Fix memory allocations on CPU

[ali-ramadhan]

My fault, must have messed up and forgot to skip _apply_*_bcs! for NotFluxBC in PR #631.

Fixes #675 and should fix failing tests on PR #671.

Before:

 ──────────────────────────────────────────────────────────────────────────────────────
        Static ocean benchmarks                Time                   Allocations      
                                       ──────────────────────   ───────────────────────
           Tot / % measured:                24.3s / 2.17%           1.83GiB / 18.1%    

 Section                       ncalls     time   %tot     avg     alloc   %tot      avg
 ──────────────────────────────────────────────────────────────────────────────────────
  32× 32× 32  [CPU, Float32]       10    256ms  48.6%  25.6ms    170MiB  50.0%  17.0MiB
  32× 32× 32  [CPU, Float64]       10    270ms  51.4%  27.0ms    170MiB  50.0%  17.0MiB
 ──────────────────────────────────────────────────────────────────────────────────────

After:

 ──────────────────────────────────────────────────────────────────────────────────────
       Static ocean benchmarks                Time                   Allocations      
                                      ──────────────────────   ───────────────────────
          Tot / % measured:                 126s / 62.8%           1.46GiB / 0.13%    

Section                       ncalls     time   %tot     avg     alloc   %tot      avg
──────────────────────────────────────────────────────────────────────────────────────
 32× 32× 32  [CPU, Float32]       10    39.4s  49.9%   3.94s   1.00MiB  50.0%   102KiB
 32× 32× 32  [CPU, Float64]       10    39.5s  50.1%   3.95s   1.00MiB  50.0%   102KiB
──────────────────────────────────────────────────────────────────────────────────────

Unfortunately still a bit higher than v0.22.0 (~50 KiB allocations) but much better and more acceptable than 17 MiB!

Remaining memory allocations seem to be occuring in fill_halo_regions.jl but tried inlining some functions and didn't help so I'll revisit the problem in the future.

julia> analyze_malloc(".")
323-element Array{CoverageTools.MallocInfo,1}:        
 ⋮                                                                                                    
 CoverageTools.MallocInfo(5008, "./benchmark/benchmark_static_ocean.jl.32885.mem", 36)                
 CoverageTools.MallocInfo(5952, "./benchmark/benchmark_utils.jl.32885.mem", 35)                       
 CoverageTools.MallocInfo(6080, "./src/TimeSteppers/time_stepping_kernels.jl.32885.mem", 139)         
 CoverageTools.MallocInfo(7136, "./src/Architectures.jl.32885.mem", 39)                               
 CoverageTools.MallocInfo(12160, "./src/Utils/tuple_utils.jl.32885.mem", 14)                          
 CoverageTools.MallocInfo(12480, "./src/BoundaryConditions/apply_no_penetration_bcs.jl.32885.mem", 20)
 CoverageTools.MallocInfo(12480, "./src/BoundaryConditions/apply_no_penetration_bcs.jl.32885.mem", 38)
 CoverageTools.MallocInfo(12800, "./src/BoundaryConditions/fill_halo_regions.jl.32885.mem", 45)       
 CoverageTools.MallocInfo(12800, "./src/BoundaryConditions/fill_halo_regions.jl.32885.mem", 49)       
 CoverageTools.MallocInfo(16640, "./src/BoundaryConditions/fill_halo_regions.jl.32885.mem", 52)       
 CoverageTools.MallocInfo(16640, "./src/BoundaryConditions/fill_halo_regions.jl.32885.mem", 53)       
 CoverageTools.MallocInfo(16640, "./src/BoundaryConditions/fill_halo_regions.jl.32885.mem", 56)       
 CoverageTools.MallocInfo(16640, "./src/BoundaryConditions/fill_halo_regions.jl.32885.mem", 57)       
 CoverageTools.MallocInfo(44000, "./src/BoundaryConditions/fill_halo_regions.jl.32885.mem", 26)       
 CoverageTools.MallocInfo(44000, "./src/BoundaryConditions/fill_halo_regions.jl.32885.mem", 27)       
 CoverageTools.MallocInfo(44000, "./src/BoundaryConditions/fill_halo_regions.jl.32885.mem", 28)       
 CoverageTools.MallocInfo(44000, "./src/BoundaryConditions/fill_halo_regions.jl.32885.mem", 29)       
 CoverageTools.MallocInfo(44000, "./src/BoundaryConditions/fill_halo_regions.jl.32885.mem", 30)       
 CoverageTools.MallocInfo(46400, "./src/BoundaryConditions/apply_flux_bcs.jl.32885.mem", 17)          
 CoverageTools.MallocInfo(46400, "./src/BoundaryConditions/apply_flux_bcs.jl.32885.mem", 30)          
 CoverageTools.MallocInfo(56320, "./src/BoundaryConditions/fill_halo_regions.jl.32885.mem", 20)       
 CoverageTools.MallocInfo(95040, "./src/BoundaryConditions/fill_halo_regions.jl.32885.mem", 25)       
 CoverageTools.MallocInfo(235751, "./benchmark/benchmark_static_ocean.jl.32885.mem", 55)              
 CoverageTools.MallocInfo(249600, "./src/BoundaryConditions/fill_halo_regions.jl.32885.mem", 70)      
 CoverageTools.MallocInfo(739712, "./benchmark/benchmark_utils.jl.32885.mem", 20)                     
 CoverageTools.MallocInfo(1686022, "./benchmark/benchmark_static_ocean.jl.32885.mem", 50) 

[codecov]

Codecov Report

Merging #685 into master will decrease coverage by 0.43%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #685      +/-   ##
==========================================
- Coverage   77.99%   77.56%   -0.44%     
==========================================
  Files         120      120              
  Lines        2413     2478      +65     
==========================================
+ Hits         1882     1922      +40     
- Misses        531      556      +25     

Impacted Files	Coverage Δ
src/BoundaryConditions/apply_flux_bcs.jl	53.84% <0.00%> (-29.49%)	⬇️
src/BoundaryConditions/boundary_function.jl	71.42% <0.00%> (-28.58%)	⬇️
src/Fields/field.jl	75.17% <0.00%> (-0.73%)	⬇️

---

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 74cfb85...6eb8db9. Read the comment docs.

[glwagner]

@views allocates because it creates a new view every time it runs. Not sure if that’s the issue. To avoid this we have to preallocate the view, somehow.