Il sistema di calcolo ad alte prestazioni del CINECA: evoluzione di ...

Recommendations

Info

MPI_INIT_Thread support ( MPI-2) MPI_INIT_THREAD (required, provided, ierr) IN: required, level of thread support (integer)). OUT: provided, level provided (integer). provided (coud be less than required). Four levels are supported: : MPI_THREAD_SINGLE: Only one thread will run. Equivalent to MPI_INIT MPI_THREAD_FUNNELED: Processes may be multithreaded, - all communication is done by one OpenMP thread, - but this may occur inside parallel regions - other threads may be computing during communication MPI_THREAD_SERIALIZED: processes could be multithreaded. More threads may make MPI calls, but only one at a time. MPI_THREAD_MULTIPLE: Multiple threads may make MPI calls, with no restrictions. www.cineca.it 22
MPI_THREAD_SINGLE (master only) !$OMP PARALLEL DO do i=1,10000 a(i)=b(i)+f*d(i) enddo !$OMP END PARALLEL DO call MPI_Xxx(...) !$OMP PARALLEL DO do i=1,10000 x(i)=a(i)+f*b(i) enddo !$OMP END PARALLEL DO #pragma omp parallel for { for (i=0; i
Page 1 and 2: CINECA Casalecchio di Reno (BO) www
Page 3 and 4: Parallel Architectures P P M P P SM
Page 5 and 6: Process A process is created by the
Page 7 and 8: MPI Execution Model - Single Progra
Page 9 and 10: Processors: trends Aggregate Number
Page 11 and 12: Hybrid Model Multi-node SMP (Symmet
Page 13 and 14: www.cineca.it 13
Page 15 and 16: Domain decomposition /1 The hybrid
Page 17 and 18: MPI plus OpenMP Pro: Better use o
Page 19 and 20: Hybrid pseudo code call MPI_INIT (i
Page 21: Mixed paradigm www.cineca.it 21
Page 25 and 26: MPI_THREAD_FUNNELED - calls outside
Page 27 and 28: MPI_THREAD_SERIALIZED • Outside t
Page 29 and 30: HPC Evolution From Herb Sutter Moor
Page 31 and 32: Real HPC Crisis is with Software A
Page 33: Some Current Unmet Needs Performanc

Il sistema di calcolo ad alte prestazioni del CINECA: evoluzione di ...

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?