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.
303 ``function_ref<Ret(Param1, Param2, ...)>`` can be implicitly constructed from
304 any callable object that can be called with arguments of type ``Param1``,
305 ``Param2``, ..., and returns a value that can be converted to type ``Ret``.
310 void visitBasicBlocks(Function *F, function_ref<bool (BasicBlock*)> Callback) {
311 for (BasicBlock &BB : *F)
320 visitBasicBlocks(F, [&](BasicBlock *BB) {
326 Note that a ``function_ref`` object contains pointers to external memory, so
327 it is not generally safe to store an instance of the class (unless you know
328 that the external storage will not be freed).
329 ``function_ref`` is small enough that it should always be passed by value.
334 You cannot use ``std::function`` within LLVM code, because it is not supported
335 by all our target toolchains.
340 The ``DEBUG()`` macro and ``-debug`` option
341 -------------------------------------------
343 Often when working on your pass you will put a bunch of debugging printouts and
344 other code into your pass. After you get it working, you want to remove it, but
345 you may need it again in the future (to work out new bugs that you run across).
347 Naturally, because of this, you don't want to delete the debug printouts, but
348 you don't want them to always be noisy. A standard compromise is to comment
349 them out, allowing you to enable them if you need them in the future.
351 The ``llvm/Support/Debug.h`` (`doxygen
352 <http://llvm.org/doxygen/Debug_8h-source.html>`__) file provides a macro named
353 ``DEBUG()`` that is a much nicer solution to this problem. Basically, you can
354 put arbitrary code into the argument of the ``DEBUG`` macro, and it is only
355 executed if '``opt``' (or any other tool) is run with the '``-debug``' command
360 DEBUG(errs() << "I am here!\n");
362 Then you can run your pass like this:
366 $ opt < a.bc > /dev/null -mypass
368 $ opt < a.bc > /dev/null -mypass -debug
371 Using the ``DEBUG()`` macro instead of a home-brewed solution allows you to not
372 have to create "yet another" command line option for the debug output for your
373 pass. Note that ``DEBUG()`` macros are disabled for optimized builds, so they
374 do not cause a performance impact at all (for the same reason, they should also
375 not contain side-effects!).
377 One additional nice thing about the ``DEBUG()`` macro is that you can enable or
378 disable it directly in gdb. Just use "``set DebugFlag=0``" or "``set
379 DebugFlag=1``" from the gdb if the program is running. If the program hasn't
380 been started yet, you can always just run it with ``-debug``.
384 Fine grained debug info with ``DEBUG_TYPE`` and the ``-debug-only`` option
385 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
387 Sometimes you may find yourself in a situation where enabling ``-debug`` just
388 turns on **too much** information (such as when working on the code generator).
389 If you want to enable debug information with more fine-grained control, you
390 can define the ``DEBUG_TYPE`` macro and use the ``-debug-only`` option as
396 DEBUG(errs() << "No debug type\n");
397 #define DEBUG_TYPE "foo"
398 DEBUG(errs() << "'foo' debug type\n");
400 #define DEBUG_TYPE "bar"
401 DEBUG(errs() << "'bar' debug type\n"));
403 #define DEBUG_TYPE ""
404 DEBUG(errs() << "No debug type (2)\n");
406 Then you can run your pass like this:
410 $ opt < a.bc > /dev/null -mypass
412 $ opt < a.bc > /dev/null -mypass -debug
417 $ opt < a.bc > /dev/null -mypass -debug-only=foo
419 $ opt < a.bc > /dev/null -mypass -debug-only=bar
422 Of course, in practice, you should only set ``DEBUG_TYPE`` at the top of a file,
423 to specify the debug type for the entire module (if you do this before you
424 ``#include "llvm/Support/Debug.h"``, you don't have to insert the ugly
425 ``#undef``'s). Also, you should use names more meaningful than "foo" and "bar",
426 because there is no system in place to ensure that names do not conflict. If
427 two different modules use the same string, they will all be turned on when the
428 name is specified. This allows, for example, all debug information for
429 instruction scheduling to be enabled with ``-debug-type=InstrSched``, even if
430 the source lives in multiple files.
432 The ``DEBUG_WITH_TYPE`` macro is also available for situations where you would
433 like to set ``DEBUG_TYPE``, but only for one specific ``DEBUG`` statement. It
434 takes an additional first parameter, which is the type to use. For example, the
435 preceding example could be written as:
439 DEBUG_WITH_TYPE("", errs() << "No debug type\n");
440 DEBUG_WITH_TYPE("foo", errs() << "'foo' debug type\n");
441 DEBUG_WITH_TYPE("bar", errs() << "'bar' debug type\n"));
442 DEBUG_WITH_TYPE("", errs() << "No debug type (2)\n");
446 The ``Statistic`` class & ``-stats`` option
447 -------------------------------------------
449 The ``llvm/ADT/Statistic.h`` (`doxygen
450 <http://llvm.org/doxygen/Statistic_8h-source.html>`__) file provides a class
451 named ``Statistic`` that is used as a unified way to keep track of what the LLVM
452 compiler is doing and how effective various optimizations are. It is useful to
453 see what optimizations are contributing to making a particular program run
456 Often you may run your pass on some big program, and you're interested to see
457 how many times it makes a certain transformation. Although you can do this with
458 hand inspection, or some ad-hoc method, this is a real pain and not very useful
459 for big programs. Using the ``Statistic`` class makes it very easy to keep
460 track of this information, and the calculated information is presented in a
461 uniform manner with the rest of the passes being executed.
463 There are many examples of ``Statistic`` uses, but the basics of using it are as
466 #. Define your statistic like this:
470 #define DEBUG_TYPE "mypassname" // This goes before any #includes.
471 STATISTIC(NumXForms, "The # of times I did stuff");
473 The ``STATISTIC`` macro defines a static variable, whose name is specified by
474 the first argument. The pass name is taken from the ``DEBUG_TYPE`` macro, and
475 the description is taken from the second argument. The variable defined
476 ("NumXForms" in this case) acts like an unsigned integer.
478 #. Whenever you make a transformation, bump the counter:
482 ++NumXForms; // I did stuff!
484 That's all you have to do. To get '``opt``' to print out the statistics
485 gathered, use the '``-stats``' option:
489 $ opt -stats -mypassname < program.bc > /dev/null
490 ... statistics output ...
492 When running ``opt`` on a C file from the SPEC benchmark suite, it gives a
493 report that looks like this:
497 7646 bitcodewriter - Number of normal instructions
498 725 bitcodewriter - Number of oversized instructions
499 129996 bitcodewriter - Number of bitcode bytes written
500 2817 raise - Number of insts DCEd or constprop'd
501 3213 raise - Number of cast-of-self removed
502 5046 raise - Number of expression trees converted
503 75 raise - Number of other getelementptr's formed
504 138 raise - Number of load/store peepholes
505 42 deadtypeelim - Number of unused typenames removed from symtab
506 392 funcresolve - Number of varargs functions resolved
507 27 globaldce - Number of global variables removed
508 2 adce - Number of basic blocks removed
509 134 cee - Number of branches revectored
510 49 cee - Number of setcc instruction eliminated
511 532 gcse - Number of loads removed
512 2919 gcse - Number of instructions removed
513 86 indvars - Number of canonical indvars added
514 87 indvars - Number of aux indvars removed
515 25 instcombine - Number of dead inst eliminate
516 434 instcombine - Number of insts combined
517 248 licm - Number of load insts hoisted
518 1298 licm - Number of insts hoisted to a loop pre-header
519 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
520 75 mem2reg - Number of alloca's promoted
521 1444 cfgsimplify - Number of blocks simplified
523 Obviously, with so many optimizations, having a unified framework for this stuff
524 is very nice. Making your pass fit well into the framework makes it more
525 maintainable and useful.
529 Viewing graphs while debugging code
530 -----------------------------------
532 Several of the important data structures in LLVM are graphs: for example CFGs
533 made out of LLVM :ref:`BasicBlocks <BasicBlock>`, CFGs made out of LLVM
534 :ref:`MachineBasicBlocks <MachineBasicBlock>`, and :ref:`Instruction Selection
535 DAGs <SelectionDAG>`. In many cases, while debugging various parts of the
536 compiler, it is nice to instantly visualize these graphs.
538 LLVM provides several callbacks that are available in a debug build to do
539 exactly that. If you call the ``Function::viewCFG()`` method, for example, the
540 current LLVM tool will pop up a window containing the CFG for the function where
541 each basic block is a node in the graph, and each node contains the instructions
542 in the block. Similarly, there also exists ``Function::viewCFGOnly()`` (does
543 not include the instructions), the ``MachineFunction::viewCFG()`` and
544 ``MachineFunction::viewCFGOnly()``, and the ``SelectionDAG::viewGraph()``
545 methods. Within GDB, for example, you can usually use something like ``call
546 DAG.viewGraph()`` to pop up a window. Alternatively, you can sprinkle calls to
547 these functions in your code in places you want to debug.
549 Getting this to work requires a small amount of setup. On Unix systems
550 with X11, install the `graphviz <http://www.graphviz.org>`_ toolkit, and make
551 sure 'dot' and 'gv' are in your path. If you are running on Mac OS X, download
552 and install the Mac OS X `Graphviz program
553 <http://www.pixelglow.com/graphviz/>`_ and add
554 ``/Applications/Graphviz.app/Contents/MacOS/`` (or wherever you install it) to
555 your path. The programs need not be present when configuring, building or
556 running LLVM and can simply be installed when needed during an active debug
559 ``SelectionDAG`` has been extended to make it easier to locate *interesting*
560 nodes in large complex graphs. From gdb, if you ``call DAG.setGraphColor(node,
561 "color")``, then the next ``call DAG.viewGraph()`` would highlight the node in
562 the specified color (choices of colors can be found at `colors
563 <http://www.graphviz.org/doc/info/colors.html>`_.) More complex node attributes
564 can be provided with ``call DAG.setGraphAttrs(node, "attributes")`` (choices can
565 be found at `Graph attributes <http://www.graphviz.org/doc/info/attrs.html>`_.)
566 If you want to restart and clear all the current graph attributes, then you can
567 ``call DAG.clearGraphAttrs()``.
569 Note that graph visualization features are compiled out of Release builds to
570 reduce file size. This means that you need a Debug+Asserts or Release+Asserts
571 build to use these features.
575 Picking the Right Data Structure for a Task
576 ===========================================
578 LLVM has a plethora of data structures in the ``llvm/ADT/`` directory, and we
579 commonly use STL data structures. This section describes the trade-offs you
580 should consider when you pick one.
582 The first step is a choose your own adventure: do you want a sequential
583 container, a set-like container, or a map-like container? The most important
584 thing when choosing a container is the algorithmic properties of how you plan to
585 access the container. Based on that, you should use:
588 * a :ref:`map-like <ds_map>` container if you need efficient look-up of a
589 value based on another value. Map-like containers also support efficient
590 queries for containment (whether a key is in the map). Map-like containers
591 generally do not support efficient reverse mapping (values to keys). If you
592 need that, use two maps. Some map-like containers also support efficient
593 iteration through the keys in sorted order. Map-like containers are the most
594 expensive sort, only use them if you need one of these capabilities.
596 * a :ref:`set-like <ds_set>` container if you need to put a bunch of stuff into
597 a container that automatically eliminates duplicates. Some set-like
598 containers support efficient iteration through the elements in sorted order.
599 Set-like containers are more expensive than sequential containers.
601 * a :ref:`sequential <ds_sequential>` container provides the most efficient way
602 to add elements and keeps track of the order they are added to the collection.
603 They permit duplicates and support efficient iteration, but do not support
604 efficient look-up based on a key.
606 * a :ref:`string <ds_string>` container is a specialized sequential container or
607 reference structure that is used for character or byte arrays.
609 * a :ref:`bit <ds_bit>` container provides an efficient way to store and
610 perform set operations on sets of numeric id's, while automatically
611 eliminating duplicates. Bit containers require a maximum of 1 bit for each
612 identifier you want to store.
614 Once the proper category of container is determined, you can fine tune the
615 memory use, constant factors, and cache behaviors of access by intelligently
616 picking a member of the category. Note that constant factors and cache behavior
617 can be a big deal. If you have a vector that usually only contains a few
618 elements (but could contain many), for example, it's much better to use
619 :ref:`SmallVector <dss_smallvector>` than :ref:`vector <dss_vector>`. Doing so
620 avoids (relatively) expensive malloc/free calls, which dwarf the cost of adding
621 the elements to the container.
625 Sequential Containers (std::vector, std::list, etc)
626 ---------------------------------------------------
628 There are a variety of sequential containers available for you, based on your
629 needs. Pick the first in this section that will do what you want.
636 The ``llvm::ArrayRef`` class is the preferred class to use in an interface that
637 accepts a sequential list of elements in memory and just reads from them. By
638 taking an ``ArrayRef``, the API can be passed a fixed size array, an
639 ``std::vector``, an ``llvm::SmallVector`` and anything else that is contiguous
647 Fixed size arrays are very simple and very fast. They are good if you know
648 exactly how many elements you have, or you have a (low) upper bound on how many
653 Heap Allocated Arrays
654 ^^^^^^^^^^^^^^^^^^^^^
656 Heap allocated arrays (``new[]`` + ``delete[]``) are also simple. They are good
657 if the number of elements is variable, if you know how many elements you will
658 need before the array is allocated, and if the array is usually large (if not,
659 consider a :ref:`SmallVector <dss_smallvector>`). The cost of a heap allocated
660 array is the cost of the new/delete (aka malloc/free). Also note that if you
661 are allocating an array of a type with a constructor, the constructor and
662 destructors will be run for every element in the array (re-sizable vectors only
663 construct those elements actually used).
665 .. _dss_tinyptrvector:
667 llvm/ADT/TinyPtrVector.h
668 ^^^^^^^^^^^^^^^^^^^^^^^^
670 ``TinyPtrVector<Type>`` is a highly specialized collection class that is
671 optimized to avoid allocation in the case when a vector has zero or one
672 elements. It has two major restrictions: 1) it can only hold values of pointer
673 type, and 2) it cannot hold a null pointer.
675 Since this container is highly specialized, it is rarely used.
679 llvm/ADT/SmallVector.h
680 ^^^^^^^^^^^^^^^^^^^^^^
682 ``SmallVector<Type, N>`` is a simple class that looks and smells just like
683 ``vector<Type>``: it supports efficient iteration, lays out elements in memory
684 order (so you can do pointer arithmetic between elements), supports efficient
685 push_back/pop_back operations, supports efficient random access to its elements,
688 The advantage of SmallVector is that it allocates space for some number of
689 elements (N) **in the object itself**. Because of this, if the SmallVector is
690 dynamically smaller than N, no malloc is performed. This can be a big win in
691 cases where the malloc/free call is far more expensive than the code that
692 fiddles around with the elements.
694 This is good for vectors that are "usually small" (e.g. the number of
695 predecessors/successors of a block is usually less than 8). On the other hand,
696 this makes the size of the SmallVector itself large, so you don't want to
697 allocate lots of them (doing so will waste a lot of space). As such,
698 SmallVectors are most useful when on the stack.
700 SmallVector also provides a nice portable and efficient replacement for
705 Prefer to use ``SmallVectorImpl<T>`` as a parameter type.
707 In APIs that don't care about the "small size" (most?), prefer to use
708 the ``SmallVectorImpl<T>`` class, which is basically just the "vector
709 header" (and methods) without the elements allocated after it. Note that
710 ``SmallVector<T, N>`` inherits from ``SmallVectorImpl<T>`` so the
711 conversion is implicit and costs nothing. E.g.
715 // BAD: Clients cannot pass e.g. SmallVector<Foo, 4>.
716 hardcodedSmallSize(SmallVector<Foo, 2> &Out);
717 // GOOD: Clients can pass any SmallVector<Foo, N>.
718 allowsAnySmallSize(SmallVectorImpl<Foo> &Out);
721 SmallVector<Foo, 8> Vec;
722 hardcodedSmallSize(Vec); // Error.
723 allowsAnySmallSize(Vec); // Works.
726 Even though it has "``Impl``" in the name, this is so widely used that
727 it really isn't "private to the implementation" anymore. A name like
728 ``SmallVectorHeader`` would be more appropriate.
735 ``std::vector`` is well loved and respected. It is useful when SmallVector
736 isn't: when the size of the vector is often large (thus the small optimization
737 will rarely be a benefit) or if you will be allocating many instances of the
738 vector itself (which would waste space for elements that aren't in the
739 container). vector is also useful when interfacing with code that expects
742 One worthwhile note about std::vector: avoid code like this:
751 Instead, write this as:
761 Doing so will save (at least) one heap allocation and free per iteration of the
769 ``std::deque`` is, in some senses, a generalized version of ``std::vector``.
770 Like ``std::vector``, it provides constant time random access and other similar
771 properties, but it also provides efficient access to the front of the list. It
772 does not guarantee continuity of elements within memory.
774 In exchange for this extra flexibility, ``std::deque`` has significantly higher
775 constant factor costs than ``std::vector``. If possible, use ``std::vector`` or
783 ``std::list`` is an extremely inefficient class that is rarely useful. It
784 performs a heap allocation for every element inserted into it, thus having an
785 extremely high constant factor, particularly for small data types.
786 ``std::list`` also only supports bidirectional iteration, not random access
789 In exchange for this high cost, std::list supports efficient access to both ends
790 of the list (like ``std::deque``, but unlike ``std::vector`` or
791 ``SmallVector``). In addition, the iterator invalidation characteristics of
792 std::list are stronger than that of a vector class: inserting or removing an
793 element into the list does not invalidate iterator or pointers to other elements
801 ``ilist<T>`` implements an 'intrusive' doubly-linked list. It is intrusive,
802 because it requires the element to store and provide access to the prev/next
803 pointers for the list.
805 ``ilist`` has the same drawbacks as ``std::list``, and additionally requires an
806 ``ilist_traits`` implementation for the element type, but it provides some novel
807 characteristics. In particular, it can efficiently store polymorphic objects,
808 the traits class is informed when an element is inserted or removed from the
809 list, and ``ilist``\ s are guaranteed to support a constant-time splice
812 These properties are exactly what we want for things like ``Instruction``\ s and
813 basic blocks, which is why these are implemented with ``ilist``\ s.
815 Related classes of interest are explained in the following subsections:
817 * :ref:`ilist_traits <dss_ilist_traits>`
819 * :ref:`iplist <dss_iplist>`
821 * :ref:`llvm/ADT/ilist_node.h <dss_ilist_node>`
823 * :ref:`Sentinels <dss_ilist_sentinel>`
825 .. _dss_packedvector:
827 llvm/ADT/PackedVector.h
828 ^^^^^^^^^^^^^^^^^^^^^^^
830 Useful for storing a vector of values using only a few number of bits for each
831 value. Apart from the standard operations of a vector-like container, it can
832 also perform an 'or' set operation.
840 FirstCondition = 0x1,
841 SecondCondition = 0x2,
846 PackedVector<State, 2> Vec1;
847 Vec1.push_back(FirstCondition);
849 PackedVector<State, 2> Vec2;
850 Vec2.push_back(SecondCondition);
853 return Vec1[0]; // returns 'Both'.
856 .. _dss_ilist_traits:
861 ``ilist_traits<T>`` is ``ilist<T>``'s customization mechanism. ``iplist<T>``
862 (and consequently ``ilist<T>``) publicly derive from this traits class.
869 ``iplist<T>`` is ``ilist<T>``'s base and as such supports a slightly narrower
870 interface. Notably, inserters from ``T&`` are absent.
872 ``ilist_traits<T>`` is a public base of this class and can be used for a wide
873 variety of customizations.
877 llvm/ADT/ilist_node.h
878 ^^^^^^^^^^^^^^^^^^^^^
880 ``ilist_node<T>`` implements a the forward and backward links that are expected
881 by the ``ilist<T>`` (and analogous containers) in the default manner.
883 ``ilist_node<T>``\ s are meant to be embedded in the node type ``T``, usually
884 ``T`` publicly derives from ``ilist_node<T>``.
886 .. _dss_ilist_sentinel:
891 ``ilist``\ s have another specialty that must be considered. To be a good
892 citizen in the C++ ecosystem, it needs to support the standard container
893 operations, such as ``begin`` and ``end`` iterators, etc. Also, the
894 ``operator--`` must work correctly on the ``end`` iterator in the case of
895 non-empty ``ilist``\ s.
897 The only sensible solution to this problem is to allocate a so-called *sentinel*
898 along with the intrusive list, which serves as the ``end`` iterator, providing
899 the back-link to the last element. However conforming to the C++ convention it
900 is illegal to ``operator++`` beyond the sentinel and it also must not be
903 These constraints allow for some implementation freedom to the ``ilist`` how to
904 allocate and store the sentinel. The corresponding policy is dictated by
905 ``ilist_traits<T>``. By default a ``T`` gets heap-allocated whenever the need
906 for a sentinel arises.
908 While the default policy is sufficient in most cases, it may break down when
909 ``T`` does not provide a default constructor. Also, in the case of many
910 instances of ``ilist``\ s, the memory overhead of the associated sentinels is
911 wasted. To alleviate the situation with numerous and voluminous
912 ``T``-sentinels, sometimes a trick is employed, leading to *ghostly sentinels*.
914 Ghostly sentinels are obtained by specially-crafted ``ilist_traits<T>`` which
915 superpose the sentinel with the ``ilist`` instance in memory. Pointer
916 arithmetic is used to obtain the sentinel, which is relative to the ``ilist``'s
917 ``this`` pointer. The ``ilist`` is augmented by an extra pointer, which serves
918 as the back-link of the sentinel. This is the only field in the ghostly
919 sentinel which can be legally accessed.
923 Other Sequential Container options
924 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
926 Other STL containers are available, such as ``std::string``.
928 There are also various STL adapter classes such as ``std::queue``,
929 ``std::priority_queue``, ``std::stack``, etc. These provide simplified access
930 to an underlying container but don't affect the cost of the container itself.
934 String-like containers
935 ----------------------
937 There are a variety of ways to pass around and use strings in C and C++, and
938 LLVM adds a few new options to choose from. Pick the first option on this list
939 that will do what you need, they are ordered according to their relative cost.
941 Note that is is generally preferred to *not* pass strings around as ``const
942 char*``'s. These have a number of problems, including the fact that they
943 cannot represent embedded nul ("\0") characters, and do not have a length
944 available efficiently. The general replacement for '``const char*``' is
947 For more information on choosing string containers for APIs, please see
948 :ref:`Passing Strings <string_apis>`.
955 The StringRef class is a simple value class that contains a pointer to a
956 character and a length, and is quite related to the :ref:`ArrayRef
957 <dss_arrayref>` class (but specialized for arrays of characters). Because
958 StringRef carries a length with it, it safely handles strings with embedded nul
959 characters in it, getting the length does not require a strlen call, and it even
960 has very convenient APIs for slicing and dicing the character range that it
963 StringRef is ideal for passing simple strings around that are known to be live,
964 either because they are C string literals, std::string, a C array, or a
965 SmallVector. Each of these cases has an efficient implicit conversion to
966 StringRef, which doesn't result in a dynamic strlen being executed.
968 StringRef has a few major limitations which make more powerful string containers
971 #. You cannot directly convert a StringRef to a 'const char*' because there is
972 no way to add a trailing nul (unlike the .c_str() method on various stronger
975 #. StringRef doesn't own or keep alive the underlying string bytes.
976 As such it can easily lead to dangling pointers, and is not suitable for
977 embedding in datastructures in most cases (instead, use an std::string or
978 something like that).
980 #. For the same reason, StringRef cannot be used as the return value of a
981 method if the method "computes" the result string. Instead, use std::string.
983 #. StringRef's do not allow you to mutate the pointed-to string bytes and it
984 doesn't allow you to insert or remove bytes from the range. For editing
985 operations like this, it interoperates with the :ref:`Twine <dss_twine>`
988 Because of its strengths and limitations, it is very common for a function to
989 take a StringRef and for a method on an object to return a StringRef that points
990 into some string that it owns.
997 The Twine class is used as an intermediary datatype for APIs that want to take a
998 string that can be constructed inline with a series of concatenations. Twine
999 works by forming recursive instances of the Twine datatype (a simple value
1000 object) on the stack as temporary objects, linking them together into a tree
1001 which is then linearized when the Twine is consumed. Twine is only safe to use
1002 as the argument to a function, and should always be a const reference, e.g.:
1006 void foo(const Twine &T);
1010 foo(X + "." + Twine(i));
1012 This example forms a string like "blarg.42" by concatenating the values
1013 together, and does not form intermediate strings containing "blarg" or "blarg.".
1015 Because Twine is constructed with temporary objects on the stack, and because
1016 these instances are destroyed at the end of the current statement, it is an
1017 inherently dangerous API. For example, this simple variant contains undefined
1018 behavior and will probably crash:
1022 void foo(const Twine &T);
1026 const Twine &Tmp = X + "." + Twine(i);
1029 ... because the temporaries are destroyed before the call. That said, Twine's
1030 are much more efficient than intermediate std::string temporaries, and they work
1031 really well with StringRef. Just be aware of their limitations.
1033 .. _dss_smallstring:
1035 llvm/ADT/SmallString.h
1036 ^^^^^^^^^^^^^^^^^^^^^^
1038 SmallString is a subclass of :ref:`SmallVector <dss_smallvector>` that adds some
1039 convenience APIs like += that takes StringRef's. SmallString avoids allocating
1040 memory in the case when the preallocated space is enough to hold its data, and
1041 it calls back to general heap allocation when required. Since it owns its data,
1042 it is very safe to use and supports full mutation of the string.
1044 Like SmallVector's, the big downside to SmallString is their sizeof. While they
1045 are optimized for small strings, they themselves are not particularly small.
1046 This means that they work great for temporary scratch buffers on the stack, but
1047 should not generally be put into the heap: it is very rare to see a SmallString
1048 as the member of a frequently-allocated heap data structure or returned
1056 The standard C++ std::string class is a very general class that (like
1057 SmallString) owns its underlying data. sizeof(std::string) is very reasonable
1058 so it can be embedded into heap data structures and returned by-value. On the
1059 other hand, std::string is highly inefficient for inline editing (e.g.
1060 concatenating a bunch of stuff together) and because it is provided by the
1061 standard library, its performance characteristics depend a lot of the host
1062 standard library (e.g. libc++ and MSVC provide a highly optimized string class,
1063 GCC contains a really slow implementation).
1065 The major disadvantage of std::string is that almost every operation that makes
1066 them larger can allocate memory, which is slow. As such, it is better to use
1067 SmallVector or Twine as a scratch buffer, but then use std::string to persist
1072 Set-Like Containers (std::set, SmallSet, SetVector, etc)
1073 --------------------------------------------------------
1075 Set-like containers are useful when you need to canonicalize multiple values
1076 into a single representation. There are several different choices for how to do
1077 this, providing various trade-offs.
1079 .. _dss_sortedvectorset:
1084 If you intend to insert a lot of elements, then do a lot of queries, a great
1085 approach is to use a vector (or other sequential container) with
1086 std::sort+std::unique to remove duplicates. This approach works really well if
1087 your usage pattern has these two distinct phases (insert then query), and can be
1088 coupled with a good choice of :ref:`sequential container <ds_sequential>`.
1090 This combination provides the several nice properties: the result data is
1091 contiguous in memory (good for cache locality), has few allocations, is easy to
1092 address (iterators in the final vector are just indices or pointers), and can be
1093 efficiently queried with a standard binary search (e.g.
1094 ``std::lower_bound``; if you want the whole range of elements comparing
1095 equal, use ``std::equal_range``).
1102 If you have a set-like data structure that is usually small and whose elements
1103 are reasonably small, a ``SmallSet<Type, N>`` is a good choice. This set has
1104 space for N elements in place (thus, if the set is dynamically smaller than N,
1105 no malloc traffic is required) and accesses them with a simple linear search.
1106 When the set grows beyond 'N' elements, it allocates a more expensive
1107 representation that guarantees efficient access (for most types, it falls back
1108 to std::set, but for pointers it uses something far better, :ref:`SmallPtrSet
1111 The magic of this class is that it handles small sets extremely efficiently, but
1112 gracefully handles extremely large sets without loss of efficiency. The
1113 drawback is that the interface is quite small: it supports insertion, queries
1114 and erasing, but does not support iteration.
1116 .. _dss_smallptrset:
1118 llvm/ADT/SmallPtrSet.h
1119 ^^^^^^^^^^^^^^^^^^^^^^
1121 SmallPtrSet has all the advantages of ``SmallSet`` (and a ``SmallSet`` of
1122 pointers is transparently implemented with a ``SmallPtrSet``), but also supports
1123 iterators. If more than 'N' insertions are performed, a single quadratically
1124 probed hash table is allocated and grows as needed, providing extremely
1125 efficient access (constant time insertion/deleting/queries with low constant
1126 factors) and is very stingy with malloc traffic.
1128 Note that, unlike ``std::set``, the iterators of ``SmallPtrSet`` are invalidated
1129 whenever an insertion occurs. Also, the values visited by the iterators are not
1130 visited in sorted order.
1137 DenseSet is a simple quadratically probed hash table. It excels at supporting
1138 small values: it uses a single allocation to hold all of the pairs that are
1139 currently inserted in the set. DenseSet is a great way to unique small values
1140 that are not simple pointers (use :ref:`SmallPtrSet <dss_smallptrset>` for
1141 pointers). Note that DenseSet has the same requirements for the value type that
1142 :ref:`DenseMap <dss_densemap>` has.
1146 llvm/ADT/SparseSet.h
1147 ^^^^^^^^^^^^^^^^^^^^
1149 SparseSet holds a small number of objects identified by unsigned keys of
1150 moderate size. It uses a lot of memory, but provides operations that are almost
1151 as fast as a vector. Typical keys are physical registers, virtual registers, or
1152 numbered basic blocks.
1154 SparseSet is useful for algorithms that need very fast clear/find/insert/erase
1155 and fast iteration over small sets. It is not intended for building composite
1158 .. _dss_sparsemultiset:
1160 llvm/ADT/SparseMultiSet.h
1161 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1163 SparseMultiSet adds multiset behavior to SparseSet, while retaining SparseSet's
1164 desirable attributes. Like SparseSet, it typically uses a lot of memory, but
1165 provides operations that are almost as fast as a vector. Typical keys are
1166 physical registers, virtual registers, or numbered basic blocks.
1168 SparseMultiSet is useful for algorithms that need very fast
1169 clear/find/insert/erase of the entire collection, and iteration over sets of
1170 elements sharing a key. It is often a more efficient choice than using composite
1171 data structures (e.g. vector-of-vectors, map-of-vectors). It is not intended for
1172 building composite data structures.
1176 llvm/ADT/FoldingSet.h
1177 ^^^^^^^^^^^^^^^^^^^^^
1179 FoldingSet is an aggregate class that is really good at uniquing
1180 expensive-to-create or polymorphic objects. It is a combination of a chained
1181 hash table with intrusive links (uniqued objects are required to inherit from
1182 FoldingSetNode) that uses :ref:`SmallVector <dss_smallvector>` as part of its ID
1185 Consider a case where you want to implement a "getOrCreateFoo" method for a
1186 complex object (for example, a node in the code generator). The client has a
1187 description of **what** it wants to generate (it knows the opcode and all the
1188 operands), but we don't want to 'new' a node, then try inserting it into a set
1189 only to find out it already exists, at which point we would have to delete it
1190 and return the node that already exists.
1192 To support this style of client, FoldingSet perform a query with a
1193 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1194 element that we want to query for. The query either returns the element
1195 matching the ID or it returns an opaque ID that indicates where insertion should
1196 take place. Construction of the ID usually does not require heap traffic.
1198 Because FoldingSet uses intrusive links, it can support polymorphic objects in
1199 the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1200 Because the elements are individually allocated, pointers to the elements are
1201 stable: inserting or removing elements does not invalidate any pointers to other
1209 ``std::set`` is a reasonable all-around set class, which is decent at many
1210 things but great at nothing. std::set allocates memory for each element
1211 inserted (thus it is very malloc intensive) and typically stores three pointers
1212 per element in the set (thus adding a large amount of per-element space
1213 overhead). It offers guaranteed log(n) performance, which is not particularly
1214 fast from a complexity standpoint (particularly if the elements of the set are
1215 expensive to compare, like strings), and has extremely high constant factors for
1216 lookup, insertion and removal.
1218 The advantages of std::set are that its iterators are stable (deleting or
1219 inserting an element from the set does not affect iterators or pointers to other
1220 elements) and that iteration over the set is guaranteed to be in sorted order.
1221 If the elements in the set are large, then the relative overhead of the pointers
1222 and malloc traffic is not a big deal, but if the elements of the set are small,
1223 std::set is almost never a good choice.
1227 llvm/ADT/SetVector.h
1228 ^^^^^^^^^^^^^^^^^^^^
1230 LLVM's ``SetVector<Type>`` is an adapter class that combines your choice of a
1231 set-like container along with a :ref:`Sequential Container <ds_sequential>` The
1232 important property that this provides is efficient insertion with uniquing
1233 (duplicate elements are ignored) with iteration support. It implements this by
1234 inserting elements into both a set-like container and the sequential container,
1235 using the set-like container for uniquing and the sequential container for
1238 The difference between SetVector and other sets is that the order of iteration
1239 is guaranteed to match the order of insertion into the SetVector. This property
1240 is really important for things like sets of pointers. Because pointer values
1241 are non-deterministic (e.g. vary across runs of the program on different
1242 machines), iterating over the pointers in the set will not be in a well-defined
1245 The drawback of SetVector is that it requires twice as much space as a normal
1246 set and has the sum of constant factors from the set-like container and the
1247 sequential container that it uses. Use it **only** if you need to iterate over
1248 the elements in a deterministic order. SetVector is also expensive to delete
1249 elements out of (linear time), unless you use its "pop_back" method, which is
1252 ``SetVector`` is an adapter class that defaults to using ``std::vector`` and a
1253 size 16 ``SmallSet`` for the underlying containers, so it is quite expensive.
1254 However, ``"llvm/ADT/SetVector.h"`` also provides a ``SmallSetVector`` class,
1255 which defaults to using a ``SmallVector`` and ``SmallSet`` of a specified size.
1256 If you use this, and if your sets are dynamically smaller than ``N``, you will
1257 save a lot of heap traffic.
1259 .. _dss_uniquevector:
1261 llvm/ADT/UniqueVector.h
1262 ^^^^^^^^^^^^^^^^^^^^^^^
1264 UniqueVector is similar to :ref:`SetVector <dss_setvector>` but it retains a
1265 unique ID for each element inserted into the set. It internally contains a map
1266 and a vector, and it assigns a unique ID for each value inserted into the set.
1268 UniqueVector is very expensive: its cost is the sum of the cost of maintaining
1269 both the map and vector, it has high complexity, high constant factors, and
1270 produces a lot of malloc traffic. It should be avoided.
1272 .. _dss_immutableset:
1274 llvm/ADT/ImmutableSet.h
1275 ^^^^^^^^^^^^^^^^^^^^^^^
1277 ImmutableSet is an immutable (functional) set implementation based on an AVL
1278 tree. Adding or removing elements is done through a Factory object and results
1279 in the creation of a new ImmutableSet object. If an ImmutableSet already exists
1280 with the given contents, then the existing one is returned; equality is compared
1281 with a FoldingSetNodeID. The time and space complexity of add or remove
1282 operations is logarithmic in the size of the original set.
1284 There is no method for returning an element of the set, you can only check for
1289 Other Set-Like Container Options
1290 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1292 The STL provides several other options, such as std::multiset and the various
1293 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1294 never use hash_set and unordered_set because they are generally very expensive
1295 (each insertion requires a malloc) and very non-portable.
1297 std::multiset is useful if you're not interested in elimination of duplicates,
1298 but has all the drawbacks of std::set. A sorted vector (where you don't delete
1299 duplicate entries) or some other approach is almost always better.
1303 Map-Like Containers (std::map, DenseMap, etc)
1304 ---------------------------------------------
1306 Map-like containers are useful when you want to associate data to a key. As
1307 usual, there are a lot of different ways to do this. :)
1309 .. _dss_sortedvectormap:
1314 If your usage pattern follows a strict insert-then-query approach, you can
1315 trivially use the same approach as :ref:`sorted vectors for set-like containers
1316 <dss_sortedvectorset>`. The only difference is that your query function (which
1317 uses std::lower_bound to get efficient log(n) lookup) should only compare the
1318 key, not both the key and value. This yields the same advantages as sorted
1323 llvm/ADT/StringMap.h
1324 ^^^^^^^^^^^^^^^^^^^^
1326 Strings are commonly used as keys in maps, and they are difficult to support
1327 efficiently: they are variable length, inefficient to hash and compare when
1328 long, expensive to copy, etc. StringMap is a specialized container designed to
1329 cope with these issues. It supports mapping an arbitrary range of bytes to an
1330 arbitrary other object.
1332 The StringMap implementation uses a quadratically-probed hash table, where the
1333 buckets store a pointer to the heap allocated entries (and some other stuff).
1334 The entries in the map must be heap allocated because the strings are variable
1335 length. The string data (key) and the element object (value) are stored in the
1336 same allocation with the string data immediately after the element object.
1337 This container guarantees the "``(char*)(&Value+1)``" points to the key string
1340 The StringMap is very fast for several reasons: quadratic probing is very cache
1341 efficient for lookups, the hash value of strings in buckets is not recomputed
1342 when looking up an element, StringMap rarely has to touch the memory for
1343 unrelated objects when looking up a value (even when hash collisions happen),
1344 hash table growth does not recompute the hash values for strings already in the
1345 table, and each pair in the map is store in a single allocation (the string data
1346 is stored in the same allocation as the Value of a pair).
1348 StringMap also provides query methods that take byte ranges, so it only ever
1349 copies a string if a value is inserted into the table.
1351 StringMap iteratation order, however, is not guaranteed to be deterministic, so
1352 any uses which require that should instead use a std::map.
1356 llvm/ADT/IndexedMap.h
1357 ^^^^^^^^^^^^^^^^^^^^^
1359 IndexedMap is a specialized container for mapping small dense integers (or
1360 values that can be mapped to small dense integers) to some other type. It is
1361 internally implemented as a vector with a mapping function that maps the keys
1362 to the dense integer range.
1364 This is useful for cases like virtual registers in the LLVM code generator: they
1365 have a dense mapping that is offset by a compile-time constant (the first
1366 virtual register ID).
1373 DenseMap is a simple quadratically probed hash table. It excels at supporting
1374 small keys and values: it uses a single allocation to hold all of the pairs
1375 that are currently inserted in the map. DenseMap is a great way to map
1376 pointers to pointers, or map other small types to each other.
1378 There are several aspects of DenseMap that you should be aware of, however.
1379 The iterators in a DenseMap are invalidated whenever an insertion occurs,
1380 unlike map. Also, because DenseMap allocates space for a large number of
1381 key/value pairs (it starts with 64 by default), it will waste a lot of space if
1382 your keys or values are large. Finally, you must implement a partial
1383 specialization of DenseMapInfo for the key that you want, if it isn't already
1384 supported. This is required to tell DenseMap about two special marker values
1385 (which can never be inserted into the map) that it needs internally.
1387 DenseMap's find_as() method supports lookup operations using an alternate key
1388 type. This is useful in cases where the normal key type is expensive to
1389 construct, but cheap to compare against. The DenseMapInfo is responsible for
1390 defining the appropriate comparison and hashing methods for each alternate key
1398 ValueMap is a wrapper around a :ref:`DenseMap <dss_densemap>` mapping
1399 ``Value*``\ s (or subclasses) to another type. When a Value is deleted or
1400 RAUW'ed, ValueMap will update itself so the new version of the key is mapped to
1401 the same value, just as if the key were a WeakVH. You can configure exactly how
1402 this happens, and what else happens on these two events, by passing a ``Config``
1403 parameter to the ValueMap template.
1405 .. _dss_intervalmap:
1407 llvm/ADT/IntervalMap.h
1408 ^^^^^^^^^^^^^^^^^^^^^^
1410 IntervalMap is a compact map for small keys and values. It maps key intervals
1411 instead of single keys, and it will automatically coalesce adjacent intervals.
1412 When then map only contains a few intervals, they are stored in the map object
1413 itself to avoid allocations.
1415 The IntervalMap iterators are quite big, so they should not be passed around as
1416 STL iterators. The heavyweight iterators allow a smaller data structure.
1423 std::map has similar characteristics to :ref:`std::set <dss_set>`: it uses a
1424 single allocation per pair inserted into the map, it offers log(n) lookup with
1425 an extremely large constant factor, imposes a space penalty of 3 pointers per
1426 pair in the map, etc.
1428 std::map is most useful when your keys or values are very large, if you need to
1429 iterate over the collection in sorted order, or if you need stable iterators
1430 into the map (i.e. they don't get invalidated if an insertion or deletion of
1431 another element takes place).
1435 llvm/ADT/MapVector.h
1436 ^^^^^^^^^^^^^^^^^^^^
1438 ``MapVector<KeyT,ValueT>`` provides a subset of the DenseMap interface. The
1439 main difference is that the iteration order is guaranteed to be the insertion
1440 order, making it an easy (but somewhat expensive) solution for non-deterministic
1441 iteration over maps of pointers.
1443 It is implemented by mapping from key to an index in a vector of key,value
1444 pairs. This provides fast lookup and iteration, but has two main drawbacks:
1445 the key is stored twice and removing elements takes linear time. If it is
1446 necessary to remove elements, it's best to remove them in bulk using
1449 .. _dss_inteqclasses:
1451 llvm/ADT/IntEqClasses.h
1452 ^^^^^^^^^^^^^^^^^^^^^^^
1454 IntEqClasses provides a compact representation of equivalence classes of small
1455 integers. Initially, each integer in the range 0..n-1 has its own equivalence
1456 class. Classes can be joined by passing two class representatives to the
1457 join(a, b) method. Two integers are in the same class when findLeader() returns
1458 the same representative.
1460 Once all equivalence classes are formed, the map can be compressed so each
1461 integer 0..n-1 maps to an equivalence class number in the range 0..m-1, where m
1462 is the total number of equivalence classes. The map must be uncompressed before
1463 it can be edited again.
1465 .. _dss_immutablemap:
1467 llvm/ADT/ImmutableMap.h
1468 ^^^^^^^^^^^^^^^^^^^^^^^
1470 ImmutableMap is an immutable (functional) map implementation based on an AVL
1471 tree. Adding or removing elements is done through a Factory object and results
1472 in the creation of a new ImmutableMap object. If an ImmutableMap already exists
1473 with the given key set, then the existing one is returned; equality is compared
1474 with a FoldingSetNodeID. The time and space complexity of add or remove
1475 operations is logarithmic in the size of the original map.
1479 Other Map-Like Container Options
1480 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1482 The STL provides several other options, such as std::multimap and the various
1483 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1484 never use hash_set and unordered_set because they are generally very expensive
1485 (each insertion requires a malloc) and very non-portable.
1487 std::multimap is useful if you want to map a key to multiple values, but has all
1488 the drawbacks of std::map. A sorted vector or some other approach is almost
1493 Bit storage containers (BitVector, SparseBitVector)
1494 ---------------------------------------------------
1496 Unlike the other containers, there are only two bit storage containers, and
1497 choosing when to use each is relatively straightforward.
1499 One additional option is ``std::vector<bool>``: we discourage its use for two
1500 reasons 1) the implementation in many common compilers (e.g. commonly
1501 available versions of GCC) is extremely inefficient and 2) the C++ standards
1502 committee is likely to deprecate this container and/or change it significantly
1503 somehow. In any case, please don't use it.
1510 The BitVector container provides a dynamic size set of bits for manipulation.
1511 It supports individual bit setting/testing, as well as set operations. The set
1512 operations take time O(size of bitvector), but operations are performed one word
1513 at a time, instead of one bit at a time. This makes the BitVector very fast for
1514 set operations compared to other containers. Use the BitVector when you expect
1515 the number of set bits to be high (i.e. a dense set).
1517 .. _dss_smallbitvector:
1522 The SmallBitVector container provides the same interface as BitVector, but it is
1523 optimized for the case where only a small number of bits, less than 25 or so,
1524 are needed. It also transparently supports larger bit counts, but slightly less
1525 efficiently than a plain BitVector, so SmallBitVector should only be used when
1526 larger counts are rare.
1528 At this time, SmallBitVector does not support set operations (and, or, xor), and
1529 its operator[] does not provide an assignable lvalue.
1531 .. _dss_sparsebitvector:
1536 The SparseBitVector container is much like BitVector, with one major difference:
1537 Only the bits that are set, are stored. This makes the SparseBitVector much
1538 more space efficient than BitVector when the set is sparse, as well as making
1539 set operations O(number of set bits) instead of O(size of universe). The
1540 downside to the SparseBitVector is that setting and testing of random bits is
1541 O(N), and on large SparseBitVectors, this can be slower than BitVector. In our
1542 implementation, setting or testing bits in sorted order (either forwards or
1543 reverse) is O(1) worst case. Testing and setting bits within 128 bits (depends
1544 on size) of the current bit is also O(1). As a general statement,
1545 testing/setting bits in a SparseBitVector is O(distance away from last set bit).
1549 Helpful Hints for Common Operations
1550 ===================================
1552 This section describes how to perform some very simple transformations of LLVM
1553 code. This is meant to give examples of common idioms used, showing the
1554 practical side of LLVM transformations.
1556 Because this is a "how-to" section, you should also read about the main classes
1557 that you will be working with. The :ref:`Core LLVM Class Hierarchy Reference
1558 <coreclasses>` contains details and descriptions of the main classes that you
1563 Basic Inspection and Traversal Routines
1564 ---------------------------------------
1566 The LLVM compiler infrastructure have many different data structures that may be
1567 traversed. Following the example of the C++ standard template library, the
1568 techniques used to traverse these various data structures are all basically the
1569 same. For a enumerable sequence of values, the ``XXXbegin()`` function (or
1570 method) returns an iterator to the start of the sequence, the ``XXXend()``
1571 function returns an iterator pointing to one past the last valid element of the
1572 sequence, and there is some ``XXXiterator`` data type that is common between the
1575 Because the pattern for iteration is common across many different aspects of the
1576 program representation, the standard template library algorithms may be used on
1577 them, and it is easier to remember how to iterate. First we show a few common
1578 examples of the data structures that need to be traversed. Other data
1579 structures are traversed in very similar ways.
1581 .. _iterate_function:
1583 Iterating over the ``BasicBlock`` in a ``Function``
1584 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1586 It's quite common to have a ``Function`` instance that you'd like to transform
1587 in some way; in particular, you'd like to manipulate its ``BasicBlock``\ s. To
1588 facilitate this, you'll need to iterate over all of the ``BasicBlock``\ s that
1589 constitute the ``Function``. The following is an example that prints the name
1590 of a ``BasicBlock`` and the number of ``Instruction``\ s it contains:
1594 // func is a pointer to a Function instance
1595 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
1596 // Print out the name of the basic block if it has one, and then the
1597 // number of instructions that it contains
1598 errs() << "Basic block (name=" << i->getName() << ") has "
1599 << i->size() << " instructions.\n";
1601 Note that i can be used as if it were a pointer for the purposes of invoking
1602 member functions of the ``Instruction`` class. This is because the indirection
1603 operator is overloaded for the iterator classes. In the above code, the
1604 expression ``i->size()`` is exactly equivalent to ``(*i).size()`` just like
1607 .. _iterate_basicblock:
1609 Iterating over the ``Instruction`` in a ``BasicBlock``
1610 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1612 Just like when dealing with ``BasicBlock``\ s in ``Function``\ s, it's easy to
1613 iterate over the individual instructions that make up ``BasicBlock``\ s. Here's
1614 a code snippet that prints out each instruction in a ``BasicBlock``:
1618 // blk is a pointer to a BasicBlock instance
1619 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
1620 // The next statement works since operator<<(ostream&,...)
1621 // is overloaded for Instruction&
1622 errs() << *i << "\n";
1625 However, this isn't really the best way to print out the contents of a
1626 ``BasicBlock``! Since the ostream operators are overloaded for virtually
1627 anything you'll care about, you could have just invoked the print routine on the
1628 basic block itself: ``errs() << *blk << "\n";``.
1630 .. _iterate_insiter:
1632 Iterating over the ``Instruction`` in a ``Function``
1633 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1635 If you're finding that you commonly iterate over a ``Function``'s
1636 ``BasicBlock``\ s and then that ``BasicBlock``'s ``Instruction``\ s,
1637 ``InstIterator`` should be used instead. You'll need to include
1638 ``llvm/IR/InstIterator.h`` (`doxygen
1639 <http://llvm.org/doxygen/InstIterator_8h.html>`__) and then instantiate
1640 ``InstIterator``\ s explicitly in your code. Here's a small example that shows
1641 how to dump all instructions in a function to the standard error stream:
1645 #include "llvm/IR/InstIterator.h"
1647 // F is a pointer to a Function instance
1648 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1649 errs() << *I << "\n";
1651 Easy, isn't it? You can also use ``InstIterator``\ s to fill a work list with
1652 its initial contents. For example, if you wanted to initialize a work list to
1653 contain all instructions in a ``Function`` F, all you would need to do is
1658 std::set<Instruction*> worklist;
1659 // or better yet, SmallPtrSet<Instruction*, 64> worklist;
1661 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1662 worklist.insert(&*I);
1664 The STL set ``worklist`` would now contain all instructions in the ``Function``
1667 .. _iterate_convert:
1669 Turning an iterator into a class pointer (and vice-versa)
1670 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1672 Sometimes, it'll be useful to grab a reference (or pointer) to a class instance
1673 when all you've got at hand is an iterator. Well, extracting a reference or a
1674 pointer from an iterator is very straight-forward. Assuming that ``i`` is a
1675 ``BasicBlock::iterator`` and ``j`` is a ``BasicBlock::const_iterator``:
1679 Instruction& inst = *i; // Grab reference to instruction reference
1680 Instruction* pinst = &*i; // Grab pointer to instruction reference
1681 const Instruction& inst = *j;
1683 However, the iterators you'll be working with in the LLVM framework are special:
1684 they will automatically convert to a ptr-to-instance type whenever they need to.
1685 Instead of derferencing the iterator and then taking the address of the result,
1686 you can simply assign the iterator to the proper pointer type and you get the
1687 dereference and address-of operation as a result of the assignment (behind the
1688 scenes, this is a result of overloading casting mechanisms). Thus the last line
1689 of the last example,
1693 Instruction *pinst = &*i;
1695 is semantically equivalent to
1699 Instruction *pinst = i;
1701 It's also possible to turn a class pointer into the corresponding iterator, and
1702 this is a constant time operation (very efficient). The following code snippet
1703 illustrates use of the conversion constructors provided by LLVM iterators. By
1704 using these, you can explicitly grab the iterator of something without actually
1705 obtaining it via iteration over some structure:
1709 void printNextInstruction(Instruction* inst) {
1710 BasicBlock::iterator it(inst);
1711 ++it; // After this line, it refers to the instruction after *inst
1712 if (it != inst->getParent()->end()) errs() << *it << "\n";
1715 Unfortunately, these implicit conversions come at a cost; they prevent these
1716 iterators from conforming to standard iterator conventions, and thus from being
1717 usable with standard algorithms and containers. For example, they prevent the
1718 following code, where ``B`` is a ``BasicBlock``, from compiling:
1722 llvm::SmallVector<llvm::Instruction *, 16>(B->begin(), B->end());
1724 Because of this, these implicit conversions may be removed some day, and
1725 ``operator*`` changed to return a pointer instead of a reference.
1727 .. _iterate_complex:
1729 Finding call sites: a slightly more complex example
1730 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1732 Say that you're writing a FunctionPass and would like to count all the locations
1733 in the entire module (that is, across every ``Function``) where a certain
1734 function (i.e., some ``Function *``) is already in scope. As you'll learn
1735 later, you may want to use an ``InstVisitor`` to accomplish this in a much more
1736 straight-forward manner, but this example will allow us to explore how you'd do
1737 it if you didn't have ``InstVisitor`` around. In pseudo-code, this is what we
1740 .. code-block:: none
1742 initialize callCounter to zero
1743 for each Function f in the Module
1744 for each BasicBlock b in f
1745 for each Instruction i in b
1746 if (i is a CallInst and calls the given function)
1747 increment callCounter
1749 And the actual code is (remember, because we're writing a ``FunctionPass``, our
1750 ``FunctionPass``-derived class simply has to override the ``runOnFunction``
1755 Function* targetFunc = ...;
1757 class OurFunctionPass : public FunctionPass {
1759 OurFunctionPass(): callCounter(0) { }
1761 virtual runOnFunction(Function& F) {
1762 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1763 for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) {
1764 if (CallInst* callInst = dyn_cast<CallInst>(&*i)) {
1765 // We know we've encountered a call instruction, so we
1766 // need to determine if it's a call to the
1767 // function pointed to by m_func or not.
1768 if (callInst->getCalledFunction() == targetFunc)
1776 unsigned callCounter;
1779 .. _calls_and_invokes:
1781 Treating calls and invokes the same way
1782 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1784 You may have noticed that the previous example was a bit oversimplified in that
1785 it did not deal with call sites generated by 'invoke' instructions. In this,
1786 and in other situations, you may find that you want to treat ``CallInst``\ s and
1787 ``InvokeInst``\ s the same way, even though their most-specific common base
1788 class is ``Instruction``, which includes lots of less closely-related things.
1789 For these cases, LLVM provides a handy wrapper class called ``CallSite``
1790 (`doxygen <http://llvm.org/doxygen/classllvm_1_1CallSite.html>`__) It is
1791 essentially a wrapper around an ``Instruction`` pointer, with some methods that
1792 provide functionality common to ``CallInst``\ s and ``InvokeInst``\ s.
1794 This class has "value semantics": it should be passed by value, not by reference
1795 and it should not be dynamically allocated or deallocated using ``operator new``
1796 or ``operator delete``. It is efficiently copyable, assignable and
1797 constructable, with costs equivalents to that of a bare pointer. If you look at
1798 its definition, it has only a single pointer member.
1802 Iterating over def-use & use-def chains
1803 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1805 Frequently, we might have an instance of the ``Value`` class (`doxygen
1806 <http://llvm.org/doxygen/classllvm_1_1Value.html>`__) and we want to determine
1807 which ``User`` s use the ``Value``. The list of all ``User``\ s of a particular
1808 ``Value`` is called a *def-use* chain. For example, let's say we have a
1809 ``Function*`` named ``F`` to a particular function ``foo``. Finding all of the
1810 instructions that *use* ``foo`` is as simple as iterating over the *def-use*
1817 for (User *U : GV->users()) {
1818 if (Instruction *Inst = dyn_cast<Instruction>(U)) {
1819 errs() << "F is used in instruction:\n";
1820 errs() << *Inst << "\n";
1823 Alternatively, it's common to have an instance of the ``User`` Class (`doxygen
1824 <http://llvm.org/doxygen/classllvm_1_1User.html>`__) and need to know what
1825 ``Value``\ s are used by it. The list of all ``Value``\ s used by a ``User`` is
1826 known as a *use-def* chain. Instances of class ``Instruction`` are common
1827 ``User`` s, so we might want to iterate over all of the values that a particular
1828 instruction uses (that is, the operands of the particular ``Instruction``):
1832 Instruction *pi = ...;
1834 for (Use &U : pi->operands()) {
1839 Declaring objects as ``const`` is an important tool of enforcing mutation free
1840 algorithms (such as analyses, etc.). For this purpose above iterators come in
1841 constant flavors as ``Value::const_use_iterator`` and
1842 ``Value::const_op_iterator``. They automatically arise when calling
1843 ``use/op_begin()`` on ``const Value*``\ s or ``const User*``\ s respectively.
1844 Upon dereferencing, they return ``const Use*``\ s. Otherwise the above patterns
1849 Iterating over predecessors & successors of blocks
1850 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1852 Iterating over the predecessors and successors of a block is quite easy with the
1853 routines defined in ``"llvm/Support/CFG.h"``. Just use code like this to
1854 iterate over all predecessors of BB:
1858 #include "llvm/Support/CFG.h"
1859 BasicBlock *BB = ...;
1861 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
1862 BasicBlock *Pred = *PI;
1866 Similarly, to iterate over successors use ``succ_iterator/succ_begin/succ_end``.
1870 Making simple changes
1871 ---------------------
1873 There are some primitive transformation operations present in the LLVM
1874 infrastructure that are worth knowing about. When performing transformations,
1875 it's fairly common to manipulate the contents of basic blocks. This section
1876 describes some of the common methods for doing so and gives example code.
1878 .. _schanges_creating:
1880 Creating and inserting new ``Instruction``\ s
1881 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1883 *Instantiating Instructions*
1885 Creation of ``Instruction``\ s is straight-forward: simply call the constructor
1886 for the kind of instruction to instantiate and provide the necessary parameters.
1887 For example, an ``AllocaInst`` only *requires* a (const-ptr-to) ``Type``. Thus:
1891 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
1893 will create an ``AllocaInst`` instance that represents the allocation of one
1894 integer in the current stack frame, at run time. Each ``Instruction`` subclass
1895 is likely to have varying default parameters which change the semantics of the
1896 instruction, so refer to the `doxygen documentation for the subclass of
1897 Instruction <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_ that
1898 you're interested in instantiating.
1902 It is very useful to name the values of instructions when you're able to, as
1903 this facilitates the debugging of your transformations. If you end up looking
1904 at generated LLVM machine code, you definitely want to have logical names
1905 associated with the results of instructions! By supplying a value for the
1906 ``Name`` (default) parameter of the ``Instruction`` constructor, you associate a
1907 logical name with the result of the instruction's execution at run time. For
1908 example, say that I'm writing a transformation that dynamically allocates space
1909 for an integer on the stack, and that integer is going to be used as some kind
1910 of index by some other code. To accomplish this, I place an ``AllocaInst`` at
1911 the first point in the first ``BasicBlock`` of some ``Function``, and I'm
1912 intending to use it within the same ``Function``. I might do:
1916 AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
1918 where ``indexLoc`` is now the logical name of the instruction's execution value,
1919 which is a pointer to an integer on the run time stack.
1921 *Inserting instructions*
1923 There are essentially three ways to insert an ``Instruction`` into an existing
1924 sequence of instructions that form a ``BasicBlock``:
1926 * Insertion into an explicit instruction list
1928 Given a ``BasicBlock* pb``, an ``Instruction* pi`` within that ``BasicBlock``,
1929 and a newly-created instruction we wish to insert before ``*pi``, we do the
1934 BasicBlock *pb = ...;
1935 Instruction *pi = ...;
1936 Instruction *newInst = new Instruction(...);
1938 pb->getInstList().insert(pi, newInst); // Inserts newInst before pi in pb
1940 Appending to the end of a ``BasicBlock`` is so common that the ``Instruction``
1941 class and ``Instruction``-derived classes provide constructors which take a
1942 pointer to a ``BasicBlock`` to be appended to. For example code that looked
1947 BasicBlock *pb = ...;
1948 Instruction *newInst = new Instruction(...);
1950 pb->getInstList().push_back(newInst); // Appends newInst to pb
1956 BasicBlock *pb = ...;
1957 Instruction *newInst = new Instruction(..., pb);
1959 which is much cleaner, especially if you are creating long instruction
1962 * Insertion into an implicit instruction list
1964 ``Instruction`` instances that are already in ``BasicBlock``\ s are implicitly
1965 associated with an existing instruction list: the instruction list of the
1966 enclosing basic block. Thus, we could have accomplished the same thing as the
1967 above code without being given a ``BasicBlock`` by doing:
1971 Instruction *pi = ...;
1972 Instruction *newInst = new Instruction(...);
1974 pi->getParent()->getInstList().insert(pi, newInst);
1976 In fact, this sequence of steps occurs so frequently that the ``Instruction``
1977 class and ``Instruction``-derived classes provide constructors which take (as
1978 a default parameter) a pointer to an ``Instruction`` which the newly-created
1979 ``Instruction`` should precede. That is, ``Instruction`` constructors are
1980 capable of inserting the newly-created instance into the ``BasicBlock`` of a
1981 provided instruction, immediately before that instruction. Using an
1982 ``Instruction`` constructor with a ``insertBefore`` (default) parameter, the
1987 Instruction* pi = ...;
1988 Instruction* newInst = new Instruction(..., pi);
1990 which is much cleaner, especially if you're creating a lot of instructions and
1991 adding them to ``BasicBlock``\ s.
1993 * Insertion using an instance of ``IRBuilder``
1995 Inserting several ``Instruction``\ s can be quite laborious using the previous
1996 methods. The ``IRBuilder`` is a convenience class that can be used to add
1997 several instructions to the end of a ``BasicBlock`` or before a particular
1998 ``Instruction``. It also supports constant folding and renaming named
1999 registers (see ``IRBuilder``'s template arguments).
2001 The example below demonstrates a very simple use of the ``IRBuilder`` where
2002 three instructions are inserted before the instruction ``pi``. The first two
2003 instructions are Call instructions and third instruction multiplies the return
2004 value of the two calls.
2008 Instruction *pi = ...;
2009 IRBuilder<> Builder(pi);
2010 CallInst* callOne = Builder.CreateCall(...);
2011 CallInst* callTwo = Builder.CreateCall(...);
2012 Value* result = Builder.CreateMul(callOne, callTwo);
2014 The example below is similar to the above example except that the created
2015 ``IRBuilder`` inserts instructions at the end of the ``BasicBlock`` ``pb``.
2019 BasicBlock *pb = ...;
2020 IRBuilder<> Builder(pb);
2021 CallInst* callOne = Builder.CreateCall(...);
2022 CallInst* callTwo = Builder.CreateCall(...);
2023 Value* result = Builder.CreateMul(callOne, callTwo);
2025 See :doc:`tutorial/LangImpl3` for a practical use of the ``IRBuilder``.
2028 .. _schanges_deleting:
2030 Deleting Instructions
2031 ^^^^^^^^^^^^^^^^^^^^^
2033 Deleting an instruction from an existing sequence of instructions that form a
2034 BasicBlock_ is very straight-forward: just call the instruction's
2035 ``eraseFromParent()`` method. For example:
2039 Instruction *I = .. ;
2040 I->eraseFromParent();
2042 This unlinks the instruction from its containing basic block and deletes it. If
2043 you'd just like to unlink the instruction from its containing basic block but
2044 not delete it, you can use the ``removeFromParent()`` method.
2046 .. _schanges_replacing:
2048 Replacing an Instruction with another Value
2049 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2051 Replacing individual instructions
2052 """""""""""""""""""""""""""""""""
2054 Including "`llvm/Transforms/Utils/BasicBlockUtils.h
2055 <http://llvm.org/doxygen/BasicBlockUtils_8h-source.html>`_" permits use of two
2056 very useful replace functions: ``ReplaceInstWithValue`` and
2057 ``ReplaceInstWithInst``.
2059 .. _schanges_deleting_sub:
2061 Deleting Instructions
2062 """""""""""""""""""""
2064 * ``ReplaceInstWithValue``
2066 This function replaces all uses of a given instruction with a value, and then
2067 removes the original instruction. The following example illustrates the
2068 replacement of the result of a particular ``AllocaInst`` that allocates memory
2069 for a single integer with a null pointer to an integer.
2073 AllocaInst* instToReplace = ...;
2074 BasicBlock::iterator ii(instToReplace);
2076 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
2077 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2079 * ``ReplaceInstWithInst``
2081 This function replaces a particular instruction with another instruction,
2082 inserting the new instruction into the basic block at the location where the
2083 old instruction was, and replacing any uses of the old instruction with the
2084 new instruction. The following example illustrates the replacement of one
2085 ``AllocaInst`` with another.
2089 AllocaInst* instToReplace = ...;
2090 BasicBlock::iterator ii(instToReplace);
2092 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
2093 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
2096 Replacing multiple uses of Users and Values
2097 """""""""""""""""""""""""""""""""""""""""""
2099 You can use ``Value::replaceAllUsesWith`` and ``User::replaceUsesOfWith`` to
2100 change more than one use at a time. See the doxygen documentation for the
2101 `Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_ and `User Class
2102 <http://llvm.org/doxygen/classllvm_1_1User.html>`_, respectively, for more
2105 .. _schanges_deletingGV:
2107 Deleting GlobalVariables
2108 ^^^^^^^^^^^^^^^^^^^^^^^^
2110 Deleting a global variable from a module is just as easy as deleting an
2111 Instruction. First, you must have a pointer to the global variable that you
2112 wish to delete. You use this pointer to erase it from its parent, the module.
2117 GlobalVariable *GV = .. ;
2119 GV->eraseFromParent();
2127 In generating IR, you may need some complex types. If you know these types
2128 statically, you can use ``TypeBuilder<...>::get()``, defined in
2129 ``llvm/Support/TypeBuilder.h``, to retrieve them. ``TypeBuilder`` has two forms
2130 depending on whether you're building types for cross-compilation or native
2131 library use. ``TypeBuilder<T, true>`` requires that ``T`` be independent of the
2132 host environment, meaning that it's built out of types from the ``llvm::types``
2133 (`doxygen <http://llvm.org/doxygen/namespacellvm_1_1types.html>`__) namespace
2134 and pointers, functions, arrays, etc. built of those. ``TypeBuilder<T, false>``
2135 additionally allows native C types whose size may depend on the host compiler.
2140 FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get();
2142 is easier to read and write than the equivalent
2146 std::vector<const Type*> params;
2147 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2148 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2150 See the `class comment
2151 <http://llvm.org/doxygen/TypeBuilder_8h-source.html#l00001>`_ for more details.
2158 This section describes the interaction of the LLVM APIs with multithreading,
2159 both on the part of client applications, and in the JIT, in the hosted
2162 Note that LLVM's support for multithreading is still relatively young. Up
2163 through version 2.5, the execution of threaded hosted applications was
2164 supported, but not threaded client access to the APIs. While this use case is
2165 now supported, clients *must* adhere to the guidelines specified below to ensure
2166 proper operation in multithreaded mode.
2168 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2169 intrinsics in order to support threaded operation. If you need a
2170 multhreading-capable LLVM on a platform without a suitably modern system
2171 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2172 using the resultant compiler to build a copy of LLVM with multithreading
2177 Ending Execution with ``llvm_shutdown()``
2178 -----------------------------------------
2180 When you are done using the LLVM APIs, you should call ``llvm_shutdown()`` to
2181 deallocate memory used for internal structures.
2185 Lazy Initialization with ``ManagedStatic``
2186 ------------------------------------------
2188 ``ManagedStatic`` is a utility class in LLVM used to implement static
2189 initialization of static resources, such as the global type tables. In a
2190 single-threaded environment, it implements a simple lazy initialization scheme.
2191 When LLVM is compiled with support for multi-threading, however, it uses
2192 double-checked locking to implement thread-safe lazy initialization.
2196 Achieving Isolation with ``LLVMContext``
2197 ----------------------------------------
2199 ``LLVMContext`` is an opaque class in the LLVM API which clients can use to
2200 operate multiple, isolated instances of LLVM concurrently within the same
2201 address space. For instance, in a hypothetical compile-server, the compilation
2202 of an individual translation unit is conceptually independent from all the
2203 others, and it would be desirable to be able to compile incoming translation
2204 units concurrently on independent server threads. Fortunately, ``LLVMContext``
2205 exists to enable just this kind of scenario!
2207 Conceptually, ``LLVMContext`` provides isolation. Every LLVM entity
2208 (``Module``\ s, ``Value``\ s, ``Type``\ s, ``Constant``\ s, etc.) in LLVM's
2209 in-memory IR belongs to an ``LLVMContext``. Entities in different contexts
2210 *cannot* interact with each other: ``Module``\ s in different contexts cannot be
2211 linked together, ``Function``\ s cannot be added to ``Module``\ s in different
2212 contexts, etc. What this means is that is is safe to compile on multiple
2213 threads simultaneously, as long as no two threads operate on entities within the
2216 In practice, very few places in the API require the explicit specification of a
2217 ``LLVMContext``, other than the ``Type`` creation/lookup APIs. Because every
2218 ``Type`` carries a reference to its owning context, most other entities can
2219 determine what context they belong to by looking at their own ``Type``. If you
2220 are adding new entities to LLVM IR, please try to maintain this interface
2223 For clients that do *not* require the benefits of isolation, LLVM provides a
2224 convenience API ``getGlobalContext()``. This returns a global, lazily
2225 initialized ``LLVMContext`` that may be used in situations where isolation is
2233 LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple
2234 threads can call ``ExecutionEngine::getPointerToFunction()`` or
2235 ``ExecutionEngine::runFunction()`` concurrently, and multiple threads can run
2236 code output by the JIT concurrently. The user must still ensure that only one
2237 thread accesses IR in a given ``LLVMContext`` while another thread might be
2238 modifying it. One way to do that is to always hold the JIT lock while accessing
2239 IR outside the JIT (the JIT *modifies* the IR by adding ``CallbackVH``\ s).
2240 Another way is to only call ``getPointerToFunction()`` from the
2241 ``LLVMContext``'s thread.
2243 When the JIT is configured to compile lazily (using
2244 ``ExecutionEngine::DisableLazyCompilation(false)``), there is currently a `race
2245 condition <http://llvm.org/bugs/show_bug.cgi?id=5184>`_ in updating call sites
2246 after a function is lazily-jitted. It's still possible to use the lazy JIT in a
2247 threaded program if you ensure that only one thread at a time can call any
2248 particular lazy stub and that the JIT lock guards any IR access, but we suggest
2249 using only the eager JIT in threaded programs.
2256 This section describes some of the advanced or obscure API's that most clients
2257 do not need to be aware of. These API's tend manage the inner workings of the
2258 LLVM system, and only need to be accessed in unusual circumstances.
2262 The ``ValueSymbolTable`` class
2263 ------------------------------
2265 The ``ValueSymbolTable`` (`doxygen
2266 <http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html>`__) class provides
2267 a symbol table that the :ref:`Function <c_Function>` and Module_ classes use for
2268 naming value definitions. The symbol table can provide a name for any Value_.
2270 Note that the ``SymbolTable`` class should not be directly accessed by most
2271 clients. It should only be used when iteration over the symbol table names
2272 themselves are required, which is very special purpose. Note that not all LLVM
2273 Value_\ s have names, and those without names (i.e. they have an empty name) do
2274 not exist in the symbol table.
2276 Symbol tables support iteration over the values in the symbol table with
2277 ``begin/end/iterator`` and supports querying to see if a specific name is in the
2278 symbol table (with ``lookup``). The ``ValueSymbolTable`` class exposes no
2279 public mutator methods, instead, simply call ``setName`` on a value, which will
2280 autoinsert it into the appropriate symbol table.
2284 The ``User`` and owned ``Use`` classes' memory layout
2285 -----------------------------------------------------
2287 The ``User`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1User.html>`__)
2288 class provides a basis for expressing the ownership of ``User`` towards other
2289 `Value instance <http://llvm.org/doxygen/classllvm_1_1Value.html>`_\ s. The
2290 ``Use`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1Use.html>`__) helper
2291 class is employed to do the bookkeeping and to facilitate *O(1)* addition and
2296 Interaction and relationship between ``User`` and ``Use`` objects
2297 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2299 A subclass of ``User`` can choose between incorporating its ``Use`` objects or
2300 refer to them out-of-line by means of a pointer. A mixed variant (some ``Use``
2301 s inline others hung off) is impractical and breaks the invariant that the
2302 ``Use`` objects belonging to the same ``User`` form a contiguous array.
2304 We have 2 different layouts in the ``User`` (sub)classes:
2308 The ``Use`` object(s) are inside (resp. at fixed offset) of the ``User``
2309 object and there are a fixed number of them.
2313 The ``Use`` object(s) are referenced by a pointer to an array from the
2314 ``User`` object and there may be a variable number of them.
2316 As of v2.4 each layout still possesses a direct pointer to the start of the
2317 array of ``Use``\ s. Though not mandatory for layout a), we stick to this
2318 redundancy for the sake of simplicity. The ``User`` object also stores the
2319 number of ``Use`` objects it has. (Theoretically this information can also be
2320 calculated given the scheme presented below.)
2322 Special forms of allocation operators (``operator new``) enforce the following
2325 * Layout a) is modelled by prepending the ``User`` object by the ``Use[]``
2328 .. code-block:: none
2330 ...---.---.---.---.-------...
2331 | P | P | P | P | User
2332 '''---'---'---'---'-------'''
2334 * Layout b) is modelled by pointing at the ``Use[]`` array.
2336 .. code-block:: none
2347 *(In the above figures* '``P``' *stands for the* ``Use**`` *that is stored in
2348 each* ``Use`` *object in the member* ``Use::Prev`` *)*
2352 The waymarking algorithm
2353 ^^^^^^^^^^^^^^^^^^^^^^^^
2355 Since the ``Use`` objects are deprived of the direct (back)pointer to their
2356 ``User`` objects, there must be a fast and exact method to recover it. This is
2357 accomplished by the following scheme:
2359 A bit-encoding in the 2 LSBits (least significant bits) of the ``Use::Prev``
2360 allows to find the start of the ``User`` object:
2362 * ``00`` --- binary digit 0
2364 * ``01`` --- binary digit 1
2366 * ``10`` --- stop and calculate (``s``)
2368 * ``11`` --- full stop (``S``)
2370 Given a ``Use*``, all we have to do is to walk till we get a stop and we either
2371 have a ``User`` immediately behind or we have to walk to the next stop picking
2372 up digits and calculating the offset:
2374 .. code-block:: none
2376 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2377 | 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
2378 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2379 |+15 |+10 |+6 |+3 |+1
2382 | | | ______________________>
2383 | | ______________________________________>
2384 | __________________________________________________________>
2386 Only the significant number of bits need to be stored between the stops, so that
2387 the *worst case is 20 memory accesses* when there are 1000 ``Use`` objects
2388 associated with a ``User``.
2392 Reference implementation
2393 ^^^^^^^^^^^^^^^^^^^^^^^^
2395 The following literate Haskell fragment demonstrates the concept:
2397 .. code-block:: haskell
2399 > import Test.QuickCheck
2401 > digits :: Int -> [Char] -> [Char]
2402 > digits 0 acc = '0' : acc
2403 > digits 1 acc = '1' : acc
2404 > digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
2406 > dist :: Int -> [Char] -> [Char]
2409 > dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
2410 > dist n acc = dist (n - 1) $ dist 1 acc
2412 > takeLast n ss = reverse $ take n $ reverse ss
2414 > test = takeLast 40 $ dist 20 []
2417 Printing <test> gives: ``"1s100000s11010s10100s1111s1010s110s11s1S"``
2419 The reverse algorithm computes the length of the string just by examining a
2422 .. code-block:: haskell
2424 > pref :: [Char] -> Int
2426 > pref ('s':'1':rest) = decode 2 1 rest
2427 > pref (_:rest) = 1 + pref rest
2429 > decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
2430 > decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
2431 > decode walk acc _ = walk + acc
2434 Now, as expected, printing <pref test> gives ``40``.
2436 We can *quickCheck* this with following property:
2438 .. code-block:: haskell
2440 > testcase = dist 2000 []
2441 > testcaseLength = length testcase
2443 > identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
2444 > where arr = takeLast n testcase
2447 As expected <quickCheck identityProp> gives:
2451 *Main> quickCheck identityProp
2452 OK, passed 100 tests.
2454 Let's be a bit more exhaustive:
2456 .. code-block:: haskell
2459 > deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
2462 And here is the result of <deepCheck identityProp>:
2466 *Main> deepCheck identityProp
2467 OK, passed 500 tests.
2471 Tagging considerations
2472 ^^^^^^^^^^^^^^^^^^^^^^
2474 To maintain the invariant that the 2 LSBits of each ``Use**`` in ``Use`` never
2475 change after being set up, setters of ``Use::Prev`` must re-tag the new
2476 ``Use**`` on every modification. Accordingly getters must strip the tag bits.
2478 For layout b) instead of the ``User`` we find a pointer (``User*`` with LSBit
2479 set). Following this pointer brings us to the ``User``. A portable trick
2480 ensures that the first bytes of ``User`` (if interpreted as a pointer) never has
2481 the LSBit set. (Portability is relying on the fact that all known compilers
2482 place the ``vptr`` in the first word of the instances.)
2486 The Core LLVM Class Hierarchy Reference
2487 =======================================
2489 ``#include "llvm/IR/Type.h"``
2491 header source: `Type.h <http://llvm.org/doxygen/Type_8h-source.html>`_
2493 doxygen info: `Type Clases <http://llvm.org/doxygen/classllvm_1_1Type.html>`_
2495 The Core LLVM classes are the primary means of representing the program being
2496 inspected or transformed. The core LLVM classes are defined in header files in
2497 the ``include/llvm/`` directory, and implemented in the ``lib/VMCore``
2502 The Type class and Derived Types
2503 --------------------------------
2505 ``Type`` is a superclass of all type classes. Every ``Value`` has a ``Type``.
2506 ``Type`` cannot be instantiated directly but only through its subclasses.
2507 Certain primitive types (``VoidType``, ``LabelType``, ``FloatType`` and
2508 ``DoubleType``) have hidden subclasses. They are hidden because they offer no
2509 useful functionality beyond what the ``Type`` class offers except to distinguish
2510 themselves from other subclasses of ``Type``.
2512 All other types are subclasses of ``DerivedType``. Types can be named, but this
2513 is not a requirement. There exists exactly one instance of a given shape at any
2514 one time. This allows type equality to be performed with address equality of
2515 the Type Instance. That is, given two ``Type*`` values, the types are identical
2516 if the pointers are identical.
2520 Important Public Methods
2521 ^^^^^^^^^^^^^^^^^^^^^^^^
2523 * ``bool isIntegerTy() const``: Returns true for any integer type.
2525 * ``bool isFloatingPointTy()``: Return true if this is one of the five
2526 floating point types.
2528 * ``bool isSized()``: Return true if the type has known size. Things
2529 that don't have a size are abstract types, labels and void.
2533 Important Derived Types
2534 ^^^^^^^^^^^^^^^^^^^^^^^
2537 Subclass of DerivedType that represents integer types of any bit width. Any
2538 bit width between ``IntegerType::MIN_INT_BITS`` (1) and
2539 ``IntegerType::MAX_INT_BITS`` (~8 million) can be represented.
2541 * ``static const IntegerType* get(unsigned NumBits)``: get an integer
2542 type of a specific bit width.
2544 * ``unsigned getBitWidth() const``: Get the bit width of an integer type.
2547 This is subclassed by ArrayType, PointerType and VectorType.
2549 * ``const Type * getElementType() const``: Returns the type of each
2550 of the elements in the sequential type.
2553 This is a subclass of SequentialType and defines the interface for array
2556 * ``unsigned getNumElements() const``: Returns the number of elements
2560 Subclass of SequentialType for pointer types.
2563 Subclass of SequentialType for vector types. A vector type is similar to an
2564 ArrayType but is distinguished because it is a first class type whereas
2565 ArrayType is not. Vector types are used for vector operations and are usually
2566 small vectors of of an integer or floating point type.
2569 Subclass of DerivedTypes for struct types.
2574 Subclass of DerivedTypes for function types.
2576 * ``bool isVarArg() const``: Returns true if it's a vararg function.
2578 * ``const Type * getReturnType() const``: Returns the return type of the
2581 * ``const Type * getParamType (unsigned i)``: Returns the type of the ith
2584 * ``const unsigned getNumParams() const``: Returns the number of formal
2589 The ``Module`` class
2590 --------------------
2592 ``#include "llvm/IR/Module.h"``
2594 header source: `Module.h <http://llvm.org/doxygen/Module_8h-source.html>`_
2596 doxygen info: `Module Class <http://llvm.org/doxygen/classllvm_1_1Module.html>`_
2598 The ``Module`` class represents the top level structure present in LLVM
2599 programs. An LLVM module is effectively either a translation unit of the
2600 original program or a combination of several translation units merged by the
2601 linker. The ``Module`` class keeps track of a list of :ref:`Function
2602 <c_Function>`\ s, a list of GlobalVariable_\ s, and a SymbolTable_.
2603 Additionally, it contains a few helpful member functions that try to make common
2608 Important Public Members of the ``Module`` class
2609 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2611 * ``Module::Module(std::string name = "")``
2613 Constructing a Module_ is easy. You can optionally provide a name for it
2614 (probably based on the name of the translation unit).
2616 * | ``Module::iterator`` - Typedef for function list iterator
2617 | ``Module::const_iterator`` - Typedef for const_iterator.
2618 | ``begin()``, ``end()``, ``size()``, ``empty()``
2620 These are forwarding methods that make it easy to access the contents of a
2621 ``Module`` object's :ref:`Function <c_Function>` list.
2623 * ``Module::FunctionListType &getFunctionList()``
2625 Returns the list of :ref:`Function <c_Function>`\ s. This is necessary to use
2626 when you need to update the list or perform a complex action that doesn't have
2627 a forwarding method.
2631 * | ``Module::global_iterator`` - Typedef for global variable list iterator
2632 | ``Module::const_global_iterator`` - Typedef for const_iterator.
2633 | ``global_begin()``, ``global_end()``, ``global_size()``, ``global_empty()``
2635 These are forwarding methods that make it easy to access the contents of a
2636 ``Module`` object's GlobalVariable_ list.
2638 * ``Module::GlobalListType &getGlobalList()``
2640 Returns the list of GlobalVariable_\ s. This is necessary to use when you
2641 need to update the list or perform a complex action that doesn't have a
2646 * ``SymbolTable *getSymbolTable()``
2648 Return a reference to the SymbolTable_ for this ``Module``.
2652 * ``Function *getFunction(StringRef Name) const``
2654 Look up the specified function in the ``Module`` SymbolTable_. If it does not
2655 exist, return ``null``.
2657 * ``Function *getOrInsertFunction(const std::string &Name, const FunctionType
2660 Look up the specified function in the ``Module`` SymbolTable_. If it does not
2661 exist, add an external declaration for the function and return it.
2663 * ``std::string getTypeName(const Type *Ty)``
2665 If there is at least one entry in the SymbolTable_ for the specified Type_,
2666 return it. Otherwise return the empty string.
2668 * ``bool addTypeName(const std::string &Name, const Type *Ty)``
2670 Insert an entry in the SymbolTable_ mapping ``Name`` to ``Ty``. If there is
2671 already an entry for this name, true is returned and the SymbolTable_ is not
2679 ``#include "llvm/IR/Value.h"``
2681 header source: `Value.h <http://llvm.org/doxygen/Value_8h-source.html>`_
2683 doxygen info: `Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_
2685 The ``Value`` class is the most important class in the LLVM Source base. It
2686 represents a typed value that may be used (among other things) as an operand to
2687 an instruction. There are many different types of ``Value``\ s, such as
2688 Constant_\ s, Argument_\ s. Even Instruction_\ s and :ref:`Function
2689 <c_Function>`\ s are ``Value``\ s.
2691 A particular ``Value`` may be used many times in the LLVM representation for a
2692 program. For example, an incoming argument to a function (represented with an
2693 instance of the Argument_ class) is "used" by every instruction in the function
2694 that references the argument. To keep track of this relationship, the ``Value``
2695 class keeps a list of all of the ``User``\ s that is using it (the User_ class
2696 is a base class for all nodes in the LLVM graph that can refer to ``Value``\ s).
2697 This use list is how LLVM represents def-use information in the program, and is
2698 accessible through the ``use_*`` methods, shown below.
2700 Because LLVM is a typed representation, every LLVM ``Value`` is typed, and this
2701 Type_ is available through the ``getType()`` method. In addition, all LLVM
2702 values can be named. The "name" of the ``Value`` is a symbolic string printed
2705 .. code-block:: llvm
2711 The name of this instruction is "foo". **NOTE** that the name of any value may
2712 be missing (an empty string), so names should **ONLY** be used for debugging
2713 (making the source code easier to read, debugging printouts), they should not be
2714 used to keep track of values or map between them. For this purpose, use a
2715 ``std::map`` of pointers to the ``Value`` itself instead.
2717 One important aspect of LLVM is that there is no distinction between an SSA
2718 variable and the operation that produces it. Because of this, any reference to
2719 the value produced by an instruction (or the value available as an incoming
2720 argument, for example) is represented as a direct pointer to the instance of the
2721 class that represents this value. Although this may take some getting used to,
2722 it simplifies the representation and makes it easier to manipulate.
2726 Important Public Members of the ``Value`` class
2727 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2729 * | ``Value::use_iterator`` - Typedef for iterator over the use-list
2730 | ``Value::const_use_iterator`` - Typedef for const_iterator over the
2732 | ``unsigned use_size()`` - Returns the number of users of the value.
2733 | ``bool use_empty()`` - Returns true if there are no users.
2734 | ``use_iterator use_begin()`` - Get an iterator to the start of the
2736 | ``use_iterator use_end()`` - Get an iterator to the end of the use-list.
2737 | ``User *use_back()`` - Returns the last element in the list.
2739 These methods are the interface to access the def-use information in LLVM.
2740 As with all other iterators in LLVM, the naming conventions follow the
2741 conventions defined by the STL_.
2743 * ``Type *getType() const``
2744 This method returns the Type of the Value.
2746 * | ``bool hasName() const``
2747 | ``std::string getName() const``
2748 | ``void setName(const std::string &Name)``
2750 This family of methods is used to access and assign a name to a ``Value``, be
2751 aware of the :ref:`precaution above <nameWarning>`.
2753 * ``void replaceAllUsesWith(Value *V)``
2755 This method traverses the use list of a ``Value`` changing all User_\ s of the
2756 current value to refer to "``V``" instead. For example, if you detect that an
2757 instruction always produces a constant value (for example through constant
2758 folding), you can replace all uses of the instruction with the constant like
2763 Inst->replaceAllUsesWith(ConstVal);
2770 ``#include "llvm/IR/User.h"``
2772 header source: `User.h <http://llvm.org/doxygen/User_8h-source.html>`_
2774 doxygen info: `User Class <http://llvm.org/doxygen/classllvm_1_1User.html>`_
2778 The ``User`` class is the common base class of all LLVM nodes that may refer to
2779 ``Value``\ s. It exposes a list of "Operands" that are all of the ``Value``\ s
2780 that the User is referring to. The ``User`` class itself is a subclass of
2783 The operands of a ``User`` point directly to the LLVM ``Value`` that it refers
2784 to. Because LLVM uses Static Single Assignment (SSA) form, there can only be
2785 one definition referred to, allowing this direct connection. This connection
2786 provides the use-def information in LLVM.
2790 Important Public Members of the ``User`` class
2791 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2793 The ``User`` class exposes the operand list in two ways: through an index access
2794 interface and through an iterator based interface.
2796 * | ``Value *getOperand(unsigned i)``
2797 | ``unsigned getNumOperands()``
2799 These two methods expose the operands of the ``User`` in a convenient form for
2802 * | ``User::op_iterator`` - Typedef for iterator over the operand list
2803 | ``op_iterator op_begin()`` - Get an iterator to the start of the operand
2805 | ``op_iterator op_end()`` - Get an iterator to the end of the operand list.
2807 Together, these methods make up the iterator based interface to the operands
2813 The ``Instruction`` class
2814 -------------------------
2816 ``#include "llvm/IR/Instruction.h"``
2818 header source: `Instruction.h
2819 <http://llvm.org/doxygen/Instruction_8h-source.html>`_
2821 doxygen info: `Instruction Class
2822 <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_
2824 Superclasses: User_, Value_
2826 The ``Instruction`` class is the common base class for all LLVM instructions.
2827 It provides only a few methods, but is a very commonly used class. The primary
2828 data tracked by the ``Instruction`` class itself is the opcode (instruction
2829 type) and the parent BasicBlock_ the ``Instruction`` is embedded into. To
2830 represent a specific type of instruction, one of many subclasses of
2831 ``Instruction`` are used.
2833 Because the ``Instruction`` class subclasses the User_ class, its operands can
2834 be accessed in the same way as for other ``User``\ s (with the
2835 ``getOperand()``/``getNumOperands()`` and ``op_begin()``/``op_end()`` methods).
2836 An important file for the ``Instruction`` class is the ``llvm/Instruction.def``
2837 file. This file contains some meta-data about the various different types of
2838 instructions in LLVM. It describes the enum values that are used as opcodes
2839 (for example ``Instruction::Add`` and ``Instruction::ICmp``), as well as the
2840 concrete sub-classes of ``Instruction`` that implement the instruction (for
2841 example BinaryOperator_ and CmpInst_). Unfortunately, the use of macros in this
2842 file confuses doxygen, so these enum values don't show up correctly in the
2843 `doxygen output <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_.
2847 Important Subclasses of the ``Instruction`` class
2848 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2852 * ``BinaryOperator``
2854 This subclasses represents all two operand instructions whose operands must be
2855 the same type, except for the comparison instructions.
2860 This subclass is the parent of the 12 casting instructions. It provides
2861 common operations on cast instructions.
2867 This subclass respresents the two comparison instructions,
2868 `ICmpInst <LangRef.html#i_icmp>`_ (integer opreands), and
2869 `FCmpInst <LangRef.html#i_fcmp>`_ (floating point operands).
2873 * ``TerminatorInst``
2875 This subclass is the parent of all terminator instructions (those which can
2880 Important Public Members of the ``Instruction`` class
2881 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2883 * ``BasicBlock *getParent()``
2885 Returns the BasicBlock_ that this
2886 ``Instruction`` is embedded into.
2888 * ``bool mayWriteToMemory()``
2890 Returns true if the instruction writes to memory, i.e. it is a ``call``,
2891 ``free``, ``invoke``, or ``store``.
2893 * ``unsigned getOpcode()``
2895 Returns the opcode for the ``Instruction``.
2897 * ``Instruction *clone() const``
2899 Returns another instance of the specified instruction, identical in all ways
2900 to the original except that the instruction has no parent (i.e. it's not
2901 embedded into a BasicBlock_), and it has no name.
2905 The ``Constant`` class and subclasses
2906 -------------------------------------
2908 Constant represents a base class for different types of constants. It is
2909 subclassed by ConstantInt, ConstantArray, etc. for representing the various
2910 types of Constants. GlobalValue_ is also a subclass, which represents the
2911 address of a global variable or function.
2915 Important Subclasses of Constant
2916 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2918 * ConstantInt : This subclass of Constant represents an integer constant of
2921 * ``const APInt& getValue() const``: Returns the underlying
2922 value of this constant, an APInt value.
2924 * ``int64_t getSExtValue() const``: Converts the underlying APInt value to an
2925 int64_t via sign extension. If the value (not the bit width) of the APInt
2926 is too large to fit in an int64_t, an assertion will result. For this
2927 reason, use of this method is discouraged.
2929 * ``uint64_t getZExtValue() const``: Converts the underlying APInt value
2930 to a uint64_t via zero extension. IF the value (not the bit width) of the
2931 APInt is too large to fit in a uint64_t, an assertion will result. For this
2932 reason, use of this method is discouraged.
2934 * ``static ConstantInt* get(const APInt& Val)``: Returns the ConstantInt
2935 object that represents the value provided by ``Val``. The type is implied
2936 as the IntegerType that corresponds to the bit width of ``Val``.
2938 * ``static ConstantInt* get(const Type *Ty, uint64_t Val)``: Returns the
2939 ConstantInt object that represents the value provided by ``Val`` for integer
2942 * ConstantFP : This class represents a floating point constant.
2944 * ``double getValue() const``: Returns the underlying value of this constant.
2946 * ConstantArray : This represents a constant array.
2948 * ``const std::vector<Use> &getValues() const``: Returns a vector of
2949 component constants that makeup this array.
2951 * ConstantStruct : This represents a constant struct.
2953 * ``const std::vector<Use> &getValues() const``: Returns a vector of
2954 component constants that makeup this array.
2956 * GlobalValue : This represents either a global variable or a function. In
2957 either case, the value is a constant fixed address (after linking).
2961 The ``GlobalValue`` class
2962 -------------------------
2964 ``#include "llvm/IR/GlobalValue.h"``
2966 header source: `GlobalValue.h
2967 <http://llvm.org/doxygen/GlobalValue_8h-source.html>`_
2969 doxygen info: `GlobalValue Class
2970 <http://llvm.org/doxygen/classllvm_1_1GlobalValue.html>`_
2972 Superclasses: Constant_, User_, Value_
2974 Global values ( GlobalVariable_\ s or :ref:`Function <c_Function>`\ s) are the
2975 only LLVM values that are visible in the bodies of all :ref:`Function
2976 <c_Function>`\ s. Because they are visible at global scope, they are also
2977 subject to linking with other globals defined in different translation units.
2978 To control the linking process, ``GlobalValue``\ s know their linkage rules.
2979 Specifically, ``GlobalValue``\ s know whether they have internal or external
2980 linkage, as defined by the ``LinkageTypes`` enumeration.
2982 If a ``GlobalValue`` has internal linkage (equivalent to being ``static`` in C),
2983 it is not visible to code outside the current translation unit, and does not
2984 participate in linking. If it has external linkage, it is visible to external
2985 code, and does participate in linking. In addition to linkage information,
2986 ``GlobalValue``\ s keep track of which Module_ they are currently part of.
2988 Because ``GlobalValue``\ s are memory objects, they are always referred to by
2989 their **address**. As such, the Type_ of a global is always a pointer to its
2990 contents. It is important to remember this when using the ``GetElementPtrInst``
2991 instruction because this pointer must be dereferenced first. For example, if
2992 you have a ``GlobalVariable`` (a subclass of ``GlobalValue)`` that is an array
2993 of 24 ints, type ``[24 x i32]``, then the ``GlobalVariable`` is a pointer to
2994 that array. Although the address of the first element of this array and the
2995 value of the ``GlobalVariable`` are the same, they have different types. The
2996 ``GlobalVariable``'s type is ``[24 x i32]``. The first element's type is
2997 ``i32.`` Because of this, accessing a global value requires you to dereference
2998 the pointer with ``GetElementPtrInst`` first, then its elements can be accessed.
2999 This is explained in the `LLVM Language Reference Manual
3000 <LangRef.html#globalvars>`_.
3004 Important Public Members of the ``GlobalValue`` class
3005 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3007 * | ``bool hasInternalLinkage() const``
3008 | ``bool hasExternalLinkage() const``
3009 | ``void setInternalLinkage(bool HasInternalLinkage)``
3011 These methods manipulate the linkage characteristics of the ``GlobalValue``.
3013 * ``Module *getParent()``
3015 This returns the Module_ that the
3016 GlobalValue is currently embedded into.
3020 The ``Function`` class
3021 ----------------------
3023 ``#include "llvm/IR/Function.h"``
3025 header source: `Function.h <http://llvm.org/doxygen/Function_8h-source.html>`_
3027 doxygen info: `Function Class
3028 <http://llvm.org/doxygen/classllvm_1_1Function.html>`_
3030 Superclasses: GlobalValue_, Constant_, User_, Value_
3032 The ``Function`` class represents a single procedure in LLVM. It is actually
3033 one of the more complex classes in the LLVM hierarchy because it must keep track
3034 of a large amount of data. The ``Function`` class keeps track of a list of
3035 BasicBlock_\ s, a list of formal Argument_\ s, and a SymbolTable_.
3037 The list of BasicBlock_\ s is the most commonly used part of ``Function``
3038 objects. The list imposes an implicit ordering of the blocks in the function,
3039 which indicate how the code will be laid out by the backend. Additionally, the
3040 first BasicBlock_ is the implicit entry node for the ``Function``. It is not
3041 legal in LLVM to explicitly branch to this initial block. There are no implicit
3042 exit nodes, and in fact there may be multiple exit nodes from a single
3043 ``Function``. If the BasicBlock_ list is empty, this indicates that the
3044 ``Function`` is actually a function declaration: the actual body of the function
3045 hasn't been linked in yet.
3047 In addition to a list of BasicBlock_\ s, the ``Function`` class also keeps track
3048 of the list of formal Argument_\ s that the function receives. This container
3049 manages the lifetime of the Argument_ nodes, just like the BasicBlock_ list does
3050 for the BasicBlock_\ s.
3052 The SymbolTable_ is a very rarely used LLVM feature that is only used when you
3053 have to look up a value by name. Aside from that, the SymbolTable_ is used
3054 internally to make sure that there are not conflicts between the names of
3055 Instruction_\ s, BasicBlock_\ s, or Argument_\ s in the function body.
3057 Note that ``Function`` is a GlobalValue_ and therefore also a Constant_. The
3058 value of the function is its address (after linking) which is guaranteed to be
3063 Important Public Members of the ``Function``
3064 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3066 * ``Function(const FunctionType *Ty, LinkageTypes Linkage,
3067 const std::string &N = "", Module* Parent = 0)``
3069 Constructor used when you need to create new ``Function``\ s to add the
3070 program. The constructor must specify the type of the function to create and
3071 what type of linkage the function should have. The FunctionType_ argument
3072 specifies the formal arguments and return value for the function. The same
3073 FunctionType_ value can be used to create multiple functions. The ``Parent``
3074 argument specifies the Module in which the function is defined. If this
3075 argument is provided, the function will automatically be inserted into that
3076 module's list of functions.
3078 * ``bool isDeclaration()``
3080 Return whether or not the ``Function`` has a body defined. If the function is
3081 "external", it does not have a body, and thus must be resolved by linking with
3082 a function defined in a different translation unit.
3084 * | ``Function::iterator`` - Typedef for basic block list iterator
3085 | ``Function::const_iterator`` - Typedef for const_iterator.
3086 | ``begin()``, ``end()``, ``size()``, ``empty()``
3088 These are forwarding methods that make it easy to access the contents of a
3089 ``Function`` object's BasicBlock_ list.
3091 * ``Function::BasicBlockListType &getBasicBlockList()``
3093 Returns the list of BasicBlock_\ s. This is necessary to use when you need to
3094 update the list or perform a complex action that doesn't have a forwarding
3097 * | ``Function::arg_iterator`` - Typedef for the argument list iterator
3098 | ``Function::const_arg_iterator`` - Typedef for const_iterator.
3099 | ``arg_begin()``, ``arg_end()``, ``arg_size()``, ``arg_empty()``
3101 These are forwarding methods that make it easy to access the contents of a
3102 ``Function`` object's Argument_ list.
3104 * ``Function::ArgumentListType &getArgumentList()``
3106 Returns the list of Argument_. This is necessary to use when you need to
3107 update the list or perform a complex action that doesn't have a forwarding
3110 * ``BasicBlock &getEntryBlock()``
3112 Returns the entry ``BasicBlock`` for the function. Because the entry block
3113 for the function is always the first block, this returns the first block of
3116 * | ``Type *getReturnType()``
3117 | ``FunctionType *getFunctionType()``
3119 This traverses the Type_ of the ``Function`` and returns the return type of
3120 the function, or the FunctionType_ of the actual function.
3122 * ``SymbolTable *getSymbolTable()``
3124 Return a pointer to the SymbolTable_ for this ``Function``.
3128 The ``GlobalVariable`` class
3129 ----------------------------
3131 ``#include "llvm/IR/GlobalVariable.h"``
3133 header source: `GlobalVariable.h
3134 <http://llvm.org/doxygen/GlobalVariable_8h-source.html>`_
3136 doxygen info: `GlobalVariable Class
3137 <http://llvm.org/doxygen/classllvm_1_1GlobalVariable.html>`_
3139 Superclasses: GlobalValue_, Constant_, User_, Value_
3141 Global variables are represented with the (surprise surprise) ``GlobalVariable``
3142 class. Like functions, ``GlobalVariable``\ s are also subclasses of
3143 GlobalValue_, and as such are always referenced by their address (global values
3144 must live in memory, so their "name" refers to their constant address). See
3145 GlobalValue_ for more on this. Global variables may have an initial value
3146 (which must be a Constant_), and if they have an initializer, they may be marked
3147 as "constant" themselves (indicating that their contents never change at
3150 .. _m_GlobalVariable:
3152 Important Public Members of the ``GlobalVariable`` class
3153 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3155 * ``GlobalVariable(const Type *Ty, bool isConstant, LinkageTypes &Linkage,
3156 Constant *Initializer = 0, const std::string &Name = "", Module* Parent = 0)``
3158 Create a new global variable of the specified type. If ``isConstant`` is true
3159 then the global variable will be marked as unchanging for the program. The
3160 Linkage parameter specifies the type of linkage (internal, external, weak,
3161 linkonce, appending) for the variable. If the linkage is InternalLinkage,
3162 WeakAnyLinkage, WeakODRLinkage, LinkOnceAnyLinkage or LinkOnceODRLinkage, then
3163 the resultant global variable will have internal linkage. AppendingLinkage
3164 concatenates together all instances (in different translation units) of the
3165 variable into a single variable but is only applicable to arrays. See the
3166 `LLVM Language Reference <LangRef.html#modulestructure>`_ for further details
3167 on linkage types. Optionally an initializer, a name, and the module to put
3168 the variable into may be specified for the global variable as well.
3170 * ``bool isConstant() const``
3172 Returns true if this is a global variable that is known not to be modified at
3175 * ``bool hasInitializer()``
3177 Returns true if this ``GlobalVariable`` has an intializer.
3179 * ``Constant *getInitializer()``
3181 Returns the initial value for a ``GlobalVariable``. It is not legal to call
3182 this method if there is no initializer.
3186 The ``BasicBlock`` class
3187 ------------------------
3189 ``#include "llvm/IR/BasicBlock.h"``
3191 header source: `BasicBlock.h
3192 <http://llvm.org/doxygen/BasicBlock_8h-source.html>`_
3194 doxygen info: `BasicBlock Class
3195 <http://llvm.org/doxygen/classllvm_1_1BasicBlock.html>`_
3199 This class represents a single entry single exit section of the code, commonly
3200 known as a basic block by the compiler community. The ``BasicBlock`` class
3201 maintains a list of Instruction_\ s, which form the body of the block. Matching
3202 the language definition, the last element of this list of instructions is always
3203 a terminator instruction (a subclass of the TerminatorInst_ class).
3205 In addition to tracking the list of instructions that make up the block, the
3206 ``BasicBlock`` class also keeps track of the :ref:`Function <c_Function>` that
3207 it is embedded into.
3209 Note that ``BasicBlock``\ s themselves are Value_\ s, because they are
3210 referenced by instructions like branches and can go in the switch tables.
3211 ``BasicBlock``\ s have type ``label``.
3215 Important Public Members of the ``BasicBlock`` class
3216 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3218 * ``BasicBlock(const std::string &Name = "", Function *Parent = 0)``
3220 The ``BasicBlock`` constructor is used to create new basic blocks for
3221 insertion into a function. The constructor optionally takes a name for the
3222 new block, and a :ref:`Function <c_Function>` to insert it into. If the
3223 ``Parent`` parameter is specified, the new ``BasicBlock`` is automatically
3224 inserted at the end of the specified :ref:`Function <c_Function>`, if not
3225 specified, the BasicBlock must be manually inserted into the :ref:`Function
3228 * | ``BasicBlock::iterator`` - Typedef for instruction list iterator
3229 | ``BasicBlock::const_iterator`` - Typedef for const_iterator.
3230 | ``begin()``, ``end()``, ``front()``, ``back()``,
3231 ``size()``, ``empty()``
3232 STL-style functions for accessing the instruction list.
3234 These methods and typedefs are forwarding functions that have the same
3235 semantics as the standard library methods of the same names. These methods
3236 expose the underlying instruction list of a basic block in a way that is easy
3237 to manipulate. To get the full complement of container operations (including
3238 operations to update the list), you must use the ``getInstList()`` method.
3240 * ``BasicBlock::InstListType &getInstList()``
3242 This method is used to get access to the underlying container that actually
3243 holds the Instructions. This method must be used when there isn't a
3244 forwarding function in the ``BasicBlock`` class for the operation that you
3245 would like to perform. Because there are no forwarding functions for
3246 "updating" operations, you need to use this if you want to update the contents
3247 of a ``BasicBlock``.
3249 * ``Function *getParent()``
3251 Returns a pointer to :ref:`Function <c_Function>` the block is embedded into,
3252 or a null pointer if it is homeless.
3254 * ``TerminatorInst *getTerminator()``
3256 Returns a pointer to the terminator instruction that appears at the end of the
3257 ``BasicBlock``. If there is no terminator instruction, or if the last
3258 instruction in the block is not a terminator, then a null pointer is returned.
3262 The ``Argument`` class
3263 ----------------------
3265 This subclass of Value defines the interface for incoming formal arguments to a
3266 function. A Function maintains a list of its formal arguments. An argument has
3267 a pointer to the parent Function.