This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Jump threading/bypassing volume 3.
On Mon, 2004-08-09 at 20:02, Kenneth Zadeck wrote:
> >
> In the old version of dom, there was one other thing that I noticed.
> The algorithm seemed to be fairly left handed. It was not clear if it
> would get all of the examples in section two of your part 1 note if the
> cfg was reflected left/right. Is this still true?
DPM certainly failed to propagate information from conditionals into
the one arm of the conditional in certain cases (which were similar
to the graphs I showed as problematical for jump bypassing). I fixed
that problem a few weeks ago (using the same methods as I had used to
fix the limitation in the jump bypassing method).
There are a couple cases where we don't try and thread jumps -- those
cases should already be handled by the standard propagation of
equivalences in DOM -- ie, trying to thread those jumps is redundant
with other actions of DOM and would just be a waste of time.
I tested that assertion a short while ago and the assertion was still
true.
Jeff