|
From: Nicholas N. <nj...@cs...> - 2006-09-04 10:07:04
|
On Mon, 4 Sep 2006, Christoph Bartoschek wrote: > what is the expected runtime degradation by using callgrind with cache > simulation? > > Currently I see that starting callgrind with --instr-atstart=no causes a > degradation by a factor of 10 (32 hours instead of 3). Using the real > instrumentation seems to cause another factor of 8. Altogether I see a > slowdown factor of 80 in the part of the programm I want to profile. > > Are the values reasonable or is there something wrong here? They seem reasonable. Cachegrind gets slowdowns anywhere from 20--100 times, and I think Callgrind is doing more than Cachegrind does. Nick |