unstructured BCs

[ranocha]

These are some first and minor performance improvements of the BCs for unstructured meshes. They only manifest for elixir_euler_unstructured_quad_wall_bc.jl since that's the only elixir with multiple BCs.

I used the following code with julia --check-bounds=no --threads=1

julia> trixi_include("examples/2d/elixir_euler_unstructured_quad_wall_bc.jl")

julia> du_ode = similar(ode.u0); du = Trixi.wrap_array(du_ode, semi); u_ode = copy(ode.u0); u = Trixi.wrap_array(u_ode, semi); Trixi.rhs!(du_ode, u_ode, semi, 0.0)

julia> @benchmark Trixi.calc_boundary_flux!(
           $(semi.cache),
           $(ode.tspan[1]),
           $(semi.boundary_conditions),
           $(mesh),
           $(equations),
           $(solver))

• Baseline 43497b2

  BenchmarkTools.Trial:
    memory estimate:  82.50 KiB
    allocs estimate:  880
    --------------
    minimum time:     17.600 μs (0.00% GC)
    median time:      19.207 μs (0.00% GC)
    mean time:        31.350 μs (18.41% GC)
    maximum time:     7.938 ms (99.44% GC)
    --------------
    samples:          10000
    evals/sample:     1
  

• Improve uniform_flow_state 17a9f92

  BenchmarkTools.Trial:
    memory estimate:  82.50 KiB
    allocs estimate:  880
    --------------
    minimum time:     16.781 μs (0.00% GC)
    median time:      17.519 μs (0.00% GC)
    mean time:        28.180 μs (17.80% GC)
    maximum time:     5.796 ms (99.35% GC)
    --------------
    samples:          10000
    evals/sample:     1
  

• Use Symbols instead of Strings for boundary names 87868b1

  BenchmarkTools.Trial:
    memory estimate:  82.50 KiB
    allocs estimate:  880
    --------------
    minimum time:     16.273 μs (0.00% GC)
    median time:      17.082 μs (0.00% GC)
    mean time:        28.010 μs (18.19% GC)
    maximum time:     5.882 ms (99.42% GC)
    --------------
    samples:          10000
    evals/sample:     1
  

• Introduce a function barrier for boundary fluxes cfec929

  BenchmarkTools.Trial:
    memory estimate:  22.50 KiB
    allocs estimate:  240
    --------------
    minimum time:     7.593 μs (0.00% GC)
    median time:      7.825 μs (0.00% GC)
    mean time:        9.900 μs (15.44% GC)
    maximum time:     1.610 ms (99.31% GC)
    --------------
    samples:          10000
    evals/sample:     4
  

Further performance improvements need some more experiments and probably a decision between at least two approaches suggested in #542 (sorting and earlier function barriers or function wrappers via ccall).

[andrewwinters5000]

Nice work @ranocha ! I just left a few comments for clarification but overall I think this is ready for merge (as long as the periodic test will now pass)

[codecov]

Codecov Report

Merging #573 (845e08e) into main (0d1bf98) will increase coverage by 15.39%.
The diff coverage is 93.75%.

@@             Coverage Diff             @@
##             main     #573       +/-   ##
===========================================
+ Coverage   78.33%   93.72%   +15.39%     
===========================================
  Files         138      138               
  Lines       13831    13837        +6     
===========================================
+ Hits        10834    12969     +2135     
+ Misses       2997      868     -2129     

Flag	Coverage Δ
unittests	93.72% <93.75%> (+15.39%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
...mples/2d/elixir_euler_unstructured_quad_wall_bc.jl	90.00% <75.00%> (-10.00%)	⬇️
src/mesh/unstructured_quad_mesh.jl	98.23% <100.00%> (ø)
src/solvers/dg_unstructured_quad/containers_2d.jl	96.42% <100.00%> (ø)
src/solvers/dg_unstructured_quad/dg_2d.jl	97.77% <100.00%> (-0.72%)	⬇️
src/callbacks_step/analysis.jl	96.84% <0.00%> (+0.90%)	⬆️
src/mesh/mesh_io.jl	95.65% <0.00%> (+1.86%)	⬆️
...emidiscretization/semidiscretization_hyperbolic.jl	90.69% <0.00%> (+2.32%)	⬆️
...discretization/semidiscretization_euler_gravity.jl	91.37% <0.00%> (+2.53%)	⬆️
src/callbacks_step/analysis_dg3d.jl	97.19% <0.00%> (+15.88%)	⬆️
... and 20 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 0d1bf98...845e08e. Read the comment docs.

[sloede]

LGTM! Just one minor question...