On Thu, 26 Sep 2002, Eric Faurot wrote:
> However, I don't quite understand the code you showed.
> The SMID you describe are not exclusive.
> unless I miss something about what you call SWAR.
I guess there is that point... if sublib A has only an SSE optimized
codepath and sublib B has only an MMX optimized code path, we should not
force one of the sublibs to turn optimizations off. So a bitflag
might make more sense, yes.
However, I'd like to see all the SWAR defines unified regardless of the
processor family, to keep the level of #ifdef and switch{} nesting low
in the source code.
The only question then is, will they all fit into a uint32 for the
forseeable future? :-)
/me has a brainflash:
If we make the return type 64-bit on 64-bit architectures, then
we could define bitflags for SWARs that are only available on 64-bit
machines in the upper dword.
... erm... wonder if that would work out well...
--
Brian
