ctxt_.comm = MPI_COMM_WORLD or can it be divided into groups
Posted: Tue Sep 03, 2013 10:02 pm
Hi,
On line 635 of FourierTransform.C, one has the following:
#if USE_MPI
int status = MPI_Alltoallv((double*)&rbuf[0],&rcounts[0],&rdispl[0], MPI_DOUBLE,(double*)&sbuf[0],&scounts[0],&sdispl[0],MPI_DOUBLE, ctxt_.comm());
assert ( status == 0 );
#else
…
Is there a parameter that modifies the last parameter, ctxt_.comm(), in some ways or another? In other words, is it always MPI_COMM_WORLD, or can we have it grouped in some ways using a certain parameter? For example, if I compare with Quantum Espresso, one can use a variable called "-ntg 4" that will subdivide the communications into 4 groups if I choose 4 as the number of groups. Is there a parameter that does a similar effect with Qbox ?
Thank you,
Pierre
On line 635 of FourierTransform.C, one has the following:
#if USE_MPI
int status = MPI_Alltoallv((double*)&rbuf[0],&rcounts[0],&rdispl[0], MPI_DOUBLE,(double*)&sbuf[0],&scounts[0],&sdispl[0],MPI_DOUBLE, ctxt_.comm());
assert ( status == 0 );
#else
…
Is there a parameter that modifies the last parameter, ctxt_.comm(), in some ways or another? In other words, is it always MPI_COMM_WORLD, or can we have it grouped in some ways using a certain parameter? For example, if I compare with Quantum Espresso, one can use a variable called "-ntg 4" that will subdivide the communications into 4 groups if I choose 4 as the number of groups. Is there a parameter that does a similar effect with Qbox ?
Thank you,
Pierre