Fix benchmark

PTsolvers · Nov 17, 2023 · 52469fd · 52469fd
1 parent 63b0f52
commit 52469fd
Showing 1 changed file with 16 additions and 12 deletions.
diff --git a/scripts_future_API/benchmark_diffusion_3D.jl b/scripts_future_API/benchmark_diffusion_3D.jl
@@ -51,7 +51,7 @@ function compute(arch, grid, hide_boundaries, bc_q, bc_c, outer_width, qC, C, dc
     return wtime
 end
 
-function diffusion_3D(ka_backend=CPU(), dTyp::DataType=Float64, dims=(0,0,0); do_visu=true)
+function diffusion_3D(ka_backend=CPU(), dTyp::DataType=Float64, dims=(0, 0, 0); do_visu=true)
     # setup arch
     arch = Architecture(ka_backend, dims)
     topo = details(arch)
@@ -85,16 +85,19 @@ function diffusion_3D(ka_backend=CPU(), dTyp::DataType=Float64, dims=(0,0,0); do
     set!(C, grid, (x, y, z) -> exp(-x^2 - y^2 - z^2))
     # boundary conditions
     zero_flux_bc = DirichletBC{FullCell}(0.0)
-    bc_q = NamedTuple(comp => BoundaryConditionsBatch((qC[comp],), (zero_flux_bc,)) for comp in eachindex(qC))
     # zero flux at physical boundaries and nothing at MPI boundaries
-    bc_q = override_boundary_conditions(arch, ((bc_q.x, bc_q.x), (bc_q.y, bc_q.y), (bc_q.z, bc_q.z)))
+    bc_q = ntuple(Val(3)) do D
+        ntuple(Val(2)) do S
+            has_neighbor(topo, D, S) ? nothing : BoundaryConditionsBatch((qC[D],), (zero_flux_bc,))
+        end
+    end
     # nothing at physical boundaries and communication at MPI boundaries
-    bc_c = BoundaryConditionsBatch((C,), nothing)
-    bc_c = override_boundary_conditions(arch, ((bc_c, bc_c), (bc_c, bc_c), (bc_c, bc_c)); exchange=true)
-    ntuple(Val(ndims(grid))) do D
-        apply_boundary_conditions!(Val(1), Val(D), arch, grid, bc_c[D][1])
-        apply_boundary_conditions!(Val(2), Val(D), arch, grid, bc_c[D][2])
+    bc_c = ntuple(Val(3)) do D
+        ntuple(Val(2)) do S
+            has_neighbor(topo, D, S) ? BoundaryConditionsBatch((C,), (ExchangeInfo(Val(S), Val(D), C),)) : nothing
+        end
     end
+    KernelLaunch.apply_all_boundary_conditions!(arch, grid, bc_c)
     # # time loop
     # for it in 1:nt
     #     (me == 0) && println("it = $it")
@@ -109,24 +112,25 @@ function diffusion_3D(ka_backend=CPU(), dTyp::DataType=Float64, dims=(0,0,0); do
     # time
     MPI.Barrier(comm)
     for ex in 1:5
-        (me==0) && (sleep(2); println("Experiment = $ex"))
+        (me == 0) && (sleep(2); println("Experiment = $ex"))
         MPI.Barrier(comm)
         wtime = compute(arch, grid, hide_boundaries, bc_q, bc_c, outer_width, qC, C, dc, dt, Δ, (iters - warmup))
         # report
         A_eff = 8 / 1e9 * prod(size) * sizeof(dTyp)
         wtime_it = wtime ./ (iters - warmup)
         T_eff = A_eff ./ wtime_it
-        @printf("  Executed %d steps in = %1.3e sec (@ T_eff = %1.2f GB/s - device %s) \n", (iters - warmup), wtime, round(T_eff, sigdigits=3), AMDGPU.device_id(AMDGPU.device()))
+        @printf("  Executed %d steps in = %1.3e sec (@ T_eff = %1.2f GB/s - device %s) \n", (iters - warmup), wtime,
+                round(T_eff, sigdigits=3), AMDGPU.device_id(AMDGPU.device()))
     end
 
     if do_visu
-        ENV["GKSwstype"]="nul"
+        ENV["GKSwstype"] = "nul"
         C_g = (me == 0) ? KernelAbstractions.allocate(CPU(), eltype(C), size_g) : nothing
         C_v = Array(C)
         gather!(arch, C_g, C_v)
         if me == 0
             p1 = heatmap(xcenters(global_grid), ycenters(global_grid), C_g[:, :, size_g[3]÷2];
-            aspect_ratio=1, xlims=extrema(xcenters(global_grid)), ylims=extrema(ycenters(global_grid)))
+                         aspect_ratio=1, xlims=extrema(xcenters(global_grid)), ylims=extrema(ycenters(global_grid)))
             png(p1, "C.png")
         end
     end