Re: [RFC] Don't schedule prefetch INSNs

[Adam Nemet <anemet at caviumnetworks dot com>]

Andreas Krebbel writes:
> > Do you mean this within the same basic block?  Prefetching something that will
> > be accessed in the same basic block is probably not very profitable (unless
> > the block is huge).
> 
> It will be worthwhile if the basic block is a loop body. It might make 
> sense to prefetch something used in the next loop iteration. Actually 
> that's what I am trying to do for memcpys.

I meant basic block in the same iteration but I see from your other mail that
the problem is that the prefetch is moved at the end of the previous
iteration, which makes it less effective.

Does the prefetch-inseration pass do something special to keep the prefetches
in place?

Anyway, I'd like to see some improvement in this area because I've seen it
many times too that prefetches are really hard to keep in place.  Your patch
might be a good step in this direction.  Thanks for working on this.

Adam