Commit Graph

  • 1c9029bc32 Merge pull request #29: integrate Adreno740 runtime changes main rponeawa 2026-11-16 10:04:55 +08:00
  • 0f150e695a Integrate qualified Adreno740 runtime changes merge/qualified-runtime-integration rponeawa 2026-11-16 10:02:18 +08:00
  • feac5a9ce0 test(vm): exercise noncoherent imports after writeback coherence succeeds fix/cached-import-visibility rponeawa 2026-11-16 01:52:46 +08:00
  • 14dea72cba fix(vm): maintain cached imports with noncoherent GPU mappings rponeawa 2026-11-16 01:37:38 +08:00
  • eb6d5af880 Preserve pending jobs during context-switch recovery fix/context-switch-recovery rponeawa 2026-11-15 16:32:30 +08:00
  • 54b41b8f59 feat(diag): retain the hung CP snapshot and locate fault address candidates rponeawa/macos-memory-diagnostics-run139 rponeawa 2026-11-12 18:19:48 +08:00
  • 031f35ccfb fix(recovery): a GPU access fault no longer fails other address spaces feat/macos rponeawa 2026-11-12 07:41:49 +08:00
  • a8fa18fa09 fix(recovery): keep the GPU usable after one process's address space fails rponeawa 2026-11-12 03:50:01 +08:00
  • 172f53f6cf fix(submit): blame no job for a ring switch the CP never finished rponeawa 2026-11-12 01:23:00 +08:00
  • 2a91aa1a14 fix(submit): count the BV thread's progress as GPU progress rponeawa 2026-11-11 23:27:40 +08:00
  • 9578923747 fix(submit): leave a working GPU alone in the stall handler rponeawa 2026-11-11 19:41:46 +08:00
  • 4a3921a2b9 feat(vm): record the mapping a failed map met rponeawa 2026-11-11 17:20:22 +08:00
  • a7c83cf765 feat(submit): keep the CP state of a ring switch that never finished rponeawa 2026-11-11 10:47:43 +08:00
  • 14dc2d7eae feat(memory): treat memory the kernel adopted outside its range as kernel memory rponeawa 2026-11-11 09:24:59 +08:00
  • 122a8f169f feat(pool): return a viewed grant's pages once the kernel reports no mapping is left rponeawa 2026-11-11 00:00:00 +08:00
  • c4dcd5391f perf(submit): run the line audits only with gpu-audit=1 and time every job rponeawa 2026-11-11 00:00:00 +08:00
  • c7bbc9cd67 fix(memory): keep the pool's zero fill out of the driver's lock rponeawa 2026-11-11 00:00:00 +08:00
  • 5282bd96c7 fix(pool): map a pool run only while it is zero-filled rponeawa 2026-11-11 00:00:00 +08:00
  • 9d99240ee4 feat(memory): take write-combined system BOs from a pool of DRAM outside the kernel's rponeawa 2026-11-11 00:00:00 +08:00
  • ae9f240e42 fix(priority): give a GPU-denied task the background priority rponeawa 2026-11-11 00:00:00 +08:00
  • db7751c3c3 perf(maintenance): leave the driver's synchronized buffers out of the per-submission cache passes rponeawa 2026-11-11 00:00:00 +08:00
  • 5559e982e4 perf(mailbox): poll only while the mailbox page is mapped rponeawa 2026-11-11 00:00:00 +08:00
  • 0eea94119e docs: describe fence waits and the memory counters rponeawa 2026-11-11 00:00:00 +08:00
  • 653e11c839 fix(wait): wake fence waiters from the completion pass rponeawa 2026-11-11 00:00:00 +08:00
  • 214133ffcc feat(diagnostics): record which ring a stall snapshot is about rponeawa 2026-11-11 00:00:00 +08:00
  • fa4f4aa98b fix(submit): look at the ring switch after every pause of a wait rponeawa 2026-11-11 00:00:00 +08:00
  • 798710e43d feat(diagnostics): count imported memory and BO views by process rponeawa 2026-11-11 00:00:00 +08:00
  • 08a32b56e1 feat(diagnostics): count the system BOs by holding process and by size bluseliu50 2026-11-10 20:19:37 +08:00
  • 7487f72ae6 feat: name WindowServer as the render server on macOS rponeawa 2026-11-10 12:00:00 +08:00
  • a517f6d336 feat: build for macOS rponeawa 2026-11-10 12:00:00 +08:00
  • 654dc40941 docs: name the boot arguments without the sheng- prefix rponeawa 2026-11-10 12:00:00 +08:00
  • 7aa02a15dc refactor: drop the sheng- prefix from boot arguments rponeawa 2026-11-10 12:00:00 +08:00
  • 6fa7bae2a5 feat(diagnostics): keep Turnip's GMEM swap trace in the GPU trace rponeawa 2026-11-03 15:11:00 +08:00
  • 1030485546 fix(diagnostics): keep each process's Metal refusal text in the GPU trace bluseliu50 2026-11-03 01:49:57 +08:00
  • 5344b299dd docs: describe the seven-issue trace selection UniXNU Docs 2026-11-03 00:48:09 +08:00
  • 4aa24c7cec fix(diagnostics): keep only seven-issue GPU trace events rponeawa 2026-11-03 00:47:46 +08:00
  • 6c28e71d4b docs: describe focused frame latency capture UniXNU Docs 2026-11-02 23:35:21 +08:00
  • cceed134e6 fix(diagnostics): filter GPU trace noise and coalesce pending queries rponeawa 2026-11-02 23:25:17 +08:00
  • 92d629318c Merge pull request #11 from UniXNU/fix/gpu-scheduling-and-metal Sharky Wang 2026-11-02 22:38:41 +08:00
  • dea9ad9754 docs: describe counter sampling and dispatch trace ownership UniXNU Docs 2026-11-02 19:43:15 +08:00
  • 3b94261d01 feat(diagnostics): capture counter validity and dispatch task events bluseliu50 2026-11-02 19:43:15 +08:00
  • 7d381d262b feat(diagnostics): retain Metal render thread identities bluseliu50 2026-11-02 18:51:52 +08:00
  • 175ff36dfb docs: describe gPU frame trace and pixel export UniXNU Docs 2026-11-02 13:08:14 +08:00
  • 0ee17b45d5 feat(gpu): add bounded frame traces and completed pixel capture rponeawa 2026-11-02 13:01:22 +08:00
  • 98967eaffc chore: refresh GPU host fixtures and verify task-role priorities rponeawa 2026-11-02 11:41:19 +08:00
  • ccabc7e446 fix(gpu): reschedule priority work and track global fence progress rponeawa 2026-11-02 11:41:19 +08:00
  • b9cc81e870 docs: describe two-ring GPU scheduling UniXNU Docs 2026-11-02 11:28:44 +08:00
  • e971110f44 fix(gpu): preserve contexts and queue ordering across priority rings rponeawa 2026-11-02 11:18:34 +08:00
  • a9a2cbb43b fix(accelerator): include the ABI header from the renamed module rpone 2026-11-01 10:04:36 +08:00
  • 7157c16ea5 Merge pull request #9 from UniXNU/merge/unmerged-branches Sharky Wang 2026-11-01 06:21:47 +08:00
  • 98396b9a28 chore: merge every remaining remote branch into main rpone 2026-11-01 06:20:58 +08:00
  • 149ef6d8ac feat(diag): record who set which GPU priority rpone 2026-11-01 05:13:28 +08:00
  • e1c0a2d73c feat(diag): let a process note its passes and pipelines to the kernel rpone 2026-11-01 05:13:27 +08:00
  • 18d79d7640 feat(preempt): give up a second ring the CP cannot reach rpone 2026-11-01 05:13:18 +08:00
  • 3f4cb94a3c feat(gpu): CP preemption between two rings (A1) rpone 2026-11-01 00:20:19 +08:00
  • 88b83dd0ef feat(timeline): owner pid and retire time per GPU job rpone 2026-10-31 19:43:24 +08:00
  • 36c4e44749 perf(vm): index mappings, not every span rpone 2026-10-31 18:32:18 +08:00
  • 14e4a7db25 perf(bo): free system BO pages after the device lock rpone 2026-10-31 18:06:36 +08:00
  • dd43abd05c perf(wait): poll fences without the device lock rpone 2026-10-31 16:59:45 +08:00
  • 01ed692a76 fix(adreno): ask for contiguous pages only up to 128 KiB, back off after a failure rpone 2026-10-31 16:12:52 +08:00
  • 59171b2bec feat(adreno): record every call into a connection rpone 2026-10-31 04:12:23 +08:00
  • e8bee0e10f feat(adreno): allocate the job timeline at attach and grow it to 16384 jobs rpone 2026-10-31 01:31:36 +08:00
  • 58ac3afda9 fix(adreno): the timeline stamps are on unless sheng-gpu-timeline=0 rpone 2026-10-31 00:03:55 +08:00
  • 90ba1e320a feat(adreno): job timeline, stall ring and CP counter stamps rpone 2026-10-30 23:54:22 +08:00
  • e0d159a868 feat(adreno): export the newest and retired fences, list of large imports with their first physical page rpone 2026-10-30 21:52:53 +08:00
  • cb9b44fbc5 feat(adreno): contiguous system BO chunks above a size threshold, trace of the failing recover step rpone 2026-10-30 19:34:40 +08:00
  • a087fcad8c perf(vm): check import aliases once per map bluseliu50 2026-10-29 21:05:15 +08:00
  • a08cf65b5f refactor(publish): drop the pipeline import method bluseliu50 2026-10-29 19:35:22 +08:00
  • 4c75ccbf42 perf(memory): make and wire BoAlloc's pages outside the lock bluseliu50 2026-10-29 18:33:17 +08:00
  • 0ca23f51c9 feat(gpu): sheng-gpu-prime=0 turns shader priming off bluseliu50 2026-10-29 16:34:13 +08:00
  • 1a9517bdef feat(telemetry): each client's own BO bytes live and at peak bluseliu50 2026-10-29 15:26:13 +08:00
  • 44e6319138 test(hostsim): answer only the GPU services, pass the rest to IOKit bluseliu50 2026-10-29 14:37:15 +08:00
  • c8806db0e5 perf(memory): user BOs from system memory, no fixed BO pool bluseliu50 2026-10-29 14:29:44 +08:00
  • 4f07dc9e71 feat(gpu): read the userdata import region for the render server bluseliu50 2026-10-29 11:44:14 +08:00
  • 9e21f6a1ec perf(adreno): no XNU contiguous search for BOs when the pool exists bluseliu50 2026-10-29 02:39:36 +08:00
  • d77091cd05 fix(adreno): keep every client's device after a GPU restart bluseliu50 2026-10-29 02:26:05 +08:00
  • b4ba1c13e8 perf(adreno): no contiguous-memory search under the publication lock bluseliu50 2026-10-29 02:13:22 +08:00
  • 3713f8a7f9 fix(adreno): end a stalled SMMU fault before a GPU restart bluseliu50 2026-10-29 00:19:47 +08:00
  • 6a3c9a8865 test(adreno): host doubles and deferred BO checks for the current driver bluseliu50 2026-10-28 20:56:12 +08:00
  • 1c184195b1 feat(adreno): grow object, queue and job tables with demand bluseliu50 2026-10-28 20:51:11 +08:00
  • bed3b6653b feat(adreno): grow the BO pool in contiguous chunks bluseliu50 2026-10-28 20:42:44 +08:00
  • 1105b0ec8c feat(adreno): msm hangcheck and GPU restart with replay bluseliu50 2026-10-28 20:40:51 +08:00
  • d834731657 feat(adreno): on-demand GPU page tables and a 2^ias address space bluseliu50 2026-10-28 20:26:11 +08:00
  • 799158a82f feat(adreno): report BO pool pressure in Query flags bluseliu50 2026-10-28 20:13:05 +08:00
  • 28c652d332 fix(adreno): let a VM's range table grow to 1048576 spans bluseliu50 2026-10-28 20:13:04 +08:00
  • bfc58413b4 fix(adreno): grant GPU connections without a per-process cap bluseliu50 2026-10-28 20:13:04 +08:00
  • 865495f2c1 feat(adreno): IO-coherent owned BOs behind sheng-gpu-owned-coherent rpone 2026-10-28 10:05:31 +08:00
  • 2c3df1e8b6 perf(adreno): count cache maintenance per phase and skip uncached imports rpone 2026-10-28 09:18:49 +08:00
  • 10ac0eaa0a feat(diag): record who set which GPU priority rpone 2026-11-01 05:13:28 +08:00
  • a423030c55 feat(diag): let a process note its passes and pipelines to the kernel rpone 2026-11-01 05:13:27 +08:00
  • 7b6245f0d5 feat(preempt): give up a second ring the CP cannot reach rpone 2026-11-01 05:13:18 +08:00
  • b8956437c7 feat(gpu): CP preemption between two rings (A1) rpone 2026-11-01 00:20:19 +08:00
  • 7fa13d4421 feat(timeline): owner pid and retire time per GPU job rpone 2026-10-31 19:43:24 +08:00
  • 92aa70fa6e perf(vm): index mappings, not every span rpone 2026-10-31 18:32:18 +08:00
  • a18a4facc3 perf(bo): free system BO pages after the device lock rpone 2026-10-31 18:06:36 +08:00
  • 17171f4c9a perf(wait): poll fences without the device lock rpone 2026-10-31 16:59:45 +08:00
  • 75e9b5f11b fix(adreno): ask for contiguous pages only up to 128 KiB, back off after a failure rpone 2026-10-31 16:12:52 +08:00
  • 759d646eca feat(adreno): record every call into a connection rpone 2026-10-31 04:12:23 +08:00
  • ff9a5ab934 feat(adreno): allocate the job timeline at attach and grow it to 16384 jobs rpone 2026-10-31 01:31:36 +08:00
  • d9f0cc831d fix(adreno): the timeline stamps are on unless sheng-gpu-timeline=0 rpone 2026-10-31 00:03:55 +08:00