Re: Running Octave from Fink?

[Sergei Steshenko]

----- Original Message -----
> From: Ev Yemini <address@hidden>
> To: Sergei Steshenko <address@hidden>
> Cc: "address@hidden" <address@hidden>
> Sent: Tuesday, November 6, 2012 4:39 PM
> Subject: Re: Running Octave from Fink?
> 
> So, since you've asked ...
> 
> In my own software (admittedly of a much smaller scale) I use terminal
> output that looks like:
> "Performing step 3/10"
> 
> To ensure the user understands how things are progressing.
> 
> With ATLAS, I repeatedly saw the line:
> "Read in L1 Cache size as = 32KB"
> 
> I don't know how frequently other people replace their CPUs during the
> install but, in my case, I'd have to rip open my laptop and rip it off
> the motherboard, while everything's running. So, when I see this line
> more than once, I'm worried.
> 
> At some point I saw a small message amongst the terminal output that read:
> "BEGIN STAGE 2-3-2"
> 
> Googling to see how many stages exist (couldn't seem to find it in the
> online ATLAS manual) I found I had a few more to go. Came back a few
> hours later to see:
> "BEGIN STAGE 2-2-1"
> 
> Now, I know the stages were progressing before because we had hit
> 2-4-1 at some point in time. Somehow we swung around to the beginning
> again. This looks exactly like a long infinite loop.
> 
> Apparently I'm in good company because, when I googled to see if other
> people had encountered issues, there were several cases where people
> had quit the install, presuming it had gotten stuck, and then
> restarted it only to learn that indeed, the install is a long
> repetitive process.
> 
> Believe me, despite this long rambling tale of my own whinging, I
> greatly appreciate all the hard work that went into ATLAS. But, would
> it kill to let the user know roughly where they are in the 5+ hour
> install process? A simple note like "performing step 14/1000000000"
> would be quite useful. That said, it's done, the programmers who came
> up with it are clearly amazing and, using this free software, I have
> no right whatsoever to complain.
> 
> Many thanks,
> Ev
> 
> 
> On Tue, Nov 6, 2012 at 1:34 PM, Sergei Steshenko <address@hidden> 
> wrote:
>> 
>> 
>> 
>> 
>> 
>>  ----- Original Message -----
>>>  From: Ev Yemini <address@hidden>
>>>  To: address@hidden
>>>  Cc:
>>>  Sent: Tuesday, November 6, 2012 2:41 PM
>>>  Subject: Re: Running Octave from Fink?
>>> 
>>>  It works! (ATLAS took 7+ hours of what looked like an infinite loop of
>>>  the same output -- at times I thought it was stuck and was tempted to
>>>  interrupt and restart the install)
>>> 
>>>  Just wanted to thank everyone for all their help!
>>> 
>>>  Many thanks & all the best,
>>>  Ev
>> 
>>  I don't know what looks to you as infinite loop.
>> 
>>  This what I have:
>> 
>>  "
>>  wc -l /mnt/sdb8/sergei/AFSWD_debug/build/atlas-3.10.0/build_here/make.log
>>  90077 /mnt/sdb8/sergei/AFSWD_debug/build/atlas-3.10.0/build_here/make.log
>>  ".
>> 
>>  ATLAS build process tries various code snippets with various parameters in 
> order to find the fastest, and these tries are reported in 'make' screen 
> output.
>> 
>>  The screen output may look stuck for a minute or two sometimes, but 
> 'top' shows high CPU utilization by ATLAS test program in such a case.
>> 
>>  Regards,
>>    Sergei.
>> 

Your insistence on "performing step 14/1000000000"-like message is based on 
assumption that the "1000000000" is known.

But it isn't.

ATLAS has a thing called "architectural defaults". For some, but not all, 
processors. And these architectural defaults can be ignored using some 
'configure' command line arguments.

For example, atlas-3.8.4 has architectural defaults for my CPU ("Athlon(tm) 64 
X2 Dual Core Processor 5000+ stepping 02"), and now (atlas-3.10.0) it doesn't. 
By the way, I sent Clint generated by my build architectural defaults, but 
apparently they are not there yet. And Clint lost the ability to generate them 
himself because his computer with the same architecture processor died.

So, total number of steps is unknown by design.

...

I am not an ATLAS developer; I think re-detection of cache size happens because 
some initial steps in which it is can sometimes be skipped, and it was easier 
and less bug prone to leave the detection in later steps (it doesn't take much 
time) rather them to complicate the flow.

...

ATLAS comes with some documents, and the build process is described there - not 
in small details, but to give overall idea. So, if one reads the documents, 
'make' screen output makes much more sense.


Regards,
  Sergei.







>