ``unsigned getOpcode()``
2983 Returns the opcode for the ``Instruction``.
2985 * ``Instruction *clone() const``
2987 Returns another instance of the specified instruction, identical in all ways
2988 to the original except that the instruction has no parent (i.e. it's not
2989 embedded into a BasicBlock_), and it has no name.
2993 The ``Constant`` class and subclasses
2994 -------------------------------------
2996 Constant represents a base class for different types of constants. It is
2997 subclassed by ConstantInt, ConstantArray, etc. for representing the various
2998 types of Constants. GlobalValue_ is also a subclass, which represents the
2999 address of a global variable or function.
3003 Important Subclasses of Constant
3004 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3006 * ConstantInt : This subclass of Constant represents an integer constant of
3009 * ``const APInt& getValue() const``: Returns the underlying
3010 value of this constant, an APInt value.
3012 * ``int64_t getSExtValue() const``: Converts the underlying APInt value to an
3013 int64_t via sign extension. If the value (not the bit width) of the APInt
3014 is too large to fit in an int64_t, an assertion will result. For this
3015 reason, use of this method is discouraged.
3017 * ``uint64_t getZExtValue() const``: Converts the underlying APInt value
3018 to a uint64_t via zero extension. IF the value (not the bit width) of the
3019 APInt is too large to fit in a uint64_t, an assertion will result. For this
3020 reason, use of this method is discouraged.
3022 * ``static ConstantInt* get(const APInt& Val)``: Returns the ConstantInt
3023 object that represents the value provided by ``Val``. The type is implied
3024 as the IntegerType that corresponds to the bit width of ``Val``.
3026 * ``static ConstantInt* get(const Type *Ty, uint64_t Val)``: Returns the
3027 ConstantInt object that represents the value provided by ``Val`` for integer
3030 * ConstantFP : This class represents a floating point constant.
3032 * ``double getValue() const``: Returns the underlying value of this constant.
3034 * ConstantArray : This represents a constant array.
3036 * ``const std::vector<Use> &getValues() const``: Returns a vector of
3037 component constants that makeup this array.
3039 * ConstantStruct : This represents a constant struct.
3041 * ``const std::vector<Use> &getValues() const``: Returns a vector of
3042 component constants that makeup this array.
3044 * GlobalValue : This represents either a global variable or a function. In
3045 either case, the value is a constant fixed address (after linking).
3049 The ``GlobalValue`` class
3050 -------------------------
3052 ``#include "llvm/IR/GlobalValue.h"``
3054 header source: `GlobalValue.h
3055 <http://llvm.org/doxygen/GlobalValue_8h-source.html>`_
3057 doxygen info: `GlobalValue Class
3058 <http://llvm.org/doxygen/classllvm_1_1GlobalValue.html>`_
3060 Superclasses: Constant_, User_, Value_
3062 Global values ( GlobalVariable_\ s or :ref:`Function <c_Function>`\ s) are the
3063 only LLVM values that are visible in the bodies of all :ref:`Function
3064 <c_Function>`\ s. Because they are visible at global scope, they are also
3065 subject to linking with other globals defined in different translation units.
3066 To control the linking process, ``GlobalValue``\ s know their linkage rules.
3067 Specifically, ``GlobalValue``\ s know whether they have internal or external
3068 linkage, as defined by the ``LinkageTypes`` enumeration.
3070 If a ``GlobalValue`` has internal linkage (equivalent to being ``static`` in C),
3071 it is not visible to code outside the current translation unit, and does not
3072 participate in linking. If it has external linkage, it is visible to external
3073 code, and does participate in linking. In addition to linkage information,
3074 ``GlobalValue``\ s keep track of which Module_ they are currently part of.
3076 Because ``GlobalValue``\ s are memory objects, they are always referred to by
3077 their **address**. As such, the Type_ of a global is always a pointer to its
3078 contents. It is important to remember this when using the ``GetElementPtrInst``
3079 instruction because this pointer must be dereferenced first. For example, if
3080 you have a ``GlobalVariable`` (a subclass of ``GlobalValue)`` that is an array
3081 of 24 ints, type ``[24 x i32]``, then the ``GlobalVariable`` is a pointer to
3082 that array. Although the address of the first element of this array and the
3083 value of the ``GlobalVariable`` are the same, they have different types. The
3084 ``GlobalVariable``'s type is ``[24 x i32]``. The first element's type is
3085 ``i32.`` Because of this, accessing a global value requires you to dereference
3086 the pointer with ``GetElementPtrInst`` first, then its elements can be accessed.
3087 This is explained in the `LLVM Language Reference Manual
3088 <LangRef.html#globalvars>`_.
3092 Important Public Members of the ``GlobalValue`` class
3093 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3095 * | ``bool hasInternalLinkage() const``
3096 | ``bool hasExternalLinkage() const``
3097 | ``void setInternalLinkage(bool HasInternalLinkage)``
3099 These methods manipulate the linkage characteristics of the ``GlobalValue``.
3101 * ``Module *getParent()``
3103 This returns the Module_ that the
3104 GlobalValue is currently embedded into.
3108 The ``Function`` class
3109 ----------------------
3111 ``#include "llvm/IR/Function.h"``
3113 header source: `Function.h <http://llvm.org/doxygen/Function_8h-source.html>`_
3115 doxygen info: `Function Class
3116 <http://llvm.org/doxygen/classllvm_1_1Function.html>`_
3118 Superclasses: GlobalValue_, Constant_, User_, Value_
3120 The ``Function`` class represents a single procedure in LLVM. It is actually
3121 one of the more complex classes in the LLVM hierarchy because it must keep track
3122 of a large amount of data. The ``Function`` class keeps track of a list of
3123 BasicBlock_\ s, a list of formal Argument_\ s, and a SymbolTable_.
3125 The list of BasicBlock_\ s is the most commonly used part of ``Function``
3126 objects. The list imposes an implicit ordering of the blocks in the function,
3127 which indicate how the code will be laid out by the backend. Additionally, the
3128 first BasicBlock_ is the implicit entry node for the ``Function``. It is not
3129 legal in LLVM to explicitly branch to this initial block. There are no implicit
3130 exit nodes, and in fact there may be multiple exit nodes from a single
3131 ``Function``. If the BasicBlock_ list is empty, this indicates that the
3132 ``Function`` is actually a function declaration: the actual body of the function
3133 hasn't been linked in yet.
3135 In addition to a list of BasicBlock_\ s, the ``Function`` class also keeps track
3136 of the list of formal Argument_\ s that the function receives. This container
3137 manages the lifetime of the Argument_ nodes, just like the BasicBlock_ list does
3138 for the BasicBlock_\ s.
3140 The SymbolTable_ is a very rarely used LLVM feature that is only used when you
3141 have to look up a value by name. Aside from that, the SymbolTable_ is used
3142 internally to make sure that there are not conflicts between the names of
3143 Instruction_\ s, BasicBlock_\ s, or Argument_\ s in the function body.
3145 Note that ``Function`` is a GlobalValue_ and therefore also a Constant_. The
3146 value of the function is its address (after linking) which is guaranteed to be
3151 Important Public Members of the ``Function``
3152 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3154 * ``Function(const FunctionType *Ty, LinkageTypes Linkage,
3155 const std::string &N = "", Module* Parent = 0)``
3157 Constructor used when you need to create new ``Function``\ s to add the
3158 program. The constructor must specify the type of the function to create and
3159 what type of linkage the function should have. The FunctionType_ argument
3160 specifies the formal arguments and return value for the function. The same
3161 FunctionType_ value can be used to create multiple functions. The ``Parent``
3162 argument specifies the Module in which the function is defined. If this
3163 argument is provided, the function will automatically be inserted into that
3164 module's list of functions.
3166 * ``bool isDeclaration()``
3168 Return whether or not the ``Function`` has a body defined. If the function is
3169 "external", it does not have a body, and thus must be resolved by linking with
3170 a function defined in a different translation unit.
3172 * | ``Function::iterator`` - Typedef for basic block list iterator
3173 | ``Function::const_iterator`` - Typedef for const_iterator.
3174 | ``begin()``, ``end()``, ``size()``, ``empty()``
3176 These are forwarding methods that make it easy to access the contents of a
3177 ``Function`` object's BasicBlock_ list.
3179 * ``Function::BasicBlockListType &getBasicBlockList()``
3181 Returns the list of BasicBlock_\ s. This is necessary to use when you need to
3182 update the list or perform a complex action that doesn't have a forwarding
3185 * | ``Function::arg_iterator`` - Typedef for the argument list iterator
3186 | ``Function::const_arg_iterator`` - Typedef for const_iterator.
3187 | ``arg_begin()``, ``arg_end()``, ``arg_size()``, ``arg_empty()``
3189 These are forwarding methods that make it easy to access the contents of a
3190 ``Function`` object's Argument_ list.
3192 * ``Function::ArgumentListType &getArgumentList()``
3194 Returns the list of Argument_. 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 * ``BasicBlock &getEntryBlock()``
3200 Returns the entry ``BasicBlock`` for the function. Because the entry block
3201 for the function is always the first block, this returns the first block of
3204 * | ``Type *getReturnType()``
3205 | ``FunctionType *getFunctionType()``
3207 This traverses the Type_ of the ``Function`` and returns the return type of
3208 the function, or the FunctionType_ of the actual function.
3210 * ``SymbolTable *getSymbolTable()``
3212 Return a pointer to the SymbolTable_ for this ``Function``.
3216 The ``GlobalVariable`` class
3217 ----------------------------
3219 ``#include "llvm/IR/GlobalVariable.h"``
3221 header source: `GlobalVariable.h
3222 <http://llvm.org/doxygen/GlobalVariable_8h-source.html>`_
3224 doxygen info: `GlobalVariable Class
3225 <http://llvm.org/doxygen/classllvm_1_1GlobalVariable.html>`_
3227 Superclasses: GlobalValue_, Constant_, User_, Value_
3229 Global variables are represented with the (surprise surprise) ``GlobalVariable``
3230 class. Like functions, ``GlobalVariable``\ s are also subclasses of
3231 GlobalValue_, and as such are always referenced by their address (global values
3232 must live in memory, so their "name" refers to their constant address). See
3233 GlobalValue_ for more on this. Global variables may have an initial value
3234 (which must be a Constant_), and if they have an initializer, they may be marked
3235 as "constant" themselves (indicating that their contents never change at
3238 .. _m_GlobalVariable:
3240 Important Public Members of the ``GlobalVariable`` class
3241 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3243 * ``GlobalVariable(const Type *Ty, bool isConstant, LinkageTypes &Linkage,
3244 Constant *Initializer = 0, const std::string &Name = "", Module* Parent = 0)``
3246 Create a new global variable of the specified type. If ``isConstant`` is true
3247 then the global variable will be marked as unchanging for the program. The
3248 Linkage parameter specifies the type of linkage (internal, external, weak,
3249 linkonce, appending) for the variable. If the linkage is InternalLinkage,
3250 WeakAnyLinkage, WeakODRLinkage, LinkOnceAnyLinkage or LinkOnceODRLinkage, then
3251 the resultant global variable will have internal linkage. AppendingLinkage
3252 concatenates together all instances (in different translation units) of the
3253 variable into a single variable but is only applicable to arrays. See the
3254 `LLVM Language Reference <LangRef.html#modulestructure>`_ for further details
3255 on linkage types. Optionally an initializer, a name, and the module to put
3256 the variable into may be specified for the global variable as well.
3258 * ``bool isConstant() const``
3260 Returns true if this is a global variable that is known not to be modified at
3263 * ``bool hasInitializer()``
3265 Returns true if this ``GlobalVariable`` has an intializer.
3267 * ``Constant *getInitializer()``
3269 Returns the initial value for a ``GlobalVariable``. It is not legal to call
3270 this method if there is no initializer.
3274 The ``BasicBlock`` class
3275 ------------------------
3277 ``#include "llvm/IR/BasicBlock.h"``
3279 header source: `BasicBlock.h
3280 <http://llvm.org/doxygen/BasicBlock_8h-source.html>`_
3282 doxygen info: `BasicBlock Class
3283 <http://llvm.org/doxygen/classllvm_1_1BasicBlock.html>`_
3287 This class represents a single entry single exit section of the code, commonly
3288 known as a basic block by the compiler community. The ``BasicBlock`` class
3289 maintains a list of Instruction_\ s, which form the body of the block. Matching
3290 the language definition, the last element of this list of instructions is always
3291 a terminator instruction (a subclass of the TerminatorInst_ class).
3293 In addition to tracking the list of instructions that make up the block, the
3294 ``BasicBlock`` class also keeps track of the :ref:`Function <c_Function>` that
3295 it is embedded into.
3297 Note that ``BasicBlock``\ s themselves are Value_\ s, because they are
3298 referenced by instructions like branches and can go in the switch tables.
3299 ``BasicBlock``\ s have type ``label``.
3303 Important Public Members of the ``BasicBlock`` class
3304 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3306 * ``BasicBlock(const std::string &Name = "", Function *Parent = 0)``
3308 The ``BasicBlock`` constructor is used to create new basic blocks for
3309 insertion into a function. The constructor optionally takes a name for the
3310 new block, and a :ref:`Function <c_Function>` to insert it into. If the
3311 ``Parent`` parameter is specified, the new ``BasicBlock`` is automatically
3312 inserted at the end of the specified :ref:`Function <c_Function>`, if not
3313 specified, the BasicBlock must be manually inserted into the :ref:`Function
3316 * | ``BasicBlock::iterator`` - Typedef for instruction list iterator
3317 | ``BasicBlock::const_iterator`` - Typedef for const_iterator.
3318 | ``begin()``, ``end()``, ``front()``, ``back()``,
3319 ``size()``, ``empty()``
3320 STL-style functions for accessing the instruction list.
3322 These methods and typedefs are forwarding functions that have the same
3323 semantics as the standard library methods of the same names. These methods
3324 expose the underlying instruction list of a basic block in a way that is easy
3325 to manipulate. To get the full complement of container operations (including
3326 operations to update the list), you must use the ``getInstList()`` method.
3328 * ``BasicBlock::InstListType &getInstList()``
3330 This method is used to get access to the underlying container that actually
3331 holds the Instructions. This method must be used when there isn't a
3332 forwarding function in the ``BasicBlock`` class for the operation that you
3333 would like to perform. Because there are no forwarding functions for
3334 "updating" operations, you need to use this if you want to update the contents
3335 of a ``BasicBlock``.
3337 * ``Function *getParent()``
3339 Returns a pointer to :ref:`Function <c_Function>` the block is embedded into,
3340 or a null pointer if it is homeless.
3342 * ``TerminatorInst *getTerminator()``
3344 Returns a pointer to the terminator instruction that appears at the end of the
3345 ``BasicBlock``. If there is no terminator instruction, or if the last
3346 instruction in the block is not a terminator, then a null pointer is returned.
3350 The ``Argument`` class
3351 ----------------------
3353 This subclass of Value defines the interface for incoming formal arguments to a
3354 function. A Function maintains a list of its formal arguments. An argument has
3355 a pointer to the parent Function.