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

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=90887

--- Comment #31 from Ilia Mirkin <imirkin at alum.mit.edu> ---
(In reply to jr from comment #30)
> (In reply to Ilia Mirkin from comment #29)
> > Hmmmm... maybe it was one of my local patches? I'll try to figure out what
> > went wrong later tonight.
> > 
> > I added logic to be clever about edge types... when splitting an edge, the
> > type should remain except a forward edge becomes a cross edge. (You can do
> > it out on paper...) Perhaps that upset things? It doesn't seem like codegen
> > is using those terms in the usual MST meanings :(
> 
> I did take a look at the edge classification. Cannot say I fully understand
> the implications yet, but I'm wondering whether the logic in
> Graph::classifyDFS is correct. Shouldn't the condition for FORWARD edge
> whenn looping over incoming edges (the second loop) be reversed?

It's definitely *not* correct. On the bright side, it's never used.

I tried using it when I was implementing a SPIR -> NV50 IR converter, and it
was a pile of fail.

Unfortunately I didn't have time last night to investigate this. But I'll
definitely get *some* fix into mesa 11. Sorry this has dragged on for so long.

Really if I could understand wtf the needNewElseBlock logic was trying to do,
and could construct a test shader to hit this in *regular* scenarios, not just
the lowered output of TXL, that would make me a lot more comfortable with any
approach that we pick.

But really I'm just going to make it return true always, and play around with
stuff.

-- 
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/20150821/c17d48f2/attachment.html>