Re: Byte swapping for doubles

[Adam Fedor]

This has been hanging around in my Inbox for a while. I found this  
about how the ARM ABI deals with doubles:

VFP is a newer floating point architecture for ARM. With respect to  
endianess, the word endianess of doubles is always the same as the  
byte endianess of words. This contrast with the older and more common  
FPA architecture, which always uses big endian for doubles. In both  
architectures, the word endianness of 64-bit ints matches the word  
endianess of doubles.

After looking at NSByteOrder.h, it appears we do it correctly, as far  
as I can tell, swapping doubles the same ways as 64-bit int.  Does  
Nikolaus (or anyone) have an actual bug report which shows the  
problem?  I'm not sure I want to change anything without at least  
some one to verify it.

On Feb 6, 2007, at 9:00 AM, Fred Kiefer wrote:

> I had an email discussion with Nikolaus Schaller about the code in
> NSByteOrder.h which does the byte swapping for GNUstep and he  
> convinced
> me that we may not handle the case of swapping doubles correctly for
> certain ARM processors. Here the word order in doubles seems to differ
> from the general byte order.
>
> I had a quick search on the web and found  
> ax_cv_c_float_words_bigendian
> (See http://autoconf-archive.cryp.to/ax_c_float_words_bigendian.html).
> Is there anybody out there, who understands word/byte order a bit  
> better
> and is able to fix this?