Directory structure

• force_constants - contains force constants files for test materials
• profile_results - contains profiling results
• profile_scripts - contains scripts for running Matlab/Python profiling

Profile results

Directory structure is euphonic_version_brille_version -> matlab/python -> machine -> results files

euphonic_version_brille_version

Each directory is named after the version of Euphonic/Brille used e.g. euphonic_1.0.0_54.g2b9c824_brille_0.6.1.dev16_g7698bb6 is Euphonic version 1.0.0_54.g2b9c824_brille_0.6.1.dev16_g7698bb6 and Brille version euphonic_1.0.0_54.g2b9c824_brille_0.6.1.dev16_g7698bb6.

old directory contains profiling results of an old version of Brille (not recorded) but is retained anyway.

matlab/python

matlab contains results of running Euphonic with Horace via Matlab, python contains Python-only results

machine

Describes the machine the profiling was run on. e.g. windows_i711800H_32GB was run on a Windows machine with an Intel i7-11800H processor and 32GB of RAM.

Results files

eubr_tictoc_Xqpts_Ychunk_Z is created by the eubr_profile.m script with do_profile = false. X is the number of q-points, Y is the chunk variable and Z is the timestamp (so subsequent runs don't overwrite the results file). To avoid the overhead of the Matlab profiler, Matlab's tic and toc were used to get the timing results. The results file contains the time taken to interpolate with Euphonic, to initialise the Brille grid, and to interpolate with Brille. Each is run N (usually 3) times.

interpolate_Z.txt is created by the interpolate_profile.py script. Z is a timestamp so subsequent runs don't overwrite the results file. Python's time is used to get timings. The results file shows the time to interpolate with Euphonic and Brille for 160801 q-points (contained in the profile_scripts/qpts_160801.txt file), the time to initialise the Brille grid, and the time taken to calculate the structure factors (this is just for a comparison to see if structure factor calculations are the bottleneck yet) for the materials in the force_constants directory. Each is run N (usually 3) times.

interpolate_cprofile_Z.txt is created by the interpolate_cprofile.py script. Z is a timestamp so subsequent runs don't overwrite the results file. This does the same thing as the interpolate_profile.py script except uses cProfile to determine specifically where timing is being spent, rather than using Python time. This may not give realistic timings overall but is useful to see where time is actually spent.

powder_map_Z.txt is created by the powder_map_profile.py script. Z is a timestamp so subsequent runs don't overwrite the results file. This results file shows the time taken to run Euphonic's euphonic-powder-map script using DOS weighting (so only calculate_qpoint_frequencies is used) and structure factor weighting (so calculate_qpoint_phonon_modes is used) for both Euphonic and Brille for different materials. The paramaters using in running the script are shown on the first line of the results file. Each is run N (usually 3) times.

powder_map_cprofile_Z.txt is created by the powder_map_profile_cprofile.py script. Z is a timestamp so subsequent runs don't overwrite the results file. This does the same thing as the powder_map_profile.py script except uses cProfile to determine specifically where timing is being spent, rather than using Python time.

brille_npts_Z.txt is created by the brille_npts_profile.py script. Z is a timestamp so subsequent runs don't overwrite the results file. Python's time is used to get timings. This results file shows the time to initialise and interpolate with the Brille grid for different grid sizes. It was used to test if more grid points would affect the interpolation time. Each was repeated N (usually 3) times.

Profile scripts

Most scripts have been briefly described above. The other contained files are:

qpts_160801.txt - a plain text file containing 160,801 q-points. These were generated with Horace 3.6.2 using lattice = [1., 1., 1., 90, 90, 90]; cut = d2d(lattice, [1 0 0 0], [-5,0.025,5], [0 1 0 0], [-5,0.025,5]);. This was used so MATLAB and Python could use the same q-points.

utils.py contains various common utilities for running the scripts. e.g. the platform name, the number of threads to use, the location and names of files and the material information.