next up previous contents
Next: Summary of Experimental Analysis Up: Experimental Analysis Previous: Profiling Results

Timing Results

Timings are approximate and were conducted using a version of the calculator compiled with optimisation flags on a Sparc Ultra 1 model 140 with 64Mb of main memory using the unix command time. The precision is specified in decimal digits after the decimal point (or binary digits past the binary point for integrations or function maximum calculations).



 
Figure 7.4: Example memory usage profiling graph, generated by the GHC profiler applied to the implemented calculator computing the expression $\ln(2)$ to 20 decimal digits
Operation Precision Time (approx.)
exp(1) 50 0.66s
ln(2) 10 1.54s
$\pi$ 50 3.88s
Logistic Map (Dyadic, 10 iterations) 10 16 mins 10s
Logistic Map (Signed Binary, 10 iterations) 10 0.10s
Logistic Map (Signed Binary, 50 iterations) 10 3.27s
ln(2) 30 28.78s
$\pi$ 200 6 mins 43s
$\mathrm{fnmax}(0.23 + 1.1x - x^2,0,1)$ 5 (binary) 34s
$\mathrm{fnmax}(0.23 + 1.1x - x^2,0,1)$ 6 (binary) 5 mins 15s
$\int_{0}^{1} x^2 dx$ 4 (binary) 8 mins 41s


The timings show that performance varies significantly with different operations. The performance of different transcendental functions is in part due to the rate of convergence of the sequences used to compute them. The timings also show that the performance of the functional operations is terrible.


next up previous contents
Next: Summary of Experimental Analysis Up: Experimental Analysis Previous: Profiling Results
Martin Escardo
5/11/2000