Perfetto Tracing ================ Mesa has experimental support for `Perfetto `__ for GPU performance monitoring. Perfetto supports multiple `producers `__ each with one or more data-sources. Perfetto already provides various producers and data-sources for things like: - CPU scheduling events (``linux.ftrace``) - CPU frequency scaling (``linux.ftrace``) - System calls (``linux.ftrace``) - Process memory utilization (``linux.process_stats``) As well as various domain specific producers. The mesa Perfetto support adds additional producers, to allow for visualizing GPU performance (frequency, utilization, performance counters, etc) on the same timeline, to better understand and tune/debug system level performance: - pps-producer: A systemwide daemon that can collect global performance counters. - mesa: Per-process producer within mesa to capture render-stage traces on the GPU timeline, track events on the CPU timeline, etc. The exact supported features vary per driver: .. list-table:: Supported data-sources :header-rows: 1 * - Driver - PPS Counters - Render Stages * - Freedreno - ``gpu.counters.msm`` - ``gpu.renderstages.msm`` * - Turnip - ``gpu.counters.msm`` - ``gpu.renderstages.msm`` * - Intel - ``gpu.counters.i915`` - ``gpu.renderstages.intel`` * - Panfrost - ``gpu.counters.panfrost`` - * - PanVK - ``gpu.counters.panfrost`` - ``gpu.renderstages.panfrost`` * - V3D - ``gpu.counters.v3d`` - * - V3DV - ``gpu.counters.v3d`` - Run --- To capture a trace with Perfetto you need to take the following steps: 1. Build Mesa with perfetto enabled. .. code-block:: sh # Configure Mesa with perfetto mesa $ meson . build -Dperfetto=true -Dvulkan-drivers=intel,broadcom -Dgallium-drivers= # Build mesa mesa $ meson compile -C build 2. Build Perfetto from sources available at ``subprojects/perfetto``. .. code-block:: sh # Within the Mesa repo, build perfetto mesa $ cd subprojects/perfetto perfetto $ ./tools/install-build-deps perfetto $ ./tools/gn gen --args='is_debug=false' out/linux perfetto $ ./tools/ninja -C out/linux # Example arm64 cross compile instead perfetto $ ./tools/install-build-deps --linux-arm perfetto $ ./tools/gn gen --args='is_debug=false target_cpu="arm64"' out/linux-arm64 More build options can be found in `this guide `__. 3. Select a `trace config `__, likely ``src/tool/pps/cfg/system.cfg`` which does whole-system including GPU profiling for any supported GPUs). Other configs are available in that directory for CPU-only or GPU-only tracing, and more examples of config files can be found in ``subprojects/perfetto/test/configs``. 4. Start the PPS producer to capture GPU performance counters. .. code-block:: sh mesa $ sudo meson devenv -C build pps-producer 5. Start your application (and any other GPU-using system components) you want to trace using the perfetto-enabled Mesa build. .. code-block:: sh mesa $ meson devenv -C build vkcube 6. Capture a perfetto trace using ``tracebox``. .. code-block:: sh mesa $ sudo ./subprojects/perfetto/out/linux/tracebox --system-sockets --txt -c src/tool/pps/cfg/system.cfg -o vkcube.trace 7. Go to `ui.perfetto.dev `__ and upload ``vkcube.trace`` by clicking on **Open trace file**. 8. Alternatively you can open the trace in `AGI `__ (which despite the name can be used to view non-android traces). CPU Tracing ~~~~~~~~~~~ Mesa's CPU tracepoints (``MESA_TRACE_*``) use Perfetto track events when Perfetto is enabled. They use ``mesa.default`` and ``mesa.slow`` categories. Currently, only EGL and the following drivers have CPU tracepoints. - Freedreno - Panfrost - Turnip - V3D - VC4 - V3DV Render stage data sources ~~~~~~~~~~~~~~~~~~~~~~~~~ The render stage data sources are the driver-specific traces of command buffer execution on the GPU. The Vulkan API gives the application control over recording of command buffers as well as when they are submitted to the hardware, and command buffers can be recorded once and reused repeatedly. Trace commands are normally only recorded into a command buffer when a perfetto trace is active. Most applications don't reuse command buffers, so you'll see traces appear shortly after the trace was started, but if you have one of the rare applications that reuses command buffers, you'll need to set the :envvar:`MESA_GPU_TRACES` environment variable before starting a Vulkan application : .. code-block:: sh MESA_GPU_TRACES=perfetto ./build/my_vulkan_app Driver Specifics ~~~~~~~~~~~~~~~~ Below is driver specific information/instructions for the PPS producer. Freedreno / Turnip ^^^^^^^^^^^^^^^^^^ The Freedreno PPS driver needs root access to read system-wide performance counters, so you can simply run it with sudo: .. code-block:: sh sudo ./build/src/tool/pps/pps-producer Intel ^^^^^ The Intel PPS driver needs root access to read system-wide `RenderBasic `__ performance counters, so you can simply run it with sudo: .. code-block:: sh sudo ./build/src/tool/pps/pps-producer Another option to enable access wide data without root permissions would be running the following: .. code-block:: sh sudo sysctl dev.i915.perf_stream_paranoid=0 Alternatively using the ``CAP_PERFMON`` permission on the binary should work too. A particular metric set can also be selected to capture a different set of HW counters : .. code-block:: sh INTEL_PERFETTO_METRIC_SET=RasterizerAndPixelBackend ./build/src/tool/pps/pps-producer Vulkan applications can also be instrumented to be Perfetto producers. To enable this for given application, set the environment variable as follow : .. code-block:: sh PERFETTO_TRACE=1 my_vulkan_app Panfrost ^^^^^^^^ The Panfrost PPS driver uses unstable ioctls that behave correctly on kernel version `5.4.23+ `__ and `5.5.7+ `__. To run the producer, follow these two simple steps: 1. Enable Panfrost unstable ioctls via kernel parameter: .. code-block:: sh modprobe panfrost unstable_ioctls=1 Alternatively you could add ``panfrost.unstable_ioctls=1`` to your kernel command line, or ``echo 1 > /sys/module/panfrost/parameters/unstable_ioctls``. 2. Run the producer: .. code-block:: sh ./build/pps-producer V3D / V3DV ^^^^^^^^^^ As we can only have one performance monitor active at a given time, we can only monitor 32 performance counters. There is a need to define the performance counters of interest for pps_producer using the environment variable ``V3D_DS_COUNTER``. .. code-block:: sh V3D_DS_COUNTER=cycle-count,CLE-bin-thread-active-cycles,CLE-render-thread-active-cycles,QPU-total-uniform-cache-hit ./src/tool/pps/pps-producer Troubleshooting --------------- Missing counter names ~~~~~~~~~~~~~~~~~~~~~ If the trace viewer shows a list of counters with a description like ``gpu_counter(#)`` instead of their proper names, maybe you had a data loss due to the trace buffer being full and wrapped. In order to prevent this loss of data you can tweak the trace config file in two different ways: - Increase the size of the buffer in use: .. code-block:: javascript buffers { size_kb: 2048, fill_policy: RING_BUFFER, } - Periodically flush the trace buffer into the output file: .. code-block:: javascript write_into_file: true file_write_period_ms: 250 - Discard new traces when the buffer fills: .. code-block:: javascript buffers { size_kb: 2048, fill_policy: DISCARD, }