Wire up Q/DQ elements in the FInAT interface #71
Conversation
| @@ -195,7 +190,7 @@ def statement_evaluate(leaf, parameters): | |||
| table[abs(table + 1.0) < epsilon] = -1.0 | |||
| table[abs(table - 0.5) < epsilon] = 0.5 | |||
| table[abs(table + 0.5) < epsilon] = -0.5 | |||
| init = coffee.SparseArrayInit(table, parameters.precision, nz_bounds) | |||
| init = coffee.ArrayInit(table, parameters.precision) | |||
There was a problem hiding this comment.
We assume this basically doesn't matter because the big zero blocks are gone in this implementation anyway.
There was a problem hiding this comment.
Yes. No zero blocks from mixed or vector. There are still zeros if many basis functions are zero at the quadrature points, like here.
There was a problem hiding this comment.
Actually what I have on the link is something I might be able to optimise away.
There was a problem hiding this comment.
Well, this has more compact tables, but code actually looks worse. I think it is better to just leave it like above for now.
| for elem in element.sub_elements()]) | ||
|
|
||
|
|
||
| quad_opc = ufl.TensorProductCell(ufl.Cell("interval"), ufl.Cell("interval")) |
There was a problem hiding this comment.
I guess you could take this opportunity to rename this variable as quad_tpc?
|
Refactoring FInAT API before merging this... |
Requires FInAT/FInAT#14, avoids coneoproject/COFFEE#97.