1======================== 2LLVM Programmer's Manual 3======================== 4 5.. contents:: 6 :local: 7 8.. warning:: 9 This is always a work in progress. 10 11.. _introduction: 12 13Introduction 14============ 15 16This document is meant to highlight some of the important classes and interfaces 17available in the LLVM source-base. This manual is not intended to explain what 18LLVM is, how it works, and what LLVM code looks like. It assumes that you know 19the basics of LLVM and are interested in writing transformations or otherwise 20analyzing or manipulating the code. 21 22This document should get you oriented so that you can find your way in the 23continuously growing source code that makes up the LLVM infrastructure. Note 24that this manual is not intended to serve as a replacement for reading the 25source code, so if you think there should be a method in one of these classes to 26do 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 28possible. 29 30The first section of this document describes general information that is useful 31to know when working in the LLVM infrastructure, and the second describes the 32Core LLVM classes. In the future this manual will be extended with information 33describing how to use extension libraries, such as dominator information, CFG 34traversal routines, and useful utilities like the ``InstVisitor`` (`doxygen 35<http://llvm.org/doxygen/InstVisitor_8h-source.html>`__) template. 36 37.. _general: 38 39General Information 40=================== 41 42This section contains general information that is useful if you are working in 43the LLVM source-base, but that isn't specific to any particular API. 44 45.. _stl: 46 47The C++ Standard Template Library 48--------------------------------- 49 50LLVM makes heavy use of the C++ Standard Template Library (STL), perhaps much 51more than you are used to, or have seen before. Because of this, you might want 52to do a little background reading in the techniques used and capabilities of the 53library. There are many good pages that discuss the STL, and several books on 54the subject that you can get, so it will not be discussed in this document. 55 56Here 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 80You 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 82than where to put your curly braces. 83 84.. _resources: 85 86Other 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 94Important and useful LLVM APIs 95============================== 96 97Here we highlight some LLVM APIs that are generally useful and good to know 98about when writing transformations. 99 100.. _isa: 101 102The ``isa<>``, ``cast<>`` and ``dyn_cast<>`` templates 103------------------------------------------------------ 104 105The LLVM source-base makes extensive use of a custom form of RTTI. These 106templates have many similarities to the C++ ``dynamic_cast<>`` operator, but 107they 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 109used so often, you must know what they do and how they work. All of these 110templates are defined in the ``llvm/Support/Casting.h`` (`doxygen 111<http://llvm.org/doxygen/Casting_8h-source.html>`__) file (note that you very 112rarely 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 179These five templates can be used with any classes, whether they have a v-table 180or 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 186Passing strings (the ``StringRef`` and ``Twine`` classes) 187--------------------------------------------------------- 188 189Although LLVM generally does not do much string manipulation, we do have several 190important APIs which take strings. Two important examples are the Value class 191-- which has names for instructions, functions, etc. -- and the ``StringMap`` 192class which is used extensively in LLVM and Clang. 193 194These are generic classes, and they need to be able to accept strings which may 195have embedded null characters. Therefore, they cannot simply take a ``const 196char *``, and taking a ``const std::string&`` requires clients to perform a heap 197allocation which is usually unnecessary. Instead, many LLVM APIs use a 198``StringRef`` or a ``const Twine&`` for passing strings efficiently. 199 200.. _StringRef: 201 202The ``StringRef`` class 203^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 204 205The ``StringRef`` data type represents a reference to a constant string (a 206character array and a length) and supports the common operations available on 207``std::string``, but does not require heap allocation. 208 209It can be implicitly constructed using a C style null-terminated string, an 210``std::string``, or explicitly with a character pointer and length. For 211example, the ``StringRef`` find function is declared as: 212 213.. code-block:: c++ 214 215 iterator find(StringRef Key); 216 217and 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 225Similarly, APIs which need to return a string may return a ``StringRef`` 226instance, which can be used directly or converted to an ``std::string`` using 227the ``str`` member function. See ``llvm/ADT/StringRef.h`` (`doxygen 228<http://llvm.org/doxygen/classllvm_1_1StringRef_8h-source.html>`__) for more 229information. 230 231You should rarely use the ``StringRef`` class directly, because it contains 232pointers to external memory it is not generally safe to store an instance of the 233class (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 235passed by value. 236 237The ``Twine`` class 238^^^^^^^^^^^^^^^^^^^ 239 240The ``Twine`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1Twine.html>`__) 241class is an efficient way for APIs to accept concatenated strings. For example, 242a common LLVM paradigm is to name one instruction based on the name of another 243instruction with a suffix, for example: 244 245.. code-block:: c++ 246 247 New = CmpInst::Create(..., SO->getName() + ".cmp"); 248 249The ``Twine`` class is effectively a lightweight `rope 250<http://en.wikipedia.org/wiki/Rope_(computer_science)>`_ which points to 251temporary (stack allocated) objects. Twines can be implicitly constructed as 252the 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 254of strings until it is actually required, at which point it can be efficiently 255rendered directly into a character array. This avoids unnecessary heap 256allocation involved in constructing the temporary results of string 257concatenation. See ``llvm/ADT/Twine.h`` (`doxygen 258<http://llvm.org/doxygen/Twine_8h_source.html>`__) and :ref:`here <dss_twine>` 259for more information. 260 261As with a ``StringRef``, ``Twine`` objects point to external memory and should 262almost never be stored or mentioned directly. They are intended solely for use 263when defining a function which should be able to efficiently accept concatenated 264strings. 265 266.. _function_apis: 267 268Passing functions and other callable objects 269-------------------------------------------- 270 271Sometimes you may want a function to be passed a callback object. In order to 272support lambda expressions and other function objects, you should not use the 273traditional C approach of taking a function pointer and an opaque cookie: 274 275.. code-block:: c++ 276 277 void takeCallback(bool (*Callback)(Function *, void *), void *Cookie); 278 279Instead, use one of the following approaches: 280 281Function template 282^^^^^^^^^^^^^^^^^ 283 284If you don't mind putting the definition of your function into a header file, 285make it a function template that is templated on the callable type. 286 287.. code-block:: c++ 288 289 template<typename Callable> 290 void takeCallback(Callable Callback) { 291 Callback(1, 2, 3); 292 } 293 294The ``function_ref`` class template 295^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 296 297The ``function_ref`` 298(`doxygen <http://llvm.org/doxygen/classllvm_1_1function_ref.html>`__) class 299template represents a reference to a callable object, templated over the type 300of the callable. This is a good choice for passing a callback to a function, 301if you don't need to hold onto the callback after the function returns. In this 302way, ``function_ref`` is to ``std::function`` as ``StringRef`` is to 303``std::string``. 304 305``function_ref<Ret(Param1, Param2, ...)>`` can be implicitly constructed from 306any callable object that can be called with arguments of type ``Param1``, 307``Param2``, ..., and returns a value that can be converted to type ``Ret``. 308For example: 309 310.. code-block:: c++ 311 312 void visitBasicBlocks(Function *F, function_ref<bool (BasicBlock*)> Callback) { 313 for (BasicBlock &BB : *F) 314 if (Callback(&BB)) 315 return; 316 } 317 318can be called using: 319 320.. code-block:: c++ 321 322 visitBasicBlocks(F, [&](BasicBlock *BB) { 323 if (process(BB)) 324 return isEmpty(BB); 325 return false; 326 }); 327 328Note that a ``function_ref`` object contains pointers to external memory, so it 329is not generally safe to store an instance of the class (unless you know that 330the external storage will not be freed). If you need this ability, consider 331using ``std::function``. ``function_ref`` is small enough that it should always 332be passed by value. 333 334.. _DEBUG: 335 336The ``DEBUG()`` macro and ``-debug`` option 337------------------------------------------- 338 339Often when working on your pass you will put a bunch of debugging printouts and 340other code into your pass. After you get it working, you want to remove it, but 341you may need it again in the future (to work out new bugs that you run across). 342 343Naturally, because of this, you don't want to delete the debug printouts, but 344you don't want them to always be noisy. A standard compromise is to comment 345them out, allowing you to enable them if you need them in the future. 346 347The ``llvm/Support/Debug.h`` (`doxygen 348<http://llvm.org/doxygen/Debug_8h-source.html>`__) file provides a macro named 349``DEBUG()`` that is a much nicer solution to this problem. Basically, you can 350put arbitrary code into the argument of the ``DEBUG`` macro, and it is only 351executed if '``opt``' (or any other tool) is run with the '``-debug``' command 352line argument: 353 354.. code-block:: c++ 355 356 DEBUG(errs() << "I am here!\n"); 357 358Then you can run your pass like this: 359 360.. code-block:: none 361 362 $ opt < a.bc > /dev/null -mypass 363 <no output> 364 $ opt < a.bc > /dev/null -mypass -debug 365 I am here! 366 367Using the ``DEBUG()`` macro instead of a home-brewed solution allows you to not 368have to create "yet another" command line option for the debug output for your 369pass. Note that ``DEBUG()`` macros are disabled for optimized builds, so they 370do not cause a performance impact at all (for the same reason, they should also 371not contain side-effects!). 372 373One additional nice thing about the ``DEBUG()`` macro is that you can enable or 374disable it directly in gdb. Just use "``set DebugFlag=0``" or "``set 375DebugFlag=1``" from the gdb if the program is running. If the program hasn't 376been started yet, you can always just run it with ``-debug``. 377 378.. _DEBUG_TYPE: 379 380Fine grained debug info with ``DEBUG_TYPE`` and the ``-debug-only`` option 381^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 382 383Sometimes you may find yourself in a situation where enabling ``-debug`` just 384turns on **too much** information (such as when working on the code generator). 385If you want to enable debug information with more fine-grained control, you 386can define the ``DEBUG_TYPE`` macro and use the ``-debug-only`` option as 387follows: 388 389.. code-block:: c++ 390 391 #undef DEBUG_TYPE 392 DEBUG(errs() << "No debug type\n"); 393 #define DEBUG_TYPE "foo" 394 DEBUG(errs() << "'foo' debug type\n"); 395 #undef DEBUG_TYPE 396 #define DEBUG_TYPE "bar" 397 DEBUG(errs() << "'bar' debug type\n")); 398 #undef DEBUG_TYPE 399 #define DEBUG_TYPE "" 400 DEBUG(errs() << "No debug type (2)\n"); 401 402Then you can run your pass like this: 403 404.. code-block:: none 405 406 $ opt < a.bc > /dev/null -mypass 407 <no output> 408 $ opt < a.bc > /dev/null -mypass -debug 409 No debug type 410 'foo' debug type 411 'bar' debug type 412 No debug type (2) 413 $ opt < a.bc > /dev/null -mypass -debug-only=foo 414 'foo' debug type 415 $ opt < a.bc > /dev/null -mypass -debug-only=bar 416 'bar' debug type 417 418Of course, in practice, you should only set ``DEBUG_TYPE`` at the top of a file, 419to specify the debug type for the entire module (if you do this before you 420``#include "llvm/Support/Debug.h"``, you don't have to insert the ugly 421``#undef``'s). Also, you should use names more meaningful than "foo" and "bar", 422because there is no system in place to ensure that names do not conflict. If 423two different modules use the same string, they will all be turned on when the 424name is specified. This allows, for example, all debug information for 425instruction scheduling to be enabled with ``-debug-only=InstrSched``, even if 426the source lives in multiple files. 427 428For performance reasons, -debug-only is not available in optimized build 429(``--enable-optimized``) of LLVM. 430 431The ``DEBUG_WITH_TYPE`` macro is also available for situations where you would 432like to set ``DEBUG_TYPE``, but only for one specific ``DEBUG`` statement. It 433takes an additional first parameter, which is the type to use. For example, the 434preceding example could be written as: 435 436.. code-block:: c++ 437 438 DEBUG_WITH_TYPE("", errs() << "No debug type\n"); 439 DEBUG_WITH_TYPE("foo", errs() << "'foo' debug type\n"); 440 DEBUG_WITH_TYPE("bar", errs() << "'bar' debug type\n")); 441 DEBUG_WITH_TYPE("", errs() << "No debug type (2)\n"); 442 443.. _Statistic: 444 445The ``Statistic`` class & ``-stats`` option 446------------------------------------------- 447 448The ``llvm/ADT/Statistic.h`` (`doxygen 449<http://llvm.org/doxygen/Statistic_8h-source.html>`__) file provides a class 450named ``Statistic`` that is used as a unified way to keep track of what the LLVM 451compiler is doing and how effective various optimizations are. It is useful to 452see what optimizations are contributing to making a particular program run 453faster. 454 455Often you may run your pass on some big program, and you're interested to see 456how many times it makes a certain transformation. Although you can do this with 457hand inspection, or some ad-hoc method, this is a real pain and not very useful 458for big programs. Using the ``Statistic`` class makes it very easy to keep 459track of this information, and the calculated information is presented in a 460uniform manner with the rest of the passes being executed. 461 462There are many examples of ``Statistic`` uses, but the basics of using it are as 463follows: 464 465#. Define your statistic like this: 466 467 .. code-block:: c++ 468 469 #define DEBUG_TYPE "mypassname" // This goes before any #includes. 470 STATISTIC(NumXForms, "The # of times I did stuff"); 471 472 The ``STATISTIC`` macro defines a static variable, whose name is specified by 473 the first argument. The pass name is taken from the ``DEBUG_TYPE`` macro, and 474 the description is taken from the second argument. The variable defined 475 ("NumXForms" in this case) acts like an unsigned integer. 476 477#. Whenever you make a transformation, bump the counter: 478 479 .. code-block:: c++ 480 481 ++NumXForms; // I did stuff! 482 483That's all you have to do. To get '``opt``' to print out the statistics 484gathered, use the '``-stats``' option: 485 486.. code-block:: none 487 488 $ opt -stats -mypassname < program.bc > /dev/null 489 ... statistics output ... 490 491When running ``opt`` on a C file from the SPEC benchmark suite, it gives a 492report that looks like this: 493 494.. code-block:: none 495 496 7646 bitcodewriter - Number of normal instructions 497 725 bitcodewriter - Number of oversized instructions 498 129996 bitcodewriter - Number of bitcode bytes written 499 2817 raise - Number of insts DCEd or constprop'd 500 3213 raise - Number of cast-of-self removed 501 5046 raise - Number of expression trees converted 502 75 raise - Number of other getelementptr's formed 503 138 raise - Number of load/store peepholes 504 42 deadtypeelim - Number of unused typenames removed from symtab 505 392 funcresolve - Number of varargs functions resolved 506 27 globaldce - Number of global variables removed 507 2 adce - Number of basic blocks removed 508 134 cee - Number of branches revectored 509 49 cee - Number of setcc instruction eliminated 510 532 gcse - Number of loads removed 511 2919 gcse - Number of instructions removed 512 86 indvars - Number of canonical indvars added 513 87 indvars - Number of aux indvars removed 514 25 instcombine - Number of dead inst eliminate 515 434 instcombine - Number of insts combined 516 248 licm - Number of load insts hoisted 517 1298 licm - Number of insts hoisted to a loop pre-header 518 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header) 519 75 mem2reg - Number of alloca's promoted 520 1444 cfgsimplify - Number of blocks simplified 521 522Obviously, with so many optimizations, having a unified framework for this stuff 523is very nice. Making your pass fit well into the framework makes it more 524maintainable and useful. 525 526.. _ViewGraph: 527 528Viewing graphs while debugging code 529----------------------------------- 530 531Several of the important data structures in LLVM are graphs: for example CFGs 532made out of LLVM :ref:`BasicBlocks <BasicBlock>`, CFGs made out of LLVM 533:ref:`MachineBasicBlocks <MachineBasicBlock>`, and :ref:`Instruction Selection 534DAGs <SelectionDAG>`. In many cases, while debugging various parts of the 535compiler, it is nice to instantly visualize these graphs. 536 537LLVM provides several callbacks that are available in a debug build to do 538exactly that. If you call the ``Function::viewCFG()`` method, for example, the 539current LLVM tool will pop up a window containing the CFG for the function where 540each basic block is a node in the graph, and each node contains the instructions 541in the block. Similarly, there also exists ``Function::viewCFGOnly()`` (does 542not include the instructions), the ``MachineFunction::viewCFG()`` and 543``MachineFunction::viewCFGOnly()``, and the ``SelectionDAG::viewGraph()`` 544methods. Within GDB, for example, you can usually use something like ``call 545DAG.viewGraph()`` to pop up a window. Alternatively, you can sprinkle calls to 546these functions in your code in places you want to debug. 547 548Getting this to work requires a small amount of setup. On Unix systems 549with X11, install the `graphviz <http://www.graphviz.org>`_ toolkit, and make 550sure 'dot' and 'gv' are in your path. If you are running on Mac OS X, download 551and install the Mac OS X `Graphviz program 552<http://www.pixelglow.com/graphviz/>`_ and add 553``/Applications/Graphviz.app/Contents/MacOS/`` (or wherever you install it) to 554your path. The programs need not be present when configuring, building or 555running LLVM and can simply be installed when needed during an active debug 556session. 557 558``SelectionDAG`` has been extended to make it easier to locate *interesting* 559nodes in large complex graphs. From gdb, if you ``call DAG.setGraphColor(node, 560"color")``, then the next ``call DAG.viewGraph()`` would highlight the node in 561the specified color (choices of colors can be found at `colors 562<http://www.graphviz.org/doc/info/colors.html>`_.) More complex node attributes 563can be provided with ``call DAG.setGraphAttrs(node, "attributes")`` (choices can 564be found at `Graph attributes <http://www.graphviz.org/doc/info/attrs.html>`_.) 565If you want to restart and clear all the current graph attributes, then you can 566``call DAG.clearGraphAttrs()``. 567 568Note that graph visualization features are compiled out of Release builds to 569reduce file size. This means that you need a Debug+Asserts or Release+Asserts 570build to use these features. 571 572.. _datastructure: 573 574Picking the Right Data Structure for a Task 575=========================================== 576 577LLVM has a plethora of data structures in the ``llvm/ADT/`` directory, and we 578commonly use STL data structures. This section describes the trade-offs you 579should consider when you pick one. 580 581The first step is a choose your own adventure: do you want a sequential 582container, a set-like container, or a map-like container? The most important 583thing when choosing a container is the algorithmic properties of how you plan to 584access the container. Based on that, you should use: 585 586 587* a :ref:`map-like <ds_map>` container if you need efficient look-up of a 588 value based on another value. Map-like containers also support efficient 589 queries for containment (whether a key is in the map). Map-like containers 590 generally do not support efficient reverse mapping (values to keys). If you 591 need that, use two maps. Some map-like containers also support efficient 592 iteration through the keys in sorted order. Map-like containers are the most 593 expensive sort, only use them if you need one of these capabilities. 594 595* a :ref:`set-like <ds_set>` container if you need to put a bunch of stuff into 596 a container that automatically eliminates duplicates. Some set-like 597 containers support efficient iteration through the elements in sorted order. 598 Set-like containers are more expensive than sequential containers. 599 600* a :ref:`sequential <ds_sequential>` container provides the most efficient way 601 to add elements and keeps track of the order they are added to the collection. 602 They permit duplicates and support efficient iteration, but do not support 603 efficient look-up based on a key. 604 605* a :ref:`string <ds_string>` container is a specialized sequential container or 606 reference structure that is used for character or byte arrays. 607 608* a :ref:`bit <ds_bit>` container provides an efficient way to store and 609 perform set operations on sets of numeric id's, while automatically 610 eliminating duplicates. Bit containers require a maximum of 1 bit for each 611 identifier you want to store. 612 613Once the proper category of container is determined, you can fine tune the 614memory use, constant factors, and cache behaviors of access by intelligently 615picking a member of the category. Note that constant factors and cache behavior 616can be a big deal. If you have a vector that usually only contains a few 617elements (but could contain many), for example, it's much better to use 618:ref:`SmallVector <dss_smallvector>` than :ref:`vector <dss_vector>`. Doing so 619avoids (relatively) expensive malloc/free calls, which dwarf the cost of adding 620the elements to the container. 621 622.. _ds_sequential: 623 624Sequential Containers (std::vector, std::list, etc) 625--------------------------------------------------- 626 627There are a variety of sequential containers available for you, based on your 628needs. Pick the first in this section that will do what you want. 629 630.. _dss_arrayref: 631 632llvm/ADT/ArrayRef.h 633^^^^^^^^^^^^^^^^^^^ 634 635The ``llvm::ArrayRef`` class is the preferred class to use in an interface that 636accepts a sequential list of elements in memory and just reads from them. By 637taking an ``ArrayRef``, the API can be passed a fixed size array, an 638``std::vector``, an ``llvm::SmallVector`` and anything else that is contiguous 639in memory. 640 641.. _dss_fixedarrays: 642 643Fixed Size Arrays 644^^^^^^^^^^^^^^^^^ 645 646Fixed size arrays are very simple and very fast. They are good if you know 647exactly how many elements you have, or you have a (low) upper bound on how many 648you have. 649 650.. _dss_heaparrays: 651 652Heap Allocated Arrays 653^^^^^^^^^^^^^^^^^^^^^ 654 655Heap allocated arrays (``new[]`` + ``delete[]``) are also simple. They are good 656if the number of elements is variable, if you know how many elements you will 657need before the array is allocated, and if the array is usually large (if not, 658consider a :ref:`SmallVector <dss_smallvector>`). The cost of a heap allocated 659array is the cost of the new/delete (aka malloc/free). Also note that if you 660are allocating an array of a type with a constructor, the constructor and 661destructors will be run for every element in the array (re-sizable vectors only 662construct those elements actually used). 663 664.. _dss_tinyptrvector: 665 666llvm/ADT/TinyPtrVector.h 667^^^^^^^^^^^^^^^^^^^^^^^^ 668 669``TinyPtrVector<Type>`` is a highly specialized collection class that is 670optimized to avoid allocation in the case when a vector has zero or one 671elements. It has two major restrictions: 1) it can only hold values of pointer 672type, and 2) it cannot hold a null pointer. 673 674Since this container is highly specialized, it is rarely used. 675 676.. _dss_smallvector: 677 678llvm/ADT/SmallVector.h 679^^^^^^^^^^^^^^^^^^^^^^ 680 681``SmallVector<Type, N>`` is a simple class that looks and smells just like 682``vector<Type>``: it supports efficient iteration, lays out elements in memory 683order (so you can do pointer arithmetic between elements), supports efficient 684push_back/pop_back operations, supports efficient random access to its elements, 685etc. 686 687The advantage of SmallVector is that it allocates space for some number of 688elements (N) **in the object itself**. Because of this, if the SmallVector is 689dynamically smaller than N, no malloc is performed. This can be a big win in 690cases where the malloc/free call is far more expensive than the code that 691fiddles around with the elements. 692 693This is good for vectors that are "usually small" (e.g. the number of 694predecessors/successors of a block is usually less than 8). On the other hand, 695this makes the size of the SmallVector itself large, so you don't want to 696allocate lots of them (doing so will waste a lot of space). As such, 697SmallVectors are most useful when on the stack. 698 699SmallVector also provides a nice portable and efficient replacement for 700``alloca``. 701 702.. note:: 703 704 Prefer to use ``SmallVectorImpl<T>`` as a parameter type. 705 706 In APIs that don't care about the "small size" (most?), prefer to use 707 the ``SmallVectorImpl<T>`` class, which is basically just the "vector 708 header" (and methods) without the elements allocated after it. Note that 709 ``SmallVector<T, N>`` inherits from ``SmallVectorImpl<T>`` so the 710 conversion is implicit and costs nothing. E.g. 711 712 .. code-block:: c++ 713 714 // BAD: Clients cannot pass e.g. SmallVector<Foo, 4>. 715 hardcodedSmallSize(SmallVector<Foo, 2> &Out); 716 // GOOD: Clients can pass any SmallVector<Foo, N>. 717 allowsAnySmallSize(SmallVectorImpl<Foo> &Out); 718 719 void someFunc() { 720 SmallVector<Foo, 8> Vec; 721 hardcodedSmallSize(Vec); // Error. 722 allowsAnySmallSize(Vec); // Works. 723 } 724 725 Even though it has "``Impl``" in the name, this is so widely used that 726 it really isn't "private to the implementation" anymore. A name like 727 ``SmallVectorHeader`` would be more appropriate. 728 729.. _dss_vector: 730 731<vector> 732^^^^^^^^ 733 734``std::vector`` is well loved and respected. It is useful when SmallVector 735isn't: when the size of the vector is often large (thus the small optimization 736will rarely be a benefit) or if you will be allocating many instances of the 737vector itself (which would waste space for elements that aren't in the 738container). vector is also useful when interfacing with code that expects 739vectors :). 740 741One worthwhile note about std::vector: avoid code like this: 742 743.. code-block:: c++ 744 745 for ( ... ) { 746 std::vector<foo> V; 747 // make use of V. 748 } 749 750Instead, write this as: 751 752.. code-block:: c++ 753 754 std::vector<foo> V; 755 for ( ... ) { 756 // make use of V. 757 V.clear(); 758 } 759 760Doing so will save (at least) one heap allocation and free per iteration of the 761loop. 762 763.. _dss_deque: 764 765<deque> 766^^^^^^^ 767 768``std::deque`` is, in some senses, a generalized version of ``std::vector``. 769Like ``std::vector``, it provides constant time random access and other similar 770properties, but it also provides efficient access to the front of the list. It 771does not guarantee continuity of elements within memory. 772 773In exchange for this extra flexibility, ``std::deque`` has significantly higher 774constant factor costs than ``std::vector``. If possible, use ``std::vector`` or 775something cheaper. 776 777.. _dss_list: 778 779<list> 780^^^^^^ 781 782``std::list`` is an extremely inefficient class that is rarely useful. It 783performs a heap allocation for every element inserted into it, thus having an 784extremely high constant factor, particularly for small data types. 785``std::list`` also only supports bidirectional iteration, not random access 786iteration. 787 788In exchange for this high cost, std::list supports efficient access to both ends 789of the list (like ``std::deque``, but unlike ``std::vector`` or 790``SmallVector``). In addition, the iterator invalidation characteristics of 791std::list are stronger than that of a vector class: inserting or removing an 792element into the list does not invalidate iterator or pointers to other elements 793in the list. 794 795.. _dss_ilist: 796 797llvm/ADT/ilist.h 798^^^^^^^^^^^^^^^^ 799 800``ilist<T>`` implements an 'intrusive' doubly-linked list. It is intrusive, 801because it requires the element to store and provide access to the prev/next 802pointers for the list. 803 804``ilist`` has the same drawbacks as ``std::list``, and additionally requires an 805``ilist_traits`` implementation for the element type, but it provides some novel 806characteristics. In particular, it can efficiently store polymorphic objects, 807the traits class is informed when an element is inserted or removed from the 808list, and ``ilist``\ s are guaranteed to support a constant-time splice 809operation. 810 811These properties are exactly what we want for things like ``Instruction``\ s and 812basic blocks, which is why these are implemented with ``ilist``\ s. 813 814Related classes of interest are explained in the following subsections: 815 816* :ref:`ilist_traits <dss_ilist_traits>` 817 818* :ref:`iplist <dss_iplist>` 819 820* :ref:`llvm/ADT/ilist_node.h <dss_ilist_node>` 821 822* :ref:`Sentinels <dss_ilist_sentinel>` 823 824.. _dss_packedvector: 825 826llvm/ADT/PackedVector.h 827^^^^^^^^^^^^^^^^^^^^^^^ 828 829Useful for storing a vector of values using only a few number of bits for each 830value. Apart from the standard operations of a vector-like container, it can 831also perform an 'or' set operation. 832 833For example: 834 835.. code-block:: c++ 836 837 enum State { 838 None = 0x0, 839 FirstCondition = 0x1, 840 SecondCondition = 0x2, 841 Both = 0x3 842 }; 843 844 State get() { 845 PackedVector<State, 2> Vec1; 846 Vec1.push_back(FirstCondition); 847 848 PackedVector<State, 2> Vec2; 849 Vec2.push_back(SecondCondition); 850 851 Vec1 |= Vec2; 852 return Vec1[0]; // returns 'Both'. 853 } 854 855.. _dss_ilist_traits: 856 857ilist_traits 858^^^^^^^^^^^^ 859 860``ilist_traits<T>`` is ``ilist<T>``'s customization mechanism. ``iplist<T>`` 861(and consequently ``ilist<T>``) publicly derive from this traits class. 862 863.. _dss_iplist: 864 865iplist 866^^^^^^ 867 868``iplist<T>`` is ``ilist<T>``'s base and as such supports a slightly narrower 869interface. Notably, inserters from ``T&`` are absent. 870 871``ilist_traits<T>`` is a public base of this class and can be used for a wide 872variety of customizations. 873 874.. _dss_ilist_node: 875 876llvm/ADT/ilist_node.h 877^^^^^^^^^^^^^^^^^^^^^ 878 879``ilist_node<T>`` implements the forward and backward links that are expected 880by the ``ilist<T>`` (and analogous containers) in the default manner. 881 882``ilist_node<T>``\ s are meant to be embedded in the node type ``T``, usually 883``T`` publicly derives from ``ilist_node<T>``. 884 885.. _dss_ilist_sentinel: 886 887Sentinels 888^^^^^^^^^ 889 890``ilist``\ s have another specialty that must be considered. To be a good 891citizen in the C++ ecosystem, it needs to support the standard container 892operations, such as ``begin`` and ``end`` iterators, etc. Also, the 893``operator--`` must work correctly on the ``end`` iterator in the case of 894non-empty ``ilist``\ s. 895 896The only sensible solution to this problem is to allocate a so-called *sentinel* 897along with the intrusive list, which serves as the ``end`` iterator, providing 898the back-link to the last element. However conforming to the C++ convention it 899is illegal to ``operator++`` beyond the sentinel and it also must not be 900dereferenced. 901 902These constraints allow for some implementation freedom to the ``ilist`` how to 903allocate and store the sentinel. The corresponding policy is dictated by 904``ilist_traits<T>``. By default a ``T`` gets heap-allocated whenever the need 905for a sentinel arises. 906 907While the default policy is sufficient in most cases, it may break down when 908``T`` does not provide a default constructor. Also, in the case of many 909instances of ``ilist``\ s, the memory overhead of the associated sentinels is 910wasted. To alleviate the situation with numerous and voluminous 911``T``-sentinels, sometimes a trick is employed, leading to *ghostly sentinels*. 912 913Ghostly sentinels are obtained by specially-crafted ``ilist_traits<T>`` which 914superpose the sentinel with the ``ilist`` instance in memory. Pointer 915arithmetic is used to obtain the sentinel, which is relative to the ``ilist``'s 916``this`` pointer. The ``ilist`` is augmented by an extra pointer, which serves 917as the back-link of the sentinel. This is the only field in the ghostly 918sentinel which can be legally accessed. 919 920.. _dss_other: 921 922Other Sequential Container options 923^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 924 925Other STL containers are available, such as ``std::string``. 926 927There are also various STL adapter classes such as ``std::queue``, 928``std::priority_queue``, ``std::stack``, etc. These provide simplified access 929to an underlying container but don't affect the cost of the container itself. 930 931.. _ds_string: 932 933String-like containers 934---------------------- 935 936There are a variety of ways to pass around and use strings in C and C++, and 937LLVM adds a few new options to choose from. Pick the first option on this list 938that will do what you need, they are ordered according to their relative cost. 939 940Note that is is generally preferred to *not* pass strings around as ``const 941char*``'s. These have a number of problems, including the fact that they 942cannot represent embedded nul ("\0") characters, and do not have a length 943available efficiently. The general replacement for '``const char*``' is 944StringRef. 945 946For more information on choosing string containers for APIs, please see 947:ref:`Passing Strings <string_apis>`. 948 949.. _dss_stringref: 950 951llvm/ADT/StringRef.h 952^^^^^^^^^^^^^^^^^^^^ 953 954The StringRef class is a simple value class that contains a pointer to a 955character and a length, and is quite related to the :ref:`ArrayRef 956<dss_arrayref>` class (but specialized for arrays of characters). Because 957StringRef carries a length with it, it safely handles strings with embedded nul 958characters in it, getting the length does not require a strlen call, and it even 959has very convenient APIs for slicing and dicing the character range that it 960represents. 961 962StringRef is ideal for passing simple strings around that are known to be live, 963either because they are C string literals, std::string, a C array, or a 964SmallVector. Each of these cases has an efficient implicit conversion to 965StringRef, which doesn't result in a dynamic strlen being executed. 966 967StringRef has a few major limitations which make more powerful string containers 968useful: 969 970#. You cannot directly convert a StringRef to a 'const char*' because there is 971 no way to add a trailing nul (unlike the .c_str() method on various stronger 972 classes). 973 974#. StringRef doesn't own or keep alive the underlying string bytes. 975 As such it can easily lead to dangling pointers, and is not suitable for 976 embedding in datastructures in most cases (instead, use an std::string or 977 something like that). 978 979#. For the same reason, StringRef cannot be used as the return value of a 980 method if the method "computes" the result string. Instead, use std::string. 981 982#. StringRef's do not allow you to mutate the pointed-to string bytes and it 983 doesn't allow you to insert or remove bytes from the range. For editing 984 operations like this, it interoperates with the :ref:`Twine <dss_twine>` 985 class. 986 987Because of its strengths and limitations, it is very common for a function to 988take a StringRef and for a method on an object to return a StringRef that points 989into some string that it owns. 990 991.. _dss_twine: 992 993llvm/ADT/Twine.h 994^^^^^^^^^^^^^^^^ 995 996The Twine class is used as an intermediary datatype for APIs that want to take a 997string that can be constructed inline with a series of concatenations. Twine 998works by forming recursive instances of the Twine datatype (a simple value 999object) on the stack as temporary objects, linking them together into a tree 1000which is then linearized when the Twine is consumed. Twine is only safe to use 1001as the argument to a function, and should always be a const reference, e.g.: 1002 1003.. code-block:: c++ 1004 1005 void foo(const Twine &T); 1006 ... 1007 StringRef X = ... 1008 unsigned i = ... 1009 foo(X + "." + Twine(i)); 1010 1011This example forms a string like "blarg.42" by concatenating the values 1012together, and does not form intermediate strings containing "blarg" or "blarg.". 1013 1014Because Twine is constructed with temporary objects on the stack, and because 1015these instances are destroyed at the end of the current statement, it is an 1016inherently dangerous API. For example, this simple variant contains undefined 1017behavior and will probably crash: 1018 1019.. code-block:: c++ 1020 1021 void foo(const Twine &T); 1022 ... 1023 StringRef X = ... 1024 unsigned i = ... 1025 const Twine &Tmp = X + "." + Twine(i); 1026 foo(Tmp); 1027 1028... because the temporaries are destroyed before the call. That said, Twine's 1029are much more efficient than intermediate std::string temporaries, and they work 1030really well with StringRef. Just be aware of their limitations. 1031 1032.. _dss_smallstring: 1033 1034llvm/ADT/SmallString.h 1035^^^^^^^^^^^^^^^^^^^^^^ 1036 1037SmallString is a subclass of :ref:`SmallVector <dss_smallvector>` that adds some 1038convenience APIs like += that takes StringRef's. SmallString avoids allocating 1039memory in the case when the preallocated space is enough to hold its data, and 1040it calls back to general heap allocation when required. Since it owns its data, 1041it is very safe to use and supports full mutation of the string. 1042 1043Like SmallVector's, the big downside to SmallString is their sizeof. While they 1044are optimized for small strings, they themselves are not particularly small. 1045This means that they work great for temporary scratch buffers on the stack, but 1046should not generally be put into the heap: it is very rare to see a SmallString 1047as the member of a frequently-allocated heap data structure or returned 1048by-value. 1049 1050.. _dss_stdstring: 1051 1052std::string 1053^^^^^^^^^^^ 1054 1055The standard C++ std::string class is a very general class that (like 1056SmallString) owns its underlying data. sizeof(std::string) is very reasonable 1057so it can be embedded into heap data structures and returned by-value. On the 1058other hand, std::string is highly inefficient for inline editing (e.g. 1059concatenating a bunch of stuff together) and because it is provided by the 1060standard library, its performance characteristics depend a lot of the host 1061standard library (e.g. libc++ and MSVC provide a highly optimized string class, 1062GCC contains a really slow implementation). 1063 1064The major disadvantage of std::string is that almost every operation that makes 1065them larger can allocate memory, which is slow. As such, it is better to use 1066SmallVector or Twine as a scratch buffer, but then use std::string to persist 1067the result. 1068 1069.. _ds_set: 1070 1071Set-Like Containers (std::set, SmallSet, SetVector, etc) 1072-------------------------------------------------------- 1073 1074Set-like containers are useful when you need to canonicalize multiple values 1075into a single representation. There are several different choices for how to do 1076this, providing various trade-offs. 1077 1078.. _dss_sortedvectorset: 1079 1080A sorted 'vector' 1081^^^^^^^^^^^^^^^^^ 1082 1083If you intend to insert a lot of elements, then do a lot of queries, a great 1084approach is to use a vector (or other sequential container) with 1085std::sort+std::unique to remove duplicates. This approach works really well if 1086your usage pattern has these two distinct phases (insert then query), and can be 1087coupled with a good choice of :ref:`sequential container <ds_sequential>`. 1088 1089This combination provides the several nice properties: the result data is 1090contiguous in memory (good for cache locality), has few allocations, is easy to 1091address (iterators in the final vector are just indices or pointers), and can be 1092efficiently queried with a standard binary search (e.g. 1093``std::lower_bound``; if you want the whole range of elements comparing 1094equal, use ``std::equal_range``). 1095 1096.. _dss_smallset: 1097 1098llvm/ADT/SmallSet.h 1099^^^^^^^^^^^^^^^^^^^ 1100 1101If you have a set-like data structure that is usually small and whose elements 1102are reasonably small, a ``SmallSet<Type, N>`` is a good choice. This set has 1103space for N elements in place (thus, if the set is dynamically smaller than N, 1104no malloc traffic is required) and accesses them with a simple linear search. 1105When the set grows beyond 'N' elements, it allocates a more expensive 1106representation that guarantees efficient access (for most types, it falls back 1107to std::set, but for pointers it uses something far better, :ref:`SmallPtrSet 1108<dss_smallptrset>`. 1109 1110The magic of this class is that it handles small sets extremely efficiently, but 1111gracefully handles extremely large sets without loss of efficiency. The 1112drawback is that the interface is quite small: it supports insertion, queries 1113and erasing, but does not support iteration. 1114 1115.. _dss_smallptrset: 1116 1117llvm/ADT/SmallPtrSet.h 1118^^^^^^^^^^^^^^^^^^^^^^ 1119 1120SmallPtrSet has all the advantages of ``SmallSet`` (and a ``SmallSet`` of 1121pointers is transparently implemented with a ``SmallPtrSet``), but also supports 1122iterators. If more than 'N' insertions are performed, a single quadratically 1123probed hash table is allocated and grows as needed, providing extremely 1124efficient access (constant time insertion/deleting/queries with low constant 1125factors) and is very stingy with malloc traffic. 1126 1127Note that, unlike ``std::set``, the iterators of ``SmallPtrSet`` are invalidated 1128whenever an insertion occurs. Also, the values visited by the iterators are not 1129visited in sorted order. 1130 1131.. _dss_denseset: 1132 1133llvm/ADT/DenseSet.h 1134^^^^^^^^^^^^^^^^^^^ 1135 1136DenseSet is a simple quadratically probed hash table. It excels at supporting 1137small values: it uses a single allocation to hold all of the pairs that are 1138currently inserted in the set. DenseSet is a great way to unique small values 1139that are not simple pointers (use :ref:`SmallPtrSet <dss_smallptrset>` for 1140pointers). Note that DenseSet has the same requirements for the value type that 1141:ref:`DenseMap <dss_densemap>` has. 1142 1143.. _dss_sparseset: 1144 1145llvm/ADT/SparseSet.h 1146^^^^^^^^^^^^^^^^^^^^ 1147 1148SparseSet holds a small number of objects identified by unsigned keys of 1149moderate size. It uses a lot of memory, but provides operations that are almost 1150as fast as a vector. Typical keys are physical registers, virtual registers, or 1151numbered basic blocks. 1152 1153SparseSet is useful for algorithms that need very fast clear/find/insert/erase 1154and fast iteration over small sets. It is not intended for building composite 1155data structures. 1156 1157.. _dss_sparsemultiset: 1158 1159llvm/ADT/SparseMultiSet.h 1160^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1161 1162SparseMultiSet adds multiset behavior to SparseSet, while retaining SparseSet's 1163desirable attributes. Like SparseSet, it typically uses a lot of memory, but 1164provides operations that are almost as fast as a vector. Typical keys are 1165physical registers, virtual registers, or numbered basic blocks. 1166 1167SparseMultiSet is useful for algorithms that need very fast 1168clear/find/insert/erase of the entire collection, and iteration over sets of 1169elements sharing a key. It is often a more efficient choice than using composite 1170data structures (e.g. vector-of-vectors, map-of-vectors). It is not intended for 1171building composite data structures. 1172 1173.. _dss_FoldingSet: 1174 1175llvm/ADT/FoldingSet.h 1176^^^^^^^^^^^^^^^^^^^^^ 1177 1178FoldingSet is an aggregate class that is really good at uniquing 1179expensive-to-create or polymorphic objects. It is a combination of a chained 1180hash table with intrusive links (uniqued objects are required to inherit from 1181FoldingSetNode) that uses :ref:`SmallVector <dss_smallvector>` as part of its ID 1182process. 1183 1184Consider a case where you want to implement a "getOrCreateFoo" method for a 1185complex object (for example, a node in the code generator). The client has a 1186description of **what** it wants to generate (it knows the opcode and all the 1187operands), but we don't want to 'new' a node, then try inserting it into a set 1188only to find out it already exists, at which point we would have to delete it 1189and return the node that already exists. 1190 1191To support this style of client, FoldingSet perform a query with a 1192FoldingSetNodeID (which wraps SmallVector) that can be used to describe the 1193element that we want to query for. The query either returns the element 1194matching the ID or it returns an opaque ID that indicates where insertion should 1195take place. Construction of the ID usually does not require heap traffic. 1196 1197Because FoldingSet uses intrusive links, it can support polymorphic objects in 1198the set (for example, you can have SDNode instances mixed with LoadSDNodes). 1199Because the elements are individually allocated, pointers to the elements are 1200stable: inserting or removing elements does not invalidate any pointers to other 1201elements. 1202 1203.. _dss_set: 1204 1205<set> 1206^^^^^ 1207 1208``std::set`` is a reasonable all-around set class, which is decent at many 1209things but great at nothing. std::set allocates memory for each element 1210inserted (thus it is very malloc intensive) and typically stores three pointers 1211per element in the set (thus adding a large amount of per-element space 1212overhead). It offers guaranteed log(n) performance, which is not particularly 1213fast from a complexity standpoint (particularly if the elements of the set are 1214expensive to compare, like strings), and has extremely high constant factors for 1215lookup, insertion and removal. 1216 1217The advantages of std::set are that its iterators are stable (deleting or 1218inserting an element from the set does not affect iterators or pointers to other 1219elements) and that iteration over the set is guaranteed to be in sorted order. 1220If the elements in the set are large, then the relative overhead of the pointers 1221and malloc traffic is not a big deal, but if the elements of the set are small, 1222std::set is almost never a good choice. 1223 1224.. _dss_setvector: 1225 1226llvm/ADT/SetVector.h 1227^^^^^^^^^^^^^^^^^^^^ 1228 1229LLVM's ``SetVector<Type>`` is an adapter class that combines your choice of a 1230set-like container along with a :ref:`Sequential Container <ds_sequential>` The 1231important property that this provides is efficient insertion with uniquing 1232(duplicate elements are ignored) with iteration support. It implements this by 1233inserting elements into both a set-like container and the sequential container, 1234using the set-like container for uniquing and the sequential container for 1235iteration. 1236 1237The difference between SetVector and other sets is that the order of iteration 1238is guaranteed to match the order of insertion into the SetVector. This property 1239is really important for things like sets of pointers. Because pointer values 1240are non-deterministic (e.g. vary across runs of the program on different 1241machines), iterating over the pointers in the set will not be in a well-defined 1242order. 1243 1244The drawback of SetVector is that it requires twice as much space as a normal 1245set and has the sum of constant factors from the set-like container and the 1246sequential container that it uses. Use it **only** if you need to iterate over 1247the elements in a deterministic order. SetVector is also expensive to delete 1248elements out of (linear time), unless you use its "pop_back" method, which is 1249faster. 1250 1251``SetVector`` is an adapter class that defaults to using ``std::vector`` and a 1252size 16 ``SmallSet`` for the underlying containers, so it is quite expensive. 1253However, ``"llvm/ADT/SetVector.h"`` also provides a ``SmallSetVector`` class, 1254which defaults to using a ``SmallVector`` and ``SmallSet`` of a specified size. 1255If you use this, and if your sets are dynamically smaller than ``N``, you will 1256save a lot of heap traffic. 1257 1258.. _dss_uniquevector: 1259 1260llvm/ADT/UniqueVector.h 1261^^^^^^^^^^^^^^^^^^^^^^^ 1262 1263UniqueVector is similar to :ref:`SetVector <dss_setvector>` but it retains a 1264unique ID for each element inserted into the set. It internally contains a map 1265and a vector, and it assigns a unique ID for each value inserted into the set. 1266 1267UniqueVector is very expensive: its cost is the sum of the cost of maintaining 1268both the map and vector, it has high complexity, high constant factors, and 1269produces a lot of malloc traffic. It should be avoided. 1270 1271.. _dss_immutableset: 1272 1273llvm/ADT/ImmutableSet.h 1274^^^^^^^^^^^^^^^^^^^^^^^ 1275 1276ImmutableSet is an immutable (functional) set implementation based on an AVL 1277tree. Adding or removing elements is done through a Factory object and results 1278in the creation of a new ImmutableSet object. If an ImmutableSet already exists 1279with the given contents, then the existing one is returned; equality is compared 1280with a FoldingSetNodeID. The time and space complexity of add or remove 1281operations is logarithmic in the size of the original set. 1282 1283There is no method for returning an element of the set, you can only check for 1284membership. 1285 1286.. _dss_otherset: 1287 1288Other Set-Like Container Options 1289^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1290 1291The STL provides several other options, such as std::multiset and the various 1292"hash_set" like containers (whether from C++ TR1 or from the SGI library). We 1293never use hash_set and unordered_set because they are generally very expensive 1294(each insertion requires a malloc) and very non-portable. 1295 1296std::multiset is useful if you're not interested in elimination of duplicates, 1297but has all the drawbacks of std::set. A sorted vector (where you don't delete 1298duplicate entries) or some other approach is almost always better. 1299 1300.. _ds_map: 1301 1302Map-Like Containers (std::map, DenseMap, etc) 1303--------------------------------------------- 1304 1305Map-like containers are useful when you want to associate data to a key. As 1306usual, there are a lot of different ways to do this. :) 1307 1308.. _dss_sortedvectormap: 1309 1310A sorted 'vector' 1311^^^^^^^^^^^^^^^^^ 1312 1313If your usage pattern follows a strict insert-then-query approach, you can 1314trivially use the same approach as :ref:`sorted vectors for set-like containers 1315<dss_sortedvectorset>`. The only difference is that your query function (which 1316uses std::lower_bound to get efficient log(n) lookup) should only compare the 1317key, not both the key and value. This yields the same advantages as sorted 1318vectors for sets. 1319 1320.. _dss_stringmap: 1321 1322llvm/ADT/StringMap.h 1323^^^^^^^^^^^^^^^^^^^^ 1324 1325Strings are commonly used as keys in maps, and they are difficult to support 1326efficiently: they are variable length, inefficient to hash and compare when 1327long, expensive to copy, etc. StringMap is a specialized container designed to 1328cope with these issues. It supports mapping an arbitrary range of bytes to an 1329arbitrary other object. 1330 1331The StringMap implementation uses a quadratically-probed hash table, where the 1332buckets store a pointer to the heap allocated entries (and some other stuff). 1333The entries in the map must be heap allocated because the strings are variable 1334length. The string data (key) and the element object (value) are stored in the 1335same allocation with the string data immediately after the element object. 1336This container guarantees the "``(char*)(&Value+1)``" points to the key string 1337for a value. 1338 1339The StringMap is very fast for several reasons: quadratic probing is very cache 1340efficient for lookups, the hash value of strings in buckets is not recomputed 1341when looking up an element, StringMap rarely has to touch the memory for 1342unrelated objects when looking up a value (even when hash collisions happen), 1343hash table growth does not recompute the hash values for strings already in the 1344table, and each pair in the map is store in a single allocation (the string data 1345is stored in the same allocation as the Value of a pair). 1346 1347StringMap also provides query methods that take byte ranges, so it only ever 1348copies a string if a value is inserted into the table. 1349 1350StringMap iteratation order, however, is not guaranteed to be deterministic, so 1351any uses which require that should instead use a std::map. 1352 1353.. _dss_indexmap: 1354 1355llvm/ADT/IndexedMap.h 1356^^^^^^^^^^^^^^^^^^^^^ 1357 1358IndexedMap is a specialized container for mapping small dense integers (or 1359values that can be mapped to small dense integers) to some other type. It is 1360internally implemented as a vector with a mapping function that maps the keys 1361to the dense integer range. 1362 1363This is useful for cases like virtual registers in the LLVM code generator: they 1364have a dense mapping that is offset by a compile-time constant (the first 1365virtual register ID). 1366 1367.. _dss_densemap: 1368 1369llvm/ADT/DenseMap.h 1370^^^^^^^^^^^^^^^^^^^ 1371 1372DenseMap is a simple quadratically probed hash table. It excels at supporting 1373small keys and values: it uses a single allocation to hold all of the pairs 1374that are currently inserted in the map. DenseMap is a great way to map 1375pointers to pointers, or map other small types to each other. 1376 1377There are several aspects of DenseMap that you should be aware of, however. 1378The iterators in a DenseMap are invalidated whenever an insertion occurs, 1379unlike map. Also, because DenseMap allocates space for a large number of 1380key/value pairs (it starts with 64 by default), it will waste a lot of space if 1381your keys or values are large. Finally, you must implement a partial 1382specialization of DenseMapInfo for the key that you want, if it isn't already 1383supported. This is required to tell DenseMap about two special marker values 1384(which can never be inserted into the map) that it needs internally. 1385 1386DenseMap's find_as() method supports lookup operations using an alternate key 1387type. This is useful in cases where the normal key type is expensive to 1388construct, but cheap to compare against. The DenseMapInfo is responsible for 1389defining the appropriate comparison and hashing methods for each alternate key 1390type used. 1391 1392.. _dss_valuemap: 1393 1394llvm/IR/ValueMap.h 1395^^^^^^^^^^^^^^^^^^^ 1396 1397ValueMap is a wrapper around a :ref:`DenseMap <dss_densemap>` mapping 1398``Value*``\ s (or subclasses) to another type. When a Value is deleted or 1399RAUW'ed, ValueMap will update itself so the new version of the key is mapped to 1400the same value, just as if the key were a WeakVH. You can configure exactly how 1401this happens, and what else happens on these two events, by passing a ``Config`` 1402parameter to the ValueMap template. 1403 1404.. _dss_intervalmap: 1405 1406llvm/ADT/IntervalMap.h 1407^^^^^^^^^^^^^^^^^^^^^^ 1408 1409IntervalMap is a compact map for small keys and values. It maps key intervals 1410instead of single keys, and it will automatically coalesce adjacent intervals. 1411When then map only contains a few intervals, they are stored in the map object 1412itself to avoid allocations. 1413 1414The IntervalMap iterators are quite big, so they should not be passed around as 1415STL iterators. The heavyweight iterators allow a smaller data structure. 1416 1417.. _dss_map: 1418 1419<map> 1420^^^^^ 1421 1422std::map has similar characteristics to :ref:`std::set <dss_set>`: it uses a 1423single allocation per pair inserted into the map, it offers log(n) lookup with 1424an extremely large constant factor, imposes a space penalty of 3 pointers per 1425pair in the map, etc. 1426 1427std::map is most useful when your keys or values are very large, if you need to 1428iterate over the collection in sorted order, or if you need stable iterators 1429into the map (i.e. they don't get invalidated if an insertion or deletion of 1430another element takes place). 1431 1432.. _dss_mapvector: 1433 1434llvm/ADT/MapVector.h 1435^^^^^^^^^^^^^^^^^^^^ 1436 1437``MapVector<KeyT,ValueT>`` provides a subset of the DenseMap interface. The 1438main difference is that the iteration order is guaranteed to be the insertion 1439order, making it an easy (but somewhat expensive) solution for non-deterministic 1440iteration over maps of pointers. 1441 1442It is implemented by mapping from key to an index in a vector of key,value 1443pairs. This provides fast lookup and iteration, but has two main drawbacks: 1444the key is stored twice and removing elements takes linear time. If it is 1445necessary to remove elements, it's best to remove them in bulk using 1446``remove_if()``. 1447 1448.. _dss_inteqclasses: 1449 1450llvm/ADT/IntEqClasses.h 1451^^^^^^^^^^^^^^^^^^^^^^^ 1452 1453IntEqClasses provides a compact representation of equivalence classes of small 1454integers. Initially, each integer in the range 0..n-1 has its own equivalence 1455class. Classes can be joined by passing two class representatives to the 1456join(a, b) method. Two integers are in the same class when findLeader() returns 1457the same representative. 1458 1459Once all equivalence classes are formed, the map can be compressed so each 1460integer 0..n-1 maps to an equivalence class number in the range 0..m-1, where m 1461is the total number of equivalence classes. The map must be uncompressed before 1462it can be edited again. 1463 1464.. _dss_immutablemap: 1465 1466llvm/ADT/ImmutableMap.h 1467^^^^^^^^^^^^^^^^^^^^^^^ 1468 1469ImmutableMap is an immutable (functional) map implementation based on an AVL 1470tree. Adding or removing elements is done through a Factory object and results 1471in the creation of a new ImmutableMap object. If an ImmutableMap already exists 1472with the given key set, then the existing one is returned; equality is compared 1473with a FoldingSetNodeID. The time and space complexity of add or remove 1474operations is logarithmic in the size of the original map. 1475 1476.. _dss_othermap: 1477 1478Other Map-Like Container Options 1479^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1480 1481The STL provides several other options, such as std::multimap and the various 1482"hash_map" like containers (whether from C++ TR1 or from the SGI library). We 1483never use hash_set and unordered_set because they are generally very expensive 1484(each insertion requires a malloc) and very non-portable. 1485 1486std::multimap is useful if you want to map a key to multiple values, but has all 1487the drawbacks of std::map. A sorted vector or some other approach is almost 1488always better. 1489 1490.. _ds_bit: 1491 1492Bit storage containers (BitVector, SparseBitVector) 1493--------------------------------------------------- 1494 1495Unlike the other containers, there are only two bit storage containers, and 1496choosing when to use each is relatively straightforward. 1497 1498One additional option is ``std::vector<bool>``: we discourage its use for two 1499reasons 1) the implementation in many common compilers (e.g. commonly 1500available versions of GCC) is extremely inefficient and 2) the C++ standards 1501committee is likely to deprecate this container and/or change it significantly 1502somehow. In any case, please don't use it. 1503 1504.. _dss_bitvector: 1505 1506BitVector 1507^^^^^^^^^ 1508 1509The BitVector container provides a dynamic size set of bits for manipulation. 1510It supports individual bit setting/testing, as well as set operations. The set 1511operations take time O(size of bitvector), but operations are performed one word 1512at a time, instead of one bit at a time. This makes the BitVector very fast for 1513set operations compared to other containers. Use the BitVector when you expect 1514the number of set bits to be high (i.e. a dense set). 1515 1516.. _dss_smallbitvector: 1517 1518SmallBitVector 1519^^^^^^^^^^^^^^ 1520 1521The SmallBitVector container provides the same interface as BitVector, but it is 1522optimized for the case where only a small number of bits, less than 25 or so, 1523are needed. It also transparently supports larger bit counts, but slightly less 1524efficiently than a plain BitVector, so SmallBitVector should only be used when 1525larger counts are rare. 1526 1527At this time, SmallBitVector does not support set operations (and, or, xor), and 1528its operator[] does not provide an assignable lvalue. 1529 1530.. _dss_sparsebitvector: 1531 1532SparseBitVector 1533^^^^^^^^^^^^^^^ 1534 1535The SparseBitVector container is much like BitVector, with one major difference: 1536Only the bits that are set, are stored. This makes the SparseBitVector much 1537more space efficient than BitVector when the set is sparse, as well as making 1538set operations O(number of set bits) instead of O(size of universe). The 1539downside to the SparseBitVector is that setting and testing of random bits is 1540O(N), and on large SparseBitVectors, this can be slower than BitVector. In our 1541implementation, setting or testing bits in sorted order (either forwards or 1542reverse) is O(1) worst case. Testing and setting bits within 128 bits (depends 1543on size) of the current bit is also O(1). As a general statement, 1544testing/setting bits in a SparseBitVector is O(distance away from last set bit). 1545 1546.. _common: 1547 1548Helpful Hints for Common Operations 1549=================================== 1550 1551This section describes how to perform some very simple transformations of LLVM 1552code. This is meant to give examples of common idioms used, showing the 1553practical side of LLVM transformations. 1554 1555Because this is a "how-to" section, you should also read about the main classes 1556that you will be working with. The :ref:`Core LLVM Class Hierarchy Reference 1557<coreclasses>` contains details and descriptions of the main classes that you 1558should know about. 1559 1560.. _inspection: 1561 1562Basic Inspection and Traversal Routines 1563--------------------------------------- 1564 1565The LLVM compiler infrastructure have many different data structures that may be 1566traversed. Following the example of the C++ standard template library, the 1567techniques used to traverse these various data structures are all basically the 1568same. For a enumerable sequence of values, the ``XXXbegin()`` function (or 1569method) returns an iterator to the start of the sequence, the ``XXXend()`` 1570function returns an iterator pointing to one past the last valid element of the 1571sequence, and there is some ``XXXiterator`` data type that is common between the 1572two operations. 1573 1574Because the pattern for iteration is common across many different aspects of the 1575program representation, the standard template library algorithms may be used on 1576them, and it is easier to remember how to iterate. First we show a few common 1577examples of the data structures that need to be traversed. Other data 1578structures are traversed in very similar ways. 1579 1580.. _iterate_function: 1581 1582Iterating over the ``BasicBlock`` in a ``Function`` 1583^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1584 1585It's quite common to have a ``Function`` instance that you'd like to transform 1586in some way; in particular, you'd like to manipulate its ``BasicBlock``\ s. To 1587facilitate this, you'll need to iterate over all of the ``BasicBlock``\ s that 1588constitute the ``Function``. The following is an example that prints the name 1589of a ``BasicBlock`` and the number of ``Instruction``\ s it contains: 1590 1591.. code-block:: c++ 1592 1593 // func is a pointer to a Function instance 1594 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i) 1595 // Print out the name of the basic block if it has one, and then the 1596 // number of instructions that it contains 1597 errs() << "Basic block (name=" << i->getName() << ") has " 1598 << i->size() << " instructions.\n"; 1599 1600Note that i can be used as if it were a pointer for the purposes of invoking 1601member functions of the ``Instruction`` class. This is because the indirection 1602operator is overloaded for the iterator classes. In the above code, the 1603expression ``i->size()`` is exactly equivalent to ``(*i).size()`` just like 1604you'd expect. 1605 1606.. _iterate_basicblock: 1607 1608Iterating over the ``Instruction`` in a ``BasicBlock`` 1609^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1610 1611Just like when dealing with ``BasicBlock``\ s in ``Function``\ s, it's easy to 1612iterate over the individual instructions that make up ``BasicBlock``\ s. Here's 1613a code snippet that prints out each instruction in a ``BasicBlock``: 1614 1615.. code-block:: c++ 1616 1617 // blk is a pointer to a BasicBlock instance 1618 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i) 1619 // The next statement works since operator<<(ostream&,...) 1620 // is overloaded for Instruction& 1621 errs() << *i << "\n"; 1622 1623 1624However, this isn't really the best way to print out the contents of a 1625``BasicBlock``! Since the ostream operators are overloaded for virtually 1626anything you'll care about, you could have just invoked the print routine on the 1627basic block itself: ``errs() << *blk << "\n";``. 1628 1629.. _iterate_insiter: 1630 1631Iterating over the ``Instruction`` in a ``Function`` 1632^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1633 1634If you're finding that you commonly iterate over a ``Function``'s 1635``BasicBlock``\ s and then that ``BasicBlock``'s ``Instruction``\ s, 1636``InstIterator`` should be used instead. You'll need to include 1637``llvm/IR/InstIterator.h`` (`doxygen 1638<http://llvm.org/doxygen/InstIterator_8h.html>`__) and then instantiate 1639``InstIterator``\ s explicitly in your code. Here's a small example that shows 1640how to dump all instructions in a function to the standard error stream: 1641 1642.. code-block:: c++ 1643 1644 #include "llvm/IR/InstIterator.h" 1645 1646 // F is a pointer to a Function instance 1647 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I) 1648 errs() << *I << "\n"; 1649 1650Easy, isn't it? You can also use ``InstIterator``\ s to fill a work list with 1651its initial contents. For example, if you wanted to initialize a work list to 1652contain all instructions in a ``Function`` F, all you would need to do is 1653something like: 1654 1655.. code-block:: c++ 1656 1657 std::set<Instruction*> worklist; 1658 // or better yet, SmallPtrSet<Instruction*, 64> worklist; 1659 1660 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I) 1661 worklist.insert(&*I); 1662 1663The STL set ``worklist`` would now contain all instructions in the ``Function`` 1664pointed to by F. 1665 1666.. _iterate_convert: 1667 1668Turning an iterator into a class pointer (and vice-versa) 1669^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1670 1671Sometimes, it'll be useful to grab a reference (or pointer) to a class instance 1672when all you've got at hand is an iterator. Well, extracting a reference or a 1673pointer from an iterator is very straight-forward. Assuming that ``i`` is a 1674``BasicBlock::iterator`` and ``j`` is a ``BasicBlock::const_iterator``: 1675 1676.. code-block:: c++ 1677 1678 Instruction& inst = *i; // Grab reference to instruction reference 1679 Instruction* pinst = &*i; // Grab pointer to instruction reference 1680 const Instruction& inst = *j; 1681 1682However, the iterators you'll be working with in the LLVM framework are special: 1683they will automatically convert to a ptr-to-instance type whenever they need to. 1684Instead of derferencing the iterator and then taking the address of the result, 1685you can simply assign the iterator to the proper pointer type and you get the 1686dereference and address-of operation as a result of the assignment (behind the 1687scenes, this is a result of overloading casting mechanisms). Thus the last line 1688of the last example, 1689 1690.. code-block:: c++ 1691 1692 Instruction *pinst = &*i; 1693 1694is semantically equivalent to 1695 1696.. code-block:: c++ 1697 1698 Instruction *pinst = i; 1699 1700It's also possible to turn a class pointer into the corresponding iterator, and 1701this is a constant time operation (very efficient). The following code snippet 1702illustrates use of the conversion constructors provided by LLVM iterators. By 1703using these, you can explicitly grab the iterator of something without actually 1704obtaining it via iteration over some structure: 1705 1706.. code-block:: c++ 1707 1708 void printNextInstruction(Instruction* inst) { 1709 BasicBlock::iterator it(inst); 1710 ++it; // After this line, it refers to the instruction after *inst 1711 if (it != inst->getParent()->end()) errs() << *it << "\n"; 1712 } 1713 1714Unfortunately, these implicit conversions come at a cost; they prevent these 1715iterators from conforming to standard iterator conventions, and thus from being 1716usable with standard algorithms and containers. For example, they prevent the 1717following code, where ``B`` is a ``BasicBlock``, from compiling: 1718 1719.. code-block:: c++ 1720 1721 llvm::SmallVector<llvm::Instruction *, 16>(B->begin(), B->end()); 1722 1723Because of this, these implicit conversions may be removed some day, and 1724``operator*`` changed to return a pointer instead of a reference. 1725 1726.. _iterate_complex: 1727 1728Finding call sites: a slightly more complex example 1729^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1730 1731Say that you're writing a FunctionPass and would like to count all the locations 1732in the entire module (that is, across every ``Function``) where a certain 1733function (i.e., some ``Function *``) is already in scope. As you'll learn 1734later, you may want to use an ``InstVisitor`` to accomplish this in a much more 1735straight-forward manner, but this example will allow us to explore how you'd do 1736it if you didn't have ``InstVisitor`` around. In pseudo-code, this is what we 1737want to do: 1738 1739.. code-block:: none 1740 1741 initialize callCounter to zero 1742 for each Function f in the Module 1743 for each BasicBlock b in f 1744 for each Instruction i in b 1745 if (i is a CallInst and calls the given function) 1746 increment callCounter 1747 1748And the actual code is (remember, because we're writing a ``FunctionPass``, our 1749``FunctionPass``-derived class simply has to override the ``runOnFunction`` 1750method): 1751 1752.. code-block:: c++ 1753 1754 Function* targetFunc = ...; 1755 1756 class OurFunctionPass : public FunctionPass { 1757 public: 1758 OurFunctionPass(): callCounter(0) { } 1759 1760 virtual runOnFunction(Function& F) { 1761 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) { 1762 for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) { 1763 if (CallInst* callInst = dyn_cast<CallInst>(&*i)) { 1764 // We know we've encountered a call instruction, so we 1765 // need to determine if it's a call to the 1766 // function pointed to by m_func or not. 1767 if (callInst->getCalledFunction() == targetFunc) 1768 ++callCounter; 1769 } 1770 } 1771 } 1772 } 1773 1774 private: 1775 unsigned callCounter; 1776 }; 1777 1778.. _calls_and_invokes: 1779 1780Treating calls and invokes the same way 1781^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1782 1783You may have noticed that the previous example was a bit oversimplified in that 1784it did not deal with call sites generated by 'invoke' instructions. In this, 1785and in other situations, you may find that you want to treat ``CallInst``\ s and 1786``InvokeInst``\ s the same way, even though their most-specific common base 1787class is ``Instruction``, which includes lots of less closely-related things. 1788For these cases, LLVM provides a handy wrapper class called ``CallSite`` 1789(`doxygen <http://llvm.org/doxygen/classllvm_1_1CallSite.html>`__) It is 1790essentially a wrapper around an ``Instruction`` pointer, with some methods that 1791provide functionality common to ``CallInst``\ s and ``InvokeInst``\ s. 1792 1793This class has "value semantics": it should be passed by value, not by reference 1794and it should not be dynamically allocated or deallocated using ``operator new`` 1795or ``operator delete``. It is efficiently copyable, assignable and 1796constructable, with costs equivalents to that of a bare pointer. If you look at 1797its definition, it has only a single pointer member. 1798 1799.. _iterate_chains: 1800 1801Iterating over def-use & use-def chains 1802^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1803 1804Frequently, we might have an instance of the ``Value`` class (`doxygen 1805<http://llvm.org/doxygen/classllvm_1_1Value.html>`__) and we want to determine 1806which ``User`` s use the ``Value``. The list of all ``User``\ s of a particular 1807``Value`` is called a *def-use* chain. For example, let's say we have a 1808``Function*`` named ``F`` to a particular function ``foo``. Finding all of the 1809instructions that *use* ``foo`` is as simple as iterating over the *def-use* 1810chain of ``F``: 1811 1812.. code-block:: c++ 1813 1814 Function *F = ...; 1815 1816 for (User *U : GV->users()) { 1817 if (Instruction *Inst = dyn_cast<Instruction>(U)) { 1818 errs() << "F is used in instruction:\n"; 1819 errs() << *Inst << "\n"; 1820 } 1821 1822Alternatively, it's common to have an instance of the ``User`` Class (`doxygen 1823<http://llvm.org/doxygen/classllvm_1_1User.html>`__) and need to know what 1824``Value``\ s are used by it. The list of all ``Value``\ s used by a ``User`` is 1825known as a *use-def* chain. Instances of class ``Instruction`` are common 1826``User`` s, so we might want to iterate over all of the values that a particular 1827instruction uses (that is, the operands of the particular ``Instruction``): 1828 1829.. code-block:: c++ 1830 1831 Instruction *pi = ...; 1832 1833 for (Use &U : pi->operands()) { 1834 Value *v = U.get(); 1835 // ... 1836 } 1837 1838Declaring objects as ``const`` is an important tool of enforcing mutation free 1839algorithms (such as analyses, etc.). For this purpose above iterators come in 1840constant flavors as ``Value::const_use_iterator`` and 1841``Value::const_op_iterator``. They automatically arise when calling 1842``use/op_begin()`` on ``const Value*``\ s or ``const User*``\ s respectively. 1843Upon dereferencing, they return ``const Use*``\ s. Otherwise the above patterns 1844remain unchanged. 1845 1846.. _iterate_preds: 1847 1848Iterating over predecessors & successors of blocks 1849^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1850 1851Iterating over the predecessors and successors of a block is quite easy with the 1852routines defined in ``"llvm/Support/CFG.h"``. Just use code like this to 1853iterate over all predecessors of BB: 1854 1855.. code-block:: c++ 1856 1857 #include "llvm/Support/CFG.h" 1858 BasicBlock *BB = ...; 1859 1860 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) { 1861 BasicBlock *Pred = *PI; 1862 // ... 1863 } 1864 1865Similarly, to iterate over successors use ``succ_iterator/succ_begin/succ_end``. 1866 1867.. _simplechanges: 1868 1869Making simple changes 1870--------------------- 1871 1872There are some primitive transformation operations present in the LLVM 1873infrastructure that are worth knowing about. When performing transformations, 1874it's fairly common to manipulate the contents of basic blocks. This section 1875describes some of the common methods for doing so and gives example code. 1876 1877.. _schanges_creating: 1878 1879Creating and inserting new ``Instruction``\ s 1880^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1881 1882*Instantiating Instructions* 1883 1884Creation of ``Instruction``\ s is straight-forward: simply call the constructor 1885for the kind of instruction to instantiate and provide the necessary parameters. 1886For example, an ``AllocaInst`` only *requires* a (const-ptr-to) ``Type``. Thus: 1887 1888.. code-block:: c++ 1889 1890 AllocaInst* ai = new AllocaInst(Type::Int32Ty); 1891 1892will create an ``AllocaInst`` instance that represents the allocation of one 1893integer in the current stack frame, at run time. Each ``Instruction`` subclass 1894is likely to have varying default parameters which change the semantics of the 1895instruction, so refer to the `doxygen documentation for the subclass of 1896Instruction <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_ that 1897you're interested in instantiating. 1898 1899*Naming values* 1900 1901It is very useful to name the values of instructions when you're able to, as 1902this facilitates the debugging of your transformations. If you end up looking 1903at generated LLVM machine code, you definitely want to have logical names 1904associated with the results of instructions! By supplying a value for the 1905``Name`` (default) parameter of the ``Instruction`` constructor, you associate a 1906logical name with the result of the instruction's execution at run time. For 1907example, say that I'm writing a transformation that dynamically allocates space 1908for an integer on the stack, and that integer is going to be used as some kind 1909of index by some other code. To accomplish this, I place an ``AllocaInst`` at 1910the first point in the first ``BasicBlock`` of some ``Function``, and I'm 1911intending to use it within the same ``Function``. I might do: 1912 1913.. code-block:: c++ 1914 1915 AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc"); 1916 1917where ``indexLoc`` is now the logical name of the instruction's execution value, 1918which is a pointer to an integer on the run time stack. 1919 1920*Inserting instructions* 1921 1922There are essentially three ways to insert an ``Instruction`` into an existing 1923sequence of instructions that form a ``BasicBlock``: 1924 1925* Insertion into an explicit instruction list 1926 1927 Given a ``BasicBlock* pb``, an ``Instruction* pi`` within that ``BasicBlock``, 1928 and a newly-created instruction we wish to insert before ``*pi``, we do the 1929 following: 1930 1931 .. code-block:: c++ 1932 1933 BasicBlock *pb = ...; 1934 Instruction *pi = ...; 1935 Instruction *newInst = new Instruction(...); 1936 1937 pb->getInstList().insert(pi, newInst); // Inserts newInst before pi in pb 1938 1939 Appending to the end of a ``BasicBlock`` is so common that the ``Instruction`` 1940 class and ``Instruction``-derived classes provide constructors which take a 1941 pointer to a ``BasicBlock`` to be appended to. For example code that looked 1942 like: 1943 1944 .. code-block:: c++ 1945 1946 BasicBlock *pb = ...; 1947 Instruction *newInst = new Instruction(...); 1948 1949 pb->getInstList().push_back(newInst); // Appends newInst to pb 1950 1951 becomes: 1952 1953 .. code-block:: c++ 1954 1955 BasicBlock *pb = ...; 1956 Instruction *newInst = new Instruction(..., pb); 1957 1958 which is much cleaner, especially if you are creating long instruction 1959 streams. 1960 1961* Insertion into an implicit instruction list 1962 1963 ``Instruction`` instances that are already in ``BasicBlock``\ s are implicitly 1964 associated with an existing instruction list: the instruction list of the 1965 enclosing basic block. Thus, we could have accomplished the same thing as the 1966 above code without being given a ``BasicBlock`` by doing: 1967 1968 .. code-block:: c++ 1969 1970 Instruction *pi = ...; 1971 Instruction *newInst = new Instruction(...); 1972 1973 pi->getParent()->getInstList().insert(pi, newInst); 1974 1975 In fact, this sequence of steps occurs so frequently that the ``Instruction`` 1976 class and ``Instruction``-derived classes provide constructors which take (as 1977 a default parameter) a pointer to an ``Instruction`` which the newly-created 1978 ``Instruction`` should precede. That is, ``Instruction`` constructors are 1979 capable of inserting the newly-created instance into the ``BasicBlock`` of a 1980 provided instruction, immediately before that instruction. Using an 1981 ``Instruction`` constructor with a ``insertBefore`` (default) parameter, the 1982 above code becomes: 1983 1984 .. code-block:: c++ 1985 1986 Instruction* pi = ...; 1987 Instruction* newInst = new Instruction(..., pi); 1988 1989 which is much cleaner, especially if you're creating a lot of instructions and 1990 adding them to ``BasicBlock``\ s. 1991 1992* Insertion using an instance of ``IRBuilder`` 1993 1994 Inserting several ``Instruction``\ s can be quite laborious using the previous 1995 methods. The ``IRBuilder`` is a convenience class that can be used to add 1996 several instructions to the end of a ``BasicBlock`` or before a particular 1997 ``Instruction``. It also supports constant folding and renaming named 1998 registers (see ``IRBuilder``'s template arguments). 1999 2000 The example below demonstrates a very simple use of the ``IRBuilder`` where 2001 three instructions are inserted before the instruction ``pi``. The first two 2002 instructions are Call instructions and third instruction multiplies the return 2003 value of the two calls. 2004 2005 .. code-block:: c++ 2006 2007 Instruction *pi = ...; 2008 IRBuilder<> Builder(pi); 2009 CallInst* callOne = Builder.CreateCall(...); 2010 CallInst* callTwo = Builder.CreateCall(...); 2011 Value* result = Builder.CreateMul(callOne, callTwo); 2012 2013 The example below is similar to the above example except that the created 2014 ``IRBuilder`` inserts instructions at the end of the ``BasicBlock`` ``pb``. 2015 2016 .. code-block:: c++ 2017 2018 BasicBlock *pb = ...; 2019 IRBuilder<> Builder(pb); 2020 CallInst* callOne = Builder.CreateCall(...); 2021 CallInst* callTwo = Builder.CreateCall(...); 2022 Value* result = Builder.CreateMul(callOne, callTwo); 2023 2024 See :doc:`tutorial/LangImpl3` for a practical use of the ``IRBuilder``. 2025 2026 2027.. _schanges_deleting: 2028 2029Deleting Instructions 2030^^^^^^^^^^^^^^^^^^^^^ 2031 2032Deleting an instruction from an existing sequence of instructions that form a 2033BasicBlock_ is very straight-forward: just call the instruction's 2034``eraseFromParent()`` method. For example: 2035 2036.. code-block:: c++ 2037 2038 Instruction *I = .. ; 2039 I->eraseFromParent(); 2040 2041This unlinks the instruction from its containing basic block and deletes it. If 2042you'd just like to unlink the instruction from its containing basic block but 2043not delete it, you can use the ``removeFromParent()`` method. 2044 2045.. _schanges_replacing: 2046 2047Replacing an Instruction with another Value 2048^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 2049 2050Replacing individual instructions 2051""""""""""""""""""""""""""""""""" 2052 2053Including "`llvm/Transforms/Utils/BasicBlockUtils.h 2054<http://llvm.org/doxygen/BasicBlockUtils_8h-source.html>`_" permits use of two 2055very useful replace functions: ``ReplaceInstWithValue`` and 2056``ReplaceInstWithInst``. 2057 2058.. _schanges_deleting_sub: 2059 2060Deleting Instructions 2061""""""""""""""""""""" 2062 2063* ``ReplaceInstWithValue`` 2064 2065 This function replaces all uses of a given instruction with a value, and then 2066 removes the original instruction. The following example illustrates the 2067 replacement of the result of a particular ``AllocaInst`` that allocates memory 2068 for a single integer with a null pointer to an integer. 2069 2070 .. code-block:: c++ 2071 2072 AllocaInst* instToReplace = ...; 2073 BasicBlock::iterator ii(instToReplace); 2074 2075 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii, 2076 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty))); 2077 2078* ``ReplaceInstWithInst`` 2079 2080 This function replaces a particular instruction with another instruction, 2081 inserting the new instruction into the basic block at the location where the 2082 old instruction was, and replacing any uses of the old instruction with the 2083 new instruction. The following example illustrates the replacement of one 2084 ``AllocaInst`` with another. 2085 2086 .. code-block:: c++ 2087 2088 AllocaInst* instToReplace = ...; 2089 BasicBlock::iterator ii(instToReplace); 2090 2091 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii, 2092 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt")); 2093 2094 2095Replacing multiple uses of Users and Values 2096""""""""""""""""""""""""""""""""""""""""""" 2097 2098You can use ``Value::replaceAllUsesWith`` and ``User::replaceUsesOfWith`` to 2099change more than one use at a time. See the doxygen documentation for the 2100`Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_ and `User Class 2101<http://llvm.org/doxygen/classllvm_1_1User.html>`_, respectively, for more 2102information. 2103 2104.. _schanges_deletingGV: 2105 2106Deleting GlobalVariables 2107^^^^^^^^^^^^^^^^^^^^^^^^ 2108 2109Deleting a global variable from a module is just as easy as deleting an 2110Instruction. First, you must have a pointer to the global variable that you 2111wish to delete. You use this pointer to erase it from its parent, the module. 2112For example: 2113 2114.. code-block:: c++ 2115 2116 GlobalVariable *GV = .. ; 2117 2118 GV->eraseFromParent(); 2119 2120 2121.. _create_types: 2122 2123How to Create Types 2124------------------- 2125 2126In generating IR, you may need some complex types. If you know these types 2127statically, you can use ``TypeBuilder<...>::get()``, defined in 2128``llvm/Support/TypeBuilder.h``, to retrieve them. ``TypeBuilder`` has two forms 2129depending on whether you're building types for cross-compilation or native 2130library use. ``TypeBuilder<T, true>`` requires that ``T`` be independent of the 2131host environment, meaning that it's built out of types from the ``llvm::types`` 2132(`doxygen <http://llvm.org/doxygen/namespacellvm_1_1types.html>`__) namespace 2133and pointers, functions, arrays, etc. built of those. ``TypeBuilder<T, false>`` 2134additionally allows native C types whose size may depend on the host compiler. 2135For example, 2136 2137.. code-block:: c++ 2138 2139 FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get(); 2140 2141is easier to read and write than the equivalent 2142 2143.. code-block:: c++ 2144 2145 std::vector<const Type*> params; 2146 params.push_back(PointerType::getUnqual(Type::Int32Ty)); 2147 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false); 2148 2149See the `class comment 2150<http://llvm.org/doxygen/TypeBuilder_8h-source.html#l00001>`_ for more details. 2151 2152.. _threading: 2153 2154Threads and LLVM 2155================ 2156 2157This section describes the interaction of the LLVM APIs with multithreading, 2158both on the part of client applications, and in the JIT, in the hosted 2159application. 2160 2161Note that LLVM's support for multithreading is still relatively young. Up 2162through version 2.5, the execution of threaded hosted applications was 2163supported, but not threaded client access to the APIs. While this use case is 2164now supported, clients *must* adhere to the guidelines specified below to ensure 2165proper operation in multithreaded mode. 2166 2167Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic 2168intrinsics in order to support threaded operation. If you need a 2169multhreading-capable LLVM on a platform without a suitably modern system 2170compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and 2171using the resultant compiler to build a copy of LLVM with multithreading 2172support. 2173 2174.. _shutdown: 2175 2176Ending Execution with ``llvm_shutdown()`` 2177----------------------------------------- 2178 2179When you are done using the LLVM APIs, you should call ``llvm_shutdown()`` to 2180deallocate memory used for internal structures. 2181 2182.. _managedstatic: 2183 2184Lazy Initialization with ``ManagedStatic`` 2185------------------------------------------ 2186 2187``ManagedStatic`` is a utility class in LLVM used to implement static 2188initialization of static resources, such as the global type tables. In a 2189single-threaded environment, it implements a simple lazy initialization scheme. 2190When LLVM is compiled with support for multi-threading, however, it uses 2191double-checked locking to implement thread-safe lazy initialization. 2192 2193.. _llvmcontext: 2194 2195Achieving Isolation with ``LLVMContext`` 2196---------------------------------------- 2197 2198``LLVMContext`` is an opaque class in the LLVM API which clients can use to 2199operate multiple, isolated instances of LLVM concurrently within the same 2200address space. For instance, in a hypothetical compile-server, the compilation 2201of an individual translation unit is conceptually independent from all the 2202others, and it would be desirable to be able to compile incoming translation 2203units concurrently on independent server threads. Fortunately, ``LLVMContext`` 2204exists to enable just this kind of scenario! 2205 2206Conceptually, ``LLVMContext`` provides isolation. Every LLVM entity 2207(``Module``\ s, ``Value``\ s, ``Type``\ s, ``Constant``\ s, etc.) in LLVM's 2208in-memory IR belongs to an ``LLVMContext``. Entities in different contexts 2209*cannot* interact with each other: ``Module``\ s in different contexts cannot be 2210linked together, ``Function``\ s cannot be added to ``Module``\ s in different 2211contexts, etc. What this means is that is is safe to compile on multiple 2212threads simultaneously, as long as no two threads operate on entities within the 2213same context. 2214 2215In practice, very few places in the API require the explicit specification of a 2216``LLVMContext``, other than the ``Type`` creation/lookup APIs. Because every 2217``Type`` carries a reference to its owning context, most other entities can 2218determine what context they belong to by looking at their own ``Type``. If you 2219are adding new entities to LLVM IR, please try to maintain this interface 2220design. 2221 2222For clients that do *not* require the benefits of isolation, LLVM provides a 2223convenience API ``getGlobalContext()``. This returns a global, lazily 2224initialized ``LLVMContext`` that may be used in situations where isolation is 2225not a concern. 2226 2227.. _jitthreading: 2228 2229Threads and the JIT 2230------------------- 2231 2232LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple 2233threads can call ``ExecutionEngine::getPointerToFunction()`` or 2234``ExecutionEngine::runFunction()`` concurrently, and multiple threads can run 2235code output by the JIT concurrently. The user must still ensure that only one 2236thread accesses IR in a given ``LLVMContext`` while another thread might be 2237modifying it. One way to do that is to always hold the JIT lock while accessing 2238IR outside the JIT (the JIT *modifies* the IR by adding ``CallbackVH``\ s). 2239Another way is to only call ``getPointerToFunction()`` from the 2240``LLVMContext``'s thread. 2241 2242When the JIT is configured to compile lazily (using 2243``ExecutionEngine::DisableLazyCompilation(false)``), there is currently a `race 2244condition <http://llvm.org/bugs/show_bug.cgi?id=5184>`_ in updating call sites 2245after a function is lazily-jitted. It's still possible to use the lazy JIT in a 2246threaded program if you ensure that only one thread at a time can call any 2247particular lazy stub and that the JIT lock guards any IR access, but we suggest 2248using only the eager JIT in threaded programs. 2249 2250.. _advanced: 2251 2252Advanced Topics 2253=============== 2254 2255This section describes some of the advanced or obscure API's that most clients 2256do not need to be aware of. These API's tend manage the inner workings of the 2257LLVM system, and only need to be accessed in unusual circumstances. 2258 2259.. _SymbolTable: 2260 2261The ``ValueSymbolTable`` class 2262------------------------------ 2263 2264The ``ValueSymbolTable`` (`doxygen 2265<http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html>`__) class provides 2266a symbol table that the :ref:`Function <c_Function>` and Module_ classes use for 2267naming value definitions. The symbol table can provide a name for any Value_. 2268 2269Note that the ``SymbolTable`` class should not be directly accessed by most 2270clients. It should only be used when iteration over the symbol table names 2271themselves are required, which is very special purpose. Note that not all LLVM 2272Value_\ s have names, and those without names (i.e. they have an empty name) do 2273not exist in the symbol table. 2274 2275Symbol tables support iteration over the values in the symbol table with 2276``begin/end/iterator`` and supports querying to see if a specific name is in the 2277symbol table (with ``lookup``). The ``ValueSymbolTable`` class exposes no 2278public mutator methods, instead, simply call ``setName`` on a value, which will 2279autoinsert it into the appropriate symbol table. 2280 2281.. _UserLayout: 2282 2283The ``User`` and owned ``Use`` classes' memory layout 2284----------------------------------------------------- 2285 2286The ``User`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1User.html>`__) 2287class provides a basis for expressing the ownership of ``User`` towards other 2288`Value instance <http://llvm.org/doxygen/classllvm_1_1Value.html>`_\ s. The 2289``Use`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1Use.html>`__) helper 2290class is employed to do the bookkeeping and to facilitate *O(1)* addition and 2291removal. 2292 2293.. _Use2User: 2294 2295Interaction and relationship between ``User`` and ``Use`` objects 2296^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 2297 2298A subclass of ``User`` can choose between incorporating its ``Use`` objects or 2299refer to them out-of-line by means of a pointer. A mixed variant (some ``Use`` 2300s inline others hung off) is impractical and breaks the invariant that the 2301``Use`` objects belonging to the same ``User`` form a contiguous array. 2302 2303We have 2 different layouts in the ``User`` (sub)classes: 2304 2305* Layout a) 2306 2307 The ``Use`` object(s) are inside (resp. at fixed offset) of the ``User`` 2308 object and there are a fixed number of them. 2309 2310* Layout b) 2311 2312 The ``Use`` object(s) are referenced by a pointer to an array from the 2313 ``User`` object and there may be a variable number of them. 2314 2315As of v2.4 each layout still possesses a direct pointer to the start of the 2316array of ``Use``\ s. Though not mandatory for layout a), we stick to this 2317redundancy for the sake of simplicity. The ``User`` object also stores the 2318number of ``Use`` objects it has. (Theoretically this information can also be 2319calculated given the scheme presented below.) 2320 2321Special forms of allocation operators (``operator new``) enforce the following 2322memory layouts: 2323 2324* Layout a) is modelled by prepending the ``User`` object by the ``Use[]`` 2325 array. 2326 2327 .. code-block:: none 2328 2329 ...---.---.---.---.-------... 2330 | P | P | P | P | User 2331 '''---'---'---'---'-------''' 2332 2333* Layout b) is modelled by pointing at the ``Use[]`` array. 2334 2335 .. code-block:: none 2336 2337 .-------... 2338 | User 2339 '-------''' 2340 | 2341 v 2342 .---.---.---.---... 2343 | P | P | P | P | 2344 '---'---'---'---''' 2345 2346*(In the above figures* '``P``' *stands for the* ``Use**`` *that is stored in 2347each* ``Use`` *object in the member* ``Use::Prev`` *)* 2348 2349.. _Waymarking: 2350 2351The waymarking algorithm 2352^^^^^^^^^^^^^^^^^^^^^^^^ 2353 2354Since the ``Use`` objects are deprived of the direct (back)pointer to their 2355``User`` objects, there must be a fast and exact method to recover it. This is 2356accomplished by the following scheme: 2357 2358A bit-encoding in the 2 LSBits (least significant bits) of the ``Use::Prev`` 2359allows to find the start of the ``User`` object: 2360 2361* ``00`` --- binary digit 0 2362 2363* ``01`` --- binary digit 1 2364 2365* ``10`` --- stop and calculate (``s``) 2366 2367* ``11`` --- full stop (``S``) 2368 2369Given a ``Use*``, all we have to do is to walk till we get a stop and we either 2370have a ``User`` immediately behind or we have to walk to the next stop picking 2371up digits and calculating the offset: 2372 2373.. code-block:: none 2374 2375 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---------------- 2376 | 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*) 2377 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---------------- 2378 |+15 |+10 |+6 |+3 |+1 2379 | | | | | __> 2380 | | | | __________> 2381 | | | ______________________> 2382 | | ______________________________________> 2383 | __________________________________________________________> 2384 2385Only the significant number of bits need to be stored between the stops, so that 2386the *worst case is 20 memory accesses* when there are 1000 ``Use`` objects 2387associated with a ``User``. 2388 2389.. _ReferenceImpl: 2390 2391Reference implementation 2392^^^^^^^^^^^^^^^^^^^^^^^^ 2393 2394The following literate Haskell fragment demonstrates the concept: 2395 2396.. code-block:: haskell 2397 2398 > import Test.QuickCheck 2399 > 2400 > digits :: Int -> [Char] -> [Char] 2401 > digits 0 acc = '0' : acc 2402 > digits 1 acc = '1' : acc 2403 > digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc 2404 > 2405 > dist :: Int -> [Char] -> [Char] 2406 > dist 0 [] = ['S'] 2407 > dist 0 acc = acc 2408 > dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r 2409 > dist n acc = dist (n - 1) $ dist 1 acc 2410 > 2411 > takeLast n ss = reverse $ take n $ reverse ss 2412 > 2413 > test = takeLast 40 $ dist 20 [] 2414 > 2415 2416Printing <test> gives: ``"1s100000s11010s10100s1111s1010s110s11s1S"`` 2417 2418The reverse algorithm computes the length of the string just by examining a 2419certain prefix: 2420 2421.. code-block:: haskell 2422 2423 > pref :: [Char] -> Int 2424 > pref "S" = 1 2425 > pref ('s':'1':rest) = decode 2 1 rest 2426 > pref (_:rest) = 1 + pref rest 2427 > 2428 > decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest 2429 > decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest 2430 > decode walk acc _ = walk + acc 2431 > 2432 2433Now, as expected, printing <pref test> gives ``40``. 2434 2435We can *quickCheck* this with following property: 2436 2437.. code-block:: haskell 2438 2439 > testcase = dist 2000 [] 2440 > testcaseLength = length testcase 2441 > 2442 > identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr 2443 > where arr = takeLast n testcase 2444 > 2445 2446As expected <quickCheck identityProp> gives: 2447 2448:: 2449 2450 *Main> quickCheck identityProp 2451 OK, passed 100 tests. 2452 2453Let's be a bit more exhaustive: 2454 2455.. code-block:: haskell 2456 2457 > 2458 > deepCheck p = check (defaultConfig { configMaxTest = 500 }) p 2459 > 2460 2461And here is the result of <deepCheck identityProp>: 2462 2463:: 2464 2465 *Main> deepCheck identityProp 2466 OK, passed 500 tests. 2467 2468.. _Tagging: 2469 2470Tagging considerations 2471^^^^^^^^^^^^^^^^^^^^^^ 2472 2473To maintain the invariant that the 2 LSBits of each ``Use**`` in ``Use`` never 2474change after being set up, setters of ``Use::Prev`` must re-tag the new 2475``Use**`` on every modification. Accordingly getters must strip the tag bits. 2476 2477For layout b) instead of the ``User`` we find a pointer (``User*`` with LSBit 2478set). Following this pointer brings us to the ``User``. A portable trick 2479ensures that the first bytes of ``User`` (if interpreted as a pointer) never has 2480the LSBit set. (Portability is relying on the fact that all known compilers 2481place the ``vptr`` in the first word of the instances.) 2482 2483.. _coreclasses: 2484 2485The Core LLVM Class Hierarchy Reference 2486======================================= 2487 2488``#include "llvm/IR/Type.h"`` 2489 2490header source: `Type.h <http://llvm.org/doxygen/Type_8h-source.html>`_ 2491 2492doxygen info: `Type Clases <http://llvm.org/doxygen/classllvm_1_1Type.html>`_ 2493 2494The Core LLVM classes are the primary means of representing the program being 2495inspected or transformed. The core LLVM classes are defined in header files in 2496the ``include/llvm/`` directory, and implemented in the ``lib/VMCore`` 2497directory. 2498 2499.. _Type: 2500 2501The Type class and Derived Types 2502-------------------------------- 2503 2504``Type`` is a superclass of all type classes. Every ``Value`` has a ``Type``. 2505``Type`` cannot be instantiated directly but only through its subclasses. 2506Certain primitive types (``VoidType``, ``LabelType``, ``FloatType`` and 2507``DoubleType``) have hidden subclasses. They are hidden because they offer no 2508useful functionality beyond what the ``Type`` class offers except to distinguish 2509themselves from other subclasses of ``Type``. 2510 2511All other types are subclasses of ``DerivedType``. Types can be named, but this 2512is not a requirement. There exists exactly one instance of a given shape at any 2513one time. This allows type equality to be performed with address equality of 2514the Type Instance. That is, given two ``Type*`` values, the types are identical 2515if the pointers are identical. 2516 2517.. _m_Type: 2518 2519Important Public Methods 2520^^^^^^^^^^^^^^^^^^^^^^^^ 2521 2522* ``bool isIntegerTy() const``: Returns true for any integer type. 2523 2524* ``bool isFloatingPointTy()``: Return true if this is one of the five 2525 floating point types. 2526 2527* ``bool isSized()``: Return true if the type has known size. Things 2528 that don't have a size are abstract types, labels and void. 2529 2530.. _derivedtypes: 2531 2532Important Derived Types 2533^^^^^^^^^^^^^^^^^^^^^^^ 2534 2535``IntegerType`` 2536 Subclass of DerivedType that represents integer types of any bit width. Any 2537 bit width between ``IntegerType::MIN_INT_BITS`` (1) and 2538 ``IntegerType::MAX_INT_BITS`` (~8 million) can be represented. 2539 2540 * ``static const IntegerType* get(unsigned NumBits)``: get an integer 2541 type of a specific bit width. 2542 2543 * ``unsigned getBitWidth() const``: Get the bit width of an integer type. 2544 2545``SequentialType`` 2546 This is subclassed by ArrayType, PointerType and VectorType. 2547 2548 * ``const Type * getElementType() const``: Returns the type of each 2549 of the elements in the sequential type. 2550 2551``ArrayType`` 2552 This is a subclass of SequentialType and defines the interface for array 2553 types. 2554 2555 * ``unsigned getNumElements() const``: Returns the number of elements 2556 in the array. 2557 2558``PointerType`` 2559 Subclass of SequentialType for pointer types. 2560 2561``VectorType`` 2562 Subclass of SequentialType for vector types. A vector type is similar to an 2563 ArrayType but is distinguished because it is a first class type whereas 2564 ArrayType is not. Vector types are used for vector operations and are usually 2565 small vectors of of an integer or floating point type. 2566 2567``StructType`` 2568 Subclass of DerivedTypes for struct types. 2569 2570.. _FunctionType: 2571 2572``FunctionType`` 2573 Subclass of DerivedTypes for function types. 2574 2575 * ``bool isVarArg() const``: Returns true if it's a vararg function. 2576 2577 * ``const Type * getReturnType() const``: Returns the return type of the 2578 function. 2579 2580 * ``const Type * getParamType (unsigned i)``: Returns the type of the ith 2581 parameter. 2582 2583 * ``const unsigned getNumParams() const``: Returns the number of formal 2584 parameters. 2585 2586.. _Module: 2587 2588The ``Module`` class 2589-------------------- 2590 2591``#include "llvm/IR/Module.h"`` 2592 2593header source: `Module.h <http://llvm.org/doxygen/Module_8h-source.html>`_ 2594 2595doxygen info: `Module Class <http://llvm.org/doxygen/classllvm_1_1Module.html>`_ 2596 2597The ``Module`` class represents the top level structure present in LLVM 2598programs. An LLVM module is effectively either a translation unit of the 2599original program or a combination of several translation units merged by the 2600linker. The ``Module`` class keeps track of a list of :ref:`Function 2601<c_Function>`\ s, a list of GlobalVariable_\ s, and a SymbolTable_. 2602Additionally, it contains a few helpful member functions that try to make common 2603operations easy. 2604 2605.. _m_Module: 2606 2607Important Public Members of the ``Module`` class 2608^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 2609 2610* ``Module::Module(std::string name = "")`` 2611 2612 Constructing a Module_ is easy. You can optionally provide a name for it 2613 (probably based on the name of the translation unit). 2614 2615* | ``Module::iterator`` - Typedef for function list iterator 2616 | ``Module::const_iterator`` - Typedef for const_iterator. 2617 | ``begin()``, ``end()``, ``size()``, ``empty()`` 2618 2619 These are forwarding methods that make it easy to access the contents of a 2620 ``Module`` object's :ref:`Function <c_Function>` list. 2621 2622* ``Module::FunctionListType &getFunctionList()`` 2623 2624 Returns the list of :ref:`Function <c_Function>`\ s. This is necessary to use 2625 when you need to update the list or perform a complex action that doesn't have 2626 a forwarding method. 2627 2628---------------- 2629 2630* | ``Module::global_iterator`` - Typedef for global variable list iterator 2631 | ``Module::const_global_iterator`` - Typedef for const_iterator. 2632 | ``global_begin()``, ``global_end()``, ``global_size()``, ``global_empty()`` 2633 2634 These are forwarding methods that make it easy to access the contents of a 2635 ``Module`` object's GlobalVariable_ list. 2636 2637* ``Module::GlobalListType &getGlobalList()`` 2638 2639 Returns the list of GlobalVariable_\ s. This is necessary to use when you 2640 need to update the list or perform a complex action that doesn't have a 2641 forwarding method. 2642 2643---------------- 2644 2645* ``SymbolTable *getSymbolTable()`` 2646 2647 Return a reference to the SymbolTable_ for this ``Module``. 2648 2649---------------- 2650 2651* ``Function *getFunction(StringRef Name) const`` 2652 2653 Look up the specified function in the ``Module`` SymbolTable_. If it does not 2654 exist, return ``null``. 2655 2656* ``Function *getOrInsertFunction(const std::string &Name, const FunctionType 2657 *T)`` 2658 2659 Look up the specified function in the ``Module`` SymbolTable_. If it does not 2660 exist, add an external declaration for the function and return it. 2661 2662* ``std::string getTypeName(const Type *Ty)`` 2663 2664 If there is at least one entry in the SymbolTable_ for the specified Type_, 2665 return it. Otherwise return the empty string. 2666 2667* ``bool addTypeName(const std::string &Name, const Type *Ty)`` 2668 2669 Insert an entry in the SymbolTable_ mapping ``Name`` to ``Ty``. If there is 2670 already an entry for this name, true is returned and the SymbolTable_ is not 2671 modified. 2672 2673.. _Value: 2674 2675The ``Value`` class 2676------------------- 2677 2678``#include "llvm/IR/Value.h"`` 2679 2680header source: `Value.h <http://llvm.org/doxygen/Value_8h-source.html>`_ 2681 2682doxygen info: `Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_ 2683 2684The ``Value`` class is the most important class in the LLVM Source base. It 2685represents a typed value that may be used (among other things) as an operand to 2686an instruction. There are many different types of ``Value``\ s, such as 2687Constant_\ s, Argument_\ s. Even Instruction_\ s and :ref:`Function 2688<c_Function>`\ s are ``Value``\ s. 2689 2690A particular ``Value`` may be used many times in the LLVM representation for a 2691program. For example, an incoming argument to a function (represented with an 2692instance of the Argument_ class) is "used" by every instruction in the function 2693that references the argument. To keep track of this relationship, the ``Value`` 2694class keeps a list of all of the ``User``\ s that is using it (the User_ class 2695is a base class for all nodes in the LLVM graph that can refer to ``Value``\ s). 2696This use list is how LLVM represents def-use information in the program, and is 2697accessible through the ``use_*`` methods, shown below. 2698 2699Because LLVM is a typed representation, every LLVM ``Value`` is typed, and this 2700Type_ is available through the ``getType()`` method. In addition, all LLVM 2701values can be named. The "name" of the ``Value`` is a symbolic string printed 2702in the LLVM code: 2703 2704.. code-block:: llvm 2705 2706 %foo = add i32 1, 2 2707 2708.. _nameWarning: 2709 2710The name of this instruction is "foo". **NOTE** that the name of any value may 2711be missing (an empty string), so names should **ONLY** be used for debugging 2712(making the source code easier to read, debugging printouts), they should not be 2713used to keep track of values or map between them. For this purpose, use a 2714``std::map`` of pointers to the ``Value`` itself instead. 2715 2716One important aspect of LLVM is that there is no distinction between an SSA 2717variable and the operation that produces it. Because of this, any reference to 2718the value produced by an instruction (or the value available as an incoming 2719argument, for example) is represented as a direct pointer to the instance of the 2720class that represents this value. Although this may take some getting used to, 2721it simplifies the representation and makes it easier to manipulate. 2722 2723.. _m_Value: 2724 2725Important Public Members of the ``Value`` class 2726^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 2727 2728* | ``Value::use_iterator`` - Typedef for iterator over the use-list 2729 | ``Value::const_use_iterator`` - Typedef for const_iterator over the 2730 use-list 2731 | ``unsigned use_size()`` - Returns the number of users of the value. 2732 | ``bool use_empty()`` - Returns true if there are no users. 2733 | ``use_iterator use_begin()`` - Get an iterator to the start of the 2734 use-list. 2735 | ``use_iterator use_end()`` - Get an iterator to the end of the use-list. 2736 | ``User *use_back()`` - Returns the last element in the list. 2737 2738 These methods are the interface to access the def-use information in LLVM. 2739 As with all other iterators in LLVM, the naming conventions follow the 2740 conventions defined by the STL_. 2741 2742* ``Type *getType() const`` 2743 This method returns the Type of the Value. 2744 2745* | ``bool hasName() const`` 2746 | ``std::string getName() const`` 2747 | ``void setName(const std::string &Name)`` 2748 2749 This family of methods is used to access and assign a name to a ``Value``, be 2750 aware of the :ref:`precaution above <nameWarning>`. 2751 2752* ``void replaceAllUsesWith(Value *V)`` 2753 2754 This method traverses the use list of a ``Value`` changing all User_\ s of the 2755 current value to refer to "``V``" instead. For example, if you detect that an 2756 instruction always produces a constant value (for example through constant 2757 folding), you can replace all uses of the instruction with the constant like 2758 this: 2759 2760 .. code-block:: c++ 2761 2762 Inst->replaceAllUsesWith(ConstVal); 2763 2764.. _User: 2765 2766The ``User`` class 2767------------------ 2768 2769``#include "llvm/IR/User.h"`` 2770 2771header source: `User.h <http://llvm.org/doxygen/User_8h-source.html>`_ 2772 2773doxygen info: `User Class <http://llvm.org/doxygen/classllvm_1_1User.html>`_ 2774 2775Superclass: Value_ 2776 2777The ``User`` class is the common base class of all LLVM nodes that may refer to 2778``Value``\ s. It exposes a list of "Operands" that are all of the ``Value``\ s 2779that the User is referring to. The ``User`` class itself is a subclass of 2780``Value``. 2781 2782The operands of a ``User`` point directly to the LLVM ``Value`` that it refers 2783to. Because LLVM uses Static Single Assignment (SSA) form, there can only be 2784one definition referred to, allowing this direct connection. This connection 2785provides the use-def information in LLVM. 2786 2787.. _m_User: 2788 2789Important Public Members of the ``User`` class 2790^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 2791 2792The ``User`` class exposes the operand list in two ways: through an index access 2793interface and through an iterator based interface. 2794 2795* | ``Value *getOperand(unsigned i)`` 2796 | ``unsigned getNumOperands()`` 2797 2798 These two methods expose the operands of the ``User`` in a convenient form for 2799 direct access. 2800 2801* | ``User::op_iterator`` - Typedef for iterator over the operand list 2802 | ``op_iterator op_begin()`` - Get an iterator to the start of the operand 2803 list. 2804 | ``op_iterator op_end()`` - Get an iterator to the end of the operand list. 2805 2806 Together, these methods make up the iterator based interface to the operands 2807 of a ``User``. 2808 2809 2810.. _Instruction: 2811 2812The ``Instruction`` class 2813------------------------- 2814 2815``#include "llvm/IR/Instruction.h"`` 2816 2817header source: `Instruction.h 2818<http://llvm.org/doxygen/Instruction_8h-source.html>`_ 2819 2820doxygen info: `Instruction Class 2821<http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_ 2822 2823Superclasses: User_, Value_ 2824 2825The ``Instruction`` class is the common base class for all LLVM instructions. 2826It provides only a few methods, but is a very commonly used class. The primary 2827data tracked by the ``Instruction`` class itself is the opcode (instruction 2828type) and the parent BasicBlock_ the ``Instruction`` is embedded into. To 2829represent a specific type of instruction, one of many subclasses of 2830``Instruction`` are used. 2831 2832Because the ``Instruction`` class subclasses the User_ class, its operands can 2833be accessed in the same way as for other ``User``\ s (with the 2834``getOperand()``/``getNumOperands()`` and ``op_begin()``/``op_end()`` methods). 2835An important file for the ``Instruction`` class is the ``llvm/Instruction.def`` 2836file. This file contains some meta-data about the various different types of 2837instructions in LLVM. It describes the enum values that are used as opcodes 2838(for example ``Instruction::Add`` and ``Instruction::ICmp``), as well as the 2839concrete sub-classes of ``Instruction`` that implement the instruction (for 2840example BinaryOperator_ and CmpInst_). Unfortunately, the use of macros in this 2841file confuses doxygen, so these enum values don't show up correctly in the 2842`doxygen output <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_. 2843 2844.. _s_Instruction: 2845 2846Important Subclasses of the ``Instruction`` class 2847^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 2848 2849.. _BinaryOperator: 2850 2851* ``BinaryOperator`` 2852 2853 This subclasses represents all two operand instructions whose operands must be 2854 the same type, except for the comparison instructions. 2855 2856.. _CastInst: 2857 2858* ``CastInst`` 2859 This subclass is the parent of the 12 casting instructions. It provides 2860 common operations on cast instructions. 2861 2862.. _CmpInst: 2863 2864* ``CmpInst`` 2865 2866 This subclass respresents the two comparison instructions, 2867 `ICmpInst <LangRef.html#i_icmp>`_ (integer opreands), and 2868 `FCmpInst <LangRef.html#i_fcmp>`_ (floating point operands). 2869 2870.. _TerminatorInst: 2871 2872* ``TerminatorInst`` 2873 2874 This subclass is the parent of all terminator instructions (those which can 2875 terminate a block). 2876 2877.. _m_Instruction: 2878 2879Important Public Members of the ``Instruction`` class 2880^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 2881 2882* ``BasicBlock *getParent()`` 2883 2884 Returns the BasicBlock_ that this 2885 ``Instruction`` is embedded into. 2886 2887* ``bool mayWriteToMemory()`` 2888 2889 Returns true if the instruction writes to memory, i.e. it is a ``call``, 2890 ``free``, ``invoke``, or ``store``. 2891 2892* ``unsigned getOpcode()`` 2893 2894 Returns the opcode for the ``Instruction``. 2895 2896* ``Instruction *clone() const`` 2897 2898 Returns another instance of the specified instruction, identical in all ways 2899 to the original except that the instruction has no parent (i.e. it's not 2900 embedded into a BasicBlock_), and it has no name. 2901 2902.. _Constant: 2903 2904The ``Constant`` class and subclasses 2905------------------------------------- 2906 2907Constant represents a base class for different types of constants. It is 2908subclassed by ConstantInt, ConstantArray, etc. for representing the various 2909types of Constants. GlobalValue_ is also a subclass, which represents the 2910address of a global variable or function. 2911 2912.. _s_Constant: 2913 2914Important Subclasses of Constant 2915^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 2916 2917* ConstantInt : This subclass of Constant represents an integer constant of 2918 any width. 2919 2920 * ``const APInt& getValue() const``: Returns the underlying 2921 value of this constant, an APInt value. 2922 2923 * ``int64_t getSExtValue() const``: Converts the underlying APInt value to an 2924 int64_t via sign extension. If the value (not the bit width) of the APInt 2925 is too large to fit in an int64_t, an assertion will result. For this 2926 reason, use of this method is discouraged. 2927 2928 * ``uint64_t getZExtValue() const``: Converts the underlying APInt value 2929 to a uint64_t via zero extension. IF the value (not the bit width) of the 2930 APInt is too large to fit in a uint64_t, an assertion will result. For this 2931 reason, use of this method is discouraged. 2932 2933 * ``static ConstantInt* get(const APInt& Val)``: Returns the ConstantInt 2934 object that represents the value provided by ``Val``. The type is implied 2935 as the IntegerType that corresponds to the bit width of ``Val``. 2936 2937 * ``static ConstantInt* get(const Type *Ty, uint64_t Val)``: Returns the 2938 ConstantInt object that represents the value provided by ``Val`` for integer 2939 type ``Ty``. 2940 2941* ConstantFP : This class represents a floating point constant. 2942 2943 * ``double getValue() const``: Returns the underlying value of this constant. 2944 2945* ConstantArray : This represents a constant array. 2946 2947 * ``const std::vector<Use> &getValues() const``: Returns a vector of 2948 component constants that makeup this array. 2949 2950* ConstantStruct : This represents a constant struct. 2951 2952 * ``const std::vector<Use> &getValues() const``: Returns a vector of 2953 component constants that makeup this array. 2954 2955* GlobalValue : This represents either a global variable or a function. In 2956 either case, the value is a constant fixed address (after linking). 2957 2958.. _GlobalValue: 2959 2960The ``GlobalValue`` class 2961------------------------- 2962 2963``#include "llvm/IR/GlobalValue.h"`` 2964 2965header source: `GlobalValue.h 2966<http://llvm.org/doxygen/GlobalValue_8h-source.html>`_ 2967 2968doxygen info: `GlobalValue Class 2969<http://llvm.org/doxygen/classllvm_1_1GlobalValue.html>`_ 2970 2971Superclasses: Constant_, User_, Value_ 2972 2973Global values ( GlobalVariable_\ s or :ref:`Function <c_Function>`\ s) are the 2974only LLVM values that are visible in the bodies of all :ref:`Function 2975<c_Function>`\ s. Because they are visible at global scope, they are also 2976subject to linking with other globals defined in different translation units. 2977To control the linking process, ``GlobalValue``\ s know their linkage rules. 2978Specifically, ``GlobalValue``\ s know whether they have internal or external 2979linkage, as defined by the ``LinkageTypes`` enumeration. 2980 2981If a ``GlobalValue`` has internal linkage (equivalent to being ``static`` in C), 2982it is not visible to code outside the current translation unit, and does not 2983participate in linking. If it has external linkage, it is visible to external 2984code, and does participate in linking. In addition to linkage information, 2985``GlobalValue``\ s keep track of which Module_ they are currently part of. 2986 2987Because ``GlobalValue``\ s are memory objects, they are always referred to by 2988their **address**. As such, the Type_ of a global is always a pointer to its 2989contents. It is important to remember this when using the ``GetElementPtrInst`` 2990instruction because this pointer must be dereferenced first. For example, if 2991you have a ``GlobalVariable`` (a subclass of ``GlobalValue)`` that is an array 2992of 24 ints, type ``[24 x i32]``, then the ``GlobalVariable`` is a pointer to 2993that array. Although the address of the first element of this array and the 2994value of the ``GlobalVariable`` are the same, they have different types. The 2995``GlobalVariable``'s type is ``[24 x i32]``. The first element's type is 2996``i32.`` Because of this, accessing a global value requires you to dereference 2997the pointer with ``GetElementPtrInst`` first, then its elements can be accessed. 2998This is explained in the `LLVM Language Reference Manual 2999<LangRef.html#globalvars>`_. 3000 3001.. _m_GlobalValue: 3002 3003Important Public Members of the ``GlobalValue`` class 3004^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 3005 3006* | ``bool hasInternalLinkage() const`` 3007 | ``bool hasExternalLinkage() const`` 3008 | ``void setInternalLinkage(bool HasInternalLinkage)`` 3009 3010 These methods manipulate the linkage characteristics of the ``GlobalValue``. 3011 3012* ``Module *getParent()`` 3013 3014 This returns the Module_ that the 3015 GlobalValue is currently embedded into. 3016 3017.. _c_Function: 3018 3019The ``Function`` class 3020---------------------- 3021 3022``#include "llvm/IR/Function.h"`` 3023 3024header source: `Function.h <http://llvm.org/doxygen/Function_8h-source.html>`_ 3025 3026doxygen info: `Function Class 3027<http://llvm.org/doxygen/classllvm_1_1Function.html>`_ 3028 3029Superclasses: GlobalValue_, Constant_, User_, Value_ 3030 3031The ``Function`` class represents a single procedure in LLVM. It is actually 3032one of the more complex classes in the LLVM hierarchy because it must keep track 3033of a large amount of data. The ``Function`` class keeps track of a list of 3034BasicBlock_\ s, a list of formal Argument_\ s, and a SymbolTable_. 3035 3036The list of BasicBlock_\ s is the most commonly used part of ``Function`` 3037objects. The list imposes an implicit ordering of the blocks in the function, 3038which indicate how the code will be laid out by the backend. Additionally, the 3039first BasicBlock_ is the implicit entry node for the ``Function``. It is not 3040legal in LLVM to explicitly branch to this initial block. There are no implicit 3041exit nodes, and in fact there may be multiple exit nodes from a single 3042``Function``. If the BasicBlock_ list is empty, this indicates that the 3043``Function`` is actually a function declaration: the actual body of the function 3044hasn't been linked in yet. 3045 3046In addition to a list of BasicBlock_\ s, the ``Function`` class also keeps track 3047of the list of formal Argument_\ s that the function receives. This container 3048manages the lifetime of the Argument_ nodes, just like the BasicBlock_ list does 3049for the BasicBlock_\ s. 3050 3051The SymbolTable_ is a very rarely used LLVM feature that is only used when you 3052have to look up a value by name. Aside from that, the SymbolTable_ is used 3053internally to make sure that there are not conflicts between the names of 3054Instruction_\ s, BasicBlock_\ s, or Argument_\ s in the function body. 3055 3056Note that ``Function`` is a GlobalValue_ and therefore also a Constant_. The 3057value of the function is its address (after linking) which is guaranteed to be 3058constant. 3059 3060.. _m_Function: 3061 3062Important Public Members of the ``Function`` 3063^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 3064 3065* ``Function(const FunctionType *Ty, LinkageTypes Linkage, 3066 const std::string &N = "", Module* Parent = 0)`` 3067 3068 Constructor used when you need to create new ``Function``\ s to add the 3069 program. The constructor must specify the type of the function to create and 3070 what type of linkage the function should have. The FunctionType_ argument 3071 specifies the formal arguments and return value for the function. The same 3072 FunctionType_ value can be used to create multiple functions. The ``Parent`` 3073 argument specifies the Module in which the function is defined. If this 3074 argument is provided, the function will automatically be inserted into that 3075 module's list of functions. 3076 3077* ``bool isDeclaration()`` 3078 3079 Return whether or not the ``Function`` has a body defined. If the function is 3080 "external", it does not have a body, and thus must be resolved by linking with 3081 a function defined in a different translation unit. 3082 3083* | ``Function::iterator`` - Typedef for basic block list iterator 3084 | ``Function::const_iterator`` - Typedef for const_iterator. 3085 | ``begin()``, ``end()``, ``size()``, ``empty()`` 3086 3087 These are forwarding methods that make it easy to access the contents of a 3088 ``Function`` object's BasicBlock_ list. 3089 3090* ``Function::BasicBlockListType &getBasicBlockList()`` 3091 3092 Returns the list of BasicBlock_\ s. This is necessary to use when you need to 3093 update the list or perform a complex action that doesn't have a forwarding 3094 method. 3095 3096* | ``Function::arg_iterator`` - Typedef for the argument list iterator 3097 | ``Function::const_arg_iterator`` - Typedef for const_iterator. 3098 | ``arg_begin()``, ``arg_end()``, ``arg_size()``, ``arg_empty()`` 3099 3100 These are forwarding methods that make it easy to access the contents of a 3101 ``Function`` object's Argument_ list. 3102 3103* ``Function::ArgumentListType &getArgumentList()`` 3104 3105 Returns the list of Argument_. This is necessary to use when you need to 3106 update the list or perform a complex action that doesn't have a forwarding 3107 method. 3108 3109* ``BasicBlock &getEntryBlock()`` 3110 3111 Returns the entry ``BasicBlock`` for the function. Because the entry block 3112 for the function is always the first block, this returns the first block of 3113 the ``Function``. 3114 3115* | ``Type *getReturnType()`` 3116 | ``FunctionType *getFunctionType()`` 3117 3118 This traverses the Type_ of the ``Function`` and returns the return type of 3119 the function, or the FunctionType_ of the actual function. 3120 3121* ``SymbolTable *getSymbolTable()`` 3122 3123 Return a pointer to the SymbolTable_ for this ``Function``. 3124 3125.. _GlobalVariable: 3126 3127The ``GlobalVariable`` class 3128---------------------------- 3129 3130``#include "llvm/IR/GlobalVariable.h"`` 3131 3132header source: `GlobalVariable.h 3133<http://llvm.org/doxygen/GlobalVariable_8h-source.html>`_ 3134 3135doxygen info: `GlobalVariable Class 3136<http://llvm.org/doxygen/classllvm_1_1GlobalVariable.html>`_ 3137 3138Superclasses: GlobalValue_, Constant_, User_, Value_ 3139 3140Global variables are represented with the (surprise surprise) ``GlobalVariable`` 3141class. Like functions, ``GlobalVariable``\ s are also subclasses of 3142GlobalValue_, and as such are always referenced by their address (global values 3143must live in memory, so their "name" refers to their constant address). See 3144GlobalValue_ for more on this. Global variables may have an initial value 3145(which must be a Constant_), and if they have an initializer, they may be marked 3146as "constant" themselves (indicating that their contents never change at 3147runtime). 3148 3149.. _m_GlobalVariable: 3150 3151Important Public Members of the ``GlobalVariable`` class 3152^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 3153 3154* ``GlobalVariable(const Type *Ty, bool isConstant, LinkageTypes &Linkage, 3155 Constant *Initializer = 0, const std::string &Name = "", Module* Parent = 0)`` 3156 3157 Create a new global variable of the specified type. If ``isConstant`` is true 3158 then the global variable will be marked as unchanging for the program. The 3159 Linkage parameter specifies the type of linkage (internal, external, weak, 3160 linkonce, appending) for the variable. If the linkage is InternalLinkage, 3161 WeakAnyLinkage, WeakODRLinkage, LinkOnceAnyLinkage or LinkOnceODRLinkage, then 3162 the resultant global variable will have internal linkage. AppendingLinkage 3163 concatenates together all instances (in different translation units) of the 3164 variable into a single variable but is only applicable to arrays. See the 3165 `LLVM Language Reference <LangRef.html#modulestructure>`_ for further details 3166 on linkage types. Optionally an initializer, a name, and the module to put 3167 the variable into may be specified for the global variable as well. 3168 3169* ``bool isConstant() const`` 3170 3171 Returns true if this is a global variable that is known not to be modified at 3172 runtime. 3173 3174* ``bool hasInitializer()`` 3175 3176 Returns true if this ``GlobalVariable`` has an intializer. 3177 3178* ``Constant *getInitializer()`` 3179 3180 Returns the initial value for a ``GlobalVariable``. It is not legal to call 3181 this method if there is no initializer. 3182 3183.. _BasicBlock: 3184 3185The ``BasicBlock`` class 3186------------------------ 3187 3188``#include "llvm/IR/BasicBlock.h"`` 3189 3190header source: `BasicBlock.h 3191<http://llvm.org/doxygen/BasicBlock_8h-source.html>`_ 3192 3193doxygen info: `BasicBlock Class 3194<http://llvm.org/doxygen/classllvm_1_1BasicBlock.html>`_ 3195 3196Superclass: Value_ 3197 3198This class represents a single entry single exit section of the code, commonly 3199known as a basic block by the compiler community. The ``BasicBlock`` class 3200maintains a list of Instruction_\ s, which form the body of the block. Matching 3201the language definition, the last element of this list of instructions is always 3202a terminator instruction (a subclass of the TerminatorInst_ class). 3203 3204In addition to tracking the list of instructions that make up the block, the 3205``BasicBlock`` class also keeps track of the :ref:`Function <c_Function>` that 3206it is embedded into. 3207 3208Note that ``BasicBlock``\ s themselves are Value_\ s, because they are 3209referenced by instructions like branches and can go in the switch tables. 3210``BasicBlock``\ s have type ``label``. 3211 3212.. _m_BasicBlock: 3213 3214Important Public Members of the ``BasicBlock`` class 3215^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 3216 3217* ``BasicBlock(const std::string &Name = "", Function *Parent = 0)`` 3218 3219 The ``BasicBlock`` constructor is used to create new basic blocks for 3220 insertion into a function. The constructor optionally takes a name for the 3221 new block, and a :ref:`Function <c_Function>` to insert it into. If the 3222 ``Parent`` parameter is specified, the new ``BasicBlock`` is automatically 3223 inserted at the end of the specified :ref:`Function <c_Function>`, if not 3224 specified, the BasicBlock must be manually inserted into the :ref:`Function 3225 <c_Function>`. 3226 3227* | ``BasicBlock::iterator`` - Typedef for instruction list iterator 3228 | ``BasicBlock::const_iterator`` - Typedef for const_iterator. 3229 | ``begin()``, ``end()``, ``front()``, ``back()``, 3230 ``size()``, ``empty()`` 3231 STL-style functions for accessing the instruction list. 3232 3233 These methods and typedefs are forwarding functions that have the same 3234 semantics as the standard library methods of the same names. These methods 3235 expose the underlying instruction list of a basic block in a way that is easy 3236 to manipulate. To get the full complement of container operations (including 3237 operations to update the list), you must use the ``getInstList()`` method. 3238 3239* ``BasicBlock::InstListType &getInstList()`` 3240 3241 This method is used to get access to the underlying container that actually 3242 holds the Instructions. This method must be used when there isn't a 3243 forwarding function in the ``BasicBlock`` class for the operation that you 3244 would like to perform. Because there are no forwarding functions for 3245 "updating" operations, you need to use this if you want to update the contents 3246 of a ``BasicBlock``. 3247 3248* ``Function *getParent()`` 3249 3250 Returns a pointer to :ref:`Function <c_Function>` the block is embedded into, 3251 or a null pointer if it is homeless. 3252 3253* ``TerminatorInst *getTerminator()`` 3254 3255 Returns a pointer to the terminator instruction that appears at the end of the 3256 ``BasicBlock``. If there is no terminator instruction, or if the last 3257 instruction in the block is not a terminator, then a null pointer is returned. 3258 3259.. _Argument: 3260 3261The ``Argument`` class 3262---------------------- 3263 3264This subclass of Value defines the interface for incoming formal arguments to a 3265function. A Function maintains a list of its formal arguments. An argument has 3266a pointer to the parent Function. 3267 3268 3269