This is a fast malloc implementation that I wrote several years ago.
I later used it as the basis of GNU malloc.  My version differs from
the GNU version in that it does not support debugging hooks, and does
not record statistics.  Therefore it is slightly faster.

In order to safely link programs using this malloc with a C library
that provides a different malloc, you need to make sure that
malloc(), free(), and realloc() are defined in a single object file.
Otherwise when linking you might get a combination of this malloc()
with the library's free().  The Makefile builds such an object file,
alloc.o.

If you are using this malloc as the allocator for a C library of your
own, and are not linking with another C library, then you don't need
alloc.o.  If you are building a C library, you should also write a
replacement for the file "morecore.c" that doesn't pollute the name
space.

The header file "malloc.h" in this directory is NOT intended to be a
public header file; it is for internal use by malloc and its
friends.  Don't install malloc.h in a public include directory!

When porting this allocator to a new machine or operating system, you
should inspect the definition of BLOCKSIZE in malloc.h to make sure
it is greater than or equal to your target machine's virtual memory
page size; otherwise valloc() won't work properly.  (If you don't
care about valloc() then BLOCKSIZE doesn't matter.)

You will also need to provide a machine-dependent _default_morecore()
function; see morecore.c for a sample version that works on Unix.
Your morecore function should return a pointer to a newly allocated
region of the given size, aligned on the most pessimistic alignment
boundary for the machine.  Subsequent calls to morecore should return
contiguous memory, and calls to morecore with a negative argument
should return memory to the system.  If no memory is available
morecore should return NULL.

Bug reports to Mike Haertel, mike@cs.uoregon.edu.
This version is dated March 26, 1993; include this
date with your bug report.