Merge branch 'develop'

configs/README.md

@@ -16,10 +16,10 @@ The beatnik spack package should enforce its build requirements appropriately; w
 
 ### Current list of tested systems targets and suggested installation method
 
-We have tested beatnik installation on the following systems via either spack install with the provided spack install flags or a spack environment. An example run script is also provided for each of these systems in the appropriate subdirectory.
-  * University of New Mexico - These systems simply use `spack install` as the UNM machines have a full global spack packages.yaml already set up. Because thse systems use spack for package maintenance, we recommend adding the provided `upstreams.yaml` file in the `unm` directory to the `.spack` directory in your home directory.
-    * Hopper V100/A100 GPU cluster system - `spack install beatnik +cuda cuda_arch=80 %gcc ^cuda@11` (or `cuda_arch=70` for the V100 nodes); ^cuda@11 is needed to avoid using CUDA 12 which the UNM cuda drivers aren't updated to support as of 10/1/23. Only tested with gcc as the compiler.
-    * General UNM (Wheeler/Hopper) CPU systems - `spack install beatnik` is sufficient
+We have tested the beatnik installation on the following systems via either spack install with the provided spack install flags or a spack environment. An example run script is also provided for each of these systems in the appropriate subdirectory.
+  * University of New Mexico - UNM systems have a full global spack packages.yaml already set up, and so it is often possible to use `spack install` with the appropriate compiler and other flags to install beatnik on these systems. We recommend adding the provided `upstreams.yaml` file in the `unm` directory to the `.spack` directory in your home directory.
+    * Hopper V100/A100 GPU cluster system - Build using the environment specification in `unm/hopper/spack.yaml`. It is sometimes possible to simply use `spack install beatnik +cuda cuda_arch=80 %gcc ^cuda@11` (or `cuda_arch=70` for the V100 nodes), but this may not always work depending on how spack chooses to concretize the spack installation.
+    * General UNM (Wheeler/Hopper) CPU systems - `spack install beatnik` is generally sufficient.
   * Lawrence Livermore National Laboratory - These systems need a spack environment (provided) to set up compilers and external packages to use spack effectively on these systems. Use `spack env create` with the provided spack.yaml to build beatnik in an environment on these systems. Simple test run scripts are also provided.
     * Lassen V100 GPU system - Build using the environment specification in llnl/lassen/spack.yaml. Other compilers besides gcc untested. Note that cuda-aware MPI support is broken in spectrum MPI with cuda versions later than 11.2; the spack configuration avoids this, but if you're building beatnik by hand on Lassen or another IBM CORAL system, be aware of this limitation.
     * Tioga MX250X GPU system - Build using the environment specification in llnl/tioga/spack.yaml; gcc should also work. Other compilers besides cce 16.0.1 untested. Note that you must run with the environment variable to enable gpu-aware cray-mpich, i.e. `export MPICH_GPU_SUPPORT_ENABLED=1`. The provided flux script (beatnik.flux) does this.

configs/unm/hopper/spack.yaml

@@ -0,0 +1,21 @@
+spack:
+  specs:
+  - matrix:
+    - - [email protected]
+      - ucx +ib_hw_tm +gdrcopy
+      - [email protected] +cuda cuda_arch=80 fabrics=ucx
+      - heffte +fftw +cuda cuda_arch=80
+      - kokkos +hwloc +cuda +cuda_lambda +wrapper cuda_arch=80
+      - cabana @master +cajita +heffte +mpi +cuda +arborx cuda_arch=80
+    - ['%[email protected]']
+  - hdf5 +mpi
+  - [email protected] +mpi
+  - blt
+  - cmake
+  concretizer:
+    unify: true
+  view: true
+
+  upstreams:
+    carc-spack:
+      install_tree: /opt/spack/opt/spack