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
Moist physics requires approximation table to be calculated to replace exact solutions.
We have ran into two cases so far, with two less than ideal solutions:
QSat: calculation of the table is done in pure Python, which requires mixed precision cast and requires every literal to be Float() casted
qs_table: calculation is done in stencil, which leads to a more readable literal code than above, but require to "misuse" the stencil system (it's not a stencil - it's off-grid)
Because we don't have off grid Fields, both solution abuse the data dimensions system.This leads to a memory usage explosion. For a 18K entry table, we need to do at minima a (1, 1, KM, 18K) field, which leads to an 18K*KM elements.
Moist physics requires approximation table to be calculated to replace exact solutions.
We have ran into two cases so far, with two less than ideal solutions:
QSat
: calculation of the table is done in pure Python, which requires mixed precision cast and requires every literal to beFloat()
castedqs_table
: calculation is done in stencil, which leads to a more readable literal code than above, but require to "misuse" the stencil system (it's not a stencil - it's off-grid)Because we don't have off grid Fields, both solution abuse the data dimensions system.This leads to a memory usage explosion. For a 18K entry table, we need to do at minima a (1, 1, KM, 18K) field, which leads to an 18K*KM elements.
Parent: #36
The text was updated successfully, but these errors were encountered: