Re: at2 swap_nibbles

On Dienstag, November 26, 2002, at 03:21  Uhr, Richard Braakman wrote:

> On Mon, Nov 25, 2002 at 07:05:23PM +0100, Andreas Fink wrote:
>> how about this:
>>
>> #define	HI_NIBBLE(a)	( (int)a >> 4 & 0x0F)
>> #define	LOW_NIBBLE(a)	( (int)a & 0x0F)
>> #define	SWAP_NIBBLE(a)	(( HI_NIBBLE(a) | (LOW_NIBBLE(a) << 4)) & 0xFF)
>>
>> This would work as a macro and has the advantage that the CPU can run
>> it pretty much native.
>
> Except that casting to int brings back the original sign-extension bug 
> :)

not really as we XOR away all the bits we don't need.

>
> If swap_nibbles is used only in the at2 driver, then you can make it a
> static function and expect the compiler to inline it.  If it's used in
> multiple files, then it's better off as a real library function in 
> gwlib
> somewhere.  (octstr_swap_nibbles perhaps?)
>
> If you need a really fast nibble swapper, then a table lookup is 
> probably
> best.  It'll only be a 256-byte table.  I wouldn't bother with that 
> kind
> of optimization unless profiling shows that it matters.
>
>

the table approach is an interesting idea.
Anyhow, Oded, finish the patch in the way it works. If macro or 
function doesn't really matter.
I really wonder why the AT driver works at all right now with this bug.


Andreas Fink
Fink Consulting GmbH

---------------------------------------------------------------
Tel: +41-61-6666332 Fax: +41-61-6666331  Mobile: +41-79-2457333
Address: Clarastrasse 3, 4058 Basel, Switzerland
E-Mail:  [email protected]
Homepage: http://www.finkconsulting.com
---------------------------------------------------------------