Compare commits

..
Author SHA1 Message Date
swung0x48 5331106e90 [CI] Use uname for Arch runner hostname 2026-10-05 08:57:17 -04:00
swung0x48 ca981a8fc1 [CI] Add Arch WSL GPU runner smoke workflow and configuration 2026-10-05 08:56:27 -04:00
swung0x48 c90383db16 [Fix] (MG_Remote): bound socket buffers in log flood tests 2026-10-05 07:13:39 -04:00
swung0x48 698f04574b [Fix] (MG_Remote): wait for records emitted while draining return events 2026-10-05 06:21:42 -04:00
swung0x48 f497ff9741 [Fix] (DirectVulkan): detach ended render passes before attachment layout reconciliation 2026-10-05 05:12:42 -04:00
swung0x48 946e63d810 [Fix] (MG_IntegrationTest): check pre-finish work completion in present credit tests 2026-10-05 04:22:18 -04:00
swung0x48 918d40dd4f [Fix] (Init): allow MobileGL initialization with LIBGL_ALWAYS_SOFTWARE 2026-10-05 02:55:35 -04:00
swung0x48 f5b8fe48aa [Fix] (MGPipe, MG_Remote): verify server pack state and reject records after guilty GPU resets 2026-10-05 02:51:19 -04:00
swung0x48 c5d25868cf [Fix] (ShaderTranspiler): preserve block identity when deduplicating image types 2026-10-05 01:27:50 -04:00
swung0x48 c464947bdd [Fix] (TraceReplay): share Android window surfaces and release current bindings before resize 2026-10-05 00:47:56 -04:00
swung0x48 4fb75f4e00 [Fix] (DirectVulkan): prepare image copy recording after uploads and share texture usage checks 2026-10-05 00:19:21 -04:00
swung0x48 4d6904184d [Fix] (ShaderTranspiler): deduplicate image types after Vulkan image rewrites 2026-10-05 00:08:32 -04:00
swung0x48 2f1a673513 [Fix] (BackendLoader, DirectGLES): check EGL extensions before resolving damage swap entry points 2026-10-05 00:08:31 -04:00
swung0x48 88bc35b711 [Refactor] (BackendLoader): share extension name matching between GLES and EGL probes 2026-10-04 20:54:28 -04:00
swung0x48 880be57bb6 [CI] (Watchdog): disable GPU hang detection in all CI runs 2026-10-04 10:39:39 -04:00
swung0x48 70e4691b3a [Docs] (Anland): refresh the handoff to the 2026-10-04 state, with the perf, CI, YUV and VA-API streams 2026-10-04 07:25:23 -04:00
Swung0x48 2d42414667 [CI] (Test): annotate failed ctest entries with their failure lines, since the run logs need a signed-in reader 2026-10-04 07:20:04 -04:00
Swung0x48 32d27b1491 [Fix] (DirectGLES): a session's backend registration no longer trips the strict apply-thread pin while another session is applying 2026-10-04 07:20:04 -04:00
Swung0x48 78c5f8dae7 [CI] (Census): record Refuse{Recover} as a local refusal; the client declines to redial and there is no peer to tell 2026-10-04 07:20:04 -04:00
swung0x48 96bd540ba1 [Docs] (Performance): record the device state at the glmark2 handoff 2026-10-04 06:46:51 -04:00
swung0x48 710beefd02 [Docs] (Performance): glmark2 optimization handoff with the landed fixes, current numbers, remaining work and the bench scripts 2026-10-04 06:44:49 -04:00
Swung0x48 c1e957cb98 [CI] (Gates): retire the P3a/P4a untouched-region gates; those phases closed and later phases legitimately changed the regions 2026-10-04 06:30:24 -04:00
Swung0x48 eb71b669b3 [Fix] (Build): make the pull and verify builds compile again 2026-10-04 06:30:24 -04:00
swung0x48 bb31f374a3 [Perf] (DirectVulkan): keep the wire draw's render pass open across draws to the same attachments, closing it only for work that cannot sit inside one 2026-10-04 06:18:35 -04:00
swung0x48 e5ee31b022 [Build] (Anland): build the anland server for android-30, the APK's minSdk, so thread_local uses native TLS 2026-10-04 06:02:45 -04:00
swung0x48 915882b293 [Perf] (Server): time records only when the per-frame statistics are on 2026-10-04 06:02:44 -04:00
swung0x48 963f4d8177 [Perf] (DirectVulkan): reuse completed command buffers instead of freeing and allocating one per submission 2026-10-04 06:02:43 -04:00
swung0x48 d627fca387 [Perf] (Transport): the server rings a parked client only once the sequence it waits for is applied 2026-10-04 05:40:14 -04:00
swung0x48 2de0d6dd25 [Perf] (Pipe): reuse client-array buffers instead of a blocking create per draw, and copy contiguous ranges in one go 2026-10-04 04:54:54 -04:00
swung0x48 66b47a101b [Fix] (Remote): glFinish waits for the GPU, Magma bounds unsubmitted clear objects, ES contexts advertise OES_mapbuffer/rgb8_rgba8/depth24 2026-10-04 04:54:53 -04:00
swung0x48 4722fa38bb [Docs] (Performance): glmark2 baseline of Espryt and Magma against the stock kgsl stack on anland, with profiles and ranked bottlenecks 2026-10-04 04:33:24 -04:00
swung0x48 3a289222af [Docs] (MobileVA): plan a vendor-agnostic VA-API driver on Android MediaCodec, H.264 first 2026-10-03 23:46:13 -04:00
swung0x48 18a5dba0e4 [Docs] (Anland): Chrome decodes video in hardware through the YUV shared images, with probes to check it 2026-10-03 22:46:01 -04:00
swung0x48 7218bacb49 [Feature] (SharedImage): sample NV12/P010 dma-bufs through GL_OES_EGL_image_external on both backends, copying foreign buffers into server YUV images 2026-10-03 22:46:00 -04:00
swung0x48 4633ca223c [Fix] (Server): log the per-frame P65ServerFrame line only when MOBILEGL_SERVER_FRAME_STATS asks for it 2026-10-03 14:20:11 -04:00
swung0x48 0ef13ab0a6 [Docs] (Anland): Chrome decodes video in software because MobileGL cannot import the container VA-API driver's NV12 frames. 2026-10-03 14:16:04 -04:00
swung0x48 287e1df8dc [Fix] (Server, Magma, Espryt): end the session whose GPU work hangs the GPU so it cannot stall the desktop again. 2026-10-03 12:28:39 -04:00
swung0x48 4f6f4125ac [Fix] (EGL, SharedImage): real EGL native fences publish a dma-buf producer's writes, so Chrome's GPU process frames reach the compositor. 2026-10-03 10:59:02 -04:00
swung0x48 2a1889b817 Discover gtest cases at ctest start so parallel links stop sharing one discovery JSON file. 2026-10-03 09:38:51 -04:00
swung0x48 07e7730085 [Docs] (Anland): kwin-recover.sh reads the latch from the server log file, not the too-small logcat buffer. 2026-10-03 09:05:55 -04:00
swung0x48 03648c2783 [Docs] (Anland): add a device test that loses KWin's own session and checks the desktop keeps compositing. 2026-10-03 09:05:54 -04:00
swung0x48 f22ae625cb [Fix] (Server): a lost session's window lease is waited for by the fresh session, and a window wait holds no backend turn. 2026-10-03 09:05:54 -04:00
swung0x48 f52a546707 [Fix] (EGL, Remote): a server-owned surface is re-created at the size it asked for, and a lost session's image reference is never released on the fresh one. 2026-10-03 09:05:53 -04:00
swung0x48 77c604061e [Docs] (Anland): glxgears is fixed-function, so it measures presentation only 2026-10-03 07:53:11 -04:00
swung0x48 36190b7cc0 [Fix] (EGL, Wayland): a swap that runs into a device loss reports it and still commits the window, instead of showing an empty frame. 2026-10-03 07:17:08 -04:00
swung0x48 cf763b4081 [Fix] (EGL, Remote): a thread that released a lost context keeps reporting the reset until it binds another. 2026-10-03 07:17:07 -04:00
swung0x48 2acc6c41bb [Fix] (Pipe): an evicted sampler CSO is deleted from every share group of its session. 2026-10-03 05:37:53 -04:00
swung0x48 a84104a40f [Fix] (Pipe): the default framebuffer's record names no placeholder surface that only one share group holds. 2026-10-03 05:37:53 -04:00
swung0x48 1dd888cb21 [Fix] (EGL, DirectGLES): keep a lost context's release off the fresh session, bind no surfaceless context natively, and keep the display while surfaces remain. 2026-10-03 05:37:52 -04:00
swung0x48 393002a46e [Fix] (Pipe): give each context's applier its own serial range so backend memos never match another context. 2026-10-03 05:37:51 -04:00
swung0x48 25a76c1b20 [Fix] (Pipe): drain a texture's pending texels only under a context of the share group that wrote them. 2026-10-03 05:37:51 -04:00
swung0x48 a9f22ebb63 [Fix] (EGL): create and end a context's objects under its own binding and the stream lock, and defer destroying a current context. 2026-10-03 05:37:50 -04:00
swung0x48 2fe011adf6 [Test] (Wire): enable the concurrent-context repro and add release, sibling, deferred-destroy and read-back-race shapes. 2026-10-03 05:37:49 -04:00
swung0x48 11668b02f2 [Docs] (Anland): Chrome runs its GPU in its own process by default and recovers from a device loss. 2026-10-03 05:21:53 -04:00
swung0x48 30977c181f [Feature] (EGL+GL): an opt-in OpenGL ES identity for ES contexts with BGRA textures, and contexts take the dispatcher's API, so Chrome's GPU process runs on ANGLE. 2026-10-03 05:21:52 -04:00
swung0x48 e5f097f9e1 [Docs] (Anland): add a Chrome device-loss recovery test and the GPU-process launch mode. 2026-10-03 04:37:34 -04:00
swung0x48 90e63c9e2a [Feature] (EGL+GL): advertise EGL_EXT_create_context_robustness and GL robustness, with per-context reset notification and a pid-targeted device-loss knob. 2026-10-03 04:37:33 -04:00
swung0x48 af2b9f4c1c [Docs] (Anland): root cause of the X11 stale-region issue 2026-10-03 04:36:24 -04:00
swung0x48 cd58f048b6 [Fix] (Shader): an inactive fragment output no longer takes location 0 from the active one 2026-10-03 04:16:41 -04:00
swung0x48 b0dad54cf1 [Docs] (Anland): device numbers and the open stale-region issue for X11 on the GPU 2026-10-03 03:20:12 -04:00
swung0x48 95aa618d22 [Docs] (Anland): x11-verify measures 1280x720 glxgears and runs Qt apps on xcb 2026-10-03 02:58:03 -04:00
swung0x48 5de6ecd4e1 [Fix] (SharedImage): uploads into a texture bound to a shared image are written into the image, and a producer's flush sends them first 2026-10-03 02:58:02 -04:00
swung0x48 7fd965ceda [Fix] (SharedImage): a session that writes images publishes every frame boundary as a write 2026-10-03 01:51:25 -04:00
swung0x48 9e7305b885 [Docs] (Anland): plan and device scripts for X11 apps on the GPU 2026-10-03 01:51:24 -04:00
swung0x48 bd4c5b5770 [Test] (Gbm): drive mobilegl_gbm through the system loader the way Xwayland's glamor does 2026-10-03 01:51:23 -04:00
swung0x48 198d6feaf3 [Feature] (GLX): present windows as shared images through DRI3 and Present, with MIT-SHM readback as the fallback 2026-10-03 01:51:22 -04:00
swung0x48 08f23543aa [Feature] (SharedImage): a glFlush publishes an image producer's accesses, so Xwayland's glamor gets implicit sync 2026-10-03 01:51:20 -04:00
swung0x48 316ffe02d0 [Fix] (X11): never open an X connection to a display this process serves 2026-10-03 01:50:58 -04:00
swung0x48 fdf9252b53 [Test] (Wire): add a disabled repro for read-backs lost when threads with their own contexts tear down. 2026-10-03 00:42:35 -04:00
swung0x48 8024b645cb [Fix] (EGL): contexts recreated naming a lost share context share one new group instead of none. 2026-10-03 00:42:34 -04:00
swung0x48 9f30fc500f [Fix] (Magma): log how many SPIR-V modules the driver screen validated at renderer shutdown. 2026-10-03 00:10:58 -04:00
swung0x48 e3120af6a9 [Fix] (EGL): name a lost context's refused bind and a refused share in the client log once. 2026-10-03 00:08:49 -04:00
swung0x48 c60fcc17c5 [Fix] (Pipe): the device-loss debug knob re-arms when set to a new value. 2026-10-02 23:59:55 -04:00
swung0x48 b06c2a9bf2 [Fix] (Remote): a surfaceless make-current's hidden pbuffer no longer latches the session offscreen. 2026-10-02 23:59:54 -04:00
swung0x48 8b5295401f [Docs] (Anland): reproduce.md brings the desktop up by opening the app. 2026-10-02 23:41:20 -04:00
swung0x48 22266c47cf Merge client-recover: open a fresh server session after a device loss. 2026-10-02 23:38:56 -04:00
swung0x48 4a4e53d9a9 Merge magma-validate: Magma SPIR-V validation declines and surfaceless EGL. 2026-10-02 23:38:19 -04:00
swung0x48 b7cc56dd44 [Docs] (Anland): record the app-driven desktop's timings, idle numbers and frame pacing 2026-10-02 23:36:47 -04:00
swung0x48 7823bcd987 [Fix] (Client): recover a fresh server session after a device loss so applications can recreate their contexts 2026-10-02 23:34:21 -04:00
swung0x48 caa00995e7 [Fix] (EGL): slow an unpaced Wayland window to one frame a second while the compositor shows none of it 2026-10-02 23:26:03 -04:00
swung0x48 1ff7ca5331 [Fix] (EGL): pace Wayland window swaps by frame callbacks so clients idle while the compositor shows nothing 2026-10-02 23:10:33 -04:00
swung0x48 6125f941d1 [Fix] (EGL): accept surfaceless make-current and config-less contexts so eglinfo prints the GL renderer lines 2026-10-02 23:09:59 -04:00
swung0x48 bfa7987ca6 [Fix] (Magma): decline programs whose final SPIR-V fails validation instead of handing the module to the driver 2026-10-02 23:09:58 -04:00
swung0x48 b1405e17ad [Docs] (Anland): the desktop starts from the app and idles while hidden, with no manual bring-up 2026-10-02 22:46:04 -04:00
swung0x48 b64c2de3c8 [Docs] (Anland): point the reproduce guide at the anland fork 2026-10-02 22:26:24 -04:00
swung0x48 2b95a9697a [Docs] (Anland): reproduce the MobileGL Plasma container setup from source 2026-10-02 22:26:11 -04:00
swung0x48 228bd24c67 [Fix] (Espryt): turn window buffer age on by default 2026-10-02 21:00:20 -04:00
swung0x48 c822a9fef7 [Fix] (Espryt): copy only the damaged rectangles into shared images again 2026-10-02 21:00:10 -04:00
swung0x48 815217cc07 [Fix] (Espryt): turn the scissor off around backend copies by what the driver has, not another session's render-state shadow 2026-10-02 20:57:18 -04:00
swung0x48 7a5b0755b3 [Docs] (Anland): install MobileGL as the container's system-wide vendor and document the fallback 2026-10-02 20:13:33 -04:00
swung0x48 3d69c37504 [Fix] (EGL): keep a process the vendor does not serve from opening its log 2026-10-02 20:07:15 -04:00
swung0x48 4f679d669d [Fix] (EGL): answer client queries and proc lookups without bringing a session up 2026-10-02 20:03:45 -04:00
swung0x48 4095624f87 [Fix] (Client): step aside for processes that ask for software rendering 2026-10-02 19:57:49 -04:00
swung0x48 321e61ca0b [Docs] (README): the target is OpenGL 4.6 Core, with the current modules and platforms 2026-10-02 19:53:29 -04:00
swung0x48 552dd3d39b [Fix] (EGL): decline another backend's GBM device and ask X for a visual only on X displays 2026-10-02 19:49:32 -04:00
swung0x48 6c742eba8a [Build] (Config): bump the version to 26.10 2026-10-02 19:43:47 -04:00
swung0x48 69e96f5b8d [Feature] (Client): read a system config file and decline EGL/GLX/GBM when no server answers 2026-10-02 19:36:38 -04:00
swung0x48 d24abe9287 [Fix] (Espryt): make window buffer age opt-in 2026-10-02 18:34:23 -04:00
swung0x48 d6d75c9951 [Fix] (Espryt): copy whole frames into shared images 2026-10-02 17:38:55 -04:00
swung0x48 bb4e3e0ec6 [Fix] (Espryt): query buffer age only after a damage region is declared 2026-10-02 17:11:31 -04:00
swung0x48 233e2f8ced [Fix] (Espryt): swap a pbuffer without damage rectangles 2026-10-02 16:27:16 -04:00
swung0x48 74919a3bc1 [Feature] (Present): buffer age and swap damage end to end on Espryt and Magma 2026-10-02 15:19:39 -04:00
swung0x48 7090e31dee [Feature] (Damage): damage-region and buffer-age bookkeeping with host tests 2026-10-02 14:51:48 -04:00
swung0x48 5accb0985c [Fix] (glslang): keep samplers of anonymous uniform structs out of the default uniform block, which made Adreno page-fault into device loss on Shadertoy 2026-10-02 14:20:36 -04:00
swung0x48 58685bc116 [Fix] (Server): wait out an apply thread that is busy in a GPU stall at Stop() instead of aborting every session 2026-10-02 13:54:26 -04:00
swung0x48 34090097a1 [Fix] (Client): make a readback on a lost device a no-op instead of a Fatal that kills the client 2026-10-02 13:43:51 -04:00
swung0x48 58c3d9bde5 [Fix] (P14): latch a lost GPU device per session on Magma and Espryt instead of aborting the whole server 2026-10-02 13:37:58 -04:00
swung0x48 d03fe4c9dd [Fix] (Client): drop the late reply of a timed-out surface op instead of reading it as the next op's 2026-10-02 13:34:17 -04:00
swung0x48 239fc228fc [Fix] (DirectVulkan): retype gl_FragCoord access chains to Private when the y-flip pass redirects them to its copy 2026-10-02 13:34:16 -04:00
swung0x48 8da082f805 [Fix] (State): start the clear colour at GL's (0, 0, 0, 0), not opaque black 2026-10-02 12:50:36 -04:00
swung0x48 258b946086 [Fix] (P12): publish the server window's extent from the display on every resize, at once, and let eglQuerySurface take it in 2026-10-02 12:03:52 -04:00
swung0x48 16b07cf4b5 [Fix] (DirectVulkan): build offscreen targets of a window-created renderer through AImageReader instead of a null headless entry point 2026-10-02 11:42:14 -04:00
swung0x48 51c37bcbc9 [Fix] (Sampler): convert sampler parameters set through the other scalar type 2026-10-02 11:22:00 -04:00
swung0x48 5a7b23f661 [Feature] (P12): suspend the on-screen session when the server window goes, resume it on the next, and republish a resized window's extent 2026-10-02 11:14:27 -04:00
swung0x48 15f9b27a42 [Fix] (P14): publish a cached sampler CSO into each share group that names it 2026-10-02 10:35:44 -04:00
swung0x48 82d96a63c9 [Docs] (Anland): runbook and skill for bringing up MobileGL with anland and KDE Plasma 2026-10-02 10:31:35 -04:00
swung0x48 7520e53b5e [Feature] (SharedImage): GPU-side fences between sessions on Espryt and Magma, no CPU wait at present 2026-10-02 10:31:34 -04:00
swung0x48 b5cab44100 [Feature] (SharedImage): sync_file write/read fences per image, so sessions order their access without CPU waits 2026-10-02 09:40:12 -04:00
swung0x48 672261764e [Fix] (SharedImage): allocate and import ARGB8888/XRGB8888 images so KWin's offscreen Quick views get buffers 2026-10-02 09:37:17 -04:00
swung0x48 0e314f645e Remove a stray build log 2026-10-02 09:18:20 -04:00
swung0x48 350fc699a1 [Fix] (SharedImage): create dma-buf wl_buffers without create_immed, drain the ring before teardown ops, and decline Magma records after the renderer is gone 2026-10-02 09:18:05 -04:00
swung0x48 3d17e8c10b [Feature] (SharedImage): zero-copy Wayland windows via linux-dmabuf, EGL dma-buf import, GBM backend, and Espryt/Magma AHB present and texture binding 2026-10-02 08:39:59 -04:00
swung0x48 38842985f8 [Feature] (SharedImage): server-allocated AHB images over a shared_image pipe verb, with dma-buf fds on the aux socket 2026-10-02 07:52:25 -04:00
swung0x48 e06aabe812 [Docs] (Anland): plan for zero-copy windows via server-allocated AHB dma-bufs 2026-10-02 07:34:45 -04:00
swung0x48 67a97065b8 [Fix] (DirectVulkan): transpose quarter-turn viewports and scissors, rebuild a resized pbuffer's target, and leave out-of-range default-framebuffer reads unfilled 2026-10-02 07:15:47 -04:00
swung0x48 ec084959fa [Docs] (Anland): record the Magma bring-up to Espryt parity 2026-10-02 06:36:20 -04:00
swung0x48 4b388950e3 [Fix] (DirectVulkan): decline draws from storeless buffers and sample with defaults when a texture's sampler CSO is gone, as DirectGLES does 2026-10-02 06:33:53 -04:00
swung0x48 7e76a98d36 [Fix] (DirectVulkan): keep one renderer per session across its surfaces, follow bind_context to each context's surface, and keep the wire resource ops registered while any session lives 2026-10-02 06:24:13 -04:00
swung0x48 15dbe78ce6 [Fix] (DirectVulkan): give each served session its own renderer and Magma state 2026-10-02 05:52:38 -04:00
swung0x48 168808d1a7 [Fix] (DirectVulkan): advertise the GLSL 1.00-era ARB strings KWin gates on, and let a debug property pick the Android backend 2026-10-02 05:42:14 -04:00
swung0x48 75e1da4d1b [Docs] (Anland): record the lock-screen corruption root cause 2026-10-02 05:07:59 -04:00
swung0x48 cd610934d7 [Fix] (DirectGLES): scope object deaths to their session, and keep container objects and staging rings per native context 2026-10-02 05:07:59 -04:00
319 changed files with 32249 additions and 1319 deletions
@@ -0,0 +1,3 @@
# Everything here is executed by bash/sh on Linux, Android or WSL: keep LF even on a Windows checkout
# (core.autocrlf=true would otherwise hand sh a script full of $'\r').
* text=auto eol=lf
@@ -0,0 +1,183 @@
---
name: anland-mobilegl-plasma
description: Bring up, rebuild, deploy, switch backends on, test and debug the "MobileGL + anland + KDE Plasma" stack - a rooted Android tablet whose anland APK embeds the MobileGL server, plus a Droidspaces Arch Linux ARM container running KWin/plasmashell/Chrome through the MobileGL client library and GBM backend - driven from this Windows machine (Git Bash, WSL archlinux, adb). Use it after a phone reboot, when changing server (NDK) or client (container glibc) MobileGL code or anland's KWin backend/patch, when switching Espryt/Magma, when validating zero-copy dma-buf presentation, and when chasing crashes, black windows, noise or software-rendering fallbacks in the Plasma session.
---
# MobileGL on anland KDE Plasma
**Setting this up on a new machine or phone from source: `reproduce.md` next to this file.**
Human-readable runbook (Chinese, same content): `docs/Disaggregated/notes/anland/runbook-plasma.md`.
Background: `docs/Disaggregated/notes/anland/handoff-legacy-mobilegl-unified.md` (state, commits, known
issues) and `plan-ahb-dmabuf.md` (zero-copy design). All scripts are in `scripts/` next to this file;
parameters are documented at the top of each script and in `scripts/env.sh`.
**The phone may be shared with another session.** Before anything that restarts the app, the daemon,
the container session or replaces a library, run the read-only `scripts/device/status.sh` and make sure
nobody else is mid-test. Use your own `ct.sh` channel names (others use c, c2, c3, xd, xr, gbm).
## Architecture (10 lines)
1. Phone: Lenovo TB321FU (Y700, Adreno 750, Android 15, kernel 6.1), rooted with **KernelSU** (`/data/adb/ksu`, `ksud`; `su -c` works). USB serial `HA27Q3LQ`, or wireless adb `ip:port`.
2. anland APK `com.anland.consumer.mobilegl` (debuggable build of anland branch `legacy-mobilegl-unified`): UI process + private service process `:mobilegl` (`MobileGLWorker`), which loads the **server** `libMobileGL.so` (NDK, from MobileGL `feat/disaggregated`) and listens on abstract socket `@anland-mobilegl` from the first attached window on. One process serves every client (KWin, plasmashell, ksplash, Chrome...), one backend per process. `MobileGLWorker` is a **foreground service** (notification "Linux desktop is running" / action "Stop desktop"), so the server and the session outlive a hidden, locked or swiped-away window.
3. Backend = the app setting "Renderer backend" (Settings > Connection > MobileGL desktop; file `/data/data/<pkg>/files/mobilegl-backend`, default DirectGLES; also `am start ... --es mobilegl_backend X`). Espryt = DirectGLES, Magma = DirectVulkan. The server reads it when its process starts; the app publishes it as `/data/local/tmp/anland-mobilegl/backend` and the session copies that into the container's `/etc/mobilegl/backend` (what every client reports; mismatch = "Hello.backendType disagrees with pinned backend"). `debug.mobilegl.backend`, if someone sets it, still overrides the setting until the next reboot.
4. Experiment display daemon `/data/adb/modules/anland-daemon/display_daemon /data/local/tmp/anland-mobilegl/display.sock` (display/input broker), started by the app (below). The module only starts the original one on `/data/local/tmp/display_daemon.sock` at boot; leave that alone.
5. Container `arch-kde-mgl` (Droidspaces 6.4.5, Arch Linux ARM aarch64 glibc, host net, `rootfs.img` mounted at `/mnt/Droidspaces/arch-kde-mgl` on the Android side). Bind mounts: `/data/local/tmp/anland-mobilegl` -> `/run/anland-mobilegl`, `/data/local/tmp/display_daemon.sock` -> `/run/display.sock`.
6. `desktop-session.service` (User=swung0x48, **enabled**: the container boots straight into it, as the original image does) has drop-in `/etc/systemd/system/desktop-session.service.d/mobilegl.conf` (anland `producers/kde/Arch_v5/desktop-session-mobilegl.conf`, installed + enabled by `sync-build-kwin.sh`): `ExecStartPre=mobilegl-startup.sh wait` (until `@anland-mobilegl` listens in `/proc/net/unix` and the daemon socket exists), `ExecStartPre=+... sync-backend`, `Restart=always`, then `ExecStart=/opt/mobilegl/bin/mobilegl-startup.sh plasma`, which deletes the sticky software-renderer key, sets `MOBILEGL_IPC_PROBE_TIMEOUT_MS=0` session-wide (MobileGL never declines, so glvnd never falls through to another GL stack), writes a user drop-in for `plasma-kwin_wayland.service` (kwin_wayland_wrapper -> `/opt/mobilegl/kwin/bin/kwin_wayland` + `/opt/mobilegl/kwin/lib/libkwin.so.6.7.4`, `ANLAND_MOBILEGL=1`, `MOBILEGL_IPC_SURFACE=server`) and runs `startplasma-wayland`.
7. KWin 6.7.4 with anland's `anland` backend renders straight into the anland Android Surface through the server. While the window is not on screen (consumer disconnected) the backend powers its output down (DPMS Off + RenderLoop inhibited: no compositing, no frame callbacks) and back up when the window returns. Every other GL process is `MOBILEGL_IPC_SURFACE=offscreen` and presents through KWin.
8. Client = `/opt/mobilegl/lib/libMobileGL.so` (glibc aarch64), the container's **system-wide** GL vendor (section below): glvnd EGL vendor `/usr/share/glvnd/egl_vendor.d/10_mobilegl.json` (sorts before the distribution's `50_*.json`, which serves whenever MobileGL declines), GLX vendor `/usr/lib/libGLX_mobilegl.so.0`, settings from `/etc/mobilegl/client.conf` (`MOBILEGL_TRANSPORT=spawn MOBILEGL_IPC_DATA=shm MOBILEGL_IPC_CONTROL=unix:@anland-mobilegl`) + `/etc/mobilegl/backend`. No per-process environment is needed.
9. Zero-copy windows: client `wl_egl_window` frames are server-allocated AHardwareBuffer "shared images" exported as dma-buf fds over `zwp_linux_dmabuf_v1`; KWin imports them with EGL and the server recognises its own buffer by inode. KWin gets a DRM device via `GBM_BACKEND=mobilegl` -> `/usr/lib/gbm/mobilegl_gbm.so` (node is only an identity, never ioctl'd). Without linux-dmabuf the client falls back to wl_shm readback.
10. X11 (`docs/Disaggregated/notes/anland/plan-x11-gpu.md`): Xwayland draws with **glamor on MobileGL** (an offscreen session; `GBM_BACKEND=mobilegl` on KWin's render node; window pixmaps are shared images handed to KWin as linux-dmabuf buffers; its `glFlush` publishes them through the server, `/tmp/mgl-xwayland.client.log`); `ANLAND_XWAYLAND_GLAMOR=0` in KWin's drop-in brings back the software `-shm` Xwayland. GLX windows present their frames as shared images through **DRI3 + Present** (log `presented through DRI3+Present shared images`), falling back to readback + MIT-SHM / `xcb_put_image` when the X server has no DRI3 (`MOBILEGL_GLX_PRESENT=readback` forces it per app). Chrome runs ANGLE-GLES over MobileGL EGL with the GPU in its own process (`mobilegl-startup.sh chrome`; `MOBILEGL_CHROME_GPU=in-process` opts out), presenting GBM dma-bufs over linux-dmabuf; after a device loss Chrome restarts that process and repaints.
## First-time prerequisites (Windows host)
- Git Bash, `adb` (SDK platform-tools), CMake >= 3.22 and Ninja on PATH (seen: CMake 3.27.6, Ninja 1.13.2), Python 3 + Pillow (only `rec.sh`).
- Android NDK **27.2.12479018** at `%LOCALAPPDATA%/Android/Sdk/ndk/27.2.12479018` (override `ANDROID_NDK`).
- WSL distro **archlinux** with clang, lld, llvm, cmake, ninja. **lld's major version must equal clang/llvm-libs'** (lld 23 on llvm 22 broke `ld.lld`; working set: all 22.1.6). Then export the sysroot once: `bash scripts/xsysroot/refresh-sysroot.sh` (-> `~/sysroots/arch-kde-mgl` + `~/sysroots/aarch64-arch.cmake`; re-run after container package upgrades, then `xbuild.sh --reconfigure`).
- MobileGL worktree (never build in the shared main checkout): `GIT_LFS_SKIP_SMUDGE=1 git worktree add ...`, `git submodule update --init --recursive`, and copy `3rdparty/glslang/External/spirv-tools` from the main checkout (the submodule leaves it empty).
- anland worktree on `legacy-mobilegl-unified` (KWin patch/backend, startup script, `mobilegl-tools/`).
- Phone side that already exists and is NOT recreated by this skill: the APK (build: anland `consumers/anland_v5/android_consumer/MOBILEGL.md`, gradle `:app:assemblePlainDebug` with `MOBILEGL_DIST`, `-PmobileglApplicationId=com.anland.consumer.mobilegl -PanlandDefaultSocket=/data/local/tmp/anland-mobilegl/display.sock`, from Git Bash with `MSYS_NO_PATHCONV=1`; signed with this machine's `~/.android/debug.keystore`, so `adb install -r` works; reinstalling with another key changes the UID and needs the KernelSU grant again - `mobilegl-tools/ksu_grant.c`), the anland KernelSU modules (`anland-daemon`, `anland-awl`), the container with KWin's configured build tree `/root/mobilegl-build/kwin-6.7.4/build` and pristine `/root/gbm-dev/kwin-a/src`.
- Install the helpers once (container must be running): `bash scripts/push-tools.sh` (device/*.sh -> `/data/local/tmp/anl`, `mgrun` -> container `/usr/local/bin`, `kwprep.sh` -> container `/root`).
Run host scripts from Git Bash as `bash scripts/<x>.sh`. Raw adb with device paths needs `MSYS_NO_PATHCONV=1` and the whole root command in one quoted string: `adb shell 'su -c "..."'`. A device script can also be run without pushing: `MSYS_NO_PATHCONV=1 adb shell 'su -c sh' < scripts/device/status.sh`.
## Starting, idling, stopping - no manual steps
Nothing starts at phone boot (container `run_at_boot=0`). **Opening the app** (launcher icon = `am start -n com.anland.consumer.mobilegl/com.anland.consumer.MainActivity`, no extras; the APK's default socket is the experiment's, `-PanlandDefaultSocket`) calls `MobileGLDesktop.ensureStarted()` - the single trigger - which starts the foreground service and runs the APK's `assets/mobilegl-desktop.sh up` as root (`su`): display daemon if absent (the script leaves the app's cgroup first, so it survives the app) -> backend published -> container started if stopped -> `systemctl start --no-block desktop-session`. The window waits for the daemon socket; the server listens once the Surface is attached; the session's `wait` step then lets Plasma start. A reopen with everything up only reattaches the window. Log: `/data/local/tmp/anland-mobilegl/desktop.log`, logcat `AnlandMobileGL` ("MobileGL desktop up done in N ms").
Hidden window, screen off, lock screen or swiped-away task: server and session stay (foreground service); KWin turns the workspace's DPMS off (logs "viewer gone: output powered down"; windows get xdg_toplevel `suspended`, no frame callbacks) and back on ("viewer back"). Client swaps are paced by frame callbacks (`MG_Impl/EGLImpl/FrameThrottle.h`): interval >= 1 waits like any Wayland EGL; interval 0 (Chrome) drops to ~1 fps once a callback goes unanswered for 200 ms.
Measured 2026-10-03 (Espryt, Chrome on webglsamples aquarium; % of one core, phone = all of Android): open-to-desktop cold 7.5-12.5 s (daemon 0.5, server 1.0, container 1.5, kwin ~5, plasmashell ~6-7), after a phone reboot 8.5 s, reopen of a running desktop 0.2 s; visible phone 249% / kwin 13 / chrome 99 / server 57 / GPU 22%, hidden 54 / 0 / 6 / 2 / 0%, locked 38 / 0 / 8 / 3 / 0%. Stop desktop 1.7 s. Wireless adb does not come back by itself after a reboot (re-enable it on the phone). **Stop desktop** (notification action, or `am startservice -n <pkg>/com.anland.consumer.MobileGLWorker -a com.anland.consumer.mobilegl.STOP`) runs `mobilegl-desktop.sh down` (session + container stopped, daemon stays), closes the windows, ends `:mobilegl`.
```sh
adb connect <ip>:<port> # only for wireless debugging
export MSYS_NO_PATHCONV=1
adb shell 'su -c "sh /data/local/tmp/anl/bringup.sh"' # after a reboot: wake, dismiss keyguard, open the app, print the timeline
bash scripts/desktop-verify.sh push cold reopen idle # timed cold start, swipe-away + reopen, CPU/GPU visible/hidden/locked
adb shell 'su -c "sh /data/local/tmp/anl/status.sh"'
```
`droidspaces` has no `status` command; `droidspaces show` lists running containers. `device/run-plasma.sh` is a developer restart (session + logind user, then the app), needed after a rebuild because KWin does not reconnect to a new server process. `device/desktop-reset.sh` takes everything down to the post-reboot state. The old manual sequence (setprop, start container, `restart-daemon.sh`, `run-plasma.sh`) is gone; `restart-daemon.sh` remains for daemon debugging only.
## Build and deploy loops
**Server (Android, NDK) -> APK lib**
```sh
bash scripts/build-android.sh # -> <worktree>/build-android/libMobileGL.so (RelWithDebInfo, unstripped)
bash scripts/cyc.sh s1 # clear sticky SW-renderer key, push, swap into APK, restart session, screenshot $ANL_DIR/s1.png, print fatals
```
`device/swap.sh` overwrites `/data/app/.../lib/arm64/libMobileGL.so` keeping owner/mode/SELinux context (original saved as `/data/local/tmp/anland-mobilegl/libMobileGL.so.orig`); no re-signing. To swap without the session restart: push to `/data/local/tmp/anland-mobilegl/libMobileGL.new.so`, `su -c "sh /data/local/tmp/anl/swap.sh"`, then `run-plasma.sh` anyway (KWin does not reconnect to a new server process).
**Client + GBM backend (container glibc) via WSL cross-build**
```sh
bash scripts/xbuild.sh [--reconfigure] # WSL ~/mgl-xbuild-a64; ~150 s full, seconds incremental
bash scripts/xdeploy.sh # rename-into-place install of libMobileGL.so + mobilegl_gbm.so, aliases, vendor JSONs, client.conf, GLX/GBM env drop-ins
```
Running clients keep the old inode; restart the client, or `run-plasma.sh` for KWin/plasmashell. Build server and client from the **same commit** (both stamp `git rev-parse HEAD`), else the logs WARN "compatible wire with different build". Fallback in-container build (slow): `scripts/sync-src.sh` + anland `mobilegl-tools/anland-build-client.sh`.
**KWin (anland backend / kwin.patch) - built inside the container**
```sh
bash scripts/kwin-stage.sh build # stage CR-stripped patch+backend+startup script, kwprep.sh, background sync-build-kwin.sh
printf 'tail -5 /root/kwin-build.log\n' | bash scripts/ct.sh kw # wait for "== ALL DONE =="
adb shell 'su -c "sh /data/local/tmp/anl/run-plasma.sh"'
```
`sync-build-kwin.sh` (anland `producers/kde/Arch_v5/`) copies only the backend sources, then `ninja -j4`, installs `kwin_wayland`, `libkwin.so.6.7.4`, the patched screencast plugin and `/opt/mobilegl/bin/mobilegl-startup.sh` by atomic rename. `kwprep.sh` is what applies changed `kwin.patch` hunks to core KWin files. Repo-root `libdisplay_producer/` is not synced by either. `kwin-stage.sh` is assembled from those two scripts and was not run as one unit yet.
**Unit tests (no device)** - in WSL (the MSVC build of this branch is broken):
```sh
MSYS2_ARG_CONV_EXCL='*' wsl -d archlinux -- bash /mnt/c/<worktree>/.claude/skills/anland-mobilegl-plasma/scripts/wsl-test.sh MultiSessionTest TwinOwnershipTest
MSYS2_ARG_CONV_EXCL='*' wsl -d archlinux -- bash -lc 'cd ~/mgl-anl-bld && ctest -R "MultiSession|TwinOwnership" -E DirectVulkan --output-on-failure'
```
`wsl-all.sh` builds and runs every unit test. Never run DirectVulkan integration tests; never two ctest runs at once.
## Switching backends
In the app: Settings > Connection > MobileGL desktop > Renderer backend, then Stop desktop and open the app. Or `adb shell 'su -c "sh /data/local/tmp/anl/switch.sh DirectGLES"'`, which does the same (saves the setting through the launch extra, STOPs the service, reopens; clears a stale `debug.mobilegl.backend`). No setprop, no file to edit in the container: the session copies the published backend into `/etc/mobilegl/backend` on its next start. `cyc.sh` keeps whatever is set.
## System-wide vendor and fallback
Installed by `xdeploy.sh` (and anland's `anland-build-client.sh`), so any process in the container gets MobileGL through the stock libglvnd/libgbm:
| File | Does |
|---|---|
| `/usr/share/glvnd/egl_vendor.d/10_mobilegl.json` | EGL vendor, asked before the system's `50_*.json` |
| `/opt/mobilegl/share/glvnd/egl_vendor.d/50_mobilegl.json` | same JSON, for per-process forcing with `__EGL_VENDOR_LIBRARY_FILENAMES` |
| `/etc/mobilegl/client.conf` | `KEY=value` defaults for every client (transport, data, endpoint); rewritten on deploy |
| `/etc/mobilegl/backend` | the backend (the session's `sync-backend` step copies the app's choice in); wins over a `MOBILEGL_BACKEND_TYPE` line in client.conf |
| `/etc/environment.d/10-mobilegl.conf` | `__GLX_VENDOR_LIBRARY_NAME=mobilegl`, `GBM_BACKEND=mobilegl` for the systemd user manager (all Plasma services and what they start, terminals included) |
| `/etc/profile.d/mobilegl.sh` | the same two for login shells (`bash -l`, `su -l`, ssh, `droidspaces run bash -l`) |
Precedence is environment > `/etc/mobilegl/backend` (backend only) > client.conf > built-in; `MOBILEGL_CONFIG_FILE` / `MOBILEGL_BACKEND_FILE` name other files (empty = off); a server process never reads them.
**Fallback.** Before bringing anything up the client probes the configured endpoint (one non-blocking connect, `MOBILEGL_IPC_PROBE_TIMEOUT_MS`, default 250, 0 = off; a refused abstract socket answers in ~0.2 ms). With no server: the EGL vendor returns `EGL_NO_DISPLAY` and no devices (glvnd moves on to the other vendor), the GLX vendor fails `__glx_Main` (libGLX falls back to the vendor Xwayland names), and `mobilegl_gbm` fails `create_device` (libgbm falls back to the device's own backend). Client queries (`eglQueryString(EGL_NO_DISPLAY)`, `eglGetProcAddress`) never bring a session up. MobileGL also steps aside for a GBM device another backend created and for `LIBGL_ALWAYS_SOFTWARE=1` - that is how a software Xwayland (`ANLAND_XWAYLAND_GLAMOR=0`, KWin's patch sets it then) keeps the software stack. MobileGL never opens an X connection to a display its own process serves (`MG_Util/X11/DisplayGuard.h`; `MOBILEGL_X11_DIAL=0` forbids all), which is what made a glamor Xwayland deadlock on itself before. Show it without stopping the real server:
```sh
sed 's|^MOBILEGL_IPC_CONTROL=.*|MOBILEGL_IPC_CONTROL=unix:@nobody|' /etc/mobilegl/client.conf > /tmp/down.conf
sudo -u swung0x48 env -i HOME=/home/swung0x48 PATH=/usr/bin XDG_RUNTIME_DIR=/run/user/1000 WAYLAND_DISPLAY=wayland-0 \
MOBILEGL_CONFIG_FILE=/tmp/down.conf eglinfo -B -p wayland # vendor = the other one, ~0.1 s
```
## Test clients and how to verify
X11 on the GPU has its own driver, `scripts/x11-verify.sh` (preflight, measure before/after with glxgears at `vblank_mode=0` on both GLX paths (presentation throughput only: glxgears is fixed-function GL, which MobileGL does not implement, so its window is black by design), deploy-client/-kwin/-server, check, apps, stop); `scripts/device/x11-measure.sh` refuses to load the phone below 40% battery.
Inside the container as root (`printf '...' | bash scripts/ct.sh <chan>`), `mgrun offscreen <cmd>` runs a client as swung0x48 in the live session through a login shell, with nothing but the surface mode and the log path set (log `/tmp/mgl-run.client.log`, override base with `MGLOG=/tmp/x.log`); a plain `env -i ... bash -l -c '<cmd>'` as swung0x48 gets MobileGL the same way:
```sh
mgrun offscreen eglinfo -B ; mgrun offscreen glxinfo -B
timeout 25 mgrun offscreen glmark2-es2-wayland -s 1280x720 --run-forever > /tmp/gm.out 2>&1 &
sleep 8; grep -a "presented through" /tmp/mgl-run.client.log
```
Chrome: `bash scripts/chrome-launch.sh <tag> [flags]` (log `/tmp/mgl-chrome-<tag>.client.log`, screenshot `ch-<tag>.png`; `CHROME_GPU=in-process` puts the GPU back in the browser, which cannot recover from a device loss).
Video: Chrome decodes in hardware. The GPU process's render node has a VA-API driver in the container (`/usr/lib/dri/msm_drm_drv_video.so`, DroidSpaces' V4L2 driver): Chrome allocates its frames with `vaCreateSurfaces` and exports them (`vaExportSurfaceHandle`, one fd, R8 + GR88 layers) - the driver's own buffers, never GBM ones - and imports each as an NV12 (P010 for 10-bit) EGLImage bound to `GL_TEXTURE_EXTERNAL_OES`. MobileGL takes such a FOREIGN YUV dma-buf through its labelled CPU-copy fallback: the server maps it and copies it into a YUV AHardwareBuffer of its own (`AHardwareBuffer_lockPlanes`) at the reader's first use in each frame (any present, publishing flush or native fence), then converts that into the texture's RGBA level on the GPU (Espryt: `GL_EXT_YUV_target` with the import's colour hints, `MOBILEGL_YUV_DRIVER_CONVERSION=1` for the driver's external sampler; Magma: `VkSamplerYcbcrConversion` on the AHB's external format). A server-allocated NV12/P010 buffer (GBM `gbm_bo_create(NV12)`) needs no copy. `MOBILEGL_CHROME_VIDEO_DECODE=software` in `mobilegl-startup.sh chrome` brings back `--disable-accelerated-video-decode`. Check a video with `scripts/probes/cdpmedia.py <port> <s>` in the container (CDP `Media.enable`: `kVideoDecoderName = VaapiVideoDecoder`, `kIsPlatformVideoDecoder = true`; `CDP_NAVIGATE=<url>` / `CDP_EXPR=<js>`), which needs Chrome started with `--remote-debugging-port` and a non-default `--user-data-dir` (a fresh profile also needs `--password-store=basic`, else page loads wait forever on KWallet), and grep Chrome's log for `exit_code` (a GPU-process restart). `scripts/probes/yuvprobe.c` pixel-checks foreign NV12/P010 imports, their per-frame refresh and a GBM NV12 buffer (`mgrun offscreen mgl-yuvprobe`, exit 0 = pass). KWin's own NV12 path (it splits a YUV buffer into R8 + GR88 imports) is not served: MobileGL imports no single-channel dma-bufs, so KWin advertises no YUV format over linux-dmabuf. A vendor-agnostic replacement driver (MobileVA: VA-API on Android MediaCodec, decoding into server YUV images with no CPU copy, the current driver kept as an A/B fallback) is planned, not built: `docs/MobileVA/plan.md`.
Device loss: `setprop debug.mobilegl.inject_device_lost_pid <pid>` (as root) loses the server session of the client whose pid - as it sees itself in the container - is `<pid>`, once per value, on any backend and whether or not it presents; `debug.mobilegl.inject_device_lost_at N` instead loses whichever session makes the N-th device check (Espryt presents and read-backs only). `bash scripts/chrome-recover.sh <in-process|process> [rounds]` checks that Chrome repaints after such a loss without a restart (battery-guarded, stops below 40%). `bash scripts/kwin-recover.sh [rounds]` loses KWin's own session (default twice) and checks that KWin restarts compositing on a fresh session - same KWin, plasmashell and client processes, the screen still animating (glmark2 on Wayland, glxgears on X11, `CHROME=1` adds Chrome) - which needs the anland backend + kwin.patch reset handling deployed (`kwin-stage.sh build`).
Probes: `scripts/probes/texfmt.c` (texture formats keep alpha?), `dmahash.c` (hash a process's dma-bufs), `yuvprobe.c` (YUV dma-buf sampling, `gcc -O1 /root/yuvprobe.c -o /usr/local/bin/mgl-yuvprobe -lEGL -lGLESv2 -lgbm -lm`), `cdpmedia.py` (which decoder a Chrome page uses); build in the container, e.g. `gcc -O1 /root/texfmt.c -o /usr/local/bin/mgl-texfmt -lEGL`, run with `mgrun offscreen`.
GPU hangs: a shader that never ends stops the WHOLE GPU (every process, MobileGL or not) until the kernel resets it - ~2 s, or ~10 s ("Preemption Fault" in dmesg) when a higher-priority context waits; nothing in user space shortens one stall. What the server does is end the guilty session so it cannot hang the GPU again (Magma: event-bracketed submissions + `mgl-gpu-watch`, budget `MOBILEGL_GPU_HANG_BUDGET_MS` / `debug.mobilegl.gpu_hang_budget_ms`, default 2000; Espryt: the driver's GUILTY reset status, which it reports after the ~2 s reset only). Repro: `scripts/probes/hangclient.c` (Wayland + GL 3.3 core; `-H 60 -N 1 -E 60` hangs every 60 frames, `-H -1` is a frame-gap observer; build with wayland-scanner's xdg-shell files, `-lwayland-client -lwayland-egl -lEGL -lOpenGL -lm`), and `gpuprobe.c` (NDK, plain adb shell; `gpuprobe 30 high` is a high-priority GLES context outside MobileGL - it both measures the device-wide stall and forces the 10 s preemption path).
A result counts only if ALL hold:
- client log says `presented through linux-dmabuf shared images` (not `wl_shm`; `could not be presented through a linux-dmabuf shared image` = fell back);
- `bash scripts/shot.sh g 2 1` gives two different hashes while the client animates (a "fast" score can mean frames are not shown);
- `adb logcat -b crash -d` and `adb logcat -d | grep -E "Fatal signal|Abort message|E MobileGL"` are clean (a frozen window is usually a dead server);
- for plasmashell verdicts, the sticky software renderer is off (below) and `/tmp/mgl-plasmashell.client.log` is freshly written.
Reference (glmark2-es2-wayland 1280x720, zero-copy vs wl_shm): Espryt 187 vs 49, Magma 158 vs 34. Rerun any surprising pass.
## Logs
| What | Where |
|---|---|
| KWin client side | container `/tmp/mobilegl-compositor.client.log` |
| plasmashell / ksplash | `/tmp/mgl-plasmashell.client.log`, `/tmp/mgl-ksplash.client.log` (user drop-ins `~/.config/systemd/user/plasma-{plasmashell,ksplash}.service.d/mobilegl-log.conf`) |
| mgrun / Chrome | `/tmp/mgl-run.client.log`, `/tmp/mgl-chrome-<tag>.client.log` (+ `/tmp/chrome-<tag>-out.log`) |
| Server | `adb logcat -s MobileGL`; file `/data/data/com.anland.consumer.mobilegl/files/mobilegl-server.server.log` |
| Native crashes | `adb logcat -b crash -d`; container `coredumpctl list` |
| Session / KWin stderr | container `journalctl -u desktop-session --no-pager`, `journalctl --no-pager --since -2min \| grep kwin_wayland` |
| Display daemon | `/data/local/tmp/anland-mobilegl/display-daemon.log` |
| Desktop start/stop (app's su script) | `/data/local/tmp/anland-mobilegl/desktop.log`, `container-start.log`; logcat `-s AnlandMobileGL` |
| Session wait / backend sync | container `journalctl -u desktop-session --no-pager` ("waiting for the MobileGL server ...") |
`MOBILEGL_LOG_FILE_PATH=/tmp/x.log` produces `/tmp/x.client.log` (and `.server.log` for forwarded lines); files are truncated on open, so one base per process. Client builds log at INFO. Symbolize server frames with `$ANDROID_NDK/toolchains/llvm/prebuilt/windows-x86_64/bin/llvm-addr2line.exe -Cfe build-android/libMobileGL.so <offset>`.
## Pitfalls
- **pkill -f self-match**: a `ct.sh`/`droidspaces run bash -lc "<script>"` process carries the script in its argv; `pkill -f chrome` kills the runner. Use `pkill -x`.
- **Root-owned /tmp files**: a log created by a root-run client cannot be reopened by swung0x48 later (no log appears). `rm -f` it as root first.
- **Sticky Plasma software renderer**: after any server crash Plasma writes `[QtQuickRendererSettings] SceneGraphBackend=software` into `~swung0x48/.config/kdeglobals`; plasmashell then renders on the CPU and screenshots still look fine (tray icon "正在使用的软件渲染器"). `mobilegl-startup.sh plasma` deletes it at every session start; within a running session that hit a crash, restart the session (`run-plasma.sh`) before any verdict.
- **Lock screen**: autolock is off (`kscreenlockerrc [Daemon] Autolock=false`); a real lock can only be left by typing the password (logind lock/unlock is not supported for this session type - `device/unlock.sh` is unverified). Test the greeter with `/usr/lib/kscreenlocker_greet --testing` and kill it.
- **CRLF**: `core.autocrlf=true` here; anland's backend sources and `sync-build-kwin.sh` are CRLF in the Windows checkout. Strip CR before anything runs on Linux/Android (`tr -d '\r'`; `ct.sh`, `push-tools.sh`, `kwin-stage.sh` do). This skill has a `.gitattributes` forcing LF.
- **Display daemon**: the app starts it; the session's `wait` step keeps KWin from starting without it (a KWin that cannot reach it logs "failed to connect to display daemon at /run/anland-mobilegl/display.sock" and exits in ~3 s). Anything started from the app's `su` must leave the app's cgroup (`echo $$ > /sys/fs/cgroup/cgroup.procs`) or a force-stop kills it.
- **Device power/connection**: low battery or the screen turning off (30 s default) can kill the SurfaceView - `svc power stayon true`; wireless adb `ip:port` changes per pairing - never pin `-s HA27Q3LQ`, use `ANDROID_SERIAL` if several devices are attached.
- **Container vs phone paths**: container `/tmp` is a different place from the phone's `/data/local/tmp`; move files via `/mnt/Droidspaces/arch-kde-mgl/root` (container `/root`) or `/data/local/tmp/anland-mobilegl` (= container `/run/anland-mobilegl`).
- **Never overwrite a mapped .so in place** (container or APK while running): copy to a new inode and `mv -f`; the APK swap force-stops the app first.
- **Input**: plain `adb shell input` is blocked - use `su -c input`; anland grabs the touchscreen, so if `input tap` does nothing use `input mouse tap x y`. Never `debuggerd -b` a container process (signal 35 kills glibc processes).
- **linux-dmabuf**: the client must use `create` + roundtrip, never `create_immed` (KWin answers a failed immediate import with a protocol error that kills the client).
- **Cross-session bugs**: noise / another window's pixels after some client starts or exits = process-global state in the multi-session server (handles are per client, FBO/VAO per context). Repro with `kscreenlocker_greet --testing`.
- `/etc/environment` leaks the baseline session's driver variables through pam_env (and, via `/usr/lib/environment.d/99-environment.conf`, into the user manager); `mobilegl-startup.sh` unsets them - keep that when editing it.
- **Shared log path**: Xwayland inherits KWin's `MOBILEGL_LOG_FILE_PATH`; any line a process logs opens (truncates) that file. MobileGL stays silent in processes it does not serve, but give any extra client its own `MOBILEGL_LOG_FILE_PATH`.
- **Stale per-process overrides**: an `__EGL_VENDOR_LIBRARY_FILENAMES` / `MOBILEGL_*` left in a unit drop-in or the user manager's environment (`systemctl --user show-environment`) still wins over the system files; `run-plasma.sh` (terminate-user) clears the manager's.
- `handoff` mentions a `kwin_wayland` logging shim and `/tmp/mgl-kwin-<pid>.log`: stale - `/opt/mobilegl/kwin/bin/kwin_wayland` is now the real binary and logs to `/tmp/mobilegl-compositor.client.log`.
## House rules
- Work in a git worktree, never in the shared main checkout or its build dirs; worktrees need `GIT_LFS_SKIP_SMUDGE=1`.
- Commit with `git commit --no-verify`, one short sentence, **no** `Co-Authored-By` trailer.
- Never run the DirectVulkan integration tests (`ctest -E DirectVulkan`).
- No external project names ("Mesa", "Zink", ...) in source code or comments; describe the technique.
- Vendor-agnostic only on the zero-copy path: AHB, EGL_ANDROID_*, VK_ANDROID AHB import, sync_file, dma-buf, libgbm ABI - no kgsl/gralloc layouts/QCOM modifiers.
- Identity caches key on monotonic lifetime ids, never GL names or pointers.
@@ -0,0 +1,183 @@
# Reproducing the MobileGL + anland + KDE Plasma setup from source
This is the from-scratch path for a NEW development machine and/or a NEW phone. `SKILL.md` covers the
day-to-day loop once everything below exists. Paths and names that are specific to the original
machine are listed in step 9: set them in `scripts/env.sh` (or the environment) before running anything.
## 0. What you need
**Phone.**
- arm64 Android 11+ with **root**. KernelSU is what this was built on; the scripts only need `su -c`.
- A Vulkan 1.1 driver with `VK_ANDROID_external_memory_android_hardware_buffer` for Magma, and/or GLES 3.2 + EGL_ANDROID_native_fence_sync for Espryt. Nothing vendor-specific is used: no kgsl and no vendor ioctls, only AHardwareBuffer, sync_file and the vendor's public GLES/Vulkan.
- About 20 GB free for the container image.
**Development host.**
- The scripts are written for **Windows + Git Bash** with WSL (`archlinux` distro) for the unit tests and the cross build.
- A Linux host works too: drop the `cygpath`/`MSYS_NO_PATHCONV` handling in `env.sh`, and run the WSL parts natively.
- Tools:
- `adb`, CMake >= 3.22, Ninja, Python 3 (+ Pillow for screenshots/recording).
- Android NDK **27.2.12479018**.
- JDK 17 and Android SDK platform 36 (for the APK).
- In the Linux environment: clang/lld/llvm of ONE major version (a mismatched lld breaks `ld.lld`), cmake, ninja.
## 1. Sources
```sh
# MobileGL: branch feat/disaggregated (everything here is on it)
git clone https://github.com/MobileGL-Dev/MobileGL.git && cd MobileGL
git checkout feat/disaggregated
GIT_LFS_SKIP_SMUDGE=1 git submodule update --init --recursive
# 3rdparty/glslang is the MobileGL-Dev fork (main); its External/spirv-tools must be populated
# (glslang's own update_glslang_sources.py, or copy it from another checkout).
# Keep builds out of a shared checkout: work in a worktree (GIT_LFS_SKIP_SMUDGE=1 git worktree add ...).
# anland: branch legacy-mobilegl-unified (KWin backend + patch, startup script, APK, daemon module)
git clone https://github.com/MobileGL-Dev/anland.git && cd anland # the fork that carries this branch
git checkout legacy-mobilegl-unified
```
Build the server (APK) and the client (container) from the **same MobileGL commit**: both stamp
`git rev-parse HEAD`, and a mismatch logs "compatible wire with different build".
## 2. Phone base: root, Droidspaces, display daemon
1. **Root** with KernelSU (or another `su` provider).
2. **Droidspaces:** install the module and its app. Containers live under `/data/local/Droidspaces/Containers/<name>/`, and the CLI is `/data/local/Droidspaces/bin/droidspaces`.
3. **anland display daemon** (KernelSU/Magisk module `anland-daemon`):
- Build it from the anland repo with `build_daemon_android.sh`, then package it with `build_magisk_module.sh`, which uses `magisk_module/`. Flash it and reboot.
- The module's `service.sh` starts the original daemon on `/data/local/tmp/display_daemon.sock`.
- The MobileGL session uses a SECOND instance of the same binary on `/data/local/tmp/anland-mobilegl/display.sock`: see `scripts/device/restart-daemon.sh`, which also chmods the directory and socket.
## 3. The container
Use an **Arch Linux ARM (aarch64, glibc) KDE Plasma Wayland rootfs** that has a systemd `desktop-session.service` running the desktop as a normal user with `PAMName=login`. A ready-made one is the Arch KDE "Wayland" image from Droidspaces RootFS Desktop Builder (see anland `doc/UserManual/anland_guide.md`, "Use prebuilt packages or a RootFS"). Or install `plasma-desktop`, `plasma-workspace`, `libglvnd`, `xorg-xwayland`, `base-devel`, `cmake` and `ninja` on plain Arch Linux ARM and write the service yourself.
Import it into Droidspaces as `arch-kde-mgl`. The default name is in `env.sh`. Reference `container.config` keys:
```
enable_gpu_mode=1
net_mode=host
bind_mounts=/data/local/tmp/display_daemon.sock:/run/display.sock,/data/local/tmp/anland-mobilegl:/run/anland-mobilegl
run_at_boot=0
```
- **Host networking:** it is required. The MobileGL server listens on the ABSTRACT socket `@anland-mobilegl`, which crosses into the container only when the network namespace is shared.
- **The anland-mobilegl directory:** create `/data/local/tmp/anland-mobilegl` on the phone before the first start, mode 777.
- **Desktop user:** the default is `swung0x48` (`MGL_DESKTOP_USER`). Create your own and point `desktop-session.service` at it.
- **Starting it:** `droidspaces -C .../container.config start`. Android root sees the rootfs at `/mnt/Droidspaces/<name>`, which `CT_ROOT` in `env.sh` relies on.
- **Container tools:** install them with `bash scripts/push-tools.sh` once the container runs.
## 4. MobileGL server and the anland APK
```sh
bash scripts/build-android.sh # -> <worktree>/build-android/libMobileGL.so
```
The script runs, with the NDK toolchain:
```
-DANDROID_ABI=arm64-v8a -DANDROID_PLATFORM=android-30 -DCMAKE_BUILD_TYPE=RelWithDebInfo
-DMOBILEGL_PIPE_PUSH=ON -DMOBILEGL_BUILD_DISAGGREGATED=ON -DMOBILEGL_BUILD_DISAGGREGATED_INPROC=ON
-DMOBILEGL_BUILD_TEST=OFF -DMOBILEGL_BUILD_BENCHMARK=OFF
```
APK, from anland `consumers/anland_v5/android_consumer` (details in its `MOBILEGL.md`):
```sh
mkdir -p dist/arm64-v8a && cp <worktree>/build-android/libMobileGL.so dist/arm64-v8a/
MOBILEGL_DIST=$PWD/dist ./gradlew :app:assemblePlainDebug \
-PanlandNdkVersion=27.2.12479018 -PmobileglApplicationId=com.anland.consumer.mobilegl \
-PanlandDefaultSocket=/data/local/tmp/anland-mobilegl/display.sock
adb install -r app/build/outputs/apk/plain/debug/app-plain-debug.apk
```
- **Grant root:** the app needs root for its helper; approve it in KernelSU.
- **Reinstall with another signing key:** that changes the UID, so grant root again (anland `mobilegl-tools/ksu_grant.c` does it from a shell). Don't uninstall to fix a signature clash: that wipes the app data.
- **Later server-only changes:** swap just the library into the installed APK (`scripts/cyc.sh` / `device/swap.sh`) instead of rebuilding the APK.
## 5. MobileGL client in the container
Pick one of these two:
- **Cross build** (fast, from WSL or Linux):
1. `bash scripts/xsysroot/refresh-sysroot.sh` exports the container's headers and libraries to `~/sysroots/arch-kde-mgl` and installs `~/sysroots/aarch64-arch.cmake`. Re-run it after container package upgrades.
2. `bash scripts/xbuild.sh`, then `bash scripts/xdeploy.sh`.
- **Native build in the container** (slow, no WSL): `scripts/sync-src.sh` copies the worktree to `/root/mobilegl-unified-src`, then run anland `producers/kde/Arch_v5/mobilegl-tools/anland-build-client.sh` inside (g++, `-DMOBILEGL_BUILD_DISAGGREGATED=ON`, Release).
Either route installs the same files, renamed into place so running clients keep their old mapping:
| File | What it is |
|---|---|
| `/opt/mobilegl/lib/libMobileGL.so` (+ `libEGL`/`libGL` aliases) | the client library |
| `/usr/share/glvnd/egl_vendor.d/10_mobilegl.json` | system-wide EGL vendor; sorts before the distribution's `50_*.json`, which serves whenever MobileGL declines (no server reachable) |
| `/opt/mobilegl/share/glvnd/egl_vendor.d/50_mobilegl.json` | per-process forcing |
| `/usr/lib/libGLX_mobilegl.so.0` | GLX vendor |
| `/usr/lib/gbm/mobilegl_gbm.so` | GBM backend (zero-copy window buffers) |
| `/etc/mobilegl/client.conf` | `MOBILEGL_TRANSPORT=spawn`, `MOBILEGL_IPC_DATA=shm`, `MOBILEGL_IPC_CONTROL=unix:@anland-mobilegl` |
| `/etc/mobilegl/backend` | `DirectVulkan` (Magma) or `DirectGLES` (Espryt) |
| `/etc/environment.d/10-mobilegl.conf`, `/etc/profile.d/mobilegl.sh` | `__GLX_VENDOR_LIBRARY_NAME=mobilegl`, `GBM_BACKEND=mobilegl` |
## 6. KWin 6.7.4 with the anland backend, and Xwayland
The MobileGL session runs a rebuilt KWin from `/opt/mobilegl/kwin`, so the packaged one stays untouched.
1. **Source tree:** in the container, unpack KWin 6.7.4 to `/root/mobilegl-build/kwin-6.7.4`, apply anland `producers/kde/Arch_v5/kwin.patch`, and overlay `producers/kde/anland_backend_Arch_v5/src/backends/anland` into `src/backends/anland`. Also take the current `common/` and `libdisplay_producer/`, as `mobilegl.md` there says.
2. **Configure:**
```sh
cmake -S kwin-6.7.4 -B kwin-6.7.4/build -G Ninja -DCMAKE_BUILD_TYPE=Release \
-DCMAKE_INSTALL_PREFIX=/usr -DCMAKE_INSTALL_LIBDIR=lib -DBUILD_TESTING=OFF
```
3. **Build and install:** run `producers/kde/Arch_v5/sync-build-kwin.sh` (`ninja -j4`). It installs `kwin_wayland` + `libkwin.so.6.7.4` into `/opt/mobilegl/kwin/{bin,lib}`, the patched screencast plugin in place (the original is kept beside it), and `/opt/mobilegl/bin/mobilegl-startup.sh`.
- From the dev host, `bash scripts/kwin-stage.sh build` does steps 1–3 incrementally: `kwprep.sh` applies changed patch hunks.
4. **Xwayland:** the anland-patched Xwayland 24.1.13 comes from `producers/kde/Arch_v5/build.sh` (makepkg, as a normal user; it builds the anland KWin and Xwayland packages). A prebuilt anland rootfs already has it.
5. **Session hook:** add `/etc/systemd/system/desktop-session.service.d/mobilegl.conf`:
```ini
[Service]
ExecStart=
ExecStart=/opt/mobilegl/bin/mobilegl-startup.sh plasma
Environment=ANLAND_SOCKET=/run/anland-mobilegl/display.sock
Environment=QT_LOGGING_RULES=kwin_*.info=true
```
`mobilegl-startup.sh plasma` writes the user drop-in for `plasma-kwin_wayland.service` (the `/opt/mobilegl/kwin` binary, `ANLAND_MOBILEGL=1`, `MOBILEGL_IPC_SURFACE=server`, `GBM_BACKEND=mobilegl`) and runs `startplasma-wayland`.
## 7. Bring it up
Nothing has to be started by hand, and the order no longer matters: **open the app** (its launcher
icon, or `am start -n com.anland.consumer.mobilegl/com.anland.consumer.MainActivity` with no extras).
That is the single trigger. The app starts its foreground service and runs its bundled
`mobilegl-desktop.sh up` as root, which:
1. starts the display daemon if it is not running;
2. publishes the backend (the app setting "Renderer backend", default DirectGLES) to
`/data/local/tmp/anland-mobilegl/backend`;
3. starts the container if it is stopped;
4. starts `desktop-session`, which waits for the server and the daemon socket, copies the backend into
`/etc/mobilegl/backend` and runs Plasma.
Cold start to a desktop takes about 8-12 s. `scripts/device/bringup.sh` wakes the phone, dismisses the
keyguard, opens the app and prints the timeline. To change the backend, use the app setting and then
Stop desktop and reopen, or run `scripts/device/switch.sh DirectGLES|DirectVulkan`. Do not use
`setprop debug.mobilegl.backend`: it overrides the setting until the next reboot. **Stop desktop**
(the notification action) stops the session and the container.
**After a phone reboot:** wireless adb does not come back by itself. Re-enable it on the phone
(Developer options > Wireless debugging), then `adb connect <ip>:<port>`; the port changes. After
that, open the app as above. See SKILL.md "Starting, idling, stopping".
## 8. Verify
- **Plasma:** it appears in the app. The tray must NOT show the "software renderer" icon; if it does, delete `SceneGraphBackend=software` from the user's `kdeglobals` and restart.
- **Plain shell:** in a plain container shell as the desktop user, `eglinfo -B` names MobileGL, and `glxinfo -B` from a login shell does too.
- **Wayland client:** `timeout 25 mgrun offscreen glmark2-es2-wayland -s 1280x720`. The log must say `presented through linux-dmabuf shared images`.
- **Chrome:** `bash scripts/chrome-launch.sh t1` gives windows with correct icons.
- **Resize and lock:** BACK toggles the extra-keys bar (resize); lock/unlock resumes the desktop. `rtest.sh` in the scratch dir checks both.
- **Unit tests:** `bash scripts/wsl-all.sh` in WSL. NEVER run the DirectVulkan integration tests.
## 9. Machine-specific values to change
| Value | Original | Where |
|---|---|---|
| MobileGL worktree | this skill's worktree | `MGL_WORKTREE` (auto) |
| scratch dir | `%LOCALAPPDATA%/Temp/anl` | `ANL_DIR` |
| NDK | `<SDK>/ndk/27.2.12479018` | `ANDROID_NDK` |
| anland worktree | `%USERPROFILE%/AndroidStudioProjects/anland/.claude/worktrees/mgl-unified` | `ANLAND_WORKTREE` (`kwin-stage.sh`) |
| APK id | `com.anland.consumer.mobilegl` | `MGL_PKG` |
| container | `arch-kde-mgl` | `MGL_CONTAINER` |
| desktop user | `swung0x48` | `MGL_DESKTOP_USER` + `desktop-session.service` |
| WSL distro | `archlinux` | `WSL_DISTRO` (`xbuild.sh`, `wsl-*.sh`) |
| cross build dir | WSL `~/mgl-xbuild-a64` | `MGL_XBUILD_DIR` |
| phone | Lenovo TB321FU, Adreno 750 | nothing: no serial is pinned; with several devices set `ANDROID_SERIAL` |
**Thermals:** a phone vendor's thermal daemon may throttle charging while the screen is on during long sessions. That is not part of this setup.
@@ -0,0 +1,25 @@
#!/usr/bin/env bash
# build-android.sh - build the Android SERVER libMobileGL.so (NDK, arm64-v8a) that the anland APK embeds.
# Output: $MGL_WORKTREE/build-android/libMobileGL.so (RelWithDebInfo, unstripped, ~320 MB; keep it for
# llvm-addr2line, it is what cyc.sh pushes). Stamp = worktree HEAD, so the container client
# (xbuild.sh, same stamp) and the server agree on the build ("compatible wire with different build"
# WARN otherwise). Env: MGL_WORKTREE, ANDROID_NDK, BUILD_DIR (default build-android), JOBS (12).
# Log level is compiled to INFO. android-30 = the APK's minSdk: thread_local is then the platform's
# own TLS rather than the emulated one (which cost 7-12% of an apply thread on the API 26 build).
set -euo pipefail
. "$(dirname "$0")/env.sh"
W=$MGL_WORKTREE
B=$W/${BUILD_DIR:-build-android}
STAMP=$(git -C "$W" rev-parse HEAD)
if [ ! -f "$B/build.ninja" ]; then
cmake -S "$W" -B "$B" -G Ninja -DCMAKE_TOOLCHAIN_FILE="$ANDROID_NDK/build/cmake/android.toolchain.cmake" \
-DANDROID_ABI=arm64-v8a -DANDROID_PLATFORM=android-30 -DCMAKE_BUILD_TYPE=RelWithDebInfo \
-DMOBILEGL_PIPE_PUSH=ON -DMOBILEGL_BUILD_DISAGGREGATED=ON -DMOBILEGL_BUILD_DISAGGREGATED_INPROC=ON \
-DMOBILEGL_BUILD_TEST=OFF -DMOBILEGL_BUILD_BENCHMARK=OFF \
-DCMAKE_CXX_FLAGS=-DMOBILEGL_LOG_ACTIVE_LEVEL=MOBILEGL_LOG_LEVEL_INFO \
-DMOBILEGL_BUILD_STAMP="$STAMP"
else
cmake "$B" -DMOBILEGL_BUILD_STAMP="$STAMP" >/dev/null
fi
cmake --build "$B" --target MobileGL -j"${JOBS:-12}"
ls -la "$B/libMobileGL.so"
@@ -0,0 +1,25 @@
#!/usr/bin/env bash
# chrome-launch.sh <tag> [extra chrome flags...] - (re)start Chrome inside the running Plasma session
# through mobilegl-startup.sh's `chrome` mode (ANGLE GLES over MobileGL EGL), client log
# /tmp/mgl-chrome-<tag>.client.log, then print its "presented through" line and save a screenshot
# to $ANL_DIR/ch-<tag>.png. Kills the running chrome first (pkill -x, never -f). MobileGL is the
# container's system-wide vendor (10_mobilegl.json + /etc/mobilegl/client.conf), so only the log is set here.
# Env: WAIT (s before checking, default 20), CHANNEL (ct.sh channel, default chrome),
# CHROME_GPU (process, the default: the GPU in its own process, which presents
# through GBM dma-bufs and is restarted by Chrome after a device loss; in-process opts out -
# needs the anland startup script that knows MOBILEGL_CHROME_GPU; in process mode the GPU process and the browser
# write the one client log, the GPU process last); CHROME_GPU_SANDBOX=on keeps its GPU sandbox.
set -euo pipefail
here=$(cd "$(dirname "$0")" && pwd)
. "$here/env.sh"
TAG=${1:?usage: chrome-launch.sh <tag> [chrome flags...]}; shift
bash "$here/ct.sh" "${CHANNEL:-chrome}" <<IN
pkill -x chrome; sleep 2
U=\$(id -u $MGL_DESKTOP_USER)
rm -f /tmp/mgl-chrome-$TAG.client.log /tmp/chrome-$TAG-out.log
sudo -u $MGL_DESKTOP_USER env XDG_RUNTIME_DIR=/run/user/\$U WAYLAND_DISPLAY=wayland-0 DBUS_SESSION_BUS_ADDRESS=unix:path=/run/user/\$U/bus MOBILEGL_LOG_FILE_PATH=/tmp/mgl-chrome-$TAG.log MOBILEGL_CHROME_GPU=${CHROME_GPU:-process} MOBILEGL_CHROME_GPU_SANDBOX=${CHROME_GPU_SANDBOX:-off} bash -c 'nohup /opt/mobilegl/bin/mobilegl-startup.sh chrome --restore-last-session $* > /tmp/chrome-$TAG-out.log 2>&1 &'
sleep ${WAIT:-20}
grep -ah 'presented through' /tmp/mgl-chrome-$TAG.client.log | head -3
IN
adb_ shell "su -c 'screencap -p $DEV_SHARE/ch-$TAG.png'"
adb_ pull "$DEV_SHARE/ch-$TAG.png" "$ANL_DIR/ch-$TAG.png" >/dev/null && echo "screenshot: $ANL_DIR/ch-$TAG.png"
@@ -0,0 +1,146 @@
#!/usr/bin/env bash
# chrome-recover.sh <in-process|process> [rounds] - does Chrome repaint after a GPU device loss,
# without a restart? Starts Chrome (chrome-launch.sh, CHROME_GPU=<mode>) on an animated WebGL page,
# then per round: screenshots (the page must animate), loses the session of Chrome's GPU client with
# the server's pid-targeted debug knob (debug.mobilegl.inject_device_lost_pid; the pid as Chrome sees
# itself in the container), waits, and screenshots again. A round PASSES when the server latched
# that pid's session, the browser process is the same one (no restart), a GPU client is drawing on a
# NEW server session, and the screen still animates. Stops early below MIN_BATTERY percent.
#
# in-process the GPU in the browser process (the startup script's default today): the target is
# the browser itself; recovery needs ANGLE to make a new native context.
# process the GPU in its own process (MOBILEGL_CHROME_GPU=process): the target is the GPU
# process; Chrome restarts it after the loss, so the next round's target is a new pid.
#
# Deploy first (same MobileGL commit on both sides; see SKILL.md):
# bash scripts/build-android.sh && bash scripts/cyc.sh pre # server -> APK, restarts the session
# bash scripts/xbuild.sh && bash scripts/xdeploy.sh # client -> container
# STARTUP=<anland worktree>/producers/kde/Arch_v5/mobilegl-startup.sh bash scripts/chrome-recover.sh deploy-startup
# Env: ROUNDS (default 2), SETTLE (s after the loss, default 8), MIN_BATTERY (default 40),
# CHANNEL (ct.sh channel, default crec), TAG (log tag, default crec-<mode>).
set -euo pipefail
here=$(cd "$(dirname "$0")" && pwd)
. "$here/env.sh"
MODE=${1:?usage: chrome-recover.sh <in-process|process|deploy-startup> [rounds]}
ROUNDS=${2:-${ROUNDS:-2}}
SETTLE=${SETTLE:-8}
MIN_BATTERY=${MIN_BATTERY:-40}
CH=${CHANNEL:-crec}
TAG=${TAG:-crec-$MODE}
PAGE=/tmp/mgl-spin.html
ct() { bash "$here/ct.sh" "$CH"; }
battery() { adb_ shell dumpsys battery | tr -d '\r' | awk '/ level:/ {print $2}'; }
check_battery() {
local level; level=$(battery)
echo "battery: $level%"
if [ -n "$level" ] && [ "$level" -lt "$MIN_BATTERY" ]; then
echo "STOP: battery below $MIN_BATTERY%"; cleanup; exit 3
fi
}
cleanup() { adb_ shell "su -c 'setprop debug.mobilegl.inject_device_lost_pid 0'" || true; }
if [ "$MODE" = deploy-startup ]; then
# The anland startup script that knows MOBILEGL_CHROME_GPU, installed by rename (a running
# session keeps its open copy).
src=${STARTUP:?STARTUP=<path to the anland mobilegl-startup.sh>}
tr -d '\r' < "$src" > "$ANL_DIR/mobilegl-startup.sh"
adb_ push "$ANL_DIR/mobilegl-startup.sh" "$DEV_SHARE/mobilegl-startup.sh.new" >/dev/null
ct <<'IN'
install -m 755 /run/anland-mobilegl/mobilegl-startup.sh.new /opt/mobilegl/bin/mobilegl-startup.sh.new
mv -f /opt/mobilegl/bin/mobilegl-startup.sh.new /opt/mobilegl/bin/mobilegl-startup.sh
grep -c MOBILEGL_CHROME_GPU /opt/mobilegl/bin/mobilegl-startup.sh
IN
exit 0
fi
case "$MODE" in in-process|process) ;; *) echo "mode must be in-process or process" >&2; exit 2;; esac
trap cleanup EXIT
check_battery
adb_ shell "su -c 'sh $DEV_TOOLS/status.sh'" || true
cleanup
# The page: WebGL clears that change colour every frame plus a moving block, so a frame that
# reaches the screen changes the screenshot.
ct <<IN
cat > $PAGE <<'HTML'
<!doctype html><html><body style="margin:0;background:#000">
<canvas id=c width=640 height=360 style="width:100vw;height:100vh"></canvas>
<script>
const gl = document.getElementById('c').getContext('webgl');
let t = 0, lost = 0;
document.getElementById('c').addEventListener('webglcontextlost', e => { e.preventDefault(); lost++; });
function frame() {
t++;
if (gl && !gl.isContextLost()) {
gl.clearColor((t % 60) / 60, ((t + 20) % 60) / 60, ((t + 40) % 60) / 60, 1);
gl.clear(gl.COLOR_BUFFER_BIT);
gl.enable(gl.SCISSOR_TEST); gl.scissor((t * 7) % 600, 100, 40, 160);
gl.clearColor(1, 1, 1, 1); gl.clear(gl.COLOR_BUFFER_BIT); gl.disable(gl.SCISSOR_TEST);
}
document.title = 'spin ' + t + ' lost ' + lost;
requestAnimationFrame(frame);
}
frame();
</script></body></html>
HTML
chmod 644 $PAGE
IN
WAIT=${WAIT:-20} CHROME_GPU=$MODE CHANNEL=$CH bash "$here/chrome-launch.sh" "$TAG" --new-window "file://$PAGE"
# The target pid inside the container: the browser (in-process) or the GPU process.
target_pid() {
ct <<'IN' | tr -d '\r' | tail -1
for p in $(pgrep -x chrome); do
args=$(tr '\0' ' ' < /proc/$p/cmdline)
case "$args" in
*--type=gpu-process*) echo "gpu $p" ;;
*--type=*) ;;
*) echo "browser $p" ;;
esac
done | sort > /tmp/mgl-crec-pids
cat /tmp/mgl-crec-pids >&2
echo "$(awk '/^browser/ {print $2; exit}' /tmp/mgl-crec-pids) $(awk '/^gpu/ {print $2; exit}' /tmp/mgl-crec-pids)"
IN
}
pass=0
for round in $(seq 1 "$ROUNDS"); do
echo "=== round $round/$ROUNDS ($MODE)"
check_battery
read -r browser gpu < <(target_pid)
if [ "$MODE" = process ]; then target=$gpu; else target=$browser; fi
echo "browser=$browser gpu=${gpu:-none} target=$target"
if [ -z "$target" ]; then echo "FAIL: no target process"; break; fi
bash "$here/shot.sh" "$TAG-r$round-pre" 2 1
adb_ logcat -c
adb_ shell "su -c 'setprop debug.mobilegl.inject_device_lost_pid $target'"
sleep "$SETTLE"
adb_ shell "su -c 'setprop debug.mobilegl.inject_device_lost_pid 0'"
bash "$here/shot.sh" "$TAG-r$round-post" 3 1 | tee "$ANL_DIR/$TAG-r$round-post.md5"
latched=$(adb_ logcat -d | grep -c "injected:pid=$target" || true)
started=$(adb_ logcat -d | grep -c "session #[0-9]* started" || true)
adb_ logcat -d | grep -E "inject_device_lost_pid|SessionLatch|session #[0-9]+ (started|pid)" | cut -c1-220 | head -12 || true
read -r browser2 gpu2 < <(target_pid)
echo "after: browser=$browser2 gpu=${gpu2:-none}"
ct <<IN || true
echo "--- client log ($TAG):"; grep -aE "DEVICE LOST|RECOVERED|GL_UNKNOWN_CONTEXT_RESET|failed to make|session #" /tmp/mgl-chrome-$TAG.client.log | tail -8
echo "--- chrome stderr:"; grep -aiE "context lost|GPU process|make.*current|lost" /tmp/chrome-$TAG-out.log | tail -8
IN
adb_ logcat -b crash -d | tail -5
hashes=$(awk '{print $1}' "$ANL_DIR/$TAG-r$round-post.md5" | sort -u | wc -l)
verdict=PASS
[ "$latched" -ge 1 ] || { echo " - the server never latched pid $target"; verdict=FAIL; }
[ "$browser2" = "$browser" ] || { echo " - the browser process changed ($browser -> $browser2): Chrome restarted"; verdict=FAIL; }
[ "$started" -ge 1 ] || { echo " - no new server session started after the loss"; verdict=FAIL; }
if [ "$MODE" = process ]; then
[ -n "$gpu2" ] && [ "$gpu2" != "$gpu" ] || { echo " - no new GPU process"; verdict=FAIL; }
fi
[ "$hashes" -ge 2 ] || { echo " - the screen does not change after the loss (frozen)"; verdict=FAIL; }
echo "round $round: $verdict"
[ "$verdict" = PASS ] && pass=$((pass + 1))
[ "$verdict" = PASS ] || break
done
echo "=== $pass/$ROUNDS rounds passed ($MODE); screenshots in $ANL_DIR/$TAG-r*-*.png"
@@ -0,0 +1,29 @@
#!/bin/bash
# kwprep.sh - (container, as root) bring the live KWin tree /root/mobilegl-build/kwin-6.7.4 in line with
# anland's kwin.patch + anland backend sources, touching ONLY files whose content changed (keeps the
# ninja build incremental). It re-applies the whole patch to a pristine copy of KWin's src/
# (/root/gbm-dev/kwin-a/src) in /root/kp and copies every patched file that differs into the live tree,
# then copies changed backend files. Build + install afterwards with anland's sync-build-kwin.sh.
#
# Input: /root/kwsync.tgz, a tar made in the anland worktree's producers/kde directory:
# tar czf kwsync.tgz Arch_v5/kwin.patch Arch_v5/mobilegl-startup.sh anland_backend_Arch_v5/src/backends/anland/
# and pushed to the container's /root (e.g. via /mnt/Droidspaces/arch-kde-mgl/root/ on the Android side).
# The pristine tree /root/gbm-dev/kwin-a/src was set up by hand earlier (unverified that it is exactly
# upstream 6.7.4); patch hunks that fail are reported by `patch`.
set -e
rm -rf /root/kwsync && mkdir -p /root/kwsync && tar xzf /root/kwsync.tgz -C /root/kwsync
P=/root/kwsync/Arch_v5/kwin.patch
rm -rf /root/kp && mkdir -p /root/kp && cp -a /root/gbm-dev/kwin-a/src /root/kp/
cd /root/kp && patch -p1 -s < $P
LIVE=/root/mobilegl-build/kwin-6.7.4
FILES=$(grep -E '^\+\+\+ b/' $P | sed 's|^+++ b/||' | awk '{print $1}' | sort -u)
for f in $FILES; do
if [ -f "/root/kp/$f" ] && ! cmp -s "/root/kp/$f" "$LIVE/$f"; then cp "/root/kp/$f" "$LIVE/$f"; echo "updated $f"; fi
done
for f in /root/kwsync/anland_backend_Arch_v5/src/backends/anland/*; do
# libdisplay_producer is a symlink in git; a Windows checkout turns it into a small text file. The live
# tree has the real directory (repo-root libdisplay_producer/), which neither script syncs.
[ -f "$f" ] && [ "$(basename $f)" != libdisplay_producer ] || continue
b=$(basename $f); if ! cmp -s "$f" "$LIVE/src/backends/anland/$b"; then cp "$f" "$LIVE/src/backends/anland/$b"; echo "updated anland/$b"; fi
done
echo PREP_DONE
@@ -0,0 +1,21 @@
#!/bin/bash
# mgrun <surface: offscreen|server> <command...> - (container, as root) run one MobileGL client as the
# desktop user inside the running Plasma session. Installed as /usr/local/bin/mgrun (push-tools.sh).
# mgrun offscreen glmark2-es2-wayland -s 1280x720 # Wayland client, presents through KWin
# mgrun offscreen eglinfo -B ; mgrun offscreen glxinfo -B # glxinfo goes through Xwayland + GLX vendor
# "server" is only for a client that owns the Android Surface itself (KWin does); everything else is
# offscreen. MobileGL is the container's system-wide vendor, so nothing selects it here: EGL comes from
# /usr/share/glvnd/egl_vendor.d/10_mobilegl.json, transport/endpoint from /etc/mobilegl/client.conf,
# the backend from /etc/mobilegl/backend (must match the server), and the GLX vendor name / GBM backend
# from /etc/profile.d/mobilegl.sh (sourced by the login shell below). Client log:
# ${MGLOG:-/tmp/mgl-run.log} -> the library writes /tmp/mgl-run.client.log (truncated per run; give
# concurrent clients different MGLOG values). Env: MGLOG, DESKTOP_USER (swung0x48).
S=$1; shift
DU=${DESKTOP_USER:-swung0x48}
U=$(id -u "$DU")
cd /tmp
exec sudo -u "$DU" env XDG_RUNTIME_DIR=/run/user/$U WAYLAND_DISPLAY=wayland-0 DISPLAY=:0 \
XAUTHORITY=$(ls /run/user/$U/xauth_* 2>/dev/null | head -1) \
DBUS_SESSION_BUS_ADDRESS=unix:path=/run/user/$U/bus QT_QPA_PLATFORM=wayland \
MOBILEGL_IPC_SURFACE=$S MOBILEGL_LOG_FILE_PATH=${MGLOG:-/tmp/mgl-run.log} \
bash -lc 'exec "$@"' mgrun "$@"
@@ -0,0 +1,26 @@
#!/usr/bin/env bash
# ct.sh <channel> [script-file] - run a bash script as ROOT inside the container (login shell),
# reading the script from the file or from stdin. Generic form of the old c.sh/c2.sh/c3.sh.
#
# printf '%s\n' 'ls -la /tmp/*.log' | bash ct.sh mychan
# bash ct.sh mychan ./something.cmd
#
# How: the script is written to $ANL_DIR/<channel>.cmd, pushed to $DEV_TOOLS/<channel>.cmd together with
# a one-line <channel>.sh that does `droidspaces --name=<container> run bash -lc "$(cat <channel>.cmd)"`,
# and that runs under `su`. Quoting through adb + su + droidspaces is thereby avoided entirely.
#
# Pick a channel name of your own: other sessions on the same phone use c, c2, c3, xd, xr, gbm...;
# writing to their channel files races with them. CR characters are stripped (Windows checkouts).
# Inside, run user-side commands with `sudo -u <desktop user> env XDG_RUNTIME_DIR=/run/user/<uid> ...`.
# Never `pkill -f <word>` in a channel script: the `bash -lc "<script>"` process carries the whole
# script on its command line and kills itself. Use `pkill -x <name>`.
set -euo pipefail
. "$(dirname "$0")/env.sh"
ch=${1:?usage: ct.sh <channel> [script-file]}
src=${2:--}
if [ "$src" = - ]; then tr -d '\r' > "$ANL_DIR/$ch.cmd"; else tr -d '\r' < "$src" > "$ANL_DIR/$ch.cmd"; fi
printf '%s\n' "$DS_BIN --name=$MGL_CONTAINER run bash -lc \"\$(cat $DEV_TOOLS/$ch.cmd)\"" > "$ANL_DIR/$ch.sh"
adb_ shell "mkdir -p $DEV_TOOLS"
adb_ push "$ANL_DIR/$ch.cmd" "$DEV_TOOLS/$ch.cmd" >/dev/null
adb_ push "$ANL_DIR/$ch.sh" "$DEV_TOOLS/$ch.sh" >/dev/null
adb_ shell "su -c 'sh $DEV_TOOLS/$ch.sh'"
@@ -0,0 +1,19 @@
#!/usr/bin/env bash
# cyc.sh <shot-name> - one server iteration: clear Plasma's sticky software-renderer key, push the
# NDK-built server lib, swap it into the installed APK (device/swap.sh), restart app + Plasma session
# (device/run-plasma.sh, ~90 s), screenshot to $ANL_DIR/<shot-name>.png, print fatal logcat lines.
# Env: SERVER_LIB (default $MGL_WORKTREE/build-android/libMobileGL.so), WAIT (s after the session
# restart, default 15), N (logcat lines, default 20), CHANNEL (ct.sh channel, default cyc).
# Needs device/swap.sh + device/run-plasma.sh in $DEV_TOOLS (push-tools.sh). RESTARTS THE SESSION.
set -euo pipefail
here=$(cd "$(dirname "$0")" && pwd)
. "$here/env.sh"
shot=${1:?usage: cyc.sh <shot-name>}
LIB=${SERVER_LIB:-$MGL_WORKTREE/build-android/libMobileGL.so}
printf '%s\n' "sudo -u $MGL_DESKTOP_USER kwriteconfig6 --file kdeglobals --group QtQuickRendererSettings --key SceneGraphBackend --delete; rm -f /tmp/mgl-plasmashell.client.log" \
| bash "$here/ct.sh" "${CHANNEL:-cyc}"
adb_ push "$LIB" "$DEV_SHARE/libMobileGL.new.so" >/dev/null
adb_ logcat -c
adb_ shell "su -c 'sh $DEV_TOOLS/swap.sh >/dev/null && sh $DEV_TOOLS/run-plasma.sh >/dev/null 2>&1; sleep ${WAIT:-15}; screencap -p $DEV_SHARE/$shot.png'"
adb_ pull "$DEV_SHARE/$shot.png" "$ANL_DIR/$shot.png" >/dev/null && echo "screenshot: $ANL_DIR/$shot.png"
adb_ logcat -d | grep -E "Fatal signal|FATAL|Abort message|Magma wire buffer|E MobileGL" | cut -c1-260 | head -"${N:-20}" || true
@@ -0,0 +1,94 @@
#!/usr/bin/env bash
# desktop-verify.sh <phase>... - check that the MobileGL desktop starts, idles and stops by itself
# (anland app: foreground service + mobilegl-desktop.sh; container: desktop-session.service).
# Phases run in the order given; screenshots go to $ANL_DIR/dv-<phase>.png.
#
# push install device/*.sh into $DEV_TOOLS (push-tools.sh's device half)
# reset everything down: app force-stopped (its service with it), container stopped,
# experiment display daemon killed - the state after a phone reboot
# cold reset, then open the app like its icon does and time each piece (desktop-timeline.sh)
# open open the app without resetting (after a reboot: the reboot was the reset)
# chrome (re)start Chrome in the session on an animated WebGL page (kills a running Chrome)
# idle CPU/GPU for visible, hidden (HOME), locked (screen off), visible again (measure.sh)
# reopen swipe the app away (remove its task), check KWin survives, open it again, time it
# stop Stop desktop from the notification (the service's STOP action) and check it all ends
# rtest lock/unlock + window resize regression ($ANL_DIR/rtest.sh)
# Env: SECS (idle window, default 20), URL (chrome page).
set -euo pipefail
here=$(cd "$(dirname "$0")" && pwd)
. "$here/env.sh"
SECS=${SECS:-20}
URL=${URL:-https://webglsamples.org/aquarium/aquarium.html}
SOCK=$DEV_SHARE/display.sock
su_() { adb_ shell "su -c '$*'"; }
shot() {
su_ "screencap -p $DEV_SHARE/dv-$1.png"
adb_ pull "$DEV_SHARE/dv-$1.png" "$ANL_DIR/dv-$1.png" > /dev/null && echo "screenshot $ANL_DIR/dv-$1.png"
}
kwin_log() { # the anland backend's lines since <time>
printf '%s\n' "journalctl --no-pager --since '$1' 2>/dev/null | grep -a -E 'viewer (gone|back)|consumer (reconnected|disconnected)|failed to connect' | tail -${2:-8}" |
bash "$here/ct.sh" dv 2>/dev/null | grep -v pushed || true
}
task_id() {
su_ "am stack list" | tr -d '\r' | grep -m1 "$MGL_PKG/" | sed -n 's/.*taskId=\([0-9]*\).*/\1/p'
}
wake() {
su_ "input keyevent KEYCODE_WAKEUP"; sleep 1; su_ "wm dismiss-keyguard"; sleep 2
}
for phase in "$@"; do
echo "=== $phase"
case $phase in
push)
for f in "$here"/device/*.sh; do
tr -d '\r' < "$f" > "$ANL_DIR/push.tmp"
adb_ push "$ANL_DIR/push.tmp" "$DEV_TOOLS/$(basename "$f")" > /dev/null
done
rm -f "$ANL_DIR/push.tmp"; echo "device scripts in $DEV_TOOLS" ;;
reset)
su_ "sh $DEV_TOOLS/desktop-reset.sh" ;;
cold)
bash "$0" reset
wake
su_ "sh $DEV_TOOLS/desktop-timeline.sh 240"
sleep 3; shot cold ;;
open)
wake
su_ "sh $DEV_TOOLS/desktop-timeline.sh 300"
sleep 3; shot open ;;
chrome)
WAIT=15 bash "$here/chrome-launch.sh" dv "$URL" ;;
idle)
su_ "sh $DEV_TOOLS/measure.sh visible $SECS"
t=$(date -u '+%Y-%m-%d %H:%M:%S UTC')
su_ "input keyevent KEYCODE_HOME"; sleep 4
su_ "sh $DEV_TOOLS/measure.sh hidden $SECS"
su_ "am start -n $MGL_PKG/com.anland.consumer.MainActivity >/dev/null"; sleep 5
su_ "input keyevent KEYCODE_SLEEP"; sleep 4
su_ "sh $DEV_TOOLS/measure.sh locked $SECS"
wake
su_ "am start -n $MGL_PKG/com.anland.consumer.MainActivity >/dev/null"; sleep 5
su_ "sh $DEV_TOOLS/measure.sh visible2 $SECS"
shot idle
kwin_log "$t" 12 ;;
reopen)
t=$(date -u '+%Y-%m-%d %H:%M:%S UTC')
id=$(task_id); echo "removing task $id"
[ -n "$id" ] && su_ "am stack remove $id"
sleep 8
su_ "echo server: \$(pidof $MGL_PKG:mobilegl) ui: \$(pidof $MGL_PKG) kwin: \$(pidof kwin_wayland) plasmashell: \$(pidof plasmashell)"
su_ "sh $DEV_TOOLS/measure.sh closed 10"
su_ "sh $DEV_TOOLS/desktop-timeline.sh 120"
sleep 3; shot reopen
kwin_log "$t" ;;
stop)
su_ "am startservice -n $MGL_PKG/com.anland.consumer.MobileGLWorker -a com.anland.consumer.mobilegl.STOP" || true
sleep 25
su_ "echo server: \$(pidof $MGL_PKG:mobilegl) ui: \$(pidof $MGL_PKG); $DS_BIN show"
su_ "tail -3 $DEV_SHARE/desktop.log" ;;
rtest)
bash "$ANL_DIR/rtest.sh" dv ;;
*)
echo "unknown phase $phase" >&2; exit 2 ;;
esac
done
@@ -0,0 +1,7 @@
# bringup.sh - (Android root shell) bring the MobileGL desktop up after a phone reboot. There is
# nothing left to do by hand: opening the anland app (its launcher icon, or this am start) starts
# the display daemon, the MobileGL server, the container and its Plasma session. Prints the timeline.
# Run: su -c "sh /data/local/tmp/anl/bringup.sh" Env: as desktop-timeline.sh.
input keyevent KEYCODE_WAKEUP
wm dismiss-keyguard
sh "$(dirname "$0")/desktop-timeline.sh" 300
@@ -0,0 +1,17 @@
# desktop-reset.sh - (Android root shell) take the MobileGL desktop fully down, as after a phone
# reboot: the app force-stopped (its foreground service and server with it), the container
# stopped, the experiment display daemon killed. Leaves the original anland daemon alone.
# Run: su -c "sh /data/local/tmp/anl/desktop-reset.sh" Env: PKG, CONTAINER, SOCK, DS.
P=${PKG:-com.anland.consumer.mobilegl}
C=${CONTAINER:-arch-kde-mgl}
SOCK=${SOCK:-/data/local/tmp/anland-mobilegl/display.sock}
D=${DS:-/data/local/Droidspaces/bin/droidspaces}
am force-stop $P
$D --name=$C stop > /dev/null 2>&1
for p in $(pidof display_daemon); do
tr '\0' ' ' < /proc/$p/cmdline | grep -qF " $SOCK " && kill $p
done
rm -f $SOCK
sleep 2
echo "app: $(pidof $P) server: $(pidof $P:mobilegl) daemon: $(ls $SOCK 2>/dev/null)"
$D show
@@ -0,0 +1,30 @@
# desktop-timeline.sh [timeout-s] - (Android root shell) open the anland app the way its launcher
# icon does (am start, no extras) and print when each piece of the MobileGL desktop comes up,
# in seconds after the open: display daemon socket, the app's MobileGL server listening, the
# container, kwin_wayland, plasmashell, and the splash (ksplashqml) gone = desktop on screen.
# Pieces that already run show 0.0. Read-only apart from the am start.
# Run: su -c "sh /data/local/tmp/anl/desktop-timeline.sh 180" Env: PKG, CONTAINER, SOCK, DS.
T=${1:-180}
P=${PKG:-com.anland.consumer.mobilegl}
C=${CONTAINER:-arch-kde-mgl}
SOCK=${SOCK:-/data/local/tmp/anland-mobilegl/display.sock}
D=${DS:-/data/local/Droidspaces/bin/droidspaces}
now() { awk '{print $1}' /proc/uptime; }
since() { awk -v a="$1" -v b="$(now)" 'BEGIN{printf "%.1f", b - a}'; }
listening() { grep -q '00010000 0001 01 [0-9]* @anland-mobilegl$' /proc/net/unix; }
t0=$(now)
am start -n $P/com.anland.consumer.MainActivity > /dev/null
for k in daemon server container kwin shell splash_seen shown; do eval "t_$k="; done
while :; do
[ -z "$t_daemon" ] && [ -S "$SOCK" ] && t_daemon=$(since $t0) && echo "daemon socket +${t_daemon}s"
[ -z "$t_server" ] && listening && t_server=$(since $t0) && echo "server listening +${t_server}s"
[ -z "$t_container" ] && [ -n "$($D --name=$C pid 2>/dev/null | tr -dc 0-9)" ] && t_container=$(since $t0) && echo "container up +${t_container}s"
[ -z "$t_kwin" ] && pidof kwin_wayland > /dev/null && t_kwin=$(since $t0) && echo "kwin_wayland +${t_kwin}s"
[ -z "$t_shell" ] && pidof plasmashell > /dev/null && t_shell=$(since $t0) && echo "plasmashell +${t_shell}s"
[ -z "$t_splash_seen" ] && pidof ksplashqml > /dev/null && t_splash_seen=1
if [ -n "$t_shell" ] && [ -z "$t_shown" ] && ! pidof ksplashqml > /dev/null; then
t_shown=$(since $t0); echo "splash gone +${t_shown}s (desktop on screen)"; break
fi
[ "$(since $t0 | cut -d. -f1)" -ge "$T" ] && { echo "TIMEOUT after ${T}s"; break; }
sleep 0.3
done
@@ -0,0 +1,47 @@
# measure.sh <label> [seconds] - (Android root shell) CPU and GPU load of the MobileGL desktop over a
# window: % of ONE core (so 800 = all 8 cores) for the whole phone, KWin (+Xwayland), plasmashell,
# Chrome (all its processes), the app's :mobilegl server and its UI process, plus the mean of the
# kgsl gpu_busy_percentage samples (1/s). The container's processes are visible from Android's
# /proc (Droidspaces only gives them their own pid namespace).
# Run: su -c "sh /data/local/tmp/anl/measure.sh visible 20" Env: PKG.
L=${1:-sample}
N=${2:-20}
P=${PKG:-com.anland.consumer.mobilegl}
GPU=/sys/class/kgsl/kgsl-3d0/gpu_busy_percentage
pids_named() { # exact process names (comm or argv0 basename)
for n in "$@"; do pidof "$n"; done
}
pids_matching() { # argv contains the word
for d in /proc/[0-9]*; do
{ tr '\0' ' ' < $d/cmdline; } 2>/dev/null | grep -q "$1" && echo ${d#/proc/}
done
}
ticks() { # utime+stime of the pids, in clock ticks
t=0
for p in "$@"; do
[ -r /proc/$p/stat ] || continue
v=$(sed 's/^.*) //' /proc/$p/stat | awk '{print $12 + $13}')
t=$((t + v))
done
echo $t
}
total() { awk '/^cpu /{print $2+$3+$4+$6+$7+$8}' /proc/stat; }
KW="$(pids_named kwin_wayland Xwayland)"
SH="$(pids_named plasmashell)"
CH="$(pids_matching /opt/google/chrome/)"
SV="$(pidof $P:mobilegl)"
UI="$(pidof $P)"
a0=$(total); k0=$(ticks $KW); s0=$(ticks $SH); c0=$(ticks $CH); v0=$(ticks $SV); u0=$(ticks $UI)
g=0; i=0
while [ $i -lt $N ]; do
sleep 1
b=$(cat $GPU 2>/dev/null | tr -dc '0-9'); g=$((g + ${b:-0})); i=$((i + 1))
done
a1=$(total); k1=$(ticks $KW); s1=$(ticks $SH); c1=$(ticks $CH); v1=$(ticks $SV); u1=$(ticks $UI)
# CLK_TCK is 100: ticks per second == percent of one core.
pct() { echo $(( ($2 - $1) / N )); }
printf '%-10s %3ss phone %4s%% kwin %3s%% plasmashell %3s%% chrome %3s%% server %3s%% app-ui %3s%% gpu %3s%%\n' \
"$L" "$N" "$(pct $a0 $a1)" "$(pct $k0 $k1)" "$(pct $s0 $s1)" "$(pct $c0 $c1)" "$(pct $v0 $v1)" "$(pct $u0 $u1)" "$((g / N))"
@@ -0,0 +1,20 @@
# restart-daemon.sh - (Android root shell) (re)start the EXPERIMENT's anland display daemon on $SOCK.
# It is not started at boot (the anland-daemon module's service.sh only starts the original daemon on
# /data/local/tmp/display_daemon.sock, which must be left alone). Without it KWin logs
# "failed to connect to display daemon at /run/anland-mobilegl/display.sock" and the session dies in ~3 s.
# The container sees $SOCK as /run/anland-mobilegl/display.sock (bind mount in container.config);
# uid 1000 in the container must connect, hence 777 dir / 666 socket.
# Env: SOCK, DAEMON (/data/adb/modules/anland-daemon/display_daemon).
SOCK=${SOCK:-/data/local/tmp/anland-mobilegl/display.sock}
DAEMON=${DAEMON:-/data/adb/modules/anland-daemon/display_daemon}
DIR=$(dirname $SOCK)
for p in $(ps -A -o PID,ARGS | grep "display_daemon $SOCK" | grep -v grep | awk '{print $1}'); do kill $p; done
sleep 1
rm -f $SOCK
mkdir -p $DIR
nohup $DAEMON $SOCK > $DIR/display-daemon.log 2>&1 &
sleep 1
chmod 777 $DIR
chmod 666 $SOCK
echo "daemon restarted:"
ps -A -o PID,ARGS | grep "display_daemon $SOCK" | grep -v grep
@@ -0,0 +1,27 @@
# run-plasma.sh - (Android root shell) DEVELOPER restart of the whole MobileGL desktop, e.g. after a
# server, client or KWin rebuild: ends the Plasma session (and the user's logind session, which takes
# a while to wind down), then restarts the anland app; opening it brings up the server, starts
# desktop-session.service again and the session waits for the server. Normal use needs none of this:
# opening the app is enough. ~60 s. Prints the unit status, plasma user units and recent coredumps.
# Run: su -c "sh /data/local/tmp/anl/run-plasma.sh"
# Env: PKG, CONTAINER, DESKTOP_USER (swung0x48), DS (droidspaces binary).
D=${DS:-/data/local/Droidspaces/bin/droidspaces}
P=${PKG:-com.anland.consumer.mobilegl}
C=${CONTAINER:-arch-kde-mgl}
U=${DESKTOP_USER:-swung0x48}
$D --name=$C run bash -lc "
systemctl stop desktop-session.service 2>/dev/null
loginctl terminate-user $U 2>/dev/null; sleep 30
systemctl reset-failed desktop-session.service
rm -f /tmp/mobilegl-compositor.log
date +%s > /tmp/plasma-start
"
am force-stop $P
am start -n $P/com.anland.consumer.MainActivity > /dev/null
sleep 30
$D --name=$C run bash -lc "
systemctl --no-pager status desktop-session.service | head -8
UID_=\$(id -u $U)
sudo -u $U XDG_RUNTIME_DIR=/run/user/\$UID_ systemctl --user --no-pager list-units --all 'plasma*' 2>&1 | head -30
coredumpctl --no-pager list --since '-2min' 2>&1 | tail -5
"
@@ -0,0 +1,13 @@
# status.sh - (Android root shell) READ-ONLY health check of the whole stack.
# Run: su -c "sh /data/local/tmp/anl/status.sh" Env: PKG, CONTAINER, DS, SOCK.
D=${DS:-/data/local/Droidspaces/bin/droidspaces}
P=${PKG:-com.anland.consumer.mobilegl}
C=${CONTAINER:-arch-kde-mgl}
SOCK=${SOCK:-/data/local/tmp/anland-mobilegl/display.sock}
echo "battery: $(dumpsys battery | grep ' level:' | tr -d ' ') $(dumpsys power | grep -o 'mStayOn=[a-z]*')"
echo "backend: app setting=$(cat /data/data/$P/files/mobilegl-backend 2>/dev/null || echo 'DirectGLES (default)') published=$(cat ${SOCK%/*}/backend 2>/dev/null) container /etc/mobilegl/backend=$(cat /mnt/Droidspaces/$C/etc/mobilegl/backend 2>/dev/null) prop override=$(getprop debug.mobilegl.backend)"
echo "desktop service log: $(tail -1 ${SOCK%/*}/desktop.log 2>/dev/null)"
echo "app pid: $(pidof $P) server (:mobilegl) pid: $(pidof $P:mobilegl)"
ps -A -o PID,ARGS | grep "display_daemon $SOCK" | grep -v grep || echo "experiment display daemon NOT running"
$D show
$D --name=$C run bash -lc 'echo "desktop-session: $(systemctl is-active desktop-session.service)"; pgrep -a -x kwin_wayland; pgrep -a -x plasmashell; grep -A1 QtQuickRendererSettings /home/*/.config/kdeglobals 2>/dev/null || echo "kdeglobals: no sticky software renderer"'
@@ -0,0 +1,17 @@
# swap.sh - (Android root shell) replace the server libMobileGL.so inside the INSTALLED anland APK with
# $SHARE/libMobileGL.new.so, keeping owner, mode and SELinux context; force-stops the app. The first
# run backs the APK's original lib up to $SHARE/libMobileGL.so.orig. No re-signing/reinstall (that
# would change the app UID and lose its root grant). Run: su -c "sh /data/local/tmp/anl/swap.sh"
# Env: PKG (com.anland.consumer.mobilegl), SHARE (/data/local/tmp/anland-mobilegl).
set -e
P=${PKG:-com.anland.consumer.mobilegl}
SHARE=${SHARE:-/data/local/tmp/anland-mobilegl}
D=$(dirname $(pm path $P | sed 's/package://'))/lib/arm64
ls -laZ $D
[ -f $SHARE/libMobileGL.so.orig ] || cp $D/libMobileGL.so $SHARE/libMobileGL.so.orig
am force-stop $P
CTX=$(ls -Z $D/libMobileGL.so | awk '{print $1}')
OWN=$(stat -c %u:%g $D/libMobileGL.so)
cp $SHARE/libMobileGL.new.so $D/libMobileGL.so
chown $OWN $D/libMobileGL.so; chmod 755 $D/libMobileGL.so; chcon $CTX $D/libMobileGL.so
ls -laZ $D/libMobileGL.so
@@ -0,0 +1,17 @@
# switch.sh <DirectGLES|DirectVulkan> - (Android root shell) change the MobileGL backend the anland
# app's server runs (Espryt = DirectGLES, Magma = DirectVulkan). The choice is the app's setting
# (Settings > Connection > MobileGL desktop, saved in the app's files/mobilegl-backend); this does
# what a user does there: save it, Stop desktop, open the app again. The new server publishes the
# backend and the container's session copies it to /etc/mobilegl/backend before it starts, so
# server and clients agree. ~30-60 s. No setprop: debug.mobilegl.backend, if set, still overrides
# the setting until the next reboot - this clears it.
# Env: PKG, SOCK.
B=${1:?usage: switch.sh DirectGLES|DirectVulkan}
P=${PKG:-com.anland.consumer.mobilegl}
[ -n "$(getprop debug.mobilegl.backend)" ] && setprop debug.mobilegl.backend ""
if pidof $P:mobilegl > /dev/null; then
am startservice -n $P/com.anland.consumer.MobileGLWorker -a com.anland.consumer.mobilegl.STOP > /dev/null
i=0; while pidof $P:mobilegl > /dev/null && [ $i -lt 60 ]; do sleep 1; i=$((i + 1)); done
fi
am start -n $P/com.anland.consumer.MainActivity --es mobilegl_backend $B > /dev/null
echo "backend $B saved; the desktop is starting (watch: sh $(dirname $0)/status.sh)"
@@ -0,0 +1,8 @@
# unlock.sh - (Android root shell) ask logind to unlock the desktop user's session. UNVERIFIED that this
# helps here: the handoff notes this logind session type does not support lock-session (a real lock
# screen then needs the password). Prevention: [Daemon] Autolock=false in ~/.config/kscreenlockerrc.
# Env: DS, CONTAINER, DESKTOP_USER.
D=${DS:-/data/local/Droidspaces/bin/droidspaces}
C=${CONTAINER:-arch-kde-mgl}
U=${DESKTOP_USER:-swung0x48}
$D --name=$C run bash -lc "S=\$(loginctl list-sessions --no-legend | awk '\$3==\"$U\"{print \$1}' | head -1); loginctl unlock-session \$S; echo unlocked \$S"
@@ -0,0 +1,47 @@
# x11-measure.sh <label> <app-process-name> [seconds] - (Android root shell) CPU of the X11 path while
# an X11 app runs: % of ONE core (800 = all 8 cores) for the whole phone, the app, Xwayland,
# kwin_wayland, the app's :mobilegl server; mean kgsl gpu_busy_percentage; and the number of dma-buf
# descriptors Xwayland and KWin hold (window pixmaps / client buffers as dma-bufs). Refuses to run
# below 40% battery: the phone drains faster under load than USB charges it.
# Run: su -c "sh /data/local/tmp/anl/x11-measure.sh glamor glxgears 10" Env: PKG, MIN_BATTERY.
L=${1:-sample}
APP=${2:-glxgears}
N=${3:-10}
P=${PKG:-com.anland.consumer.mobilegl}
GPU=/sys/class/kgsl/kgsl-3d0/gpu_busy_percentage
level=$(dumpsys battery | sed -n 's/^ *level: *//p')
if [ "${level:-0}" -lt "${MIN_BATTERY:-40}" ]; then
echo "battery ${level}% < ${MIN_BATTERY:-40}%: not loading the phone; charge first" >&2
exit 3
fi
ticks() {
t=0
for p in "$@"; do
[ -r /proc/$p/stat ] || continue
v=$(sed 's/^.*) //' /proc/$p/stat | awk '{print $12 + $13}')
t=$((t + v))
done
echo $t
}
total() { awk '/^cpu /{print $2+$3+$4+$6+$7+$8}' /proc/stat; }
dmabufs() { # dma-buf descriptors held by the pids (Android kernels link them to /dmabuf:...)
n=0
for p in "$@"; do n=$((n + $(ls -l /proc/$p/fd 2>/dev/null | grep -c dmabuf))); done
echo $n
}
AP="$(pidof $APP)"
XW="$(pidof Xwayland)"
KW="$(pidof kwin_wayland)"
SV="$(pidof $P:mobilegl)"
[ -n "$AP" ] || { echo "no running process named $APP" >&2; exit 2; }
a0=$(total); p0=$(ticks $AP); x0=$(ticks $XW); k0=$(ticks $KW); v0=$(ticks $SV)
g=0; i=0
while [ $i -lt $N ]; do
sleep 1
b=$(cat $GPU 2>/dev/null | tr -dc '0-9'); g=$((g + ${b:-0})); i=$((i + 1))
done
a1=$(total); p1=$(ticks $AP); x1=$(ticks $XW); k1=$(ticks $KW); v1=$(ticks $SV)
pct() { echo $(( ($2 - $1) / N )); }
printf '%-12s %3ss phone %4s%% %s %3s%% Xwayland %3s%% kwin %3s%% server %3s%% gpu %3s%% dmabufs Xwayland %s kwin %s battery %s%%\n' \
"$L" "$N" "$(pct $a0 $a1)" "$APP" "$(pct $p0 $p1)" "$(pct $x0 $x1)" "$(pct $k0 $k1)" "$(pct $v0 $v1)" "$((g / N))" \
"$(dmabufs $XW)" "$(dmabufs $KW)" "$level"
@@ -0,0 +1,41 @@
# env.sh - shared parameters for the host-side (Git Bash) scripts of the anland-mobilegl-plasma skill.
# Sourced, never run. Every value can be overridden from the environment before calling a script.
#
# MGL_WORKTREE MobileGL worktree (Windows form C:/...). Default: the worktree this skill lives in.
# ANL_DIR Local scratch dir: channel files, screenshots, xdist/. Default: %LOCALAPPDATA%/Temp/anl
# ANDROID_NDK NDK used for the Android server lib. Default: <SDK>/ndk/27.2.12479018
# MGL_PKG Anland APK that embeds the server. Default: com.anland.consumer.mobilegl
# MGL_CONTAINER Droidspaces container. Default: arch-kde-mgl
# MGL_DESKTOP_USER Desktop user inside the container. Default: swung0x48
# DS_BIN droidspaces binary on the phone. Default: /data/local/Droidspaces/bin/droidspaces
# DEV_TOOLS Phone dir for device/*.sh and channel files. Default: /data/local/tmp/anl
# DEV_SHARE Phone dir bind-mounted into the container as /run/anland-mobilegl (screenshots,
# server lib staging, display daemon socket). Default: /data/local/tmp/anland-mobilegl
# ANDROID_SERIAL Honoured by adb itself; leave unset when only one device is attached
# (USB serial or a wireless "ip:port").
_skill_scripts=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
if [ -z "${MGL_WORKTREE:-}" ]; then
MGL_WORKTREE=$(cd "$_skill_scripts/../../../.." && pwd -W 2>/dev/null || pwd)
fi
: "${ANL_DIR:=$(cygpath -m "${LOCALAPPDATA:-C:/Users/$USERNAME/AppData/Local}" 2>/dev/null || echo /tmp)/Temp/anl}"
: "${ANDROID_NDK:=$(cygpath -m "${LOCALAPPDATA:-C:/Users/$USERNAME/AppData/Local}" 2>/dev/null)/Android/Sdk/ndk/27.2.12479018}"
: "${MGL_PKG:=com.anland.consumer.mobilegl}"
: "${MGL_CONTAINER:=arch-kde-mgl}"
: "${MGL_DESKTOP_USER:=swung0x48}"
: "${DS_BIN:=/data/local/Droidspaces/bin/droidspaces}"
: "${DEV_TOOLS:=/data/local/tmp/anl}"
: "${DEV_SHARE:=/data/local/tmp/anland-mobilegl}"
# The container rootfs as Android root sees it (the container's /tmp is NOT visible here).
CT_ROOT=/mnt/Droidspaces/$MGL_CONTAINER
export MGL_WORKTREE ANL_DIR ANDROID_NDK MGL_PKG MGL_CONTAINER MGL_DESKTOP_USER DS_BIN DEV_TOOLS DEV_SHARE CT_ROOT
# Git Bash would rewrite /data/... arguments into C:/Program Files/Git/data/...; keep them literal.
# Consequence: pass LOCAL paths to Windows tools (adb, cmake, git) in C:/ form, never /c/.
export MSYS_NO_PATHCONV=1
adb_() { MSYS2_ARG_CONV_EXCL='*' adb "$@"; }
# C:/x/y -> /mnt/c/x/y (for WSL).
win2wsl() { local p=$1 d; d=$(printf '%s' "${p:0:1}" | tr 'A-Z' 'a-z'); printf '/mnt/%s%s' "$d" "${p:2}"; }
mkdir -p "$ANL_DIR"
@@ -0,0 +1,146 @@
#!/usr/bin/env bash
# kwin-recover.sh [rounds] - does the desktop keep working when KWin's OWN MobileGL session is lost?
# Starts animated clients (glmark2-es2-wayland: Wayland, zero-copy shared images; glxgears: X11 through
# Xwayland's GLX; with CHROME=1 also Chrome on an animated WebGL page), then per round loses the server
# session of KWin with the pid-targeted debug knob (debug.mobilegl.inject_device_lost_pid, KWin's pid as
# it sees itself in the container), waits, and checks. A round PASSES when:
# - the server latched KWin's session (the server log file: "injected:pid=<kwin>"; logcat is too small);
# - KWin, plasmashell and every test client are the SAME processes (nothing restarted);
# - KWin restarted compositing on a fresh session (its journal: "compositing restarted on a fresh
# MobileGL session"; its client log: one more "RECOVERED");
# - the screen still animates (shot.sh hashes differ) and the clients keep presenting;
# - no native crash (logcat -b crash).
# Between rounds the knob is reset to 0 while the clients run, which re-arms it for the same pid.
# Stops early below MIN_BATTERY percent (the phone drains faster under load than USB charges it).
#
# Deploy first (same MobileGL commit on both sides; see SKILL.md):
# bash scripts/build-android.sh && bash scripts/cyc.sh pre # server -> APK, restarts the session
# bash scripts/xbuild.sh && bash scripts/xdeploy.sh # client -> container
# ANLAND_WORKTREE=<anland worktree> bash scripts/kwin-stage.sh build # kwin.patch + anland backend
# (wait for "== ALL DONE ==" in /root/kwin-build.log), then: adb shell 'su -c "sh /data/local/tmp/anl/run-plasma.sh"'
# Env: ROUNDS (default 2), SETTLE (s after the loss, default 10), REARM (s with the knob at 0 between
# rounds, default 4), MIN_BATTERY (default 40), CHROME (1: add Chrome), CHANNEL (ct.sh channel,
# default krec), TAG (screenshot/log tag, default krec).
set -euo pipefail
here=$(cd "$(dirname "$0")" && pwd)
. "$here/env.sh"
ROUNDS=${1:-${ROUNDS:-2}}
SETTLE=${SETTLE:-10}
REARM=${REARM:-4}
MIN_BATTERY=${MIN_BATTERY:-40}
CH=${CHANNEL:-krec}
TAG=${TAG:-krec}
COMPOSITOR_LOG=/tmp/mobilegl-compositor.client.log
SERVER_LOG=/data/data/$MGL_PKG/files/mobilegl-server.server.log
ct() { bash "$here/ct.sh" "$CH"; }
battery() { adb_ shell dumpsys battery | tr -d '\r' | awk '/ level:/ {print $2}'; }
stop_clients() {
ct <<'IN' || true
pkill -x glmark2-es2-way 2>/dev/null; pkill -x glmark2-es2-wayland 2>/dev/null; pkill -x glxgears 2>/dev/null
true
IN
}
cleanup() {
adb_ shell "su -c 'setprop debug.mobilegl.inject_device_lost_pid 0'" || true
stop_clients
}
check_battery() {
local level; level=$(battery)
echo "battery: $level%"
if [ -n "$level" ] && [ "$level" -lt "$MIN_BATTERY" ]; then
echo "STOP: battery below $MIN_BATTERY%"; exit 3
fi
}
trap cleanup EXIT
check_battery
adb_ shell "su -c 'sh $DEV_TOOLS/status.sh'" || true
adb_ shell "su -c 'setprop debug.mobilegl.inject_device_lost_pid 0'"
# The clients, each with its own log (mgrun's MGLOG), in the background for the whole run.
ct <<'IN'
rm -f /tmp/mgl-krec-gm.client.log /tmp/mgl-krec-gx.client.log /tmp/krec-gm.out /tmp/krec-gx.out
MGLOG=/tmp/mgl-krec-gm.log nohup timeout 900 mgrun offscreen glmark2-es2-wayland -s 960x540 --run-forever \
> /tmp/krec-gm.out 2>&1 < /dev/null &
MGLOG=/tmp/mgl-krec-gx.log nohup timeout 900 mgrun offscreen glxgears -info > /tmp/krec-gx.out 2>&1 < /dev/null &
sleep 8
echo "glmark2: $(pgrep -x glmark2-es2-way || pgrep -x glmark2-es2-wayland) glxgears: $(pgrep -x glxgears)"
grep -aE "presented through|could not be presented" /tmp/mgl-krec-gm.client.log | head -2
grep -aE "presented through|readback" /tmp/mgl-krec-gx.client.log | head -2
IN
if [ "${CHROME:-0}" = 1 ]; then
ct <<'IN'
cat > /tmp/mgl-krec-spin.html <<'HTML'
<!doctype html><html><body style="margin:0;background:#000">
<canvas id=c width=640 height=360 style="width:100vw;height:100vh"></canvas>
<script>
const gl = document.getElementById('c').getContext('webgl'); let t = 0;
function frame() { t++; if (gl && !gl.isContextLost()) { gl.clearColor((t % 60) / 60, 0.2, 0.6, 1); gl.clear(gl.COLOR_BUFFER_BIT); }
requestAnimationFrame(frame); }
frame();
</script></body></html>
HTML
chmod 644 /tmp/mgl-krec-spin.html
IN
WAIT=${WAIT:-20} CHANNEL=$CH bash "$here/chrome-launch.sh" "$TAG" --new-window "file:///tmp/mgl-krec-spin.html"
fi
# KWin, plasmashell and the clients, as pids inside the container; how often KWin recovered so far.
snapshot() {
ct <<IN | tr -d '\r' | tail -1
k=\$(pgrep -x kwin_wayland | head -1); p=\$(pgrep -x plasmashell | head -1)
g=\$(pgrep -x glmark2-es2-way || pgrep -x glmark2-es2-wayland); x=\$(pgrep -x glxgears | head -1)
c=\$(pgrep -x chrome | head -1)
r=\$(grep -ac "RECOVERED" $COMPOSITOR_LOG 2>/dev/null || true)
f=\$(grep -ac "FPS" /tmp/krec-gm.out 2>/dev/null || true)
echo "\${k:-none} \${p:-none} \${g:-none} \${x:-none} \${c:-none} \${r:-0} \${f:-0}"
IN
}
pass=0
for round in $(seq 1 "$ROUNDS"); do
echo "=== round $round/$ROUNDS"
check_battery
read -r kwin plasma gm gx chrome recovered fps < <(snapshot)
echo "kwin=$kwin plasmashell=$plasma glmark2=$gm glxgears=$gx chrome=$chrome recoveries=$recovered fps-lines=$fps"
if [ "$kwin" = none ]; then echo "FAIL: KWin is not running"; break; fi
bash "$here/shot.sh" "$TAG-r$round-pre" 2 1
srvlines=$(adb_ shell "su -c 'wc -l < $SERVER_LOG'" | tr -dc '0-9')
adb_ logcat -b crash -c || true
adb_ shell "su -c 'setprop debug.mobilegl.inject_device_lost_pid $kwin'"
sleep "$SETTLE"
adb_ shell "su -c 'setprop debug.mobilegl.inject_device_lost_pid 0'"
bash "$here/shot.sh" "$TAG-r$round-post" 3 1 | tee "$ANL_DIR/$TAG-r$round-post.md5"
read -r kwin2 plasma2 gm2 gx2 chrome2 recovered2 fps2 < <(snapshot)
echo "after: kwin=$kwin2 plasmashell=$plasma2 glmark2=$gm2 glxgears=$gx2 chrome=$chrome2 recoveries=$recovered2 fps-lines=$fps2"
adb_ shell "su -c 'tail -n +$((srvlines + 1)) $SERVER_LOG'" > "$ANL_DIR/$TAG-r$round-server.log"
latched=$(grep -c "injected:pid=$kwin" "$ANL_DIR/$TAG-r$round-server.log" || true)
grep -aE "inject_device_lost_pid|SessionLatch|session #[0-9]+ (started|pid)|owner=server|lease" \
"$ANL_DIR/$TAG-r$round-server.log" | cut -c1-220 | head -14 || true
ct <<IN || true
echo "--- KWin's MobileGL client log:"; grep -aE "DEVICE LOST|RECOVERED|GL_UNKNOWN_CONTEXT_RESET|re-created on the fresh|Refuse" $COMPOSITOR_LOG | tail -6
echo "--- KWin journal:"; journalctl --no-pager --since -40s 2>/dev/null | grep -aE "kwin_wayland" | grep -aiE "reset|restarted|MobileGL swap|context|fatal|abort" | tail -10
echo "--- clients:"; grep -aE "DEVICE LOST|lost|Fatal" /tmp/mgl-krec-gm.client.log /tmp/mgl-krec-gx.client.log | tail -4
IN
crashes=$(adb_ logcat -b crash -d | grep -c "Fatal signal\|Abort message" || true)
hashes=$(awk '{print $1}' "$ANL_DIR/$TAG-r$round-post.md5" | sort -u | wc -l)
verdict=PASS
[ "$latched" -ge 1 ] || { echo " - the server never latched KWin's session (pid $kwin)"; verdict=FAIL; }
[ "$kwin2" = "$kwin" ] || { echo " - KWin is a different process ($kwin -> $kwin2): it restarted"; verdict=FAIL; }
[ "$plasma2" = "$plasma" ] || { echo " - plasmashell restarted ($plasma -> $plasma2)"; verdict=FAIL; }
[ "$gm2" = "$gm" ] && [ "$gx2" = "$gx" ] || { echo " - a test client restarted or died"; verdict=FAIL; }
[ "${CHROME:-0}" != 1 ] || [ "$chrome2" = "$chrome" ] || { echo " - Chrome restarted"; verdict=FAIL; }
[ "$recovered2" -gt "$recovered" ] || { echo " - KWin's client log shows no new RECOVERED session"; verdict=FAIL; }
[ "$fps2" -gt "$fps" ] || { echo " - glmark2 stopped reporting frames"; verdict=FAIL; }
[ "$hashes" -ge 2 ] || { echo " - the screen does not change after the loss (frozen)"; verdict=FAIL; }
[ "$crashes" -eq 0 ] || { echo " - native crash(es):"; adb_ logcat -b crash -d | tail -8; verdict=FAIL; }
echo "round $round: $verdict"
[ "$verdict" = PASS ] && pass=$((pass + 1))
[ "$verdict" = PASS ] || break
# Re-arm the knob for the same pid: a session must observe 0 before the next value counts.
sleep "$REARM"
done
echo "=== $pass/$ROUNDS rounds passed; screenshots in $ANL_DIR/$TAG-r*-*.png"
[ "$pass" = "$ROUNDS" ]
@@ -0,0 +1,46 @@
#!/usr/bin/env bash
# kwin-stage.sh [build] - stage anland's KWin changes (kwin.patch, anland backend sources,
# mobilegl-startup.sh, sync-build-kwin.sh) for the in-container KWin rebuild; with "build", also run
# kwprep.sh (apply changed patch hunks / backend files to the live tree) and start sync-build-kwin.sh
# (ninja -j4 + atomic install into /opt/mobilegl/kwin, screencast plugin, startup script) in the
# background, log /root/kwin-build.log. KWin is NOT restarted: run device/run-plasma.sh afterwards.
# Poll: printf 'tail -5 /root/kwin-build.log\n' | bash ct.sh kw
#
# Two staging places, because the two container scripts read different ones:
# /root/kwsync.tgz <- kwprep.sh (installed by push-tools.sh)
# $DEV_SHARE/kwin-sync = /run/anland-mobilegl/kwin-sync <- sync-build-kwin.sh (as kwin-sync/sync-build.sh)
# All files are CR-stripped: the Windows checkout of anland has CRLF backend sources and a CRLF
# sync-build-kwin.sh (bash then fails on $'\r').
# Assembled from kwprep.sh + sync-build-kwin.sh; not exercised as one unit yet.
# Env: ANLAND_WORKTREE (anland checkout on branch legacy-mobilegl-unified, Windows form),
# CHANNEL (ct.sh channel, default kw).
set -euo pipefail
here=$(cd "$(dirname "$0")" && pwd)
. "$here/env.sh"
: "${ANLAND_WORKTREE:=$(cygpath -m "$USERPROFILE" 2>/dev/null)/AndroidStudioProjects/anland/.claude/worktrees/mgl-unified}"
K=$ANLAND_WORKTREE/producers/kde
st=$ANL_DIR/kwstage
rm -rf "$st"; mkdir -p "$st/kwsync/Arch_v5" "$st/kwsync/anland_backend_Arch_v5/src/backends/anland" "$st/kwin-sync/anland" "$st/kwin-sync/misc"
lf() { tr -d '\r' < "$1" > "$2"; }
for f in "$K"/anland_backend_Arch_v5/src/backends/anland/*; do
b=$(basename "$f"); [ "$b" = libdisplay_producer ] && continue
lf "$f" "$st/kwsync/anland_backend_Arch_v5/src/backends/anland/$b"
lf "$f" "$st/kwin-sync/anland/$b"
done
for f in kwin.patch mobilegl-startup.sh desktop-session-mobilegl.conf; do
lf "$K/Arch_v5/$f" "$st/kwsync/Arch_v5/$f"
lf "$K/Arch_v5/$f" "$st/kwin-sync/misc/$f"
done
lf "$K/Arch_v5/sync-build-kwin.sh" "$st/kwin-sync/sync-build.sh"
(cd "$st/kwsync" && tar --force-local -czf ../kwsync.tgz Arch_v5 anland_backend_Arch_v5)
tar --force-local -C "$st" -czf "$st/kwin-sync.tgz" kwin-sync
adb_ push "$st/kwsync.tgz" "$DEV_TOOLS/kwsync.tgz" >/dev/null
adb_ push "$st/kwin-sync.tgz" "$DEV_TOOLS/kwin-sync.tgz" >/dev/null
adb_ shell "su -c 'cp $DEV_TOOLS/kwsync.tgz $CT_ROOT/root/kwsync.tgz && tar xzf $DEV_TOOLS/kwin-sync.tgz -C $DEV_SHARE && chmod -R a+rX $DEV_SHARE/kwin-sync && rm -f $DEV_TOOLS/kwsync.tgz $DEV_TOOLS/kwin-sync.tgz && echo staged'"
if [ "${1:-}" = build ]; then
bash "$here/ct.sh" "${CHANNEL:-kw}" <<'IN'
bash /root/kwprep.sh || exit 1
nohup bash /run/anland-mobilegl/kwin-sync/sync-build.sh > /root/kwin-build.log 2>&1 < /dev/null &
echo "kwin build started (pid $!), log /root/kwin-build.log; it ends with '== ALL DONE =='"
IN
fi
@@ -0,0 +1,94 @@
#!/usr/bin/env python3
# cdpmedia.py [port] [seconds] - ask a Chrome started with --remote-debugging-port=<port> (and a
# non-default --user-data-dir) which video decoder its first page's players use, and evaluate
# `window.__mglStats` / the element `#t` text of the page. Standard library only (no websocket
# module in the container): a minimal RFC 6455 client.
# python3 cdpmedia.py 9333 6
# Prints kVideoDecoderName / kIsPlatformVideoDecoder / kVideoTracks properties and media errors as
# they arrive, then the page's stats text.
import base64, json, os, socket, struct, sys, time, urllib.request
port = int(sys.argv[1]) if len(sys.argv) > 1 else 9333
seconds = float(sys.argv[2]) if len(sys.argv) > 2 else 6.0
targets = json.load(urllib.request.urlopen(f"http://127.0.0.1:{port}/json"))
page = next(t for t in targets if t.get("type") == "page")
url = page["webSocketDebuggerUrl"]
path = url.split(f"127.0.0.1:{port}", 1)[1]
sock = socket.create_connection(("127.0.0.1", port))
key = base64.b64encode(os.urandom(16)).decode()
sock.sendall((f"GET {path} HTTP/1.1\r\nHost: 127.0.0.1:{port}\r\nUpgrade: websocket\r\nConnection: Upgrade\r\n"
f"Sec-WebSocket-Key: {key}\r\nSec-WebSocket-Version: 13\r\n\r\n").encode())
response = b""
while b"\r\n\r\n" not in response:
response += sock.recv(4096)
buffer = response.split(b"\r\n\r\n", 1)[1]
def send(message):
data = json.dumps(message).encode()
header = bytes([0x81])
mask = os.urandom(4)
if len(data) < 126:
header += bytes([0x80 | len(data)])
elif len(data) < 65536:
header += bytes([0x80 | 126]) + struct.pack(">H", len(data))
else:
header += bytes([0x80 | 127]) + struct.pack(">Q", len(data))
sock.sendall(header + mask + bytes(b ^ mask[i % 4] for i, b in enumerate(data)))
def recv_exact(n):
global buffer
while len(buffer) < n:
chunk = sock.recv(65536)
if not chunk:
raise EOFError
buffer += chunk
out, buffer = buffer[:n], buffer[n:]
return out
def recv():
first, second = recv_exact(2)
length = second & 0x7F
if length == 126:
length = struct.unpack(">H", recv_exact(2))[0]
elif length == 127:
length = struct.unpack(">Q", recv_exact(8))[0]
payload = recv_exact(length)
return json.loads(payload) if first & 0x0F == 1 else None
send({"id": 1, "method": "Media.enable"})
# Players created before the domain was enabled report nothing: reload so the page makes new ones.
if os.environ.get("CDP_NAVIGATE"):
send({"id": 3, "method": "Page.navigate", "params": {"url": os.environ["CDP_NAVIGATE"]}})
elif os.environ.get("CDP_RELOAD", "1") == "1":
send({"id": 3, "method": "Page.reload"})
deadline = time.time() + seconds
sock.settimeout(0.5)
seen = set()
while time.time() < deadline:
try:
message = recv()
except socket.timeout:
continue
if not message or "method" not in message:
continue
if message["method"] == "Media.playerPropertiesChanged":
for prop in message["params"]["properties"]:
if prop["name"] in ("kVideoDecoderName", "kIsPlatformVideoDecoder", "kVideoTracks", "kResolution"):
line = f"{prop['name']} = {prop['value']}"
if line not in seen:
seen.add(line)
print(line)
elif message["method"] in ("Media.playerErrorsRaised",):
print("media error:", json.dumps(message["params"])[:300])
send({"id": 2, "method": "Runtime.evaluate",
"params": {"expression": os.environ.get("CDP_EXPR", "(document.getElementById('t')||{}).textContent"), "returnByValue": True}})
sock.settimeout(3)
while True:
message = recv()
if message and message.get("id") == 2:
print("page:", message.get("result", {}).get("result", {}).get("value"))
break
@@ -0,0 +1,34 @@
// dmahash <pid> : for each dma-buf fd of <pid> at least 1 MB, print inode, size and a sampled checksum.
#include <dirent.h>
#include <fcntl.h>
#include <stdint.h>
#include <stdio.h>
#include <string.h>
#include <sys/mman.h>
#include <sys/stat.h>
#include <unistd.h>
int main(int argc, char** argv) {
char dir[64];
snprintf(dir, sizeof dir, "/proc/%s/fd", argv[1]);
DIR* d = opendir(dir);
if (!d) { perror("opendir"); return 1; }
struct dirent* e;
while ((e = readdir(d))) {
char p[128], l[128] = {0};
snprintf(p, sizeof p, "%s/%s", dir, e->d_name);
if (readlink(p, l, sizeof l - 1) <= 0 || strncmp(l, "/dmabuf", 7) != 0) continue;
int fd = open(p, O_RDONLY);
if (fd < 0) continue;
struct stat st; fstat(fd, &st);
if (st.st_size < (1 << 20)) { close(fd); continue; }
const uint8_t* m = mmap(NULL, st.st_size, PROT_READ, MAP_SHARED, fd, 0);
if (m == MAP_FAILED) { printf("fd %s ino %lu mmap failed\n", e->d_name, (unsigned long)st.st_ino); close(fd); continue; }
uint64_t h = 1469598103934665603ull; unsigned nz = 0;
for (off_t i = 0; i < st.st_size; i += 64) { h = (h ^ m[i]) * 1099511628211ull; nz += m[i] != 0; }
printf("fd %s ino %lu size %ld hash %016llx nz %u\n", e->d_name, (unsigned long)st.st_ino, (long)st.st_size,
(unsigned long long)h, nz);
munmap((void*)m, st.st_size); close(fd);
}
return 0;
}
@@ -0,0 +1,55 @@
// gpuprobe - a plain Android process (outside MobileGL) that measures whether the GPU keeps executing
// other contexts' work: a pbuffer GLES context clears + draws and glFinish()es every ~10 ms, printing any
// round trip over 50 ms with a CLOCK_REALTIME timestamp. Usage: gpuprobe <seconds> [priority: low|med|high]
#include <EGL/egl.h>
#include <EGL/eglext.h>
#include <GLES3/gl3.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <unistd.h>
#ifndef EGL_CONTEXT_PRIORITY_LEVEL_IMG
#define EGL_CONTEXT_PRIORITY_LEVEL_IMG 0x3100
#define EGL_CONTEXT_PRIORITY_HIGH_IMG 0x3101
#define EGL_CONTEXT_PRIORITY_MEDIUM_IMG 0x3102
#define EGL_CONTEXT_PRIORITY_LOW_IMG 0x3103
#endif
static double now_ms(void) { struct timespec ts; clock_gettime(CLOCK_REALTIME, &ts); return ts.tv_sec * 1e3 + ts.tv_nsec / 1e6; }
int main(int argc, char** argv) {
int seconds = argc > 1 ? atoi(argv[1]) : 20;
EGLint prio = EGL_CONTEXT_PRIORITY_MEDIUM_IMG;
if (argc > 2 && !strcmp(argv[2], "high")) prio = EGL_CONTEXT_PRIORITY_HIGH_IMG;
if (argc > 2 && !strcmp(argv[2], "low")) prio = EGL_CONTEXT_PRIORITY_LOW_IMG;
setvbuf(stdout, NULL, _IOLBF, 0);
EGLDisplay d = eglGetDisplay(EGL_DEFAULT_DISPLAY);
eglInitialize(d, NULL, NULL);
const EGLint ca[] = {EGL_SURFACE_TYPE, EGL_PBUFFER_BIT, EGL_RENDERABLE_TYPE, EGL_OPENGL_ES3_BIT, EGL_RED_SIZE, 8, EGL_NONE};
EGLConfig c; EGLint n;
eglChooseConfig(d, ca, &c, 1, &n);
const EGLint pa[] = {EGL_WIDTH, 256, EGL_HEIGHT, 256, EGL_NONE};
EGLSurface s = eglCreatePbufferSurface(d, c, pa);
const EGLint xa[] = {EGL_CONTEXT_CLIENT_VERSION, 3, EGL_CONTEXT_PRIORITY_LEVEL_IMG, prio, EGL_NONE};
EGLContext x = eglCreateContext(d, c, EGL_NO_CONTEXT, xa);
if (x == EGL_NO_CONTEXT) { printf("no ctx 0x%x\n", eglGetError()); return 1; }
eglMakeCurrent(d, s, s, x);
EGLint got = 0; eglQueryContext(d, x, EGL_CONTEXT_PRIORITY_LEVEL_IMG, &got);
const double t0 = now_ms(); double maxRt = 0; long iters = 0, slow = 0;
printf("[probe] start %.0f prio=0x%x\n", t0, got);
while (now_ms() - t0 < seconds * 1000.0) {
const double a = now_ms();
glClearColor((iters & 1) ? 1.f : 0.f, 0.f, 0.f, 1.f);
glClear(GL_COLOR_BUFFER_BIT);
glFinish();
const double rt = now_ms() - a;
if (rt > maxRt) maxRt = rt;
if (rt > 50) { ++slow; printf("[probe] glFinish %.0f ms ending at %.0f\n", rt, now_ms()); }
++iters;
usleep(10000);
}
printf("[probe] END iters=%ld max=%.0f ms slow=%ld\n", iters, maxRt, slow);
return 0;
}
@@ -0,0 +1,150 @@
// hangclient - a Wayland xdg-toplevel, EGL + desktop GL 3.3 core client that animates and, on request,
// submits fragment work that never ends (the driver has to reset it). Prints per-frame gaps so the same
// binary measures "other app" frame pacing (frame callbacks come from the compositor) when run with -H -1.
//
// hangclient -T tag [-H hang_at_frame] [-N hang_repeat] [-I iters] [-S seconds] [-W w] [-X h] [-F]
// -H n frame at which the hanging draw is submitted (-1: never; default -1)
// -N k how many consecutive hanging frames (default 1)
// -E e re-arm a hang every e frames after the first (0: once)
// -I i loop count of the hanging shader (default 2^31-1)
// -S s run time in seconds (default 30)
// -F glFinish after every frame (puts the wait inside the client's own session)
#define GL_GLEXT_PROTOTYPES 1
#include <EGL/egl.h>
#include <EGL/eglext.h>
#include <GL/glcorearb.h>
#include <wayland-client.h>
#include <wayland-egl.h>
#include "xdg-shell-client-protocol.h"
#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <unistd.h>
static struct wl_compositor* g_comp;
static struct xdg_wm_base* g_wm;
static int g_configured, g_closed;
static int g_w = 480, g_h = 360;
static void reg_global(void* d, struct wl_registry* r, uint32_t name, const char* iface, uint32_t ver) {
(void)d;
if (!strcmp(iface, "wl_compositor")) g_comp = wl_registry_bind(r, name, &wl_compositor_interface, 4);
else if (!strcmp(iface, "xdg_wm_base")) g_wm = wl_registry_bind(r, name, &xdg_wm_base_interface, 1);
}
static void reg_remove(void* d, struct wl_registry* r, uint32_t n) { (void)d; (void)r; (void)n; }
static const struct wl_registry_listener reg_l = {reg_global, reg_remove};
static void wm_ping(void* d, struct xdg_wm_base* wm, uint32_t s) { (void)d; xdg_wm_base_pong(wm, s); }
static const struct xdg_wm_base_listener wm_l = {wm_ping};
static void xs_conf(void* d, struct xdg_surface* s, uint32_t serial) { (void)d; xdg_surface_ack_configure(s, serial); g_configured = 1; }
static const struct xdg_surface_listener xs_l = {xs_conf};
static void tl_conf(void* d, struct xdg_toplevel* t, int32_t w, int32_t h, struct wl_array* st) { (void)d; (void)t; (void)st; (void)w; (void)h; }
static void tl_close(void* d, struct xdg_toplevel* t) { (void)d; (void)t; g_closed = 1; }
static const struct xdg_toplevel_listener tl_l = {tl_conf, tl_close};
static double now_ms(void) { struct timespec ts; clock_gettime(CLOCK_REALTIME, &ts); return ts.tv_sec * 1e3 + ts.tv_nsec / 1e6; }
static GLuint sh(GLenum type, const char* src) {
GLuint s = glCreateShader(type);
glShaderSource(s, 1, &src, NULL);
glCompileShader(s);
GLint ok = 0; glGetShaderiv(s, GL_COMPILE_STATUS, &ok);
if (!ok) { char log[2048]; glGetShaderInfoLog(s, sizeof log, NULL, log); fprintf(stderr, "shader: %s\n", log); exit(2); }
return s;
}
int main(int argc, char** argv) {
const char* tag = "hc"; int hangAt = -1, hangN = 1, every = 0, seconds = 30, finish = 0; long iters = 2147483647L;
for (int i = 1; i < argc; ++i) {
if (!strcmp(argv[i], "-T") && i + 1 < argc) tag = argv[++i];
else if (!strcmp(argv[i], "-H") && i + 1 < argc) hangAt = atoi(argv[++i]);
else if (!strcmp(argv[i], "-N") && i + 1 < argc) hangN = atoi(argv[++i]);
else if (!strcmp(argv[i], "-E") && i + 1 < argc) every = atoi(argv[++i]);
else if (!strcmp(argv[i], "-I") && i + 1 < argc) iters = atol(argv[++i]);
else if (!strcmp(argv[i], "-S") && i + 1 < argc) seconds = atoi(argv[++i]);
else if (!strcmp(argv[i], "-W") && i + 1 < argc) g_w = atoi(argv[++i]);
else if (!strcmp(argv[i], "-X") && i + 1 < argc) g_h = atoi(argv[++i]);
else if (!strcmp(argv[i], "-F")) finish = 1;
}
setvbuf(stdout, NULL, _IOLBF, 0);
struct wl_display* dpy = wl_display_connect(NULL);
if (!dpy) { fprintf(stderr, "no wayland\n"); return 1; }
struct wl_registry* reg = wl_display_get_registry(dpy);
wl_registry_add_listener(reg, &reg_l, NULL);
wl_display_roundtrip(dpy);
if (!g_comp || !g_wm) { fprintf(stderr, "no compositor/xdg_wm_base\n"); return 1; }
xdg_wm_base_add_listener(g_wm, &wm_l, NULL);
struct wl_surface* surf = wl_compositor_create_surface(g_comp);
struct xdg_surface* xs = xdg_wm_base_get_xdg_surface(g_wm, surf);
xdg_surface_add_listener(xs, &xs_l, NULL);
struct xdg_toplevel* tl = xdg_surface_get_toplevel(xs);
xdg_toplevel_add_listener(tl, &tl_l, NULL);
xdg_toplevel_set_title(tl, tag);
wl_surface_commit(surf);
while (!g_configured) wl_display_dispatch(dpy);
EGLDisplay ed = eglGetPlatformDisplay(EGL_PLATFORM_WAYLAND_KHR, dpy, NULL);
if (!eglInitialize(ed, NULL, NULL)) { fprintf(stderr, "eglInitialize\n"); return 1; }
eglBindAPI(EGL_OPENGL_API);
const EGLint ca[] = {EGL_SURFACE_TYPE, EGL_WINDOW_BIT, EGL_RENDERABLE_TYPE, EGL_OPENGL_BIT, EGL_RED_SIZE, 8, EGL_GREEN_SIZE, 8, EGL_BLUE_SIZE, 8, EGL_NONE};
EGLConfig cfg; EGLint n = 0;
if (!eglChooseConfig(ed, ca, &cfg, 1, &n) || n < 1) { fprintf(stderr, "no config\n"); return 1; }
const EGLint xa[] = {EGL_CONTEXT_MAJOR_VERSION, 3, EGL_CONTEXT_MINOR_VERSION, 3, EGL_CONTEXT_OPENGL_PROFILE_MASK, EGL_CONTEXT_OPENGL_CORE_PROFILE_BIT, EGL_NONE};
EGLContext ctx = eglCreateContext(ed, cfg, EGL_NO_CONTEXT, xa);
if (ctx == EGL_NO_CONTEXT) { fprintf(stderr, "no context 0x%x\n", eglGetError()); return 1; }
struct wl_egl_window* ew = wl_egl_window_create(surf, g_w, g_h);
EGLSurface es = eglCreatePlatformWindowSurface(ed, cfg, ew, NULL);
eglMakeCurrent(ed, es, es, ctx);
eglSwapInterval(ed, 1);
const char* vs = "#version 330 core\nlayout(location=0) in vec2 p; uniform float t; out vec2 uv;\n"
"void main(){ uv=p*0.5+0.5; gl_Position=vec4(p*0.4+vec2(0.5*sin(t),0.0),0.0,1.0);}";
const char* fs = "#version 330 core\nin vec2 uv; uniform int iters; uniform float t; out vec4 c;\n"
"void main(){ float a=uv.x; for(int i=0;i<iters;++i){ a=fract(a*1.0001+0.37+uv.y); }\n"
" c=vec4(uv, 0.5+0.5*sin(t)+a*1e-6, 1.0);}";
GLuint prog = glCreateProgram();
glAttachShader(prog, sh(GL_VERTEX_SHADER, vs));
glAttachShader(prog, sh(GL_FRAGMENT_SHADER, fs));
glLinkProgram(prog);
GLint ok = 0; glGetProgramiv(prog, GL_LINK_STATUS, &ok);
if (!ok) { fprintf(stderr, "link failed\n"); return 2; }
GLint uIt = glGetUniformLocation(prog, "iters"), uT = glGetUniformLocation(prog, "t");
GLuint vao, vbo; glGenVertexArrays(1, &vao); glBindVertexArray(vao);
const float q[] = {-1, -1, 1, -1, -1, 1, 1, 1};
glGenBuffers(1, &vbo); glBindBuffer(GL_ARRAY_BUFFER, vbo); glBufferData(GL_ARRAY_BUFFER, sizeof q, q, GL_STATIC_DRAW);
glEnableVertexAttribArray(0); glVertexAttribPointer(0, 2, GL_FLOAT, GL_FALSE, 0, 0);
const double t0 = now_ms(); double last = t0, maxGap = 0; long frames = 0, gaps100 = 0;
printf("[%s] start %.0f renderer=%s hangAt=%d N=%d every=%d iters=%ld\n", tag, t0, glGetString(GL_RENDERER), hangAt, hangN, every, iters);
while (!g_closed && now_ms() - t0 < seconds * 1000.0) {
wl_display_dispatch_pending(dpy);
const double t = (now_ms() - t0) / 1000.0;
int hang = 0;
if (hangAt >= 0 && frames >= hangAt) {
long rel = frames - hangAt;
if (every > 0) rel %= every;
hang = rel < hangN;
}
glViewport(0, 0, g_w, g_h);
glClearColor(0.5f + 0.5f * (float)sin(t * 3.0), 0.2f, 0.4f, 1.0f);
glClear(GL_COLOR_BUFFER_BIT);
glUseProgram(prog);
glUniform1i(uIt, hang ? (int)iters : 0);
glUniform1f(uT, (float)t);
glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
if (hang) printf("[%s] HANG submitted frame %ld at %.0f\n", tag, frames, now_ms());
if (finish) glFinish();
if (!eglSwapBuffers(ed, es)) printf("[%s] swap failed 0x%x at %.0f\n", tag, eglGetError(), now_ms());
const GLenum rs = glGetError();
if (rs != GL_NO_ERROR) printf("[%s] glGetError 0x%x at %.0f\n", tag, rs, now_ms());
const double nowv = now_ms(), gap = nowv - last;
last = nowv; ++frames;
if (gap > maxGap) maxGap = gap;
if (gap > 100.0) { ++gaps100; printf("[%s] GAP %.0f ms ending at %.0f (frame %ld)\n", tag, gap, nowv, frames); }
if (frames % 120 == 0) printf("[%s] frame %ld t=%.1fs fps=%.1f\n", tag, frames, (nowv - t0) / 1e3, frames * 1e3 / (nowv - t0));
}
const double total = now_ms() - t0;
printf("[%s] END frames=%ld fps=%.1f maxGap=%.0f ms gaps>100ms=%ld\n", tag, frames, frames * 1e3 / total, maxGap, gaps100);
return 0;
}
@@ -0,0 +1,95 @@
/* texfmt: does a texture keep its alpha? One opaque red texel and one transparent texel per
* format, sampled across a 64x64 ES3 pbuffer cleared white with SRC_ALPHA blending. Expect the
* left half red and the right half WHITE; a black right half means the alpha was lost. */
#include <EGL/egl.h>
#include <GLES3/gl3.h>
#include <GLES2/gl2ext.h>
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#define P(T, n) T n = (T)eglGetProcAddress(#n)
static const char* vs = "#version 300 es\nlayout(location=0) in vec2 p; out vec2 uv; void main(){uv=p*0.5+0.5; gl_Position=vec4(p,0,1);}";
static const char* fs = "#version 300 es\nprecision mediump float; uniform sampler2D t; uniform int mode; in vec2 uv; out vec4 o;"
"void main(){vec4 c=texture(t,uv); o = mode==1 ? vec4(1.0,0.0,0.0,c.a) : (mode==2 ? vec4(c.r,0.0,0.0,c.r) : c);}";
int main(void) {
EGLDisplay d = eglGetDisplay(EGL_DEFAULT_DISPLAY); eglInitialize(d, 0, 0); eglBindAPI(EGL_OPENGL_ES_API);
EGLint ca[] = {EGL_SURFACE_TYPE, EGL_PBUFFER_BIT, EGL_RENDERABLE_TYPE, EGL_OPENGL_ES3_BIT, EGL_RED_SIZE, 8, EGL_GREEN_SIZE, 8, EGL_BLUE_SIZE, 8, EGL_ALPHA_SIZE, 8, EGL_NONE};
EGLConfig c; EGLint n; eglChooseConfig(d, ca, &c, 1, &n);
EGLint xa[] = {EGL_CONTEXT_CLIENT_VERSION, 3, EGL_NONE};
EGLContext x = eglCreateContext(d, c, EGL_NO_CONTEXT, xa);
EGLint pa[] = {EGL_WIDTH, 64, EGL_HEIGHT, 64, EGL_NONE};
EGLSurface s = eglCreatePbufferSurface(d, c, pa); eglMakeCurrent(d, s, s, x);
P(PFNGLCREATESHADERPROC, glCreateShader); P(PFNGLSHADERSOURCEPROC, glShaderSource); P(PFNGLCOMPILESHADERPROC, glCompileShader);
P(PFNGLCREATEPROGRAMPROC, glCreateProgram); P(PFNGLATTACHSHADERPROC, glAttachShader); P(PFNGLLINKPROGRAMPROC, glLinkProgram);
P(PFNGLUSEPROGRAMPROC, glUseProgram); P(PFNGLGETUNIFORMLOCATIONPROC, glGetUniformLocation); P(PFNGLUNIFORM1IPROC, glUniform1i);
P(PFNGLGENBUFFERSPROC, glGenBuffers); P(PFNGLBINDBUFFERPROC, glBindBuffer); P(PFNGLBUFFERDATAPROC, glBufferData);
P(PFNGLVERTEXATTRIBPOINTERPROC, glVertexAttribPointer); P(PFNGLENABLEVERTEXATTRIBARRAYPROC, glEnableVertexAttribArray);
P(PFNGLDRAWARRAYSPROC, glDrawArrays); P(PFNGLCLEARCOLORPROC, glClearColor); P(PFNGLCLEARPROC, glClear);
P(PFNGLREADPIXELSPROC, glReadPixels); P(PFNGLVIEWPORTPROC, glViewport); P(PFNGLGENTEXTURESPROC, glGenTextures);
P(PFNGLBINDTEXTUREPROC, glBindTexture); P(PFNGLTEXIMAGE2DPROC, glTexImage2D); P(PFNGLTEXPARAMETERIPROC, glTexParameteri);
P(PFNGLENABLEPROC, glEnable); P(PFNGLBLENDFUNCPROC, glBlendFunc); P(PFNGLGETERRORPROC, glGetError);
P(PFNGLTEXSTORAGE2DPROC, glTexStorage2D); P(PFNGLTEXSUBIMAGE2DPROC, glTexSubImage2D); P(PFNGLGETSTRINGPROC, glGetString);
printf("GL_VERSION %s\n", glGetString(GL_VERSION));
printf("BGRA ext: %s\n", strstr((const char*)glGetString(GL_EXTENSIONS), "GL_EXT_texture_format_BGRA8888") ? "yes" : "no");
GLuint v = glCreateShader(GL_VERTEX_SHADER), f = glCreateShader(GL_FRAGMENT_SHADER);
glShaderSource(v, 1, &vs, 0); glCompileShader(v); glShaderSource(f, 1, &fs, 0); glCompileShader(f);
GLuint pr = glCreateProgram(); glAttachShader(pr, v); glAttachShader(pr, f); glLinkProgram(pr); glUseProgram(pr);
float q[] = {-1,-1, 1,-1, -1,1, 1,1};
GLuint vbo; glGenBuffers(1, &vbo); glBindBuffer(GL_ARRAY_BUFFER, vbo);
glBufferData(GL_ARRAY_BUFFER, sizeof q, q, GL_STATIC_DRAW); glVertexAttribPointer(0, 2, GL_FLOAT, 0, 0, 0); glEnableVertexAttribArray(0);
GLint lm = glGetUniformLocation(pr, "mode");
glEnable(GL_BLEND); glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
struct { const char* name; GLint ifmt; GLenum fmt; int bpp; int mode; int storage; GLenum sized; unsigned char px[8]; } cases[] = {
{"RGBA/TexImage", GL_RGBA, GL_RGBA, 4, 0, 0, 0, {255,0,0,255, 0,0,0,0}},
{"BGRA_EXT/TexImage", GL_BGRA_EXT, GL_BGRA_EXT, 4, 0, 0, 0, {0,0,255,255, 0,0,0,0}},
{"BGRA8_EXT/TexStorage+Sub", 0, GL_BGRA_EXT, 4, 0, 1, GL_BGRA8_EXT, {0,0,255,255, 0,0,0,0}},
{"RGBA8/TexStorage+Sub", 0, GL_RGBA, 4, 0, 1, GL_RGBA8, {255,0,0,255, 0,0,0,0}},
{"ALPHA", GL_ALPHA, GL_ALPHA, 1, 1, 0, 0, {255, 0}},
{"LUMINANCE_ALPHA", GL_LUMINANCE_ALPHA, GL_LUMINANCE_ALPHA, 2, 1, 0, 0, {255,255, 0,0}},
{"R8/TexStorage+Sub", 0, GL_RED, 1, 2, 1, GL_R8, {255, 0}},
};
for (unsigned i = 0; i < sizeof cases / sizeof cases[0]; ++i) {
GLuint t; glGenTextures(1, &t); glBindTexture(GL_TEXTURE_2D, t);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST); glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
while (glGetError()) {}
if (cases[i].storage) {
glTexStorage2D(GL_TEXTURE_2D, 1, cases[i].sized, 2, 1);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 2, 1, cases[i].fmt, GL_UNSIGNED_BYTE, cases[i].px);
} else {
glTexImage2D(GL_TEXTURE_2D, 0, cases[i].ifmt, 2, 1, 0, cases[i].fmt, GL_UNSIGNED_BYTE, cases[i].px);
}
GLenum err = glGetError();
glUniform1i(lm, cases[i].mode);
glViewport(0, 0, 64, 64); glClearColor(1, 1, 1, 1); glClear(GL_COLOR_BUFFER_BIT);
glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
unsigned char l[4], r[4];
glReadPixels(16, 32, 1, 1, GL_RGBA, GL_UNSIGNED_BYTE, l); glReadPixels(48, 32, 1, 1, GL_RGBA, GL_UNSIGNED_BYTE, r);
const int ok = l[0] > 200 && l[1] < 60 && l[2] < 60 && r[0] > 200 && r[1] > 200 && r[2] > 200;
printf("%-26s err=0x%04x left=%3d,%3d,%3d right=%3d,%3d,%3d %s\n", cases[i].name, err, l[0], l[1], l[2], r[0], r[1], r[2], ok ? "ok" : "BAD");
}
/* ANGLE's emulation of ALPHA8 / LUMINANCE_ALPHA8 on a desktop-GL driver: R8 / RG8 + swizzle. */
P(PFNGLTEXPARAMETERIPROC, glTexParameteri2);
struct { const char* name; GLenum sized, fmt; int bpp; GLint sw[4]; unsigned char px[4]; unsigned char wantL[3], wantR[3]; } sz[] = {
{"R8 swizzled as ALPHA8", GL_R8, GL_RED, 1, {GL_ZERO, GL_ZERO, GL_ZERO, GL_RED}, {255, 0}, {0,0,0}, {255,255,255}},
{"RG8 swizzled as LUM_ALPHA", GL_RG8, GL_RG, 2, {GL_RED, GL_RED, GL_RED, GL_GREEN}, {255,255, 0,0}, {255,255,255}, {255,255,255}},
{"RG8 as LUM_ALPHA (lum 0)", GL_RG8, GL_RG, 2, {GL_RED, GL_RED, GL_RED, GL_GREEN}, {0,255, 0,0}, {0,0,0}, {255,255,255}},
};
glUniform1i(lm, 0);
for (unsigned i = 0; i < sizeof sz / sizeof sz[0]; ++i) {
GLuint t; glGenTextures(1, &t); glBindTexture(GL_TEXTURE_2D, t);
glTexParameteri2(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST); glTexParameteri2(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
glTexStorage2D(GL_TEXTURE_2D, 1, sz[i].sized, 2, 1);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 2, 1, sz[i].fmt, GL_UNSIGNED_BYTE, sz[i].px);
glTexParameteri2(GL_TEXTURE_2D, GL_TEXTURE_SWIZZLE_R, sz[i].sw[0]); glTexParameteri2(GL_TEXTURE_2D, GL_TEXTURE_SWIZZLE_G, sz[i].sw[1]);
glTexParameteri2(GL_TEXTURE_2D, GL_TEXTURE_SWIZZLE_B, sz[i].sw[2]); glTexParameteri2(GL_TEXTURE_2D, GL_TEXTURE_SWIZZLE_A, sz[i].sw[3]);
GLenum err = glGetError();
glClearColor(1, 1, 1, 1); glClear(GL_COLOR_BUFFER_BIT); glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
unsigned char l[4], r[4];
glReadPixels(16, 32, 1, 1, GL_RGBA, GL_UNSIGNED_BYTE, l); glReadPixels(48, 32, 1, 1, GL_RGBA, GL_UNSIGNED_BYTE, r);
int ok = 1;
for (int k = 0; k < 3; ++k) ok = ok && abs(l[k] - sz[i].wantL[k]) < 40 && abs(r[k] - sz[i].wantR[k]) < 40;
printf("%-26s err=0x%04x left=%3d,%3d,%3d right=%3d,%3d,%3d %s\n", sz[i].name, err, l[0], l[1], l[2], r[0], r[1], r[2], ok ? "ok" : "BAD");
}
return 0;
}
@@ -0,0 +1,410 @@
// yuvprobe.c - pixel check of YUV dma-buf sampling through MobileGL (GL_OES_EGL_image_external).
//
// gcc -O1 yuvprobe.c -o /usr/local/bin/mgl-yuvprobe -lEGL -lGLESv2 -lgbm
// mgrun offscreen mgl-yuvprobe # exit 0 = every check passed
//
// What it does, in an ES 3.0 context on a pbuffer:
// 1. FOREIGN NV12 (and P010): a dma-buf from /dev/dma_heap/system - a buffer the MobileGL server
// did not allocate, as a video decoder's would be - with padded pitches and a gap before the
// chroma plane, filled with four vertical bars (red, green, blue, white) in BT.709 narrow
// range. Imported with eglCreateImage(EGL_LINUX_DMA_BUF_EXT) and its colour hints, bound to
// GL_TEXTURE_EXTERNAL_OES, drawn through a samplerExternalOES shader into an RGBA8 FBO and read
// back: each bar must come out its colour. The buffer is then rewritten with the bars in
// reverse order and, after a native fence (what ends a frame for an offscreen producer), drawn
// again: the new order must show (the server refreshes a foreign buffer at each frame).
// 2. GBM NV12: gbm_bo_create(NV12) through the mobilegl GBM backend (a server-allocated YUV
// buffer, zero-copy), its planes printed, imported and sampled once (content undefined; the
// check is that every call succeeds).
// Also prints the dma-buf formats/modifiers EGL lists for NV12/P010.
#define _GNU_SOURCE
#include <EGL/egl.h>
#include <EGL/eglext.h>
#include <GLES3/gl3.h>
#include <GLES2/gl2ext.h>
#include <fcntl.h>
#include <gbm.h>
#include <linux/dma-buf.h>
#include <linux/dma-heap.h>
#include <math.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/ioctl.h>
#include <sys/mman.h>
#include <unistd.h>
#ifndef DRM_FORMAT_NV12
#define DRM_FORMAT_NV12 0x3231564E
#endif
#ifndef DRM_FORMAT_P010
#define DRM_FORMAT_P010 0x30313050
#endif
static int g_failures = 0;
#define CHECK(cond, ...) \
do { \
if (!(cond)) { \
printf("FAIL: " __VA_ARGS__); \
printf("\n"); \
++g_failures; \
} \
} while (0)
static PFNEGLCREATEIMAGEKHRPROC pCreateImage;
static PFNEGLDESTROYIMAGEKHRPROC pDestroyImage;
static PFNGLEGLIMAGETARGETTEXTURE2DOESPROC pTargetTexture;
static PFNEGLQUERYDMABUFFORMATSEXTPROC pQueryFormats;
static PFNEGLQUERYDMABUFMODIFIERSEXTPROC pQueryModifiers;
static PFNEGLCREATESYNCKHRPROC pCreateSync;
static PFNEGLDESTROYSYNCKHRPROC pDestroySync;
static EGLDisplay g_display;
enum { W = 256, H = 128, BARS = 4 };
static const unsigned char kBars[BARS][3] = {{255, 0, 0}, {0, 255, 0}, {0, 0, 255}, {255, 255, 255}};
// RGB -> BT.709 narrow-range codes at `bits` (8 or 10).
static void Encode709(const unsigned char rgb[3], int bits, int out[3]) {
const double kr = 0.2126, kb = 0.0722, kg = 1.0 - kr - kb;
const double r = rgb[0] / 255.0, g = rgb[1] / 255.0, b = rgb[2] / 255.0;
const double y = kr * r + kg * g + kb * b;
const double cb = (b - y) / (2.0 * (1.0 - kb)), cr = (r - y) / (2.0 * (1.0 - kr));
const double scale = (double)(1 << (bits - 8));
out[0] = (int)lround((16.0 + 219.0 * y) * scale);
out[1] = (int)lround((128.0 + 224.0 * cb) * scale);
out[2] = (int)lround((128.0 + 224.0 * cr) * scale);
}
// Writes the bars (`reverse`d or not) into a semi-planar buffer: 1 byte samples, or P010's 16-bit
// samples with the 10 significant bits high.
static void FillBars(uint8_t* base, uint32_t pitchY, uint32_t offsetC, uint32_t pitchC, int bytes, int reverse) {
const int bits = bytes == 1 ? 8 : 10;
for (int y = 0; y < H; ++y) {
for (int x = 0; x < W; ++x) {
int bar = x * BARS / W;
if (reverse) bar = BARS - 1 - bar;
int code[3];
Encode709(kBars[bar], bits, code);
uint8_t* py = base + (size_t)y * pitchY + (size_t)x * bytes;
if (bytes == 1) py[0] = (uint8_t)code[0];
else *(uint16_t*)py = (uint16_t)(code[0] << 6);
if ((x & 1) == 0 && (y & 1) == 0) {
uint8_t* pc = base + offsetC + (size_t)(y / 2) * pitchC + (size_t)(x / 2) * 2 * bytes;
if (bytes == 1) {
pc[0] = (uint8_t)code[1];
pc[1] = (uint8_t)code[2];
} else {
((uint16_t*)pc)[0] = (uint16_t)(code[1] << 6);
((uint16_t*)pc)[1] = (uint16_t)(code[2] << 6);
}
}
}
}
}
static GLuint Compile(GLenum type, const char* source) {
GLuint shader = glCreateShader(type);
glShaderSource(shader, 1, &source, NULL);
glCompileShader(shader);
GLint ok = 0;
glGetShaderiv(shader, GL_COMPILE_STATUS, &ok);
if (!ok) {
char log[2048];
glGetShaderInfoLog(shader, sizeof(log), NULL, log);
printf("shader: %s\n", log);
}
return shader;
}
static GLuint g_program, g_fbo, g_rgba;
static void SetUpDraw(void) {
const char* vs = "#version 300 es\n"
"out vec2 uv;\n"
"void main() {\n"
" vec2 p = vec2(float((gl_VertexID << 1) & 2), float(gl_VertexID & 2));\n"
" gl_Position = vec4(p * 2.0 - 1.0, 0.0, 1.0);\n"
" uv = p;\n"
"}\n";
const char* fs = "#version 300 es\n"
"#extension GL_OES_EGL_image_external_essl3 : require\n"
"precision mediump float;\n"
"uniform samplerExternalOES tex;\n"
"in vec2 uv;\n"
"out vec4 color;\n"
"void main() { color = texture(tex, uv); }\n";
g_program = glCreateProgram();
glAttachShader(g_program, Compile(GL_VERTEX_SHADER, vs));
glAttachShader(g_program, Compile(GL_FRAGMENT_SHADER, fs));
glLinkProgram(g_program);
GLint linked = 0;
glGetProgramiv(g_program, GL_LINK_STATUS, &linked);
CHECK(linked, "the samplerExternalOES program did not link");
GLint type = 0, size = 0;
char name[64];
glGetActiveUniform(g_program, 0, sizeof(name), NULL, &size, (GLenum*)&type, name);
printf("uniform 0: %s type 0x%04x\n", name, type);
CHECK(type == GL_SAMPLER_EXTERNAL_OES, "the sampler reports type 0x%04x, not GL_SAMPLER_EXTERNAL_OES", type);
glGenTextures(1, &g_rgba);
glBindTexture(GL_TEXTURE_2D, g_rgba);
glTexStorage2D(GL_TEXTURE_2D, 1, GL_RGBA8, W, H);
glGenFramebuffers(1, &g_fbo);
glBindFramebuffer(GL_FRAMEBUFFER, g_fbo);
glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, g_rgba, 0);
CHECK(glCheckFramebufferStatus(GL_FRAMEBUFFER) == GL_FRAMEBUFFER_COMPLETE, "the readback FBO is incomplete");
GLuint vao;
glGenVertexArrays(1, &vao);
glBindVertexArray(vao);
}
// Draws the external texture `texture` and checks each bar's center (row `y` of the image).
static void DrawAndCheck(GLuint texture, const char* what, int reverse, int tolerance) {
glBindFramebuffer(GL_FRAMEBUFFER, g_fbo);
glViewport(0, 0, W, H);
glUseProgram(g_program);
glActiveTexture(GL_TEXTURE0);
glBindTexture(GL_TEXTURE_EXTERNAL_OES, texture);
glUniform1i(glGetUniformLocation(g_program, "tex"), 0);
glDrawArrays(GL_TRIANGLES, 0, 3);
static unsigned char pixels[W * H * 4];
glReadPixels(0, 0, W, H, GL_RGBA, GL_UNSIGNED_BYTE, pixels);
GLenum error = glGetError();
CHECK(error == GL_NO_ERROR, "%s: GL error 0x%04x", what, error);
if (tolerance < 0) {
printf("%s: center pixel %d %d %d %d (content undefined)\n", what, pixels[(H / 2 * W + W / 2) * 4],
pixels[(H / 2 * W + W / 2) * 4 + 1], pixels[(H / 2 * W + W / 2) * 4 + 2],
pixels[(H / 2 * W + W / 2) * 4 + 3]);
return;
}
// Rows: GL row 0 is the image's first row (the uv mapping above), and the bars are vertical,
// so any row does; take two to catch a broken chroma row pitch.
const int rows[2] = {H / 4, (3 * H) / 4};
for (int r = 0; r < 2; ++r) {
for (int bar = 0; bar < BARS; ++bar) {
const int x = bar * (W / BARS) + W / (2 * BARS);
const unsigned char* p = &pixels[(rows[r] * W + x) * 4];
const unsigned char* want = kBars[reverse ? BARS - 1 - bar : bar];
int worst = 0;
for (int c = 0; c < 3; ++c) {
int d = abs((int)p[c] - (int)want[c]);
if (d > worst) worst = d;
}
printf("%s: row %d bar %d -> %3d %3d %3d %3d (want %3d %3d %3d)%s\n", what, rows[r], bar, p[0], p[1], p[2],
p[3], want[0], want[1], want[2], worst > tolerance ? " <-- off" : "");
CHECK(worst <= tolerance, "%s: bar %d is off by %d", what, bar, worst);
}
}
}
static int AllocateDmaHeap(size_t size) {
int heap = open("/dev/dma_heap/system", O_RDWR | O_CLOEXEC);
if (heap < 0) {
perror("open /dev/dma_heap/system");
return -1;
}
struct dma_heap_allocation_data data = {.len = size, .fd_flags = O_RDWR | O_CLOEXEC};
if (ioctl(heap, DMA_HEAP_IOCTL_ALLOC, &data) != 0) {
perror("DMA_HEAP_IOCTL_ALLOC");
close(heap);
return -1;
}
close(heap);
return (int)data.fd;
}
static void Sync(int fd, int start) {
struct dma_buf_sync sync = {.flags = DMA_BUF_SYNC_WRITE | (start ? DMA_BUF_SYNC_START : DMA_BUF_SYNC_END)};
ioctl(fd, DMA_BUF_IOCTL_SYNC, &sync);
}
static EGLImageKHR Import(int fd, uint32_t fourcc, uint32_t pitchY, uint32_t offsetC, uint32_t pitchC, int modifier) {
EGLint attribs[64];
int n = 0;
#define A(k, v) (attribs[n++] = (k), attribs[n++] = (EGLint)(v))
A(EGL_WIDTH, W);
A(EGL_HEIGHT, H);
A(EGL_LINUX_DRM_FOURCC_EXT, fourcc);
A(EGL_DMA_BUF_PLANE0_FD_EXT, fd);
A(EGL_DMA_BUF_PLANE0_OFFSET_EXT, 0);
A(EGL_DMA_BUF_PLANE0_PITCH_EXT, pitchY);
A(EGL_DMA_BUF_PLANE1_FD_EXT, fd);
A(EGL_DMA_BUF_PLANE1_OFFSET_EXT, offsetC);
A(EGL_DMA_BUF_PLANE1_PITCH_EXT, pitchC);
if (modifier) {
A(EGL_DMA_BUF_PLANE0_MODIFIER_LO_EXT, 0);
A(EGL_DMA_BUF_PLANE0_MODIFIER_HI_EXT, 0);
A(EGL_DMA_BUF_PLANE1_MODIFIER_LO_EXT, 0);
A(EGL_DMA_BUF_PLANE1_MODIFIER_HI_EXT, 0);
}
A(EGL_YUV_COLOR_SPACE_HINT_EXT, EGL_ITU_REC709_EXT);
A(EGL_SAMPLE_RANGE_HINT_EXT, EGL_YUV_NARROW_RANGE_EXT);
attribs[n] = EGL_NONE;
#undef A
EGLImageKHR image = pCreateImage(g_display, EGL_NO_CONTEXT, EGL_LINUX_DMA_BUF_EXT, NULL, attribs);
if (image == EGL_NO_IMAGE_KHR) printf("eglCreateImage failed: 0x%04x\n", eglGetError());
return image;
}
static GLuint ExternalTexture(EGLImageKHR image) {
GLuint texture;
glGenTextures(1, &texture);
glBindTexture(GL_TEXTURE_EXTERNAL_OES, texture);
pTargetTexture(GL_TEXTURE_EXTERNAL_OES, image);
GLenum error = glGetError();
CHECK(error == GL_NO_ERROR, "glEGLImageTargetTexture2DOES(GL_TEXTURE_EXTERNAL_OES) -> 0x%04x", error);
GLint bound = 0;
glGetIntegerv(GL_TEXTURE_BINDING_EXTERNAL_OES, &bound);
CHECK((GLuint)bound == texture, "GL_TEXTURE_BINDING_EXTERNAL_OES reads %d", bound);
return texture;
}
// A frame boundary as an offscreen producer makes one: a native fence.
static void EndFrame(void) {
EGLint attribs[] = {EGL_NONE};
EGLSyncKHR sync = pCreateSync(g_display, EGL_SYNC_NATIVE_FENCE_ANDROID, attribs);
CHECK(sync != EGL_NO_SYNC_KHR, "eglCreateSync(native fence) failed");
if (sync != EGL_NO_SYNC_KHR) pDestroySync(g_display, sync);
}
static void ForeignTest(uint32_t fourcc, int bytes, int tolerance) {
const char* what = fourcc == DRM_FORMAT_NV12 ? "foreign NV12" : "foreign P010";
// Padded pitches and a gap before the chroma plane, as decoders lay buffers out.
const uint32_t pitchY = W * bytes + 64, pitchC = W * bytes + 128;
const uint32_t offsetC = pitchY * (H + 16);
const size_t size = offsetC + (size_t)pitchC * (H / 2) + 4096;
int fd = AllocateDmaHeap(size);
if (fd < 0) {
CHECK(0, "%s: no dma-buf to test with", what);
return;
}
uint8_t* map = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
CHECK(map != MAP_FAILED, "%s: mmap failed", what);
if (map == MAP_FAILED) return;
Sync(fd, 1);
FillBars(map, pitchY, offsetC, pitchC, bytes, 0);
Sync(fd, 0);
EGLImageKHR image = Import(fd, fourcc, pitchY, offsetC, pitchC, fourcc == DRM_FORMAT_NV12);
CHECK(image != EGL_NO_IMAGE_KHR, "%s: import refused", what);
if (image == EGL_NO_IMAGE_KHR) return;
GLuint texture = ExternalTexture(image);
DrawAndCheck(texture, what, 0, tolerance);
EndFrame();
// New content, same buffer: shows after the frame boundary.
Sync(fd, 1);
FillBars(map, pitchY, offsetC, pitchC, bytes, 1);
Sync(fd, 0);
EndFrame();
DrawAndCheck(texture, what, 1, tolerance);
glDeleteTextures(1, &texture);
pDestroyImage(g_display, image);
munmap(map, size);
close(fd);
}
static void GbmTest(void) {
int node = open("/dev/dri/renderD128", O_RDWR | O_CLOEXEC);
if (node < 0) node = open("/dev/null", O_RDWR | O_CLOEXEC);
struct gbm_device* gbm = gbm_create_device(node);
if (gbm == NULL) {
CHECK(0, "gbm_create_device failed");
return;
}
printf("gbm backend: %s\n", gbm_device_get_backend_name(gbm));
const int supported = gbm_device_is_format_supported(gbm, DRM_FORMAT_NV12, 0);
printf("gbm NV12 supported: %d, P010: %d\n", supported,
gbm_device_is_format_supported(gbm, DRM_FORMAT_P010, 0));
CHECK(supported == 1, "gbm does not support NV12");
struct gbm_bo* bo = gbm_bo_create(gbm, W, H, DRM_FORMAT_NV12, 0);
CHECK(bo != NULL, "gbm_bo_create(NV12) failed");
if (bo != NULL) {
const int planes = gbm_bo_get_plane_count(bo);
printf("gbm NV12 bo: planes %d, stride %u/%u, offset %u/%u, modifier 0x%llx\n", planes,
gbm_bo_get_stride_for_plane(bo, 0), gbm_bo_get_stride_for_plane(bo, 1), gbm_bo_get_offset(bo, 0),
gbm_bo_get_offset(bo, 1), (unsigned long long)gbm_bo_get_modifier(bo));
CHECK(planes == 2, "the NV12 bo has %d planes", planes);
int fd = gbm_bo_get_fd(bo);
EGLImageKHR image =
Import(fd, DRM_FORMAT_NV12, gbm_bo_get_stride_for_plane(bo, 0), gbm_bo_get_offset(bo, 1),
gbm_bo_get_stride_for_plane(bo, 1), 0);
CHECK(image != EGL_NO_IMAGE_KHR, "the server's own NV12 buffer was not imported");
if (image != EGL_NO_IMAGE_KHR) {
GLuint texture = ExternalTexture(image);
DrawAndCheck(texture, "gbm NV12", 0, -1);
glDeleteTextures(1, &texture);
pDestroyImage(g_display, image);
}
close(fd);
gbm_bo_destroy(bo);
}
gbm_device_destroy(gbm);
close(node);
}
int main(void) {
setvbuf(stdout, NULL, _IONBF, 0);
g_display = eglGetDisplay(EGL_DEFAULT_DISPLAY);
EGLint major, minor;
if (!eglInitialize(g_display, &major, &minor)) {
printf("FAIL: eglInitialize\n");
return 2;
}
const char* eglExtensions = eglQueryString(g_display, EGL_EXTENSIONS);
printf("EGL %d.%d dma_buf_import: %s\n", major, minor,
strstr(eglExtensions, "EGL_EXT_image_dma_buf_import") ? "yes" : "no");
eglBindAPI(EGL_OPENGL_ES_API);
const EGLint configAttribs[] = {EGL_SURFACE_TYPE, EGL_PBUFFER_BIT, EGL_RENDERABLE_TYPE, EGL_OPENGL_ES3_BIT,
EGL_RED_SIZE, 8, EGL_GREEN_SIZE, 8, EGL_BLUE_SIZE, 8, EGL_NONE};
EGLConfig config;
EGLint count = 0;
eglChooseConfig(g_display, configAttribs, &config, 1, &count);
if (count < 1) {
printf("FAIL: no config\n");
return 2;
}
const EGLint pbufferAttribs[] = {EGL_WIDTH, 16, EGL_HEIGHT, 16, EGL_NONE};
EGLSurface surface = eglCreatePbufferSurface(g_display, config, pbufferAttribs);
const EGLint contextAttribs[] = {EGL_CONTEXT_MAJOR_VERSION, 3, EGL_NONE};
EGLContext context = eglCreateContext(g_display, config, EGL_NO_CONTEXT, contextAttribs);
if (context == EGL_NO_CONTEXT || !eglMakeCurrent(g_display, surface, surface, context)) {
printf("FAIL: no ES 3 context\n");
return 2;
}
pCreateImage = (PFNEGLCREATEIMAGEKHRPROC)eglGetProcAddress("eglCreateImageKHR");
pDestroyImage = (PFNEGLDESTROYIMAGEKHRPROC)eglGetProcAddress("eglDestroyImageKHR");
pTargetTexture = (PFNGLEGLIMAGETARGETTEXTURE2DOESPROC)eglGetProcAddress("glEGLImageTargetTexture2DOES");
pQueryFormats = (PFNEGLQUERYDMABUFFORMATSEXTPROC)eglGetProcAddress("eglQueryDmaBufFormatsEXT");
pQueryModifiers = (PFNEGLQUERYDMABUFMODIFIERSEXTPROC)eglGetProcAddress("eglQueryDmaBufModifiersEXT");
pCreateSync = (PFNEGLCREATESYNCKHRPROC)eglGetProcAddress("eglCreateSyncKHR");
pDestroySync = (PFNEGLDESTROYSYNCKHRPROC)eglGetProcAddress("eglDestroySyncKHR");
printf("GL_RENDERER: %s\n", glGetString(GL_RENDERER));
const char* glExtensions = (const char*)glGetString(GL_EXTENSIONS);
const int external = strstr(glExtensions, "GL_OES_EGL_image_external_essl3") != NULL;
printf("GL_OES_EGL_image_external_essl3: %s\n", external ? "yes" : "no");
CHECK(external, "GL_OES_EGL_image_external_essl3 is not advertised");
EGLint formats[32], formatCount = 0;
pQueryFormats(g_display, 32, formats, &formatCount);
int nv12 = 0;
for (int i = 0; i < formatCount; ++i) {
printf("dma-buf format %.4s", (const char*)&formats[i]);
EGLuint64KHR modifiers[4];
EGLBoolean externalOnly[4];
EGLint modifierCount = 0;
pQueryModifiers(g_display, formats[i], 4, modifiers, externalOnly, &modifierCount);
for (int m = 0; m < modifierCount; ++m)
printf(" [modifier 0x%llx%s]", (unsigned long long)modifiers[m], externalOnly[m] ? " external-only" : "");
printf("\n");
nv12 |= formats[i] == (EGLint)DRM_FORMAT_NV12;
}
CHECK(nv12, "NV12 is not a listed dma-buf format");
SetUpDraw();
ForeignTest(DRM_FORMAT_NV12, 1, 12);
ForeignTest(DRM_FORMAT_P010, 2, 16);
GbmTest();
printf("%s (%d failure(s))\n", g_failures ? "FAILED" : "PASSED", g_failures);
eglMakeCurrent(g_display, EGL_NO_SURFACE, EGL_NO_SURFACE, EGL_NO_CONTEXT);
eglTerminate(g_display);
return g_failures ? 1 : 0;
}
@@ -0,0 +1,22 @@
#!/usr/bin/env bash
# push-tools.sh - install this skill's device- and container-side helpers on the phone.
# device/*.sh -> $DEV_TOOLS (run as root: adb shell 'su -c "sh $DEV_TOOLS/<x>.sh"')
# container/mgrun -> <container>/usr/local/bin/mgrun
# container/kwprep.sh -> <container>/root/kwprep.sh
# Needs the container running (its rootfs is mounted at $CT_ROOT only then).
# Overwrites same-named files other sessions may be using in $DEV_TOOLS; check first on a shared phone.
set -euo pipefail
here=$(cd "$(dirname "$0")" && pwd)
. "$here/env.sh"
adb_ shell "mkdir -p $DEV_TOOLS"
for f in "$here"/device/*.sh; do
tr -d '\r' < "$f" > "$ANL_DIR/push.tmp"
adb_ push "$ANL_DIR/push.tmp" "$DEV_TOOLS/$(basename "$f")" >/dev/null
echo "pushed $DEV_TOOLS/$(basename "$f")"
done
for f in mgrun kwprep.sh; do
tr -d '\r' < "$here/container/$f" > "$ANL_DIR/push.tmp"
adb_ push "$ANL_DIR/push.tmp" "$DEV_TOOLS/$f" >/dev/null
done
rm -f "$ANL_DIR/push.tmp"
adb_ shell "su -c 'cp $DEV_TOOLS/mgrun $CT_ROOT/usr/local/bin/mgrun && chmod 755 $CT_ROOT/usr/local/bin/mgrun && cp $DEV_TOOLS/kwprep.sh $CT_ROOT/root/kwprep.sh && echo installed mgrun kwprep.sh'"
@@ -0,0 +1,29 @@
#!/usr/bin/env bash
# rec.sh <name> - record 3 s of the screen while opening Kickoff, extract frames inside the container
# with ffmpeg, pull them to $ANL_DIR/af_<name>/ and build a 3x4 contact sheet $ANL_DIR/grid_<name>.png
# (needs Python + Pillow on Windows). Use it to catch flicker/black frames a single screenshot misses.
# Env: TAP="x y" Kickoff button position (default "69 1349" = bottom-left on the 2560x1600 tablet;
# device specific). Note: anland grabs the touchscreen (EVIOCGRAB); if `input tap` does nothing, use
# `input mouse tap x y`. Env CHANNEL (default rec).
set -euo pipefail
here=$(cd "$(dirname "$0")" && pwd)
. "$here/env.sh"
N=${1:?usage: rec.sh <name>}; TAP=${TAP:-69 1349}
adb_ shell "su -c 'input keyevent 111; sleep 1.5; (screenrecord --time-limit 3 --bit-rate 20000000 $DEV_SHARE/$N.mp4 &); sleep 0.7; input tap $TAP; sleep 3; cp $DEV_SHARE/$N.mp4 $CT_ROOT/root/$N.mp4'"
bash "$here/ct.sh" "${CHANNEL:-rec}" <<IN
rm -rf /root/af_$N; mkdir -p /root/af_$N; ffmpeg -loglevel quiet -i /root/$N.mp4 -vf scale=1000:-1 -fps_mode passthrough /root/af_$N/f%03d.png; ls /root/af_$N | wc -l
IN
adb_ shell "su -c 'rm -rf $DEV_SHARE/af_$N; cp -r $CT_ROOT/root/af_$N $DEV_SHARE/ && chmod -R 777 $DEV_SHARE/af_$N'"
rm -rf "$ANL_DIR/af_$N"; adb_ pull "$DEV_SHARE/af_$N" "$ANL_DIR/" >/dev/null
python - "$ANL_DIR/af_$N" "$ANL_DIR/grid_$N.png" <<'P'
import sys, os
from PIL import Image
d, out = sys.argv[1], sys.argv[2]
fs = sorted(os.listdir(d)); n = len(fs)
pick = [fs[min(n-1, int(i * n / 12))] for i in range(12)]
ims = [Image.open(os.path.join(d, f)) for f in pick]
w, h = ims[0].size; tw, th = w // 2, h // 2
g = Image.new('RGB', (tw * 3, th * 4))
for i, im in enumerate(ims): g.paste(im.resize((tw, th)), ((i % 3) * tw, (i // 3) * th))
g.save(out); print(n, pick)
P
@@ -0,0 +1,15 @@
#!/usr/bin/env bash
# shot.sh <name> [count] [interval-s] - take <count> (default 2) phone screenshots <interval> s apart
# (default 1), pull them to $ANL_DIR/<name>-<i>.png and print their md5. Identical hashes while an
# animated client (glmark2, a video) runs mean the frames are NOT reaching the screen.
set -euo pipefail
. "$(dirname "$0")/env.sh"
name=${1:?usage: shot.sh <name> [count] [interval]}; n=${2:-2}; dt=${3:-1}
cmd=""; i=1
while [ "$i" -le "$n" ]; do cmd="$cmd screencap -p $DEV_SHARE/$name-$i.png; sleep $dt;"; i=$((i+1)); done
adb_ shell "su -c '$cmd'"
i=1
while [ "$i" -le "$n" ]; do
adb_ pull "$DEV_SHARE/$name-$i.png" "$ANL_DIR/$name-$i.png" >/dev/null
md5sum "$ANL_DIR/$name-$i.png"; i=$((i+1))
done
@@ -0,0 +1,19 @@
#!/usr/bin/env bash
# sync-src.sh - FALLBACK path only (the WSL cross build, xbuild.sh, replaced it). Pushes the files the
# worktree changed since the container's source stamp into the container's MobileGL source tree
# (/root/mobilegl-unified-src), keeping mtimes so the in-container ninja build stays incremental;
# then build/install inside the container with anland's anland-build-client.sh
# (producers/kde/Arch_v5/mobilegl-tools/, gcc, -j4, slow on the phone).
# A full re-sync of the tree is anland's anland-sync-source.sh (MOBILEGL_REPO=<worktree>).
set -euo pipefail
. "$(dirname "$0")/env.sh"
SRC=$CT_ROOT/root/mobilegl-unified-src
STAMP=$(adb_ shell "su -c 'cat $SRC/.anland-build-stamp'" | tr -d '\r\n')
cd "$MGL_WORKTREE"
FILES=$( (git diff --name-only --diff-filter=d "$STAMP"; git ls-files --others --exclude-standard) \
| grep -v -E '^(3rdparty/|build-android|\.anland-build-stamp|\.claude/)' | sort -u)
echo "$(echo "$FILES" | wc -l) files since $STAMP"
tar --force-local -czf "$ANL_DIR/src-delta.tgz" $FILES
adb_ push "$ANL_DIR/src-delta.tgz" /data/local/tmp/src-delta.tgz >/dev/null
HEAD=$(git rev-parse HEAD)
adb_ shell "su -c 'tar xzf /data/local/tmp/src-delta.tgz -C $SRC && echo $HEAD > $SRC/.anland-build-stamp && echo synced'"
@@ -0,0 +1,9 @@
#!/bin/bash
# wsl-all.sh - run INSIDE WSL archlinux after wsl-test.sh configured $B: build every *Test target except
# DirectVulkan*/Benchmark* and run the unit label. NEVER run the DirectVulkan integration tests
# (they need a live device + manual intervention); never run two ctest suites at once.
B=${B:-$HOME/mgl-anl-bld}
cd $B
T=$(ninja -t targets all 2>/dev/null | grep -oE "^[A-Za-z]+Test:" | tr -d : | sort -u | grep -v -E "^(DirectVulkan|Benchmark)" | tr "\n" " ")
ninja -j14 MobileGLServer $T 2>&1 | grep -E "error:|FAILED" | head -10
ctest -L unit -j10 --timeout 300 -E "DirectVulkan" 2>&1 | tail -25
@@ -0,0 +1,13 @@
#!/bin/bash
# wsl-test.sh <targets...> - run INSIDE WSL archlinux: configure (once) and build MobileGL unit-test
# targets for the host (clang, disaggregated + in-process server ON). From Git Bash:
# MSYS2_ARG_CONV_EXCL='*' wsl -d archlinux -- bash /mnt/c/<worktree>/.claude/skills/anland-mobilegl-plasma/scripts/wsl-test.sh MultiSessionTest
# then ctest in $B (see wsl-all.sh). Env: W (worktree, WSL path; default: derived from this file),
# B (build dir, default ~/mgl-anl-bld). The Windows MSVC build of feat/disaggregated is broken; use this.
set -e
W=${W:-$(cd "$(dirname "$0")/../../../.." && pwd)}
B=${B:-$HOME/mgl-anl-bld}
if [ ! -f $B/build.ninja ]; then
cmake -S $W -B $B -G Ninja -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DMOBILEGL_BUILD_TEST=ON -DMOBILEGL_BUILD_DISAGGREGATED=ON -DMOBILEGL_BUILD_DISAGGREGATED_INPROC=ON -DMOBILEGL_BUILD_STAMP=wsl
fi
cmake --build $B --target "$@" -j16
@@ -0,0 +1,135 @@
#!/usr/bin/env bash
# x11-verify.sh <step> [...] - device verification of X11 on the GPU (docs/Disaggregated/notes/anland/
# plan-x11-gpu.md): Xwayland drawing with glamor on MobileGL, and GLX windows presented through
# DRI3 + Present from shared images. Steps, in the order a full run takes them:
#
# preflight read-only: battery, stack status, Xwayland package, render node, xcb libs,
# test apps, Xwayland's extensions and its GLX vendor
# measure <label> [s] glxgears 1280x720 (GEARS_ARGS; vblank_mode=0) for s seconds (10): its fps, then CPU/GPU/dma-buf
# counts (device/x11-measure.sh); also with MOBILEGL_GLX_PRESENT=readback
# deploy-client xbuild.sh + xdeploy.sh of this worktree's client (MGL_XBUILD_DIR)
# deploy-kwin kwin-stage.sh build from ANLAND_WORKTREE; waits for "== ALL DONE =="
# deploy-server build-android.sh + cyc.sh x11-server (swaps the APK lib, RESTARTS the session)
# check the evidence: Xwayland took glamor (journal), its MobileGL log, GLX's path
# line, dma-buf counts in Xwayland/KWin, the server's import lines
# apps X11 clients on screen: konsole and kate on Qt's xcb platform (glamor-drawn
# core X), glxgears (GLX), xterm/xeyes when installed; two screenshots
# stop kill the test apps
#
# Typical run: preflight; measure before; deploy-client; deploy-kwin; deploy-server; check; measure
# after; apps; stop. Battery: every load step refuses below MIN_BATTERY (40) %.
# Env: CHANNEL (ct.sh channel, default x11), ANLAND_WORKTREE (anland x11-gpu worktree, Windows
# form), MGL_XBUILD_DIR (default /home/swung/mgl-xbuild-x11), MIN_BATTERY.
set -euo pipefail
here=$(cd "$(dirname "$0")" && pwd)
. "$here/env.sh"
CH=${CHANNEL:-x11}
: "${MGL_XBUILD_DIR:=/home/swung/mgl-xbuild-x11}"
: "${ANLAND_WORKTREE:=$(cygpath -m "$USERPROFILE" 2>/dev/null)/AndroidStudioProjects/anland/.claude/worktrees/x11-gpu}"
export MGL_XBUILD_DIR ANLAND_WORKTREE
battery_ok() {
local level
level=$(adb_ shell "dumpsys battery" | sed -n 's/^ *level: *//p' | tr -d '\r')
echo "battery: ${level}%"
if [ "${level:-0}" -lt "${MIN_BATTERY:-40}" ]; then
echo "below ${MIN_BATTERY:-40}%: stopping GPU-heavy work; charge the phone first" >&2
return 1
fi
}
push_measure() {
tr -d '\r' < "$here/device/x11-measure.sh" > "$ANL_DIR/x11-measure.sh"
adb_ push "$ANL_DIR/x11-measure.sh" "$DEV_TOOLS/x11-measure.sh" >/dev/null
}
step=${1:?usage: x11-verify.sh <preflight|measure|deploy-client|deploy-kwin|deploy-server|check|apps|stop>}
shift
case "$step" in
preflight)
battery_ok || true
adb_ shell "su -c 'sh $DEV_TOOLS/status.sh'" || true
bash "$here/ct.sh" "$CH" <<'IN'
echo "== packages"; pacman -Q xorg-xwayland libxcb mesa-utils xorg-xdpyinfo qt6-base 2>&1 | sed 's/^/ /'
echo "== render nodes"; ls -l /dev/dri/ 2>&1 | sed 's/^/ /'
U=swung0x48; for n in /dev/dri/renderD*; do sudo -u $U test -r "$n" -a -w "$n" && echo " $U can open $n"; done
echo "== KWin's thin node"; tr '\0' '\n' < /proc/$(pgrep -x kwin_wayland)/environ 2>/dev/null | grep -E "MOBILEGL_GBM_NODE|ANLAND_XWAYLAND_GLAMOR|GBM_BACKEND" | sed 's/^/ /'
echo "== xcb libraries"; for l in libxcb-dri3.so.0 libxcb-present.so.0 libxcb-xfixes.so.0 libxcb-shm.so.0; do ls /usr/lib/$l >/dev/null 2>&1 && echo " $l ok" || echo " $l MISSING"; done
echo "== Xwayland"; pgrep -a -x Xwayland | sed 's/^/ /'
mgrun offscreen xdpyinfo 2>/dev/null | grep -E "^ +(DRI3|Present|MIT-SHM|XFIXES|GLX)$" | sed 's/^/ ext/'
mgrun offscreen glxinfo -B 2>&1 | grep -E "vendor|renderer|direct rendering" | sed 's/^/ /'
IN
;;
measure)
label=${1:?usage: x11-verify.sh measure <label> [seconds]}
secs=${2:-10}
battery_ok
push_measure
for mode in auto readback; do
printf '%s\n' \
"pkill -x glxgears; rm -f /tmp/mgl-x11-gears.client.log /tmp/x11-gears.out" \
"MGLOG=/tmp/mgl-x11-gears.log timeout $((secs + 8)) mgrun offscreen env vblank_mode=${VBLANK_MODE:-0} MOBILEGL_GLX_PRESENT=$mode glxgears ${GEARS_ARGS:--geometry 1280x720} > /tmp/x11-gears.out 2>&1 &" \
"sleep 3" | bash "$here/ct.sh" "$CH" >/dev/null
adb_ shell "su -c 'sh $DEV_TOOLS/x11-measure.sh $label-$mode glxgears $secs'" || true
sleep 6
printf '%s\n' "grep -a 'frames in' /tmp/x11-gears.out | tail -3; grep -a 'presented through' /tmp/mgl-x11-gears.client.log | head -1; pkill -x glxgears; true" \
| bash "$here/ct.sh" "$CH"
done
;;
deploy-client)
bash "$here/xbuild.sh"
CHANNEL=$CH bash "$here/xdeploy.sh"
;;
deploy-kwin)
CHANNEL=$CH bash "$here/kwin-stage.sh" build
for _ in $(seq 1 120); do
out=$(printf 'tail -3 /root/kwin-build.log\n' | bash "$here/ct.sh" "$CH" 2>/dev/null || true)
echo "$out" | tail -1
echo "$out" | grep -q "== ALL DONE ==" && exit 0
echo "$out" | grep -qiE "error|failed" && { echo "KWin build failed" >&2; exit 1; }
sleep 30
done
echo "KWin build did not finish in an hour" >&2; exit 1
;;
deploy-server)
battery_ok
bash "$here/build-android.sh"
CHANNEL=$CH bash "$here/cyc.sh" x11-server
;;
check)
bash "$here/ct.sh" "$CH" <<'IN'
echo "== Xwayland's command line (no -shm with glamor)"; pgrep -a -x Xwayland | sed 's/^/ /'
echo "== Xwayland's glamor (journal)"
journalctl --no-pager --since -30min 2>/dev/null | grep -aiE "glamor|Xwayland.*(EGL|gbm|dri3|falling back)" | tail -10 | sed 's/^/ /'
echo "== Xwayland's MobileGL client log"; grep -aE "presented|shared image|declin|Fatal|ERROR" /tmp/mgl-xwayland.client.log /tmp/mgl-xwayland.client 2>/dev/null | tail -8 | sed 's/^/ /' || echo " (none)"
X=$(pgrep -x Xwayland); K=$(pgrep -x kwin_wayland)
echo "== dma-bufs held: Xwayland $(ls -l /proc/$X/fd 2>/dev/null | grep -c dmabuf), kwin $(ls -l /proc/$K/fd 2>/dev/null | grep -c dmabuf)"
echo "== Xwayland's GLX vendor and extensions"
mgrun offscreen xdpyinfo 2>/dev/null | grep -E "^ +(DRI3|Present|MIT-SHM)$" | sed 's/^/ ext/'
echo "== a GLX window's path"
rm -f /tmp/mgl-x11-check.client.log
MGLOG=/tmp/mgl-x11-check.log timeout 6 mgrun offscreen glxgears >/dev/null 2>&1
grep -a "presented through\|leaves DRI3" /tmp/mgl-x11-check.client.log | sed 's/^/ /'
IN
adb_ logcat -d | grep -aE "MobileGL.*(shared image|dma-buf import|refused)" | tail -8 || true
adb_ logcat -b crash -d | tail -5 || true
;;
apps)
battery_ok
printf '%s\n' \
"pkill -x xterm; pkill -x xeyes; pkill -x glxgears; pkill -x konsole; pkill -x kate" \
"MGLOG=/tmp/mgl-x11-konsole.log mgrun offscreen env QT_QPA_PLATFORM=xcb konsole >/dev/null 2>&1 &" \
"MGLOG=/tmp/mgl-x11-kate.log mgrun offscreen env QT_QPA_PLATFORM=xcb kate >/dev/null 2>&1 &" \
"command -v xterm >/dev/null && MGLOG=/tmp/mgl-x11-xterm.log mgrun offscreen xterm >/dev/null 2>&1 &" \
"command -v xeyes >/dev/null && MGLOG=/tmp/mgl-x11-xeyes.log mgrun offscreen xeyes >/dev/null 2>&1 &" \
"MGLOG=/tmp/mgl-x11-gears.log mgrun offscreen glxgears >/dev/null 2>&1 &" \
"sleep 6; pgrep -a -x konsole; pgrep -a -x kate; pgrep -a -x xterm; pgrep -a -x xeyes; pgrep -a -x glxgears; true" | bash "$here/ct.sh" "$CH"
bash "$here/shot.sh" x11-apps 2 1
;;
stop)
printf '%s\n' "pkill -x xterm; pkill -x xeyes; pkill -x glxgears; pkill -x konsole; pkill -x kate; true" | bash "$here/ct.sh" "$CH"
;;
*)
echo "unknown step $step" >&2; exit 2
;;
esac
@@ -0,0 +1,42 @@
#!/usr/bin/env bash
# xbuild.sh [--reconfigure] [extra ninja targets...] - cross-build the container CLIENT: libMobileGL.so
# (glibc aarch64, glvnd EGL/GLX vendor) + mobilegl_gbm.so (GBM backend), in WSL archlinux with
# clang + lld against a sysroot exported from the container (xsysroot/refresh-sysroot.sh).
# Full build ~150 s, incremental seconds. Deploy with xdeploy.sh.
# Env: MGL_WORKTREE (source, Windows form), MGL_XBUILD_DIR (WSL build dir, default ~/mgl-xbuild-a64),
# MGL_JOBS (16), WSL_DISTRO (archlinux).
# Prereq in WSL: clang, lld, llvm (lld major MUST equal clang/llvm-libs major - lld 23 on llvm 22
# broke ld.lld), cmake, ninja; ~/sysroots/aarch64-arch.cmake + ~/sysroots/arch-kde-mgl.
set -euo pipefail
. "$(dirname "$0")/env.sh"
STAMP=$(git -C "$MGL_WORKTREE" rev-parse HEAD)
RECONF=0
if [ "${1:-}" = "--reconfigure" ]; then RECONF=1; shift; fi
EXTRA="$*"
SRC_WSL=$(win2wsl "$MGL_WORKTREE")
wsl -d "${WSL_DISTRO:-archlinux}" -- env SRC="$SRC_WSL" STAMP="$STAMP" RECONF="$RECONF" EXTRA="$EXTRA" \
OUT="${MGL_XBUILD_DIR:-}" JOBS="${MGL_JOBS:-16}" bash -s <<'IN' 2>&1 | tr -d '\0' | grep -av '^wsl:'
set -euo pipefail
OUT=${OUT:-$HOME/mgl-xbuild-a64}
TC=$HOME/sysroots/aarch64-arch.cmake
[ -f "$TC" ] || { echo "missing toolchain $TC (run xsysroot/refresh-sysroot.sh)"; exit 1; }
ld.lld --version >/dev/null || { echo "ld.lld is not runnable in WSL (lld/llvm version mismatch?)"; exit 1; }
t0=$(date +%s)
if [ "$RECONF" = 1 ] || [ ! -f "$OUT/build.ninja" ]; then
rm -rf "$OUT/CMakeCache.txt" "$OUT/CMakeFiles"
cmake -S "$SRC" -B "$OUT" -G Ninja --toolchain "$TC" -DMOBILEGL_BUILD_STAMP="$STAMP" \
-DCMAKE_BUILD_TYPE=Release -DMOBILEGL_BUILD_TEST=OFF -DMOBILEGL_BUILD_BENCHMARK=OFF \
-DMOBILEGL_BUILD_DISAGGREGATED=ON -DMOBILEGL_ENABLE_LTO=OFF
elif ! grep -qx "MOBILEGL_BUILD_STAMP:STRING=$STAMP" "$OUT/CMakeCache.txt"; then
cmake -S "$SRC" -B "$OUT" -DMOBILEGL_BUILD_STAMP="$STAMP" >/dev/null
fi
t1=$(date +%s)
cmake --build "$OUT" --target MobileGL mobilegl_gbm $EXTRA -j"$JOBS"
t2=$(date +%s)
echo "stamp: $STAMP"
echo "configure: $((t1-t0))s build: $((t2-t1))s"
for f in "$OUT/libMobileGL.so" "$OUT/MobileGL/MG_Gbm/mobilegl_gbm.so"; do
ls -la "$f"; echo " windows: $(wslpath -w "$f")"
done
IN
exit "${PIPESTATUS[0]}"
@@ -0,0 +1,65 @@
#!/usr/bin/env bash
# xdeploy.sh - install the WSL cross-built libMobileGL.so + mobilegl_gbm.so into the container, exactly
# like anland's anland-build-client.sh: copy to a NEW inode and rename into place (overwriting a mapped
# .so in place crashes every running client, KWin included), refresh the libEGL/libGL aliases, the
# glvnd GLX vendor link, the EGL vendor JSON (system-wide 10_mobilegl.json + the /opt copy), the client
# config /etc/mobilegl/client.conf and the GLX/GBM selection drop-ins. Does not restart anything: already-running clients
# keep the old inode; restart the client (or the session) to pick up the new library.
# Env: MGL_XBUILD_DIR (WSL build dir), CHANNEL (ct.sh channel, default xd), WSL_DISTRO.
set -euo pipefail
here=$(cd "$(dirname "$0")" && pwd)
. "$here/env.sh"
mkdir -p "$ANL_DIR/xdist"
# 1. Copy the build outputs out of WSL.
wsl -d "${WSL_DISTRO:-archlinux}" -- env OUT="${MGL_XBUILD_DIR:-}" D="$(win2wsl "$ANL_DIR")/xdist" bash -s <<'IN' 2>&1 | tr -d '\0' | grep -av '^wsl:'
set -e; OUT=${OUT:-$HOME/mgl-xbuild-a64}
cp -f "$OUT/libMobileGL.so" "$D/libMobileGL.so"; cp -f "$OUT/MobileGL/MG_Gbm/mobilegl_gbm.so" "$D/mobilegl_gbm.so"
sha256sum "$D/libMobileGL.so" "$D/mobilegl_gbm.so"
IN
# 2. Push to the phone and stage them in the container's /root/xdeploy (the container's /tmp and the
# phone's /data/local/tmp are different places; the rootfs is reachable at $CT_ROOT).
adb_ shell "mkdir -p $DEV_TOOLS/xdist"
adb_ push "$ANL_DIR/xdist/libMobileGL.so" "$DEV_TOOLS/xdist/libMobileGL.so"
adb_ push "$ANL_DIR/xdist/mobilegl_gbm.so" "$DEV_TOOLS/xdist/mobilegl_gbm.so"
adb_ shell "su -c 'mkdir -p $CT_ROOT/root/xdeploy && cp $DEV_TOOLS/xdist/libMobileGL.so $DEV_TOOLS/xdist/mobilegl_gbm.so $CT_ROOT/root/xdeploy/ && rm -f $DEV_TOOLS/xdist/*.so'"
# 3. Install inside the container.
bash "$here/ct.sh" "${CHANNEL:-xd}" <<'IN'
set -euo pipefail
in=/root/xdeploy
mkdir -p /opt/mobilegl/lib /opt/mobilegl/share/glvnd/egl_vendor.d
cp "$in/libMobileGL.so" /opt/mobilegl/lib/libMobileGL.so.new
mv -f /opt/mobilegl/lib/libMobileGL.so.new /opt/mobilegl/lib/libMobileGL.so
for lib in libEGL.so libEGL.so.1 libGL.so libGL.so.1; do ln -sf libMobileGL.so "/opt/mobilegl/lib/$lib"; done
ln -sf /opt/mobilegl/lib/libMobileGL.so /usr/lib/libGLX_mobilegl.so.0
vendor='{"file_format_version":"1.0.0","ICD":{"library_path":"/opt/mobilegl/lib/libMobileGL.so"}}'
# Per-process forcing (__EGL_VENDOR_LIBRARY_FILENAMES=<this file>) keeps working off the /opt copy.
printf '%s\n' "$vendor" >/opt/mobilegl/share/glvnd/egl_vendor.d/50_mobilegl.json
# System-wide vendor: 10_ sorts ahead of the system's 50_mesa.json, so stock libglvnd asks MobileGL
# first and moves on to the other vendor whenever MobileGL declines (no server reachable).
mkdir -p /usr/share/glvnd/egl_vendor.d
printf '%s\n' "$vendor" >/usr/share/glvnd/egl_vendor.d/10_mobilegl.json.new
mv -f /usr/share/glvnd/egl_vendor.d/10_mobilegl.json.new /usr/share/glvnd/egl_vendor.d/10_mobilegl.json
# Client settings every process reads when its environment does not say otherwise (the backend
# comes from /etc/mobilegl/backend, which switch.sh rewrites).
mkdir -p /etc/mobilegl
cat >/etc/mobilegl/client.conf <<'CONF'
# MobileGL client configuration, read by every process that loads libMobileGL.so.
# KEY=value lines; a MOBILEGL_* variable in a process's environment overrides the line here.
# The backend is /etc/mobilegl/backend. Written by xdeploy.sh / anland-build-client.sh.
MOBILEGL_TRANSPORT=spawn
MOBILEGL_IPC_DATA=shm
MOBILEGL_IPC_CONTROL=unix:@anland-mobilegl
CONF
# What glvnd's libGLX and libgbm cannot infer: environment.d reaches the systemd user manager (every
# Plasma service and what it launches, terminals included), profile.d reaches login shells.
mkdir -p /etc/environment.d /etc/profile.d
printf '%s\n' '__GLX_VENDOR_LIBRARY_NAME=mobilegl' 'GBM_BACKEND=mobilegl' >/etc/environment.d/10-mobilegl.conf
printf '%s\n' 'export __GLX_VENDOR_LIBRARY_NAME=mobilegl' 'export GBM_BACKEND=mobilegl' >/etc/profile.d/mobilegl.sh
mkdir -p /usr/lib/gbm
cp "$in/mobilegl_gbm.so" /usr/lib/gbm/mobilegl_gbm.so.new
mv -f /usr/lib/gbm/mobilegl_gbm.so.new /usr/lib/gbm/mobilegl_gbm.so
rm -f "$in/libMobileGL.so" "$in/mobilegl_gbm.so"
sha256sum /opt/mobilegl/lib/libMobileGL.so /usr/lib/gbm/mobilegl_gbm.so
ls /usr/share/glvnd/egl_vendor.d/ /etc/mobilegl/ /etc/environment.d/
echo DEPLOYED
IN
@@ -0,0 +1,47 @@
# Cross toolchain: WSL clang -> aarch64 Arch Linux ARM (glibc) container "arch-kde-mgl".
# Sysroot exported from the container by refresh-sysroot.sh (default $HOME/sysroots/arch-kde-mgl).
set(CMAKE_SYSTEM_NAME Linux)
set(CMAKE_SYSTEM_PROCESSOR aarch64)
set(_triple aarch64-unknown-linux-gnu)
if (DEFINED ENV{MGL_SYSROOT})
set(_sysroot "$ENV{MGL_SYSROOT}")
else()
set(_sysroot "$ENV{HOME}/sysroots/arch-kde-mgl")
endif()
# The container's GCC install dir (crtbegin/crtend, libgcc, libstdc++ via /usr/include/c++/<ver>).
file(GLOB _gcc_dirs LIST_DIRECTORIES true "${_sysroot}/usr/lib/gcc/${_triple}/*")
list(SORT _gcc_dirs COMPARE NATURAL ORDER DESCENDING)
list(GET _gcc_dirs 0 _gcc_install_dir)
set(CMAKE_SYSROOT "${_sysroot}")
set(CMAKE_C_COMPILER clang)
set(CMAKE_CXX_COMPILER clang++)
set(CMAKE_ASM_COMPILER clang)
set(CMAKE_C_COMPILER_TARGET ${_triple})
set(CMAKE_CXX_COMPILER_TARGET ${_triple})
set(CMAKE_ASM_COMPILER_TARGET ${_triple})
set(CMAKE_AR llvm-ar)
set(CMAKE_RANLIB llvm-ranlib)
set(CMAKE_NM llvm-nm)
set(CMAKE_OBJCOPY llvm-objcopy)
set(CMAKE_STRIP llvm-strip)
set(_xflags "--gcc-install-dir=${_gcc_install_dir}")
set(CMAKE_C_FLAGS_INIT "${_xflags}")
set(CMAKE_CXX_FLAGS_INIT "${_xflags}")
set(CMAKE_ASM_FLAGS_INIT "${_xflags}")
set(CMAKE_EXE_LINKER_FLAGS_INIT "-fuse-ld=lld")
set(CMAKE_SHARED_LINKER_FLAGS_INIT "-fuse-ld=lld")
set(CMAKE_MODULE_LINKER_FLAGS_INIT "-fuse-ld=lld")
set(CMAKE_LINKER_TYPE LLD)
set(CMAKE_FIND_ROOT_PATH "${_sysroot}")
set(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
set(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
set(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)
set(CMAKE_FIND_ROOT_PATH_MODE_PACKAGE ONLY)
set(ENV{PKG_CONFIG_SYSROOT_DIR} "${_sysroot}")
set(ENV{PKG_CONFIG_LIBDIR} "${_sysroot}/usr/lib/pkgconfig:${_sysroot}/usr/share/pkgconfig")
set(ENV{PKG_CONFIG_PATH} "")
@@ -0,0 +1,26 @@
#!/usr/bin/env bash
# xsysroot/refresh-sysroot.sh - (re)export the cross sysroot from the container into WSL
# ~/sysroots/arch-kde-mgl and install the toolchain file as ~/sysroots/aarch64-arch.cmake.
# Read-only on the container side (it only tars headers/libs into /root and moves the tarball out).
# Run once on a new machine, and again after the container upgrades glibc/gcc/vulkan/libdrm/gbm;
# then `bash xbuild.sh --reconfigure`. Env: CHANNEL (default xr), WSL_DISTRO.
set -euo pipefail
here=$(cd "$(dirname "$0")" && pwd)
. "$here/../env.sh"
cp -f "$here/aarch64-arch.cmake" "$ANL_DIR/aarch64-arch.cmake"
bash "$here/../ct.sh" "${CHANNEL:-xr}" <<'IN'
set -e; cd /
L=$(ls -d usr/lib/*crt*.o usr/lib/lib{c,m,mvec,pthread,dl,rt,util,resolv,anl,stdc++,stdc++fs,stdc++exp,supc++,gcc_s,atomic,vulkan,drm,drm_*,gbm,z}.* usr/lib/lib{c,m}_nonshared* usr/lib/ld-linux* usr/lib/pkgconfig usr/share/pkgconfig usr/share/cmake/VulkanHeaders usr/lib/cmake/Vulkan* 2>/dev/null)
tar czf /root/xsysroot.tar.gz --exclude='usr/lib/gcc/*/*/cc1' --exclude='usr/lib/gcc/*/*/cc1plus' --exclude='usr/lib/gcc/*/*/lto1' --exclude='usr/lib/gcc/*/*/lto-wrapper' --exclude='usr/lib/gcc/*/*/collect2' --exclude='usr/lib/gcc/*/*/plugin' --exclude='usr/lib/gcc/*/*/install-tools' --exclude='usr/lib/gcc/*/*/liblto_plugin.so' lib usr/include usr/lib/gcc $L
echo SYSROOT_TAR_DONE
IN
adb_ shell "su -c 'mv $CT_ROOT/root/xsysroot.tar.gz $DEV_TOOLS/xsysroot.tar.gz && chmod 644 $DEV_TOOLS/xsysroot.tar.gz'"
adb_ pull "$DEV_TOOLS/xsysroot.tar.gz" "$ANL_DIR/xsysroot.tar.gz"
adb_ shell "su -c 'rm -f $DEV_TOOLS/xsysroot.tar.gz'"
wsl -d "${WSL_DISTRO:-archlinux}" -- env A="$(win2wsl "$ANL_DIR")" bash -s <<'IN' 2>&1 | tr -d '\0' | grep -av '^wsl:'
set -e; S=$HOME/sysroots/arch-kde-mgl; rm -rf "$S.new"; mkdir -p "$S.new"
tar xzf "$A/xsysroot.tar.gz" -C "$S.new"
rm -rf "$S"; mv "$S.new" "$S"; cp -f "$A/aarch64-arch.cmake" "$HOME/sysroots/aarch64-arch.cmake"
du -sh "$S"; ls "$S/usr/lib/gcc/aarch64-unknown-linux-gnu/"
IN
rm -f "$ANL_DIR/xsysroot.tar.gz"
+7 -3
View File
@@ -18,6 +18,10 @@ on:
- '**/*.md'
workflow_dispatch:
env:
# Software-rendered CI submissions may exceed the interactive GPU hang budget.
MOBILEGL_GPU_HANG_BUDGET_MS: "0"
# ONE LIVE RUN PER BRANCH, as in test.yml: a superseded push is cancelled rather than left queued
# behind its successor; a manual dispatch gets a group of its own; dev queues rather than cancels.
concurrency:
@@ -477,7 +481,7 @@ jobs:
extra_retrace_args=()
if [ "${MOBILEGL_CI_TRANSPORT}" = "inproc" ]; then
extra_retrace_args+=(
--env "MOBILEGL_TRANSPORT=inproc;MOBILEGL_IPC_ROLE_SPLIT_STATE=1;MOBILEGL_IPC_STRICT_ERRORS=1;MOBILEGL_IPC_RUN_AHEAD=1"
--env "MOBILEGL_TRANSPORT=inproc;MOBILEGL_IPC_ROLE_SPLIT_STATE=1;MOBILEGL_IPC_STRICT_ERRORS=1;MOBILEGL_IPC_RUN_AHEAD=1;MOBILEGL_GPU_HANG_BUDGET_MS=${MOBILEGL_GPU_HANG_BUDGET_MS}"
--require-inproc
)
elif [ "${MOBILEGL_CI_TRANSPORT}" = "spawn" ]; then
@@ -493,12 +497,12 @@ jobs:
# into the CLIENT's process and SetWindowHandle is refused on the way across with
# Fatal{UnmigratedSurface, "AndroidNativeWindow@P12"} (Rule H).
extra_retrace_args+=(
--env "MOBILEGL_TRANSPORT=spawn;MOBILEGL_IPC_ROLE_SPLIT_STATE=1;MOBILEGL_IPC_STRICT_ERRORS=1;MOBILEGL_IPC_RUN_AHEAD=1"
--env "MOBILEGL_TRANSPORT=spawn;MOBILEGL_IPC_ROLE_SPLIT_STATE=1;MOBILEGL_IPC_STRICT_ERRORS=1;MOBILEGL_IPC_RUN_AHEAD=1;MOBILEGL_GPU_HANG_BUDGET_MS=${MOBILEGL_GPU_HANG_BUDGET_MS}"
--use-pbuffer
--require-spawn
)
else
extra_retrace_args+=(--env "MOBILEGL_TRANSPORT=monolith")
extra_retrace_args+=(--env "MOBILEGL_TRANSPORT=monolith;MOBILEGL_GPU_HANG_BUDGET_MS=${MOBILEGL_GPU_HANG_BUDGET_MS}")
fi
# Bliss needs the newer signed ANGLE variant plus sampler mipmap
# min-filter downgrading on ANGLE llvmpipe.
+35 -90
View File
@@ -21,6 +21,10 @@ on:
required: false
default: "087685d1"
env:
# Software-rendered CI submissions may exceed the interactive GPU hang budget.
MOBILEGL_GPU_HANG_BUDGET_MS: "0"
# ONE LIVE RUN PER BRANCH. The disaggregation branch pushes (often docs-only) commits minutes apart
# and each push fans out to ~330 retrace legs here and ~275 jobs in apk.yml; with no group, runs
# sat queued for 10+ hours behind their own successors. A manual dispatch gets a group of its own,
@@ -413,6 +417,12 @@ jobs:
--discovery "${RUNNER_TEMP}/primary-inproc-lane.json" \
--junit "${RUNNER_TEMP}/primary-inproc.xml" --output "${RUNNER_TEMP}/primary-inproc-proof.json"
# The run log needs a signed-in reader; annotations do not. Name every failed ctest entry,
# with the lines of its output that say why, where anyone reading the run can see them.
- name: Annotate failed entries
if: failure()
run: python3 scripts/ci/junit_annotate.py "${RUNNER_TEMP}"
- name: Upload integration lane evidence
if: always()
uses: actions/upload-artifact@v7
@@ -1169,6 +1179,12 @@ jobs:
# arm proof read; verify-split-logs/ is whatever the control steps left behind, which for
# twelve entries is a deliberately red re-run under the same name. Uploading only the
# second - which is what this step did before the copy above existed - loses the first.
# The run log needs a signed-in reader; annotations do not. Name every failed ctest entry,
# with the lines of its output that say why, where anyone reading the run can see them.
- name: Annotate failed entries
if: failure()
run: python3 scripts/ci/junit_annotate.py "${RUNNER_TEMP}"
- name: Upload verify-split lane logs
if: always()
uses: actions/upload-artifact@v7
@@ -1481,7 +1497,7 @@ jobs:
working-directory: build-split
env:
MOBILEGL_ITEST_REQUIRE_GPU: "1"
run: ctest --output-on-failure -L unit --no-tests=error -j "$(nproc)"
run: ctest --output-on-failure -L unit --no-tests=error -j "$(nproc)" --output-junit "${RUNNER_TEMP}/unit-split.xml"
# P7 wave 2-F, PH-7 (1). THE TCP SUPERVISOR LANE NEEDS flatc, AND CI HAD NONE.
#
@@ -1648,6 +1664,12 @@ jobs:
CONTROL_TMPDIR: ${{ runner.temp }}
run: bash "${GITHUB_WORKSPACE}/scripts/ci/split_negative_controls.sh"
# The run log needs a signed-in reader; annotations do not. Name every failed ctest entry,
# with the lines of its output that say why, where anyone reading the run can see them.
- name: Annotate failed entries
if: failure()
run: python3 scripts/ci/junit_annotate.py "${RUNNER_TEMP}"
- name: Upload split lane logs
if: always()
uses: actions/upload-artifact@v7
@@ -1737,7 +1759,7 @@ jobs:
env:
MOBILEGL_ITEST_REQUIRE_GPU: "1"
MOBILEGL_IPC_STRICT_ERRORS: "1"
run: ctest --output-on-failure -L unit --no-tests=error -j "$(nproc)"
run: ctest --output-on-failure -L unit --no-tests=error -j "$(nproc)" --output-junit "${RUNNER_TEMP}/unit-strict.xml"
# ---- P5e (ra), MG_Remote/CONTRACT-P5E.md §7: ONE STEP, TWO ARMS, AND THE TREE PICKS ----
#
@@ -2246,6 +2268,12 @@ jobs:
slp.require_green(lane, sys.argv[3], names)
PY
# The run log needs a signed-in reader; annotations do not. Name every failed ctest entry,
# with the lines of its output that say why, where anyone reading the run can see them.
- name: Annotate failed entries
if: failure()
run: python3 scripts/ci/junit_annotate.py "${RUNNER_TEMP}"
- name: Upload strict lane logs
if: always()
uses: actions/upload-artifact@v7
@@ -2751,7 +2779,11 @@ jobs:
&& [ '${{ matrix.case }}' = 'improved-transparency-minecraft-26.3' ]; then
export MOBILEGL_MAGMA_DISABLE_BLENDED_DEPTH_WRITE=1
fi
ctest -V --no-tests=error -R '^MobileGLTraceReplay\.${{ matrix.case }}\.${{ matrix.backend }}$'
ctest -V --no-tests=error -R '^MobileGLTraceReplay\.${{ matrix.case }}\.${{ matrix.backend }}$' --output-junit "${RUNNER_TEMP}/retrace.xml"
- name: Annotate failed entries
if: failure()
run: python3 scripts/ci/junit_annotate.py --excerpts 1 "${RUNNER_TEMP}"
- name: Upload core dumps
if: failure()
@@ -3506,27 +3538,10 @@ jobs:
runs-on: ubuntu-latest
# Deliberately independent of build-linux: these are source-level gates, they take
# seconds, and a broken build must not hide a drifted interface.
env:
# THE CURRENT PHASE's base ref, for the two G5 region gates below. It is 37da3c3a - P4a's
# base ref, INTEGRATOR-DECISIONS ID-1 - and NOT the workflow's baseline_sha input: that input
# is the SYMBOL baseline (087685d1, P1's G1 reading) and it is empty on a push, whereas these
# gates ask "did the do-not-touch list move since the phase started".
#
# IT MOVED FROM P3a's 44c2b5cf TO P4a's 37da3c3a WITH THE PHASE, and that is a deliberate
# narrowing rather than a loss: P3a's eleven functions were compared against 44c2b5cf at P3a's
# own exit and were byte-identical there, so 37da3c3a carries the same bodies (measured: the
# eleven shas at 37da3c3a are the eleven shas at 44c2b5cf, and FlushPendingRangesFrom's is
# still the sha pinned in the script at 3e298c9a). What the two gates now both answer is "did
# anything on the list move during P4a", which is the question this phase can act on.
BASELINE: "37da3c3a"
steps:
- name: Checkout repo
uses: actions/checkout@v6
with:
# The G5 gate reads Managers.cpp at BASELINE with `git show`, which a depth-1 checkout
# does not have. Nothing else in this job needs history.
fetch-depth: 0
# The seven generators all read MG_Pipe/*.def, so regenerating and diffing is what
# keeps the two interface tables, the wire records, the verify comparators, the
@@ -3655,76 +3670,6 @@ jobs:
python3 tools/trace_replay/test_run_tcp_matrix.py
python3 tools/trace_replay/test_compare_actuals.py
# A GATE as of P3a (G5). "pool 与延迟释放原样搬" (ROADMAP.md:19) is meant literally: the
# buffer pool, the deferred-release drain and the three persistently mapped rings move
# VERBATIM, and ARCHITECTURE.md:515 says why - their retire happens only inside Present, so
# a batching or ordering change there starves them, and nothing else in this workflow can
# see it. P3a rewrites the rest of Managers.cpp by design, so a file diff says nothing; the
# script extracts the ELEVEN named bodies and compares their hashes on their own.
#
# Eleven and not ten (ID-15): Managers.cpp carries the three-tier flush drain TWICE, once
# per preprocessor arm, and a push build compiles only FlushPendingRangesFrom while the
# untouched FlushPendingRangesNow lives in the `#else`. Hashing the pull name alone would
# protect text no shipping build compiles, so both are hashed - the pull ladder against
# BASELINE, the push ladder against a sha pinned in the script at 3dadd4c1 (ID-41), because
# that one was born in P3a and is compared against the reviewed body rather than against the
# base ref.
#
# Scoped to the disaggregation branch and to a manual dispatch, deliberately: the question
# is "did these eleven move since P3a started", and BASELINE is P3a's base ref. On dev,
# where unrelated buffer fixes land on their own schedule, the same comparison would be
# asking a question nobody posed - it belongs with the TEMPORARY trigger lines at the top
# of this file and retires with them.
#
# --self-test is the half that keeps it honest, and it is not optional: a comparison that
# silently stopped comparing produces exactly the same green as eleven untouched bodies. It
# runs eight canned controls - eleven bodies extracted, an untouched copy compared equal, an
# edit OUTSIDE them ignored, each of the three perturbation targets (ClearBufferPool and BOTH
# flush ladders) reported BY NAME, plus the pin-precedence control and a one-token edit to the
# pinned ladder compared against the pin - and fails if any of them does not answer.
# Same shape as gen_pipe.py --self-test above.
- name: P3a regions unchanged (G5)
if: ${{ github.ref == 'refs/heads/feat/disaggregated' || github.event_name == 'workflow_dispatch' }}
run: bash scripts/p3a_untouched_regions.sh "${BASELINE}" HEAD
- name: P3a untouched-region self-test (G5)
if: ${{ github.ref == 'refs/heads/feat/disaggregated' || github.event_name == 'workflow_dispatch' }}
run: bash scripts/p3a_untouched_regions.sh --self-test
# A GATE AS OF P4a (G5), and a SECOND script rather than an edit to the one above. P4a extends
# the same claim to the rest of ARCHITECTURE.md:318's do-not-touch list - the unpack PBO ring's
# staging repack and its two helpers, the attachment permutation, the D24S8 sampling-emulation
# core and the format-caveat handler - which is SEVENTEEN regions across THREE files
# (BRIEF-P4A.md D-N: P3a's eleven, which P4a must not touch either, plus P4a's six). The
# parent's SOURCE_PATH is a single file, so the extension needed a per-region source path and
# a region KIND (DepthStencilSamplingReadImpl is a namespace, not a function); everything else
# about the extraction is its parent's, verbatim.
#
# Both scripts run. The parent keeps answering its own question against its own eleven, so a
# regression in either half names itself, and neither gate can be silenced by editing the
# other's list.
#
# Same feat/disaggregated-or-dispatch guard as the P3a step, for the same reason: the question
# is "did these move since the phase started", and on dev - where unrelated buffer and texture
# fixes land on their own schedule - it would be a question nobody posed. It belongs with the
# TEMPORARY trigger lines at the top of this file and retires with them.
#
# --self-test is the half that keeps it honest and is not optional: a comparison that silently
# stopped comparing produces exactly the same green as seventeen untouched regions. It runs
# three positive controls (seventeen regions extracted, an untouched copy compared equal, an
# edit OUTSIDE them invisible in all three files) and NINE negative ones - ClearBufferPool,
# FlushPendingRangesNow, RecomputeBackendColorSlots and StageBlocksIntoUnpackRing, each
# perturbed at its HEAD and at its TAIL and each required to be named BY NAME, plus a
# one-token edit to the pinned FlushPendingRangesFrom body compared against the pin - and
# fails if any of them does not answer.
- name: P4a regions unchanged (G5)
if: ${{ github.ref == 'refs/heads/feat/disaggregated' || github.event_name == 'workflow_dispatch' }}
run: bash scripts/p4a_untouched_regions.sh "${BASELINE}" HEAD
- name: P4a untouched-region self-test (G5)
if: ${{ github.ref == 'refs/heads/feat/disaggregated' || github.event_name == 'workflow_dispatch' }}
run: bash scripts/p4a_untouched_regions.sh --self-test
# Warning only for now: the disaggregation documents are still being written, and a
# lint that fails a rewrite in progress teaches people to ignore it. It becomes
# --strict when the documents settle.
+31
View File
@@ -0,0 +1,31 @@
name: Arch WSL GPU smoke
on:
workflow_dispatch:
push:
branches:
- ci/arch-wsl-gpu-runner
paths:
- .github/workflows/wsl-gpu-smoke.yml
- docs/ci/arch-wsl-gpu-runner.md
permissions: {}
concurrency:
group: arch-wsl-gpu-smoke
cancel-in-progress: false
jobs:
gpu-smoke:
runs-on: [self-hosted, Linux, X64, arch, wsl, gpu, gles, vulkan]
timeout-minutes: 5
steps:
- name: Report runner identity
shell: bash
run: |
printf 'Runner: %s\n' "$RUNNER_NAME"
uname -n
uname -a
- name: Verify hardware GLES and Vulkan
shell: bash
run: /usr/local/libexec/mobilegl-runner/gpu-smoke
+33
View File
@@ -304,6 +304,7 @@ set(SOURCE_FILES
MobileGL/ConfigLoader.cpp
MobileGL/MG_Util/Debug/Log.cpp
MobileGL/MG_Util/X11/DisplayGuard.cpp
MobileGL/MG_Util/Async/JobNode.cpp
MobileGL/MG_Util/Async/ShaderCompilePool.cpp
@@ -371,6 +372,7 @@ set(SOURCE_FILES
MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/FixIterationRPSubgroupScratchPass.cpp
MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/EmulateSubgroupsPass.cpp
MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/NormalizeRectCoordinatesPass.cpp
MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/DeduplicateImageTypesPass.cpp
MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/Lower1DArrayImagesPass.cpp
MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/Lower1DSampledImagesPass.cpp
MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/BakeImageFormatsPass.cpp
@@ -401,6 +403,8 @@ set(SOURCE_FILES
MobileGL/MG_Impl/GLXImpl/Exporting/Definitions.cpp
MobileGL/MG_Impl/GLXImpl/Exporting/GlvndVendor.cpp
MobileGL/MG_Impl/GLXImpl/GLXImpl.cpp
MobileGL/MG_Impl/GLXImpl/X11Present.cpp
MobileGL/MG_Impl/GLXImpl/X11PresentXcb.cpp
MobileGL/MG_Impl/GLXImpl/LookUp/LookUp.cpp
MobileGL/MG_Impl/EGLImpl/Exporting/Definitions.cpp
@@ -437,6 +441,7 @@ set(SOURCE_FILES
MobileGL/MG_Backend/Init.cpp
MobileGL/MG_Backend/BackendObject.cpp
MobileGL/MG_Backend/GpuHangWatch.cpp
MobileGL/MG_Backend/DirectGLES/DirectGLES.cpp
MobileGL/MG_Backend/DirectGLES/BackendObject_DirectGLES.cpp
@@ -446,6 +451,7 @@ set(SOURCE_FILES
MobileGL/MG_Backend/DirectVulkan/DirectVulkan.cpp
MobileGL/MG_Backend/DirectVulkan/BackendObject_DirectVulkan.cpp
MobileGL/MG_Backend/DirectVulkan/MagmaSession.cpp
MobileGL/MG_Backend/DirectVulkan/VmaImpl.cpp
MobileGL/MG_Backend/DirectVulkan/Renderer/VulkanRenderer.cpp
MobileGL/MG_Backend/DirectVulkan/Renderer/SwapchainObject.cpp
@@ -463,6 +469,7 @@ set(SOURCE_FILES
MobileGL/MG_Backend/DirectVulkan/Renderer/VkSamplerManager.cpp
MobileGL/MG_Backend/DirectVulkan/Renderer/VkClearManager.cpp
MobileGL/MG_Backend/DirectVulkan/Renderer/VkRenderPassManager.cpp
MobileGL/MG_Backend/DirectVulkan/Renderer/GpuProgressMarkers.cpp
MobileGL/MG_State/GLState/Core.cpp
MobileGL/MG_State/GLState/ShareGroupState.cpp
@@ -645,6 +652,9 @@ if (MOBILEGL_BUILD_DISAGGREGATED)
MobileGL/MG_Remote/FatalFunnel.cpp
MobileGL/MG_Remote/Wire/PipeWireCodec.cpp
MobileGL/MG_Remote/Client/ClientSession.cpp
# Whether the configured server is there at all, asked before a display is offered so the
# system's EGL/GLX/GBM loaders can move on to another implementation when it is not.
MobileGL/MG_Remote/Client/ServerProbe.cpp
MobileGL/MG_Remote/Client/EmitTables.cpp
MobileGL/MG_Remote/Client/CapsMirror.cpp
# P5 c1: the client role's BackendObject. pActiveBackendObject holds one of these
@@ -674,6 +684,9 @@ if (MOBILEGL_BUILD_DISAGGREGATED)
MobileGL/MG_Remote/Server/PairAcceptor.cpp
# P11 B2: T0's Offers on the aux socket, matched to their map_persistent records by seq.
MobileGL/MG_Remote/Server/AdoptInbox.cpp
# Shared images: server-allocated AHardwareBuffers exported to clients as dma-buf descriptors
# (docs/Disaggregated/notes/anland/plan-ahb-dmabuf.md).
MobileGL/MG_Remote/Server/SharedImageRegistry.cpp
MobileGL/MG_Remote/Server/ServerSpawn.cpp
# P5f fc: the value frame that replaced the control mailbox's function-pointer payload,
# and its codec against protocol.fbs's SurfaceOp/SurfaceReply (the spawn arm's seam).
@@ -915,6 +928,15 @@ if (UNIX AND NOT APPLE)
target_link_options(${CMAKE_PROJECT_NAME} PRIVATE "LINKER:-Bsymbolic-functions")
endif()
# ---- Shared images outside GL: the C ABI and the GBM backend (Linux client, split builds) ----
# docs/Disaggregated/notes/anland/plan-ahb-dmabuf.md. libMobileGL.so exports the shared-image C ABI
# (MG_Impl/SharedImageApi.h); mobilegl_gbm.so (MG_Gbm/) is a separate GBM loader module that reaches
# it through dlsym. Self-contained: nothing outside this block refers to either.
if (MOBILEGL_BUILD_DISAGGREGATED AND UNIX AND NOT APPLE AND NOT ANDROID)
target_sources(${CMAKE_PROJECT_NAME} PRIVATE MobileGL/MG_Impl/SharedImageApi.cpp)
add_subdirectory(MobileGL/MG_Gbm)
endif()
if (ANDROID)
target_link_libraries(${CMAKE_PROJECT_NAME} PUBLIC
android
@@ -988,6 +1010,17 @@ if (NOT ANDROID AND NOT MOBILEGL_IOS)
endif ()
endif ()
# Discover gtest cases when ctest starts, one binary after another, instead of in a post-build
# step. The post-build discovery of the CMake releases that never pass the target name down
# writes every binary's `--gtest_output=json` result to ONE per-directory file
# (cmake_test_discovery_<hash of an empty name>.json), so binaries linked in parallel in the same
# directory overwrite and delete each other's list: the registered count then differs between
# runs of the same tree and a binary ends up with none or only some of its cases. The pre-test
# mode reads the list from stdout, serially, and also works when the binary is cross-built.
if (NOT DEFINED CMAKE_GTEST_DISCOVER_TESTS_DISCOVERY_MODE)
set(CMAKE_GTEST_DISCOVER_TESTS_DISCOVERY_MODE PRE_TEST)
endif()
if (NOT ANDROID)
# Enable testing in the top-level scope so a CTestTestfile.cmake is emitted
# at the build-tree root. This lets `ctest` be invoked from the top-level
+13 -1
View File
@@ -14,7 +14,7 @@ namespace MobileGL::MG_Config {
inline const String ProjectName = "MobileGL";
inline const String CoreName = "MobileGL Core";
inline const String CoreVendor = "MobileGL-Dev (BZLZHH, Swung0x48, Tungsten)";
inline const Version CoreVersion = {26, 9, 0, "-dev", VersionType::Development};
inline const Version CoreVersion = {26, 10, 0, "-dev", VersionType::Development};
inline const VersionStringFormatAttrib DefaultVersionStringFormatAttrib = {2, 2, 0, true, true};
inline const Uint64 CacheVersion = 0;
@@ -130,6 +130,11 @@ namespace MobileGL::MG_Config {
Bool AdvertiseFp64 = false;
// MOBILEGL_MAGMA_R11G11B10F_FALLBACK: use fallback format for R11G11B10F on Vulkan.
Bool MagmaR11G11B10FFallback = false;
// MOBILEGL_MAGMA_ALLOW_INVALID_SPIRV: hand a shader module that fails SPIR-V validation to
// the driver anyway. Off by default: Magma declines such a program (no shader module, its
// draws refused by name), because a driver that accepts invalid SPIR-V may fault the GPU
// into device loss. For diagnosing a driver that copes with a specific module only.
Bool MagmaAllowInvalidSpirv = false;
// MOBILEGL_MAGMA_FRAMESINFLIGHT: requested Magma frames in flight, defaulting to 3.
Uint32 MagmaFramesInFlight = 3;
// MOBILEGL_MAGMA_MAX_DRAWS_PER_COMMAND_BUFFER: GL draws and dispatches Magma records into
@@ -629,6 +634,13 @@ namespace MobileGL::MG_Config {
// poll loop cannot outlive a late report. 0 = every poll crosses, as before P10: the A/B
// arm and the red-once of the no-reply gate (FencePollScenario).
Uint32 PollEscalate = 64;
// MOBILEGL_IPC_RECOVER (client): 1 (default) = after a device loss ends this client's
// session, the application's next eglCreateContext / eglCreate*Surface / eglInitialize
// dials the server for a FRESH session, and contexts created from then on work. Contexts
// (and their objects) from before the loss stay lost, as EGL's power-management event
// has them; surfaces survive and are re-created on the new session when next used.
// 0 = the loss is final for the process (every verb DECLINED for the rest of its life).
Uint32 RecoverAfterLoss = 1;
// MOBILEGL_IPC_ADOPT_TIER, THE ZERO-COPY SWITCH (ruling ID-P11-14, CONTRACT-P11 B2):
// UNSET (kAdoptTierUnset, the default since 2026-09-29, the user's decision) = T0 where the
// session can run it, T2 QUIETLY where it cannot (MGLOG_D only, still counted);
+137 -4
View File
@@ -6,6 +6,9 @@
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#if defined(__ANDROID__)
#include <sys/system_properties.h>
#endif
#include "Config.h"
#if MOBILEGL_PIPE_PUSH
// For kMGPipeSubsystemsMigratedAtP5e, the push build's PipePush default (the P2, P3a and P4a
@@ -15,12 +18,10 @@
#endif
#include <cerrno>
#include <cstdio>
#include <cstdlib>
#if MOBILEGL_BUILD_DISAGGREGATED
// std::strcmp, for InitIpc's server-role check (P12). Split-only, so the pull build's TU is
// unchanged (G1).
// std::strcmp: InitIpc's server-role check (P12) and the system configuration's.
#include <cstring>
#endif
#ifndef _WIN32
extern char** environ;
@@ -49,6 +50,115 @@ namespace MobileGL::MG_ConfigLoader {
return (key.compare(0, 6, "LIBGL_") == 0 || key.compare(0, 9, "MOBILEGL_") == 0);
}
// ---- the system configuration -----------------------------------------------------------
//
// A library that is every process's GL implementation (installed as the system's EGL/GLX
// vendor) cannot rely on each process having been started with MOBILEGL_* in its environment:
// most are started by something that has never heard of it. So the settings a whole machine
// shares live in files, and the environment only overrides them:
//
// environment > /etc/mobilegl/backend (MOBILEGL_BACKEND_TYPE only) > client.conf > built-in
//
// client.conf is `KEY=value` lines; blank lines and lines starting with '#' are skipped, the
// key and the value are trimmed, and one pair of matching quotes around the value is removed.
// Only MOBILEGL_* / LIBGL_* keys are taken, exactly as from the environment. The backend file
// holds just the backend's name (the file the backend switch rewrites), and names it over a
// MOBILEGL_BACKEND_TYPE line in client.conf so that switch cannot be silently shadowed.
//
// MOBILEGL_CONFIG_FILE / MOBILEGL_BACKEND_FILE name other files; set to empty they turn the
// file off. The defaults exist only on desktop Linux: an Android app configures its process
// itself, and a server process (MOBILEGL_IPC_ROLE=server) never reads the clients' file.
namespace {
#if defined(__linux__) && !defined(__ANDROID__)
constexpr const char* kDefaultConfigFile = "/etc/mobilegl/client.conf";
constexpr const char* kDefaultBackendFile = "/etc/mobilegl/backend";
#else
constexpr const char* kDefaultConfigFile = "";
constexpr const char* kDefaultBackendFile = "";
#endif
String FilePathSetting(const char* variable, const char* fallback) {
const char* value = std::getenv(variable);
return value != nullptr ? String(value) : String(fallback);
}
String Trimmed(const String& text) {
const auto first = text.find_first_not_of(" \t\r\n");
if (first == String::npos) return String();
const auto last = text.find_last_not_of(" \t\r\n");
return text.substr(first, last - first + 1);
}
Bool IsServerProcess() {
const char* role = std::getenv("MOBILEGL_IPC_ROLE");
return role != nullptr && std::strcmp(role, "server") == 0;
}
// Reads `path` into `out` without replacing what `out` already holds. A missing file is
// not an error: the defaults are the absence of one.
void ReadConfigFile(const String& path, UnorderedMap<String, String>& out, Bool log) {
if (path.empty()) return;
std::FILE* file = std::fopen(path.c_str(), "r");
if (file == nullptr) return;
char buffer[4096];
Uint32 lineNumber = 0;
while (std::fgets(buffer, sizeof(buffer), file) != nullptr) {
++lineNumber;
const String line = Trimmed(buffer);
if (line.empty() || line[0] == '#') continue;
const auto equals = line.find('=');
const String key = equals == String::npos ? String() : Trimmed(line.substr(0, equals));
if (key.empty() || !IsAcceptedPrefix(key)) {
if (log) {
MGLOG_W("Config: %s:%u is not a MOBILEGL_*/LIBGL_* KEY=value line; ignored", path.c_str(),
lineNumber);
}
continue;
}
String value = Trimmed(line.substr(equals + 1));
if (value.size() >= 2 && (value.front() == '"' || value.front() == '\'') && value.back() == value.front()) {
value = value.substr(1, value.size() - 2);
}
out.emplace(key, value);
}
std::fclose(file);
}
void ReadBackendFile(const String& path, UnorderedMap<String, String>& out) {
if (path.empty()) return;
std::FILE* file = std::fopen(path.c_str(), "r");
if (file == nullptr) return;
char buffer[256];
const Bool read = std::fgets(buffer, sizeof(buffer), file) != nullptr;
std::fclose(file);
if (!read) return;
const String name = Trimmed(buffer);
if (!name.empty()) out.emplace("MOBILEGL_BACKEND_TYPE", name);
}
// Everything the system files say, the backend file ahead of client.conf (first emplace
// wins). Re-read on every call: callers are few (Init, and a handful of lookups before it),
// and a file that changed between two processes' starts must not be served stale.
void ReadSystemConfig(UnorderedMap<String, String>& out, Bool log) {
if (IsServerProcess()) return;
ReadBackendFile(FilePathSetting("MOBILEGL_BACKEND_FILE", kDefaultBackendFile), out);
ReadConfigFile(FilePathSetting("MOBILEGL_CONFIG_FILE", kDefaultConfigFile), out, log);
}
} // namespace
Bool LookupSetting(const char* key, String& outValue) {
if (const char* value = std::getenv(key)) {
outValue = value;
return true;
}
UnorderedMap<String, String> system;
ReadSystemConfig(system, false);
const auto it = system.find(key);
if (it == system.end()) return false;
outValue = it->second;
return true;
}
inline void InitializeAcceptedEnvVariables() {
if (!acceptedEnvVariablesMap) {
acceptedEnvVariablesMap = MakeUnique<UnorderedMap<String, String>>();
@@ -79,6 +189,15 @@ namespace MobileGL::MG_ConfigLoader {
}
}
}
// Below the environment: a key the environment set is never replaced.
UnorderedMap<String, String> system;
ReadSystemConfig(system, true);
for (const auto& [key, value] : system) {
if (acceptedEnvVariablesMap->emplace(key, value).second) {
MGLOG_D("Config: Accepted system setting: %s=%s", key.c_str(), value.c_str());
}
}
}
inline void QueryEnvVariable(const String& key, String& outValue, const String& defaultValue) {
@@ -229,6 +348,7 @@ namespace MobileGL::MG_ConfigLoader {
features.MagmaDeriveNumSubgroups = QueryEnvQuirkOverride("MOBILEGL_MAGMA_DERIVE_NUM_SUBGROUPS");
features.AdvertiseFp64 = QueryEnvFlag("MOBILEGL_ADVERTISE_FP64");
features.MagmaR11G11B10FFallback = QueryEnvFlag("MOBILEGL_MAGMA_R11G11B10F_FALLBACK");
features.MagmaAllowInvalidSpirv = QueryEnvFlag("MOBILEGL_MAGMA_ALLOW_INVALID_SPIRV");
features.MagmaFramesInFlight = QueryEnvUint32("MOBILEGL_MAGMA_FRAMESINFLIGHT", 3, 1, 64);
features.MagmaMaxDrawsPerCommandBuffer =
QueryEnvUint32("MOBILEGL_MAGMA_MAX_DRAWS_PER_COMMAND_BUFFER", 16384, 0, 1u << 24);
@@ -302,6 +422,17 @@ namespace MobileGL::MG_ConfigLoader {
inline void InitBackendType() {
String backendTypeStr;
QueryEnvVariable("MOBILEGL_BACKEND_TYPE", backendTypeStr, "DirectGLES");
#if defined(__ANDROID__)
// An Android app that hosts the library sets the environment itself, before the library
// loads; the debug property is how a developer picks the backend for such a process
// without rebuilding it (`adb shell setprop debug.mobilegl.backend DirectVulkan`).
char property[PROP_VALUE_MAX] = {};
if (__system_property_get("debug.mobilegl.backend", property) > 0) {
MGLOG_I("Config: debug.mobilegl.backend=%s overrides MOBILEGL_BACKEND_TYPE=%s", property,
backendTypeStr.c_str());
backendTypeStr = property;
}
#endif
#define ENTRY(backendType) \
if (backendTypeStr == #backendType) { \
MG_Config::ActiveBackendType = BackendType::backendType; \
@@ -428,6 +559,8 @@ namespace MobileGL::MG_ConfigLoader {
ipc.PboReadbackSync = QueryEnvUint32("MOBILEGL_IPC_PBO_READBACK_SYNC", 0, 0, 1);
// P10: fence polls answered locally; 0 is the A/B arm (Config.h).
ipc.PollEscalate = QueryEnvUint32("MOBILEGL_IPC_POLL_ESCALATE", 64, 0, 1000000);
// A lost session is replaced by a fresh one on the application's next create (Config.h).
ipc.RecoverAfterLoss = QueryEnvUint32("MOBILEGL_IPC_RECOVER", 1, 0, 1);
// The verify harness compares the pushed block against the applier per verb; a
// batched queue lets the comparer read a supplied field mid-apply, which is a
// torn read rather than a divergence. The batch is therefore off whenever the
+26
View File
@@ -24,6 +24,9 @@
#if MOBILEGL_PIPE_VERIFY
#include <MG_Backend/MGPipe/PipeInputs.h> // MGPipeVerifyFlushSummary, before Debug::Close in DestroyImpl
#endif
#if MOBILEGL_BUILD_DISAGGREGATED
#include <MG_Remote/Client/ServerProbe.h>
#endif
#include <atomic>
#include <mutex>
@@ -102,6 +105,10 @@ namespace MobileGL {
MG_State::GLState::ClearProgramTranslationCache();
MG_Backend::gBackendFunctionsTable = {};
g_isInitialized = false;
#if MOBILEGL_BUILD_DISAGGREGATED
// A later re-initialization asks whether the server is there again.
MG_Remote::Client::ResetServerProbeCache();
#endif
#if MOBILEGL_PIPE_VERIFY
// BEFORE Close, and from here rather than from a static destructor (V1 fix round 2):
// Close() nulls the role's sink and Log.cpp's next write reopens it with "w", so the
@@ -220,6 +227,25 @@ namespace MobileGL {
DestroyImpl(true);
}
Bool ImplementationAvailable() {
if (g_isInitialized.load(std::memory_order_acquire)) {
#if MOBILEGL_BUILD_DISAGGREGATED
// Initialized: under a split transport the session either came up (a backend object
// exists) or it did not, and a display on a session that never came up cannot draw.
if (MG_Config::Transport != MG_Config::TransportMode::Monolith) {
return MG_Backend::pActiveBackendObject != nullptr;
}
#endif
return true;
}
#if MOBILEGL_BUILD_DISAGGREGATED
// Not yet: the bring-up would dial the configured server and wait for it. Ask first.
return MG_Remote::Client::ConfiguredServerAvailable();
#else
return true;
#endif
}
// MobileGL's lifecycle is owned entirely by the host-API layers
// (EGL/WGL/CGL): initialization happens lazily on the first entry point
// via EnsureInitialized(), and full teardown happens deterministically
+16
View File
@@ -26,13 +26,29 @@ namespace MobileGL {
namespace MG_ConfigLoader {
void Init();
// One setting resolved the way Init() resolves it - the environment, then the system
// configuration files (ConfigLoader.cpp) - for the few readers that must decide something
// before MobileGL initializes. False when neither names it.
Bool LookupSetting(const char* key, String& outValue);
} // namespace MG_ConfigLoader
// Whether this process can be served right now. False when MobileGL is configured to reach a
// server that is not there (or a split session already failed to come up), or when the process
// asked for software rendering (LIBGL_ALWAYS_SOFTWARE, desktop Linux), so a
// caller that can choose another implementation - the system's EGL/GLX/GBM loaders - should
// decline rather than hand out something that cannot draw. Cheap: once true it stays true
// until the library is torn down, and a probe is one non-blocking connect with a short bound.
Bool ImplementationAvailable();
namespace MG_Backend {
void Init();
} // namespace MG_Backend
namespace MG_Impl {
void Init();
// GL framebuffer 0, built into the calling thread's GLContext (the process default at
// Init) and installed as the one every later context starts with. Init's step, and the
// rebuild after a recovered device loss, whose fresh session holds none of the old one's.
void BuildDefaultFramebuffer();
} // namespace MG_Impl
} // namespace MobileGL
+12
View File
@@ -97,6 +97,8 @@ namespace MobileGL::MG_Backend {
static_cast<SizeT>(target) >= kFormatCapabilityTextureTargetCount) {
return kFormatCapabilityTargetCount;
}
// An external texture is a 2D one wherever a format can go.
if (target == TextureTarget::External) return static_cast<SizeT>(TextureTarget::Texture2D);
return static_cast<SizeT>(target);
}
@@ -388,6 +390,16 @@ namespace MobileGL::MG_Backend {
m_windowHandle = {};
}
namespace {
MG_Util::Damage::Region& PresentDamageSlot() {
thread_local MG_Util::Damage::Region damage = MG_Util::Damage::Region::Full();
return damage;
}
} // namespace
const MG_Util::Damage::Region& CurrentPresentDamage() { return PresentDamageSlot(); }
void SetCurrentPresentDamage(const MG_Util::Damage::Region& damage) { PresentDamageSlot() = damage; }
Bool BackendObject::SwapEGLBuffers(EGLDisplay dpy, EGLSurface draw) {
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
if (!m_eglDisplayInitialized || m_eglDisplay != dpy) {
+231
View File
@@ -9,6 +9,7 @@
#pragma once
#include <Includes.h>
#include "MG_State/GLState/TextureState/TextureEnum.h"
#include <MG_Util/Damage/Damage.h>
#if MOBILEGL_BUILD_DISAGGREGATED
#include <MG_Pipe/MGPipeHandles.h>
#endif
@@ -583,26 +584,256 @@ namespace MobileGL {
Uint32 Height = 0;
};
// SHARED IMAGES (docs/Disaggregated/notes/anland/plan-ahb-dmabuf.md): server-allocated colour
// buffers exported to clients as dma-buf descriptors.
//
// The server-side view of one image, as a backend sees it: `NativeBuffer` is the image's
// AHardwareBuffer (null where there is none), `Fourcc` a DRM fourcc - ABGR8888 (R, G, B, A in
// memory, GL RGBA8) or XBGR8888 (alpha ignored); ARGB8888 / XRGB8888 images are stored the
// same way (SharedImageRegistry.h says why), so a backend binds every one as RGBA8.
struct SharedImageView {
Uint64 Id = 0;
Uint32 Width = 0;
Uint32 Height = 0;
Uint32 Fourcc = 0;
void* NativeBuffer = nullptr;
};
// The client-side answer of an allocation: the image's id and its exported descriptor
// (`Fd`, owned by the caller), with the plane layout a linux-dmabuf wl_buffer is created with.
struct SharedImageExport {
Uint64 Id = 0;
int Fd = -1;
Uint32 Width = 0;
Uint32 Height = 0;
Uint32 Fourcc = 0;
Uint32 Stride = 0;
Uint32 Offset = 0;
Uint64 Modifier = 0;
// A YUV image's Cb Cr plane: in the same descriptor, at this offset and pitch.
Uint32 Plane1Stride = 0;
Uint32 Plane1Offset = 0;
};
// An import of a two-plane (YUV) dma-buf: where its planes are in the one descriptor, and
// EGL_EXT_image_dma_buf_import's colour hints (0 = not given).
struct SharedImageImportLayout {
Uint32 PlaneCount = 1;
Uint32 Offset[2] = {0, 0};
Uint32 Pitch[2] = {0, 0};
Uint32 ColorSpace = 0;
Uint32 Range = 0;
Uint32 SitingX = 0;
Uint32 SitingY = 0;
// The DRM format modifier the import named; HasModifier false = none (the implicit one).
Bool HasModifier = false;
Uint64 Modifier = 0;
};
// THE DAMAGE OF THE PRESENT BEING APPLIED on this thread: the server's apply thread sets it
// around the backend's Present (from the present record), and a backend that can hand damage
// to its native present reads it there. Full whenever nothing set it.
const MG_Util::Damage::Region& CurrentPresentDamage();
void SetCurrentPresentDamage(const MG_Util::Damage::Region& damage);
class BackendObject {
public:
virtual ~BackendObject() = default;
// ---- shared images, SERVER side (a backend implements these) ---------------------
//
// Copies the default framebuffer of the surface current on this (the apply) thread into
// `image`, rows TOP FIRST - GL's top row (y = height-1) lands in the image's first memory
// row, which is what a wl_buffer's first row is - scaled to the image's extent if the two
// differ. The image is ORDERED THROUGH ITS SYNC STATE (SharedImageRegistry.h): the copy
// waits for the image's pending read fences before it writes, and its own completion
// is published with SharedImages::PublishWrite as a sync_file - the call returns once
// the copy is submitted, never waiting for it on the CPU (a backend with no fence export
// waits and publishes -1). A reader binding the image waits for that fence on its GPU
// when the write generation moved, and at its frame boundary publishes its frame's fence
// with PublishRead / ReadTracker. The application's GL state is untouched. False = not
// supported / failed (logged).
//
// `region` (GL window coordinates, origin bottom-left) is the part of the frame to copy;
// Full copies all of it. The rest of the image is left as it is - the client sends a
// region only for an image that already holds the frame outside it - and a backend that
// cannot be sure the image's contents survived (a target it has never written) copies
// all of it regardless.
virtual Bool BlitDefaultFramebufferToSharedImage(const SharedImageView& image,
const MG_Util::Damage::Region& region) {
(void)image, (void)region;
return false;
}
// EGL_BUFFER_AGE_EXT of the draw surface of the context current on this (the apply)
// thread: how many presents ago the buffer the next frame draws into held the frame then
// presented; 0 = unknown (the application repaints everything). Asking says the client
// relies on the answer, so from then on the backend keeps what it presents.
//
// `damageRegionFollows`: the client declares the region it draws (SetCurrentDamageRegion)
// before drawing. A driver whose only buffer age is EGL_KHR_partial_update's keeps a
// buffer's content only outside that region, so without one coming it is not asked.
virtual Int32 QueryCurrentBufferAge(Bool damageRegionFollows) {
(void)damageRegionFollows;
return 0;
}
// eglSetDamageRegionKHR for that draw surface: the frame draws `region` (GL window
// coordinates; Full = all of it) and nothing outside it. A backend without the notion
// ignores it.
virtual Bool SetCurrentDamageRegion(const MG_Util::Damage::Region& region) {
(void)region;
return true;
}
// IMPLICIT SYNC BY FLUSH (docs/Disaggregated/notes/anland/plan-x11-gpu.md): the session
// on this (the apply) thread ends a frame at a glFlush. Every shared image it used since
// its last boundary is published as written and read by one fence of everything it
// submitted (SharedImages::ReadTracker::PublishFrameAsWrite), and from now on its first
// use of an image in a frame waits for that image's pending reads (it may be about to
// overwrite what another session still samples). A backend without shared images has
// nothing to publish.
virtual Bool PublishSharedImageAccesses() { return true; }
// EGL_ANDROID_native_fence_sync's fence command, on the server: in `*fence`, a sync_file
// (the caller's to close) that signals once everything the session submitted so far has
// completed - and, for a session that writes images, what PublishSharedImageAccesses
// does. -1 when there is none to give: the backend cannot export one, or the work has
// already completed - in both cases it HAS completed when this returns.
virtual Bool ExportNativeFence(int* fence) {
*fence = -1;
return PublishSharedImageAccesses();
}
// ---- shared images, CLIENT side (BackendObject_Remote implements these) ----------
//
// A new image, allocated by the server; `out->Fd` is the caller's to close.
virtual Bool AllocateSharedImage(Uint32 width, Uint32 height, Uint32 fourcc, SharedImageExport* out) {
(void)width, (void)height, (void)fourcc, (void)out;
return false;
}
// Names the image `fd` was exported from (the caller keeps `fd`). False: not one of this
// server's images, or the size/format disagree.
virtual Bool ImportSharedImage(int fd, Uint32 width, Uint32 height, Uint32 fourcc, Uint64* outId) {
(void)fd, (void)width, (void)height, (void)fourcc, (void)outId;
return false;
}
// The same for a YUV fourcc, whose planes and hints `layout` gives. A YUV buffer the
// server did not export is taken too (its labelled CPU-copy fallback).
virtual Bool ImportSharedImagePlanes(int fd, Uint32 width, Uint32 height, Uint32 fourcc,
const SharedImageImportLayout& layout, Uint64* outId) {
(void)fd, (void)width, (void)height, (void)fourcc, (void)layout, (void)outId;
return false;
}
// Drops this client's reference (an allocation's or an import's).
virtual Bool ReleaseSharedImage(Uint64 id) {
(void)id;
return false;
}
// In the GL stream: the current context's default framebuffer, copied top row first into
// the image (BlitDefaultFramebufferToSharedImage on the server) - `region` of it, Full for
// all. Returns when it completed.
virtual Bool PresentToSharedImage(Uint64 id, const MG_Util::Damage::Region& region) {
(void)id, (void)region;
return false;
}
// In the GL stream: EGL_BUFFER_AGE_EXT of the current draw surface (QueryCurrentBufferAge
// on the server). False: the backend has no answer (the caller reports 0).
virtual Bool QueryBufferAge(Bool damageRegionFollows, EGLint* age) {
(void)damageRegionFollows, (void)age;
return false;
}
// In the GL stream: eglSetDamageRegionKHR for the current draw surface
// (SetCurrentDamageRegion on the server).
virtual Bool SetDamageRegion(const MG_Util::Damage::Region& region) {
(void)region;
return false;
}
// In the GL stream: the texture whose lifetime id is `textureLifetimeId` takes the image as
// its level-0 storage (glEGLImageTargetTexture2DOES), until its next storage definition.
// The caller has already defined level 0 as Width x Height RGBA8 with no data.
virtual Bool AttachSharedImageToTexture(Uint64 textureLifetimeId, Uint64 id) {
(void)textureLifetimeId, (void)id;
return false;
}
// In the GL stream, from glFlush/glFinish: PublishSharedImageAccesses on the server,
// returning once the fence is published. Skipped (true) when nothing reached the stream
// since the last one.
virtual Bool FlushSharedImageAccesses() { return false; }
// In the GL stream, from eglCreateSync(EGL_SYNC_NATIVE_FENCE_ANDROID) with no descriptor:
// ExportNativeFence on the server. `*fence` is the sync_file it handed back (the
// caller's to close), or -1 when it had none - the fenced work has then completed.
virtual Bool CreateNativeFence(int* fence) {
*fence = -1;
return false;
}
virtual void Initialize() = 0;
virtual Bool InitCapabilities() = 0;
virtual Bool InitWindowSurface() = 0;
// A served session's apply thread - the one thread that runs this backend from here on -
// calls this first, in a process that may serve several sessions at once. A backend whose
// native state is process-global by construction gives this object its own copy and binds
// it to the calling thread. The default has nothing per process to separate.
virtual void BindSessionStateToThisThread() {}
// A served session's apply thread asks this each time it wakes, BEFORE it applies another
// record: true when this session's own GPU work faulted or hung the GPU and the backend has
// latched the session for it (MGPipeSessionLatch, family DeviceLost), so the next frame -
// which would very likely stop the GPU again - is never applied. Cheap by contract: no
// wait, at most a status read. The default knows of no fault.
virtual Bool LatchIfGpuFaulted() { return false; }
// A served session's ring moved to another of its client's contexts (bind_context). A
// backend whose surface follows the native context it binds needs nothing; one that keeps
// a single native surface binding re-binds the surface that context was last made current
// with. `token` is the client's context token (the EGLContext MakeEGLCurrent was given).
virtual void OnClientContextBound(Uint64 token) { (void)token; }
virtual Bool InitializeEGLDisplay(EGLDisplay dpy, EGLint* major, EGLint* minor);
virtual Bool CreateEGLWindowSurface(EGLSurface surface, const WindowHandle& handle);
virtual Bool ResizeEGLWindowSurface(EGLSurface surface, Uint32 width, Uint32 height);
virtual Bool CreateEGLPbufferSurface(EGLSurface surface, EGLint width, EGLint height);
virtual Bool MakeEGLCurrent(EGLDisplay dpy, EGLSurface draw, EGLSurface read, EGLContext ctx);
virtual Bool SwapEGLBuffers(EGLDisplay dpy, EGLSurface draw);
// eglSwapBuffersWithDamage: `damage` (GL window coordinates) is a hint for the native
// present. A backend with nothing to hand it to swaps as SwapEGLBuffers does.
virtual Bool SwapEGLBuffersWithDamage(EGLDisplay dpy, EGLSurface draw, const MG_Util::Damage::Region& damage) {
(void)damage;
return SwapEGLBuffers(dpy, draw);
}
// Forwards the app-requested eglSwapInterval to the backend's native
// presentation path (no-op for backends without a SetSwapInterval hook).
virtual void SetEGLSwapInterval(Int interval);
virtual void ReleaseEGLSurface(EGLSurface surface);
virtual void ReleaseEGLResources();
// ---- the server's own display window going away and coming back -------------------
//
// `window` (a window surface's native window) is about to be destroyed - the display's
// screen went off. Release ONLY what is bound to it: every window surface on it keeps its
// client handle and draws into a placeholder from here on, and every context, object and
// device stays. Called on the thread that runs this backend; nothing may reference the
// window once it returns true. False: not supported (the caller ends the session instead).
virtual Bool SuspendServerWindow(void* window) {
(void)window;
return false;
}
// The surfaces SuspendServerWindow parked are rebuilt on `window` (a new native window,
// `width`x`height`), and their new extent is published as a creation publishes it. False:
// they could not be (the caller ends the session).
virtual Bool ResumeServerWindow(void* window, Uint32 width, Uint32 height) {
(void)window, (void)width, (void)height;
return false;
}
// `window` changed size to `width`x`height` (the display's report - the one size to publish).
// Its surfaces publish that extent to the client now, without waiting for a present; a
// swapchain that has to follow is rebuilt when no frame is half-recorded, else at the
// present. False: nothing was published (no surface on it, or not supported).
virtual Bool ServerWindowResized(void* window, Uint32 width, Uint32 height) {
(void)window, (void)width, (void)height;
return false;
}
// CLIENT side, from eglQuerySurface(EGL_WIDTH/EGL_HEIGHT): a backend whose surface extents
// change under the application (a remote server's window) applies what it has received.
virtual void RefreshSurfaceExtent(EGLSurface surface) { (void)surface; }
void SetWindowHandle(const WindowHandle& handle);
virtual const RendererInfo& GetRendererInfo() const = 0;
@@ -936,6 +936,19 @@ namespace MobileGL::MG_Backend::DirectGLES {
Bool BackendObject_DirectGLES::InitWindowSurface() {
if (RebindNativeSurfaceFor(m_activatingSurface)) return true;
// A suspended surface whose placeholder went: another placeholder - never the window, which
// is gone.
if (std::find(m_suspendedWindowSurfaces.begin(), m_suspendedWindowSurfaces.end(), m_activatingSurface) !=
m_suspendedWindowSurfaces.end()) {
const EGLSurface placeholder = DirectGLES::CreateSessionPbuffer(1, 1);
if (placeholder == EGL_NO_SURFACE || !DirectGLES::BindSessionSurface(placeholder)) {
MGLOG_E("DirectGLES: no placeholder for the suspended window surface %p",
reinterpret_cast<void*>(m_activatingSurface));
return false;
}
m_nativeSurfaces[m_activatingSurface] = placeholder;
return true;
}
// Only use EGL for now
auto nativeWindow = reinterpret_cast<NativeWindowType>(m_windowHandle.Handle);
if (!DirectGLES::InitWindowSurface(nativeWindow)) {
@@ -948,6 +961,14 @@ namespace MobileGL::MG_Backend::DirectGLES {
return true;
}
Bool BackendObject_DirectGLES::LatchIfGpuFaulted() {
#if MOBILEGL_BUILD_DISAGGREGATED
return LatchIfGuiltyBeforeApply();
#else
return false;
#endif
}
void BackendObject_DirectGLES::Initialize() {
MG_Util::BackendLoader::AcquireEGLFunctions(m_EGLFunctions);
MG_Util::BackendLoader::AcquireGLESFunctions(m_GLESFunctions, m_EGLFunctions.eglGetProcAddress);
@@ -1199,6 +1220,41 @@ namespace MobileGL::MG_Backend::DirectGLES {
return true;
}
Bool BackendObject_DirectGLES::BlitDefaultFramebufferToSharedImage(const SharedImageView& image,
const MG_Util::Damage::Region& region) {
#if MOBILEGL_BUILD_DISAGGREGATED
return DirectGLES::SharedImageImpl::BlitDefaultFramebufferTo(image, region);
#else
(void)image, (void)region;
return false;
#endif
}
Int32 BackendObject_DirectGLES::QueryCurrentBufferAge(Bool damageRegionFollows) {
return DirectGLES::CurrentDrawBufferAge(damageRegionFollows);
}
Bool BackendObject_DirectGLES::SetCurrentDamageRegion(const MG_Util::Damage::Region& region) {
return DirectGLES::SetCurrentDrawDamageRegion(region);
}
Bool BackendObject_DirectGLES::PublishSharedImageAccesses() {
#if MOBILEGL_BUILD_DISAGGREGATED
return DirectGLES::SharedImageImpl::PublishPendingAccesses();
#else
return true;
#endif
}
Bool BackendObject_DirectGLES::ExportNativeFence(int* fence) {
*fence = -1;
#if MOBILEGL_BUILD_DISAGGREGATED
return DirectGLES::SharedImageImpl::ExportNativeFence(fence);
#else
return true;
#endif
}
Bool BackendObject_DirectGLES::SwapEGLBuffers(EGLDisplay dpy, EGLSurface draw) {
return BackendObject::SwapEGLBuffers(dpy, draw);
}
@@ -1214,9 +1270,96 @@ namespace MobileGL::MG_Backend::DirectGLES {
#if MOBILEGL_BUILD_DISAGGREGATED
NoteNativeContextGone();
#endif
m_suspendedWindowSurfaces.clear();
BackendObject::ReleaseEGLResources();
}
Bool BackendObject_DirectGLES::SuspendServerWindow(void* window) {
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
if (window == nullptr) return false;
Vector<EGLSurface> onWindow;
for (const auto& [client, state] : m_eglSurfaces) {
if (state.Kind != SurfaceKind::Window || state.Window.Handle != window) continue;
if (std::find(m_suspendedWindowSurfaces.begin(), m_suspendedWindowSurfaces.end(), client) !=
m_suspendedWindowSurfaces.end())
continue;
onWindow.push_back(client);
}
for (const EGLSurface client : onWindow) {
// A surface never activated has no native surface: only its record names the window.
const auto native = m_nativeSurfaces.find(client);
if (native != m_nativeSurfaces.end()) {
const EGLSurface placeholder = DirectGLES::SuspendSessionWindowSurface(native->second);
if (placeholder == EGL_NO_SURFACE) return false;
native->second = placeholder;
}
m_suspendedWindowSurfaces.push_back(client);
}
#if MOBILEGL_BUILD_DISAGGREGATED
// The native binding moved onto the placeholder: the next make-current binds for real.
NoteNativeContextGone();
#endif
MGLOG_I("DirectGLES: %zu window surface(s) on server window %p now draw into placeholder pbuffers; the "
"contexts and their objects stay",
onWindow.size(), window);
return true;
}
Bool BackendObject_DirectGLES::ResumeServerWindow(void* window, Uint32 width, Uint32 height) {
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
if (window == nullptr) return false;
Vector<EGLSurface> suspended;
suspended.swap(m_suspendedWindowSurfaces);
for (const EGLSurface client : suspended) {
const auto state = m_eglSurfaces.find(client);
if (state == m_eglSurfaces.end()) continue; // released while suspended
state->second.Window.Handle = window;
state->second.Window.Width = width;
state->second.Window.Height = height;
state->second.Width = static_cast<EGLint>(std::max<Uint32>(width, 1));
state->second.Height = static_cast<EGLint>(std::max<Uint32>(height, 1));
if (m_eglSurface == client) m_windowHandle = state->second.Window;
// Never activated: its first activation creates it on the new window.
const auto native = m_nativeSurfaces.find(client);
if (native == m_nativeSurfaces.end()) continue;
const EGLSurface surface =
DirectGLES::ResumeSessionWindowSurface(native->second, reinterpret_cast<NativeWindowType>(window));
if (surface == EGL_NO_SURFACE) return false;
native->second = surface;
}
#if MOBILEGL_BUILD_DISAGGREGATED
NoteNativeContextGone();
#endif
MGLOG_I("DirectGLES: %zu suspended window surface(s) rebuilt on server window %p (%ux%u)", suspended.size(),
window, width, height);
return true;
}
Bool BackendObject_DirectGLES::ServerWindowResized(void* window, Uint32 width, Uint32 height) {
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
if (window == nullptr) return false;
Bool published = false;
for (auto& [client, state] : m_eglSurfaces) {
if (state.Kind != SurfaceKind::Window || state.Window.Handle != window) continue;
if (std::find(m_suspendedWindowSurfaces.begin(), m_suspendedWindowSurfaces.end(), client) !=
m_suspendedWindowSurfaces.end())
continue;
state.Window.Width = width;
state.Window.Height = height;
state.Width = static_cast<EGLint>(std::max<Uint32>(width, 1));
state.Height = static_cast<EGLint>(std::max<Uint32>(height, 1));
if (m_eglSurface == client) m_windowHandle = state.Window;
// The window surface's buffers follow the window by themselves (from the next one the
// driver dequeues); only the client has to be told.
const auto native = m_nativeSurfaces.find(client);
if (native == m_nativeSurfaces.end()) continue;
published = DirectGLES::RepublishWindowSurfaceShape(native->second,
reinterpret_cast<NativeWindowType>(window)) ||
published;
}
return published;
}
void BackendObject_DirectGLES::OnEGLSurfaceReleased(EGLSurface surface) {
// P14: releasing a surface releases THAT SURFACE (OnEGLSurfaceForgotten), never the
// contexts. This used to be DestroyEGLContext() - every native context of the session, with
@@ -1229,6 +1372,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
}
void BackendObject_DirectGLES::OnEGLSurfaceForgotten(EGLSurface surface) {
m_suspendedWindowSurfaces.erase(
std::remove(m_suspendedWindowSurfaces.begin(), m_suspendedWindowSurfaces.end(), surface),
m_suspendedWindowSurfaces.end());
const auto it = m_nativeSurfaces.find(surface);
if (it == m_nativeSurfaces.end()) return;
const EGLSurface native = it->second;
@@ -32,6 +32,8 @@ namespace MobileGL::MG_Backend::DirectGLES {
public:
~BackendObject_DirectGLES() override;
// The driver's verdict on this session's context (DirectGLES::LatchIfDeviceLost).
Bool LatchIfGpuFaulted() override;
void Initialize() override;
Bool InitCapabilities() override;
Bool InitWindowSurface() override;
@@ -41,8 +43,27 @@ namespace MobileGL::MG_Backend::DirectGLES {
Bool ResizeEGLWindowSurface(EGLSurface surface, Uint32 width, Uint32 height) override;
Bool MakeEGLCurrent(EGLDisplay dpy, EGLSurface draw, EGLSurface read, EGLContext ctx) override;
Bool SwapEGLBuffers(EGLDisplay dpy, EGLSurface draw) override;
// Shared images: a GL blit of default framebuffer 0 into a renderbuffer whose storage is
// the image's EGLImage, ordered after the image's pending reads and published as a native
// fence without a CPU wait (DirectGLES.h, SharedImageImpl).
Bool BlitDefaultFramebufferToSharedImage(const SharedImageView& image,
const MG_Util::Damage::Region& region) override;
// A window surface's age is the driver's (EGL_EXT_buffer_age); a pbuffer keeps its content,
// so it is 1 once it has been presented (DirectGLES::CurrentDrawBufferAge).
Int32 QueryCurrentBufferAge(Bool damageRegionFollows) override;
// eglSetDamageRegionKHR on a window surface whose driver has EGL_KHR_partial_update.
Bool SetCurrentDamageRegion(const MG_Util::Damage::Region& region) override;
// A glFlush that ends a shared-image producer's frame (SharedImageImpl::PublishPendingAccesses).
Bool PublishSharedImageAccesses() override;
// An EGL native fence of the session's work (SharedImageImpl::ExportNativeFence).
Bool ExportNativeFence(int* fence) override;
void ReleaseEGLSurface(EGLSurface surface) override;
void ReleaseEGLResources() override;
// The server window going and coming back: each window surface on it swaps its native EGL
// window surface for a placeholder pbuffer and back (DirectGLES::Suspend/ResumeSessionWindowSurface).
Bool SuspendServerWindow(void* window) override;
Bool ResumeServerWindow(void* window, Uint32 width, Uint32 height) override;
Bool ServerWindowResized(void* window, Uint32 width, Uint32 height) override;
const RendererInfo& GetRendererInfo() const override;
String GetBackendAPIVersionString() const override;
@@ -71,6 +92,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
// on every later one. A multi-window client switches surfaces every frame; making a new
// native surface each time leaked them and lost what the window had drawn.
UnorderedMap<EGLSurface, EGLSurface> m_nativeSurfaces;
// Window surfaces whose window is gone (SuspendServerWindow): their native surface is a
// placeholder pbuffer until ResumeServerWindow.
Vector<EGLSurface> m_suspendedWindowSurfaces;
MG_External::EGLFunctionsTable m_EGLFunctions;
MG_External::GLESFunctionsTable m_GLESFunctions;
MG_External::GLESCapabilities m_GLESCapabilities;
File diff suppressed because it is too large Load Diff
+203
View File
@@ -116,6 +116,16 @@ namespace MobileGL::MG_Backend::DirectGLES {
EGLSurface CurrentSessionDrawSurface();
// The surface the calling thread's ACTIVE context last drew to (EGL_NO_SURFACE for none).
EGLSurface CurrentContextDrawSurface();
// EGL_BUFFER_AGE_EXT of the session's current draw surface for the next frame: the driver's
// answer for a window surface, 1 for a pbuffer that has been presented (a pbuffer's content stays
// across swaps), 0 otherwise. A driver whose buffer age is only EGL_KHR_partial_update's keeps
// nothing inside the frame's damage region - the whole surface unless one is set - so it is asked
// only when `damageRegionFollows` (SetCurrentDrawDamageRegion before the first draw).
Int32 CurrentDrawBufferAge(Bool damageRegionFollows);
// eglSetDamageRegionKHR(region) on the session's current window surface (a pbuffer, or a driver
// without EGL_KHR_partial_update, ignores it). The driver wants the age asked first in the frame;
// when the client was answered without asking it, it is asked here.
Bool SetCurrentDrawDamageRegion(const MG_Util::Damage::Region& region);
// Makes an EXISTING native surface of the calling thread's session the one it draws to, and
// binds it to the active context. False for a surface the session does not own.
Bool BindSessionSurface(EGLSurface surface);
@@ -130,11 +140,43 @@ namespace MobileGL::MG_Backend::DirectGLES {
// Every context of the session that drew to `previous` draws to `replacement` instead, and
// `previous` is destroyed (a pbuffer-backed window's resize).
void ReplaceSessionSurface(EGLSurface previous, EGLSurface replacement);
// The server window going away (screen off): every context of the session that drew to the
// window surface `windowSurface` draws to a new 1x1 placeholder pbuffer instead, and the window
// surface is destroyed - no longer current anywhere once this returns. The contexts and their
// objects stay. Returns the placeholder, or EGL_NO_SURFACE (nothing changed) when none could be
// made.
EGLSurface SuspendSessionWindowSurface(EGLSurface windowSurface);
// The way back: a window surface on `window` replaces `placeholder` (EGL_NO_SURFACE: none to
// replace) for every context that drew to it, is bound, and its extent is published as a
// creation publishes it. Returns the new surface, or EGL_NO_SURFACE.
EGLSurface ResumeSessionWindowSurface(EGLSurface placeholder, NativeWindowType window);
// The window under `surface` changed size: its extent (the display's) is published again, at once.
Bool RepublishWindowSurfaceShape(EGLSurface surface, NativeWindowType window);
Bool MakeCurrent();
Bool ReleaseCurrent();
// True when the backend ES context is current on the calling thread, i.e.
// immediate buffer ops may issue GL calls right now.
Bool IsBackendContextCurrentOnThisThread();
#if MOBILEGL_BUILD_DISAGGREGATED
// DEVICE LOSS, Espryt's half (Magma's is VulkanRenderer::LatchWireDeviceLoss). Asks the
// driver whether the calling thread's context was reset (glGetGraphicsResetStatus); a reset -
// or the debug knob MGPipeDebugDeviceLossDue - latches THIS session through
// MGPipeSessionLatch and answers true, and the caller returns from its verb without issuing
// more GL. Asked at the frame boundaries (Present, a shared-image present), before a readback,
// and by the apply loop each time it wakes (LatchIfGpuFaulted), never per draw. The contexts ask
// for reset notification (NativeContextsNotifyResets) so the driver can name the GUILTY one;
// only GUILTY (or a reset of unknown cause) ends the session - an INNOCENT report, a context
// whose work merely waited behind another's hang (the compositor's among them), is only noted.
// `debugKnob` false: the knob is not consulted here (the on-screen Present, so an injected loss
// lands on an offscreen client's shared-image present or readback, never on the compositor's).
Bool LatchIfDeviceLost(const char* site, Bool debugKnob = true);
// The apply loop's form of it (BackendObject::LatchIfGpuFaulted), asked each time the apply thread
// wakes: the driver's report only, no debug knob, and no EGL ground-truth verification spent.
Bool LatchIfGuiltyBeforeApply();
#endif
// True once the process's native contexts were created with EGL_LOSE_CONTEXT_ON_RESET (decided
// by the first one; the driver must offer EGL_EXT_create_context_robustness).
Bool NativeContextsNotifyResets();
// GL fence sync objects, backed by native ES fences. FenceSync returns null
// (the frontend then falls back to an always-signaled sync) when the calling
// thread does not own the ES context. Waits/queries degrade to "signaled" in
@@ -293,6 +335,98 @@ namespace MobileGL::MG_Backend::DirectGLES {
// has gone.
void DestroyNativeContextFor(Uint64 contextToken);
// ---- objects that belong to ONE native context ----
//
// Framebuffers, vertex arrays, transform feedbacks and queries are container objects: GL never
// shares them, not even between contexts of one share group, and every context numbers its own
// from 1. A process-wide scratch FBO id minted in one context and bound in another therefore
// names THAT context's own framebuffer - often an application's - and the scratch path's
// attach/scrub then rewires it onto staging textures. With one process serving several
// sessions (each with its own native contexts) that silently corrupted other clients' frames.
//
// The calling thread's current native context as a lifetime serial: assigned when the context
// is created and never reused (a recycled EGLContext address is a new serial). 0 when none was
// made current through MakeCurrent.
Uint64 CurrentNativeContextSerial();
class PerNativeContextBase {
public:
PerNativeContextBase();
PerNativeContextBase(const PerNativeContextBase&) = delete;
PerNativeContextBase& operator=(const PerNativeContextBase&) = delete;
// The context died, and its names with it: forget them (no GL call - there is no context
// left to make one in).
virtual void ForgetContext(Uint64 serial) = 0;
protected:
~PerNativeContextBase();
};
// Called where a native context is destroyed.
void ForgetNativeContextObjects(Uint64 serial);
// Moves whenever any per-context entry is dropped; a thread's cached entry is valid while
// it has not moved.
Uint64 PerNativeContextEpoch();
void BumpPerNativeContextEpoch();
// One T per native context, created on first use from that context.
template <typename T>
class PerNativeContext final : public PerNativeContextBase {
public:
T& Current() {
const Uint64 serial = CurrentNativeContextSerial();
// The draw path asks for the same entry over and over from one thread: a one-entry
// thread-local cache, invalidated by any drop anywhere (rare).
thread_local CurrentCache cache;
const Uint64 epoch = PerNativeContextEpoch();
if (cache.Owner == this && cache.Serial == serial && cache.Epoch == epoch) return *cache.Entry;
const std::lock_guard<std::mutex> lock(m_mutex);
T& entry = m_bySerial[serial]; // node-based: the reference survives other contexts' inserts
cache = {this, serial, epoch, &entry};
return entry;
}
template <typename F>
void ForEach(F&& visit) {
const std::lock_guard<std::mutex> lock(m_mutex);
for (auto& entry : m_bySerial) visit(entry.second);
}
void Clear() {
const std::lock_guard<std::mutex> lock(m_mutex);
BumpPerNativeContextEpoch();
m_bySerial.clear();
}
void ForgetContext(Uint64 serial) override {
const std::lock_guard<std::mutex> lock(m_mutex);
BumpPerNativeContextEpoch();
m_bySerial.erase(serial);
}
private:
struct CurrentCache {
const PerNativeContext* Owner = nullptr;
Uint64 Serial = 0;
Uint64 Epoch = 0;
T* Entry = nullptr;
};
std::mutex m_mutex;
std::unordered_map<Uint64, T> m_bySerial;
};
// The GL name of one per-context object, spelled like the plain `GLuint` it replaces: it reads
// as the CURRENT context's name, `= 0` resets that one, and `&name` is where glGen* writes it.
class NativeContextName {
public:
operator GLuint() const { return m_names.Current(); }
NativeContextName& operator=(GLuint name) {
m_names.Current() = name;
return *this;
}
GLuint* operator&() { return &m_names.Current(); }
private:
mutable PerNativeContext<GLuint> m_names;
};
// P14 S4. THE NATIVE HALF OF A CONTEXT SWITCH. `bind_context` moves which context the session's
// records belong to; this makes the calling thread's native tuple follow it, binding the
// session's current surface to the newly named context's native EGLContext. A no-op when the
@@ -368,8 +502,77 @@ namespace MobileGL::MG_Backend::DirectGLES {
// answers from the frontend state, which this function only reads.
void SyncRenderState(Bool forColorClear = false);
void InvalidateSyncedRenderState();
// GL_SCISSOR_TEST as the driver has it in the native context current on this thread -
// what a guard that turns the test off around its own blit, and puts it back, needs to
// know. The shadow answers only while it describes this context (synced since the last
// invalidation, in this context epoch); otherwise the driver is asked. The shadow is one
// per process: once another session's context synced into it, or a context switch
// invalidated it, its bits are some other context's.
Bool DriverScissorTestEnabled();
// TESTING: the shadow as SyncRenderState leaves it in the current context epoch, with
// GL_SCISSOR_TEST `enabled`.
void SetSyncedScissorTestForTesting(Bool enabled);
// TESTING: `inside` runs within the guard the shared-image present and the readbacks
// copy under (ScopedScissorDisable).
void RunScissorDisabledForTesting(void (*inside)(void*), void* data);
} // namespace RenderStateImpl
#if MOBILEGL_BUILD_DISAGGREGATED
// ---- SHARED IMAGES (docs/Disaggregated/notes/anland/plan-ahb-dmabuf.md) ----
//
// A shared image reaches the driver only through the Android import path: its AHardwareBuffer
// becomes an EGLImage (EGL_ANDROID_get_native_client_buffer + EGL_ANDROID_image_native_buffer)
// and that EGLImage the storage of a texture or renderbuffer (GL_OES_EGL_image). On a host
// build there is no such path and every entry below declines.
namespace SharedImageImpl {
// One EGLImage over one shared image, holding the image's registry reference for as long
// as the EGLImage lives (SharedImageRegistry.h's lifetime rule). The last reference
// destroys the EGLImage, unless the display that made it has been terminated since (which
// destroyed it already). Opaque outside DirectGLES.cpp.
struct EglImage;
using EglImageRef = SharedPtr<EglImage>;
// An EGLImage for the live shared image `id`, or null with `why`. Needs a current context
// (the first call resolves the extension entry points).
EglImageRef CreateEglImage(Uint64 id, String& why);
// glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, image) on what the active unit has bound to
// GL_TEXTURE_2D. False when the driver refused it (its GL error is consumed).
Bool TargetBoundTexture2D(const EglImageRef& image);
// BackendObject::BlitDefaultFramebufferToSharedImage, for the calling thread's context.
Bool BlitDefaultFramebufferTo(const SharedImageView& image, const MG_Util::Damage::Region& region);
// THE READER'S HALF of the image's sync state (SharedImageRegistry.h). Called on every use
// of a name whose level 0 is `image`, before the command that samples it: when a write
// landed since this context last waited, the context's later commands wait for its fence
// on the GPU; and the image is noted as read by the calling session's current frame.
void AcquireForSampling(const EglImageRef& image);
// YUV IMAGES (NV12/P010, MG_Remote/Server/SharedImageYuv.h) are never a name's level 0: the
// name keeps its own RGBA8 storage `target` (width x height), into which the image is
// converted at its first use in each of the calling session's frames - the image bound as
// an external texture, one triangle drawn - after the registry's CPU copy for a foreign
// buffer. Every touched binding is put back. Called after AcquireForSampling.
Bool IsYuv(const EglImageRef& image);
void ConvertYuvForSampling(const EglImageRef& image, GLuint target, Uint32 width, Uint32 height);
// The calling session's frame boundary for what it has read: hands a fence for everything
// the current context has submitted to every image noted since the last boundary. Called
// at Present and wherever the current native context stops being current (its reads can
// only be fenced from it). Nothing noted = nothing done.
void PublishPendingReads();
// IMPLICIT SYNC BY FLUSH (BackendObject::PublishSharedImageAccesses): everything the
// calling session used since its last boundary is published as written and read by one
// fence, and the session's first use of an image in each frame from now on also waits for
// the image's pending reads.
Bool PublishPendingAccesses();
// EGL_ANDROID_native_fence_sync's fence command (BackendObject::ExportNativeFence): a
// sync_file of everything the current context submitted in `*fence` (-1: none, and the work
// is waited out here), and - for a session that writes images - the boundary
// PublishPendingAccesses makes.
Bool ExportNativeFence(int* fence);
} // namespace SharedImageImpl
#endif
extern MG_External::EGLFunctionsTable g_EGLFuncs;
extern MG_External::GLESFunctionsTable g_GLESFuncs;
extern MG_External::GLESCapabilities g_GLESCapabilities;
+343 -72
View File
@@ -206,6 +206,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
#else
Bool InProcessTeardown() { return g_processTeardown; }
#endif
Bool InProcessExit() { return g_processTeardown; }
void EnsureProcessTeardownSentinel() {
std::call_once(g_teardownSentinelOnce,
[] { std::atexit(+[] { g_processTeardown = true; }); });
@@ -1083,14 +1084,14 @@ namespace MobileGL::MG_Backend::DirectGLES {
const char* label = "";
};
PersistentRing g_uboRing{{}, {}, {}, kUboRingInitialBytes, kUboRingMaxBytes, 0, "Global-UBO ring"};
PersistentRing g_unpackRing{{},
{},
{},
kUnpackRingInitialBytes,
kUnpackRingMaxBytes,
kUnpackRingAlignment,
"Texture unpack ring"};
// THE RINGS ARE PER NATIVE CONTEXT. Their slots are reclaimed by frame marks recorded at
// Present, which say "everything this context wrote before its fence N is done once N is".
// One process-wide set used to be shared by every session's contexts: a fence from one
// context retired bytes another context's not-yet-executed copies still had to read
// (a glTexSubImage from the unpack ring is a deferred GPU read), and the next session's
// staging overwrote them - its pixels landed, at the wrong offsets, in the first one's
// textures. With one set per context, a context's bytes wait for that context's own
// fence. (The buffers themselves are shared objects; only who stages into them is split.)
// Staging ring for app buffer updates whose destination store may still be
// referenced by in-flight GPU work. Mali's glBufferSubData resolves that WAR
// hazard by BLOCKING in the call (osup_sync_object_wait) until every
@@ -1099,13 +1100,67 @@ namespace MobileGL::MG_Backend::DirectGLES {
// chunks stream in). Staging the bytes here and issuing a
// glCopyBufferSubData instead keeps the hazard on the GPU timeline where it
// is just job ordering, and the CPU never waits.
PersistentRing g_uploadRing{{},
{},
{},
kUploadRingInitialBytes,
kUploadRingMaxBytes,
kUploadRingAlignment,
"Buffer upload ring"};
struct RingSet {
PersistentRing ubo{{}, {}, {}, kUboRingInitialBytes, kUboRingMaxBytes, 0, "Global-UBO ring"};
PersistentRing unpack{{},
{},
{},
kUnpackRingInitialBytes,
kUnpackRingMaxBytes,
kUnpackRingAlignment,
"Texture unpack ring"};
PersistentRing upload{{},
{},
{},
kUploadRingInitialBytes,
kUploadRingMaxBytes,
kUploadRingAlignment,
"Buffer upload ring"};
RingSet();
~RingSet();
RingSet(const RingSet&) = delete;
RingSet& operator=(const RingSet&) = delete;
};
// A dead context's ring buffers: shared objects, so they outlive it; deleted at the next
// Present of any context of the same share-group generation.
struct OrphanRingBuffer {
Uint id = 0;
Uint contextGeneration = 0;
};
std::mutex g_orphanRingBuffersMutex;
Vector<OrphanRingBuffer>& OrphanRingBuffers() {
static auto* orphans = new Vector<OrphanRingBuffer>();
return *orphans;
}
// Each set's generations start in a range of their own: a per-program slot cache keyed
// by (generation, offset) must never take one context's ring for another's.
std::atomic<Uint32> g_ringGenerationBases{0};
RingSet::RingSet() {
const Uint32 base = (g_ringGenerationBases.fetch_add(1, std::memory_order_relaxed) + 1) << 12;
ubo.store.generation = base;
unpack.store.generation = base;
upload.store.generation = base;
}
RingSet::~RingSet() {
const std::lock_guard<std::mutex> lock(g_orphanRingBuffersMutex);
for (const PersistentRing* ring : {&ubo, &unpack, &upload}) {
if (ring->store.id != 0) {
OrphanRingBuffers().push_back({ring->store.id, ring->store.contextGeneration});
}
for (const auto& retired : ring->retired) {
if (retired.id != 0) OrphanRingBuffers().push_back({retired.id, retired.contextGeneration});
}
}
}
DirectGLES::PerNativeContext<RingSet> g_ringSets;
PersistentRing& UboRing() { return g_ringSets.Current().ubo; }
PersistentRing& UnpackRing() { return g_ringSets.Current().unpack; }
PersistentRing& UploadRing() { return g_ringSets.Current().upload; }
// The ES context the ring's id/map belonged to is gone (or was never
// seen): drop every handle without GL calls and re-arm creation. The
@@ -1383,7 +1438,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
Bool UploadRingUsableNow() {
if (MG_Config::Features.EsprytDisableUploadRing) return false;
if (!g_GLESFuncs.glCopyBufferSubData) return false;
return RingAvailable(g_uploadRing);
return RingAvailable(UploadRing());
}
// A partial range below this goes through the staging ring instead of a
@@ -1532,9 +1587,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
}
SizeT ringOffset = 0;
if (ringUsable && size <= kUploadRingMaxBytes &&
RingAllocate(g_uploadRing, size, ringOffset)) {
Memcpy(g_uploadRing.store.mappedPtr + ringOffset, hostBase + start, size);
BindBufferId(GL_COPY_READ_BUFFER, g_uploadRing.store.id);
RingAllocate(UploadRing(), size, ringOffset)) {
Memcpy(UploadRing().store.mappedPtr + ringOffset, hostBase + start, size);
BindBufferId(GL_COPY_READ_BUFFER, UploadRing().store.id);
BindBufferId(GL_COPY_WRITE_BUFFER, resource.id);
g_GLESFuncs.glCopyBufferSubData(GL_COPY_READ_BUFFER, GL_COPY_WRITE_BUFFER,
(GLintptr)ringOffset, (GLintptr)start, (GLsizeiptr)size);
@@ -1611,9 +1666,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
}
SizeT ringOffset = 0;
if (ringUsable && size <= kUploadRingMaxBytes &&
RingAllocate(g_uploadRing, size, ringOffset)) {
Memcpy(g_uploadRing.store.mappedPtr + ringOffset, write.bytes.data(), size);
BindBufferId(GL_COPY_READ_BUFFER, g_uploadRing.store.id);
RingAllocate(UploadRing(), size, ringOffset)) {
Memcpy(UploadRing().store.mappedPtr + ringOffset, write.bytes.data(), size);
BindBufferId(GL_COPY_READ_BUFFER, UploadRing().store.id);
BindBufferId(GL_COPY_WRITE_BUFFER, resource.id);
g_GLESFuncs.glCopyBufferSubData(GL_COPY_READ_BUFFER, GL_COPY_WRITE_BUFFER,
(GLintptr)ringOffset, (GLintptr)write.offset,
@@ -1714,7 +1769,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
Bool UploadRingUsableNow() {
if (MG_Config::Features.EsprytDisableUploadRing) return false;
if (!g_GLESFuncs.glCopyBufferSubData) return false;
return RingAvailable(g_uploadRing);
return RingAvailable(UploadRing());
}
// A partial range below this goes through the staging ring instead of a
@@ -1820,9 +1875,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
}
SizeT ringOffset = 0;
if (ringUsable && size <= kUploadRingMaxBytes &&
RingAllocate(g_uploadRing, size, ringOffset)) {
Memcpy(g_uploadRing.store.mappedPtr + ringOffset, bufferObject.MappedData() + start, size);
BindBufferId(GL_COPY_READ_BUFFER, g_uploadRing.store.id);
RingAllocate(UploadRing(), size, ringOffset)) {
Memcpy(UploadRing().store.mappedPtr + ringOffset, bufferObject.MappedData() + start, size);
BindBufferId(GL_COPY_READ_BUFFER, UploadRing().store.id);
BindBufferId(GL_COPY_WRITE_BUFFER, resource.id);
g_GLESFuncs.glCopyBufferSubData(GL_COPY_READ_BUFFER, GL_COPY_WRITE_BUFFER,
(GLintptr)ringOffset, (GLintptr)start, (GLsizeiptr)size);
@@ -1866,9 +1921,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
}
SizeT ringOffset = 0;
if (ringUsable && size <= kUploadRingMaxBytes &&
RingAllocate(g_uploadRing, size, ringOffset)) {
Memcpy(g_uploadRing.store.mappedPtr + ringOffset, write.bytes.data(), size);
BindBufferId(GL_COPY_READ_BUFFER, g_uploadRing.store.id);
RingAllocate(UploadRing(), size, ringOffset)) {
Memcpy(UploadRing().store.mappedPtr + ringOffset, write.bytes.data(), size);
BindBufferId(GL_COPY_READ_BUFFER, UploadRing().store.id);
BindBufferId(GL_COPY_WRITE_BUFFER, resource.id);
g_GLESFuncs.glCopyBufferSubData(GL_COPY_READ_BUFFER, GL_COPY_WRITE_BUFFER,
(GLintptr)ringOffset, (GLintptr)write.offset,
@@ -2843,16 +2898,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
kT0MapReadBit | kT0MapWriteBit | kMapPersistentBit | kMapCoherentBit | kDynamicStorageBit;
constexpr GLbitfield kT0MapFlags = kT0MapReadBit | kT0MapWriteBit | kMapPersistentBit | kMapCoherentBit;
Bool HasExtensionWord(const char* list, const char* word) {
if (list == nullptr) return false;
const SizeT n = std::strlen(word);
for (const char* at = std::strstr(list, word); at != nullptr; at = std::strstr(at + 1, word)) {
const Bool startOk = at == list || at[-1] == ' ';
const Bool endOk = at[n] == '\0' || at[n] == ' ';
if (startOk && endOk) return true;
}
return false;
}
using MG_Util::BackendLoader::HasExtensionWord;
Bool ResolveT0EntryPoints(String& why) {
if (g_t0GetNativeClientBuffer != nullptr && g_t0BufferStorageExternal != nullptr) return true;
@@ -3848,8 +3894,13 @@ namespace MobileGL::MG_Backend::DirectGLES {
MG_Pipe::MGPipeSetResourceOps(&g_glesResourceOps);
#endif
// Frontend writes issued while ops were unregistered advanced change
// serials with no per-op bump; re-open every draw-clean memo.
BumpBufferMutationEpoch();
// serials with no per-op bump; re-open every draw-clean memo. Advanced directly,
// past BumpBufferMutationEpoch's apply-thread pin: a registration mutates no buffer
// and only makes every memo probe again, and with several sessions in one process
// a new session's backend registers on its own thread while another session's
// apply thread is already running - which the pin would otherwise read as a
// client-thread producer and, under the strict lane, abort the server for.
g_bufferMutationEpoch.fetch_add(1, std::memory_order_release);
}
void UnregisterBufferBackendOps() {
@@ -3895,8 +3946,8 @@ namespace MobileGL::MG_Backend::DirectGLES {
InvalidatePixelBufferBindingCaches();
// The rings' ids and persistent maps died with the context; drop the
// handles (no GL) and let the next draw / texture upload recreate them.
ResetRingForNewContext(g_uboRing);
ResetRingForNewContext(g_unpackRing);
ResetRingForNewContext(UboRing());
ResetRingForNewContext(UnpackRing());
#if MOBILEGL_PIPE_PUSH
// R-11's server copies die with the context for the same reason the rings' ids do:
// the resource twins they are keyed by are about to be rebuilt against a new
@@ -4905,9 +4956,23 @@ namespace MobileGL::MG_Backend::DirectGLES {
return true;
}
void DeleteOrphanRingBuffers() {
const std::lock_guard<std::mutex> lock(g_orphanRingBuffersMutex);
auto& orphans = OrphanRingBuffers();
if (orphans.empty()) return;
for (const OrphanRingBuffer& orphan : orphans) {
// A buffer of a share group that is gone was reclaimed with it.
if (orphan.contextGeneration != g_bufferContextGeneration) continue;
ScrubBufferBindingShadowsForId(orphan.id);
g_GLESFuncs.glDeleteBuffers(1, &orphan.id);
}
orphans.clear();
}
// Present()-time upkeep shared by both rings.
void RingOnPresent(PersistentRing& ring) {
if (!CanTouchGLNow()) return;
DeleteOrphanRingBuffers();
// Delete grown-away stores the GPU is provably done with.
const Uint64 completed = DirectGLES::CompletedFrameSerial();
@@ -4952,36 +5017,36 @@ namespace MobileGL::MG_Backend::DirectGLES {
Bool UboRingAvailable() {
if (MG_Config::Features.EsprytDisableUboRing) return false;
return RingAvailable(g_uboRing);
return RingAvailable(UboRing());
}
Bool UboRingAllocate(SizeT size, SizeT& outOffset) { return RingAllocate(g_uboRing, size, outOffset); }
Bool UboRingAllocate(SizeT size, SizeT& outOffset) { return RingAllocate(UboRing(), size, outOffset); }
void* UboRingMappedPtr() { return g_uboRing.store.mappedPtr; }
Uint UboRingBufferId() { return g_uboRing.store.id; }
Uint32 UboRingGeneration() { return g_uboRing.store.generation; }
void* UboRingMappedPtr() { return UboRing().store.mappedPtr; }
Uint UboRingBufferId() { return UboRing().store.id; }
Uint32 UboRingGeneration() { return UboRing().store.generation; }
void UboRingOnPresent() { RingOnPresent(g_uboRing); }
void UboRingOnPresent() { RingOnPresent(UboRing()); }
Bool UnpackRingAvailable() {
if (MG_Config::Features.EsprytDisableUnpackRing) return false;
return RingAvailable(g_unpackRing);
return RingAvailable(UnpackRing());
}
Bool UnpackRingAllocate(SizeT size, SizeT& outOffset) {
// A request the ring could never satisfy even empty would otherwise walk
// the whole grow/drain ladder before failing.
if (size > kUnpackRingMaxBytes) return false;
return RingAllocate(g_unpackRing, size, outOffset);
return RingAllocate(UnpackRing(), size, outOffset);
}
void* UnpackRingMappedPtr() { return g_unpackRing.store.mappedPtr; }
Uint UnpackRingBufferId() { return g_unpackRing.store.id; }
void* UnpackRingMappedPtr() { return UnpackRing().store.mappedPtr; }
Uint UnpackRingBufferId() { return UnpackRing().store.id; }
SizeT UnpackRingMaxBytes() { return kUnpackRingMaxBytes; }
void UnpackRingOnPresent() { RingOnPresent(g_unpackRing); }
void UnpackRingOnPresent() { RingOnPresent(UnpackRing()); }
void UploadRingOnPresent() { RingOnPresent(g_uploadRing); }
void UploadRingOnPresent() { RingOnPresent(UploadRing()); }
} // namespace BufferImpl
#if MOBILEGL_PIPE_PUSH
@@ -7519,6 +7584,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
}
}
#if MOBILEGL_BUILD_DISAGGREGATED
// A shared image's EGLImage lives exactly as long as the name it was bound to.
m_sharedImage = nullptr;
#endif
g_GLESFuncs.glGenTextures(1, &m_backendTextureId);
m_contextGeneration = g_backendContextGeneration;
if (m_backendTextureId == 0) {
@@ -8769,6 +8838,167 @@ namespace MobileGL::MG_Backend::DirectGLES {
#define MGB_STORAGE_KIND(obj) ((obj)->GetStorageType())
#endif
#if MOBILEGL_BUILD_DISAGGREGATED
// SHARED IMAGES (docs/Disaggregated/notes/anland/plan-ahb-dmabuf.md). The client defined
// level 0 as WxH RGBA8 with no data and then attached the image; the record's descriptor
// still says so, but the texels are the image's, so nothing below may allocate or upload.
Bool BackendTextureObject::SyncSharedImageStorage(const MG_Pipe::MGPipeResourceRecord& record) {
const Uint64 imageId = record.SharedImageId;
if (imageId == 0) {
// Respecified away from the image. The ES name is an EGLImage sibling, which must not
// be redefined in place: the immutable flag sends the normal path through
// RecreateBackendTexture, and the EGLImage goes with the name it was bound to.
m_sharedImageId = 0;
m_isInitialized = false;
m_backendStorageImmutable = true;
return false;
}
// Steady state: bound to this image already (the serial moves with sub-data and with a
// metadata respecify, neither of which changes what the name samples). A failed bind is
// retried only once the record moves. Pending sub-data is the caller's
// (ReplaySharedImageUploads), after the image's sync state has been acquired.
if (m_sharedImageId == imageId && m_isInitialized &&
(m_sharedImage != nullptr || m_syncedResourceSerial == record.Serial)) {
m_syncedResourceSerial = record.Serial;
return true;
}
// A fresh mutable name: the current one may hold immutable storage (which no EGLImage
// can replace) or another image's sibling. RecreateBackendTexture also moves the FBO
// twins' attachment generation and forces the parameter resync the new name needs.
RecreateBackendTexture();
m_sharedImage = nullptr;
m_sharedImageId = imageId;
m_isInitialized = true;
// Whatever comes next redefines storage through a re-mint (see the imageId == 0 arm).
m_backendStorageImmutable = true;
m_syncedResourceSerial = record.Serial;
const TextureTarget target = BufferImpl::StagedTextureTargetForPipeTarget(record.Desc.Target);
if (ConvertTextureTargetToBackendGLEnum(target) != GL_TEXTURE_2D) {
MGLOG_E_ONCE("Shared image texture %u: target %s cannot take an EGLImage; left empty",
record.Desc.GlNameForDiag, MG_Util::ConvertTextureTargetToString(target).c_str());
return true;
}
String why;
SharedImageImpl::EglImageRef image = SharedImageImpl::CreateEglImage(imageId, why);
if (image == nullptr) {
MGLOG_E_ONCE("Shared image texture %u: image %llu cannot be bound (%s); left empty",
record.Desc.GlNameForDiag, static_cast<unsigned long long>(imageId), why.c_str());
return true;
}
Bind(GL_TEXTURE_2D);
if (SharedImageImpl::IsYuv(image)) {
// A YUV image is CONVERTED into the name's own RGBA8 level at each frame's first use
// (SharedImageImpl::ConvertYuvForSampling), never made its storage.
while (g_GLESFuncs.glGetError() != GL_NO_ERROR) {
}
g_GLESFuncs.glTexStorage2D(GL_TEXTURE_2D, 1, GL_RGBA8, static_cast<GLsizei>(record.Desc.Width),
static_cast<GLsizei>(record.Desc.Height));
if (g_GLESFuncs.glGetError() != GL_NO_ERROR) {
MGLOG_E_ONCE("Shared image texture %u: no %ux%u RGBA8 storage for YUV image %llu; left empty",
record.Desc.GlNameForDiag, record.Desc.Width, record.Desc.Height,
static_cast<unsigned long long>(imageId));
return true;
}
} else if (!SharedImageImpl::TargetBoundTexture2D(image)) {
MGLOG_E_ONCE("Shared image texture %u: glEGLImageTargetTexture2DOES refused image %llu; left empty",
record.Desc.GlNameForDiag, static_cast<unsigned long long>(imageId));
return true;
}
m_sharedImage = std::move(image);
m_sharedImageUploadsRgb = false;
// The shape the name now has: the descriptor's level 0, and that one level only.
m_prevTextureInfo = {static_cast<TextureInternalFormat>(record.Desc.InternalFormat),
record.Desc.Width,
record.Desc.Height,
record.Desc.Depth,
1,
0,
record.Desc.Samples,
record.Desc.FixedSampleLocations != 0};
return true;
}
// SUB-DATA ON A SHARED IMAGE IS WRITTEN INTO IT. An X server drawing with glamor uploads
// its clients' pixels (MIT-SHM / PutImage) into window pixmaps whose level 0 is a shared
// image, with glTexSubImage2D: the texels are in the staged store (the client defined the
// level without data, so only the uploaded boxes are covered), and they go into the
// EGLImage-backed name box by box. Every pending entry leaves the set either way, or the
// per-draw clean gate would never close. Called after AcquireForSampling, so the upload is
// ordered after the image's pending reads and writes like any other use.
void BackendTextureObject::ReplaySharedImageUploads(const MG_Pipe::MGPipeResourceRecord& record) {
if (record.PendingUploads.empty()) return;
Vector<std::pair<Uint16, Uint16>> keys;
Vector<std::pair<Uint16, MG_Pipe::MGPBox>> boxes;
keys.reserve(record.PendingUploads.size());
for (const auto& pending : record.PendingUploads) {
const Uint16 uploadHalf = MG_Pipe::MGPipeSubDataUploadTargetOf(pending.UploadTarget);
keys.emplace_back(uploadHalf, pending.Level);
if (pending.Level != 0) continue; // an image has one level
if (pending.Regions.empty()) {
boxes.emplace_back(uploadHalf, pending.UnionBox);
} else {
for (const auto& region : pending.Regions)
boxes.emplace_back(uploadHalf, MG_Pipe::MGPBox{region.X, region.Y, region.Z, region.W, region.H,
region.D});
}
}
if (m_sharedImage != nullptr && !boxes.empty()) {
auto& store = MG_Remote::Server::ServerStagedTexture();
const Uint64 key = MG_Remote::Server::StagedTextureStore::KeyForHandle(record.Desc.Resource);
Bind(GL_TEXTURE_2D);
ScopedDefaultUnpackState unpackState;
BufferImpl::BindPixelUnpackBufferId(0);
for (const auto& [uploadHalf, box] : boxes) {
Int rowTexels = 0;
// A shared image is RGBA8 to every client that names it (the level the client
// defined over it): four bytes a texel.
const Uint8* bytes =
store.CoveredBoxBytes(key, uploadHalf, 0, box.X, box.Y, box.W, box.H, 4, &rowTexels);
if (bytes == nullptr) {
MGLOG_E_ONCE("Shared image texture %u: an upload of %ux%u at (%d,%d) has no staged texels; "
"not written",
record.Desc.GlNameForDiag, box.W, box.H, box.X, box.Y);
continue;
}
const Uint8* origin =
bytes + (static_cast<SizeT>(box.Y) * static_cast<SizeT>(rowTexels) + static_cast<SizeT>(box.X)) * 4;
if (!m_sharedImageUploadsRgb) {
while (g_GLESFuncs.glGetError() != GL_NO_ERROR) {
}
g_GLESFuncs.glPixelStorei(GL_UNPACK_ROW_LENGTH, rowTexels);
g_GLESFuncs.glTexSubImage2D(GL_TEXTURE_2D, 0, box.X, box.Y, static_cast<GLsizei>(box.W),
static_cast<GLsizei>(box.H), GL_RGBA, GL_UNSIGNED_BYTE, origin);
g_GLESFuncs.glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
// An image of an X format (no alpha) is an RGB8 texture to the driver, which
// takes only RGB texels: this image's uploads are converted from now on.
if (g_GLESFuncs.glGetError() != GL_INVALID_OPERATION) continue;
m_sharedImageUploadsRgb = true;
}
Vector<Uint8> rgb(static_cast<SizeT>(box.W) * box.H * 3);
for (Uint32 row = 0; row < box.H; ++row) {
const Uint8* src = origin + static_cast<SizeT>(row) * static_cast<SizeT>(rowTexels) * 4;
Uint8* dst = rgb.data() + static_cast<SizeT>(row) * box.W * 3;
for (Uint32 x = 0; x < box.W; ++x) {
dst[x * 3 + 0] = src[x * 4 + 0];
dst[x * 3 + 1] = src[x * 4 + 1];
dst[x * 3 + 2] = src[x * 4 + 2];
}
}
g_GLESFuncs.glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
g_GLESFuncs.glTexSubImage2D(GL_TEXTURE_2D, 0, box.X, box.Y, static_cast<GLsizei>(box.W),
static_cast<GLsizei>(box.H), GL_RGB, GL_UNSIGNED_BYTE, rgb.data());
g_GLESFuncs.glPixelStorei(GL_UNPACK_ALIGNMENT, 4);
}
MGLOG_I_ONCE("Shared image texture %u: sub-data written into its image (%s texels)",
record.Desc.GlNameForDiag, m_sharedImageUploadsRgb ? "RGB" : "RGBA");
}
for (const auto& [uploadHalf, level] : keys) ConsumePipeTextureUpload(record.Desc.Resource, uploadHalf, level);
}
#endif
void BackendTextureObject::SyncMipmapsToBackend(
const SharedPtr<MG_State::GLState::ITextureObject>& stateTextureObject) {
// P5e (tx2): null IS the by-handle arm once a handle has been noted on this twin;
@@ -8811,6 +9041,22 @@ namespace MobileGL::MG_Backend::DirectGLES {
}
#endif
#if MOBILEGL_BUILD_DISAGGREGATED
// SHARED IMAGES: ahead of the view test and of the staged-store and serial gates below,
// none of which knows the texels belong to an image - the staged store holds none of
// them, and the storage path would allocate fresh, empty storage over the image.
if (pushedStorage != nullptr && (pushedStorage->SharedImageId != 0 || m_sharedImageId != 0) &&
SyncSharedImageStorage(*pushedStorage)) {
// EVERY use, not only the attach (the draw-clean gates never pass a bound image):
// a write that landed since the last one is waited for on this context's GPU
// before the command this sync is for, and the session's frame notes the read.
if (m_sharedImage != nullptr) SharedImageImpl::AcquireForSampling(m_sharedImage);
ReplaySharedImageUploads(*pushedStorage);
// A YUV image is converted into the name at the frame's first use.
if (m_sharedImage != nullptr && SharedImageImpl::IsYuv(m_sharedImage))
SharedImageImpl::ConvertYuvForSampling(m_sharedImage, GetBackendTextureId(),
pushedStorage->Desc.Width, pushedStorage->Desc.Height);
return;
}
// P5e (tx2), CONTRACT-P5E §5.2 (scout G-S2-2): THE VIEW TEST MOVES BEHIND THE
// RECORD. `IsTextureView()` was the one frontend read this function made BEFORE it
// had resolved anything at all, and Desc.ViewOf is the carrier - it names the storage
@@ -11078,6 +11324,20 @@ namespace MobileGL::MG_Backend::DirectGLES {
g_boundTexturesCache;
TwinRegistry<MG_State::GLState::ITextureObject, BackendTextureObject, MG_Pipe::MGPipeKind::Texture> g_backendTextureObjects;
#if MOBILEGL_BUILD_DISAGGREGATED
void WritePendingSharedImageUploads() {
auto& applier = MG_Pipe::MGPipeApplier();
for (SizeT slot = 0; slot < applier.TextureResources.size(); ++slot) {
const auto& record = applier.TextureResources[slot];
if (!record.Live || record.SharedImageId == 0 || record.PendingUploads.empty()) continue;
const MG_Pipe::MGPipeHandle handle{static_cast<Uint32>(slot), record.Gen};
auto* twin = g_backendTextureObjects.FindByHandle(handle);
if (twin == nullptr || !*twin) continue;
(*twin)->SyncMipmapsToBackendByHandle(handle);
}
}
#endif
void NoteDriverSideTextureWrite(const SharedPtr<MG_State::GLState::ITextureObject>& textureObject) {
SharedPtr<MG_State::GLState::ITextureObject> storage = textureObject;
// Bounded: GL resolves a view of a view to the original storage, so this takes one step.
@@ -12891,11 +13151,16 @@ namespace MobileGL::MG_Backend::DirectGLES {
namespace ScratchFBOImpl {
namespace {
ScratchFramebuffer g_tempFramebuffer;
ScratchFramebuffer g_blitReadFramebuffer;
ScratchFramebuffer g_blitDrawFramebuffer;
Uint g_completeTinyFBOId = 0;
Uint g_completeTinyRBOId = 0;
// ONE SET PER NATIVE CONTEXT: a framebuffer belongs to the context that made it
// (DirectGLES.h, PerNativeContext), and so does the attachment shadow kept for it.
struct ScratchSet {
ScratchFramebuffer temp;
ScratchFramebuffer blitRead;
ScratchFramebuffer blitDraw;
Uint completeTinyFBOId = 0;
Uint completeTinyRBOId = 0;
};
PerNativeContext<ScratchSet> g_scratchSets;
// Detach every point the shadow no longer vouches for. Used when the
// shadow is unknown (context reset, texture id deleted while attached).
@@ -12950,13 +13215,13 @@ namespace MobileGL::MG_Backend::DirectGLES {
} // namespace
ScratchFramebuffer& TempFramebuffer() {
return g_tempFramebuffer;
return g_scratchSets.Current().temp;
}
ScratchFramebuffer& BlitReadFramebuffer() {
return g_blitReadFramebuffer;
return g_scratchSets.Current().blitRead;
}
ScratchFramebuffer& BlitDrawFramebuffer() {
return g_blitDrawFramebuffer;
return g_scratchSets.Current().blitDraw;
}
Uint EnsureId(ScratchFramebuffer& fb) {
@@ -13125,6 +13390,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
}
Uint EnsureCompleteTinyFramebufferId() {
ScratchSet& set = g_scratchSets.Current();
Uint& g_completeTinyFBOId = set.completeTinyFBOId;
Uint& g_completeTinyRBOId = set.completeTinyRBOId;
if (g_completeTinyFBOId != 0) {
return g_completeTinyFBOId;
}
@@ -13152,19 +13420,18 @@ namespace MobileGL::MG_Backend::DirectGLES {
if (textureId == 0) {
return;
}
for (ScratchFramebuffer* fb : {&g_tempFramebuffer, &g_blitReadFramebuffer, &g_blitDrawFramebuffer}) {
if (fb->colorTex == textureId || fb->depthTex == textureId) {
fb->attachmentsKnown = false;
// Textures ARE shared, so the deleted id may sit in any context's scratch FBO.
g_scratchSets.ForEach([textureId](ScratchSet& set) {
for (ScratchFramebuffer* fb : {&set.temp, &set.blitRead, &set.blitDraw}) {
if (fb->colorTex == textureId || fb->depthTex == textureId) {
fb->attachmentsKnown = false;
}
}
}
});
}
void OnBackendContextDestroyed() {
g_tempFramebuffer = {};
g_blitReadFramebuffer = {};
g_blitDrawFramebuffer = {};
g_completeTinyFBOId = 0;
g_completeTinyRBOId = 0;
g_scratchSets.Clear();
}
} // namespace ScratchFBOImpl
@@ -16760,6 +17027,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
// PipeTextureStorageRecordForRecord and CONTRACT-P5E.md §5.2.
const MG_Pipe::MGPipeResourceRecord* storage = PipeTextureStorageRecordForRecord(record);
if (storage == nullptr) return false;
#if MOBILEGL_BUILD_DISAGGREGATED
// Never clean while bound to a shared image: see IsDrawSyncClean.
if (m_sharedImage != nullptr) return false;
#endif
// SyncMipmapsToBackend's own early-out, verbatim.
if (!m_isInitialized || m_syncedResourceSerial == 0 || m_syncedResourceSerial != storage->Serial ||
!storage->PendingUploads.empty()) {
+37
View File
@@ -74,6 +74,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
// flag while the process is very much alive.
Bool InProcessTeardown();
void EnsureProcessTeardownSentinel();
// The exit() half of InProcessTeardown alone, for an object whose release needs only the
// display, not a current context (an EGLImage): a server session's twin teardown has no
// context current but a live display, and skipping the release there would leak the image.
Bool InProcessExit();
// Generation of the backend ES context that owns the driver ids currently handed
// out. Bumped exactly once per DestroyEGLContext. Every backend twin that owns a
@@ -1940,6 +1944,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
switch (target) {
case TextureTarget::Texture1D:
case TextureTarget::TextureRectangle:
case TextureTarget::External:
return TextureTarget::Texture2D;
case TextureTarget::Texture1DArray:
return TextureTarget::Texture2DArray;
@@ -2099,6 +2104,18 @@ namespace MobileGL::MG_Backend::DirectGLES {
// level when the store covers it (notes/p9/W2-REMINT.md).
void RequireImageBindableStorageByHandle(MG_Pipe::MGPipeHandle res,
const MG_Pipe::MGPipeResourceRecord& record);
#endif
#if MOBILEGL_BUILD_DISAGGREGATED
// SHARED IMAGES (MGPipeResourceRecord::SharedImageId). True when this call settled the
// storage - the ES name's level 0 IS the image (glEGLImageTargetTexture2DOES), or the
// image could not be bound and the name is left empty - and SyncMipmapsToBackend must
// not go on to allocate or upload. False when the record names no image: a twin that
// was bound to one has been told to re-mint, and the normal storage path takes over.
Bool SyncSharedImageStorage(const MG_Pipe::MGPipeResourceRecord& record);
// Pending sub-data written into the attached shared image (Managers.cpp).
void ReplaySharedImageUploads(const MG_Pipe::MGPipeResourceRecord& record);
// The attached image takes RGB texels only (an X format): uploads are converted.
Bool m_sharedImageUploadsRgb = false;
#endif
// Whether this texture's ES storage was minted in an image carrier rather than in the
// frontend format's own layout - the readback has to ask, because for a NORMALIZED
@@ -2133,6 +2150,12 @@ namespace MobileGL::MG_Backend::DirectGLES {
Bool IsDrawSyncClean(const MG_State::GLState::ITextureObject* t, Uint64 contextId,
Uint64 samplingGeneration) const {
#if MOBILEGL_BUILD_DISAGGREGATED
// A shared image's texels move with its producer's presents, which nothing here
// sees: every use goes through SyncMipmapsToBackend's shared-image arm, which waits
// for a write that landed since the last one (SharedImageImpl::AcquireForSampling).
if (m_sharedImage != nullptr) return false;
#endif
if (!m_isInitialized || m_syncedShapeContextId == 0 || m_syncedShapeContextId != contextId ||
m_syncedShapeGeneration != samplingGeneration) {
return false;
@@ -2224,6 +2247,14 @@ namespace MobileGL::MG_Backend::DirectGLES {
Bool m_isInitialized = false;
Bool m_imageBindableStorageRequired = false;
Bool m_backendStorageImmutable = false;
#if MOBILEGL_BUILD_DISAGGREGATED
// The shared image the ES name's level 0 was last bound to (0 = none; image ids are
// never reused, so this is a lifetime key), and the EGLImage that binding went through,
// which holds the image for as long as the name samples it. Null with a nonzero id when
// the image could not be bound: the name is left empty until the record moves.
Uint64 m_sharedImageId = 0;
SharedImageImpl::EglImageRef m_sharedImage;
#endif
// Whether the driver texture may hold texels the shadow never saw. Set the first time
// anything other than a texel upload can write it - an application framebuffer
// attachment, a writable image binding, a copy or blit into it, a driver-side mipmap
@@ -2457,6 +2488,12 @@ namespace MobileGL::MG_Backend::DirectGLES {
// FBO's texture attachments - onto the backend, through the two borrowed-pair memos
// documented at their definitions. Declared here so tests can drive those memos directly.
void SyncNeccessaryTextures();
#if MOBILEGL_BUILD_DISAGGREGATED
// Every texture of this session bound to a shared image with sub-data still pending is
// synced now, so the uploads are in the images before a flush publishes them (an X
// server's PutImage into a window pixmap no later draw of its own samples).
void WritePendingSharedImageUploads();
#endif
extern Array<Array<BackendTextureObject*, (SizeT)TextureTarget::TextureTargetCount>,
MG_State::GLState::TextureState::MAX_TEXTURE_IMAGE_UNITS>
g_boundTexturesCache;
+23 -1
View File
@@ -224,6 +224,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
return found->second;
}
SharedPtr<Table> fresh = MakeShared<Table>();
fresh->SetOwnerSession(key.SessionKey);
m_buckets.emplace(key, fresh);
m_memoKey = key;
m_memoTable = fresh;
@@ -458,6 +459,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
// ENTRIES and the memo; the links are the table's own and are never copied.
BackendSlotTable() { LinkHolder(); }
BackendSlotTable(const BackendSlotTable& other):
m_ownerSession(other.m_ownerSession),
m_slots(other.m_slots),
m_band(other.m_band),
m_nullTwin(other.m_nullTwin),
@@ -466,6 +468,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
LinkHolder();
}
BackendSlotTable(BackendSlotTable&& other) noexcept:
m_ownerSession(other.m_ownerSession),
m_slots(std::move(other.m_slots)),
m_band(std::move(other.m_band)),
m_nullTwin(std::move(other.m_nullTwin)),
@@ -501,6 +504,12 @@ namespace MobileGL::MG_Backend::DirectGLES {
}
~BackendSlotTable() { UnlinkHolder(); }
// The served session whose bucket this table is (SlotTableRegistry::Shared), 0 for a
// table that belongs to none (the process-wide registry on a sessionless thread, a
// fixture's copy of one).
void SetOwnerSession(Uint64 session) { m_ownerSession = session; }
Uint64 OwnerSession() const { return m_ownerSession; }
// Resolve-or-create. The handle comes from the client allocator keyed on the frontend
// object's lifetime id, so two calls for the same live object always land on the same
// slot, and a successor object at the same heap address never does.
@@ -802,12 +811,24 @@ namespace MobileGL::MG_Backend::DirectGLES {
// double Free would be refused by generation anyway. Idempotent against the kind's
// own delete opcode, exactly as the notice arm is: a twin the delete already released
// fails ReleaseTwinAt's generation check and the walk moves on.
//
// ONE SESSION'S HOLDERS, WHEN THE DEATH IS ONE SESSION'S. A handle is numbered by its own
// client's allocator, so with several clients served by one process the same {slot, gen}
// names a live, unrelated object in every other session: walking every holder released
// THOSE twins too - a client deleting a texture (or exiting) deleted the compositor's
// driver texture with the same handle, and its windows sampled freed memory. On a thread
// that belongs to a session only that session's buckets answer; a sessionless thread
// (the in-library backend, a unit fixture) keeps the process-wide walk.
static Bool ReleaseTwinByHandle(MG_Pipe::MGPipeHandle handle) {
if (MG_Pipe::MGPipeHandleIsNull(handle)) return false;
TwinKey key;
const Bool sessionScoped = CurrentTwinKey(&key) && key.SessionKey != 0;
Bool released = false;
for (BackendSlotTable* holder = s_firstHolder; holder != nullptr;) {
BackendSlotTable* const next = holder->m_nextHolder;
released = holder->ReleaseTwinAt(handle) || released;
if (!sessionScoped || holder->m_ownerSession == key.SessionKey) {
released = holder->ReleaseTwinAt(handle) || released;
}
holder = next;
}
return released;
@@ -1012,6 +1033,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
static inline BackendSlotTable* s_firstHolder = nullptr;
BackendSlotTable* m_prevHolder = nullptr;
BackendSlotTable* m_nextHolder = nullptr;
Uint64 m_ownerSession = 0;
// Indexed by MGPipeHandle::Slot; [0] is the reserved slot and is never live.
Vector<Entry> m_slots;
+4 -1
View File
@@ -572,7 +572,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
ZoneScopedC(TRACY_ZONECOLOR_BACKEND);
#endif
// The name is the marker: ShaderSourceProcessor only emits it when the source
// wrote gl_FragColor, and such a shader can have no other output.
// names gl_FragColor, and a shader that writes it can have no other output. One that
// names it only in an inactive preprocessor arm keeps the declaration unwritten, and
// the IO resolver moves that inactive output off location 0, so the match below
// (which wants location 0) does not take it.
static const char* const kLoweredName = "mg_FragColor";
if (shaderType != GL_FRAGMENT_SHADER || drawBufferCount <= 1) {
return glslCode;
@@ -332,6 +332,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
#if MOBILEGL_BUILD_DISAGGREGATED
BackendObject_DirectVulkan::~BackendObject_DirectVulkan() {
// Everything below reaches this session's renderer slot, whichever thread destroys it.
MagmaSession* const ownSession = m_magmaSession.get();
MagmaSession* const previousSession = BoundMagmaSession();
if (ownSession != nullptr) BindMagmaSessionToThisThread(ownSession);
// P12 review fix: A SERVER SESSION THAT ENDS WITHOUT A CLEAN TEARDOWN LEAVES NO RENDERER ON
// THE WINDOW. The renderer - swapchain and VkSurfaceKHR on the window - is a process global
// that only ReleaseEGLResources / a released surface drops, and a client that went away
@@ -349,6 +353,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
if (MG_Config::Transport != MG_Config::TransportMode::Monolith) {
ForgetRequestedSwapInterval();
}
// The session's state goes with it, and no thread may keep answering with it.
m_magmaSession.reset();
if (ownSession != nullptr) {
BindMagmaSessionToThisThread(previousSession == ownSession ? nullptr : previousSession);
}
}
#else
BackendObject_DirectVulkan::~BackendObject_DirectVulkan() = default;
@@ -357,6 +366,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
BackendObject_DirectVulkan::BackendObject_DirectVulkan() : m_rendererInfo{GetRendererIdentity()} {}
Bool BackendObject_DirectVulkan::InitWindowSurface() {
// A suspended surface activates its placeholder - never the window, which is gone.
const auto suspended = m_suspendedWindowSurfaces.find(m_activatingSurface);
if (suspended != m_suspendedWindowSurfaces.end()) {
if (UsesSurfaceTargets() && pVulkanRenderer) {
return ActivateSuspendedPlaceholder(m_activatingSurface, suspended->second);
}
return InitPbufferSurface(static_cast<EGLint>(suspended->second.Width),
static_cast<EGLint>(suspended->second.Height));
}
if (!m_windowHandle.Handle) {
MGLOG_E("Cannot initialize DirectVulkan window surface: native window handle is null");
return false;
@@ -364,14 +382,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
auto nativeWindow = reinterpret_cast<NativeWindowType>(m_windowHandle.Handle);
VulkanRendererConfig config;
config.SwapInterval = GetRequestedSwapInterval();
if (UsesSurfaceTargets() && pVulkanRenderer) {
return pVulkanRenderer->ActivateSurfaceTarget(SurfaceTargetKey(m_activatingSurface), nativeWindow, config);
}
// Any renderer instance this assignment replaces is destroyed here;
// fence/timer-query handles stamped with the old generation go stale.
BumpRendererGeneration();
VulkanRendererConfig config;
config.SwapInterval = GetRequestedSwapInterval();
pVulkanRenderer = MakeUnique<MG_Backend::DirectVulkan::VulkanRenderer>(nativeWindow, config);
MOBILEGL_ASSERT(pVulkanRenderer != nullptr, "InitWindowSurface: VulkanRenderer creation failed");
pVulkanRenderer->Initialize();
pVulkanRenderer->SetActiveSurfaceTargetKey(SurfaceTargetKey(m_activatingSurface));
return true;
}
@@ -379,15 +401,35 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VulkanRendererConfig config;
config.SurfaceWidth = static_cast<Uint32>(std::max<EGLint>(width, 1));
config.SurfaceHeight = static_cast<Uint32>(std::max<EGLint>(height, 1));
if (UsesSurfaceTargets() && pVulkanRenderer) {
return pVulkanRenderer->ActivateSurfaceTarget(SurfaceTargetKey(m_activatingSurface), NativeWindowType{},
config);
}
// Any renderer instance this assignment replaces is destroyed here;
// fence/timer-query handles stamped with the old generation go stale.
BumpRendererGeneration();
pVulkanRenderer = MakeUnique<MG_Backend::DirectVulkan::VulkanRenderer>(NativeWindowType{}, config);
MOBILEGL_ASSERT(pVulkanRenderer != nullptr, "InitPbufferSurface: VulkanRenderer creation failed");
pVulkanRenderer->Initialize();
pVulkanRenderer->SetActiveSurfaceTargetKey(SurfaceTargetKey(m_activatingSurface));
return true;
}
void BackendObject_DirectVulkan::BindSessionStateToThisThread() {
if (m_magmaSession == nullptr) m_magmaSession = MakeUnique<MagmaSession>();
BindMagmaSessionToThisThread(m_magmaSession.get());
MGLOG_I("DirectVulkan: session state %llu bound to this apply thread",
static_cast<unsigned long long>(m_magmaSession->Id()));
}
Bool BackendObject_DirectVulkan::LatchIfGpuFaulted() {
#if MOBILEGL_BUILD_DISAGGREGATED
return pVulkanRenderer != nullptr && pVulkanRenderer->LatchIfGpuHung("apply");
#else
return false;
#endif
}
void BackendObject_DirectVulkan::Initialize() {
#if MOBILEGL_BUILD_DISAGGREGATED
VkBufferManager::RegisterWireResourceOps();
@@ -463,6 +505,22 @@ namespace MobileGL::MG_Backend::DirectVulkan {
if (!BackendObject::ResizeEGLWindowSurface(surface, width, height)) {
return false;
}
const EGLSurfaceState* state = GetRegisteredEGLSurface(surface);
if (UsesSurfaceTargets() && pVulkanRenderer && state != nullptr && state->Kind == SurfaceKind::Pbuffer) {
// A pbuffer-backed window (a Wayland window presented through wl_shm) whose target is an
// image reader of FIXED size: a swapchain rebuild cannot follow the resize, so the target
// is rebuilt at the new size - as DirectGLES replaces its native pbuffer. A parked one is
// rebuilt when it is next activated, from the size registered above.
const Bool active = m_eglSurfaceInitialized && m_eglSurface == surface;
pVulkanRenderer->DestroySurfaceTarget(SurfaceTargetKey(surface));
if (active) {
VulkanRendererConfig config;
config.SurfaceWidth = static_cast<Uint32>(state->Width);
config.SurfaceHeight = static_cast<Uint32>(state->Height);
(void)pVulkanRenderer->ActivateSurfaceTarget(SurfaceTargetKey(surface), NativeWindowType{}, config);
}
return true;
}
if (pVulkanRenderer && m_eglSurface == surface) {
pVulkanRenderer->RequestSwapchainResize(width, height);
}
@@ -480,7 +538,33 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Bool BackendObject_DirectVulkan::MakeEGLCurrent(EGLDisplay dpy, EGLSurface draw, EGLSurface read, EGLContext ctx) {
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
return BackendObject::MakeEGLCurrent(dpy, draw, read, ctx);
const Bool ok = BackendObject::MakeEGLCurrent(dpy, draw, read, ctx);
if (ok && UsesSurfaceTargets() && ctx != EGL_NO_CONTEXT && draw != EGL_NO_SURFACE) {
m_contextBindings[ctx] = EGLCurrentState{.Display = dpy, .DrawSurface = draw, .ReadSurface = read,
.Context = ctx};
}
return ok;
}
void BackendObject_DirectVulkan::OnClientContextBound(Uint64 token) {
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
if (!UsesSurfaceTargets() || token == 0) return;
// Each of a multi-window client's contexts draws to the surface IT was made current with
// (Qt renders every window on its own thread and context); the ring interleaves them with
// a bind_context between, and no make-current. Follow the context back to its surface.
const auto found = m_contextBindings.find(reinterpret_cast<EGLContext>(token));
if (found == m_contextBindings.end()) return;
const EGLCurrentState binding = found->second;
if (!GetRegisteredEGLSurface(binding.DrawSurface)) {
m_contextBindings.erase(found);
return;
}
const auto current = m_eglCurrentThreads.find(std::this_thread::get_id());
if (current != m_eglCurrentThreads.end() && current->second.Context == binding.Context &&
current->second.DrawSurface == binding.DrawSurface && m_eglSurface == binding.DrawSurface) {
return;
}
(void)BackendObject::MakeEGLCurrent(binding.Display, binding.DrawSurface, binding.ReadSurface, binding.Context);
}
Bool BackendObject_DirectVulkan::SwapEGLBuffers(EGLDisplay dpy, EGLSurface draw) {
@@ -492,6 +576,50 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return BackendObject::SwapEGLBuffers(dpy, draw);
}
Bool BackendObject_DirectVulkan::BlitDefaultFramebufferToSharedImage(const SharedImageView& image,
const MG_Util::Damage::Region& region) {
#if MOBILEGL_BUILD_DISAGGREGATED
// Same lock as SwapEGLBuffers: the copy reads the current surface target's image.
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
if (!pVulkanRenderer) {
MGLOG_E("DirectVulkan: shared image %llu: the renderer is not initialized",
static_cast<unsigned long long>(image.Id));
return false;
}
return pVulkanRenderer->BlitDefaultFramebufferToSharedImage(image, region);
#else
(void)image, (void)region;
return false;
#endif
}
Bool BackendObject_DirectVulkan::PublishSharedImageAccesses() {
#if MOBILEGL_BUILD_DISAGGREGATED
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
return pVulkanRenderer ? pVulkanRenderer->PublishSharedImageAccesses() : true;
#else
return true;
#endif
}
Bool BackendObject_DirectVulkan::ExportNativeFence(int* fence) {
*fence = -1;
#if MOBILEGL_BUILD_DISAGGREGATED
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
return pVulkanRenderer ? pVulkanRenderer->ExportNativeFence(fence) : true;
#else
return true;
#endif
}
// A swapchain image keeps its content through a present whatever region is drawn next: the
// damage region the client may declare is not needed here.
Int32 BackendObject_DirectVulkan::QueryCurrentBufferAge(Bool damageRegionFollows) {
(void)damageRegionFollows;
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
return pVulkanRenderer ? pVulkanRenderer->CurrentDrawBufferAge() : 0;
}
void BackendObject_DirectVulkan::SetEGLSwapInterval(Int interval) {
// Same lock as SwapEGLBuffers: the renderer reads the request in Present.
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
@@ -512,11 +640,131 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// The reflection cache is file-scope, not renderer-owned; without this the
// deleted programs' reflection strings survive full context teardown.
ClearProgramResourceCaches();
m_suspendedWindowSurfaces.clear();
BackendObject::ReleaseEGLResources();
}
Bool BackendObject_DirectVulkan::ActivateSuspendedPlaceholder(EGLSurface surface,
const SuspendedWindowSurface& extent) {
// A pbuffer target at the window's extent: every default-framebuffer path (render passes,
// read-back, the shared-image blit, the acquire and the present) keeps a live swapchain to
// work on, and the extent it publishes is the one the client already has.
VulkanRendererConfig config;
config.SurfaceWidth = std::max<Uint32>(extent.Width, 1);
config.SurfaceHeight = std::max<Uint32>(extent.Height, 1);
return pVulkanRenderer->ActivateSurfaceTarget(SurfaceTargetKey(surface), NativeWindowType{}, config);
}
Bool BackendObject_DirectVulkan::SuspendServerWindow(void* window) {
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
// The monolith's one renderer IS its window; only surface targets can be swapped under it.
if (window == nullptr || !UsesSurfaceTargets()) return false;
Vector<EGLSurface> onWindow;
for (const auto& [client, state] : m_eglSurfaces) {
if (state.Kind != SurfaceKind::Window || state.Window.Handle != window) continue;
if (m_suspendedWindowSurfaces.find(client) != m_suspendedWindowSurfaces.end()) continue;
onWindow.push_back(client);
}
for (const EGLSurface client : onWindow) {
const auto state = m_eglSurfaces.find(client);
SuspendedWindowSurface extent;
extent.Width = static_cast<Uint32>(std::max<EGLint>(state->second.Width, 1));
extent.Height = static_cast<Uint32>(std::max<EGLint>(state->second.Height, 1));
if (pVulkanRenderer) {
const Uint64 key = SurfaceTargetKey(client);
const VkExtent2D built = pVulkanRenderer->SurfaceTargetExtent(key);
if (built.width != 0 && built.height != 0) {
extent.Width = built.width;
extent.Height = built.height;
}
const Bool active = pVulkanRenderer->IsActiveSurfaceTarget(key);
// Flushes, waits the device idle, purges the render passes on the target and drops its
// swapchain and VkSurfaceKHR - nothing references the window after this.
pVulkanRenderer->DestroySurfaceTarget(key);
m_suspendedWindowSurfaces[client] = extent;
// A placeholder that cannot be built is a suspend that failed: the window is already let
// go of, and the caller ends the session by name instead (its EGL resources go).
if (active && !ActivateSuspendedPlaceholder(client, extent)) return false;
} else {
m_suspendedWindowSurfaces[client] = extent;
}
}
MGLOG_I("DirectVulkan: %zu window surface(s) on server window %p now draw into placeholder pbuffer targets; "
"the renderer, the device and every object stay",
onWindow.size(), window);
return true;
}
Bool BackendObject_DirectVulkan::ResumeServerWindow(void* window, Uint32 width, Uint32 height) {
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
if (window == nullptr || !UsesSurfaceTargets()) return false;
UnorderedMap<EGLSurface, SuspendedWindowSurface> suspended;
suspended.swap(m_suspendedWindowSurfaces);
for (const auto& [client, extent] : suspended) {
(void)extent;
const auto state = m_eglSurfaces.find(client);
if (state == m_eglSurfaces.end()) continue; // released while suspended
state->second.Window.Handle = window;
state->second.Window.Width = width;
state->second.Window.Height = height;
state->second.Width = static_cast<EGLint>(std::max<Uint32>(width, 1));
state->second.Height = static_cast<EGLint>(std::max<Uint32>(height, 1));
if (m_eglSurface == client) m_windowHandle = state->second.Window;
if (!pVulkanRenderer) continue;
const Uint64 key = SurfaceTargetKey(client);
const Bool active = pVulkanRenderer->IsActiveSurfaceTarget(key);
pVulkanRenderer->DestroySurfaceTarget(key); // the placeholder
// A parked surface is built on the new window when it is next activated (InitWindowSurface).
if (!active) continue;
// The creation's own path: a new VkSurfaceKHR and swapchain with the surface's current
// pre-transform, a new target serial (render passes rebuilt), and the extent published.
VulkanRendererConfig config;
config.SurfaceWidth = std::max<Uint32>(width, 1);
config.SurfaceHeight = std::max<Uint32>(height, 1);
config.SwapInterval = GetRequestedSwapInterval();
if (!pVulkanRenderer->ActivateSurfaceTarget(key, reinterpret_cast<NativeWindowType>(window), config)) {
return false;
}
}
MGLOG_I("DirectVulkan: %zu suspended window surface(s) rebuilt on server window %p (%ux%u)", suspended.size(),
window, width, height);
return true;
}
Bool BackendObject_DirectVulkan::ServerWindowResized(void* window, Uint32 width, Uint32 height) {
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
if (window == nullptr || !UsesSurfaceTargets()) return false;
Bool followed = false;
for (auto& [client, state] : m_eglSurfaces) {
if (state.Kind != SurfaceKind::Window || state.Window.Handle != window) continue;
if (m_suspendedWindowSurfaces.find(client) != m_suspendedWindowSurfaces.end()) continue;
state.Window.Width = width;
state.Window.Height = height;
state.Width = static_cast<EGLint>(std::max<Uint32>(width, 1));
state.Height = static_cast<EGLint>(std::max<Uint32>(height, 1));
if (m_eglSurface == client) m_windowHandle = state.Window;
// Magma's extent is its swapchain's, published when it is built: the active target follows
// the window now (or at the half-recorded frame's present); a parked one at its next present.
if (pVulkanRenderer && pVulkanRenderer->IsActiveSurfaceTarget(SurfaceTargetKey(client))) {
followed = pVulkanRenderer->FollowActiveWindowResize() || followed;
}
}
return followed;
}
Bool BackendObject_DirectVulkan::UsesSurfaceTargets() const {
return MG_Config::Transport != MG_Config::TransportMode::Monolith;
}
void BackendObject_DirectVulkan::OnEGLSurfaceForgotten(EGLSurface surface) {
m_suspendedWindowSurfaces.erase(surface);
if (UsesSurfaceTargets() && pVulkanRenderer) pVulkanRenderer->DestroySurfaceTarget(SurfaceTargetKey(surface));
}
void BackendObject_DirectVulkan::OnEGLSurfaceReleased(EGLSurface surface) {
(void)surface;
// The renderer outlives its surfaces here: only this surface's target goes (in
// OnEGLSurfaceForgotten, which follows), and every fence stays valid.
if (UsesSurfaceTargets() && pVulkanRenderer) return;
// Outstanding fence/timer-query handles now refer to a dead renderer;
// treat them as signaled/available with zero results from here on.
BumpRendererGeneration();
@@ -565,6 +813,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
V_OpenGL30, V_OpenGL31, V_OpenGL32, V_OpenGL33, V_OpenGL40, V_OpenGL41, V_OpenGL42, V_OpenGL43,
V_OpenGL44, V_OpenGL45, V_OpenGL46,
E_GL_ARB_draw_buffers_blend,
// The GLSL 1.00/1.10-era ARB trio plus NPOT textures: all core since GL 2.0 and
// trivially true for every context this backend claims (4.6). KWin 6's
// EglContext::checkSupported() gates desktop-GL compositing on exactly these four
// strings, so withholding them fails its renderer bring-up. Same block as DirectGLES.
E_GL_ARB_shader_objects, E_GL_ARB_vertex_shader, E_GL_ARB_fragment_shader,
E_GL_ARB_texture_non_power_of_two,
E_GL_ARB_compute_shader, E_GL_ARB_shader_storage_buffer_object, E_GL_ARB_shader_image_load_store,
E_GL_ARB_clear_buffer_object, E_GL_ARB_program_interface_query, E_GL_ARB_framebuffer_object, E_GL_ARB_draw_indirect,
E_GL_ARB_multi_draw_indirect,
@@ -9,6 +9,7 @@
#pragma once
#include <Includes.h>
#include "../BackendObject.h"
#include "MagmaSession.h"
#include <MG_Util/BackendLoaders/Vulkan/Loader.h>
namespace MobileGL::MG_Backend::DirectVulkan {
@@ -26,6 +27,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
~BackendObject_DirectVulkan() override;
void Initialize() override;
void BindSessionStateToThisThread() override;
// The GPU hang watch's verdict on this session's device (VulkanRenderer::LatchIfGpuHung).
Bool LatchIfGpuFaulted() override;
void OnClientContextBound(Uint64 token) override;
Bool InitWindowSurface() override;
Bool InitCapabilities() override;
Bool InitializeEGLDisplay(EGLDisplay dpy, EGLint* major, EGLint* minor) override;
@@ -35,8 +40,21 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Bool MakeEGLCurrent(EGLDisplay dpy, EGLSurface draw, EGLSurface read, EGLContext ctx) override;
Bool SwapEGLBuffers(EGLDisplay dpy, EGLSurface draw) override;
void SetEGLSwapInterval(Int interval) override;
Bool BlitDefaultFramebufferToSharedImage(const SharedImageView& image,
const MG_Util::Damage::Region& region) override;
// The active surface target's swapchain image ages (VulkanRenderer::CurrentDrawBufferAge).
Int32 QueryCurrentBufferAge(Bool damageRegionFollows) override;
// A glFlush that ends a shared-image producer's frame (VulkanRenderer::PublishSharedImageAccesses).
Bool PublishSharedImageAccesses() override;
// An EGL native fence of the session's work (VulkanRenderer::ExportNativeFence).
Bool ExportNativeFence(int* fence) override;
void ReleaseEGLSurface(EGLSurface surface) override;
void ReleaseEGLResources() override;
// The server window going and coming back: the window surface's target (VkSurfaceKHR,
// swapchain) is destroyed and a pbuffer target of the same extent stands in for it, and back.
Bool SuspendServerWindow(void* window) override;
Bool ResumeServerWindow(void* window, Uint32 width, Uint32 height) override;
Bool ServerWindowResized(void* window, Uint32 width, Uint32 height) override;
const RendererInfo& GetRendererInfo() const override;
String GetBackendAPIVersionString() const override;
@@ -48,6 +66,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
private:
Bool InitPbufferSurface(EGLint width, EGLint height) override;
void OnEGLSurfaceReleased(EGLSurface surface) override;
void OnEGLSurfaceForgotten(EGLSurface surface) override;
// A served session keeps one renderer across all its client's surfaces and switches the
// renderer's surface target instead (VulkanRenderer::ActivateSurfaceTarget).
Bool UsesSurfaceTargets() const;
static Uint64 SurfaceTargetKey(EGLSurface surface) { return reinterpret_cast<Uint64>(surface); }
void UpdateAdvertisedExtensions();
void UpdateDynamicBackendParameters();
@@ -55,6 +78,21 @@ namespace MobileGL::MG_Backend::DirectVulkan {
DynamicBackendParameters m_dynamicParameters;
MG_External::VulkanCapabilities m_vulkanCaps;
RendererInfo m_rendererInfo;
// A served session's own copy of Magma's per-process state (MagmaSession.h), bound to its
// apply thread. Null in the monolith and every other single-session shape.
UniquePtr<MagmaSession> m_magmaSession;
// The binding each client context was last made current with, for OnClientContextBound.
UnorderedMap<EGLContext, EGLCurrentState> m_contextBindings;
// Window surfaces whose window is gone (SuspendServerWindow), with the extent their
// placeholder pbuffer target is built at - the window's, so the default framebuffer the
// client knows keeps its size.
struct SuspendedWindowSurface {
Uint32 Width = 1;
Uint32 Height = 1;
};
UnorderedMap<EGLSurface, SuspendedWindowSurface> m_suspendedWindowSurfaces;
// The placeholder pbuffer target of a suspended surface, made active. False: it could not be built.
Bool ActivateSuspendedPlaceholder(EGLSurface surface, const SuspendedWindowSurface& extent);
};
// Single-source-of-truth helpers shared with the driver POST
@@ -35,8 +35,9 @@
#include <spirv_reflect.h>
namespace MobileGL::MG_Backend::DirectVulkan {
// Leak-at-exit storage; see GlobalObjects.cpp.
UniquePtr<VulkanRenderer>& pVulkanRenderer = *new UniquePtr<VulkanRenderer>();
// Per Magma session (MagmaSession.h); the process-wide session is leak-at-exit storage, see
// GlobalObjects.cpp.
RendererSlot pVulkanRenderer;
namespace {
#if MOBILEGL_BUILD_DISAGGREGATED
@@ -59,7 +60,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// current one (a new renderer restarts its frame-serial counter and
// reuses pool indices). Atomic because handles may be polled from a
// thread other than the EGL thread that recreates the renderer.
std::atomic<Uint64> g_rendererGeneration{1};
//
// One per Magma session, drawn from one process-wide source: a handle stamped by one
// session's renderer must read as stale against every other session's, never equal.
std::atomic<Uint64> g_rendererGenerationSource{1};
struct RendererGeneration {
std::atomic<Uint64> value{g_rendererGenerationSource.fetch_add(1, std::memory_order_acq_rel)};
};
SessionLocal<RendererGeneration> g_rendererGeneration;
} // namespace
#if MOBILEGL_BUILD_DISAGGREGATED && MOBILEGL_PIPE_PUSH
@@ -123,11 +131,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
#endif // MOBILEGL_BUILD_DISAGGREGATED && MOBILEGL_PIPE_PUSH
Uint64 GetRendererGeneration() {
return g_rendererGeneration.load(std::memory_order_acquire);
return g_rendererGeneration->value.load(std::memory_order_acquire);
}
void BumpRendererGeneration() {
g_rendererGeneration.fetch_add(1, std::memory_order_acq_rel);
g_rendererGeneration->value.store(g_rendererGenerationSource.fetch_add(1, std::memory_order_acq_rel),
std::memory_order_release);
}
namespace {
@@ -1274,7 +1283,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
payload.mode = mode;
// TODO: allocate draw cmd buf elsewhere
static Vector<DrawCmdParam> params;
static thread_local Vector<DrawCmdParam> params;
params.clear();
params.resize(drawcount);
@@ -1347,7 +1356,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
const Uint32 indexSizeShift = static_cast<Uint32>(std::countr_zero(indexSize));
// TODO: allocate draw cmd buf elsewhere
static Vector<DrawIndexedCmdParam> params;
static thread_local Vector<DrawIndexedCmdParam> params;
params.clear();
params.resize(drawcount);
@@ -1751,7 +1760,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
void Present() {
MOBILEGL_ASSERT(pVulkanRenderer, "DirectVulkan::Present called with null VulkanRenderer");
// A served session's renderer is gone once its client released its resources; a present
// that still reaches it has nothing to show and is dropped rather than dereferenced.
if (!pVulkanRenderer) {
MGLOG_E_ONCE("DirectVulkan: a present reached a session whose renderer is already released; dropped");
return;
}
pVulkanRenderer->Present();
// THE frame boundary for the MGPipe counters, at the backend entry point rather
// than inside VulkanRenderer::Present: that function has an early return for the
@@ -1763,23 +1777,24 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
namespace {
Optional<Int> g_requestedSwapInterval;
// Per session: each client asks for its own interval.
SessionLocal<Optional<Int>> g_requestedSwapInterval;
} // namespace
void SetSwapInterval(Int interval) {
g_requestedSwapInterval = interval;
*g_requestedSwapInterval = interval;
if (pVulkanRenderer) {
pVulkanRenderer->SetSwapInterval(interval);
}
}
Optional<Int> GetRequestedSwapInterval() {
return g_requestedSwapInterval;
return *g_requestedSwapInterval;
}
#if MOBILEGL_BUILD_DISAGGREGATED
void ForgetRequestedSwapInterval() {
g_requestedSwapInterval.reset();
g_requestedSwapInterval->reset();
}
#endif
@@ -9,10 +9,32 @@
#pragma once
#include <Includes.h>
#include <MG_Backend/BackendObject.h>
#include "MagmaSession.h"
#include "Renderer/VulkanRenderer.h"
namespace MobileGL::MG_Backend::DirectVulkan {
extern UniquePtr<VulkanRenderer>& pVulkanRenderer;
// The renderer of the calling thread's Magma session (MagmaSession.h): a served session's own on
// its apply thread, the process-wide one everywhere else. Spelled like the UniquePtr it used to
// be so the call sites read the same.
class RendererSlot {
public:
UniquePtr<VulkanRenderer>& Slot() const { return m_slot.Get(); }
VulkanRenderer* get() const { return Slot().get(); }
VulkanRenderer* operator->() const { return get(); }
VulkanRenderer& operator*() const { return *get(); }
explicit operator bool() const { return get() != nullptr; }
RendererSlot& operator=(UniquePtr<VulkanRenderer>&& renderer) {
Slot() = std::move(renderer);
return *this;
}
void reset() { Slot().reset(); }
friend Bool operator==(const RendererSlot& slot, std::nullptr_t) { return slot.get() == nullptr; }
friend Bool operator!=(const RendererSlot& slot, std::nullptr_t) { return slot.get() != nullptr; }
private:
SessionLocal<UniquePtr<VulkanRenderer>> m_slot;
};
extern RendererSlot pVulkanRenderer;
// Generation of the live VulkanRenderer instance, mirroring DirectGLES's
// g_syncContextGeneration. BackendObject_DirectVulkan bumps it wherever
@@ -0,0 +1,87 @@
// MobileGL - MobileGL/MG_Backend/DirectVulkan/MagmaSession.cpp
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#include "MagmaSession.h"
#include <atomic>
namespace MobileGL::MG_Backend::DirectVulkan {
namespace {
struct SessionLocalEntry {
Detail::SessionLocalConstruct construct;
Detail::SessionLocalDestroy destroy;
};
// Construct-on-first-use: SessionLocals are namespace-scope objects in several translation
// units and register during static initialization, in no particular order.
Vector<SessionLocalEntry>& Registry() {
static auto* registry = new Vector<SessionLocalEntry>();
return *registry;
}
std::atomic<Uint64> g_nextSessionId{1};
thread_local MagmaSession* t_boundSession = nullptr;
// Leak-at-exit, like the renderer it holds (GlobalObjects.cpp): nothing may tear the
// monolith's renderer down from a static destructor after the driver is gone.
MagmaSession& ProcessSession() {
static auto* session = new MagmaSession();
return *session;
}
} // namespace
Uint32 Detail::RegisterSessionLocal(SessionLocalConstruct construct, SessionLocalDestroy destroy) {
auto& registry = Registry();
registry.push_back({construct, destroy});
return static_cast<Uint32>(registry.size() - 1);
}
MagmaSession& Detail::CurrentMagmaSession() {
MagmaSession* bound = t_boundSession;
return bound != nullptr ? *bound : ProcessSession();
}
void* Detail::SessionLocalSlot(MagmaSession& session, Uint32 index) {
if (index >= session.m_slots.size()) session.m_slots.resize(Registry().size(), nullptr);
void*& slot = session.m_slots[index];
if (slot == nullptr) slot = Registry()[index].construct();
return slot;
}
MagmaSession::MagmaSession() : m_id(g_nextSessionId.fetch_add(1, std::memory_order_relaxed)) {
// Every slot is built up front: the process-wide session is reached from whichever thread
// the app renders on, and a lazily built slot would be a first-touch race between two.
const auto& registry = Registry();
m_slots.resize(registry.size(), nullptr);
for (SizeT i = 0; i < registry.size(); ++i) m_slots[i] = registry[i].construct();
}
MagmaSession::~MagmaSession() {
MagmaSession* const previous = t_boundSession;
t_boundSession = this;
const auto& registry = Registry();
for (SizeT i = m_slots.size(); i-- > 0;) {
if (m_slots[i] != nullptr) registry[i].destroy(m_slots[i]);
m_slots[i] = nullptr;
}
// Never back to this session: it is gone.
t_boundSession = previous == this ? nullptr : previous;
}
MagmaSessionScope::MagmaSessionScope(MagmaSession* session) : m_previous(t_boundSession) {
t_boundSession = session;
}
MagmaSessionScope::~MagmaSessionScope() { t_boundSession = m_previous; }
void BindMagmaSessionToThisThread(MagmaSession* session) { t_boundSession = session; }
MagmaSession* BoundMagmaSession() { return t_boundSession; }
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -0,0 +1,100 @@
// MobileGL - MobileGL/MG_Backend/DirectVulkan/MagmaSession.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
// MAGMA'S PER-SESSION STATE. Magma was written for one renderer per process: the renderer slot,
// the active buffer manager, the dynamic-state shadow, the render-pass manager's statics and the
// renderer generation are all file-scope. A multi-session server (several clients served from one
// process, each on its own apply thread with its own BackendObject_DirectVulkan) needs one copy of
// each per session, or one session's work lands in another's tables.
//
// SessionLocal<T> is that copy. It resolves through the CALLING THREAD's bound MagmaSession, and to
// the process-wide default session on every thread nothing bound - which is the monolith, the client
// role and every unit case, i.e. exactly the old behaviour, on whatever thread the app renders from.
// A served session's backend owns its MagmaSession and binds it on its apply thread (the only thread
// that runs that session's backend), so the two never mix.
#pragma once
#include <Includes.h>
#include <cstddef>
#include <new>
#include <utility>
namespace MobileGL::MG_Backend::DirectVulkan {
class MagmaSession;
namespace Detail {
using SessionLocalConstruct = void* (*)();
using SessionLocalDestroy = void (*)(void*);
// Registers one SessionLocal<T> and returns its slot. Called from static initialization.
Uint32 RegisterSessionLocal(SessionLocalConstruct construct, SessionLocalDestroy destroy);
// The calling thread's session, or the process-wide default.
MagmaSession& CurrentMagmaSession();
void* SessionLocalSlot(MagmaSession& session, Uint32 index);
} // namespace Detail
// One served session's copy of every SessionLocal. Slots are built lazily on first use and
// destroyed in reverse registration order with the session still bound, so a destructor that
// reaches another SessionLocal finds this session's copy and not the binding thread's.
class MagmaSession {
public:
MagmaSession();
~MagmaSession();
MagmaSession(const MagmaSession&) = delete;
MagmaSession& operator=(const MagmaSession&) = delete;
// A monotonic id, unique for the process lifetime. Never an address: a later session can
// be allocated where an ended one lived.
Uint64 Id() const { return m_id; }
private:
friend void* Detail::SessionLocalSlot(MagmaSession& session, Uint32 index);
Vector<void*> m_slots;
Uint64 m_id = 0;
};
// Makes `session` the calling thread's Magma session for the scope's lifetime, restoring the
// previous binding (usually none) afterwards.
class MagmaSessionScope {
public:
explicit MagmaSessionScope(MagmaSession* session);
~MagmaSessionScope();
MagmaSessionScope(const MagmaSessionScope&) = delete;
MagmaSessionScope& operator=(const MagmaSessionScope&) = delete;
private:
MagmaSession* m_previous = nullptr;
};
// Binds `session` to the calling thread until the next call (nullptr unbinds). For the apply
// thread, which binds its backend once and runs it for the rest of its life.
void BindMagmaSessionToThisThread(MagmaSession* session);
// The calling thread's bound session, or nullptr (the process-wide default answers).
MagmaSession* BoundMagmaSession();
template <class T>
class SessionLocal {
public:
SessionLocal() : m_index(Detail::RegisterSessionLocal(&Construct, &Destroy)) {}
SessionLocal(const SessionLocal&) = delete;
SessionLocal& operator=(const SessionLocal&) = delete;
T& Get() const {
return *static_cast<T*>(Detail::SessionLocalSlot(Detail::CurrentMagmaSession(), m_index));
}
T& operator*() const { return Get(); }
T* operator->() const { return &Get(); }
private:
static void* Construct() { return new T(); }
static void Destroy(void* object) { delete static_cast<T*>(object); }
const Uint32 m_index;
};
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -66,7 +66,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
FreeRetiredCommandBuffers(frame);
}
vkFreeCommandBuffers(device, commandPool, frameCount * 2, commandBuffers.data());
if (!m_spareCommandBuffers.empty())
vkFreeCommandBuffers(device, commandPool, static_cast<Uint32>(m_spareCommandBuffers.size()),
m_spareCommandBuffers.data());
}
m_spareCommandBuffers.clear();
m_frames.clear();
currentFrameIndex = 0;
m_device = VK_NULL_HANDLE;
@@ -300,6 +304,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return packet;
}
VkResult FrameContext::AcquireNextImageInCurrentSlot(VkDevice device, VkSwapchainKHR swapchain,
Uint32& outImageIndex) {
auto& frame = GetCurrent();
const VkResult result = vkAcquireNextImageKHR(device, swapchain, UINT64_MAX, frame.imageAvailableSemaphore,
VK_NULL_HANDLE, &outImageIndex);
if (result == VK_SUCCESS || result == VK_SUBOPTIMAL_KHR) frame.imageAvailableSemaphoreConsumed = false;
return result;
}
VkResult FrameContext::WaitAndAcquireNextImage(VkDevice device, VkSwapchainKHR swapchain, Uint32& outImageIndex,
Uint64 timeout, VkFence acquireFence) {
auto& frame = GetCurrent();
@@ -353,21 +366,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
MOBILEGL_ASSERT(!frame.isPreCommandRecording,
"RetireCurrentCommandBuffer called while the pre-pass stream is still recording");
VkCommandBufferAllocateInfo allocInfo{};
allocInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO;
allocInfo.commandPool = m_commandPool;
allocInfo.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY;
allocInfo.commandBufferCount = 1;
VkCommandBuffer replacement = VK_NULL_HANDLE;
VkResult result = vkAllocateCommandBuffers(m_device, &allocInfo, &replacement);
VkResult result = TakeCommandBuffer(replacement);
if (result != VK_SUCCESS) {
return result;
}
if (retirePreCommandBuffer) {
VkCommandBuffer preReplacement = VK_NULL_HANDLE;
result = vkAllocateCommandBuffers(m_device, &allocInfo, &preReplacement);
result = TakeCommandBuffer(preReplacement);
if (result != VK_SUCCESS) {
vkFreeCommandBuffers(m_device, m_commandPool, 1, &replacement);
RecycleCommandBuffer(replacement);
return result;
}
frame.retiredCommandBuffers.push_back({frame.preCommandBuffer, frame.lastSubmitIndex});
@@ -402,8 +410,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
SizeT completedCount = 0;
while (completedCount < frame.retiredCommandBuffers.size() &&
frame.retiredCommandBuffers[completedCount].submitIndex <= completedSubmitIndex) {
vkFreeCommandBuffers(m_device, m_commandPool, 1,
&frame.retiredCommandBuffers[completedCount].commandBuffer);
RecycleCommandBuffer(frame.retiredCommandBuffers[completedCount].commandBuffer);
++completedCount;
}
if (completedCount > 0) {
@@ -413,6 +420,31 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
}
VkResult FrameContext::TakeCommandBuffer(VkCommandBuffer& out) {
if (!m_spareCommandBuffers.empty()) {
out = m_spareCommandBuffers.back();
m_spareCommandBuffers.pop_back();
return VK_SUCCESS;
}
VkCommandBufferAllocateInfo allocInfo{};
allocInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO;
allocInfo.commandPool = m_commandPool;
allocInfo.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY;
allocInfo.commandBufferCount = 1;
return vkAllocateCommandBuffers(m_device, &allocInfo, &out);
}
void FrameContext::RecycleCommandBuffer(VkCommandBuffer commandBuffer) {
// A handful covers the buffers in flight between two completions; beyond that, free.
constexpr SizeT kSpareCommandBuffers = 8;
if (commandBuffer == VK_NULL_HANDLE) return;
if (m_spareCommandBuffers.size() < kSpareCommandBuffers) {
m_spareCommandBuffers.push_back(commandBuffer);
return;
}
vkFreeCommandBuffers(m_device, m_commandPool, 1, &commandBuffer);
}
void FrameContext::FreeAllRetiredCommandBuffers() {
for (auto& frame : m_frames) {
FreeRetiredCommandBuffers(frame);
@@ -101,6 +101,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkImageLayout presentLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR);
SubmitInfoPacket GetSubmitInfo(Bool shouldSubmitCommandBuffer, Uint32 swapchainImageIndex) const;
PresentInfoPacket GetPresentInfo(VkSwapchainKHR swapchain, Uint32 imageIndex) const;
// Acquires into the current slot's semaphore WITHOUT the slot-fence wait: for a surface
// brought up mid-frame, whose slot fence this frame has already waited and reset. The
// slot's semaphore must be unsignaled with no pending signal (consumed).
VkResult AcquireNextImageInCurrentSlot(VkDevice device, VkSwapchainKHR swapchain, Uint32& outImageIndex);
// The per-swapchain-image render-finished semaphores belong to one swapchain; a renderer
// that parks a surface parks them with it.
void ExchangeSwapchainSemaphores(Vector<VkSemaphore>& other) { std::swap(m_swapchainImageRenderFinishedSemaphores, other); }
VkResult WaitAndAcquireNextImage(VkDevice device, VkSwapchainKHR swapchain, Uint32& outImageIndex,
Uint64 timeout = UINT64_MAX, VkFence acquireFence = VK_NULL_HANDLE);
@@ -143,5 +150,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// Stored at Initialize for retired-command-buffer management.
VkDevice m_device = VK_NULL_HANDLE;
VkCommandPool m_commandPool = VK_NULL_HANDLE;
// Completed command buffers kept for the next retirement instead of being freed and
// reallocated: the pool resets a buffer implicitly at its next vkBeginCommandBuffer.
Vector<VkCommandBuffer> m_spareCommandBuffers;
VkResult TakeCommandBuffer(VkCommandBuffer& out);
void RecycleCommandBuffer(VkCommandBuffer commandBuffer);
};
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -0,0 +1,202 @@
// MobileGL - MobileGL/MG_Backend/DirectVulkan/Renderer/GpuProgressMarkers.cpp
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#include "GpuProgressMarkers.h"
#include <MG_Util/Debug/Log.h>
namespace MobileGL::MG_Backend::DirectVulkan {
namespace Watch = MG_Backend::GpuHangWatch;
void GpuProgressMarkers::Init(VkDevice device, Uint32 queueFamilyIndex) {
if (m_device != VK_NULL_HANDLE || device == VK_NULL_HANDLE) return;
if (Watch::BudgetMs() == 0) {
MGLOG_I("Magma: GPU hang watch off (MOBILEGL_GPU_HANG_BUDGET_MS=0)");
return;
}
VkCommandPoolCreateInfo poolInfo{VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO};
poolInfo.queueFamilyIndex = queueFamilyIndex;
if (vkCreateCommandPool(device, &poolInfo, nullptr, &m_pool) != VK_SUCCESS) {
MGLOG_W("Magma: GPU hang watch off - no command pool for its markers");
m_pool = VK_NULL_HANDLE;
return;
}
m_device = device;
{
const std::lock_guard<std::mutex> lock(m_mutex);
if (!AddSlot()) {
MGLOG_W("Magma: GPU hang watch off - the device cannot make its progress markers");
vkDestroyCommandPool(m_device, m_pool, nullptr);
m_pool = VK_NULL_HANDLE;
m_device = VK_NULL_HANDLE;
return;
}
}
m_watch = Watch::Register(this, &Sample, &OnHung);
MGLOG_I("Magma: GPU hang watch on - a submission the GPU started and has not finished after %u ms ends "
"its session",
Watch::BudgetMs());
}
void GpuProgressMarkers::Shutdown() {
// The watch first, outside m_mutex: Unregister waits out a sample in progress, which takes it.
Watch::Unregister(m_watch);
m_watch = 0;
if (m_device == VK_NULL_HANDLE) return;
const std::lock_guard<std::mutex> lock(m_mutex);
for (Slot& slot : m_slots) {
if (slot.Started != VK_NULL_HANDLE) vkDestroyEvent(m_device, slot.Started, nullptr);
if (slot.Finished != VK_NULL_HANDLE) vkDestroyEvent(m_device, slot.Finished, nullptr);
}
m_slots.clear();
if (m_pool != VK_NULL_HANDLE) vkDestroyCommandPool(m_device, m_pool, nullptr); // frees the buffers
m_pool = VK_NULL_HANDLE;
m_device = VK_NULL_HANDLE;
}
Bool GpuProgressMarkers::AddSlot() {
if (m_slots.size() >= kMaxSlots) return false;
Slot slot;
const VkEventCreateInfo eventInfo{VK_STRUCTURE_TYPE_EVENT_CREATE_INFO};
VkCommandBuffer buffers[2] = {VK_NULL_HANDLE, VK_NULL_HANDLE};
VkCommandBufferAllocateInfo allocInfo{VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO};
allocInfo.commandPool = m_pool;
allocInfo.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY;
allocInfo.commandBufferCount = 2;
const auto fail = [&] {
if (slot.Started != VK_NULL_HANDLE) vkDestroyEvent(m_device, slot.Started, nullptr);
if (slot.Finished != VK_NULL_HANDLE) vkDestroyEvent(m_device, slot.Finished, nullptr);
if (buffers[0] != VK_NULL_HANDLE) vkFreeCommandBuffers(m_device, m_pool, 2, buffers);
return false;
};
if (vkCreateEvent(m_device, &eventInfo, nullptr, &slot.Started) != VK_SUCCESS ||
vkCreateEvent(m_device, &eventInfo, nullptr, &slot.Finished) != VK_SUCCESS ||
vkAllocateCommandBuffers(m_device, &allocInfo, buffers) != VK_SUCCESS)
return fail();
// Prerecorded once and resubmitted for every submission the slot brackets; SIMULTANEOUS_USE
// because the slot is reused once its `finished` event is set, which can be a moment before
// the submission that set it formally leaves the pending state.
VkCommandBufferBeginInfo beginInfo{VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO};
beginInfo.flags = VK_COMMAND_BUFFER_USAGE_SIMULTANEOUS_USE_BIT;
if (vkBeginCommandBuffer(buffers[0], &beginInfo) != VK_SUCCESS) return fail();
vkCmdSetEvent(buffers[0], slot.Started, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT);
if (vkEndCommandBuffer(buffers[0]) != VK_SUCCESS) return fail();
if (vkBeginCommandBuffer(buffers[1], &beginInfo) != VK_SUCCESS) return fail();
vkCmdSetEvent(buffers[1], slot.Finished, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT);
if (vkEndCommandBuffer(buffers[1]) != VK_SUCCESS) return fail();
slot.Begin = buffers[0];
slot.End = buffers[1];
m_slots.push_back(slot);
return true;
}
GpuProgressMarkers::Bracket GpuProgressMarkers::Acquire() {
Bracket bracket;
if (m_device == VK_NULL_HANDLE) return bracket;
const std::lock_guard<std::mutex> lock(m_mutex);
if (m_lost) return bracket;
Slot* free = nullptr;
for (Slot& slot : m_slots) {
if (slot.Claimed) continue;
// Retire what finished since the watch last looked: frames come faster than its polls.
if (slot.InFlight && vkGetEventStatus(m_device, slot.Finished) == VK_EVENT_SET) slot.InFlight = false;
if (!slot.InFlight && free == nullptr) free = &slot;
}
if (free == nullptr) {
if (!AddSlot()) return bracket;
free = &m_slots.back();
}
if (free->Dirty) {
if (vkResetEvent(m_device, free->Started) != VK_SUCCESS ||
vkResetEvent(m_device, free->Finished) != VK_SUCCESS)
return bracket;
free->Dirty = false;
}
free->Claimed = true;
bracket.Begin = free->Begin;
bracket.End = free->End;
bracket.Slot = static_cast<Uint32>(free - m_slots.data());
return bracket;
}
void GpuProgressMarkers::Submitted(const Bracket& bracket) {
if (!bracket.Valid()) return;
const std::lock_guard<std::mutex> lock(m_mutex);
if (bracket.Slot >= m_slots.size()) return;
Slot& slot = m_slots[bracket.Slot];
slot.Claimed = false;
slot.InFlight = true;
slot.Dirty = true;
slot.Id = m_nextId++;
}
void GpuProgressMarkers::Abandon(const Bracket& bracket) {
if (!bracket.Valid()) return;
const std::lock_guard<std::mutex> lock(m_mutex);
if (bracket.Slot >= m_slots.size()) return;
// A failed vkQueueSubmit may still have consumed nothing or something: reset before reuse.
m_slots[bracket.Slot].Claimed = false;
m_slots[bracket.Slot].Dirty = true;
}
Bool GpuProgressMarkers::Sample(void* self, Vector<MG_Backend::GpuHangWatch::SubmissionProgress>& out) {
auto* markers = static_cast<GpuProgressMarkers*>(self);
const std::lock_guard<std::mutex> lock(markers->m_mutex);
if (markers->m_lost || markers->m_device == VK_NULL_HANDLE) return false;
for (Slot& slot : markers->m_slots) {
if (!slot.InFlight) continue;
const VkResult finished = vkGetEventStatus(markers->m_device, slot.Finished);
const VkResult started = vkGetEventStatus(markers->m_device, slot.Started);
if ((finished != VK_EVENT_SET && finished != VK_EVENT_RESET) ||
(started != VK_EVENT_SET && started != VK_EVENT_RESET)) {
// VK_ERROR_DEVICE_LOST: the driver already reset the device and says so itself.
markers->m_lost = true;
return false;
}
if (finished == VK_EVENT_SET) {
slot.InFlight = false;
continue;
}
out.push_back({slot.Id, started == VK_EVENT_SET, false});
}
return true;
}
void GpuProgressMarkers::OnHung(void* self, Uint64 submissionId, Uint64 runningMs) {
auto* markers = static_cast<GpuProgressMarkers*>(self);
markers->m_hungRunningMs.store(runningMs, std::memory_order_release);
markers->m_hung.store(true, std::memory_order_release);
MGLOG_E("Magma: submission %llu of this session's VkDevice has run %llu ms on the GPU and has not finished - "
"it is what stops the GPU for every process; the session ends at its next record and submits "
"nothing more (budget %u ms, MOBILEGL_GPU_HANG_BUDGET_MS)",
static_cast<unsigned long long>(submissionId), static_cast<unsigned long long>(runningMs),
MG_Backend::GpuHangWatch::BudgetMs());
}
VkResult GpuProgressMarkers::SubmitBracketed(VkQueue queue, VkSubmitInfo& info, VkFence fence,
const Bracket& bracket) {
constexpr Uint32 kMaxOwn = 6;
if (!bracket.Valid() || info.commandBufferCount == 0 || info.commandBufferCount > kMaxOwn)
return vkQueueSubmit(queue, 1, &info, fence);
VkCommandBuffer buffers[kMaxOwn + 2];
Uint32 count = 0;
buffers[count++] = bracket.Begin;
for (Uint32 i = 0; i < info.commandBufferCount; ++i) buffers[count++] = info.pCommandBuffers[i];
buffers[count++] = bracket.End;
const VkCommandBuffer* own = info.pCommandBuffers;
const Uint32 ownCount = info.commandBufferCount;
info.pCommandBuffers = buffers;
info.commandBufferCount = count;
const VkResult result = vkQueueSubmit(queue, 1, &info, fence);
info.pCommandBuffers = own;
info.commandBufferCount = ownCount;
return result;
}
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -0,0 +1,95 @@
// MobileGL - MobileGL/MG_Backend/DirectVulkan/Renderer/GpuProgressMarkers.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#pragma once
#include "../VkIncludes.h"
#include <Includes.h>
#include <MG_Backend/GpuHangWatch.h>
#include <atomic>
#include <mutex>
namespace MobileGL::MG_Backend::DirectVulkan {
// WHICH OF THIS DEVICE'S SUBMISSIONS THE GPU HAS STARTED, AND WHICH IT HAS FINISHED
// (GpuHangWatch.h says why that is the whole attribution of a GPU hang).
//
// A frame's submission is bracketed by two prerecorded one-command buffers: the first sets a
// `started` event at TOP_OF_PIPE, the last sets a `finished` event once everything before it in
// the queue has completed. The host reads both with vkGetEventStatus from the watch thread - no
// fence is shared with the renderer's own waits and resets. On Adreno 750 a submission whose
// fragment shader never ends reads started=SET from its first poll until the kernel resets the
// GPU, and a submission queued behind it reads started=RESET for the whole stall.
//
// The renderer's thread brackets submissions (Acquire / Submitted / Abandon) and asks Hung(); the
// watch thread samples. Everything Vulkan the two share - the events - is touched under m_mutex.
class GpuProgressMarkers {
public:
struct Bracket {
VkCommandBuffer Begin = VK_NULL_HANDLE;
VkCommandBuffer End = VK_NULL_HANDLE;
Uint32 Slot = ~0u;
Bool Valid() const { return Slot != ~0u; }
};
GpuProgressMarkers() = default;
~GpuProgressMarkers() { Shutdown(); }
GpuProgressMarkers(const GpuProgressMarkers&) = delete;
GpuProgressMarkers& operator=(const GpuProgressMarkers&) = delete;
// Off (no brackets, never hung) when the budget is 0 or the device cannot make events.
void Init(VkDevice device, Uint32 queueFamilyIndex);
// Before the device goes; its queue must be idle. Stops the watch first.
void Shutdown();
// The two buffers to put first and last in one submission (an invalid bracket when off or
// when every slot is in flight: the submission then goes unwatched).
Bracket Acquire();
// The submission carrying `bracket` was accepted by vkQueueSubmit / was not submitted.
void Submitted(const Bracket& bracket);
void Abandon(const Bracket& bracket);
// The watch named one of this device's submissions as hanging the GPU.
Bool Hung() const { return m_hung.load(std::memory_order_acquire); }
Uint64 HungRunningMs() const { return m_hungRunningMs.load(std::memory_order_acquire); }
Bool Active() const { return m_watch != 0; }
// vkQueueSubmit of `info` with `bracket` wrapped around its command buffers (which it puts
// back before returning). At most 6 buffers of the caller's.
static VkResult SubmitBracketed(VkQueue queue, VkSubmitInfo& info, VkFence fence, const Bracket& bracket);
private:
struct Slot {
VkEvent Started = VK_NULL_HANDLE;
VkEvent Finished = VK_NULL_HANDLE;
VkCommandBuffer Begin = VK_NULL_HANDLE;
VkCommandBuffer End = VK_NULL_HANDLE;
Uint64 Id = 0;
Bool InFlight = false; // submitted and not yet seen finished
Bool Claimed = false; // handed out by Acquire, not yet submitted or abandoned
Bool Dirty = false; // its events were set by a past submission: reset before reuse
};
static constexpr Uint32 kMaxSlots = 16;
Bool AddSlot();
static Bool Sample(void* self, Vector<MG_Backend::GpuHangWatch::SubmissionProgress>& out);
static void OnHung(void* self, Uint64 submissionId, Uint64 runningMs);
std::mutex m_mutex;
VkDevice m_device = VK_NULL_HANDLE;
VkCommandPool m_pool = VK_NULL_HANDLE;
Vector<Slot> m_slots;
Uint64 m_nextId = 1;
Uint64 m_watch = 0;
Bool m_lost = false;
std::atomic<Bool> m_hung{false};
std::atomic<Uint64> m_hungRunningMs{0};
};
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -0,0 +1,37 @@
// MobileGL - MobileGL/MG_Backend/DirectVulkan/Renderer/OffscreenSurfaceRoute.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
// How an offscreen (pbuffer) surface target gets its VkSurfaceKHR on Android, decided from what the
// INSTANCE has, not from what the loader could offer. A renderer created for a window enabled
// VK_KHR_android_surface and nothing headless, so its later offscreen targets (a second client
// surface, or the placeholder a suspended server window draws into) must take the AImageReader
// route; asking that instance for vkCreateHeadlessSurfaceEXT answers null. Pure, so a host case pins
// it without a driver.
#pragma once
#include <Includes.h>
namespace MobileGL::MG_Backend::DirectVulkan {
enum class OffscreenSurfaceRoute : Uint8 {
Headless, // vkCreateHeadlessSurfaceEXT
AndroidImageReader, // vkCreateAndroidSurfaceKHR on an AImageReader's window nobody displays
Unavailable, // neither: the target cannot be built (the caller refuses, by name)
};
// `headlessEnabled`: the instance was created with VK_EXT_headless_surface.
// `headlessEntryPoint`: vkGetInstanceProcAddr answered vkCreateHeadlessSurfaceEXT.
// `androidSurfaceEnabled`: the instance was created with VK_KHR_android_surface.
constexpr OffscreenSurfaceRoute ChooseOffscreenSurfaceRoute(Bool headlessEnabled, Bool headlessEntryPoint,
Bool androidSurfaceEnabled) {
if (headlessEnabled && headlessEntryPoint) return OffscreenSurfaceRoute::Headless;
if (androidSurfaceEnabled) return OffscreenSurfaceRoute::AndroidImageReader;
return OffscreenSurfaceRoute::Unavailable;
}
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -391,6 +391,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkPipeline PipelineFactory::CreatePipeline(const PipelineCreatePayload& payload) const {
MOBILEGL_ASSERT(payload.stages != nullptr && !payload.stages->empty(), "PipelineFactory: stages are empty");
// The assert compiles out of shipping builds, and a program declined for an invalid
// module (VkProgramObject::invalidSpirvDeclined) has no stages at all: a caller that
// forgot to refuse it must get a skipped draw, not a stageCount=0 pipeline.
if (payload.stages == nullptr || payload.stages->empty()) {
MGLOG_E_ONCE("PipelineFactory::CreatePipeline: refusing a pipeline with no shader stages "
"(programHash=0x%llx); its draws are skipped",
static_cast<unsigned long long>(payload.programHash));
return VK_NULL_HANDLE;
}
MOBILEGL_ASSERT(payload.vertexInputState != nullptr, "PipelineFactory: vertexInputState is null");
MOBILEGL_ASSERT(payload.pipelineLayout != VK_NULL_HANDLE, "PipelineFactory: pipelineLayout is null");
MOBILEGL_ASSERT(payload.renderPass != VK_NULL_HANDLE, "PipelineFactory: renderPass is null");
@@ -199,7 +199,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
UnorderedMap<HashType, PipelineCacheEntry> m_cache;
// Monotonic frame-boundary counter (bumped in OnFrameBoundary) for cache aging.
Uint64 m_frameCounter = 0;
static inline XXH64_state_t* m_hashState = XXH64_createState();
static inline thread_local XXH64_state_t* m_hashState = XXH64_createState();
static inline Bool s_suppressBlendedDepthWrite = false;
};
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -12,7 +12,9 @@
#include "MG_Util/ShaderTranspiler/ShaderCompiler.h"
#include "MG_Util/ShaderTranspiler/SpvcSession.h"
#include "MG_Util/ShaderTranspiler/Types.h"
#include <Config.h>
#include <algorithm>
#include <atomic>
#include <bit>
#include <cmath>
#include <cstring>
@@ -1313,6 +1315,30 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
});
if (!ok) return false;
// An access chain into the builtin (gl_FragCoord.y is one) carries an Input
// pointer type. Based on the Private copy it has to carry a Private one, and so
// does every chain built on top of it - otherwise the module hands the driver a
// pointer whose storage class disagrees with its base, which is invalid SPIR-V
// and reads an undefined value. Plan every retype before touching anything, so a
// shape this pass does not model declines the whole rewrite.
Vector<Pair<Instruction*, Uint32>> retypes;
Vector<Instruction*> chains;
for (Instruction* user : users) {
if (IsAccessChain(user->opcode())) chains.push_back(user);
}
while (!chains.empty()) {
Instruction* chain = chains.back();
chains.pop_back();
const Uint32 privateType = PrivatePointerTypeFor(chain->type_id());
if (privateType == 0) return false;
retypes.emplace_back(chain, privateType);
context()->get_def_use_mgr()->ForEachUser(chain, [&](Instruction* next) {
if (IsAccessChain(next->opcode()) && next->NumInOperands() > 0 &&
next->GetSingleWordInOperand(0) == chain->result_id()) {
chains.push_back(next);
}
});
}
for (Instruction* user : users) {
for (Uint32 i = 0; i < user->NumInOperands(); ++i) {
auto& operand = user->GetInOperand(i);
@@ -1323,9 +1349,32 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
context()->AnalyzeUses(user);
}
for (auto& [chain, privateType] : retypes) {
chain->SetResultType(privateType);
context()->AnalyzeUses(chain);
}
return true;
}
static Bool IsAccessChain(spv::Op opcode) {
return opcode == spv::Op::OpAccessChain || opcode == spv::Op::OpInBoundsAccessChain ||
opcode == spv::Op::OpPtrAccessChain || opcode == spv::Op::OpInBoundsPtrAccessChain;
}
// The Private-storage twin of an Input pointer type, or 0 when `pointerTypeId` is not
// an Input pointer (then the chain is not the shape this pass rewrites).
Uint32 PrivatePointerTypeFor(Uint32 pointerTypeId) {
using namespace spvtools::opt;
Instruction* pointerType = context()->get_def_use_mgr()->GetDef(pointerTypeId);
if (!pointerType || pointerType->opcode() != spv::Op::OpTypePointer) return 0;
if (static_cast<spv::StorageClass>(pointerType->GetSingleWordInOperand(0)) !=
spv::StorageClass::Input) {
return 0;
}
return context()->get_type_mgr()->FindPointerToType(pointerType->GetSingleWordInOperand(1),
spv::StorageClass::Private);
}
Uint32 m_framebufferHeight = 0;
};
@@ -1861,35 +1910,38 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return spvtools::Optimizer::PassToken(MakeUnique<GlToVulkanPositionFixPass>(transformFlags));
}
Bool TransformSpirvForFragCoordYFlip(const Vector<Uint>& input, Vector<Uint>& output,
Uint32 framebufferHeight) {
if (input.empty()) {
output.clear();
return true;
}
if (framebufferHeight == 0) {
output = input;
return true;
}
} // namespace
spvtools::Optimizer optimizer(SPV_ENV_VULKAN_1_3);
spvtools::OptimizerOptions options;
options.set_run_validator(false); // see TransformSpirvForExplicitLod0Sampling
optimizer.SetMessageConsumer([](spv_message_level_t, const char*, const spv_position_t&,
const char* message) {
MGLOG_E("Vulkan: fragcoord y-flip pass: %s", message != nullptr ? message : "");
});
optimizer.RegisterPass(
spvtools::Optimizer::PassToken(MakeUnique<GlFragCoordYFlipPass>(framebufferHeight)));
const Bool success = optimizer.Run(input.data(), input.size(), &output, options);
if (!success) {
MGLOG_E("Vulkan: failed to run the gl_FragCoord y-flip pass; keeping the original module");
output = input;
}
return success;
Bool TransformSpirvForFragCoordYFlip(const Vector<Uint>& input, Vector<Uint>& output,
Uint32 framebufferHeight) {
if (input.empty()) {
output.clear();
return true;
}
if (framebufferHeight == 0) {
output = input;
return true;
}
spvtools::Optimizer optimizer(SPV_ENV_VULKAN_1_3);
spvtools::OptimizerOptions options;
options.set_run_validator(false); // see TransformSpirvForExplicitLod0Sampling
optimizer.SetMessageConsumer([](spv_message_level_t, const char*, const spv_position_t&,
const char* message) {
MGLOG_E("Vulkan: fragcoord y-flip pass: %s", message != nullptr ? message : "");
});
optimizer.RegisterPass(
spvtools::Optimizer::PassToken(MakeUnique<GlFragCoordYFlipPass>(framebufferHeight)));
const Bool success = optimizer.Run(input.data(), input.size(), &output, options);
if (!success) {
MGLOG_E("Vulkan: failed to run the gl_FragCoord y-flip pass; keeping the original module");
output = input;
}
return success;
}
namespace {
Bool TransformSpirvForXfbCapture(const Vector<Uint>& input, Vector<Uint>& output,
const MagmaProgramSource& program) {
if (input.empty()) {
@@ -2223,8 +2275,84 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return false;
}
}
// ScreenModuleForDriver's memory. Keyed by content, never by program: the same bytes get
// the same answer whichever program, variant or session carries them.
struct DriverModuleVerdictCache {
struct Entry {
Uint64 wordCount = 0;
Bool valid = true;
String reason;
};
// Generous for any one application's distinct stage modules; past it the whole table
// is dropped rather than tracked per entry - a re-validation is the only cost.
static constexpr SizeT kMaxEntries = 8192;
std::mutex mutex;
UnorderedMap<Uint64, Entry> entries;
std::atomic<Uint64> validatorRuns{0};
};
DriverModuleVerdictCache& GetDriverModuleVerdictCache() {
// Leaked on purpose: screened from session threads that may outlive static destruction.
static auto* cache = new DriverModuleVerdictCache();
return *cache;
}
} // namespace
ProgramFactory::DriverModuleVerdict ProgramFactory::ScreenModuleForDriver(const Vector<Uint>& spirv,
Bool allowInvalid, String* outReason) {
if (outReason != nullptr) outReason->clear();
if (spirv.empty()) return DriverModuleVerdict::Valid;
auto& cache = GetDriverModuleVerdictCache();
const Uint64 key = XXH64(spirv.data(), spirv.size() * sizeof(Uint), 0);
Bool valid = true;
String reason;
Bool known = false;
{
const std::lock_guard<std::mutex> lock(cache.mutex);
const auto it = cache.entries.find(key);
if (it != cache.entries.end() && it->second.wordCount == spirv.size()) {
known = true;
valid = it->second.valid;
reason = it->second.reason;
}
}
if (!known) {
// Outside the lock: two threads screening the same new module both validate, and
// agree; nothing waits on another session's spirv-val.
MG_Util::ShaderTranspiler::ShaderCompiler::PrepareSpirvValidation();
cache.validatorRuns.fetch_add(1, std::memory_order_relaxed);
spv_context context = spvContextCreate(GetSpirvTargetEnv(spirv));
spv_validator_options options = spvValidatorOptionsCreate();
spvValidatorOptionsSetFriendlyNames(options, true);
spv_const_binary_t binary = {spirv.data(), spirv.size()};
spv_diagnostic diagnostic = nullptr;
const spv_result_t result = spvValidateWithOptions(context, options, &binary, &diagnostic);
valid = result == SPV_SUCCESS;
if (!valid) {
reason = diagnostic != nullptr && diagnostic->error != nullptr ? diagnostic->error : "<no diagnostic>";
// The process-wide counter the validating test lanes assert on.
MG_Util::ShaderTranspiler::ShaderCompiler::NoteSpirvValidationFailure();
}
spvDiagnosticDestroy(diagnostic);
spvValidatorOptionsDestroy(options);
spvContextDestroy(context);
const std::lock_guard<std::mutex> lock(cache.mutex);
if (cache.entries.size() >= DriverModuleVerdictCache::kMaxEntries) cache.entries.clear();
cache.entries[key] = DriverModuleVerdictCache::Entry{spirv.size(), valid, reason};
}
if (valid) return DriverModuleVerdict::Valid;
if (outReason != nullptr) *outReason = Move(reason);
return allowInvalid ? DriverModuleVerdict::AdmittedInvalid : DriverModuleVerdict::DeclinedInvalid;
}
Uint64 ProgramFactory::DriverModuleValidatorRuns() {
return GetDriverModuleVerdictCache().validatorRuns.load(std::memory_order_relaxed);
}
// A shader that assigns gl_FragDepth (SPIR-V DepthReplacing) supplies depth itself
// instead of taking the pipeline's interpolated Z, so a driver that varies the vertex
// position math between pipelines cannot desynchronize it; the blended depth-write
@@ -2401,6 +2529,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return SamplerNumericDomain::UnsignedInteger;
case GL_SAMPLER_1D:
case GL_SAMPLER_2D:
case GL_SAMPLER_EXTERNAL_OES:
case GL_SAMPLER_3D:
case GL_SAMPLER_CUBE:
case GL_SAMPLER_2D_RECT:
@@ -2591,6 +2720,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
case GL_UNSIGNED_INT_IMAGE_CUBE_MAP_ARRAY:
return TextureTarget::TextureCubeMapArray;
case GL_SAMPLER_2D:
case GL_SAMPLER_EXTERNAL_OES: // a 2D texture below the frontend
case GL_SAMPLER_2D_SHADOW:
case GL_INT_SAMPLER_2D:
case GL_UNSIGNED_INT_SAMPLER_2D:
@@ -3764,27 +3894,42 @@ namespace MobileGL::MG_Backend::DirectVulkan {
const Bool remapOk = RemapDescriptorBindingsForVulkan(moduleSpirvs, m_maxBindings, moduleSpirvs);
MOBILEGL_ASSERT(remapOk, "ProgramFactory::GetOrCreateProgram: descriptor binding remap failed");
// Last look at the exact bytes the driver receives, in EVERY build rather than only in
// DEBUG or with MOBILEGL_ENABLE_SPIRV_VALIDATION armed, and for every stage BEFORE any
// module is created. By here the descriptor bindings have been remapped and the layout
// about to be reflected describes the remapped module, so there is no module left that
// is both valid and consistent with it to fall back to (the recovery that exists lives
// one step earlier, at the clip/XFB fixups). What is left is to keep the bytes away from
// the driver: a driver is not a validating entry point, and an invalid module it accepts
// can fault the GPU into device loss (a frontend translation once put a sampler inside
// the default uniform block, and Adreno page-faulted reading its "descriptor" out of
// uniform-buffer bytes). One invalid stage declines the whole program: no shader module
// of it is created, and every draw or dispatch of it is refused by name
// (VkProgramObject::invalidSpirvDeclined). MOBILEGL_MAGMA_ALLOW_INVALID_SPIRV hands the
// bytes over anyway, for diagnosing a driver that copes.
for (SizeT i = 0; i < moduleCount; ++i) {
String reason;
const DriverModuleVerdict verdict = ScreenModuleForDriver(
moduleSpirvs[i], MG_Config::Features.MagmaAllowInvalidSpirv, &reason);
if (verdict == DriverModuleVerdict::DeclinedInvalid) {
MGLOG_E_ONCE("ProgramFactory: declining program %u - its stage %d module fails SPIR-V validation (%s); "
"no shader module is created and its draws are refused. "
"MOBILEGL_MAGMA_ALLOW_INVALID_SPIRV=1 hands such modules to the driver anyway.",
program.GetExternalIndex(), static_cast<Int>(stages[i]), reason.c_str());
entry.invalidSpirvDeclined = true;
break;
}
if (verdict == DriverModuleVerdict::AdmittedInvalid) {
MGLOG_E_ONCE("ProgramFactory: handing vkCreateShaderModule an INVALID module for program %u stage %d "
"because MOBILEGL_MAGMA_ALLOW_INVALID_SPIRV is set (%s)",
program.GetExternalIndex(), static_cast<Int>(stages[i]), reason.c_str());
}
}
for (SizeT i = 0; i < moduleCount && !entry.invalidSpirvDeclined; ++i) {
auto& moduleSpv = moduleSpirvs[i];
if (moduleSpv.empty()) continue;
// Last look at the exact bytes the driver receives, in EVERY build rather than only
// in DEBUG or with MOBILEGL_ENABLE_SPIRV_VALIDATION armed. This one only reports:
// by here the descriptor bindings have been remapped and the layout about to be
// reflected describes the remapped module, so there is no module left that is both
// valid and consistent with it to fall back to. The recovery lives one step earlier,
// at the clip/XFB fixups (see the revert there) - which is where a transform can
// introduce a reference to a delisted interface variable, the failure this whole
// guard exists for. Anything that reaches this line names itself in the log of a
// shipping build instead of dying anonymously inside the driver.
SpirvValidationFailure finalFailure{};
if (!ValidateTransformedSpirv(moduleSpv, stages[i], program.GetExternalIndex(), &finalFailure)) {
MGLOG_E_ONCE("ProgramFactory: handing vkCreateShaderModule an INVALID module for program %u stage %d - "
"a backend transform after the clip/XFB fixups broke it (%s)",
program.GetExternalIndex(), static_cast<Int>(stages[i]),
finalFailure.message.c_str());
}
// Does the stage the driver will treat as the last pre-rasterization one actually
// carry Xfb? Asked of the FINAL bytes, so it answers for whatever the whole transform
// chain produced - a rewound clip/XFB backstop, a capture pass that resolved no
@@ -9,6 +9,7 @@
#pragma once
#include "../VkIncludes.h"
#include "../MagmaSession.h"
#include "PipelineFactory.h"
#include "MagmaProgramSource.h"
#include "MG_State/GLState/ProgramState/ShaderObject.h"
@@ -228,6 +229,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// the cache entry, so every later draw of the same program variant reads the same
// answer instead of re-deciding it.
Bool pointSizeCapabilityUnsupported = false;
// One of the program's final modules fails SPIR-V validation, so none of its modules
// was created and every draw or dispatch of it is refused by name (the ProgramInvalidSpirv /
// ComputeProgramInvalidSpirv wire declines) - see ProgramFactory::ScreenModuleForDriver.
// Sticky like the flag above: it lives on the cache entry, so the one log line is honest.
Bool invalidSpirvDeclined = false;
Bool needsPassthroughTessControl = false;
// ...and the pass-through this renderer can synthesize carries gl_Position and
// nothing else, so it is only correct when the evaluation stage's inputs are
@@ -251,7 +257,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// entry pointer re-stamps use through a const reference (StampProgramUse).
mutable Uint64 lastUsedFrame = 0;
static inline VkDevice s_device = VK_NULL_HANDLE;
static inline SessionLocal<VkDevice> s_device; // per Magma session (MagmaSession.h)
VkProgramObject() = default;
VkProgramObject(const VkProgramObject&) = delete;
@@ -288,6 +294,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
storageBlockIndexByBinding = std::move(other.storageBlockIndexByBinding);
hasStorageImages = other.hasStorageImages;
declinedDescriptors = other.declinedDescriptors;
// The three refusal flags travel like every other field: a robin-hood insert swaps
// entries, and a flag left behind would refuse (or admit) the wrong program.
invalidSpirvDeclined = other.invalidSpirvDeclined;
xfbCaptureDeclined = other.xfbCaptureDeclined;
pointSizeCapabilityUnsupported = other.pointSizeCapabilityUnsupported;
globalUboBinding = other.globalUboBinding;
activeVertexInputLocationMask = other.activeVertexInputLocationMask;
vertexInputTypes = other.vertexInputTypes;
@@ -309,6 +320,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
other.pipelineLayout = VK_NULL_HANDLE;
other.hasStorageImages = false;
other.declinedDescriptors = false;
other.invalidSpirvDeclined = false;
other.xfbCaptureDeclined = false;
other.pointSizeCapabilityUnsupported = false;
other.globalUboBinding = -1;
other.activeVertexInputLocationMask = 0;
other.activeFragmentOutputLocationMask = 0;
@@ -352,6 +366,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
storageBlockIndexByBinding = std::move(other.storageBlockIndexByBinding);
hasStorageImages = other.hasStorageImages;
declinedDescriptors = other.declinedDescriptors;
// The three refusal flags travel like every other field: a robin-hood insert swaps
// entries, and a flag left behind would refuse (or admit) the wrong program.
invalidSpirvDeclined = other.invalidSpirvDeclined;
xfbCaptureDeclined = other.xfbCaptureDeclined;
pointSizeCapabilityUnsupported = other.pointSizeCapabilityUnsupported;
globalUboBinding = other.globalUboBinding;
activeVertexInputLocationMask = other.activeVertexInputLocationMask;
vertexInputTypes = other.vertexInputTypes;
@@ -373,6 +392,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
other.pipelineLayout = VK_NULL_HANDLE;
other.hasStorageImages = false;
other.declinedDescriptors = false;
other.invalidSpirvDeclined = false;
other.xfbCaptureDeclined = false;
other.pointSizeCapabilityUnsupported = false;
other.globalUboBinding = -1;
other.activeVertexInputLocationMask = 0;
other.activeFragmentOutputLocationMask = 0;
@@ -395,18 +417,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
private:
void Destroy() {
if (s_device != VK_NULL_HANDLE) {
if (s_device.Get() != VK_NULL_HANDLE) {
if (pipelineLayout != VK_NULL_HANDLE) {
vkDestroyPipelineLayout(s_device, pipelineLayout, nullptr);
vkDestroyPipelineLayout(s_device.Get(), pipelineLayout, nullptr);
pipelineLayout = VK_NULL_HANDLE;
}
if (descriptorSetLayout != VK_NULL_HANDLE) {
vkDestroyDescriptorSetLayout(s_device, descriptorSetLayout, nullptr);
vkDestroyDescriptorSetLayout(s_device.Get(), descriptorSetLayout, nullptr);
descriptorSetLayout = VK_NULL_HANDLE;
}
for (auto module : modules) {
if (module != VK_NULL_HANDLE) {
vkDestroyShaderModule(s_device, module, nullptr);
vkDestroyShaderModule(s_device.Get(), module, nullptr);
}
}
}
@@ -463,7 +485,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
m_enableSpirvValidation(enableSpirvValidation),
m_updateAfterBindLimits(updateAfterBindLimits),
m_subgroupPolicy(subgroupPolicy) {
VkProgramObject::s_device = device;
VkProgramObject::s_device.Get() = device;
}
// Destroys the pass-through tessellation control modules. Runs while the device is
// still alive for the same reason ~VkProgramObject's does: this factory outlives
@@ -503,6 +525,29 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// VkRenderPassManager::OnPresent's sweep.
void OnFrameBoundary();
// What the last look at a module's final bytes - the exact words vkCreateShaderModule
// would receive - decided.
enum class DriverModuleVerdict : Uint8 {
Valid, // validates: handed to the driver
DeclinedInvalid, // fails validation: never handed over, the program's work is declined
AdmittedInvalid, // fails validation, handed over anyway (MOBILEGL_MAGMA_ALLOW_INVALID_SPIRV)
};
// Validates `spirv` for the Vulkan environment its version word names and turns the
// result into a verdict. A driver is not a validating entry point: Adreno page-faulted
// into device loss on a module that carried a sampler inside a uniform block, Mali
// SIGSEGVs inside pipeline creation. So an invalid module is declined unless
// `allowInvalid` (the escape hatch) says to hand it over regardless.
//
// The validator's answer is remembered per module content (64-bit hash plus word count),
// process-wide: a program variant, an evicted-then-rebuilt entry and two programs that
// share a stage all re-screen the same bytes, and only the first pays for spirv-val.
// `outReason` receives the validator's message for an invalid module.
static DriverModuleVerdict ScreenModuleForDriver(const Vector<Uint>& spirv, Bool allowInvalid,
String* outReason = nullptr);
// How many times ScreenModuleForDriver actually ran the validator (cache misses), for
// the tests that pin the cache.
static Uint64 DriverModuleValidatorRuns();
static VkShaderStageFlagBits ToVkStage(ShaderStage stage);
static VkFormat ConvertSpirvImageFormatToVkFormat(SpvImageFormat format);
static SamplerNumericDomain UniformTypeToSamplerNumericDomain(GLenum glType);
@@ -655,6 +700,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// and adds only a handful of values, so it does not move the cap in practice.
static constexpr SizeT kMaxPassthroughTessControlStages = 64;
UnorderedMap<Uint64, VkPipelineShaderStageCreateInfo> m_passthroughTessControlStages;
static inline XXH64_state_t* m_hashState = XXH64_createState();
static inline thread_local XXH64_state_t* m_hashState = XXH64_createState();
};
// The FragCoordYFlip variant's module transform: every read of gl_FragCoord goes through a
// Private copy whose y is (framebufferHeight - y). Fails open (output = input, false). Exposed
// so a unit case can validate the bytes it hands the driver.
Bool TransformSpirvForFragCoordYFlip(const Vector<Uint>& input, Vector<Uint>& output, Uint32 framebufferHeight);
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -0,0 +1,27 @@
// MobileGL - MobileGL/MG_Backend/DirectVulkan/Renderer/RenderPassGuard.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#pragma once
#include "../VkIncludes.h"
#include <Includes.h>
namespace MobileGL::MG_Backend::DirectVulkan {
// Ends the render pass that is active on `commandBuffer`, if there is one. Called right before
// recording a command a render pass may not contain (a barrier, a copy, a blit, a clear of an
// image) into a command buffer that may hold an open pass: the wire draw path keeps its pass
// open across consecutive draws to the same attachments, so the resource work that falls
// between two draws must close it first. A pass on another command buffer is left alone.
void EndActiveRenderPassOn(VkCommandBuffer commandBuffer);
// A counter that moves whenever images may have been written since it was last read: a render
// pass ended (its attachments), resource work recorded outside a pass (copies, clears, blits,
// dispatches all close the pass first), or a draw with storage images. A sampled image needs
// the memory barrier that makes earlier writes visible only when this moved since the last one.
Uint64 WireImageWriteEpoch();
void BumpWireImageWriteEpoch();
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -292,6 +292,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// Fresh swapchain images hold garbage until a render pass stores into them.
m_imageContentDefined.assign(imageCount, false);
m_depthStencilContentDefined.assign(imageCount, false);
// Nothing presented from this chain yet: every buffer age is unknown.
m_ages.Reset(imageCount);
CreateImageViews(device);
CreateDepthStencilResources(device, physicalDevice);
@@ -299,6 +301,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
MGLOG_I("Swapchain created, extent = %dx%d, swapchain imageCount = %d", m_extent.width, m_extent.height,
imageCount);
PublishDefaultFramebufferInfo();
}
void SwapchainObject::PublishDefaultFramebufferInfo() const {
const VkExtent2D defaultFramebufferExtent = m_surfaceExtent;
// Properly initialize Default FBO here
auto& defaultFBOInfo = MG_Impl::GLImpl::FramebufferImpl::pDefaultFramebufferInfo;
const Int extentWidth = static_cast<Int>(defaultFramebufferExtent.width);
@@ -488,6 +495,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
m_imageLayouts.clear();
m_imageContentDefined.clear();
m_depthStencilContentDefined.clear();
m_ages.Reset(0);
m_preTransform = VK_SURFACE_TRANSFORM_IDENTITY_BIT_KHR;
}
@@ -9,6 +9,7 @@
#pragma once
#include <Includes.h>
#include <MG_Util/Damage/Damage.h>
#include "../VkIncludes.h"
namespace MobileGL::MG_Backend::DirectVulkan {
@@ -38,6 +39,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Uint32 presentQueueFamily, Uint32 minImageCountHint, VkExtent2D desiredExtent,
Optional<Int> swapInterval);
void Shutdown(VkDevice device);
// Tells the frontend (or, under a transport, the client) the default framebuffer's size
// and formats. Create does this; a renderer switching between surfaces repeats it.
void PublishDefaultFramebufferInfo() const;
VkSwapchainKHR GetHandle() const { return m_swapchain; }
VkPresentModeKHR GetPresentMode() const { return m_presentMode; }
@@ -77,6 +81,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
void SetDepthStencilContentDefined(Uint32 index, Bool defined);
void SetAllDepthStencilContentUndefined();
// EGL_EXT_buffer_age over this chain's images. Ages start unknown with the chain (Create).
// Until a client asks for an age (NoteBufferAgeAsked) a presented image's content is
// treated as undefined, as before; once one has, a presented image keeps it - its age is
// counted and its content stays defined, so the next frame loads it.
void NoteBufferAgeAsked() { m_bufferAgeAsked = true; }
Bool BufferAgeAsked() const { return m_bufferAgeAsked; }
Int32 BufferAgeOf(Uint32 index) const { return m_ages.AgeOf(index); }
// Image `index` was presented.
void OnImagePresented(Uint32 index) { m_ages.OnPresented(index, m_bufferAgeAsked); }
// A swapped frame reached no image: every age is unknown.
void OnFrameDropped() { m_ages.OnFrameDropped(); }
private:
void CreateImageViews(VkDevice device);
void CreateDepthStencilResources(VkDevice device, VkPhysicalDevice physicalDevice);
@@ -106,5 +122,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Vector<VkImageLayout> m_depthStencilImageLayouts;
Vector<Bool> m_imageContentDefined;
Vector<Bool> m_depthStencilContentDefined;
MG_Util::Damage::SwapchainAgeTracker m_ages;
// Kept across a rebuild: it is the application's habit, not the chain's.
Bool m_bufferAgeAsked = false;
};
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -6,6 +6,7 @@
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#include "RenderPassGuard.h"
#include "UniformManager.h"
#include "MG_Backend/DirectVulkan/DirectVulkanResourceState.h"
@@ -251,18 +252,26 @@ namespace MobileGL::MG_Backend::DirectVulkan {
WireDescriptorFatal("image-descriptor-transition");
// A second descriptor/draw can keep GENERAL after a shader write. A layout
// equality fast return is not a memory dependency for that prior write.
VkMemoryBarrier memory{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
memory.srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT;
memory.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
vkCmdPipelineBarrier(commandBuffer, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
0, 1, &memory, 0, nullptr, 0, nullptr);
// Only when something may have written an image since the last such barrier
// (WireImageWriteEpoch): with nothing written in between, consecutive draws keep their
// render pass open instead of closing it for a barrier that orders nothing.
if (WireImageWriteEpoch() != m_wireImageBarrierEpoch) {
VkMemoryBarrier memory{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
memory.srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT;
memory.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
EndActiveRenderPassOn(commandBuffer);
vkCmdPipelineBarrier(commandBuffer, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
0, 1, &memory, 0, nullptr, 0, nullptr);
m_wireImageBarrierEpoch = WireImageWriteEpoch();
}
VkImageViewCreateInfo info{VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO};
info.image = resource->image;
info.viewType = type;
info.format = format;
info.subresourceRange = {aspect, level, levels, layer, layers};
if (!storage) {
const Bool alphaIsOne = ResolveTextureFormatInfo(
// An X-format shared image carries undefined alpha bytes, exactly like an expanded RGB.
const Bool alphaIsOne = resource->sharedImageAlphaOne || ResolveTextureFormatInfo(
static_cast<TextureInternalFormat>(record.Desc.InternalFormat)).expandRgbToRgba;
const auto swizzle = [&](Uint32 channel) {
const auto value = record.Params.Swizzle[channel];
@@ -293,11 +302,22 @@ namespace MobileGL::MG_Backend::DirectVulkan {
} else {
auto samplerHandle = state.BoundSamplerStates[unit];
if (MG_Pipe::MGPipeHandleIsNull(samplerHandle)) samplerHandle = record.Params.BuiltinSampler;
if (MG_Pipe::MGPipeHandleIsNull(samplerHandle) || samplerHandle.Slot >= state.SamplerCsos.size())
WireDescriptorFatal("sampler-record");
const auto& sampler = state.SamplerCsos[samplerHandle.Slot];
if (!sampler.Live || sampler.Gen != samplerHandle.Gen) WireDescriptorFatal("sampler-record-generation");
out.sampler = m_samplerManager->GetOrCreateSamplerFromParameters(sampler.Params,
// A texture's built-in sampler CSO can be missing or gone while the texture still names
// it: sampler CSOs are content-addressed and recycled at capacity, and the texture's
// record is only re-pointed by its next set_texture_params (a texture whose params never
// crossed names none at all). Sample with GL's default parameters rather than end the
// session - the DirectGLES arm declines the push in both cases.
using SamplerCsoRecord = std::decay_t<decltype(state.SamplerCsos[0])>;
static const decltype(SamplerCsoRecord::Params) kDefaultSamplerParams{};
const Bool known = !MG_Pipe::MGPipeHandleIsNull(samplerHandle) &&
samplerHandle.Slot < state.SamplerCsos.size();
const SamplerCsoRecord* sampler = known ? &state.SamplerCsos[samplerHandle.Slot] : nullptr;
const Bool stale = sampler == nullptr || !sampler->Live || sampler->Gen != samplerHandle.Gen;
if (stale) {
MGLOG_E_ONCE("Magma wire: sampler CSO {%u, %u} named by texture unit %u is missing or no longer "
"live; sampling with default parameters", samplerHandle.Slot, samplerHandle.Gen, unit);
}
out.sampler = m_samplerManager->GetOrCreateSamplerFromParameters(stale ? kDefaultSamplerParams : sampler->Params,
static_cast<TextureInternalFormat>(record.Desc.InternalFormat),
domain == SamplerNumericDomain::SignedInteger || domain == SamplerNumericDomain::UnsignedInteger,
levels);
@@ -3183,7 +3203,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// preparing the texture would decline a storage-less one and lose the pass.
if (storage && WireShaderImageNamesNoTexel(*m_textureManager, state.BoundShaderImages[unit]))
continue;
if (!m_textureManager->SyncTextureResourceByHandle(handle, false, storage)) return false;
const auto* resource = m_textureManager->SyncTextureResourceByHandle(handle, false, storage);
if (resource == nullptr) return false;
// Another session writes it: the renderer acquires it before this use.
if (resource->sharedImageId != 0 || resource->yuvImageId != 0)
m_textureManager->NoteSharedImageUse(*resource);
}
}
}
@@ -660,5 +660,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
m_samplerResolveMemoHighWater = binding + 1;
}
}
// WireImageWriteEpoch() when ResolveWireImageDescriptor last made image writes visible.
mutable Uint64 m_wireImageBarrierEpoch = 0;
};
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -19,6 +19,7 @@
#include "VertexInputStateBuilder.h"
#include "MG_State/GLState/VertexArrayState/VertexArrayObject.h"
#include <Includes.h>
#include <atomic>
#include "../VkIncludes.h"
namespace MobileGL::MG_Backend::DirectVulkan {
@@ -217,11 +218,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// memo table is owned by this factory and dies with it, so a per-instance counter is
// enough there and the epoch shrinks back to what it looks like it should be.
#if MOBILEGL_PIPE_LEGACY_MEMOS
static inline Uint64 s_evictionEpochSource = 0;
static inline std::atomic<Uint64> s_evictionEpochSource{0};
Uint64 m_evictionEpoch = ++s_evictionEpochSource;
#else
Uint64 m_evictionEpoch = 1;
#endif
static inline XXH64_state_t* m_hashState = XXH64_createState();
static inline thread_local XXH64_state_t* m_hashState = XXH64_createState();
};
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -6,6 +6,7 @@
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#include "RenderPassGuard.h"
#include "VkBufferManager.h"
#include "../DirectVulkan.h"
#include "VulkanRenderer.h"
@@ -113,30 +114,30 @@ namespace MobileGL::MG_Backend::DirectVulkan {
using MG_State::GLState::BufferObject;
// The manager owned by the active VulkanRenderer; immediate ops route here.
VkBufferManager* g_activeBufferManager = nullptr;
SessionLocal<VkBufferManager*> g_activeBufferManager; // per Magma session (MagmaSession.h)
void Ops_Respecify(BufferObject& bufferObject) {
if (g_activeBufferManager) {
g_activeBufferManager->OnRespecify(bufferObject);
if (g_activeBufferManager.Get()) {
g_activeBufferManager.Get()->OnRespecify(bufferObject);
}
}
void Ops_SubData(BufferObject& bufferObject, SizeT offset, SizeT size) {
if (g_activeBufferManager) {
g_activeBufferManager->OnSubData(bufferObject, offset, size);
if (g_activeBufferManager.Get()) {
g_activeBufferManager.Get()->OnSubData(bufferObject, offset, size);
}
}
void Ops_ResidentSubData(BufferObject& bufferObject, SizeT offset, DataPtr data) {
if (g_activeBufferManager) {
g_activeBufferManager->OnResidentSubData(bufferObject, offset, data);
if (g_activeBufferManager.Get()) {
g_activeBufferManager.Get()->OnResidentSubData(bufferObject, offset, data);
}
}
void Ops_FlushMappedRange(BufferObject& bufferObject, Range1D range,
Flags<BufferMappingAccessBit> appAccess) {
if (g_activeBufferManager) {
g_activeBufferManager->OnFlushMappedRange(bufferObject, range, appAccess);
if (g_activeBufferManager.Get()) {
g_activeBufferManager.Get()->OnFlushMappedRange(bufferObject, range, appAccess);
}
}
@@ -154,15 +155,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
void* Ops_AcquirePersistentMap(BufferObject& bufferObject) {
if (g_activeBufferManager) {
return g_activeBufferManager->AcquirePersistentMap(bufferObject);
if (g_activeBufferManager.Get()) {
return g_activeBufferManager.Get()->AcquirePersistentMap(bufferObject);
}
return nullptr;
}
void Ops_OnDestroy(SharedPtr<BackendBufferResource>&& resource) {
if (g_activeBufferManager) {
g_activeBufferManager->OnResourceDestroyed(std::move(resource));
if (g_activeBufferManager.Get()) {
g_activeBufferManager.Get()->OnResourceDestroyed(std::move(resource));
}
// No active manager: the device/allocator is gone or going away and
// Shutdown() already destroyed the storage; dropping the handle here
@@ -181,34 +182,39 @@ namespace MobileGL::MG_Backend::DirectVulkan {
};
#if MOBILEGL_BUILD_DISAGGREGATED
VkBufferManager& WireManager() {
if (g_activeBufferManager == nullptr) {
MGLOG_F("Magma: Fatal{ResourceUnavailable, \"wire-buffer-manager\"}");
std::abort();
// The session's manager, or null once its renderer is gone - a client's records can still
// arrive after its eglTerminate released it (or, surfaceless, before its first surface built
// one). Such a record is declined with one line rather than taking the whole server down.
VkBufferManager* WireManager(const char* op) {
VkBufferManager* manager = g_activeBufferManager.Get();
if (manager == nullptr) {
MGLOG_E_ONCE("Magma: a wire buffer %s reached a session with no renderer; declined", op);
}
return *g_activeBufferManager;
return manager;
}
const MG_Pipe::MGPipeResourceOps g_vulkanWireResourceOps = {
.Create = [](auto res, const auto& desc) { WireManager().CreateWireBuffer(res, desc); },
.Create = [](auto res, const auto& desc) { if (auto* m = WireManager("create")) m->CreateWireBuffer(res, desc); },
.Respecify = [](auto res, const auto& desc, const void* bytes) {
WireManager().RespecifyWireBuffer(res, desc, bytes);
if (auto* m = WireManager("respecify")) m->RespecifyWireBuffer(res, desc, bytes);
},
.SubData = [](auto res, const auto& record, const void* bytes) {
WireManager().WriteWireBuffer(res, MG_Pipe::MGPipeSubDataBufferOffset(record),
MG_Pipe::MGPipeSubDataBufferSize(record), bytes);
if (auto* m = WireManager("write"))
m->WriteWireBuffer(res, MG_Pipe::MGPipeSubDataBufferOffset(record),
MG_Pipe::MGPipeSubDataBufferSize(record), bytes);
},
.SubDataResident = [](auto res, const auto& record, const void* bytes) {
WireManager().WriteWireBuffer(res, MG_Pipe::MGPipeSubDataBufferOffset(record),
MG_Pipe::MGPipeSubDataBufferSize(record), bytes);
if (auto* m = WireManager("write"))
m->WriteWireBuffer(res, MG_Pipe::MGPipeSubDataBufferOffset(record),
MG_Pipe::MGPipeSubDataBufferSize(record), bytes);
},
.FlushRange = [](auto res, const auto& record, const void* bytes) {
WireManager().FlushWireBuffer(res, record.Offset, record.Size, bytes);
if (auto* m = WireManager("flush")) m->FlushWireBuffer(res, record.Offset, record.Size, bytes);
},
.Readback = [](auto res, const auto& record) {
WireManager().ReadbackWireBuffer(res, record.Offset, record.Size);
if (auto* m = WireManager("readback")) m->ReadbackWireBuffer(res, record.Offset, record.Size);
},
.Destroy = [](auto res) { WireManager().DestroyWireBuffer(res); },
.Destroy = [](auto res) { if (auto* m = WireManager("destroy")) m->DestroyWireBuffer(res); },
// Split mapping remains T2: the client owns its map and pushes exact
// modified ranges. A server pointer is never donated across the wire.
.MapPersistent = [](MG_Pipe::MGPipeHandle, Uint64, const void*) -> void* { return nullptr; },
@@ -217,7 +223,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// P11 B2 (T0): the one door a server pointer is still never donated through stays shut;
// what T0 adds is the other direction - the CLIENT's pages imported as the store.
.ImportExternal = [](MG_Pipe::MGPipeHandle res, void* ahb, Uint64 size) -> Bool {
return WireManager().ImportWireBuffer(res, ahb, size);
auto* m = WireManager("import");
return m != nullptr && m->ImportWireBuffer(res, ahb, size);
},
.SelfTestExternal = [](char* why, Uint64 whyBytes) -> Bool {
return VkBufferManager::SelfTestWireImport(why, whyBytes);
@@ -234,6 +241,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
}
namespace {
// Diagnostic memory for DescribeWireBufferHistory: the last handles created and destroyed.
struct WireBufferHistory {
static constexpr SizeT kDepth = 256;
MG_Pipe::MGPipeHandle created[kDepth]{};
MG_Pipe::MGPipeHandle destroyed[kDepth]{};
SizeT createdCount = 0;
SizeT destroyedCount = 0;
};
SessionLocal<WireBufferHistory> g_wireBufferHistory;
} // namespace
VkBufferManager::WireBufferResource* VkBufferManager::FindWireBuffer(MG_Pipe::MGPipeHandle res) {
const auto found = m_wireBuffers.find(WireBufferKey(res));
return found == m_wireBuffers.end() ? nullptr : &found->second;
@@ -246,6 +265,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
std::abort();
}
m_wireBuffers.try_emplace(WireBufferKey(res));
{
auto& history = *g_wireBufferHistory;
history.created[history.createdCount++ % 256] = res;
}
RespecifyWireBuffer(res, desc, nullptr);
}
@@ -315,6 +338,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkMemoryBarrier barrier{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
barrier.srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT;
barrier.dstAccessMask = VK_ACCESS_HOST_READ_BIT | VK_ACCESS_HOST_WRITE_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_HOST_BIT,
0, 1, &barrier, 0, nullptr, 0, nullptr);
// P8-D: counted here, past the idle early-out, so `whw` is waits TAKEN, not calls.
@@ -403,7 +427,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
uploaded = StagedWireRangeCopy(*resource, bytes, static_cast<SizeT>(offset),
static_cast<SizeT>(size));
}
if (!uploaded && !WaitForWireBufferHostAccess(*resource)) WireBufferSyncFatal("host-write");
if (!uploaded && !WaitForWireBufferHostAccess(*resource)) {
if (LatchedOnDeviceLoss("buffer-host-write")) return;
WireBufferSyncFatal("host-write");
}
}
if (!uploaded) {
uploaded = resource->buffer.Upload(bytes, size, offset);
@@ -412,6 +439,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
}
if (!uploaded) {
if (LatchedOnDeviceLoss("buffer-upload")) return;
MGLOG_F("Magma: Fatal{ResourceUnavailable, \"buffer-upload\"}");
std::abort();
}
@@ -505,13 +533,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkMemoryBarrier before{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
before.srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT | VK_ACCESS_HOST_WRITE_BIT;
before.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT | VK_PIPELINE_STAGE_HOST_BIT,
VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 1, &before, 0, nullptr, 0, nullptr);
const VkBufferCopy copy{offset, dst.offset, size};
EndActiveRenderPassOn(commands);
vkCmdCopyBuffer(commands, resource->buffer.GetHandle(), dst.buffer, 1, &copy);
VkMemoryBarrier after{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
after.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
after.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
0, 1, &after, 0, nullptr, 0, nullptr);
resource->lastUseSerial = m_frameSerial;
@@ -549,22 +580,27 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkMemoryBarrier before{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
before.srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT | VK_ACCESS_HOST_WRITE_BIT;
before.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT | VK_PIPELINE_STAGE_HOST_BIT,
VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 1, &before, 0, nullptr, 0, nullptr);
const VkBufferCopy widen{alignedOffset, staging.offset, stagedSize};
EndActiveRenderPassOn(commands);
vkCmdCopyBuffer(commands, resource->buffer.GetHandle(), staging.buffer, 1, &widen);
// The shift reads what the widening copy just wrote, so it needs its own edge even
// though both halves are transfers on one command buffer.
VkMemoryBarrier staged{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
staged.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
staged.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
0, 1, &staged, 0, nullptr, 0, nullptr);
const VkBufferCopy shift{staging.offset + headPad, dst.offset + dstSkip, size};
EndActiveRenderPassOn(commands);
vkCmdCopyBuffer(commands, staging.buffer, dst.buffer, 1, &shift);
VkMemoryBarrier after{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
after.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
after.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
0, 1, &after, 0, nullptr, 0, nullptr);
resource->lastUseSerial = m_frameSerial;
@@ -582,6 +618,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
const Bool t0 = resource != nullptr && resource->imported;
Vector<Uint8> bytes(t0 ? 0 : static_cast<SizeT>(size));
if (t0 ? !WaitForWireBufferHostAccess(*resource) : !ReadWireBuffer(res, offset, size, bytes.data())) {
// A lost device: the session is latched and closing; its client reads a lost context.
if (LatchedOnDeviceLoss("buffer-readback")) return;
MGLOG_F("Magma: Fatal{ResourceUnavailable, \"buffer-readback\"} {slot=%u, gen=%u}", res.Slot, res.Gen);
std::abort();
}
@@ -880,14 +918,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkMemoryBarrier hostWrites{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
hostWrites.srcAccessMask = VK_ACCESS_HOST_WRITE_BIT;
hostWrites.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 1,
&hostWrites, 0, nullptr, 0, nullptr);
VkBufferCopy region{0, 0, kWords * 4};
EndActiveRenderPassOn(commands);
vkCmdCopyBuffer(commands, imported.buffer, staging.buffer, 1, &region);
EndActiveRenderPassOn(commands);
vkCmdFillBuffer(commands, imported.buffer, kFillAt, kWords * 4, kFillWord);
VkMemoryBarrier toHost{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
toHost.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
toHost.dstAccessMask = VK_ACCESS_HOST_READ_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, 0, 1, &toHost,
0, nullptr, 0, nullptr);
vkEndCommandBuffer(commands);
@@ -937,11 +979,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
verdict = "steady: the second submission could not be recorded";
break;
}
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 1,
&hostWrites, 0, nullptr, 0, nullptr);
VkBufferCopy steadyRegion{kSteadyAt, 0, kWords * 4};
EndActiveRenderPassOn(commands);
vkCmdCopyBuffer(commands, imported.buffer, staging.buffer, 1, &steadyRegion);
EndActiveRenderPassOn(commands);
vkCmdFillBuffer(commands, imported.buffer, kFillAt, kWords * 4, kSteadyFillWord);
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, 0, 1, &toHost,
0, nullptr, 0, nullptr);
vkEndCommandBuffer(commands);
@@ -972,7 +1018,29 @@ namespace MobileGL::MG_Backend::DirectVulkan {
#endif
}
void VkBufferManager::DescribeWireBufferHistory(MG_Pipe::MGPipeHandle res) const {
const auto& history = *g_wireBufferHistory;
Int createdAgo = -1, destroyedAgo = -1;
for (SizeT i = 0; i < std::min(history.createdCount, WireBufferHistory::kDepth); ++i) {
const auto& h = history.created[(history.createdCount - 1 - i) % WireBufferHistory::kDepth];
if (h.Slot == res.Slot && h.Gen == res.Gen) { createdAgo = static_cast<Int>(i); break; }
}
for (SizeT i = 0; i < std::min(history.destroyedCount, WireBufferHistory::kDepth); ++i) {
const auto& h = history.destroyed[(history.destroyedCount - 1 - i) % WireBufferHistory::kDepth];
if (h.Slot == res.Slot && h.Gen == res.Gen) { destroyedAgo = static_cast<Int>(i); break; }
}
MGLOG_E("Magma wire buffer {slot=%u, gen=%u}: live stores %zu; created %d creates ago, destroyed %d "
"destroys ago (-1 = not in the last %zu); %zu creates, %zu destroys in this session",
res.Slot, res.Gen, m_wireBuffers.size(), createdAgo, destroyedAgo, WireBufferHistory::kDepth,
history.createdCount, history.destroyedCount);
}
void VkBufferManager::DestroyWireBuffer(MG_Pipe::MGPipeHandle res) {
{
auto& history = *g_wireBufferHistory;
history.destroyed[history.destroyedCount++ % WireBufferHistory::kDepth] = res;
}
const auto found = m_wireBuffers.find(WireBufferKey(res));
if (found == m_wireBuffers.end()) return;
DeferWireRelease(std::move(found->second.buffer),
@@ -1030,6 +1098,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// budget's worth of orphans, not one per glBufferData.
if (pVulkanRenderer == nullptr ||
!pVulkanRenderer->WaitForSubmitIndex(pVulkanRenderer->GetSyncPointSubmitIndex(), UINT64_MAX, true)) {
if (LatchedOnDeviceLoss("buffer-deferred-watermark")) return;
WireBufferSyncFatal("deferred-watermark");
}
++m_wireDeferredSyncs;
@@ -1050,6 +1119,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
MG_Util::PipeStats::PublishGauge(Gauge::WireDeferredSyncs, m_wireDeferredSyncs);
}
Bool VkBufferManager::LatchedOnDeviceLoss(const char* site) {
return pVulkanRenderer != nullptr && pVulkanRenderer->LatchWireDeviceLoss(site);
}
void VkBufferManager::WireBufferSyncFatal(const char* site) {
MGLOG_F("Magma: Fatal{ResourceUnavailable, \"buffer-write-sync\"} {site=%s}", site);
std::abort();
@@ -1123,7 +1196,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
if (!InitializeTransientArenas()) {
return false;
}
g_activeBufferManager = this;
g_activeBufferManager.Get() = this;
MG_State::GLState::SetBufferBackendOps(&g_vulkanBufferBackendOps);
#if MOBILEGL_BUILD_DISAGGREGATED
RegisterWireResourceOps();
@@ -1132,10 +1205,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
void VkBufferManager::Shutdown() {
if (g_activeBufferManager == this) {
g_activeBufferManager = nullptr;
if (g_activeBufferManager.Get() == this) {
g_activeBufferManager.Get() = nullptr;
#if MOBILEGL_BUILD_DISAGGREGATED
if (MG_Pipe::MGPipeGetResourceOps() == &g_vulkanWireResourceOps) {
// Under a transport the table is the PROCESS's and serves every session: it dispatches
// to the calling session's own manager (WireManager), so one session's renderer going
// away must not take it from the others - their buffer creates would be dropped from
// then on, and every buffer born after that would be a handle the server never saw.
if (MG_Config::Transport == MG_Config::TransportMode::Monolith &&
MG_Pipe::MGPipeGetResourceOps() == &g_vulkanWireResourceOps) {
MG_Pipe::MGPipeSetResourceOps(nullptr);
}
#endif
@@ -1428,6 +1506,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkMemoryBarrier beforeBarrier{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
beforeBarrier.srcAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
beforeBarrier.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
EndActiveRenderPassOn(commandBuffer);
vkCmdPipelineBarrier(commandBuffer, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 1,
&beforeBarrier, 0, nullptr, 0, nullptr);
@@ -1435,11 +1514,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
region.srcOffset = staging.offset;
region.dstOffset = static_cast<VkDeviceSize>(offset);
region.size = static_cast<VkDeviceSize>(size);
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyBuffer(commandBuffer, staging.buffer, resource.buffer.GetHandle(), 1, &region);
VkMemoryBarrier afterBarrier{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
afterBarrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
afterBarrier.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
EndActiveRenderPassOn(commandBuffer);
vkCmdPipelineBarrier(commandBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, 1,
&afterBarrier, 0, nullptr, 0, nullptr);
@@ -1474,6 +1555,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkMemoryBarrier beforeBarrier{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
beforeBarrier.srcAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
beforeBarrier.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
EndActiveRenderPassOn(commandBuffer);
vkCmdPipelineBarrier(commandBuffer, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 1,
&beforeBarrier, 0, nullptr, 0, nullptr);
@@ -1481,11 +1563,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
region.srcOffset = staging.offset;
region.dstOffset = static_cast<VkDeviceSize>(offset);
region.size = static_cast<VkDeviceSize>(size);
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyBuffer(commandBuffer, staging.buffer, resource.buffer.GetHandle(), 1, &region);
VkMemoryBarrier afterBarrier{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
afterBarrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
afterBarrier.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
EndActiveRenderPassOn(commandBuffer);
vkCmdPipelineBarrier(commandBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, 1,
&afterBarrier, 0, nullptr, 0, nullptr);
@@ -110,6 +110,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// frontend BufferObject. Acquires expose the full GPU store, without a CPU
// pointer: CPU consumers must use ReadWireBuffer for ordered, current bytes.
Bool AcquireWireSlice(BufferKind kind, MG_Pipe::MGPipeHandle res, BufferSlice& outSlice);
// The handle names a store this session created (whatever its size, zero included).
Bool IsKnownWireBuffer(MG_Pipe::MGPipeHandle res) { return FindWireBuffer(res) != nullptr; }
// Names, for a Fatal about an unknown handle, whether this manager ever saw it.
void DescribeWireBufferHistory(MG_Pipe::MGPipeHandle res) const;
Bool ReadWireBuffer(MG_Pipe::MGPipeHandle res, Uint64 offset, Uint64 size, void* dst);
Bool CopyWireBufferRangeToSlice(MG_Pipe::MGPipeHandle res, Uint64 offset, Uint64 size,
const BufferSlice& dst);
@@ -358,6 +362,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// for the census to count), shared by the host write that cannot wait for the GPU and
// the watermark that cannot. `site` says which.
[[noreturn]] static void WireBufferSyncFatal(const char* site);
// True when the wait or copy that just failed failed because the device is lost: the
// session has been latched (VulkanRenderer::LatchWireDeviceLoss) and the caller returns
// instead of taking its Fatal.
static Bool LatchedOnDeviceLoss(const char* site);
// The wbuf[] gauges (PipeStats.h, Gauge::WireBuffers..WireDeferredSyncs). The peaks are
// taken at the two points the numbers can rise - a park and a mint - and published when
// the stats channel is on; MagmaWireReclaimScenario reads them off the server's line.
@@ -7,6 +7,8 @@
// End of Source File Header
#include "VkRenderPassManager.h"
#include "RenderPassGuard.h"
#include <atomic>
#include "MG_Impl/GLImpl/Framebuffer/GL_Framebuffer.h"
#include "MG_State/GLState/TextureState/TextureObject2D.h"
@@ -235,11 +237,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkClearManager& clearManager, VkTextureManager& textureManager, SwapchainObject& swapchainObject):
m_device(device), m_physicalDevice(physicalDevice), m_allocator(allocator), m_config(config),
m_clearManager(clearManager), m_textureManager(textureManager), m_swapchainObject(swapchainObject) {
RenderPassEntry::s_device = m_device;
s_clearManager = &m_clearManager;
s_textureManager = &m_textureManager;
s_swapchainObject = &m_swapchainObject;
s_renderPassManager = this;
RenderPassEntry::s_device.Get() = m_device;
s_clearManager.Get() = &m_clearManager;
s_textureManager.Get() = &m_textureManager;
s_swapchainObject.Get() = &m_swapchainObject;
s_renderPassManager.Get() = this;
}
VkRenderPassManager::~VkRenderPassManager() {}
@@ -257,8 +259,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
CollectDeferredRenderbufferReleases(/*destroyAll=*/true); // caller guarantees device idle
m_pendingRenderbufferClears.clear();
RenderPassEntry::s_textureResourcesScratch.clear();
s_activeRenderPass = {};
s_hasActiveRenderPass = false;
s_activeRenderPass.Get() = {};
s_hasActiveRenderPass.Get() = false;
m_rpFastValid = false;
}
@@ -607,6 +609,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
const Bool isDefaultFbo = fbo.IsDefaultFramebuffer();
if (isDefaultFbo) {
XXHASH_VERIFY(XXH64_update(m_hashState, &swapchainImageIndex, sizeof(swapchainImageIndex)));
XXHASH_VERIFY(XXH64_update(m_hashState, &m_defaultFramebufferTarget, sizeof(m_defaultFramebufferTarget)));
}
// sRGB attachments switch between their sRGB and UNORM-twin views with this
// capability (ResolveSrgbAttachmentWriteFormat), changing the render pass formats.
@@ -1528,6 +1531,17 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return &insertedIt->second;
}
void VkRenderPassManager::PurgeRenderPasses() {
Vector<VkRenderPass> destroyedRenderPasses;
destroyedRenderPasses.reserve(m_renderPasses.size());
for (auto& [_, entry] : m_renderPasses) destroyedRenderPasses.push_back(entry.renderPass);
m_renderPasses.clear();
m_rpFastValid = false;
if (!destroyedRenderPasses.empty() && m_evictionObserver != nullptr) {
m_evictionObserver->OnRenderPassesDestroyed(destroyedRenderPasses);
}
}
void VkRenderPassManager::OnPresent() {
++m_frameCounter;
@@ -1554,9 +1568,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// immediately - and a single batched notification costs one pipeline-cache
// scan instead of one per evicted pass.
Vector<VkRenderPass> destroyedRenderPasses;
const Uint64 activeHash = s_hasActiveRenderPass ? s_activeRenderPass.hash : 0;
const Uint64 activeHash = s_hasActiveRenderPass.Get() ? s_activeRenderPass.Get().hash : 0;
for (auto it = m_renderPasses.begin(); it != m_renderPasses.end();) {
const Bool isActive = s_hasActiveRenderPass && it->first == activeHash;
const Bool isActive = s_hasActiveRenderPass.Get() && it->first == activeHash;
if (!isActive && m_frameCounter - it->second.lastUsedFrame > kRetireAgeFrames) {
if (m_rpFastValid && m_rpFastRenderPassHash == it->first) {
m_rpFastValid = false;
@@ -1601,8 +1615,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// texture path below is immune because it re-reads the live payload).
// Same defense as ClearAttachmentsOnActiveRenderPass: prefer the live
// pending clear, fall back to the snapshot only when none is queued.
if (s_renderPassManager != nullptr &&
s_renderPassManager->GetPendingRenderbufferClear(pending.renderbuffer, clearPayload)) {
if (s_renderPassManager.Get() != nullptr &&
s_renderPassManager.Get()->GetPendingRenderbufferClear(pending.renderbuffer, clearPayload)) {
if ((clearPayload.mask & GL_COLOR_BUFFER_BIT) != 0 && pending.renderbuffer != nullptr &&
MG_Util::GetBaseInternalFormatComponentCount(pending.renderbuffer->GetInternalFormat()) ==
3) {
@@ -1614,7 +1628,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
} else {
if (pending.key.texture == nullptr ||
!s_clearManager->GetPendingClear(pending.key, clearPayload, liveTexture)) {
!s_clearManager.Get()->GetPendingClear(pending.key, clearPayload, liveTexture)) {
continue;
}
}
@@ -1636,48 +1650,58 @@ namespace MobileGL::MG_Backend::DirectVulkan {
vkCmdBeginRenderPass(commandBuffer, &renderPassBeginInfo, VK_SUBPASS_CONTENTS_INLINE);
// Pre-pass stream bookkeeping: this pass's attachment images are now
// referenced by the open frame recording.
if (s_textureManager != nullptr) {
if (s_textureManager.Get() != nullptr) {
for (const auto& tracked : renderPassEntry.trackedAttachmentLayouts) {
if (tracked.target == TrackedAttachmentTarget::Texture) {
if (const auto texture = tracked.texture.lock()) {
s_textureManager->StampTextureRecordingUse(texture.get());
s_textureManager.Get()->StampTextureRecordingUse(texture.get());
}
}
}
}
for (const auto& pending: renderPassEntry.pendingClearAttachments) {
if (pending.hasInlinePayload) {
if (s_renderPassManager != nullptr) {
s_renderPassManager->PopPendingRenderbufferClear(pending.renderbuffer);
if (s_renderPassManager.Get() != nullptr) {
s_renderPassManager.Get()->PopPendingRenderbufferClear(pending.renderbuffer);
}
} else {
s_clearManager->PopPendingClear(pending.key);
s_clearManager.Get()->PopPendingClear(pending.key);
}
}
s_activeRenderPass.hash = renderPassEntry.hash;
s_activeRenderPass.compatibilityHash = renderPassEntry.compatibilityHash;
s_activeRenderPass.trackedAttachmentLayouts = renderPassEntry.trackedAttachmentLayouts;
s_activeRenderPass.extent = renderPassEntry.extent;
s_hasActiveRenderPass = true;
s_activeRenderPass.Get().commandBuffer = commandBuffer;
s_activeRenderPass.Get().hash = renderPassEntry.hash;
s_activeRenderPass.Get().compatibilityHash = renderPassEntry.compatibilityHash;
s_activeRenderPass.Get().trackedAttachmentLayouts = renderPassEntry.trackedAttachmentLayouts;
s_activeRenderPass.Get().extent = renderPassEntry.extent;
s_hasActiveRenderPass.Get() = true;
return true;
}
Bool VkRenderPassManager::EndRenderPass(VkCommandBuffer commandBuffer) {
BumpWireImageWriteEpoch();
auto* activeRenderPass = GetActiveRenderPass();
Vector<TrackedAttachmentLayoutInfo> trackedAttachmentLayouts;
if (activeRenderPass != nullptr) {
trackedAttachmentLayouts = Move(activeRenderPass->trackedAttachmentLayouts);
}
vkCmdEndRenderPass(commandBuffer);
// Layout reconciliation below may record barriers, whose guard closes an active
// pass. Detach the ended pass first so that guard cannot recursively end it again.
s_activeRenderPass.Get() = {};
s_hasActiveRenderPass.Get() = false;
// The fast-path memo reuses the ACTIVE render pass; once the pass ends it must not carry
// over (the next span may be a different FBO resolved before its render pass is begun).
if (s_renderPassManager != nullptr) {
s_renderPassManager->m_rpFastValid = false;
if (s_renderPassManager.Get() != nullptr) {
s_renderPassManager.Get()->m_rpFastValid = false;
}
if (activeRenderPass != nullptr && !activeRenderPass->trackedAttachmentLayouts.empty()) {
for (const auto& trackedAttachment : activeRenderPass->trackedAttachmentLayouts) {
if (!trackedAttachmentLayouts.empty()) {
for (const auto& trackedAttachment : trackedAttachmentLayouts) {
switch (trackedAttachment.target) {
case TrackedAttachmentTarget::Texture:
MOBILEGL_ASSERT(s_textureManager != nullptr, "EndRenderPass: texture manager is null");
MOBILEGL_ASSERT(s_textureManager.Get() != nullptr, "EndRenderPass: texture manager is null");
if (const auto texture = trackedAttachment.texture.lock()) {
s_textureManager->UpdateTrackedImageLayoutAfterAttachmentWrite(
s_textureManager.Get()->UpdateTrackedImageLayoutAfterAttachmentWrite(
commandBuffer,
texture.get(),
trackedAttachment.textureMipLevel,
@@ -1685,27 +1709,27 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
break;
case TrackedAttachmentTarget::Renderbuffer:
MOBILEGL_ASSERT(s_renderPassManager != nullptr, "EndRenderPass: render pass manager is null");
MOBILEGL_ASSERT(s_renderPassManager.Get() != nullptr, "EndRenderPass: render pass manager is null");
if (const auto renderbuffer = trackedAttachment.renderbuffer.lock()) {
auto resourceIt =
s_renderPassManager->m_renderbufferResources.find(renderbuffer.get());
if (resourceIt != s_renderPassManager->m_renderbufferResources.end()) {
s_renderPassManager.Get()->m_renderbufferResources.find(renderbuffer.get());
if (resourceIt != s_renderPassManager.Get()->m_renderbufferResources.end()) {
resourceIt->second.layout = trackedAttachment.finalLayout;
}
}
break;
case TrackedAttachmentTarget::SwapchainColor:
MOBILEGL_ASSERT(s_swapchainObject != nullptr, "EndRenderPass: swapchain object is null");
s_swapchainObject->SetImageLayout(trackedAttachment.swapchainImageIndex, trackedAttachment.finalLayout);
MOBILEGL_ASSERT(s_swapchainObject.Get() != nullptr, "EndRenderPass: swapchain object is null");
s_swapchainObject.Get()->SetImageLayout(trackedAttachment.swapchainImageIndex, trackedAttachment.finalLayout);
// The pass stored into the attachment: its content is defined
// until the image is next presented.
s_swapchainObject->SetImageContentDefined(trackedAttachment.swapchainImageIndex, true);
s_swapchainObject.Get()->SetImageContentDefined(trackedAttachment.swapchainImageIndex, true);
break;
case TrackedAttachmentTarget::SwapchainDepthStencil:
MOBILEGL_ASSERT(s_swapchainObject != nullptr, "EndRenderPass: swapchain object is null");
s_swapchainObject->SetDepthStencilImageLayout(trackedAttachment.swapchainImageIndex,
MOBILEGL_ASSERT(s_swapchainObject.Get() != nullptr, "EndRenderPass: swapchain object is null");
s_swapchainObject.Get()->SetDepthStencilImageLayout(trackedAttachment.swapchainImageIndex,
trackedAttachment.finalLayout);
s_swapchainObject->SetDepthStencilContentDefined(trackedAttachment.swapchainImageIndex, true);
s_swapchainObject.Get()->SetDepthStencilContentDefined(trackedAttachment.swapchainImageIndex, true);
break;
default:
MOBILEGL_ASSERT(false, "EndRenderPass: unsupported tracked attachment target=%d",
@@ -1714,12 +1738,23 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
}
}
s_activeRenderPass = {};
s_hasActiveRenderPass = false;
return true;
}
ActiveRenderPassInfo* VkRenderPassManager::GetActiveRenderPass() {
return s_hasActiveRenderPass ? &s_activeRenderPass : nullptr;
return s_hasActiveRenderPass.Get() ? &s_activeRenderPass.Get() : nullptr;
}
namespace {
std::atomic<Uint64> g_wireImageWriteEpoch{1};
}
Uint64 WireImageWriteEpoch() { return g_wireImageWriteEpoch.load(std::memory_order_relaxed); }
void BumpWireImageWriteEpoch() { g_wireImageWriteEpoch.fetch_add(1, std::memory_order_relaxed); }
void EndActiveRenderPassOn(VkCommandBuffer commandBuffer) {
BumpWireImageWriteEpoch();
const auto* active = VkRenderPassManager::GetActiveRenderPass();
if (active != nullptr && active->commandBuffer == commandBuffer && commandBuffer != VK_NULL_HANDLE)
VkRenderPassManager::EndRenderPass(commandBuffer);
}
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -66,8 +66,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkExtent2D swapchainExtent);
struct RenderPassEntry {
static inline VkDevice s_device;
static inline Vector<VkTextureManager::TextureResource*> s_textureResourcesScratch;
// Per Magma session (MagmaSession.h): each session's renderer is its own device.
static inline SessionLocal<VkDevice> s_device;
static inline thread_local Vector<VkTextureManager::TextureResource*> s_textureResourcesScratch;
Uint64 hash = 0;
VkRenderPass renderPass = VK_NULL_HANDLE;
VkFramebuffer framebuffer = VK_NULL_HANDLE;
@@ -165,10 +166,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
~RenderPassEntry() {
if (renderPass != VK_NULL_HANDLE) {
vkDestroyRenderPass(s_device, renderPass, nullptr);
vkDestroyRenderPass(s_device.Get(), renderPass, nullptr);
}
if (framebuffer != VK_NULL_HANDLE) {
vkDestroyFramebuffer(s_device, framebuffer, nullptr);
vkDestroyFramebuffer(s_device.Get(), framebuffer, nullptr);
}
}
@@ -184,6 +185,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
struct ActiveRenderPassInfo {
Uint64 hash = 0;
Uint64 compatibilityHash = 0;
VkCommandBuffer commandBuffer = VK_NULL_HANDLE;
Vector<TrackedAttachmentLayoutInfo> trackedAttachmentLayouts;
IntVec2 extent = {0, 0};
@@ -226,6 +228,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Bool Initialize();
void Shutdown();
// The surface target the default framebuffer currently names (VulkanRenderer's surface
// targets): a default-framebuffer pass of one surface is not another's.
void SetDefaultFramebufferTarget(Uint64 serial) {
m_defaultFramebufferTarget = serial;
m_rpFastValid = false;
}
// Destroys every cached pass and framebuffer (not the renderbuffers). The device must be
// idle and no pass may be open: a surface target's image views are going away.
void PurgeRenderPasses();
HashType ComputeHash(
const MG_State::GLState::FramebufferObject& fbo,
@@ -302,6 +313,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// is reused WITHOUT recomputing the expensive per-draw hash. Invalidated by FBO switch /
// version change, swapchain rotation, any attachment image recreation (the two epochs),
// or a pending clear. Portable to Vulkan 1.1 (no dynamic_rendering / imageless FB needed).
Uint64 m_defaultFramebufferTarget = 0;
Bool m_rpFastValid = false;
const MG_State::GLState::FramebufferObject* m_rpFastFbo = nullptr;
// The FBO's never-reused lifetime id joins the raw pointer + Uint16 version:
@@ -411,12 +423,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
void DeferRenderbufferBackingRelease(RenderbufferResource& resource);
void CollectDeferredRenderbufferReleases(Bool destroyAll);
static inline XXH64_state_t* m_hashState = XXH64_createState();
static inline ActiveRenderPassInfo s_activeRenderPass{};
static inline Bool s_hasActiveRenderPass = false;
static inline VkClearManager* s_clearManager = nullptr;
static inline VkTextureManager* s_textureManager = nullptr;
static inline SwapchainObject* s_swapchainObject = nullptr;
static inline VkRenderPassManager* s_renderPassManager = nullptr;
static inline thread_local XXH64_state_t* m_hashState = XXH64_createState();
// Per Magma session (MagmaSession.h): one session's open render pass and managers are not
// another's.
static inline SessionLocal<ActiveRenderPassInfo> s_activeRenderPass;
static inline SessionLocal<Bool> s_hasActiveRenderPass;
static inline SessionLocal<VkClearManager*> s_clearManager;
static inline SessionLocal<VkTextureManager*> s_textureManager;
static inline SessionLocal<SwapchainObject*> s_swapchainObject;
static inline SessionLocal<VkRenderPassManager*> s_renderPassManager;
};
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -138,6 +138,6 @@ private:
UnorderedMap<Uint64, SamplerCacheEntry> m_samplers;
// Monotonic frame-boundary counter (bumped in OnFrameBoundary) for cache aging.
Uint64 m_frameBoundaryCounter = 0;
static inline XXH64_state_t* m_hashState = XXH64_createState();
static inline thread_local XXH64_state_t* m_hashState = XXH64_createState();
};
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -6,6 +6,7 @@
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#include "RenderPassGuard.h"
#include "VkTextureManager.h"
#include "ProgramFactory.h"
@@ -22,10 +23,16 @@
// (shape) and the server's staged-texture store (texels).
#include <MG_Pipe/PipeApply.h>
#include <MG_Remote/Server/StagedTextureStore.h>
// Shared images: a texture whose level 0 is a server-allocated AHardwareBuffer.
#include <MG_Remote/Server/SharedImageRegistry.h>
#include "../DirectVulkan.h"
// P7 wave 2 package B3: rule I's tally for the silent exits on this file's wire arm.
#include "WireDeclineTally.h"
#if defined(__ANDROID__)
#include <android/hardware_buffer.h>
#endif
#endif
#include <string>
#include <algorithm>
#include <cstdio>
#include <cstdlib>
@@ -42,9 +49,28 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// their feature is off (VUID-vkCmdPipelineBarrier-srcStageMask-04090/-04091), and ALL_GRAPHICS
// would also serialize against non-shader stages. The default only matters before a device
// exists, when nothing records barriers.
static VkPipelineStageFlags s_sampledReadStages =
VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
// Per Magma session: each session's renderer is its own device.
struct SampledReadStages {
VkPipelineStageFlags value = VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
};
static SessionLocal<SampledReadStages> s_sampledReadStagesSlot;
static VkImageUsageFlags ResolveTextureUsage(VkImageAspectFlags aspect,
VkFormatFeatureFlags formatFeatures,
Bool supportsStorageImage) {
VkImageUsageFlags usage = VK_IMAGE_USAGE_SAMPLED_BIT;
if (supportsStorageImage) usage |= VK_IMAGE_USAGE_STORAGE_BIT;
// A color aspect does not imply renderability (for example, RGB9_E5).
if ((aspect & VK_IMAGE_ASPECT_COLOR_BIT) &&
(formatFeatures & VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BIT)) {
usage |= VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
}
if (aspect & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
usage |= VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT;
}
return usage;
}
static Uint32 ComputeFullMipLevelCount(const IntVec3& baseTexelSize) {
Int maxDimension = std::max<Int>(baseTexelSize.x(),
@@ -203,7 +229,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
case VK_IMAGE_LAYOUT_DEPTH_READ_ONLY_STENCIL_ATTACHMENT_OPTIMAL:
case VK_IMAGE_LAYOUT_DEPTH_ATTACHMENT_STENCIL_READ_ONLY_OPTIMAL:
case VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL:
outSrcStageMask = s_sampledReadStages;
outSrcStageMask = s_sampledReadStagesSlot->value;
outSrcAccessMask = VK_ACCESS_SHADER_READ_BIT;
return;
case VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL:
@@ -267,7 +293,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
case VK_IMAGE_LAYOUT_DEPTH_READ_ONLY_STENCIL_ATTACHMENT_OPTIMAL:
case VK_IMAGE_LAYOUT_DEPTH_ATTACHMENT_STENCIL_READ_ONLY_OPTIMAL:
case VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL:
outDstStageMask = s_sampledReadStages;
outDstStageMask = s_sampledReadStagesSlot->value;
outDstAccessMask = VK_ACCESS_SHADER_READ_BIT;
return;
case VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL:
@@ -367,6 +393,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
copyRegions.push_back(copy);
}
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyImage(commandBuffer,
oldResource.image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
newResource.image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
@@ -646,7 +673,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
m_commandPool = initInfo.commandPool;
m_graphicsQueue = initInfo.graphicsQueue;
m_imageFormatListSupported = initInfo.imageFormatListSupported;
s_sampledReadStages = initInfo.sampledReadStageMask;
s_sampledReadStagesSlot->value = initInfo.sampledReadStageMask;
m_currentFrameIndex = 0;
m_deferredReleases.clear();
m_deferredReleases.resize(initInfo.frameCount);
@@ -659,8 +686,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
MOBILEGL_ASSERT(initInfo.frameCount > 0,
"VkTextureManager::Initialize failed: frameCount must be > 0");
TextureResource::s_device = m_device;
TextureResource::s_allocator = m_allocator;
TextureResource::s_device.Get() = m_device;
TextureResource::s_allocator.Get() = m_allocator;
// Own pool for the recycled upload-batch command buffers. Parking a
// dozen reset-but-alive command buffers in the renderer's shared pool
@@ -698,6 +725,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// Destroy images while the device/allocator still exist, including after context death.
m_wireTextureResources.clear();
m_wireRenderbufferResources.clear();
m_sharedImageUses.clear();
#endif
m_aliveObjects.clear();
m_storageImageTextures.clear();
@@ -1477,7 +1505,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
const Bool ok = TransitionImageLayout(commandBuffer, resource->image, resource->layout, targetLayout, srcStageMask,
s_sampledReadStages, srcAccessMask,
s_sampledReadStagesSlot->value, srcAccessMask,
VK_ACCESS_SHADER_READ_BIT, resource->aspect, 0, resource->mipLevels);
MOBILEGL_ASSERT(ok, "TransitionTextureForSampling: transition failed for textureId=%d", texture.GetExternalIndex());
// Pre-pass stream bookkeeping: a command referencing the image was recorded.
@@ -1648,6 +1676,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
std::max(source->depth >> level, 1u)};
copyRegions.push_back(copy);
}
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyImage(commandBuffer, source->image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, snapshot.image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, static_cast<Uint32>(copyRegions.size()), copyRegions.data());
@@ -1762,6 +1791,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// slices" but will read as "depth slice 0" once VK_KHR_maintenance9 is enabled. The
// validation layer warns about that literal 1 by name.
barrier.subresourceRange.layerCount = VK_REMAINING_ARRAY_LAYERS;
EndActiveRenderPassOn(commandBuffer);
vkCmdPipelineBarrier(commandBuffer, srcStageMask, dstStageMask, 0, 0, nullptr, 0, nullptr, 1, &barrier);
trackedLayout = newLayout;
@@ -2105,6 +2135,44 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Bool VkTextureManager::SyncWireTextureShape(const MG_Pipe::MGPipeResourceRecord& record,
TextureResource& resource, Bool requireStorage) {
const MG_Pipe::MGPResourceDesc& desc = record.Desc;
// A YUV shared image is converted INTO the texture's own RGBA8 storage, which the rest of
// this function defines as for any texture; the image is only named here, and the
// renderer converts it at its first use in each frame (WireYuvImage.inc).
if (record.SharedImageId != resource.yuvImageId && !m_syncingYuvTextureStorage) {
// Respecified away from (or to another) YUV image: its source goes with it, once the
// GPU is done with it (a released resource's Reset is deferred).
if (resource.yuvSource != nullptr) {
TextureResource retired;
retired.yuvSource = std::move(resource.yuvSource);
DeferResourceRelease(Move(retired));
}
resource.yuvImageId = 0;
resource.yuvOwner.reset();
resource.yuvSource.reset();
resource.yuvConvertedFrame = 0;
}
if (record.SharedImageId != 0 && !m_syncingYuvTextureStorage) {
const auto yuv = MG_Remote::Server::SharedImages::Find(record.SharedImageId);
if (yuv != nullptr && MG_Remote::Server::SharedImages::FourccIsYuv(yuv->Fourcc)) {
m_syncingYuvTextureStorage = true;
const Bool defined = SyncWireTextureShape(record, resource, requireStorage);
m_syncingYuvTextureStorage = false;
if (!defined) return false;
if (resource.yuvImageId != record.SharedImageId) {
resource.yuvImageId = record.SharedImageId;
resource.yuvOwner = yuv;
resource.yuvSource.reset();
resource.yuvConvertedFrame = 0;
}
return true;
}
}
// Level 0 is a shared image: bind its buffer. An image that cannot be bound here leaves
// the texture the storage its NULL-data definition asked for, below.
if (record.SharedImageId != 0 && !m_syncingYuvTextureStorage) {
const SharedImageBind bound = SyncWireSharedImage(record, resource, requireStorage);
if (bound != SharedImageBind::Unresolved) return bound == SharedImageBind::Bound;
}
if (desc.StorageKind != static_cast<Uint8>(TextureStorageType::Mipmap)) {
MGLOG_W_ONCE("Magma wire texture {slot=%u, gen=%u}: buffer-backed textures are not migrated; declined",
desc.Resource.Slot, desc.Resource.Gen);
@@ -2204,13 +2272,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
VkImageUsageFlags desiredUsage =
VK_IMAGE_USAGE_SAMPLED_BIT | (supportsStorageImage ? VK_IMAGE_USAGE_STORAGE_BIT : 0) |
((aspect & VK_IMAGE_ASPECT_COLOR_BIT) &&
(formatProperties.optimalTilingFeatures & VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BIT)
? VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT : 0) |
(((aspect & VK_IMAGE_ASPECT_DEPTH_BIT) || (aspect & VK_IMAGE_ASPECT_STENCIL_BIT))
? VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT
: 0);
ResolveTextureUsage(aspect, formatProperties.optimalTilingFeatures, supportsStorageImage);
// Multisample attachments are also transfer sources for vkCmdResolveImage.
desiredUsage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
@@ -2273,7 +2335,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
: (desc.Levels > 1 ? std::min(std::max<Uint32>(desc.Levels, fullMipLevels), fullMipLevels)
: 1u);
const Bool compatible = resource.image != VK_NULL_HANDLE && resource.format == format &&
// A shared image never stands in for allocated storage: the attach is over (or never
// resolved), so the texture gets storage of its own, and none of the image's texels.
const Bool compatible = resource.image != VK_NULL_HANDLE && resource.sharedImageId == 0 &&
resource.format == format &&
resource.extent.width == desc.Width && resource.extent.height == imageHeight &&
resource.depth == shapeInfo.depth && resource.arrayLayers == shapeInfo.arrayLayers &&
resource.viewType == shapeInfo.viewType &&
@@ -2288,7 +2353,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// A named-level respecify changes ONLY that level; a BindMask update changes no
// storage at all. Preserve every compatible old level when growing/upgrading the
// allocation. A whole-store redefinition permits (but does not require) old texels.
const Bool preserve = resource.image != VK_NULL_HANDLE && resource.format == format &&
const Bool preserve = resource.image != VK_NULL_HANDLE && resource.sharedImageId == 0 &&
resource.format == format &&
resource.extent.width == desc.Width && resource.extent.height == imageHeight &&
resource.depth == shapeInfo.depth && resource.arrayLayers == shapeInfo.arrayLayers &&
resource.viewType == shapeInfo.viewType &&
@@ -2682,9 +2748,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
stencilCopy.imageSubresource.aspectMask = VK_IMAGE_ASPECT_STENCIL_BIT;
stencilCopy.bufferOffset = stagingBase + item.offset + static_cast<VkDeviceSize>(texelCount) * 4;
const VkBufferImageCopy copies[2] = {depthCopy, stencilCopy};
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyBufferToImage(commandBuffer, stagingBuffer, resource.image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 2, copies);
} else {
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyBufferToImage(commandBuffer, stagingBuffer, resource.image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &copy);
}
@@ -2693,7 +2761,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
const VkImageLayout finalLayout = ResolveSampledReadOnlyLayout(aspectMask);
VkImageLayout uploadLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
ok = TransitionImageLayout(commandBuffer, resource.image, uploadLayout, finalLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT, s_sampledReadStages,
VK_PIPELINE_STAGE_TRANSFER_BIT, s_sampledReadStagesSlot->value,
VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_SHADER_READ_BIT, aspectMask, 0,
resource.mipLevels);
MOBILEGL_ASSERT(ok, "UploadPendingWireLevels: transition to the sampled layout failed");
@@ -2766,6 +2834,201 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return MG_Pipe::kMGPipeNullHandle;
}
Bool VkTextureManager::ImportSharedImage(void* nativeBuffer, Uint32 width, Uint32 height,
ImportedSharedImage& out, String& why) {
out = ImportedSharedImage{};
#if defined(__ANDROID__)
auto* ahb = static_cast<AHardwareBuffer*>(nativeBuffer);
if (ahb == nullptr) {
why = "the image has no AHardwareBuffer";
return false;
}
VulkanRenderer::WireAhbImport ctx;
if (pVulkanRenderer == nullptr || !pVulkanRenderer->GetWireAhbImport(ctx)) {
why = "the device did not take VK_ANDROID_external_memory_android_hardware_buffer";
return false;
}
AHardwareBuffer_Desc desc{};
AHardwareBuffer_describe(ahb, &desc);
if (desc.width != width || desc.height != height || desc.layers != 1) {
why = "the buffer is " + std::to_string(desc.width) + "x" + std::to_string(desc.height) + "x" +
std::to_string(desc.layers) + ", not " + std::to_string(width) + "x" + std::to_string(height);
return false;
}
const auto getProperties =
reinterpret_cast<PFN_vkGetAndroidHardwareBufferPropertiesANDROID>(ctx.getAhbProperties);
VkAndroidHardwareBufferFormatPropertiesANDROID formatProperties{
VK_STRUCTURE_TYPE_ANDROID_HARDWARE_BUFFER_FORMAT_PROPERTIES_ANDROID};
VkAndroidHardwareBufferPropertiesANDROID properties{VK_STRUCTURE_TYPE_ANDROID_HARDWARE_BUFFER_PROPERTIES_ANDROID};
properties.pNext = &formatProperties;
VkResult result = getProperties(ctx.device, ahb, &properties);
if (result != VK_SUCCESS) {
why = "vkGetAndroidHardwareBufferPropertiesANDROID -> " + std::to_string(static_cast<int>(result));
return false;
}
// The RGBA8/RGBX8 buffers the registry allocates have a Vulkan format; an external-only
// format would need a YCbCr conversion and is not a colour buffer this path can write.
if (formatProperties.format == VK_FORMAT_UNDEFINED) {
why = "the buffer has no Vulkan format (external format only)";
return false;
}
// Usage is bounded by the buffer's own usage (the AHardwareBuffer usage equivalence);
// transfers need no buffer usage bit.
VkImageUsageFlags usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT;
if ((desc.usage & AHARDWAREBUFFER_USAGE_GPU_SAMPLED_IMAGE) != 0 &&
(formatProperties.formatFeatures & VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT) != 0)
usage |= VK_IMAGE_USAGE_SAMPLED_BIT;
if ((desc.usage & AHARDWAREBUFFER_USAGE_GPU_COLOR_OUTPUT) != 0 &&
(formatProperties.formatFeatures & VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BIT) != 0)
usage |= VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
VkExternalMemoryImageCreateInfo external{VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_IMAGE_CREATE_INFO};
external.handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_ANDROID_HARDWARE_BUFFER_BIT_ANDROID;
VkImageCreateInfo imageInfo{VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO};
imageInfo.pNext = &external;
imageInfo.imageType = VK_IMAGE_TYPE_2D;
imageInfo.format = formatProperties.format;
imageInfo.extent = {width, height, 1};
imageInfo.mipLevels = 1;
imageInfo.arrayLayers = 1;
imageInfo.samples = VK_SAMPLE_COUNT_1_BIT;
imageInfo.tiling = VK_IMAGE_TILING_OPTIMAL;
imageInfo.usage = usage;
imageInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
imageInfo.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
result = vkCreateImage(ctx.device, &imageInfo, nullptr, &out.image);
if (result != VK_SUCCESS) {
out.image = VK_NULL_HANDLE;
why = "vkCreateImage(external AHB) -> " + std::to_string(static_cast<int>(result));
return false;
}
Int32 type = -1;
for (Uint32 i = 0; i < ctx.memory.memoryTypeCount && type < 0; ++i) {
if ((properties.memoryTypeBits & (1u << i)) != 0 &&
(ctx.memory.memoryTypes[i].propertyFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT) != 0)
type = static_cast<Int32>(i);
}
for (Uint32 i = 0; i < ctx.memory.memoryTypeCount && type < 0; ++i) {
if ((properties.memoryTypeBits & (1u << i)) != 0) type = static_cast<Int32>(i);
}
if (type < 0) {
why = "no memory type can import it";
DestroyImportedSharedImage(ctx.device, out);
return false;
}
VkMemoryDedicatedAllocateInfo dedicated{VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO};
dedicated.image = out.image;
VkImportAndroidHardwareBufferInfoANDROID import{VK_STRUCTURE_TYPE_IMPORT_ANDROID_HARDWARE_BUFFER_INFO_ANDROID};
import.pNext = &dedicated;
import.buffer = ahb;
VkMemoryAllocateInfo allocateInfo{VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO};
allocateInfo.pNext = &import;
allocateInfo.allocationSize = properties.allocationSize;
allocateInfo.memoryTypeIndex = static_cast<Uint32>(type);
result = vkAllocateMemory(ctx.device, &allocateInfo, nullptr, &out.memory);
if (result != VK_SUCCESS) {
out.memory = VK_NULL_HANDLE;
why = "vkAllocateMemory(import AHB) -> " + std::to_string(static_cast<int>(result));
DestroyImportedSharedImage(ctx.device, out);
return false;
}
result = vkBindImageMemory(ctx.device, out.image, out.memory, 0);
if (result != VK_SUCCESS) {
why = "vkBindImageMemory -> " + std::to_string(static_cast<int>(result));
DestroyImportedSharedImage(ctx.device, out);
return false;
}
out.format = formatProperties.format;
out.usage = usage;
out.features = formatProperties.formatFeatures;
out.queueFamily = ctx.queueFamily;
return true;
#else
(void)nativeBuffer, (void)width, (void)height;
why = "shared images are imported on Android only";
return false;
#endif
}
void VkTextureManager::DestroyImportedSharedImage(VkDevice device, ImportedSharedImage& image) {
if (image.image != VK_NULL_HANDLE) vkDestroyImage(device, image.image, nullptr);
if (image.memory != VK_NULL_HANDLE) vkFreeMemory(device, image.memory, nullptr);
image = ImportedSharedImage{};
}
VkTextureManager::SharedImageBind VkTextureManager::SyncWireSharedImage(
const MG_Pipe::MGPipeResourceRecord& record, TextureResource& resource, Bool requireStorage) {
const MG_Pipe::MGPResourceDesc& desc = record.Desc;
const Uint64 id = record.SharedImageId;
if (requireStorage) {
// The buffer's usage has no storage equivalent; a reallocation would drop the image.
MGLOG_E_ONCE("Magma wire texture {slot=%u, gen=%u}: shared image %llu is bound as a shader image, "
"which it cannot back; declined",
desc.Resource.Slot, desc.Resource.Gen, static_cast<unsigned long long>(id));
return SharedImageBind::Failed;
}
// Already this image: the serial moved for something else (params, the bind mask).
if (resource.sharedImageId == id && resource.image != VK_NULL_HANDLE) return SharedImageBind::Bound;
const auto image = MG_Remote::Server::SharedImages::Find(id);
if (image == nullptr) {
MGLOG_E_ONCE("Magma wire texture {slot=%u, gen=%u}: shared image %llu is no longer live; the texture "
"keeps its own (undefined) level 0",
desc.Resource.Slot, desc.Resource.Gen, static_cast<unsigned long long>(id));
return SharedImageBind::Unresolved;
}
const Bool flat2D = desc.Target == static_cast<Uint8>(MG_Pipe::MGPipeResourceTarget::Tex2D) ||
desc.Target == static_cast<Uint8>(MG_Pipe::MGPipeResourceTarget::TexRect);
if (!flat2D || desc.Width != image->Width || desc.Height != image->Height ||
!MG_Pipe::MGPipeHandleIsNull(desc.ViewOf)) {
MGLOG_E_ONCE("Magma wire texture {slot=%u, gen=%u}: pipe target %u at %ux%u cannot take shared image "
"%llu (%ux%u); the texture keeps its own level 0",
desc.Resource.Slot, desc.Resource.Gen, static_cast<Uint32>(desc.Target), desc.Width,
desc.Height, static_cast<unsigned long long>(id), image->Width, image->Height);
return SharedImageBind::Unresolved;
}
ImportedSharedImage imported;
String why;
if (!ImportSharedImage(image->Native, image->Width, image->Height, imported, why)) {
MGLOG_E_ONCE("Magma wire texture {slot=%u, gen=%u}: shared image %llu not imported - %s; the texture "
"keeps its own level 0",
desc.Resource.Slot, desc.Resource.Gen, static_cast<unsigned long long>(id), why.c_str());
return SharedImageBind::Unresolved;
}
// NOT ACQUIRED HERE. The writer (a client session's present, on its own device) releases
// the image to VK_QUEUE_FAMILY_FOREIGN_EXT in GENERAL - also the layout every wire
// descriptor samples in - once per frame it writes. Each use acquires it again when the
// write generation moved, waiting for that write's fence on the GPU, and the frame boundary
// releases it back (VulkanRenderer::AcquireSharedImage, WireSharedImage.inc).
TextureResource replacement;
replacement.image = imported.image;
replacement.importedMemory = imported.memory;
replacement.layout = VK_IMAGE_LAYOUT_GENERAL;
replacement.extent = {image->Width, image->Height};
replacement.depth = 1;
replacement.arrayLayers = 1;
replacement.mipLevels = 1;
replacement.sampledBaseMipLevel = 0;
replacement.sampledLevelCount = 1;
replacement.format = imported.format;
replacement.aspect = VK_IMAGE_ASPECT_COLOR_BIT;
replacement.viewType = VK_IMAGE_VIEW_TYPE_2D;
replacement.sampleCount = VK_SAMPLE_COUNT_1_BIT;
replacement.imageCreateFlags = 0;
replacement.usageFlags = imported.usage;
replacement.sharedImageId = id;
replacement.sharedImageOwner = image;
replacement.sharedImageAlphaOne = MG_Remote::Server::SharedImages::FourccIgnoresAlpha(image->Fourcc);
++m_textureImageEpoch; // a new attachment image invalidates cached render passes
DeferResourceRelease(Move(resource));
std::destroy_at(&resource);
std::construct_at(&resource, Move(replacement));
MGLOG_D("Magma wire texture {slot=%u, gen=%u}: level 0 is shared image %llu (%ux%u)", desc.Resource.Slot,
desc.Resource.Gen, static_cast<unsigned long long>(id), image->Width, image->Height);
return SharedImageBind::Bound;
}
VkTextureManager::TextureResource* VkTextureManager::SyncTextureResourceByHandle(
MG_Pipe::MGPipeHandle handle, Bool renderbuffer, Bool requireStorage) {
MOBILEGL_ASSERT(m_device != VK_NULL_HANDLE, "SyncTextureResourceByHandle: m_device == VK_NULL_HANDLE");
@@ -2842,6 +3105,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
handle.Slot, handle.Gen);
return nullptr;
}
resource.sharedImageKey = resource.sharedImageId != 0 || resource.yuvImageId != 0 ? key : 0;
resource.syncedWireSerial = record.Serial;
return &resource;
}
@@ -2963,6 +3227,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
if ((aspect & copyAspect) == 0) continue;
copy.srcSubresource.aspectMask = copyAspect;
copy.dstSubresource.aspectMask = copyAspect;
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyImage(commandBuffer, old.image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, grown.image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &copy);
}
@@ -3258,12 +3523,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
VkImageUsageFlags desiredUsage =
VK_IMAGE_USAGE_SAMPLED_BIT |
(supportsStorageImage ? VK_IMAGE_USAGE_STORAGE_BIT : 0) |
((aspect & VK_IMAGE_ASPECT_COLOR_BIT) ? VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT : 0) |
(((aspect & VK_IMAGE_ASPECT_DEPTH_BIT) || (aspect & VK_IMAGE_ASPECT_STENCIL_BIT)) ?
VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT :
0);
ResolveTextureUsage(aspect, formatProperties.optimalTilingFeatures, supportsStorageImage);
if (!isMultisampleTexture) {
desiredUsage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
}
@@ -3558,7 +3818,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
resource.sampledView == VK_NULL_HANDLE &&
resource.perMipViews.empty() && resource.perMipSampledViews.empty() &&
resource.attachmentViews.empty() && resource.alternateSampledViews.empty() &&
resource.storageImageViews.empty()) {
resource.storageImageViews.empty()
#if MOBILEGL_BUILD_DISAGGREGATED
&& resource.yuvSource == nullptr
#endif
) {
return;
}
@@ -4518,6 +4782,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
rectCopies[rectCopyCount++] = rectCopy;
runningOffset += static_cast<VkDeviceSize>(rect.TexelCount() * item.texelBytes);
}
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyBufferToImage(commandBuffer, stagingBuffer, outResource.image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, rectCopyCount, rectCopies.data());
continue;
@@ -4559,10 +4824,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
stencilCopy.imageSubresource.aspectMask = VK_IMAGE_ASPECT_STENCIL_BIT;
stencilCopy.bufferOffset = stagingBase + item.offset + static_cast<VkDeviceSize>(texelCount) * 4;
const VkBufferImageCopy copies[2] = {depthCopy, stencilCopy};
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyBufferToImage(commandBuffer, stagingBuffer, outResource.image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 2, copies);
continue;
}
EndActiveRenderPassOn(commandBuffer);
vkCmdCopyBufferToImage(commandBuffer, stagingBuffer, outResource.image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
1, &copy);
}
@@ -4573,7 +4840,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
uploadLayout,
finalLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT,
s_sampledReadStages,
s_sampledReadStagesSlot->value,
VK_ACCESS_TRANSFER_WRITE_BIT,
VK_ACCESS_SHADER_READ_BIT,
aspectMask, 0, outResource.mipLevels);
@@ -9,6 +9,7 @@
#pragma once
#include "../VkIncludes.h"
#include "../MagmaSession.h"
#include <Includes.h>
#include <MG_State/GLState/FramebufferState/FramebufferObject.h>
#include <MG_State/GLState/TextureState/TextureObject.h>
@@ -356,6 +357,30 @@ public:
// on every mutation, so the first handle sync of a fresh image always re-checks.
#if MOBILEGL_BUILD_DISAGGREGATED
Uint64 syncedWireSerial = 0;
// SHARED IMAGES: level 0 is a server shared image (an imported AHardwareBuffer), not a
// VMA allocation. The resource owns the image and its dedicated imported memory, and
// holds the registry's reference so the buffer outlives every use (deferred releases
// included). `sharedImageAlphaOne`: an X-format image, whose alpha bytes are undefined.
VkDeviceMemory importedMemory = VK_NULL_HANDLE;
Uint64 sharedImageId = 0;
SharedPtr<const void> sharedImageOwner;
Bool sharedImageAlphaOne = false;
// The reading side's ownership of the image (VulkanRenderer::AcquireSharedImage): whether
// this device holds it now - acquired from the foreign family and not yet released back at
// the frame boundary - the writer's generation it last acquired, and the
// m_wireTextureResources key it lives under, by which the frame boundary finds it again.
Bool sharedImageHeld = false;
Uint64 sharedImageGeneration = 0;
Uint64 sharedImageKey = 0;
// A YUV shared image (NV12/P010) is never the texture's image: the texture keeps its own
// RGBA8 level 0, into which the renderer converts the image at its first use in each frame
// (WireYuvImage.inc) - `yuvSource` being that conversion's import of the buffer, made on
// first use, and `yuvConvertedFrame` the frame it last ran in. `yuvOwner` holds the
// registry's image for as long as the texture names it.
Uint64 yuvImageId = 0;
SharedPtr<const void> yuvOwner;
SharedPtr<void> yuvSource;
Uint64 yuvConvertedFrame = 0;
#endif
TextureResource() = default;
@@ -392,44 +417,72 @@ public:
std::swap(this->syncedShapeVersion, that.syncedShapeVersion);
#if MOBILEGL_BUILD_DISAGGREGATED
std::swap(this->syncedWireSerial, that.syncedWireSerial);
std::swap(this->importedMemory, that.importedMemory);
std::swap(this->sharedImageId, that.sharedImageId);
std::swap(this->sharedImageOwner, that.sharedImageOwner);
std::swap(this->sharedImageAlphaOne, that.sharedImageAlphaOne);
std::swap(this->sharedImageHeld, that.sharedImageHeld);
std::swap(this->sharedImageGeneration, that.sharedImageGeneration);
std::swap(this->sharedImageKey, that.sharedImageKey);
std::swap(this->yuvImageId, that.yuvImageId);
std::swap(this->yuvOwner, that.yuvOwner);
std::swap(this->yuvSource, that.yuvSource);
std::swap(this->yuvConvertedFrame, that.yuvConvertedFrame);
#endif
}
void Reset() {
if (fullView != VK_NULL_HANDLE) {
vkDestroyImageView(s_device, fullView, nullptr);
vkDestroyImageView(s_device.Get(), fullView, nullptr);
}
if (sampledView != VK_NULL_HANDLE) {
vkDestroyImageView(s_device, sampledView, nullptr);
vkDestroyImageView(s_device.Get(), sampledView, nullptr);
}
for (const auto attachmentView : perMipViews) {
if (attachmentView != VK_NULL_HANDLE) {
vkDestroyImageView(s_device, attachmentView, nullptr);
vkDestroyImageView(s_device.Get(), attachmentView, nullptr);
}
}
for (const auto sampledView : perMipSampledViews) {
if (sampledView != VK_NULL_HANDLE) {
vkDestroyImageView(s_device, sampledView, nullptr);
vkDestroyImageView(s_device.Get(), sampledView, nullptr);
}
}
for (const auto& [_, attachmentView] : attachmentViews) {
if (attachmentView != VK_NULL_HANDLE) {
vkDestroyImageView(s_device, attachmentView, nullptr);
vkDestroyImageView(s_device.Get(), attachmentView, nullptr);
}
}
for (const auto& [_, sampledView] : alternateSampledViews) {
if (sampledView != VK_NULL_HANDLE) {
vkDestroyImageView(s_device, sampledView, nullptr);
vkDestroyImageView(s_device.Get(), sampledView, nullptr);
}
}
for (const auto& [_, storageImageView] : storageImageViews) {
if (storageImageView != VK_NULL_HANDLE) {
vkDestroyImageView(s_device, storageImageView, nullptr);
vkDestroyImageView(s_device.Get(), storageImageView, nullptr);
}
}
if (image != VK_NULL_HANDLE && allocation != nullptr) {
vmaDestroyImage(s_allocator, image, allocation);
vmaDestroyImage(s_allocator.Get(), image, allocation);
}
#if MOBILEGL_BUILD_DISAGGREGATED
if (importedMemory != VK_NULL_HANDLE) {
if (image != VK_NULL_HANDLE) vkDestroyImage(s_device.Get(), image, nullptr);
vkFreeMemory(s_device.Get(), importedMemory, nullptr);
}
importedMemory = VK_NULL_HANDLE;
sharedImageId = 0;
sharedImageOwner.reset();
sharedImageAlphaOne = false;
sharedImageHeld = false;
sharedImageGeneration = 0;
sharedImageKey = 0;
yuvImageId = 0;
yuvOwner.reset();
yuvSource.reset();
yuvConvertedFrame = 0;
#endif
fullView = VK_NULL_HANDLE;
sampledView = VK_NULL_HANDLE;
perMipViews.clear();
@@ -466,8 +519,8 @@ public:
Reset();
}
static inline VkDevice s_device = VK_NULL_HANDLE;
static inline VmaAllocator s_allocator = VK_NULL_HANDLE;
static inline SessionLocal<VkDevice> s_device; // per Magma session (MagmaSession.h)
static inline SessionLocal<VmaAllocator> s_allocator;
};
struct SampledTextureSnapshot {
@@ -591,6 +644,41 @@ public:
// ranges select cube faces; array and 3D levels share one staged level shadow.
void MarkWireTextureGpuWritten(MG_Pipe::MGPipeHandle handle, Uint32 mipLevel,
Uint32 baseArrayLayer = 0, Uint32 layerCount = 1);
// SHARED IMAGES (docs/Disaggregated/notes/anland/plan-ahb-dmabuf.md): a server shared image's
// AHardwareBuffer bound as a single-level 2D colour image on THIS session's device, on dedicated
// memory imported from the buffer. Android only (false with `why` elsewhere, or when the device
// did not take the AHardwareBuffer extension). The image is born UNDEFINED; whoever uses it
// acquires it from the foreign queue family first, and a writer releases it back after.
struct ImportedSharedImage {
VkImage image = VK_NULL_HANDLE;
VkDeviceMemory memory = VK_NULL_HANDLE;
VkFormat format = VK_FORMAT_UNDEFINED;
VkImageUsageFlags usage = 0;
VkFormatFeatureFlags features = 0;
Uint32 queueFamily = 0;
};
static Bool ImportSharedImage(void* nativeBuffer, Uint32 width, Uint32 height, ImportedSharedImage& out,
String& why);
static void DestroyImportedSharedImage(VkDevice device, ImportedSharedImage& image);
// The reading side's per-use hook: PrepareWireTextureResources notes every shared-image
// texture a draw or dispatch samples, and the renderer takes the list right after
// (VulkanRenderer::AcquireNotedSharedImages) to acquire each one ahead of the sampling.
void NoteSharedImageUse(const TextureResource& resource) {
if (std::find(m_sharedImageUses.begin(), m_sharedImageUses.end(), resource.sharedImageKey) ==
m_sharedImageUses.end())
m_sharedImageUses.push_back(resource.sharedImageKey);
}
Bool HasNotedSharedImageUses() const { return !m_sharedImageUses.empty(); }
void TakeNotedSharedImageUses(Vector<Uint64>& out) {
out.clear();
std::swap(out, m_sharedImageUses);
}
// The live handle-keyed texture resource under `key`, or null.
TextureResource* FindWireTextureResource(Uint64 key) {
const auto it = m_wireTextureResources.find(key);
return it != m_wireTextureResources.end() && it->second.image != VK_NULL_HANDLE ? &it->second : nullptr;
}
#endif
VkImageView GetOrCreateViewAtMipLevel(MG_State::GLState::ITextureObject& texture, Uint32 mipLevel);
VkImageView GetOrCreateAttachmentViewAtMipLevel(MG_State::GLState::ITextureObject& texture, Uint32 mipLevel,
@@ -777,10 +865,18 @@ private:
Bool requireStorage);
Bool UploadPendingWireLevels(MG_Pipe::MGPipeHandle handle, MG_Pipe::MGPipeResourceRecord& record,
TextureResource& resource);
// SyncWireTextureShape's arm for a record whose level 0 IS a shared image (SharedImageId):
// binds the imported buffer instead of allocating. Unresolved = the image is gone or cannot be
// imported here; `resource` is then untouched and the caller allocates as for any texture.
enum class SharedImageBind : Uint8 { Bound, Failed, Unresolved };
SharedImageBind SyncWireSharedImage(const MG_Pipe::MGPipeResourceRecord& record, TextureResource& resource,
Bool requireStorage);
// The handle-keyed twin of m_textureResources, keyed by StagedTextureStore::KeyForHandle.
// Node-based for the same reason m_renderbufferResources is: callers hold TextureResource*
// across map-touching calls.
std::unordered_map<Uint64, TextureResource> m_wireTextureResources;
// Shared-image textures sampled since the renderer last took the list (NoteSharedImageUse).
Vector<Uint64> m_sharedImageUses;
// Handles are per-kind; a renderbuffer and texture may have identical slot/gen.
std::unordered_map<Uint64, TextureResource> m_wireRenderbufferResources;
#endif
@@ -803,6 +899,8 @@ private:
// Active only between BeginDrawSyncScope/EndDrawSyncScope; identities of
// textures already fully synced in the current draw (small N -> flat scan).
Bool m_drawSyncScopeActive = false;
// SyncWireTextureShape defining the RGBA8 storage a YUV shared image is converted into.
Bool m_syncingYuvTextureStorage = false;
// Per-draw sync memo: the identity plus the resolved resource pointer. The pointer is stable
// across rehash in the node-based m_textureResources and stays valid for the draw (a texture
// synced this draw is alive and is not erased mid-draw), so a repeat sync of the same texture
File diff suppressed because it is too large Load Diff
@@ -23,7 +23,9 @@
#include "VkTextureManager.h"
#include "VkTimerQueryManager.h"
#if MOBILEGL_BUILD_DISAGGREGATED
#include "GpuProgressMarkers.h"
#include "WireRenderPassCompatibility.h"
#include <MG_Remote/Server/SharedImageRegistry.h>
#endif
#include "MG_Util/Math/VectorTypes.h"
#include <Includes.h>
@@ -144,6 +146,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// IBufferCopyCommandProvider: recording command buffer, outside any
// render pass, for immediate staged buffer copies.
VkCommandBuffer AcquireBufferCopyCommandBuffer() override;
// EGL_BUFFER_AGE_EXT of the active target's default framebuffer for the next frame: the age
// of the swapchain image acquired for it (SwapchainObject::BufferAgeOf). Asking makes the
// target keep its presented images' content from then on.
Int32 CurrentDrawBufferAge();
#if MOBILEGL_BUILD_DISAGGREGATED
VkBufferManager& GetWireBufferManager() { return m_bufferManager; }
Bool FlushWirePendingCommandsForTextureUpdate() {
@@ -179,6 +185,36 @@ namespace MobileGL::MG_Backend::DirectVulkan {
void* getAhbProperties = nullptr; // PFN_vkGetAndroidHardwareBufferPropertiesANDROID
};
Bool GetWireAhbImport(WireAhbImport& out) const;
// SHARED IMAGES (BackendObject::BlitDefaultFramebufferToSharedImage): the active surface
// target's default framebuffer, copied into `image` top row first and scaled to its
// extent. Returns once the copy is submitted, with its fence published to the image
// (SharedImages::PublishWrite); waits for it on the CPU only on a device without sync_file
// export. False (logged) when it cannot be done here: no AHardwareBuffer import, a
// quarter-turned surface, an unblittable format.
//
// `region` (GL window coordinates; Full = all) is the part to copy into an image this
// renderer has written before - the rest of it is left as it is; a first write, a scaled
// or turned copy copies all of it.
Bool BlitDefaultFramebufferToSharedImage(const SharedImageView& image, const MG_Util::Damage::Region& region);
// IMPLICIT SYNC BY FLUSH (BackendObject::PublishSharedImageAccesses): every image acquired
// since the last boundary goes back to the foreign family, the recording is submitted with
// the exportable semaphore, and its sync_file is published to each as a write and a read.
// From then on an image's first acquire per frame also waits for its pending reads.
Bool PublishSharedImageAccesses();
// EGL_ANDROID_native_fence_sync's fence command (BackendObject::ExportNativeFence): the
// submission's sync_file in `*fence` (-1: none, and the work is waited out here), and - for
// a session that writes images - the boundary PublishSharedImageAccesses makes.
Bool ExportNativeFence(int* fence);
// The reading side's per-use hook (WireSharedImage.inc): every shared-image texture the
// wire draw or dispatch being set up samples (VkTextureManager::NoteSharedImageUse) is
// acquired from the foreign family before it, when it moved or is not held.
void AcquireNotedSharedImages() {
if (m_textureManager && m_textureManager->HasNotedSharedImageUses()) AcquireNotedSharedImagesSlow();
}
// One shared-image texture, before a use: a new write generation, or an image released at
// the last frame boundary, is acquired here - an ownership barrier recorded ahead of the
// use, the write's fence a wait of the submission carrying it - and noted as read.
void AcquireSharedImage(VkTextureManager::TextureResource& resource);
#endif
// FrameContext::IRecordingObserver: prepares the frame's timer-query
@@ -358,6 +394,27 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// one is retired until the frame slot's fence is next waited. Returns
// true when a submission was made.
Bool FlushPendingCommands();
// DEVICE LOSS. A GPU fault or hang the driver reset leaves this renderer's VkDevice lost:
// every later submit and wait answers VK_ERROR_DEVICE_LOST. NoteDeviceLoss records it at
// the submit/wait sites (cheap, any result); IsDeviceLost answers from the record and,
// when nothing was recorded, asks the device once (only failure paths call it, so the
// probe's wait costs nothing on a healthy frame). Each session owns its own VkDevice, so a
// loss here is this session's alone.
void NoteDeviceLoss(VkResult result, const char* where);
Bool IsDeviceLost();
#if MOBILEGL_BUILD_DISAGGREGATED
// For a wire site whose submit or wait just failed: true when the failure is a lost
// device, in which case this session has been latched (MGPipeSessionLatch - it ends, the
// display server's other sessions do not) and the caller must return from its verb
// without recording or submitting anything more. False: an ordinary failure, the site
// keeps the death it had. `site` names it in the latch line.
Bool LatchWireDeviceLoss(const char* site);
// THE GPU HANG WATCH (GpuProgressMarkers.h): true once the watch named one of this device's
// submissions as the one that stops the GPU. The device is then treated as lost, this session
// is latched, and the caller returns from its verb without submitting anything more. Asked
// before every frame submission and by the apply loop before it applies another record.
Bool LatchIfGpuHung(const char* site);
#endif
// Flush gated on usefulness: only flushes when `submitIndex` is still
// unsubmitted, so poll loops on already-submitted fences do not split
// the frame's render pass (a full tile load/store on TBDR GPUs).
@@ -427,7 +484,65 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// no new swapchain is installed and presentation must stay suspended.
Bool RecreateSwapchain();
// SURFACE TARGETS. One renderer - one device and every GL object's resources - serves
// every EGL surface of its context's client; only the presentation side (VkSurfaceKHR,
// swapchain, its render-finished semaphores, the acquired image, the pbuffer's image
// reader) belongs to a surface. `key` names the client surface. The renderer starts with
// the surface it was constructed for as its active target (SetActiveSurfaceTargetKey).
// Activating another key parks the active target - its recorded work submitted, its
// pending acquire consumed - and brings the other back, or builds it on first use.
void SetActiveSurfaceTargetKey(Uint64 key) { m_activeTargetKey = key; }
// False when a NEW target's surface could not be built (logged); the renderer is then left
// with no active target - every present drops its frame - and the caller refuses by name.
Bool ActivateSurfaceTarget(Uint64 key, NativeWindowType window, const VulkanRendererConfig& surfaceConfig);
// Drops a surface's target, active or parked. A no-op for a key that has none.
void DestroySurfaceTarget(Uint64 key);
// The default framebuffer's extent (surface space, what PublishDefaultFramebufferInfo
// publishes) of `key`'s target, active or parked; 0x0 when it has none or no swapchain.
VkExtent2D SurfaceTargetExtent(Uint64 key) const;
// The active WINDOW target's window changed size. Between frames (nothing recorded since the last
// present) its swapchain is rebuilt now - which publishes the new extent - so a client that has
// nothing to redraw learns it without presenting; with a frame half-recorded the rebuild waits
// for that frame's present (it would otherwise land half in each swapchain). False: no active
// window target.
Bool FollowActiveWindowResize();
// Whether `key` names the active target.
Bool IsActiveSurfaceTarget(Uint64 key) const { return key == m_activeTargetKey && m_surface != VK_NULL_HANDLE; }
private:
// The presentation half of the renderer, parked while another surface is active. Its
// fields mirror the renderer members ExchangeActiveTarget swaps them with.
struct SurfaceTarget {
Uint64 serial = 0;
NativeWindowType window = 0;
void* platformDisplay = nullptr;
void* platformLibrary = nullptr;
void* platformCloseDisplay = nullptr;
void* fallbackImageReader = nullptr;
Uint32 surfaceWidth = 0;
Uint32 surfaceHeight = 0;
Optional<Int> swapInterval;
Bool presentsToAppWindow = false;
Optional<Int> swapchainSwapInterval;
Bool swapchainResizeRequested = false;
Bool presentSuspended = false;
VkSurfaceKHR surface = VK_NULL_HANDLE;
SwapchainObject swapchain;
Vector<VkSemaphore> renderFinishedSemaphores;
Uint imageIndexAcquired = 0;
Uint defaultFramebufferImageIndex = 0;
};
void ExchangeActiveTarget(SurfaceTarget& target);
// Submits the active target's work, consumes its pending acquire and moves it into
// m_parkedTargets. Leaves the renderer with no target.
void ParkActiveTarget();
void OnActiveTargetChanged();
void DestroyParkedTarget(SurfaceTarget& target);
UnorderedMap<Uint64, SurfaceTarget> m_parkedTargets;
Uint64 m_activeTargetKey = 0;
Uint64 m_activeTargetSerial = 1;
Uint64 m_nextTargetSerial = 2;
// Tiered emission for an already-set-up multi-draw batch (state bound, index
// buffer bound for the indexed form). Tier 1: VK_EXT_multi_draw. Tier 2: one
// vkCmdDraw(Indexed)Indirect over a transient command array. Tier 3: unrolled
@@ -513,7 +628,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
};
void QueueClearBufferPayload(GLenum buffer, GLint drawbuffer, const ClearAttachmentPayload& clearPayload);
// See NoteDeviceLoss. Set once, never cleared: a lost VkDevice stays lost.
Bool m_deviceLost = false;
Bool m_deviceLossLatched = false;
#if MOBILEGL_BUILD_DISAGGREGATED
// Brackets this device's frame submissions for the GPU hang watch (LatchIfGpuHung).
GpuProgressMarkers m_progressMarkers;
struct WireImage {
VkImage image = VK_NULL_HANDLE;
VkFormat format = VK_FORMAT_UNDEFINED;
@@ -597,6 +717,94 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// one entry point the import needs (PFN_vkGetAndroidHardwareBufferPropertiesANDROID).
Bool m_wireAhbImport = false;
void* m_wireGetAhbProperties = nullptr;
// YUV shared images (WireYuvImage.inc): the device enabled samplerYcbcrConversion, and the
// conversion pipelines made so far (one per buffer format and colour conversion).
Bool m_samplerYcbcrConversion = false;
struct WireYuvResources;
WireYuvResources* m_wireYuvResources = nullptr;
void DestroyWireYuvResources();
// A YUV texture's first use in a frame converts its image (yuvConvertedFrame against this,
// which every frame boundary - present, publishing flush, native fence - moves).
Uint64 m_yuvFrameSerial = 1;
Bool ConvertYuvSharedImage(VkTextureManager::TextureResource& resource);
// Shared images this session presents into, imported once each and keyed by the image's
// id (monotonic, never reused). `owner` watches the registry's image: an entry whose image
// died is dropped at a later present once `lastSubmit`, its last copy, has completed.
struct SharedImagePresentTarget {
VkTextureManager::ImportedSharedImage image;
WeakPtr<const void> owner;
Uint64 lastSubmit = 0;
// A copy has written the image through this import: what it holds is a frame.
Bool written = false;
};
UnorderedMap<Uint64, SharedImagePresentTarget> m_sharedImagePresentTargets;
void DestroySharedImagePresentTargets(Bool onlyDead);
// SHARED-IMAGE SYNCHRONISATION (WireSharedImage.inc). The sync_file fences other sessions
// publish become binary semaphores: a temporary SYNC_FD import is a WAIT, an exportable
// semaphore a SIGNAL, of the next frame submission - whichever of SubmitPendingCommandBuffer
// and Present's submit comes first, which is the one carrying what was recorded since.
// Each semaphore is recycled once the submission that used it has completed.
// m_sharedImageSyncFd: the device imports and exports SYNC_FD binary semaphores; without
// it the waits happen on the CPU and nothing is exported (see WireSharedImage.inc).
Bool m_sharedImageSyncFd = false;
// The session writes images - it ends frames at publishing flushes
// (PublishSharedImageAccesses), or it reached one through a framebuffer (ResolveWireImage):
// a first acquire waits for the image's pending reads as well, and its boundaries publish
// what it used as written.
Bool m_sharedImageImplicitSync = false;
PFN_vkImportSemaphoreFdKHR m_importSemaphoreFd = nullptr;
PFN_vkGetSemaphoreFdKHR m_getSemaphoreFd = nullptr;
struct SharedImageWait {
VkSemaphore semaphore = VK_NULL_HANDLE;
VkPipelineStageFlags stages = 0;
};
Vector<SharedImageWait> m_sharedImageWaits; // staged for the next submission
VkSemaphore m_sharedImageSignal = VK_NULL_HANDLE; // staged for the next submission
VkSemaphore m_sharedImageSignalSubmitted = VK_NULL_HANDLE; // the last one submitted, to export
struct SharedImageSemaphoreInFlight {
VkSemaphore semaphore = VK_NULL_HANDLE;
Uint64 submitIndex = 0;
Bool exportable = false;
Bool exported = false; // a signal whose sync_file was taken (which unsignals it)
};
Vector<SharedImageSemaphoreInFlight> m_sharedImageSemaphoresInFlight; // submit order
Vector<VkSemaphore> m_freeSharedImageWaitSemaphores;
Vector<VkSemaphore> m_freeSharedImageSignalSemaphores;
// The reading side, per session: the shared-image textures this frame holds (acquired, not
// yet released; keyed as in VkTextureManager's handle map, with the submission that carries
// the acquire), and the frame's ReadTracker, published at Present.
struct SharedImageHold {
Uint64 key = 0;
Uint64 acquireSubmit = 0;
};
Vector<SharedImageHold> m_sharedImagesHeld;
MG_Remote::Server::SharedImages::ReadTracker m_sharedImageReads;
Vector<Uint64> m_sharedImageUseScratch;
// What one vkQueueSubmit carries beyond its own semaphores (AttachSharedImageSync).
struct SharedImageSubmitSync {
Bool attached = false;
Vector<VkSemaphore> waits;
Vector<VkPipelineStageFlags> stages;
Vector<VkSemaphore> signals;
};
void AttachSharedImageSync(VkSubmitInfo& info, SharedImageSubmitSync& storage);
// After the submission `storage` was attached to was submitted and registered.
void CommitSharedImageSync(const SharedImageSubmitSync& storage);
// Takes ownership of `fence` (a sync_file; -1 = nothing to wait for).
void StageSharedImageWait(int fence, VkPipelineStageFlags stages);
VkSemaphore TakeSharedImageSemaphore(Bool exportable);
// The submitted signal's sync_file (-1: it has already signaled). False: not exported.
Bool ExportSharedImageSignal(VkSemaphore semaphore, int& fence);
void RecycleSharedImageSemaphores(Uint64 completedSubmit);
void DestroySharedImageSync();
void AcquireNotedSharedImagesSlow();
// The reading side's frame boundary: release every held image back to the foreign family
// in the open recording, and - after the submission carrying it - publish its fence.
void ReleaseHeldSharedImages();
void PublishSharedImageReads();
// A frame dropped unsubmitted (Present with no usable swapchain).
void AbandonSharedImageReads();
// Runs the probe (memoized per device identity) and sets the member above. Called at the end of
// device creation, after ArmPrimGenReroute: it records on m_graphicsQueue.
void ArmWireDepthResolveOrder();
@@ -696,7 +904,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// P8-SV: glDispatchComputeIndirect on the wire arm - vkCmdDispatchIndirect from the
// dispatch-indirect store, as the monolith arm issues it (WireDraw.inc).
void DispatchWireComputeIndirect(GLintptr offset);
void RewindWireDescriptorSetsIfDue();
// False: the device was lost and the session latched; the caller drops its draw/dispatch.
Bool RewindWireDescriptorSetsIfDue();
UniquePtr<RenderPassEntry> m_wireDrawPass;
Vector<VkImageView> m_wireDrawViews;
VkPipeline GetOrCreatePipelineWithInput(GLenum mode, const MagmaProgramSource& program,
@@ -805,7 +1014,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// a substituted window. (Android is the one exception and has its own path -
// no Mali/Adreno driver seen so far exposes the extension, so a windowless
// context is given an AImageReader ANativeWindow that is never displayed.)
Bool m_headlessSurfaceSupported = true;
// ENABLED ON THIS INSTANCE, not merely offered: a renderer created for a window never asks
// for the extension, and its later offscreen targets must not call an entry point the
// instance does not have (OffscreenSurfaceRoute.h). Hence false until CreateInstance says so.
Bool m_headlessSurfaceSupported = false;
// VK_KHR_android_surface was enabled on this instance (the AImageReader route needs it).
Bool m_androidSurfaceEnabled = false;
// Android has the same shortfall: no Mali/Adreno driver seen so far exposes
// VK_EXT_headless_surface, so a windowless (EGL pbuffer) context gets an
// AImageReader's ANativeWindow to hand the WSI instead. Nothing is ever
@@ -935,17 +1149,17 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkDeviceSize offset, VkBuffer countBuffer,
VkDeviceSize countBufferOffset, Uint32 maxDrawCount,
Uint32 stride);
static inline PFNDrawIndexedIndirectCountFunc s_vkCmdDrawIndexedIndirectCount = nullptr;
PFNDrawIndexedIndirectCountFunc s_vkCmdDrawIndexedIndirectCount = nullptr;
#if MOBILEGL_BUILD_DISAGGREGATED
// P8-D: vkCmdDrawIndirectCount (same signature, same extension), for the wire arm's
// glMultiDrawArraysIndirectCount. The monolith arm never loads it: it reads that count on
// the CPU (DirectVulkan.cpp's MultiDrawArraysIndirectCount).
static inline PFNDrawIndexedIndirectCountFunc s_vkCmdWireDrawIndirectCount = nullptr;
PFNDrawIndexedIndirectCountFunc s_vkCmdWireDrawIndirectCount = nullptr;
#endif
// VK_EXT_multi_draw entry points, loaded at device creation when the extension
// (and its multiDraw feature) is enabled; null otherwise.
static inline PFN_vkCmdDrawMultiEXT s_vkCmdDrawMultiEXT = nullptr;
static inline PFN_vkCmdDrawMultiIndexedEXT s_vkCmdDrawMultiIndexedEXT = nullptr;
PFN_vkCmdDrawMultiEXT s_vkCmdDrawMultiEXT = nullptr;
PFN_vkCmdDrawMultiIndexedEXT s_vkCmdDrawMultiIndexedEXT = nullptr;
// VK_EXT_transform_feedback (GL transform feedback capture)
Bool m_transformFeedbackFeatureEnabled = false;
@@ -976,9 +1190,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// VK_EXT_vertex_attribute_divisor: without it every non-zero glVertexAttribDivisor
// behaves as 1, because that is all Vulkan's instance input rate can express.
Bool m_vertexAttributeDivisorEnabled = false;
static inline PFN_vkCmdBindTransformFeedbackBuffersEXT s_vkCmdBindTransformFeedbackBuffersEXT = nullptr;
static inline PFN_vkCmdBeginTransformFeedbackEXT s_vkCmdBeginTransformFeedbackEXT = nullptr;
static inline PFN_vkCmdEndTransformFeedbackEXT s_vkCmdEndTransformFeedbackEXT = nullptr;
PFN_vkCmdBindTransformFeedbackBuffersEXT s_vkCmdBindTransformFeedbackBuffersEXT = nullptr;
PFN_vkCmdBeginTransformFeedbackEXT s_vkCmdBeginTransformFeedbackEXT = nullptr;
PFN_vkCmdEndTransformFeedbackEXT s_vkCmdEndTransformFeedbackEXT = nullptr;
// Counter buffers (one 4-byte slot per capture binding) let consecutive
// draws within one glBeginTransformFeedback append GL-style. Transform feedback
// objects can each hold an open, paused span at the same time, so the counters are
@@ -1824,7 +2038,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkResult SetupDebugReportCallback();
void DestroyDebugReportCallback();
VkDebugUtilsMessengerCreateInfoEXT PopulateDebugMessengerCreateInfo();
void CreateSurface();
// False (logged) when the target's surface cannot be built; the renderer then has no surface.
Bool CreateSurface();
void PickPhysicalDevice();
void CreateLogicalDeviceAndQueues();
void CreateAllocator();
@@ -378,6 +378,7 @@
begin.renderPass = target->renderPass;
begin.framebuffer = framebuffer;
begin.renderArea = {{0, 0}, destination.extent};
EndActiveRenderPassOn(frame.commandBuffer);
vkCmdBeginRenderPass(frame.commandBuffer, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdBindPipeline(frame.commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, target->pipeline);
const VkViewport viewport{0.0f, 0.0f, static_cast<float>(destination.extent.width),
@@ -126,3 +126,11 @@ MGL_WIRE_DECLINE(MipmapDepthStencilAspect)
// is missing, or the three group-count words lie outside it or are misaligned.
MGL_WIRE_DECLINE(DispatchIndirectBufferUnbound)
MGL_WIRE_DECLINE(DispatchIndirectRange)
// ---- ProgramFactory::ScreenModuleForDriver ------------------------------------------------
// One of the bound program's final modules fails SPIR-V validation, so no shader module of it was
// created (VkProgramObject::invalidSpirvDeclined): SetupWireDraw refuses the draw, and the two
// dispatch arms the dispatch. A driver handed the module could fault the GPU into device loss.
// MOBILEGL_MAGMA_ALLOW_INVALID_SPIRV=1 builds the module anyway, and then neither row moves.
MGL_WIRE_DECLINE(ProgramInvalidSpirv)
MGL_WIRE_DECLINE(ComputeProgramInvalidSpirv)
@@ -299,6 +299,7 @@
begin.renderPass = target->renderPass;
begin.framebuffer = framebuffer;
begin.renderArea = {{0, 0}, destination.extent};
EndActiveRenderPassOn(frame.commandBuffer);
vkCmdBeginRenderPass(frame.commandBuffer, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdBindPipeline(frame.commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, target->pipeline);
const VkViewport viewport{0.0f, 0.0f, static_cast<float>(destination.extent.width),
@@ -11,6 +11,7 @@
// PrimitivesGeneratedNoXfbProbe.cpp): one throwaway command buffer, a bounded fence wait that
// leaks rather than idle-waits a hung GPU, teardown on every other path.
#include "RenderPassGuard.h"
#include "WireDepthResolveProbe.h"
#include <MG_Util/Debug/Log.h>
@@ -237,6 +238,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// The RECTANGLE is the render area, exactly as the render-pass arm's renderArea is - the
// scratch outside it is never read.
begin.renderArea = rect;
EndActiveRenderPassOn(commandBuffer);
vkCmdBeginRenderPass(commandBuffer, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdBindPipeline(commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
// THE VIEWPORT IS THE WHOLE ATTACHMENT AND THE SCISSOR IS THE RECT, and the two are not
@@ -746,6 +748,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
begin.renderPass = clear.pass;
begin.framebuffer = clear.framebuffer;
begin.renderArea = fullRect;
EndActiveRenderPassOn(cmd);
vkCmdBeginRenderPass(cmd, &begin, VK_SUBPASS_CONTENTS_INLINE);
VkClearAttachment attachment{};
attachment.aspectMask = clear.aspects;
@@ -772,6 +775,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
begin.renderPass = item.resolvePass;
begin.framebuffer = item.resolveFramebuffer;
begin.renderArea = fullRect;
EndActiveRenderPassOn(cmd);
vkCmdBeginRenderPass(cmd, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdEndRenderPass(cmd);
}
@@ -215,7 +215,10 @@
MakeXfbWritesVisible();
const auto* fbo = state.DrawFramebuffer();
if (!fbo) MagmaWireFatal("draw-framebuffer-record");
RetireWireDrawPass();
// THE PREVIOUS DRAW'S RENDER PASS STAYS OPEN until this draw's attachments are known: a
// draw to the same attachments continues it (below), and on a tiler a pass per draw also
// loaded and stored every attachment once per draw. Resource work in between that may not
// sit inside a pass (a barrier, a copy, a clear) closes it itself (RenderPassGuard.h).
// CPU preparation may synchronize GPU-produced vertex/index bytes. Do
// not rewind an arena while this draw owns as-yet-unrecorded slices.
WirePreparationScope preparing(m_wirePreparationDepth);
@@ -230,6 +233,10 @@
flags |= ProgramFactory::CompileOptionBit::ZeroBaseVertex;
programObject = &m_programFactory->GetOrCreateProgram(program, flags);
}
if (programObject->invalidSpirvDeclined) {
MGL_WIRE_DECLINE_AT(ProgramInvalidSpirv, "a module of the program fails SPIR-V validation and was never built");
return false;
}
if (programObject->pointSizeCapabilityUnsupported) {
MGL_WIRE_DECLINE_AT(PointSizeCapability, "the program writes gl_PointSize and the device cannot express it");
return false;
@@ -240,6 +247,9 @@
MGL_WIRE_DECLINE_AT(TextureResourcesUnprepared, "a sampled/storage image could not be made ready; see the Tex* tally");
return false;
}
// Shared images another session writes: acquired ahead of the recording that samples them
// (WireSharedImage.inc).
AcquireNotedSharedImages();
// Lazy image uploads/promotions can submit the old recording. Resolve
// ALL attachments before descriptor binding captures the current cmd.
for (Uint32 i = 0; i < MG_Pipe::kMGPipeMaxColorAttachments; ++i) {
@@ -306,8 +316,20 @@
"MGPipe: Fatal{ProtocolCorruption, \"Magma:client-vertex-array\"} location=%u: an "
"attribute the program reads carries no buffer handle", location);
BufferSlice slice{};
if (!m_bufferManager.AcquireWireSlice(BufferKind::Vertex, resource.Res, slice))
MagmaWireFatal("vertex-buffer-resource");
if (!m_bufferManager.AcquireWireSlice(BufferKind::Vertex, resource.Res, slice)) {
// A buffer with no data store is legal GL to bind (glGenBuffers + glBindBuffer, no
// glBufferData), and the client mints its handle at construction while the create
// only crosses with the first respecify (PipeFill.cpp's M-1) - so the handle may be
// one this server never saw, or one it holds with zero bytes. Either way the draw
// reads no storage: it is declined, as the GL draws nothing defined from it.
static Bool s_described = false;
if (!s_described && !m_bufferManager.IsKnownWireBuffer(resource.Res)) {
s_described = true;
m_bufferManager.DescribeWireBufferHistory(resource.Res);
}
MGL_WIRE_DECLINE_AT(VertexAttributeWindow, "an enabled attribute's buffer has no storage");
return false;
}
const VkDeviceSize offset = input.bindingBaseOffsets[binding];
const auto type = static_cast<DataType>(attr.Type);
const SizeT elementSize = VertexInputStateFactory::GetAttributeByteSize(type, attr.Size, attr.IsBgra);
@@ -387,8 +409,16 @@
const void* bytes = nullptr;
if (cpuIndices) bytes = reinterpret_cast<const void*>(indices->indexByteOffset);
else {
if (!m_bufferManager.AcquireWireSlice(BufferKind::Index, state.IndexBuffer.Res, indexSlice))
MagmaWireFatal("index-buffer-resource");
if (!m_bufferManager.AcquireWireSlice(BufferKind::Index, state.IndexBuffer.Res, indexSlice)) {
// As for a vertex buffer: a bound element buffer with no data store yet.
static Bool s_described = false;
if (!s_described && !m_bufferManager.IsKnownWireBuffer(state.IndexBuffer.Res)) {
s_described = true;
m_bufferManager.DescribeWireBufferHistory(state.IndexBuffer.Res);
}
MGL_WIRE_DECLINE_AT(IndexWindow, "the element buffer has no storage");
return false;
}
const Uint64 offset = state.IndexBuffer.Offset + indices->indexByteOffset;
if (offset < state.IndexBuffer.Offset || offset > indexSlice.size ||
indices->indexByteSize > indexSlice.size - offset) {
@@ -463,6 +493,7 @@
// attachments below resolve the same index, so this key has to be taken after it.
passKey.swapchainImageIndex = DefaultFramebufferWriteIndex(true);
}
Vector<WireImage> sameLayoutAttachments;
const auto add = [&](const MG_Pipe::MGPSurface& surface, VkImageAspectFlags aspect) -> Uint32 {
WireImage image = ResolveWireImage(*fbo, surface, aspect, /*isWriteTarget=*/true);
if (!image.image) return VK_ATTACHMENT_UNUSED;
@@ -472,7 +503,10 @@
MGB_CTX->IsCapabilityEnabled(CapabilityInput::FramebufferSrgb));
const VkImageLayout layout = aspect == VK_IMAGE_ASPECT_COLOR_BIT ?
VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_GENERAL;
TransitionWireImage(image, layout);
// Already in its attachment layout: its barrier (which closes any open pass) waits until
// it is known whether the open pass continues with this very attachment.
if (image.layout == layout) sameLayoutAttachments.push_back(image);
else TransitionWireImage(image, layout);
VkImageViewCreateInfo info{VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO};
info.image = image.image;
info.viewType = image.viewType;
@@ -538,7 +572,17 @@
passInfo.pAttachments = attachments.data();
passInfo.subpassCount = 1;
passInfo.pSubpasses = &subpass;
if (!AcquireCachedWireDrawPass(passKey)) {
const Bool continuing = m_wireDrawPass && m_wireDrawPassKey == passKey;
{
const auto* active = VkRenderPassManager::GetActiveRenderPass();
const Bool stillOpen = continuing && active != nullptr && active->commandBuffer == frame.commandBuffer;
// A new pass orders itself after earlier writes to its attachments; a continued one
// is the same pass, whose attachment writes are ordered by rasterization.
if (!stillOpen)
for (auto& image : sameLayoutAttachments) TransitionWireImage(image, VK_IMAGE_LAYOUT_GENERAL);
}
if (!continuing) RetireWireDrawPass();
if (!continuing && !AcquireCachedWireDrawPass(passKey)) {
m_wireDrawPassKey = Move(passKey);
Vector<VkImageView> views;
views.reserve(viewInfos.size());
@@ -602,13 +646,20 @@
vkCmdBindVertexBuffers(frame.commandBuffer, binding++, 1, &slice.buffer, &slice.offset);
}
if (indexSlice.IsValid()) vkCmdBindIndexBuffer(frame.commandBuffer, indexSlice.buffer, indexSlice.offset, indexType);
if (!VkRenderPassManager::BeginRenderPass(frame.commandBuffer, pass)) {
// Still open from the previous draw (nothing closed it, same command buffer): continue.
const auto* active = VkRenderPassManager::GetActiveRenderPass();
const Bool open = continuing && active != nullptr && active->commandBuffer == frame.commandBuffer &&
active->hash == pass.hash;
if (!open && active != nullptr) VkRenderPassManager::EndRenderPass(frame.commandBuffer);
if (!open && !VkRenderPassManager::BeginRenderPass(frame.commandBuffer, pass)) {
MGL_WIRE_DECLINE_AT(BeginRenderPass, "the render pass for this draw could not be begun");
return false;
}
vkCmdBindPipeline(frame.commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
ApplyDynamicDrawStateTail(frame, pass.extent, fbo->IsDefault,
ResolveDrawViewportCount(programObj.writesViewportIndexBuiltin));
// Storage images this draw writes are what the next sampled-image barrier must cover.
if (programObj.hasStorageImages) BumpWireImageWriteEpoch();
if (MG_Util::PipeStats::Enabled()) MG_Util::PipeStats::AddCalls(MG_Util::PipeStats::CallClass::Draws, 1);
return true;
}
@@ -752,10 +803,15 @@
return;
}
const auto& programObj = m_programFactory->GetOrCreateProgram(program, {});
if (programObj.invalidSpirvDeclined) {
MGL_WIRE_DECLINE_AT(ComputeProgramInvalidSpirv, "a module of the compute program fails SPIR-V validation and was never built");
return;
}
if (!m_uniformManager->PrepareWireTextureResources(program, programObj)) {
MGL_WIRE_DECLINE_AT(ComputeTextureResourcesUnprepared, "a sampled/storage image could not be made ready for the dispatch");
return;
}
AcquireNotedSharedImages(); // shared images another session writes (WireSharedImage.inc)
const auto pipeline = GetOrCreateComputePipeline(programObj);
if (!pipeline) {
MGL_WIRE_DECLINE_AT(ComputePipelineUnavailable, "no compute pipeline could be created for this dispatch");
@@ -768,6 +824,7 @@
return;
}
vkCmdBindPipeline(frame.commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
EndActiveRenderPassOn(frame.commandBuffer);
vkCmdDispatch(frame.commandBuffer, x, y, z);
}
@@ -823,10 +880,15 @@
return;
}
const auto& programObj = m_programFactory->GetOrCreateProgram(program, {});
if (programObj.invalidSpirvDeclined) {
MGL_WIRE_DECLINE_AT(ComputeProgramInvalidSpirv, "a module of the compute program fails SPIR-V validation and was never built");
return;
}
if (!m_uniformManager->PrepareWireTextureResources(program, programObj)) {
MGL_WIRE_DECLINE_AT(ComputeTextureResourcesUnprepared, "a sampled/storage image could not be made ready for the dispatch");
return;
}
AcquireNotedSharedImages(); // shared images another session writes (WireSharedImage.inc)
const auto pipeline = GetOrCreateComputePipeline(programObj);
if (!pipeline) {
MGL_WIRE_DECLINE_AT(ComputePipelineUnavailable, "no compute pipeline could be created for this dispatch");
@@ -841,6 +903,7 @@
return;
}
vkCmdBindPipeline(dispatchFrame.commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
EndActiveRenderPassOn(dispatchFrame.commandBuffer);
vkCmdDispatchIndirect(dispatchFrame.commandBuffer, parameters.buffer,
parameters.offset + static_cast<VkDeviceSize>(offset));
if (MG_Util::PipeStats::Enabled())
@@ -160,6 +160,13 @@
if (MG_Pipe::MGPipeHandleIsNull(out.storage)) MagmaWireFatal("framebuffer-view-record");
auto* resource = m_textureManager->SyncTextureResourceByHandle(out.storage, renderbuffer);
if (!resource) MagmaWireFatal("framebuffer-resource");
// Another session writes it: acquired (WireSharedImage.inc) before the layout is read below.
// An image reached through a framebuffer is one this session renders into (or copies
// through), which makes it a writer of images from now on.
if (resource->sharedImageId != 0) {
m_sharedImageImplicitSync = true;
AcquireSharedImage(*resource);
}
out.image = resource->image;
out.format = viewFormat != VK_FORMAT_UNDEFINED ? viewFormat : resource->format;
out.extent = {std::max(resource->extent.width >> level, 1u),
@@ -226,6 +233,13 @@
void VulkanRenderer::ClearWireFramebuffer(const MG_Pipe::MGPFramebufferState& fbo,
const ClearAttachmentPayload& payload, GLint drawbuffer) {
RetireWireDrawPass();
// A stream that neither presents nor finishes (an off-screen loop) never submits, so its
// recording and the per-clear objects retired below would grow until host memory runs
// out. Past a bound, submit what is recorded: the objects then retire with that work.
constexpr SizeT kRetiredObjectsBeforeSubmit = 1024;
if (m_wireRetiredObjects.size() >= kRetiredObjectsBeforeSubmit && m_wirePreparationDepth == 0 &&
HasPendingRecordedWork())
(void)FlushPendingCommands();
if (MGB_CTX->IsCapabilityEnabled(CapabilityInput::RasterizerDiscard)) return;
const auto clearSurface = [&](const MG_Pipe::MGPSurface& surface, VkImageAspectFlags aspect,
Uint32 colorIndex) {
@@ -312,6 +326,7 @@
if (!color) clear.clearValue.depthStencil = {payload.depth, payload.stencil};
VkClearRect rect{area, 0, image.layers};
auto& frame = m_frameContext.GetCurrent();
EndActiveRenderPassOn(frame.commandBuffer);
vkCmdBeginRenderPass(frame.commandBuffer, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdClearAttachments(frame.commandBuffer, 1, &clear, 1, &rect);
vkCmdEndRenderPass(frame.commandBuffer);
@@ -379,7 +394,13 @@
void VulkanRenderer::ReadWirePixels(GLint x, GLint y, GLsizei width, GLsizei height,
GLenum format, GLenum type, void* pixels) {
if (width <= 0 || height <= 0) return;
if (HasPendingRecordedWork() && !FlushPendingCommands()) MagmaWireFatal("wire-verb-flush");
if (MG_Pipe::MGPipeDebugDeviceLossDue()) NoteDeviceLoss(VK_ERROR_DEVICE_LOST, "the debug knob (read-pixels)");
if (m_deviceLost && LatchWireDeviceLoss("read-pixels")) return;
if (HasPendingRecordedWork() && !FlushPendingCommands()) {
// A lost device: the session is latched and closing, the pixels stay unfilled.
if (LatchWireDeviceLoss("wire-verb-flush")) return;
MagmaWireFatal("wire-verb-flush");
}
const auto* fbo = MG_Pipe::MGPipeApplier().ReadFramebuffer();
if (!fbo) MagmaWireFatal("read-framebuffer-record");
VkImageAspectFlags aspect = format == GL_DEPTH_COMPONENT ? VK_IMAGE_ASPECT_DEPTH_BIT :
@@ -400,6 +421,22 @@
"default framebuffer; the readback returns its raw pixels",
source.swapchainImageIndex);
}
if (source.isDefault && aspect == VK_IMAGE_ASPECT_COLOR_BIT) {
VkOffset2D probeOffset{};
VkExtent2D probeExtent{};
if (!MapDefaultFramebufferReadbackRect(x, y, width, height, source.extent,
m_swapchainObject.GetPreTransform(), &probeOffset, &probeExtent)) {
// A rectangle reaching past the surface: GL leaves those pixels undefined rather
// than raising an error, and the usual cause is a client reading back at its
// window's new size one frame before its surface follows the resize. Nothing is
// copied, and nothing is recorded.
MGLOG_E_ONCE("Magma wire: default-framebuffer read %dx%d at (%d,%d) does not fit surface %llx's "
"%ux%u (transform %u); the read is left unfilled", width, height, x, y,
static_cast<unsigned long long>(m_activeTargetKey), source.extent.width,
source.extent.height, static_cast<Uint32>(m_swapchainObject.GetPreTransform()));
return;
}
}
m_textureManager->FlushPendingUploads();
RecordWireReadbackWriteBarrier();
TransitionWireImage(source, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL);
@@ -428,7 +465,10 @@
auto& frame = m_frameContext.GetCurrent();
vkCmdCopyImageToBuffer(frame.commandBuffer,source.image,VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
buffer.GetHandle(),1,&region);
if (!SubmitReadbackCommandsAndWait(frame)) MagmaWireFatal("read-submit");
if (!SubmitReadbackCommandsAndWait(frame)) {
if (LatchWireDeviceLoss("read-submit")) return;
MagmaWireFatal("read-submit");
}
const auto* bytes = static_cast<const Uint8*>(buffer.Map());
if (!bytes || !buffer.Invalidate(size)) MagmaWireFatal("read-map");
Vector<Uint8> remapped;
@@ -798,6 +838,7 @@
begin.renderPass = pass;
begin.framebuffer = framebuffer;
begin.renderArea = {{sxMin, syMin}, {static_cast<Uint32>(width), static_cast<Uint32>(height)}};
EndActiveRenderPassOn(m_frameContext.GetCurrent().commandBuffer);
vkCmdBeginRenderPass(m_frameContext.GetCurrent().commandBuffer, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdEndRenderPass(m_frameContext.GetCurrent().commandBuffer);
} else if (!resolvedByShader) {
@@ -1374,6 +1415,14 @@
&viewFormat, &layerCount);
auto* resource = m_textureManager->SyncTextureResourceByHandle(handle);
if (!resource) MagmaWireFatal("mipmap-resource");
// A shared image is one level of a buffer someone else allocated; growing the chain would
// reallocate it away from the buffer.
if (resource->sharedImageId != 0) {
MGLOG_E_ONCE("DirectVulkan::GenerateWireMipmap: the texture's level 0 is shared image %llu, which has "
"no mip chain; levels %u..%u are left as they are",
static_cast<unsigned long long>(resource->sharedImageId), logicalBase + 1, logicalEnd - 1);
return;
}
m_textureManager->FlushPendingUploads();
RetireWireDrawPass();
auto& frame = m_frameContext.GetCurrent();
@@ -215,6 +215,7 @@
VkRenderPassBeginInfo begin{VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO};
begin.renderPass = image.clearPass; begin.framebuffer = image.clearFramebuffer;
begin.renderArea.extent = {1, 1}; begin.clearValueCount = 1; begin.pClearValues = &clear;
EndActiveRenderPassOn(commands);
vkCmdBeginRenderPass(commands, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdEndRenderPass(commands);
image.layout = VK_IMAGE_LAYOUT_GENERAL;
@@ -226,6 +227,7 @@
const VkImageSubresourceRange range{image.aspect, 0, 1, 0, image.layers};
if (shadow) {
const VkClearDepthStencilValue depth{0.0f, 0};
EndActiveRenderPassOn(commands);
vkCmdClearDepthStencilImage(commands, image.image, image.layout, &depth, 1, &range);
} else vkCmdClearColorImage(commands, image.image, image.layout, &clear.color, 1, &range);
if (!VkTextureManager::TransitionImageLayout(commands, image.image, image.layout,
@@ -236,6 +238,7 @@
VkMemoryBarrier visible{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
visible.srcAccessMask = multisample ? VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT : VK_ACCESS_TRANSFER_WRITE_BIT;
visible.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT;
EndActiveRenderPassOn(commands);
vkCmdPipelineBarrier(commands, multisample ? VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT : VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, 1, &visible, 0, nullptr, 0, nullptr);
}
@@ -0,0 +1,718 @@
// MobileGL - MobileGL/MG_Backend/DirectVulkan/Renderer/WireSharedImage.inc
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
// SHARED IMAGES (docs/Disaggregated/notes/anland/plan-ahb-dmabuf.md). Included by VulkanRenderer.cpp
// inside the backend namespace, after WireFramebuffer.inc, whose readback the present side's copy
// is modelled on: the same flush, the same write-visibility barrier, the same transition of the
// default framebuffer's image.
//
// OWNERSHIP. The image lives on more than one device: a client session writes it, the
// compositor's session (another VulkanRenderer, another VkDevice) samples it. Each present
// acquires it from VK_QUEUE_FAMILY_FOREIGN_EXT (UNDEFINED: the copy overwrites every texel) and
// releases it back in GENERAL. A reader acquires it (GENERAL) before the first use of a frame and
// again whenever the write generation moved, and releases it back at its frame boundary, so every
// acquire on either side matches a release on the other.
//
// ORDER (MG_Remote/Server/SharedImageRegistry.h). Every fence is a sync_file. The writer's copy
// waits on its GPU for the read fences still pending, signals a semaphore exported as the write
// fence, and the present returns without waiting. A reader's use waits on its GPU for the write
// fence of the generation it acquires; its frame's submission signals a semaphore exported as the
// read fence of every image the frame used. A device without SYNC_FD semaphore import and export
// waits on the CPU instead where it would have waited on the GPU - before the writer's copy, after
// it, and before the reader's use - and exports nothing for its reads: there is no fence to give,
// and waiting for the frame on the CPU only to say it is over would stall every frame of the
// compositor. The writer's next copy is then ordered after the reads by nothing but the client's
// own buffer rotation (a Wayland client writes a buffer only after its release) - which is all
// this path had before - on a device Android's Vulkan 1.1 requirements leave rare.
#if MOBILEGL_BUILD_DISAGGREGATED
// A CPU wait on another session's fence is a fallback, and bounded: a peer that never signals
// costs a frame's worth of stall, not a hang.
static constexpr Uint32 kSharedImageCpuWaitMs = 1000;
// Every stage a reader's use of a shared image can reach: sampling (draws and dispatches), a
// blit or readback source, an attachment. The write fence's wait blocks these, and the
// acquire's first scope is the same set, so it is ordered after that wait.
static constexpr VkPipelineStageFlags kSharedImageUseStages =
VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT |
VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
static void CloseSyncFile(int fd) {
#if !defined(_WIN32)
if (fd >= 0) ::close(fd);
#else
(void)fd;
#endif
}
void VulkanRenderer::DestroySharedImagePresentTargets(Bool onlyDead) {
if (onlyDead) RefreshCompletedSubmits();
for (auto it = m_sharedImagePresentTargets.begin(); it != m_sharedImagePresentTargets.end();) {
// A dead image's last copy may still be executing: the present no longer waits for it.
if (onlyDead && (!it->second.owner.expired() || it->second.lastSubmit > m_completedSubmitCounter)) {
++it;
continue;
}
VkTextureManager::DestroyImportedSharedImage(m_device, it->second.image);
it = m_sharedImagePresentTargets.erase(it);
}
}
// A copy moves bytes; a blit converts through the format. The bytes are right only between
// formats with one memory layout, which for the RGBA8 buffers means RGBA8 itself or its sRGB
// twin (a blit would decode the sRGB bytes on the way).
static Bool SharedImageCopiesAsIs(VkFormat source, VkFormat destination) {
if (source == destination) return true;
return destination == VK_FORMAT_R8G8B8A8_UNORM && source == VK_FORMAT_R8G8B8A8_SRGB;
}
Bool VulkanRenderer::BlitDefaultFramebufferToSharedImage(const SharedImageView& view,
const MG_Util::Damage::Region& region) {
if (m_device == VK_NULL_HANDLE || view.Width == 0 || view.Height == 0) return false;
// The debug knob's device check (MGPipeDebugDeviceLossDue) rides the frame boundary here.
if (MG_Pipe::MGPipeDebugDeviceLossDue()) NoteDeviceLoss(VK_ERROR_DEVICE_LOST, "the debug knob (shared-image present)");
if (m_deviceLost && LatchWireDeviceLoss("shared-image-present")) return false;
DestroySharedImagePresentTargets(true);
// The default framebuffer's image is stored in display orientation (row 0 is the top of
// the frame) on an identity surface, and turned half round on a 180 one; a blit expresses
// both. A quarter turn transposes, which no copy or blit can - and only the compositor's
// own window is ever quarter-turned, which presents to the display rather than here.
const VkSurfaceTransformFlagBitsKHR preTransform = m_swapchainObject.GetPreTransform();
if (IsQuarterTurnPreTransform(preTransform)) {
MGLOG_E_ONCE("Magma: shared image %llu: the current surface %llx is quarter-turned (transform %u); a "
"present into a shared image is declined there",
static_cast<unsigned long long>(view.Id), static_cast<unsigned long long>(m_activeTargetKey),
static_cast<Uint32>(preTransform));
return false;
}
auto found = m_sharedImagePresentTargets.find(view.Id);
if (found == m_sharedImagePresentTargets.end()) {
const auto owner = MG_Remote::Server::SharedImages::Find(view.Id);
if (owner == nullptr) {
MGLOG_E_ONCE("Magma: shared image %llu is not live; the present is declined",
static_cast<unsigned long long>(view.Id));
return false;
}
SharedImagePresentTarget target;
String why;
if (!VkTextureManager::ImportSharedImage(view.NativeBuffer, view.Width, view.Height, target.image, why)) {
MGLOG_E_ONCE("Magma: shared image %llu (%ux%u) not imported - %s; the present is declined",
static_cast<unsigned long long>(view.Id), view.Width, view.Height, why.c_str());
return false;
}
target.owner = owner;
found = m_sharedImagePresentTargets.emplace(view.Id, Move(target)).first;
}
const VkTextureManager::ImportedSharedImage& destination = found->second.image;
// The registry's record carries the image's fences. The server holds a reference for as
// long as the present runs; a lapsed one is an image released under it.
const auto image =
std::static_pointer_cast<const MG_Remote::Server::SharedImages::Image>(found->second.owner.lock());
if (image == nullptr) {
MGLOG_E_ONCE("Magma: shared image %llu is not live; the present is declined",
static_cast<unsigned long long>(view.Id));
return false;
}
// The frame's draws reach the queue first; the copy below is a later submission that the
// readback barrier orders behind them.
if (HasPendingRecordedWork() && !FlushPendingCommands()) {
MGLOG_E("Magma: shared image %llu: the frame's recorded work could not be submitted",
static_cast<unsigned long long>(view.Id));
return false;
}
MG_Pipe::MGPFramebufferState defaultFramebuffer{};
defaultFramebuffer.IsDefault = 1;
const MG_Pipe::MGPSurface noSurface{};
WireImage source = ResolveWireImage(defaultFramebuffer, noSurface, VK_IMAGE_ASPECT_COLOR_BIT,
/*isWriteTarget=*/false);
if (source.image == VK_NULL_HANDLE || source.extent.width == 0 || source.extent.height == 0) {
MGLOG_E_ONCE("Magma: shared image %llu: no default framebuffer is current; the present is declined",
static_cast<unsigned long long>(view.Id));
return false;
}
if (!m_swapchainObject.IsImageContentDefined(source.swapchainImageIndex)) {
MGLOG_W_ONCE("Magma: shared image present: no write is recorded for image %u of the default "
"framebuffer; its raw pixels are copied",
source.swapchainImageIndex);
}
const Bool sameExtent = source.extent.width == view.Width && source.extent.height == view.Height;
const Bool halfTurn = preTransform == VK_SURFACE_TRANSFORM_ROTATE_180_BIT_KHR;
const Bool copy = sameExtent && !halfTurn && SharedImageCopiesAsIs(source.format, destination.format);
// ONLY WHERE THE IMAGE DIFFERS FROM THE FRAME: an image this renderer wrote before, copied
// one to one (no scaling, no half turn). Its other texels must survive, so the acquire then
// keeps the layout the last reader released it in rather than discarding it.
MG_Util::Damage::Region part = region;
if (!part.IsFull()) part.Normalize(static_cast<Int32>(view.Width), static_cast<Int32>(view.Height));
const Bool partial = !part.IsFull() && found->second.written && sameExtent && !halfTurn;
// The region's rectangles, top row first and made disjoint: a copy's regions may not overlap.
Vector<MG_Util::Damage::Rect> rows;
if (partial) {
for (const MG_Util::Damage::Rect& rect : MG_Util::Damage::Disjoint(part.Rects()))
rows.push_back(MG_Util::Damage::FlipY(rect, static_cast<Int32>(view.Height)));
}
VkFormatProperties sourceProperties{};
vkGetPhysicalDeviceFormatProperties(m_physicalDevice.handle, source.format, &sourceProperties);
if (!copy) {
if ((sourceProperties.optimalTilingFeatures & VK_FORMAT_FEATURE_BLIT_SRC_BIT) == 0 ||
(destination.features & VK_FORMAT_FEATURE_BLIT_DST_BIT) == 0) {
MGLOG_E_ONCE("Magma: shared image %llu: a %ux%u format-%d default framebuffer needs a blit into "
"the %ux%u format-%d image, which this device cannot do; the present is declined",
static_cast<unsigned long long>(view.Id), source.extent.width, source.extent.height,
static_cast<Int>(source.format), view.Width, view.Height,
static_cast<Int>(destination.format));
return false;
}
if (vkuFormatIsSRGB(source.format) && !vkuFormatIsSRGB(destination.format)) {
MGLOG_W_ONCE("Magma: shared image present: the sRGB default framebuffer is blitted into a UNORM "
"image, which stores linear values");
}
}
m_textureManager->FlushPendingUploads();
RecordWireReadbackWriteBarrier();
TransitionWireImage(source, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL);
auto& frame = m_frameContext.GetCurrent();
// NOT OVER WHAT IS STILL READ. The reads published since this image's last write (a
// compositor frame sampling it) end before the copy starts: on the GPU, as waits of the
// submission below at the transfer stage; on the CPU, where they cannot be imported.
// The frame's draws were flushed above, so the next submission is the copy's.
namespace SI = MG_Remote::Server::SharedImages;
if (m_sharedImageSyncFd) {
for (const int fence : SI::DupPendingReadFences(*image))
StageSharedImageWait(fence, VK_PIPELINE_STAGE_TRANSFER_BIT);
} else if (!SI::WaitForReads(*image, kSharedImageCpuWaitMs)) {
MGLOG_W_ONCE("Magma: shared image %llu: a read of the previous frame did not finish within %u ms; it is "
"overwritten anyway",
static_cast<unsigned long long>(view.Id), kSharedImageCpuWaitMs);
}
// The acquire's layout transition runs after those waits: its first scope is the stage
// they block.
VkImageMemoryBarrier acquire{VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER};
acquire.srcAccessMask = 0;
acquire.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
// Every present releases the image in GENERAL, and so does every reader.
acquire.oldLayout = partial ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED;
acquire.newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
acquire.srcQueueFamilyIndex = VK_QUEUE_FAMILY_FOREIGN_EXT;
acquire.dstQueueFamilyIndex = destination.queueFamily;
acquire.image = destination.image;
acquire.subresourceRange = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1};
vkCmdPipelineBarrier(frame.commandBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
0, nullptr, 0, nullptr, 1, &acquire);
const VkImageSubresourceLayers layers{VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, 1};
if (partial) {
// An empty region copies nothing: the image already is the frame.
if (copy && !rows.empty()) {
Vector<VkImageCopy> regions;
regions.reserve(rows.size());
for (const MG_Util::Damage::Rect& row : rows) {
VkImageCopy one{};
one.srcSubresource = layers;
one.dstSubresource = layers;
one.srcOffset = {row.X, row.Y, 0};
one.dstOffset = {row.X, row.Y, 0};
one.extent = {static_cast<Uint32>(row.Width), static_cast<Uint32>(row.Height), 1};
regions.push_back(one);
}
vkCmdCopyImage(frame.commandBuffer, source.image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
destination.image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
static_cast<Uint32>(regions.size()), regions.data());
} else if (!rows.empty()) {
Vector<VkImageBlit> regions;
regions.reserve(rows.size());
for (const MG_Util::Damage::Rect& row : rows) {
VkImageBlit one{};
one.srcSubresource = layers;
one.dstSubresource = layers;
one.srcOffsets[0] = {row.X, row.Y, 0};
one.srcOffsets[1] = {row.X + row.Width, row.Y + row.Height, 1};
one.dstOffsets[0] = one.srcOffsets[0];
one.dstOffsets[1] = one.srcOffsets[1];
regions.push_back(one);
}
vkCmdBlitImage(frame.commandBuffer, source.image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
destination.image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
static_cast<Uint32>(regions.size()), regions.data(), VK_FILTER_NEAREST);
}
} else if (copy) {
VkImageCopy region{};
region.srcSubresource = layers;
region.dstSubresource = layers;
region.extent = {view.Width, view.Height, 1};
vkCmdCopyImage(frame.commandBuffer, source.image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, destination.image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &region);
} else {
// Identity: row 0 is already the top. 180: the stored image is the frame turned half
// round, so the destination rectangle runs backwards on both axes.
const Int32 width = static_cast<Int32>(view.Width);
const Int32 height = static_cast<Int32>(view.Height);
VkImageBlit region{};
region.srcSubresource = layers;
region.srcOffsets[1] = {static_cast<Int32>(source.extent.width), static_cast<Int32>(source.extent.height),
1};
region.dstSubresource = layers;
region.dstOffsets[0] = halfTurn ? VkOffset3D{width, height, 0} : VkOffset3D{0, 0, 0};
region.dstOffsets[1] = halfTurn ? VkOffset3D{0, 0, 1} : VkOffset3D{width, height, 1};
const VkFilter filter =
!sameExtent && (sourceProperties.optimalTilingFeatures &
VK_FORMAT_FEATURE_SAMPLED_IMAGE_FILTER_LINEAR_BIT) != 0
? VK_FILTER_LINEAR
: VK_FILTER_NEAREST;
vkCmdBlitImage(frame.commandBuffer, source.image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, destination.image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &region, filter);
}
VkImageMemoryBarrier release{VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER};
release.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
release.dstAccessMask = 0;
release.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
release.newLayout = VK_IMAGE_LAYOUT_GENERAL;
release.srcQueueFamilyIndex = destination.queueFamily;
release.dstQueueFamilyIndex = VK_QUEUE_FAMILY_FOREIGN_EXT;
release.image = destination.image;
release.subresourceRange = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1};
vkCmdPipelineBarrier(frame.commandBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT,
0, 0, nullptr, 0, nullptr, 1, &release);
// THE WRITE FENCE. The client attaches the buffer to its surface as soon as this returns;
// what orders the compositor's read after the copy is the fence published here, which the
// compositor's session waits for on its GPU when it next acquires the image. The copy's
// submission signals a semaphore exported as that fence, and nothing waits for it here.
// The submission is an ordinary pooled-fence flush: the renderer's in-flight tracking keeps
// the command buffer and every resource it names alive until it has completed.
if (m_sharedImageSyncFd && m_sharedImageSignal == VK_NULL_HANDLE)
m_sharedImageSignal = TakeSharedImageSemaphore(/*exportable=*/true);
if (m_sharedImageSignal != VK_NULL_HANDLE) {
m_sharedImageSignalSubmitted = VK_NULL_HANDLE;
if (!FlushPendingCommands()) {
MGLOG_E("Magma: shared image %llu: the copy could not be submitted",
static_cast<unsigned long long>(view.Id));
return false;
}
found->second.lastSubmit = m_submitCounter;
found->second.written = true;
const VkSemaphore signal = std::exchange(m_sharedImageSignalSubmitted, VK_NULL_HANDLE);
int fence = -1;
if (signal != VK_NULL_HANDLE && ExportSharedImageSignal(signal, fence)) {
SI::PublishWrite(*image, fence);
return true;
}
// Submitted, but no fence came out of it: the copy is waited for, as on a device that
// exports none.
if (!WaitForSubmitsUpTo(m_submitCounter, UINT64_MAX)) {
MGLOG_E("Magma: shared image %llu: the copy could not be waited for",
static_cast<unsigned long long>(view.Id));
return false;
}
SI::PublishWrite(*image, -1);
return true;
}
if (!SubmitReadbackCommandsAndWait(frame)) {
MGLOG_E("Magma: shared image %llu: the copy could not be submitted or waited for",
static_cast<unsigned long long>(view.Id));
return false;
}
found->second.lastSubmit = m_submitCounter;
found->second.written = true;
SI::PublishWrite(*image, -1);
return true;
}
// ---- fences as semaphores ------------------------------------------------------------------
VkSemaphore VulkanRenderer::TakeSharedImageSemaphore(Bool exportable) {
auto& pool = exportable ? m_freeSharedImageSignalSemaphores : m_freeSharedImageWaitSemaphores;
if (!pool.empty()) {
const VkSemaphore semaphore = pool.back();
pool.pop_back();
return semaphore;
}
VkExportSemaphoreCreateInfo exportInfo{VK_STRUCTURE_TYPE_EXPORT_SEMAPHORE_CREATE_INFO};
exportInfo.handleTypes = VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT;
VkSemaphoreCreateInfo info{VK_STRUCTURE_TYPE_SEMAPHORE_CREATE_INFO};
if (exportable) info.pNext = &exportInfo;
VkSemaphore semaphore = VK_NULL_HANDLE;
const VkResult result = vkCreateSemaphore(m_device, &info, nullptr, &semaphore);
if (result != VK_SUCCESS) {
MGLOG_E_ONCE("Magma: a shared-image semaphore could not be created (vkCreateSemaphore -> %d)",
static_cast<Int>(result));
return VK_NULL_HANDLE;
}
return semaphore;
}
void VulkanRenderer::StageSharedImageWait(int fence, VkPipelineStageFlags stages) {
if (fence < 0) return;
if (m_sharedImageSyncFd) {
const VkSemaphore semaphore = TakeSharedImageSemaphore(/*exportable=*/false);
if (semaphore != VK_NULL_HANDLE) {
// A sync_file import is always temporary: the wait consumes it, and the semaphore
// is a plain unsignaled one again once that wait has executed.
VkImportSemaphoreFdInfoKHR import{VK_STRUCTURE_TYPE_IMPORT_SEMAPHORE_FD_INFO_KHR};
import.semaphore = semaphore;
import.flags = VK_SEMAPHORE_IMPORT_TEMPORARY_BIT;
import.handleType = VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT;
import.fd = fence;
const VkResult result = m_importSemaphoreFd(m_device, &import);
if (result == VK_SUCCESS) {
m_sharedImageWaits.push_back({semaphore, stages}); // the import owns `fence` now
return;
}
m_freeSharedImageWaitSemaphores.push_back(semaphore);
MGLOG_W_ONCE("Magma: a shared-image fence could not be imported (vkImportSemaphoreFdKHR -> %d); it "
"is waited for on the CPU",
static_cast<Int>(result));
}
}
// Nothing recorded since reaches the queue before this returns.
if (!MG_Remote::Server::SharedImages::SyncFile::Wait(fence, kSharedImageCpuWaitMs)) {
MGLOG_W_ONCE("Magma: a shared-image fence did not signal within %u ms; the image is used anyway",
kSharedImageCpuWaitMs);
}
CloseSyncFile(fence);
}
Bool VulkanRenderer::ExportSharedImageSignal(VkSemaphore semaphore, int& fence) {
fence = -1;
VkSemaphoreGetFdInfoKHR info{VK_STRUCTURE_TYPE_SEMAPHORE_GET_FD_INFO_KHR};
info.semaphore = semaphore;
info.handleType = VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT;
// A sync_file export has a wait's effect on the semaphore: it is unsignaled again, and
// reusable once the submission that signaled it has completed. -1 back means it had
// already signaled.
const VkResult result = m_getSemaphoreFd(m_device, &info, &fence);
if (result != VK_SUCCESS) {
fence = -1;
MGLOG_W_ONCE("Magma: a shared-image fence could not be exported (vkGetSemaphoreFdKHR -> %d)",
static_cast<Int>(result));
return false;
}
for (auto& entry : m_sharedImageSemaphoresInFlight) {
if (entry.semaphore == semaphore) entry.exported = true;
}
return true;
}
void VulkanRenderer::AttachSharedImageSync(VkSubmitInfo& info, SharedImageSubmitSync& storage) {
storage.attached = false;
if (m_sharedImageWaits.empty() && m_sharedImageSignal == VK_NULL_HANDLE) return;
storage.waits.assign(info.pWaitSemaphores, info.pWaitSemaphores + info.waitSemaphoreCount);
storage.stages.assign(info.pWaitDstStageMask, info.pWaitDstStageMask + info.waitSemaphoreCount);
for (const auto& wait : m_sharedImageWaits) {
storage.waits.push_back(wait.semaphore);
storage.stages.push_back(wait.stages);
}
storage.signals.assign(info.pSignalSemaphores, info.pSignalSemaphores + info.signalSemaphoreCount);
if (m_sharedImageSignal != VK_NULL_HANDLE) storage.signals.push_back(m_sharedImageSignal);
info.waitSemaphoreCount = static_cast<Uint32>(storage.waits.size());
info.pWaitSemaphores = storage.waits.data();
info.pWaitDstStageMask = storage.stages.data();
info.signalSemaphoreCount = static_cast<Uint32>(storage.signals.size());
info.pSignalSemaphores = storage.signals.data();
storage.attached = true;
}
void VulkanRenderer::CommitSharedImageSync(const SharedImageSubmitSync& storage) {
if (!storage.attached) return;
for (const auto& wait : m_sharedImageWaits)
m_sharedImageSemaphoresInFlight.push_back({wait.semaphore, m_submitCounter, false, false});
m_sharedImageWaits.clear();
if (m_sharedImageSignal != VK_NULL_HANDLE) {
m_sharedImageSemaphoresInFlight.push_back({m_sharedImageSignal, m_submitCounter, true, false});
m_sharedImageSignalSubmitted = std::exchange(m_sharedImageSignal, VK_NULL_HANDLE);
}
}
void VulkanRenderer::RecycleSharedImageSemaphores(Uint64 completedSubmit) {
SizeT done = 0;
for (; done < m_sharedImageSemaphoresInFlight.size() &&
m_sharedImageSemaphoresInFlight[done].submitIndex <= completedSubmit;
++done) {
const auto& entry = m_sharedImageSemaphoresInFlight[done];
if (entry.semaphore == m_sharedImageSignalSubmitted) m_sharedImageSignalSubmitted = VK_NULL_HANDLE;
if (!entry.exportable) {
m_freeSharedImageWaitSemaphores.push_back(entry.semaphore);
} else if (entry.exported) {
m_freeSharedImageSignalSemaphores.push_back(entry.semaphore);
} else if (m_device != VK_NULL_HANDLE) {
// Signaled and never exported, so still signaled: a binary semaphore cannot be
// signaled again until something waits on it, and nothing will.
vkDestroySemaphore(m_device, entry.semaphore, nullptr);
}
}
if (done != 0) {
m_sharedImageSemaphoresInFlight.erase(
m_sharedImageSemaphoresInFlight.begin(),
m_sharedImageSemaphoresInFlight.begin() + static_cast<std::ptrdiff_t>(done));
}
}
void VulkanRenderer::DestroySharedImageSync() {
// The device is idle: every semaphore is inert.
const auto destroy = [this](VkSemaphore semaphore) {
if (semaphore != VK_NULL_HANDLE && m_device != VK_NULL_HANDLE)
vkDestroySemaphore(m_device, semaphore, nullptr);
};
for (const auto& entry : m_sharedImageSemaphoresInFlight) destroy(entry.semaphore);
m_sharedImageSemaphoresInFlight.clear();
for (const auto& wait : m_sharedImageWaits) destroy(wait.semaphore);
m_sharedImageWaits.clear();
destroy(std::exchange(m_sharedImageSignal, VK_NULL_HANDLE));
m_sharedImageSignalSubmitted = VK_NULL_HANDLE;
for (const VkSemaphore semaphore : m_freeSharedImageWaitSemaphores) destroy(semaphore);
m_freeSharedImageWaitSemaphores.clear();
for (const VkSemaphore semaphore : m_freeSharedImageSignalSemaphores) destroy(semaphore);
m_freeSharedImageSignalSemaphores.clear();
m_sharedImagesHeld.clear();
m_sharedImageReads.PublishFrame(-1); // drops the frame's references; publishes nothing
}
// ---- the reading side ------------------------------------------------------------------------
//
// REACHED ON EVERY USE. A wire draw or dispatch prepares its sampled textures through
// PrepareWireTextureResources on every call - the wire arm keeps no sampled-set cache across
// draws (the walk-skip is the monolith arm's SetupDraw) - and that notes each shared-image
// texture it resolves; SetupWireDraw and both dispatch paths acquire the noted ones straight
// after (AcquireNotedSharedImages). ResolveWireImage, through which every attachment, blit,
// copy and readback reaches a texture's image, acquires directly. The test is the write
// generation against the one last acquired, so a generation that moves between two draws of
// one frame is caught at the second.
void VulkanRenderer::AcquireNotedSharedImagesSlow() {
m_textureManager->TakeNotedSharedImageUses(m_sharedImageUseScratch);
for (const Uint64 key : m_sharedImageUseScratch) {
if (auto* resource = m_textureManager->FindWireTextureResource(key)) AcquireSharedImage(*resource);
}
m_sharedImageUseScratch.clear();
}
void VulkanRenderer::AcquireSharedImage(VkTextureManager::TextureResource& resource) {
namespace SI = MG_Remote::Server::SharedImages;
// A YUV image is converted into the texture's own image instead (WireYuvImage.inc).
if (resource.yuvImageId != 0) {
(void)ConvertYuvSharedImage(resource);
return;
}
if (resource.sharedImageId == 0 || resource.image == VK_NULL_HANDLE || resource.sharedImageOwner == nullptr)
return;
const auto image = std::static_pointer_cast<const SI::Image>(resource.sharedImageOwner);
Uint64 generation = 0;
const int fence = SI::DupWriteFence(*image, &generation);
// Held, and not written since it was acquired: every use after a frame's first.
if (resource.sharedImageHeld && generation == resource.sharedImageGeneration) {
CloseSyncFile(fence);
return;
}
// Out of any render pass, in the recording the use goes into (or one ahead of it).
auto& frame = m_frameContext.GetCurrent();
if (!frame.isCommandRecording) m_frameContext.BeginCommandRecording();
if (VkRenderPassManager::GetActiveRenderPass()) VkRenderPassManager::EndRenderPass(frame.commandBuffer);
// The write's fence gates the submission carrying the acquire: this recording's, or an
// earlier one if a flush takes it first - either way ahead of every later command.
StageSharedImageWait(fence, kSharedImageUseStages);
// A session that writes images (implicit sync by flush) may be about to overwrite what
// another session still samples: those reads go first.
if (m_sharedImageImplicitSync && !resource.sharedImageHeld) {
for (const int read : SI::DupPendingReadFences(*image)) StageSharedImageWait(read, kSharedImageUseStages);
}
const Uint32 family = static_cast<Uint32>(m_physicalDevice.queueFamilies.graphicsFamily);
VkImageMemoryBarrier barrier{VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER};
barrier.image = resource.image;
barrier.subresourceRange = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1};
VkPipelineStageFlags acquireAfter = kSharedImageUseStages;
if (resource.sharedImageHeld) {
// Written again while this frame holds it - a buffer reused before the compositor let
// it go. It is handed back first, so the acquire below matches a release.
barrier.srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT;
barrier.dstAccessMask = 0;
barrier.oldLayout = resource.layout;
barrier.newLayout = VK_IMAGE_LAYOUT_GENERAL;
barrier.srcQueueFamilyIndex = family;
barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_FOREIGN_EXT;
vkCmdPipelineBarrier(frame.commandBuffer, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, 0, nullptr, 0, nullptr, 1, &barrier);
acquireAfter = VK_PIPELINE_STAGE_ALL_COMMANDS_BIT;
}
// GENERAL: the layout the writer releases in, and the one every wire descriptor samples
// in. An image no present has written yet (generation 0) has nothing to keep.
barrier.srcAccessMask = 0;
barrier.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
barrier.oldLayout = generation == 0 ? VK_IMAGE_LAYOUT_UNDEFINED : VK_IMAGE_LAYOUT_GENERAL;
barrier.newLayout = VK_IMAGE_LAYOUT_GENERAL;
barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_FOREIGN_EXT;
barrier.dstQueueFamilyIndex = family;
vkCmdPipelineBarrier(frame.commandBuffer, acquireAfter, kSharedImageUseStages, 0, 0, nullptr, 0, nullptr, 1,
&barrier);
// The open recording names the image now: its out-of-pass work may not move ahead of it.
m_textureManager->StampResourceRecordingUse(resource);
resource.layout = VK_IMAGE_LAYOUT_GENERAL;
resource.sharedImageGeneration = generation;
if (!resource.sharedImageHeld) {
resource.sharedImageHeld = true;
// The current recording is the next submission (RetireWireObjects' rule).
m_sharedImagesHeld.push_back({resource.sharedImageKey, m_submitCounter + 1});
}
m_sharedImageReads.NoteRead(image);
}
// THE FRAME BOUNDARY (Present). Every image this frame acquired goes back to the foreign
// family in the frame's last recording, and that recording's submission signals the semaphore
// whose sync_file is published as the frame's read fence for all of them at once.
void VulkanRenderer::ReleaseHeldSharedImages() {
// A frame boundary: a YUV texture's next use converts its image again.
++m_yuvFrameSerial;
if (m_sharedImagesHeld.empty()) return;
const Uint32 family = static_cast<Uint32>(m_physicalDevice.queueFamilies.graphicsFamily);
Vector<VkImageMemoryBarrier> releases;
releases.reserve(m_sharedImagesHeld.size());
for (const auto& hold : m_sharedImagesHeld) {
// Gone or re-imaged since: a destroyed image needs no release, and a replacement
// acquired this frame has an entry of its own.
auto* resource = m_textureManager->FindWireTextureResource(hold.key);
if (resource == nullptr || !resource->sharedImageHeld) continue;
VkImageMemoryBarrier barrier{VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER};
barrier.srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT;
barrier.dstAccessMask = 0;
barrier.oldLayout = resource->layout;
barrier.newLayout = VK_IMAGE_LAYOUT_GENERAL;
barrier.srcQueueFamilyIndex = family;
barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_FOREIGN_EXT;
barrier.image = resource->image;
barrier.subresourceRange = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1};
releases.push_back(barrier);
resource->sharedImageHeld = false;
resource->layout = VK_IMAGE_LAYOUT_GENERAL;
m_textureManager->StampResourceRecordingUse(*resource);
}
m_sharedImagesHeld.clear();
if (releases.empty()) return;
auto& frame = m_frameContext.GetCurrent();
if (!frame.isCommandRecording) m_frameContext.BeginCommandRecording();
if (VkRenderPassManager::GetActiveRenderPass()) VkRenderPassManager::EndRenderPass(frame.commandBuffer);
vkCmdPipelineBarrier(frame.commandBuffer, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, 0, 0, nullptr, 0, nullptr,
static_cast<Uint32>(releases.size()), releases.data());
}
Bool VulkanRenderer::PublishSharedImageAccesses() {
m_sharedImageImplicitSync = true;
if (m_sharedImageReads.Pending() == 0 && m_sharedImagesHeld.empty()) return true;
ReleaseHeldSharedImages();
if (m_sharedImageSyncFd && m_sharedImageSignal == VK_NULL_HANDLE)
m_sharedImageSignal = TakeSharedImageSemaphore(/*exportable=*/true);
m_sharedImageSignalSubmitted = VK_NULL_HANDLE;
if (!FlushPendingCommands()) {
MGLOG_E_ONCE("Magma: a flush's shared-image accesses could not be submitted");
m_sharedImageReads.PublishFrame(-1); // drops the references; nothing to order against
return false;
}
const VkSemaphore signal = std::exchange(m_sharedImageSignalSubmitted, VK_NULL_HANDLE);
int fence = -1;
if (signal == VK_NULL_HANDLE || !ExportSharedImageSignal(signal, fence)) {
// No fence came out of it: the submission is waited for and published as completed.
fence = -1;
if (!WaitForSubmitsUpTo(m_submitCounter, UINT64_MAX)) {
MGLOG_E_ONCE("Magma: a flush's shared-image accesses could not be waited for");
}
}
m_sharedImageReads.PublishFrameAsWrite(fence);
return true;
}
// THE NATIVE FENCE. A sync_file of everything the session submitted so far, for the client's
// EGL_ANDROID_native_fence_sync: the application hands the descriptor on (Chrome's GPU process,
// to the browser that commits the buffer), so it has to signal no earlier than the last of that
// work. With nothing recorded, an empty
// recording is submitted to carry the signal: the in-order queue runs it after every earlier
// submission, and it is tracked like any other (the semaphore's reuse waits for it).
Bool VulkanRenderer::ExportNativeFence(int* fence) {
*fence = -1;
// Every fence the application asks for ends one of its frames as far as a YUV texture is
// concerned (an offscreen producer - a browser's GPU process - has no present to do it).
++m_yuvFrameSerial;
if (m_device == VK_NULL_HANDLE || m_graphicsQueue == VK_NULL_HANDLE || m_frameContext.GetFrameCount() == 0)
return true; // nothing was ever submitted: nothing to fence
// A session that writes images ends its frame here, as at a flush; a compositor that only
// samples them keeps its reads for its present, which follows the fence's work anyway.
const Bool boundary = m_sharedImageImplicitSync;
if (boundary) ReleaseHeldSharedImages();
if (m_sharedImageSyncFd && m_sharedImageSignal == VK_NULL_HANDLE)
m_sharedImageSignal = TakeSharedImageSemaphore(/*exportable=*/true);
m_sharedImageSignalSubmitted = VK_NULL_HANDLE;
if (!HasPendingRecordedWork() && m_sharedImageSignal != VK_NULL_HANDLE)
m_frameContext.BeginCommandRecording();
const Bool submitted = HasPendingRecordedWork() && FlushPendingCommands();
if (HasPendingRecordedWork()) MGLOG_E_ONCE("Magma: a native fence's commands could not be submitted");
const VkSemaphore signal =
submitted ? std::exchange(m_sharedImageSignalSubmitted, VK_NULL_HANDLE) : VK_NULL_HANDLE;
int exported = -1;
if (signal == VK_NULL_HANDLE || !ExportSharedImageSignal(signal, exported)) exported = -1;
if (exported < 0) {
// No descriptor: the work is waited out, so the caller may report it completed.
if (!WaitForSubmitsUpTo(m_submitCounter, UINT64_MAX))
MGLOG_E_ONCE("Magma: a native fence's work could not be waited for");
}
if (boundary && m_sharedImageReads.Pending() != 0) {
int published = -1;
#if !defined(_WIN32)
if (exported >= 0) published = ::dup(exported);
#endif
m_sharedImageReads.PublishFrameAsWrite(published);
}
*fence = exported;
return true;
}
void VulkanRenderer::PublishSharedImageReads() {
const VkSemaphore signal = std::exchange(m_sharedImageSignalSubmitted, VK_NULL_HANDLE);
int fence = -1;
// An export also unsignals the semaphore, which is what lets it be signaled again: a
// submitted signal is exported even when nothing is left to publish it to.
if (signal != VK_NULL_HANDLE && !ExportSharedImageSignal(signal, fence)) fence = -1;
if (m_sharedImageReads.Pending() == 0) {
CloseSyncFile(fence);
return;
}
// No fence (no export on this device): the frame's reads go unpublished - the file
// comment says why that is the choice.
// A session that writes images publishes every boundary as a write (PublishSharedImageAccesses).
if (m_sharedImageImplicitSync)
m_sharedImageReads.PublishFrameAsWrite(fence);
else
m_sharedImageReads.PublishFrame(fence);
}
void VulkanRenderer::AbandonSharedImageReads() {
// Acquires recorded into the dropped recording never ran, so those images are not held;
// ones an earlier flush submitted are, and the next frame boundary releases them.
SizeT kept = 0;
for (const auto& hold : m_sharedImagesHeld) {
if (hold.acquireSubmit <= m_submitCounter) {
m_sharedImagesHeld[kept++] = hold;
continue;
}
if (auto* resource = m_textureManager->FindWireTextureResource(hold.key))
resource->sharedImageHeld = false;
}
m_sharedImagesHeld.resize(kept);
// Its waits and signal were never submitted. A wait semaphore's unconsumed temporary
// payload is replaced by its next import; the signal semaphore was never signaled.
for (const auto& wait : m_sharedImageWaits) m_freeSharedImageWaitSemaphores.push_back(wait.semaphore);
m_sharedImageWaits.clear();
if (m_sharedImageSignal != VK_NULL_HANDLE)
m_freeSharedImageSignalSemaphores.push_back(std::exchange(m_sharedImageSignal, VK_NULL_HANDLE));
}
#endif // MOBILEGL_BUILD_DISAGGREGATED
@@ -0,0 +1,10 @@
#version 450
// The image is bound through an immutable sampler that carries its VkSamplerYcbcrConversion, so
// a plain sample already returns RGB.
layout(set = 0, binding = 0) uniform sampler2D yuvImage;
layout(location = 0) in vec2 texCoord;
layout(location = 0) out vec4 outColor;
void main() {
outColor = vec4(textureLod(yuvImage, texCoord, 0.0).rgb, 1.0);
}

Some files were not shown because too many files have changed in this diff Show More