Port-macppc archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: Confidence: Chopping between 5.2 and 6.0.1

Replying to my own message.

I lied.

After re-testing the "-D 2011-05-03" version, I DID get it to fail.

The test case I've been using runs 20 iterations of about 45 seconds each.
Normally, one run is enough, but sometimes it seems to run several iterations
clean, and then fail.  I **think* that something about the load on the
machine helps trigger the error.  I have tried to isolate this, but the
pattern is not clear.

In any case, I ran the test 3 more times on the 2011-05-03 build, and it
failed 2 of the 3 runs.

-D 2011-05-01 looks clean
-D 2011-05-03 looks bad.

Looks to me like matt broke something. ;->

I kept the several of the relevant build directories, so I am in a good
position to test fixes if they should appear.  I don't have a build dir
for HEAD, but could set one up.  My test scripts require some
relatively large data files, so they are not so easy to "deliver" to
someone else.


At 5:21 PM -0500 3/24/13, Donald Lee wrote:
>I saw this commit and thought that would be the one:
>powerpc: Converted PowerPC to use PCU to manage FPU/AltiVec/SPE. [matt 
>I then did a checkout of -D 2011-05-03 and tested it.  It came through
>CLEAN.  (I am testing some more just to be sure.)
>-D 2011-05-03 looks clean
>-D 2011-05-05 looks BAD
>-D 2011-05-04 is being built as we speak....
>Matt's commit above allegedly went in on May 2nd, but I am cautious about my
>conclusions, because I don't remember the tome zone rules with CVS.
>As I recall, everything in CVS is done in UTC, so the commit above
>could still be the culprit, depending on what time of day it was committed,
>and in what time zone.
>In any case, I'm testing "-D 2011-05-04" and will report back when done.
>At 12:09 PM +0000 3/24/13, David Brownlee wrote:
>>Looks like you're getting close to identifying the offending commit :)
>>I wonder if it could be something in the pcu changes (
>>On 23 March 2013 22:28, Donald Lee <MacPPC2%c.icompute.com@localhost> wrote:
>>> At 2:12 PM -0400 3/23/13, Michael wrote:
>>>>On Fri, 22 Mar 2013 23:38:17 -0500
>>>>Donald Lee <MacPPC2%c.icompute.com@localhost> wrote:
>>>>> I have converged on a 99% solution for my "chop".  Here is my
>>>>> data:
>>>>> 2011-04-15 - usr9 - netbsd 5.99.49 - clean - clean2 - c3 c4 c5
>>>>> 2011-04-25 - usrd - (buildfail)      netbsd 5.99.50
>>>>> 2011-05-05 - usra - netbsd 5.99.51 - clean - bad2 - bad3
>>>>> 2011-05-11 - usrc - netbsd 5.99.51b - clean - clean bad1 bad2 bad3
>>>>> 2011-05-17 - usrb - netbsd 5.99.51a - bad - bad - bad - bad
>>>>> These are the checkout dates I tried, and the test results.
>>>>> The 2011-04-15 version was the latest one where I saw no errors, even 
>>>>> though
>>>>> I ran the test case (at least) 5 times.  No errors on versions prior to 
>>>>> that
>>>>> date.
>>>>> The 2011-04-25 checkout did not build, unfortunately.
>>>>> The 2011-05-05 and 2011-05-11 builds looked OK, but when I re-ran the
>>>>> tests, I did see failures.  For the 2011-05-11 version especially, it
>>>>> was hard to get an error, but I did get them.  2011-05-17 (and later)
>>>>> seems to fail freely
>>>>> This puts the timeframe about 3 weeks - somewhere between 2011-04-15
>>>>> and 2011-05-05, code went in to cause our badness.
>>>>There was a bunch of altivec and fpu handling related commits on May 2nd:
>>>>Might be worth to check right before / after that.
>>>>have fun
>>> I'll see if May 1 builds.
>>> -dgl-

Home | Main Index | Thread Index | Old Index