Merge pull request #1872 from teojgo/tests/remove_custom_dom

[test] Align Daint and Dom versions of tests

Author: Vasileios Karakasis

cscs-checks/apps/jupyter/check_ipcmagic.py

@@ -15,19 +15,10 @@ def __init__(self):
         self.descr = 'Distributed training with TensorFlow using ipyparallel'
         self.valid_systems = ['daint:gpu', 'dom:gpu']
         self.valid_prog_environs = ['PrgEnv-gnu']
-        cray_cdt_version = osext.cray_cdt_version()
-        # FIXME: The following will not be needed after the Daint upgrade
-        if self.current_system.name == 'dom':
-            self.modules = [
-                'ipcmagic',
-                f'Horovod/0.21.0-CrayGNU-{cray_cdt_version}-tf-2.4.0'
-            ]
-        else:
-            self.modules = [
-                'ipcmagic',
-                'Horovod/0.19.1-CrayGNU-20.08-tf-2.2.0'
-            ]
-
+        self.modules = [
+            f'ipcmagic',
+            f'Horovod/0.21.0-CrayGNU-{osext.cray_cdt_version()}-tf-2.4.0'
+        ]
         self.num_tasks = 2
         self.num_tasks_per_node = 1
         self.executable = 'ipython'

cscs-checks/apps/quantumespresso/quantumespresso_check.py

@@ -8,6 +8,9 @@
 
 
 class QuantumESPRESSOCheck(rfm.RunOnlyRegressionTest):
+    scale = parameter(['small', 'large'])
+    variant = parameter(['maint', 'prod'])
+
     def __init__(self):
         if self.current_system.name == 'pilatus':
             self.valid_prog_environs = ['cpeGNU']
@@ -37,15 +40,14 @@ def __init__(self):
         }
 
 
-@rfm.parameterized_test(*([s, v]
-                          for s in ['small', 'large']
-                          for v in ['maint', 'prod']))
+@rfm.simple_test
 class QuantumESPRESSOCpuCheck(QuantumESPRESSOCheck):
-    def __init__(self, scale, variant):
+    def __init__(self):
         super().__init__()
-        self.descr = f'QuantumESPRESSO CPU check (version: {scale}, {variant})'
+        self.descr = (f'QuantumESPRESSO CPU check (version: {self.scale}, '
+                      f'{self.variant})')
         self.valid_systems = ['daint:mc', 'eiger:mc', 'pilatus:mc']
-        if scale == 'small':
+        if self.scale == 'small':
             self.valid_systems += ['dom:mc']
             energy_reference = -11427.09017218
             if self.current_system.name in ['daint', 'dom']:
@@ -86,8 +88,6 @@ def __init__(self, scale, variant):
         energy_diff = sn.abs(energy-energy_reference)
         self.sanity_patterns = sn.all([
             self.sanity_patterns,
-            # FIXME temporarily increase energy difference
-            # (different QE default on Dom and Daint)
             sn.assert_lt(energy_diff, 1e-6)
         ])
 
@@ -120,8 +120,10 @@ def __init__(self, scale, variant):
             }
         }
 
-        self.reference = references[variant][scale]
-        self.tags |= {'maintenance' if variant == 'maint' else 'production'}
+        self.reference = references[self.variant][self.scale]
+        self.tags |= {
+            'maintenance' if self.variant == 'maint' else 'production'
+        }
 
     @rfm.run_before('run')
     def set_task_distribution(self):
@@ -132,16 +134,15 @@ def set_cpu_binding(self):
         self.job.launcher.options = ['--cpu-bind=cores']
 
 
-@rfm.parameterized_test(*([s, v]
-                          for s in ['small', 'large']
-                          for v in ['maint', 'prod']))
+@rfm.simple_test
 class QuantumESPRESSOGpuCheck(QuantumESPRESSOCheck):
-    def __init__(self, scale, variant):
+    def __init__(self):
         super().__init__()
-        self.descr = f'QuantumESPRESSO GPU check (version: {scale}, {variant})'
+        self.descr = (f'QuantumESPRESSO GPU check (version: {self.scale}, '
+                      f'{self.variant})')
         self.valid_systems = ['daint:gpu']
         self.num_gpus_per_node = 1
-        if scale == 'small':
+        if self.scale == 'small':
             self.valid_systems += ['dom:gpu']
             self.num_tasks = 6
             energy_reference = -11427.09017168
@@ -157,8 +158,6 @@ def __init__(self, scale, variant):
         energy_diff = sn.abs(energy-energy_reference)
         self.sanity_patterns = sn.all([
             self.sanity_patterns,
-            # FIXME temporarily increase energy difference
-            # (different CUDA default on Dom and Daint)
             sn.assert_lt(energy_diff, 1e-7)
         ])
 
@@ -183,5 +182,7 @@ def __init__(self, scale, variant):
             }
         }
 
-        self.reference = references[variant][scale]
-        self.tags |= {'maintenance' if variant == 'maint' else 'production'}
+        self.reference = references[self.variant][self.scale]
+        self.tags |= {
+            'maintenance' if self.variant == 'maint' else 'production'
+        }

cscs-checks/apps/tensorflow/tf2_horovod_check.py

@@ -8,26 +8,21 @@
 import reframe.utility.osext as osext
 
 
-@rfm.parameterized_test(['small'], ['large'])
+@rfm.simple_test
 class TensorFlow2HorovodTest(rfm.RunOnlyRegressionTest):
-    def __init__(self, variant):
+    variant = parameter(['small', 'large'])
+
+    def __init__(self):
         self.descr = 'Distributed training with TensorFlow2 and Horovod'
         self.valid_systems = ['daint:gpu']
         self.valid_prog_environs = ['builtin']
-
-        cray_cdt_version = osext.cray_cdt_version()
-        # FIXME: The following will not be needed after the Daint upgrade
-        if self.current_system.name == 'dom':
-            self.modules = [
-                f'Horovod/0.21.0-CrayGNU-{cray_cdt_version}-tf-2.4.0'
-            ]
-        else:
-            self.modules = ['Horovod/0.19.1-CrayGNU-20.08-tf-2.2.0']
-
+        self.modules = [
+            f'Horovod/0.21.0-CrayGNU-{osext.cray_cdt_version()}-tf-2.4.0'
+        ]
         self.sourcesdir = None
         self.num_tasks_per_node = 1
         self.num_cpus_per_task = 12
-        if variant == 'small':
+        if self.variant == 'small':
             self.valid_systems += ['dom:gpu']
             self.num_tasks = 8
             self.reference = {

cscs-checks/compile/libsci_acc_symlink.py

@@ -9,29 +9,30 @@
 import reframe.utility.sanity as sn
 
 
-# FIXME: Remove the 71 entries after daint upgrade
-@rfm.parameterized_test(['libsci_acc_gnu_81_nv35'],
-                        ['libsci_acc_gnu_81_nv60'],
-                        ['libsci_acc_cray_nv35'],
-                        ['libsci_acc_cray_nv60'],
-                        ['libsci_acc_cray_nv35_openacc'],
-                        ['libsci_acc_cray_nv60_openacc'])
+@rfm.simple_test
 class LibSciAccSymLinkTest(rfm.RunOnlyRegressionTest):
-    def __init__(self, lib_name):
-        self.descr = 'LibSciAcc symlink check of %s' % lib_name
+    lib_name = parameter([
+        'libsci_acc_gnu_81_nv35', 'libsci_acc_gnu_81_nv60',
+        'libsci_acc_cray_nv35', 'libsci_acc_cray_nv60',
+        'libsci_acc_cray_nv35_openacc', 'libsci_acc_cray_nv60_openacc'
+    ])
+
+    def __init__(self):
+        self.descr = f'LibSciAcc symlink check of {self.lib_name}'
         self.valid_systems = [
             'daint:login', 'daint:gpu',
             'dom:login', 'dom:gpu',
         ]
         regex = (r'libsci_acc_(?P<prgenv>[A-Za-z]+)_((?P<cver>[A-Za-z0-9]+)_)'
                  r'?(?P<version>\S+)')
-        prgenv = re.match(regex, lib_name).group('prgenv')
+        prgenv = re.match(regex, self.lib_name).group('prgenv')
 
         # The prgenv is irrelevant for this case, so just chose one
         self.valid_prog_environs = ['builtin']
         self.executable = 'ls'
         self.executable_opts = ['-al', '/opt/cray/pe/lib64/libsci_a*']
-        self.sanity_patterns = sn.assert_found(f'{lib_name}.so', self.stdout)
+        self.sanity_patterns = sn.assert_found(f'{self.lib_name}.so',
+                                               self.stdout)
 
         self.maintainers = ['AJ', 'LM']
         self.tags = {'production', 'craype', 'health'}

cscs-checks/mch/collectives_halo.py

@@ -8,13 +8,15 @@
 
 
 class CollectivesBaseTest(rfm.RegressionTest):
-    def __init__(self, variant, bench_reference):
+    variant = parameter(['default', 'nocomm', 'nocomp'])
+
+    def __init__(self, bench_reference):
         self.valid_systems = ['dom:gpu', 'daint:gpu', 'arolla:cn', 'tsa:cn']
         self.valid_prog_environs = ['PrgEnv-gnu']
         self.variables = {'G2G': '1'}
         self.executable = 'build/src/comm_overlap_benchmark'
-        if variant != 'default':
-            self.executable_opts = ['--' + variant]
+        if self.variant != 'default':
+            self.executable_opts = [f'--{self.variant}']
 
         self.build_system = 'CMake'
         self.build_system.builddir = 'build'
@@ -37,7 +39,7 @@ def __init__(self, variant, bench_reference):
             self.num_tasks = 4
             self.num_gpus_per_node = 1
             self.num_tasks_per_node = 1
-            self.modules = ['craype-accel-nvidia60', 'CMake']
+            self.modules = ['craype-accel-nvidia60', 'CMake', 'cdt-cuda']
             self.variables['MPICH_RDMA_ENABLED_CUDA'] = '1'
             self.build_system.config_opts += [
                 '-DCUDA_COMPUTE_CAPABILITY="sm_60"'
@@ -68,7 +70,7 @@ def __init__(self, variant, bench_reference):
             sysname = self.current_system.name
 
         try:
-            ref = bench_reference[sysname][variant]
+            ref = bench_reference[sysname][self.variant]
         except KeyError:
             ref = 0.0
 
@@ -87,44 +89,41 @@ def __init__(self, variant, bench_reference):
         else:
             self.tags = {'production', 'mch', 'craype'}
 
-    @rfm.run_before('compile')
-    def dom_set_cuda_cdt(self):
-        if self.current_system.name == 'dom':
-            self.modules += ['cdt-cuda']
-
     @rfm.run_before('run')
     def set_launcher_options(self):
         if self.current_system.name in ['arolla', 'tsa']:
             self.job.launcher.options = ['--distribution=block:block',
                                          '--cpu_bind=q']
 
 
-@rfm.parameterized_test(['default'], ['nocomm'], ['nocomp'])
+@rfm.simple_test
 class AlltoallvTest(CollectivesBaseTest):
-    def __init__(self, variant):
-        super().__init__(variant,
-                         {
-                             'daint': {
-                                 'nocomm':  0.0171947,
-                                 'nocomp':  0.0137893,
-                                 'default': 0.0138493
-                             },
-                         })
+    def __init__(self):
+        super().__init__(
+            {
+                'daint': {
+                    'nocomm':  0.0171947,
+                    'nocomp':  0.0137893,
+                    'default': 0.0138493
+                },
+            }
+        )
         self.strict_check = False
         self.sourcesdir = 'https://github.com/eth-cscs/comm_overlap_bench.git'
         self.prebuild_cmds = ['git checkout alltoallv']
 
 
-@rfm.parameterized_test(['default'], ['nocomm'], ['nocomp'])
+@rfm.simple_test
 class HaloExchangeTest(CollectivesBaseTest):
-    def __init__(self, variant):
-        super().__init__(variant,
-                         {
-                             'daint': {
-                                 'nocomm':  0.978306,
-                                 'nocomp':  1.36716,
-                                 'default': 2.53509
-                             },
-                         })
+    def __init__(self):
+        super().__init__(
+            {
+                'daint': {
+                    'nocomm':  0.978306,
+                    'nocomp':  1.36716,
+                    'default': 2.53509
+                },
+            }
+        )
         self.sourcesdir = 'https://github.com/eth-cscs/comm_overlap_bench.git'
         self.prebuild_cmds = ['git checkout barebones']

cscs-checks/mch/cuda_stress_test.py

@@ -19,7 +19,7 @@ def __init__(self):
             self.modules = ['cuda/10.1.243']
         else:
             self.valid_prog_environs = ['PrgEnv-gnu']
-            self.modules = ['craype-accel-nvidia60']
+            self.modules = ['craype-accel-nvidia60', 'cdt-cuda']
 
         self.sourcepath = 'cuda_stencil_test.cu'
         self.build_system = 'SingleSource'
@@ -40,8 +40,3 @@ def __init__(self):
         }
         self.tags = {'production', 'mch', 'craype'}
         self.maintainers = ['MKr', 'AJ']
-
-    @rfm.run_before('compile')
-    def dom_set_cuda_cdt(self):
-        if self.current_system.name == 'dom':
-            self.modules += ['cdt-cuda']

cscs-checks/mch/gpu_direct_cuda.py

@@ -17,7 +17,7 @@ def __init__(self):
         self.build_system = 'SingleSource'
         self.build_system.ldflags = ['-lcublas', '-lcudart']
         if self.current_system.name in ['daint', 'dom']:
-            self.modules = ['craype-accel-nvidia60']
+            self.modules = ['craype-accel-nvidia60', 'cdt-cuda']
             self.variables = {'MPICH_RDMA_ENABLED_CUDA': '1'}
             self.build_system.cxxflags = ['-ccbin CC', '-arch=sm_60']
         elif self.current_system.name in ['arolla', 'tsa']:
@@ -36,8 +36,3 @@ def __init__(self):
         self.sanity_patterns = sn.assert_reference(result, 1., -1e-5, 1e-5)
         self.maintainers = ['AJ', 'MKr']
         self.tags = {'production', 'mch', 'craype'}
-
-    @rfm.run_before('compile')
-    def dom_set_cuda_cdt(self):
-        if self.current_system.name == 'dom':
-            self.modules += ['cdt-cuda']

cscs-checks/microbenchmarks/gpu/dgemm/dgmemm.py

@@ -78,10 +78,7 @@ def set_gpu_arch(self):
         if nvidia_sm:
             self.build_system.cxxflags += [f'-arch=sm_{nvidia_sm}']
             if cp in {'dom:gpu', 'daint:gpu'}:
-                self.modules += ['craype-accel-nvidia60']
-                if cp == 'dom:gpu':
-                    self.modules += ['cdt-cuda']
-
+                self.modules += ['craype-accel-nvidia60', 'cdt-cuda']
             else:
                 self.modules += ['cuda']
 

cscs-checks/microbenchmarks/gpu/gpu_burn/gpu_burn_test.py

@@ -83,10 +83,7 @@ def set_gpu_arch(self):
         # Nvidia options
         if cs in {'dom', 'daint'}:
             gpu_arch = '60'
-            self.modules = ['craype-accel-nvidia60']
-            if cs == 'dom':
-                self.modules += ['cdt-cuda']
-
+            self.modules = ['craype-accel-nvidia60', 'cdt-cuda']
         elif cs in {'arola', 'tsa'}:
             gpu_arch = '70'
             self.modules = ['cuda/10.1.243']

cscs-checks/microbenchmarks/gpu/memory_bandwidth/memory_bandwidth.py

@@ -103,10 +103,7 @@ def set_gpu_arch(self):
         if nvidia_sm:
             self.build_system.cxxflags += [f'-arch=sm_{nvidia_sm}']
             if cp in {'dom:gpu', 'daint:gpu'}:
-                self.modules += ['craype-accel-nvidia60']
-                if cp == 'dom:gpu':
-                    self.modules += ['cdt-cuda']
-
+                self.modules += ['craype-accel-nvidia60', 'cdt-cuda']
             else:
                 self.modules += ['cuda']