document LUMI installation

docs/getting-started/index.rst

@@ -3,9 +3,10 @@ Getting started
 
 .. toctree::
    :hidden:
+   :maxdepth: 1
 
    quick-start
-   installation
+   installation/index
    examples
 
 Welcome to tandem!
@@ -16,6 +17,6 @@ to compiled from source.
 If you have Docker installed, you might want to try the
 :doc:`quick start <quick-start>` procedure.
 Here, all required dependencies are already contained in a Docker image.
-Otherwise, follow the :doc:`regular installation <installation>` procedure.
+Otherwise, follow the :doc:`regular installation <installation/index>` procedure.
 
 Once tandem is installed, try to run the :doc:`example problems <examples>`.

docs/getting-started/installation/gpus.rst

@@ -0,0 +1,39 @@
+Tandem with GPUs
+================
+
+installation with spack
+-----------------------
+
+First add to you `~/.spack/packages.yaml`
+
+.. code-block:: yaml
+
+    packages:
+      all:
+        variants: cuda_arch=xxx
+
+With xxx you cuda architecture (e.g. 86). For AMD GPUs, use the keyword `amdgpu_target` (e.g. =gfx90a).
+This will ensure that the architecture is propagated to all dependencies.
+
+Then, installing the GPU version tandem is (in theory) as simple as adding the GPU variant (+cuda for Nvidia GPUs or +rocm for AMD GPUs), for example:
+
+.. code-block:: bash
+
+    spack install -j 20 tandem@main polynomial_degree=4 domain_dimension=3 +cuda
+
+
+Using a GPU version
+-------------------
+
+To ensure that tandem utilizes GPUs for computation, additionnal PETSc arguments are required when running the application.
+For Cuda, use:
+
+.. code-block:: bash
+
+    -vec_type cuda -mat_type aijcusparse
+
+For ROCM, use:
+
+.. code-block:: bash
+
+    -vec_type hip -mat_type aijhipsparse

docs/getting-started/installation/index.rst

@@ -0,0 +1,15 @@
+Installation
+============
+
+.. toctree::
+   :hidden:
+
+   spack_installation
+   installation_supermucNG
+   installation_lumi
+   installation_leonardo
+   gpus
+   manual_installation
+   installing_lua_modules
+
+Tandem and its dependencies can be installed automatically with `Spack <https://github.com/spack/spack/wiki>`_, or manually.