DCPerf: Adding support for lower granularity to topdown monitors (#190)

Summary:

This diff introduces support for specifying a lower granularity interval in Perfutil performance data collection, which is necessary for DCPerf mini, where the execution time is less than 45 seconds

Key changes include:
Added an interval parameter with a default of 5 seconds to various Perfutil classes (BasePerfUtil, AMDPerfUtil, ARMPerfUtil, NVPerfUtil).
Updated constructors and run methods to accept and use the interval parameter for controlling data collection frequency.
Modified configuration dictionaries to include interval values for the "topdown" monitor.
Enhanced subprocess command construction to pass the interval argument to the underlying perf collection scripts.

Reviewed By: YifanYuan3

Differential Revision: D79523911

Author: marziehlenjaniMeta

benchpress/plugins/hooks/perf.py

@@ -39,7 +39,7 @@
     "perfstat": {"interval": 5, "additional_events": []},
     "netstat": {"interval": 5, "additional_counters": []},
     "memstat": {"interval": 5, "additional_counters": []},
-    "topdown": {},
+    "topdown": {"interval": 5},
     "power": {"interval": 1},
 }
 

benchpress/plugins/hooks/perf_monitors/topdown.py

@@ -141,9 +141,12 @@ def get_os_release():
 
 
 class IntelPerfSpect(Monitor):
-    def __init__(self, job_uuid, mux_interval_msecs=125, perfspect_path=None):
-        # PerfSpect 1.x does not support specifying interval
-        super(IntelPerfSpect, self).__init__(1, "perfspect", job_uuid)
+    def __init__(self, interval, job_uuid, mux_interval_msecs=125, perfspect_path=None):
+        # NOTE: PerfSpect 1.x does not support configurable sampling intervals.
+        # The 'interval' parameter is accepted here only for API compatibility
+        # with the DEFAULT_OPTIONS in perf.py, but it is not actually used
+        # by this implementation.
+        super(IntelPerfSpect, self).__init__(interval, "perfspect", job_uuid)
         self.mux_interval_msecs = mux_interval_msecs
         if perfspect_path is None:
             self.perfspect_path = os.path.join(BP_BASEPATH, "perfspect")
@@ -216,14 +219,12 @@ def write_csv(self):
 class IntelPerfSpect3(Monitor):
     def __init__(
         self,
+        interval,
         job_uuid,
-        report_interval_secs=5,
         mux_interval_msecs=125,
         perfspect_path=None,
     ):
-        super(IntelPerfSpect3, self).__init__(
-            report_interval_secs, "perfspect3", job_uuid
-        )
+        super(IntelPerfSpect3, self).__init__(interval, "perfspect3", job_uuid)
         self.mux_interval_msecs = mux_interval_msecs
         if perfspect_path is None:
             self.perfspect_path = os.path.join(BP_BASEPATH, "perfspect")
@@ -282,14 +283,15 @@ def write_csv(self):
 class BasePerfUtil(Monitor):
     def __init__(
         self,
+        interval,
         job_uuid,
         name,
         perf_collect_script_name,
         perf_postproc_script_name,
         perfutils_path=None,
         perf_postproc_args=None,
     ):
-        super(BasePerfUtil, self).__init__(0, name, job_uuid)
+        super(BasePerfUtil, self).__init__(interval, name, job_uuid)
         if perfutils_path is None:
             self.perfutils_path = os.path.join(BP_BASEPATH, "perfutils")
         else:
@@ -314,6 +316,8 @@ def run(self):
             logger.warning(f"{perf_collect_script} does not exist")
             return
         cmd = [perf_collect_script]
+        if self.interval is not None:
+            cmd.append(str(self.interval))
         self.proc = subprocess.Popen(cmd, stdout=subprocess.PIPE, encoding="utf-8")
         super(BasePerfUtil, self).run()
 
@@ -355,20 +359,22 @@ def write_csv(self):
 
 
 class AMDPerfUtil:
-    def __init__(self, job_uuid, **kwargs):
+    def __init__(self, interval, job_uuid, **kwargs):
         self.cpuinfo = get_cpuinfo()
         self.cpu_vendor = get_cpu_vendor(self.cpuinfo)
         if self.cpu_vendor != "amd":
             raise Exception("Not an AMD processor!")
         self.amd_gen = get_amd_zen_generation(self.cpuinfo)
         self.perfutil = BasePerfUtil(
+            interval,
             job_uuid,
             "amd-perf-collector",
             perf_collect_script_name="collect_amd_perf_counters.sh",
             perf_postproc_script_name="generate_amd_perf_report.py",
         )
         if self.amd_gen == "zen4":
             self.perfutil_zen4 = BasePerfUtil(
+                interval,
                 job_uuid,
                 "amd-zen4-perf-collector",
                 perf_collect_script_name="collect_amd_zen4_perf_counters.sh",
@@ -377,6 +383,7 @@ def __init__(self, job_uuid, **kwargs):
             )
         elif self.amd_gen == "zen5":
             self.perfutil = BasePerfUtil(
+                interval,
                 job_uuid,
                 "amd-zen5-perf-collector",
                 perf_collect_script_name="collect_amd_zen5_perf_counters.sh",
@@ -385,6 +392,7 @@ def __init__(self, job_uuid, **kwargs):
             )
         elif self.amd_gen == "zen5es":
             self.perfutil = BasePerfUtil(
+                interval,
                 job_uuid,
                 "amd-zen5-perf-collector",
                 perf_collect_script_name="collect_amd_zen5_perf_counters.sh",
@@ -416,7 +424,7 @@ class ARMPerfUtil(Monitor):
         "https://git.gitlab.arm.com/telemetry-solution/telemetry-solution.git"
     )
 
-    def __init__(self, job_uuid, interval=5):
+    def __init__(self, interval, job_uuid, **kwargs):
         super(ARMPerfUtil, self).__init__(interval, "arm-perf-collector", job_uuid)
         self.avail = self.install_if_not_available()
         if not self.avail:
@@ -502,17 +510,21 @@ def write_csv(self):
 
 
 class NVPerfUtil(BasePerfUtil):
-    def __init__(self, job_uuid, **kwargs):
+    def __init__(self, interval, job_uuid, **kwargs):
         super(NVPerfUtil, self).__init__(
+            interval,
             job_uuid,
             "nv-perf-collector",
             perf_collect_script_name="collect_nvda_neoversev2_perf_counters.sh",
             perf_postproc_script_name="generate_arm_perf_report.py",
         )
 
+    def run(self):
+        super(NVPerfUtil, self).run()
+
 
 class DummyPerfUtil:
-    def __init__(self, job_uuid, **kwargs):
+    def __init__(self, interval, job_uuid, **kwargs):
         pass
 
     def run(self):
@@ -538,6 +550,12 @@ def choose_perfspect():
     elif os.path.exists(perfspect1_bin1) and os.path.exists(perfspect1_bin2):
         return IntelPerfSpect
     else:
+        logger.warning(f"Neither perfspect 1.x nor perfspect 3.x is available.\n \
+        None of the followings exist:\n \
+        {perfspect3_bin}\n \
+        {perfspect1_bin1} \n \
+        {perfspect1_bin2} \n \
+        .")
         return DummyPerfUtil
 
 

perfutils/generate_amd_perf_report.py

@@ -649,6 +649,7 @@ def zen4_frontend_bound(grouped_df):
     return {"name": "Zen4 frontend Bound %", "series": frontend_bound_pct_series}
 
 
+@skip_if_missing
 def zen4_backend_bound(grouped_df):
     de_no_dispatch_per_slot_backend_stalls_series = grouped_df.get_group(
         "de_no_dispatch_per_slot.backend_stalls"
@@ -2580,6 +2581,13 @@ def main(
         metrics.append(zen4_backend_bound(grouped_df))
 
     filtered_metrics = list(itertools.filterfalse(lambda x: x is None, metrics))
+
+    if not filtered_metrics:
+        click.echo(
+            "No metrics could be calculated. All required counters are missing from the input data."
+        )
+        return
+
     shortest_series = max(filtered_metrics, key=lambda m: m["series"].size)
     df_metrics = concat_series(filtered_metrics, shortest_series)
     if series: