[Benchmarks] Pin benchmarks to small set of cores

For better results stability, pin benchmark binaries to four cores with the maximum available frequency.

Author: PatKamin

devops/actions/run-tests/benchmark/action.yml

@@ -79,14 +79,14 @@ runs:
     shell: bash
     run: |
       # Compute the core range for the first NUMA node; second node is used by
-      # UMF. Skip the first 4 cores as the kernel is likely to schedule more
+      # UMF. Skip the first 3 cores as the kernel is likely to schedule more
       # work on these.
       CORES="$(lscpu | awk '
         /NUMA node0 CPU|On-line CPU/ {line=$0}
         END {
           split(line, a, " ")
           split(a[4], b, ",")
-          sub(/^0/, "4", b[1])
+          sub(/^0/, "3", b[1])
           print b[1]
         }')"
       echo "CPU core range to use: $CORES"

devops/scripts/benchmarks/benches/compute.py

@@ -10,6 +10,7 @@
 from enum import Enum
 from itertools import product
 from pathlib import Path
+from psutil import Process
 
 from git_project import GitProject
 from options import options
@@ -417,6 +418,24 @@ def run(
         command += self.bin_args(run_trace)
         env_vars.update(self.extra_env_vars())
 
+        # Pin compute benchmarks to a CPU cores set to ensure consistent results
+        # and non-zero CPU count measurements (e.g. avoid E-cores). 4 max freq cores
+        # are pinned by default to satisfy multiple threads benchmarks.
+
+        available_cores = Process().cpu_affinity()
+        # Get 4 cores with the highest available frequency.
+        core_frequencies = []
+        for core in available_cores:
+            with open(
+                f"/sys/devices/system/cpu/cpu{core}/cpufreq/cpuinfo_max_freq"
+            ) as f:
+                freq = int(f.read().strip())
+                core_frequencies.append((core, freq))
+        core_frequencies.sort(key=lambda x: x[1], reverse=True)
+        available_cores = [core for core, _ in core_frequencies[:4]]
+
+        command = ["taskset", "-c"] + [str(core) for core in available_cores] + command
+
         result = self.run_bench(
             command, env_vars, run_trace=run_trace, force_trace=force_trace
         )