Feature/qgpu benchmark script

benchmark-qgpu/benchmark_report.html.j2

@@ -23,9 +23,9 @@
   <small>Logs root: {{ logs_root }}</small>
 
   <div class="chart">
-  <h2>Simulation performance (ns/day)</h2>
+  <h2>Simulation throughput (total ns/day)</h2>
   <div id="nsday-chart"></div>
-  <div class="note">Simulated nanoseconds per wall-clock day.</div>
+  <div class="note">Total simulated nanoseconds per wall-clock day across all concurrent processes.</div>
   </div>
 
   <div class="chart">
@@ -209,8 +209,8 @@ function lineChart(containerId, cfg) {
   });
 
   lineChart("nsday-chart", {
-    xLabel: "Processes", yLabel: "ns/day", xs: payload.procs,
-    series: [{ label: "Performance (ns/day)", ys: payload.ns_per_day }]
+    xLabel: "Processes", yLabel: "total ns/day", xs: payload.procs,
+    series: [{ label: "Total throughput (ns/day)", ys: payload.ns_per_day }]
   });
 
 
@@ -233,7 +233,7 @@ function lineChart(containerId, cfg) {
       <th>GPU util mean (%)</th><th>GPU util peak (%)</th>
       <th>VRAM util mean (%)</th><th>VRAM util peak (%)</th>
       <th>Speedup (×)</th>
-      <th>ns/day</th>
+      <th>Total ns/day</th>
     </tr>`;
   table.appendChild(thead);
 

benchmark-qgpu/benchmark_report.py

@@ -85,6 +85,7 @@ def fget(k):
         mem_means = [v["mem_util_mean"] for v in vals if math.isfinite(v["mem_util_mean"])]
         mem_peaks = [v["mem_util_peak"] for v in vals if math.isfinite(v["mem_util_peak"])]
         tmp_ns_per_day = [v["ns_per_day"] for v in vals if math.isfinite(v["ns_per_day"])]
+        total_ns_per_day = statistics.mean(tmp_ns_per_day) * p if tmp_ns_per_day else float("nan")
 
 
         rc_bad = sum(1 for v in vals if v["rc"] != 0)
@@ -99,7 +100,7 @@ def fget(k):
         gpu_util_peak.append(statistics.mean(util_peak) if util_peak else float("nan"))
         util_mem_mean.append(statistics.mean(mem_means) if mem_means else float("nan"))
         util_mem_peak.append(statistics.mean(mem_peaks) if mem_peaks else float("nan"))
-        ns_per_day.append(statistics.mean(tmp_ns_per_day) if tmp_ns_per_day else float("nan"))
+        ns_per_day.append(total_ns_per_day)
 
 
         Tn = max(walls) if walls else float("nan")
@@ -126,7 +127,7 @@ def fget(k):
             "vram_util_peak": statistics.mean(mem_peaks) if mem_peaks else float("nan"),
             "Tn": Tn,
             "speedup": speedup,
-            "ns_per_day": statistics.mean(tmp_ns_per_day) if tmp_ns_per_day else float("nan"),
+            "ns_per_day": total_ns_per_day,
         })
 
 
@@ -156,4 +157,3 @@ def fget(k):
     with open(out_html, "w", encoding="utf-8") as f:
         f.write(html_out)
     print(f"Report written to: {out_html}")
-

benchmark-qgpu/benchmark_run.py

@@ -350,7 +350,12 @@ def _get(d, dotted, default=None):
 def run(args):
     data_dir = os.path.expanduser(args.data_dir)   # e.g., TEST/water
     bin_path = os.path.expanduser(args.bin)        # e.g., /path/to/qdyn
-    max_procs = int(args.max_processes)
+    if getattr(args, "concurrency", None):
+        concurrency = sorted(dict.fromkeys(int(value) for value in args.concurrency))
+    elif args.max_processes is not None:
+        concurrency = list(range(1, int(args.max_processes) + 1))
+    else:
+        raise ValueError("Pass --concurrency or --max_processes.")
 
     if not os.path.isdir(data_dir):
         raise FileNotFoundError(f"data_dir not found: {data_dir}")
@@ -386,7 +391,7 @@ def run(args):
     os.makedirs(logs_dir, exist_ok=True)
     work(1, logs_dir, f'"{bin_path}" "{data_dir}"', steps)
 
-    for process_num in range(1, max_procs + 1):
+    for process_num in concurrency:
         print(f"Will run {process_num} processes in parallel:")
         logs_dir = os.path.join(current_dir, f"benchmark_logs/{process_num:02d}_procs")
         os.makedirs(logs_dir, exist_ok=True)
@@ -405,4 +410,3 @@ def run(args):
     # generate report
     out_html = os.path.join(current_dir, "benchmark_report.html")
     make_html_report(logs_root, out_html)
-