1 ========================
2 LLVM Programmer's Manual
3 ========================
9 This is always a work in progress.
16 This document is meant to highlight some of the important classes and interfaces
17 available in the LLVM source-base. This manual is not intended to explain what
18 LLVM is, how it works, and what LLVM code looks like. It assumes that you know
19 the basics of LLVM and are interested in writing transformations or otherwise
20 analyzing or manipulating the code.
22 This document should get you oriented so that you can find your way in the
23 continuously growing source code that makes up the LLVM infrastructure. Note
24 that this manual is not intended to serve as a replacement for reading the
25 source code, so if you think there should be a method in one of these classes to
26 do something, but it's not listed, check the source. Links to the `doxygen
27 <http://llvm.org/doxygen/>`__ sources are provided to make this as easy as
30 The first section of this document describes general information that is useful
31 to know when working in the LLVM infrastructure, and the second describes the
32 Core LLVM classes. In the future this manual will be extended with information
33 describing how to use extension libraries, such as dominator information, CFG
34 traversal routines, and useful utilities like the ``InstVisitor`` (`doxygen
35 <http://llvm.org/doxygen/InstVisitor_8h-source.html>`__) template.
42 This section contains general information that is useful if you are working in
43 the LLVM source-base, but that isn't specific to any particular API.
47 The C++ Standard Template Library
48 ---------------------------------
50 LLVM makes heavy use of the C++ Standard Template Library (STL), perhaps much
51 more than you are used to, or have seen before. Because of this, you might want
52 to do a little background reading in the techniques used and capabilities of the
53 library. There are many good pages that discuss the STL, and several books on
54 the subject that you can get, so it will not be discussed in this document.
56 Here are some useful links:
59 <http://en.cppreference.com/w/>`_ - an excellent
60 reference for the STL and other parts of the standard C++ library.
62 #. `C++ In a Nutshell <http://www.tempest-sw.com/cpp/>`_ - This is an O'Reilly
63 book in the making. It has a decent Standard Library Reference that rivals
64 Dinkumware's, and is unfortunately no longer free since the book has been
67 #. `C++ Frequently Asked Questions <http://www.parashift.com/c++-faq-lite/>`_.
69 #. `SGI's STL Programmer's Guide <http://www.sgi.com/tech/stl/>`_ - Contains a
70 useful `Introduction to the STL
71 <http://www.sgi.com/tech/stl/stl_introduction.html>`_.
73 #. `Bjarne Stroustrup's C++ Page
74 <http://www.research.att.com/%7Ebs/C++.html>`_.
76 #. `Bruce Eckel's Thinking in C++, 2nd ed. Volume 2 Revision 4.0
77 (even better, get the book)
78 <http://www.mindview.net/Books/TICPP/ThinkingInCPP2e.html>`_.
80 You are also encouraged to take a look at the :doc:`LLVM Coding Standards
81 <CodingStandards>` guide which focuses on how to write maintainable code more
82 than where to put your curly braces.
86 Other useful references
87 -----------------------
89 #. `Using static and shared libraries across platforms
90 <http://www.fortran-2000.com/ArnaudRecipes/sharedlib.html>`_
94 Important and useful LLVM APIs
95 ==============================
97 Here we highlight some LLVM APIs that are generally useful and good to know
98 about when writing transformations.
102 The ``isa<>``, ``cast<>`` and ``dyn_cast<>`` templates
103 ------------------------------------------------------
105 The LLVM source-base makes extensive use of a custom form of RTTI. These
106 templates have many similarities to the C++ ``dynamic_cast<>`` operator, but
107 they don't have some drawbacks (primarily stemming from the fact that
108 ``dynamic_cast<>`` only works on classes that have a v-table). Because they are
109 used so often, you must know what they do and how they work. All of these
110 templates are defined in the ``llvm/Support/Casting.h`` (`doxygen
111 <http://llvm.org/doxygen/Casting_8h-source.html>`__) file (note that you very
112 rarely have to include this file directly).
115 The ``isa<>`` operator works exactly like the Java "``instanceof``" operator.
116 It returns true or false depending on whether a reference or pointer points to
117 an instance of the specified class. This can be very useful for constraint
118 checking of various sorts (example below).
121 The ``cast<>`` operator is a "checked cast" operation. It converts a pointer
122 or reference from a base class to a derived class, causing an assertion
123 failure if it is not really an instance of the right type. This should be
124 used in cases where you have some information that makes you believe that
125 something is of the right type. An example of the ``isa<>`` and ``cast<>``
130 static bool isLoopInvariant(const Value *V, const Loop *L) {
131 if (isa<Constant>(V) || isa<Argument>(V) || isa<GlobalValue>(V))
134 // Otherwise, it must be an instruction...
135 return !L->contains(cast<Instruction>(V)->getParent());
138 Note that you should **not** use an ``isa<>`` test followed by a ``cast<>``,
139 for that use the ``dyn_cast<>`` operator.
142 The ``dyn_cast<>`` operator is a "checking cast" operation. It checks to see
143 if the operand is of the specified type, and if so, returns a pointer to it
144 (this operator does not work with references). If the operand is not of the
145 correct type, a null pointer is returned. Thus, this works very much like
146 the ``dynamic_cast<>`` operator in C++, and should be used in the same
147 circumstances. Typically, the ``dyn_cast<>`` operator is used in an ``if``
148 statement or some other flow control statement like this:
152 if (AllocationInst *AI = dyn_cast<AllocationInst>(Val)) {
156 This form of the ``if`` statement effectively combines together a call to
157 ``isa<>`` and a call to ``cast<>`` into one statement, which is very
160 Note that the ``dyn_cast<>`` operator, like C++'s ``dynamic_cast<>`` or Java's
161 ``instanceof`` operator, can be abused. In particular, you should not use big
162 chained ``if/then/else`` blocks to check for lots of different variants of
163 classes. If you find yourself wanting to do this, it is much cleaner and more
164 efficient to use the ``InstVisitor`` class to dispatch over the instruction
168 The ``cast_or_null<>`` operator works just like the ``cast<>`` operator,
169 except that it allows for a null pointer as an argument (which it then
170 propagates). This can sometimes be useful, allowing you to combine several
171 null checks into one.
173 ``dyn_cast_or_null<>``:
174 The ``dyn_cast_or_null<>`` operator works just like the ``dyn_cast<>``
175 operator, except that it allows for a null pointer as an argument (which it
176 then propagates). This can sometimes be useful, allowing you to combine
177 several null checks into one.
179 These five templates can be used with any classes, whether they have a v-table
180 or not. If you want to add support for these templates, see the document
181 :doc:`How to set up LLVM-style RTTI for your class hierarchy
182 <HowToSetUpLLVMStyleRTTI>`
186 Passing strings (the ``StringRef`` and ``Twine`` classes)
187 ---------------------------------------------------------
189 Although LLVM generally does not do much string manipulation, we do have several
190 important APIs which take strings. Two important examples are the Value class
191 -- which has names for instructions, functions, etc. -- and the ``StringMap``
192 class which is used extensively in LLVM and Clang.
194 These are generic classes, and they need to be able to accept strings which may
195 have embedded null characters. Therefore, they cannot simply take a ``const
196 char *``, and taking a ``const std::string&`` requires clients to perform a heap
197 allocation which is usually unnecessary. Instead, many LLVM APIs use a
198 ``StringRef`` or a ``const Twine&`` for passing strings efficiently.
202 The ``StringRef`` class
203 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
205 The ``StringRef`` data type represents a reference to a constant string (a
206 character array and a length) and supports the common operations available on
207 ``std::string``, but does not require heap allocation.
209 It can be implicitly constructed using a C style null-terminated string, an
210 ``std::string``, or explicitly with a character pointer and length. For
211 example, the ``StringRef`` find function is declared as:
215 iterator find(StringRef Key);
217 and clients can call it using any one of:
221 Map.find("foo"); // Lookup "foo"
222 Map.find(std::string("bar")); // Lookup "bar"
223 Map.find(StringRef("\0baz", 4)); // Lookup "\0baz"
225 Similarly, APIs which need to return a string may return a ``StringRef``
226 instance, which can be used directly or converted to an ``std::string`` using
227 the ``str`` member function. See ``llvm/ADT/StringRef.h`` (`doxygen
228 <http://llvm.org/doxygen/classllvm_1_1StringRef_8h-source.html>`__) for more
231 You should rarely use the ``StringRef`` class directly, because it contains
232 pointers to external memory it is not generally safe to store an instance of the
233 class (unless you know that the external storage will not be freed).
234 ``StringRef`` is small and pervasive enough in LLVM that it should always be
240 The ``Twine`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1Twine.html>`__)
241 class is an efficient way for APIs to accept concatenated strings. For example,
242 a common LLVM paradigm is to name one instruction based on the name of another
243 instruction with a suffix, for example:
247 New = CmpInst::Create(..., SO->getName() + ".cmp");
249 The ``Twine`` class is effectively a lightweight `rope
250 <http://en.wikipedia.org/wiki/Rope_(computer_science)>`_ which points to
251 temporary (stack allocated) objects. Twines can be implicitly constructed as
252 the result of the plus operator applied to strings (i.e., a C strings, an
253 ``std::string``, or a ``StringRef``). The twine delays the actual concatenation
254 of strings until it is actually required, at which point it can be efficiently
255 rendered directly into a character array. This avoids unnecessary heap
256 allocation involved in constructing the temporary results of string
257 concatenation. See ``llvm/ADT/Twine.h`` (`doxygen
258 <http://llvm.org/doxygen/Twine_8h_source.html>`__) and :ref:`here <dss_twine>`
259 for more information.
261 As with a ``StringRef``, ``Twine`` objects point to external memory and should
262 almost never be stored or mentioned directly. They are intended solely for use
263 when defining a function which should be able to efficiently accept concatenated
268 Passing functions and other callable objects
269 --------------------------------------------
271 Sometimes you may want a function to be passed a callback object. In order to
272 support lambda expressions and other function objects, you should not use the
273 traditional C approach of taking a function pointer and an opaque cookie:
277 void takeCallback(bool (*Callback)(Function *, void *), void *Cookie);
279 Instead, use one of the following approaches:
284 If you don't mind putting the definition of your function into a header file,
285 make it a function template that is templated on the callable type.
289 template<typename Callable>
290 void takeCallback(Callable Callback) {
294 The ``function_ref`` class template
295 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
298 (`doxygen <http://llvm.org/doxygen/classllvm_1_1function_ref.html>`__) class
299 template represents a reference to a callable object, templated over the type
300 of the callable. This is a good choice for passing a callback to a function,
301 if you don't need to hold onto the callback after the function returns.
303 ``function_ref<Ret(Param1, Param2, ...)>`` can be implicitly constructed from
304 any callable object that can be called with arguments of type ``Param1``,
305 ``Param2``, ..., and returns a value that can be converted to type ``Ret``.
310 void visitBasicBlocks(Function *F, function_ref<bool (BasicBlock*)> Callback) {
311 for (BasicBlock &BB : *F)
320 visitBasicBlocks(F, [&](BasicBlock *BB) {
326 Note that a ``function_ref`` object contains pointers to external memory, so
327 it is not generally safe to store an instance of the class (unless you know
328 that the external storage will not be freed).
329 ``function_ref`` is small enough that it should always be passed by value.
334 You cannot use ``std::function`` within LLVM code, because it is not supported
335 by all our target toolchains.
340 The ``DEBUG()`` macro and ``-debug`` option
341 -------------------------------------------
343 Often when working on your pass you will put a bunch of debugging printouts and
344 other code into your pass. After you get it working, you want to remove it, but
345 you may need it again in the future (to work out new bugs that you run across).
347 Naturally, because of this, you don't want to delete the debug printouts, but
348 you don't want them to always be noisy. A standard compromise is to comment
349 them out, allowing you to enable them if you need them in the future.
351 The ``llvm/Support/Debug.h`` (`doxygen
352 <http://llvm.org/doxygen/Debug_8h-source.html>`__) file provides a macro named
353 ``DEBUG()`` that is a much nicer solution to this problem. Basically, you can
354 put arbitrary code into the argument of the ``DEBUG`` macro, and it is only
355 executed if '``opt``' (or any other tool) is run with the '``-debug``' command
360 DEBUG(errs() << "I am here!\n");
362 Then you can run your pass like this:
366 $ opt < a.bc > /dev/null -mypass
368 $ opt < a.bc > /dev/null -mypass -debug
371 Using the ``DEBUG()`` macro instead of a home-brewed solution allows you to not
372 have to create "yet another" command line option for the debug output for your
373 pass. Note that ``DEBUG()`` macros are disabled for optimized builds, so they
374 do not cause a performance impact at all (for the same reason, they should also
375 not contain side-effects!).
377 One additional nice thing about the ``DEBUG()`` macro is that you can enable or
378 disable it directly in gdb. Just use "``set DebugFlag=0``" or "``set
379 DebugFlag=1``" from the gdb if the program is running. If the program hasn't
380 been started yet, you can always just run it with ``-debug``.
384 Fine grained debug info with ``DEBUG_TYPE`` and the ``-debug-only`` option
385 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
387 Sometimes you may find yourself in a situation where enabling ``-debug`` just
388 turns on **too much** information (such as when working on the code generator).
389 If you want to enable debug information with more fine-grained control, you
390 define the ``DEBUG_TYPE`` macro and the ``-debug`` only option as follows:
395 DEBUG(errs() << "No debug type\n");
396 #define DEBUG_TYPE "foo"
397 DEBUG(errs() << "'foo' debug type\n");
399 #define DEBUG_TYPE "bar"
400 DEBUG(errs() << "'bar' debug type\n"));
402 #define DEBUG_TYPE ""
403 DEBUG(errs() << "No debug type (2)\n");
405 Then you can run your pass like this:
409 $ opt < a.bc > /dev/null -mypass
411 $ opt < a.bc > /dev/null -mypass -debug
416 $ opt < a.bc > /dev/null -mypass -debug-only=foo
418 $ opt < a.bc > /dev/null -mypass -debug-only=bar
421 Of course, in practice, you should only set ``DEBUG_TYPE`` at the top of a file,
422 to specify the debug type for the entire module (if you do this before you
423 ``#include "llvm/Support/Debug.h"``, you don't have to insert the ugly
424 ``#undef``'s). Also, you should use names more meaningful than "foo" and "bar",
425 because there is no system in place to ensure that names do not conflict. If
426 two different modules use the same string, they will all be turned on when the
427 name is specified. This allows, for example, all debug information for
428 instruction scheduling to be enabled with ``-debug-type=InstrSched``, even if
429 the source lives in multiple files.
431 The ``DEBUG_WITH_TYPE`` macro is also available for situations where you would
432 like to set ``DEBUG_TYPE``, but only for one specific ``DEBUG`` statement. It
433 takes an additional first parameter, which is the type to use. For example, the
434 preceding example could be written as:
438 DEBUG_WITH_TYPE("", errs() << "No debug type\n");
439 DEBUG_WITH_TYPE("foo", errs() << "'foo' debug type\n");
440 DEBUG_WITH_TYPE("bar", errs() << "'bar' debug type\n"));
441 DEBUG_WITH_TYPE("", errs() << "No debug type (2)\n");
445 The ``Statistic`` class & ``-stats`` option
446 -------------------------------------------
448 The ``llvm/ADT/Statistic.h`` (`doxygen
449 <http://llvm.org/doxygen/Statistic_8h-source.html>`__) file provides a class
450 named ``Statistic`` that is used as a unified way to keep track of what the LLVM
451 compiler is doing and how effective various optimizations are. It is useful to
452 see what optimizations are contributing to making a particular program run
455 Often you may run your pass on some big program, and you're interested to see
456 how many times it makes a certain transformation. Although you can do this with
457 hand inspection, or some ad-hoc method, this is a real pain and not very useful
458 for big programs. Using the ``Statistic`` class makes it very easy to keep
459 track of this information, and the calculated information is presented in a
460 uniform manner with the rest of the passes being executed.
462 There are many examples of ``Statistic`` uses, but the basics of using it are as
465 #. Define your statistic like this:
469 #define DEBUG_TYPE "mypassname" // This goes before any #includes.
470 STATISTIC(NumXForms, "The # of times I did stuff");
472 The ``STATISTIC`` macro defines a static variable, whose name is specified by
473 the first argument. The pass name is taken from the ``DEBUG_TYPE`` macro, and
474 the description is taken from the second argument. The variable defined
475 ("NumXForms" in this case) acts like an unsigned integer.
477 #. Whenever you make a transformation, bump the counter:
481 ++NumXForms; // I did stuff!
483 That's all you have to do. To get '``opt``' to print out the statistics
484 gathered, use the '``-stats``' option:
488 $ opt -stats -mypassname < program.bc > /dev/null
489 ... statistics output ...
491 When running ``opt`` on a C file from the SPEC benchmark suite, it gives a
492 report that looks like this:
496 7646 bitcodewriter - Number of normal instructions
497 725 bitcodewriter - Number of oversized instructions
498 129996 bitcodewriter - Number of bitcode bytes written
499 2817 raise - Number of insts DCEd or constprop'd
500 3213 raise - Number of cast-of-self removed
501 5046 raise - Number of expression trees converted
502 75 raise - Number of other getelementptr's formed
503 138 raise - Number of load/store peepholes
504 42 deadtypeelim - Number of unused typenames removed from symtab
505 392 funcresolve - Number of varargs functions resolved
506 27 globaldce - Number of global variables removed
507 2 adce - Number of basic blocks removed
508 134 cee - Number of branches revectored
509 49 cee - Number of setcc instruction eliminated
510 532 gcse - Number of loads removed
511 2919 gcse - Number of instructions removed
512 86 indvars - Number of canonical indvars added
513 87 indvars - Number of aux indvars removed
514 25 instcombine - Number of dead inst eliminate
515 434 instcombine - Number of insts combined
516 248 licm - Number of load insts hoisted
517 1298 licm - Number of insts hoisted to a loop pre-header
518 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
519 75 mem2reg - Number of alloca's promoted
520 1444 cfgsimplify - Number of blocks simplified
522 Obviously, with so many optimizations, having a unified framework for this stuff
523 is very nice. Making your pass fit well into the framework makes it more
524 maintainable and useful.
528 Viewing graphs while debugging code
529 -----------------------------------
531 Several of the important data structures in LLVM are graphs: for example CFGs
532 made out of LLVM :ref:`BasicBlocks <BasicBlock>`, CFGs made out of LLVM
533 :ref:`MachineBasicBlocks <MachineBasicBlock>`, and :ref:`Instruction Selection
534 DAGs <SelectionDAG>`. In many cases, while debugging various parts of the
535 compiler, it is nice to instantly visualize these graphs.
537 LLVM provides several callbacks that are available in a debug build to do
538 exactly that. If you call the ``Function::viewCFG()`` method, for example, the
539 current LLVM tool will pop up a window containing the CFG for the function where
540 each basic block is a node in the graph, and each node contains the instructions
541 in the block. Similarly, there also exists ``Function::viewCFGOnly()`` (does
542 not include the instructions), the ``MachineFunction::viewCFG()`` and
543 ``MachineFunction::viewCFGOnly()``, and the ``SelectionDAG::viewGraph()``
544 methods. Within GDB, for example, you can usually use something like ``call
545 DAG.viewGraph()`` to pop up a window. Alternatively, you can sprinkle calls to
546 these functions in your code in places you want to debug.
548 Getting this to work requires a small amount of setup. On Unix systems
549 with X11, install the `graphviz <http://www.graphviz.org>`_ toolkit, and make
550 sure 'dot' and 'gv' are in your path. If you are running on Mac OS X, download
551 and install the Mac OS X `Graphviz program
552 <http://www.pixelglow.com/graphviz/>`_ and add
553 ``/Applications/Graphviz.app/Contents/MacOS/`` (or wherever you install it) to
554 your path. The programs need not be present when configuring, building or
555 running LLVM and can simply be installed when needed during an active debug
558 ``SelectionDAG`` has been extended to make it easier to locate *interesting*
559 nodes in large complex graphs. From gdb, if you ``call DAG.setGraphColor(node,
560 "color")``, then the next ``call DAG.viewGraph()`` would highlight the node in
561 the specified color (choices of colors can be found at `colors
562 <http://www.graphviz.org/doc/info/colors.html>`_.) More complex node attributes
563 can be provided with ``call DAG.setGraphAttrs(node, "attributes")`` (choices can
564 be found at `Graph attributes <http://www.graphviz.org/doc/info/attrs.html>`_.)
565 If you want to restart and clear all the current graph attributes, then you can
566 ``call DAG.clearGraphAttrs()``.
568 Note that graph visualization features are compiled out of Release builds to
569 reduce file size. This means that you need a Debug+Asserts or Release+Asserts
570 build to use these features.
574 Picking the Right Data Structure for a Task
575 ===========================================
577 LLVM has a plethora of data structures in the ``llvm/ADT/`` directory, and we
578 commonly use STL data structures. This section describes the trade-offs you
579 should consider when you pick one.
581 The first step is a choose your own adventure: do you want a sequential
582 container, a set-like container, or a map-like container? The most important
583 thing when choosing a container is the algorithmic properties of how you plan to
584 access the container. Based on that, you should use:
587 * a :ref:`map-like <ds_map>` container if you need efficient look-up of a
588 value based on another value. Map-like containers also support efficient
589 queries for containment (whether a key is in the map). Map-like containers
590 generally do not support efficient reverse mapping (values to keys). If you
591 need that, use two maps. Some map-like containers also support efficient
592 iteration through the keys in sorted order. Map-like containers are the most
593 expensive sort, only use them if you need one of these capabilities.
595 * a :ref:`set-like <ds_set>` container if you need to put a bunch of stuff into
596 a container that automatically eliminates duplicates. Some set-like
597 containers support efficient iteration through the elements in sorted order.
598 Set-like containers are more expensive than sequential containers.
600 * a :ref:`sequential <ds_sequential>` container provides the most efficient way
601 to add elements and keeps track of the order they are added to the collection.
602 They permit duplicates and support efficient iteration, but do not support
603 efficient look-up based on a key.
605 * a :ref:`string <ds_string>` container is a specialized sequential container or
606 reference structure that is used for character or byte arrays.
608 * a :ref:`bit <ds_bit>` container provides an efficient way to store and
609 perform set operations on sets of numeric id's, while automatically
610 eliminating duplicates. Bit containers require a maximum of 1 bit for each
611 identifier you want to store.
613 Once the proper category of container is determined, you can fine tune the
614 memory use, constant factors, and cache behaviors of access by intelligently
615 picking a member of the category. Note that constant factors and cache behavior
616 can be a big deal. If you have a vector that usually only contains a few
617 elements (but could contain many), for example, it's much better to use
618 :ref:`SmallVector <dss_smallvector>` than :ref:`vector <dss_vector>`. Doing so
619 avoids (relatively) expensive malloc/free calls, which dwarf the cost of adding
620 the elements to the container.
624 Sequential Containers (std::vector, std::list, etc)
625 ---------------------------------------------------
627 There are a variety of sequential containers available for you, based on your
628 needs. Pick the first in this section that will do what you want.
635 The ``llvm::ArrayRef`` class is the preferred class to use in an interface that
636 accepts a sequential list of elements in memory and just reads from them. By
637 taking an ``ArrayRef``, the API can be passed a fixed size array, an
638 ``std::vector``, an ``llvm::SmallVector`` and anything else that is contiguous
646 Fixed size arrays are very simple and very fast. They are good if you know
647 exactly how many elements you have, or you have a (low) upper bound on how many
652 Heap Allocated Arrays
653 ^^^^^^^^^^^^^^^^^^^^^
655 Heap allocated arrays (``new[]`` + ``delete[]``) are also simple. They are good
656 if the number of elements is variable, if you know how many elements you will
657 need before the array is allocated, and if the array is usually large (if not,
658 consider a :ref:`SmallVector <dss_smallvector>`). The cost of a heap allocated
659 array is the cost of the new/delete (aka malloc/free). Also note that if you
660 are allocating an array of a type with a constructor, the constructor and
661 destructors will be run for every element in the array (re-sizable vectors only
662 construct those elements actually used).
664 .. _dss_tinyptrvector:
666 llvm/ADT/TinyPtrVector.h
667 ^^^^^^^^^^^^^^^^^^^^^^^^
669 ``TinyPtrVector<Type>`` is a highly specialized collection class that is
670 optimized to avoid allocation in the case when a vector has zero or one
671 elements. It has two major restrictions: 1) it can only hold values of pointer
672 type, and 2) it cannot hold a null pointer.
674 Since this container is highly specialized, it is rarely used.
678 llvm/ADT/SmallVector.h
679 ^^^^^^^^^^^^^^^^^^^^^^
681 ``SmallVector<Type, N>`` is a simple class that looks and smells just like
682 ``vector<Type>``: it supports efficient iteration, lays out elements in memory
683 order (so you can do pointer arithmetic between elements), supports efficient
684 push_back/pop_back operations, supports efficient random access to its elements,
687 The advantage of SmallVector is that it allocates space for some number of
688 elements (N) **in the object itself**. Because of this, if the SmallVector is
689 dynamically smaller than N, no malloc is performed. This can be a big win in
690 cases where the malloc/free call is far more expensive than the code that
691 fiddles around with the elements.
693 This is good for vectors that are "usually small" (e.g. the number of
694 predecessors/successors of a block is usually less than 8). On the other hand,
695 this makes the size of the SmallVector itself large, so you don't want to
696 allocate lots of them (doing so will waste a lot of space). As such,
697 SmallVectors are most useful when on the stack.
699 SmallVector also provides a nice portable and efficient replacement for
704 Prefer to use ``SmallVectorImpl<T>`` as a parameter type.
706 In APIs that don't care about the "small size" (most?), prefer to use
707 the ``SmallVectorImpl<T>`` class, which is basically just the "vector
708 header" (and methods) without the elements allocated after it. Note that
709 ``SmallVector<T, N>`` inherits from ``SmallVectorImpl<T>`` so the
710 conversion is implicit and costs nothing. E.g.
714 // BAD: Clients cannot pass e.g. SmallVector<Foo, 4>.
715 hardcodedSmallSize(SmallVector<Foo, 2> &Out);
716 // GOOD: Clients can pass any SmallVector<Foo, N>.
717 allowsAnySmallSize(SmallVectorImpl<Foo> &Out);
720 SmallVector<Foo, 8> Vec;
721 hardcodedSmallSize(Vec); // Error.
722 allowsAnySmallSize(Vec); // Works.
725 Even though it has "``Impl``" in the name, this is so widely used that
726 it really isn't "private to the implementation" anymore. A name like
727 ``SmallVectorHeader`` would be more appropriate.
734 ``std::vector`` is well loved and respected. It is useful when SmallVector
735 isn't: when the size of the vector is often large (thus the small optimization
736 will rarely be a benefit) or if you will be allocating many instances of the
737 vector itself (which would waste space for elements that aren't in the
738 container). vector is also useful when interfacing with code that expects
741 One worthwhile note about std::vector: avoid code like this:
750 Instead, write this as:
760 Doing so will save (at least) one heap allocation and free per iteration of the
768 ``std::deque`` is, in some senses, a generalized version of ``std::vector``.
769 Like ``std::vector``, it provides constant time random access and other similar
770 properties, but it also provides efficient access to the front of the list. It
771 does not guarantee continuity of elements within memory.
773 In exchange for this extra flexibility, ``std::deque`` has significantly higher
774 constant factor costs than ``std::vector``. If possible, use ``std::vector`` or
782 ``std::list`` is an extremely inefficient class that is rarely useful. It
783 performs a heap allocation for every element inserted into it, thus having an
784 extremely high constant factor, particularly for small data types.
785 ``std::list`` also only supports bidirectional iteration, not random access
788 In exchange for this high cost, std::list supports efficient access to both ends
789 of the list (like ``std::deque``, but unlike ``std::vector`` or
790 ``SmallVector``). In addition, the iterator invalidation characteristics of
791 std::list are stronger than that of a vector class: inserting or removing an
792 element into the list does not invalidate iterator or pointers to other elements
800 ``ilist<T>`` implements an 'intrusive' doubly-linked list. It is intrusive,
801 because it requires the element to store and provide access to the prev/next
802 pointers for the list.
804 ``ilist`` has the same drawbacks as ``std::list``, and additionally requires an
805 ``ilist_traits`` implementation for the element type, but it provides some novel
806 characteristics. In particular, it can efficiently store polymorphic objects,
807 the traits class is informed when an element is inserted or removed from the
808 list, and ``ilist``\ s are guaranteed to support a constant-time splice
811 These properties are exactly what we want for things like ``Instruction``\ s and
812 basic blocks, which is why these are implemented with ``ilist``\ s.
814 Related classes of interest are explained in the following subsections:
816 * :ref:`ilist_traits <dss_ilist_traits>`
818 * :ref:`iplist <dss_iplist>`
820 * :ref:`llvm/ADT/ilist_node.h <dss_ilist_node>`
822 * :ref:`Sentinels <dss_ilist_sentinel>`
824 .. _dss_packedvector:
826 llvm/ADT/PackedVector.h
827 ^^^^^^^^^^^^^^^^^^^^^^^
829 Useful for storing a vector of values using only a few number of bits for each
830 value. Apart from the standard operations of a vector-like container, it can
831 also perform an 'or' set operation.
839 FirstCondition = 0x1,
840 SecondCondition = 0x2,
845 PackedVector<State, 2> Vec1;
846 Vec1.push_back(FirstCondition);
848 PackedVector<State, 2> Vec2;
849 Vec2.push_back(SecondCondition);
852 return Vec1[0]; // returns 'Both'.
855 .. _dss_ilist_traits:
860 ``ilist_traits<T>`` is ``ilist<T>``'s customization mechanism. ``iplist<T>``
861 (and consequently ``ilist<T>``) publicly derive from this traits class.
868 ``iplist<T>`` is ``ilist<T>``'s base and as such supports a slightly narrower
869 interface. Notably, inserters from ``T&`` are absent.
871 ``ilist_traits<T>`` is a public base of this class and can be used for a wide
872 variety of customizations.
876 llvm/ADT/ilist_node.h
877 ^^^^^^^^^^^^^^^^^^^^^
879 ``ilist_node<T>`` implements a the forward and backward links that are expected
880 by the ``ilist<T>`` (and analogous containers) in the default manner.
882 ``ilist_node<T>``\ s are meant to be embedded in the node type ``T``, usually
883 ``T`` publicly derives from ``ilist_node<T>``.
885 .. _dss_ilist_sentinel:
890 ``ilist``\ s have another specialty that must be considered. To be a good
891 citizen in the C++ ecosystem, it needs to support the standard container
892 operations, such as ``begin`` and ``end`` iterators, etc. Also, the
893 ``operator--`` must work correctly on the ``end`` iterator in the case of
894 non-empty ``ilist``\ s.
896 The only sensible solution to this problem is to allocate a so-called *sentinel*
897 along with the intrusive list, which serves as the ``end`` iterator, providing
898 the back-link to the last element. However conforming to the C++ convention it
899 is illegal to ``operator++`` beyond the sentinel and it also must not be
902 These constraints allow for some implementation freedom to the ``ilist`` how to
903 allocate and store the sentinel. The corresponding policy is dictated by
904 ``ilist_traits<T>``. By default a ``T`` gets heap-allocated whenever the need
905 for a sentinel arises.
907 While the default policy is sufficient in most cases, it may break down when
908 ``T`` does not provide a default constructor. Also, in the case of many
909 instances of ``ilist``\ s, the memory overhead of the associated sentinels is
910 wasted. To alleviate the situation with numerous and voluminous
911 ``T``-sentinels, sometimes a trick is employed, leading to *ghostly sentinels*.
913 Ghostly sentinels are obtained by specially-crafted ``ilist_traits<T>`` which
914 superpose the sentinel with the ``ilist`` instance in memory. Pointer
915 arithmetic is used to obtain the sentinel, which is relative to the ``ilist``'s
916 ``this`` pointer. The ``ilist`` is augmented by an extra pointer, which serves
917 as the back-link of the sentinel. This is the only field in the ghostly
918 sentinel which can be legally accessed.
922 Other Sequential Container options
923 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
925 Other STL containers are available, such as ``std::string``.
927 There are also various STL adapter classes such as ``std::queue``,
928 ``std::priority_queue``, ``std::stack``, etc. These provide simplified access
929 to an underlying container but don't affect the cost of the container itself.
933 String-like containers
934 ----------------------
936 There are a variety of ways to pass around and use strings in C and C++, and
937 LLVM adds a few new options to choose from. Pick the first option on this list
938 that will do what you need, they are ordered according to their relative cost.
940 Note that is is generally preferred to *not* pass strings around as ``const
941 char*``'s. These have a number of problems, including the fact that they
942 cannot represent embedded nul ("\0") characters, and do not have a length
943 available efficiently. The general replacement for '``const char*``' is
946 For more information on choosing string containers for APIs, please see
947 :ref:`Passing Strings <string_apis>`.
954 The StringRef class is a simple value class that contains a pointer to a
955 character and a length, and is quite related to the :ref:`ArrayRef
956 <dss_arrayref>` class (but specialized for arrays of characters). Because
957 StringRef carries a length with it, it safely handles strings with embedded nul
958 characters in it, getting the length does not require a strlen call, and it even
959 has very convenient APIs for slicing and dicing the character range that it
962 StringRef is ideal for passing simple strings around that are known to be live,
963 either because they are C string literals, std::string, a C array, or a
964 SmallVector. Each of these cases has an efficient implicit conversion to
965 StringRef, which doesn't result in a dynamic strlen being executed.
967 StringRef has a few major limitations which make more powerful string containers
970 #. You cannot directly convert a StringRef to a 'const char*' because there is
971 no way to add a trailing nul (unlike the .c_str() method on various stronger
974 #. StringRef doesn't own or keep alive the underlying string bytes.
975 As such it can easily lead to dangling pointers, and is not suitable for
976 embedding in datastructures in most cases (instead, use an std::string or
977 something like that).
979 #. For the same reason, StringRef cannot be used as the return value of a
980 method if the method "computes" the result string. Instead, use std::string.
982 #. StringRef's do not allow you to mutate the pointed-to string bytes and it
983 doesn't allow you to insert or remove bytes from the range. For editing
984 operations like this, it interoperates with the :ref:`Twine <dss_twine>`
987 Because of its strengths and limitations, it is very common for a function to
988 take a StringRef and for a method on an object to return a StringRef that points
989 into some string that it owns.
996 The Twine class is used as an intermediary datatype for APIs that want to take a
997 string that can be constructed inline with a series of concatenations. Twine
998 works by forming recursive instances of the Twine datatype (a simple value
999 object) on the stack as temporary objects, linking them together into a tree
1000 which is then linearized when the Twine is consumed. Twine is only safe to use
1001 as the argument to a function, and should always be a const reference, e.g.:
1005 void foo(const Twine &T);
1009 foo(X + "." + Twine(i));
1011 This example forms a string like "blarg.42" by concatenating the values
1012 together, and does not form intermediate strings containing "blarg" or "blarg.".
1014 Because Twine is constructed with temporary objects on the stack, and because
1015 these instances are destroyed at the end of the current statement, it is an
1016 inherently dangerous API. For example, this simple variant contains undefined
1017 behavior and will probably crash:
1021 void foo(const Twine &T);
1025 const Twine &Tmp = X + "." + Twine(i);
1028 ... because the temporaries are destroyed before the call. That said, Twine's
1029 are much more efficient than intermediate std::string temporaries, and they work
1030 really well with StringRef. Just be aware of their limitations.
1032 .. _dss_smallstring:
1034 llvm/ADT/SmallString.h
1035 ^^^^^^^^^^^^^^^^^^^^^^
1037 SmallString is a subclass of :ref:`SmallVector <dss_smallvector>` that adds some
1038 convenience APIs like += that takes StringRef's. SmallString avoids allocating
1039 memory in the case when the preallocated space is enough to hold its data, and
1040 it calls back to general heap allocation when required. Since it owns its data,
1041 it is very safe to use and supports full mutation of the string.
1043 Like SmallVector's, the big downside to SmallString is their sizeof. While they
1044 are optimized for small strings, they themselves are not particularly small.
1045 This means that they work great for temporary scratch buffers on the stack, but
1046 should not generally be put into the heap: it is very rare to see a SmallString
1047 as the member of a frequently-allocated heap data structure or returned
1055 The standard C++ std::string class is a very general class that (like
1056 SmallString) owns its underlying data. sizeof(std::string) is very reasonable
1057 so it can be embedded into heap data structures and returned by-value. On the
1058 other hand, std::string is highly inefficient for inline editing (e.g.
1059 concatenating a bunch of stuff together) and because it is provided by the
1060 standard library, its performance characteristics depend a lot of the host
1061 standard library (e.g. libc++ and MSVC provide a highly optimized string class,
1062 GCC contains a really slow implementation).
1064 The major disadvantage of std::string is that almost every operation that makes
1065 them larger can allocate memory, which is slow. As such, it is better to use
1066 SmallVector or Twine as a scratch buffer, but then use std::string to persist
1071 Set-Like Containers (std::set, SmallSet, SetVector, etc)
1072 --------------------------------------------------------
1074 Set-like containers are useful when you need to canonicalize multiple values
1075 into a single representation. There are several different choices for how to do
1076 this, providing various trade-offs.
1078 .. _dss_sortedvectorset:
1083 If you intend to insert a lot of elements, then do a lot of queries, a great
1084 approach is to use a vector (or other sequential container) with
1085 std::sort+std::unique to remove duplicates. This approach works really well if
1086 your usage pattern has these two distinct phases (insert then query), and can be
1087 coupled with a good choice of :ref:`sequential container <ds_sequential>`.
1089 This combination provides the several nice properties: the result data is
1090 contiguous in memory (good for cache locality), has few allocations, is easy to
1091 address (iterators in the final vector are just indices or pointers), and can be
1092 efficiently queried with a standard binary search (e.g.
1093 ``std::lower_bound``; if you want the whole range of elements comparing
1094 equal, use ``std::equal_range``).
1101 If you have a set-like data structure that is usually small and whose elements
1102 are reasonably small, a ``SmallSet<Type, N>`` is a good choice. This set has
1103 space for N elements in place (thus, if the set is dynamically smaller than N,
1104 no malloc traffic is required) and accesses them with a simple linear search.
1105 When the set grows beyond 'N' elements, it allocates a more expensive
1106 representation that guarantees efficient access (for most types, it falls back
1107 to std::set, but for pointers it uses something far better, :ref:`SmallPtrSet
1110 The magic of this class is that it handles small sets extremely efficiently, but
1111 gracefully handles extremely large sets without loss of efficiency. The
1112 drawback is that the interface is quite small: it supports insertion, queries
1113 and erasing, but does not support iteration.
1115 .. _dss_smallptrset:
1117 llvm/ADT/SmallPtrSet.h
1118 ^^^^^^^^^^^^^^^^^^^^^^
1120 SmallPtrSet has all the advantages of ``SmallSet`` (and a ``SmallSet`` of
1121 pointers is transparently implemented with a ``SmallPtrSet``), but also supports
1122 iterators. If more than 'N' insertions are performed, a single quadratically
1123 probed hash table is allocated and grows as needed, providing extremely
1124 efficient access (constant time insertion/deleting/queries with low constant
1125 factors) and is very stingy with malloc traffic.
1127 Note that, unlike ``std::set``, the iterators of ``SmallPtrSet`` are invalidated
1128 whenever an insertion occurs. Also, the values visited by the iterators are not
1129 visited in sorted order.
1136 DenseSet is a simple quadratically probed hash table. It excels at supporting
1137 small values: it uses a single allocation to hold all of the pairs that are
1138 currently inserted in the set. DenseSet is a great way to unique small values
1139 that are not simple pointers (use :ref:`SmallPtrSet <dss_smallptrset>` for
1140 pointers). Note that DenseSet has the same requirements for the value type that
1141 :ref:`DenseMap <dss_densemap>` has.
1145 llvm/ADT/SparseSet.h
1146 ^^^^^^^^^^^^^^^^^^^^
1148 SparseSet holds a small number of objects identified by unsigned keys of
1149 moderate size. It uses a lot of memory, but provides operations that are almost
1150 as fast as a vector. Typical keys are physical registers, virtual registers, or
1151 numbered basic blocks.
1153 SparseSet is useful for algorithms that need very fast clear/find/insert/erase
1154 and fast iteration over small sets. It is not intended for building composite
1157 .. _dss_sparsemultiset:
1159 llvm/ADT/SparseMultiSet.h
1160 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1162 SparseMultiSet adds multiset behavior to SparseSet, while retaining SparseSet's
1163 desirable attributes. Like SparseSet, it typically uses a lot of memory, but
1164 provides operations that are almost as fast as a vector. Typical keys are
1165 physical registers, virtual registers, or numbered basic blocks.
1167 SparseMultiSet is useful for algorithms that need very fast
1168 clear/find/insert/erase of the entire collection, and iteration over sets of
1169 elements sharing a key. It is often a more efficient choice than using composite
1170 data structures (e.g. vector-of-vectors, map-of-vectors). It is not intended for
1171 building composite data structures.
1175 llvm/ADT/FoldingSet.h
1176 ^^^^^^^^^^^^^^^^^^^^^
1178 FoldingSet is an aggregate class that is really good at uniquing
1179 expensive-to-create or polymorphic objects. It is a combination of a chained
1180 hash table with intrusive links (uniqued objects are required to inherit from
1181 FoldingSetNode) that uses :ref:`SmallVector <dss_smallvector>` as part of its ID
1184 Consider a case where you want to implement a "getOrCreateFoo" method for a
1185 complex object (for example, a node in the code generator). The client has a
1186 description of **what** it wants to generate (it knows the opcode and all the
1187 operands), but we don't want to 'new' a node, then try inserting it into a set
1188 only to find out it already exists, at which point we would have to delete it
1189 and return the node that already exists.
1191 To support this style of client, FoldingSet perform a query with a
1192 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1193 element that we want to query for. The query either returns the element
1194 matching the ID or it returns an opaque ID that indicates where insertion should
1195 take place. Construction of the ID usually does not require heap traffic.
1197 Because FoldingSet uses intrusive links, it can support polymorphic objects in
1198 the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1199 Because the elements are individually allocated, pointers to the elements are
1200 stable: inserting or removing elements does not invalidate any pointers to other
1208 ``std::set`` is a reasonable all-around set class, which is decent at many
1209 things but great at nothing. std::set allocates memory for each element
1210 inserted (thus it is very malloc intensive) and typically stores three pointers
1211 per element in the set (thus adding a large amount of per-element space
1212 overhead). It offers guaranteed log(n) performance, which is not particularly
1213 fast from a complexity standpoint (particularly if the elements of the set are
1214 expensive to compare, like strings), and has extremely high constant factors for
1215 lookup, insertion and removal.
1217 The advantages of std::set are that its iterators are stable (deleting or
1218 inserting an element from the set does not affect iterators or pointers to other
1219 elements) and that iteration over the set is guaranteed to be in sorted order.
1220 If the elements in the set are large, then the relative overhead of the pointers
1221 and malloc traffic is not a big deal, but if the elements of the set are small,
1222 std::set is almost never a good choice.
1226 llvm/ADT/SetVector.h
1227 ^^^^^^^^^^^^^^^^^^^^
1229 LLVM's ``SetVector<Type>`` is an adapter class that combines your choice of a
1230 set-like container along with a :ref:`Sequential Container <ds_sequential>` The
1231 important property that this provides is efficient insertion with uniquing
1232 (duplicate elements are ignored) with iteration support. It implements this by
1233 inserting elements into both a set-like container and the sequential container,
1234 using the set-like container for uniquing and the sequential container for
1237 The difference between SetVector and other sets is that the order of iteration
1238 is guaranteed to match the order of insertion into the SetVector. This property
1239 is really important for things like sets of pointers. Because pointer values
1240 are non-deterministic (e.g. vary across runs of the program on different
1241 machines), iterating over the pointers in the set will not be in a well-defined
1244 The drawback of SetVector is that it requires twice as much space as a normal
1245 set and has the sum of constant factors from the set-like container and the
1246 sequential container that it uses. Use it **only** if you need to iterate over
1247 the elements in a deterministic order. SetVector is also expensive to delete
1248 elements out of (linear time), unless you use its "pop_back" method, which is
1251 ``SetVector`` is an adapter class that defaults to using ``std::vector`` and a
1252 size 16 ``SmallSet`` for the underlying containers, so it is quite expensive.
1253 However, ``"llvm/ADT/SetVector.h"`` also provides a ``SmallSetVector`` class,
1254 which defaults to using a ``SmallVector`` and ``SmallSet`` of a specified size.
1255 If you use this, and if your sets are dynamically smaller than ``N``, you will
1256 save a lot of heap traffic.
1258 .. _dss_uniquevector:
1260 llvm/ADT/UniqueVector.h
1261 ^^^^^^^^^^^^^^^^^^^^^^^
1263 UniqueVector is similar to :ref:`SetVector <dss_setvector>` but it retains a
1264 unique ID for each element inserted into the set. It internally contains a map
1265 and a vector, and it assigns a unique ID for each value inserted into the set.
1267 UniqueVector is very expensive: its cost is the sum of the cost of maintaining
1268 both the map and vector, it has high complexity, high constant factors, and
1269 produces a lot of malloc traffic. It should be avoided.
1271 .. _dss_immutableset:
1273 llvm/ADT/ImmutableSet.h
1274 ^^^^^^^^^^^^^^^^^^^^^^^
1276 ImmutableSet is an immutable (functional) set implementation based on an AVL
1277 tree. Adding or removing elements is done through a Factory object and results
1278 in the creation of a new ImmutableSet object. If an ImmutableSet already exists
1279 with the given contents, then the existing one is returned; equality is compared
1280 with a FoldingSetNodeID. The time and space complexity of add or remove
1281 operations is logarithmic in the size of the original set.
1283 There is no method for returning an element of the set, you can only check for
1288 Other Set-Like Container Options
1289 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1291 The STL provides several other options, such as std::multiset and the various
1292 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1293 never use hash_set and unordered_set because they are generally very expensive
1294 (each insertion requires a malloc) and very non-portable.
1296 std::multiset is useful if you're not interested in elimination of duplicates,
1297 but has all the drawbacks of std::set. A sorted vector (where you don't delete
1298 duplicate entries) or some other approach is almost always better.
1302 Map-Like Containers (std::map, DenseMap, etc)
1303 ---------------------------------------------
1305 Map-like containers are useful when you want to associate data to a key. As
1306 usual, there are a lot of different ways to do this. :)
1308 .. _dss_sortedvectormap:
1313 If your usage pattern follows a strict insert-then-query approach, you can
1314 trivially use the same approach as :ref:`sorted vectors for set-like containers
1315 <dss_sortedvectorset>`. The only difference is that your query function (which
1316 uses std::lower_bound to get efficient log(n) lookup) should only compare the
1317 key, not both the key and value. This yields the same advantages as sorted
1322 llvm/ADT/StringMap.h
1323 ^^^^^^^^^^^^^^^^^^^^
1325 Strings are commonly used as keys in maps, and they are difficult to support
1326 efficiently: they are variable length, inefficient to hash and compare when
1327 long, expensive to copy, etc. StringMap is a specialized container designed to
1328 cope with these issues. It supports mapping an arbitrary range of bytes to an
1329 arbitrary other object.
1331 The StringMap implementation uses a quadratically-probed hash table, where the
1332 buckets store a pointer to the heap allocated entries (and some other stuff).
1333 The entries in the map must be heap allocated because the strings are variable
1334 length. The string data (key) and the element object (value) are stored in the
1335 same allocation with the string data immediately after the element object.
1336 This container guarantees the "``(char*)(&Value+1)``" points to the key string
1339 The StringMap is very fast for several reasons: quadratic probing is very cache
1340 efficient for lookups, the hash value of strings in buckets is not recomputed
1341 when looking up an element, StringMap rarely has to touch the memory for
1342 unrelated objects when looking up a value (even when hash collisions happen),
1343 hash table growth does not recompute the hash values for strings already in the
1344 table, and each pair in the map is store in a single allocation (the string data
1345 is stored in the same allocation as the Value of a pair).
1347 StringMap also provides query methods that take byte ranges, so it only ever
1348 copies a string if a value is inserted into the table.
1350 StringMap iteratation order, however, is not guaranteed to be deterministic, so
1351 any uses which require that should instead use a std::map.
1355 llvm/ADT/IndexedMap.h
1356 ^^^^^^^^^^^^^^^^^^^^^
1358 IndexedMap is a specialized container for mapping small dense integers (or
1359 values that can be mapped to small dense integers) to some other type. It is
1360 internally implemented as a vector with a mapping function that maps the keys
1361 to the dense integer range.
1363 This is useful for cases like virtual registers in the LLVM code generator: they
1364 have a dense mapping that is offset by a compile-time constant (the first
1365 virtual register ID).
1372 DenseMap is a simple quadratically probed hash table. It excels at supporting
1373 small keys and values: it uses a single allocation to hold all of the pairs
1374 that are currently inserted in the map. DenseMap is a great way to map
1375 pointers to pointers, or map other small types to each other.
1377 There are several aspects of DenseMap that you should be aware of, however.
1378 The iterators in a DenseMap are invalidated whenever an insertion occurs,
1379 unlike map. Also, because DenseMap allocates space for a large number of
1380 key/value pairs (it starts with 64 by default), it will waste a lot of space if
1381 your keys or values are large. Finally, you must implement a partial
1382 specialization of DenseMapInfo for the key that you want, if it isn't already
1383 supported. This is required to tell DenseMap about two special marker values
1384 (which can never be inserted into the map) that it needs internally.
1386 DenseMap's find_as() method supports lookup operations using an alternate key
1387 type. This is useful in cases where the normal key type is expensive to
1388 construct, but cheap to compare against. The DenseMapInfo is responsible for
1389 defining the appropriate comparison and hashing methods for each alternate key
1397 ValueMap is a wrapper around a :ref:`DenseMap <dss_densemap>` mapping
1398 ``Value*``\ s (or subclasses) to another type. When a Value is deleted or
1399 RAUW'ed, ValueMap will update itself so the new version of the key is mapped to
1400 the same value, just as if the key were a WeakVH. You can configure exactly how
1401 this happens, and what else happens on these two events, by passing a ``Config``
1402 parameter to the ValueMap template.
1404 .. _dss_intervalmap:
1406 llvm/ADT/IntervalMap.h
1407 ^^^^^^^^^^^^^^^^^^^^^^
1409 IntervalMap is a compact map for small keys and values. It maps key intervals
1410 instead of single keys, and it will automatically coalesce adjacent intervals.
1411 When then map only contains a few intervals, they are stored in the map object
1412 itself to avoid allocations.
1414 The IntervalMap iterators are quite big, so they should not be passed around as
1415 STL iterators. The heavyweight iterators allow a smaller data structure.
1422 std::map has similar characteristics to :ref:`std::set <dss_set>`: it uses a
1423 single allocation per pair inserted into the map, it offers log(n) lookup with
1424 an extremely large constant factor, imposes a space penalty of 3 pointers per
1425 pair in the map, etc.
1427 std::map is most useful when your keys or values are very large, if you need to
1428 iterate over the collection in sorted order, or if you need stable iterators
1429 into the map (i.e. they don't get invalidated if an insertion or deletion of
1430 another element takes place).
1434 llvm/ADT/MapVector.h
1435 ^^^^^^^^^^^^^^^^^^^^
1437 ``MapVector<KeyT,ValueT>`` provides a subset of the DenseMap interface. The
1438 main difference is that the iteration order is guaranteed to be the insertion
1439 order, making it an easy (but somewhat expensive) solution for non-deterministic
1440 iteration over maps of pointers.
1442 It is implemented by mapping from key to an index in a vector of key,value
1443 pairs. This provides fast lookup and iteration, but has two main drawbacks: The
1444 key is stored twice and it doesn't support removing elements.
1446 .. _dss_inteqclasses:
1448 llvm/ADT/IntEqClasses.h
1449 ^^^^^^^^^^^^^^^^^^^^^^^
1451 IntEqClasses provides a compact representation of equivalence classes of small
1452 integers. Initially, each integer in the range 0..n-1 has its own equivalence
1453 class. Classes can be joined by passing two class representatives to the
1454 join(a, b) method. Two integers are in the same class when findLeader() returns
1455 the same representative.
1457 Once all equivalence classes are formed, the map can be compressed so each
1458 integer 0..n-1 maps to an equivalence class number in the range 0..m-1, where m
1459 is the total number of equivalence classes. The map must be uncompressed before
1460 it can be edited again.
1462 .. _dss_immutablemap:
1464 llvm/ADT/ImmutableMap.h
1465 ^^^^^^^^^^^^^^^^^^^^^^^
1467 ImmutableMap is an immutable (functional) map implementation based on an AVL
1468 tree. Adding or removing elements is done through a Factory object and results
1469 in the creation of a new ImmutableMap object. If an ImmutableMap already exists
1470 with the given key set, then the existing one is returned; equality is compared
1471 with a FoldingSetNodeID. The time and space complexity of add or remove
1472 operations is logarithmic in the size of the original map.
1476 Other Map-Like Container Options
1477 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1479 The STL provides several other options, such as std::multimap and the various
1480 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1481 never use hash_set and unordered_set because they are generally very expensive
1482 (each insertion requires a malloc) and very non-portable.
1484 std::multimap is useful if you want to map a key to multiple values, but has all
1485 the drawbacks of std::map. A sorted vector or some other approach is almost
1490 Bit storage containers (BitVector, SparseBitVector)
1491 ---------------------------------------------------
1493 Unlike the other containers, there are only two bit storage containers, and
1494 choosing when to use each is relatively straightforward.
1496 One additional option is ``std::vector<bool>``: we discourage its use for two
1497 reasons 1) the implementation in many common compilers (e.g. commonly
1498 available versions of GCC) is extremely inefficient and 2) the C++ standards
1499 committee is likely to deprecate this container and/or change it significantly
1500 somehow. In any case, please don't use it.
1507 The BitVector container provides a dynamic size set of bits for manipulation.
1508 It supports individual bit setting/testing, as well as set operations. The set
1509 operations take time O(size of bitvector), but operations are performed one word
1510 at a time, instead of one bit at a time. This makes the BitVector very fast for
1511 set operations compared to other containers. Use the BitVector when you expect
1512 the number of set bits to be high (i.e. a dense set).
1514 .. _dss_smallbitvector:
1519 The SmallBitVector container provides the same interface as BitVector, but it is
1520 optimized for the case where only a small number of bits, less than 25 or so,
1521 are needed. It also transparently supports larger bit counts, but slightly less
1522 efficiently than a plain BitVector, so SmallBitVector should only be used when
1523 larger counts are rare.
1525 At this time, SmallBitVector does not support set operations (and, or, xor), and
1526 its operator[] does not provide an assignable lvalue.
1528 .. _dss_sparsebitvector:
1533 The SparseBitVector container is much like BitVector, with one major difference:
1534 Only the bits that are set, are stored. This makes the SparseBitVector much
1535 more space efficient than BitVector when the set is sparse, as well as making
1536 set operations O(number of set bits) instead of O(size of universe). The
1537 downside to the SparseBitVector is that setting and testing of random bits is
1538 O(N), and on large SparseBitVectors, this can be slower than BitVector. In our
1539 implementation, setting or testing bits in sorted order (either forwards or
1540 reverse) is O(1) worst case. Testing and setting bits within 128 bits (depends
1541 on size) of the current bit is also O(1). As a general statement,
1542 testing/setting bits in a SparseBitVector is O(distance away from last set bit).
1546 Helpful Hints for Common Operations
1547 ===================================
1549 This section describes how to perform some very simple transformations of LLVM
1550 code. This is meant to give examples of common idioms used, showing the
1551 practical side of LLVM transformations.
1553 Because this is a "how-to" section, you should also read about the main classes
1554 that you will be working with. The :ref:`Core LLVM Class Hierarchy Reference
1555 <coreclasses>` contains details and descriptions of the main classes that you
1560 Basic Inspection and Traversal Routines
1561 ---------------------------------------
1563 The LLVM compiler infrastructure have many different data structures that may be
1564 traversed. Following the example of the C++ standard template library, the
1565 techniques used to traverse these various data structures are all basically the
1566 same. For a enumerable sequence of values, the ``XXXbegin()`` function (or
1567 method) returns an iterator to the start of the sequence, the ``XXXend()``
1568 function returns an iterator pointing to one past the last valid element of the
1569 sequence, and there is some ``XXXiterator`` data type that is common between the
1572 Because the pattern for iteration is common across many different aspects of the
1573 program representation, the standard template library algorithms may be used on
1574 them, and it is easier to remember how to iterate. First we show a few common
1575 examples of the data structures that need to be traversed. Other data
1576 structures are traversed in very similar ways.
1578 .. _iterate_function:
1580 Iterating over the ``BasicBlock`` in a ``Function``
1581 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1583 It's quite common to have a ``Function`` instance that you'd like to transform
1584 in some way; in particular, you'd like to manipulate its ``BasicBlock``\ s. To
1585 facilitate this, you'll need to iterate over all of the ``BasicBlock``\ s that
1586 constitute the ``Function``. The following is an example that prints the name
1587 of a ``BasicBlock`` and the number of ``Instruction``\ s it contains:
1591 // func is a pointer to a Function instance
1592 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
1593 // Print out the name of the basic block if it has one, and then the
1594 // number of instructions that it contains
1595 errs() << "Basic block (name=" << i->getName() << ") has "
1596 << i->size() << " instructions.\n";
1598 Note that i can be used as if it were a pointer for the purposes of invoking
1599 member functions of the ``Instruction`` class. This is because the indirection
1600 operator is overloaded for the iterator classes. In the above code, the
1601 expression ``i->size()`` is exactly equivalent to ``(*i).size()`` just like
1604 .. _iterate_basicblock:
1606 Iterating over the ``Instruction`` in a ``BasicBlock``
1607 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1609 Just like when dealing with ``BasicBlock``\ s in ``Function``\ s, it's easy to
1610 iterate over the individual instructions that make up ``BasicBlock``\ s. Here's
1611 a code snippet that prints out each instruction in a ``BasicBlock``:
1615 // blk is a pointer to a BasicBlock instance
1616 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
1617 // The next statement works since operator<<(ostream&,...)
1618 // is overloaded for Instruction&
1619 errs() << *i << "\n";
1622 However, this isn't really the best way to print out the contents of a
1623 ``BasicBlock``! Since the ostream operators are overloaded for virtually
1624 anything you'll care about, you could have just invoked the print routine on the
1625 basic block itself: ``errs() << *blk << "\n";``.
1627 .. _iterate_insiter:
1629 Iterating over the ``Instruction`` in a ``Function``
1630 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1632 If you're finding that you commonly iterate over a ``Function``'s
1633 ``BasicBlock``\ s and then that ``BasicBlock``'s ``Instruction``\ s,
1634 ``InstIterator`` should be used instead. You'll need to include
1635 ``llvm/IR/InstIterator.h`` (`doxygen
1636 <http://llvm.org/doxygen/InstIterator_8h.html>`__) and then instantiate
1637 ``InstIterator``\ s explicitly in your code. Here's a small example that shows
1638 how to dump all instructions in a function to the standard error stream:
1642 #include "llvm/IR/InstIterator.h"
1644 // F is a pointer to a Function instance
1645 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1646 errs() << *I << "\n";
1648 Easy, isn't it? You can also use ``InstIterator``\ s to fill a work list with
1649 its initial contents. For example, if you wanted to initialize a work list to
1650 contain all instructions in a ``Function`` F, all you would need to do is
1655 std::set<Instruction*> worklist;
1656 // or better yet, SmallPtrSet<Instruction*, 64> worklist;
1658 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1659 worklist.insert(&*I);
1661 The STL set ``worklist`` would now contain all instructions in the ``Function``
1664 .. _iterate_convert:
1666 Turning an iterator into a class pointer (and vice-versa)
1667 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1669 Sometimes, it'll be useful to grab a reference (or pointer) to a class instance
1670 when all you've got at hand is an iterator. Well, extracting a reference or a
1671 pointer from an iterator is very straight-forward. Assuming that ``i`` is a
1672 ``BasicBlock::iterator`` and ``j`` is a ``BasicBlock::const_iterator``:
1676 Instruction& inst = *i; // Grab reference to instruction reference
1677 Instruction* pinst = &*i; // Grab pointer to instruction reference
1678 const Instruction& inst = *j;
1680 However, the iterators you'll be working with in the LLVM framework are special:
1681 they will automatically convert to a ptr-to-instance type whenever they need to.
1682 Instead of derferencing the iterator and then taking the address of the result,
1683 you can simply assign the iterator to the proper pointer type and you get the
1684 dereference and address-of operation as a result of the assignment (behind the
1685 scenes, this is a result of overloading casting mechanisms). Thus the last line
1686 of the last example,
1690 Instruction *pinst = &*i;
1692 is semantically equivalent to
1696 Instruction *pinst = i;
1698 It's also possible to turn a class pointer into the corresponding iterator, and
1699 this is a constant time operation (very efficient). The following code snippet
1700 illustrates use of the conversion constructors provided by LLVM iterators. By
1701 using these, you can explicitly grab the iterator of something without actually
1702 obtaining it via iteration over some structure:
1706 void printNextInstruction(Instruction* inst) {
1707 BasicBlock::iterator it(inst);
1708 ++it; // After this line, it refers to the instruction after *inst
1709 if (it != inst->getParent()->end()) errs() << *it << "\n";
1712 Unfortunately, these implicit conversions come at a cost; they prevent these
1713 iterators from conforming to standard iterator conventions, and thus from being
1714 usable with standard algorithms and containers. For example, they prevent the
1715 following code, where ``B`` is a ``BasicBlock``, from compiling:
1719 llvm::SmallVector<llvm::Instruction *, 16>(B->begin(), B->end());
1721 Because of this, these implicit conversions may be removed some day, and
1722 ``operator*`` changed to return a pointer instead of a reference.
1724 .. _iterate_complex:
1726 Finding call sites: a slightly more complex example
1727 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1729 Say that you're writing a FunctionPass and would like to count all the locations
1730 in the entire module (that is, across every ``Function``) where a certain
1731 function (i.e., some ``Function *``) is already in scope. As you'll learn
1732 later, you may want to use an ``InstVisitor`` to accomplish this in a much more
1733 straight-forward manner, but this example will allow us to explore how you'd do
1734 it if you didn't have ``InstVisitor`` around. In pseudo-code, this is what we
1737 .. code-block:: none
1739 initialize callCounter to zero
1740 for each Function f in the Module
1741 for each BasicBlock b in f
1742 for each Instruction i in b
1743 if (i is a CallInst and calls the given function)
1744 increment callCounter
1746 And the actual code is (remember, because we're writing a ``FunctionPass``, our
1747 ``FunctionPass``-derived class simply has to override the ``runOnFunction``
1752 Function* targetFunc = ...;
1754 class OurFunctionPass : public FunctionPass {
1756 OurFunctionPass(): callCounter(0) { }
1758 virtual runOnFunction(Function& F) {
1759 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1760 for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) {
1761 if (CallInst* callInst = dyn_cast<CallInst>(&*i)) {
1762 // We know we've encountered a call instruction, so we
1763 // need to determine if it's a call to the
1764 // function pointed to by m_func or not.
1765 if (callInst->getCalledFunction() == targetFunc)
1773 unsigned callCounter;
1776 .. _calls_and_invokes:
1778 Treating calls and invokes the same way
1779 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1781 You may have noticed that the previous example was a bit oversimplified in that
1782 it did not deal with call sites generated by 'invoke' instructions. In this,
1783 and in other situations, you may find that you want to treat ``CallInst``\ s and
1784 ``InvokeInst``\ s the same way, even though their most-specific common base
1785 class is ``Instruction``, which includes lots of less closely-related things.
1786 For these cases, LLVM provides a handy wrapper class called ``CallSite``
1787 (`doxygen <http://llvm.org/doxygen/classllvm_1_1CallSite.html>`__) It is
1788 essentially a wrapper around an ``Instruction`` pointer, with some methods that
1789 provide functionality common to ``CallInst``\ s and ``InvokeInst``\ s.
1791 This class has "value semantics": it should be passed by value, not by reference
1792 and it should not be dynamically allocated or deallocated using ``operator new``
1793 or ``operator delete``. It is efficiently copyable, assignable and
1794 constructable, with costs equivalents to that of a bare pointer. If you look at
1795 its definition, it has only a single pointer member.
1799 Iterating over def-use & use-def chains
1800 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1802 Frequently, we might have an instance of the ``Value`` class (`doxygen
1803 <http://llvm.org/doxygen/classllvm_1_1Value.html>`__) and we want to determine
1804 which ``User`` s use the ``Value``. The list of all ``User``\ s of a particular
1805 ``Value`` is called a *def-use* chain. For example, let's say we have a
1806 ``Function*`` named ``F`` to a particular function ``foo``. Finding all of the
1807 instructions that *use* ``foo`` is as simple as iterating over the *def-use*
1814 for (User *U : GV->users()) {
1815 if (Instruction *Inst = dyn_cast<Instruction>(U)) {
1816 errs() << "F is used in instruction:\n";
1817 errs() << *Inst << "\n";
1820 Alternatively, it's common to have an instance of the ``User`` Class (`doxygen
1821 <http://llvm.org/doxygen/classllvm_1_1User.html>`__) and need to know what
1822 ``Value``\ s are used by it. The list of all ``Value``\ s used by a ``User`` is
1823 known as a *use-def* chain. Instances of class ``Instruction`` are common
1824 ``User`` s, so we might want to iterate over all of the values that a particular
1825 instruction uses (that is, the operands of the particular ``Instruction``):
1829 Instruction *pi = ...;
1831 for (Use &U : pi->operands()) {
1836 Declaring objects as ``const`` is an important tool of enforcing mutation free
1837 algorithms (such as analyses, etc.). For this purpose above iterators come in
1838 constant flavors as ``Value::const_use_iterator`` and
1839 ``Value::const_op_iterator``. They automatically arise when calling
1840 ``use/op_begin()`` on ``const Value*``\ s or ``const User*``\ s respectively.
1841 Upon dereferencing, they return ``const Use*``\ s. Otherwise the above patterns
1846 Iterating over predecessors & successors of blocks
1847 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1849 Iterating over the predecessors and successors of a block is quite easy with the
1850 routines defined in ``"llvm/Support/CFG.h"``. Just use code like this to
1851 iterate over all predecessors of BB:
1855 #include "llvm/Support/CFG.h"
1856 BasicBlock *BB = ...;
1858 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
1859 BasicBlock *Pred = *PI;
1863 Similarly, to iterate over successors use ``succ_iterator/succ_begin/succ_end``.
1867 Making simple changes
1868 ---------------------
1870 There are some primitive transformation operations present in the LLVM
1871 infrastructure that are worth knowing about. When performing transformations,
1872 it's fairly common to manipulate the contents of basic blocks. This section
1873 describes some of the common methods for doing so and gives example code.
1875 .. _schanges_creating:
1877 Creating and inserting new ``Instruction``\ s
1878 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1880 *Instantiating Instructions*
1882 Creation of ``Instruction``\ s is straight-forward: simply call the constructor
1883 for the kind of instruction to instantiate and provide the necessary parameters.
1884 For example, an ``AllocaInst`` only *requires* a (const-ptr-to) ``Type``. Thus:
1888 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
1890 will create an ``AllocaInst`` instance that represents the allocation of one
1891 integer in the current stack frame, at run time. Each ``Instruction`` subclass
1892 is likely to have varying default parameters which change the semantics of the
1893 instruction, so refer to the `doxygen documentation for the subclass of
1894 Instruction <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_ that
1895 you're interested in instantiating.
1899 It is very useful to name the values of instructions when you're able to, as
1900 this facilitates the debugging of your transformations. If you end up looking
1901 at generated LLVM machine code, you definitely want to have logical names
1902 associated with the results of instructions! By supplying a value for the
1903 ``Name`` (default) parameter of the ``Instruction`` constructor, you associate a
1904 logical name with the result of the instruction's execution at run time. For
1905 example, say that I'm writing a transformation that dynamically allocates space
1906 for an integer on the stack, and that integer is going to be used as some kind
1907 of index by some other code. To accomplish this, I place an ``AllocaInst`` at
1908 the first point in the first ``BasicBlock`` of some ``Function``, and I'm
1909 intending to use it within the same ``Function``. I might do:
1913 AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
1915 where ``indexLoc`` is now the logical name of the instruction's execution value,
1916 which is a pointer to an integer on the run time stack.
1918 *Inserting instructions*
1920 There are essentially two ways to insert an ``Instruction`` into an existing
1921 sequence of instructions that form a ``BasicBlock``:
1923 * Insertion into an explicit instruction list
1925 Given a ``BasicBlock* pb``, an ``Instruction* pi`` within that ``BasicBlock``,
1926 and a newly-created instruction we wish to insert before ``*pi``, we do the
1931 BasicBlock *pb = ...;
1932 Instruction *pi = ...;
1933 Instruction *newInst = new Instruction(...);
1935 pb->getInstList().insert(pi, newInst); // Inserts newInst before pi in pb
1937 Appending to the end of a ``BasicBlock`` is so common that the ``Instruction``
1938 class and ``Instruction``-derived classes provide constructors which take a
1939 pointer to a ``BasicBlock`` to be appended to. For example code that looked
1944 BasicBlock *pb = ...;
1945 Instruction *newInst = new Instruction(...);
1947 pb->getInstList().push_back(newInst); // Appends newInst to pb
1953 BasicBlock *pb = ...;
1954 Instruction *newInst = new Instruction(..., pb);
1956 which is much cleaner, especially if you are creating long instruction
1959 * Insertion into an implicit instruction list
1961 ``Instruction`` instances that are already in ``BasicBlock``\ s are implicitly
1962 associated with an existing instruction list: the instruction list of the
1963 enclosing basic block. Thus, we could have accomplished the same thing as the
1964 above code without being given a ``BasicBlock`` by doing:
1968 Instruction *pi = ...;
1969 Instruction *newInst = new Instruction(...);
1971 pi->getParent()->getInstList().insert(pi, newInst);
1973 In fact, this sequence of steps occurs so frequently that the ``Instruction``
1974 class and ``Instruction``-derived classes provide constructors which take (as
1975 a default parameter) a pointer to an ``Instruction`` which the newly-created
1976 ``Instruction`` should precede. That is, ``Instruction`` constructors are
1977 capable of inserting the newly-created instance into the ``BasicBlock`` of a
1978 provided instruction, immediately before that instruction. Using an
1979 ``Instruction`` constructor with a ``insertBefore`` (default) parameter, the
1984 Instruction* pi = ...;
1985 Instruction* newInst = new Instruction(..., pi);
1987 which is much cleaner, especially if you're creating a lot of instructions and
1988 adding them to ``BasicBlock``\ s.
1990 .. _schanges_deleting:
1992 Deleting Instructions
1993 ^^^^^^^^^^^^^^^^^^^^^
1995 Deleting an instruction from an existing sequence of instructions that form a
1996 BasicBlock_ is very straight-forward: just call the instruction's
1997 ``eraseFromParent()`` method. For example:
2001 Instruction *I = .. ;
2002 I->eraseFromParent();
2004 This unlinks the instruction from its containing basic block and deletes it. If
2005 you'd just like to unlink the instruction from its containing basic block but
2006 not delete it, you can use the ``removeFromParent()`` method.
2008 .. _schanges_replacing:
2010 Replacing an Instruction with another Value
2011 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2013 Replacing individual instructions
2014 """""""""""""""""""""""""""""""""
2016 Including "`llvm/Transforms/Utils/BasicBlockUtils.h
2017 <http://llvm.org/doxygen/BasicBlockUtils_8h-source.html>`_" permits use of two
2018 very useful replace functions: ``ReplaceInstWithValue`` and
2019 ``ReplaceInstWithInst``.
2021 .. _schanges_deleting_sub:
2023 Deleting Instructions
2024 """""""""""""""""""""
2026 * ``ReplaceInstWithValue``
2028 This function replaces all uses of a given instruction with a value, and then
2029 removes the original instruction. The following example illustrates the
2030 replacement of the result of a particular ``AllocaInst`` that allocates memory
2031 for a single integer with a null pointer to an integer.
2035 AllocaInst* instToReplace = ...;
2036 BasicBlock::iterator ii(instToReplace);
2038 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
2039 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2041 * ``ReplaceInstWithInst``
2043 This function replaces a particular instruction with another instruction,
2044 inserting the new instruction into the basic block at the location where the
2045 old instruction was, and replacing any uses of the old instruction with the
2046 new instruction. The following example illustrates the replacement of one
2047 ``AllocaInst`` with another.
2051 AllocaInst* instToReplace = ...;
2052 BasicBlock::iterator ii(instToReplace);
2054 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
2055 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
2058 Replacing multiple uses of Users and Values
2059 """""""""""""""""""""""""""""""""""""""""""
2061 You can use ``Value::replaceAllUsesWith`` and ``User::replaceUsesOfWith`` to
2062 change more than one use at a time. See the doxygen documentation for the
2063 `Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_ and `User Class
2064 <http://llvm.org/doxygen/classllvm_1_1User.html>`_, respectively, for more
2067 .. _schanges_deletingGV:
2069 Deleting GlobalVariables
2070 ^^^^^^^^^^^^^^^^^^^^^^^^
2072 Deleting a global variable from a module is just as easy as deleting an
2073 Instruction. First, you must have a pointer to the global variable that you
2074 wish to delete. You use this pointer to erase it from its parent, the module.
2079 GlobalVariable *GV = .. ;
2081 GV->eraseFromParent();
2089 In generating IR, you may need some complex types. If you know these types
2090 statically, you can use ``TypeBuilder<...>::get()``, defined in
2091 ``llvm/Support/TypeBuilder.h``, to retrieve them. ``TypeBuilder`` has two forms
2092 depending on whether you're building types for cross-compilation or native
2093 library use. ``TypeBuilder<T, true>`` requires that ``T`` be independent of the
2094 host environment, meaning that it's built out of types from the ``llvm::types``
2095 (`doxygen <http://llvm.org/doxygen/namespacellvm_1_1types.html>`__) namespace
2096 and pointers, functions, arrays, etc. built of those. ``TypeBuilder<T, false>``
2097 additionally allows native C types whose size may depend on the host compiler.
2102 FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get();
2104 is easier to read and write than the equivalent
2108 std::vector<const Type*> params;
2109 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2110 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2112 See the `class comment
2113 <http://llvm.org/doxygen/TypeBuilder_8h-source.html#l00001>`_ for more details.
2120 This section describes the interaction of the LLVM APIs with multithreading,
2121 both on the part of client applications, and in the JIT, in the hosted
2124 Note that LLVM's support for multithreading is still relatively young. Up
2125 through version 2.5, the execution of threaded hosted applications was
2126 supported, but not threaded client access to the APIs. While this use case is
2127 now supported, clients *must* adhere to the guidelines specified below to ensure
2128 proper operation in multithreaded mode.
2130 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2131 intrinsics in order to support threaded operation. If you need a
2132 multhreading-capable LLVM on a platform without a suitably modern system
2133 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2134 using the resultant compiler to build a copy of LLVM with multithreading
2137 .. _startmultithreaded:
2139 Entering and Exiting Multithreaded Mode
2140 ---------------------------------------
2142 In order to properly protect its internal data structures while avoiding
2143 excessive locking overhead in the single-threaded case, the LLVM must intialize
2144 certain data structures necessary to provide guards around its internals. To do
2145 so, the client program must invoke ``llvm_start_multithreaded()`` before making
2146 any concurrent LLVM API calls. To subsequently tear down these structures, use
2147 the ``llvm_stop_multithreaded()`` call. You can also use the
2148 ``llvm_is_multithreaded()`` call to check the status of multithreaded mode.
2150 Note that both of these calls must be made *in isolation*. That is to say that
2151 no other LLVM API calls may be executing at any time during the execution of
2152 ``llvm_start_multithreaded()`` or ``llvm_stop_multithreaded``. It is the
2153 client's responsibility to enforce this isolation.
2155 The return value of ``llvm_start_multithreaded()`` indicates the success or
2156 failure of the initialization. Failure typically indicates that your copy of
2157 LLVM was built without multithreading support, typically because GCC atomic
2158 intrinsics were not found in your system compiler. In this case, the LLVM API
2159 will not be safe for concurrent calls. However, it *will* be safe for hosting
2160 threaded applications in the JIT, though :ref:`care must be taken
2161 <jitthreading>` to ensure that side exits and the like do not accidentally
2162 result in concurrent LLVM API calls.
2166 Ending Execution with ``llvm_shutdown()``
2167 -----------------------------------------
2169 When you are done using the LLVM APIs, you should call ``llvm_shutdown()`` to
2170 deallocate memory used for internal structures. This will also invoke
2171 ``llvm_stop_multithreaded()`` if LLVM is operating in multithreaded mode. As
2172 such, ``llvm_shutdown()`` requires the same isolation guarantees as
2173 ``llvm_stop_multithreaded()``.
2175 Note that, if you use scope-based shutdown, you can use the
2176 ``llvm_shutdown_obj`` class, which calls ``llvm_shutdown()`` in its destructor.
2180 Lazy Initialization with ``ManagedStatic``
2181 ------------------------------------------
2183 ``ManagedStatic`` is a utility class in LLVM used to implement static
2184 initialization of static resources, such as the global type tables. Before the
2185 invocation of ``llvm_shutdown()``, it implements a simple lazy initialization
2186 scheme. Once ``llvm_start_multithreaded()`` returns, however, it uses
2187 double-checked locking to implement thread-safe lazy initialization.
2189 Note that, because no other threads are allowed to issue LLVM API calls before
2190 ``llvm_start_multithreaded()`` returns, it is possible to have
2191 ``ManagedStatic``\ s of ``llvm::sys::Mutex``\ s.
2193 The ``llvm_acquire_global_lock()`` and ``llvm_release_global_lock`` APIs provide
2194 access to the global lock used to implement the double-checked locking for lazy
2195 initialization. These should only be used internally to LLVM, and only if you
2196 know what you're doing!
2200 Achieving Isolation with ``LLVMContext``
2201 ----------------------------------------
2203 ``LLVMContext`` is an opaque class in the LLVM API which clients can use to
2204 operate multiple, isolated instances of LLVM concurrently within the same
2205 address space. For instance, in a hypothetical compile-server, the compilation
2206 of an individual translation unit is conceptually independent from all the
2207 others, and it would be desirable to be able to compile incoming translation
2208 units concurrently on independent server threads. Fortunately, ``LLVMContext``
2209 exists to enable just this kind of scenario!
2211 Conceptually, ``LLVMContext`` provides isolation. Every LLVM entity
2212 (``Module``\ s, ``Value``\ s, ``Type``\ s, ``Constant``\ s, etc.) in LLVM's
2213 in-memory IR belongs to an ``LLVMContext``. Entities in different contexts
2214 *cannot* interact with each other: ``Module``\ s in different contexts cannot be
2215 linked together, ``Function``\ s cannot be added to ``Module``\ s in different
2216 contexts, etc. What this means is that is is safe to compile on multiple
2217 threads simultaneously, as long as no two threads operate on entities within the
2220 In practice, very few places in the API require the explicit specification of a
2221 ``LLVMContext``, other than the ``Type`` creation/lookup APIs. Because every
2222 ``Type`` carries a reference to its owning context, most other entities can
2223 determine what context they belong to by looking at their own ``Type``. If you
2224 are adding new entities to LLVM IR, please try to maintain this interface
2227 For clients that do *not* require the benefits of isolation, LLVM provides a
2228 convenience API ``getGlobalContext()``. This returns a global, lazily
2229 initialized ``LLVMContext`` that may be used in situations where isolation is
2237 LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple
2238 threads can call ``ExecutionEngine::getPointerToFunction()`` or
2239 ``ExecutionEngine::runFunction()`` concurrently, and multiple threads can run
2240 code output by the JIT concurrently. The user must still ensure that only one
2241 thread accesses IR in a given ``LLVMContext`` while another thread might be
2242 modifying it. One way to do that is to always hold the JIT lock while accessing
2243 IR outside the JIT (the JIT *modifies* the IR by adding ``CallbackVH``\ s).
2244 Another way is to only call ``getPointerToFunction()`` from the
2245 ``LLVMContext``'s thread.
2247 When the JIT is configured to compile lazily (using
2248 ``ExecutionEngine::DisableLazyCompilation(false)``), there is currently a `race
2249 condition <http://llvm.org/bugs/show_bug.cgi?id=5184>`_ in updating call sites
2250 after a function is lazily-jitted. It's still possible to use the lazy JIT in a
2251 threaded program if you ensure that only one thread at a time can call any
2252 particular lazy stub and that the JIT lock guards any IR access, but we suggest
2253 using only the eager JIT in threaded programs.
2260 This section describes some of the advanced or obscure API's that most clients
2261 do not need to be aware of. These API's tend manage the inner workings of the
2262 LLVM system, and only need to be accessed in unusual circumstances.
2266 The ``ValueSymbolTable`` class
2267 ------------------------------
2269 The ``ValueSymbolTable`` (`doxygen
2270 <http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html>`__) class provides
2271 a symbol table that the :ref:`Function <c_Function>` and Module_ classes use for
2272 naming value definitions. The symbol table can provide a name for any Value_.
2274 Note that the ``SymbolTable`` class should not be directly accessed by most
2275 clients. It should only be used when iteration over the symbol table names
2276 themselves are required, which is very special purpose. Note that not all LLVM
2277 Value_\ s have names, and those without names (i.e. they have an empty name) do
2278 not exist in the symbol table.
2280 Symbol tables support iteration over the values in the symbol table with
2281 ``begin/end/iterator`` and supports querying to see if a specific name is in the
2282 symbol table (with ``lookup``). The ``ValueSymbolTable`` class exposes no
2283 public mutator methods, instead, simply call ``setName`` on a value, which will
2284 autoinsert it into the appropriate symbol table.
2288 The ``User`` and owned ``Use`` classes' memory layout
2289 -----------------------------------------------------
2291 The ``User`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1User.html>`__)
2292 class provides a basis for expressing the ownership of ``User`` towards other
2293 `Value instance <http://llvm.org/doxygen/classllvm_1_1Value.html>`_\ s. The
2294 ``Use`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1Use.html>`__) helper
2295 class is employed to do the bookkeeping and to facilitate *O(1)* addition and
2300 Interaction and relationship between ``User`` and ``Use`` objects
2301 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2303 A subclass of ``User`` can choose between incorporating its ``Use`` objects or
2304 refer to them out-of-line by means of a pointer. A mixed variant (some ``Use``
2305 s inline others hung off) is impractical and breaks the invariant that the
2306 ``Use`` objects belonging to the same ``User`` form a contiguous array.
2308 We have 2 different layouts in the ``User`` (sub)classes:
2312 The ``Use`` object(s) are inside (resp. at fixed offset) of the ``User``
2313 object and there are a fixed number of them.
2317 The ``Use`` object(s) are referenced by a pointer to an array from the
2318 ``User`` object and there may be a variable number of them.
2320 As of v2.4 each layout still possesses a direct pointer to the start of the
2321 array of ``Use``\ s. Though not mandatory for layout a), we stick to this
2322 redundancy for the sake of simplicity. The ``User`` object also stores the
2323 number of ``Use`` objects it has. (Theoretically this information can also be
2324 calculated given the scheme presented below.)
2326 Special forms of allocation operators (``operator new``) enforce the following
2329 * Layout a) is modelled by prepending the ``User`` object by the ``Use[]``
2332 .. code-block:: none
2334 ...---.---.---.---.-------...
2335 | P | P | P | P | User
2336 '''---'---'---'---'-------'''
2338 * Layout b) is modelled by pointing at the ``Use[]`` array.
2340 .. code-block:: none
2351 *(In the above figures* '``P``' *stands for the* ``Use**`` *that is stored in
2352 each* ``Use`` *object in the member* ``Use::Prev`` *)*
2356 The waymarking algorithm
2357 ^^^^^^^^^^^^^^^^^^^^^^^^
2359 Since the ``Use`` objects are deprived of the direct (back)pointer to their
2360 ``User`` objects, there must be a fast and exact method to recover it. This is
2361 accomplished by the following scheme:
2363 A bit-encoding in the 2 LSBits (least significant bits) of the ``Use::Prev``
2364 allows to find the start of the ``User`` object:
2366 * ``00`` --- binary digit 0
2368 * ``01`` --- binary digit 1
2370 * ``10`` --- stop and calculate (``s``)
2372 * ``11`` --- full stop (``S``)
2374 Given a ``Use*``, all we have to do is to walk till we get a stop and we either
2375 have a ``User`` immediately behind or we have to walk to the next stop picking
2376 up digits and calculating the offset:
2378 .. code-block:: none
2380 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2381 | 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
2382 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2383 |+15 |+10 |+6 |+3 |+1
2386 | | | ______________________>
2387 | | ______________________________________>
2388 | __________________________________________________________>
2390 Only the significant number of bits need to be stored between the stops, so that
2391 the *worst case is 20 memory accesses* when there are 1000 ``Use`` objects
2392 associated with a ``User``.
2396 Reference implementation
2397 ^^^^^^^^^^^^^^^^^^^^^^^^
2399 The following literate Haskell fragment demonstrates the concept:
2401 .. code-block:: haskell
2403 > import Test.QuickCheck
2405 > digits :: Int -> [Char] -> [Char]
2406 > digits 0 acc = '0' : acc
2407 > digits 1 acc = '1' : acc
2408 > digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
2410 > dist :: Int -> [Char] -> [Char]
2413 > dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
2414 > dist n acc = dist (n - 1) $ dist 1 acc
2416 > takeLast n ss = reverse $ take n $ reverse ss
2418 > test = takeLast 40 $ dist 20 []
2421 Printing <test> gives: ``"1s100000s11010s10100s1111s1010s110s11s1S"``
2423 The reverse algorithm computes the length of the string just by examining a
2426 .. code-block:: haskell
2428 > pref :: [Char] -> Int
2430 > pref ('s':'1':rest) = decode 2 1 rest
2431 > pref (_:rest) = 1 + pref rest
2433 > decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
2434 > decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
2435 > decode walk acc _ = walk + acc
2438 Now, as expected, printing <pref test> gives ``40``.
2440 We can *quickCheck* this with following property:
2442 .. code-block:: haskell
2444 > testcase = dist 2000 []
2445 > testcaseLength = length testcase
2447 > identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
2448 > where arr = takeLast n testcase
2451 As expected <quickCheck identityProp> gives:
2455 *Main> quickCheck identityProp
2456 OK, passed 100 tests.
2458 Let's be a bit more exhaustive:
2460 .. code-block:: haskell
2463 > deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
2466 And here is the result of <deepCheck identityProp>:
2470 *Main> deepCheck identityProp
2471 OK, passed 500 tests.
2475 Tagging considerations
2476 ^^^^^^^^^^^^^^^^^^^^^^
2478 To maintain the invariant that the 2 LSBits of each ``Use**`` in ``Use`` never
2479 change after being set up, setters of ``Use::Prev`` must re-tag the new
2480 ``Use**`` on every modification. Accordingly getters must strip the tag bits.
2482 For layout b) instead of the ``User`` we find a pointer (``User*`` with LSBit
2483 set). Following this pointer brings us to the ``User``. A portable trick
2484 ensures that the first bytes of ``User`` (if interpreted as a pointer) never has
2485 the LSBit set. (Portability is relying on the fact that all known compilers
2486 place the ``vptr`` in the first word of the instances.)
2490 The Core LLVM Class Hierarchy Reference
2491 =======================================
2493 ``#include "llvm/IR/Type.h"``
2495 header source: `Type.h <http://llvm.org/doxygen/Type_8h-source.html>`_
2497 doxygen info: `Type Clases <http://llvm.org/doxygen/classllvm_1_1Type.html>`_
2499 The Core LLVM classes are the primary means of representing the program being
2500 inspected or transformed. The core LLVM classes are defined in header files in
2501 the ``include/llvm/`` directory, and implemented in the ``lib/VMCore``
2506 The Type class and Derived Types
2507 --------------------------------
2509 ``Type`` is a superclass of all type classes. Every ``Value`` has a ``Type``.
2510 ``Type`` cannot be instantiated directly but only through its subclasses.
2511 Certain primitive types (``VoidType``, ``LabelType``, ``FloatType`` and
2512 ``DoubleType``) have hidden subclasses. They are hidden because they offer no
2513 useful functionality beyond what the ``Type`` class offers except to distinguish
2514 themselves from other subclasses of ``Type``.
2516 All other types are subclasses of ``DerivedType``. Types can be named, but this
2517 is not a requirement. There exists exactly one instance of a given shape at any
2518 one time. This allows type equality to be performed with address equality of
2519 the Type Instance. That is, given two ``Type*`` values, the types are identical
2520 if the pointers are identical.
2524 Important Public Methods
2525 ^^^^^^^^^^^^^^^^^^^^^^^^
2527 * ``bool isIntegerTy() const``: Returns true for any integer type.
2529 * ``bool isFloatingPointTy()``: Return true if this is one of the five
2530 floating point types.
2532 * ``bool isSized()``: Return true if the type has known size. Things
2533 that don't have a size are abstract types, labels and void.
2537 Important Derived Types
2538 ^^^^^^^^^^^^^^^^^^^^^^^
2541 Subclass of DerivedType that represents integer types of any bit width. Any
2542 bit width between ``IntegerType::MIN_INT_BITS`` (1) and
2543 ``IntegerType::MAX_INT_BITS`` (~8 million) can be represented.
2545 * ``static const IntegerType* get(unsigned NumBits)``: get an integer
2546 type of a specific bit width.
2548 * ``unsigned getBitWidth() const``: Get the bit width of an integer type.
2551 This is subclassed by ArrayType, PointerType and VectorType.
2553 * ``const Type * getElementType() const``: Returns the type of each
2554 of the elements in the sequential type.
2557 This is a subclass of SequentialType and defines the interface for array
2560 * ``unsigned getNumElements() const``: Returns the number of elements
2564 Subclass of SequentialType for pointer types.
2567 Subclass of SequentialType for vector types. A vector type is similar to an
2568 ArrayType but is distinguished because it is a first class type whereas
2569 ArrayType is not. Vector types are used for vector operations and are usually
2570 small vectors of of an integer or floating point type.
2573 Subclass of DerivedTypes for struct types.
2578 Subclass of DerivedTypes for function types.
2580 * ``bool isVarArg() const``: Returns true if it's a vararg function.
2582 * ``const Type * getReturnType() const``: Returns the return type of the
2585 * ``const Type * getParamType (unsigned i)``: Returns the type of the ith
2588 * ``const unsigned getNumParams() const``: Returns the number of formal
2593 The ``Module`` class
2594 --------------------
2596 ``#include "llvm/IR/Module.h"``
2598 header source: `Module.h <http://llvm.org/doxygen/Module_8h-source.html>`_
2600 doxygen info: `Module Class <http://llvm.org/doxygen/classllvm_1_1Module.html>`_
2602 The ``Module`` class represents the top level structure present in LLVM
2603 programs. An LLVM module is effectively either a translation unit of the
2604 original program or a combination of several translation units merged by the
2605 linker. The ``Module`` class keeps track of a list of :ref:`Function
2606 <c_Function>`\ s, a list of GlobalVariable_\ s, and a SymbolTable_.
2607 Additionally, it contains a few helpful member functions that try to make common
2612 Important Public Members of the ``Module`` class
2613 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2615 * ``Module::Module(std::string name = "")``
2617 Constructing a Module_ is easy. You can optionally provide a name for it
2618 (probably based on the name of the translation unit).
2620 * | ``Module::iterator`` - Typedef for function list iterator
2621 | ``Module::const_iterator`` - Typedef for const_iterator.
2622 | ``begin()``, ``end()``, ``size()``, ``empty()``
2624 These are forwarding methods that make it easy to access the contents of a
2625 ``Module`` object's :ref:`Function <c_Function>` list.
2627 * ``Module::FunctionListType &getFunctionList()``
2629 Returns the list of :ref:`Function <c_Function>`\ s. This is necessary to use
2630 when you need to update the list or perform a complex action that doesn't have
2631 a forwarding method.
2635 * | ``Module::global_iterator`` - Typedef for global variable list iterator
2636 | ``Module::const_global_iterator`` - Typedef for const_iterator.
2637 | ``global_begin()``, ``global_end()``, ``global_size()``, ``global_empty()``
2639 These are forwarding methods that make it easy to access the contents of a
2640 ``Module`` object's GlobalVariable_ list.
2642 * ``Module::GlobalListType &getGlobalList()``
2644 Returns the list of GlobalVariable_\ s. This is necessary to use when you
2645 need to update the list or perform a complex action that doesn't have a
2650 * ``SymbolTable *getSymbolTable()``
2652 Return a reference to the SymbolTable_ for this ``Module``.
2656 * ``Function *getFunction(StringRef Name) const``
2658 Look up the specified function in the ``Module`` SymbolTable_. If it does not
2659 exist, return ``null``.
2661 * ``Function *getOrInsertFunction(const std::string &Name, const FunctionType
2664 Look up the specified function in the ``Module`` SymbolTable_. If it does not
2665 exist, add an external declaration for the function and return it.
2667 * ``std::string getTypeName(const Type *Ty)``
2669 If there is at least one entry in the SymbolTable_ for the specified Type_,
2670 return it. Otherwise return the empty string.
2672 * ``bool addTypeName(const std::string &Name, const Type *Ty)``
2674 Insert an entry in the SymbolTable_ mapping ``Name`` to ``Ty``. If there is
2675 already an entry for this name, true is returned and the SymbolTable_ is not
2683 ``#include "llvm/IR/Value.h"``
2685 header source: `Value.h <http://llvm.org/doxygen/Value_8h-source.html>`_
2687 doxygen info: `Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_
2689 The ``Value`` class is the most important class in the LLVM Source base. It
2690 represents a typed value that may be used (among other things) as an operand to
2691 an instruction. There are many different types of ``Value``\ s, such as
2692 Constant_\ s, Argument_\ s. Even Instruction_\ s and :ref:`Function
2693 <c_Function>`\ s are ``Value``\ s.
2695 A particular ``Value`` may be used many times in the LLVM representation for a
2696 program. For example, an incoming argument to a function (represented with an
2697 instance of the Argument_ class) is "used" by every instruction in the function
2698 that references the argument. To keep track of this relationship, the ``Value``
2699 class keeps a list of all of the ``User``\ s that is using it (the User_ class
2700 is a base class for all nodes in the LLVM graph that can refer to ``Value``\ s).
2701 This use list is how LLVM represents def-use information in the program, and is
2702 accessible through the ``use_*`` methods, shown below.
2704 Because LLVM is a typed representation, every LLVM ``Value`` is typed, and this
2705 Type_ is available through the ``getType()`` method. In addition, all LLVM
2706 values can be named. The "name" of the ``Value`` is a symbolic string printed
2709 .. code-block:: llvm
2715 The name of this instruction is "foo". **NOTE** that the name of any value may
2716 be missing (an empty string), so names should **ONLY** be used for debugging
2717 (making the source code easier to read, debugging printouts), they should not be
2718 used to keep track of values or map between them. For this purpose, use a
2719 ``std::map`` of pointers to the ``Value`` itself instead.
2721 One important aspect of LLVM is that there is no distinction between an SSA
2722 variable and the operation that produces it. Because of this, any reference to
2723 the value produced by an instruction (or the value available as an incoming
2724 argument, for example) is represented as a direct pointer to the instance of the
2725 class that represents this value. Although this may take some getting used to,
2726 it simplifies the representation and makes it easier to manipulate.
2730 Important Public Members of the ``Value`` class
2731 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2733 * | ``Value::use_iterator`` - Typedef for iterator over the use-list
2734 | ``Value::const_use_iterator`` - Typedef for const_iterator over the
2736 | ``unsigned use_size()`` - Returns the number of users of the value.
2737 | ``bool use_empty()`` - Returns true if there are no users.
2738 | ``use_iterator use_begin()`` - Get an iterator to the start of the
2740 | ``use_iterator use_end()`` - Get an iterator to the end of the use-list.
2741 | ``User *use_back()`` - Returns the last element in the list.
2743 These methods are the interface to access the def-use information in LLVM.
2744 As with all other iterators in LLVM, the naming conventions follow the
2745 conventions defined by the STL_.
2747 * ``Type *getType() const``
2748 This method returns the Type of the Value.
2750 * | ``bool hasName() const``
2751 | ``std::string getName() const``
2752 | ``void setName(const std::string &Name)``
2754 This family of methods is used to access and assign a name to a ``Value``, be
2755 aware of the :ref:`precaution above <nameWarning>`.
2757 * ``void replaceAllUsesWith(Value *V)``
2759 This method traverses the use list of a ``Value`` changing all User_\ s of the
2760 current value to refer to "``V``" instead. For example, if you detect that an
2761 instruction always produces a constant value (for example through constant
2762 folding), you can replace all uses of the instruction with the constant like
2767 Inst->replaceAllUsesWith(ConstVal);
2774 ``#include "llvm/IR/User.h"``
2776 header source: `User.h <http://llvm.org/doxygen/User_8h-source.html>`_
2778 doxygen info: `User Class <http://llvm.org/doxygen/classllvm_1_1User.html>`_
2782 The ``User`` class is the common base class of all LLVM nodes that may refer to
2783 ``Value``\ s. It exposes a list of "Operands" that are all of the ``Value``\ s
2784 that the User is referring to. The ``User`` class itself is a subclass of
2787 The operands of a ``User`` point directly to the LLVM ``Value`` that it refers
2788 to. Because LLVM uses Static Single Assignment (SSA) form, there can only be
2789 one definition referred to, allowing this direct connection. This connection
2790 provides the use-def information in LLVM.
2794 Important Public Members of the ``User`` class
2795 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2797 The ``User`` class exposes the operand list in two ways: through an index access
2798 interface and through an iterator based interface.
2800 * | ``Value *getOperand(unsigned i)``
2801 | ``unsigned getNumOperands()``
2803 These two methods expose the operands of the ``User`` in a convenient form for
2806 * | ``User::op_iterator`` - Typedef for iterator over the operand list
2807 | ``op_iterator op_begin()`` - Get an iterator to the start of the operand
2809 | ``op_iterator op_end()`` - Get an iterator to the end of the operand list.
2811 Together, these methods make up the iterator based interface to the operands
2817 The ``Instruction`` class
2818 -------------------------
2820 ``#include "llvm/IR/Instruction.h"``
2822 header source: `Instruction.h
2823 <http://llvm.org/doxygen/Instruction_8h-source.html>`_
2825 doxygen info: `Instruction Class
2826 <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_
2828 Superclasses: User_, Value_
2830 The ``Instruction`` class is the common base class for all LLVM instructions.
2831 It provides only a few methods, but is a very commonly used class. The primary
2832 data tracked by the ``Instruction`` class itself is the opcode (instruction
2833 type) and the parent BasicBlock_ the ``Instruction`` is embedded into. To
2834 represent a specific type of instruction, one of many subclasses of
2835 ``Instruction`` are used.
2837 Because the ``Instruction`` class subclasses the User_ class, its operands can
2838 be accessed in the same way as for other ``User``\ s (with the
2839 ``getOperand()``/``getNumOperands()`` and ``op_begin()``/``op_end()`` methods).
2840 An important file for the ``Instruction`` class is the ``llvm/Instruction.def``
2841 file. This file contains some meta-data about the various different types of
2842 instructions in LLVM. It describes the enum values that are used as opcodes
2843 (for example ``Instruction::Add`` and ``Instruction::ICmp``), as well as the
2844 concrete sub-classes of ``Instruction`` that implement the instruction (for
2845 example BinaryOperator_ and CmpInst_). Unfortunately, the use of macros in this
2846 file confuses doxygen, so these enum values don't show up correctly in the
2847 `doxygen output <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_.
2851 Important Subclasses of the ``Instruction`` class
2852 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2856 * ``BinaryOperator``
2858 This subclasses represents all two operand instructions whose operands must be
2859 the same type, except for the comparison instructions.
2864 This subclass is the parent of the 12 casting instructions. It provides
2865 common operations on cast instructions.
2871 This subclass respresents the two comparison instructions,
2872 `ICmpInst <LangRef.html#i_icmp>`_ (integer opreands), and
2873 `FCmpInst <LangRef.html#i_fcmp>`_ (floating point operands).
2877 * ``TerminatorInst``
2879 This subclass is the parent of all terminator instructions (those which can
2884 Important Public Members of the ``Instruction`` class
2885 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2887 * ``BasicBlock *getParent()``
2889 Returns the BasicBlock_ that this
2890 ``Instruction`` is embedded into.
2892 * ``bool mayWriteToMemory()``
2894 Returns true if the instruction writes to memory, i.e. it is a ``call``,
2895 ``free``, ``invoke``, or ``store``.
2897 * ``unsigned getOpcode()``
2899 Returns the opcode for the ``Instruction``.
2901 * ``Instruction *clone() const``
2903 Returns another instance of the specified instruction, identical in all ways
2904 to the original except that the instruction has no parent (i.e. it's not
2905 embedded into a BasicBlock_), and it has no name.
2909 The ``Constant`` class and subclasses
2910 -------------------------------------
2912 Constant represents a base class for different types of constants. It is
2913 subclassed by ConstantInt, ConstantArray, etc. for representing the various
2914 types of Constants. GlobalValue_ is also a subclass, which represents the
2915 address of a global variable or function.
2919 Important Subclasses of Constant
2920 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2922 * ConstantInt : This subclass of Constant represents an integer constant of
2925 * ``const APInt& getValue() const``: Returns the underlying
2926 value of this constant, an APInt value.
2928 * ``int64_t getSExtValue() const``: Converts the underlying APInt value to an
2929 int64_t via sign extension. If the value (not the bit width) of the APInt
2930 is too large to fit in an int64_t, an assertion will result. For this
2931 reason, use of this method is discouraged.
2933 * ``uint64_t getZExtValue() const``: Converts the underlying APInt value
2934 to a uint64_t via zero extension. IF the value (not the bit width) of the
2935 APInt is too large to fit in a uint64_t, an assertion will result. For this
2936 reason, use of this method is discouraged.
2938 * ``static ConstantInt* get(const APInt& Val)``: Returns the ConstantInt
2939 object that represents the value provided by ``Val``. The type is implied
2940 as the IntegerType that corresponds to the bit width of ``Val``.
2942 * ``static ConstantInt* get(const Type *Ty, uint64_t Val)``: Returns the
2943 ConstantInt object that represents the value provided by ``Val`` for integer
2946 * ConstantFP : This class represents a floating point constant.
2948 * ``double getValue() const``: Returns the underlying value of this constant.
2950 * ConstantArray : This represents a constant array.
2952 * ``const std::vector<Use> &getValues() const``: Returns a vector of
2953 component constants that makeup this array.
2955 * ConstantStruct : This represents a constant struct.
2957 * ``const std::vector<Use> &getValues() const``: Returns a vector of
2958 component constants that makeup this array.
2960 * GlobalValue : This represents either a global variable or a function. In
2961 either case, the value is a constant fixed address (after linking).
2965 The ``GlobalValue`` class
2966 -------------------------
2968 ``#include "llvm/IR/GlobalValue.h"``
2970 header source: `GlobalValue.h
2971 <http://llvm.org/doxygen/GlobalValue_8h-source.html>`_
2973 doxygen info: `GlobalValue Class
2974 <http://llvm.org/doxygen/classllvm_1_1GlobalValue.html>`_
2976 Superclasses: Constant_, User_, Value_
2978 Global values ( GlobalVariable_\ s or :ref:`Function <c_Function>`\ s) are the
2979 only LLVM values that are visible in the bodies of all :ref:`Function
2980 <c_Function>`\ s. Because they are visible at global scope, they are also
2981 subject to linking with other globals defined in different translation units.
2982 To control the linking process, ``GlobalValue``\ s know their linkage rules.
2983 Specifically, ``GlobalValue``\ s know whether they have internal or external
2984 linkage, as defined by the ``LinkageTypes`` enumeration.
2986 If a ``GlobalValue`` has internal linkage (equivalent to being ``static`` in C),
2987 it is not visible to code outside the current translation unit, and does not
2988 participate in linking. If it has external linkage, it is visible to external
2989 code, and does participate in linking. In addition to linkage information,
2990 ``GlobalValue``\ s keep track of which Module_ they are currently part of.
2992 Because ``GlobalValue``\ s are memory objects, they are always referred to by
2993 their **address**. As such, the Type_ of a global is always a pointer to its
2994 contents. It is important to remember this when using the ``GetElementPtrInst``
2995 instruction because this pointer must be dereferenced first. For example, if
2996 you have a ``GlobalVariable`` (a subclass of ``GlobalValue)`` that is an array
2997 of 24 ints, type ``[24 x i32]``, then the ``GlobalVariable`` is a pointer to
2998 that array. Although the address of the first element of this array and the
2999 value of the ``GlobalVariable`` are the same, they have different types. The
3000 ``GlobalVariable``'s type is ``[24 x i32]``. The first element's type is
3001 ``i32.`` Because of this, accessing a global value requires you to dereference
3002 the pointer with ``GetElementPtrInst`` first, then its elements can be accessed.
3003 This is explained in the `LLVM Language Reference Manual
3004 <LangRef.html#globalvars>`_.
3008 Important Public Members of the ``GlobalValue`` class
3009 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3011 * | ``bool hasInternalLinkage() const``
3012 | ``bool hasExternalLinkage() const``
3013 | ``void setInternalLinkage(bool HasInternalLinkage)``
3015 These methods manipulate the linkage characteristics of the ``GlobalValue``.
3017 * ``Module *getParent()``
3019 This returns the Module_ that the
3020 GlobalValue is currently embedded into.
3024 The ``Function`` class
3025 ----------------------
3027 ``#include "llvm/IR/Function.h"``
3029 header source: `Function.h <http://llvm.org/doxygen/Function_8h-source.html>`_
3031 doxygen info: `Function Class
3032 <http://llvm.org/doxygen/classllvm_1_1Function.html>`_
3034 Superclasses: GlobalValue_, Constant_, User_, Value_
3036 The ``Function`` class represents a single procedure in LLVM. It is actually
3037 one of the more complex classes in the LLVM hierarchy because it must keep track
3038 of a large amount of data. The ``Function`` class keeps track of a list of
3039 BasicBlock_\ s, a list of formal Argument_\ s, and a SymbolTable_.
3041 The list of BasicBlock_\ s is the most commonly used part of ``Function``
3042 objects. The list imposes an implicit ordering of the blocks in the function,
3043 which indicate how the code will be laid out by the backend. Additionally, the
3044 first BasicBlock_ is the implicit entry node for the ``Function``. It is not
3045 legal in LLVM to explicitly branch to this initial block. There are no implicit
3046 exit nodes, and in fact there may be multiple exit nodes from a single
3047 ``Function``. If the BasicBlock_ list is empty, this indicates that the
3048 ``Function`` is actually a function declaration: the actual body of the function
3049 hasn't been linked in yet.
3051 In addition to a list of BasicBlock_\ s, the ``Function`` class also keeps track
3052 of the list of formal Argument_\ s that the function receives. This container
3053 manages the lifetime of the Argument_ nodes, just like the BasicBlock_ list does
3054 for the BasicBlock_\ s.
3056 The SymbolTable_ is a very rarely used LLVM feature that is only used when you
3057 have to look up a value by name. Aside from that, the SymbolTable_ is used
3058 internally to make sure that there are not conflicts between the names of
3059 Instruction_\ s, BasicBlock_\ s, or Argument_\ s in the function body.
3061 Note that ``Function`` is a GlobalValue_ and therefore also a Constant_. The
3062 value of the function is its address (after linking) which is guaranteed to be
3067 Important Public Members of the ``Function``
3068 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3070 * ``Function(const FunctionType *Ty, LinkageTypes Linkage,
3071 const std::string &N = "", Module* Parent = 0)``
3073 Constructor used when you need to create new ``Function``\ s to add the
3074 program. The constructor must specify the type of the function to create and
3075 what type of linkage the function should have. The FunctionType_ argument
3076 specifies the formal arguments and return value for the function. The same
3077 FunctionType_ value can be used to create multiple functions. The ``Parent``
3078 argument specifies the Module in which the function is defined. If this
3079 argument is provided, the function will automatically be inserted into that
3080 module's list of functions.
3082 * ``bool isDeclaration()``
3084 Return whether or not the ``Function`` has a body defined. If the function is
3085 "external", it does not have a body, and thus must be resolved by linking with
3086 a function defined in a different translation unit.
3088 * | ``Function::iterator`` - Typedef for basic block list iterator
3089 | ``Function::const_iterator`` - Typedef for const_iterator.
3090 | ``begin()``, ``end()``, ``size()``, ``empty()``
3092 These are forwarding methods that make it easy to access the contents of a
3093 ``Function`` object's BasicBlock_ list.
3095 * ``Function::BasicBlockListType &getBasicBlockList()``
3097 Returns the list of BasicBlock_\ s. This is necessary to use when you need to
3098 update the list or perform a complex action that doesn't have a forwarding
3101 * | ``Function::arg_iterator`` - Typedef for the argument list iterator
3102 | ``Function::const_arg_iterator`` - Typedef for const_iterator.
3103 | ``arg_begin()``, ``arg_end()``, ``arg_size()``, ``arg_empty()``
3105 These are forwarding methods that make it easy to access the contents of a
3106 ``Function`` object's Argument_ list.
3108 * ``Function::ArgumentListType &getArgumentList()``
3110 Returns the list of Argument_. This is necessary to use when you need to
3111 update the list or perform a complex action that doesn't have a forwarding
3114 * ``BasicBlock &getEntryBlock()``
3116 Returns the entry ``BasicBlock`` for the function. Because the entry block
3117 for the function is always the first block, this returns the first block of
3120 * | ``Type *getReturnType()``
3121 | ``FunctionType *getFunctionType()``
3123 This traverses the Type_ of the ``Function`` and returns the return type of
3124 the function, or the FunctionType_ of the actual function.
3126 * ``SymbolTable *getSymbolTable()``
3128 Return a pointer to the SymbolTable_ for this ``Function``.
3132 The ``GlobalVariable`` class
3133 ----------------------------
3135 ``#include "llvm/IR/GlobalVariable.h"``
3137 header source: `GlobalVariable.h
3138 <http://llvm.org/doxygen/GlobalVariable_8h-source.html>`_
3140 doxygen info: `GlobalVariable Class
3141 <http://llvm.org/doxygen/classllvm_1_1GlobalVariable.html>`_
3143 Superclasses: GlobalValue_, Constant_, User_, Value_
3145 Global variables are represented with the (surprise surprise) ``GlobalVariable``
3146 class. Like functions, ``GlobalVariable``\ s are also subclasses of
3147 GlobalValue_, and as such are always referenced by their address (global values
3148 must live in memory, so their "name" refers to their constant address). See
3149 GlobalValue_ for more on this. Global variables may have an initial value
3150 (which must be a Constant_), and if they have an initializer, they may be marked
3151 as "constant" themselves (indicating that their contents never change at
3154 .. _m_GlobalVariable:
3156 Important Public Members of the ``GlobalVariable`` class
3157 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3159 * ``GlobalVariable(const Type *Ty, bool isConstant, LinkageTypes &Linkage,
3160 Constant *Initializer = 0, const std::string &Name = "", Module* Parent = 0)``
3162 Create a new global variable of the specified type. If ``isConstant`` is true
3163 then the global variable will be marked as unchanging for the program. The
3164 Linkage parameter specifies the type of linkage (internal, external, weak,
3165 linkonce, appending) for the variable. If the linkage is InternalLinkage,
3166 WeakAnyLinkage, WeakODRLinkage, LinkOnceAnyLinkage or LinkOnceODRLinkage, then
3167 the resultant global variable will have internal linkage. AppendingLinkage
3168 concatenates together all instances (in different translation units) of the
3169 variable into a single variable but is only applicable to arrays. See the
3170 `LLVM Language Reference <LangRef.html#modulestructure>`_ for further details
3171 on linkage types. Optionally an initializer, a name, and the module to put
3172 the variable into may be specified for the global variable as well.
3174 * ``bool isConstant() const``
3176 Returns true if this is a global variable that is known not to be modified at
3179 * ``bool hasInitializer()``
3181 Returns true if this ``GlobalVariable`` has an intializer.
3183 * ``Constant *getInitializer()``
3185 Returns the initial value for a ``GlobalVariable``. It is not legal to call
3186 this method if there is no initializer.
3190 The ``BasicBlock`` class
3191 ------------------------
3193 ``#include "llvm/IR/BasicBlock.h"``
3195 header source: `BasicBlock.h
3196 <http://llvm.org/doxygen/BasicBlock_8h-source.html>`_
3198 doxygen info: `BasicBlock Class
3199 <http://llvm.org/doxygen/classllvm_1_1BasicBlock.html>`_
3203 This class represents a single entry single exit section of the code, commonly
3204 known as a basic block by the compiler community. The ``BasicBlock`` class
3205 maintains a list of Instruction_\ s, which form the body of the block. Matching
3206 the language definition, the last element of this list of instructions is always
3207 a terminator instruction (a subclass of the TerminatorInst_ class).
3209 In addition to tracking the list of instructions that make up the block, the
3210 ``BasicBlock`` class also keeps track of the :ref:`Function <c_Function>` that
3211 it is embedded into.
3213 Note that ``BasicBlock``\ s themselves are Value_\ s, because they are
3214 referenced by instructions like branches and can go in the switch tables.
3215 ``BasicBlock``\ s have type ``label``.
3219 Important Public Members of the ``BasicBlock`` class
3220 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3222 * ``BasicBlock(const std::string &Name = "", Function *Parent = 0)``
3224 The ``BasicBlock`` constructor is used to create new basic blocks for
3225 insertion into a function. The constructor optionally takes a name for the
3226 new block, and a :ref:`Function <c_Function>` to insert it into. If the
3227 ``Parent`` parameter is specified, the new ``BasicBlock`` is automatically
3228 inserted at the end of the specified :ref:`Function <c_Function>`, if not
3229 specified, the BasicBlock must be manually inserted into the :ref:`Function
3232 * | ``BasicBlock::iterator`` - Typedef for instruction list iterator
3233 | ``BasicBlock::const_iterator`` - Typedef for const_iterator.
3234 | ``begin()``, ``end()``, ``front()``, ``back()``,
3235 ``size()``, ``empty()``
3236 STL-style functions for accessing the instruction list.
3238 These methods and typedefs are forwarding functions that have the same
3239 semantics as the standard library methods of the same names. These methods
3240 expose the underlying instruction list of a basic block in a way that is easy
3241 to manipulate. To get the full complement of container operations (including
3242 operations to update the list), you must use the ``getInstList()`` method.
3244 * ``BasicBlock::InstListType &getInstList()``
3246 This method is used to get access to the underlying container that actually
3247 holds the Instructions. This method must be used when there isn't a
3248 forwarding function in the ``BasicBlock`` class for the operation that you
3249 would like to perform. Because there are no forwarding functions for
3250 "updating" operations, you need to use this if you want to update the contents
3251 of a ``BasicBlock``.
3253 * ``Function *getParent()``
3255 Returns a pointer to :ref:`Function <c_Function>` the block is embedded into,
3256 or a null pointer if it is homeless.
3258 * ``TerminatorInst *getTerminator()``
3260 Returns a pointer to the terminator instruction that appears at the end of the
3261 ``BasicBlock``. If there is no terminator instruction, or if the last
3262 instruction in the block is not a terminator, then a null pointer is returned.
3266 The ``Argument`` class
3267 ----------------------
3269 This subclass of Value defines the interface for incoming formal arguments to a
3270 function. A Function maintains a list of its formal arguments. An argument has
3271 a pointer to the parent Function.