This is the mail archive of the
mailing list for the GCC project.
Re: IRA patch: move clobbers downwards
On 07/13/2010 12:15 PM, Steven Bosscher wrote:
> Anyway, it seems to me that the scheduler should not move the clobber
> up for a case like this, and that this is the real problem. Perhaps
> the scheduler can be taught to give clobbers a low-enough priority to
> schedule them no earlier than the first instruction that depends on it
> (assuming the scheduler creates dependencies of the SETs on the
I looked at that. The problem is that you need to schedule the clobber
to make the instruction that follows it ready, so you'd want to schedule
them as soon as possible, really. I tried putting them in a little
queue on the side and emitting them only once they became necessary
(i.e. we're about to schedule an insn that sets the reg), but that got
quite ugly and didn't work very well. Hence, the IRA patch.
If you have a nice clean scheduler patch, feel free to post it.