Commit Graph
4451 Commits
Author SHA1 Message Date
swung0x48 dce02ae80e Build: -Pmobilegl.androidApi=29 builds a separate API 29 library and plugin; the default stays 26. 2026-10-09 20:27:53 -04:00
swung0x48 e70925d35b Pipe: per-program global-constants key and frontend-written spans replace the client shadow diff. 2026-10-09 18:26:43 -04:00
swung0x48 e9201891f5 Docs (P15): Espryt BSL depth-copy and attachment-shadow items measured on device, no gain (x 51.4 vs b 51.9 fps, MobileGlues 55.8); not landed. 2026-10-09 18:26:42 -04:00
swung0x48 cee9178e22 Bench: bench_session stop restores the pinned clocks. 2026-10-09 18:26:41 -04:00
swung0x48 0650608ddc User guide: env fallback now read consistently. 2026-10-09 14:29:53 -04:00
swung0x48 0b536de3fb Plugin build: honour the MOBILEGL_BUILD_DISAGGREGATED env fallbacks. 2026-10-09 14:25:56 -04:00
swung0x48 e072176c24 Plugin manifest: add the relaxed-semantics title meta-data. 2026-10-09 14:24:56 -04:00
swung0x48 d4688d0037 README: document MOBILEGL_ENABLE_LTO and FRAMES_IN_FLIGHT, link the user guide. 2026-10-09 14:24:35 -04:00
swung0x48 13e8e095b4 Fix the unix:/pipe: transport comment in ConfigLoader. 2026-10-09 14:24:19 -04:00
swung0x48 286e4af582 Docs: MobileGL no longer steps aside for LIBGL_ALWAYS_SOFTWARE. 2026-10-09 14:23:56 -04:00
swung0x48 4e055e4686 Reword the permanent client-window refusal text and drop the stale P12 tag. 2026-10-09 14:23:32 -04:00
swung0x48 9f61016776 Add an end-user guide under userguide/. 2026-10-09 13:36:15 -04:00
swung0x48 9f4d9095a9 Docs (P15): windowed out-of-process smoke of the LTO plugin passes on both backends; aux-socket shared-image send error noted. 2026-10-09 12:54:30 -04:00
swung0x48 6ffd2084fb Docs (P15): open queue puts the CI fix and the user guide first, plus the clock-restore and stale surface-message follow-ups. 2026-10-09 12:54:29 -04:00
swung0x48 c0ea142955 Docs (P15): hand-off of the running m2ab job (2a/2b A/B) and the LTO server-link check. 2026-10-09 12:54:28 -04:00
swung0x48 fe748773f3 Docs (P15): hand-off for the next agents (LTO GL CPU Magma 2.52 -> 2.36, Espryt 2.99 -> 2.90; levers table; WIP branches; harness rules). 2026-10-09 12:54:27 -04:00
swung0x48 1b2e817a97 Magma: wire-store rename spares 4 -> 12 with one fence question per rename, and a 50 us quiet window on unsignaled fence polls (FCL monolith GL CPU 2.44 -> 2.38 ms, 272 -> 278 fps at cpuhunt). 2026-10-09 12:54:27 -04:00
swung0x48 c4032d2022 Bench: freqcheck judges clocks over the measurement windows and thermal signals over the whole run (BSL's idle-GPU warmup sample no longer invalidates a pinned run). 2026-10-09 12:54:25 -04:00
swung0x48 09ae776836 Docs (P15): BSL 3 reps (1c +3.4 % fps on the GPU-bound scene; Magma ~0.99 of MobileGlues, whose runs fail only an out-of-window clock sample). 2026-10-09 12:54:24 -04:00
swung0x48 2b8c04373d Gradle builds the library with LTO by default, and the LTO block adds probed ICF, gc and interposition flags per toolchain. 2026-10-09 12:54:24 -04:00
swung0x48 fcf5ae5659 Magma: an enabled scissor test over a never-written box clips nothing instead of everything (DefaultScissorFollowsSurfaceScenario runs on DirectVulkan instead of skipping). 2026-10-09 12:54:23 -04:00
swung0x48 2bbb67e528 DirectGLES: emulate a same-image glBlitFramebuffer through a scratch copy (red-once: without it ATextureBlittedWithinItselfIsSampledByTheNextDraw fails on DirectGLES and DirectGLES.Split). 2026-10-09 12:54:22 -04:00
swung0x48 db346f2844 Docs (P15): fbm2 row (pipe batch: Magma GL CPU 2.68 -> 2.62, Espryt 3.03 -> 3.00) and BSL check note. 2026-10-09 07:42:24 -04:00
swung0x48 c85c9439a5 Magma: the pipeline, program and sampler hashes read their thread_local state once per hash (GetOrCreatePipeline 0.026 -> 0.013 ms/frame). 2026-10-09 07:42:03 -04:00
swung0x48 485dff1421 Pipe: per-framebuffer state memo, object-held handle hints, per-buffer handle reuse and a cached per-verb gate (ValidateForVerb 0.328 -> 0.282 ms/frame; FCL GL CPU Magma 2.68 -> 2.62, Espryt 3.03 -> 3.00; red-once per piece in its unit case). 2026-10-09 07:42:03 -04:00
swung0x48 9ff97c2e7e CI: the control smoke test's stub ctest follows E1 to ApplierReset as well. 2026-10-09 06:52:45 -04:00
swung0x48 afcc313d9f Docs (P15): fbm row (first FB-state memo within noise, per-target entry misses on A/B/A) and handoff; BSL gpuhunt shows no GPU regression from 1c (53.8 -> 55.7 fps). 2026-10-09 06:41:45 -04:00
swung0x48 9eed5a27d7 CI: the E1 wait-boundary control follows its unit case to ApplierReset (GenerateMipmap went kWaitNone in 2fbcc224), and ScenarioIsolationScenario's four verify-split entries are listed as reaching no verb. 2026-10-09 06:41:45 -04:00
swung0x48 8941eae008 Docs (P15): 1c rows (Magma inproc +5.0 % fps), PARITY-AUDIT M1 cross-FBO deferral, Espryt attrib-format split parked. 2026-10-09 04:50:18 -04:00
swung0x48 533a256482 Magma: wire clears defer into the next pass's load op and pass-begin barriers coalesce (passes 11.5 -> 4.3, barriers 15.5 -> 9.0 per frame; inproc apply 1.41 -> 1.24 ms; red-once: no materialize fails 4 ClearOrder cases, a dropped fold fails 2). 2026-10-09 04:50:17 -04:00
swung0x48 65d4b808c1 Espryt: the surface size is memoized per session until a swap or a draw-surface change (FCL monolith CPU 3.18 -> 3.05 ms/frame; red-once: an uninvalidated memo fails DefaultScissorFollowsSurfaceScenario). 2026-10-09 04:10:20 -04:00
swung0x48 98b2c37384 Docs (P15): monolith diff vs dev and H2 rows; Magma records 12.6 passes and 27.5 barriers/frame against dev's 4.1 and 3.4. 2026-10-09 04:10:18 -04:00
swung0x48 d2acb75766 Test: UploadOrderScenario checks a texture upload lands where the app issued it (red-once: hoisting it past recorded uses fails all 4 Magma cases). 2026-10-09 04:10:17 -04:00
swung0x48 6933228d59 Docs (P15): scoreboard with the fixed dev reference and harness H2 (pinned A720, SF frame source; spread 9.3% -> 1.2-2.9%). 2026-10-09 04:10:16 -04:00
swung0x48 d786f1c072 Magma: the shader blit keeps its views and framebuffer across frames (two views + one VkFramebuffer created and retired per blit before, ~0.03 ms/frame). 2026-10-09 04:10:15 -04:00
swung0x48 79b6deacaf CI: ScenarioIsolationScenario is exempt from split coverage as a mechanism case (its pair only means something run back to back in one process). 2026-10-09 04:09:48 -04:00
swung0x48 262c8985a9 Merge the integration-test scenario isolation fix. 2026-10-08 23:10:12 -04:00
swung0x48 e5ab9d4f06 CI: barriered-validate wrapper builds without the transport, and the field-ownership self-test expects CopyTexImage2D unbarriered since P15. 2026-10-08 23:04:48 -04:00
swung0x48 129355961e [Fix] (MG_IntegrationTest): each scenario starts with no program current and no GL error pending, so a case deleting its current program no longer breaks the next one in the same process. 2026-10-08 21:54:10 -04:00
swung0x48 6cecfcd22f Docs (P15): handoff for the async default and frames in flight. 2026-10-08 19:49:55 -04:00
swung0x48 16867c5c64 Docs (P15): final wait-class verdicts. 2026-10-08 19:49:06 -04:00
swung0x48 0b96a3efbf MOBILEGL_FRAMES_IN_FLIGHT drives both backends and derives the split present credit (default 3 keeps credit 1; the Magma name stays as an alias). 2026-10-08 19:49:05 -04:00
swung0x48 2fbcc22410 Pipe: fire-and-forget is async - mipmap, copies, XFB span and storage-block rows to kWaitNone, inproc creates and PBO readbacks stop waiting, and the default dual block lets a barriered verb cost one wait (FCL BSL: ~40 sync points/frame). 2026-10-08 19:49:04 -04:00
swung0x48 1771381057 Docs (P15): wait-class audit for review and the inproc census and credit results. 2026-10-08 18:46:44 -04:00
swung0x48 2a1f1df6d1 Docs (P15): handoff notes on the inproc serialization and the anland root cause. 2026-10-08 18:02:05 -04:00
swung0x48 cc5a4c0a60 Bench session retries a slow service restart for up to 30 s before reporting it not restored. 2026-10-08 18:02:04 -04:00
swung0x48 7acf3ea8e1 Transport: MOBILEGL_IPC_WAIT_STATS logs the run-ahead client's waits per frame by record op, quiesce reason and barriered verb; the per-frame P65ClientPace line drops to debug. 2026-10-08 18:02:03 -04:00
swung0x48 db915f76f0 Transport: a client parked on the present credit is rung only by the credit, not by every retired drain batch (apply-thread futex wakes 0.056 ms/frame on FCL inproc). 2026-10-08 18:02:03 -04:00
swung0x48 657722ff57 Pipe: a buffer-only glBindVertexBuffer no longer rebuilds and compares the vertex-elements blob (VAO elements version; red-once: BindVertexBufferSwitchesAndOffsets fails on both backends). 2026-10-08 18:02:02 -04:00
swung0x48 58a97bb7b8 Bench session stops the anland app at start and relaunches it at stop, so FCL keeps the foreground. 2026-10-08 18:02:01 -04:00