Swung 0x48 swung0x48
  • Joined on 2026-07-03
swung0x48 synced commits to refs/heads/feat/disaggregated at swung0x48/MobileGL from mirror 2026-10-05 18:03:24 +09:00
946e63d810 [Fix] (MG_IntegrationTest): check pre-finish work completion in present credit tests
918d40dd4f [Fix] (Init): allow MobileGL initialization with LIBGL_ALWAYS_SOFTWARE
f5b8fe48aa [Fix] (MGPipe, MG_Remote): verify server pack state and reject records after guilty GPU resets
c5d25868cf [Fix] (ShaderTranspiler): preserve block identity when deduplicating image types
c464947bdd [Fix] (TraceReplay): share Android window surfaces and release current bindings before resize
Compare 5 commits »
swung0x48 synced commits to refs/heads/feat/disaggregated at swung0x48/MobileGL from mirror 2026-10-05 01:43:24 +09:00
880be57bb6 [CI] (Watchdog): disable GPU hang detection in all CI runs
70e4691b3a [Docs] (Anland): refresh the handoff to the 2026-10-04 state, with the perf, CI, YUV and VA-API streams
2d42414667 [CI] (Test): annotate failed ctest entries with their failure lines, since the run logs need a signed-in reader
32d27b1491 [Fix] (DirectGLES): a session's backend registration no longer trips the strict apply-thread pin while another session is applying
78c5f8dae7 [CI] (Census): record Refuse{Recover} as a local refusal; the client declines to redial and there is no peer to tell
Compare 5 commits »
swung0x48 synced commits to refs/heads/feat/disaggregated at swung0x48/MobileGL from mirror 2026-10-04 17:33:23 +09:00
3a289222af [Docs] (MobileVA): plan a vendor-agnostic VA-API driver on Android MediaCodec, H.264 first
18a5dba0e4 [Docs] (Anland): Chrome decodes video in hardware through the YUV shared images, with probes to check it
7218bacb49 [Feature] (SharedImage): sample NV12/P010 dma-bufs through GL_OES_EGL_image_external on both backends, copying foreign buffers into server YUV images
Compare 3 commits »
swung0x48 synced commits to refs/heads/feat/disaggregated at swung0x48/MobileGL from mirror 2026-10-04 09:23:24 +09:00
4633ca223c [Fix] (Server): log the per-frame P65ServerFrame line only when MOBILEGL_SERVER_FRAME_STATS asks for it
0ef13ab0a6 [Docs] (Anland): Chrome decodes video in software because MobileGL cannot import the container VA-API driver's NV12 frames.
287e1df8dc [Fix] (Server, Magma, Espryt): end the session whose GPU work hangs the GPU so it cannot stall the desktop again.
Compare 3 commits »
swung0x48 synced commits to refs/heads/feat/disaggregated at swung0x48/MobileGL from mirror 2026-10-04 01:13:24 +09:00
4f6f4125ac [Fix] (EGL, SharedImage): real EGL native fences publish a dma-buf producer's writes, so Chrome's GPU process frames reach the compositor.
2a1889b817 Discover gtest cases at ctest start so parallel links stop sharing one discovery JSON file.
07e7730085 [Docs] (Anland): kwin-recover.sh reads the latch from the server log file, not the too-small logcat buffer.
03648c2783 [Docs] (Anland): add a device test that loses KWin's own session and checks the desktop keeps compositing.
f22ae625cb [Fix] (Server): a lost session's window lease is waited for by the fresh session, and a window wait holds no backend turn.
Compare 5 commits »
swung0x48 synced commits to refs/heads/feat/disaggregated at swung0x48/MobileGL from mirror 2026-10-03 17:03:24 +09:00
fdf9252b53 [Test] (Wire): add a disabled repro for read-backs lost when threads with their own contexts tear down.
8024b645cb [Fix] (EGL): contexts recreated naming a lost share context share one new group instead of none.
9f30fc500f [Fix] (Magma): log how many SPIR-V modules the driver screen validated at renderer shutdown.
e3120af6a9 [Fix] (EGL): name a lost context's refused bind and a refused share in the client log once.
c60fcc17c5 [Fix] (Pipe): the device-loss debug knob re-arms when set to a new value.
Compare 5 commits »
swung0x48 synced commits to refs/heads/feat/disaggregated at swung0x48/MobileGL from mirror 2026-10-03 08:53:23 +09:00
6c742eba8a [Build] (Config): bump the version to 26.10
69e96f5b8d [Feature] (Client): read a system config file and decline EGL/GLX/GBM when no server answers
d24abe9287 [Fix] (Espryt): make window buffer age opt-in
d6d75c9951 [Fix] (Espryt): copy whole frames into shared images
bb4e3e0ec6 [Fix] (Espryt): query buffer age only after a damage region is declared
Compare 5 commits »
swung0x48 synced commits to refs/heads/feat/disaggregated at swung0x48/MobileGL from mirror 2026-10-03 00:43:25 +09:00
15f9b27a42 [Fix] (P14): publish a cached sampler CSO into each share group that names it
82d96a63c9 [Docs] (Anland): runbook and skill for bringing up MobileGL with anland and KDE Plasma
7520e53b5e [Feature] (SharedImage): GPU-side fences between sessions on Espryt and Magma, no CPU wait at present
b5cab44100 [Feature] (SharedImage): sync_file write/read fences per image, so sessions order their access without CPU waits
672261764e [Fix] (SharedImage): allocate and import ARGB8888/XRGB8888 images so KWin's offscreen Quick views get buffers
Compare 5 commits »
swung0x48 synced commits to refs/heads/feat/disaggregated at swung0x48/MobileGL from mirror 2026-10-02 16:33:23 +09:00
335637af38 [Docs] (Anland): record the GLX, Chrome and server-config fixes
4236b42a57 [Fix] (Server+Transpiler): configure the server once per process, and enable blend_func_extended for index-1 outputs
0b6842ac8b [Fix] (EGL+GL): let ANGLE's GL backend run on MobileGL (GBM displays, foreign wl_display, precision query, ES-free GL_VERSION)
47db68d557 [Feature] (GLX): serve as a glvnd GLX vendor and present remote X windows
ce11c9accd [Docs] (Anland): record the Chrome/dma-buf feedback fix
Compare 5 commits »
swung0x48 synced commits to refs/heads/feat/disaggregated at swung0x48/MobileGL from mirror 2026-10-02 08:23:25 +09:00
4ed1c9cef2 [Feature] (P14): state ownership - multi-session, multi-context, share groups
ebc552a048 [Fix] (WSI): report Wayland present failure and keep GLX drawables off XIDs
6fb880cdf2 [Build] (Linux): alias libGL.so and libGL.so.1 to libMobileGL
d52c031c50 [Docs] (Design): state ownership for multi-session/context/share-group (P14)
4156ca6e36 [Docs] (Anland): hand off unified runtime work and record corrected state ownership requirements
Compare 5 commits »
swung0x48 synced new reference refs/heads/feat/disaggregated-container-gbm to swung0x48/MobileGL from mirror 2026-10-02 00:13:24 +09:00
swung0x48 synced commits to refs/heads/feat/disaggregated-container-gbm at swung0x48/MobileGL from mirror 2026-10-02 00:13:24 +09:00
swung0x48 synced commits to refs/heads/feat/disaggregated at swung0x48/MobileGL from mirror 2026-10-01 07:53:24 +09:00
7308668488 [Docs] (Anland): kgsl-as-server.md - read with the kernel as the server, routes A and B are about who owns the buffer (6.x + Mesa is A, 5.x is B), plus the split-level analogy to virtio-gpu native context vs virgl / venus; linked from the README
134791b4b9 [Docs] (Anland): compare today's container Mesa (freedreno / turnip over kgsl) with routes A and B - it is route A's no-change dma-buf tier with the renderer in the app process, Anland's dma-buf ingress was shaped around it, and 5.x's private protocol sat between A and B
c24926cafe [Docs] (Anland): add route B - Anland hands the window's Surface to the MobileGL server (the FCL shape: SC mode leaves the parent Surface's queue free, waylandbridge itself receives it over binder), with its MobileGL items B1-B4, the Anland extension, its limits, and the phasing reordered around it
c5fa722be7 [Docs] (Anland): correct the window model - a toplevel attaches on demand to at most one Android window and several can be attached at once (popups and subsurfaces composite into the parent); P12 does not fit because the Android window and its input belong to Anland, not because of the window count
b7aa8d2011 [Docs] (Anland): research note - what MobileGL and Anland would each need for MobileGL to render Anland's Linux apps (glibc split client in the container, bionic render server, native AHB ingress into waylandbridge), with the two source-audit reports, their briefs and the notes index entry
Compare 5 commits »
swung0x48 synced commits to refs/heads/feat/disaggregated at swung0x48/MobileGL from mirror 2026-09-30 23:48:23 +09:00
378020adf3 [Fix] (MG_Remote, P11 T0): the T0 POST also checks what the CPU writes through the held lock after the import and its first GPU use, in both directions - Mali-G1-Ultra passed the old single round (the import cleans the CPU cache) while its GPU saw 0/1024 post-import words, so T0 was granted and create-indirect-in-world-align1024 x Magma lost every Flywheel object on inproc/spawn (0.867/0.848); it now fails there and runs T2 (0.999984), and the Redmi passes both rounds and keeps T0
028cf4eb62 [Merge] (trace): take dev's create-indirect-in-world-align1024 fixture (a derived trace replacing the original create-indirect capture) and register its documentation-only derived_from key
08124c99f1 [Test] (trace): replace the create-indirect fixture with minecraft-1.21.1-neoforge-create-indirect-in-world-align1024, a derived trace and not the original capture - the desktop capture bound Flywheel's scatter lists at 16-byte SSBO offsets that Adreno 830 and Mali's Vulkan driver (64) reject, and fixture_tools/realign_buffer_ranges.cpp moves the 1,270 binds off a 1024-byte boundary onto 1024-aligned slots (3,812 calls inserted, target call 504631 -> 508443, host SSIM unchanged to 9 digits on both backends)
Compare 3 commits »
swung0x48 synced commits to refs/heads/dev at swung0x48/MobileGL from mirror 2026-09-30 23:48:23 +09:00
08124c99f1 [Test] (trace): replace the create-indirect fixture with minecraft-1.21.1-neoforge-create-indirect-in-world-align1024, a derived trace and not the original capture - the desktop capture bound Flywheel's scatter lists at 16-byte SSBO offsets that Adreno 830 and Mali's Vulkan driver (64) reject, and fixture_tools/realign_buffer_ranges.cpp moves the 1,270 binds off a 1024-byte boundary onto 1024-aligned slots (3,812 calls inserted, target call 504631 -> 508443, host SSIM unchanged to 9 digits on both backends)
swung0x48 synced commits to refs/heads/dev at swung0x48/MobileGL from mirror 2026-09-30 23:13:24 +09:00
f973008c4d [Fix] (Espryt): a partial buffer flush no longer takes the range-invalidating map while an upload-ring copy into the same store may still be queued - Adreno 830 loses the queued copy, and create-instancing drew whole Embeddium terrain sections empty
swung0x48 synced commits to refs/heads/feat/disaggregated at swung0x48/MobileGL from mirror 2026-09-30 23:13:24 +09:00
648800a2da [Docs] (P8): close the phase - ruling ID-P8-19 (the exit device check and the P12 cross-host rerun), the P8 README and PLAN, the roadmap row and header, MEASUREMENTS row 19, the status page and the last debt
b4bee9dd38 [Docs] (P8): CONTRACT-P8 (the rules P8 landed, each with its code anchor and its test), rulings ID-P8-14..18 (MD + MF and the create-indirect fixture's alignment, SE, SV, the dev merge, the wave 2 integration gate) and the debts wave 2 hands on
494305381e [Merge] (P8): integrate wave 2 package MF - the push flush ladder gets dev's Adreno 830 queued-ring-copy guard (FlushPendingRangesFrom, create-instancing x Espryt x monolith 0.870 -> 0.99998 on the Redmi), and both G5 scripts pin the two reviewed ladder bodies under ID-P8-14
fb106a682e [Docs] (P8-MF): the package note - the push-ladder mirror of dev f973008c, create-instancing Espryt monolith 0.870 -> 0.99998 on the Redmi with spawn unchanged, why spawn was unaffected (the same two partial maps behind queued ring copies, the copy already submitted by the server's intervening whole-buffer maps), the red-onces and the gate
c8c4397475 [Merge] (P8): integrate wave 2 package SV - Magma's 3D native mip blit uses one layer, the depth/stencil mip decline gets its tally row, 1D depth mips decline by name, and the wire arm issues glDispatchComputeIndirect natively
Compare 5 commits »
swung0x48 synced commits to refs/heads/dev at swung0x48/MobileGL from mirror 2026-09-30 15:03:24 +09:00
7030d307be [Test] (Magma): UnboundUniformBlockScenario - a dispatch and a draw whose program declares a uniform block at an empty binding point (never bound, or its buffer deleted while bound) run and produce their fallback, and a buffer bound there afterwards is read, on both backends
853c5f1264 [Fix] (Magma): a uniform block declared at an empty binding point reads a zeroed block instead of dereferencing a null buffer
Compare 2 commits »
swung0x48 synced commits to refs/heads/feat/disaggregated at swung0x48/MobileGL from mirror 2026-09-30 15:03:24 +09:00
8d3e2317a6 [Docs] (P8): re-scope the phase - strike the original scope in the roadmap and the P8 README, add PLAN-P8 (six items closed, packages P8-0 and A-G, the rewritten gates) and rulings ID-P8-1..4, and mark P8 in progress on the status page
0b55b55d4d [Docs] (P11): the status page lists the decisions waiting on the user - the P8 start and the six debt rulings
642ac891af [Docs] (P11): ruling ID-P11-20 - the closing head's gate, lanes and retrace numbers
592558f5fc [Docs] (Debts): the debt review - drop the closed E1, DualBlock and GetCaps-blobref rows (open question 24 closed, the blobref rule now a CONTRACT-P5 row-18 invariant), move the unmeasured M2 note to MEASUREMENTS, rewrite HyperOS / P11 RSS / named refusals, split the P12 row, merge the two CI rows, add three B2 rows, and mark each destination as needing a ruling or ready
0a117243d5 [Docs] (P11): close the phase - rulings ID-P11-17..19 (B2 T0 zero-copy, the stale indexed binding and Magma empty-uniform fixes with the skip-hole, T0 on by default), B2-DEVICE, the P11 README, the roadmap row and header, the status page and the debts P11 hands on
Compare 5 commits »
swung0x48 synced commits to refs/heads/feat/disaggregated at swung0x48/MobileGL from mirror 2026-09-30 06:53:25 +09:00
b9b14c4775 [Docs] (P11): status page - the two handed-on dev bugs are merged, the gate row points at the B1 and M2 rulings
37c7aede31 [Docs] (P11): ruling ID-P11-16 on B1 (fd: endpoint, token broker pairing at hand-off, app_process helper, per-peer apply affinity, the HyperOS battery requirement, the device pass), the B1 debts, the status page, the phase README and the roadmap row
64b5794bc9 [Docs] (P11 B1): CONTRACT-P11 B1 - under the EOF Hello wait a second client slower than 2 s sees a send failure rather than Refuse{Busy}, and the unix-reach argument is the device's
ce4dca2835 [Docs] (P11 B1): CONTRACT-P11 B1 - the rebased head's device pass: rd12 shm vs tcp on both routes, Magma openra re-measured clean, the 10 s slow start and the plugin flavour
df826f426f [Docs] (P11 B1): CONTRACT-P11 B1 - the broker pairs at hand-off, a paired unix session waits for its Hello until EOF (the decision and its file:line), the lone-prime guard, the supervisor's announced core as the one source, and the follow-up gates
Compare 5 commits »