This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Graph coloring for register allocation?
On Wed, Jan 24, 2001 at 10:47:14AM -0800, Zack Weinberg wrote:
> I had thought this was a problem with all SMALL_REGISTER_CLASSES
> machines, but it seems not.
Actually, we're nearly at the point where we could run scheduling
before reload and not abort.
A while ago I solved the problem of moving the copy from hard
return register to pseudo away from the call instruction. What's
left is making sure that copies from hard argument register to
pseudo don't move from the start of the function.
Of course, that doesn't address the fact that the current
scheduler doesn't even pretend to consider register pressure,
so it's a questionable thing whether sched1 would be a win.
r~