[Nouveau] [Bug 90887] PhiMovesPass in register allocator broken

bugzilla-daemon at freedesktop.org bugzilla-daemon at freedesktop.org
Tue Jun 16 04:28:57 PDT 2015


--- Comment #4 from jr <j-r at online.de> ---
Thanks for reviewing and fixing the other patch! Meant to send it to the list,
but was too busy elsewhere last week.

I've been thinking a little about a better way to fix this inbound edge
ordering dependency. Unfortunately it seems that unless there is a way to
efficiently recompute the information there always will be an invariant that
has to be respected by graph transformations (and which attach and detach
simply cannot guarantee on their own).

And because the PhiMoves pass actually needs the corresponding inbound edge
instead of the BB the current encoding makes sense.

The conclusion being that neither the encoding of the information nor graph
transformations during SSA per se are the real problem. But attach and detach
are not the right tools to do graph transformations (at least during SSA). It
may be an improvement to add some more high level helpers, e.g. an Edge::split
for this particular problem. Would there be interest in a patch in that
direction? (Seems easy enough, but I cannot promise I'm up to the task,

You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20150616/f8ed546e/attachment.html>

More information about the Nouveau mailing list