Re: Extreme PRE compile times

[Richard Henderson <rth at redhat dot com>]

On Thu, Nov 13, 2003 at 11:51:45AM -0500, Daniel Berlin wrote:
> It also looks like it's the splitting of critical edges that exposes them,
> at first glance.

This is true; I've noticed it myself before as well.

The dom1 pass will for example fail to thread edges when we
cross PHIs at the beginning of the block.  Split the edge and
we'll not cross the PHI and so then dom2 will thread the edge.

I forget which tree-ssa testcase shows this, but you can find it
easily enough by running the testsuite with -fno-tree-pre and
looking at the new failures.


r~