Results

Table [[*]] compares of the vanilla version of DL_POLY on the XT4 and XE6 with DL_POLY with an optimised version of the routine ewald_spme_forces. This evaluates the reciprocal space contribution to the Ewald energy and forces, and uses the SPME algorithm [Darden et al., J. Chem. Phys., 103, 19, (1995)]. A small decrease in timings can be observed with the Opt ewald version.


Table: Timing comparison of different runs with Bak on XT4, XE6 and optimised ewald_spme_forces on XE6
Nb. Procs Bak XT4 Bak XE6 Opt ewald
16 199.154 218.722 216.172
32 106.790 98.113 94.852
64 63.129 57.494 53.736
128 42.036 39.360 35.639
256 27.471 29.492 27.062
512 22.137 25.951 21.960


In the next table [[*]], as before with link_cell_pairs, I show the percentage speed up due to the optimised code. In this case no improvement is found for the 16 processes run, but in all other cases there is a small improvement a with a peak of 13 less time for the 256 processes run.


Table: Variation rate of different runs of optimised ewald_spme_forces on XE6 with Bak on XT4 and XE6
<#954#> Opt ewald comp. to  
  Bak XT4 Bak XE6
16 8.55 -1.17  
32 -11.18 -3.32  
64 -14.88 -6.54  
128 -15.22 -9.45  
256 -1.49 -8.24  
512 -0.80 -15.38  
Average -5.84 -7.35  


Valène Pellissier 2011-08-24