[uClibc]Starting port of uClib for x86_64 (AMD64)

[Erik Andersen]

On Mon Apr 28, 2003 at 04:25:56PM +0800, Andrew Derrick Balsa wrote:
> Hi,
> 
> I am starting a port of ucLib to the AMD64 architecture and I wanted to ask 
> you for some hints, tips, advice and constructive criticism.

Hint #1, please send me an AMD64 machine, preferable with several
GB of ram and a few TB of disk space.  ;-)

> First thing, I would prefer to call it x86_64 in the uClib tree, even though 
> AMD has noted its preference for the AMD64 moniker.

The "x86_64" naming convention matches the naming used in the
linux kernel port, so that seems best....

> Second, I guess it will be almost a copy of the i386 tree, but any hints will 
> be welcome as to what I should pay attention to in particular.

Starting with a copy of the i386 tree would be an excellent
start.  One advantage of x86_64 though, is you _know_ that more
advanced opcodes are available.  So, for example, for
x86_64/bits/byteswap.h you can just use the bswap opcode, rather
than using rotates to accomadate legacy cpus.  Many similar
examples exist...

Another place to watch out is full 64-bit compatibility.  In
general, uClibc should be ready for 64-bit.  I did a quick port
to alpha a while back and worked out the most obvious problems.
But some problems may have crept back into the source tree, and
some problems may never have been fixed, since I didn't spend the
time to get alpha working particularly well,

 -Erik

--
Erik B. Andersen             http://codepoet-consulting.com/
--This message was written using 73% post-consumer electrons--