iencinas10 / cpds Goto Github PK

Report: Good work. However, even when we "do not expect you to fully optimize the codes and get high performance (mainly for MPI)", it'd be good that you evaluate and report the speedup (or slowdown) observed for each code.

• columna en la tabla del latex

Part 2:- Copy of uhelp into u perfectly optimized. Thumbs up.- Sticking to NB equal to 8 for Jacobi is not good for certain values of the number of threads. You could use omp_get_max_threads() as you did for Gauss-Seidel. Also, you could try collapsing the two outer loops.

• finge que haces al go para que se calle

Part 3:- in relax_gauss, in general processes need to receive the corresponding part of the last row computed in the previous block (above) before performing the computation on the current block. Such communication is missing. Are 4 loops needed now?

• La comunicación de los cojones está, no sé que se ha fumado

Part 4:- Well done!

gracias un saludo

ggwp