[pull] develop from isaac-sim:develop

docs/source/api/lab/isaaclab.benchmark.rst

@@ -150,6 +150,10 @@ Workflow Functions
 
 .. autofunction:: run_runtime_benchmark
 
+.. autofunction:: isaaclab.benchmark.stepping.profile_physics_steps
+
+.. autofunction:: isaaclab.benchmark.stepping.profile_renderers
+
 .. autofunction:: run_startup_benchmark
 
 .. autofunction:: run_training_benchmark

docs/source/developer-tools/benchmarking/benchmark_api.rst

@@ -58,6 +58,15 @@ The command prints the summary report. The paths in ``result.output_paths``
 identify the schema and summary JSON files that were written. Use these paths in
 automation instead of reconstructing the timestamped names.
 
+With a non-``None`` task ``benchmark_mode`` and ``ISAACLAB_RENDER_PROFILE=1`` or
+``ISAACLAB_PHYSICS_PROFILE=1``, the runtime workflow also includes scalar profiling
+summaries in ``result.bundle.extra``: ``physics_mean_ms``, ``physics_std_ms``,
+``physics_max_ms``, ``physics_calls``, and the corresponding ``render_*`` keys.
+These report mean, standard deviation, maximum time per call [ms], and call count
+over the captured samples. Disabled scopes contribute no keys. The summaries are
+included in schema and OmniPerf output without changing schema version 1.4.
+Raw ordered samples remain in ``<output_path>/profile_timings.json`` for local analysis.
+
 Choose a request
 ----------------
 

docs/source/developer-tools/benchmarking/run_benchmarks.rst

@@ -78,6 +78,34 @@ environment-step rate. Runtime samples random actions before starting the
 runtime run, ``runtime.collection_fps`` and ``runtime.total_fps`` describe the
 same random-action stepping workload.
 
+Render and physics scope profiling requires a non-``None`` ``benchmark_mode`` in the
+task configuration. If the field is absent or ``None``, both scopes remain disabled
+even when their profiling flags are set; standard runtime reports are still produced.
+
+For ``Isaac-RenderBenchmark-Franka-Cabinet``, ``BENCHMARK_MODE=render`` (the default)
+writes analytic joint poses after physics, while ``BENCHMARK_MODE=physics_render``
+sets actuator targets before physics. Both modes still step physics. Direct posing
+requires ``scene.lazy_sensor_update=True``. With Isaac RTX, use ``--visualizer none``
+in this mode: a Kit visualizer would render before the pose write.
+
+Set ``ISAACLAB_PHYSICS_PROFILE=1`` to collect synchronized physics-step timings during
+the runtime measurement loop. The benchmark wraps the selected physics
+manager's ``step`` through :func:`~isaaclab.benchmark.stepping.profile_physics_steps`
+during the measurement loop. Similarly, ``ISAACLAB_RENDER_PROFILE=1`` wraps registered
+renderers through :func:`~isaaclab.benchmark.stepping.profile_renderers`, timing only
+``render()`` and excluding scene updates and output readback. Both context managers install
+wrappers after warmup and restore the original methods when measurement ends, including on
+failure, so subsequent benchmark runs are unaffected. The render sweep
+enables both flags automatically. Ordered ``[scope, elapsed_ms]`` samples are written under
+``timings_ms`` in the local ``<output_path>/profile_timings.json`` file. The benchmark bundle's
+``extra`` dictionary holds scalar ``physics_mean_ms``, ``physics_std_ms``, ``physics_max_ms``,
+``physics_calls``, and corresponding ``render_*`` summaries. Disabled scopes contribute no keys.
+Schema version 1.4 remains unchanged, and OmniPerf output includes the same summaries.
+These statistics describe individual calls after warmup; the render sweep instead groups physics
+calls by rendered frame and discards its padding frames. The sweep reads the local profiling file
+from a separate directory for each profile. Device synchronization changes execution overlap,
+so these profiled runs are diagnostics rather than throughput measurements.
+
 .. dropdown:: Canonical workstation output and provenance
 
    Headless runtime summary (output abbreviated):