On Thu, Oct 10, 2013 at 04:48:05PM +0530, Umesh Kalappa wrote:
> where A has value oxff and B has value 0xee.
If the 16-bit registers overlap the 8-bit ones, then IMNSHO you don't
want to use different register numbers for the pairs, instead you just tell
GCC that HImode values must be put into even register numbers. So
you would have say (reg:QI 0 a), (reg:QI 1 b) and (reg:HI 0 a) where
the last one just would be printed as ab rather than a. You can find
many ports where register names differ depending on mode, including e.g.
i386; on say sparc the first 32 floating point registers can be used either
individually as SFmode (32-bit), or even pairs of them as DFmode (64-bit),
etc.
Jakub
