Seems that 2x has the best effect for now but this is not the best way
of implementing AA function because it requires much greater video card
memory and memory bandwidth than other avaible methods like MSAA, CSAA
etc.
Add missing #include to various files.
strings.h isn't provided by MSVC so we don't include it there; it's needed
for other OSes/compilers.
Get rid of pre-ISO-C malloc.h includes; malloc is provided by stdlib.h.
Fixes some linuxisms.
Prepend __builtin_ to __clear_cache, calling it without the prefix is a
GNU extension.
