Merge branch 'main' into cp2k-2025.2

Author: RMeli

.github/pull_request_template.md

@@ -16,7 +16,7 @@ cscs-ci run alps-starlex-uenv;MY_UENV=prgenv-gnu/25.11:rc1
 cscs-ci run alps-beverin-uenv;MY_UENV=prgenv-gnu/25.07-6.3.3:v9
 ```
 
-- You can also test from your terminal, for example, on beverin:
+- You can also test from your terminal:
     - install: https://confluence.cscs.ch/spaces/reframe/pages/886276110/Installing+ReFrame
     - run test: for example, cp2k on beverin:
 
@@ -31,3 +31,4 @@ UENV=prgenv-gnu/25.07-6.3.3:v8 \
 ```
 
 Thank you for taking the time to contribute to `cscs-reframe-tests` !
+- CSCS staff: More info in https://confluence.cscs.ch/spaces/reframe --> Contributing+Pull+Requests

checks/apps/lammps/lammps.py

@@ -78,7 +78,7 @@ class lammps_build_test(rfm.CompileOnlyRegressionTest):
     sourcesdir = None
     lammps_sources = fixture(lammps_download, scope='environment')
     build_system = 'CMake'
-    tags = {'uenv', 'production'}
+    tags = {'uenv'}
     build_locally = False
 
     @run_before('compile')

checks/apps/q-e-sirius/q-e-sirius_check_uenv.py

@@ -0,0 +1,174 @@
+# Copyright Swiss National Supercomputing Centre (CSCS/ETH Zurich)
+# ReFrame Project Developers. See the top-level LICENSE file for details.
+#
+# SPDX-License-Identifier: BSD-3-Clause
+
+import reframe as rfm
+import reframe.utility.sanity as sn
+import uenv
+import os
+import reframe.utility.osext as osext
+
+
+qe_references = {
+    'Au surf': {
+        'gh200': {'time_run': (50, None, 0.05, 's')},
+        'mi300': {'time_run': (112, None, 0.05, 's')},
+        'mi200': {'time_run': (78, None, 0.05, 's')},
+    },
+}
+
+
+slurm_config = {
+    'Au surf': {
+        'gh200': {
+            'nodes': 1,
+            'ntasks-per-node': 4,
+            'cpus-per-task': 72,
+            'walltime': '0d0h20m0s',
+            'extra_job_options': [],
+            'gpu': True,
+        },
+        'mi200': {
+            'nodes': 1,
+            'ntasks-per-node': 8,
+            'cpus-per-task': 8,
+            'walltime': '0d0h20m0s',
+            'extra_job_options': [],
+            'gpu': True,
+        },
+        'mi300': {
+            'nodes': 1,
+            'ntasks-per-node': 12,
+            'gpus-per-task': 1,
+            'cpus-per-task': 8,
+            'extra_job_options': ['--constraint=amdgpu_tpx'],
+            'walltime': '0d0h20m0s',
+            'gpu': True,
+        },
+        'zen2': {
+            'nodes': 1,
+            'ntasks-per-node': 16,
+            'cpus-per-task': 8,
+            'walltime': '0d0h20m0s',
+            'extra_job_options': [],
+            'gpu': False,
+        },
+    },
+}
+
+
+def uenv_uarch():
+    uenv_name = os.environ.get("UENV", None)
+    if not uenv_name:
+        return uenv_name
+    uenv_inspect_cmd = f"uenv image inspect --format={{uarch}} {uenv_name}"
+    return osext.run_command(uenv_inspect_cmd, shell=True).stdout.strip()
+
+
+class QeSiriusCheckUENV(rfm.RunOnlyRegressionTest):
+    pwx_executable = 'pw.x'
+    maintainers = ['simonpintarelli', 'SSA']
+    valid_systems = ['+uenv +amdgpu', '+uenv +nvgpu']
+
+    @run_after('setup')
+    def skip_unsupported_uenv(self):
+        _uarch = uenv.uarch(self.current_partition)
+        self.skip_if(_uarch != uenv_uarch(),
+                     f'this uenv does not support {_uarch}')
+
+    @run_before('run')
+    def prepare_run(self):
+        self.uarch = uenv.uarch(self.current_partition)
+        config = slurm_config[self.test_name][self.uarch]
+        # sbatch options
+        self.job.options = [
+            f'--nodes={config["nodes"]}',
+        ]
+        self.job.options += config['extra_job_options']
+
+        self.num_tasks_per_node = config['ntasks-per-node']
+        self.num_tasks = config['nodes'] * self.num_tasks_per_node
+        self.num_cpus_per_task = config['cpus-per-task']
+        self.ntasks_per_core = 1
+        self.time_limit = config['walltime']
+
+        # environment variables
+        self.env_vars['OMP_NUM_THREADS'] = config['cpus-per-task']
+        self.env_vars['SIRIUS_PRINT_TIMING'] = '5'
+        self.env_vars['SIRIUS_PRINT_MPI_LAYOUT'] = '1'
+        self.env_vars['SIRIUS_VERBOSITY'] = '2'
+        self.env_vars['SLURM_HINT'] = 'nomultithread'
+        if self.uarch == 'gh200':
+            self.env_vars['MPICH_GPU_SUPPORT_ENABLED'] = '1'
+            self.env_vars['OMP_NUM_THREADS'] = str(20)
+        if self.uarch == 'mi200':
+            self.env_vars['MPICH_GPU_SUPPORT_ENABLED'] = '1'
+            self.env_vars['OMP_NUM_THREADS'] = str(8)
+        if self.uarch == 'mi300':
+            self.env_vars['MPICH_GPU_SUPPORT_ENABLED'] = '1'
+            self.env_vars['OMP_NUM_THREADS'] = str(8)
+        if self.uarch in ('mi300', 'mi200'):
+            self.env_vars['PIKA_MPI_ENABLE_POOL'] = '1'
+            self.env_vars['PIKA_MPI_COMPLETION_MODE'] = '28'
+            self.env_vars['DLAF_BAND_TO_TRIDIAG_1D_BLOCK_SIZE_BASE'] = '2048'
+            self.env_vars['DLAF_NUM_NP_GPU_STREAMS'] = '4'
+            self.env_vars['DLAF_NUM_HP_GPU_STREAMS'] = '4'
+
+        # set reference
+        if self.uarch is not None and self.uarch in qe_references[self.test_name]:
+            self.reference = {
+                self.current_partition.fullname:
+                qe_references[self.test_name][self.uarch]
+            }
+
+    @sanity_function
+    def assert_energy_diff(self):
+        energy = sn.extractsingle(
+            r'^!\s+total energy\s+=\s+(?P<energy>\S+)',
+            self.stdout,
+            'energy',
+            float,
+            item=-1,
+        )
+        energy_diff = sn.abs(energy - self.energy_reference)
+        successful_termination = sn.assert_found(r'JOB DONE', self.stdout)
+        correct_energy = sn.assert_lt(energy_diff, 5e-4)
+        return sn.all(
+            [
+                successful_termination,
+                correct_energy,
+            ]
+        )
+
+    # INFO: The name of this function needs to match with the reference dict!
+    @performance_function('s')
+    def time_run(self):
+        return sn.extractsingle(
+            r'electrons.+\s(?P<wtime>\S+)s WALL', self.stdout, 'wtime', float
+        )
+
+
+class QeSiriusCheckAuSurfUENV(QeSiriusCheckUENV):
+    test_name = 'Au surf'
+    executable_opts = ['-i', 'ausurf.in']
+    energy_reference = -11427.09017218
+
+
+@rfm.simple_test
+class QeCheckAuSurfUENVExec(QeSiriusCheckAuSurfUENV):
+    valid_prog_environs = ['+uenv +q-e-sirius']
+    tags = {'uenv', 'production', 'bencher'}
+
+    @run_after('setup')
+    def setup_executable(self):
+        self.executable = f'pw.x'
+        uarch = uenv.uarch(self.current_partition)
+        if uarch == 'gh200':
+            self.executable = f'./mps-wrapper.sh pw.x'
+        if uarch == 'mi200':
+            self.executable = \
+                f'./amdgpu-wrapper.sh pw.x -sirius_cfg sirius-amd.json'
+        if uarch == 'mi300':
+            self.executable = \
+                f'./tpx-wrapper.sh pw.x -sirius_cfg sirius-amd.json'