This document is meant to highlight some of the important classes and interfaces available in the LLVM source-base. This manual is not @@ -223,24 +240,22 @@ href="/doxygen/InstVisitor_8h-source.html">InstVisitor template.
This section contains general information that is useful if you are working in the LLVM source-base, but that isn't specific to any particular API.
-LLVM makes heavy use of the C++ Standard Template Library (STL), perhaps much more than you are used to, or have seen before. Because of @@ -253,9 +268,9 @@ can get, so it will not be discussed in this document.
Here we highlight some LLVM APIs that are generally useful and good to know about when writing transformations.
-The LLVM source-base makes extensive use of a custom form of RTTI. These templates have many similarities to the C++ dynamic_cast<> @@ -423,12 +436,111 @@ are lots of examples in the LLVM source base.
Although LLVM generally does not do much string manipulation, we do have +several important APIs which take strings. Two important examples are the +Value class -- which has names for instructions, functions, etc. -- and the +StringMap class which is used extensively in LLVM and Clang.
+ +These are generic classes, and they need to be able to accept strings which +may have embedded null characters. Therefore, they cannot simply take +a const char *, and taking a const std::string& requires +clients to perform a heap allocation which is usually unnecessary. Instead, +many LLVM APIs use a StringRef or a const Twine& for +passing strings efficiently.
+ + +The StringRef data type represents a reference to a constant string +(a character array and a length) and supports the common operations available +on std:string, but does not require heap allocation.
+ +It can be implicitly constructed using a C style null-terminated string, +an std::string, or explicitly with a character pointer and length. +For example, the StringRef find function is declared as:
+ ++ iterator find(StringRef Key); ++ +
and clients can call it using any one of:
+ +
+ Map.find("foo"); // Lookup "foo"
+ Map.find(std::string("bar")); // Lookup "bar"
+ Map.find(StringRef("\0baz", 4)); // Lookup "\0baz"
+
+
+Similarly, APIs which need to return a string may return a StringRef +instance, which can be used directly or converted to an std::string +using the str member function. See +"llvm/ADT/StringRef.h" +for more information.
+ +You should rarely use the StringRef class directly, because it contains +pointers to external memory it is not generally safe to store an instance of the +class (unless you know that the external storage will not be freed). StringRef is +small and pervasive enough in LLVM that it should always be passed by value.
+The Twine class is an efficient way for APIs to accept concatenated +strings. For example, a common LLVM paradigm is to name one instruction based on +the name of another instruction with a suffix, for example:
+ ++ New = CmpInst::Create(..., SO->getName() + ".cmp"); ++
The Twine class is effectively a +lightweight rope +which points to temporary (stack allocated) objects. Twines can be implicitly +constructed as the result of the plus operator applied to strings (i.e., a C +strings, an std::string, or a StringRef). The twine delays the +actual concatenation of strings until it is actually required, at which point +it can be efficiently rendered directly into a character array. This avoids +unnecessary heap allocation involved in constructing the temporary results of +string concatenation. See +"llvm/ADT/Twine.h" +for more information.
+ +As with a StringRef, Twine objects point to external memory +and should almost never be stored or mentioned directly. They are intended +solely for use when defining a function which should be able to efficiently +accept concatenated strings.
+ +Often when working on your pass you will put a bunch of debugging printouts and other code into your pass. After you get it working, you want to remove @@ -447,7 +559,7 @@ tool) is run with the '-debug' command line argument:
-DOUT << "I am here!\n"; +DEBUG(errs() << "I am here!\n");
Sometimes you may find yourself in a situation where enabling -debug just turns on too much information (such as when working on the code @@ -492,16 +602,16 @@ option as follows:
-DOUT << "No debug type\n"; #undef DEBUG_TYPE +DEBUG(errs() << "No debug type\n"); #define DEBUG_TYPE "foo" -DOUT << "'foo' debug type\n"; +DEBUG(errs() << "'foo' debug type\n"); #undef DEBUG_TYPE #define DEBUG_TYPE "bar" -DOUT << "'bar' debug type\n"; +DEBUG(errs() << "'bar' debug type\n")); #undef DEBUG_TYPE #define DEBUG_TYPE "" -DOUT << "No debug type (2)\n"; +DEBUG(errs() << "No debug type (2)\n");
The DEBUG_WITH_TYPE macro is also available for situations where you +would like to set DEBUG_TYPE, but only for one specific DEBUG +statement. It takes an additional first parameter, which is the type to use. For +example, the preceding example could be written as:
+ + +
+DEBUG_WITH_TYPE("", errs() << "No debug type\n");
+DEBUG_WITH_TYPE("foo", errs() << "'foo' debug type\n");
+DEBUG_WITH_TYPE("bar", errs() << "'bar' debug type\n"));
+DEBUG_WITH_TYPE("", errs() << "No debug type (2)\n");
+
+The "llvm/ADT/Statistic.h" file @@ -636,11 +763,11 @@ maintainable and useful.
Several of the important data structures in LLVM are graphs: for example CFGs made out of LLVM BasicBlocks, CFGs made out of @@ -682,15 +809,21 @@ found at Graph Attributes.) If you want to restart and clear all the current graph attributes, then you can call DAG.clearGraphAttrs().
+Note that graph visualization features are compiled out of Release builds +to reduce file size. This means that you need a Debug+Asserts or +Release+Asserts build to use these features.
+ +LLVM has a plethora of data structures in the llvm/ADT/ directory, and we commonly use STL data structures. This section describes the trade-offs @@ -725,6 +858,10 @@ access the container. Based on that, you should use:
iteration, but do not support efficient look-up based on a key. +The llvm::ArrayRef class is the preferred class to use in an interface that + accepts a sequential list of elements in memory and just reads from them. By + taking an ArrayRef, the API can be passed a fixed size array, an std::vector, + an llvm::SmallVector and anything else that is contiguous in memory. +
Fixed size arrays are very simple and very fast. They are good if you know exactly how many elements you have, or you have a (low) upper bound on how many you have.
Heap allocated arrays (new[] + delete[]) are also simple. They are good if the number of elements is variable, if you know how many elements you will need before the array is allocated, and if the array is usually large (if not, @@ -782,11 +934,27 @@ construct those elements actually used).
TinyPtrVector<Type> is a highly specialized collection class +that is optimized to avoid allocation in the case when a vector has zero or one +elements. It has two major restrictions: 1) it can only hold values of pointer +type, and 2) it cannot hold a null pointer.
+ +Since this container is highly specialized, it is rarely used.
+SmallVector<Type, N> is a simple class that looks and smells just like vector<Type>: it supports efficient iteration, lays out elements in memory order (so you can @@ -811,11 +979,11 @@ SmallVectors are most useful when on the stack.
std::vector is well loved and respected. It is useful when SmallVector isn't: when the size of the vector is often large (thus the small optimization will @@ -853,11 +1021,11 @@ the loop.
std::deque is, in some senses, a generalized version of std::vector. Like std::vector, it provides constant time random access and other similar properties, but it also provides efficient access to the front of the list. It @@ -869,11 +1037,11 @@ something cheaper.
std::list is an extremely inefficient class that is rarely useful. It performs a heap allocation for every element inserted into it, thus having an extremely high constant factor, particularly for small data types. std::list @@ -887,11 +1055,11 @@ not invalidate iterator or pointers to other elements in the list.
ilist<T> implements an 'intrusive' doubly-linked list. It is intrusive, because it requires the element to store and provide access to the prev/next pointers for the list.
@@ -917,22 +1085,60 @@ Related classes of interest are explained in the following subsections:+Useful for storing a vector of values using only a few number of bits for each +value. Apart from the standard operations of a vector-like container, it can +also perform an 'or' set operation. +
+ +For example:
+ +
+enum State {
+ None = 0x0,
+ FirstCondition = 0x1,
+ SecondCondition = 0x2,
+ Both = 0x3
+};
+
+State get() {
+ PackedVector<State, 2> Vec1;
+ Vec1.push_back(FirstCondition);
+
+ PackedVector<State, 2> Vec2;
+ Vec2.push_back(SecondCondition);
+
+ Vec1 |= Vec2;
+ return Vec1[0]; // returns 'Both'.
+}
+
ilist_traits<T> is ilist<T>'s customization mechanism. iplist<T> (and consequently ilist<T>) publicly derive from this traits class.
iplist<T> is ilist<T>'s base and as such supports a slightly narrower interface. Notably, inserters from T& are absent.
@@ -942,11 +1148,11 @@ used for a wide variety of customizations.ilist_node<T> implements a the forward and backward links that are expected by the ilist<T> (and analogous containers) in the default manner.
@@ -957,12 +1163,12 @@ in the default manner.ilists have another speciality that must be considered. To be a good +
ilists have another specialty that must be considered. To be a good citizen in the C++ ecosystem, it needs to support the standard container operations, such as begin and end iterators, etc. Also, the operator-- must work correctly on the end iterator in the @@ -995,11 +1201,11 @@ field in the ghostly sentinel which can be legally accessed.
Other STL containers are available, such as std::string.
There are also various STL adapter classes such as std::queue, @@ -1007,28 +1213,204 @@ std::priority_queue, std::stack, etc. These provide simplified access to an underlying container but don't affect the cost of the container itself.
+There are a variety of ways to pass around and use strings in C and C++, and +LLVM adds a few new options to choose from. Pick the first option on this list +that will do what you need, they are ordered according to their relative cost. +
++Note that is is generally preferred to not pass strings around as +"const char*"'s. These have a number of problems, including the fact +that they cannot represent embedded nul ("\0") characters, and do not have a +length available efficiently. The general replacement for 'const +char*' is StringRef. +
+ +For more information on choosing string containers for APIs, please see +Passing strings.
+ + + ++The StringRef class is a simple value class that contains a pointer to a +character and a length, and is quite related to the ArrayRef class (but specialized for arrays of +characters). Because StringRef carries a length with it, it safely handles +strings with embedded nul characters in it, getting the length does not require +a strlen call, and it even has very convenient APIs for slicing and dicing the +character range that it represents. +
+ ++StringRef is ideal for passing simple strings around that are known to be live, +either because they are C string literals, std::string, a C array, or a +SmallVector. Each of these cases has an efficient implicit conversion to +StringRef, which doesn't result in a dynamic strlen being executed. +
+ +StringRef has a few major limitations which make more powerful string +containers useful:
+ +Because of its strengths and limitations, it is very common for a function to +take a StringRef and for a method on an object to return a StringRef that +points into some string that it owns.
+ ++ The Twine class is used as an intermediary datatype for APIs that want to take + a string that can be constructed inline with a series of concatenations. + Twine works by forming recursive instances of the Twine datatype (a simple + value object) on the stack as temporary objects, linking them together into a + tree which is then linearized when the Twine is consumed. Twine is only safe + to use as the argument to a function, and should always be a const reference, + e.g.: +
+ ++ void foo(const Twine &T); + ... + StringRef X = ... + unsigned i = ... + foo(X + "." + Twine(i)); ++ +
This example forms a string like "blarg.42" by concatenating the values + together, and does not form intermediate strings containing "blarg" or + "blarg.". +
+ +Because Twine is constructed with temporary objects on the stack, and + because these instances are destroyed at the end of the current statement, + it is an inherently dangerous API. For example, this simple variant contains + undefined behavior and will probably crash:
+ ++ void foo(const Twine &T); + ... + StringRef X = ... + unsigned i = ... + const Twine &Tmp = X + "." + Twine(i); + foo(Tmp); ++ +
... because the temporaries are destroyed before the call. That said, + Twine's are much more efficient than intermediate std::string temporaries, and + they work really well with StringRef. Just be aware of their limitations.
+ +SmallString is a subclass of SmallVector that +adds some convenience APIs like += that takes StringRef's. SmallString avoids +allocating memory in the case when the preallocated space is enough to hold its +data, and it calls back to general heap allocation when required. Since it owns +its data, it is very safe to use and supports full mutation of the string.
+ +Like SmallVector's, the big downside to SmallString is their sizeof. While +they are optimized for small strings, they themselves are not particularly +small. This means that they work great for temporary scratch buffers on the +stack, but should not generally be put into the heap: it is very rare to +see a SmallString as the member of a frequently-allocated heap data structure +or returned by-value. +
+ +The standard C++ std::string class is a very general class that (like + SmallString) owns its underlying data. sizeof(std::string) is very reasonable + so it can be embedded into heap data structures and returned by-value. + On the other hand, std::string is highly inefficient for inline editing (e.g. + concatenating a bunch of stuff together) and because it is provided by the + standard library, its performance characteristics depend a lot of the host + standard library (e.g. libc++ and MSVC provide a highly optimized string + class, GCC contains a really slow implementation). +
+ +The major disadvantage of std::string is that almost every operation that + makes them larger can allocate memory, which is slow. As such, it is better + to use SmallVector or Twine as a scratch buffer, but then use std::string to + persist the result.
+ + +Set-like containers are useful when you need to canonicalize multiple values into a single representation. There are several different choices for how to do this, providing various trade-offs.
-If you intend to insert a lot of elements, then do a lot of queries, a great approach is to use a vector (or other sequential container) with @@ -1046,11 +1428,11 @@ efficiently queried with a standard binary or radix search.
If you have a set-like data structure that is usually small and whose elements are reasonably small, a SmallSet<Type, N> is a good choice. This set @@ -1069,31 +1451,31 @@ and erasing, but does not support iteration.
SmallPtrSet has all the advantages of SmallSet (and a SmallSet of pointers is -transparently implemented with a SmallPtrSet), but also supports iterators. If +
SmallPtrSet has all the advantages of SmallSet (and a SmallSet of pointers is +transparently implemented with a SmallPtrSet), but also supports iterators. If more than 'N' insertions are performed, a single quadratically probed hash table is allocated and grows as needed, providing extremely efficient access (constant time insertion/deleting/queries with low constant factors) and is very stingy with malloc traffic.
-Note that, unlike std::set, the iterators of SmallPtrSet are invalidated +
Note that, unlike std::set, the iterators of SmallPtrSet are invalidated whenever an insertion occurs. Also, the values visited by the iterators are not visited in sorted order.
DenseSet is a simple quadratically probed hash table. It excels at supporting @@ -1108,11 +1490,11 @@ href="#dss_densemap">DenseMap has.
FoldingSet is an aggregate class that is really good at uniquing @@ -1145,11 +1527,11 @@ elements.
std::set is a reasonable all-around set class, which is decent at many things but great at nothing. std::set allocates memory for each element @@ -1170,11 +1552,11 @@ std::set is almost never a good choice.
LLVM's SetVector<Type> is an adapter class that combines your choice of a set-like container along with a Sequential Container. The important property @@ -1210,11 +1592,11 @@ heap traffic.
UniqueVector is similar to SetVector, but it @@ -1230,11 +1612,11 @@ factors, and produces a lot of malloc traffic. It should be avoided.
- + -The STL provides several other options, such as std::multiset and the various @@ -1250,22 +1632,23 @@ better.
If your usage pattern follows a strict insert-then-query approach, you can @@ -1278,11 +1661,11 @@ vectors for sets.
Strings are commonly used as keys in maps, and they are difficult to support @@ -1301,7 +1684,7 @@ to the key string for a value.
The StringMap is very fast for several reasons: quadratic probing is very cache efficient for lookups, the hash value of strings in buckets is not -recomputed when lookup up an element, StringMap rarely has to touch the +recomputed when looking up an element, StringMap rarely has to touch the memory for unrelated objects when looking up a value (even when hash collisions happen), hash table growth does not recompute the hash values for strings already in the table, and each pair in the map is store in a single allocation @@ -1312,11 +1695,11 @@ copies a string if a value is inserted into the table.
IndexedMap is a specialized container for mapping small dense integers (or values that can be mapped to small dense integers) to some other type. It is @@ -1332,11 +1715,11 @@ virtual register ID).
DenseMap is a simple quadratically probed hash table. It excels at supporting @@ -1358,11 +1741,45 @@ inserted into the map) that it needs internally.
+ValueMap is a wrapper around a DenseMap mapping
+Value*s (or subclasses) to another type. When a Value is deleted or RAUW'ed,
+ValueMap will update itself so the new version of the key is mapped to the same
+value, just as if the key were a WeakVH. You can configure exactly how this
+happens, and what else happens on these two events, by passing
+a Config parameter to the ValueMap template.
IntervalMap is a compact map for small keys and values. It maps key +intervals instead of single keys, and it will automatically coalesce adjacent +intervals. When then map only contains a few intervals, they are stored in the +map object itself to avoid allocations.
+ +The IntervalMap iterators are quite big, so they should not be passed around +as STL iterators. The heavyweight iterators allow a smaller data structure.
+ +std::map has similar characteristics to std::set: it uses @@ -1378,11 +1795,31 @@ another element takes place).
IntEqClasses provides a compact representation of equivalence classes of +small integers. Initially, each integer in the range 0..n-1 has its own +equivalence class. Classes can be joined by passing two class representatives to +the join(a, b) method. Two integers are in the same class when findLeader() +returns the same representative.
+ +Once all equivalence classes are formed, the map can be compressed so each +integer 0..n-1 maps to an equivalence class number in the range 0..m-1, where m +is the total number of equivalence classes. The map must be uncompressed before +it can be edited again.
+The STL provides several other options, such as std::multimap and the various @@ -1396,12 +1833,14 @@ always better.
Unlike the other containers, there are only two bit storage containers, and choosing when to use each is relatively straightforward.
@@ -1411,15 +1850,14 @@ implementation in many common compilers (e.g. commonly available versions of GCC) is extremely inefficient and 2) the C++ standards committee is likely to deprecate this container and/or change it significantly somehow. In any case, please don't use it. -The BitVector container provides a fixed size set of bits for manipulation. +
The BitVector container provides a dynamic size set of bits for manipulation. It supports individual bit setting/testing, as well as set operations. The set operations take time O(size of bitvector), but operations are performed one word at a time, instead of one bit at a time. This makes the BitVector very fast for @@ -1429,11 +1867,30 @@ the number of set bits to be high (IE a dense set).
The SmallBitVector container provides the same interface as BitVector, but +it is optimized for the case where only a small number of bits, less than +25 or so, are needed. It also transparently supports larger bit counts, but +slightly less efficiently than a plain BitVector, so SmallBitVector should +only be used when larger counts are rare. +
+ ++At this time, SmallBitVector does not support set operations (and, or, xor), +and its operator[] does not provide an assignable lvalue. +
The SparseBitVector container is much like BitVector, with one major difference: Only the bits that are set, are stored. This makes the SparseBitVector much more space efficient than BitVector when the set is sparse, @@ -1443,13 +1900,17 @@ universe). The downside to the SparseBitVector is that setting and testing of r
This section describes how to perform some very simple transformations of LLVM code. This is meant to give examples of common idioms used, showing the @@ -1458,15 +1919,13 @@ you should also read about the main classes that you will be working with. The Core LLVM Class Hierarchy Reference contains details and descriptions of the main classes that you should know about.
-The LLVM compiler infrastructure have many different data structures that may be traversed. Following the example of the C++ standard template library, the @@ -1483,16 +1942,14 @@ on them, and it is easier to remember how to iterate. First we show a few common examples of the data structures that need to be traversed. Other data structures are traversed in very similar ways.
-It's quite common to have a Function instance that you'd like to transform in some way; in particular, you'd like to manipulate its @@ -1507,7 +1964,7 @@ an example that prints the name of a BasicBlock and the number of for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i) // Print out the name of the basic block if it has one, and then the // number of instructions that it contains - llvm::cerr << "Basic block (name=" << i->getName() << ") has " + errs() << "Basic block (name=" << i->getName() << ") has " << i->size() << " instructions.\n";
Just like when dealing with BasicBlocks in Functions, it's easy to iterate over the individual instructions that make up @@ -1540,25 +1997,25 @@ a BasicBlock:
for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i) // The next statement works since operator<<(ostream&,...) // is overloaded for Instruction& - llvm::cerr << *i << "\n"; + errs() << *i << "\n";However, this isn't really the best way to print out the contents of a BasicBlock! Since the ostream operators are overloaded for virtually anything you'll care about, you could have just invoked the print routine on the -basic block itself: llvm::cerr << *blk << "\n";.
+basic block itself: errs() << *blk << "\n";.If you're finding that you commonly iterate over a Function's BasicBlocks and then that BasicBlock's Instructions, @@ -1573,7 +2030,7 @@ small example that shows how to dump all instructions in a function to the stand // F is a pointer to a Function instance for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I) - llvm::cerr << *I << "\n"; + errs() << *I << "\n";
Sometimes, it'll be useful to grab a reference (or pointer) to a class instance when all you've got at hand is an iterator. Well, extracting @@ -1652,20 +2109,35 @@ without actually obtaining it via iteration over some structure:
void printNextInstruction(Instruction* inst) { BasicBlock::iterator it(inst); ++it; // After this line, it refers to the instruction after *inst - if (it != inst->getParent()->end()) llvm::cerr << *it << "\n"; + if (it != inst->getParent()->end()) errs() << *it << "\n"; }Unfortunately, these implicit conversions come at a cost; they prevent +these iterators from conforming to standard iterator conventions, and thus +from being usable with standard algorithms and containers. For example, they +prevent the following code, where B is a BasicBlock, +from compiling:
+ ++ llvm::SmallVector<llvm::Instruction *, 16>(B->begin(), B->end()); ++
Because of this, these implicit conversions may be removed some day, +and operator* changed to return a pointer instead of a reference.
+Say that you're writing a FunctionPass and would like to count all the locations in the entire module (that is, across every Function) where a @@ -1722,11 +2194,11 @@ class OurFunctionPass : public FunctionPass {
You may have noticed that the previous example was a bit oversimplified in that it did not deal with call sites generated by 'invoke' instructions. In @@ -1749,11 +2221,11 @@ If you look at its definition, it has only a single pointer member.
Frequently, we might have an instance of the Value Class and we want to @@ -1770,13 +2242,17 @@ Function *F = ...; for (Value::use_iterator i = F->use_begin(), e = F->use_end(); i != e; ++i) if (Instruction *Inst = dyn_cast<Instruction>(*i)) { - llvm::cerr << "F is used in instruction:\n"; - llvm::cerr << *Inst << "\n"; + errs() << "F is used in instruction:\n"; + errs() << *Inst << "\n"; }
Alternately, it's common to have an instance of the Note that dereferencing a Value::use_iterator is not a very cheap +operation. Instead of performing *i above several times, consider +doing it only once in the loop body and reusing its result.
+ +Alternatively, it's common to have an instance of the User Class and need to know what Values are used by it. The list of all Values used by a User is known as a use-def chain. Instances of class @@ -1795,20 +2271,23 @@ for (User::op_iterator i = pi->op_begin(), e = pi->op_end(); i != e; ++i)
Declaring objects as const is an important tool of enforcing +mutation free algorithms (such as analyses, etc.). For this purpose above +iterators come in constant flavors as Value::const_use_iterator +and Value::const_op_iterator. They automatically arise when +calling use/op_begin() on const Value*s or +const User*s respectively. Upon dereferencing, they return +const Use*s. Otherwise the above patterns remain unchanged.
Iterating over the predecessors and successors of a block is quite easy with the routines defined in "llvm/Support/CFG.h". Just use code like @@ -1831,13 +2310,14 @@ succ_iterator/succ_begin/succ_end.
There are some primitive transformation operations present in the LLVM infrastructure that are worth knowing about. When performing @@ -1845,15 +2325,13 @@ transformations, it's fairly common to manipulate the contents of basic blocks. This section describes some of the common methods for doing so and gives example code.
-Instantiating Instructions
@@ -1989,18 +2467,15 @@ Instruction* newInst = new Instruction(..., pi);Deleting an instruction from an existing sequence of instructions that form a -BasicBlock is very straight-forward. First, -you must have a pointer to the instruction that you wish to delete. Second, you -need to obtain the pointer to that instruction's basic block. You use the -pointer to the basic block to get its list of instructions and then use the -erase function to remove your instruction. For example:
+BasicBlock is very straight-forward: just +call the instruction's eraseFromParent() method. For example:@@ -2009,15 +2484,19 @@ I->eraseFromParent();
This unlinks the instruction from its containing basic block and deletes +it. If you'd just like to unlink the instruction from its containing basic +block but not delete it, you can use the removeFromParent() method.
+Replacing individual instructions
@@ -2025,7 +2504,7 @@ I->eraseFromParent(); permits use of two very useful replace functions: ReplaceInstWithValue and ReplaceInstWithInst. -Deleting a global variable from a module is just as easy as deleting an Instruction. First, you must have a pointer to the global variable that you wish @@ -2099,12 +2578,14 @@ GV->eraseFromParent();
In generating IR, you may need some complex types. If you know these types statically, you can use TypeBuilder<...>::get(), defined @@ -2139,13 +2620,15 @@ comment for more details.
This section describes the interaction of the LLVM APIs with multithreading, both on the part of client applications, and in the JIT, in the hosted @@ -2168,45 +2651,56 @@ compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and using the resultant compiler to build a copy of LLVM with multithreading support.
-In order to properly protect its internal data structures while avoiding -excessive locking overhead in the single-threaded case, the LLVM APIs require -that the client invoke llvm_start_multithreaded(). This call must -complete before any other threads attempt to invoke LLVM APIs. Any -attempts to call LLVM APIs from multiple threads before -llvm_start_multithreaded returns can and will cause corruption of -LLVM's internal data. +excessive locking overhead in the single-threaded case, the LLVM must intialize +certain data structures necessary to provide guards around its internals. To do +so, the client program must invoke llvm_start_multithreaded() before +making any concurrent LLVM API calls. To subsequently tear down these +structures, use the llvm_stop_multithreaded() call. You can also use +the llvm_is_multithreaded() call to check the status of multithreaded +mode.
-A caveat: before llvm_start_multithreaded() has been invoked, all -llvm::sys::Mutex acquisitions and releases will become no-ops. This -means that llvm_start_multithreaded() must be invoked before a threaded -application can be executed in the JIT. +Note that both of these calls must be made in isolation. That is to +say that no other LLVM API calls may be executing at any time during the +execution of llvm_start_multithreaded() or llvm_stop_multithreaded +. It's is the client's responsibility to enforce this isolation. +
+ ++The return value of llvm_start_multithreaded() indicates the success or +failure of the initialization. Failure typically indicates that your copy of +LLVM was built without multithreading support, typically because GCC atomic +intrinsics were not found in your system compiler. In this case, the LLVM API +will not be safe for concurrent calls. However, it will be safe for +hosting threaded applications in the JIT, though care +must be taken to ensure that side exits and the like do not accidentally +result in concurrent LLVM API calls.
When you are done using the LLVM APIs, you should call llvm_shutdown() -to deallocate memory used for internal structures. This call must not begin -while any other threads are still issuing LLVM API calls. Doing so is likely -to result in garbage data or crashes. +to deallocate memory used for internal structures. This will also invoke +llvm_stop_multithreaded() if LLVM is operating in multithreaded mode. +As such, llvm_shutdown() requires the same isolation guarantees as +llvm_stop_multithreaded().
@@ -2216,11 +2710,11 @@ destructor.
ManagedStatic is a utility class in LLVM used to implement static initialization of static resources, such as the global type tables. Before the @@ -2235,201 +2729,115 @@ Note that, because no other threads are allowed to issue LLVM API calls before llvm_start_multithreaded() returns, it is possible to have ManagedStatics of llvm::sys::Mutexs.
--This section describes some of the advanced or obscure API's that most clients -do not need to be aware of. These API's tend manage the inner workings of the -LLVM system, and only need to be accessed in unusual circumstances. +The llvm_acquire_global_lock() and llvm_release_global_lock +APIs provide access to the global lock used to implement the double-checked +locking for lazy initialization. These should only be used internally to LLVM, +and only if you know what you're doing!
-The LLVM type system has a very simple goal: allow clients to compare types for -structural equality with a simple pointer comparison (aka a shallow compare). -This goal makes clients much simpler and faster, and is used throughout the LLVM -system. -
- --Unfortunately achieving this goal is not a simple matter. In particular, -recursive types and late resolution of opaque types makes the situation very -difficult to handle. Fortunately, for the most part, our implementation makes -most clients able to be completely unaware of the nasty internal details. The -primary case where clients are exposed to the inner workings of it are when -building a recursive type. In addition to this case, the LLVM bitcode reader, -assembly parser, and linker also have to be aware of the inner workings of this -system. -
- --For our purposes below, we need three concepts. First, an "Opaque Type" is -exactly as defined in the language -reference. Second an "Abstract Type" is any type which includes an -opaque type as part of its type graph (for example "{ opaque, i32 }"). -Third, a concrete type is a type that is not an abstract type (e.g. "{ i32, -float }"). -
- --Because the most common question is "how do I build a recursive type with LLVM", -we answer it now and explain it as we go. Here we include enough to cause this -to be emitted to an output .ll file: +LLVMContext is an opaque class in the LLVM API which clients can use +to operate multiple, isolated instances of LLVM concurrently within the same +address space. For instance, in a hypothetical compile-server, the compilation +of an individual translation unit is conceptually independent from all the +others, and it would be desirable to be able to compile incoming translation +units concurrently on independent server threads. Fortunately, +LLVMContext exists to enable just this kind of scenario!
-
-%mylist = type { %mylist*, i32 }
-
--To build this, use the following LLVM APIs: -
- --// Create the initial outer struct -PATypeHolder StructTy = OpaqueType::get(); -std::vector<const Type*> Elts; -Elts.push_back(PointerType::getUnqual(StructTy)); -Elts.push_back(Type::Int32Ty); -StructType *NewSTy = StructType::get(Elts); - -// At this point, NewSTy = "{ opaque*, i32 }". Tell VMCore that -// the struct and the opaque type are actually the same. -cast<OpaqueType>(StructTy.get())->refineAbstractTypeTo(NewSTy); - -// NewSTy is potentially invalidated, but StructTy (a PATypeHolder) is -// kept up-to-date -NewSTy = cast<StructType>(StructTy.get()); - -// Add a name for the type to the module symbol table (optional) -MyModule->addTypeName("mylist", NewSTy); --
-This code shows the basic approach used to build recursive types: build a -non-recursive type using 'opaque', then use type unification to close the cycle. -The type unification step is performed by the refineAbstractTypeTo method, which is -described next. After that, we describe the PATypeHolder class. +Conceptually, LLVMContext provides isolation. Every LLVM entity +(Modules, Values, Types, Constants, etc.) +in LLVM's in-memory IR belongs to an LLVMContext. Entities in +different contexts cannot interact with each other: Modules in +different contexts cannot be linked together, Functions cannot be added +to Modules in different contexts, etc. What this means is that is is +safe to compile on multiple threads simultaneously, as long as no two threads +operate on entities within the same context.
--The refineAbstractTypeTo method starts the type unification process. -While this method is actually a member of the DerivedType class, it is most -often used on OpaqueType instances. Type unification is actually a recursive -process. After unification, types can become structurally isomorphic to -existing types, and all duplicates are deleted (to preserve pointer equality). +In practice, very few places in the API require the explicit specification of a +LLVMContext, other than the Type creation/lookup APIs. +Because every Type carries a reference to its owning context, most +other entities can determine what context they belong to by looking at their +own Type. If you are adding new entities to LLVM IR, please try to +maintain this interface design.
-In the example above, the OpaqueType object is definitely deleted. -Additionally, if there is an "{ \2*, i32}" type already created in the system, -the pointer and struct type created are also deleted. Obviously whenever -a type is deleted, any "Type*" pointers in the program are invalidated. As -such, it is safest to avoid having any "Type*" pointers to abstract types -live across a call to refineAbstractTypeTo (note that non-abstract -types can never move or be deleted). To deal with this, the PATypeHolder class is used to maintain a stable -reference to a possibly refined type, and the AbstractTypeUser class is used to update more -complex datastructures. +For clients that do not require the benefits of isolation, LLVM +provides a convenience API getGlobalContext(). This returns a global, +lazily initialized LLVMContext that may be used in situations where +isolation is not a concern.
--PATypeHolder is a form of a "smart pointer" for Type objects. When VMCore -happily goes about nuking types that become isomorphic to existing types, it -automatically updates all PATypeHolder objects to point to the new type. In the -example above, this allows the code to maintain a pointer to the resultant -resolved recursive type, even though the Type*'s are potentially invalidated. +LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple +threads can call ExecutionEngine::getPointerToFunction() or +ExecutionEngine::runFunction() concurrently, and multiple threads can +run code output by the JIT concurrently. The user must still ensure that only +one thread accesses IR in a given LLVMContext while another thread +might be modifying it. One way to do that is to always hold the JIT lock while +accessing IR outside the JIT (the JIT modifies the IR by adding +CallbackVHs). Another way is to only +call getPointerToFunction() from the LLVMContext's thread.
--PATypeHolder is an extremely light-weight object that uses a lazy union-find -implementation to update pointers. For example the pointer from a Value to its -Type is maintained by PATypeHolder objects. +
When the JIT is configured to compile lazily (using +ExecutionEngine::DisableLazyCompilation(false)), there is currently a +race condition in +updating call sites after a function is lazily-jitted. It's still possible to +use the lazy JIT in a threaded program if you ensure that only one thread at a +time can call any particular lazy stub and that the JIT lock guards any IR +access, but we suggest using only the eager JIT in threaded programs.
--Some data structures need more to perform more complex updates when types get -resolved. To support this, a class can derive from the AbstractTypeUser class. -This class -allows it to get callbacks when certain types are resolved. To register to get -callbacks for a particular type, the DerivedType::{add/remove}AbstractTypeUser -methods can be called on a type. Note that these methods only work for - abstract types. Concrete types (those that do not include any opaque -objects) can never be refined. +This section describes some of the advanced or obscure API's that most clients +do not need to be aware of. These API's tend manage the inner workings of the +LLVM system, and only need to be accessed in unusual circumstances.
-The ValueSymbolTable class provides a symbol table that the Function and Module classes use for naming value definitions. The symbol table can provide a name for any Value. -The -TypeSymbolTable class is used by the Module class to store -names for types.
+Note that the SymbolTable class should not be directly accessed by most clients. It should only be used when iteration over the symbol table @@ -2439,24 +2847,23 @@ all LLVM an empty name) do not exist in the symbol table.
-These symbol tables support iteration over the values/types in the symbol +
Symbol tables support iteration over the values in the symbol table with begin/end/iterator and supports querying to see if a specific name is in the symbol table (with lookup). The ValueSymbolTable class exposes no public mutator methods, instead, simply call setName on a value, which will autoinsert it into the -appropriate symbol table. For types, use the Module::addTypeName method to -insert entries into the symbol table.
+appropriate symbol table.The
User class provides a basis for expressing the ownership of User
towards other
@@ -2465,18 +2872,19 @@ Use helper class is employed to do the bookkeeping and to facilitate
-
+
A subclass of User can choose between incorporating its Use objects
or refer to them out-of-line by means of a pointer. A mixed variant
(some Uses inline others hung off) is impractical and breaks the invariant
that the Use objects belonging to the same User form a contiguous array.
We have 2 different layouts in the User (sub)classes:
@@ -2525,17 +2933,18 @@ enforce the following memory layouts:
Since the Use objects are deprived of the direct (back)pointer to
their User objects, there must be a fast and exact method to
recover it. This is accomplished by the following scheme:
The following literate Haskell fragment demonstrates the concept:
To maintain the invariant that the 2 LSBits of each Use** in Use
@@ -2675,13 +3089,17 @@ the LSBit set. (Portability is relying on the fact that all known compilers plac
#include "llvm/Type.h"
Type is a superclass of all type classes. Every Value has
a Type. Type cannot be instantiated directly but only
@@ -2712,24 +3128,20 @@ the lib/VMCore directory. #include "llvm/Module.h" #include "llvm/Value.h"
#include "llvm/User.h" The User class exposes the operand list in two ways: through
an index access interface and through an iterator based interface. #include "llvm/Instruction.h" This subclasses represents all two operand instructions whose operands
@@ -3157,12 +3559,13 @@ this file confuses doxygen, so these enum values don't show up correctly in the
Constant represents a base class for different types of constants. It
is subclassed by ConstantInt, ConstantArray, etc. for representing
@@ -3195,11 +3600,9 @@ the various types of Constants. GlobalValue is also
a subclass, which represents the address of a global variable or function.
#include "llvm/GlobalValue.h" #include "llvm/Function.h" The Function class represents a single procedure in LLVM. It is
-actually one of the more complex classes in the LLVM heirarchy because it must
+actually one of the more complex classes in the LLVM hierarchy because it must
keep track of a large amount of data. The Function class keeps track
of a list of BasicBlocks, a list of formal
Arguments, and a
@@ -3342,7 +3747,7 @@ of a list of BasicBlocks, a list of formal
The list of BasicBlocks is the most
commonly used part of Function objects. The list imposes an implicit
ordering of the blocks in the function, which indicate how the code will be
-layed out by the backend. Additionally, the first BasicBlock is the implicit entry node for the
Function. It is not legal in LLVM to explicitly branch to this initial
block. There are no implicit exit nodes, and in fact there may be multiple exit
@@ -3369,15 +3774,15 @@ href="#Argument">Arguments in the function body. Note that Function is a GlobalValue
and therefore also a Constant. The value of the function
is its address (after linking) which is guaranteed to be constant. #include "llvm/GlobalVariable.h"
@@ -3472,7 +3879,7 @@ Superclasses: GlobalValue,
User,
Value Global variables are represented with the (suprise suprise)
+ Global variables are represented with the (surprise surprise)
GlobalVariable class. Like functions, GlobalVariables are also
subclasses of GlobalValue, and as such are
always referenced by their address (global values must live in memory, so their
@@ -3482,15 +3889,15 @@ variables may have an initial value (which must be a
Constant), and if they have an initializer,
they may be marked as "constant" themselves (indicating that their contents
never change at runtime). #include "llvm/BasicBlock.h" This class represents a single entry multiple exit section of the code,
+ This class represents a single entry single exit section of the code,
commonly known as a basic block by the compiler community. The
BasicBlock class maintains a list of Instructions, which form the body of the block.
@@ -3559,15 +3967,14 @@ href="#Value">Values, because they are referenced by instructions
like branches and can go in the switch tables. BasicBlocks have type
label. This subclass of Value defines the interface for incoming formal
arguments to a function. A Function maintains a list of its formal
@@ -3633,6 +4041,8 @@ arguments. An argument has a pointer to the parent Function.
+
+ Interaction and relationship between User and Use objects
+
+
-
@@ -2656,10 +3066,14 @@ And here is the result of <deepCheck identityProp>:
OK, passed 500 tests.
+
+ The Core LLVM Class Hierarchy Reference
+
-
doxygen info: Type Class
-
-
The Module class
-
doxygen info:
@@ -2824,14 +3228,12 @@ href="#GlobalVariable">GlobalVariables, and a SymbolTable. Additionally, it contains a few
helpful member functions that try to make common operations easy.
The Value class
-
@@ -2987,19 +3390,17 @@ the class that
represents this value. Although this may take some getting used to, it
simplifies the representation and makes it easier to manipulate.
- Value::use_const_iterator - Typedef for const_iterator over
+ Value::const_use_iterator - Typedef for const_iterator over
the use-list
unsigned use_size() - Returns the number of users of the
value.
@@ -3041,12 +3442,14 @@ Inst->replaceAllUsesWith(ConstVal);
The User class
-
@@ -3065,14 +3468,12 @@ Single Assignment (SSA) form, there can only be one definition referred to,
allowing this direct connection. This connection provides the use-def
information in LLVM.
@@ -3131,14 +3534,13 @@ href="#CmpInst">CmpInst). Unfortunately, the use of macros in
this file confuses doxygen, so these enum values don't show up correctly in the
doxygen output.
+
+ Important Subclasses of the Instruction class
+
+
+
+
+ Important Public Members of the Instruction class
+
+
-
Important Subclasses of Constant
+
@@ -3293,15 +3697,14 @@ dereference the pointer with GetElementPtrInst first, then its elements
can be accessed. This is explained in the LLVM
Language Reference Manual.
+
+ Important Public Members of the GlobalValue class
+
+
-
@@ -3317,12 +3720,14 @@ GlobalValue is currently embedded into.
doxygen
@@ -3333,7 +3738,7 @@ Superclasses: GlobalValue,
Value
+
+ Important Public Members of the Function class
+
+
-
+
+ Important Public Members of the GlobalVariable class
+
+
-
-doxygen info: BasicBlock
+doxygen info: BasicBlock
Class
Superclass: Value
+
+ Important Public Members of the BasicBlock class
+
+
-
-
+
-
@@ -3643,7 +4053,7 @@ arguments. An argument has a pointer to the parent Function.
Dinakar Dhurjati and
Chris Lattner
- The LLVM Compiler Infrastructure
+ The LLVM Compiler Infrastructure
Last modified: $Date$