enhance the perf value update

Author: RUIJIEZHONG66166

.github/scripts/op_calculate_best_perf.py

@@ -14,22 +14,48 @@
 updated_cases = []
 removed_cases = []
 
+def safe_float_convert(value):
+    try:
+        return float(value) if value.strip() else None
+    except (ValueError, AttributeError):
+        return None
+
 def update_baseline(xpu_file, baseline_file, remove_missing=False):
     with open(xpu_file) as f:
         xpu_reader = csv.DictReader(f, delimiter=';')
         xpu_rows = list(xpu_reader)
-        xpu_fieldnames = xpu_reader.fieldnames  # Keep original field order
-        fieldnames = [f for f in xpu_fieldnames if f not in ['time(us)', 'E2E total time(us)', 'E2E forward time(us)']]
-        xpu_data = {make_key(row, fieldnames): (float(row['time(us)']), row) for row in xpu_rows}
+        xpu_fieldnames = xpu_reader.fieldnames
+        time_fields = ['time(us)', 'E2E total time(us)', 'E2E forward time(us)']
+        fieldnames = [f for f in xpu_fieldnames if f not in time_fields]
+        xpu_data = {}
+        for row in xpu_rows:
+            key = make_key(row, fieldnames)
+            time_values = {}
+            if 'time(us)' in row:
+                time_val = safe_float_convert(row['time(us)'])
+                if time_val is not None:
+                    time_values['time(us)'] = time_val
+            if 'E2E total time(us)' in row:
+                e2e_val = safe_float_convert(row['E2E total time(us)'])
+                if e2e_val is not None:
+                    time_values['E2E total time(us)'] = e2e_val
+            xpu_data[key] = (time_values, row)
 
     with open(baseline_file) as f:
         baseline_reader = csv.DictReader(f, delimiter=';')
         baseline_rows = list(baseline_reader)
         baseline_fieldnames = baseline_reader.fieldnames
 
     # To add new parameter of new ops into baseline file
-    all_fieldnames = xpu_fieldnames + [f for f in baseline_fieldnames if f not in xpu_fieldnames]
-    fieldnames = [f for f in all_fieldnames if f not in ['time(us)', 'E2E total time(us)', 'E2E forward time(us)']]
+    all_fieldnames = list(set(xpu_fieldnames + baseline_fieldnames))
+    # Maintain original order as much as possible
+    ordered_fieldnames = []
+    for f in xpu_fieldnames:
+        if f in all_fieldnames and f not in ordered_fieldnames:
+            ordered_fieldnames.append(f)
+    for f in baseline_fieldnames:
+        if f in all_fieldnames and f not in ordered_fieldnames:
+            ordered_fieldnames.append(f)
 
     baseline_keys = {make_key(row, fieldnames) for row in baseline_rows}
     xpu_keys = set(xpu_data.keys())
@@ -38,75 +64,117 @@ def update_baseline(xpu_file, baseline_file, remove_missing=False):
     for row in baseline_rows:
         key = make_key(row, fieldnames)
         if key in xpu_data:
-            xpu_time, xpu_row = xpu_data[key]
-            baseline_time = float(row['time(us)'])
-
-            if xpu_time < baseline_time:
-                updated_row = {}
-                for field in all_fieldnames:
-                    updated_row[field] = xpu_row.get(field, row.get(field, ''))
-                updated_row['time(us)'] = str(xpu_time)
-                if 'E2E total time(us)' in row:
-                    updated_row['E2E total time(us)'] = row['E2E total time(us)']
-                updated_cases.append((key, baseline_time, xpu_time, updated_row))
-                updated_rows.append(updated_row)
-            else:
-                ordered_row = {}
-                for field in all_fieldnames:
-                    ordered_row[field] = row.get(field, '')
-                updated_rows.append(ordered_row)
+            xpu_times, xpu_row = xpu_data[key]
+            updated_row = {}
+
+            # Copy all fields from baseline first
+            for field in ordered_fieldnames:
+                updated_row[field] = row.get(field, '')
+
+            # Update with xpu values where they exist
+            for field in ordered_fieldnames:
+                if field in xpu_row and xpu_row[field]:
+                    updated_row[field] = xpu_row[field]
+
+            # Handle time fields
+            updated = False
+            if 'time(us)' in xpu_times and 'time(us)' in row:
+                baseline_time = safe_float_convert(row['time(us)'])
+                if baseline_time is not None:
+                    xpu_time = xpu_times['time(us)']
+                    if xpu_time < baseline_time:
+                        updated_row['time(us)'] = str(xpu_time)
+                        updated = True
+
+            if 'E2E total time(us)' in xpu_times and 'E2E total time(us)' in row:
+                baseline_e2e = safe_float_convert(row['E2E total time(us)'])
+                if baseline_e2e is not None:
+                    xpu_e2e = xpu_times['E2E total time(us)']
+                    if xpu_e2e < baseline_e2e:
+                        updated_row['E2E total time(us)'] = str(xpu_e2e)
+                        updated = True
+
+            if updated:
+                updated_cases.append((key, row, updated_row))
+            updated_rows.append(updated_row)
         elif not remove_missing:
             ordered_row = {}
-            for field in all_fieldnames:
+            for field in ordered_fieldnames:
                 ordered_row[field] = row.get(field, '')
             updated_rows.append(ordered_row)
 
     # Add new cases
     for key in xpu_keys - baseline_keys:
-        xpu_time, xpu_row = xpu_data[key]
+        xpu_times, xpu_row = xpu_data[key]
         new_row = {}
-        for field in all_fieldnames:
+        for field in ordered_fieldnames:
             new_row[field] = xpu_row.get(field, '')
-        new_row['time(us)'] = str(xpu_time)
+
+        if 'time(us)' in xpu_times:
+            new_row['time(us)'] = str(xpu_times['time(us)'])
+        if 'E2E total time(us)' in xpu_times:
+            new_row['E2E total time(us)'] = str(xpu_times['E2E total time(us)'])
+
         updated_rows.append(new_row)
-        added_cases.append((key, xpu_time, new_row))
+        added_cases.append((key, xpu_times, new_row))
 
     # Resolve removed cases
     if remove_missing:
         for key in baseline_keys - xpu_keys:
             removed_case = next(row for row in baseline_rows if make_key(row, fieldnames) == key)
-            removed_cases.append((key, float(removed_case['time(us)']), removed_case))
+            removed_cases.append((key, removed_case))
 
     if added_cases:
         print(f"\nAdded {len(added_cases)} new case(s):")
-        for key, time, row in added_cases:
+        for key, times, row in added_cases:
             print(f"\n[New Case] {format_case(key)}")
-            print(f"Time: {time} us")
+            if 'time(us)' in times:
+                print(f"Time: {times['time(us)']} us")
+            if 'E2E total time(us)' in times:
+                print(f"E2E Time: {times['E2E total time(us)']} us")
             print("Parameters:")
             for k, v in row.items():
-                if k not in ['time(us)', 'E2E total time(us)', 'E2E forward time(us)']:
+                if k not in time_fields:
                     print(f"  {k}: {v}")
         print("-" * 60)
 
     if updated_cases:
         print(f"\nUpdated {len(updated_cases)} case(s):")
-        for key, old_time, new_time, row in updated_cases:
+        for key, old_row, new_row in updated_cases:
             print(f"\n[Updated] {format_case(key)}")
-            print(f"Time: {old_time} us → {new_time} us")
+            if 'time(us)' in old_row and 'time(us)' in new_row:
+                old_time = safe_float_convert(old_row['time(us)'])
+                new_time = safe_float_convert(new_row['time(us)'])
+                if old_time is not None and new_time is not None and old_time != new_time:
+                    print(f"Time: {old_time} us → {new_time} us")
+
+            if 'E2E total time(us)' in old_row and 'E2E total time(us)' in new_row:
+                old_e2e = safe_float_convert(old_row['E2E total time(us)'])
+                new_e2e = safe_float_convert(new_row['E2E total time(us)'])
+                if old_e2e is not None and new_e2e is not None and old_e2e != new_e2e:
+                    print(f"E2E Time: {old_e2e} us → {new_e2e} us")
+
             print("Parameters:")
-            for k, v in row.items():
-                if k not in ['time(us)', 'E2E total time(us)', 'E2E forward time(us)']:
+            for k, v in new_row.items():
+                if k not in time_fields:
                     print(f"  {k}: {v}")
         print("-" * 60)
 
     if remove_missing and removed_cases:
         print(f"\nRemoved {len(removed_cases)} case(s):")
-        for key, time, row in removed_cases:
+        for key, row in removed_cases:
             print(f"\n[Removed] {format_case(key)}")
-            print(f"Time: {time} us")
+            if 'time(us)' in row:
+                time_val = safe_float_convert(row['time(us)'])
+                if time_val is not None:
+                    print(f"Time: {time_val} us")
+            if 'E2E total time(us)' in row:
+                e2e_val = safe_float_convert(row['E2E total time(us)'])
+                if e2e_val is not None:
+                    print(f"E2E Time: {e2e_val} us")
             print("Parameters:")
             for k, v in row.items():
-                if k not in ['time(us)', 'E2E total time(us)', 'E2E forward time(us)']:
+                if k not in time_fields:
                     print(f"  {k}: {v}")
         print("-" * 60)
 
@@ -117,7 +185,7 @@ def update_baseline(xpu_file, baseline_file, remove_missing=False):
     Path(baseline_file).rename(backup_file)
 
     with open(baseline_file, 'w', newline='') as f:
-        writer = csv.DictWriter(f, fieldnames=all_fieldnames, delimiter=';')
+        writer = csv.DictWriter(f, fieldnames=ordered_fieldnames, delimiter=';')
         writer.writeheader()
         writer.writerows(updated_rows)