* Improve diffexp for tiledb
- The rows from the A and B sets are gathered and processed at the same time. In this
way the matrix is only accessed once instead of twice for each tile.
- There is now a single thread queue that gets shared between all callers of the diffexp.
This will slow down work if diffexp gets too busy.
- There is a target_workunit amount of work given to each thread. Previously the
workunit was (rows selected * width of tile), which could be small. Now multiple
column tiles can be combined into one workunit. If the target is too small then
thread and other overheads may reduce performance. If target_workunit is too large
then the size of the gathered sub matrix may take up too much memory.
- add configuration parameters (max_workers, cpu_multiplier, and target_workunit)
* Specialize diffexp for tiledb
This patch adds a new diffexp algorithm which is tuned for tiledb.
This algorithm was written by Bruce and is adapted here to plug into the
current framework. The anndata_adaptor still calls the original
algotithm (which was move from diffexp.py to diffexp_generic.py).
The cxg_adaptor now calls the new diffexp_tiledb version. Some
code is shared between the two.
This is part 1 of the diffexp for tiledb. Further tuning and
global throttles are still needed.
A script to run and time diffexp with various options is also
added: test/run_diffexp.py.
