Fix: unify --rounds flag and disable profiling when rounds > 1

Remove -n short form for --rounds in run_example.py and scene_test.py
standalone entry to match pytest's conftest.py (which only defines --rounds).
Automatically disable --enable-profiling when --rounds > 1 with a
logger.warning, since profiling only captures the first round and
multi-round mode is for benchmarking.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: ChaoWao

examples/scripts/run_example.py

@@ -171,7 +171,6 @@ def compute_golden(tensors: dict, params: dict) -> None:
     )
 
     parser.add_argument(
-        "-n",
         "--rounds",
         type=int,
         default=None,
@@ -205,6 +204,10 @@ def compute_golden(tensors: dict, params: dict) -> None:
 
     configure_logging(args.log_level)
 
+    if args.rounds is not None and args.rounds > 1 and args.enable_profiling:
+        logger.warning("Profiling disabled: --rounds > 1")
+        args.enable_profiling = False
+
     # Validate paths
     kernels_path = Path(args.kernels)
     golden_path = Path(args.golden)

simpler_setup/scene_test.py

@@ -23,6 +23,7 @@
 from __future__ import annotations
 
 import inspect
+import logging
 import os
 import sys
 from contextlib import contextmanager
@@ -31,6 +32,8 @@
 
 from .log_config import DEFAULT_LOG_LEVEL, LOG_LEVEL_CHOICES, configure_logging
 
+logger = logging.getLogger(__name__)
+
 _compile_cache: dict[tuple[str, str, str], object] = {}
 
 
@@ -874,6 +877,9 @@ def test_run(self, st_platform, st_worker, request):
         skip_golden = request.config.getoption("--skip-golden", default=False)
         enable_profiling = request.config.getoption("--enable-profiling", default=False)
         enable_dump_tensor = request.config.getoption("--dump-tensor", default=False)
+        if rounds > 1 and enable_profiling:
+            logger.warning("Profiling disabled: --rounds > 1")
+            enable_profiling = False
 
         cls_name = type(self).__name__
         callable_obj = self.build_callable(st_platform)
@@ -946,7 +952,7 @@ def run_module(module_name):
             default="exclude",
             help="Manual case handling: exclude (default), include, only",
         )
-        parser.add_argument("-n", "--rounds", type=int, default=1, help="Run each case N times (default: 1)")
+        parser.add_argument("--rounds", type=int, default=1, help="Run each case N times (default: 1)")
         parser.add_argument("--skip-golden", action="store_true", help="Skip golden comparison (benchmark mode)")
         parser.add_argument("--enable-profiling", action="store_true", help="Enable profiling (first round only)")
         parser.add_argument("--dump-tensor", action="store_true", help="Dump per-task tensor I/O at runtime")
@@ -958,6 +964,10 @@ def run_module(module_name):
             help=f"Root logger level (default: {DEFAULT_LOG_LEVEL})",
         )
         args = parser.parse_args()
+        if args.rounds > 1 and args.enable_profiling:
+            logger.warning("Profiling disabled: --rounds > 1")
+            args.enable_profiling = False
+
         configure_logging(args.log_level)
 
         module = sys.modules[module_name]