I have a routine that my my measurements takes 393219 clock cycles to excute. If I need this to run as quickly as possible and as often as possible is this no. of clock cycles acceptable. I.e I'm using a 1GHz Athlon, how would this run on other machines. I'm also just curious how this compares to other languages. You see I've only recently began to measure clock cycles so at the moment they just seem like numbers, they have no meaning. I suppose its like frame rates in 3d games, At first a statement like 60 fps ment nothing but now that I'm used to it I realise this is quite a good figure.
Posted on 2001-03-17 06:39:00 by Zadkiel
Sorry that should have read 253179 clock cycles, I was calling an extra routine I had written to display ten floating point values in edit boxes for debuging purposes. Its hard to believe that those ten values took approx 140000 clock cycles to excute about 60% of the actual routine
Posted on 2001-03-17 06:45:00 by Zadkiel
Zadkiel, Benchmarking contrary to popular opinion is a very useful technique where performance matters. When you are targetting a particular algorithm, the requirements tend to be speed of repetition of the test and convenience of use. This allows you to modify parts of the code and re-test it to see if the changes are faster or slower. My own preference is real time testing where the code is set up to run for over a half a second so that the variations drop below 1%. As long as you are satisfied with the method you are using to get the timings and it is convenient to use, you can set up a test piece and hammer it until it does what you are after. It may sound like a lot of hassle but the results justify the extra work that you do in performance terms. Regards, hutch@pbq.com.au
Posted on 2001-03-17 14:57:00 by hutch--