Performance of a Code Migration for the Simulation of Supersonic Ejector Flow to SMP, MIC, and GPU Using OpenMP, OpenMP+LEO, and OpenACC Directives

Table 2

Computing times and speed-up factors obtained using OpenMP on the Xeon CPU with HT disabled. The speed-up factors are calculated with best optimized serial version as reference. With HT: hyper threading, V: vectorization, D: deactivated, and A: activated.

Threads	HT-D V-A	Speed-up
1	34 h 48 m 32 s	(1.00X)
2	19 h 26 m 58 s	(1.79X)
3	12 h 20 m 02 s	(2.82X)
4	09 h 35 m 11 s	(3.63X)
5	07 h 58 m 53 s	(4.36X)
6	07 h 15 m 54 s	(4.79X)
7	06 h 58 m 14 s	(4.99X)
8	06 h 40 m 30 s	(5.21X)
9	06 h 18 m 21 s	(5.52X)
10	05 h 55 m 56 s	(5.87X)
11	05 h 41 m 52 s	(6.11X)
12	05 h 40 m 23 s	(6.14X)