[abc] Compile time benchmarks

Hi all,

I assembled a script to measure compile times of ajc and abc. The script
is in the cvs folder 'benches' and is called buildall.pl.
It compiles the specified benchmark a number of times and computes the
average (removing smallest and greates values first, based on total time
(user+sys), not elapsed time).
Before each compilation, it cleans the output folders. I've re-used the
existing scripts to compile and clean the benchmarks, so I assumed these
work correctly.

This is the command line:
./buildall.pl `cat paper_benchmarks` > buildtimes

The results of a first test run are shown below.

ajc: bean: Total(sys+user) 3.69 Elapsed 3.68 (chopped times (total):
3.68, 3.69)
abc: bean: Total(sys+user) 4.26 Elapsed 7.93 (chopped times (total):
3.64, 4.28)

ajc: figure: Total(sys+user) 3.37 Elapsed 11.3 (chopped times (total):
3.36, 3.37)
abc: figure: Total(sys+user) 3.835 Elapsed 15.125 (chopped times
(total): 3.74, 3.85)

ajc: dcm_aspects: Total(sys+user) 5.68 Elapsed 20.795 (chopped times
(total): 5.68, 5.69)
abc: dcm_aspects: Total(sys+user) 8.82 Elapsed 29.59 (chopped times
(total): 8.73, 8.86)

ajc: simulator_null_checks: Total(sys+user) 4.925 Elapsed 34.515
(chopped times (total): 4.81, 4.93)
abc: simulator_null_checks: Total(sys+user) 8.18 Elapsed 42.685 (chopped
times (total): 8.07, 8.3)

ajc: adaptj-pooling-FFT: Total(sys+user) 8.915 Elapsed 51.55 (chopped
times (total): 7.67, 8.98)
abc: adaptj-pooling-FFT: Total(sys+user) 50.07 Elapsed 102.865 (chopped
times (total): 49.86, 51.31)

If everybody is happy with the format and method, I will run it again
for all benchmarks and with a higher number of rounds.

Sascha
Received on Thu Dec 16 17:38:57 2004