Scheduling of asm()

[Paul Koning]

Excerpt of message (sent 4 January 2002) by Richard Henderson:
> On Fri, Jan 04, 2002 at 05:55:52PM -0500, Paul Koning wrote:
> > Supplying the schedule class seems like the right answer..
> 
> Until you consider how much compiler internals that exposes
> to the user's source code, at which point we can't reasonably
> change anything in the backend any more.

Hm.  True...

My example of the prefetch instruction was wrong, but I think it is
still true that you may be supplying instructions that the compiler
doesn't know about.  

So how about saying "schedule this asm the same way you would schedule
an XYZ opcode" where XYZ is an opcode that the compiler does know?

That doesn't sound as easy as simply supplying the scheduling class,
since you have to do a mapping from opcode to class (correct me if I'm
wrong, but that doesn't sound like a mapping that's currently readily
available).  It has the advantage of tying the user source code to
something that is more likely to be constant over time.

	  paul