74060e8b5d
[Fix, Test] (DirectGLES/BackendObject, MG_Remote/Server, MG_Test/Wire): ID-67 - the native skip in BackendObject_DirectGLES::MakeEGLCurrent is keyed on the VIRTUAL (dpy, draw, read, ctx) tuple, not the surface: the surface's own creation bind is adopted by the first tuple, an identical tuple is a native no-op and publishes no caps snapshot (the client's mirror generation does not move), and a DIFFERENT tuple onto the same surface binds natively again (DirectGLES's invalidations describe the frontend context that changed) and republishes (R-12 arm (a)); ServerLoop tallies MakeCurrentRepublishCount and the control asserts 1 bind + 1 republish for a different tuple and 0 + 0 for a repeat at the EGL table and the tally; the EGL fixture gives the session its backend so the republish is real
swung0x482026-09-16 10:43:33 -04:00
edcdf771db
[Merge] (Docs, P5): land the P5 sections of README, ROADMAP, ARCHITECTURE and MEASUREMENTS
swung0x482026-09-16 10:36:11 -04:00
fbeea876b3
[Merge] (MGPipe, P5): land c1f - the client's missing gates and controls (ID-64)
swung0x482026-09-16 10:31:33 -04:00
1646049ac9
[Fix, Test] (DirectGLES/Managers, MG_Test/Wire): M-3's two whole-store refusals key on the DESCRIPTOR's HasDefinedContent - content the application supplied (glBufferData(size, data)) must be fully staged by the draw, a gap is a missing record and Fatal{StageSnapshotTooNarrow} by name; a store the application orphaned (glBufferData(size, NULL)) has undefined bytes by its own declaration, so the streaming idiom (orphan, partial glBufferSubData, draw) uploads the staged bytes plus zero-fill exactly as the monolith arm uploads MappedData(); round 2 refused both shapes and aborted six LargeArenaAdoption / ResourceSubsystemControl joint entries by name (hidden under c1 round 2's BlobMissing abort until c1 round 3); two positive gates run the idiom end to end in a forked child through each reader and red on their own string when the refusal is made unconditional again
swung0x482026-09-16 10:31:29 -04:00
13043e3ff4
[Test] (MG_Test/Wire): ServerLoopTest round 3 - an EGL fixture brings the server's REAL DirectGLES backend up on the apply thread through the twelve forwarders alone and counts native eglMakeCurrent at the EGL function table; the review-v2 N-5 gates each red on their own string: C7 one native bind / zero native releases / owner unchanged, N-3 destroy-recreate with the same handle values binds again, ID-49 the reply is the tight extent whatever DstSize the client sent, M-3 sparse shadow through pool-reuse and through respecify dies by name, M-1 a queued range outside the staged coverage dies by name at the drain, m-5 a twin surviving context loss drops its freed base, M-2 the ensure uploads the server shadow not MappedData (item 10), M-6 ShutdownSplitRoles after an early Start failure drops the backend, codex-11 the effective mask via an absent cpu, C6 the seven reads answered by two objects; N-9 the park case bounds its own post and the tally case names its tally
swung0x482026-09-16 10:17:29 -04:00
49fecc1686
[Fix] (DirectGLES/Managers, MG_Backend/Init, MG_Remote/Server): N-7 m-5's freed-base null is transport-guarded like the other two split hunks so monolith transport in a split build changes nothing by construction; N-8 the second consumer-mask check compares MGPipeGetResourceOps() against the table step 2 saw and refuses a replaced table as Fatal{ConsumerMaskLie, resource ops table replaced} (the re-run alone could not see a client-registered table); ServerVerbSink::ReadbackScratchBytes for ID-49's tight-size gate
swung0x482026-09-16 10:17:29 -04:00
a838c7879d
[Fix] (MG_Remote/Server, DirectGLES/BackendObject): ID-54 bind-once at the native layer - under an active transport BackendObject_DirectGLES skips the native eglMakeCurrent for a surface already current on the apply thread (its own creation bound it; measured 2 -> 1 native binds per process, 0 native releases, owner unchanged), and N-3 the held (dpy, draw, read, ctx) tuple is FORGOTTEN on ReleaseEGLResources, ReleaseEGLSurface of a surface it names, surface (re)creation and backend destruction so a recycled handle value after a destroy is a real bind again; ServerLoop.h's ownership prose states what is measured rather than what was hoped (review v2 item 10, N-3)
swung0x482026-09-16 10:17:29 -04:00
6d86f9d9be
[Fix] (Client): share teardown refusal and gate live wire producers
swung0x482026-09-16 10:16:11 -04:00
eec0e836c6
[Build] (Android): map mobilegl.buildDisaggregated and mobilegl.buildDisaggregatedInproc to the CMake options so the trace APK has a split flavour beside pull and push
swung0x482026-09-16 10:15:58 -04:00
fa924d34fb
[Merge] (MGPipe, P5): land j0 rounds 2-3 - the split-lane private logs wired into CI and the skipped-selection control failure (ID-62)
swung0x482026-09-16 10:15:58 -04:00
ca92308bc0
Merge branch 'feat/disaggregated' into p5/v1
swung0x482026-09-16 09:27:25 -04:00
68d8ef9f34
[Fix] (IntegrationTest): gate private split logs and repair control red-checks
swung0x482026-09-16 09:23:27 -04:00
e61d00123c
[Feat] (MG_Backend, MG_Remote/Server): construct BackendObject_Remote directly in the split hook and bracket PipeApplier::ApplyOne with ScopedApplierEntry - the two merge-time hunks (ID-59)
swung0x482026-09-16 09:17:30 -04:00
5682f429d0
[Fix, Test] (MG_Remote/Client, MG_Impl/Pipe): c1 round 3 - route the five by-address delete sites (B1); observe the client arm (B3); readback requires Status=OK and the exact extent and refuses a pack-PBO read by name (M2/M8/codex11/ID-57); one ERROR->Fatal{ReplyError} rule across the two escapes and an honest ShutDown decline (M4/codex5); InitialBytesNotCarried is role-aware (M5); a routed call during teardown refuses by name not runs the applier (codex4); caps adopted on every make-current (codex12); set_dynamic_state carries an optional blob; PACK_SKIP_IMAGES ignored for a 2-D read (codex6); one tight-size function for production and test (M3/codex10); redcheck 9/9 red each on its own string (M6)
swung0x482026-09-16 09:11:47 -04:00
c9d84e33c5
[Fix] (DirectGLES/Managers): under an active transport a resource_subdata that finds no twin MINTS one (GetOrCreateBufferResourceForHandle) so its bytes reach the server StagedShadowStore - the twin is otherwise lazy until the first draw (D-A2) and the reduced-path VBO bytes were lost, so the milestone frame rendered a shifted store; monolith keeps the lazy twin (R-11/ID-52 item 3)
swung0x482026-09-16 08:11:05 -04:00
f280baf2b3
Merge branch 'feat/disaggregated' into p5/c1
swung0x482026-09-16 08:07:59 -04:00
455b475c5d
[Fix] (DirectGLES/Managers, MG_Remote/Server): m-5 nulls a stale hostBytes only when its server shadow is actually gone (StagedShadowStore::HasShadow), not on a twins first-ensure generation sync - the unguarded null dropped a subdatas freshly staged base and made the reduced-path VBO read shifted; HasShadow unit control added
swung0x482026-09-16 08:03:52 -04:00
a85aa95401
[Test] (MG_Test/Wire): ServerLoopTest - the R-11 production gate through the real ops table (M-1), the C2/M-6/M-7/C6/C7 controls, actual-blocked-park arming (C10), an explicit-mask affinity case (M-4), and the coverage death control naming SIGABRT (m-2)
swung0x482026-09-16 07:45:29 -04:00
05bcb73477
[Fix] (MGPipe): resource_flush_range carries no companion bytes under an active transport - the consumer reads the server-owned StagedShadowStore instead of faulting the null carrier as corruption (item-12, R-13.2/ID-37)
swung0x482026-09-16 07:45:29 -04:00
61d6cd8e30
[Fix] (MG_Remote/Server): OnReadPixels reads with neutral pack state into a tight w*h*bpp reply so a server-visible pack state cannot overflow the slot (ID-49); the present<->swap 1:1 is a convention not structural (m-1); the reply-ordering invariant is documented against ReplySlot.h (m-7)
swung0x482026-09-16 07:45:29 -04:00
7a925b3259
[Fix] (DirectGLES): the seven backend format-capability reads use the server-private cache via ActiveBackendFormatCaps under split, not pActiveBackendObject (C6); guarded so the pull build is byte-identical
swung0x482026-09-16 07:45:29 -04:00
0fde95357c
[Fix] (DirectGLES/Managers): under an active transport the server staged copy is the only base - liveHostBase no longer falls back to the client object (M-2); the pool-reuse and whole-store respecify readers go through RequireCoverage (M-3); a twin surviving DropAll has its freed hostBytes nulled (m-5)
swung0x482026-09-16 07:45:29 -04:00
0244fc4a50
[Fix] (MG_Backend/Init): ShutdownSplitRoles stops the apply thread and drops the server backend on the early ClientSession::Start-failure path (M-6), and the consumer-mask cross-check re-runs after the client object exists (m-6)
swung0x482026-09-16 07:45:29 -04:00
e7fe3b9770
[Fix] (MG_Remote/Server): ServerLoop - clear m_running under m_controlMutex and re-check before publishing (C2); a backend with no thread is Fatal{ApplyThreadNotRunning} not an inline app-thread fallback (M-7); ApplyAffinity logs the effective sched_getaffinity mask (M-4/codex-11); ServerMakeEGLCurrent binds once per tuple and records a client release without unbinding (C7/ID-54); Start resets its own diagnostics (m-3); doorbell Fatal text corrected (m-4)
swung0x482026-09-16 07:45:29 -04:00
41992649c9
Merge branch 'feat/disaggregated' into p5/c1
swung0x482026-09-16 07:23:35 -04:00
c3e736a343
[Docs] (Disaggregated): SEG_REPLY is 16 MiB in eight 2 MiB slots (ID-47) - the segment table still said 8 MiB with 4 KiB slots
swung0x482026-09-16 07:14:26 -04:00
02f15bb675
[Feat] (MG_Pipe, MG_Remote/Client, MG_Impl/Pipe): R-17 - install the two generated MGPipe tables over the 37 apply entry points, 40 call sites onto the thunks, 37 wire emitters, the reply mailbox, and ID-47/ID-49 readback
swung0x482026-09-16 07:10:51 -04:00
949ba509f3
[Fix, Test] (MG_Remote, MG_Test, CONTRACT): wave 1.5 s1 - ID-47 SEG_REPLY is 16 MiB / 8 slots of 2 MiB and ReplySlotPool::CanHold+RequireReadPixelsFits refuse an oversize ReadPixels at the CLIENT by name (ProtocolSmokeTest pin, CONTRACT row 23); CapsAbiFingerprint IS MixAbiFingerprint over CapsAbiFingerprintInputs and the sensitivity case drives it; the null-union Hello/Welcome frames go THROUGH Accept/StartOverTransportPair; the pad-bit case stamps the bit past Reserve; both death controls name their diagnostic via WireLogFatal (stderr echo); the RSS ledger reads /proc once and keeps its own running peak (GitHub run 35079459114)
swung0x482026-09-16 07:04:08 -04:00
0a22f363ff
[Doc] (workflows): correct the two G5 self-test descriptions to b1 round 2 - the eleventh row is pinned in the script at 3dadd4c1 and compared against the reviewed body, p3a runs eight canned controls and p4a nine negative ones
swung0x482026-09-16 06:58:58 -04:00
2e2b41ab49
[Fix, Test] (workflows, scripts/ci): make the three CI negative controls assert their own failure reason instead of accepting any non-zero ctest exit, count the arming baseline by PASSED rather than by not-skipped, and put both control bodies in files a smoke test can drive against a stubbed ctest
swung0x482026-09-16 06:58:34 -04:00
aa7ad670c4
Merge branch 'p5/v1' into p5/c1
swung0x482026-09-16 06:43:30 -04:00
3fb97cc536
Merge branch 'feat/disaggregated' into p5/v1
swung0x482026-09-16 06:42:04 -04:00
bb6ba28241
[Fix] (MG_Remote, Wire): send the codec's MapPersistent arm through the same adoption-tier gate the applier uses, so MOBILEGL_IPC_ADOPT_TIER=0/1 die on their named diagnostic
swung0x482026-09-16 06:41:01 -04:00
ba6dc4f7f1
[Fix] (MG_Remote, Wire): require every client-server content blob to declare SEG_STAGE and refuse any other carrier by name, which also puts the audit's run bookkeeping out of reach
swung0x482026-09-16 06:41:00 -04:00
1a06b40a58
[Fix] (MG_Remote, Wire): let an empty SEG_STAGE take a blob that fits by rebasing its cursors to zero, so the wrap skip is only ever charged against bytes still in flight
swung0x482026-09-16 06:40:55 -04:00
7110cdb61a
[Fix] (scripts): make gen_pipe.py's expect_trip require the tripped guard's own message and a non-zero exit code, with an R-16 meta-control that an unrelated SystemExit or sys.exit(0) cannot pass as a trip - the any-SystemExit defect fb70704e fixed in the field-ownership generator, confirmed still present here by codex review finding 9
swung0x482026-09-16 06:33:07 -04:00
6cabc6b070
[Fix, Test] (MG_IntegrationTest, MG_Test): make the persistent-map membership expectation a function of the arm the map landed in, and pin the arm-independent half on IsLivePersistentMap where both answers can be produced
swung0x482026-09-16 06:03:52 -04:00
f63c9483de
[Fix] (scripts): compare G5's eleventh row against its pin ALWAYS and re-pin it on the reviewed P5 body, with a negative control that the pin itself can fail
swung0x482026-09-16 05:56:08 -04:00
571cbabb53
[Fix] (MobileGL/Init, MG_Backend): ShutdownSplitRoles exists only under MOBILEGL_BUILD_DISAGGREGATED - G1 admits no new pull symbol and the unconditional call moved DestroyImpl by 32 bytes
swung0x482026-09-16 05:54:33 -04:00
7ea3b1d1f1
[Test] (MG_Test/Wire): ServerLoopTest - the apply thread, the control mailbox, the verb stamp, the audit poison, the bounded shutdown, and StagedShadowTest
swung0x482026-09-16 05:45:15 -04:00
d623118485
[Feat] (MG_Backend, MobileGL/Init): the single split hook in MG_Backend::Init - server backend, CallMask halves, ClientSession::Start, remote object - and ShutdownSplitRoles at the top of Destroy
swung0x482026-09-16 05:45:15 -04:00
23c0468775
[Feat] (MG_Remote/Server): ServerLoop - the mgl-srv-apply thread, its affinity mask, the three-way park predicate, the blocking control mailbox, the bounded Stop, and the twelve EGL forwarders
swung0x482026-09-16 05:45:15 -04:00
8dbe230ae3
[Feat] (MG_Remote/Server): PipeApplier - attach the decoder on the apply thread, stamp/apply/clear per record, and ServerVerbSink for the five class-B verbs
swung0x482026-09-16 05:45:15 -04:00
377356f8ca
[Feat] (DirectGLES/Managers, MG_Remote/Server): R-11 - copy staged bytes into a server-owned StagedShadowStore with exact coverage; Fatal{StageSnapshotTooNarrow} before the flush drain instead of a widened INVALIDATE_RANGE
swung0x482026-09-16 05:45:15 -04:00
4ba14fa4ed
[Test] (MG_Test, Wire): ID-43 - RingTest.TheTwoFlagSpacesAreDisjointByTranslation states the tree: a header stamped with the CALL flags outside Reserve is POPPED with its lying kRecPad/kRecBorrowSlot bits visible, and the control that a genuine filler still vanishes names Pop kind check as the defence
swung0x482026-09-16 05:42:12 -04:00
87c1e1bebd
Merge branch 'feat/disaggregated' into p5/c1
swung0x482026-09-16 05:35:48 -04:00
5e5bf7b9d6
[Merge] (dev): bring fff9d639 - the DirectVulkan resident-buffer ordering and coherent trace-copy fix for Minecraft 26.3-rc-3 - into feat/disaggregated
swung0x482026-09-16 05:19:41 -04:00
fff9d639f0
[Fix] (DirectVulkan, tools/trace_replay): preserve resident buffer ordering and coherent trace copies - fix intermittent geometry and UI corruption in Minecraft 26.3-rc-3
swung0x482026-09-16 01:04:31 -04:00
d628d906d1
[Feat] (MG_Remote, P5): land the client role - BackendObject_Remote, the 71-slot emit table, the caps mirror and the verb barrier
swung0x482026-09-11 16:46:02 -04:00
505b5084cf
[Test] (MG_Test): drive the live-host-writes probe through the production chain instead of writing the record by hand, and pin both map edges and the disabled-push control by serial arithmetic
swung0x482026-09-11 15:48:28 -04:00
6afa044701
[Docs] (PipeStats): state in both entries that persistent-map-push and stage-buffer cover the same bytes under split, because at tier T2 a pushed block IS an ordinary resource_subdata
swung0x482026-09-11 15:48:28 -04:00
f858841c5f
[Fix] (MG_Remote, GLImpl): gate the transform-feedback reconciliation on the transport like every other new site, and give the unwritten writeback wait a named Fatal instead of an empty body
swung0x482026-09-11 15:48:28 -04:00
3dadd4c1e0
[Fix] (DirectGLES): make the extent hostBase is good for a parameter of the flush ladder, so tier 1's widening refusal is live code the moment a SEG_STAGE snapshot is narrower than the queued range
swung0x482026-09-11 15:48:28 -04:00
0598cfffe0
[Fix, Test] (MG_Pipe): replace the always-false live-host-writes wire, which no producer can survive, with the one statement that stays true - the bit is the buffer family's
swung0x482026-09-11 15:48:28 -04:00
950968044c
[Fix] (MG_Impl, PipeFill): make buffer_subdata_resident declare the live-host-writes bit - ApplyBufferWrite assigns it, so a silent content record writes false over a live write map
swung0x482026-09-11 15:48:28 -04:00
3a5e26215d
[Fix] (MG_State, BufferObject): publish the live-host-writes bit on the RISING edge too - the only content record a coherent persistent map emits is the push, and a draw-clean answer skips the ensure the push lives in and then latches it
swung0x482026-09-11 15:48:28 -04:00
6e858fa3f6
[Docs] (MG_Remote): record R-16 and R-17 - the fourteen reply-slot rows and why the barrier makes them cost no block, and the two flag spaces being disjoint by translation
swung0x482026-09-11 15:44:24 -04:00
ab112eb0ed
[Fix, Test] (gen_pipe, Wire): stop the generated header comment inviting a stamp of the call flags into the ring's framing flags - three of the six bits mean something else there, and only one of the three is masked on one path
swung0x482026-09-11 15:44:24 -04:00
d55e37337b
[Fix] (MG_Pipe): give the four acceptance calls the kReplySlot their applier entry points have always answered through, so the flags table is the single thing that sizes the reply pool and the decoder posts against
swung0x482026-09-11 15:44:24 -04:00
f33e5d6090
[Fix] (MG_Remote, Transport, Client): SEG_STAGE is w1's encoder-local linear allocator and not a ring - drop the stage RingProducer and its power-of-two rounding, pin the stage cursor triple dead, require a wrap filler to carry kind kRingPadRecordKind as well as the flag it shares with kVarTail, and name every kRecBorrowSlot sighting
swung0x482026-09-11 15:41:44 -04:00
949ed1f037
[Test, CI] (tools/trace_replay, workflows): give retrace-split the negative control it was cloned without, match ConfigLoader's whole INFO sentence rather than a KEY=VALUE an env dump also prints, read the three-arm names file the identity step was writing and discarding, and refuse an unknown manifest key that silently empties a CI matrix
swung0x482026-09-11 15:30:17 -04:00
8b425a9718
[Test] (MG_IntegrationTest): arm the split lanes from the running process - MG_Config::Transport, ClientSession::Active() and ImplementedVerbCount() - instead of from a grep over c0's stub messages, which a one-word rename disarmed into eight green monolith runs, and require the encoder's record ordinal to move before any split case may pass
swung0x482026-09-11 15:30:17 -04:00
79b7866511
[Fix] (MG_Remote, Wire): close R-2 arm 4 over MGHostSpan, stop the producer writing SEG_STAGEs consumer cursors, bound the stage-mark queue, and answer a reply slot only for rows the catalogue flags kReplySlot
swung0x482026-09-11 15:20:15 -04:00
3f1eaee787
[Test] (MG_Test, Wire): pin the seven findings of fix round one - the frozen CapsSnapshot field ids, the verifiable frame with a null union, the borrowed slot that is not handed back, the peer's own mapping, the backwards-watermark abort, the lagging watermark and the refused reply geometry
swung0x482026-09-11 15:04:54 -04:00
8024dab9f4
[Fix] (MG_Remote, Client): fold the null union payload into both handshake guards, close the server session on every Start failure that ran after Accept succeeded, and drain teardown against the producer's own published seq rather than a watermark the ring lets lag
swung0x482026-09-11 15:04:54 -04:00
da8f030459
[Fix] (MG_Remote, Transport): the knob names the ring and the segment adds a control page, a backwards watermark is Fatal rather than a permanent hang, RetireThrough stops at a borrowed slot, the inproc peer gets a real second mapping through dup and Adopt, and the producer remembers its own published seq
swung0x482026-09-11 15:04:54 -04:00
ed60c06b98
[Fix] (MG_Remote, Protocol): mark CapsSnapshot's four retired fields deprecated instead of deleting them - a deletion FREES the vtable slot, so callMask took slot 14 from an [int] vector as an 8-byte ulong with no ABI-major bump and a fingerprint that mixes struct sizes rather than the schema
swung0x482026-09-11 15:04:54 -04:00
c5c12b489a
[Fix] (MG_Remote, Server): delete CallMask's two derived defaults and make an unset mask Fatal{UnsetCallMask} - the consumer default read MGPipeGetResourceOps, a process-wide global, so under inproc the server answered R-8's client-side gate with the client's own registration and under spawn it collapsed to P2's mask, dropping five P4a families with the dirty flags cleared anyway and the lane green
swung0x482026-09-11 15:04:54 -04:00
3aa48bccdc
[Test] (CI): run the field-ownership generator's --check and --self-test in pipe-gates - R-7.1 rests on a generated table that nothing in CI diffed, and the build assertion only catches an unclassified row
swung0x482026-09-11 14:53:39 -04:00
126bd9aa62
[Test] (MG_Test, Pipe): pin the applier's own clear on the path a spawned server takes, and four named fields the stamp case must not re-derive from the table it is testing
swung0x482026-09-11 14:53:39 -04:00
3cb2c4aa9a
[Fix] (MGPipe): make the applier's own verb-boundary clear mandatory rather than inherited from the client's fill, and name the argument in an argument-narrowed abort so the two halves of the pixel store do not print the same line
swung0x482026-09-11 14:53:39 -04:00
c16411aa33
[Feat] (MG_Pipe, scripts): refuse an OMITTED stamp row and not only a typo - every verb-shaped call now has a row or a named exemption, because a record with no stamp point applies under the previous verb's serial, mask and name
swung0x482026-09-11 14:53:39 -04:00
fb70704e74
[Fix] (scripts): assert every field-ownership negative control against its OWN message - one of the eleven was a silent duplicate of another because the harness asked only whether something exited, and the harness now has a control of its own
swung0x482026-09-11 14:53:39 -04:00
7ad0ff3c33
[Test] (MG_IntegrationTest): assert which persistent-map arm the scenario landed in through a peek at IsBackendPersistentMapped, not by inferring it from the push byte counter - "the push was never wired" and "the store was adopted" are different defects with different owners and the counter reads the same for both
swung0x482026-09-11 14:40:30 -04:00
31de23142f
[Test] (MG_Test, Wire): assert a payload crosses byte for byte including the bytes spelled Pad - a codec that normalised one would silently delete the fields P5 and b1 have just put there
swung0x482026-09-11 14:37:43 -04:00
db1a5dbc27
[Test] (MG_Test): one case per GPU-write row, the lifted live-host-writes pin from both sides, the widened INVALIDATE_RANGE edge and a non-zero persistent-map-push
swung0x482026-09-11 14:31:46 -04:00
0b847d462a
[Feat] (GLImpl): mark the pack PBO after a glReadPixels and the capture targets at glEndTransformFeedback, in place of a stall and an unbounded fence wait
swung0x482026-09-11 14:31:46 -04:00
d9ade1e7cd
[Feat, Docs] (PipeStats, MEASUREMENTS): wire persistent-map-push to the bytes the client ships, rewrite the inventory line that called it unwired, and correct the persistent-map census from 20 to 21
swung0x482026-09-11 14:31:46 -04:00
142bf921d4
[Fix] (DirectGLES): retire the probe's last frontend read under split and refuse tier 1 of the flush ladder when the map and the queued range disagree in either direction
swung0x482026-09-11 14:31:46 -04:00
13827e376f
[Feat] (MG_Pipe): give HasLiveHostWrites its producer, decline every persistent-map acquisition under split, and retire the verify pin that was waiting for exactly this
swung0x482026-09-11 14:31:46 -04:00
feca3994d1
[Feat] (MG_State, BufferObject): under split cut the mapped span into blocks, publish the live-host-writes edge, and let the writeback rather than the request clear a GPU write
swung0x482026-09-11 14:31:46 -04:00
851a0b6f76
[Feat] (MG_Remote, Client): the conservative GPU-write set and the persistent-map block tracker, on the client, where the decision SyncGpuWrites makes synchronously actually lives
swung0x482026-09-11 14:31:46 -04:00
eaf2b28b8c
[Feat] (MG_Pipe): give MGPSubData's pad byte to HasLiveHostWrites - the live-host-writer fact is content-shaped, and MGPResourceDesc's only carrier costs a buffer a spurious reallocation on every map
swung0x482026-09-11 14:31:46 -04:00
76140c33eb
[Fix] (MG_Remote, Wire): stop the decoder writing RingControl - appliedSeq has exactly one writer and it is the session, so the decoder keeps its own tally and the two are compared instead
swung0x482026-09-11 14:28:34 -04:00
75c0bcae80
[Docs] (MG_Remote): correct the sparse-caps comment - 13 targets by 77 internal formats is ~16 KiB of dense table, not half a megabyte
swung0x482026-09-11 14:28:34 -04:00
swung0x48