--- /dev/null
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
+<html xmlns="http://www.w3.org/1999/xhtml"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" /><title>bitmap_allocator</title><meta name="generator" content="DocBook XSL Stylesheets V1.74.0" /><meta name="keywords" content=" ISO C++ , allocator " /><meta name="keywords" content=" ISO C++ , library " /><link rel="home" href="../spine.html" title="The GNU C++ Library Documentation" /><link rel="up" href="ext_allocators.html" title="Chapter 32. Allocators" /><link rel="prev" href="ext_allocators.html" title="Chapter 32. Allocators" /><link rel="next" href="ext_containers.html" title="Chapter 33. Containers" /></head><body><div class="navheader"><table width="100%" summary="Navigation header"><tr><th colspan="3" align="center">bitmap_allocator</th></tr><tr><td width="20%" align="left"><a accesskey="p" href="ext_allocators.html">Prev</a> </td><th width="60%" align="center">Chapter 32. Allocators</th><td width="20%" align="right"> <a accesskey="n" href="ext_containers.html">Next</a></td></tr></table><hr /></div><div class="sect1" lang="en" xml:lang="en"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="manual.ext.allocator.bitmap"></a>bitmap_allocator</h2></div></div></div><p>
+</p><div class="sect2" lang="en" xml:lang="en"><div class="titlepage"><div><div><h3 class="title"><a id="allocator.bitmap.design"></a>Design</h3></div></div></div><p>
+ As this name suggests, this allocator uses a bit-map to keep track
+ of the used and unused memory locations for it's book-keeping
+ purposes.
+ </p><p>
+ This allocator will make use of 1 single bit to keep track of
+ whether it has been allocated or not. A bit 1 indicates free,
+ while 0 indicates allocated. This has been done so that you can
+ easily check a collection of bits for a free block. This kind of
+ Bitmapped strategy works best for single object allocations, and
+ with the STL type parameterized allocators, we do not need to
+ choose any size for the block which will be represented by a
+ single bit. This will be the size of the parameter around which
+ the allocator has been parameterized. Thus, close to optimal
+ performance will result. Hence, this should be used for node based
+ containers which call the allocate function with an argument of 1.
+ </p><p>
+ The bitmapped allocator's internal pool is exponentially growing.
+ Meaning that internally, the blocks acquired from the Free List
+ Store will double every time the bitmapped allocator runs out of
+ memory.
+ </p><p>
+ The macro <code class="literal">__GTHREADS</code> decides whether to use
+ Mutex Protection around every allocation/deallocation. The state
+ of the macro is picked up automatically from the gthr abstraction
+ layer.
+ </p></div><div class="sect2" lang="en" xml:lang="en"><div class="titlepage"><div><div><h3 class="title"><a id="allocator.bitmap.impl"></a>Implementation</h3></div></div></div><div class="sect3" lang="en" xml:lang="en"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.free_list_store"></a>Free List Store</h4></div></div></div><p>
+ The Free List Store (referred to as FLS for the remaining part of this
+ document) is the Global memory pool that is shared by all instances of
+ the bitmapped allocator instantiated for any type. This maintains a
+ sorted order of all free memory blocks given back to it by the
+ bitmapped allocator, and is also responsible for giving memory to the
+ bitmapped allocator when it asks for more.
+ </p><p>
+ Internally, there is a Free List threshold which indicates the
+ Maximum number of free lists that the FLS can hold internally
+ (cache). Currently, this value is set at 64. So, if there are
+ more than 64 free lists coming in, then some of them will be given
+ back to the OS using operator delete so that at any given time the
+ Free List's size does not exceed 64 entries. This is done because
+ a Binary Search is used to locate an entry in a free list when a
+ request for memory comes along. Thus, the run-time complexity of
+ the search would go up given an increasing size, for 64 entries
+ however, lg(64) == 6 comparisons are enough to locate the correct
+ free list if it exists.
+ </p><p>
+ Suppose the free list size has reached it's threshold, then the
+ largest block from among those in the list and the new block will
+ be selected and given back to the OS. This is done because it
+ reduces external fragmentation, and allows the OS to use the
+ larger blocks later in an orderly fashion, possibly merging them
+ later. Also, on some systems, large blocks are obtained via calls
+ to mmap, so giving them back to free system resources becomes most
+ important.
+ </p><p>
+ The function _S_should_i_give decides the policy that determines
+ whether the current block of memory should be given to the
+ allocator for the request that it has made. That's because we may
+ not always have exact fits for the memory size that the allocator
+ requests. We do this mainly to prevent external fragmentation at
+ the cost of a little internal fragmentation. Now, the value of
+ this internal fragmentation has to be decided by this function. I
+ can see 3 possibilities right now. Please add more as and when you
+ find better strategies.
+ </p><div class="orderedlist"><ol type="1"><li><p>Equal size check. Return true only when the 2 blocks are of equal
+size.</p></li><li><p>Difference Threshold: Return true only when the _block_size is
+greater than or equal to the _required_size, and if the _BS is > _RS
+by a difference of less than some THRESHOLD value, then return true,
+else return false. </p></li><li><p>Percentage Threshold. Return true only when the _block_size is
+greater than or equal to the _required_size, and if the _BS is > _RS
+by a percentage of less than some THRESHOLD value, then return true,
+else return false.</p></li></ol></div><p>
+ Currently, (3) is being used with a value of 36% Maximum wastage per
+ Super Block.
+ </p></div><div class="sect3" lang="en" xml:lang="en"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.super_block"></a>Super Block</h4></div></div></div><p>
+ A super block is the block of memory acquired from the FLS from
+ which the bitmap allocator carves out memory for single objects
+ and satisfies the user's requests. These super blocks come in
+ sizes that are powers of 2 and multiples of 32
+ (_Bits_Per_Block). Yes both at the same time! That's because the
+ next super block acquired will be 2 times the previous one, and
+ also all super blocks have to be multiples of the _Bits_Per_Block
+ value.
+ </p><p>
+ How does it interact with the free list store?
+ </p><p>
+ The super block is contained in the FLS, and the FLS is responsible for
+ getting / returning Super Bocks to and from the OS using operator new
+ as defined by the C++ standard.
+ </p></div><div class="sect3" lang="en" xml:lang="en"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.super_block_data"></a>Super Block Data Layout</h4></div></div></div><p>
+ Each Super Block will be of some size that is a multiple of the
+ number of Bits Per Block. Typically, this value is chosen as
+ Bits_Per_Byte x sizeof(size_t). On an x86 system, this gives the
+ figure 8 x 4 = 32. Thus, each Super Block will be of size 32
+ x Some_Value. This Some_Value is sizeof(value_type). For now, let
+ it be called 'K'. Thus, finally, Super Block size is 32 x K bytes.
+ </p><p>
+ This value of 32 has been chosen because each size_t has 32-bits
+ and Maximum use of these can be made with such a figure.
+ </p><p>
+ Consider a block of size 64 ints. In memory, it would look like this:
+ (assume a 32-bit system where, size_t is a 32-bit entity).
+ </p><div class="table"><a id="id435115"></a><p class="title"><b>Table 32.1. Bitmap Allocator Memory Map</b></p><div class="table-contents"><table summary="Bitmap Allocator Memory Map" border="1"><colgroup><col align="left" /><col align="left" /><col align="left" /><col align="left" /><col align="left" /></colgroup><tbody><tr><td align="left">268</td><td align="left">0</td><td align="left">4294967295</td><td align="left">4294967295</td><td align="left">Data -> Space for 64 ints</td></tr></tbody></table></div></div><br class="table-break" /><p>
+ The first Column(268) represents the size of the Block in bytes as
+ seen by the Bitmap Allocator. Internally, a global free list is
+ used to keep track of the free blocks used and given back by the
+ bitmap allocator. It is this Free List Store that is responsible
+ for writing and managing this information. Actually the number of
+ bytes allocated in this case would be: 4 + 4 + (4x2) + (64x4) =
+ 272 bytes, but the first 4 bytes are an addition by the Free List
+ Store, so the Bitmap Allocator sees only 268 bytes. These first 4
+ bytes about which the bitmapped allocator is not aware hold the
+ value 268.
+ </p><p>
+ What do the remaining values represent?</p><p>
+ The 2nd 4 in the expression is the sizeof(size_t) because the
+ Bitmapped Allocator maintains a used count for each Super Block,
+ which is initially set to 0 (as indicated in the diagram). This is
+ incremented every time a block is removed from this super block
+ (allocated), and decremented whenever it is given back. So, when
+ the used count falls to 0, the whole super block will be given
+ back to the Free List Store.
+ </p><p>
+ The value 4294967295 represents the integer corresponding to the bit
+ representation of all bits set: 11111111111111111111111111111111.
+ </p><p>
+ The 3rd 4x2 is size of the bitmap itself, which is the size of 32-bits
+ x 2,
+ which is 8-bytes, or 2 x sizeof(size_t).
+ </p></div><div class="sect3" lang="en" xml:lang="en"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.max_wasted"></a>Maximum Wasted Percentage</h4></div></div></div><p>
+ This has nothing to do with the algorithm per-se,
+ only with some vales that must be chosen correctly to ensure that the
+ allocator performs well in a real word scenario, and maintains a good
+ balance between the memory consumption and the allocation/deallocation
+ speed.
+ </p><p>
+ The formula for calculating the maximum wastage as a percentage:
+ </p><p>
+(32 x k + 1) / (2 x (32 x k + 1 + 32 x c)) x 100.
+ </p><p>
+ where k is the constant overhead per node (e.g., for list, it is
+ 8 bytes, and for map it is 12 bytes) and c is the size of the
+ base type on which the map/list is instantiated. Thus, suppose the
+ type1 is int and type2 is double, they are related by the relation
+ sizeof(double) == 2*sizeof(int). Thus, all types must have this
+ double size relation for this formula to work properly.
+ </p><p>
+ Plugging-in: For List: k = 8 and c = 4 (int and double), we get:
+ 33.376%
+ </p><p>
+For map/multimap: k = 12, and c = 4 (int and double), we get: 37.524%
+ </p><p>
+ Thus, knowing these values, and based on the sizeof(value_type), we may
+ create a function that returns the Max_Wastage_Percentage for us to use.
+ </p></div><div class="sect3" lang="en" xml:lang="en"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.allocate"></a><code class="function">allocate</code></h4></div></div></div><p>
+ The allocate function is specialized for single object allocation
+ ONLY. Thus, ONLY if n == 1, will the bitmap_allocator's
+ specialized algorithm be used. Otherwise, the request is satisfied
+ directly by calling operator new.
+ </p><p>
+ Suppose n == 1, then the allocator does the following:
+ </p><div class="orderedlist"><ol type="1"><li><p>
+ Checks to see whether a free block exists somewhere in a region
+ of memory close to the last satisfied request. If so, then that
+ block is marked as allocated in the bit map and given to the
+ user. If not, then (2) is executed.
+ </p></li><li><p>
+ Is there a free block anywhere after the current block right
+ up to the end of the memory that we have? If so, that block is
+ found, and the same procedure is applied as above, and
+ returned to the user. If not, then (3) is executed.
+ </p></li><li><p>
+ Is there any block in whatever region of memory that we own
+ free? This is done by checking
+ </p><div class="itemizedlist"><ul type="disc"><li><p>
+ The use count for each super block, and if that fails then
+ </p></li><li><p>
+ The individual bit-maps for each super block.
+ </p></li></ul></div><p>
+ Note: Here we are never touching any of the memory that the
+ user will be given, and we are confining all memory accesses
+ to a small region of memory! This helps reduce cache
+ misses. If this succeeds then we apply the same procedure on
+ that bit-map as (1), and return that block of memory to the
+ user. However, if this process fails, then we resort to (4).
+ </p></li><li><p>
+ This process involves Refilling the internal exponentially
+ growing memory pool. The said effect is achieved by calling
+ _S_refill_pool which does the following:
+ </p><div class="itemizedlist"><ul type="disc"><li><p>
+ Gets more memory from the Global Free List of the Required
+ size.
+ </p></li><li><p>
+ Adjusts the size for the next call to itself.
+ </p></li><li><p>
+ Writes the appropriate headers in the bit-maps.
+ </p></li><li><p>
+ Sets the use count for that super-block just allocated to 0
+ (zero).
+ </p></li><li><p>
+ All of the above accounts to maintaining the basic invariant
+ for the allocator. If the invariant is maintained, we are
+ sure that all is well. Now, the same process is applied on
+ the newly acquired free blocks, which are dispatched
+ accordingly.
+ </p></li></ul></div></li></ol></div><p>
+Thus, you can clearly see that the allocate function is nothing but a
+combination of the next-fit and first-fit algorithm optimized ONLY for
+single object allocations.
+</p></div><div class="sect3" lang="en" xml:lang="en"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.deallocate"></a><code class="function">deallocate</code></h4></div></div></div><p>
+ The deallocate function again is specialized for single objects ONLY.
+ For all n belonging to > 1, the operator delete is called without
+ further ado, and the deallocate function returns.
+ </p><p>
+ However for n == 1, a series of steps are performed:
+ </p><div class="orderedlist"><ol type="1"><li><p>
+ We first need to locate that super-block which holds the memory
+ location given to us by the user. For that purpose, we maintain
+ a static variable _S_last_dealloc_index, which holds the index
+ into the vector of block pairs which indicates the index of the
+ last super-block from which memory was freed. We use this
+ strategy in the hope that the user will deallocate memory in a
+ region close to what he/she deallocated the last time around. If
+ the check for belongs_to succeeds, then we determine the bit-map
+ for the given pointer, and locate the index into that bit-map,
+ and mark that bit as free by setting it.
+ </p></li><li><p>
+ If the _S_last_dealloc_index does not point to the memory block
+ that we're looking for, then we do a linear search on the block
+ stored in the vector of Block Pairs. This vector in code is
+ called _S_mem_blocks. When the corresponding super-block is
+ found, we apply the same procedure as we did for (1) to mark the
+ block as free in the bit-map.
+ </p></li></ol></div><p>
+ Now, whenever a block is freed, the use count of that particular
+ super block goes down by 1. When this use count hits 0, we remove
+ that super block from the list of all valid super blocks stored in
+ the vector. While doing this, we also make sure that the basic
+ invariant is maintained by making sure that _S_last_request and
+ _S_last_dealloc_index point to valid locations within the vector.
+ </p></div><div class="sect3" lang="en" xml:lang="en"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.questions"></a>Questions</h4></div></div></div><div class="sect4" lang="en" xml:lang="en"><div class="titlepage"><div><div><h5 class="title"><a id="bitmap.impl.question.1"></a>1</h5></div></div></div><p>
+Q1) The "Data Layout" section is
+cryptic. I have no idea of what you are trying to say. Layout of what?
+The free-list? Each bitmap? The Super Block?
+ </p><p>
+ The layout of a Super Block of a given
+size. In the example, a super block of size 32 x 1 is taken. The
+general formula for calculating the size of a super block is
+32 x sizeof(value_type) x 2^n, where n ranges from 0 to 32 for 32-bit
+systems.
+ </p></div><div class="sect4" lang="en" xml:lang="en"><div class="titlepage"><div><div><h5 class="title"><a id="bitmap.impl.question.2"></a>2</h5></div></div></div><p>
+ And since I just mentioned the
+term `each bitmap', what in the world is meant by it? What does each
+bitmap manage? How does it relate to the super block? Is the Super
+Block a bitmap as well?
+ </p><p>
+ Each bitmap is part of a Super Block which is made up of 3 parts
+ as I have mentioned earlier. Re-iterating, 1. The use count,
+ 2. The bit-map for that Super Block. 3. The actual memory that
+ will be eventually given to the user. Each bitmap is a multiple
+ of 32 in size. If there are 32 x (2^3) blocks of single objects
+ to be given, there will be '32 x (2^3)' bits present. Each 32
+ bits managing the allocated / free status for 32 blocks. Since
+ each size_t contains 32-bits, one size_t can manage up to 32
+ blocks' status. Each bit-map is made up of a number of size_t,
+ whose exact number for a super-block of a given size I have just
+ mentioned.
+ </p></div><div class="sect4" lang="en" xml:lang="en"><div class="titlepage"><div><div><h5 class="title"><a id="bitmap.impl.question.3"></a>3</h5></div></div></div><p>
+ How do the allocate and deallocate functions work in regard to
+ bitmaps?
+ </p><p>
+ The allocate and deallocate functions manipulate the bitmaps and
+ have nothing to do with the memory that is given to the user. As
+ I have earlier mentioned, a 1 in the bitmap's bit field
+ indicates free, while a 0 indicates allocated. This lets us
+ check 32 bits at a time to check whether there is at lease one
+ free block in those 32 blocks by testing for equality with
+ (0). Now, the allocate function will given a memory block find
+ the corresponding bit in the bitmap, and will reset it (i.e.,
+ make it re-set (0)). And when the deallocate function is called,
+ it will again set that bit after locating it to indicate that
+ that particular block corresponding to this bit in the bit-map
+ is not being used by anyone, and may be used to satisfy future
+ requests.
+ </p><p>
+ e.g.: Consider a bit-map of 64-bits as represented below:
+ 1111111111111111111111111111111111111111111111111111111111111111
+ </p><p>
+ Now, when the first request for allocation of a single object
+ comes along, the first block in address order is returned. And
+ since the bit-maps in the reverse order to that of the address
+ order, the last bit (LSB if the bit-map is considered as a
+ binary word of 64-bits) is re-set to 0.
+ </p><p>
+ The bit-map now looks like this:
+ 1111111111111111111111111111111111111111111111111111111111111110
+ </p></div></div><div class="sect3" lang="en" xml:lang="en"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.locality"></a>Locality</h4></div></div></div><p>
+ Another issue would be whether to keep the all bitmaps in a
+ separate area in memory, or to keep them near the actual blocks
+ that will be given out or allocated for the client. After some
+ testing, I've decided to keep these bitmaps close to the actual
+ blocks. This will help in 2 ways.
+ </p><div class="orderedlist"><ol type="1"><li><p>Constant time access for the bitmap themselves, since no kind of
+look up will be needed to find the correct bitmap list or it's
+equivalent.</p></li><li><p>And also this would preserve the cache as far as possible.</p></li></ol></div><p>
+ So in effect, this kind of an allocator might prove beneficial from a
+ purely cache point of view. But this allocator has been made to try and
+ roll out the defects of the node_allocator, wherein the nodes get
+ skewed about in memory, if they are not returned in the exact reverse
+ order or in the same order in which they were allocated. Also, the
+ new_allocator's book keeping overhead is too much for small objects and
+ single object allocations, though it preserves the locality of blocks
+ very well when they are returned back to the allocator.
+ </p></div><div class="sect3" lang="en" xml:lang="en"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.grow_policy"></a>Overhead and Grow Policy</h4></div></div></div><p>
+ Expected overhead per block would be 1 bit in memory. Also, once
+ the address of the free list has been found, the cost for
+ allocation/deallocation would be negligible, and is supposed to be
+ constant time. For these very reasons, it is very important to
+ minimize the linear time costs, which include finding a free list
+ with a free block while allocating, and finding the corresponding
+ free list for a block while deallocating. Therefore, I have
+ decided that the growth of the internal pool for this allocator
+ will be exponential as compared to linear for
+ node_allocator. There, linear time works well, because we are
+ mainly concerned with speed of allocation/deallocation and memory
+ consumption, whereas here, the allocation/deallocation part does
+ have some linear/logarithmic complexity components in it. Thus, to
+ try and minimize them would be a good thing to do at the cost of a
+ little bit of memory.
+ </p><p>
+ Another thing to be noted is the pool size will double every time
+ the internal pool gets exhausted, and all the free blocks have
+ been given away. The initial size of the pool would be
+ sizeof(size_t) x 8 which is the number of bits in an integer,
+ which can fit exactly in a CPU register. Hence, the term given is
+ exponential growth of the internal pool.
+ </p></div></div></div><div class="navfooter"><hr /><table width="100%" summary="Navigation footer"><tr><td width="40%" align="left"><a accesskey="p" href="ext_allocators.html">Prev</a> </td><td width="20%" align="center"><a accesskey="u" href="ext_allocators.html">Up</a></td><td width="40%" align="right"> <a accesskey="n" href="ext_containers.html">Next</a></td></tr><tr><td width="40%" align="left" valign="top">Chapter 32. Allocators </td><td width="20%" align="center"><a accesskey="h" href="../spine.html">Home</a></td><td width="40%" align="right" valign="top"> Chapter 33. Containers</td></tr></table></div></body></html>
projects / msp430-gcc.git / blobdiff