Closed cwschilly closed 11 months ago
Here is the current comparison between the reader (input JSON) and writer (output JSON)
The user_defined data is now written out, but the collection_id
and index
still need to be transferred.
In the case of the memory-toy-problem, the index is [rank_index, decomp_index_on_rank, task_index_on_decomp]
. I don't actually see the index being read in by the VTDataReader.
rank_index
is output with "home": rank_id
, and decomp_index_on_rank
is used to calculate shared id
,[^1] but the VTDataReader doesn't actually read it in. Similarly, I can't find any use of task_index_on_decomp
.
It seems therefore that we may not have to write out index
, since it isn't read in. Is this accurate @ppebay ?
I also don't see the collection_id
being read in, and I can't tell why it is set to 7 everywhere.
[^1]: shared_id = decomp_index_on_rank * num_ranks + rank_index
Fixes #421