Time of execution increase more and more

Hello!,

i have implemented a DT-CWT (Dual Tree - Complex Wavelet Transform) on CUDA recently. For tunig purpose, i have been making differents proof and i obtained a curious results. For measuring time of execution, i launched the analisis part of the DT-CWT for 100 times to produce the average of it, the time of successive for loop is increasing the time more and more and i dont know te reason of it. The time is measured from Matlab, the code of DT-CWT is implemented as library, and the for loop is as this:

[i]top = 100;
tic;
for i = 0:top -1;

[indent] [YlC,YhC,Y3C] = dtwavexfm2_cuda_sp(Xs,6);[/indent]
end
toc / top[/i]

Now, the average time, on seconds, obtained for each excution of the same for loop:

0,0168; 0,0177; 0,0186; 0,0197; 0,0200; 0,0208; 0,0220; 0,0224; 0,0231; 0,0241; 0,0246; 0,0253; 0,0268; … and more and more

I need retart Matlab to get the initial good times.

Any idea?

Thanks a lot!.

Richard.