From mboxrd@z Thu Jan 1 00:00:00 1970 From: bugzilla-daemon-CC+yJ3UmIYqDUpFQwHEjaQ@public.gmane.org Subject: [Bug 90887] PhiMovesPass in register allocator broken Date: Sat, 22 Aug 2015 22:41:56 +0000 Message-ID: References: Mime-Version: 1.0 Content-Type: multipart/mixed; boundary="===============1751857651==" Return-path: In-Reply-To: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: nouveau-bounces-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org Sender: "Nouveau" To: nouveau-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org List-Id: nouveau.vger.kernel.org --===============1751857651== Content-Type: multipart/alternative; boundary="1440283316.6dcD2.11222"; charset="UTF-8" --1440283316.6dcD2.11222 Date: Sat, 22 Aug 2015 22:41:56 +0000 MIME-Version: 1.0 Content-Type: text/plain; charset="UTF-8" https://bugs.freedesktop.org/show_bug.cgi?id=90887 --- Comment #38 from jr --- (In reply to Ilia Mirkin from comment #37) > I think this is too big of a problem to fix in a fixup, esp since we don't > really know what these are used for. > > I'm thinking the following: > > (a) Push my fixup patch with the hash map, cc'd to stable. It's a little > inefficient, but it *works*, and is (in my mind) quite simple. I'll > double-check that it doesn't affect perf too much. > (b) Figure out this edge type insanity, including fixing things up, and > maybe adding a validator that makes sure that the edge types are correct. > (c) Revert my fixup, and implement actual edge splitting, (i.e. what you > did) and based on critical edges rather than the current logic. > (d) Lots and lots of testing. > > Does that sound reasonable? One of the considerations here is that neither > you nor I have a *ton* of time to play around with this, and Mesa 11.0.0 > will be released in mid-September, and I'd like to have *some* fix for nv50 > in there. Looks like a good plan. Your patch shouldn't make anything worse and has been tested to fix the reported problems, so this and the other bug can be closed. > As for your patch, the edge is preserved in all cases except forward, which > becomes cross. (draw it out, should make sense, unless I messed up) I'm afk for a week, but will take a longer look. Actually it shouldn't be too difficult, since there are not that many places creating edges. -- You are receiving this mail because: You are the QA Contact for the bug. You are the assignee for the bug. --1440283316.6dcD2.11222 Date: Sat, 22 Aug 2015 22:41:56 +0000 MIME-Version: 1.0 Content-Type: text/html; charset="UTF-8"

Comment # 38 on bug 90887 from 
(In reply to Ilia Mirkin from comment #37)
> I think this is too big of a problem to fix in a fixup, esp since we don't
> really know what these are used for.
> 
> I'm thinking the following:
> 
> (a) Push my fixup patch with the hash map, cc'd to stable. It's a little
> inefficient, but it *works*, and is (in my mind) quite simple. I'll
> double-check that it doesn't affect perf too much.
> (b) Figure out this edge type insanity, including fixing things up, and
> maybe adding a validator that makes sure that the edge types are correct.
> (c) Revert my fixup, and implement actual edge splitting, (i.e. what you
> did) and based on critical edges rather than the current logic.
> (d) Lots and lots of testing.
> 
> Does that sound reasonable? One of the considerations here is that neither
> you nor I have a *ton* of time to play around with this, and Mesa 11.0.0
> will be released in mid-September, and I'd like to have *some* fix for nv50
> in there.

Looks like a good plan. Your patch shouldn't make anything worse and has been
tested to fix the reported problems, so this and the other bug can be closed.

> As for your patch, the edge is preserved in all cases except forward, which
> becomes cross. (draw it out, should make sense, unless I messed up)

I'm afk for a week, but will take a longer look. Actually it shouldn't be too
difficult, since there are not that many places creating edges.

You are receiving this mail because:
  • You are the QA Contact for the bug.
  • You are the assignee for the bug.
--1440283316.6dcD2.11222-- --===============1751857651== Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: base64 Content-Disposition: inline X19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX18KTm91dmVhdSBt YWlsaW5nIGxpc3QKTm91dmVhdUBsaXN0cy5mcmVlZGVza3RvcC5vcmcKaHR0cDovL2xpc3RzLmZy ZWVkZXNrdG9wLm9yZy9tYWlsbWFuL2xpc3RpbmZvL25vdXZlYXUK --===============1751857651==--