[WIP] Add gradient subsampling test

[navjotk]

This PR is the next milestone leading up to the compression followup work. It adds a test that computes the gradient normally first, then does it again for dt/2 and then a third time for dt/2 but subsampling 1-in-2 in time. It then asserts equality between the three, ensuring that the subsampling implementation makes sense.

Currently, the norms for the three are:

grad_base 1498.496997261012 grad_2t 2996.1773245927657 grad_sub 1498.087825898719

which means the second case (where dt/2 is used without subsampling) probably needs a dt-related scaling factor somewhere. Can @mloubout point me to the best way to incorporate this factor so this test makes sense?

Also, even though norms are fairly close, test does not pass at tolerance=1e-3. Might be a silly mistake that crept in. Looking for overall feedback at this point.

[codecov[bot]]

Codecov Report

Merging #1532 (7083017) into master (97570bb) will increase coverage by 6.91%. The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #1532      +/-   ##
==========================================
+ Coverage   61.23%   68.15%   +6.91%     
==========================================
  Files         203      124      -79     
  Lines       27608    16501   -11107     
  Branches     3910     2984     -926     
==========================================
- Hits        16906    11246    -5660     
+ Misses      10170     4740    -5430     
+ Partials      532      515      -17     

Impacted Files	Coverage Δ
devito/builtins/initializers.py	10.29% <0.00%> (-73.53%)	:arrow_down:
examples/seismic/model.py	25.00% <0.00%> (-63.05%)	:arrow_down:
devito/data/utils.py	26.20% <0.00%> (-60.27%)	:arrow_down:
devito/passes/clusters/aliases.py	33.13% <0.00%> (-55.89%)	:arrow_down:
devito/finite_differences/coefficients.py	19.29% <0.00%> (-52.64%)	:arrow_down:
examples/seismic/preset_models.py	2.90% <0.00%> (-51.17%)	:arrow_down:
devito/mpi/routines.py	47.31% <0.00%> (-47.64%)	:arrow_down:
devito/builtins/utils.py	49.05% <0.00%> (-47.17%)	:arrow_down:
examples/seismic/utils.py	26.89% <0.00%> (-47.06%)	:arrow_down:
devito/core/autotuning.py	13.04% <0.00%> (-46.09%)	:arrow_down:
... and 149 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 97570bb...7083017. Read the comment docs.

[mloubout]

Stale, closing