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
C --> access by rows (`i' cycle over the rows, `j' cycle over the columns)
Fortran --> access by column (`i' cycle over the cols, `j' cycle over the rows)
Advices for assignments and other things:
1) run tests for ~ 3 secs
2) plot the flop/s, and the L1, L2, L3 sizes.
3) use a batch script (given by professor) to run an executable on the cluster with reserved resources (in order to avoid having a single cpu reserved on the cluster).
4) in assignment 1 there's a file matmult_X.compiler: if we access the matrix with [i][j] notation use the C version in place of the X, if we use [i+..j] use the F in place of the X. Use .studio if we use the Sun Studio compiler, otherwise the .gcc in case of the GCC comp.
In GPUs the access to matrices has to be done column-wise, otherwise the access could be strided.