Re: floating-point loads vs dreamcast

>> seems..odd..to design an FPU which requires changing settings to
>> both do arithmetic on double-precision numbers and load and store
>> double-precision numbers.
> Note that in little endian mode 64-bit FMOV will store the registers
> in the wrong order anyway.

Wrong only in that the more-significant half comes out before the
less-significant half.  This is "wrong" only if you care about the
structure of a double as stored in memory, which many programs don't
(the major exception that comes to mind being dev tools generating
precomputed nonzero values).

It is, at least, called out in the documentation, though that's kind of
"be grateful for small favours". :/

/~\ The ASCII				  Mouse
\ / Ribbon Campaign
 X  Against HTML		[email protected]
/ \ Email!	     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B