From 1c86f39d32b95fe9de306d82f02a982bbad778b4 Mon Sep 17 00:00:00 2001 From: Benjamin Kosnik Date: Thu, 12 Feb 2004 01:11:48 +0000 Subject: [multiple changes] 2004-02-11 Stefan Olsson * docs/html/ext/mt_allocator.html: New. 2004-02-11 Benjamin Kosnik * docs/html/20_util/allocator.html: New file, consolidate allocator information here. Revamp. * docs/html/documentation.html: Change links. * docs/html/20_util/howto.html: Same. * docs/html/ext/howto.html: Same. From-SVN: r77687 --- libstdc++-v3/docs/html/ext/howto.html | 244 +--------------- libstdc++-v3/docs/html/ext/mt_allocator.html | 414 +++++++++++++++++++++++++++ 2 files changed, 415 insertions(+), 243 deletions(-) create mode 100644 libstdc++-v3/docs/html/ext/mt_allocator.html (limited to 'libstdc++-v3/docs/html/ext') diff --git a/libstdc++-v3/docs/html/ext/howto.html b/libstdc++-v3/docs/html/ext/howto.html index 2ce76ee..3e5c35c 100644 --- a/libstdc++-v3/docs/html/ext/howto.html +++ b/libstdc++-v3/docs/html/ext/howto.html @@ -48,8 +48,7 @@

Ropes and trees and hashes, oh my!
Added members and types
Allocators (versions 3.0, 3.1, 3.2, 3.3)
Allocators (version 3.4)
__mt_alloc
Compile-time checks
LWG Issues
Demangling

Allocators (versions 3.0, 3.1, 3.2, 3.3)

Thread-safety, space efficiency, high speed, portability... this is a - mess. Where to begin? -

The Rules

The C++ standard only gives a few directives in this area: -

When you add elements to a container, and the container must allocate - more memory to hold them, the container makes the request via its - Allocator template parameter. This includes adding - char's to the string class, which acts as a regular STL container - in this respect. -
The default Allocator of every container-of-T is - std::allocator<T>. -
The interface of the allocator<T> class is - extremely simple. It has about 20 public declarations (nested - typedefs, member functions, etc), but the two which concern us most - are: -
```
-      T*    allocate   (size_type n, const void* hint = 0);
-      void  deallocate (T* p, size_type n);
```
- (This is a simplicifcation; the real signatures use nested typedefs.) - The "n" arguments in both those functions is a - count of the number of T's to allocate space for, - not their total size. -
"The storage is obtained by calling - ::operator new(size_t), but it is unspecified when or - how often this function is called. The use of hint - is unspecified, but intended as an aid to locality if an - implementation so desires." [20.4.1.1]/6 -

Problems and Possibilities

The easiest way of fulfilling the requirements is to call operator new - each time a container needs memory, and to call operator delete each - time the container releases memory. BUT - this - method is horribly slow. -

Or we can keep old memory around, and reuse it in a pool to save time. - The old libstdc++-v2 used a memory pool, and so do we. As of 3.0, - it's - on by default. The pool is shared among all the containers in the - program: when your program's std::vector<int> gets cut in half - and frees a bunch of its storage, that memory can be reused by the - private std::list<WonkyWidget> brought in from a KDE library - that you linked against. And we don't have to call operators new and - delete to pass the memory on, either, which is a speed bonus. - BUT... -

What about threads? No problem: in a threadsafe environment, the - memory pool is manipulated atomically, so you can grow a container in - one thread and shrink it in another, etc. BUT what - if threads in libstdc++-v3 aren't set up properly? - That's been answered already. -

BUT what if you want to use your own allocator? What - if you plan on using a runtime-loadable version of malloc() which uses - shared telepathic anonymous mmap'd sections serializable over a - network, so that memory requests should go through malloc? - And what if you need to debug it? -

Well then: -

Available allocators in namespace std

First I'll describe the situation as it exists for the code which - was released in GCC 3.1 and 3.2. Then I'll describe the differences - for 3.0. The allocator classes also have source documentation, - which is described here (you - will need to retrieve the maintainer-level docs, as almost none of - these entities are in the ISO standard). -

As a general rule of thumb, users are not allowed to use names which - begin with an underscore. This means that to be portable between - compilers, none of the following may be used in your program directly. - (If you decide to be unportable, then you're free do do what you want, - but it's not our fault if stuff breaks.) They are presented here for - information for maintainers and contributors in addition to users. -

These classes are always available: -

__new_alloc simply wraps ::operator new - and ::operator delete. -
__malloc_alloc_template<int inst> simply wraps - malloc and free. There is also a hook - for an out-of-memory handler (for new/delete this is taken care of - elsewhere). The inst parameter is described below. - This class was called malloc_alloc in earlier versions. -
allocator<T> has already been described; it is - The Standard Allocator for instances of T. It uses the internal - __alloc typedef (see below) to satisy its requests. -
__simple_alloc<T,A> is a wrapper around another - allocator, A, which itself is an allocator for instances of T. - This is primarily used in an internal "allocator traits" - class which helps encapsulate the different styles of allocators. -
__debug_alloc<A> is also a wrapper around an - arbitrary allocator A. It passes on slightly increased size - requests to A, and uses the extra memory to store size information. - When a pointer is passed to deallocate(), the stored - size is checked, and assert() is used to guarantee they match. -
__allocator<T,A> is an adaptor. Many of these - allocator classes have a consistent yet non-standard interface. - Such classes can be changed to a conforming interface with this - wrapper: __allocator<T, __alloc> is thus the - same as allocator<T>. -

Normally, - __default_alloc_template<bool thr, int inst> - is also available. This is the high-speed pool, called the default - node allocator. The reusable memory is shared among identical - instantiations of - this type. It calls through __new_alloc to obtain - new memory when its lists run out. If a client container requests a - block larger than a certain threshold size, then the pool is bypassed, - and the allocate/deallocate request is passed to - __new_alloc directly. -

Its inst parameter is described below. The - thr boolean determines whether the pool should be - manipulated atomically or not. Two typedefs are provided: - __alloc is defined as this node allocator with thr=true, - and therefore is threadsafe, while __single_client_alloc - defines thr=false, and is slightly faster but unsafe for multiple - threads. -

(Note that the GCC thread abstraction layer allows us to provide safe - zero-overhead stubs for the threading routines, if threads were - disabled at configuration time. In this situation, - __alloc should not be noticably slower than - __single_client_alloc.) -

[Another threadsafe allocator where each thread keeps its own free - list, so that no locking is needed, might be described here.] -

A cannon to swat a fly: `__USE_MALLOC`

If you've already read this - advice but still think you remember how to use this macro from - SGI STL days. We have removed it in gcc 3.3. See next section - for the new way to get the same effect. -

Globally disabling memory caching: `GLIBCXX_FORCE_NEW`

Starting with gcc 3.3, if you want to globally disable memory - caching within the library for the default allocator (i.e. - the one you get for all library objects when you do not specify - which one to use), merely set GLIBCXX_FORCE_NEW (at this time, - with any value) into your environment before running the - program. You will obtain a similar effect without having to - recompile your entire program and the entire library (the new - operator in gcc is a light wrapper around malloc). If your - program crashes with GLIBCXX_FORCE_NEW in the environment, - it likely means that you linked against objects built against - the older library. Code to support this extension is fully - compatible with 3.2 code if GLIBCXX_FORCE_NEW is not in the - environment. Prior to GCC 3.4, this variable was spelt - GLIBCPP_FORCE_NEW. -

Writing your own allocators

Depending on your application (a specific program, a generic library, - etc), allocator classes tend to be one of two styles: "SGI" - or "standard". See the comments in stl_alloc.h for more - information on this crucial difference. -

At the bottom of that header is a helper type, - _Alloc_traits, and various specializations of it. This - allows the container classes to make possible compile-time - optimizations based on features of the allocator. You should provide - a specialization of this type for your allocator (doing so takes only - two or three statements). -

Using non-default allocators

You can specify different memory management schemes on a per-container - basis, by overriding the default Allocator template - parameter. For example, an easy - (but nonportable) - method of specifying that only malloc/free should be used instead of - the default node allocator is: -

-    std::list <my_type, std::__malloc_alloc_template<0> >  my_malloc_based_list;

-    std::deque <my_type, std::__debug_alloc<std::__alloc> >  debug_deque;

`inst`

The __malloc_alloc_template and - __default_alloc_template classes take an integer parameter, - called inst here. This number is completely unused. -

The point of the number is to allow multiple instantiations of the - classes without changing the semantics at all. All three of -

-    typedef  __default_alloc_template<true,0>    normal;
-    typedef  __default_alloc_template<true,1>    private;
-    typedef  __default_alloc_template<true,42>   also_private;

behave exactly the same way. However, the memory pool for each type - (and remember that different instantiations result in different types) - remains separate. -

The library uses 0 in all its instantiations. If you - wish to keep separate free lists for a particular purpose, use a - different number. -

3.0.x

For 3.0.x, many of the names were incorrectly not prefixed - with underscores. So symbols such as "std::single_client_alloc" - are present. Be very careful to not depend on these names any more - than you would depend on implementation-only names. -

Certain macros like _NOTHREADS and __STL_THREADS - can affect the 3.0.x allocators. Do not use them. Those macros have - been completely removed for 3.1. -

Return to top of page or - to the FAQ. -

Allocators (version 3.4)

Changes are coming... -

If you plan on writing your own allocators, - source documentation is - available. You'll need to get the "maintainers" collection - in order to see the helper classes and extra notes. -

Return to top of page or - to the FAQ. -

Compile-time checks

Currently libstdc++-v3 uses the concept checkers from the Boost library to perform optional diff --git a/libstdc++-v3/docs/html/ext/mt_allocator.html b/libstdc++-v3/docs/html/ext/mt_allocator.html new file mode 100644 index 0000000..93a5bfb --- /dev/null +++ b/libstdc++-v3/docs/html/ext/mt_allocator.html @@ -0,0 +1,414 @@ + + + + + + + + + + A fixed-size, multi-thread optimized allocator + + + + + +

A fixed-size, multi-thread optimized allocator

+ The latest version of this document is always available at + + http://gcc.gnu.org/onlinedocs/libstdc++/ext/mt_allocator.html. +

+ To the libstdc++-v3 homepage. +

+ Introduction +

The mt allocator [hereinafter referred to simply as "the +allocator"] is a fixed size (power of two) allocator that was +initially developed specifically to suit the needs of multi threaded +applications [hereinafter referred to as an MT application]. Over time +the allocator has evolved and been improved in many ways, one of the +being that it now also does a good job in single threaded applications +[hereinafter referred to as a ST application]. (Note: In this +document, when referring to single threaded applications this also +includes applications that are compiled with gcc without thread +support enabled. This is accomplished using ifdef's on __GTHREADS) +

+The aim of this document is to describe - from a application point of +view - the "inner workings" of the allocator. +

+ Initialization +

+The static variables (pointers to freelists, tuning parameters etc) +are initialized to their default values at file scope, i.e.: +

+  template size_t
+  __mt_alloc<_Tp>::_S_freelist_headroom = 10;
+

+The very first allocate() call will always call the _S_init() function. +In order to make sure that this function is called exactly once we make use +of a __gthread_once (with _S_once_mt and _S_init as arguments) call in MT +applications and check a static bool (_S_initialized) in ST applications. +

+The _S_init() function: +- If the GLIBCXX_FORCE_NEW environment variable is set, it sets the bool + _S_force_new to true and then returns. This will cause subsequent calls to + allocate() to return memory directly from a new() call, and deallocate will + only do a delete() call. +

+- If the GLIBCXX_FORCE_NEW environment variable is not set, both ST and MT + applications will: + - Calculate the number of bins needed. A bin is a specific power of two size + of bytes. I.e., by default the allocator will deal with requests of up to + 128 bytes (or whatever the value of _S_max_bytes is when _S_init() is + called). This means that there will be bins of the following sizes + (in bytes): 1, 2, 4, 8, 16, 32, 64, 128. + + - Create the _S_binmap array. All requests are rounded up to the next + "large enough" bin. I.e., a request for 29 bytes will cause a block from + the "32 byte bin" to be returned to the application. The purpose of + _S_binmap is to speed up the process of finding out which bin to use. + I.e., the value of _S_binmap[ 29 ] is initialized to 5 (bin 5 = 32 bytes). +

+ - Create the _S_bin array. This array consists of bin_records. There will be + as many bin_records in this array as the number of bins that we calculated + earlier. I.e., if _S_max_bytes = 128 there will be 8 entries. + Each bin_record is then initialized: + - bin_record->first = An array of pointers to block_records. There will be + as many block_records pointers as there are maximum number of threads + (in a ST application there is only 1 thread, in a MT application there + are _S_max_threads). + This holds the pointer to the first free block for each thread in this + bin. I.e., if we would like to know where the first free block of size 32 + for thread number 3 is we would look this up by: _S_bin[ 5 ].first[ 3 ] + - bin_record->last = See above, the only difference being that this points + to the last record on the same freelist. + + The above created block_record pointers members are now initialized to + their initial values. I.e. _S_bin[ n ].first[ n ] = NULL; +

+- Additionally a MT application will: + - Create a list of free thread id's. The pointer to the first entry + is stored in _S_thread_freelist_first. The reason for this approach is + that the __gthread_self() call will not return a value that corresponds to + the maximum number of threads allowed but rather a process id number or + something else. So what we do is that we create a list of thread_records. + This list is _S_max_threads long and each entry holds a size_t thread_id + which is initialized to 1, 2, 3, 4, 5 and so on up to _S_max_threads. + Each time a thread calls allocate() or deallocate() we call + _S_get_thread_id() which looks at the value of _S_thread_key which is a + thread local storage pointer. If this is NULL we know that this is a newly + created thread and we pop the first entry from this list and saves the + pointer to this record in the _S_thread_key variable. The next time + we will get the pointer to the thread_record back and we use the + thread_record->thread_id as identification. I.e., the first thread that + calls allocate will get the first record in this list and thus be thread + number 1 and will then find the pointer to its first free 32 byte block + in _S_bin[ 5 ].first[ 1 ] + When we create the _S_thread_key we also define a destructor + (_S_thread_key_destr) which means that when the thread dies, this + thread_record is returned to the front of this list and the thread id + can then be reused if a new thread is created. + This list is protected by a mutex (_S_thread_freelist_mutex) which is only + locked when records are removed/added to the list. +

+ - Initialize the free and used counters of each bin_record: + - bin_record->free = An array of size_t. This keeps track of the number + of blocks on a specific thread's freelist in each bin. I.e., if a thread + has 12 32-byte blocks on it's freelists and allocates one of these, this + counter would be decreased to 11. + + - bin_record->used = An array of size_t. This keeps track of the number + of blocks currently in use of this size by this thread. I.e., if a thread + has made 678 requests (and no deallocations...) of 32-byte blocks this + counter will read 678. + + The above created arrays are now initialized with their initial values. + I.e. _S_bin[ n ].free[ n ] = 0; +

+ - Initialize the mutex of each bin_record: + The bin_record->mutex is used to protect the global freelist. This concept + of a global freelist is explained in more detail in the section + "A multi threaded example", but basically this mutex is locked whenever + a block of memory is retrieved or returned to the global freelist for this + specific bin. This only occurs when a number of blocks are grabbed from the + global list to a thread specific list or when a thread decides to return + some blocks to the global freelist. +

+ A single threaded example (and a primer for the multi threaded example!) +

+Let's start by describing how the data on a freelist is laid out in memory. +This is the first two blocks in freelist for thread id 3 in bin 3 (8 bytes): +

++----------------+
+| next* ---------|--+  (_S_bin[ 3 ].first[ 3 ] points here)
+|                |  |
+|                |  |
+|                |  |
++----------------+  |
+| thread_id = 3  |  |
+|                |  |
+|                |  |
+|                |  |
++----------------+  |
+| DATA           |  |  (A pointer to here is what is returned to the
+|                |  |   the application when needed)
+|                |  |
+|                |  |
+|                |  |
+|                |  |
+|                |  |
+|                |  |
++----------------+  |
++----------------+  |
+| next*          |<-+  (If next == NULL it's the last one on the list and
+|                |      then the _S_bin[ 3 ].last[ 3 ] pointer points to
+|                |      here as well)
+|                |
++----------------+
+| thread_id = 3  |
+|                |
+|                |
+|                |
++----------------+
+| DATA           |
+|                |
+|                |
+|                |
+|                |
+|                |
+|                |
+|                |
++----------------+
+

+With this in mind we simplify things a bit for a while and say that there is +only one thread (a ST application). In this case all operations are made to +what is referred to as the global pool - thread id 0 (No thread may be +assigned this id since they span from 1 to _S_max_threads in a MT application). +

+When the application requests memory (calling allocate()) we first look at the +requested size and if this is > _S_max_bytes we call new() directly and return. +

+If the requested size is within limits we start by finding out from which +bin we should serve this request by looking in _S_binmap. +

+A quick look at _S_bin[ bin ].first[ 0 ] tells us if there are any blocks of +this size on the freelist (0). If this is not NULL - fine, just remove the +block that _S_bin[ bin ].first[ 0 ] points to from the list, +update _S_bin[ bin ].first[ 0 ] and return a pointer to that blocks data. +

+If the freelist is empty (the pointer is NULL) we must get memory from the +system and build us a freelist within this memory. All requests for new memory +is made in chunks of _S_chunk_size. Knowing the size of a block_record and +the bytes that this bin stores we then calculate how many blocks we can create +within this chunk, build the list, remove the first block, update the pointers +(_S_bin[ bin ].first[ 0 ] and _S_bin[ bin ].last[ 0 ]) and return a pointer +to that blocks data. +

+Deallocation is equally simple; the pointer is casted back to a block_record +pointer, lookup which bin to use based on the size, add the block to the end +of the global freelist (with the next pointer set to NULL) and update the +pointers as needed (_S_bin[ bin ].first[ 0 ] and _S_bin[ bin ].last[ 0 ]). +

+ A multi threaded example +

+In the ST example we never used the thread_id variable present in each block. +Let's start by explaining the purpose of this in a MT application. +

+The concept of "ownership" was introduced since many MT applications +allocate and deallocate memory to shared containers from different +threads (such as a cache shared amongst all threads). This introduces +a problem if the allocator only returns memory to the current threads +freelist (I.e., there might be one thread doing all the allocation and +thus obtaining ever more memory from the system and another thread +that is getting a longer and longer freelist - this will in the end +consume all available memory). +

+Each time a block is moved from the global list (where ownership is +irrelevant), to a threads freelist (or when a new freelist is built +from a chunk directly onto a threads freelist or when a deallocation +occurs on a block which was not allocated by the same thread id as the +one doing the deallocation) the thread id is set to the current one. +

+What's the use? Well, when a deallocation occurs we can now look at +the thread id and find out if it was allocated by another thread id +and decrease the used counter of that thread instead, thus keeping the +free and used counters correct. And keeping the free and used counters +corrects is very important since the relationship between these two +variables decides if memory should be returned to the global pool or +not when a deallocation occurs. +

+When the application requests memory (calling allocate()) we first +look at the requested size and if this is > _S_max_bytes we call new() +directly and return. +

+If the requested size is within limits we start by finding out from which +bin we should serve this request by looking in _S_binmap. +

+A call to _S_get_thread_id() returns the thread id for the calling thread +(and if no value has been set in _S_thread_key, a new id is assigned and +returned). +

+A quick look at _S_bin[ bin ].first[ thread_id ] tells us if there are +any blocks of this size on the current threads freelist. If this is +not NULL - fine, just remove the block that _S_bin[ bin ].first[ +thread_id ] points to from the list, update _S_bin[ bin ].first[ +thread_id ], update the free and used counters and return a pointer to +that blocks data. +

+If the freelist is empty (the pointer is NULL) we start by looking at +the global freelist (0). If there are blocks available on the global +freelist we lock this bins mutex and move up to block_count (the +number of blocks of this bins size that will fit into a _S_chunk_size) +or until end of list - whatever comes first - to the current threads +freelist and at the same time change the thread_id ownership and +update the counters and pointers. When the bins mutex has been +unlocked, we remove the block that _S_bin[ bin ].first[ thread_id ] +points to from the list, update _S_bin[ bin ].first[ thread_id ], +update the free and used counters, and return a pointer to that blocks +data. +

+The reason that the number of blocks moved to the current threads +freelist is limited to block_count is to minimize the chance that a +subsequent deallocate() call will return the excess blocks to the +global freelist (based on the _S_freelist_headroom calculation, see +below). +

+However if there isn't any memory on the global pool we need to get +memory from the system - this is done in exactly the same way as in a +single threaded application with one major difference; the list built +in the newly allocated memory (of _S_chunk_size size) is added to the +current threads freelist instead of to the global. +

+The basic process of a deallocation call is simple: always add the +block to the end of the current threads freelist and update the +counters and pointers (as described earlier with the specific check of +ownership that causes the used counter of the thread that originally +allocated the block to be decreased instead of the current threads +counter). +

+And here comes the free and used counters to service. Each time a +deallocation() call is made, the length of the current threads +freelist is compared to the amount memory in use by this thread. +

+Let's go back to the example of an application that has one thread +that does all the allocations and one that deallocates. Both these +threads use say 516 32-byte blocks that was allocated during thread +creation for example. Their used counters will both say 516 at this +point. The allocation thread now grabs 1000 32-byte blocks and puts +them in a shared container. The used counter for this thread is now +1516. +

+The deallocation thread now deallocates 500 of these blocks. For each +deallocation made the used counter of the allocating thread is +decreased and the freelist of the deallocation thread gets longer and +longer. But the calculation made in deallocate() will limit the length +of the freelist in the deallocation thread to _S_freelist_headroom % +of it's used counter. In this case, when the freelist (given that the +_S_freelist_headroom is at it's default value of 10%) exceeds 52 +(516/10) blocks will be returned to the global pool where the +allocating thread may pick them up and reuse them. +

+In order to reduce lock contention (since this requires this bins +mutex to be locked) this operation is also made in chunks of blocks +(just like when chunks of blocks are moved from the global freelist to +a threads freelist mentioned above). The "formula" used can probably +be improved to further reduce the risk of blocks being "bounced back +and forth" between freelists. +

Return to the top of the page or + to the libstdc++ homepage. +

+See license.html for copying conditions. +Comments and suggestions are welcome, and may be sent to +the libstdc++ mailing list. +