Add Blocked timestepping support to PyRevolve #55
Conversation
|
Can you please clean up the commit history on this a bit? e.g. I see some commits undoing things done by previous commits. |
speglich
force-pushed
the
blocked_timestepping
branch
from
90c0b7f to
58fd043
Compare
Done. |
|
This is showing a merge conflict after merging the previous PR. Could you please do a rebase on current master? |
speglich
force-pushed
the
blocked_timestepping
branch
from
58fd043 to
6844553
Compare
Done. |
|
Help me to remember... there is something missing @navjotk? |
| assert(t_start <= t_end) | ||
| if self.direction == 1: | ||
| self.u[:] = self.u[:] + self.direction * abs(t_start - t_end) | ||
| for t in range(t_start, t_end): |
There was a problem hiding this comment.
We added a new dimension to the u array. This dimension is responsible to save the block_size time steps. That before are not saved because isn't necessary. This loop is responsible to write each time step in the right position.
| if self.direction == 1: | ||
| self.u[:] = self.u[:] + self.direction * abs(t_start - t_end) | ||
| for t in range(t_start, t_end): | ||
| idx = t % np.shape(self.u)[0] |
There was a problem hiding this comment.
There must be a better way of getting block_size
| self.u[:] = self.u[:] + self.direction * abs(t_start - t_end) | ||
| for t in range(t_start, t_end): | ||
| idx = t % np.shape(self.u)[0] | ||
| past_idx = (t-1) % np.shape(self.u)[0] |
There was a problem hiding this comment.
You're assuming that the operator moves one time step at a time, when this has is supposed to go from t_start to t_end.
| for t in range(t_start, t_end): | ||
| idx = t % np.shape(self.u)[0] | ||
| past_idx = (t-1) % np.shape(self.u)[0] | ||
| self.u[idx][:] = self.u[past_idx][:] + self.direction * 1 |
There was a problem hiding this comment.
Again, assuming one time step at a time. Not true.
| self.u[idx][:] = self.u[past_idx][:] + self.direction * 1 | ||
| else: | ||
| self.v[:] = (self.u[:]*(-1) + 1) | ||
| idx = (t_start) % np.shape(self.u)[0] |
There was a problem hiding this comment.
You have to fill in all elements from t_start to t_end.
|
@speglich any progress here? |
speglich
force-pushed
the
blocked_timestepping
branch
from
6844553 to
573ae8e
Compare
2 participants
Closes #54