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 The ``DEBUG()`` macro and ``-debug`` option
269 -------------------------------------------
271 Often when working on your pass you will put a bunch of debugging printouts and
272 other code into your pass. After you get it working, you want to remove it, but
273 you may need it again in the future (to work out new bugs that you run across).
275 Naturally, because of this, you don't want to delete the debug printouts, but
276 you don't want them to always be noisy. A standard compromise is to comment
277 them out, allowing you to enable them if you need them in the future.
279 The ``llvm/Support/Debug.h`` (`doxygen
280 <http://llvm.org/doxygen/Debug_8h-source.html>`__) file provides a macro named
281 ``DEBUG()`` that is a much nicer solution to this problem. Basically, you can
282 put arbitrary code into the argument of the ``DEBUG`` macro, and it is only
283 executed if '``opt``' (or any other tool) is run with the '``-debug``' command
288 DEBUG(errs() << "I am here!\n");
290 Then you can run your pass like this:
294 $ opt < a.bc > /dev/null -mypass
296 $ opt < a.bc > /dev/null -mypass -debug
299 Using the ``DEBUG()`` macro instead of a home-brewed solution allows you to not
300 have to create "yet another" command line option for the debug output for your
301 pass. Note that ``DEBUG()`` macros are disabled for optimized builds, so they
302 do not cause a performance impact at all (for the same reason, they should also
303 not contain side-effects!).
305 One additional nice thing about the ``DEBUG()`` macro is that you can enable or
306 disable it directly in gdb. Just use "``set DebugFlag=0``" or "``set
307 DebugFlag=1``" from the gdb if the program is running. If the program hasn't
308 been started yet, you can always just run it with ``-debug``.
312 Fine grained debug info with ``DEBUG_TYPE`` and the ``-debug-only`` option
313 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
315 Sometimes you may find yourself in a situation where enabling ``-debug`` just
316 turns on **too much** information (such as when working on the code generator).
317 If you want to enable debug information with more fine-grained control, you
318 define the ``DEBUG_TYPE`` macro and the ``-debug`` only option as follows:
323 DEBUG(errs() << "No debug type\n");
324 #define DEBUG_TYPE "foo"
325 DEBUG(errs() << "'foo' debug type\n");
327 #define DEBUG_TYPE "bar"
328 DEBUG(errs() << "'bar' debug type\n"));
330 #define DEBUG_TYPE ""
331 DEBUG(errs() << "No debug type (2)\n");
333 Then you can run your pass like this:
337 $ opt < a.bc > /dev/null -mypass
339 $ opt < a.bc > /dev/null -mypass -debug
344 $ opt < a.bc > /dev/null -mypass -debug-only=foo
346 $ opt < a.bc > /dev/null -mypass -debug-only=bar
349 Of course, in practice, you should only set ``DEBUG_TYPE`` at the top of a file,
350 to specify the debug type for the entire module (if you do this before you
351 ``#include "llvm/Support/Debug.h"``, you don't have to insert the ugly
352 ``#undef``'s). Also, you should use names more meaningful than "foo" and "bar",
353 because there is no system in place to ensure that names do not conflict. If
354 two different modules use the same string, they will all be turned on when the
355 name is specified. This allows, for example, all debug information for
356 instruction scheduling to be enabled with ``-debug-type=InstrSched``, even if
357 the source lives in multiple files.
359 The ``DEBUG_WITH_TYPE`` macro is also available for situations where you would
360 like to set ``DEBUG_TYPE``, but only for one specific ``DEBUG`` statement. It
361 takes an additional first parameter, which is the type to use. For example, the
362 preceding example could be written as:
366 DEBUG_WITH_TYPE("", errs() << "No debug type\n");
367 DEBUG_WITH_TYPE("foo", errs() << "'foo' debug type\n");
368 DEBUG_WITH_TYPE("bar", errs() << "'bar' debug type\n"));
369 DEBUG_WITH_TYPE("", errs() << "No debug type (2)\n");
373 The ``Statistic`` class & ``-stats`` option
374 -------------------------------------------
376 The ``llvm/ADT/Statistic.h`` (`doxygen
377 <http://llvm.org/doxygen/Statistic_8h-source.html>`__) file provides a class
378 named ``Statistic`` that is used as a unified way to keep track of what the LLVM
379 compiler is doing and how effective various optimizations are. It is useful to
380 see what optimizations are contributing to making a particular program run
383 Often you may run your pass on some big program, and you're interested to see
384 how many times it makes a certain transformation. Although you can do this with
385 hand inspection, or some ad-hoc method, this is a real pain and not very useful
386 for big programs. Using the ``Statistic`` class makes it very easy to keep
387 track of this information, and the calculated information is presented in a
388 uniform manner with the rest of the passes being executed.
390 There are many examples of ``Statistic`` uses, but the basics of using it are as
393 #. Define your statistic like this:
397 #define DEBUG_TYPE "mypassname" // This goes before any #includes.
398 STATISTIC(NumXForms, "The # of times I did stuff");
400 The ``STATISTIC`` macro defines a static variable, whose name is specified by
401 the first argument. The pass name is taken from the ``DEBUG_TYPE`` macro, and
402 the description is taken from the second argument. The variable defined
403 ("NumXForms" in this case) acts like an unsigned integer.
405 #. Whenever you make a transformation, bump the counter:
409 ++NumXForms; // I did stuff!
411 That's all you have to do. To get '``opt``' to print out the statistics
412 gathered, use the '``-stats``' option:
416 $ opt -stats -mypassname < program.bc > /dev/null
417 ... statistics output ...
419 When running ``opt`` on a C file from the SPEC benchmark suite, it gives a
420 report that looks like this:
424 7646 bitcodewriter - Number of normal instructions
425 725 bitcodewriter - Number of oversized instructions
426 129996 bitcodewriter - Number of bitcode bytes written
427 2817 raise - Number of insts DCEd or constprop'd
428 3213 raise - Number of cast-of-self removed
429 5046 raise - Number of expression trees converted
430 75 raise - Number of other getelementptr's formed
431 138 raise - Number of load/store peepholes
432 42 deadtypeelim - Number of unused typenames removed from symtab
433 392 funcresolve - Number of varargs functions resolved
434 27 globaldce - Number of global variables removed
435 2 adce - Number of basic blocks removed
436 134 cee - Number of branches revectored
437 49 cee - Number of setcc instruction eliminated
438 532 gcse - Number of loads removed
439 2919 gcse - Number of instructions removed
440 86 indvars - Number of canonical indvars added
441 87 indvars - Number of aux indvars removed
442 25 instcombine - Number of dead inst eliminate
443 434 instcombine - Number of insts combined
444 248 licm - Number of load insts hoisted
445 1298 licm - Number of insts hoisted to a loop pre-header
446 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
447 75 mem2reg - Number of alloca's promoted
448 1444 cfgsimplify - Number of blocks simplified
450 Obviously, with so many optimizations, having a unified framework for this stuff
451 is very nice. Making your pass fit well into the framework makes it more
452 maintainable and useful.
456 Viewing graphs while debugging code
457 -----------------------------------
459 Several of the important data structures in LLVM are graphs: for example CFGs
460 made out of LLVM :ref:`BasicBlocks <BasicBlock>`, CFGs made out of LLVM
461 :ref:`MachineBasicBlocks <MachineBasicBlock>`, and :ref:`Instruction Selection
462 DAGs <SelectionDAG>`. In many cases, while debugging various parts of the
463 compiler, it is nice to instantly visualize these graphs.
465 LLVM provides several callbacks that are available in a debug build to do
466 exactly that. If you call the ``Function::viewCFG()`` method, for example, the
467 current LLVM tool will pop up a window containing the CFG for the function where
468 each basic block is a node in the graph, and each node contains the instructions
469 in the block. Similarly, there also exists ``Function::viewCFGOnly()`` (does
470 not include the instructions), the ``MachineFunction::viewCFG()`` and
471 ``MachineFunction::viewCFGOnly()``, and the ``SelectionDAG::viewGraph()``
472 methods. Within GDB, for example, you can usually use something like ``call
473 DAG.viewGraph()`` to pop up a window. Alternatively, you can sprinkle calls to
474 these functions in your code in places you want to debug.
476 Getting this to work requires a small amount of configuration. On Unix systems
477 with X11, install the `graphviz <http://www.graphviz.org>`_ toolkit, and make
478 sure 'dot' and 'gv' are in your path. If you are running on Mac OS/X, download
479 and install the Mac OS/X `Graphviz program
480 <http://www.pixelglow.com/graphviz/>`_ and add
481 ``/Applications/Graphviz.app/Contents/MacOS/`` (or wherever you install it) to
482 your path. Once in your system and path are set up, rerun the LLVM configure
483 script and rebuild LLVM to enable this functionality.
485 ``SelectionDAG`` has been extended to make it easier to locate *interesting*
486 nodes in large complex graphs. From gdb, if you ``call DAG.setGraphColor(node,
487 "color")``, then the next ``call DAG.viewGraph()`` would highlight the node in
488 the specified color (choices of colors can be found at `colors
489 <http://www.graphviz.org/doc/info/colors.html>`_.) More complex node attributes
490 can be provided with ``call DAG.setGraphAttrs(node, "attributes")`` (choices can
491 be found at `Graph attributes <http://www.graphviz.org/doc/info/attrs.html>`_.)
492 If you want to restart and clear all the current graph attributes, then you can
493 ``call DAG.clearGraphAttrs()``.
495 Note that graph visualization features are compiled out of Release builds to
496 reduce file size. This means that you need a Debug+Asserts or Release+Asserts
497 build to use these features.
501 Picking the Right Data Structure for a Task
502 ===========================================
504 LLVM has a plethora of data structures in the ``llvm/ADT/`` directory, and we
505 commonly use STL data structures. This section describes the trade-offs you
506 should consider when you pick one.
508 The first step is a choose your own adventure: do you want a sequential
509 container, a set-like container, or a map-like container? The most important
510 thing when choosing a container is the algorithmic properties of how you plan to
511 access the container. Based on that, you should use:
514 * a :ref:`map-like <ds_map>` container if you need efficient look-up of a
515 value based on another value. Map-like containers also support efficient
516 queries for containment (whether a key is in the map). Map-like containers
517 generally do not support efficient reverse mapping (values to keys). If you
518 need that, use two maps. Some map-like containers also support efficient
519 iteration through the keys in sorted order. Map-like containers are the most
520 expensive sort, only use them if you need one of these capabilities.
522 * a :ref:`set-like <ds_set>` container if you need to put a bunch of stuff into
523 a container that automatically eliminates duplicates. Some set-like
524 containers support efficient iteration through the elements in sorted order.
525 Set-like containers are more expensive than sequential containers.
527 * a :ref:`sequential <ds_sequential>` container provides the most efficient way
528 to add elements and keeps track of the order they are added to the collection.
529 They permit duplicates and support efficient iteration, but do not support
530 efficient look-up based on a key.
532 * a :ref:`string <ds_string>` container is a specialized sequential container or
533 reference structure that is used for character or byte arrays.
535 * a :ref:`bit <ds_bit>` container provides an efficient way to store and
536 perform set operations on sets of numeric id's, while automatically
537 eliminating duplicates. Bit containers require a maximum of 1 bit for each
538 identifier you want to store.
540 Once the proper category of container is determined, you can fine tune the
541 memory use, constant factors, and cache behaviors of access by intelligently
542 picking a member of the category. Note that constant factors and cache behavior
543 can be a big deal. If you have a vector that usually only contains a few
544 elements (but could contain many), for example, it's much better to use
545 :ref:`SmallVector <dss_smallvector>` than :ref:`vector <dss_vector>`. Doing so
546 avoids (relatively) expensive malloc/free calls, which dwarf the cost of adding
547 the elements to the container.
551 Sequential Containers (std::vector, std::list, etc)
552 ---------------------------------------------------
554 There are a variety of sequential containers available for you, based on your
555 needs. Pick the first in this section that will do what you want.
562 The ``llvm::ArrayRef`` class is the preferred class to use in an interface that
563 accepts a sequential list of elements in memory and just reads from them. By
564 taking an ``ArrayRef``, the API can be passed a fixed size array, an
565 ``std::vector``, an ``llvm::SmallVector`` and anything else that is contiguous
573 Fixed size arrays are very simple and very fast. They are good if you know
574 exactly how many elements you have, or you have a (low) upper bound on how many
579 Heap Allocated Arrays
580 ^^^^^^^^^^^^^^^^^^^^^
582 Heap allocated arrays (``new[]`` + ``delete[]``) are also simple. They are good
583 if the number of elements is variable, if you know how many elements you will
584 need before the array is allocated, and if the array is usually large (if not,
585 consider a :ref:`SmallVector <dss_smallvector>`). The cost of a heap allocated
586 array is the cost of the new/delete (aka malloc/free). Also note that if you
587 are allocating an array of a type with a constructor, the constructor and
588 destructors will be run for every element in the array (re-sizable vectors only
589 construct those elements actually used).
591 .. _dss_tinyptrvector:
593 llvm/ADT/TinyPtrVector.h
594 ^^^^^^^^^^^^^^^^^^^^^^^^
596 ``TinyPtrVector<Type>`` is a highly specialized collection class that is
597 optimized to avoid allocation in the case when a vector has zero or one
598 elements. It has two major restrictions: 1) it can only hold values of pointer
599 type, and 2) it cannot hold a null pointer.
601 Since this container is highly specialized, it is rarely used.
605 llvm/ADT/SmallVector.h
606 ^^^^^^^^^^^^^^^^^^^^^^
608 ``SmallVector<Type, N>`` is a simple class that looks and smells just like
609 ``vector<Type>``: it supports efficient iteration, lays out elements in memory
610 order (so you can do pointer arithmetic between elements), supports efficient
611 push_back/pop_back operations, supports efficient random access to its elements,
614 The advantage of SmallVector is that it allocates space for some number of
615 elements (N) **in the object itself**. Because of this, if the SmallVector is
616 dynamically smaller than N, no malloc is performed. This can be a big win in
617 cases where the malloc/free call is far more expensive than the code that
618 fiddles around with the elements.
620 This is good for vectors that are "usually small" (e.g. the number of
621 predecessors/successors of a block is usually less than 8). On the other hand,
622 this makes the size of the SmallVector itself large, so you don't want to
623 allocate lots of them (doing so will waste a lot of space). As such,
624 SmallVectors are most useful when on the stack.
626 SmallVector also provides a nice portable and efficient replacement for
634 ``std::vector`` is well loved and respected. It is useful when SmallVector
635 isn't: when the size of the vector is often large (thus the small optimization
636 will rarely be a benefit) or if you will be allocating many instances of the
637 vector itself (which would waste space for elements that aren't in the
638 container). vector is also useful when interfacing with code that expects
641 One worthwhile note about std::vector: avoid code like this:
650 Instead, write this as:
660 Doing so will save (at least) one heap allocation and free per iteration of the
668 ``std::deque`` is, in some senses, a generalized version of ``std::vector``.
669 Like ``std::vector``, it provides constant time random access and other similar
670 properties, but it also provides efficient access to the front of the list. It
671 does not guarantee continuity of elements within memory.
673 In exchange for this extra flexibility, ``std::deque`` has significantly higher
674 constant factor costs than ``std::vector``. If possible, use ``std::vector`` or
682 ``std::list`` is an extremely inefficient class that is rarely useful. It
683 performs a heap allocation for every element inserted into it, thus having an
684 extremely high constant factor, particularly for small data types.
685 ``std::list`` also only supports bidirectional iteration, not random access
688 In exchange for this high cost, std::list supports efficient access to both ends
689 of the list (like ``std::deque``, but unlike ``std::vector`` or
690 ``SmallVector``). In addition, the iterator invalidation characteristics of
691 std::list are stronger than that of a vector class: inserting or removing an
692 element into the list does not invalidate iterator or pointers to other elements
700 ``ilist<T>`` implements an 'intrusive' doubly-linked list. It is intrusive,
701 because it requires the element to store and provide access to the prev/next
702 pointers for the list.
704 ``ilist`` has the same drawbacks as ``std::list``, and additionally requires an
705 ``ilist_traits`` implementation for the element type, but it provides some novel
706 characteristics. In particular, it can efficiently store polymorphic objects,
707 the traits class is informed when an element is inserted or removed from the
708 list, and ``ilist``\ s are guaranteed to support a constant-time splice
711 These properties are exactly what we want for things like ``Instruction``\ s and
712 basic blocks, which is why these are implemented with ``ilist``\ s.
714 Related classes of interest are explained in the following subsections:
716 * :ref:`ilist_traits <dss_ilist_traits>`
718 * :ref:`iplist <dss_iplist>`
720 * :ref:`llvm/ADT/ilist_node.h <dss_ilist_node>`
722 * :ref:`Sentinels <dss_ilist_sentinel>`
724 .. _dss_packedvector:
726 llvm/ADT/PackedVector.h
727 ^^^^^^^^^^^^^^^^^^^^^^^
729 Useful for storing a vector of values using only a few number of bits for each
730 value. Apart from the standard operations of a vector-like container, it can
731 also perform an 'or' set operation.
739 FirstCondition = 0x1,
740 SecondCondition = 0x2,
745 PackedVector<State, 2> Vec1;
746 Vec1.push_back(FirstCondition);
748 PackedVector<State, 2> Vec2;
749 Vec2.push_back(SecondCondition);
752 return Vec1[0]; // returns 'Both'.
755 .. _dss_ilist_traits:
760 ``ilist_traits<T>`` is ``ilist<T>``'s customization mechanism. ``iplist<T>``
761 (and consequently ``ilist<T>``) publicly derive from this traits class.
768 ``iplist<T>`` is ``ilist<T>``'s base and as such supports a slightly narrower
769 interface. Notably, inserters from ``T&`` are absent.
771 ``ilist_traits<T>`` is a public base of this class and can be used for a wide
772 variety of customizations.
776 llvm/ADT/ilist_node.h
777 ^^^^^^^^^^^^^^^^^^^^^
779 ``ilist_node<T>`` implements a the forward and backward links that are expected
780 by the ``ilist<T>`` (and analogous containers) in the default manner.
782 ``ilist_node<T>``\ s are meant to be embedded in the node type ``T``, usually
783 ``T`` publicly derives from ``ilist_node<T>``.
785 .. _dss_ilist_sentinel:
790 ``ilist``\ s have another specialty that must be considered. To be a good
791 citizen in the C++ ecosystem, it needs to support the standard container
792 operations, such as ``begin`` and ``end`` iterators, etc. Also, the
793 ``operator--`` must work correctly on the ``end`` iterator in the case of
794 non-empty ``ilist``\ s.
796 The only sensible solution to this problem is to allocate a so-called *sentinel*
797 along with the intrusive list, which serves as the ``end`` iterator, providing
798 the back-link to the last element. However conforming to the C++ convention it
799 is illegal to ``operator++`` beyond the sentinel and it also must not be
802 These constraints allow for some implementation freedom to the ``ilist`` how to
803 allocate and store the sentinel. The corresponding policy is dictated by
804 ``ilist_traits<T>``. By default a ``T`` gets heap-allocated whenever the need
805 for a sentinel arises.
807 While the default policy is sufficient in most cases, it may break down when
808 ``T`` does not provide a default constructor. Also, in the case of many
809 instances of ``ilist``\ s, the memory overhead of the associated sentinels is
810 wasted. To alleviate the situation with numerous and voluminous
811 ``T``-sentinels, sometimes a trick is employed, leading to *ghostly sentinels*.
813 Ghostly sentinels are obtained by specially-crafted ``ilist_traits<T>`` which
814 superpose the sentinel with the ``ilist`` instance in memory. Pointer
815 arithmetic is used to obtain the sentinel, which is relative to the ``ilist``'s
816 ``this`` pointer. The ``ilist`` is augmented by an extra pointer, which serves
817 as the back-link of the sentinel. This is the only field in the ghostly
818 sentinel which can be legally accessed.
822 Other Sequential Container options
823 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
825 Other STL containers are available, such as ``std::string``.
827 There are also various STL adapter classes such as ``std::queue``,
828 ``std::priority_queue``, ``std::stack``, etc. These provide simplified access
829 to an underlying container but don't affect the cost of the container itself.
833 String-like containers
834 ----------------------
836 There are a variety of ways to pass around and use strings in C and C++, and
837 LLVM adds a few new options to choose from. Pick the first option on this list
838 that will do what you need, they are ordered according to their relative cost.
840 Note that is is generally preferred to *not* pass strings around as ``const
841 char*``'s. These have a number of problems, including the fact that they
842 cannot represent embedded nul ("\0") characters, and do not have a length
843 available efficiently. The general replacement for '``const char*``' is
846 For more information on choosing string containers for APIs, please see
847 :ref:`Passing Strings <string_apis>`.
854 The StringRef class is a simple value class that contains a pointer to a
855 character and a length, and is quite related to the :ref:`ArrayRef
856 <dss_arrayref>` class (but specialized for arrays of characters). Because
857 StringRef carries a length with it, it safely handles strings with embedded nul
858 characters in it, getting the length does not require a strlen call, and it even
859 has very convenient APIs for slicing and dicing the character range that it
862 StringRef is ideal for passing simple strings around that are known to be live,
863 either because they are C string literals, std::string, a C array, or a
864 SmallVector. Each of these cases has an efficient implicit conversion to
865 StringRef, which doesn't result in a dynamic strlen being executed.
867 StringRef has a few major limitations which make more powerful string containers
870 #. You cannot directly convert a StringRef to a 'const char*' because there is
871 no way to add a trailing nul (unlike the .c_str() method on various stronger
874 #. StringRef doesn't own or keep alive the underlying string bytes.
875 As such it can easily lead to dangling pointers, and is not suitable for
876 embedding in datastructures in most cases (instead, use an std::string or
877 something like that).
879 #. For the same reason, StringRef cannot be used as the return value of a
880 method if the method "computes" the result string. Instead, use std::string.
882 #. StringRef's do not allow you to mutate the pointed-to string bytes and it
883 doesn't allow you to insert or remove bytes from the range. For editing
884 operations like this, it interoperates with the :ref:`Twine <dss_twine>`
887 Because of its strengths and limitations, it is very common for a function to
888 take a StringRef and for a method on an object to return a StringRef that points
889 into some string that it owns.
896 The Twine class is used as an intermediary datatype for APIs that want to take a
897 string that can be constructed inline with a series of concatenations. Twine
898 works by forming recursive instances of the Twine datatype (a simple value
899 object) on the stack as temporary objects, linking them together into a tree
900 which is then linearized when the Twine is consumed. Twine is only safe to use
901 as the argument to a function, and should always be a const reference, e.g.:
905 void foo(const Twine &T);
909 foo(X + "." + Twine(i));
911 This example forms a string like "blarg.42" by concatenating the values
912 together, and does not form intermediate strings containing "blarg" or "blarg.".
914 Because Twine is constructed with temporary objects on the stack, and because
915 these instances are destroyed at the end of the current statement, it is an
916 inherently dangerous API. For example, this simple variant contains undefined
917 behavior and will probably crash:
921 void foo(const Twine &T);
925 const Twine &Tmp = X + "." + Twine(i);
928 ... because the temporaries are destroyed before the call. That said, Twine's
929 are much more efficient than intermediate std::string temporaries, and they work
930 really well with StringRef. Just be aware of their limitations.
934 llvm/ADT/SmallString.h
935 ^^^^^^^^^^^^^^^^^^^^^^
937 SmallString is a subclass of :ref:`SmallVector <dss_smallvector>` that adds some
938 convenience APIs like += that takes StringRef's. SmallString avoids allocating
939 memory in the case when the preallocated space is enough to hold its data, and
940 it calls back to general heap allocation when required. Since it owns its data,
941 it is very safe to use and supports full mutation of the string.
943 Like SmallVector's, the big downside to SmallString is their sizeof. While they
944 are optimized for small strings, they themselves are not particularly small.
945 This means that they work great for temporary scratch buffers on the stack, but
946 should not generally be put into the heap: it is very rare to see a SmallString
947 as the member of a frequently-allocated heap data structure or returned
955 The standard C++ std::string class is a very general class that (like
956 SmallString) owns its underlying data. sizeof(std::string) is very reasonable
957 so it can be embedded into heap data structures and returned by-value. On the
958 other hand, std::string is highly inefficient for inline editing (e.g.
959 concatenating a bunch of stuff together) and because it is provided by the
960 standard library, its performance characteristics depend a lot of the host
961 standard library (e.g. libc++ and MSVC provide a highly optimized string class,
962 GCC contains a really slow implementation).
964 The major disadvantage of std::string is that almost every operation that makes
965 them larger can allocate memory, which is slow. As such, it is better to use
966 SmallVector or Twine as a scratch buffer, but then use std::string to persist
971 Set-Like Containers (std::set, SmallSet, SetVector, etc)
972 --------------------------------------------------------
974 Set-like containers are useful when you need to canonicalize multiple values
975 into a single representation. There are several different choices for how to do
976 this, providing various trade-offs.
978 .. _dss_sortedvectorset:
983 If you intend to insert a lot of elements, then do a lot of queries, a great
984 approach is to use a vector (or other sequential container) with
985 std::sort+std::unique to remove duplicates. This approach works really well if
986 your usage pattern has these two distinct phases (insert then query), and can be
987 coupled with a good choice of :ref:`sequential container <ds_sequential>`.
989 This combination provides the several nice properties: the result data is
990 contiguous in memory (good for cache locality), has few allocations, is easy to
991 address (iterators in the final vector are just indices or pointers), and can be
992 efficiently queried with a standard binary or radix search.
999 If you have a set-like data structure that is usually small and whose elements
1000 are reasonably small, a ``SmallSet<Type, N>`` is a good choice. This set has
1001 space for N elements in place (thus, if the set is dynamically smaller than N,
1002 no malloc traffic is required) and accesses them with a simple linear search.
1003 When the set grows beyond 'N' elements, it allocates a more expensive
1004 representation that guarantees efficient access (for most types, it falls back
1005 to std::set, but for pointers it uses something far better, :ref:`SmallPtrSet
1008 The magic of this class is that it handles small sets extremely efficiently, but
1009 gracefully handles extremely large sets without loss of efficiency. The
1010 drawback is that the interface is quite small: it supports insertion, queries
1011 and erasing, but does not support iteration.
1013 .. _dss_smallptrset:
1015 llvm/ADT/SmallPtrSet.h
1016 ^^^^^^^^^^^^^^^^^^^^^^
1018 SmallPtrSet has all the advantages of ``SmallSet`` (and a ``SmallSet`` of
1019 pointers is transparently implemented with a ``SmallPtrSet``), but also supports
1020 iterators. If more than 'N' insertions are performed, a single quadratically
1021 probed hash table is allocated and grows as needed, providing extremely
1022 efficient access (constant time insertion/deleting/queries with low constant
1023 factors) and is very stingy with malloc traffic.
1025 Note that, unlike ``std::set``, the iterators of ``SmallPtrSet`` are invalidated
1026 whenever an insertion occurs. Also, the values visited by the iterators are not
1027 visited in sorted order.
1034 DenseSet is a simple quadratically probed hash table. It excels at supporting
1035 small values: it uses a single allocation to hold all of the pairs that are
1036 currently inserted in the set. DenseSet is a great way to unique small values
1037 that are not simple pointers (use :ref:`SmallPtrSet <dss_smallptrset>` for
1038 pointers). Note that DenseSet has the same requirements for the value type that
1039 :ref:`DenseMap <dss_densemap>` has.
1043 llvm/ADT/SparseSet.h
1044 ^^^^^^^^^^^^^^^^^^^^
1046 SparseSet holds a small number of objects identified by unsigned keys of
1047 moderate size. It uses a lot of memory, but provides operations that are almost
1048 as fast as a vector. Typical keys are physical registers, virtual registers, or
1049 numbered basic blocks.
1051 SparseSet is useful for algorithms that need very fast clear/find/insert/erase
1052 and fast iteration over small sets. It is not intended for building composite
1055 .. _dss_sparsemultiset:
1057 llvm/ADT/SparseMultiSet.h
1058 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1060 SparseMultiSet adds multiset behavior to SparseSet, while retaining SparseSet's
1061 desirable attributes. Like SparseSet, it typically uses a lot of memory, but
1062 provides operations that are almost as fast as a vector. Typical keys are
1063 physical registers, virtual registers, or numbered basic blocks.
1065 SparseMultiSet is useful for algorithms that need very fast
1066 clear/find/insert/erase of the entire collection, and iteration over sets of
1067 elements sharing a key. It is often a more efficient choice than using composite
1068 data structures (e.g. vector-of-vectors, map-of-vectors). It is not intended for
1069 building composite data structures.
1073 llvm/ADT/FoldingSet.h
1074 ^^^^^^^^^^^^^^^^^^^^^
1076 FoldingSet is an aggregate class that is really good at uniquing
1077 expensive-to-create or polymorphic objects. It is a combination of a chained
1078 hash table with intrusive links (uniqued objects are required to inherit from
1079 FoldingSetNode) that uses :ref:`SmallVector <dss_smallvector>` as part of its ID
1082 Consider a case where you want to implement a "getOrCreateFoo" method for a
1083 complex object (for example, a node in the code generator). The client has a
1084 description of **what** it wants to generate (it knows the opcode and all the
1085 operands), but we don't want to 'new' a node, then try inserting it into a set
1086 only to find out it already exists, at which point we would have to delete it
1087 and return the node that already exists.
1089 To support this style of client, FoldingSet perform a query with a
1090 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1091 element that we want to query for. The query either returns the element
1092 matching the ID or it returns an opaque ID that indicates where insertion should
1093 take place. Construction of the ID usually does not require heap traffic.
1095 Because FoldingSet uses intrusive links, it can support polymorphic objects in
1096 the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1097 Because the elements are individually allocated, pointers to the elements are
1098 stable: inserting or removing elements does not invalidate any pointers to other
1106 ``std::set`` is a reasonable all-around set class, which is decent at many
1107 things but great at nothing. std::set allocates memory for each element
1108 inserted (thus it is very malloc intensive) and typically stores three pointers
1109 per element in the set (thus adding a large amount of per-element space
1110 overhead). It offers guaranteed log(n) performance, which is not particularly
1111 fast from a complexity standpoint (particularly if the elements of the set are
1112 expensive to compare, like strings), and has extremely high constant factors for
1113 lookup, insertion and removal.
1115 The advantages of std::set are that its iterators are stable (deleting or
1116 inserting an element from the set does not affect iterators or pointers to other
1117 elements) and that iteration over the set is guaranteed to be in sorted order.
1118 If the elements in the set are large, then the relative overhead of the pointers
1119 and malloc traffic is not a big deal, but if the elements of the set are small,
1120 std::set is almost never a good choice.
1124 llvm/ADT/SetVector.h
1125 ^^^^^^^^^^^^^^^^^^^^
1127 LLVM's ``SetVector<Type>`` is an adapter class that combines your choice of a
1128 set-like container along with a :ref:`Sequential Container <ds_sequential>` The
1129 important property that this provides is efficient insertion with uniquing
1130 (duplicate elements are ignored) with iteration support. It implements this by
1131 inserting elements into both a set-like container and the sequential container,
1132 using the set-like container for uniquing and the sequential container for
1135 The difference between SetVector and other sets is that the order of iteration
1136 is guaranteed to match the order of insertion into the SetVector. This property
1137 is really important for things like sets of pointers. Because pointer values
1138 are non-deterministic (e.g. vary across runs of the program on different
1139 machines), iterating over the pointers in the set will not be in a well-defined
1142 The drawback of SetVector is that it requires twice as much space as a normal
1143 set and has the sum of constant factors from the set-like container and the
1144 sequential container that it uses. Use it **only** if you need to iterate over
1145 the elements in a deterministic order. SetVector is also expensive to delete
1146 elements out of (linear time), unless you use it's "pop_back" method, which is
1149 ``SetVector`` is an adapter class that defaults to using ``std::vector`` and a
1150 size 16 ``SmallSet`` for the underlying containers, so it is quite expensive.
1151 However, ``"llvm/ADT/SetVector.h"`` also provides a ``SmallSetVector`` class,
1152 which defaults to using a ``SmallVector`` and ``SmallSet`` of a specified size.
1153 If you use this, and if your sets are dynamically smaller than ``N``, you will
1154 save a lot of heap traffic.
1156 .. _dss_uniquevector:
1158 llvm/ADT/UniqueVector.h
1159 ^^^^^^^^^^^^^^^^^^^^^^^
1161 UniqueVector is similar to :ref:`SetVector <dss_setvector>` but it retains a
1162 unique ID for each element inserted into the set. It internally contains a map
1163 and a vector, and it assigns a unique ID for each value inserted into the set.
1165 UniqueVector is very expensive: its cost is the sum of the cost of maintaining
1166 both the map and vector, it has high complexity, high constant factors, and
1167 produces a lot of malloc traffic. It should be avoided.
1169 .. _dss_immutableset:
1171 llvm/ADT/ImmutableSet.h
1172 ^^^^^^^^^^^^^^^^^^^^^^^
1174 ImmutableSet is an immutable (functional) set implementation based on an AVL
1175 tree. Adding or removing elements is done through a Factory object and results
1176 in the creation of a new ImmutableSet object. If an ImmutableSet already exists
1177 with the given contents, then the existing one is returned; equality is compared
1178 with a FoldingSetNodeID. The time and space complexity of add or remove
1179 operations is logarithmic in the size of the original set.
1181 There is no method for returning an element of the set, you can only check for
1186 Other Set-Like Container Options
1187 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1189 The STL provides several other options, such as std::multiset and the various
1190 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1191 never use hash_set and unordered_set because they are generally very expensive
1192 (each insertion requires a malloc) and very non-portable.
1194 std::multiset is useful if you're not interested in elimination of duplicates,
1195 but has all the drawbacks of std::set. A sorted vector (where you don't delete
1196 duplicate entries) or some other approach is almost always better.
1200 Map-Like Containers (std::map, DenseMap, etc)
1201 ---------------------------------------------
1203 Map-like containers are useful when you want to associate data to a key. As
1204 usual, there are a lot of different ways to do this. :)
1206 .. _dss_sortedvectormap:
1211 If your usage pattern follows a strict insert-then-query approach, you can
1212 trivially use the same approach as :ref:`sorted vectors for set-like containers
1213 <dss_sortedvectorset>`. The only difference is that your query function (which
1214 uses std::lower_bound to get efficient log(n) lookup) should only compare the
1215 key, not both the key and value. This yields the same advantages as sorted
1220 llvm/ADT/StringMap.h
1221 ^^^^^^^^^^^^^^^^^^^^
1223 Strings are commonly used as keys in maps, and they are difficult to support
1224 efficiently: they are variable length, inefficient to hash and compare when
1225 long, expensive to copy, etc. StringMap is a specialized container designed to
1226 cope with these issues. It supports mapping an arbitrary range of bytes to an
1227 arbitrary other object.
1229 The StringMap implementation uses a quadratically-probed hash table, where the
1230 buckets store a pointer to the heap allocated entries (and some other stuff).
1231 The entries in the map must be heap allocated because the strings are variable
1232 length. The string data (key) and the element object (value) are stored in the
1233 same allocation with the string data immediately after the element object.
1234 This container guarantees the "``(char*)(&Value+1)``" points to the key string
1237 The StringMap is very fast for several reasons: quadratic probing is very cache
1238 efficient for lookups, the hash value of strings in buckets is not recomputed
1239 when looking up an element, StringMap rarely has to touch the memory for
1240 unrelated objects when looking up a value (even when hash collisions happen),
1241 hash table growth does not recompute the hash values for strings already in the
1242 table, and each pair in the map is store in a single allocation (the string data
1243 is stored in the same allocation as the Value of a pair).
1245 StringMap also provides query methods that take byte ranges, so it only ever
1246 copies a string if a value is inserted into the table.
1248 StringMap iteratation order, however, is not guaranteed to be deterministic, so
1249 any uses which require that should instead use a std::map.
1253 llvm/ADT/IndexedMap.h
1254 ^^^^^^^^^^^^^^^^^^^^^
1256 IndexedMap is a specialized container for mapping small dense integers (or
1257 values that can be mapped to small dense integers) to some other type. It is
1258 internally implemented as a vector with a mapping function that maps the keys
1259 to the dense integer range.
1261 This is useful for cases like virtual registers in the LLVM code generator: they
1262 have a dense mapping that is offset by a compile-time constant (the first
1263 virtual register ID).
1270 DenseMap is a simple quadratically probed hash table. It excels at supporting
1271 small keys and values: it uses a single allocation to hold all of the pairs
1272 that are currently inserted in the map. DenseMap is a great way to map
1273 pointers to pointers, or map other small types to each other.
1275 There are several aspects of DenseMap that you should be aware of, however.
1276 The iterators in a DenseMap are invalidated whenever an insertion occurs,
1277 unlike map. Also, because DenseMap allocates space for a large number of
1278 key/value pairs (it starts with 64 by default), it will waste a lot of space if
1279 your keys or values are large. Finally, you must implement a partial
1280 specialization of DenseMapInfo for the key that you want, if it isn't already
1281 supported. This is required to tell DenseMap about two special marker values
1282 (which can never be inserted into the map) that it needs internally.
1284 DenseMap's find_as() method supports lookup operations using an alternate key
1285 type. This is useful in cases where the normal key type is expensive to
1286 construct, but cheap to compare against. The DenseMapInfo is responsible for
1287 defining the appropriate comparison and hashing methods for each alternate key
1295 ValueMap is a wrapper around a :ref:`DenseMap <dss_densemap>` mapping
1296 ``Value*``\ s (or subclasses) to another type. When a Value is deleted or
1297 RAUW'ed, ValueMap will update itself so the new version of the key is mapped to
1298 the same value, just as if the key were a WeakVH. You can configure exactly how
1299 this happens, and what else happens on these two events, by passing a ``Config``
1300 parameter to the ValueMap template.
1302 .. _dss_intervalmap:
1304 llvm/ADT/IntervalMap.h
1305 ^^^^^^^^^^^^^^^^^^^^^^
1307 IntervalMap is a compact map for small keys and values. It maps key intervals
1308 instead of single keys, and it will automatically coalesce adjacent intervals.
1309 When then map only contains a few intervals, they are stored in the map object
1310 itself to avoid allocations.
1312 The IntervalMap iterators are quite big, so they should not be passed around as
1313 STL iterators. The heavyweight iterators allow a smaller data structure.
1320 std::map has similar characteristics to :ref:`std::set <dss_set>`: it uses a
1321 single allocation per pair inserted into the map, it offers log(n) lookup with
1322 an extremely large constant factor, imposes a space penalty of 3 pointers per
1323 pair in the map, etc.
1325 std::map is most useful when your keys or values are very large, if you need to
1326 iterate over the collection in sorted order, or if you need stable iterators
1327 into the map (i.e. they don't get invalidated if an insertion or deletion of
1328 another element takes place).
1332 llvm/ADT/MapVector.h
1333 ^^^^^^^^^^^^^^^^^^^^
1335 ``MapVector<KeyT,ValueT>`` provides a subset of the DenseMap interface. The
1336 main difference is that the iteration order is guaranteed to be the insertion
1337 order, making it an easy (but somewhat expensive) solution for non-deterministic
1338 iteration over maps of pointers.
1340 It is implemented by mapping from key to an index in a vector of key,value
1341 pairs. This provides fast lookup and iteration, but has two main drawbacks: The
1342 key is stored twice and it doesn't support removing elements.
1344 .. _dss_inteqclasses:
1346 llvm/ADT/IntEqClasses.h
1347 ^^^^^^^^^^^^^^^^^^^^^^^
1349 IntEqClasses provides a compact representation of equivalence classes of small
1350 integers. Initially, each integer in the range 0..n-1 has its own equivalence
1351 class. Classes can be joined by passing two class representatives to the
1352 join(a, b) method. Two integers are in the same class when findLeader() returns
1353 the same representative.
1355 Once all equivalence classes are formed, the map can be compressed so each
1356 integer 0..n-1 maps to an equivalence class number in the range 0..m-1, where m
1357 is the total number of equivalence classes. The map must be uncompressed before
1358 it can be edited again.
1360 .. _dss_immutablemap:
1362 llvm/ADT/ImmutableMap.h
1363 ^^^^^^^^^^^^^^^^^^^^^^^
1365 ImmutableMap is an immutable (functional) map implementation based on an AVL
1366 tree. Adding or removing elements is done through a Factory object and results
1367 in the creation of a new ImmutableMap object. If an ImmutableMap already exists
1368 with the given key set, then the existing one is returned; equality is compared
1369 with a FoldingSetNodeID. The time and space complexity of add or remove
1370 operations is logarithmic in the size of the original map.
1374 Other Map-Like Container Options
1375 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1377 The STL provides several other options, such as std::multimap and the various
1378 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1379 never use hash_set and unordered_set because they are generally very expensive
1380 (each insertion requires a malloc) and very non-portable.
1382 std::multimap is useful if you want to map a key to multiple values, but has all
1383 the drawbacks of std::map. A sorted vector or some other approach is almost
1388 Bit storage containers (BitVector, SparseBitVector)
1389 ---------------------------------------------------
1391 Unlike the other containers, there are only two bit storage containers, and
1392 choosing when to use each is relatively straightforward.
1394 One additional option is ``std::vector<bool>``: we discourage its use for two
1395 reasons 1) the implementation in many common compilers (e.g. commonly
1396 available versions of GCC) is extremely inefficient and 2) the C++ standards
1397 committee is likely to deprecate this container and/or change it significantly
1398 somehow. In any case, please don't use it.
1405 The BitVector container provides a dynamic size set of bits for manipulation.
1406 It supports individual bit setting/testing, as well as set operations. The set
1407 operations take time O(size of bitvector), but operations are performed one word
1408 at a time, instead of one bit at a time. This makes the BitVector very fast for
1409 set operations compared to other containers. Use the BitVector when you expect
1410 the number of set bits to be high (i.e. a dense set).
1412 .. _dss_smallbitvector:
1417 The SmallBitVector container provides the same interface as BitVector, but it is
1418 optimized for the case where only a small number of bits, less than 25 or so,
1419 are needed. It also transparently supports larger bit counts, but slightly less
1420 efficiently than a plain BitVector, so SmallBitVector should only be used when
1421 larger counts are rare.
1423 At this time, SmallBitVector does not support set operations (and, or, xor), and
1424 its operator[] does not provide an assignable lvalue.
1426 .. _dss_sparsebitvector:
1431 The SparseBitVector container is much like BitVector, with one major difference:
1432 Only the bits that are set, are stored. This makes the SparseBitVector much
1433 more space efficient than BitVector when the set is sparse, as well as making
1434 set operations O(number of set bits) instead of O(size of universe). The
1435 downside to the SparseBitVector is that setting and testing of random bits is
1436 O(N), and on large SparseBitVectors, this can be slower than BitVector. In our
1437 implementation, setting or testing bits in sorted order (either forwards or
1438 reverse) is O(1) worst case. Testing and setting bits within 128 bits (depends
1439 on size) of the current bit is also O(1). As a general statement,
1440 testing/setting bits in a SparseBitVector is O(distance away from last set bit).
1444 Helpful Hints for Common Operations
1445 ===================================
1447 This section describes how to perform some very simple transformations of LLVM
1448 code. This is meant to give examples of common idioms used, showing the
1449 practical side of LLVM transformations.
1451 Because this is a "how-to" section, you should also read about the main classes
1452 that you will be working with. The :ref:`Core LLVM Class Hierarchy Reference
1453 <coreclasses>` contains details and descriptions of the main classes that you
1458 Basic Inspection and Traversal Routines
1459 ---------------------------------------
1461 The LLVM compiler infrastructure have many different data structures that may be
1462 traversed. Following the example of the C++ standard template library, the
1463 techniques used to traverse these various data structures are all basically the
1464 same. For a enumerable sequence of values, the ``XXXbegin()`` function (or
1465 method) returns an iterator to the start of the sequence, the ``XXXend()``
1466 function returns an iterator pointing to one past the last valid element of the
1467 sequence, and there is some ``XXXiterator`` data type that is common between the
1470 Because the pattern for iteration is common across many different aspects of the
1471 program representation, the standard template library algorithms may be used on
1472 them, and it is easier to remember how to iterate. First we show a few common
1473 examples of the data structures that need to be traversed. Other data
1474 structures are traversed in very similar ways.
1476 .. _iterate_function:
1478 Iterating over the ``BasicBlock`` in a ``Function``
1479 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1481 It's quite common to have a ``Function`` instance that you'd like to transform
1482 in some way; in particular, you'd like to manipulate its ``BasicBlock``\ s. To
1483 facilitate this, you'll need to iterate over all of the ``BasicBlock``\ s that
1484 constitute the ``Function``. The following is an example that prints the name
1485 of a ``BasicBlock`` and the number of ``Instruction``\ s it contains:
1489 // func is a pointer to a Function instance
1490 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
1491 // Print out the name of the basic block if it has one, and then the
1492 // number of instructions that it contains
1493 errs() << "Basic block (name=" << i->getName() << ") has "
1494 << i->size() << " instructions.\n";
1496 Note that i can be used as if it were a pointer for the purposes of invoking
1497 member functions of the ``Instruction`` class. This is because the indirection
1498 operator is overloaded for the iterator classes. In the above code, the
1499 expression ``i->size()`` is exactly equivalent to ``(*i).size()`` just like
1502 .. _iterate_basicblock:
1504 Iterating over the ``Instruction`` in a ``BasicBlock``
1505 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1507 Just like when dealing with ``BasicBlock``\ s in ``Function``\ s, it's easy to
1508 iterate over the individual instructions that make up ``BasicBlock``\ s. Here's
1509 a code snippet that prints out each instruction in a ``BasicBlock``:
1513 // blk is a pointer to a BasicBlock instance
1514 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
1515 // The next statement works since operator<<(ostream&,...)
1516 // is overloaded for Instruction&
1517 errs() << *i << "\n";
1520 However, this isn't really the best way to print out the contents of a
1521 ``BasicBlock``! Since the ostream operators are overloaded for virtually
1522 anything you'll care about, you could have just invoked the print routine on the
1523 basic block itself: ``errs() << *blk << "\n";``.
1525 .. _iterate_insiter:
1527 Iterating over the ``Instruction`` in a ``Function``
1528 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1530 If you're finding that you commonly iterate over a ``Function``'s
1531 ``BasicBlock``\ s and then that ``BasicBlock``'s ``Instruction``\ s,
1532 ``InstIterator`` should be used instead. You'll need to include
1533 ``llvm/Support/InstIterator.h`` (`doxygen
1534 <http://llvm.org/doxygen/InstIterator_8h-source.html>`__) and then instantiate
1535 ``InstIterator``\ s explicitly in your code. Here's a small example that shows
1536 how to dump all instructions in a function to the standard error stream:
1540 #include "llvm/Support/InstIterator.h"
1542 // F is a pointer to a Function instance
1543 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1544 errs() << *I << "\n";
1546 Easy, isn't it? You can also use ``InstIterator``\ s to fill a work list with
1547 its initial contents. For example, if you wanted to initialize a work list to
1548 contain all instructions in a ``Function`` F, all you would need to do is
1553 std::set<Instruction*> worklist;
1554 // or better yet, SmallPtrSet<Instruction*, 64> worklist;
1556 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1557 worklist.insert(&*I);
1559 The STL set ``worklist`` would now contain all instructions in the ``Function``
1562 .. _iterate_convert:
1564 Turning an iterator into a class pointer (and vice-versa)
1565 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1567 Sometimes, it'll be useful to grab a reference (or pointer) to a class instance
1568 when all you've got at hand is an iterator. Well, extracting a reference or a
1569 pointer from an iterator is very straight-forward. Assuming that ``i`` is a
1570 ``BasicBlock::iterator`` and ``j`` is a ``BasicBlock::const_iterator``:
1574 Instruction& inst = *i; // Grab reference to instruction reference
1575 Instruction* pinst = &*i; // Grab pointer to instruction reference
1576 const Instruction& inst = *j;
1578 However, the iterators you'll be working with in the LLVM framework are special:
1579 they will automatically convert to a ptr-to-instance type whenever they need to.
1580 Instead of derferencing the iterator and then taking the address of the result,
1581 you can simply assign the iterator to the proper pointer type and you get the
1582 dereference and address-of operation as a result of the assignment (behind the
1583 scenes, this is a result of overloading casting mechanisms). Thus the last line
1584 of the last example,
1588 Instruction *pinst = &*i;
1590 is semantically equivalent to
1594 Instruction *pinst = i;
1596 It's also possible to turn a class pointer into the corresponding iterator, and
1597 this is a constant time operation (very efficient). The following code snippet
1598 illustrates use of the conversion constructors provided by LLVM iterators. By
1599 using these, you can explicitly grab the iterator of something without actually
1600 obtaining it via iteration over some structure:
1604 void printNextInstruction(Instruction* inst) {
1605 BasicBlock::iterator it(inst);
1606 ++it; // After this line, it refers to the instruction after *inst
1607 if (it != inst->getParent()->end()) errs() << *it << "\n";
1610 Unfortunately, these implicit conversions come at a cost; they prevent these
1611 iterators from conforming to standard iterator conventions, and thus from being
1612 usable with standard algorithms and containers. For example, they prevent the
1613 following code, where ``B`` is a ``BasicBlock``, from compiling:
1617 llvm::SmallVector<llvm::Instruction *, 16>(B->begin(), B->end());
1619 Because of this, these implicit conversions may be removed some day, and
1620 ``operator*`` changed to return a pointer instead of a reference.
1622 .. _iterate_complex:
1624 Finding call sites: a slightly more complex example
1625 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1627 Say that you're writing a FunctionPass and would like to count all the locations
1628 in the entire module (that is, across every ``Function``) where a certain
1629 function (i.e., some ``Function *``) is already in scope. As you'll learn
1630 later, you may want to use an ``InstVisitor`` to accomplish this in a much more
1631 straight-forward manner, but this example will allow us to explore how you'd do
1632 it if you didn't have ``InstVisitor`` around. In pseudo-code, this is what we
1635 .. code-block:: none
1637 initialize callCounter to zero
1638 for each Function f in the Module
1639 for each BasicBlock b in f
1640 for each Instruction i in b
1641 if (i is a CallInst and calls the given function)
1642 increment callCounter
1644 And the actual code is (remember, because we're writing a ``FunctionPass``, our
1645 ``FunctionPass``-derived class simply has to override the ``runOnFunction``
1650 Function* targetFunc = ...;
1652 class OurFunctionPass : public FunctionPass {
1654 OurFunctionPass(): callCounter(0) { }
1656 virtual runOnFunction(Function& F) {
1657 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1658 for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) {
1659 if (CallInst* callInst = dyn_cast<CallInst>(&*i)) {
1660 // We know we've encountered a call instruction, so we
1661 // need to determine if it's a call to the
1662 // function pointed to by m_func or not.
1663 if (callInst->getCalledFunction() == targetFunc)
1671 unsigned callCounter;
1674 .. _calls_and_invokes:
1676 Treating calls and invokes the same way
1677 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1679 You may have noticed that the previous example was a bit oversimplified in that
1680 it did not deal with call sites generated by 'invoke' instructions. In this,
1681 and in other situations, you may find that you want to treat ``CallInst``\ s and
1682 ``InvokeInst``\ s the same way, even though their most-specific common base
1683 class is ``Instruction``, which includes lots of less closely-related things.
1684 For these cases, LLVM provides a handy wrapper class called ``CallSite``
1685 (`doxygen <http://llvm.org/doxygen/classllvm_1_1CallSite.html>`__) It is
1686 essentially a wrapper around an ``Instruction`` pointer, with some methods that
1687 provide functionality common to ``CallInst``\ s and ``InvokeInst``\ s.
1689 This class has "value semantics": it should be passed by value, not by reference
1690 and it should not be dynamically allocated or deallocated using ``operator new``
1691 or ``operator delete``. It is efficiently copyable, assignable and
1692 constructable, with costs equivalents to that of a bare pointer. If you look at
1693 its definition, it has only a single pointer member.
1697 Iterating over def-use & use-def chains
1698 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1700 Frequently, we might have an instance of the ``Value`` class (`doxygen
1701 <http://llvm.org/doxygen/classllvm_1_1Value.html>`__) and we want to determine
1702 which ``User`` s use the ``Value``. The list of all ``User``\ s of a particular
1703 ``Value`` is called a *def-use* chain. For example, let's say we have a
1704 ``Function*`` named ``F`` to a particular function ``foo``. Finding all of the
1705 instructions that *use* ``foo`` is as simple as iterating over the *def-use*
1712 for (Value::use_iterator i = F->use_begin(), e = F->use_end(); i != e; ++i)
1713 if (Instruction *Inst = dyn_cast<Instruction>(*i)) {
1714 errs() << "F is used in instruction:\n";
1715 errs() << *Inst << "\n";
1718 Note that dereferencing a ``Value::use_iterator`` is not a very cheap operation.
1719 Instead of performing ``*i`` above several times, consider doing it only once in
1720 the loop body and reusing its result.
1722 Alternatively, it's common to have an instance of the ``User`` Class (`doxygen
1723 <http://llvm.org/doxygen/classllvm_1_1User.html>`__) and need to know what
1724 ``Value``\ s are used by it. The list of all ``Value``\ s used by a ``User`` is
1725 known as a *use-def* chain. Instances of class ``Instruction`` are common
1726 ``User`` s, so we might want to iterate over all of the values that a particular
1727 instruction uses (that is, the operands of the particular ``Instruction``):
1731 Instruction *pi = ...;
1733 for (User::op_iterator i = pi->op_begin(), e = pi->op_end(); i != e; ++i) {
1738 Declaring objects as ``const`` is an important tool of enforcing mutation free
1739 algorithms (such as analyses, etc.). For this purpose above iterators come in
1740 constant flavors as ``Value::const_use_iterator`` and
1741 ``Value::const_op_iterator``. They automatically arise when calling
1742 ``use/op_begin()`` on ``const Value*``\ s or ``const User*``\ s respectively.
1743 Upon dereferencing, they return ``const Use*``\ s. Otherwise the above patterns
1748 Iterating over predecessors & successors of blocks
1749 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1751 Iterating over the predecessors and successors of a block is quite easy with the
1752 routines defined in ``"llvm/Support/CFG.h"``. Just use code like this to
1753 iterate over all predecessors of BB:
1757 #include "llvm/Support/CFG.h"
1758 BasicBlock *BB = ...;
1760 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
1761 BasicBlock *Pred = *PI;
1765 Similarly, to iterate over successors use ``succ_iterator/succ_begin/succ_end``.
1769 Making simple changes
1770 ---------------------
1772 There are some primitive transformation operations present in the LLVM
1773 infrastructure that are worth knowing about. When performing transformations,
1774 it's fairly common to manipulate the contents of basic blocks. This section
1775 describes some of the common methods for doing so and gives example code.
1777 .. _schanges_creating:
1779 Creating and inserting new ``Instruction``\ s
1780 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1782 *Instantiating Instructions*
1784 Creation of ``Instruction``\ s is straight-forward: simply call the constructor
1785 for the kind of instruction to instantiate and provide the necessary parameters.
1786 For example, an ``AllocaInst`` only *requires* a (const-ptr-to) ``Type``. Thus:
1790 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
1792 will create an ``AllocaInst`` instance that represents the allocation of one
1793 integer in the current stack frame, at run time. Each ``Instruction`` subclass
1794 is likely to have varying default parameters which change the semantics of the
1795 instruction, so refer to the `doxygen documentation for the subclass of
1796 Instruction <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_ that
1797 you're interested in instantiating.
1801 It is very useful to name the values of instructions when you're able to, as
1802 this facilitates the debugging of your transformations. If you end up looking
1803 at generated LLVM machine code, you definitely want to have logical names
1804 associated with the results of instructions! By supplying a value for the
1805 ``Name`` (default) parameter of the ``Instruction`` constructor, you associate a
1806 logical name with the result of the instruction's execution at run time. For
1807 example, say that I'm writing a transformation that dynamically allocates space
1808 for an integer on the stack, and that integer is going to be used as some kind
1809 of index by some other code. To accomplish this, I place an ``AllocaInst`` at
1810 the first point in the first ``BasicBlock`` of some ``Function``, and I'm
1811 intending to use it within the same ``Function``. I might do:
1815 AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
1817 where ``indexLoc`` is now the logical name of the instruction's execution value,
1818 which is a pointer to an integer on the run time stack.
1820 *Inserting instructions*
1822 There are essentially two ways to insert an ``Instruction`` into an existing
1823 sequence of instructions that form a ``BasicBlock``:
1825 * Insertion into an explicit instruction list
1827 Given a ``BasicBlock* pb``, an ``Instruction* pi`` within that ``BasicBlock``,
1828 and a newly-created instruction we wish to insert before ``*pi``, we do the
1833 BasicBlock *pb = ...;
1834 Instruction *pi = ...;
1835 Instruction *newInst = new Instruction(...);
1837 pb->getInstList().insert(pi, newInst); // Inserts newInst before pi in pb
1839 Appending to the end of a ``BasicBlock`` is so common that the ``Instruction``
1840 class and ``Instruction``-derived classes provide constructors which take a
1841 pointer to a ``BasicBlock`` to be appended to. For example code that looked
1846 BasicBlock *pb = ...;
1847 Instruction *newInst = new Instruction(...);
1849 pb->getInstList().push_back(newInst); // Appends newInst to pb
1855 BasicBlock *pb = ...;
1856 Instruction *newInst = new Instruction(..., pb);
1858 which is much cleaner, especially if you are creating long instruction
1861 * Insertion into an implicit instruction list
1863 ``Instruction`` instances that are already in ``BasicBlock``\ s are implicitly
1864 associated with an existing instruction list: the instruction list of the
1865 enclosing basic block. Thus, we could have accomplished the same thing as the
1866 above code without being given a ``BasicBlock`` by doing:
1870 Instruction *pi = ...;
1871 Instruction *newInst = new Instruction(...);
1873 pi->getParent()->getInstList().insert(pi, newInst);
1875 In fact, this sequence of steps occurs so frequently that the ``Instruction``
1876 class and ``Instruction``-derived classes provide constructors which take (as
1877 a default parameter) a pointer to an ``Instruction`` which the newly-created
1878 ``Instruction`` should precede. That is, ``Instruction`` constructors are
1879 capable of inserting the newly-created instance into the ``BasicBlock`` of a
1880 provided instruction, immediately before that instruction. Using an
1881 ``Instruction`` constructor with a ``insertBefore`` (default) parameter, the
1886 Instruction* pi = ...;
1887 Instruction* newInst = new Instruction(..., pi);
1889 which is much cleaner, especially if you're creating a lot of instructions and
1890 adding them to ``BasicBlock``\ s.
1892 .. _schanges_deleting:
1894 Deleting Instructions
1895 ^^^^^^^^^^^^^^^^^^^^^
1897 Deleting an instruction from an existing sequence of instructions that form a
1898 BasicBlock_ is very straight-forward: just call the instruction's
1899 ``eraseFromParent()`` method. For example:
1903 Instruction *I = .. ;
1904 I->eraseFromParent();
1906 This unlinks the instruction from its containing basic block and deletes it. If
1907 you'd just like to unlink the instruction from its containing basic block but
1908 not delete it, you can use the ``removeFromParent()`` method.
1910 .. _schanges_replacing:
1912 Replacing an Instruction with another Value
1913 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1915 Replacing individual instructions
1916 """""""""""""""""""""""""""""""""
1918 Including "`llvm/Transforms/Utils/BasicBlockUtils.h
1919 <http://llvm.org/doxygen/BasicBlockUtils_8h-source.html>`_" permits use of two
1920 very useful replace functions: ``ReplaceInstWithValue`` and
1921 ``ReplaceInstWithInst``.
1923 .. _schanges_deleting_sub:
1925 Deleting Instructions
1926 """""""""""""""""""""
1928 * ``ReplaceInstWithValue``
1930 This function replaces all uses of a given instruction with a value, and then
1931 removes the original instruction. The following example illustrates the
1932 replacement of the result of a particular ``AllocaInst`` that allocates memory
1933 for a single integer with a null pointer to an integer.
1937 AllocaInst* instToReplace = ...;
1938 BasicBlock::iterator ii(instToReplace);
1940 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
1941 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
1943 * ``ReplaceInstWithInst``
1945 This function replaces a particular instruction with another instruction,
1946 inserting the new instruction into the basic block at the location where the
1947 old instruction was, and replacing any uses of the old instruction with the
1948 new instruction. The following example illustrates the replacement of one
1949 ``AllocaInst`` with another.
1953 AllocaInst* instToReplace = ...;
1954 BasicBlock::iterator ii(instToReplace);
1956 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
1957 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
1960 Replacing multiple uses of Users and Values
1961 """""""""""""""""""""""""""""""""""""""""""
1963 You can use ``Value::replaceAllUsesWith`` and ``User::replaceUsesOfWith`` to
1964 change more than one use at a time. See the doxygen documentation for the
1965 `Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_ and `User Class
1966 <http://llvm.org/doxygen/classllvm_1_1User.html>`_, respectively, for more
1969 .. _schanges_deletingGV:
1971 Deleting GlobalVariables
1972 ^^^^^^^^^^^^^^^^^^^^^^^^
1974 Deleting a global variable from a module is just as easy as deleting an
1975 Instruction. First, you must have a pointer to the global variable that you
1976 wish to delete. You use this pointer to erase it from its parent, the module.
1981 GlobalVariable *GV = .. ;
1983 GV->eraseFromParent();
1991 In generating IR, you may need some complex types. If you know these types
1992 statically, you can use ``TypeBuilder<...>::get()``, defined in
1993 ``llvm/Support/TypeBuilder.h``, to retrieve them. ``TypeBuilder`` has two forms
1994 depending on whether you're building types for cross-compilation or native
1995 library use. ``TypeBuilder<T, true>`` requires that ``T`` be independent of the
1996 host environment, meaning that it's built out of types from the ``llvm::types``
1997 (`doxygen <http://llvm.org/doxygen/namespacellvm_1_1types.html>`__) namespace
1998 and pointers, functions, arrays, etc. built of those. ``TypeBuilder<T, false>``
1999 additionally allows native C types whose size may depend on the host compiler.
2004 FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get();
2006 is easier to read and write than the equivalent
2010 std::vector<const Type*> params;
2011 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2012 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2014 See the `class comment
2015 <http://llvm.org/doxygen/TypeBuilder_8h-source.html#l00001>`_ for more details.
2022 This section describes the interaction of the LLVM APIs with multithreading,
2023 both on the part of client applications, and in the JIT, in the hosted
2026 Note that LLVM's support for multithreading is still relatively young. Up
2027 through version 2.5, the execution of threaded hosted applications was
2028 supported, but not threaded client access to the APIs. While this use case is
2029 now supported, clients *must* adhere to the guidelines specified below to ensure
2030 proper operation in multithreaded mode.
2032 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2033 intrinsics in order to support threaded operation. If you need a
2034 multhreading-capable LLVM on a platform without a suitably modern system
2035 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2036 using the resultant compiler to build a copy of LLVM with multithreading
2039 .. _startmultithreaded:
2041 Entering and Exiting Multithreaded Mode
2042 ---------------------------------------
2044 In order to properly protect its internal data structures while avoiding
2045 excessive locking overhead in the single-threaded case, the LLVM must intialize
2046 certain data structures necessary to provide guards around its internals. To do
2047 so, the client program must invoke ``llvm_start_multithreaded()`` before making
2048 any concurrent LLVM API calls. To subsequently tear down these structures, use
2049 the ``llvm_stop_multithreaded()`` call. You can also use the
2050 ``llvm_is_multithreaded()`` call to check the status of multithreaded mode.
2052 Note that both of these calls must be made *in isolation*. That is to say that
2053 no other LLVM API calls may be executing at any time during the execution of
2054 ``llvm_start_multithreaded()`` or ``llvm_stop_multithreaded``. It's is the
2055 client's responsibility to enforce this isolation.
2057 The return value of ``llvm_start_multithreaded()`` indicates the success or
2058 failure of the initialization. Failure typically indicates that your copy of
2059 LLVM was built without multithreading support, typically because GCC atomic
2060 intrinsics were not found in your system compiler. In this case, the LLVM API
2061 will not be safe for concurrent calls. However, it *will* be safe for hosting
2062 threaded applications in the JIT, though :ref:`care must be taken
2063 <jitthreading>` to ensure that side exits and the like do not accidentally
2064 result in concurrent LLVM API calls.
2068 Ending Execution with ``llvm_shutdown()``
2069 -----------------------------------------
2071 When you are done using the LLVM APIs, you should call ``llvm_shutdown()`` to
2072 deallocate memory used for internal structures. This will also invoke
2073 ``llvm_stop_multithreaded()`` if LLVM is operating in multithreaded mode. As
2074 such, ``llvm_shutdown()`` requires the same isolation guarantees as
2075 ``llvm_stop_multithreaded()``.
2077 Note that, if you use scope-based shutdown, you can use the
2078 ``llvm_shutdown_obj`` class, which calls ``llvm_shutdown()`` in its destructor.
2082 Lazy Initialization with ``ManagedStatic``
2083 ------------------------------------------
2085 ``ManagedStatic`` is a utility class in LLVM used to implement static
2086 initialization of static resources, such as the global type tables. Before the
2087 invocation of ``llvm_shutdown()``, it implements a simple lazy initialization
2088 scheme. Once ``llvm_start_multithreaded()`` returns, however, it uses
2089 double-checked locking to implement thread-safe lazy initialization.
2091 Note that, because no other threads are allowed to issue LLVM API calls before
2092 ``llvm_start_multithreaded()`` returns, it is possible to have
2093 ``ManagedStatic``\ s of ``llvm::sys::Mutex``\ s.
2095 The ``llvm_acquire_global_lock()`` and ``llvm_release_global_lock`` APIs provide
2096 access to the global lock used to implement the double-checked locking for lazy
2097 initialization. These should only be used internally to LLVM, and only if you
2098 know what you're doing!
2102 Achieving Isolation with ``LLVMContext``
2103 ----------------------------------------
2105 ``LLVMContext`` is an opaque class in the LLVM API which clients can use to
2106 operate multiple, isolated instances of LLVM concurrently within the same
2107 address space. For instance, in a hypothetical compile-server, the compilation
2108 of an individual translation unit is conceptually independent from all the
2109 others, and it would be desirable to be able to compile incoming translation
2110 units concurrently on independent server threads. Fortunately, ``LLVMContext``
2111 exists to enable just this kind of scenario!
2113 Conceptually, ``LLVMContext`` provides isolation. Every LLVM entity
2114 (``Module``\ s, ``Value``\ s, ``Type``\ s, ``Constant``\ s, etc.) in LLVM's
2115 in-memory IR belongs to an ``LLVMContext``. Entities in different contexts
2116 *cannot* interact with each other: ``Module``\ s in different contexts cannot be
2117 linked together, ``Function``\ s cannot be added to ``Module``\ s in different
2118 contexts, etc. What this means is that is is safe to compile on multiple
2119 threads simultaneously, as long as no two threads operate on entities within the
2122 In practice, very few places in the API require the explicit specification of a
2123 ``LLVMContext``, other than the ``Type`` creation/lookup APIs. Because every
2124 ``Type`` carries a reference to its owning context, most other entities can
2125 determine what context they belong to by looking at their own ``Type``. If you
2126 are adding new entities to LLVM IR, please try to maintain this interface
2129 For clients that do *not* require the benefits of isolation, LLVM provides a
2130 convenience API ``getGlobalContext()``. This returns a global, lazily
2131 initialized ``LLVMContext`` that may be used in situations where isolation is
2139 LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple
2140 threads can call ``ExecutionEngine::getPointerToFunction()`` or
2141 ``ExecutionEngine::runFunction()`` concurrently, and multiple threads can run
2142 code output by the JIT concurrently. The user must still ensure that only one
2143 thread accesses IR in a given ``LLVMContext`` while another thread might be
2144 modifying it. One way to do that is to always hold the JIT lock while accessing
2145 IR outside the JIT (the JIT *modifies* the IR by adding ``CallbackVH``\ s).
2146 Another way is to only call ``getPointerToFunction()`` from the
2147 ``LLVMContext``'s thread.
2149 When the JIT is configured to compile lazily (using
2150 ``ExecutionEngine::DisableLazyCompilation(false)``), there is currently a `race
2151 condition <http://llvm.org/bugs/show_bug.cgi?id=5184>`_ in updating call sites
2152 after a function is lazily-jitted. It's still possible to use the lazy JIT in a
2153 threaded program if you ensure that only one thread at a time can call any
2154 particular lazy stub and that the JIT lock guards any IR access, but we suggest
2155 using only the eager JIT in threaded programs.
2162 This section describes some of the advanced or obscure API's that most clients
2163 do not need to be aware of. These API's tend manage the inner workings of the
2164 LLVM system, and only need to be accessed in unusual circumstances.
2168 The ``ValueSymbolTable`` class
2169 ------------------------------
2171 The ``ValueSymbolTable`` (`doxygen
2172 <http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html>`__) class provides
2173 a symbol table that the :ref:`Function <c_Function>` and Module_ classes use for
2174 naming value definitions. The symbol table can provide a name for any Value_.
2176 Note that the ``SymbolTable`` class should not be directly accessed by most
2177 clients. It should only be used when iteration over the symbol table names
2178 themselves are required, which is very special purpose. Note that not all LLVM
2179 Value_\ s have names, and those without names (i.e. they have an empty name) do
2180 not exist in the symbol table.
2182 Symbol tables support iteration over the values in the symbol table with
2183 ``begin/end/iterator`` and supports querying to see if a specific name is in the
2184 symbol table (with ``lookup``). The ``ValueSymbolTable`` class exposes no
2185 public mutator methods, instead, simply call ``setName`` on a value, which will
2186 autoinsert it into the appropriate symbol table.
2190 The ``User`` and owned ``Use`` classes' memory layout
2191 -----------------------------------------------------
2193 The ``User`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1User.html>`__)
2194 class provides a basis for expressing the ownership of ``User`` towards other
2195 `Value instance <http://llvm.org/doxygen/classllvm_1_1Value.html>`_\ s. The
2196 ``Use`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1Use.html>`__) helper
2197 class is employed to do the bookkeeping and to facilitate *O(1)* addition and
2202 Interaction and relationship between ``User`` and ``Use`` objects
2203 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2205 A subclass of ``User`` can choose between incorporating its ``Use`` objects or
2206 refer to them out-of-line by means of a pointer. A mixed variant (some ``Use``
2207 s inline others hung off) is impractical and breaks the invariant that the
2208 ``Use`` objects belonging to the same ``User`` form a contiguous array.
2210 We have 2 different layouts in the ``User`` (sub)classes:
2214 The ``Use`` object(s) are inside (resp. at fixed offset) of the ``User``
2215 object and there are a fixed number of them.
2219 The ``Use`` object(s) are referenced by a pointer to an array from the
2220 ``User`` object and there may be a variable number of them.
2222 As of v2.4 each layout still possesses a direct pointer to the start of the
2223 array of ``Use``\ s. Though not mandatory for layout a), we stick to this
2224 redundancy for the sake of simplicity. The ``User`` object also stores the
2225 number of ``Use`` objects it has. (Theoretically this information can also be
2226 calculated given the scheme presented below.)
2228 Special forms of allocation operators (``operator new``) enforce the following
2231 * Layout a) is modelled by prepending the ``User`` object by the ``Use[]``
2234 .. code-block:: none
2236 ...---.---.---.---.-------...
2237 | P | P | P | P | User
2238 '''---'---'---'---'-------'''
2240 * Layout b) is modelled by pointing at the ``Use[]`` array.
2242 .. code-block:: none
2253 *(In the above figures* '``P``' *stands for the* ``Use**`` *that is stored in
2254 each* ``Use`` *object in the member* ``Use::Prev`` *)*
2258 The waymarking algorithm
2259 ^^^^^^^^^^^^^^^^^^^^^^^^
2261 Since the ``Use`` objects are deprived of the direct (back)pointer to their
2262 ``User`` objects, there must be a fast and exact method to recover it. This is
2263 accomplished by the following scheme:
2265 A bit-encoding in the 2 LSBits (least significant bits) of the ``Use::Prev``
2266 allows to find the start of the ``User`` object:
2268 * ``00`` --- binary digit 0
2270 * ``01`` --- binary digit 1
2272 * ``10`` --- stop and calculate (``s``)
2274 * ``11`` --- full stop (``S``)
2276 Given a ``Use*``, all we have to do is to walk till we get a stop and we either
2277 have a ``User`` immediately behind or we have to walk to the next stop picking
2278 up digits and calculating the offset:
2280 .. code-block:: none
2282 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2283 | 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
2284 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2285 |+15 |+10 |+6 |+3 |+1
2288 | | | ______________________>
2289 | | ______________________________________>
2290 | __________________________________________________________>
2292 Only the significant number of bits need to be stored between the stops, so that
2293 the *worst case is 20 memory accesses* when there are 1000 ``Use`` objects
2294 associated with a ``User``.
2298 Reference implementation
2299 ^^^^^^^^^^^^^^^^^^^^^^^^
2301 The following literate Haskell fragment demonstrates the concept:
2303 .. code-block:: haskell
2305 > import Test.QuickCheck
2307 > digits :: Int -> [Char] -> [Char]
2308 > digits 0 acc = '0' : acc
2309 > digits 1 acc = '1' : acc
2310 > digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
2312 > dist :: Int -> [Char] -> [Char]
2315 > dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
2316 > dist n acc = dist (n - 1) $ dist 1 acc
2318 > takeLast n ss = reverse $ take n $ reverse ss
2320 > test = takeLast 40 $ dist 20 []
2323 Printing <test> gives: ``"1s100000s11010s10100s1111s1010s110s11s1S"``
2325 The reverse algorithm computes the length of the string just by examining a
2328 .. code-block:: haskell
2330 > pref :: [Char] -> Int
2332 > pref ('s':'1':rest) = decode 2 1 rest
2333 > pref (_:rest) = 1 + pref rest
2335 > decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
2336 > decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
2337 > decode walk acc _ = walk + acc
2340 Now, as expected, printing <pref test> gives ``40``.
2342 We can *quickCheck* this with following property:
2344 .. code-block:: haskell
2346 > testcase = dist 2000 []
2347 > testcaseLength = length testcase
2349 > identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
2350 > where arr = takeLast n testcase
2353 As expected <quickCheck identityProp> gives:
2357 *Main> quickCheck identityProp
2358 OK, passed 100 tests.
2360 Let's be a bit more exhaustive:
2362 .. code-block:: haskell
2365 > deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
2368 And here is the result of <deepCheck identityProp>:
2372 *Main> deepCheck identityProp
2373 OK, passed 500 tests.
2377 Tagging considerations
2378 ^^^^^^^^^^^^^^^^^^^^^^
2380 To maintain the invariant that the 2 LSBits of each ``Use**`` in ``Use`` never
2381 change after being set up, setters of ``Use::Prev`` must re-tag the new
2382 ``Use**`` on every modification. Accordingly getters must strip the tag bits.
2384 For layout b) instead of the ``User`` we find a pointer (``User*`` with LSBit
2385 set). Following this pointer brings us to the ``User``. A portable trick
2386 ensures that the first bytes of ``User`` (if interpreted as a pointer) never has
2387 the LSBit set. (Portability is relying on the fact that all known compilers
2388 place the ``vptr`` in the first word of the instances.)
2392 The Core LLVM Class Hierarchy Reference
2393 =======================================
2395 ``#include "llvm/Type.h"``
2397 header source: `Type.h <http://llvm.org/doxygen/Type_8h-source.html>`_
2399 doxygen info: `Type Clases <http://llvm.org/doxygen/classllvm_1_1Type.html>`_
2401 The Core LLVM classes are the primary means of representing the program being
2402 inspected or transformed. The core LLVM classes are defined in header files in
2403 the ``include/llvm/`` directory, and implemented in the ``lib/VMCore``
2408 The Type class and Derived Types
2409 --------------------------------
2411 ``Type`` is a superclass of all type classes. Every ``Value`` has a ``Type``.
2412 ``Type`` cannot be instantiated directly but only through its subclasses.
2413 Certain primitive types (``VoidType``, ``LabelType``, ``FloatType`` and
2414 ``DoubleType``) have hidden subclasses. They are hidden because they offer no
2415 useful functionality beyond what the ``Type`` class offers except to distinguish
2416 themselves from other subclasses of ``Type``.
2418 All other types are subclasses of ``DerivedType``. Types can be named, but this
2419 is not a requirement. There exists exactly one instance of a given shape at any
2420 one time. This allows type equality to be performed with address equality of
2421 the Type Instance. That is, given two ``Type*`` values, the types are identical
2422 if the pointers are identical.
2426 Important Public Methods
2427 ^^^^^^^^^^^^^^^^^^^^^^^^
2429 * ``bool isIntegerTy() const``: Returns true for any integer type.
2431 * ``bool isFloatingPointTy()``: Return true if this is one of the five
2432 floating point types.
2434 * ``bool isSized()``: Return true if the type has known size. Things
2435 that don't have a size are abstract types, labels and void.
2439 Important Derived Types
2440 ^^^^^^^^^^^^^^^^^^^^^^^
2443 Subclass of DerivedType that represents integer types of any bit width. Any
2444 bit width between ``IntegerType::MIN_INT_BITS`` (1) and
2445 ``IntegerType::MAX_INT_BITS`` (~8 million) can be represented.
2447 * ``static const IntegerType* get(unsigned NumBits)``: get an integer
2448 type of a specific bit width.
2450 * ``unsigned getBitWidth() const``: Get the bit width of an integer type.
2453 This is subclassed by ArrayType, PointerType and VectorType.
2455 * ``const Type * getElementType() const``: Returns the type of each
2456 of the elements in the sequential type.
2459 This is a subclass of SequentialType and defines the interface for array
2462 * ``unsigned getNumElements() const``: Returns the number of elements
2466 Subclass of SequentialType for pointer types.
2469 Subclass of SequentialType for vector types. A vector type is similar to an
2470 ArrayType but is distinguished because it is a first class type whereas
2471 ArrayType is not. Vector types are used for vector operations and are usually
2472 small vectors of of an integer or floating point type.
2475 Subclass of DerivedTypes for struct types.
2480 Subclass of DerivedTypes for function types.
2482 * ``bool isVarArg() const``: Returns true if it's a vararg function.
2484 * ``const Type * getReturnType() const``: Returns the return type of the
2487 * ``const Type * getParamType (unsigned i)``: Returns the type of the ith
2490 * ``const unsigned getNumParams() const``: Returns the number of formal
2495 The ``Module`` class
2496 --------------------
2498 ``#include "llvm/Module.h"``
2500 header source: `Module.h <http://llvm.org/doxygen/Module_8h-source.html>`_
2502 doxygen info: `Module Class <http://llvm.org/doxygen/classllvm_1_1Module.html>`_
2504 The ``Module`` class represents the top level structure present in LLVM
2505 programs. An LLVM module is effectively either a translation unit of the
2506 original program or a combination of several translation units merged by the
2507 linker. The ``Module`` class keeps track of a list of :ref:`Function
2508 <c_Function>`\ s, a list of GlobalVariable_\ s, and a SymbolTable_.
2509 Additionally, it contains a few helpful member functions that try to make common
2514 Important Public Members of the ``Module`` class
2515 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2517 * ``Module::Module(std::string name = "")``
2519 Constructing a Module_ is easy. You can optionally provide a name for it
2520 (probably based on the name of the translation unit).
2522 * | ``Module::iterator`` - Typedef for function list iterator
2523 | ``Module::const_iterator`` - Typedef for const_iterator.
2524 | ``begin()``, ``end()``, ``size()``, ``empty()``
2526 These are forwarding methods that make it easy to access the contents of a
2527 ``Module`` object's :ref:`Function <c_Function>` list.
2529 * ``Module::FunctionListType &getFunctionList()``
2531 Returns the list of :ref:`Function <c_Function>`\ s. This is necessary to use
2532 when you need to update the list or perform a complex action that doesn't have
2533 a forwarding method.
2537 * | ``Module::global_iterator`` - Typedef for global variable list iterator
2538 | ``Module::const_global_iterator`` - Typedef for const_iterator.
2539 | ``global_begin()``, ``global_end()``, ``global_size()``, ``global_empty()``
2541 These are forwarding methods that make it easy to access the contents of a
2542 ``Module`` object's GlobalVariable_ list.
2544 * ``Module::GlobalListType &getGlobalList()``
2546 Returns the list of GlobalVariable_\ s. This is necessary to use when you
2547 need to update the list or perform a complex action that doesn't have a
2552 * ``SymbolTable *getSymbolTable()``
2554 Return a reference to the SymbolTable_ for this ``Module``.
2558 * ``Function *getFunction(StringRef Name) const``
2560 Look up the specified function in the ``Module`` SymbolTable_. If it does not
2561 exist, return ``null``.
2563 * ``Function *getOrInsertFunction(const std::string &Name, const FunctionType
2566 Look up the specified function in the ``Module`` SymbolTable_. If it does not
2567 exist, add an external declaration for the function and return it.
2569 * ``std::string getTypeName(const Type *Ty)``
2571 If there is at least one entry in the SymbolTable_ for the specified Type_,
2572 return it. Otherwise return the empty string.
2574 * ``bool addTypeName(const std::string &Name, const Type *Ty)``
2576 Insert an entry in the SymbolTable_ mapping ``Name`` to ``Ty``. If there is
2577 already an entry for this name, true is returned and the SymbolTable_ is not
2585 ``#include "llvm/Value.h"``
2587 header source: `Value.h <http://llvm.org/doxygen/Value_8h-source.html>`_
2589 doxygen info: `Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_
2591 The ``Value`` class is the most important class in the LLVM Source base. It
2592 represents a typed value that may be used (among other things) as an operand to
2593 an instruction. There are many different types of ``Value``\ s, such as
2594 Constant_\ s, Argument_\ s. Even Instruction_\ s and :ref:`Function
2595 <c_Function>`\ s are ``Value``\ s.
2597 A particular ``Value`` may be used many times in the LLVM representation for a
2598 program. For example, an incoming argument to a function (represented with an
2599 instance of the Argument_ class) is "used" by every instruction in the function
2600 that references the argument. To keep track of this relationship, the ``Value``
2601 class keeps a list of all of the ``User``\ s that is using it (the User_ class
2602 is a base class for all nodes in the LLVM graph that can refer to ``Value``\ s).
2603 This use list is how LLVM represents def-use information in the program, and is
2604 accessible through the ``use_*`` methods, shown below.
2606 Because LLVM is a typed representation, every LLVM ``Value`` is typed, and this
2607 Type_ is available through the ``getType()`` method. In addition, all LLVM
2608 values can be named. The "name" of the ``Value`` is a symbolic string printed
2611 .. code-block:: llvm
2617 The name of this instruction is "foo". **NOTE** that the name of any value may
2618 be missing (an empty string), so names should **ONLY** be used for debugging
2619 (making the source code easier to read, debugging printouts), they should not be
2620 used to keep track of values or map between them. For this purpose, use a
2621 ``std::map`` of pointers to the ``Value`` itself instead.
2623 One important aspect of LLVM is that there is no distinction between an SSA
2624 variable and the operation that produces it. Because of this, any reference to
2625 the value produced by an instruction (or the value available as an incoming
2626 argument, for example) is represented as a direct pointer to the instance of the
2627 class that represents this value. Although this may take some getting used to,
2628 it simplifies the representation and makes it easier to manipulate.
2632 Important Public Members of the ``Value`` class
2633 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2635 * | ``Value::use_iterator`` - Typedef for iterator over the use-list
2636 | ``Value::const_use_iterator`` - Typedef for const_iterator over the
2638 | ``unsigned use_size()`` - Returns the number of users of the value.
2639 | ``bool use_empty()`` - Returns true if there are no users.
2640 | ``use_iterator use_begin()`` - Get an iterator to the start of the
2642 | ``use_iterator use_end()`` - Get an iterator to the end of the use-list.
2643 | ``User *use_back()`` - Returns the last element in the list.
2645 These methods are the interface to access the def-use information in LLVM.
2646 As with all other iterators in LLVM, the naming conventions follow the
2647 conventions defined by the STL_.
2649 * ``Type *getType() const``
2650 This method returns the Type of the Value.
2652 * | ``bool hasName() const``
2653 | ``std::string getName() const``
2654 | ``void setName(const std::string &Name)``
2656 This family of methods is used to access and assign a name to a ``Value``, be
2657 aware of the :ref:`precaution above <nameWarning>`.
2659 * ``void replaceAllUsesWith(Value *V)``
2661 This method traverses the use list of a ``Value`` changing all User_\ s of the
2662 current value to refer to "``V``" instead. For example, if you detect that an
2663 instruction always produces a constant value (for example through constant
2664 folding), you can replace all uses of the instruction with the constant like
2669 Inst->replaceAllUsesWith(ConstVal);
2676 ``#include "llvm/User.h"``
2678 header source: `User.h <http://llvm.org/doxygen/User_8h-source.html>`_
2680 doxygen info: `User Class <http://llvm.org/doxygen/classllvm_1_1User.html>`_
2684 The ``User`` class is the common base class of all LLVM nodes that may refer to
2685 ``Value``\ s. It exposes a list of "Operands" that are all of the ``Value``\ s
2686 that the User is referring to. The ``User`` class itself is a subclass of
2689 The operands of a ``User`` point directly to the LLVM ``Value`` that it refers
2690 to. Because LLVM uses Static Single Assignment (SSA) form, there can only be
2691 one definition referred to, allowing this direct connection. This connection
2692 provides the use-def information in LLVM.
2696 Important Public Members of the ``User`` class
2697 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2699 The ``User`` class exposes the operand list in two ways: through an index access
2700 interface and through an iterator based interface.
2702 * | ``Value *getOperand(unsigned i)``
2703 | ``unsigned getNumOperands()``
2705 These two methods expose the operands of the ``User`` in a convenient form for
2708 * | ``User::op_iterator`` - Typedef for iterator over the operand list
2709 | ``op_iterator op_begin()`` - Get an iterator to the start of the operand
2711 | ``op_iterator op_end()`` - Get an iterator to the end of the operand list.
2713 Together, these methods make up the iterator based interface to the operands
2719 The ``Instruction`` class
2720 -------------------------
2722 ``#include "llvm/Instruction.h"``
2724 header source: `Instruction.h
2725 <http://llvm.org/doxygen/Instruction_8h-source.html>`_
2727 doxygen info: `Instruction Class
2728 <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_
2730 Superclasses: User_, Value_
2732 The ``Instruction`` class is the common base class for all LLVM instructions.
2733 It provides only a few methods, but is a very commonly used class. The primary
2734 data tracked by the ``Instruction`` class itself is the opcode (instruction
2735 type) and the parent BasicBlock_ the ``Instruction`` is embedded into. To
2736 represent a specific type of instruction, one of many subclasses of
2737 ``Instruction`` are used.
2739 Because the ``Instruction`` class subclasses the User_ class, its operands can
2740 be accessed in the same way as for other ``User``\ s (with the
2741 ``getOperand()``/``getNumOperands()`` and ``op_begin()``/``op_end()`` methods).
2742 An important file for the ``Instruction`` class is the ``llvm/Instruction.def``
2743 file. This file contains some meta-data about the various different types of
2744 instructions in LLVM. It describes the enum values that are used as opcodes
2745 (for example ``Instruction::Add`` and ``Instruction::ICmp``), as well as the
2746 concrete sub-classes of ``Instruction`` that implement the instruction (for
2747 example BinaryOperator_ and CmpInst_). Unfortunately, the use of macros in this
2748 file confuses doxygen, so these enum values don't show up correctly in the
2749 `doxygen output <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_.
2753 Important Subclasses of the ``Instruction`` class
2754 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2758 * ``BinaryOperator``
2760 This subclasses represents all two operand instructions whose operands must be
2761 the same type, except for the comparison instructions.
2766 This subclass is the parent of the 12 casting instructions. It provides
2767 common operations on cast instructions.
2773 This subclass respresents the two comparison instructions,
2774 `ICmpInst <LangRef.html#i_icmp>`_ (integer opreands), and
2775 `FCmpInst <LangRef.html#i_fcmp>`_ (floating point operands).
2779 * ``TerminatorInst``
2781 This subclass is the parent of all terminator instructions (those which can
2786 Important Public Members of the ``Instruction`` class
2787 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2789 * ``BasicBlock *getParent()``
2791 Returns the BasicBlock_ that this
2792 ``Instruction`` is embedded into.
2794 * ``bool mayWriteToMemory()``
2796 Returns true if the instruction writes to memory, i.e. it is a ``call``,
2797 ``free``, ``invoke``, or ``store``.
2799 * ``unsigned getOpcode()``
2801 Returns the opcode for the ``Instruction``.
2803 * ``Instruction *clone() const``
2805 Returns another instance of the specified instruction, identical in all ways
2806 to the original except that the instruction has no parent (i.e. it's not
2807 embedded into a BasicBlock_), and it has no name.
2811 The ``Constant`` class and subclasses
2812 -------------------------------------
2814 Constant represents a base class for different types of constants. It is
2815 subclassed by ConstantInt, ConstantArray, etc. for representing the various
2816 types of Constants. GlobalValue_ is also a subclass, which represents the
2817 address of a global variable or function.
2821 Important Subclasses of Constant
2822 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2824 * ConstantInt : This subclass of Constant represents an integer constant of
2827 * ``const APInt& getValue() const``: Returns the underlying
2828 value of this constant, an APInt value.
2830 * ``int64_t getSExtValue() const``: Converts the underlying APInt value to an
2831 int64_t via sign extension. If the value (not the bit width) of the APInt
2832 is too large to fit in an int64_t, an assertion will result. For this
2833 reason, use of this method is discouraged.
2835 * ``uint64_t getZExtValue() const``: Converts the underlying APInt value
2836 to a uint64_t via zero extension. IF the value (not the bit width) of the
2837 APInt is too large to fit in a uint64_t, an assertion will result. For this
2838 reason, use of this method is discouraged.
2840 * ``static ConstantInt* get(const APInt& Val)``: Returns the ConstantInt
2841 object that represents the value provided by ``Val``. The type is implied
2842 as the IntegerType that corresponds to the bit width of ``Val``.
2844 * ``static ConstantInt* get(const Type *Ty, uint64_t Val)``: Returns the
2845 ConstantInt object that represents the value provided by ``Val`` for integer
2848 * ConstantFP : This class represents a floating point constant.
2850 * ``double getValue() const``: Returns the underlying value of this constant.
2852 * ConstantArray : This represents a constant array.
2854 * ``const std::vector<Use> &getValues() const``: Returns a vector of
2855 component constants that makeup this array.
2857 * ConstantStruct : This represents a constant struct.
2859 * ``const std::vector<Use> &getValues() const``: Returns a vector of
2860 component constants that makeup this array.
2862 * GlobalValue : This represents either a global variable or a function. In
2863 either case, the value is a constant fixed address (after linking).
2867 The ``GlobalValue`` class
2868 -------------------------
2870 ``#include "llvm/GlobalValue.h"``
2872 header source: `GlobalValue.h
2873 <http://llvm.org/doxygen/GlobalValue_8h-source.html>`_
2875 doxygen info: `GlobalValue Class
2876 <http://llvm.org/doxygen/classllvm_1_1GlobalValue.html>`_
2878 Superclasses: Constant_, User_, Value_
2880 Global values ( GlobalVariable_\ s or :ref:`Function <c_Function>`\ s) are the
2881 only LLVM values that are visible in the bodies of all :ref:`Function
2882 <c_Function>`\ s. Because they are visible at global scope, they are also
2883 subject to linking with other globals defined in different translation units.
2884 To control the linking process, ``GlobalValue``\ s know their linkage rules.
2885 Specifically, ``GlobalValue``\ s know whether they have internal or external
2886 linkage, as defined by the ``LinkageTypes`` enumeration.
2888 If a ``GlobalValue`` has internal linkage (equivalent to being ``static`` in C),
2889 it is not visible to code outside the current translation unit, and does not
2890 participate in linking. If it has external linkage, it is visible to external
2891 code, and does participate in linking. In addition to linkage information,
2892 ``GlobalValue``\ s keep track of which Module_ they are currently part of.
2894 Because ``GlobalValue``\ s are memory objects, they are always referred to by
2895 their **address**. As such, the Type_ of a global is always a pointer to its
2896 contents. It is important to remember this when using the ``GetElementPtrInst``
2897 instruction because this pointer must be dereferenced first. For example, if
2898 you have a ``GlobalVariable`` (a subclass of ``GlobalValue)`` that is an array
2899 of 24 ints, type ``[24 x i32]``, then the ``GlobalVariable`` is a pointer to
2900 that array. Although the address of the first element of this array and the
2901 value of the ``GlobalVariable`` are the same, they have different types. The
2902 ``GlobalVariable``'s type is ``[24 x i32]``. The first element's type is
2903 ``i32.`` Because of this, accessing a global value requires you to dereference
2904 the pointer with ``GetElementPtrInst`` first, then its elements can be accessed.
2905 This is explained in the `LLVM Language Reference Manual
2906 <LangRef.html#globalvars>`_.
2910 Important Public Members of the ``GlobalValue`` class
2911 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2913 * | ``bool hasInternalLinkage() const``
2914 | ``bool hasExternalLinkage() const``
2915 | ``void setInternalLinkage(bool HasInternalLinkage)``
2917 These methods manipulate the linkage characteristics of the ``GlobalValue``.
2919 * ``Module *getParent()``
2921 This returns the Module_ that the
2922 GlobalValue is currently embedded into.
2926 The ``Function`` class
2927 ----------------------
2929 ``#include "llvm/Function.h"``
2931 header source: `Function.h <http://llvm.org/doxygen/Function_8h-source.html>`_
2933 doxygen info: `Function Class
2934 <http://llvm.org/doxygen/classllvm_1_1Function.html>`_
2936 Superclasses: GlobalValue_, Constant_, User_, Value_
2938 The ``Function`` class represents a single procedure in LLVM. It is actually
2939 one of the more complex classes in the LLVM hierarchy because it must keep track
2940 of a large amount of data. The ``Function`` class keeps track of a list of
2941 BasicBlock_\ s, a list of formal Argument_\ s, and a SymbolTable_.
2943 The list of BasicBlock_\ s is the most commonly used part of ``Function``
2944 objects. The list imposes an implicit ordering of the blocks in the function,
2945 which indicate how the code will be laid out by the backend. Additionally, the
2946 first BasicBlock_ is the implicit entry node for the ``Function``. It is not
2947 legal in LLVM to explicitly branch to this initial block. There are no implicit
2948 exit nodes, and in fact there may be multiple exit nodes from a single
2949 ``Function``. If the BasicBlock_ list is empty, this indicates that the
2950 ``Function`` is actually a function declaration: the actual body of the function
2951 hasn't been linked in yet.
2953 In addition to a list of BasicBlock_\ s, the ``Function`` class also keeps track
2954 of the list of formal Argument_\ s that the function receives. This container
2955 manages the lifetime of the Argument_ nodes, just like the BasicBlock_ list does
2956 for the BasicBlock_\ s.
2958 The SymbolTable_ is a very rarely used LLVM feature that is only used when you
2959 have to look up a value by name. Aside from that, the SymbolTable_ is used
2960 internally to make sure that there are not conflicts between the names of
2961 Instruction_\ s, BasicBlock_\ s, or Argument_\ s in the function body.
2963 Note that ``Function`` is a GlobalValue_ and therefore also a Constant_. The
2964 value of the function is its address (after linking) which is guaranteed to be
2969 Important Public Members of the ``Function``
2970 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2972 * ``Function(const FunctionType *Ty, LinkageTypes Linkage,
2973 const std::string &N = "", Module* Parent = 0)``
2975 Constructor used when you need to create new ``Function``\ s to add the
2976 program. The constructor must specify the type of the function to create and
2977 what type of linkage the function should have. The FunctionType_ argument
2978 specifies the formal arguments and return value for the function. The same
2979 FunctionType_ value can be used to create multiple functions. The ``Parent``
2980 argument specifies the Module in which the function is defined. If this
2981 argument is provided, the function will automatically be inserted into that
2982 module's list of functions.
2984 * ``bool isDeclaration()``
2986 Return whether or not the ``Function`` has a body defined. If the function is
2987 "external", it does not have a body, and thus must be resolved by linking with
2988 a function defined in a different translation unit.
2990 * | ``Function::iterator`` - Typedef for basic block list iterator
2991 | ``Function::const_iterator`` - Typedef for const_iterator.
2992 | ``begin()``, ``end()``, ``size()``, ``empty()``
2994 These are forwarding methods that make it easy to access the contents of a
2995 ``Function`` object's BasicBlock_ list.
2997 * ``Function::BasicBlockListType &getBasicBlockList()``
2999 Returns the list of BasicBlock_\ s. This is necessary to use when you need to
3000 update the list or perform a complex action that doesn't have a forwarding
3003 * | ``Function::arg_iterator`` - Typedef for the argument list iterator
3004 | ``Function::const_arg_iterator`` - Typedef for const_iterator.
3005 | ``arg_begin()``, ``arg_end()``, ``arg_size()``, ``arg_empty()``
3007 These are forwarding methods that make it easy to access the contents of a
3008 ``Function`` object's Argument_ list.
3010 * ``Function::ArgumentListType &getArgumentList()``
3012 Returns the list of Argument_. This is necessary to use when you need to
3013 update the list or perform a complex action that doesn't have a forwarding
3016 * ``BasicBlock &getEntryBlock()``
3018 Returns the entry ``BasicBlock`` for the function. Because the entry block
3019 for the function is always the first block, this returns the first block of
3022 * | ``Type *getReturnType()``
3023 | ``FunctionType *getFunctionType()``
3025 This traverses the Type_ of the ``Function`` and returns the return type of
3026 the function, or the FunctionType_ of the actual function.
3028 * ``SymbolTable *getSymbolTable()``
3030 Return a pointer to the SymbolTable_ for this ``Function``.
3034 The ``GlobalVariable`` class
3035 ----------------------------
3037 ``#include "llvm/GlobalVariable.h"``
3039 header source: `GlobalVariable.h
3040 <http://llvm.org/doxygen/GlobalVariable_8h-source.html>`_
3042 doxygen info: `GlobalVariable Class
3043 <http://llvm.org/doxygen/classllvm_1_1GlobalVariable.html>`_
3045 Superclasses: GlobalValue_, Constant_, User_, Value_
3047 Global variables are represented with the (surprise surprise) ``GlobalVariable``
3048 class. Like functions, ``GlobalVariable``\ s are also subclasses of
3049 GlobalValue_, and as such are always referenced by their address (global values
3050 must live in memory, so their "name" refers to their constant address). See
3051 GlobalValue_ for more on this. Global variables may have an initial value
3052 (which must be a Constant_), and if they have an initializer, they may be marked
3053 as "constant" themselves (indicating that their contents never change at
3056 .. _m_GlobalVariable:
3058 Important Public Members of the ``GlobalVariable`` class
3059 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3061 * ``GlobalVariable(const Type *Ty, bool isConstant, LinkageTypes &Linkage,
3062 Constant *Initializer = 0, const std::string &Name = "", Module* Parent = 0)``
3064 Create a new global variable of the specified type. If ``isConstant`` is true
3065 then the global variable will be marked as unchanging for the program. The
3066 Linkage parameter specifies the type of linkage (internal, external, weak,
3067 linkonce, appending) for the variable. If the linkage is InternalLinkage,
3068 WeakAnyLinkage, WeakODRLinkage, LinkOnceAnyLinkage or LinkOnceODRLinkage, then
3069 the resultant global variable will have internal linkage. AppendingLinkage
3070 concatenates together all instances (in different translation units) of the
3071 variable into a single variable but is only applicable to arrays. See the
3072 `LLVM Language Reference <LangRef.html#modulestructure>`_ for further details
3073 on linkage types. Optionally an initializer, a name, and the module to put
3074 the variable into may be specified for the global variable as well.
3076 * ``bool isConstant() const``
3078 Returns true if this is a global variable that is known not to be modified at
3081 * ``bool hasInitializer()``
3083 Returns true if this ``GlobalVariable`` has an intializer.
3085 * ``Constant *getInitializer()``
3087 Returns the initial value for a ``GlobalVariable``. It is not legal to call
3088 this method if there is no initializer.
3092 The ``BasicBlock`` class
3093 ------------------------
3095 ``#include "llvm/BasicBlock.h"``
3097 header source: `BasicBlock.h
3098 <http://llvm.org/doxygen/BasicBlock_8h-source.html>`_
3100 doxygen info: `BasicBlock Class
3101 <http://llvm.org/doxygen/classllvm_1_1BasicBlock.html>`_
3105 This class represents a single entry single exit section of the code, commonly
3106 known as a basic block by the compiler community. The ``BasicBlock`` class
3107 maintains a list of Instruction_\ s, which form the body of the block. Matching
3108 the language definition, the last element of this list of instructions is always
3109 a terminator instruction (a subclass of the TerminatorInst_ class).
3111 In addition to tracking the list of instructions that make up the block, the
3112 ``BasicBlock`` class also keeps track of the :ref:`Function <c_Function>` that
3113 it is embedded into.
3115 Note that ``BasicBlock``\ s themselves are Value_\ s, because they are
3116 referenced by instructions like branches and can go in the switch tables.
3117 ``BasicBlock``\ s have type ``label``.
3121 Important Public Members of the ``BasicBlock`` class
3122 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3124 * ``BasicBlock(const std::string &Name = "", Function *Parent = 0)``
3126 The ``BasicBlock`` constructor is used to create new basic blocks for
3127 insertion into a function. The constructor optionally takes a name for the
3128 new block, and a :ref:`Function <c_Function>` to insert it into. If the
3129 ``Parent`` parameter is specified, the new ``BasicBlock`` is automatically
3130 inserted at the end of the specified :ref:`Function <c_Function>`, if not
3131 specified, the BasicBlock must be manually inserted into the :ref:`Function
3134 * | ``BasicBlock::iterator`` - Typedef for instruction list iterator
3135 | ``BasicBlock::const_iterator`` - Typedef for const_iterator.
3136 | ``begin()``, ``end()``, ``front()``, ``back()``,
3137 ``size()``, ``empty()``
3138 STL-style functions for accessing the instruction list.
3140 These methods and typedefs are forwarding functions that have the same
3141 semantics as the standard library methods of the same names. These methods
3142 expose the underlying instruction list of a basic block in a way that is easy
3143 to manipulate. To get the full complement of container operations (including
3144 operations to update the list), you must use the ``getInstList()`` method.
3146 * ``BasicBlock::InstListType &getInstList()``
3148 This method is used to get access to the underlying container that actually
3149 holds the Instructions. This method must be used when there isn't a
3150 forwarding function in the ``BasicBlock`` class for the operation that you
3151 would like to perform. Because there are no forwarding functions for
3152 "updating" operations, you need to use this if you want to update the contents
3153 of a ``BasicBlock``.
3155 * ``Function *getParent()``
3157 Returns a pointer to :ref:`Function <c_Function>` the block is embedded into,
3158 or a null pointer if it is homeless.
3160 * ``TerminatorInst *getTerminator()``
3162 Returns a pointer to the terminator instruction that appears at the end of the
3163 ``BasicBlock``. If there is no terminator instruction, or if the last
3164 instruction in the block is not a terminator, then a null pointer is returned.
3168 The ``Argument`` class
3169 ----------------------
3171 This subclass of Value defines the interface for incoming formal arguments to a
3172 function. A Function maintains a list of its formal arguments. An argument has
3173 a pointer to the parent Function.