Xarray optimizations

Some of our xarray handling is taking a lot of the total simulation time in the minimal runs in benchmarking and profiling.

Potential optimization candidates:

• to_xarray()
  • Partially done in !222 (merged) (800 us -> 180 us)
  • Further optimized in !223 (merged) and !224 (merged) (-> 90 us)
• check() (30%)
  • Partially done in !222 (merged) (500 us -> 170 us)
  • Further optimized in !223 (merged) and !224 (merged) (-> 80 us)
  • No call to equivalent_units() in !228 (merged) (-> 35 us)
• get_time() (12%) (>10% from to_datetime())
  • Done in !222 (merged) (230 us -> 16 us)
• assign_time() (37%) (used in component)
  • Done in !222 (merged) (680 us -> 16 us with existing axis)
• to_units()
  • Done in !222 (merged) (320 us -> 12 us for no-op conversion)
• equivalent_units() (13%)
  • Partially done in !222 (merged) (150 us -> 105 70 us)
  • Reduced number of calls in !228 (merged)
• _get_pint_units() (7%)
• is_quantified() (7%)
• Component.logger (5%) -> full name could be cached instead of repeatedly reconstructed, logging.getLogger() also takes 3.6%

Further optimization ideas:

• Store units in info as pint objects instead of strings -> avoid re-constructing

added performance label

changed the description

changed the description

changed the description

changed the description

changed the description

marked the checklist item _get_pint_units() (7%) as completed

marked the checklist item is_quantified() (7%) as completed

added in-progress label

marked the checklist item to_xarray() as completed

marked the checklist item check() (30%) as completed

marked the checklist item Component.logger (5%) -> full name could be cached instead of repeatedly reconstructed, logging.getLogger() also takes 3.6% as completed

changed the description

changed the description

Mostly done, with over 90% gain for the minimal simulation.

Tracking equivalent_units() in #91 (closed).

closed

changed milestone to %release-0.4