📄 bget.doc
字号:
BGET -- Memory Allocator
==========================
by John Walker
kelvin@fourmilab.ch
http://www.fourmilab.ch/
BGET is a comprehensive memory allocation package which is easily
configured to the needs of an application. BGET is efficient in both
the time needed to allocate and release buffers and in the memory
overhead required for buffer pool management. It automatically
consolidates contiguous space to minimise fragmentation. BGET is
configured by compile-time definitions, Major options include:
* A built-in test program to exercise BGET and
demonstrate how the various functions are used.
* Allocation by either the "first fit" or "best fit"
method.
* Wiping buffers at release time to catch code which
references previously released storage.
* Built-in routines to dump individual buffers or the
entire buffer pool.
* Retrieval of allocation and pool size statistics.
* Quantisation of buffer sizes to a power of two to
satisfy hardware alignment constraints.
* Automatic pool compaction, growth, and shrinkage by
means of call-backs to user defined functions.
Applications of BGET can range from storage management in ROM-based
embedded programs to providing the framework upon which a multitasking
system incorporating garbage collection is constructed. BGET
incorporates extensive internal consistency checking using the
<assert.h> mechanism; all these checks can be turned off by compiling
with NDEBUG defined, yielding a version of BGET with minimal size and
maximum speed.
The basic algorithm underlying BGET has withstood the test of time; more
than 25 years have passed since the first implementation of this code.
And yet, it is substantially more efficient than the native allocation
schemes of many operating systems: the Macintosh and Microsoft Windows
to name two, on which programs have obtained substantial speed-ups by
layering BGET as an application level memory manager atop the underlying
system's.
BGET has been implemented on the largest mainframes and the lowest of
microprocessors. It has served as the core for multitasking operating
systems, multi-thread applications, embedded software in data network
switching processors, and a host of C programs. And while it has
accreted flexibility and additional options over the years, it remains
fast, memory efficient, portable, and easy to integrate into your
program.
BGET IMPLEMENTATION ASSUMPTIONS
===============================
BGET is written in as portable a dialect of C as possible. The only
fundamental assumption about the underlying hardware architecture is
that memory is allocated is a linear array which can be addressed as a
vector of C "char" objects. On segmented address space architectures,
this generally means that BGET should be used to allocate storage within
a single segment (although some compilers simulate linear address spaces
on segmented architectures). On segmented architectures, then, BGET
buffer pools may not be larger than a segment, but since BGET allows any
number of separate buffer pools, there is no limit on the total storage
which can be managed, only on the largest individual object which can be
allocated. Machines with a linear address architecture, such as the
VAX, 680x0, Sparc, MIPS, or the Intel 80386 and above in native mode,
may use BGET without restriction.
GETTING STARTED WITH BGET
=========================
Although BGET can be configured in a multitude of fashions, there are
three basic ways of working with BGET. The functions mentioned below
are documented in the following section. Please excuse the forward
references which are made in the interest of providing a roadmap to
guide you to the BGET functions you're likely to need.
Embedded Applications
---------------------
Embedded applications typically have a fixed area of memory dedicated to
buffer allocation (often in a separate RAM address space distinct from
the ROM that contains the executable code). To use BGET in such an
environment, simply call bpool() with the start address and length of
the buffer pool area in RAM, then allocate buffers with bget() and
release them with brel(). Embedded applications with very limited RAM
but abundant CPU speed may benefit by configuring BGET for BestFit
allocation (which is usually not worth it in other environments).
Malloc() Emulation
------------------
If the C library malloc() function is too slow, not present in your
development environment (for example, an a native Windows or Macintosh
program), or otherwise unsuitable, you can replace it with BGET.
Initially define a buffer pool of an appropriate size with
bpool()--usually obtained by making a call to the operating system's
low-level memory allocator. Then allocate buffers with bget(), bgetz(),
and bgetr() (the last two permit the allocation of buffers initialised
to zero and [inefficient] re-allocation of existing buffers for
compatibility with C library functions). Release buffers by calling
brel(). If a buffer allocation request fails, obtain more storage from
the underlying operating system, add it to the buffer pool by another
call to bpool(), and continue execution.
Automatic Storage Management
----------------------------
You can use BGET as your application's native memory manager and
implement automatic storage pool expansion, contraction, and optionally
application-specific memory compaction by compiling BGET with the BECtl
variable defined, then calling bectl() and supplying functions for
storage compaction, acquisition, and release, as well as a standard pool
expansion increment. All of these functions are optional (although it
doesn't make much sense to provide a release function without an
acquisition function, does it?). Once the call-back functions have been
defined with bectl(), you simply use bget() and brel() to allocate and
release storage as before. You can supply an initial buffer pool with
bpool() or rely on automatic allocation to acquire the entire pool.
When a call on bget() cannot be satisfied, BGET first checks if a
compaction function has been supplied. If so, it is called (with the
space required to satisfy the allocation request and a sequence number
to allow the compaction routine to be called successively without
looping). If the compaction function is able to free any storage (it
needn't know whether the storage it freed was adequate) it should return
a nonzero value, whereupon BGET will retry the allocation request and,
if it fails again, call the compaction function again with the
next-higher sequence number.
If the compaction function returns zero, indicating failure to free
space, or no compaction function is defined, BGET next tests whether a
non-NULL allocation function was supplied to bectl(). If so, that
function is called with an argument indicating how many bytes of
additional space are required. This will be the standard pool expansion
increment supplied in the call to bectl() unless the original bget()
call requested a buffer larger than this; buffers larger than the
standard pool block can be managed "off the books" by BGET in this mode.
If the allocation function succeeds in obtaining the storage, it returns
a pointer to the new block and BGET expands the buffer pool; if it
fails, the allocation request fails and returns NULL to the caller. If
a non-NULL release function is supplied, expansion blocks which become
totally empty are released to the global free pool by passing their
addresses to the release function.
Equipped with appropriate allocation, release, and compaction functions,
BGET can be used as part of very sophisticated memory management
strategies, including garbage collection. (Note, however, that BGET is
*not* a garbage collector by itself, and that developing such a system
requires much additional logic and careful design of the application's
memory allocation strategy.)
BGET FUNCTION DESCRIPTIONS
==========================
Functions implemented by BGET (some are enabled by certain of the
optional settings below):
void bpool(void *buffer, bufsize len);
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -