Test failing

[giuliojiang]

/usr/bin/timeout: the monitored command dumped core
echo parallelism = X X
(standard_in) 1: syntax error
gj414, 33, Timing hpce.gj414.fast_fourier_transform_taskgroup with 4 CPUs and n=2^20: check observed parallelism (totalTime/wallTime) of is more than 2.5., FAIL

Seems my executable segfaults? But it doesn't segfault when I test it with all the parameters required for the graphs.

The same happens for all tests 33-38

[m8pple]

Remember that I am testing in a "clean" environment, not your environment.

I strongly suggest that you try it in a clean environment (e.g. a fresh AWS instance,
or a fresh mingw install), and you should see that bin/test_fourier_transform crashes
for you too. So get a completely clean build environment and do:

git clone https://github.com/HPCE/hpce-2017-cw2-$LOGIN.git
cd hpce-2017-cw2-$LOGIN
make bin/test_fourier_transform
bin/test_fourier_transform  hpce.$LOGIN.fast_fourier_transform_combined

I'm pretty sure it will crash (well, I know it will).

Think about it - what did you change in your environment during the experiments,
that won't have been changed in a clean env?

[giuliojiang]

ok

I'm now able to get the segfault on a different machine, I should be able to fix this.

thakns

[peterith]

@jjd06 Do we need to include default parameter values for HPCE_FFT_RECURSION_K and HPCE_FFT_LOOP_K in order to pass test cases 33-38? Because at the moment, I get

pr615, 33, Timing hpce.pr615.fast_fourier_transform_taskgroup with 4 CPUs and n=2^20: check observed parallelism (totalTime/wallTime) of NoOutput is more than 2.5., FAIL
pr615, 34, Timing hpce.pr615.fast_fourier_transform_parfor with 4 CPUs and n=2^24: check observed parallelism (totalTime/wallTime) of NoOutput is more than 1.1., FAIL
pr615, 35, Timing hpce.pr615.fast_fourier_transform_combined with 4 CPUs and n=2^20: check observed parallelism (totalTime/wallTime) of NoOutput is more than 2.5., FAIL
pr615, 36, For 4 CPUs, n=2^12, direct: check time for serial (7.4549179880) is greater than parfor (CommandFailed)., FAIL
pr615, 37, For 4 CPUs, n=2^22, fast: check time for serial (4.2280187170) is at least 2x that of taskgroup (CommandFailed)., FAIL
pr615, 38, For 4 CPUs, n=2^22, fast: check time for parfor (CommandFailed) is at least 1.5x that of combined (CommandFailed)., FAIL

My code currently looks like this:

const size_t HPCE_FFT_RECURSION_K = atoi(getenv("HPCE_FFT_RECURSION_K"));
const size_t HPCE_FFT_LOOP_K = atoi(getenv("HPCE_FFT_LOOP_K"));

[jjd06]

Yes. These parameters are for you to experiment with during development (environment variables make this easy), then fix to the values you find to be "best."