Freeze constant arrays #354
Conversation
| @@ -232,6 +232,11 @@ class _FromGroupPickData(Generic[ArrayT]): | |||
| from_element_indices: ArrayT | |||
| is_surjective: bool | |||
|
|
|||
| @memoize_method | |||
| def indexed_dof_pick_lists(self, actx): | |||
There was a problem hiding this comment.
Not completely sure if this is the right object to maintain the memoization dict. (Lmk if you think something's not ideal here)
kaushikcfd
force-pushed
the
freeze_constants
branch
from
11339a4
to
d27c898
Compare
|
Rerunning to check how things go now that inducer/arraycontext#203 is in. |
kaushikcfd
force-pushed
the
freeze_constants
branch
from
d27c898
to
0591cbb
Compare
| @@ -232,6 +232,12 @@ class _FromGroupPickData(Generic[ArrayT]): | |||
| from_element_indices: ArrayT | |||
| is_surjective: bool | |||
|
|
|||
| @keyed_memoize_method(key=lambda actx: type(actx)) | |||
| def indexed_dof_pick_lists(self, actx): | |||
There was a problem hiding this comment.
Could you explain why this is needed? I would have figured that DataWrapper deduplication would take care of this.
There was a problem hiding this comment.
This is computing pt_data[pt_data_0[i, j]] i.e. a constant that should be computed AOT rather than in every compiled_rhs. See all these terms.
|
As discussed, I think I would prefer to do this as part of a more general constant folding pass in pytato. (And in this particular case, I think the temporaries being produced are |
No description provided.