Christoph:
As far as I understand it, Vampir is a trace-based profiling technique,
whereas TAU is not. Also, I think there are licensing differences
between the two. I think one needs to buy a Vampir license for this to work.
Thanks!
On 12/9/2010 12:00 PM, pycuda-requ...@tiker.net wrote:
------------------------------
Message: 2
Date: Thu, 09 Dec 2010 18:18:40 +0100
From: Christoph Statz<christoph.st...@mailbox.tu-dresden.de>
To: pycuda@tiker.net
Subject: Re: [PyCUDA] Using TAU to profile CUDA
Message-ID:<4d010f70.9070...@mailbox.tu-dresden.de>
Content-Type: text/plain; charset="iso-8859-1"
Hi;
Same thing with vampirtrace:
http://tu-dresden.de/die_tu_dresden/zentrale_einrichtungen/zih/forschung/software_werkzeuge_zur_unterstuetzung_von_programmierung_und_optimierung/vampirtrace/index_html/document_view?set_language=en
It can be used to trace CUDA CallTree/Communication and Performance.
Unfortunately a VT-instrumented pycuda triggers "CUDA Errors".
Basically tau instrumentation may be used with vampirtrace, so it might
be a starting point in getting this working.
Has there already been effort on this topic?
Sincerely,
Christoph
_______________________________________________
PyCUDA mailing list
PyCUDA@tiker.net
http://lists.tiker.net/listinfo/pycuda