Investigate GrGeom*Loop uses for parallelizability

[ian-bertolacci]

We want to know what GrGeom*Loops (right now, GrGeomInLoops) are parallelizable, and what are common features that make them not parallelizable currently.

[ian-bertolacci]

Current notes on

• compute_total_concentration.c:L90
  • Reduction of field sum
• l2_error_norm.c:L127 138 153 168 183 199 215
  • reduction of err
• pfield.c:L272
  • has while loop inside
  • some really odd indexing
• problem_ic_phase_pressure.c:L408, 420, 630, 643
  • reduction of nonlin_resid
• problem_ic_phase_pressure.c:L572
  • gather stencil, probably fine
• problem_ic_phase_pressure.mask.c:L384, 396, 601, 614
  • reduction of nonlin_resid
  • Likely same issues in problem_ic_phase_pressure.org.c, but I don't think that file is used.
• problem_mannings.c:L192
  • x is assigned but unused. is it used later? Unlikely, but concerning.
• problem_phase_rel_perm.c:L1060 1163
  • some VanGLookupSpline thing
• problem_phase_rel_perm.c:L1655
  • use of for loop; looks like reduction to index owned array variable
• problem_phsae_rel_permc:L199
  • x is assigned but unused. is it used later? Unlikely, but concerning.
  • this kind of situation happens a lot, and I am likely to miss them.
• problem_real_space_z.c:L133
  • use of goto (which is immediately below), but is unconditional. Seems like this could be rewritten as something else.
• problem_saturation.c:L565 584
  • use of for loop; looks like reduction to index owned array variable. does not use variables that are (at least from this perspective) also written by this loop.
• richards_jacobia_eval.c:L596
  • Scatter pattern similar to NL_function_eval
• solver_richards.c:L2635
  • side effect: printf
• tuning_bandsRF.c:L243
  • doing_TB is assigned (unconditionally to 1) and used after loop