Merge pull request #3134 from Crivella/feature-qe_scaling_test

[testlib] Added QuantumESPRESSO scalability test for `pw.x`

docs/hpctestlib.rst

@@ -102,6 +102,9 @@ Scientific Applications
    :members:
    :show-inheritance:
 
+.. automodule:: hpctestlib.sciapps.qespresso.benchmarks
+   :members:
+   :show-inheritance:
 
 System
 =======================

hpctestlib/sciapps/qespresso/benchmarks.py

@@ -0,0 +1,202 @@
+"""ReFrame benchmark for QuantumESPRESSO"""
+import os
+from typing import TypeVar
+
+import reframe as rfm
+import reframe.utility.sanity as sn
+from reframe.core.builtins import (performance_function, run_after, run_before,
+                                   sanity_function)
+from reframe.core.exceptions import SanityError
+from reframe.core.parameters import TestParam as parameter
+from reframe.core.variables import TestVar as variable
+
+R = TypeVar('R')
+
+INPUT_TEMPLATE = """&CONTROL
+  calculation  = "scf",
+  prefix       = "Si",
+  pseudo_dir   = ".",
+  outdir       = "./out",
+  restart_mode = "from_scratch"
+  verbosity    = 'high'
+/
+&SYSTEM
+  ibrav     = 2,
+  celldm(1) = 10.2,
+  nat       = 2,
+  ntyp      = 1,
+  nbnd      = {nbnd}
+  ecutwfc   = {ecut}
+/
+&ELECTRONS
+  conv_thr    = 1.D-8,
+  mixing_beta = 0.7D0,
+/
+ATOMIC_SPECIES
+ Si  28.086  {pseudo}
+ATOMIC_POSITIONS
+ Si 0.00 0.00 0.00
+ Si 0.25 0.25 0.25
+K_POINTS {{automatic}}
+ 15 15 15   0 0 0
+
+"""
+
+
+@rfm.simple_test
+class QEspressoPWCheck(rfm.RunOnlyRegressionTest):
+    """QuantumESPRESSO benchmark test.
+
+    `QuantumESPRESSO <https://www.quantum-espresso.org/>`__ is an integrated
+    suite of Open-Source computer codes for electronic-structure calculations
+    and materials modeling at the nanoscale.
+
+    The benchmarks consist of one input file templated inside the code and
+    a pseudo-potential file that is downloaded from the official repository.
+
+    This tests aims at measuring the scalability of the pw.x executable, in
+    particular the FFT and diagonalization algorithms, by running a simple
+    silicon calculation with high `ecut` (increases size of FFTs) and `nbnd`
+    (increases size of matrices to diagonalize) values."""
+
+    #: Parametert to tests the performance of the FFTW algorithm,
+    #: higher `ecut` implicates more FFTs
+    #:
+    #: :type: :class:`int`
+    #: :values: ``[50, 150]``
+    ecut = parameter([50, 150], loggable=True)
+
+    #: Parameter to Tests the performance of the diagonalization algorithm,
+    #: higher `nbnd` implicates bigger matrices
+    #:
+    #: :type: :class:`int`
+    #: :values: ``[10, 200]``
+    nbnd = parameter([10, 200], loggable=True)
+
+    executable = 'pw.x'
+    tags = {'sciapp', 'chemistry'}
+
+    descr = 'QuantumESPRESSO pw.x benchmark'
+
+    #: The name of the input file used.
+    #:
+    #: :type: :class:`str`
+    #: :default: ``'Si.scf.in'``
+    input_name: str = variable(str, value='Si.scf.in')
+
+    #: The pseudo-potential file to be used check
+    #: https://www.quantum-espresso.org/pseudopotentials/ for more info
+    #:
+    #: :type: :class:`str`
+    #: :default: ``'Si.pbe-n-kjpaw_psl.1.0.0.UPF'``
+    pp_name: str = variable(str, value='Si.pbe-n-kjpaw_psl.1.0.0.UPF')
+
+    @run_after('init')
+    def prepare_test(self):
+        """Hook to the set the downloading of the pseudo-potentials"""
+        self.prerun_cmds = [(
+            'wget -q http://pseudopotentials.quantum-espresso.org/upf_files/'
+            f'{self.pp_name}'
+        )]
+        self.executable_opts += [f'-in {self.input_name}']
+
+    @run_after('setup')
+    def write_input(self):
+        """Write the input file for the calculation"""
+        inp_file = os.path.join(self.stagedir, self.input_name)
+        with open(inp_file, 'w', encoding='utf-8') as file:
+            file.write(
+                INPUT_TEMPLATE.format(
+                    ecut=self.ecut,
+                    nbnd=self.nbnd,
+                    pseudo=self.pp_name,
+                ))
+
+    @staticmethod
+    @sn.deferrable
+    def extractsingle_or_val(*args, on_except_value: str = '0s') -> str:
+        """Wrap extractsingle_or_val to return a default value if the regex is
+        not found.
+
+        Returns:
+            str: The value of the regular expression
+        """
+        try:
+            res = sn.extractsingle(*args).evaluate()
+        except SanityError:
+            res = on_except_value
+
+        return res
+
+    @staticmethod
+    @sn.deferrable
+    def convert_timings(timing: str) -> float:
+        """Convert timings to seconds"""
+
+        if timing is None:
+            return 0
+
+        days, timing = (['0', '0'] + timing.split('d'))[-2:]
+        hours, timing = (['0', '0'] + timing.split('h'))[-2:]
+        minutes, timing = (['0', '0'] + timing.split('m'))[-2:]
+        seconds = timing.split('s')[0]
+
+        return (
+            float(days) * 86400 +
+            float(hours) * 3600 +
+            float(minutes) * 60 +
+            float(seconds)
+        )
+
+    @performance_function('s')
+    def extract_report_time(self, name: str = None, kind: str = None) -> float:
+        """Extract timings from pw.x stdout
+
+        Args:
+            name (str, optional): Name of the timing to extract.
+                                  Defaults to None.
+            kind (str, optional): Kind ('cpu' or 'wall) of timing to extract.
+                                  Defaults to None.
+
+        Raises:
+            ValueError: If the kind is not 'cpu' or 'wall'
+
+        Returns:
+            float: The timing in seconds
+        """
+        if kind is None:
+            return 0
+        kind = kind.lower()
+        if kind == 'cpu':
+            tag = 1
+        elif kind == 'wall':
+            tag = 2
+        else:
+            raise ValueError(f'unknown kind: {kind}')
+
+        # Possible formats
+        #       PWSCF        :   4d 6h19m CPU  10d14h38m WALL
+        # --> (Should also catch spaces)
+        return self.convert_timings(
+            self.extractsingle_or_val(
+                fr'{name}\s+:\s+(.+)\s+CPU\s+(.+)\s+WALL',
+                self.stdout, tag, str
+            ))
+
+    @run_before('performance')
+    def set_perf_variables(self):
+        """Build a dictionary of performance variables"""
+
+        timings = [
+            'PWSCF', 'electrons', 'c_bands', 'cegterg', 'calbec',
+            'fft', 'ffts', 'fftw'
+        ]
+        for name in timings:
+            for kind in ['cpu', 'wall']:
+                res = self.extract_report_time(name, kind)
+                self.perf_variables[f'{name}_{kind}'] = res
+
+    @sanity_function
+    def assert_job_finished(self):
+        """Check if the job finished successfully"""
+        return sn.assert_found(r'JOB DONE', self.stdout)