Merge pull request #6 from PTsolvers/lr/revision

GMD 2021-411 revision

Project.toml

@@ -1,7 +1,7 @@
 name = "PseudoTransientDiffusion"
 uuid = "b5b0327e-567e-4ee6-9713-fc28d852c7ee"
 authors = ["Ludovic Raess <[email protected]>, Ivan Utkin and contributors"]
-version = "0.1.0"
+version = "1.0.0"
 
 [deps]
 BSON = "fbb218c0-5317-5bc6-957e-2ee96dd4b1f0"

README.md

@@ -5,13 +5,19 @@
 
 Parallel (multi-) XPU iterative 1D, 2D and 3D diffusion solvers (resolving linear, step-function and nonlinear diffusion coefficient). This software is part of the [**PTsolvers project**](https://ptsolvers.github.io/).
 
-The aim of this project is to provide iterative solvers **assessing the scalability, performance, and robustness of the accelerated pseudo-transient method** with application to diffusion processes. The solution strategy characterises as semi-iterative, implementing the second-order convergence acceleration as introduced by, e.g., \[[Frankel, 1950](https://doi.org/10.2307/2002770)\].
+The aim of this project is to provide iterative solvers **assessing the scalability, performance, and robustness of the accelerated pseudo-transient method** with application to diffusion processes. The solution strategy characterises as semi-iterative, implementing the second-order convergence acceleration as introduced by, e.g., [Frankel (1950)](https://doi.org/10.2307/2002770).
 
-This repository, together with [**PseudoTransientStokes.jl**](https://github.com/PTsolvers/PseudoTransientStokes.jl/), relates to the original research article draft submitted to the _**Journal XXX**_:
+This repository, together with [**PseudoTransientStokes.jl**](https://github.com/PTsolvers/PseudoTransientStokes.jl/), relates to the original research article draft submitted to the _**Geoscientific Model Development**_ journal:
 ```tex
-@article{raess2022,
-    title = {{ }},
-    journal = {Journal XXX}
+@Article{raess2022,
+    AUTHOR = {R\"ass, L. and Utkin, I. and Duretz, T. and Omlin, S. and Podladchikov, Y. Y.},
+    TITLE = {Assessing the robustness and scalability of the accelerated pseudo-transient method towards exascale computing},
+    JOURNAL = {Geoscientific Model Development Discussions},
+    VOLUME = {2022},
+    YEAR = {2022},
+    PAGES = {1--46},
+    URL = {https://gmd.copernicus.org/preprints/gmd-2021-411/},
+    DOI = {10.5194/gmd-2021-411}
 }
 ```
 

scripts/diff_3D/diff_3D_lin_multixpu.jl

@@ -53,7 +53,6 @@ end
     nout       = 10               # tol check
     CFL        = 1/sqrt(3)        # CFL number
     me, dims   = init_global_grid(nx, ny, nz) # MPI initialisation
-    @static if USE_GPU select_device() end    # select one GPU per MPI local rank (if >1 GPU per node)
     b_width    = (8, 4, 4)       # boundary width for comm/comp overlap
     # Derived numerics    
     dx, dy, dz = lx/nx_g(), ly/ny_g(), lz/nz_g() # cell sizes
@@ -109,7 +108,7 @@ end
     if (me==0) @printf("Total time = %1.2f, time steps = %d, nx = %d, iterations tot = %d \n", round(ttot, sigdigits=2), it, nx_g(), ittot) end
     # Visualise
     if do_viz || do_save_viz
-        H_inn .= inn(H); gather!(H_inn, H_v)
+        H_inn .= Array(inn(H)); gather!(H_inn, H_v)
         if me==0 && do_viz
             heatmap(Xi_g, Yi_g, H_v[:,:,z_sl]', dpi=150, aspect_ratio=1, framestyle=:box, xlims=(Xi_g[1],Xi_g[end]), ylims=(Yi_g[1],Yi_g[end]), xlabel="lx", ylabel="ly", c=:viridis, clims=(0,1), title="linear diffusion (nt=$it, iters=$ittot)")
             savefig("../../figures/diff_3D_lin_$(nx_g()).png")

scripts/diff_3D/diff_3D_linstep_multixpu.jl

@@ -77,7 +77,6 @@ end
     nout       = 10               # tol check
     CFL        = 1/sqrt(3)        # CFL number
     me, dims   = init_global_grid(nx, ny, nz) # MPI initialisation
-    @static if USE_GPU select_device() end    # select one GPU per MPI local rank (if >1 GPU per node)
     b_width    = (8, 4, 4)       # boundary width for comm/comp overlap
     # Derived numerics    
     dx, dy, dz = lx/nx_g(), ly/ny_g(), lz/nz_g() # cell sizes
@@ -141,7 +140,7 @@ end
     if (me==0) @printf("Total time = %1.2f, time steps = %d, nx = %d, iterations tot = %d \n", round(ttot, sigdigits=2), it, nx_g(), ittot) end
     # Visualise
     if do_viz || do_save_viz
-        H_inn .= inn(H); gather!(H_inn, H_v)
+        H_inn .= Array(inn(H)); gather!(H_inn, H_v)
         if me==0 && do_viz
             heatmap(Xi_g, Yi_g, H_v[:,:,z_sl]', dpi=150, aspect_ratio=1, framestyle=:box, xlims=(Xi_g[1],Xi_g[end]), ylims=(Yi_g[1],Yi_g[end]), xlabel="lx", ylabel="ly", c=:viridis, clims=(0,1), title="linear step diffusion (nt=$it, iters=$ittot)")
             savefig("../../figures/diff_3D_linstep_$(nx_g()).png")

scripts/diff_3D/diff_3D_nonlin_multixpu.jl

@@ -66,7 +66,6 @@ end
     CFL        = 1 / sqrt(3)      # CFL number
     Resc       = 1 / 1.2          # iteration parameter scaling
     me, dims   = init_global_grid(nx, ny, nz) # MPI initialisation
-    @static if USE_GPU select_device() end    # select one GPU per MPI local rank (if >1 GPU per node)
     b_width    = (8, 4, 4)       # boundary width for comm/comp overlap
     # Derived numerics    
     dx, dy, dz = lx/nx_g(), ly/ny_g(), lz/nz_g() # cell sizes
@@ -121,7 +120,7 @@ end
     if (me==0) @printf("Total time = %1.2f, time steps = %d, nx = %d, iterations tot = %d \n", round(ttot, sigdigits=2), it, nx_g(), ittot) end
     # Visualise
     if do_viz || do_save_viz
-        H_inn .= inn(H); gather!(H_inn, H_v)
+        H_inn .= Array(inn(H)); gather!(H_inn, H_v)
         if me==0 && do_viz
             heatmap(Xi_g, Yi_g, H_v[:,:,z_sl]', dpi=150, aspect_ratio=1, framestyle=:box, xlims=(Xi_g[1],Xi_g[end]), ylims=(Yi_g[1],Yi_g[end]), xlabel="lx", ylabel="ly", c=:viridis, clims=(0,1), title="nonlinear diffusion (nt=$it, iters=$ittot)")
             savefig("../../figures/diff_3D_nonlin_$(nx_g()).png")

scripts/diff_3D/diff_3D_nonlin_multixpu_perf.jl

@@ -75,7 +75,6 @@ end
     CFL        = 1 / sqrt(3)      # CFL number
     Resc       = 1 / 1.2          # iteration parameter scaling
     me, dims, nprocs = init_global_grid(nx, ny, nz) # MPI initialisation
-    @static if USE_GPU select_device() end    # select one GPU per MPI local rank (if >1 GPU per node)
     b_width    = (16, 4, 4)       # boundary width for comm/comp overlap
     # Derived numerics    
     dx, dy, dz = lx/nx_g(), ly/ny_g(), lz/nz_g() # cell sizes
@@ -139,7 +138,7 @@ end
     if (me==0) @printf("Total time = %1.2f, time steps = %d, nx = %d, iterations tot = %d \n", round(ttot, sigdigits=2), it, nx_g(), ittot) end
     # Visualise
     if do_viz || do_save_viz
-        H_inn .= inn(H); gather!(H_inn, H_v)
+        H_inn .= Array(inn(H)); gather!(H_inn, H_v)
         if me==0 && do_viz
             heatmap(Xi_g, Yi_g, H_v[:,:,z_sl]', dpi=150, aspect_ratio=1, framestyle=:box, xlims=(Xi_g[1],Xi_g[end]), ylims=(Yi_g[1],Yi_g[end]), xlabel="lx", ylabel="ly", c=:viridis, clims=(0,1), title="nonlinear diffusion (nt=$it, iters=$ittot)")
             savefig("../../figures/diff_3D_nonlin_multixpu_perf_$(nx_g()).png")