Text archives Help
- From: "Rocky Rhodes" <rhodes@sgi.com>
- To: "Hansong Zhang" <hansong@sgi.com>, "Abe Stephens" <abe@sci.utah.edu>
- Cc: <manta@sci.utah.edu>
- Subject: RE: [MANTA] trunk performance on Altix
- Date: Tue, 13 Dec 2005 08:13:50 -0800
Yep. There was a "-p" in there. never mind...
>
-----Original Message-----
>
From: Hansong Zhang [mailto:hansong@sgi.com]
>
Sent: Monday, December 12, 2005 5:21 PM
>
To: Abe Stephens
>
Cc: Rocky Rhodes; 'manta@sci.utah.edu'
>
Subject: Re: [MANTA] trunk performance on Altix
>
>
There appears to be a glitch in Rocky's environment (possibly -p). I
>
compiled the main branch and got the same performance as the Itanium
>
branch with "bin/manta -np 4".
>
Hansong
>
>
Abe Stephens wrote:
>
>
> Is it possible you are compiling with -p?
>
>
>
> Abe
>
>
>
> Rocky Rhodes wrote:
>
>
>
>> When I run bin/manta in the trunk directory, someone creates a
>
>> "gmon.out" file. Whoever is doing this must be calling mcount, but
I
>
>> don't see any obvious calls in the code to anything like gmon or
gprof
>
>> or similar...
>
>>
>
>> Hansong is checking out the trunk and will see if my results are
>
>> reproducible or just somehow related to the history of my tree.
>
>>
>
>> Rocky
>
>>
>
>>
>
>>
>
>>> -----Original Message-----
>
>>> From: Abe Stephens [mailto:abe@sci.utah.edu]
>
>>> Sent: Monday, December 12, 2005 4:55 PM
>
>>> To: Rocky Rhodes
>
>>> Subject: Re: [MANTA] trunk performance on Altix
>
>>>
>
>>> Maybe decrease the sampling frequency? I'm not sure why this
function
>
>>> would show up in one build and not the other. Is it labeled as
being
>
>>>
>
>>
>
>> in
>
>>
>
>>
>
>>> kernel or user space?
>
>>>
>
>>> Abe
>
>>>
>
>>> Rocky Rhodes wrote:
>
>>>
>
>>>
>
>>>
>
>>>> Histx shows over 50% of the trunk time being spent in mcount.
Google
>
>>>> shows this to have something to do with profiling. Does this
ring a
>
>>>> bell?
>
>>>>
>
>>>> Rocky
>
>>>>
>
>>>>
>
>>>>
>
>>>>
>
>>>>
>
>>>>> -----Original Message-----
>
>>>>> From: Abe Stephens [mailto:abe@sci.utah.edu]
>
>>>>> Sent: Monday, December 12, 2005 4:25 PM
>
>>>>> To: Rocky Rhodes
>
>>>>> Cc: Steven G. Parker
>
>>>>> Subject: Re: [MANTA] trunk performance on Altix
>
>>>>>
>
>>>>> Wow, we'll certainlly have to look into this-- I will try to run
a
>
>>>>> profile tonight--
>
>>>>>
>
>>>>> Abe
>
>>>>>
>
>>>>> Rocky Rhodes wrote:
>
>>>>>
>
>>>>>
>
>>>>>
>
>>>>>
>
>>>>>
>
>>>>>> For bin/manta -np4 with cockpit.v3c1 and cockpit-path1, itanium
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>
>
>>>> branch
>
>>>>
>
>>>>
>
>>>>
>
>>>>
>
>>>>>> gives me 3.75 fps and the trunk code gives me 1.34 fps.
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>> -----Original Message-----
>
>>>>>>> From: Abe Stephens [mailto:abe@sci.utah.edu]
>
>>>>>>> Sent: Monday, December 12, 2005 4:01 PM
>
>>>>>>> To: Rocky Rhodes
>
>>>>>>> Subject: Re: [MANTA] trunk performance on Altix
>
>>>>>>>
>
>>>>>>> There were quite a few changes merged across and many style
>
>>>>>>>
>
>>>>>>
>
>> changes
>
>>
>
>>
>
>>>>>>>
>
>>>>>>>
>
>>>>>>>
>
>>>>>>
>
>>>>>> made
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>> during the merge so this type of thing isn't unexpected.
>
>>>>>>>
>
>>>>>>> I am working on merging the last itanium2 changes over (most
deal
>
>>>>>>>
>
>>>>>>>
>
>>>>>>>
>
>>>>>>
>
>>>> with
>
>>>>
>
>>>>
>
>>>>
>
>>>>
>
>>>>>>> texturing) and then I'll do a comprehensive performance
>
>>>>>>>
>
>>>>>>
>
>> comparison.
>
>>
>
>>
>
>>>>>>> Have you tried any of the kdtree scenes?
>
>>>>>>>
>
>>>>>>> Abe
>
>>>>>>>
>
>>>>>>> Rocky Rhodes wrote:
>
>>>>>>>
>
>>>>>>>
>
>>>>>>>
>
>>>>>>>
>
>>>>>>>
>
>>>>>>>
>
>>>>>>>
>
>>>>>>>> I have updated from the trunk and compiled on my Altix.
Bin/manta
>
>>>>>>>> shows me the nice shiny board and checkerboard ground plane,
but
>
>>>>>>>> performance is not what it is in the itanium branch. On a
machine
>
>>>>>>>>
>
>>>>>>>>
>
>>>>>>>>
>
>>>>>>>>
>
>>>>>>>>
>
>>>>>>>
>
>>>>>> with
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>>> 1.6 GHz processors, "bin/manta" gives me 1.4 fps on the
itanium
>
>>>>>>>>
>
>>>>>>>>
>
>>>>>>>>
>
>>>>>>>>
>
>>>>>>>>
>
>>>>>>>
>
>>>>>> branch
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>>> and 0.99 fps on the main trunk. "bin/manta -np 4" gives me
5.5
>
>>>>>>>>
>
>>>>>>>
>
>> fps
>
>>
>
>>
>
>>>>>>>>
>
>>>>>>>>
>
>>>>>>>>
>
>>>>>>>
>
>>>>>> on
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>>>> the itanium branch and 1.6 fps on the main trunk.
>
>>>>>>>>
>
>>>>>>>> I think both are being compiled with "Release" flags (-O3
-IPO
>
>>>>>>>>
>
>>>>>>>
>
>> -g).
>
>>
>
>>
>
>>>>>>>> Rocky
>
>>>>>>>>
>
>>>>>>>>
>
>>>>>>>>
>
>>>>>>>>
>
>>>>>>>>
>
>>>>>>>>
>
>>>>>>>
>
>>>>>>
>
>>>>>>
>
>>>>>
>
>>>>
>
>>>>
>
>>>
>
>>
>
>>
>
>>
>
>
>
>
Archive powered by MHonArc 2.6.16.