Home | History | Annotate | Download | only in docs
      1 ========================
      2 LLVM Programmer's Manual
      3 ========================
      4 
      5 .. contents::
      6    :local:
      7 
      8 .. warning::
      9    This is always a work in progress.
     10 
     11 .. _introduction:
     12 
     13 Introduction
     14 ============
     15 
     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.
     21 
     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
     28 possible.
     29 
     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.
     36 
     37 .. _general:
     38 
     39 General Information
     40 ===================
     41 
     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.
     44 
     45 .. _stl:
     46 
     47 The C++ Standard Template Library
     48 ---------------------------------
     49 
     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.
     55 
     56 Here are some useful links:
     57 
     58 #. `cppreference.com
     59    <http://en.cppreference.com/w/>`_ - an excellent
     60    reference for the STL and other parts of the standard C++ library.
     61 
     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
     65    published.
     66 
     67 #. `C++ Frequently Asked Questions <http://www.parashift.com/c++-faq-lite/>`_.
     68 
     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>`_.
     72 
     73 #. `Bjarne Stroustrup's C++ Page
     74    <http://www.research.att.com/%7Ebs/C++.html>`_.
     75 
     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>`_.
     79 
     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.
     83 
     84 .. _resources:
     85 
     86 Other useful references
     87 -----------------------
     88 
     89 #. `Using static and shared libraries across platforms
     90    <http://www.fortran-2000.com/ArnaudRecipes/sharedlib.html>`_
     91 
     92 .. _apis:
     93 
     94 Important and useful LLVM APIs
     95 ==============================
     96 
     97 Here we highlight some LLVM APIs that are generally useful and good to know
     98 about when writing transformations.
     99 
    100 .. _isa:
    101 
    102 The ``isa<>``, ``cast<>`` and ``dyn_cast<>`` templates
    103 ------------------------------------------------------
    104 
    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).
    113 
    114 ``isa<>``:
    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).
    119 
    120 ``cast<>``:
    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<>``
    126   template is:
    127 
    128   .. code-block:: c++
    129 
    130     static bool isLoopInvariant(const Value *V, const Loop *L) {
    131       if (isa<Constant>(V) || isa<Argument>(V) || isa<GlobalValue>(V))
    132         return true;
    133 
    134       // Otherwise, it must be an instruction...
    135       return !L->contains(cast<Instruction>(V)->getParent());
    136     }
    137 
    138   Note that you should **not** use an ``isa<>`` test followed by a ``cast<>``,
    139   for that use the ``dyn_cast<>`` operator.
    140 
    141 ``dyn_cast<>``:
    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:
    149 
    150   .. code-block:: c++
    151 
    152     if (AllocationInst *AI = dyn_cast<AllocationInst>(Val)) {
    153       // ...
    154     }
    155 
    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
    158   convenient.
    159 
    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
    165   type directly.
    166 
    167 ``cast_or_null<>``:
    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.
    172 
    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.
    178 
    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>`
    183 
    184 .. _string_apis:
    185 
    186 Passing strings (the ``StringRef`` and ``Twine`` classes)
    187 ---------------------------------------------------------
    188 
    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.
    193 
    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.
    199 
    200 .. _StringRef:
    201 
    202 The ``StringRef`` class
    203 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    204 
    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.
    208 
    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:
    212 
    213 .. code-block:: c++
    214 
    215   iterator find(StringRef Key);
    216 
    217 and clients can call it using any one of:
    218 
    219 .. code-block:: c++
    220 
    221   Map.find("foo");                 // Lookup "foo"
    222   Map.find(std::string("bar"));    // Lookup "bar"
    223   Map.find(StringRef("\0baz", 4)); // Lookup "\0baz"
    224 
    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
    229 information.
    230 
    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
    235 passed by value.
    236 
    237 The ``Twine`` class
    238 ^^^^^^^^^^^^^^^^^^^
    239 
    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:
    244 
    245 .. code-block:: c++
    246 
    247     New = CmpInst::Create(..., SO->getName() + ".cmp");
    248 
    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.
    260 
    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
    264 strings.
    265 
    266 .. _error_apis:
    267 
    268 Error handling
    269 --------------
    270 
    271 Proper error handling helps us identify bugs in our code, and helps end-users
    272 understand errors in their tool usage. Errors fall into two broad categories:
    273 *programmatic* and *recoverable*, with different strategies for handling and
    274 reporting.
    275 
    276 Programmatic Errors
    277 ^^^^^^^^^^^^^^^^^^^
    278 
    279 Programmatic errors are violations of program invariants or API contracts, and
    280 represent bugs within the program itself. Our aim is to document invariants, and
    281 to abort quickly at the point of failure (providing some basic diagnostic) when
    282 invariants are broken at runtime.
    283 
    284 The fundamental tools for handling programmatic errors are assertions and the
    285 llvm_unreachable function. Assertions are used to express invariant conditions,
    286 and should include a message describing the invariant:
    287 
    288 .. code-block:: c++
    289 
    290   assert(isPhysReg(R) && "All virt regs should have been allocated already.");
    291 
    292 The llvm_unreachable function can be used to document areas of control flow
    293 that should never be entered if the program invariants hold:
    294 
    295 .. code-block:: c++
    296 
    297   enum { Foo, Bar, Baz } X = foo();
    298 
    299   switch (X) {
    300     case Foo: /* Handle Foo */; break;
    301     case Bar: /* Handle Bar */; break;
    302     default:
    303       llvm_unreachable("X should be Foo or Bar here");
    304   }
    305 
    306 Recoverable Errors
    307 ^^^^^^^^^^^^^^^^^^
    308 
    309 Recoverable errors represent an error in the program's environment, for example
    310 a resource failure (a missing file, a dropped network connection, etc.), or
    311 malformed input. These errors should be detected and communicated to a level of
    312 the program where they can be handled appropriately. Handling the error may be
    313 as simple as reporting the issue to the user, or it may involve attempts at
    314 recovery.
    315 
    316 Recoverable errors are modeled using LLVM's ``Error`` scheme. This scheme
    317 represents errors using function return values, similar to classic C integer
    318 error codes, or C++'s ``std::error_code``. However, the ``Error`` class is
    319 actually a lightweight wrapper for user-defined error types, allowing arbitrary
    320 information to be attached to describe the error. This is similar to the way C++
    321 exceptions allow throwing of user-defined types.
    322 
    323 Success values are created by calling ``Error::success()``:
    324 
    325 .. code-block:: c++
    326 
    327   Error foo() {
    328     // Do something.
    329     // Return success.
    330     return Error::success();
    331   }
    332 
    333 Success values are very cheap to construct and return - they have minimal
    334 impact on program performance.
    335 
    336 Failure values are constructed using ``make_error<T>``, where ``T`` is any class
    337 that inherits from the ErrorInfo utility:
    338 
    339 .. code-block:: c++
    340 
    341   class MyError : public ErrorInfo<MyError> {
    342   public:
    343     MyError(std::string Msg) : Msg(Msg) {}
    344     void log(OStream &OS) const override { OS << "MyError - " << Msg; }
    345     static char ID;
    346   private:
    347     std::string Msg;
    348   };
    349 
    350   char MyError::ID = 0; // In MyError.cpp
    351 
    352   Error bar() {
    353     if (checkErrorCondition)
    354       return make_error<MyError>("Error condition detected");
    355 
    356     // No error - proceed with bar.
    357 
    358     // Return success value.
    359     return Error::success();
    360   }
    361 
    362 Error values can be implicitly converted to bool: true for error, false for
    363 success, enabling the following idiom:
    364 
    365 .. code-block:: c++
    366 
    367   Error mayFail();
    368 
    369   Error foo() {
    370     if (auto Err = mayFail())
    371       return Err;
    372     // Success! We can proceed.
    373     ...
    374 
    375 For functions that can fail but need to return a value the ``Expected<T>``
    376 utility can be used. Values of this type can be constructed with either a
    377 ``T``, or a ``Error``. Expected<T> values are also implicitly convertible to
    378 boolean, but with the opposite convention to Error: true for success, false for
    379 error. If success, the ``T`` value can be accessed via the dereference operator.
    380 If failure, the ``Error`` value can be extracted using the ``takeError()``
    381 method. Idiomatic usage looks like:
    382 
    383 .. code-block:: c++
    384 
    385   Expected<float> parseAndSquareRoot(IStream &IS) {
    386     float f;
    387     OS >> f;
    388     if (f < 0)
    389       return make_error<FloatingPointError>(...);
    390     return sqrt(f);
    391   }
    392 
    393   Error foo(IStream &IS) {
    394     if (auto SqrtOrErr = parseAndSquartRoot(IS)) {
    395       float Sqrt = *SqrtOrErr;
    396       // ...
    397     } else
    398       return SqrtOrErr.takeError();
    399   }
    400 
    401 All Error instances, whether success or failure, must be either checked or
    402 moved from (via std::move or a return) before they are destructed. Accidentally
    403 discarding an unchecked error will cause a program abort at the point where the
    404 unchecked value's destructor is run, making it easy to identify and fix
    405 violations of this rule.
    406 
    407 Success values are considered checked once they have been tested (by invoking
    408 the boolean conversion operator):
    409 
    410 .. code-block:: c++
    411 
    412   if (auto Err = canFail(...))
    413     return Err; // Failure value - move error to caller.
    414 
    415   // Safe to continue: Err was checked.
    416 
    417 In contrast, the following code will always cause an abort, regardless of the
    418 return value of ``foo``:
    419 
    420 .. code-block:: c++
    421 
    422     canFail();
    423     // Program will always abort here, even if canFail() returns Success, since
    424     // the value is not checked.
    425 
    426 Failure values are considered checked once a handler for the error type has
    427 been activated:
    428 
    429 .. code-block:: c++
    430 
    431   auto Err = canFail(...);
    432   if (auto Err2 =
    433        handleErrors(std::move(Err),
    434          [](std::unique_ptr<MyError> M) {
    435            // Try to handle 'M'. If successful, return a success value from
    436            // the handler.
    437            if (tryToHandle(M))
    438              return Error::success();
    439 
    440            // We failed to handle 'M' - return it from the handler.
    441            // This value will be passed back from catchErrors and
    442            // wind up in Err2, where it will be returned from this function.
    443            return Error(std::move(M));
    444          })))
    445     return Err2;
    446 
    447 
    448 .. _function_apis:
    449 
    450 More information on Error and its related utilities can be found in the
    451 Error.h header file.
    452 
    453 Passing functions and other callable objects
    454 --------------------------------------------
    455 
    456 Sometimes you may want a function to be passed a callback object. In order to
    457 support lambda expressions and other function objects, you should not use the
    458 traditional C approach of taking a function pointer and an opaque cookie:
    459 
    460 .. code-block:: c++
    461 
    462     void takeCallback(bool (*Callback)(Function *, void *), void *Cookie);
    463 
    464 Instead, use one of the following approaches:
    465 
    466 Function template
    467 ^^^^^^^^^^^^^^^^^
    468 
    469 If you don't mind putting the definition of your function into a header file,
    470 make it a function template that is templated on the callable type.
    471 
    472 .. code-block:: c++
    473 
    474     template<typename Callable>
    475     void takeCallback(Callable Callback) {
    476       Callback(1, 2, 3);
    477     }
    478 
    479 The ``function_ref`` class template
    480 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    481 
    482 The ``function_ref``
    483 (`doxygen <http://llvm.org/docs/doxygen/html/classllvm_1_1function__ref_3_01Ret_07Params_8_8_8_08_4.html>`__) class
    484 template represents a reference to a callable object, templated over the type
    485 of the callable. This is a good choice for passing a callback to a function,
    486 if you don't need to hold onto the callback after the function returns. In this
    487 way, ``function_ref`` is to ``std::function`` as ``StringRef`` is to
    488 ``std::string``.
    489 
    490 ``function_ref<Ret(Param1, Param2, ...)>`` can be implicitly constructed from
    491 any callable object that can be called with arguments of type ``Param1``,
    492 ``Param2``, ..., and returns a value that can be converted to type ``Ret``.
    493 For example:
    494 
    495 .. code-block:: c++
    496 
    497     void visitBasicBlocks(Function *F, function_ref<bool (BasicBlock*)> Callback) {
    498       for (BasicBlock &BB : *F)
    499         if (Callback(&BB))
    500           return;
    501     }
    502 
    503 can be called using:
    504 
    505 .. code-block:: c++
    506 
    507     visitBasicBlocks(F, [&](BasicBlock *BB) {
    508       if (process(BB))
    509         return isEmpty(BB);
    510       return false;
    511     });
    512 
    513 Note that a ``function_ref`` object contains pointers to external memory, so it
    514 is not generally safe to store an instance of the class (unless you know that
    515 the external storage will not be freed). If you need this ability, consider
    516 using ``std::function``. ``function_ref`` is small enough that it should always
    517 be passed by value.
    518 
    519 .. _DEBUG:
    520 
    521 The ``DEBUG()`` macro and ``-debug`` option
    522 -------------------------------------------
    523 
    524 Often when working on your pass you will put a bunch of debugging printouts and
    525 other code into your pass.  After you get it working, you want to remove it, but
    526 you may need it again in the future (to work out new bugs that you run across).
    527 
    528 Naturally, because of this, you don't want to delete the debug printouts, but
    529 you don't want them to always be noisy.  A standard compromise is to comment
    530 them out, allowing you to enable them if you need them in the future.
    531 
    532 The ``llvm/Support/Debug.h`` (`doxygen
    533 <http://llvm.org/doxygen/Debug_8h-source.html>`__) file provides a macro named
    534 ``DEBUG()`` that is a much nicer solution to this problem.  Basically, you can
    535 put arbitrary code into the argument of the ``DEBUG`` macro, and it is only
    536 executed if '``opt``' (or any other tool) is run with the '``-debug``' command
    537 line argument:
    538 
    539 .. code-block:: c++
    540 
    541   DEBUG(errs() << "I am here!\n");
    542 
    543 Then you can run your pass like this:
    544 
    545 .. code-block:: none
    546 
    547   $ opt < a.bc > /dev/null -mypass
    548   <no output>
    549   $ opt < a.bc > /dev/null -mypass -debug
    550   I am here!
    551 
    552 Using the ``DEBUG()`` macro instead of a home-brewed solution allows you to not
    553 have to create "yet another" command line option for the debug output for your
    554 pass.  Note that ``DEBUG()`` macros are disabled for non-asserts builds, so they
    555 do not cause a performance impact at all (for the same reason, they should also
    556 not contain side-effects!).
    557 
    558 One additional nice thing about the ``DEBUG()`` macro is that you can enable or
    559 disable it directly in gdb.  Just use "``set DebugFlag=0``" or "``set
    560 DebugFlag=1``" from the gdb if the program is running.  If the program hasn't
    561 been started yet, you can always just run it with ``-debug``.
    562 
    563 .. _DEBUG_TYPE:
    564 
    565 Fine grained debug info with ``DEBUG_TYPE`` and the ``-debug-only`` option
    566 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    567 
    568 Sometimes you may find yourself in a situation where enabling ``-debug`` just
    569 turns on **too much** information (such as when working on the code generator).
    570 If you want to enable debug information with more fine-grained control, you
    571 should define the ``DEBUG_TYPE`` macro and use the ``-debug-only`` option as
    572 follows:
    573 
    574 .. code-block:: c++
    575 
    576   #define DEBUG_TYPE "foo"
    577   DEBUG(errs() << "'foo' debug type\n");
    578   #undef  DEBUG_TYPE
    579   #define DEBUG_TYPE "bar"
    580   DEBUG(errs() << "'bar' debug type\n"));
    581   #undef  DEBUG_TYPE
    582 
    583 Then you can run your pass like this:
    584 
    585 .. code-block:: none
    586 
    587   $ opt < a.bc > /dev/null -mypass
    588   <no output>
    589   $ opt < a.bc > /dev/null -mypass -debug
    590   'foo' debug type
    591   'bar' debug type
    592   $ opt < a.bc > /dev/null -mypass -debug-only=foo
    593   'foo' debug type
    594   $ opt < a.bc > /dev/null -mypass -debug-only=bar
    595   'bar' debug type
    596   $ opt < a.bc > /dev/null -mypass -debug-only=foo,bar
    597   'foo' debug type
    598   'bar' debug type
    599 
    600 Of course, in practice, you should only set ``DEBUG_TYPE`` at the top of a file,
    601 to specify the debug type for the entire module. Be careful that you only do
    602 this after including Debug.h and not around any #include of headers. Also, you
    603 should use names more meaningful than "foo" and "bar", because there is no
    604 system in place to ensure that names do not conflict. If two different modules
    605 use the same string, they will all be turned on when the name is specified.
    606 This allows, for example, all debug information for instruction scheduling to be
    607 enabled with ``-debug-only=InstrSched``, even if the source lives in multiple
    608 files. The name must not include a comma (,) as that is used to separate the
    609 arguments of the ``-debug-only`` option.
    610 
    611 For performance reasons, -debug-only is not available in optimized build
    612 (``--enable-optimized``) of LLVM.
    613 
    614 The ``DEBUG_WITH_TYPE`` macro is also available for situations where you would
    615 like to set ``DEBUG_TYPE``, but only for one specific ``DEBUG`` statement.  It
    616 takes an additional first parameter, which is the type to use.  For example, the
    617 preceding example could be written as:
    618 
    619 .. code-block:: c++
    620 
    621   DEBUG_WITH_TYPE("foo", errs() << "'foo' debug type\n");
    622   DEBUG_WITH_TYPE("bar", errs() << "'bar' debug type\n"));
    623 
    624 .. _Statistic:
    625 
    626 The ``Statistic`` class & ``-stats`` option
    627 -------------------------------------------
    628 
    629 The ``llvm/ADT/Statistic.h`` (`doxygen
    630 <http://llvm.org/doxygen/Statistic_8h-source.html>`__) file provides a class
    631 named ``Statistic`` that is used as a unified way to keep track of what the LLVM
    632 compiler is doing and how effective various optimizations are.  It is useful to
    633 see what optimizations are contributing to making a particular program run
    634 faster.
    635 
    636 Often you may run your pass on some big program, and you're interested to see
    637 how many times it makes a certain transformation.  Although you can do this with
    638 hand inspection, or some ad-hoc method, this is a real pain and not very useful
    639 for big programs.  Using the ``Statistic`` class makes it very easy to keep
    640 track of this information, and the calculated information is presented in a
    641 uniform manner with the rest of the passes being executed.
    642 
    643 There are many examples of ``Statistic`` uses, but the basics of using it are as
    644 follows:
    645 
    646 #. Define your statistic like this:
    647 
    648   .. code-block:: c++
    649 
    650     #define DEBUG_TYPE "mypassname"   // This goes before any #includes.
    651     STATISTIC(NumXForms, "The # of times I did stuff");
    652 
    653   The ``STATISTIC`` macro defines a static variable, whose name is specified by
    654   the first argument.  The pass name is taken from the ``DEBUG_TYPE`` macro, and
    655   the description is taken from the second argument.  The variable defined
    656   ("NumXForms" in this case) acts like an unsigned integer.
    657 
    658 #. Whenever you make a transformation, bump the counter:
    659 
    660   .. code-block:: c++
    661 
    662     ++NumXForms;   // I did stuff!
    663 
    664 That's all you have to do.  To get '``opt``' to print out the statistics
    665 gathered, use the '``-stats``' option:
    666 
    667 .. code-block:: none
    668 
    669   $ opt -stats -mypassname < program.bc > /dev/null
    670   ... statistics output ...
    671 
    672 Note that in order to use the '``-stats``' option, LLVM must be
    673 compiled with assertions enabled.
    674 
    675 When running ``opt`` on a C file from the SPEC benchmark suite, it gives a
    676 report that looks like this:
    677 
    678 .. code-block:: none
    679 
    680    7646 bitcodewriter   - Number of normal instructions
    681     725 bitcodewriter   - Number of oversized instructions
    682  129996 bitcodewriter   - Number of bitcode bytes written
    683    2817 raise           - Number of insts DCEd or constprop'd
    684    3213 raise           - Number of cast-of-self removed
    685    5046 raise           - Number of expression trees converted
    686      75 raise           - Number of other getelementptr's formed
    687     138 raise           - Number of load/store peepholes
    688      42 deadtypeelim    - Number of unused typenames removed from symtab
    689     392 funcresolve     - Number of varargs functions resolved
    690      27 globaldce       - Number of global variables removed
    691       2 adce            - Number of basic blocks removed
    692     134 cee             - Number of branches revectored
    693      49 cee             - Number of setcc instruction eliminated
    694     532 gcse            - Number of loads removed
    695    2919 gcse            - Number of instructions removed
    696      86 indvars         - Number of canonical indvars added
    697      87 indvars         - Number of aux indvars removed
    698      25 instcombine     - Number of dead inst eliminate
    699     434 instcombine     - Number of insts combined
    700     248 licm            - Number of load insts hoisted
    701    1298 licm            - Number of insts hoisted to a loop pre-header
    702       3 licm            - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
    703      75 mem2reg         - Number of alloca's promoted
    704    1444 cfgsimplify     - Number of blocks simplified
    705 
    706 Obviously, with so many optimizations, having a unified framework for this stuff
    707 is very nice.  Making your pass fit well into the framework makes it more
    708 maintainable and useful.
    709 
    710 .. _ViewGraph:
    711 
    712 Viewing graphs while debugging code
    713 -----------------------------------
    714 
    715 Several of the important data structures in LLVM are graphs: for example CFGs
    716 made out of LLVM :ref:`BasicBlocks <BasicBlock>`, CFGs made out of LLVM
    717 :ref:`MachineBasicBlocks <MachineBasicBlock>`, and :ref:`Instruction Selection
    718 DAGs <SelectionDAG>`.  In many cases, while debugging various parts of the
    719 compiler, it is nice to instantly visualize these graphs.
    720 
    721 LLVM provides several callbacks that are available in a debug build to do
    722 exactly that.  If you call the ``Function::viewCFG()`` method, for example, the
    723 current LLVM tool will pop up a window containing the CFG for the function where
    724 each basic block is a node in the graph, and each node contains the instructions
    725 in the block.  Similarly, there also exists ``Function::viewCFGOnly()`` (does
    726 not include the instructions), the ``MachineFunction::viewCFG()`` and
    727 ``MachineFunction::viewCFGOnly()``, and the ``SelectionDAG::viewGraph()``
    728 methods.  Within GDB, for example, you can usually use something like ``call
    729 DAG.viewGraph()`` to pop up a window.  Alternatively, you can sprinkle calls to
    730 these functions in your code in places you want to debug.
    731 
    732 Getting this to work requires a small amount of setup.  On Unix systems
    733 with X11, install the `graphviz <http://www.graphviz.org>`_ toolkit, and make
    734 sure 'dot' and 'gv' are in your path.  If you are running on Mac OS X, download
    735 and install the Mac OS X `Graphviz program
    736 <http://www.pixelglow.com/graphviz/>`_ and add
    737 ``/Applications/Graphviz.app/Contents/MacOS/`` (or wherever you install it) to
    738 your path. The programs need not be present when configuring, building or
    739 running LLVM and can simply be installed when needed during an active debug
    740 session.
    741 
    742 ``SelectionDAG`` has been extended to make it easier to locate *interesting*
    743 nodes in large complex graphs.  From gdb, if you ``call DAG.setGraphColor(node,
    744 "color")``, then the next ``call DAG.viewGraph()`` would highlight the node in
    745 the specified color (choices of colors can be found at `colors
    746 <http://www.graphviz.org/doc/info/colors.html>`_.) More complex node attributes
    747 can be provided with ``call DAG.setGraphAttrs(node, "attributes")`` (choices can
    748 be found at `Graph attributes <http://www.graphviz.org/doc/info/attrs.html>`_.)
    749 If you want to restart and clear all the current graph attributes, then you can
    750 ``call DAG.clearGraphAttrs()``.
    751 
    752 Note that graph visualization features are compiled out of Release builds to
    753 reduce file size.  This means that you need a Debug+Asserts or Release+Asserts
    754 build to use these features.
    755 
    756 .. _datastructure:
    757 
    758 Picking the Right Data Structure for a Task
    759 ===========================================
    760 
    761 LLVM has a plethora of data structures in the ``llvm/ADT/`` directory, and we
    762 commonly use STL data structures.  This section describes the trade-offs you
    763 should consider when you pick one.
    764 
    765 The first step is a choose your own adventure: do you want a sequential
    766 container, a set-like container, or a map-like container?  The most important
    767 thing when choosing a container is the algorithmic properties of how you plan to
    768 access the container.  Based on that, you should use:
    769 
    770 
    771 * a :ref:`map-like <ds_map>` container if you need efficient look-up of a
    772   value based on another value.  Map-like containers also support efficient
    773   queries for containment (whether a key is in the map).  Map-like containers
    774   generally do not support efficient reverse mapping (values to keys).  If you
    775   need that, use two maps.  Some map-like containers also support efficient
    776   iteration through the keys in sorted order.  Map-like containers are the most
    777   expensive sort, only use them if you need one of these capabilities.
    778 
    779 * a :ref:`set-like <ds_set>` container if you need to put a bunch of stuff into
    780   a container that automatically eliminates duplicates.  Some set-like
    781   containers support efficient iteration through the elements in sorted order.
    782   Set-like containers are more expensive than sequential containers.
    783 
    784 * a :ref:`sequential <ds_sequential>` container provides the most efficient way
    785   to add elements and keeps track of the order they are added to the collection.
    786   They permit duplicates and support efficient iteration, but do not support
    787   efficient look-up based on a key.
    788 
    789 * a :ref:`string <ds_string>` container is a specialized sequential container or
    790   reference structure that is used for character or byte arrays.
    791 
    792 * a :ref:`bit <ds_bit>` container provides an efficient way to store and
    793   perform set operations on sets of numeric id's, while automatically
    794   eliminating duplicates.  Bit containers require a maximum of 1 bit for each
    795   identifier you want to store.
    796 
    797 Once the proper category of container is determined, you can fine tune the
    798 memory use, constant factors, and cache behaviors of access by intelligently
    799 picking a member of the category.  Note that constant factors and cache behavior
    800 can be a big deal.  If you have a vector that usually only contains a few
    801 elements (but could contain many), for example, it's much better to use
    802 :ref:`SmallVector <dss_smallvector>` than :ref:`vector <dss_vector>`.  Doing so
    803 avoids (relatively) expensive malloc/free calls, which dwarf the cost of adding
    804 the elements to the container.
    805 
    806 .. _ds_sequential:
    807 
    808 Sequential Containers (std::vector, std::list, etc)
    809 ---------------------------------------------------
    810 
    811 There are a variety of sequential containers available for you, based on your
    812 needs.  Pick the first in this section that will do what you want.
    813 
    814 .. _dss_arrayref:
    815 
    816 llvm/ADT/ArrayRef.h
    817 ^^^^^^^^^^^^^^^^^^^
    818 
    819 The ``llvm::ArrayRef`` class is the preferred class to use in an interface that
    820 accepts a sequential list of elements in memory and just reads from them.  By
    821 taking an ``ArrayRef``, the API can be passed a fixed size array, an
    822 ``std::vector``, an ``llvm::SmallVector`` and anything else that is contiguous
    823 in memory.
    824 
    825 .. _dss_fixedarrays:
    826 
    827 Fixed Size Arrays
    828 ^^^^^^^^^^^^^^^^^
    829 
    830 Fixed size arrays are very simple and very fast.  They are good if you know
    831 exactly how many elements you have, or you have a (low) upper bound on how many
    832 you have.
    833 
    834 .. _dss_heaparrays:
    835 
    836 Heap Allocated Arrays
    837 ^^^^^^^^^^^^^^^^^^^^^
    838 
    839 Heap allocated arrays (``new[]`` + ``delete[]``) are also simple.  They are good
    840 if the number of elements is variable, if you know how many elements you will
    841 need before the array is allocated, and if the array is usually large (if not,
    842 consider a :ref:`SmallVector <dss_smallvector>`).  The cost of a heap allocated
    843 array is the cost of the new/delete (aka malloc/free).  Also note that if you
    844 are allocating an array of a type with a constructor, the constructor and
    845 destructors will be run for every element in the array (re-sizable vectors only
    846 construct those elements actually used).
    847 
    848 .. _dss_tinyptrvector:
    849 
    850 llvm/ADT/TinyPtrVector.h
    851 ^^^^^^^^^^^^^^^^^^^^^^^^
    852 
    853 ``TinyPtrVector<Type>`` is a highly specialized collection class that is
    854 optimized to avoid allocation in the case when a vector has zero or one
    855 elements.  It has two major restrictions: 1) it can only hold values of pointer
    856 type, and 2) it cannot hold a null pointer.
    857 
    858 Since this container is highly specialized, it is rarely used.
    859 
    860 .. _dss_smallvector:
    861 
    862 llvm/ADT/SmallVector.h
    863 ^^^^^^^^^^^^^^^^^^^^^^
    864 
    865 ``SmallVector<Type, N>`` is a simple class that looks and smells just like
    866 ``vector<Type>``: it supports efficient iteration, lays out elements in memory
    867 order (so you can do pointer arithmetic between elements), supports efficient
    868 push_back/pop_back operations, supports efficient random access to its elements,
    869 etc.
    870 
    871 The advantage of SmallVector is that it allocates space for some number of
    872 elements (N) **in the object itself**.  Because of this, if the SmallVector is
    873 dynamically smaller than N, no malloc is performed.  This can be a big win in
    874 cases where the malloc/free call is far more expensive than the code that
    875 fiddles around with the elements.
    876 
    877 This is good for vectors that are "usually small" (e.g. the number of
    878 predecessors/successors of a block is usually less than 8).  On the other hand,
    879 this makes the size of the SmallVector itself large, so you don't want to
    880 allocate lots of them (doing so will waste a lot of space).  As such,
    881 SmallVectors are most useful when on the stack.
    882 
    883 SmallVector also provides a nice portable and efficient replacement for
    884 ``alloca``.
    885 
    886 .. note::
    887 
    888    Prefer to use ``SmallVectorImpl<T>`` as a parameter type.
    889 
    890    In APIs that don't care about the "small size" (most?), prefer to use
    891    the ``SmallVectorImpl<T>`` class, which is basically just the "vector
    892    header" (and methods) without the elements allocated after it. Note that
    893    ``SmallVector<T, N>`` inherits from ``SmallVectorImpl<T>`` so the
    894    conversion is implicit and costs nothing. E.g.
    895 
    896    .. code-block:: c++
    897 
    898       // BAD: Clients cannot pass e.g. SmallVector<Foo, 4>.
    899       hardcodedSmallSize(SmallVector<Foo, 2> &Out);
    900       // GOOD: Clients can pass any SmallVector<Foo, N>.
    901       allowsAnySmallSize(SmallVectorImpl<Foo> &Out);
    902 
    903       void someFunc() {
    904         SmallVector<Foo, 8> Vec;
    905         hardcodedSmallSize(Vec); // Error.
    906         allowsAnySmallSize(Vec); // Works.
    907       }
    908 
    909    Even though it has "``Impl``" in the name, this is so widely used that
    910    it really isn't "private to the implementation" anymore. A name like
    911    ``SmallVectorHeader`` would be more appropriate.
    912 
    913 .. _dss_vector:
    914 
    915 <vector>
    916 ^^^^^^^^
    917 
    918 ``std::vector`` is well loved and respected.  It is useful when SmallVector
    919 isn't: when the size of the vector is often large (thus the small optimization
    920 will rarely be a benefit) or if you will be allocating many instances of the
    921 vector itself (which would waste space for elements that aren't in the
    922 container).  vector is also useful when interfacing with code that expects
    923 vectors :).
    924 
    925 One worthwhile note about std::vector: avoid code like this:
    926 
    927 .. code-block:: c++
    928 
    929   for ( ... ) {
    930      std::vector<foo> V;
    931      // make use of V.
    932   }
    933 
    934 Instead, write this as:
    935 
    936 .. code-block:: c++
    937 
    938   std::vector<foo> V;
    939   for ( ... ) {
    940      // make use of V.
    941      V.clear();
    942   }
    943 
    944 Doing so will save (at least) one heap allocation and free per iteration of the
    945 loop.
    946 
    947 .. _dss_deque:
    948 
    949 <deque>
    950 ^^^^^^^
    951 
    952 ``std::deque`` is, in some senses, a generalized version of ``std::vector``.
    953 Like ``std::vector``, it provides constant time random access and other similar
    954 properties, but it also provides efficient access to the front of the list.  It
    955 does not guarantee continuity of elements within memory.
    956 
    957 In exchange for this extra flexibility, ``std::deque`` has significantly higher
    958 constant factor costs than ``std::vector``.  If possible, use ``std::vector`` or
    959 something cheaper.
    960 
    961 .. _dss_list:
    962 
    963 <list>
    964 ^^^^^^
    965 
    966 ``std::list`` is an extremely inefficient class that is rarely useful.  It
    967 performs a heap allocation for every element inserted into it, thus having an
    968 extremely high constant factor, particularly for small data types.
    969 ``std::list`` also only supports bidirectional iteration, not random access
    970 iteration.
    971 
    972 In exchange for this high cost, std::list supports efficient access to both ends
    973 of the list (like ``std::deque``, but unlike ``std::vector`` or
    974 ``SmallVector``).  In addition, the iterator invalidation characteristics of
    975 std::list are stronger than that of a vector class: inserting or removing an
    976 element into the list does not invalidate iterator or pointers to other elements
    977 in the list.
    978 
    979 .. _dss_ilist:
    980 
    981 llvm/ADT/ilist.h
    982 ^^^^^^^^^^^^^^^^
    983 
    984 ``ilist<T>`` implements an 'intrusive' doubly-linked list.  It is intrusive,
    985 because it requires the element to store and provide access to the prev/next
    986 pointers for the list.
    987 
    988 ``ilist`` has the same drawbacks as ``std::list``, and additionally requires an
    989 ``ilist_traits`` implementation for the element type, but it provides some novel
    990 characteristics.  In particular, it can efficiently store polymorphic objects,
    991 the traits class is informed when an element is inserted or removed from the
    992 list, and ``ilist``\ s are guaranteed to support a constant-time splice
    993 operation.
    994 
    995 These properties are exactly what we want for things like ``Instruction``\ s and
    996 basic blocks, which is why these are implemented with ``ilist``\ s.
    997 
    998 Related classes of interest are explained in the following subsections:
    999 
   1000 * :ref:`ilist_traits <dss_ilist_traits>`
   1001 
   1002 * :ref:`iplist <dss_iplist>`
   1003 
   1004 * :ref:`llvm/ADT/ilist_node.h <dss_ilist_node>`
   1005 
   1006 * :ref:`Sentinels <dss_ilist_sentinel>`
   1007 
   1008 .. _dss_packedvector:
   1009 
   1010 llvm/ADT/PackedVector.h
   1011 ^^^^^^^^^^^^^^^^^^^^^^^
   1012 
   1013 Useful for storing a vector of values using only a few number of bits for each
   1014 value.  Apart from the standard operations of a vector-like container, it can
   1015 also perform an 'or' set operation.
   1016 
   1017 For example:
   1018 
   1019 .. code-block:: c++
   1020 
   1021   enum State {
   1022       None = 0x0,
   1023       FirstCondition = 0x1,
   1024       SecondCondition = 0x2,
   1025       Both = 0x3
   1026   };
   1027 
   1028   State get() {
   1029       PackedVector<State, 2> Vec1;
   1030       Vec1.push_back(FirstCondition);
   1031 
   1032       PackedVector<State, 2> Vec2;
   1033       Vec2.push_back(SecondCondition);
   1034 
   1035       Vec1 |= Vec2;
   1036       return Vec1[0]; // returns 'Both'.
   1037   }
   1038 
   1039 .. _dss_ilist_traits:
   1040 
   1041 ilist_traits
   1042 ^^^^^^^^^^^^
   1043 
   1044 ``ilist_traits<T>`` is ``ilist<T>``'s customization mechanism. ``iplist<T>``
   1045 (and consequently ``ilist<T>``) publicly derive from this traits class.
   1046 
   1047 .. _dss_iplist:
   1048 
   1049 iplist
   1050 ^^^^^^
   1051 
   1052 ``iplist<T>`` is ``ilist<T>``'s base and as such supports a slightly narrower
   1053 interface.  Notably, inserters from ``T&`` are absent.
   1054 
   1055 ``ilist_traits<T>`` is a public base of this class and can be used for a wide
   1056 variety of customizations.
   1057 
   1058 .. _dss_ilist_node:
   1059 
   1060 llvm/ADT/ilist_node.h
   1061 ^^^^^^^^^^^^^^^^^^^^^
   1062 
   1063 ``ilist_node<T>`` implements the forward and backward links that are expected
   1064 by the ``ilist<T>`` (and analogous containers) in the default manner.
   1065 
   1066 ``ilist_node<T>``\ s are meant to be embedded in the node type ``T``, usually
   1067 ``T`` publicly derives from ``ilist_node<T>``.
   1068 
   1069 .. _dss_ilist_sentinel:
   1070 
   1071 Sentinels
   1072 ^^^^^^^^^
   1073 
   1074 ``ilist``\ s have another specialty that must be considered.  To be a good
   1075 citizen in the C++ ecosystem, it needs to support the standard container
   1076 operations, such as ``begin`` and ``end`` iterators, etc.  Also, the
   1077 ``operator--`` must work correctly on the ``end`` iterator in the case of
   1078 non-empty ``ilist``\ s.
   1079 
   1080 The only sensible solution to this problem is to allocate a so-called *sentinel*
   1081 along with the intrusive list, which serves as the ``end`` iterator, providing
   1082 the back-link to the last element.  However conforming to the C++ convention it
   1083 is illegal to ``operator++`` beyond the sentinel and it also must not be
   1084 dereferenced.
   1085 
   1086 These constraints allow for some implementation freedom to the ``ilist`` how to
   1087 allocate and store the sentinel.  The corresponding policy is dictated by
   1088 ``ilist_traits<T>``.  By default a ``T`` gets heap-allocated whenever the need
   1089 for a sentinel arises.
   1090 
   1091 While the default policy is sufficient in most cases, it may break down when
   1092 ``T`` does not provide a default constructor.  Also, in the case of many
   1093 instances of ``ilist``\ s, the memory overhead of the associated sentinels is
   1094 wasted.  To alleviate the situation with numerous and voluminous
   1095 ``T``-sentinels, sometimes a trick is employed, leading to *ghostly sentinels*.
   1096 
   1097 Ghostly sentinels are obtained by specially-crafted ``ilist_traits<T>`` which
   1098 superpose the sentinel with the ``ilist`` instance in memory.  Pointer
   1099 arithmetic is used to obtain the sentinel, which is relative to the ``ilist``'s
   1100 ``this`` pointer.  The ``ilist`` is augmented by an extra pointer, which serves
   1101 as the back-link of the sentinel.  This is the only field in the ghostly
   1102 sentinel which can be legally accessed.
   1103 
   1104 .. _dss_other:
   1105 
   1106 Other Sequential Container options
   1107 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   1108 
   1109 Other STL containers are available, such as ``std::string``.
   1110 
   1111 There are also various STL adapter classes such as ``std::queue``,
   1112 ``std::priority_queue``, ``std::stack``, etc.  These provide simplified access
   1113 to an underlying container but don't affect the cost of the container itself.
   1114 
   1115 .. _ds_string:
   1116 
   1117 String-like containers
   1118 ----------------------
   1119 
   1120 There are a variety of ways to pass around and use strings in C and C++, and
   1121 LLVM adds a few new options to choose from.  Pick the first option on this list
   1122 that will do what you need, they are ordered according to their relative cost.
   1123 
   1124 Note that it is generally preferred to *not* pass strings around as ``const
   1125 char*``'s.  These have a number of problems, including the fact that they
   1126 cannot represent embedded nul ("\0") characters, and do not have a length
   1127 available efficiently.  The general replacement for '``const char*``' is
   1128 StringRef.
   1129 
   1130 For more information on choosing string containers for APIs, please see
   1131 :ref:`Passing Strings <string_apis>`.
   1132 
   1133 .. _dss_stringref:
   1134 
   1135 llvm/ADT/StringRef.h
   1136 ^^^^^^^^^^^^^^^^^^^^
   1137 
   1138 The StringRef class is a simple value class that contains a pointer to a
   1139 character and a length, and is quite related to the :ref:`ArrayRef
   1140 <dss_arrayref>` class (but specialized for arrays of characters).  Because
   1141 StringRef carries a length with it, it safely handles strings with embedded nul
   1142 characters in it, getting the length does not require a strlen call, and it even
   1143 has very convenient APIs for slicing and dicing the character range that it
   1144 represents.
   1145 
   1146 StringRef is ideal for passing simple strings around that are known to be live,
   1147 either because they are C string literals, std::string, a C array, or a
   1148 SmallVector.  Each of these cases has an efficient implicit conversion to
   1149 StringRef, which doesn't result in a dynamic strlen being executed.
   1150 
   1151 StringRef has a few major limitations which make more powerful string containers
   1152 useful:
   1153 
   1154 #. You cannot directly convert a StringRef to a 'const char*' because there is
   1155    no way to add a trailing nul (unlike the .c_str() method on various stronger
   1156    classes).
   1157 
   1158 #. StringRef doesn't own or keep alive the underlying string bytes.
   1159    As such it can easily lead to dangling pointers, and is not suitable for
   1160    embedding in datastructures in most cases (instead, use an std::string or
   1161    something like that).
   1162 
   1163 #. For the same reason, StringRef cannot be used as the return value of a
   1164    method if the method "computes" the result string.  Instead, use std::string.
   1165 
   1166 #. StringRef's do not allow you to mutate the pointed-to string bytes and it
   1167    doesn't allow you to insert or remove bytes from the range.  For editing
   1168    operations like this, it interoperates with the :ref:`Twine <dss_twine>`
   1169    class.
   1170 
   1171 Because of its strengths and limitations, it is very common for a function to
   1172 take a StringRef and for a method on an object to return a StringRef that points
   1173 into some string that it owns.
   1174 
   1175 .. _dss_twine:
   1176 
   1177 llvm/ADT/Twine.h
   1178 ^^^^^^^^^^^^^^^^
   1179 
   1180 The Twine class is used as an intermediary datatype for APIs that want to take a
   1181 string that can be constructed inline with a series of concatenations.  Twine
   1182 works by forming recursive instances of the Twine datatype (a simple value
   1183 object) on the stack as temporary objects, linking them together into a tree
   1184 which is then linearized when the Twine is consumed.  Twine is only safe to use
   1185 as the argument to a function, and should always be a const reference, e.g.:
   1186 
   1187 .. code-block:: c++
   1188 
   1189   void foo(const Twine &T);
   1190   ...
   1191   StringRef X = ...
   1192   unsigned i = ...
   1193   foo(X + "." + Twine(i));
   1194 
   1195 This example forms a string like "blarg.42" by concatenating the values
   1196 together, and does not form intermediate strings containing "blarg" or "blarg.".
   1197 
   1198 Because Twine is constructed with temporary objects on the stack, and because
   1199 these instances are destroyed at the end of the current statement, it is an
   1200 inherently dangerous API.  For example, this simple variant contains undefined
   1201 behavior and will probably crash:
   1202 
   1203 .. code-block:: c++
   1204 
   1205   void foo(const Twine &T);
   1206   ...
   1207   StringRef X = ...
   1208   unsigned i = ...
   1209   const Twine &Tmp = X + "." + Twine(i);
   1210   foo(Tmp);
   1211 
   1212 ... because the temporaries are destroyed before the call.  That said, Twine's
   1213 are much more efficient than intermediate std::string temporaries, and they work
   1214 really well with StringRef.  Just be aware of their limitations.
   1215 
   1216 .. _dss_smallstring:
   1217 
   1218 llvm/ADT/SmallString.h
   1219 ^^^^^^^^^^^^^^^^^^^^^^
   1220 
   1221 SmallString is a subclass of :ref:`SmallVector <dss_smallvector>` that adds some
   1222 convenience APIs like += that takes StringRef's.  SmallString avoids allocating
   1223 memory in the case when the preallocated space is enough to hold its data, and
   1224 it calls back to general heap allocation when required.  Since it owns its data,
   1225 it is very safe to use and supports full mutation of the string.
   1226 
   1227 Like SmallVector's, the big downside to SmallString is their sizeof.  While they
   1228 are optimized for small strings, they themselves are not particularly small.
   1229 This means that they work great for temporary scratch buffers on the stack, but
   1230 should not generally be put into the heap: it is very rare to see a SmallString
   1231 as the member of a frequently-allocated heap data structure or returned
   1232 by-value.
   1233 
   1234 .. _dss_stdstring:
   1235 
   1236 std::string
   1237 ^^^^^^^^^^^
   1238 
   1239 The standard C++ std::string class is a very general class that (like
   1240 SmallString) owns its underlying data.  sizeof(std::string) is very reasonable
   1241 so it can be embedded into heap data structures and returned by-value.  On the
   1242 other hand, std::string is highly inefficient for inline editing (e.g.
   1243 concatenating a bunch of stuff together) and because it is provided by the
   1244 standard library, its performance characteristics depend a lot of the host
   1245 standard library (e.g. libc++ and MSVC provide a highly optimized string class,
   1246 GCC contains a really slow implementation).
   1247 
   1248 The major disadvantage of std::string is that almost every operation that makes
   1249 them larger can allocate memory, which is slow.  As such, it is better to use
   1250 SmallVector or Twine as a scratch buffer, but then use std::string to persist
   1251 the result.
   1252 
   1253 .. _ds_set:
   1254 
   1255 Set-Like Containers (std::set, SmallSet, SetVector, etc)
   1256 --------------------------------------------------------
   1257 
   1258 Set-like containers are useful when you need to canonicalize multiple values
   1259 into a single representation.  There are several different choices for how to do
   1260 this, providing various trade-offs.
   1261 
   1262 .. _dss_sortedvectorset:
   1263 
   1264 A sorted 'vector'
   1265 ^^^^^^^^^^^^^^^^^
   1266 
   1267 If you intend to insert a lot of elements, then do a lot of queries, a great
   1268 approach is to use a vector (or other sequential container) with
   1269 std::sort+std::unique to remove duplicates.  This approach works really well if
   1270 your usage pattern has these two distinct phases (insert then query), and can be
   1271 coupled with a good choice of :ref:`sequential container <ds_sequential>`.
   1272 
   1273 This combination provides the several nice properties: the result data is
   1274 contiguous in memory (good for cache locality), has few allocations, is easy to
   1275 address (iterators in the final vector are just indices or pointers), and can be
   1276 efficiently queried with a standard binary search (e.g.
   1277 ``std::lower_bound``; if you want the whole range of elements comparing
   1278 equal, use ``std::equal_range``).
   1279 
   1280 .. _dss_smallset:
   1281 
   1282 llvm/ADT/SmallSet.h
   1283 ^^^^^^^^^^^^^^^^^^^
   1284 
   1285 If you have a set-like data structure that is usually small and whose elements
   1286 are reasonably small, a ``SmallSet<Type, N>`` is a good choice.  This set has
   1287 space for N elements in place (thus, if the set is dynamically smaller than N,
   1288 no malloc traffic is required) and accesses them with a simple linear search.
   1289 When the set grows beyond N elements, it allocates a more expensive
   1290 representation that guarantees efficient access (for most types, it falls back
   1291 to :ref:`std::set <dss_set>`, but for pointers it uses something far better,
   1292 :ref:`SmallPtrSet <dss_smallptrset>`.
   1293 
   1294 The magic of this class is that it handles small sets extremely efficiently, but
   1295 gracefully handles extremely large sets without loss of efficiency.  The
   1296 drawback is that the interface is quite small: it supports insertion, queries
   1297 and erasing, but does not support iteration.
   1298 
   1299 .. _dss_smallptrset:
   1300 
   1301 llvm/ADT/SmallPtrSet.h
   1302 ^^^^^^^^^^^^^^^^^^^^^^
   1303 
   1304 ``SmallPtrSet`` has all the advantages of ``SmallSet`` (and a ``SmallSet`` of
   1305 pointers is transparently implemented with a ``SmallPtrSet``), but also supports
   1306 iterators.  If more than N insertions are performed, a single quadratically
   1307 probed hash table is allocated and grows as needed, providing extremely
   1308 efficient access (constant time insertion/deleting/queries with low constant
   1309 factors) and is very stingy with malloc traffic.
   1310 
   1311 Note that, unlike :ref:`std::set <dss_set>`, the iterators of ``SmallPtrSet``
   1312 are invalidated whenever an insertion occurs.  Also, the values visited by the
   1313 iterators are not visited in sorted order.
   1314 
   1315 .. _dss_stringset:
   1316 
   1317 llvm/ADT/StringSet.h
   1318 ^^^^^^^^^^^^^^^^^^^^
   1319 
   1320 ``StringSet`` is a thin wrapper around :ref:`StringMap\<char\> <dss_stringmap>`,
   1321 and it allows efficient storage and retrieval of unique strings.
   1322 
   1323 Functionally analogous to ``SmallSet<StringRef>``, ``StringSet`` also supports
   1324 iteration. (The iterator dereferences to a ``StringMapEntry<char>``, so you
   1325 need to call ``i->getKey()`` to access the item of the StringSet.)  On the
   1326 other hand, ``StringSet`` doesn't support range-insertion and
   1327 copy-construction, which :ref:`SmallSet <dss_smallset>` and :ref:`SmallPtrSet
   1328 <dss_smallptrset>` do support.
   1329 
   1330 .. _dss_denseset:
   1331 
   1332 llvm/ADT/DenseSet.h
   1333 ^^^^^^^^^^^^^^^^^^^
   1334 
   1335 DenseSet is a simple quadratically probed hash table.  It excels at supporting
   1336 small values: it uses a single allocation to hold all of the pairs that are
   1337 currently inserted in the set.  DenseSet is a great way to unique small values
   1338 that are not simple pointers (use :ref:`SmallPtrSet <dss_smallptrset>` for
   1339 pointers).  Note that DenseSet has the same requirements for the value type that
   1340 :ref:`DenseMap <dss_densemap>` has.
   1341 
   1342 .. _dss_sparseset:
   1343 
   1344 llvm/ADT/SparseSet.h
   1345 ^^^^^^^^^^^^^^^^^^^^
   1346 
   1347 SparseSet holds a small number of objects identified by unsigned keys of
   1348 moderate size.  It uses a lot of memory, but provides operations that are almost
   1349 as fast as a vector.  Typical keys are physical registers, virtual registers, or
   1350 numbered basic blocks.
   1351 
   1352 SparseSet is useful for algorithms that need very fast clear/find/insert/erase
   1353 and fast iteration over small sets.  It is not intended for building composite
   1354 data structures.
   1355 
   1356 .. _dss_sparsemultiset:
   1357 
   1358 llvm/ADT/SparseMultiSet.h
   1359 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   1360 
   1361 SparseMultiSet adds multiset behavior to SparseSet, while retaining SparseSet's
   1362 desirable attributes. Like SparseSet, it typically uses a lot of memory, but
   1363 provides operations that are almost as fast as a vector.  Typical keys are
   1364 physical registers, virtual registers, or numbered basic blocks.
   1365 
   1366 SparseMultiSet is useful for algorithms that need very fast
   1367 clear/find/insert/erase of the entire collection, and iteration over sets of
   1368 elements sharing a key. It is often a more efficient choice than using composite
   1369 data structures (e.g. vector-of-vectors, map-of-vectors). It is not intended for
   1370 building composite data structures.
   1371 
   1372 .. _dss_FoldingSet:
   1373 
   1374 llvm/ADT/FoldingSet.h
   1375 ^^^^^^^^^^^^^^^^^^^^^
   1376 
   1377 FoldingSet is an aggregate class that is really good at uniquing
   1378 expensive-to-create or polymorphic objects.  It is a combination of a chained
   1379 hash table with intrusive links (uniqued objects are required to inherit from
   1380 FoldingSetNode) that uses :ref:`SmallVector <dss_smallvector>` as part of its ID
   1381 process.
   1382 
   1383 Consider a case where you want to implement a "getOrCreateFoo" method for a
   1384 complex object (for example, a node in the code generator).  The client has a
   1385 description of **what** it wants to generate (it knows the opcode and all the
   1386 operands), but we don't want to 'new' a node, then try inserting it into a set
   1387 only to find out it already exists, at which point we would have to delete it
   1388 and return the node that already exists.
   1389 
   1390 To support this style of client, FoldingSet perform a query with a
   1391 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
   1392 element that we want to query for.  The query either returns the element
   1393 matching the ID or it returns an opaque ID that indicates where insertion should
   1394 take place.  Construction of the ID usually does not require heap traffic.
   1395 
   1396 Because FoldingSet uses intrusive links, it can support polymorphic objects in
   1397 the set (for example, you can have SDNode instances mixed with LoadSDNodes).
   1398 Because the elements are individually allocated, pointers to the elements are
   1399 stable: inserting or removing elements does not invalidate any pointers to other
   1400 elements.
   1401 
   1402 .. _dss_set:
   1403 
   1404 <set>
   1405 ^^^^^
   1406 
   1407 ``std::set`` is a reasonable all-around set class, which is decent at many
   1408 things but great at nothing.  std::set allocates memory for each element
   1409 inserted (thus it is very malloc intensive) and typically stores three pointers
   1410 per element in the set (thus adding a large amount of per-element space
   1411 overhead).  It offers guaranteed log(n) performance, which is not particularly
   1412 fast from a complexity standpoint (particularly if the elements of the set are
   1413 expensive to compare, like strings), and has extremely high constant factors for
   1414 lookup, insertion and removal.
   1415 
   1416 The advantages of std::set are that its iterators are stable (deleting or
   1417 inserting an element from the set does not affect iterators or pointers to other
   1418 elements) and that iteration over the set is guaranteed to be in sorted order.
   1419 If the elements in the set are large, then the relative overhead of the pointers
   1420 and malloc traffic is not a big deal, but if the elements of the set are small,
   1421 std::set is almost never a good choice.
   1422 
   1423 .. _dss_setvector:
   1424 
   1425 llvm/ADT/SetVector.h
   1426 ^^^^^^^^^^^^^^^^^^^^
   1427 
   1428 LLVM's ``SetVector<Type>`` is an adapter class that combines your choice of a
   1429 set-like container along with a :ref:`Sequential Container <ds_sequential>` The
   1430 important property that this provides is efficient insertion with uniquing
   1431 (duplicate elements are ignored) with iteration support.  It implements this by
   1432 inserting elements into both a set-like container and the sequential container,
   1433 using the set-like container for uniquing and the sequential container for
   1434 iteration.
   1435 
   1436 The difference between SetVector and other sets is that the order of iteration
   1437 is guaranteed to match the order of insertion into the SetVector.  This property
   1438 is really important for things like sets of pointers.  Because pointer values
   1439 are non-deterministic (e.g. vary across runs of the program on different
   1440 machines), iterating over the pointers in the set will not be in a well-defined
   1441 order.
   1442 
   1443 The drawback of SetVector is that it requires twice as much space as a normal
   1444 set and has the sum of constant factors from the set-like container and the
   1445 sequential container that it uses.  Use it **only** if you need to iterate over
   1446 the elements in a deterministic order.  SetVector is also expensive to delete
   1447 elements out of (linear time), unless you use its "pop_back" method, which is
   1448 faster.
   1449 
   1450 ``SetVector`` is an adapter class that defaults to using ``std::vector`` and a
   1451 size 16 ``SmallSet`` for the underlying containers, so it is quite expensive.
   1452 However, ``"llvm/ADT/SetVector.h"`` also provides a ``SmallSetVector`` class,
   1453 which defaults to using a ``SmallVector`` and ``SmallSet`` of a specified size.
   1454 If you use this, and if your sets are dynamically smaller than ``N``, you will
   1455 save a lot of heap traffic.
   1456 
   1457 .. _dss_uniquevector:
   1458 
   1459 llvm/ADT/UniqueVector.h
   1460 ^^^^^^^^^^^^^^^^^^^^^^^
   1461 
   1462 UniqueVector is similar to :ref:`SetVector <dss_setvector>` but it retains a
   1463 unique ID for each element inserted into the set.  It internally contains a map
   1464 and a vector, and it assigns a unique ID for each value inserted into the set.
   1465 
   1466 UniqueVector is very expensive: its cost is the sum of the cost of maintaining
   1467 both the map and vector, it has high complexity, high constant factors, and
   1468 produces a lot of malloc traffic.  It should be avoided.
   1469 
   1470 .. _dss_immutableset:
   1471 
   1472 llvm/ADT/ImmutableSet.h
   1473 ^^^^^^^^^^^^^^^^^^^^^^^
   1474 
   1475 ImmutableSet is an immutable (functional) set implementation based on an AVL
   1476 tree.  Adding or removing elements is done through a Factory object and results
   1477 in the creation of a new ImmutableSet object.  If an ImmutableSet already exists
   1478 with the given contents, then the existing one is returned; equality is compared
   1479 with a FoldingSetNodeID.  The time and space complexity of add or remove
   1480 operations is logarithmic in the size of the original set.
   1481 
   1482 There is no method for returning an element of the set, you can only check for
   1483 membership.
   1484 
   1485 .. _dss_otherset:
   1486 
   1487 Other Set-Like Container Options
   1488 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   1489 
   1490 The STL provides several other options, such as std::multiset and the various
   1491 "hash_set" like containers (whether from C++ TR1 or from the SGI library).  We
   1492 never use hash_set and unordered_set because they are generally very expensive
   1493 (each insertion requires a malloc) and very non-portable.
   1494 
   1495 std::multiset is useful if you're not interested in elimination of duplicates,
   1496 but has all the drawbacks of :ref:`std::set <dss_set>`.  A sorted vector
   1497 (where you don't delete duplicate entries) or some other approach is almost
   1498 always better.
   1499 
   1500 .. _ds_map:
   1501 
   1502 Map-Like Containers (std::map, DenseMap, etc)
   1503 ---------------------------------------------
   1504 
   1505 Map-like containers are useful when you want to associate data to a key.  As
   1506 usual, there are a lot of different ways to do this. :)
   1507 
   1508 .. _dss_sortedvectormap:
   1509 
   1510 A sorted 'vector'
   1511 ^^^^^^^^^^^^^^^^^
   1512 
   1513 If your usage pattern follows a strict insert-then-query approach, you can
   1514 trivially use the same approach as :ref:`sorted vectors for set-like containers
   1515 <dss_sortedvectorset>`.  The only difference is that your query function (which
   1516 uses std::lower_bound to get efficient log(n) lookup) should only compare the
   1517 key, not both the key and value.  This yields the same advantages as sorted
   1518 vectors for sets.
   1519 
   1520 .. _dss_stringmap:
   1521 
   1522 llvm/ADT/StringMap.h
   1523 ^^^^^^^^^^^^^^^^^^^^
   1524 
   1525 Strings are commonly used as keys in maps, and they are difficult to support
   1526 efficiently: they are variable length, inefficient to hash and compare when
   1527 long, expensive to copy, etc.  StringMap is a specialized container designed to
   1528 cope with these issues.  It supports mapping an arbitrary range of bytes to an
   1529 arbitrary other object.
   1530 
   1531 The StringMap implementation uses a quadratically-probed hash table, where the
   1532 buckets store a pointer to the heap allocated entries (and some other stuff).
   1533 The entries in the map must be heap allocated because the strings are variable
   1534 length.  The string data (key) and the element object (value) are stored in the
   1535 same allocation with the string data immediately after the element object.
   1536 This container guarantees the "``(char*)(&Value+1)``" points to the key string
   1537 for a value.
   1538 
   1539 The StringMap is very fast for several reasons: quadratic probing is very cache
   1540 efficient for lookups, the hash value of strings in buckets is not recomputed
   1541 when looking up an element, StringMap rarely has to touch the memory for
   1542 unrelated objects when looking up a value (even when hash collisions happen),
   1543 hash table growth does not recompute the hash values for strings already in the
   1544 table, and each pair in the map is store in a single allocation (the string data
   1545 is stored in the same allocation as the Value of a pair).
   1546 
   1547 StringMap also provides query methods that take byte ranges, so it only ever
   1548 copies a string if a value is inserted into the table.
   1549 
   1550 StringMap iteratation order, however, is not guaranteed to be deterministic, so
   1551 any uses which require that should instead use a std::map.
   1552 
   1553 .. _dss_indexmap:
   1554 
   1555 llvm/ADT/IndexedMap.h
   1556 ^^^^^^^^^^^^^^^^^^^^^
   1557 
   1558 IndexedMap is a specialized container for mapping small dense integers (or
   1559 values that can be mapped to small dense integers) to some other type.  It is
   1560 internally implemented as a vector with a mapping function that maps the keys
   1561 to the dense integer range.
   1562 
   1563 This is useful for cases like virtual registers in the LLVM code generator: they
   1564 have a dense mapping that is offset by a compile-time constant (the first
   1565 virtual register ID).
   1566 
   1567 .. _dss_densemap:
   1568 
   1569 llvm/ADT/DenseMap.h
   1570 ^^^^^^^^^^^^^^^^^^^
   1571 
   1572 DenseMap is a simple quadratically probed hash table.  It excels at supporting
   1573 small keys and values: it uses a single allocation to hold all of the pairs
   1574 that are currently inserted in the map.  DenseMap is a great way to map
   1575 pointers to pointers, or map other small types to each other.
   1576 
   1577 There are several aspects of DenseMap that you should be aware of, however.
   1578 The iterators in a DenseMap are invalidated whenever an insertion occurs,
   1579 unlike map.  Also, because DenseMap allocates space for a large number of
   1580 key/value pairs (it starts with 64 by default), it will waste a lot of space if
   1581 your keys or values are large.  Finally, you must implement a partial
   1582 specialization of DenseMapInfo for the key that you want, if it isn't already
   1583 supported.  This is required to tell DenseMap about two special marker values
   1584 (which can never be inserted into the map) that it needs internally.
   1585 
   1586 DenseMap's find_as() method supports lookup operations using an alternate key
   1587 type.  This is useful in cases where the normal key type is expensive to
   1588 construct, but cheap to compare against.  The DenseMapInfo is responsible for
   1589 defining the appropriate comparison and hashing methods for each alternate key
   1590 type used.
   1591 
   1592 .. _dss_valuemap:
   1593 
   1594 llvm/IR/ValueMap.h
   1595 ^^^^^^^^^^^^^^^^^^^
   1596 
   1597 ValueMap is a wrapper around a :ref:`DenseMap <dss_densemap>` mapping
   1598 ``Value*``\ s (or subclasses) to another type.  When a Value is deleted or
   1599 RAUW'ed, ValueMap will update itself so the new version of the key is mapped to
   1600 the same value, just as if the key were a WeakVH.  You can configure exactly how
   1601 this happens, and what else happens on these two events, by passing a ``Config``
   1602 parameter to the ValueMap template.
   1603 
   1604 .. _dss_intervalmap:
   1605 
   1606 llvm/ADT/IntervalMap.h
   1607 ^^^^^^^^^^^^^^^^^^^^^^
   1608 
   1609 IntervalMap is a compact map for small keys and values.  It maps key intervals
   1610 instead of single keys, and it will automatically coalesce adjacent intervals.
   1611 When the map only contains a few intervals, they are stored in the map object
   1612 itself to avoid allocations.
   1613 
   1614 The IntervalMap iterators are quite big, so they should not be passed around as
   1615 STL iterators.  The heavyweight iterators allow a smaller data structure.
   1616 
   1617 .. _dss_map:
   1618 
   1619 <map>
   1620 ^^^^^
   1621 
   1622 std::map has similar characteristics to :ref:`std::set <dss_set>`: it uses a
   1623 single allocation per pair inserted into the map, it offers log(n) lookup with
   1624 an extremely large constant factor, imposes a space penalty of 3 pointers per
   1625 pair in the map, etc.
   1626 
   1627 std::map is most useful when your keys or values are very large, if you need to
   1628 iterate over the collection in sorted order, or if you need stable iterators
   1629 into the map (i.e. they don't get invalidated if an insertion or deletion of
   1630 another element takes place).
   1631 
   1632 .. _dss_mapvector:
   1633 
   1634 llvm/ADT/MapVector.h
   1635 ^^^^^^^^^^^^^^^^^^^^
   1636 
   1637 ``MapVector<KeyT,ValueT>`` provides a subset of the DenseMap interface.  The
   1638 main difference is that the iteration order is guaranteed to be the insertion
   1639 order, making it an easy (but somewhat expensive) solution for non-deterministic
   1640 iteration over maps of pointers.
   1641 
   1642 It is implemented by mapping from key to an index in a vector of key,value
   1643 pairs.  This provides fast lookup and iteration, but has two main drawbacks:
   1644 the key is stored twice and removing elements takes linear time.  If it is
   1645 necessary to remove elements, it's best to remove them in bulk using
   1646 ``remove_if()``.
   1647 
   1648 .. _dss_inteqclasses:
   1649 
   1650 llvm/ADT/IntEqClasses.h
   1651 ^^^^^^^^^^^^^^^^^^^^^^^
   1652 
   1653 IntEqClasses provides a compact representation of equivalence classes of small
   1654 integers.  Initially, each integer in the range 0..n-1 has its own equivalence
   1655 class.  Classes can be joined by passing two class representatives to the
   1656 join(a, b) method.  Two integers are in the same class when findLeader() returns
   1657 the same representative.
   1658 
   1659 Once all equivalence classes are formed, the map can be compressed so each
   1660 integer 0..n-1 maps to an equivalence class number in the range 0..m-1, where m
   1661 is the total number of equivalence classes.  The map must be uncompressed before
   1662 it can be edited again.
   1663 
   1664 .. _dss_immutablemap:
   1665 
   1666 llvm/ADT/ImmutableMap.h
   1667 ^^^^^^^^^^^^^^^^^^^^^^^
   1668 
   1669 ImmutableMap is an immutable (functional) map implementation based on an AVL
   1670 tree.  Adding or removing elements is done through a Factory object and results
   1671 in the creation of a new ImmutableMap object.  If an ImmutableMap already exists
   1672 with the given key set, then the existing one is returned; equality is compared
   1673 with a FoldingSetNodeID.  The time and space complexity of add or remove
   1674 operations is logarithmic in the size of the original map.
   1675 
   1676 .. _dss_othermap:
   1677 
   1678 Other Map-Like Container Options
   1679 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   1680 
   1681 The STL provides several other options, such as std::multimap and the various
   1682 "hash_map" like containers (whether from C++ TR1 or from the SGI library).  We
   1683 never use hash_set and unordered_set because they are generally very expensive
   1684 (each insertion requires a malloc) and very non-portable.
   1685 
   1686 std::multimap is useful if you want to map a key to multiple values, but has all
   1687 the drawbacks of std::map.  A sorted vector or some other approach is almost
   1688 always better.
   1689 
   1690 .. _ds_bit:
   1691 
   1692 Bit storage containers (BitVector, SparseBitVector)
   1693 ---------------------------------------------------
   1694 
   1695 Unlike the other containers, there are only two bit storage containers, and
   1696 choosing when to use each is relatively straightforward.
   1697 
   1698 One additional option is ``std::vector<bool>``: we discourage its use for two
   1699 reasons 1) the implementation in many common compilers (e.g.  commonly
   1700 available versions of GCC) is extremely inefficient and 2) the C++ standards
   1701 committee is likely to deprecate this container and/or change it significantly
   1702 somehow.  In any case, please don't use it.
   1703 
   1704 .. _dss_bitvector:
   1705 
   1706 BitVector
   1707 ^^^^^^^^^
   1708 
   1709 The BitVector container provides a dynamic size set of bits for manipulation.
   1710 It supports individual bit setting/testing, as well as set operations.  The set
   1711 operations take time O(size of bitvector), but operations are performed one word
   1712 at a time, instead of one bit at a time.  This makes the BitVector very fast for
   1713 set operations compared to other containers.  Use the BitVector when you expect
   1714 the number of set bits to be high (i.e. a dense set).
   1715 
   1716 .. _dss_smallbitvector:
   1717 
   1718 SmallBitVector
   1719 ^^^^^^^^^^^^^^
   1720 
   1721 The SmallBitVector container provides the same interface as BitVector, but it is
   1722 optimized for the case where only a small number of bits, less than 25 or so,
   1723 are needed.  It also transparently supports larger bit counts, but slightly less
   1724 efficiently than a plain BitVector, so SmallBitVector should only be used when
   1725 larger counts are rare.
   1726 
   1727 At this time, SmallBitVector does not support set operations (and, or, xor), and
   1728 its operator[] does not provide an assignable lvalue.
   1729 
   1730 .. _dss_sparsebitvector:
   1731 
   1732 SparseBitVector
   1733 ^^^^^^^^^^^^^^^
   1734 
   1735 The SparseBitVector container is much like BitVector, with one major difference:
   1736 Only the bits that are set, are stored.  This makes the SparseBitVector much
   1737 more space efficient than BitVector when the set is sparse, as well as making
   1738 set operations O(number of set bits) instead of O(size of universe).  The
   1739 downside to the SparseBitVector is that setting and testing of random bits is
   1740 O(N), and on large SparseBitVectors, this can be slower than BitVector.  In our
   1741 implementation, setting or testing bits in sorted order (either forwards or
   1742 reverse) is O(1) worst case.  Testing and setting bits within 128 bits (depends
   1743 on size) of the current bit is also O(1).  As a general statement,
   1744 testing/setting bits in a SparseBitVector is O(distance away from last set bit).
   1745 
   1746 .. _common:
   1747 
   1748 Helpful Hints for Common Operations
   1749 ===================================
   1750 
   1751 This section describes how to perform some very simple transformations of LLVM
   1752 code.  This is meant to give examples of common idioms used, showing the
   1753 practical side of LLVM transformations.
   1754 
   1755 Because this is a "how-to" section, you should also read about the main classes
   1756 that you will be working with.  The :ref:`Core LLVM Class Hierarchy Reference
   1757 <coreclasses>` contains details and descriptions of the main classes that you
   1758 should know about.
   1759 
   1760 .. _inspection:
   1761 
   1762 Basic Inspection and Traversal Routines
   1763 ---------------------------------------
   1764 
   1765 The LLVM compiler infrastructure have many different data structures that may be
   1766 traversed.  Following the example of the C++ standard template library, the
   1767 techniques used to traverse these various data structures are all basically the
   1768 same.  For a enumerable sequence of values, the ``XXXbegin()`` function (or
   1769 method) returns an iterator to the start of the sequence, the ``XXXend()``
   1770 function returns an iterator pointing to one past the last valid element of the
   1771 sequence, and there is some ``XXXiterator`` data type that is common between the
   1772 two operations.
   1773 
   1774 Because the pattern for iteration is common across many different aspects of the
   1775 program representation, the standard template library algorithms may be used on
   1776 them, and it is easier to remember how to iterate.  First we show a few common
   1777 examples of the data structures that need to be traversed.  Other data
   1778 structures are traversed in very similar ways.
   1779 
   1780 .. _iterate_function:
   1781 
   1782 Iterating over the ``BasicBlock`` in a ``Function``
   1783 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   1784 
   1785 It's quite common to have a ``Function`` instance that you'd like to transform
   1786 in some way; in particular, you'd like to manipulate its ``BasicBlock``\ s.  To
   1787 facilitate this, you'll need to iterate over all of the ``BasicBlock``\ s that
   1788 constitute the ``Function``.  The following is an example that prints the name
   1789 of a ``BasicBlock`` and the number of ``Instruction``\ s it contains:
   1790 
   1791 .. code-block:: c++
   1792 
   1793   // func is a pointer to a Function instance
   1794   for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
   1795     // Print out the name of the basic block if it has one, and then the
   1796     // number of instructions that it contains
   1797     errs() << "Basic block (name=" << i->getName() << ") has "
   1798                << i->size() << " instructions.\n";
   1799 
   1800 Note that i can be used as if it were a pointer for the purposes of invoking
   1801 member functions of the ``Instruction`` class.  This is because the indirection
   1802 operator is overloaded for the iterator classes.  In the above code, the
   1803 expression ``i->size()`` is exactly equivalent to ``(*i).size()`` just like
   1804 you'd expect.
   1805 
   1806 .. _iterate_basicblock:
   1807 
   1808 Iterating over the ``Instruction`` in a ``BasicBlock``
   1809 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   1810 
   1811 Just like when dealing with ``BasicBlock``\ s in ``Function``\ s, it's easy to
   1812 iterate over the individual instructions that make up ``BasicBlock``\ s.  Here's
   1813 a code snippet that prints out each instruction in a ``BasicBlock``:
   1814 
   1815 .. code-block:: c++
   1816 
   1817   // blk is a pointer to a BasicBlock instance
   1818   for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
   1819      // The next statement works since operator<<(ostream&,...)
   1820      // is overloaded for Instruction&
   1821      errs() << *i << "\n";
   1822 
   1823 
   1824 However, this isn't really the best way to print out the contents of a
   1825 ``BasicBlock``!  Since the ostream operators are overloaded for virtually
   1826 anything you'll care about, you could have just invoked the print routine on the
   1827 basic block itself: ``errs() << *blk << "\n";``.
   1828 
   1829 .. _iterate_insiter:
   1830 
   1831 Iterating over the ``Instruction`` in a ``Function``
   1832 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   1833 
   1834 If you're finding that you commonly iterate over a ``Function``'s
   1835 ``BasicBlock``\ s and then that ``BasicBlock``'s ``Instruction``\ s,
   1836 ``InstIterator`` should be used instead.  You'll need to include
   1837 ``llvm/IR/InstIterator.h`` (`doxygen
   1838 <http://llvm.org/doxygen/InstIterator_8h.html>`__) and then instantiate
   1839 ``InstIterator``\ s explicitly in your code.  Here's a small example that shows
   1840 how to dump all instructions in a function to the standard error stream:
   1841 
   1842 .. code-block:: c++
   1843 
   1844   #include "llvm/IR/InstIterator.h"
   1845 
   1846   // F is a pointer to a Function instance
   1847   for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
   1848     errs() << *I << "\n";
   1849 
   1850 Easy, isn't it?  You can also use ``InstIterator``\ s to fill a work list with
   1851 its initial contents.  For example, if you wanted to initialize a work list to
   1852 contain all instructions in a ``Function`` F, all you would need to do is
   1853 something like:
   1854 
   1855 .. code-block:: c++
   1856 
   1857   std::set<Instruction*> worklist;
   1858   // or better yet, SmallPtrSet<Instruction*, 64> worklist;
   1859 
   1860   for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
   1861     worklist.insert(&*I);
   1862 
   1863 The STL set ``worklist`` would now contain all instructions in the ``Function``
   1864 pointed to by F.
   1865 
   1866 .. _iterate_convert:
   1867 
   1868 Turning an iterator into a class pointer (and vice-versa)
   1869 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   1870 
   1871 Sometimes, it'll be useful to grab a reference (or pointer) to a class instance
   1872 when all you've got at hand is an iterator.  Well, extracting a reference or a
   1873 pointer from an iterator is very straight-forward.  Assuming that ``i`` is a
   1874 ``BasicBlock::iterator`` and ``j`` is a ``BasicBlock::const_iterator``:
   1875 
   1876 .. code-block:: c++
   1877 
   1878   Instruction& inst = *i;   // Grab reference to instruction reference
   1879   Instruction* pinst = &*i; // Grab pointer to instruction reference
   1880   const Instruction& inst = *j;
   1881 
   1882 However, the iterators you'll be working with in the LLVM framework are special:
   1883 they will automatically convert to a ptr-to-instance type whenever they need to.
   1884 Instead of dereferencing the iterator and then taking the address of the result,
   1885 you can simply assign the iterator to the proper pointer type and you get the
   1886 dereference and address-of operation as a result of the assignment (behind the
   1887 scenes, this is a result of overloading casting mechanisms).  Thus the second
   1888 line of the last example,
   1889 
   1890 .. code-block:: c++
   1891 
   1892   Instruction *pinst = &*i;
   1893 
   1894 is semantically equivalent to
   1895 
   1896 .. code-block:: c++
   1897 
   1898   Instruction *pinst = i;
   1899 
   1900 It's also possible to turn a class pointer into the corresponding iterator, and
   1901 this is a constant time operation (very efficient).  The following code snippet
   1902 illustrates use of the conversion constructors provided by LLVM iterators.  By
   1903 using these, you can explicitly grab the iterator of something without actually
   1904 obtaining it via iteration over some structure:
   1905 
   1906 .. code-block:: c++
   1907 
   1908   void printNextInstruction(Instruction* inst) {
   1909     BasicBlock::iterator it(inst);
   1910     ++it; // After this line, it refers to the instruction after *inst
   1911     if (it != inst->getParent()->end()) errs() << *it << "\n";
   1912   }
   1913 
   1914 Unfortunately, these implicit conversions come at a cost; they prevent these
   1915 iterators from conforming to standard iterator conventions, and thus from being
   1916 usable with standard algorithms and containers.  For example, they prevent the
   1917 following code, where ``B`` is a ``BasicBlock``, from compiling:
   1918 
   1919 .. code-block:: c++
   1920 
   1921   llvm::SmallVector<llvm::Instruction *, 16>(B->begin(), B->end());
   1922 
   1923 Because of this, these implicit conversions may be removed some day, and
   1924 ``operator*`` changed to return a pointer instead of a reference.
   1925 
   1926 .. _iterate_complex:
   1927 
   1928 Finding call sites: a slightly more complex example
   1929 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   1930 
   1931 Say that you're writing a FunctionPass and would like to count all the locations
   1932 in the entire module (that is, across every ``Function``) where a certain
   1933 function (i.e., some ``Function *``) is already in scope.  As you'll learn
   1934 later, you may want to use an ``InstVisitor`` to accomplish this in a much more
   1935 straight-forward manner, but this example will allow us to explore how you'd do
   1936 it if you didn't have ``InstVisitor`` around.  In pseudo-code, this is what we
   1937 want to do:
   1938 
   1939 .. code-block:: none
   1940 
   1941   initialize callCounter to zero
   1942   for each Function f in the Module
   1943     for each BasicBlock b in f
   1944       for each Instruction i in b
   1945         if (i is a CallInst and calls the given function)
   1946           increment callCounter
   1947 
   1948 And the actual code is (remember, because we're writing a ``FunctionPass``, our
   1949 ``FunctionPass``-derived class simply has to override the ``runOnFunction``
   1950 method):
   1951 
   1952 .. code-block:: c++
   1953 
   1954   Function* targetFunc = ...;
   1955 
   1956   class OurFunctionPass : public FunctionPass {
   1957     public:
   1958       OurFunctionPass(): callCounter(0) { }
   1959 
   1960       virtual runOnFunction(Function& F) {
   1961         for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
   1962           for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) {
   1963             if (CallInst* callInst = dyn_cast<CallInst>(&*i)) {
   1964               // We know we've encountered a call instruction, so we
   1965               // need to determine if it's a call to the
   1966               // function pointed to by m_func or not.
   1967               if (callInst->getCalledFunction() == targetFunc)
   1968                 ++callCounter;
   1969             }
   1970           }
   1971         }
   1972       }
   1973 
   1974     private:
   1975       unsigned callCounter;
   1976   };
   1977 
   1978 .. _calls_and_invokes:
   1979 
   1980 Treating calls and invokes the same way
   1981 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   1982 
   1983 You may have noticed that the previous example was a bit oversimplified in that
   1984 it did not deal with call sites generated by 'invoke' instructions.  In this,
   1985 and in other situations, you may find that you want to treat ``CallInst``\ s and
   1986 ``InvokeInst``\ s the same way, even though their most-specific common base
   1987 class is ``Instruction``, which includes lots of less closely-related things.
   1988 For these cases, LLVM provides a handy wrapper class called ``CallSite``
   1989 (`doxygen <http://llvm.org/doxygen/classllvm_1_1CallSite.html>`__) It is
   1990 essentially a wrapper around an ``Instruction`` pointer, with some methods that
   1991 provide functionality common to ``CallInst``\ s and ``InvokeInst``\ s.
   1992 
   1993 This class has "value semantics": it should be passed by value, not by reference
   1994 and it should not be dynamically allocated or deallocated using ``operator new``
   1995 or ``operator delete``.  It is efficiently copyable, assignable and
   1996 constructable, with costs equivalents to that of a bare pointer.  If you look at
   1997 its definition, it has only a single pointer member.
   1998 
   1999 .. _iterate_chains:
   2000 
   2001 Iterating over def-use & use-def chains
   2002 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   2003 
   2004 Frequently, we might have an instance of the ``Value`` class (`doxygen
   2005 <http://llvm.org/doxygen/classllvm_1_1Value.html>`__) and we want to determine
   2006 which ``User`` s use the ``Value``.  The list of all ``User``\ s of a particular
   2007 ``Value`` is called a *def-use* chain.  For example, let's say we have a
   2008 ``Function*`` named ``F`` to a particular function ``foo``.  Finding all of the
   2009 instructions that *use* ``foo`` is as simple as iterating over the *def-use*
   2010 chain of ``F``:
   2011 
   2012 .. code-block:: c++
   2013 
   2014   Function *F = ...;
   2015 
   2016   for (User *U : F->users()) {
   2017     if (Instruction *Inst = dyn_cast<Instruction>(U)) {
   2018       errs() << "F is used in instruction:\n";
   2019       errs() << *Inst << "\n";
   2020     }
   2021 
   2022 Alternatively, it's common to have an instance of the ``User`` Class (`doxygen
   2023 <http://llvm.org/doxygen/classllvm_1_1User.html>`__) and need to know what
   2024 ``Value``\ s are used by it.  The list of all ``Value``\ s used by a ``User`` is
   2025 known as a *use-def* chain.  Instances of class ``Instruction`` are common
   2026 ``User`` s, so we might want to iterate over all of the values that a particular
   2027 instruction uses (that is, the operands of the particular ``Instruction``):
   2028 
   2029 .. code-block:: c++
   2030 
   2031   Instruction *pi = ...;
   2032 
   2033   for (Use &U : pi->operands()) {
   2034     Value *v = U.get();
   2035     // ...
   2036   }
   2037 
   2038 Declaring objects as ``const`` is an important tool of enforcing mutation free
   2039 algorithms (such as analyses, etc.).  For this purpose above iterators come in
   2040 constant flavors as ``Value::const_use_iterator`` and
   2041 ``Value::const_op_iterator``.  They automatically arise when calling
   2042 ``use/op_begin()`` on ``const Value*``\ s or ``const User*``\ s respectively.
   2043 Upon dereferencing, they return ``const Use*``\ s.  Otherwise the above patterns
   2044 remain unchanged.
   2045 
   2046 .. _iterate_preds:
   2047 
   2048 Iterating over predecessors & successors of blocks
   2049 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   2050 
   2051 Iterating over the predecessors and successors of a block is quite easy with the
   2052 routines defined in ``"llvm/IR/CFG.h"``.  Just use code like this to
   2053 iterate over all predecessors of BB:
   2054 
   2055 .. code-block:: c++
   2056 
   2057   #include "llvm/Support/CFG.h"
   2058   BasicBlock *BB = ...;
   2059 
   2060   for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
   2061     BasicBlock *Pred = *PI;
   2062     // ...
   2063   }
   2064 
   2065 Similarly, to iterate over successors use ``succ_iterator/succ_begin/succ_end``.
   2066 
   2067 .. _simplechanges:
   2068 
   2069 Making simple changes
   2070 ---------------------
   2071 
   2072 There are some primitive transformation operations present in the LLVM
   2073 infrastructure that are worth knowing about.  When performing transformations,
   2074 it's fairly common to manipulate the contents of basic blocks.  This section
   2075 describes some of the common methods for doing so and gives example code.
   2076 
   2077 .. _schanges_creating:
   2078 
   2079 Creating and inserting new ``Instruction``\ s
   2080 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   2081 
   2082 *Instantiating Instructions*
   2083 
   2084 Creation of ``Instruction``\ s is straight-forward: simply call the constructor
   2085 for the kind of instruction to instantiate and provide the necessary parameters.
   2086 For example, an ``AllocaInst`` only *requires* a (const-ptr-to) ``Type``.  Thus:
   2087 
   2088 .. code-block:: c++
   2089 
   2090   AllocaInst* ai = new AllocaInst(Type::Int32Ty);
   2091 
   2092 will create an ``AllocaInst`` instance that represents the allocation of one
   2093 integer in the current stack frame, at run time.  Each ``Instruction`` subclass
   2094 is likely to have varying default parameters which change the semantics of the
   2095 instruction, so refer to the `doxygen documentation for the subclass of
   2096 Instruction <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_ that
   2097 you're interested in instantiating.
   2098 
   2099 *Naming values*
   2100 
   2101 It is very useful to name the values of instructions when you're able to, as
   2102 this facilitates the debugging of your transformations.  If you end up looking
   2103 at generated LLVM machine code, you definitely want to have logical names
   2104 associated with the results of instructions!  By supplying a value for the
   2105 ``Name`` (default) parameter of the ``Instruction`` constructor, you associate a
   2106 logical name with the result of the instruction's execution at run time.  For
   2107 example, say that I'm writing a transformation that dynamically allocates space
   2108 for an integer on the stack, and that integer is going to be used as some kind
   2109 of index by some other code.  To accomplish this, I place an ``AllocaInst`` at
   2110 the first point in the first ``BasicBlock`` of some ``Function``, and I'm
   2111 intending to use it within the same ``Function``.  I might do:
   2112 
   2113 .. code-block:: c++
   2114 
   2115   AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
   2116 
   2117 where ``indexLoc`` is now the logical name of the instruction's execution value,
   2118 which is a pointer to an integer on the run time stack.
   2119 
   2120 *Inserting instructions*
   2121 
   2122 There are essentially three ways to insert an ``Instruction`` into an existing
   2123 sequence of instructions that form a ``BasicBlock``:
   2124 
   2125 * Insertion into an explicit instruction list
   2126 
   2127   Given a ``BasicBlock* pb``, an ``Instruction* pi`` within that ``BasicBlock``,
   2128   and a newly-created instruction we wish to insert before ``*pi``, we do the
   2129   following:
   2130 
   2131   .. code-block:: c++
   2132 
   2133       BasicBlock *pb = ...;
   2134       Instruction *pi = ...;
   2135       Instruction *newInst = new Instruction(...);
   2136 
   2137       pb->getInstList().insert(pi, newInst); // Inserts newInst before pi in pb
   2138 
   2139   Appending to the end of a ``BasicBlock`` is so common that the ``Instruction``
   2140   class and ``Instruction``-derived classes provide constructors which take a
   2141   pointer to a ``BasicBlock`` to be appended to.  For example code that looked
   2142   like:
   2143 
   2144   .. code-block:: c++
   2145 
   2146     BasicBlock *pb = ...;
   2147     Instruction *newInst = new Instruction(...);
   2148 
   2149     pb->getInstList().push_back(newInst); // Appends newInst to pb
   2150 
   2151   becomes:
   2152 
   2153   .. code-block:: c++
   2154 
   2155     BasicBlock *pb = ...;
   2156     Instruction *newInst = new Instruction(..., pb);
   2157 
   2158   which is much cleaner, especially if you are creating long instruction
   2159   streams.
   2160 
   2161 * Insertion into an implicit instruction list
   2162 
   2163   ``Instruction`` instances that are already in ``BasicBlock``\ s are implicitly
   2164   associated with an existing instruction list: the instruction list of the
   2165   enclosing basic block.  Thus, we could have accomplished the same thing as the
   2166   above code without being given a ``BasicBlock`` by doing:
   2167 
   2168   .. code-block:: c++
   2169 
   2170     Instruction *pi = ...;
   2171     Instruction *newInst = new Instruction(...);
   2172 
   2173     pi->getParent()->getInstList().insert(pi, newInst);
   2174 
   2175   In fact, this sequence of steps occurs so frequently that the ``Instruction``
   2176   class and ``Instruction``-derived classes provide constructors which take (as
   2177   a default parameter) a pointer to an ``Instruction`` which the newly-created
   2178   ``Instruction`` should precede.  That is, ``Instruction`` constructors are
   2179   capable of inserting the newly-created instance into the ``BasicBlock`` of a
   2180   provided instruction, immediately before that instruction.  Using an
   2181   ``Instruction`` constructor with a ``insertBefore`` (default) parameter, the
   2182   above code becomes:
   2183 
   2184   .. code-block:: c++
   2185 
   2186     Instruction* pi = ...;
   2187     Instruction* newInst = new Instruction(..., pi);
   2188 
   2189   which is much cleaner, especially if you're creating a lot of instructions and
   2190   adding them to ``BasicBlock``\ s.
   2191 
   2192 * Insertion using an instance of ``IRBuilder``
   2193 
   2194   Inserting several ``Instruction``\ s can be quite laborious using the previous
   2195   methods. The ``IRBuilder`` is a convenience class that can be used to add
   2196   several instructions to the end of a ``BasicBlock`` or before a particular
   2197   ``Instruction``. It also supports constant folding and renaming named
   2198   registers (see ``IRBuilder``'s template arguments).
   2199 
   2200   The example below demonstrates a very simple use of the ``IRBuilder`` where
   2201   three instructions are inserted before the instruction ``pi``. The first two
   2202   instructions are Call instructions and third instruction multiplies the return
   2203   value of the two calls.
   2204 
   2205   .. code-block:: c++
   2206 
   2207     Instruction *pi = ...;
   2208     IRBuilder<> Builder(pi);
   2209     CallInst* callOne = Builder.CreateCall(...);
   2210     CallInst* callTwo = Builder.CreateCall(...);
   2211     Value* result = Builder.CreateMul(callOne, callTwo);
   2212 
   2213   The example below is similar to the above example except that the created
   2214   ``IRBuilder`` inserts instructions at the end of the ``BasicBlock`` ``pb``.
   2215 
   2216   .. code-block:: c++
   2217 
   2218     BasicBlock *pb = ...;
   2219     IRBuilder<> Builder(pb);
   2220     CallInst* callOne = Builder.CreateCall(...);
   2221     CallInst* callTwo = Builder.CreateCall(...);
   2222     Value* result = Builder.CreateMul(callOne, callTwo);
   2223 
   2224   See :doc:`tutorial/LangImpl03` for a practical use of the ``IRBuilder``.
   2225 
   2226 
   2227 .. _schanges_deleting:
   2228 
   2229 Deleting Instructions
   2230 ^^^^^^^^^^^^^^^^^^^^^
   2231 
   2232 Deleting an instruction from an existing sequence of instructions that form a
   2233 BasicBlock_ is very straight-forward: just call the instruction's
   2234 ``eraseFromParent()`` method.  For example:
   2235 
   2236 .. code-block:: c++
   2237 
   2238   Instruction *I = .. ;
   2239   I->eraseFromParent();
   2240 
   2241 This unlinks the instruction from its containing basic block and deletes it.  If
   2242 you'd just like to unlink the instruction from its containing basic block but
   2243 not delete it, you can use the ``removeFromParent()`` method.
   2244 
   2245 .. _schanges_replacing:
   2246 
   2247 Replacing an Instruction with another Value
   2248 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   2249 
   2250 Replacing individual instructions
   2251 """""""""""""""""""""""""""""""""
   2252 
   2253 Including "`llvm/Transforms/Utils/BasicBlockUtils.h
   2254 <http://llvm.org/doxygen/BasicBlockUtils_8h-source.html>`_" permits use of two
   2255 very useful replace functions: ``ReplaceInstWithValue`` and
   2256 ``ReplaceInstWithInst``.
   2257 
   2258 .. _schanges_deleting_sub:
   2259 
   2260 Deleting Instructions
   2261 """""""""""""""""""""
   2262 
   2263 * ``ReplaceInstWithValue``
   2264 
   2265   This function replaces all uses of a given instruction with a value, and then
   2266   removes the original instruction.  The following example illustrates the
   2267   replacement of the result of a particular ``AllocaInst`` that allocates memory
   2268   for a single integer with a null pointer to an integer.
   2269 
   2270   .. code-block:: c++
   2271 
   2272     AllocaInst* instToReplace = ...;
   2273     BasicBlock::iterator ii(instToReplace);
   2274 
   2275     ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
   2276                          Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
   2277 
   2278 * ``ReplaceInstWithInst``
   2279 
   2280   This function replaces a particular instruction with another instruction,
   2281   inserting the new instruction into the basic block at the location where the
   2282   old instruction was, and replacing any uses of the old instruction with the
   2283   new instruction.  The following example illustrates the replacement of one
   2284   ``AllocaInst`` with another.
   2285 
   2286   .. code-block:: c++
   2287 
   2288     AllocaInst* instToReplace = ...;
   2289     BasicBlock::iterator ii(instToReplace);
   2290 
   2291     ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
   2292                         new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
   2293 
   2294 
   2295 Replacing multiple uses of Users and Values
   2296 """""""""""""""""""""""""""""""""""""""""""
   2297 
   2298 You can use ``Value::replaceAllUsesWith`` and ``User::replaceUsesOfWith`` to
   2299 change more than one use at a time.  See the doxygen documentation for the
   2300 `Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_ and `User Class
   2301 <http://llvm.org/doxygen/classllvm_1_1User.html>`_, respectively, for more
   2302 information.
   2303 
   2304 .. _schanges_deletingGV:
   2305 
   2306 Deleting GlobalVariables
   2307 ^^^^^^^^^^^^^^^^^^^^^^^^
   2308 
   2309 Deleting a global variable from a module is just as easy as deleting an
   2310 Instruction.  First, you must have a pointer to the global variable that you
   2311 wish to delete.  You use this pointer to erase it from its parent, the module.
   2312 For example:
   2313 
   2314 .. code-block:: c++
   2315 
   2316   GlobalVariable *GV = .. ;
   2317 
   2318   GV->eraseFromParent();
   2319 
   2320 
   2321 .. _create_types:
   2322 
   2323 How to Create Types
   2324 -------------------
   2325 
   2326 In generating IR, you may need some complex types.  If you know these types
   2327 statically, you can use ``TypeBuilder<...>::get()``, defined in
   2328 ``llvm/Support/TypeBuilder.h``, to retrieve them.  ``TypeBuilder`` has two forms
   2329 depending on whether you're building types for cross-compilation or native
   2330 library use.  ``TypeBuilder<T, true>`` requires that ``T`` be independent of the
   2331 host environment, meaning that it's built out of types from the ``llvm::types``
   2332 (`doxygen <http://llvm.org/doxygen/namespacellvm_1_1types.html>`__) namespace
   2333 and pointers, functions, arrays, etc. built of those.  ``TypeBuilder<T, false>``
   2334 additionally allows native C types whose size may depend on the host compiler.
   2335 For example,
   2336 
   2337 .. code-block:: c++
   2338 
   2339   FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get();
   2340 
   2341 is easier to read and write than the equivalent
   2342 
   2343 .. code-block:: c++
   2344 
   2345   std::vector<const Type*> params;
   2346   params.push_back(PointerType::getUnqual(Type::Int32Ty));
   2347   FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
   2348 
   2349 See the `class comment
   2350 <http://llvm.org/doxygen/TypeBuilder_8h-source.html#l00001>`_ for more details.
   2351 
   2352 .. _threading:
   2353 
   2354 Threads and LLVM
   2355 ================
   2356 
   2357 This section describes the interaction of the LLVM APIs with multithreading,
   2358 both on the part of client applications, and in the JIT, in the hosted
   2359 application.
   2360 
   2361 Note that LLVM's support for multithreading is still relatively young.  Up
   2362 through version 2.5, the execution of threaded hosted applications was
   2363 supported, but not threaded client access to the APIs.  While this use case is
   2364 now supported, clients *must* adhere to the guidelines specified below to ensure
   2365 proper operation in multithreaded mode.
   2366 
   2367 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
   2368 intrinsics in order to support threaded operation.  If you need a
   2369 multhreading-capable LLVM on a platform without a suitably modern system
   2370 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
   2371 using the resultant compiler to build a copy of LLVM with multithreading
   2372 support.
   2373 
   2374 .. _shutdown:
   2375 
   2376 Ending Execution with ``llvm_shutdown()``
   2377 -----------------------------------------
   2378 
   2379 When you are done using the LLVM APIs, you should call ``llvm_shutdown()`` to
   2380 deallocate memory used for internal structures.
   2381 
   2382 .. _managedstatic:
   2383 
   2384 Lazy Initialization with ``ManagedStatic``
   2385 ------------------------------------------
   2386 
   2387 ``ManagedStatic`` is a utility class in LLVM used to implement static
   2388 initialization of static resources, such as the global type tables.  In a
   2389 single-threaded environment, it implements a simple lazy initialization scheme.
   2390 When LLVM is compiled with support for multi-threading, however, it uses
   2391 double-checked locking to implement thread-safe lazy initialization.
   2392 
   2393 .. _llvmcontext:
   2394 
   2395 Achieving Isolation with ``LLVMContext``
   2396 ----------------------------------------
   2397 
   2398 ``LLVMContext`` is an opaque class in the LLVM API which clients can use to
   2399 operate multiple, isolated instances of LLVM concurrently within the same
   2400 address space.  For instance, in a hypothetical compile-server, the compilation
   2401 of an individual translation unit is conceptually independent from all the
   2402 others, and it would be desirable to be able to compile incoming translation
   2403 units concurrently on independent server threads.  Fortunately, ``LLVMContext``
   2404 exists to enable just this kind of scenario!
   2405 
   2406 Conceptually, ``LLVMContext`` provides isolation.  Every LLVM entity
   2407 (``Module``\ s, ``Value``\ s, ``Type``\ s, ``Constant``\ s, etc.) in LLVM's
   2408 in-memory IR belongs to an ``LLVMContext``.  Entities in different contexts
   2409 *cannot* interact with each other: ``Module``\ s in different contexts cannot be
   2410 linked together, ``Function``\ s cannot be added to ``Module``\ s in different
   2411 contexts, etc.  What this means is that is is safe to compile on multiple
   2412 threads simultaneously, as long as no two threads operate on entities within the
   2413 same context.
   2414 
   2415 In practice, very few places in the API require the explicit specification of a
   2416 ``LLVMContext``, other than the ``Type`` creation/lookup APIs.  Because every
   2417 ``Type`` carries a reference to its owning context, most other entities can
   2418 determine what context they belong to by looking at their own ``Type``.  If you
   2419 are adding new entities to LLVM IR, please try to maintain this interface
   2420 design.
   2421 
   2422 .. _jitthreading:
   2423 
   2424 Threads and the JIT
   2425 -------------------
   2426 
   2427 LLVM's "eager" JIT compiler is safe to use in threaded programs.  Multiple
   2428 threads can call ``ExecutionEngine::getPointerToFunction()`` or
   2429 ``ExecutionEngine::runFunction()`` concurrently, and multiple threads can run
   2430 code output by the JIT concurrently.  The user must still ensure that only one
   2431 thread accesses IR in a given ``LLVMContext`` while another thread might be
   2432 modifying it.  One way to do that is to always hold the JIT lock while accessing
   2433 IR outside the JIT (the JIT *modifies* the IR by adding ``CallbackVH``\ s).
   2434 Another way is to only call ``getPointerToFunction()`` from the
   2435 ``LLVMContext``'s thread.
   2436 
   2437 When the JIT is configured to compile lazily (using
   2438 ``ExecutionEngine::DisableLazyCompilation(false)``), there is currently a `race
   2439 condition <http://llvm.org/bugs/show_bug.cgi?id=5184>`_ in updating call sites
   2440 after a function is lazily-jitted.  It's still possible to use the lazy JIT in a
   2441 threaded program if you ensure that only one thread at a time can call any
   2442 particular lazy stub and that the JIT lock guards any IR access, but we suggest
   2443 using only the eager JIT in threaded programs.
   2444 
   2445 .. _advanced:
   2446 
   2447 Advanced Topics
   2448 ===============
   2449 
   2450 This section describes some of the advanced or obscure API's that most clients
   2451 do not need to be aware of.  These API's tend manage the inner workings of the
   2452 LLVM system, and only need to be accessed in unusual circumstances.
   2453 
   2454 .. _SymbolTable:
   2455 
   2456 The ``ValueSymbolTable`` class
   2457 ------------------------------
   2458 
   2459 The ``ValueSymbolTable`` (`doxygen
   2460 <http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html>`__) class provides
   2461 a symbol table that the :ref:`Function <c_Function>` and Module_ classes use for
   2462 naming value definitions.  The symbol table can provide a name for any Value_.
   2463 
   2464 Note that the ``SymbolTable`` class should not be directly accessed by most
   2465 clients.  It should only be used when iteration over the symbol table names
   2466 themselves are required, which is very special purpose.  Note that not all LLVM
   2467 Value_\ s have names, and those without names (i.e. they have an empty name) do
   2468 not exist in the symbol table.
   2469 
   2470 Symbol tables support iteration over the values in the symbol table with
   2471 ``begin/end/iterator`` and supports querying to see if a specific name is in the
   2472 symbol table (with ``lookup``).  The ``ValueSymbolTable`` class exposes no
   2473 public mutator methods, instead, simply call ``setName`` on a value, which will
   2474 autoinsert it into the appropriate symbol table.
   2475 
   2476 .. _UserLayout:
   2477 
   2478 The ``User`` and owned ``Use`` classes' memory layout
   2479 -----------------------------------------------------
   2480 
   2481 The ``User`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1User.html>`__)
   2482 class provides a basis for expressing the ownership of ``User`` towards other
   2483 `Value instance <http://llvm.org/doxygen/classllvm_1_1Value.html>`_\ s.  The
   2484 ``Use`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1Use.html>`__) helper
   2485 class is employed to do the bookkeeping and to facilitate *O(1)* addition and
   2486 removal.
   2487 
   2488 .. _Use2User:
   2489 
   2490 Interaction and relationship between ``User`` and ``Use`` objects
   2491 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   2492 
   2493 A subclass of ``User`` can choose between incorporating its ``Use`` objects or
   2494 refer to them out-of-line by means of a pointer.  A mixed variant (some ``Use``
   2495 s inline others hung off) is impractical and breaks the invariant that the
   2496 ``Use`` objects belonging to the same ``User`` form a contiguous array.
   2497 
   2498 We have 2 different layouts in the ``User`` (sub)classes:
   2499 
   2500 * Layout a)
   2501 
   2502   The ``Use`` object(s) are inside (resp. at fixed offset) of the ``User``
   2503   object and there are a fixed number of them.
   2504 
   2505 * Layout b)
   2506 
   2507   The ``Use`` object(s) are referenced by a pointer to an array from the
   2508   ``User`` object and there may be a variable number of them.
   2509 
   2510 As of v2.4 each layout still possesses a direct pointer to the start of the
   2511 array of ``Use``\ s.  Though not mandatory for layout a), we stick to this
   2512 redundancy for the sake of simplicity.  The ``User`` object also stores the
   2513 number of ``Use`` objects it has. (Theoretically this information can also be
   2514 calculated given the scheme presented below.)
   2515 
   2516 Special forms of allocation operators (``operator new``) enforce the following
   2517 memory layouts:
   2518 
   2519 * Layout a) is modelled by prepending the ``User`` object by the ``Use[]``
   2520   array.
   2521 
   2522   .. code-block:: none
   2523 
   2524     ...---.---.---.---.-------...
   2525       | P | P | P | P | User
   2526     '''---'---'---'---'-------'''
   2527 
   2528 * Layout b) is modelled by pointing at the ``Use[]`` array.
   2529 
   2530   .. code-block:: none
   2531 
   2532     .-------...
   2533     | User
   2534     '-------'''
   2535         |
   2536         v
   2537         .---.---.---.---...
   2538         | P | P | P | P |
   2539         '---'---'---'---'''
   2540 
   2541 *(In the above figures* '``P``' *stands for the* ``Use**`` *that is stored in
   2542 each* ``Use`` *object in the member* ``Use::Prev`` *)*
   2543 
   2544 .. _Waymarking:
   2545 
   2546 The waymarking algorithm
   2547 ^^^^^^^^^^^^^^^^^^^^^^^^
   2548 
   2549 Since the ``Use`` objects are deprived of the direct (back)pointer to their
   2550 ``User`` objects, there must be a fast and exact method to recover it.  This is
   2551 accomplished by the following scheme:
   2552 
   2553 A bit-encoding in the 2 LSBits (least significant bits) of the ``Use::Prev``
   2554 allows to find the start of the ``User`` object:
   2555 
   2556 * ``00`` --- binary digit 0
   2557 
   2558 * ``01`` --- binary digit 1
   2559 
   2560 * ``10`` --- stop and calculate (``s``)
   2561 
   2562 * ``11`` --- full stop (``S``)
   2563 
   2564 Given a ``Use*``, all we have to do is to walk till we get a stop and we either
   2565 have a ``User`` immediately behind or we have to walk to the next stop picking
   2566 up digits and calculating the offset:
   2567 
   2568 .. code-block:: none
   2569 
   2570   .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
   2571   | 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
   2572   '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
   2573       |+15                |+10            |+6         |+3     |+1
   2574       |                   |               |           |       | __>
   2575       |                   |               |           | __________>
   2576       |                   |               | ______________________>
   2577       |                   | ______________________________________>
   2578       | __________________________________________________________>
   2579 
   2580 Only the significant number of bits need to be stored between the stops, so that
   2581 the *worst case is 20 memory accesses* when there are 1000 ``Use`` objects
   2582 associated with a ``User``.
   2583 
   2584 .. _ReferenceImpl:
   2585 
   2586 Reference implementation
   2587 ^^^^^^^^^^^^^^^^^^^^^^^^
   2588 
   2589 The following literate Haskell fragment demonstrates the concept:
   2590 
   2591 .. code-block:: haskell
   2592 
   2593   > import Test.QuickCheck
   2594   >
   2595   > digits :: Int -> [Char] -> [Char]
   2596   > digits 0 acc = '0' : acc
   2597   > digits 1 acc = '1' : acc
   2598   > digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
   2599   >
   2600   > dist :: Int -> [Char] -> [Char]
   2601   > dist 0 [] = ['S']
   2602   > dist 0 acc = acc
   2603   > dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
   2604   > dist n acc = dist (n - 1) $ dist 1 acc
   2605   >
   2606   > takeLast n ss = reverse $ take n $ reverse ss
   2607   >
   2608   > test = takeLast 40 $ dist 20 []
   2609   >
   2610 
   2611 Printing <test> gives: ``"1s100000s11010s10100s1111s1010s110s11s1S"``
   2612 
   2613 The reverse algorithm computes the length of the string just by examining a
   2614 certain prefix:
   2615 
   2616 .. code-block:: haskell
   2617 
   2618   > pref :: [Char] -> Int
   2619   > pref "S" = 1
   2620   > pref ('s':'1':rest) = decode 2 1 rest
   2621   > pref (_:rest) = 1 + pref rest
   2622   >
   2623   > decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
   2624   > decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
   2625   > decode walk acc _ = walk + acc
   2626   >
   2627 
   2628 Now, as expected, printing <pref test> gives ``40``.
   2629 
   2630 We can *quickCheck* this with following property:
   2631 
   2632 .. code-block:: haskell
   2633 
   2634   > testcase = dist 2000 []
   2635   > testcaseLength = length testcase
   2636   >
   2637   > identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
   2638   >     where arr = takeLast n testcase
   2639   >
   2640 
   2641 As expected <quickCheck identityProp> gives:
   2642 
   2643 ::
   2644 
   2645   *Main> quickCheck identityProp
   2646   OK, passed 100 tests.
   2647 
   2648 Let's be a bit more exhaustive:
   2649 
   2650 .. code-block:: haskell
   2651 
   2652   >
   2653   > deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
   2654   >
   2655 
   2656 And here is the result of <deepCheck identityProp>:
   2657 
   2658 ::
   2659 
   2660   *Main> deepCheck identityProp
   2661   OK, passed 500 tests.
   2662 
   2663 .. _Tagging:
   2664 
   2665 Tagging considerations
   2666 ^^^^^^^^^^^^^^^^^^^^^^
   2667 
   2668 To maintain the invariant that the 2 LSBits of each ``Use**`` in ``Use`` never
   2669 change after being set up, setters of ``Use::Prev`` must re-tag the new
   2670 ``Use**`` on every modification.  Accordingly getters must strip the tag bits.
   2671 
   2672 For layout b) instead of the ``User`` we find a pointer (``User*`` with LSBit
   2673 set).  Following this pointer brings us to the ``User``.  A portable trick
   2674 ensures that the first bytes of ``User`` (if interpreted as a pointer) never has
   2675 the LSBit set. (Portability is relying on the fact that all known compilers
   2676 place the ``vptr`` in the first word of the instances.)
   2677 
   2678 .. _polymorphism:
   2679 
   2680 Designing Type Hiercharies and Polymorphic Interfaces
   2681 -----------------------------------------------------
   2682 
   2683 There are two different design patterns that tend to result in the use of
   2684 virtual dispatch for methods in a type hierarchy in C++ programs. The first is
   2685 a genuine type hierarchy where different types in the hierarchy model
   2686 a specific subset of the functionality and semantics, and these types nest
   2687 strictly within each other. Good examples of this can be seen in the ``Value``
   2688 or ``Type`` type hierarchies.
   2689 
   2690 A second is the desire to dispatch dynamically across a collection of
   2691 polymorphic interface implementations. This latter use case can be modeled with
   2692 virtual dispatch and inheritance by defining an abstract interface base class
   2693 which all implementations derive from and override. However, this
   2694 implementation strategy forces an **"is-a"** relationship to exist that is not
   2695 actually meaningful. There is often not some nested hierarchy of useful
   2696 generalizations which code might interact with and move up and down. Instead,
   2697 there is a singular interface which is dispatched across a range of
   2698 implementations.
   2699 
   2700 The preferred implementation strategy for the second use case is that of
   2701 generic programming (sometimes called "compile-time duck typing" or "static
   2702 polymorphism"). For example, a template over some type parameter ``T`` can be
   2703 instantiated across any particular implementation that conforms to the
   2704 interface or *concept*. A good example here is the highly generic properties of
   2705 any type which models a node in a directed graph. LLVM models these primarily
   2706 through templates and generic programming. Such templates include the
   2707 ``LoopInfoBase`` and ``DominatorTreeBase``. When this type of polymorphism
   2708 truly needs **dynamic** dispatch you can generalize it using a technique
   2709 called *concept-based polymorphism*. This pattern emulates the interfaces and
   2710 behaviors of templates using a very limited form of virtual dispatch for type
   2711 erasure inside its implementation. You can find examples of this technique in
   2712 the ``PassManager.h`` system, and there is a more detailed introduction to it
   2713 by Sean Parent in several of his talks and papers:
   2714 
   2715 #. `Inheritance Is The Base Class of Evil
   2716    <http://channel9.msdn.com/Events/GoingNative/2013/Inheritance-Is-The-Base-Class-of-Evil>`_
   2717    - The GoingNative 2013 talk describing this technique, and probably the best
   2718    place to start.
   2719 #. `Value Semantics and Concepts-based Polymorphism
   2720    <http://www.youtube.com/watch?v=_BpMYeUFXv8>`_ - The C++Now! 2012 talk
   2721    describing this technique in more detail.
   2722 #. `Sean Parent's Papers and Presentations
   2723    <http://github.com/sean-parent/sean-parent.github.com/wiki/Papers-and-Presentations>`_
   2724    - A Github project full of links to slides, video, and sometimes code.
   2725 
   2726 When deciding between creating a type hierarchy (with either tagged or virtual
   2727 dispatch) and using templates or concepts-based polymorphism, consider whether
   2728 there is some refinement of an abstract base class which is a semantically
   2729 meaningful type on an interface boundary. If anything more refined than the
   2730 root abstract interface is meaningless to talk about as a partial extension of
   2731 the semantic model, then your use case likely fits better with polymorphism and
   2732 you should avoid using virtual dispatch. However, there may be some exigent
   2733 circumstances that require one technique or the other to be used.
   2734 
   2735 If you do need to introduce a type hierarchy, we prefer to use explicitly
   2736 closed type hierarchies with manual tagged dispatch and/or RTTI rather than the
   2737 open inheritance model and virtual dispatch that is more common in C++ code.
   2738 This is because LLVM rarely encourages library consumers to extend its core
   2739 types, and leverages the closed and tag-dispatched nature of its hierarchies to
   2740 generate significantly more efficient code. We have also found that a large
   2741 amount of our usage of type hierarchies fits better with tag-based pattern
   2742 matching rather than dynamic dispatch across a common interface. Within LLVM we
   2743 have built custom helpers to facilitate this design. See this document's
   2744 section on :ref:`isa and dyn_cast <isa>` and our :doc:`detailed document
   2745 <HowToSetUpLLVMStyleRTTI>` which describes how you can implement this
   2746 pattern for use with the LLVM helpers.
   2747 
   2748 .. _abi_breaking_checks:
   2749 
   2750 ABI Breaking Checks
   2751 -------------------
   2752 
   2753 Checks and asserts that alter the LLVM C++ ABI are predicated on the
   2754 preprocessor symbol `LLVM_ENABLE_ABI_BREAKING_CHECKS` -- LLVM
   2755 libraries built with `LLVM_ENABLE_ABI_BREAKING_CHECKS` are not ABI
   2756 compatible LLVM libraries built without it defined.  By default,
   2757 turning on assertions also turns on `LLVM_ENABLE_ABI_BREAKING_CHECKS`
   2758 so a default +Asserts build is not ABI compatible with a
   2759 default -Asserts build.  Clients that want ABI compatibility
   2760 between +Asserts and -Asserts builds should use the CMake or autoconf
   2761 build systems to set `LLVM_ENABLE_ABI_BREAKING_CHECKS` independently
   2762 of `LLVM_ENABLE_ASSERTIONS`.
   2763 
   2764 .. _coreclasses:
   2765 
   2766 The Core LLVM Class Hierarchy Reference
   2767 =======================================
   2768 
   2769 ``#include "llvm/IR/Type.h"``
   2770 
   2771 header source: `Type.h <http://llvm.org/doxygen/Type_8h-source.html>`_
   2772 
   2773 doxygen info: `Type Clases <http://llvm.org/doxygen/classllvm_1_1Type.html>`_
   2774 
   2775 The Core LLVM classes are the primary means of representing the program being
   2776 inspected or transformed.  The core LLVM classes are defined in header files in
   2777 the ``include/llvm/IR`` directory, and implemented in the ``lib/IR``
   2778 directory. It's worth noting that, for historical reasons, this library is
   2779 called ``libLLVMCore.so``, not ``libLLVMIR.so`` as you might expect.
   2780 
   2781 .. _Type:
   2782 
   2783 The Type class and Derived Types
   2784 --------------------------------
   2785 
   2786 ``Type`` is a superclass of all type classes.  Every ``Value`` has a ``Type``.
   2787 ``Type`` cannot be instantiated directly but only through its subclasses.
   2788 Certain primitive types (``VoidType``, ``LabelType``, ``FloatType`` and
   2789 ``DoubleType``) have hidden subclasses.  They are hidden because they offer no
   2790 useful functionality beyond what the ``Type`` class offers except to distinguish
   2791 themselves from other subclasses of ``Type``.
   2792 
   2793 All other types are subclasses of ``DerivedType``.  Types can be named, but this
   2794 is not a requirement.  There exists exactly one instance of a given shape at any
   2795 one time.  This allows type equality to be performed with address equality of
   2796 the Type Instance.  That is, given two ``Type*`` values, the types are identical
   2797 if the pointers are identical.
   2798 
   2799 .. _m_Type:
   2800 
   2801 Important Public Methods
   2802 ^^^^^^^^^^^^^^^^^^^^^^^^
   2803 
   2804 * ``bool isIntegerTy() const``: Returns true for any integer type.
   2805 
   2806 * ``bool isFloatingPointTy()``: Return true if this is one of the five
   2807   floating point types.
   2808 
   2809 * ``bool isSized()``: Return true if the type has known size.  Things
   2810   that don't have a size are abstract types, labels and void.
   2811 
   2812 .. _derivedtypes:
   2813 
   2814 Important Derived Types
   2815 ^^^^^^^^^^^^^^^^^^^^^^^
   2816 
   2817 ``IntegerType``
   2818   Subclass of DerivedType that represents integer types of any bit width.  Any
   2819   bit width between ``IntegerType::MIN_INT_BITS`` (1) and
   2820   ``IntegerType::MAX_INT_BITS`` (~8 million) can be represented.
   2821 
   2822   * ``static const IntegerType* get(unsigned NumBits)``: get an integer
   2823     type of a specific bit width.
   2824 
   2825   * ``unsigned getBitWidth() const``: Get the bit width of an integer type.
   2826 
   2827 ``SequentialType``
   2828   This is subclassed by ArrayType, PointerType and VectorType.
   2829 
   2830   * ``const Type * getElementType() const``: Returns the type of each
   2831     of the elements in the sequential type.
   2832 
   2833 ``ArrayType``
   2834   This is a subclass of SequentialType and defines the interface for array
   2835   types.
   2836 
   2837   * ``unsigned getNumElements() const``: Returns the number of elements
   2838     in the array.
   2839 
   2840 ``PointerType``
   2841   Subclass of SequentialType for pointer types.
   2842 
   2843 ``VectorType``
   2844   Subclass of SequentialType for vector types.  A vector type is similar to an
   2845   ArrayType but is distinguished because it is a first class type whereas
   2846   ArrayType is not.  Vector types are used for vector operations and are usually
   2847   small vectors of an integer or floating point type.
   2848 
   2849 ``StructType``
   2850   Subclass of DerivedTypes for struct types.
   2851 
   2852 .. _FunctionType:
   2853 
   2854 ``FunctionType``
   2855   Subclass of DerivedTypes for function types.
   2856 
   2857   * ``bool isVarArg() const``: Returns true if it's a vararg function.
   2858 
   2859   * ``const Type * getReturnType() const``: Returns the return type of the
   2860     function.
   2861 
   2862   * ``const Type * getParamType (unsigned i)``: Returns the type of the ith
   2863     parameter.
   2864 
   2865   * ``const unsigned getNumParams() const``: Returns the number of formal
   2866     parameters.
   2867 
   2868 .. _Module:
   2869 
   2870 The ``Module`` class
   2871 --------------------
   2872 
   2873 ``#include "llvm/IR/Module.h"``
   2874 
   2875 header source: `Module.h <http://llvm.org/doxygen/Module_8h-source.html>`_
   2876 
   2877 doxygen info: `Module Class <http://llvm.org/doxygen/classllvm_1_1Module.html>`_
   2878 
   2879 The ``Module`` class represents the top level structure present in LLVM
   2880 programs.  An LLVM module is effectively either a translation unit of the
   2881 original program or a combination of several translation units merged by the
   2882 linker.  The ``Module`` class keeps track of a list of :ref:`Function
   2883 <c_Function>`\ s, a list of GlobalVariable_\ s, and a SymbolTable_.
   2884 Additionally, it contains a few helpful member functions that try to make common
   2885 operations easy.
   2886 
   2887 .. _m_Module:
   2888 
   2889 Important Public Members of the ``Module`` class
   2890 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   2891 
   2892 * ``Module::Module(std::string name = "")``
   2893 
   2894   Constructing a Module_ is easy.  You can optionally provide a name for it
   2895   (probably based on the name of the translation unit).
   2896 
   2897 * | ``Module::iterator`` - Typedef for function list iterator
   2898   | ``Module::const_iterator`` - Typedef for const_iterator.
   2899   | ``begin()``, ``end()``, ``size()``, ``empty()``
   2900 
   2901   These are forwarding methods that make it easy to access the contents of a
   2902   ``Module`` object's :ref:`Function <c_Function>` list.
   2903 
   2904 * ``Module::FunctionListType &getFunctionList()``
   2905 
   2906   Returns the list of :ref:`Function <c_Function>`\ s.  This is necessary to use
   2907   when you need to update the list or perform a complex action that doesn't have
   2908   a forwarding method.
   2909 
   2910 ----------------
   2911 
   2912 * | ``Module::global_iterator`` - Typedef for global variable list iterator
   2913   | ``Module::const_global_iterator`` - Typedef for const_iterator.
   2914   | ``global_begin()``, ``global_end()``, ``global_size()``, ``global_empty()``
   2915 
   2916   These are forwarding methods that make it easy to access the contents of a
   2917   ``Module`` object's GlobalVariable_ list.
   2918 
   2919 * ``Module::GlobalListType &getGlobalList()``
   2920 
   2921   Returns the list of GlobalVariable_\ s.  This is necessary to use when you
   2922   need to update the list or perform a complex action that doesn't have a
   2923   forwarding method.
   2924 
   2925 ----------------
   2926 
   2927 * ``SymbolTable *getSymbolTable()``
   2928 
   2929   Return a reference to the SymbolTable_ for this ``Module``.
   2930 
   2931 ----------------
   2932 
   2933 * ``Function *getFunction(StringRef Name) const``
   2934 
   2935   Look up the specified function in the ``Module`` SymbolTable_.  If it does not
   2936   exist, return ``null``.
   2937 
   2938 * ``Function *getOrInsertFunction(const std::string &Name, const FunctionType
   2939   *T)``
   2940 
   2941   Look up the specified function in the ``Module`` SymbolTable_.  If it does not
   2942   exist, add an external declaration for the function and return it.
   2943 
   2944 * ``std::string getTypeName(const Type *Ty)``
   2945 
   2946   If there is at least one entry in the SymbolTable_ for the specified Type_,
   2947   return it.  Otherwise return the empty string.
   2948 
   2949 * ``bool addTypeName(const std::string &Name, const Type *Ty)``
   2950 
   2951   Insert an entry in the SymbolTable_ mapping ``Name`` to ``Ty``.  If there is
   2952   already an entry for this name, true is returned and the SymbolTable_ is not
   2953   modified.
   2954 
   2955 .. _Value:
   2956 
   2957 The ``Value`` class
   2958 -------------------
   2959 
   2960 ``#include "llvm/IR/Value.h"``
   2961 
   2962 header source: `Value.h <http://llvm.org/doxygen/Value_8h-source.html>`_
   2963 
   2964 doxygen info: `Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_
   2965 
   2966 The ``Value`` class is the most important class in the LLVM Source base.  It
   2967 represents a typed value that may be used (among other things) as an operand to
   2968 an instruction.  There are many different types of ``Value``\ s, such as
   2969 Constant_\ s, Argument_\ s.  Even Instruction_\ s and :ref:`Function
   2970 <c_Function>`\ s are ``Value``\ s.
   2971 
   2972 A particular ``Value`` may be used many times in the LLVM representation for a
   2973 program.  For example, an incoming argument to a function (represented with an
   2974 instance of the Argument_ class) is "used" by every instruction in the function
   2975 that references the argument.  To keep track of this relationship, the ``Value``
   2976 class keeps a list of all of the ``User``\ s that is using it (the User_ class
   2977 is a base class for all nodes in the LLVM graph that can refer to ``Value``\ s).
   2978 This use list is how LLVM represents def-use information in the program, and is
   2979 accessible through the ``use_*`` methods, shown below.
   2980 
   2981 Because LLVM is a typed representation, every LLVM ``Value`` is typed, and this
   2982 Type_ is available through the ``getType()`` method.  In addition, all LLVM
   2983 values can be named.  The "name" of the ``Value`` is a symbolic string printed
   2984 in the LLVM code:
   2985 
   2986 .. code-block:: llvm
   2987 
   2988   %foo = add i32 1, 2
   2989 
   2990 .. _nameWarning:
   2991 
   2992 The name of this instruction is "foo". **NOTE** that the name of any value may
   2993 be missing (an empty string), so names should **ONLY** be used for debugging
   2994 (making the source code easier to read, debugging printouts), they should not be
   2995 used to keep track of values or map between them.  For this purpose, use a
   2996 ``std::map`` of pointers to the ``Value`` itself instead.
   2997 
   2998 One important aspect of LLVM is that there is no distinction between an SSA
   2999 variable and the operation that produces it.  Because of this, any reference to
   3000 the value produced by an instruction (or the value available as an incoming
   3001 argument, for example) is represented as a direct pointer to the instance of the
   3002 class that represents this value.  Although this may take some getting used to,
   3003 it simplifies the representation and makes it easier to manipulate.
   3004 
   3005 .. _m_Value:
   3006 
   3007 Important Public Members of the ``Value`` class
   3008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   3009 
   3010 * | ``Value::use_iterator`` - Typedef for iterator over the use-list
   3011   | ``Value::const_use_iterator`` - Typedef for const_iterator over the
   3012     use-list
   3013   | ``unsigned use_size()`` - Returns the number of users of the value.
   3014   | ``bool use_empty()`` - Returns true if there are no users.
   3015   | ``use_iterator use_begin()`` - Get an iterator to the start of the
   3016     use-list.
   3017   | ``use_iterator use_end()`` - Get an iterator to the end of the use-list.
   3018   | ``User *use_back()`` - Returns the last element in the list.
   3019 
   3020   These methods are the interface to access the def-use information in LLVM.
   3021   As with all other iterators in LLVM, the naming conventions follow the
   3022   conventions defined by the STL_.
   3023 
   3024 * ``Type *getType() const``
   3025   This method returns the Type of the Value.
   3026 
   3027 * | ``bool hasName() const``
   3028   | ``std::string getName() const``
   3029   | ``void setName(const std::string &Name)``
   3030 
   3031   This family of methods is used to access and assign a name to a ``Value``, be
   3032   aware of the :ref:`precaution above <nameWarning>`.
   3033 
   3034 * ``void replaceAllUsesWith(Value *V)``
   3035 
   3036   This method traverses the use list of a ``Value`` changing all User_\ s of the
   3037   current value to refer to "``V``" instead.  For example, if you detect that an
   3038   instruction always produces a constant value (for example through constant
   3039   folding), you can replace all uses of the instruction with the constant like
   3040   this:
   3041 
   3042   .. code-block:: c++
   3043 
   3044     Inst->replaceAllUsesWith(ConstVal);
   3045 
   3046 .. _User:
   3047 
   3048 The ``User`` class
   3049 ------------------
   3050 
   3051 ``#include "llvm/IR/User.h"``
   3052 
   3053 header source: `User.h <http://llvm.org/doxygen/User_8h-source.html>`_
   3054 
   3055 doxygen info: `User Class <http://llvm.org/doxygen/classllvm_1_1User.html>`_
   3056 
   3057 Superclass: Value_
   3058 
   3059 The ``User`` class is the common base class of all LLVM nodes that may refer to
   3060 ``Value``\ s.  It exposes a list of "Operands" that are all of the ``Value``\ s
   3061 that the User is referring to.  The ``User`` class itself is a subclass of
   3062 ``Value``.
   3063 
   3064 The operands of a ``User`` point directly to the LLVM ``Value`` that it refers
   3065 to.  Because LLVM uses Static Single Assignment (SSA) form, there can only be
   3066 one definition referred to, allowing this direct connection.  This connection
   3067 provides the use-def information in LLVM.
   3068 
   3069 .. _m_User:
   3070 
   3071 Important Public Members of the ``User`` class
   3072 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   3073 
   3074 The ``User`` class exposes the operand list in two ways: through an index access
   3075 interface and through an iterator based interface.
   3076 
   3077 * | ``Value *getOperand(unsigned i)``
   3078   | ``unsigned getNumOperands()``
   3079 
   3080   These two methods expose the operands of the ``User`` in a convenient form for
   3081   direct access.
   3082 
   3083 * | ``User::op_iterator`` - Typedef for iterator over the operand list
   3084   | ``op_iterator op_begin()`` - Get an iterator to the start of the operand
   3085     list.
   3086   | ``op_iterator op_end()`` - Get an iterator to the end of the operand list.
   3087 
   3088   Together, these methods make up the iterator based interface to the operands
   3089   of a ``User``.
   3090 
   3091 
   3092 .. _Instruction:
   3093 
   3094 The ``Instruction`` class
   3095 -------------------------
   3096 
   3097 ``#include "llvm/IR/Instruction.h"``
   3098 
   3099 header source: `Instruction.h
   3100 <http://llvm.org/doxygen/Instruction_8h-source.html>`_
   3101 
   3102 doxygen info: `Instruction Class
   3103 <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_
   3104 
   3105 Superclasses: User_, Value_
   3106 
   3107 The ``Instruction`` class is the common base class for all LLVM instructions.
   3108 It provides only a few methods, but is a very commonly used class.  The primary
   3109 data tracked by the ``Instruction`` class itself is the opcode (instruction
   3110 type) and the parent BasicBlock_ the ``Instruction`` is embedded into.  To
   3111 represent a specific type of instruction, one of many subclasses of
   3112 ``Instruction`` are used.
   3113 
   3114 Because the ``Instruction`` class subclasses the User_ class, its operands can
   3115 be accessed in the same way as for other ``User``\ s (with the
   3116 ``getOperand()``/``getNumOperands()`` and ``op_begin()``/``op_end()`` methods).
   3117 An important file for the ``Instruction`` class is the ``llvm/Instruction.def``
   3118 file.  This file contains some meta-data about the various different types of
   3119 instructions in LLVM.  It describes the enum values that are used as opcodes
   3120 (for example ``Instruction::Add`` and ``Instruction::ICmp``), as well as the
   3121 concrete sub-classes of ``Instruction`` that implement the instruction (for
   3122 example BinaryOperator_ and CmpInst_).  Unfortunately, the use of macros in this
   3123 file confuses doxygen, so these enum values don't show up correctly in the
   3124 `doxygen output <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_.
   3125 
   3126 .. _s_Instruction:
   3127 
   3128 Important Subclasses of the ``Instruction`` class
   3129 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   3130 
   3131 .. _BinaryOperator:
   3132 
   3133 * ``BinaryOperator``
   3134 
   3135   This subclasses represents all two operand instructions whose operands must be
   3136   the same type, except for the comparison instructions.
   3137 
   3138 .. _CastInst:
   3139 
   3140 * ``CastInst``
   3141   This subclass is the parent of the 12 casting instructions.  It provides
   3142   common operations on cast instructions.
   3143 
   3144 .. _CmpInst:
   3145 
   3146 * ``CmpInst``
   3147 
   3148   This subclass respresents the two comparison instructions,
   3149   `ICmpInst <LangRef.html#i_icmp>`_ (integer opreands), and
   3150   `FCmpInst <LangRef.html#i_fcmp>`_ (floating point operands).
   3151 
   3152 .. _TerminatorInst:
   3153 
   3154 * ``TerminatorInst``
   3155 
   3156   This subclass is the parent of all terminator instructions (those which can
   3157   terminate a block).
   3158 
   3159 .. _m_Instruction:
   3160 
   3161 Important Public Members of the ``Instruction`` class
   3162 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   3163 
   3164 * ``BasicBlock *getParent()``
   3165 
   3166   Returns the BasicBlock_ that this
   3167   ``Instruction`` is embedded into.
   3168 
   3169 * ``bool mayWriteToMemory()``
   3170 
   3171   Returns true if the instruction writes to memory, i.e. it is a ``call``,
   3172   ``free``, ``invoke``, or ``store``.
   3173 
   3174 * ``unsigned getOpcode()``
   3175 
   3176   Returns the opcode for the ``Instruction``.
   3177 
   3178 * ``Instruction *clone() const``
   3179 
   3180   Returns another instance of the specified instruction, identical in all ways
   3181   to the original except that the instruction has no parent (i.e. it's not
   3182   embedded into a BasicBlock_), and it has no name.
   3183 
   3184 .. _Constant:
   3185 
   3186 The ``Constant`` class and subclasses
   3187 -------------------------------------
   3188 
   3189 Constant represents a base class for different types of constants.  It is
   3190 subclassed by ConstantInt, ConstantArray, etc. for representing the various
   3191 types of Constants.  GlobalValue_ is also a subclass, which represents the
   3192 address of a global variable or function.
   3193 
   3194 .. _s_Constant:
   3195 
   3196 Important Subclasses of Constant
   3197 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   3198 
   3199 * ConstantInt : This subclass of Constant represents an integer constant of
   3200   any width.
   3201 
   3202   * ``const APInt& getValue() const``: Returns the underlying
   3203     value of this constant, an APInt value.
   3204 
   3205   * ``int64_t getSExtValue() const``: Converts the underlying APInt value to an
   3206     int64_t via sign extension.  If the value (not the bit width) of the APInt
   3207     is too large to fit in an int64_t, an assertion will result.  For this
   3208     reason, use of this method is discouraged.
   3209 
   3210   * ``uint64_t getZExtValue() const``: Converts the underlying APInt value
   3211     to a uint64_t via zero extension.  IF the value (not the bit width) of the
   3212     APInt is too large to fit in a uint64_t, an assertion will result.  For this
   3213     reason, use of this method is discouraged.
   3214 
   3215   * ``static ConstantInt* get(const APInt& Val)``: Returns the ConstantInt
   3216     object that represents the value provided by ``Val``.  The type is implied
   3217     as the IntegerType that corresponds to the bit width of ``Val``.
   3218 
   3219   * ``static ConstantInt* get(const Type *Ty, uint64_t Val)``: Returns the
   3220     ConstantInt object that represents the value provided by ``Val`` for integer
   3221     type ``Ty``.
   3222 
   3223 * ConstantFP : This class represents a floating point constant.
   3224 
   3225   * ``double getValue() const``: Returns the underlying value of this constant.
   3226 
   3227 * ConstantArray : This represents a constant array.
   3228 
   3229   * ``const std::vector<Use> &getValues() const``: Returns a vector of
   3230     component constants that makeup this array.
   3231 
   3232 * ConstantStruct : This represents a constant struct.
   3233 
   3234   * ``const std::vector<Use> &getValues() const``: Returns a vector of
   3235     component constants that makeup this array.
   3236 
   3237 * GlobalValue : This represents either a global variable or a function.  In
   3238   either case, the value is a constant fixed address (after linking).
   3239 
   3240 .. _GlobalValue:
   3241 
   3242 The ``GlobalValue`` class
   3243 -------------------------
   3244 
   3245 ``#include "llvm/IR/GlobalValue.h"``
   3246 
   3247 header source: `GlobalValue.h
   3248 <http://llvm.org/doxygen/GlobalValue_8h-source.html>`_
   3249 
   3250 doxygen info: `GlobalValue Class
   3251 <http://llvm.org/doxygen/classllvm_1_1GlobalValue.html>`_
   3252 
   3253 Superclasses: Constant_, User_, Value_
   3254 
   3255 Global values ( GlobalVariable_\ s or :ref:`Function <c_Function>`\ s) are the
   3256 only LLVM values that are visible in the bodies of all :ref:`Function
   3257 <c_Function>`\ s.  Because they are visible at global scope, they are also
   3258 subject to linking with other globals defined in different translation units.
   3259 To control the linking process, ``GlobalValue``\ s know their linkage rules.
   3260 Specifically, ``GlobalValue``\ s know whether they have internal or external
   3261 linkage, as defined by the ``LinkageTypes`` enumeration.
   3262 
   3263 If a ``GlobalValue`` has internal linkage (equivalent to being ``static`` in C),
   3264 it is not visible to code outside the current translation unit, and does not
   3265 participate in linking.  If it has external linkage, it is visible to external
   3266 code, and does participate in linking.  In addition to linkage information,
   3267 ``GlobalValue``\ s keep track of which Module_ they are currently part of.
   3268 
   3269 Because ``GlobalValue``\ s are memory objects, they are always referred to by
   3270 their **address**.  As such, the Type_ of a global is always a pointer to its
   3271 contents.  It is important to remember this when using the ``GetElementPtrInst``
   3272 instruction because this pointer must be dereferenced first.  For example, if
   3273 you have a ``GlobalVariable`` (a subclass of ``GlobalValue)`` that is an array
   3274 of 24 ints, type ``[24 x i32]``, then the ``GlobalVariable`` is a pointer to
   3275 that array.  Although the address of the first element of this array and the
   3276 value of the ``GlobalVariable`` are the same, they have different types.  The
   3277 ``GlobalVariable``'s type is ``[24 x i32]``.  The first element's type is
   3278 ``i32.`` Because of this, accessing a global value requires you to dereference
   3279 the pointer with ``GetElementPtrInst`` first, then its elements can be accessed.
   3280 This is explained in the `LLVM Language Reference Manual
   3281 <LangRef.html#globalvars>`_.
   3282 
   3283 .. _m_GlobalValue:
   3284 
   3285 Important Public Members of the ``GlobalValue`` class
   3286 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   3287 
   3288 * | ``bool hasInternalLinkage() const``
   3289   | ``bool hasExternalLinkage() const``
   3290   | ``void setInternalLinkage(bool HasInternalLinkage)``
   3291 
   3292   These methods manipulate the linkage characteristics of the ``GlobalValue``.
   3293 
   3294 * ``Module *getParent()``
   3295 
   3296   This returns the Module_ that the
   3297   GlobalValue is currently embedded into.
   3298 
   3299 .. _c_Function:
   3300 
   3301 The ``Function`` class
   3302 ----------------------
   3303 
   3304 ``#include "llvm/IR/Function.h"``
   3305 
   3306 header source: `Function.h <http://llvm.org/doxygen/Function_8h-source.html>`_
   3307 
   3308 doxygen info: `Function Class
   3309 <http://llvm.org/doxygen/classllvm_1_1Function.html>`_
   3310 
   3311 Superclasses: GlobalValue_, Constant_, User_, Value_
   3312 
   3313 The ``Function`` class represents a single procedure in LLVM.  It is actually
   3314 one of the more complex classes in the LLVM hierarchy because it must keep track
   3315 of a large amount of data.  The ``Function`` class keeps track of a list of
   3316 BasicBlock_\ s, a list of formal Argument_\ s, and a SymbolTable_.
   3317 
   3318 The list of BasicBlock_\ s is the most commonly used part of ``Function``
   3319 objects.  The list imposes an implicit ordering of the blocks in the function,
   3320 which indicate how the code will be laid out by the backend.  Additionally, the
   3321 first BasicBlock_ is the implicit entry node for the ``Function``.  It is not
   3322 legal in LLVM to explicitly branch to this initial block.  There are no implicit
   3323 exit nodes, and in fact there may be multiple exit nodes from a single
   3324 ``Function``.  If the BasicBlock_ list is empty, this indicates that the
   3325 ``Function`` is actually a function declaration: the actual body of the function
   3326 hasn't been linked in yet.
   3327 
   3328 In addition to a list of BasicBlock_\ s, the ``Function`` class also keeps track
   3329 of the list of formal Argument_\ s that the function receives.  This container
   3330 manages the lifetime of the Argument_ nodes, just like the BasicBlock_ list does
   3331 for the BasicBlock_\ s.
   3332 
   3333 The SymbolTable_ is a very rarely used LLVM feature that is only used when you
   3334 have to look up a value by name.  Aside from that, the SymbolTable_ is used
   3335 internally to make sure that there are not conflicts between the names of
   3336 Instruction_\ s, BasicBlock_\ s, or Argument_\ s in the function body.
   3337 
   3338 Note that ``Function`` is a GlobalValue_ and therefore also a Constant_.  The
   3339 value of the function is its address (after linking) which is guaranteed to be
   3340 constant.
   3341 
   3342 .. _m_Function:
   3343 
   3344 Important Public Members of the ``Function``
   3345 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   3346 
   3347 * ``Function(const FunctionType *Ty, LinkageTypes Linkage,
   3348   const std::string &N = "", Module* Parent = 0)``
   3349 
   3350   Constructor used when you need to create new ``Function``\ s to add the
   3351   program.  The constructor must specify the type of the function to create and
   3352   what type of linkage the function should have.  The FunctionType_ argument
   3353   specifies the formal arguments and return value for the function.  The same
   3354   FunctionType_ value can be used to create multiple functions.  The ``Parent``
   3355   argument specifies the Module in which the function is defined.  If this
   3356   argument is provided, the function will automatically be inserted into that
   3357   module's list of functions.
   3358 
   3359 * ``bool isDeclaration()``
   3360 
   3361   Return whether or not the ``Function`` has a body defined.  If the function is
   3362   "external", it does not have a body, and thus must be resolved by linking with
   3363   a function defined in a different translation unit.
   3364 
   3365 * | ``Function::iterator`` - Typedef for basic block list iterator
   3366   | ``Function::const_iterator`` - Typedef for const_iterator.
   3367   | ``begin()``, ``end()``, ``size()``, ``empty()``
   3368 
   3369   These are forwarding methods that make it easy to access the contents of a
   3370   ``Function`` object's BasicBlock_ list.
   3371 
   3372 * ``Function::BasicBlockListType &getBasicBlockList()``
   3373 
   3374   Returns the list of BasicBlock_\ s.  This is necessary to use when you need to
   3375   update the list or perform a complex action that doesn't have a forwarding
   3376   method.
   3377 
   3378 * | ``Function::arg_iterator`` - Typedef for the argument list iterator
   3379   | ``Function::const_arg_iterator`` - Typedef for const_iterator.
   3380   | ``arg_begin()``, ``arg_end()``, ``arg_size()``, ``arg_empty()``
   3381 
   3382   These are forwarding methods that make it easy to access the contents of a
   3383   ``Function`` object's Argument_ list.
   3384 
   3385 * ``Function::ArgumentListType &getArgumentList()``
   3386 
   3387   Returns the list of Argument_.  This is necessary to use when you need to
   3388   update the list or perform a complex action that doesn't have a forwarding
   3389   method.
   3390 
   3391 * ``BasicBlock &getEntryBlock()``
   3392 
   3393   Returns the entry ``BasicBlock`` for the function.  Because the entry block
   3394   for the function is always the first block, this returns the first block of
   3395   the ``Function``.
   3396 
   3397 * | ``Type *getReturnType()``
   3398   | ``FunctionType *getFunctionType()``
   3399 
   3400   This traverses the Type_ of the ``Function`` and returns the return type of
   3401   the function, or the FunctionType_ of the actual function.
   3402 
   3403 * ``SymbolTable *getSymbolTable()``
   3404 
   3405   Return a pointer to the SymbolTable_ for this ``Function``.
   3406 
   3407 .. _GlobalVariable:
   3408 
   3409 The ``GlobalVariable`` class
   3410 ----------------------------
   3411 
   3412 ``#include "llvm/IR/GlobalVariable.h"``
   3413 
   3414 header source: `GlobalVariable.h
   3415 <http://llvm.org/doxygen/GlobalVariable_8h-source.html>`_
   3416 
   3417 doxygen info: `GlobalVariable Class
   3418 <http://llvm.org/doxygen/classllvm_1_1GlobalVariable.html>`_
   3419 
   3420 Superclasses: GlobalValue_, Constant_, User_, Value_
   3421 
   3422 Global variables are represented with the (surprise surprise) ``GlobalVariable``
   3423 class.  Like functions, ``GlobalVariable``\ s are also subclasses of
   3424 GlobalValue_, and as such are always referenced by their address (global values
   3425 must live in memory, so their "name" refers to their constant address).  See
   3426 GlobalValue_ for more on this.  Global variables may have an initial value
   3427 (which must be a Constant_), and if they have an initializer, they may be marked
   3428 as "constant" themselves (indicating that their contents never change at
   3429 runtime).
   3430 
   3431 .. _m_GlobalVariable:
   3432 
   3433 Important Public Members of the ``GlobalVariable`` class
   3434 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   3435 
   3436 * ``GlobalVariable(const Type *Ty, bool isConstant, LinkageTypes &Linkage,
   3437   Constant *Initializer = 0, const std::string &Name = "", Module* Parent = 0)``
   3438 
   3439   Create a new global variable of the specified type.  If ``isConstant`` is true
   3440   then the global variable will be marked as unchanging for the program.  The
   3441   Linkage parameter specifies the type of linkage (internal, external, weak,
   3442   linkonce, appending) for the variable.  If the linkage is InternalLinkage,
   3443   WeakAnyLinkage, WeakODRLinkage, LinkOnceAnyLinkage or LinkOnceODRLinkage, then
   3444   the resultant global variable will have internal linkage.  AppendingLinkage
   3445   concatenates together all instances (in different translation units) of the
   3446   variable into a single variable but is only applicable to arrays.  See the
   3447   `LLVM Language Reference <LangRef.html#modulestructure>`_ for further details
   3448   on linkage types.  Optionally an initializer, a name, and the module to put
   3449   the variable into may be specified for the global variable as well.
   3450 
   3451 * ``bool isConstant() const``
   3452 
   3453   Returns true if this is a global variable that is known not to be modified at
   3454   runtime.
   3455 
   3456 * ``bool hasInitializer()``
   3457 
   3458   Returns true if this ``GlobalVariable`` has an intializer.
   3459 
   3460 * ``Constant *getInitializer()``
   3461 
   3462   Returns the initial value for a ``GlobalVariable``.  It is not legal to call
   3463   this method if there is no initializer.
   3464 
   3465 .. _BasicBlock:
   3466 
   3467 The ``BasicBlock`` class
   3468 ------------------------
   3469 
   3470 ``#include "llvm/IR/BasicBlock.h"``
   3471 
   3472 header source: `BasicBlock.h
   3473 <http://llvm.org/doxygen/BasicBlock_8h-source.html>`_
   3474 
   3475 doxygen info: `BasicBlock Class
   3476 <http://llvm.org/doxygen/classllvm_1_1BasicBlock.html>`_
   3477 
   3478 Superclass: Value_
   3479 
   3480 This class represents a single entry single exit section of the code, commonly
   3481 known as a basic block by the compiler community.  The ``BasicBlock`` class
   3482 maintains a list of Instruction_\ s, which form the body of the block.  Matching
   3483 the language definition, the last element of this list of instructions is always
   3484 a terminator instruction (a subclass of the TerminatorInst_ class).
   3485 
   3486 In addition to tracking the list of instructions that make up the block, the
   3487 ``BasicBlock`` class also keeps track of the :ref:`Function <c_Function>` that
   3488 it is embedded into.
   3489 
   3490 Note that ``BasicBlock``\ s themselves are Value_\ s, because they are
   3491 referenced by instructions like branches and can go in the switch tables.
   3492 ``BasicBlock``\ s have type ``label``.
   3493 
   3494 .. _m_BasicBlock:
   3495 
   3496 Important Public Members of the ``BasicBlock`` class
   3497 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   3498 
   3499 * ``BasicBlock(const std::string &Name = "", Function *Parent = 0)``
   3500 
   3501   The ``BasicBlock`` constructor is used to create new basic blocks for
   3502   insertion into a function.  The constructor optionally takes a name for the
   3503   new block, and a :ref:`Function <c_Function>` to insert it into.  If the
   3504   ``Parent`` parameter is specified, the new ``BasicBlock`` is automatically
   3505   inserted at the end of the specified :ref:`Function <c_Function>`, if not
   3506   specified, the BasicBlock must be manually inserted into the :ref:`Function
   3507   <c_Function>`.
   3508 
   3509 * | ``BasicBlock::iterator`` - Typedef for instruction list iterator
   3510   | ``BasicBlock::const_iterator`` - Typedef for const_iterator.
   3511   | ``begin()``, ``end()``, ``front()``, ``back()``,
   3512     ``size()``, ``empty()``
   3513     STL-style functions for accessing the instruction list.
   3514 
   3515   These methods and typedefs are forwarding functions that have the same
   3516   semantics as the standard library methods of the same names.  These methods
   3517   expose the underlying instruction list of a basic block in a way that is easy
   3518   to manipulate.  To get the full complement of container operations (including
   3519   operations to update the list), you must use the ``getInstList()`` method.
   3520 
   3521 * ``BasicBlock::InstListType &getInstList()``
   3522 
   3523   This method is used to get access to the underlying container that actually
   3524   holds the Instructions.  This method must be used when there isn't a
   3525   forwarding function in the ``BasicBlock`` class for the operation that you
   3526   would like to perform.  Because there are no forwarding functions for
   3527   "updating" operations, you need to use this if you want to update the contents
   3528   of a ``BasicBlock``.
   3529 
   3530 * ``Function *getParent()``
   3531 
   3532   Returns a pointer to :ref:`Function <c_Function>` the block is embedded into,
   3533   or a null pointer if it is homeless.
   3534 
   3535 * ``TerminatorInst *getTerminator()``
   3536 
   3537   Returns a pointer to the terminator instruction that appears at the end of the
   3538   ``BasicBlock``.  If there is no terminator instruction, or if the last
   3539   instruction in the block is not a terminator, then a null pointer is returned.
   3540 
   3541 .. _Argument:
   3542 
   3543 The ``Argument`` class
   3544 ----------------------
   3545 
   3546 This subclass of Value defines the interface for incoming formal arguments to a
   3547 function.  A Function maintains a list of its formal arguments.  An argument has
   3548 a pointer to the parent Function.
   3549 
   3550 
   3551