You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In process grid load balance strategy, P * Q == NP(number of processes). All processes are shaped into P rows and Q columns. Sparse blocks of distributed matrix are placed on P*Q processes.
For example, there is a matrix, N==6(order is 6), set NB=2(order of each non-zero block is 2), then block_length==3 (ceil(N/NB)). block-level sparse pattern is shown below:
0 1 2
0 * *
1 * * *
2 * *
Numbering each non-zero blocks row by row, then numbers of non-zero blocks are 0~6.
If P==Q==2, numbering processes row by row from 0 to 3, then:
block 0,6 are placed on process 0;
block 1,5 are placed on process 1;
block 2,4 are placed on process 2;
block 3 is placed on process 3.
No description provided.
The text was updated successfully, but these errors were encountered: