Merge branch 'main' into quasi_shallow_water_1d

Project.toml

@@ -1,7 +1,7 @@
 name = "Trixi"
 uuid = "a7f1ee26-1774-49b1-8366-f1abc58fbfcb"
 authors = ["Michael Schlottke-Lakemper <[email protected]>", "Gregor Gassner <[email protected]>", "Hendrik Ranocha <[email protected]>", "Andrew R. Winters <[email protected]>", "Jesse Chan <[email protected]>"]
-version = "0.5.41-pre"
+version = "0.5.42-pre"
 
 [deps]
 CodeTracking = "da1fd8a2-8d9e-5ec2-8556-3022fb5608a2"
@@ -56,7 +56,7 @@ DiffEqCallbacks = "2.25"
 EllipsisNotation = "1.0"
 FillArrays = "0.13.2, 1"
 ForwardDiff = "0.10.18"
-HDF5 = "0.14, 0.15, 0.16"
+HDF5 = "0.14, 0.15, 0.16, 0.17"
 IfElse = "0.1"
 LinearMaps = "2.7, 3.0"
 LoopVectorization = "0.12.118"

docs/src/parallelization.md

@@ -166,17 +166,36 @@ section, specifically at the descriptions of the performance index (PID).
 
 
 ### Using error-based step size control with MPI
-If you use error-based step size control (see also the section on [error-based adaptive step sizes](@ref adaptive_step_sizes))
-together with MPI you need to pass `internalnorm=ode_norm` and you should pass
-`unstable_check=ode_unstable_check` to OrdinaryDiffEq's [`solve`](https://docs.sciml.ai/DiffEqDocs/latest/basics/common_solver_opts/),
+If you use error-based step size control (see also the section on
+[error-based adaptive step sizes](@ref adaptive_step_sizes)) together with MPI you need to pass
+`internalnorm=ode_norm` and you should pass `unstable_check=ode_unstable_check` to
+OrdinaryDiffEq's [`solve`](https://docs.sciml.ai/DiffEqDocs/latest/basics/common_solver_opts/),
 which are both included in [`ode_default_options`](@ref).
 
 ### Using parallel input and output
-Trixi.jl allows parallel I/O using MPI by leveraging parallel HDF5.jl. To enable this, you first need
-to use a system-provided MPI library, see also [here](@ref parallel_system_MPI) and you need to tell
-[HDF5.jl](https://github.com/JuliaIO/HDF5.jl) to use this library.
-To do so, set the environment variable `JULIA_HDF5_PATH` to the local path
-that contains the `libhdf5.so` shared object file and build HDF5.jl by executing `using Pkg; Pkg.build("HDF5")`.
-For more information see also the [documentation of HDF5.jl](https://juliaio.github.io/HDF5.jl/stable/mpi/).
-
-If you do not perform these steps to use parallel HDF5 or if the HDF5 is not MPI-enabled, Trixi.jl will fall back on a less efficient I/O mechanism. In that case, all disk I/O is performed only on rank zero and data is distributed to/gathered from the other ranks using regular MPI communication.
+Trixi.jl allows parallel I/O using MPI by leveraging parallel HDF5.jl. On most systems, this is
+enabled by default. Additionally, you can also use a local installation of the HDF5 library
+(with MPI support). For this, you first need to use a system-provided MPI library, see also
+[here](@ref parallel_system_MPI) and you need to tell [HDF5.jl](https://github.com/JuliaIO/HDF5.jl)
+to use this library. To do so with HDF5.jl v0.17 and newer, set the preferences `libhdf5` and
+`libhdf5_hl` to the local paths of the libraries `libhdf5` and `libhdf5_hl`, which can be done by
+```julia
+julia> using Preferences, UUIDs
+julia> set_preferences!(
+           UUID("f67ccb44-e63f-5c2f-98bd-6dc0ccc4ba2f"), # UUID of HDF5.jl
+           "libhdf5" => "/path/to/your/libhdf5.so",
+           "libhdf5_hl" => "/path/to/your/libhdf5_hl.so", force = true)
+```
+For more information see also the
+[documentation of HDF5.jl](https://juliaio.github.io/HDF5.jl/stable/mpi/). In total, you should
+have a file called LocalPreferences.toml in the project directory that contains a section
+`[MPIPreferences]`, a section `[HDF5]` with entries `libhdf5` and `libhdf5_hl`, a section `[P4est]`
+with the entry `libp4est` as well as a section `[T8code]` with the entries `libt8`, `libp4est`
+and `libsc`.
+If you use HDF5.jl v0.16 or older, instead of setting the preferences for HDF5.jl, you need to set
+the environment variable `JULIA_HDF5_PATH` to the path, where the HDF5 binaries are located and
+then call `]build HDF5` from Julia.
+
+If HDF5 is not MPI-enabled, Trixi.jl will fall back on a less efficient I/O mechanism. In that
+case, all disk I/O is performed only on rank zero and data is distributed to/gathered from the
+other ranks using regular MPI communication.

test/test_tree_1d_shallowwater.jl

@@ -102,7 +102,8 @@ EXAMPLES_DIR = pkgdir(Trixi, "examples", "tree_1d_dgsem")
     @test_trixi_include(joinpath(EXAMPLES_DIR, "elixir_shallowwater_beach.jl"),
       l2   = [0.17979210479598923, 1.2377495706611434, 6.289818963361573e-8],
       linf = [0.845938394800688, 3.3740800777086575, 4.4541473087633676e-7],
-      tspan = (0.0, 0.05))
+      tspan = (0.0, 0.05),
+      atol = 3e-10) # see https://github.com/trixi-framework/Trixi.jl/issues/1617
   end
 
   @trixi_testset "elixir_shallowwater_parabolic_bowl.jl" begin

test/test_trixi.jl

@@ -5,7 +5,7 @@ import Trixi
 # inside an elixir.
 """
     @test_trixi_include(elixir; l2=nothing, linf=nothing,
-                                atol=10*eps(), rtol=0.001,
+                                atol=500*eps(), rtol=sqrt(eps()),
                                 parameters...)
 
 Test Trixi by calling `trixi_include(elixir; parameters...)`.