Text archives Help
- From: Rocky Rhodes <rhodes@sgi.com>
- To: "'James Bigler'" <bigler@cs.utah.edu>
- Cc: "'Hansong Zhang'" <hansong@sgi.com>, manta@sci.utah.edu
- Subject: RE: [MANTA] RE: Compiler flags and performance on altix?
- Date: Fri, 13 May 2005 08:13:49 -0700
That's exactly what it was doing. I had a hard time finding documentation
on those intrinsics as well. I finally searched for them on the Intel site
and found a few scraps of information. ;^)
Rocky
>
-----Original Message-----
>
From: owner-manta@sci.utah.edu [mailto:owner-manta@sci.utah.edu] On Behalf
>
Of James Bigler
>
Sent: Thursday, May 12, 2005 8:11 PM
>
Cc: 'Hansong Zhang'; manta@sci.utah.edu
>
Subject: Re: [MANTA] RE: Compiler flags and performance on altix?
>
>
It's very likely that this code was using the assumptions that if
>
__fetchadd8_acq returned the new value that __sync_fetch_and_add_di
>
would as well. Glad you were able to catch this.
>
>
BTW, I can't find any documentation on the __sync_fetch_and_add functions.
>
>
James
>
>
Rocky Rhodes wrote:
>
> The reason the -imagedisplay null flag wasn't working was that the ia64
>
> AtomicCounter implementation in SCIRun/Core/Threads/Thread_pthreads.cc
>
> was broken. Seems like a bunch of stuff wouldn't have worked...
>
>
>
> I checked in a fix and the "-imagedisplay null" thing now works. Wonder
>
if
>
> anything else is now broken, though...
>
>
>
> Rocky
>
>
>
>
>
>>-----Original Message-----
>
>>From: Steven G. Parker [mailto:sparker@cs.utah.edu]
>
>>Sent: Thursday, May 12, 2005 4:05 PM
>
>>To: Rocky Rhodes
>
>>Cc: 'Abe Stephens'; 'Hansong Zhang'
>
>>Subject: Re: Compiler flags and performance on altix?
>
>>
>
>>I actually get 4.5 frames per second on the G5, and 8.3 fps for two
>
>>processors. The altix gets only .77 fps for 1p.
>
>>
>
>>I did:
>
>>bin/manta -np 1 -imagedisplay null -bench
>
>>
>
>>The default bench is equivalent to -bench 100 10. Also, for some
>
>>reason the altix won't parse the -imagedisplay null flag correctly...
>
>>
>
>>Steve
>
>>
>
>>
>
>>On May 12, 2005, at 4:40 PM, Rocky Rhodes wrote:
>
>>
>
>>
>
>>>This is running "bin/manta -bench 10 10"? By "3.5" do you mean 3.5
>
>>>seconds
>
>>>for 10 frames on the g5?
>
>>>
>
>>>We'll think about some compiler flags...
>
>>>
>
>>>
>
>>>
>
>>>>-----Original Message-----
>
>>>>From: Abe Stephens [mailto:abe@sci.utah.edu]
>
>>>>Sent: Thursday, May 12, 2005 3:28 PM
>
>>>>To: Rocky Rhodes; 'Hansong Zhang'
>
>>>>Cc: Steven Parker
>
>>>>Subject: Compiler flags and performance on altix?
>
>>>>
>
>>>>
>
>>>>Hi-
>
>>>>What compilers and compiler option flags have you tried? I have
>
>>>>used the
>
>>>>intel compiler, with pretty much just -O3. I think we are looking
>
>>>>for
>
>>>>some good ones to hard code into the default cmake build on the
>
>>>>different platforms.
>
>>>>
>
>>>>1 Processor -- Benchmark completed in 12.9411 seconds (10 frames,
>
>>>>0.772732 frames per second)
>
>>>>
>
>>>>Steve reports 3.5 on one processor of his g5.
>
>>>>
>
>>>>2 Processors -- Benchmark completed in 6.59063 seconds (10 frames,
>
>>>>1.51731 frames per second)
>
>>>>4 Processors -- Benchmark completed in 3.38019 seconds (10 frames,
>
>>>>2.95841 frames per second)
>
>>>>8 Processors -- Benchmark completed in 1.70183 seconds (10 frames,
>
>>>>5.87604 frames per second)
>
>>>>16 Processors -- Benchmark completed in 0.892033 seconds (10 frames,
>
>>>>11.2103 frames per second)
>
>>>>
>
>>>>
>
>>>>Abe
>
>>>
Archive powered by MHonArc 2.6.16.