Merge pull request #647 from sebkelle1/checks/cpu-bandwidth

[test] Add Likwid benchmarks for cache, memory and cross-socket bandwidths

Author: vkarak

cscs-checks/microbenchmarks/likwid/likwid.py

@@ -0,0 +1,160 @@
+import reframe as rfm
+import reframe.utility.sanity as sn
+
+
+class MemBandwidthTest(rfm.RunOnlyRegressionTest):
+    def __init__(self):
+        super().__init__()
+
+        self.modules = ['likwid']
+        self.valid_prog_environs = ['PrgEnv-gnu']
+        self.sourcesdir = None
+
+        self.executable = 'likwid-bench'
+
+        self.num_tasks = 0
+        self.num_tasks_per_core = 2
+        self.system_num_cpus = {
+            'daint:mc':  72,
+            'daint:gpu': 24,
+            'dom:mc':  72,
+            'dom:gpu': 24,
+        }
+        self.system_numa_domains = {
+            'daint:mc':  ['S0', 'S1'],
+            'daint:gpu': ['S0'],
+            'dom:mc':  ['S0', 'S1'],
+            'dom:gpu': ['S0'],
+        }
+
+        # Test each level at half capacity times nthreads per domain
+        self.system_cache_sizes = {
+            'daint:mc':  {'L1': '288kB', 'L2': '2304kB', 'L3': '23MB',
+                          'memory': '1800MB'},
+            'daint:gpu': {'L1': '192kB', 'L2': '1536kB', 'L3': '15MB',
+                          'memory': '1200MB'},
+            'dom:mc':    {'L1': '288kB', 'L2': '2304kB', 'L3': '23MB',
+                          'memory': '1800MB'},
+            'dom:gpu':   {'L1': '192kB', 'L2': '1536kB', 'L3': '15MB',
+                          'memory': '1200MB'},
+        }
+
+        self.maintainers = ['SK']
+        self.tags = {'benchmark', 'diagnostic'}
+
+        bw_pattern = sn.extractsingle(r'MByte/s:\s*(?P<bw>\S+)',
+                                      self.stdout, 'bw',  float)
+
+        self.sanity_patterns = sn.assert_ge(bw_pattern, 0.0)
+        self.perf_patterns = {
+            'bandwidth': bw_pattern
+        }
+
+
+@rfm.required_version('>=2.16-dev0')
+@rfm.parameterized_test(*[[l, k] for l in ['L1', 'L2', 'L3']
+                          for k in ['load_avx', 'store_avx']],
+                        ['memory', 'load_avx'],
+                        ['memory', 'store_mem_avx'])
+class CPUBandwidth(MemBandwidthTest):
+    def __init__(self, mem_level, kernel_name):
+        super().__init__()
+
+        self.descr = 'CPU <- %s %s benchmark' % (mem_level, kernel_name)
+        self.valid_systems = ['daint:mc', 'daint:gpu', 'dom:gpu', 'dom:mc']
+
+        # the kernel to run in likwid
+        self.kernel_name = kernel_name
+        self.mem_level = mem_level
+
+        self.refs = {
+            'mc':  {
+                'load_avx': {'L1': 5100000, 'L2': 2000000, 'L3': 900000,
+                             'memory': 130000},
+                'store_avx': {'L1': 2800000, 'L2': 900000, 'L3': 480000},
+                'store_mem_avx': {'memory': 85000},
+            },
+            'gpu': {
+                'load_avx': {'L1': 2100000, 'L2': 850000, 'L3': 360000,
+                             'memory': 65000},
+                'store_avx': {'L1': 1200000, 'L2': 340000, 'L3': 210000},
+                'store_mem_avx': {'memory': 42500},
+            }
+        }
+        ref_proxy = {part: self.refs[part][kernel_name][mem_level]
+                     for part in self.refs.keys()}
+
+        self.reference = {
+            'daint:gpu': {
+                'bandwidth': (ref_proxy['gpu'], -0.1, None, 'MB/s')
+            },
+            'daint:mc': {
+                'bandwidth': (ref_proxy['mc'], -0.1, None, 'MB/s')
+            },
+            'dom:gpu': {
+                'bandwidth': (ref_proxy['gpu'], -0.1, None, 'MB/s')
+            },
+            'dom:mc': {
+                'bandwidth': (ref_proxy['mc'], -0.1, None, 'MB/s')
+            },
+        }
+
+    def setup(self, partition, environ, **job_opts):
+        pfn = parition.fullname
+        self.data_size = self.system_cache_sizes[pfn][self.mem_level]
+        self.num_cpus_per_task = self.system_num_cpus[partition.fullname]
+        numa_domains = self.system_numa_domains[partition.fullname]
+        num_cpu_domain = self.num_cpus_per_task / (len(numa_domains) *
+                                                   self.num_tasks_per_core)
+        # result for daint:mc: '-w S0:100MB:18:1:2 -w S1:100MB:18:1:2'
+        # format: -w domain:data_size:nthreads:chunk_size:stride
+        # chunk_size and stride affect which cpus from <domain> are selected
+        workgroups = ['-w %s:%s:%d:1:2' %
+                      (dom, self.data_size, num_cpu_domain)
+                      for dom in numa_domains]
+
+        self.executable_opts = ['-t %s' % self.kernel_name] + workgroups
+
+        super().setup(partition, environ, **job_opts)
+
+
+@rfm.required_version('>=2.16-dev0')
+@rfm.simple_test
+class CPUBandwidthCrossSocket(MemBandwidthTest):
+    def __init__(self):
+        super().__init__()
+
+        self.descr = ("CPU S0 <- main memory S1 read " +
+                      "CPU S1 <- main memory S0 read")
+
+        self.valid_systems = ['daint:mc', 'dom:mc']
+        self.kernel_name = 'load_avx'
+        self.reference = {
+            'daint:mc': {
+                'bandwidth': (56000, -0.1, None, 'MB/s')
+            },
+            'dom:mc': {
+                'bandwidth': (56000, -0.1, None, 'MB/s')
+            },
+        }
+
+    def setup(self, partition, environ, **job_opts):
+
+        self.num_cpus_per_task = self.system_num_cpus[partition.fullname]
+        numa_domains = self.system_numa_domains[partition.fullname]
+
+        num_cpu_domain = (self.num_cpus_per_task /
+                          (len(numa_domains) * self.num_tasks_per_core))
+
+        # daint:mc: '-w S0:100MB:18:1:2-0:S1 -w S1:100MB:18:1:2-0:S0'
+        # format:
+        # -w domain:data_size:nthreads:chunk_size:stride-stream_nr:mem_domain
+        # chunk_size and stride affect which cpus from <domain> are selected
+        workgroups = ['-w %s:100MB:%d:1:2-0:%s' %
+                      (dom_cpu, num_cpu_domain, dom_mem)
+                      for dom_cpu, dom_mem in
+                      zip(numa_domains[:2], reversed(numa_domains[:2]))]
+
+        self.executable_opts = ['-t %s' % self.kernel_name] + workgroups
+
+        super().setup(partition, environ, **job_opts)