MobileGL/tools/trace_replay/README.md

MobileGL trace replay

This directory builds a Linux command line replay runner for apitrace files. It is an integration testing infrastructure of MobileGL.

The bundled fixtures cover:

• OpenRA: sourced from GL4ES' apitrace corpus.
• minecraft-1.21.4-startup: captured from Minecraft 1.21.4's startup screen.
• minecraft-1.21.4-main-menu: captured from Minecraft 1.21.4's main menu.
• minecraft-1.21.11-main-menu: captured from Minecraft 1.21.11's main menu on a Pixel 8 Pro through FCL MobileGL.
• minecraft-1.17-main-menu-854: captured from Minecraft 1.17's 854x480 main menu through FCL MobileGL capture.
• minecraft-1.21.4-in-world: captured from Minecraft 1.21.4 after entering a singleplayer world.
• minecraft-1.21.4-rd12-odinlite-in-world: captured on an Android device (FCL MobileGL Magma capture, 854x480) from vanilla Minecraft 1.21.4 at render distance 12, drifting down a river valley in a boat. Unlike the other in-world fixtures this one is a 251-frame window (gltrim -f 1094-1343) rather than a single frame, so it doubles as the campaign's benchmark scene: benchmark mode replays the whole window and the tail frames measure steady-state in-world frame time. The golden is still the final frame, so it works as an ordinary correctness case too.
• minecraft-1.21.4-fabric-sodium-in-world: captured from Minecraft 1.21.4 Fabric with Sodium after entering a singleplayer world with Fancy graphics.
• improved-transparency-minecraft-26.3: captured from the Minecraft 26.3 improved-transparency scene.
• minecraft-1.21.4-fabric-common-mods-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, REI, Xaero's Minimap, Xaero's World Map, JourneyMap, and Modern UI, with shader packs disabled.
• minecraft-1.21.4-fabric-common-mods-inventory: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, REI, Xaero's Minimap, Xaero's World Map, JourneyMap, and Modern UI with the creative inventory and REI item list open.
• minecraft-1.21.4-fabric-rei-inventory: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and REI, with shader packs disabled and the creative inventory and REI item list open.
• minecraft-1.21.4-fabric-xaero-minimap-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and Xaero's Minimap after entering a singleplayer world with shader packs disabled.
• minecraft-1.21.4-fabric-xaero-world-map-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and Xaero's World Map, with shader packs disabled and the world map screen open.
• minecraft-1.21.4-fabric-journeymap-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and JourneyMap after entering a singleplayer world with shader packs disabled.
• minecraft-1.21.4-fabric-modernui-inventory: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and Modern UI, with shader packs disabled and the creative inventory open.
• minecraft-1.21.4-fabric-rei-inventory-normal-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and REI in a normal singleplayer world, with shader packs disabled and the creative inventory and REI item list open.
• minecraft-1.21.4-fabric-xaero-minimap-in-world-normal-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and Xaero's Minimap after entering a normal singleplayer world with shader packs disabled.
• minecraft-1.21.4-fabric-xaero-world-map-in-world-normal-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and Xaero's World Map in a normal singleplayer world, with shader packs disabled and the world map screen open.
• minecraft-1.21.4-fabric-journeymap-in-world-normal-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and JourneyMap after entering a normal singleplayer world with shader packs disabled.
• minecraft-1.21.4-fabric-modernui-inventory-normal-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and Modern UI in a normal singleplayer world, with shader packs disabled and the creative inventory open.
• minecraft-1.21.4-fabric-iris-bsl-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and BSL Shaders after entering a singleplayer world.
• minecraft-1.21.4-fabric-iris-bsl-esc-menu-854: captured on an Android device (Mali-G77, FCL MobileGL capture) from Minecraft 1.21.4 Fabric with Sodium, Iris, and BSL Shaders, at the pause menu over a BSL-blurred world. The frame pins glyph rendering: every menu label, the menu title and the tutorial toast must be present. Regressions in the DirectGLES per-draw texture memo have made the whole text path disappear here while sprites kept rendering, so a failure that leaves the buttons but empties them is the signature to look for in the diff.
• minecraft-1.21.4-fabric-iris-makeup-ultrafast-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and MakeUP UltraFast after entering a singleplayer world.
• minecraft-1.21.4-fabric-iris-super-duper-vanilla-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and Super Duper Vanilla after entering a singleplayer world.
• minecraft-1.21.4-fabric-iris-complementary-reimagined-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and Complementary Reimagined after entering a singleplayer world.
• minecraft-1.21.4-fabric-iris-complementary-unbound-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and Complementary Unbound after entering a singleplayer world.
• minecraft-1.21.4-fabric-iris-mellow-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and Mellow after entering a singleplayer world.
• minecraft-1.21.4-fabric-iris-nostalgia-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and Nostalgia after entering a singleplayer world.
• minecraft-1.21.4-fabric-iris-bliss-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and Bliss after entering a singleplayer world.
• minecraft-1.21.4-fabric-iris-chocapic-v6-lite-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and Chocapic V6 Lite after entering a singleplayer world.
• minecraft-1.21.4-fabric-iris-iterationt-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and iterationT after entering a singleplayer world.
• minecraft-1.21.4-fabric-iris-iterationt-nodsa-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and iterationT after entering a singleplayer world, with Iris' DSA path disabled.
• minecraft-1.21.4-fabric-iris-iterationrp-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and iterationRP after entering a singleplayer world, framing the iterationRP name overlay over a lake with far-shore tree reflections. iterationRP's temporal auto-exposure makes a single-frame trim overexpose and drop the overlay, so the fixture is a prefix trace (all calls up to the target frame) that replays the temporal state. The pack also gates an NVIDIA-only shadow path (subgroupPartitionNV, GL_NV_shader_subgroup_partitioned) on the GL vendor string, so the capture reports a masked vendor and the trace carries the portable subgroupShuffleXor path that non-NVIDIA GPUs take. The trace archive and golden are not committed yet (the repository's Git LFS quota rejects new objects with GH009); the case stays registered and its fixture files are hydrated from the trace fixture mirror.
• minecraft-1.21.4-fabric-iris-photon-v1.1-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and Photon v1.1 after entering a singleplayer world.
• minecraft-1.21.4-fabric-iris-photon-v1.3b-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and Photon v1.3b after entering a singleplayer world.
• minecraft-1.21.4-fabric-iris-derivative-main-d24.4.14-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and Derivative Main d24.4.14 after entering a singleplayer world.
• minecraft-1.21.4-fabric-iris-sundial-lite-in-world: captured from Minecraft 1.21.4 Fabric with Sodium, Iris, and Sundial Lite after entering a singleplayer world.
• minecraft-1.21.1-neoforge-create-indirect-in-world: captured from Minecraft 1.21.1 NeoForge with Create, Sodium, and Iris (no shader pack) in a world facing Create water wheels and a large cogwheel, with Flywheel's flywheel:indirect backend (compute-shader culling, glMultiDrawElementsIndirect, persistent-mapped staging).
• minecraft-1.21.1-neoforge-create-instancing-in-world: same world and camera as the indirect case, with Flywheel's flywheel:instancing backend (texture-buffer instance data, glDrawElementsInstancedBaseVertex).

Build from the MobileGL repository root:

cmake -S . -B build-test -G Ninja \
  -DMOBILEGL_BUILD_TEST=ON \
  -DMOBILEGL_BUILD_BENCHMARK=OFF \
  -DMOBILEGL_BUILD_TRACE_REPLAY=ON
cmake --build build-test

Run the fixture tests:

ctest --test-dir build-test -V -R 'MobileGLTraceReplay\.'

Run the CLI directly:

build-test/tools/trace_replay/mobilegl_trace_replay \
  --trace openra.trace \
  --golden openra.0000031249.png \
  --output out/openra \
  --backend DirectGLES \
  --mobilegl-library build-test/libMobileGL.so \
  --target-call 31249 \
  --width 640 \
  --height 480 \
  --crop-x 1 \
  --crop-y 1 \
  --crop-width 638 \
  --crop-height 478 \
  --ssim-threshold 0.99

Dumping framebuffer attachments mid-frame

--target-call snapshots one framebuffer. To see inside a frame - which intermediate render target a pass actually produced - pass --dump-fbo-attachments CALL:DIR[:FBO,FBO,...], repeatably:

build-test/tools/trace_replay/mobilegl_trace_replay \
  --trace trace.trace --golden golden.png --output out --target-call 2667619 \
  --dump-fbo-attachments 2666231:out/fbos-before \
  --dump-fbo-attachments 2666232:out/fbos-after

At each call boundary it walks every live framebuffer object (or only the named ones), reads back every colour attachment and the depth attachment, and writes fbo<N>-att<M>.png / fbo<N>-depth.png plus a manifest.txt line per attachment recording the attached object, size, internal format, component type and per-channel min/max/mean and a content hash. Attachments are read as floats whatever their storage, so HDR accumulation buffers stay legible in the statistics even though the PNG has to clamp.

The manifest is the useful part when comparing two drivers: dump the same call on both stacks and diff/paste the two manifests, and the first attachment whose hash differs names the pass that diverged. Read-side and pixel-pack state is saved and restored, so the replay continues unperturbed; without the flag nothing is installed and the replay is byte-for-byte what it was.

Run the macOS native-window DirectVulkan retrace matrix and render the same HTML overview shape as CI:

cmake -S . -B cmake-build-macos-trace-arm64 -G Ninja \
  -DCMAKE_BUILD_TYPE=Release \
  -DCMAKE_OSX_ARCHITECTURES=arm64 \
  -DMOBILEGL_BUILD_TEST=OFF \
  -DMOBILEGL_BUILD_BENCHMARK=OFF \
  -DMOBILEGL_BUILD_TRACE_REPLAY=ON
cmake --build cmake-build-macos-trace-arm64 --target MobileGL mobilegl_trace_replay
python3 tools/trace_replay/run_macos_window_retrace_local.py --ci --all
open .trace-work/macos-window-retrace-summary/mobilegl-macos-window-vulkan-retrace-overview.html

The macOS runner hydrates missing fixtures from the trace fixture mirror with parallel downloads before falling back to Git LFS. It reuses the cmake-build-macos-trace-arm64 harness by default on Apple Silicon, passes --window-surface, and defaults to DirectVulkan only. If a native-window replay hits a fatal assertion, the runner writes a failure result and stops before launching later cases; use --continue-after-fatal to collect the full matrix, or --skip-case NAME for known fatal cases.

Android device replay

Build and install the generic trace APK from the repository root. Both DirectGLES and DirectVulkan use the same APK and package; select the backend with the intent's backend extra.

gradle --no-daemon -p android-plugin :app:assembleTraceDebug
TRACE_APK=$(find android-plugin/app/build/outputs/apk/trace/debug -maxdepth 1 -name '*.apk' -print -quit)
adb install -r "$TRACE_APK"

Prepare a fixture and copy it into the app-private directory:

mkdir -p /tmp/mobilegl-openra
tar -xzf tools/trace_replay/fixtures/openra.tgz -C /tmp/mobilegl-openra
adb push /tmp/mobilegl-openra/openra.trace /data/local/tmp/mobilegl-openra.trace
adb push tools/trace_replay/fixtures/openra.0000031249.png /data/local/tmp/mobilegl-openra.golden.png

PKG=top.mobilegl.plugin.trace
APP_DIR=/data/user/0/$PKG/files/trace-replay
adb shell run-as $PKG rm -rf files/trace-replay
adb shell run-as $PKG mkdir -p files/trace-replay/input files/trace-replay/output
adb shell run-as $PKG cp /data/local/tmp/mobilegl-openra.trace files/trace-replay/input/openra.trace
adb shell run-as $PKG cp /data/local/tmp/mobilegl-openra.golden.png files/trace-replay/input/openra.golden.png

Launch the standalone trace runner Activity:

adb shell am force-stop $PKG
adb shell am start -W -a top.mobilegl.plugin.TRACE_REPLAY \
  -n $PKG/top.mobilegl.plugin.trace.TraceReplayActivity \
  --es trace_path $APP_DIR/input/openra.trace \
  --es golden_path $APP_DIR/input/openra.golden.png \
  --es output_dir $APP_DIR/output \
  --es diff_path $APP_DIR/output/openra-diff.png \
  --es backend DirectGLES \
  --el target_call 31249 \
  --ei width 640 \
  --ei height 480 \
  --ei crop_x 1 \
  --ei crop_y 1 \
  --ei crop_width 638 \
  --ei crop_height 478 \
  --es ssim_threshold 0.99

Read back the result and images:

adb shell run-as $PKG cat files/trace-replay/output/result.json
adb exec-out run-as $PKG cat files/trace-replay/output/actual.png > openra-actual.png
adb exec-out run-as $PKG cat files/trace-replay/output/openra-diff.png > openra-diff.png

For Vulkan replay, keep the same APK and $PKG, then pass --es backend DirectVulkan. DirectGLES also renders to the Activity surface by default; pass --ez use_pbuffer true to use the offscreen pbuffer path. Always adb shell am force-stop $PKG before another replay: apitrace snapshot state is process-local. For cases registered with coherent_as_flush (Flywheel-style unflushed persistent maps, e.g. the Create fixtures), pass --ez coherent_as_flush true so the replay runs with MOBILEGL_COHERENT_AS_FLUSH=1. For cases registered with avoid_angle_llvmpipe_explicit_lod_bias (DirectGLES on ANGLE llvmpipe, e.g. the sundial-lite fixture), pass --ez avoid_angle_llvmpipe_explicit_lod_bias true so the replay runs with MOBILEGL_ESPRYT_AVOID_EXPLICIT_LOD_BIAS=1.

Repeats, the CI split subset, and keeping every repeat's evidence

run_android_retrace_local.py drives the above through trace-replay-ci.sh. Three options exist for the P7-7 device diagnostics, where the question is not "did this case pass" but "which cases diverge, and is the divergence reproducible":

option	what it does
--matrix	run the CI split subset (the set retrace-split runs; 39 cases today) instead of the whole manifest. --all additionally includes the non-CI workloads such as rd12, which ID-P7-4 excludes from the exit-gate-3 denominator. Each case runs only the backends its own ci_backends names - and a case NAMED with --case that does not run the requested backend is an error, not a silent skip.
--repeat N	replay each case/backend N times. Repeats after the first pass --reuse-fixture, so they measure the replay and not the push.
--archive-dir DIR	copy each repeat's result.json, actual PNG, both role logs, transport-proof.json and logcat.txt to DIR/<case>-<backend>/repeat-NN/ before the next run overwrites them, plus DIR/run.json recording the APK SHA-256, the arm and the environment.

The archive exists because .trace-work/android-retrace-result/<case>-<backend>/ is keyed by case and backend and by nothing else: without --archive-dir the only artefact that survives a three-repeat run is the last one, and "were the three repeats the same picture?" becomes unanswerable after the fact.

python tools/trace_replay/run_android_retrace_local.py \
  --matrix --backend DirectVulkan --use-pbuffer --transport inproc --repeat 3 \
  --archive-dir .trace-work/p7w1/E0-inproc
python3 tools/trace_replay/compare_actuals.py summary .trace-work/p7w1/E0-inproc \
  --json E0-inproc-summary.json

compare_actuals.py computes the pair result.json cannot: SSIM between two actual PNGs. It is the retrace gate's OWN SSIM - a transcription of ComputeChannelSsim / ComputeRgbSsim (android-plugin/app/src/trace/cpp/trace_replay_core.cpp:630 and :671), one global window per channel, C1 = (0.01255)², C2 = (0.03255)², alpha ignored - and the crop comes from each repeat's own result.json, so the ssim_vs_first column describes the same rectangle as the recorded ssim_vs_golden. summary prints case × repeat × ssim_vs_golden × ssim_vs_first × bit-identical; compare A.png B.png scores one pair; --self-test checks the transcription against closed-form values. Unit tests: python3 tools/trace_replay/test_compare_actuals.py.

Benchmark mode (frame timing)

Benchmark mode reuses the same fixtures as a performance harness instead of a correctness one: it replays the trace from the first call to the last, takes a wall-clock timestamp at every frame boundary, and takes no snapshot and runs no SSIM comparison. passed then only means the replay reached the end of the trace without an error.

Frame times include GPU completion by default, because a swap boundary on a tiled mobile GPU returns long before the tiler is done and would otherwise time CPU submission alone. That is what benchmark_finish / --benchmark-finish controls: on (the default) it issues a glFinish through the replayed context at every frame boundary, which serializes CPU/GPU overlap - pessimistic against a real running game, but deterministic and comparable between backends and revisions. Turn it off to measure CPU-side submission only.

The mean/median/p95 are computed over the last benchmark_tail_frames frames (default 200, clamped to the frames actually recorded); the head of a trace is dominated by shader compiles and first-use uploads. The full per-frame array is in the timing JSON, next to the headline numbers, which also appear in result.json.

Whole device runs, one line per run plus the best of the repeats:

python tools/trace_replay/run_android_retrace_local.py --benchmark \
  --case minecraft-1.21.4-fabric-iris-photon-in-world --backend DirectVulkan \
  --benchmark-repeats 3

--benchmark-tail-frames N, --benchmark-no-finish and --benchmark-timeout-seconds N are available; only the first repeat installs the APK and pushes the trace. Each run's benchmark.json is kept next to the case result as benchmark-run<N>.json.

The Activity takes the same settings directly:

adb shell am start -a top.mobilegl.plugin.TRACE_REPLAY \
  -n $PKG/top.mobilegl.plugin.trace.TraceReplayActivity \
  --es trace_path $APP_DIR/input/openra.trace \
  --es output_dir $APP_DIR/output \
  --es backend DirectGLES \
  --ei width 640 --ei height 480 \
  --ez benchmark true \
  --ei benchmark_tail_frames 200 \
  --ez benchmark_finish true \
  --es benchmark_result_path $APP_DIR/output/benchmark.json
adb exec-out run-as $PKG cat files/trace-replay/output/benchmark.json > benchmark.json

golden_path and target_call are not needed in benchmark mode. The Linux CLI takes the same options:

./mobilegl_trace_replay --trace openra.trace --output out --backend DirectGLES \
  --benchmark --benchmark-tail-frames=200 --benchmark-finish=1 \
  --benchmark-result=out/benchmark.json

Two caveats when reading the numbers. Frame times are taken at eglSwapBuffers, so a trace that ends frames with glFrameTerminatorGREMEDY instead is not timed. And on a window surface the swap can block on the compositor, which pins frame times to the display refresh; replay against the pbuffer surface (--ez use_pbuffer true) to measure the renderer rather than the presentation path.

Reproducing the Android DirectGLES lane on Linux (ANGLE on lavapipe)

The APK workflow's DirectGLES lane is not the same stack as the Linux one, which is why a case can be green here and red there:

lane	stack
Linux Test retrace, DirectGLES	Espryt -> Mesa GLES -> llvmpipe
Android APK retrace, DirectGLES	Espryt -> ANGLE -> Mesa Vulkan (lavapipe)
Android APK retrace, DirectVulkan	Magma -> lavapipe (no ANGLE)

Only the Android DirectGLES lane puts ANGLE in the middle, so an ANGLE translation difference shows up in exactly one of the six combinations. That stack can be reproduced on Linux without an emulator, which is far faster to iterate on than a CI round trip. The Android emulator SDK ships a glibc ANGLE:

ANGLE=$ANDROID_SDK_ROOT/emulator/lib64/gles_angle
mkdir -p ~/angle-farm && cd ~/angle-farm
# MobileGL dlopens these two names; ANGLE's own libEGL then dlopens the
# unsuffixed libGLESv2.so from the same directory - without that symlink it
# loads a truncated entry-point table and dies on a missing EGL function.
ln -sf $ANGLE/libEGL.so    libEGL_angle.so
ln -sf $ANGLE/libGLESv2.so libGLESv2_angle.so
ln -sf $ANGLE/libEGL.so    libEGL.so
ln -sf $ANGLE/libGLESv2.so libGLESv2.so
ln -sf $ANGLE/libvulkan.so.1 libvulkan.so.1   # else eglInitialize fails

MOBILEGL_ESPRYT_USE_ANGLE=1 \
LD_LIBRARY_PATH=~/angle-farm:/path/to/build/ \
VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/lvp_icd.json \
ANGLE_DEFAULT_PLATFORM=vulkan \
  ./mobilegl_trace_replay --trace trace.trace --golden golden.png \
    --target-call N --width 854 --height 480 --backend DirectGLES \
    --output outdir --pbuffer-surface

ANGLE_DEFAULT_PLATFORM=vulkan is required: ANGLE otherwise picks its OpenGL backend and you get ANGLE (Mesa, llvmpipe ..., OpenGL 4.6 (Core Profile)) instead of the CI-shaped ANGLE (Mesa, Vulkan 1.x (llvmpipe ...)). Check MOBILEGL_TRACE_GL_RENDERER in outdir/retrace.log before trusting a result. Run the binary directly rather than through ctest, whose ENVIRONMENT property overrides these variables. Build with clang, not gcc: gcc rejects GLXImpl.cpp under -Wchanges-meaning.

One more caveat before attributing anything: the emulator SDK's ANGLE is not the ANGLE the Android lane runs. The CI lane uses a pinned build (MOBILEGL_TRACE_ANGLE_VARIANT, default ec889e6ea831) whose version and extension set differ from the SDK copy (GL_EXT_texture_buffer support, ES 3.2 entry points). Compare GL_RENDERER and the relevant extension lists on both stacks before treating a local result as a statement about CI.

TCP matrix driver (Linux / WSL)

run_tcp_matrix.py uses a CTest JSON catalog plus the current trace_cases.json. It does not build, install an APK, or launch the remote server. For the formal P6.5 run-ahead matrix, use --require-run-ahead. It forces MOBILEGL_IPC_RUN_AHEAD=1 and MOBILEGL_IPC_VERB_BARRIER=1, then requires run-ahead ARMED in each private client log and rejects any running lockstep / DISARMED marker. The checkpoint records requested_arm and actual_arm; strict resume cannot reuse an earlier lockstep image pass. Start the intended server first, then run serially:

ctest --test-dir /path/to/retrace-build --show-only=json-v1 > /tmp/trace-catalog.json
python3 tools/trace_replay/run_tcp_matrix.py \
  --catalog /tmp/trace-catalog.json \
  --runner /path/to/mobilegl_trace_replay --library /path/to/libMobileGL.so \
  --endpoint tcp://192.168.21.181:40613 --token devtoken \
  --backend DirectGLES --credit 2 --require-run-ahead --out /tmp/tcp-matrix
# Continue the same frozen library/runner/peer:
python3 tools/trace_replay/run_tcp_matrix.py \
  --catalog /tmp/trace-catalog.json \
  --runner /path/to/mobilegl_trace_replay --library /path/to/libMobileGL.so \
  --endpoint tcp://192.168.21.181:40613 --token devtoken \
  --backend DirectGLES --credit 2 --require-run-ahead --out /tmp/tcp-matrix --resume

--build-dir can replace --catalog to obtain the JSON without building. --runner overrides the catalog runner; fixture paths default to the catalog's fixture directory and can be overridden with --fixtures. Repeat --case or --backend to select a subset. The default is the manifest's CI split subset (currently 39 cases), honoring each case's backend list. Explicit --case minecraft-1.21.4-rd12-odinlite-in-world permits that non-CI workload separately; an explicit split:false always remains excluded.

Both backends, one driver

--backend DirectVulkan runs P7 exit gate 3's device matrix with the same run-ahead proof, resume and watchdog the DirectGLES matrix uses. Three things are per backend and the rest is deliberately shared:

• selection - a case is planned for a backend only if its own ci_backends names it, so minecraft-1.21.4-fabric-iris-iterationrp-in-world appears in a DirectVulkan sweep (39 cases) and not in a DirectGLES one (38). A --case named explicitly that does not run a requested backend is a refusal that names the case, not one plan fewer and no message;
• golden and threshold - from backend_overrides.<backend> in trace_cases.json when the case declares one. A per-backend golden REPLACES the shared golden/alternate_golden pair for that backend rather than adding to it, and a run that matched an image another backend declares is refused by name. (alternate_golden alone cannot express this: it is an OR across backends, so a DirectVulkan run matching the DirectGLES picture passes.) The block may restate only golden, alternate_golden and ssim_threshold - restating target_call or a crop would stop the two backends retracing the same frame through the same window, which is what makes the two numbers comparable;
• result naming and checkpoint key - <backend>/credit-<n>/<case>, plus the golden, alternate golden and threshold recorded in each results.json row.

A case with no backend_overrides is resolved to itself, so DirectGLES plans, identities and checkpoint keys are unchanged by this option's existence - and the key is dropped from the fingerprint even for the backend that overrides nothing, so giving DirectVulkan its own golden does not invalidate the DirectGLES checkpoints. The CTest catalog emitter and the APK matrix refuse a case that declares backend_overrides: both carry one golden path for both backends, so emitting such a case through them would register one backend's arm against the other's image. Teach them the key in the same commit that adds an override.

The watchdog defaults to 300 seconds without log progress and a separate 7200-second absolute ceiling (--idle-seconds, --max-seconds). It observes runner output and trace/client/server log changes. It deliberately ignores the old local CTest/manifest timeout_seconds: TCP main-menu kept progressing beyond its old local 180-second budget. A growing log does not bypass the explicit absolute ceiling. On timeout the driver terminates and reaps the entire case process group, including a child that ignores TERM after CMake exits, and stops the matrix to avoid cascading Busy failures. Check the peer before resuming. Ctrl-C/TERM/HUP also clean up the group and save a cancelled checkpoint.

Each attempt has a new directory. checkpoint.json is written atomically, and only a zero exit plus current-attempt successful result, matching backend/call/ golden, adequate SSIM, an actual image, and the requested TCP arm marker count as passed. Resume also checks request/artifact identity and stored evidence hashes. Failed, timed-out, interrupted, changed, or missing evidence is rerun; old scratch results.json files are retained but never silently promoted into successful checkpoints. Replacing the remote deployment requires a new output root unless it is the same frozen peer. results.json summarizes the currently selected runs and marks reused evidence with resumed:true.

--wake-adb-serial 2f7cbe2e explicitly opts into adb shell input keyevent KEYCODE_WAKEUP every 15 seconds during a case. It does not install anything or change global power/idle settings. Its own adb log is excluded from the progress watchdog, so wakeup messages cannot conceal a stalled trace.

Tool regression tests use short process stubs, never a GPU/device/trace. They run in CI beside the other executable negative controls (test.yml, the R-16 step):

python3 tools/trace_replay/test_run_tcp_matrix.py

Device servers: offscreen or on the phone's screen

tcp_device_server.py starts one of the trace APK's two TCP servers; only one is active at a time (whichever starts last stops the other). stop force-stops the package.

# Offscreen (default): MobileGLServerService exec's the supervisor, one child per session.
python3 tools/trace_replay/tcp_device_server.py start --serial 2f7cbe2e \
  --listen tcp://127.0.0.1:40613 --token devtoken-of-16-bytes --forward
# On-screen: MobileGLDisplayActivity (process :mglwin) serves in its own process and a
# client's window surface renders on its SurfaceView. The screen must be on and unlocked.
python3 tools/trace_replay/tcp_device_server.py start --surface window [--backend DirectVulkan] \
  --serial 2f7cbe2e --listen tcp://127.0.0.1:40613 --token devtoken-of-16-bytes --forward

A headless client replays onto the phone's screen with --window-surface and MOBILEGL_IPC_SURFACE=server (plus the usual MOBILEGL_TRANSPORT=spawn, MOBILEGL_IPC_CONTROL=tcp://127.0.0.1:40613, MOBILEGL_IPC_DATA=stream, MOBILEGL_IPC_TOKEN). The server sizes its window to the trace's EGL_WIDTH/EGL_HEIGHT (letterboxed on screen) and the client's eglQuerySurface answers that size. Without --backend the on-screen server pins the first session's backend for its process; restart it (stop, then start) to switch. The same knob against the offscreen service is refused by name (NoServerDisplay) and that service goes on serving.

TCP credit and Stage measurements

For a paired frame-time breakdown, enable MOBILEGL_PIPE_STATS=1 in both roles and keep their logs. The following reducer matches client and server present serials, drops the first (unpaired) server interval and one warmup frame, and reports avg/p50/p99 in milliseconds plus percentages of client present time:

python3 tools/trace_replay/frame_stats.py \
  --client-log mobilegl.client.log --server-log mobilegl.server.log \
  --warmup-frames 1 --output frame-stats.json

transport_wait_ms measures wall time in the client producer's transport park (flush, spin and sleep). It includes waiting for replies, present credit and ring space; it is not socket latency. Client and server CPU can overlap with each other and with a transport wait, so each percentage uses the client present interval as its denominator and the three percentages need not sum to 100%. A missing frame is rejected. Percentiles use nearest rank over per-frame samples.

benchmark_tcp_credits.py runs the complete OpenRA and rd12 traces at credits 1, 2, and 3, then selects matching client/server tail frame IDs (100 and 200 frames respectively). Extract the unchanged trace files as inputs/OpenRA/openra.trace and inputs/rd12/trace.trace. Keep the server and library frozen, enable server MOBILEGL_PIPE_STATS=1 and MOBILEGL_PIPE_STATS_PERIOD=1, and leave the device connection free:

python3 tools/trace_replay/benchmark_tcp_credits.py \
  --runner /path/to/mobilegl_trace_replay --library /path/to/libMobileGL.so \
  --inputs /path/to/inputs --out /tmp/tcp-credits \
  --endpoint tcp://192.168.21.181:40613 --token devtoken --serial 2f7cbe2e
python3 tools/trace_replay/measure_tcp_stage_burst.py \
  --library /path/to/libMobileGL.so --out /tmp/tcp-stage-burst \
  --endpoint tcp://192.168.21.181:40613 --token devtoken

The credit runner serializes all six runs, sends the same explicit awake key as the matrix driver, and preserves complete benchmark and role logs. It requires the requested credit and continuous run-ahead ARMED proof in the client log, matching complete tail frames, and positive per-thread CPU observations. --case and --credit select a subset; --resume reuses only matching identity and intact evidence. FPS, kWaitReply counts, bucketed RTT distribution, Stage bytes, and the two role CPU clocks come from the selected steady window. Completing this benchmark does not perform or replace a golden image comparison.

The Stage tool uploads exactly 64 MiB after one warmup frame, consumes the dirty buffer in a draw, and waits for apply. It requires actual TCP run-ahead and at least 64 MiB of observed Stage bytes in frame 2, preserves artifact hashes, and refuses to overwrite an existing output directory. Its throughput includes client copies and backend work; it is not raw Wi-Fi bandwidth. Divide steady frame Stage traffic by this separately measured throughput only with that interpretation. Device matrices, benchmarks, and the burst must run one at a time.

Offline evidence controls (no device):

python3 tools/trace_replay/test_benchmark_tcp_credits.py