- New CMake option MOBILEGL_BUILD_DISAGGREGATED (default OFF, plan B appendix B). ON appends the MG_Remote sources to SOURCE_FILES, puts 3rdparty/flatbuffers/include on the include path and defines MOBILEGL_BUILD_DISAGGREGATED=1. OFF compiles nothing from MG_Remote and adds no include path and no library, which is one of the two byte-level equalities plan B section 10.3 keeps: measured `nm --defined-only build-linux/libMobileGL.so | grep -ic MG_Remote` = 0 with the option OFF and 93 with it ON, with an identical ldd set in both configurations.
- The option forces itself OFF with message(WARNING) when 3rdparty/flatbuffers/include/flatbuffers/flatbuffers.h is missing, so a checkout without the submodule still configures and builds rather than failing a hundred lines later on a missing header.
- ITransport.h: SendFrame / ReceiveFrame / PeekFrameSize / ShareFd / ReceiveFd / Shutdown. ReceiveFrame's contract is the fix for a defect of the earlier branch: a destination buffer smaller than the pending message returns MOBILEGL_ERR_BUFFER_TOO_SMALL with the required size and KEEPS the message queued. The earlier transport failed the call and popped the message anyway, which wedges the stream permanently the first time a reader guesses a size wrong.
- Framing.h: [u32 'MGLF'][u32 len][payload], 64 MiB cap, validated on read. A bad magic or an oversized length latches the reader dead, is logged at ERROR and makes every later call return MOBILEGL_ERR_PROTOCOL_MISMATCH. This is the second inherited defect: Feat/CS-Delta-IPC's Framing.h:41-45 Feed() always returned OK and its header peek merely returned false, so a desynchronized stream became a silent permanent hang; its LocalSocketTransport.cpp:232-236 then allocated on the peer-supplied length with no cap. The magic is also byte-ordered so it reads "MGLF" on the wire, where the earlier constant spelled "FLGM".
- Ring.h/.cpp: RingControl exactly as the inherited design (plan B section 8.1 -> earlier section 6.2): two independent cursor triples (cmd and stage, each {head, appliedTail, retiredTail}), appliedSeq / submittedSeq / retiredSeq / completedFrameSerial / presentAckSerial, serverEpoch, ringGeneration, consumerParked, producerParked, eventRingFull, eventDropped. One 4096-byte page with each contended group on its own cache line, pinned by static_assert on size and alignment and by an offset test.
- The SPSC pair uses monotonic byte cursors and a power-of-two mask, so a torn read can never look like a valid earlier position. A record never straddles the wrap: the producer emits a kPad filler to the boundary, which is always a multiple of 8 and therefore always has room for a header. The producer reclaims against retiredTail rather than appliedTail, so the day the server borrows a ring slot into the GPU timeline (kRecBorrowSlot) it does not silently degrade to early recycling. HardDrainRing bumps ringGeneration only when the ring is quiesced, and leaves the cursors monotonic so cached offsets are recognisably stale.
- The consumer bounds-checks every record header before dispatch (8-aligned, at least a header, no larger than what the producer published, contiguous inside the mapping) and reports corruption to the caller instead of dispatching into undefined behaviour. That is the runtime half of the earlier section 6.3 discipline: SEG_CMD is written by another process, so a compile-time static_assert on record sizes proves nothing about what is in the mapping.
- Doorbell.h/.cpp: spin then park, both directions (earlier section 6.2a). CondVarDoorbell for inproc, SocketDoorbell for spawn (one byte, codes 0x01 ring-advanced and 0x02 watermark-advanced) - no futex, eventfd or named event anywhere. The lost-wakeup window is closed by ordering: the waiter stores its park flag seq_cst and then re-tests the condition, NotifyIfParked loads the same flag seq_cst after the watermark is published, so one of the two always sees the other. kDefaultSpinUs is 50, the MOBILEGL_IPC_SPIN_US default.
- ShmSegment.{h,cpp} + ShmSegmentPosix.cpp: memfd_create by raw syscall on desktop Linux (the glibc wrapper is too recent to rely on), ASharedMemory_create on Android (API 26; libc's memfd_create wrapper is API 30, above MobileGL's floor), shm_open + immediate shm_unlink as the fallback. Adopt() refuses a descriptor whose fstat size is smaller than the size the peer announced, so a short segment cannot turn every later offset into an out-of-bounds map. ShmSegmentWin32.cpp (CreateFileMappingW in Local\) is compile-guarded and untested - this project's Windows machine is not a correctness gate.
- The Android path is compile-verified, not just written: an arm64-v8a NDK build with the option ON links, ShmSegmentPosix.cpp.o carries an undefined ASharedMemory_create, and ShmSegmentWin32.cpp.o is empty there. That build is also what caught the missing <cstddef> in ShmSegment.h and Framing.h, where std::size_t / std::ptrdiff_t only resolved through a transitive include on the host sysroot.
- FdPassing.{h,cpp}: SCM_RIGHTS in the FIRST transport commit, as plan B section 8.1 demands, over a dedicated AF_UNIX SOCK_DGRAM socketpair rather than the control byte stream (message boundaries survive on every POSIX - SOCK_SEQPACKET does not exist on macOS - and ancillary data can never be split from its payload). The third inherited defect this replaces: Feat/CS-Delta-IPC deferred fd passing and hardcoded `out->fd = -1` in LocalSocketTransport.cpp:296, so on the only platform that matters its data plane could not move one byte between processes. MSG_CTRUNC, an unexpected descriptor count and a malformed sideband header all close every descriptor received before failing, and a too-small sideband buffer is refused before the recvmsg so a datagram is never half-consumed.
- InProcessTransport.{h,cpp}: two in-memory queues plus the condvar doorbell pair. It keeps the frame size cap so nothing that passes in inproc becomes illegal after the switch to spawn, hands descriptors over with dup() under the same ownership rule as SCM_RIGHTS, and lets a peer's queued messages be drained after Shutdown - usually the last one says why it is going away.
- Not done here on purpose: the MOBILEGL_IPC_* environment variables are parsed in ConfigLoader.cpp, which belongs to another P0 work package running in parallel; this commit only exposes the constants (kDefaultSpinUs, kMaxFramePayloadSize, FdPassing::kMaxSidebandBytes) so that plumbing has something to set.
- Plan B section 11 P0 asks for TracyPlot per-frame counters on both sides of the
boundary, and section 2.3.1 adds the deliverable v1 did not have: DYNAMIC call
counters. The static call-site counts everyone quotes (Espryt 124 / Magma 169) are
not the per-draw cost, because every one of those paths is memo-gated; without a
dynamic counter P2's verdict stays a guess. The tree had no per-frame byte or call
measurement at all - MG_Util/Metrics is format arithmetic, and Tracy has zones but
no plots.
- Cost when off: g_pipeStatsEnabled is a plain global Bool latched once in
Initialize() right after MG_ConfigLoader::Init(), and every counting site is
`if (PipeStats::Enabled()) ...` - one load of a hot global and one never-taken
branch. The counters are relaxed atomics because buffer and texture staging are
reachable from more than one thread; the off path never touches them.
- Eight byte classes (stage-buffer, stage-texture, stage-ubo-global, stage-ubo-named,
stage-vertex-client, stage-index-client, persistent-map-push and the
residual-value-block placeholder of section 6.3), six call classes and the six memo
gates of section 2.3.1, each as a hit/miss pair. Names are minted now, including the
two that stay 0 in P0, so no recorded baseline is invalidated by a later rename -
which is why the name set is pinned by a test.
- Reporting: TracyPlot per counter per frame when TRACY_ENABLE; with
MOBILEGL_PIPE_STATS=1 one MGLOG_I summary line every 120 frames and at teardown, plus
a JSON dump to MOBILEGL_PIPE_STATS_FILE when that is set. MGLOG_I against the usual
"MGLOG_D for non-critical" rule on purpose: the line has to survive an INFO build -
the only build a device runs - it is at most one line per 120 frames, and it exists
only when the operator asked for it.
- Summary lines report disjoint WINDOWS, not run totals: a run total over a workload
that changes shape (load, then steady state) averages away the very number section
2.3.1 wants an absolute value for.
- MOBILEGL_PIPE_STATS_FILE joins the six switches parsed in e48a3582; like them it
needs no allow-list entry because InitializeAcceptedEnvVariables accepts every
MOBILEGL_-prefixed variable by construction.
- Plan B section 4 makes the frontend/backend boundary explicit and gives the backend its
own state machine. This is the P0 deliverable of section 11: the whole catalogue exists
from day one, placeholders included, because the wire opcode is a call's position in
PipeCalls.def and record numbering must never churn.
- MG_Pipe/PipeCalls.def is the single source of truth: 68 unique calls as
X(Name, Payload, Class, Flags). Its header reconciles that number with the plan's
headline counts (section 4.4, appendix A), which double count - "CSO 15" names
bind_sampler_states and set_sampler_views that the "set_* 17" list also names, "screen
14" tabulates the query family that section 4.3 assigns to the context, and the
"transfer 12" row enumerates 11 calls. Each reconciliation is written down next to the
count rather than resolved silently.
- MGPipeHandles.h: the 8-byte {slot, gen} pair, dense per-kind slots, the reserved null
and default-framebuffer handles, and the ShaderCso composite band (sections 4.2, 5.6.3).
The two generations are documented as strictly separate, with the interface rule that no
call may require the client to know MGGen.
- MGPipeTypes.h: every payload of section 4.5 as a flat POD with explicit padding, a
trivial-copyability assertion and an exact sizeof assertion, because the wire records are
memcpy'd and a field silently changing width is a protocol break no test would see.
MGPCaps embeds DynamicBackendParameters by inclusion so a caps field added there needs no
second edit here; its assertion is stated as a composition because that struct still
carries SizeT. ResidualValueBlock is pinned at MGL_RESIDUAL_BLOCK_SIZE 1248, the ratchet
that only ever goes down and reaches static_assert(... == 0) in P13 (section 6.3).
- MGPipeHostSpan.h keeps the one shape that changes with the transport isolated behind one
predictable branch, with the kFromServerIndexMirror sentinel D-B7 needs.
- MGPipeCallbacks.h names the reverse channel as ten callbacks plus the forward terminator
in the context table, replacing 95 poke sites across 17 methods (section 7.1).
- scripts/gen_pipe.py runs G1-G7 off those .def files. G1 asserts each table is EXACTLY its
call count of function pointers; G3 pads every wire record to the stream's 8-byte
granularity and checks size >= sizeof && size <= remaining && size % 8 == 0 before
dispatch, fatally; G4 compares field by field (padding excluded, floats by bits) because
a comparator with false positives is one nobody reads - DirectGLES.cpp says the same
thing about its own memcmp of RenderStateParameters; G5 turns the accessor list into
per-verb poison generations rather than a written-once bitmap, which is the only version
that can see a field left over from the previous draw (section 6.2.2); G6 joins the 477
read points of the vendored backend_read_inventory.md against Coverage.def and reports
0 UNMAPPED (299 to a call, 167 signatures that become handle parameters, 6 reverse
channel, 5 client-resolved); G7 pins the pipeline subset BY MEMBER NAME from what
VulkanRenderer::ComputePipelineStateHash hashes today, computing no offsets in python.
- The generated files are committed so the build never depends on python; CI regenerates
and diffs them.
Program 203 reuses prefixSumCache for a second subgroup reduction before every workgroup invocation has consumed the first result. Add a fingerprint-gated SPIR-V pass that inserts the missing Workgroup acquire-release barrier while preserving native subgroup operations.
Keep the repair opt-in behind MOBILEGL_ITERATIONRP_FIX_BARRIER, cover insertion, pass-through, and idempotence, and enable it together with the existing iterationRP subgroup repairs for the matching Linux and Android CI retraces.
Add a deterministic iterationRP Program 203 fixture that dispatches the original shader and compares every RG16F texel against fixed half-float golden bits. This catches both a wrong exposure result and collateral writes without retaining a serial reference shader.
Make MobileGLIntegrationTest runnable as a standalone Android executable by linking the shared MobileGL library and backing EGL with an AImageReader window; desktop keeps its static-library pbuffer path.
Validation: Adreno 830 passes with 0/262656 mismatches; lavapipe reproduces the current reduction defect with 1/262656 mismatches at the exposure texel.
iterationRP's Program 203 declares shared vec2 prefixSumCache[32] for a
512-invocation workgroup indexed by gl_SubgroupID; any device narrower
than 16 lanes partitions into more than 32 subgroups and the pack writes
shared memory out of bounds (heap corruption on lavapipe's CPU
rasterizer, ssim 0.028 on the CI retrace). Fix it where the fault lies -
in the fixture - and keep the GL contract sound everywhere else:
- FixIterationRPSubgroupScratchPass: fingerprint-gated SPIR-V pass that
grows exactly that array to ceil(invocations/width) entries on sub-16-lane devices; every other module passes through byte-identical.
- DeriveNumSubgroupsPass stays default-on for the Adreno topology bug
and is made spec-sound: pipelines request REQUIRE_FULL_SUBGROUPS
whenever the workgroup shape makes the flag legal (computeFullSubgroups
enabled, local_size_x a multiple of the native width, subgroup count
within maxComputeWorkgroupSubgroups).
- EmulateSubgroupsPass: 32-lane virtual-subgroup lowering kept in-tree
as a last resort, enabled only by MOBILEGL_MAGMA_EMULATE_SUBGROUP=1 on
devices with no native subgroup support; fails closed on extended
subgroup instructions and on modules whose added scratch would exceed
maxComputeSharedMemorySize.
- IterationRPFirstReductionScenario skips gracefully outside the pack's
16..256-lane source domain; the new IterationRPScratchFixScenario runs
the fixture-shaped reduction on any width and asserts the exact
width-independent total. DriverPost keeps reporting FAIL on
out-of-domain devices.
- Program203 -> IterationRP rename throughout; the per-trace
num_subgroups_quirk plumbing is removed from the trace replayer, JNI
chain, and CI workflows.
Wave 2 of the advertised-extension conformance campaign.
PROGRAM INTERFACE QUERIES (the load-bearing piece). glGetProgramInterfaceiv
and the five glGetProgramResource* entry points were answered by the
BACKENDS - Espryt asked the real driver about SPIRV-Cross-generated ESSL
whose namespace is not the GL one (default-block uniforms live in
MGL_GLOBAL_UBO there), and Magma kept a second, partial reflection that
hardcoded types and diverged from the frontend. Both are now deleted; a new
frontend resource-model layer (ProgramInterface.{h,cpp}) answers every
interface - uniforms, uniform blocks, atomic-counter buffers (recovered
from glslang's synthesized gl_AtomicCounterBlock_<binding> lowering),
buffer variables, shader-storage blocks (classified by TType storage
qualifier since glslang reflects them as uniform blocks), program inputs/
outputs (built-ins' layoutLocationEnd sentinel mapped to -1), and
transform-feedback varyings including the gl_NextBuffer/gl_SkipComponentsN
pseudo-varyings - from the glslang reflection the frontend already trusts
for glGetActiveUniform. Name/index round-tripping, the "[0]" array
spelling, and the GL 4.6 table 7.2 prop/error matrix live in the new layer
only; GetActiveUniform*/GetActiveAttrib* are untouched.
glShaderStorageBlockBinding now takes the interface-layer index (the one
GetProgramResourceIndex returns, with a range check it never had), records
the binding on the program keyed by block NAME - the one coordinate all
three index spaces agree on - and delegates by name across the backend
boundary. Both backends reseed the recorded bindings on their own program
rebuilds, so an unrelated resync can no longer silently revert a rebound
block, and GL_BUFFER_BINDING reports the live binding, not the declared
one. The Espryt delegate applies only to an already-synced twin and can no
longer trigger SyncToBackend from a getter; the sync path's GL query
out-params are initialized and clamped (a load-dependent stack-garbage
Vector size crash caught by the gate, reproduced 3/50 pre-fix, 100/100
post-fix under saturating load).
ESPRYT RENDER-STATE SHADOW RESET (rides along because it shares
DirectGLES.cpp): the render-state shadow is file-static and survives
MobileGL context switches, so GL_FRAMEBUFFER_SRGB (and the whole synced-cap
class) leaked between contexts - the cross-test leakage class the CTS maps
have carried for a week. MakeCurrent now invalidates the shadow like it
already invalidates the program/FBO/buffer caches, and the resync resolves
a never-set scissor box to the current surface instead of pushing the
(0,0,0,0) sentinel verbatim (which scissored everything away - caught by
the retrace gate, bisected to the exact field via a bitmask probe, and
fixed by resolving like the viewport path rather than reverting).
Frontend riders exposed by the layer: glGetUniformLocation resolves
arrays-of-arrays element addressing ("a[2][1]"); transform-feedback capture
accepts element-addressed varying names ("b[1]") and snapshots the request
verbatim for the interface (Magma's decorate pass logs loudly that element
capture is unimplemented there - follow-up).
KNOWN GAPS, documented in code and tests: the 6 subroutines-* cases
(glslang refuses subroutine for SPIR-V; wave 3), the 5 separate-programs-*
cases (glslang's pipe-I/O reflection cannot see a separable non-vertex
stage's own inputs; needs stage-aware output validation first), and
uniform-block-types' per-instance stage masks (not derivable from the
reflection).
Gate: 593/593 unit at default and kill-switch, x10 each, plus the SSB race
case 100/100 under 20-way CPU load; ext caselist Espryt 77.87% -> 80.42%,
Magma 77.58% -> 79.39% (+212 fixed, 0 newly broken); program_interface_query
2/43 -> 31/43 unique on Espryt, 9/43 -> 31/43 on Magma, backends now
byte-identical; KHR-GL45.direct_state_access 370/371 + 371/371 with the 4
sRGB leak victims recovered in cross-test ordering; KHR-GL33 held at
9884/9886; full 39x2 CI retrace with zero wave-attributable failures (the
3 failing newly-added fixtures are bit-identical on the pristine baseline).
~21% of a shaderpack's glCompileShader calls hand different shader objects
byte-identical source; the P0b cache only helps after one finishes, so under
async two workers would run the whole pipeline twice. Now the GL thread
consults a per-context (stage, hash, length, envFingerprint) -> weak-node
map at enqueue and ADOPTS the in-flight (or completed) node instead of
posting a duplicate - a hit is honored only after a full byte comparison
(the hash never decides), a cancel-requested or settled-cancelled node is
never adopted, and no worker ever waits.
Sharing a node makes the unconditional cancel wrong, so release is now
adopter-counted: a plain GL-thread Int (every mutation site is a GL entry
point; the single-threadedness argument and the terminal-early-out that
keeps the count exact are in the header), and the cancel fires only at
count zero AND with no pending link pinning the node (the stage-4
MarkLinkReferenced precedence). Adoption also re-points the object's source
at the node's snapshot so the layer-1 memo's pointer compare stays armed -
without that, an adopter's next glCompileShader would re-enqueue the very
duplicate this stage removes. Both guards are negative-control-proven: each
removed guard fails exactly its own tests. Count discipline was proven with
a temporary hard-abort on underflow/leak across the full suite and retrace
corpus - zero hits.
18 new tests (13 GL-surface incl. shared-node re-source/delete/orphan-sweep
isolation, shared failure logs, 48-over-6 stress with a deterministic
adoption count, flag-off and KHR-suspended zero-adoption guards; 5 direct
map cases incl. fingerprint mismatch and cancelled/expired pruning).
Gates: 538/538 unit both flag states, async suites x5 no flakes, NVIDIA
DirectGLES retrace identical sets both states. Timing: 2-worker
(Android-shaped) 1-3% faster consistently on complementary and BSL;
4-worker unchanged - the win this stage exists for lands where CPU is
scarce.
glLinkProgram with the flag on snapshots its inputs in a GL-thread prologue
(stage-sorted shaders with their compile nodes taken without joining, env,
explicit locations/fragdata/xfb, draw-buffer count), then runs the whole
link body - glslang link/mapIO, SPIR-V, reflection, routing tables - as a
ProgramLinkTask that auto-posts when its last compile dependency settles
(+1-guarded countdown; no worker ever waits on another job). The publish is
one move of the LinkArtifacts block at the join, with the second version
bump so nothing memoized during the pending window survives.
The consume-once TShader claim moved onto the shared compile node as a CAS:
two link jobs racing for one shader resolve to winner-takes-the-parse,
loser re-parses the preprocessed source against the node's own env -
identical SPIR-V pinned by test for 2 and for 12 sharing programs.
Two deliberate corrections to the design's cancel matrix, both test-proven:
attach/detach do NOT cancel a pending link (the snapshot isolates it, and
glCreateShaderProgramv's link-then-detach would otherwise discard its own
result before anyone read it); and a compile node a pending link depends on
is pinned against the orphan-name sweep - the ordinary LWJGL teardown
compile/attach/link/detach/delete used to cancel the dependency and turn a
must-pass link into GL_FALSE.
Continuations are now throw-contained per-item (a stage-3 leftover made
load-bearing by the first real continuation), and the review's deadlock
find is fixed: the dispatch loop no longer cancels a node while holding the
pool mutex, since that cancel can run OnDepSettled -> Post -> same mutex.
Explicit joins: the draw path (GetProgramForDraw, both the pipeline stage
loop and the plain-UseProgram half) and the composite-link site; destroy
paths cancel-not-join; COMPLETION_STATUS readers stay non-joining.
Gates: 506/506 unit both flag states; AsyncCompile/AsyncLink/AsyncTeardown
suites x10 repeats clean both states (teardown with 128 jobs in flight,
then re-Initialize); full NVIDIA DirectGLES retrace flag on twice - result
sets identical to flag off, zero new deltas. Compile-phase prefix-diff,
flag on vs off: complementary-reimagined 5.21s -> 2.16s, BSL 1.72s ->
0.90s - past the design's final acceptance targets before the KHR
extension is even advertised. Default remains OFF until stage 5+7.
glCompileShader with MOBILEGL_ASYNC_SHADER_COMPILE=1 snapshots its inputs on
the GL thread (source SharedPtr, CompileEnv, cache handle) and runs the whole
pure pipeline - preprocess, validators, extractors, glslang parse - as a
ShaderCompileTask on the worker pool, returning immediately. Every read of
compile-produced state joins through the single Compiled() gate; links stay
synchronous this stage and join their attached shaders at the top of the
body. Flag off, the path is the same code run inline.
Mechanics: the job node owns all its inputs (no back-pointer, no lifetime
tie to the shader object), so re-sourcing or deleting a pending shader is
cancel-and-drop, never a wait; glslang worker hygiene is a TLS-allocator
scope guard plus GL-thread builtin prewarm (gated on the flag, latch reset
on Destroy so re-initialization re-warms); worker-side diagnostics defer
through the job and replay on the GL thread at the join, enforced by
IsPoolThread asserts in RecordError and an empty-deferred-errors tripwire.
A body that throws publishes a COMPLETE failed compile (status false, real
info log) rather than an abandoned node, and never memoizes away the retry;
a failed enqueue (OOM) cancels the node instead of stranding the joiner -
including inside the dispatch loop, where the in-flight slot is repaid.
The pool StopAndDrains from an atexit sentinel too: workers still inside
glslang parse while exit() ran static destructors was a real 2-in-5 SIGSEGV,
reproduced and fixed (15/15 clean after).
Backend-internal shader objects (default FS, DirectVulkan blit/mipmap) are
cache-less and always compile inline - compile-and-read-in-one-breath needs
no round trip.
Gates: unit suite 488/488 with the flag off AND on (x5); AsyncCompileTest
(12 e2e cases: pending re-source/delete/recompile, byte-identical failure
logs across modes, 48-compile cache stress) x10 repeats clean both modes;
full NVIDIA DirectGLES retrace identical result sets flag off/on (zero new
deltas); compile-phase timing flat as designed (links still serial - the
parallel win arrives with stage 4's async link + stage 5's
KHR_parallel_shader_compile).
Still fully synchronous - EnsureLinkJoined()/EnsureCompileJoined() are empty
inline no-ops (verified to fold away at every one of the ~1200 call sites;
this project builds without LTO) - but every read of link- or compile-produced
state now goes through a private accessor the compiler enforces, so when
stage 4 moves the bodies onto pool workers, 'which reads must join' is a
type-system fact instead of a 400-line audit.
- ProgramObject: the 31 fields ResetLinkArtifacts clears plus the 5 link
outputs it forgot (infoLog, linkedFragData{Location,Index}, the geometry
strip-capture pair) move into a nested LinkArtifacts behind Artifacts().
ResetLinkArtifacts is now a worker-safe pure clear; the link-observable
version bumps (backendState/link/uboContent) move to a GL-thread-only
BumpLinkObservableVersions() called once from Link()'s prologue and from
glProgramBinary's mandated failure - the link body never writes them, so
a stage-4 worker cannot lose an invalidation against the draw path.
- ShaderObject: compile artifacts (TShader, preprocessed source, side-channel
maps, status/log, consume-once flag) behind Compiled(); the P0b layer-1
memo trio deliberately stays outside as the future non-joining
COMPLETION_STATUS_KHR fast path.
- CompileEnv (new): a GL-thread snapshot of everything the compile pipeline
used to read live from the backend mid-parse - compute limits (the
GetIntegeri_v reach-back is gone from the worker path), advertised
extensions, device quirks, TBuiltInResource inputs. Captured lazily per
backend activation; the consume-once re-parse now runs against the same
env as the original parse.
- The GL-thread prologue / worker-body boundary is marked in Link() where
the stage sort ends; everything below is a pure function of the snapshot.
Public getter signatures unchanged - MG_Impl and both backends compile
untouched. Unit 476/476, Program suites 117/117, DirectGLES retrace 38/39 on
llvmpipe (the one failure is the known pre-existing non-CI iterationrp case;
the NVIDIA userspace driver was updated out from under the running kernel
module mid-session, so GLX there is down until a reboot).
Standalone Asio (submodule, asio-1-38-2 @ 8806a680, ASIO_STANDALONE +
ASIO_NO_DEPRECATED, header-only - no linked artifact) and the job machinery
the async shader pipeline will run on: JobNode (state machine with deferred
errors, continuations firing exactly once, dependency counters, cancel
semantics split into request vs outcome) and ShaderCompilePool
(asio::thread_pool behind a pimpl so no header leaks asio; big-core count
via cpufreq at >=85% of peak clamped to [1,4]; lazily constructed, so with
the flag off no worker thread ever exists; StopAndDrain leads DestroyImpl).
MOBILEGL_ASYNC_SHADER_COMPILE / _THREADS config knobs, default OFF. Nothing
in the GL pipeline references the pool yet - grep-verified; the full
DirectGLES retrace and compile benches are byte- and time-identical. 25
threaded unit tests, clean across 20x gtest_repeat.
Iris-style packs hand MobileGL the same source text repeatedly: probed across
three shaderpack traces, 28-32% of all glCompileShader work was redundant -
~9% same-object recompiles with byte-identical source, ~21% distinct shader
objects sharing identical source (the same common GLSL chunk glued into many
program stages). Two layers, both keyed by XXH64 + length with a full byte
compare on every hit (correctness never rides on the hash):
- Per-object: a successful (or failed) compile remembers its source hash;
glShaderSource with byte-identical text keeps the compiled state and
glCompileShader on unchanged source returns immediately. Deterministic
(stage, source) pipeline makes the memo observationally identical to
recompiling; the consume-once TakeShaderForLink re-parse path is untouched.
- Cross-object: a per-context bounded cache (ProgramState-owned, declared to
outlive every shader object) shares the preprocessed source, both explicit
side-channel maps, and the validation verdicts between objects with equal
source; only the glslang parse stays per-object. Single-GL-thread today;
flagged for a mutex when compiles go async (P1).
Interleaved A/B on the iterationrp trace (the recompile-heavy pack):
5.65s -> 5.46s median total replay, every round faster; BSL/complementary
stay flat (their duplicate sources are the small common shaders, so calls
drop but wall time is parse-bound on unique sources). Full DirectGLES
retrace, 445-test unit suite, and dedupe-semantics tests (no-op recompile,
invalidation on new source, failed-compile memo, cache bounds) all green.
RenameBuiltinShadowingFunction probed the whole source ten times per compile
(5 names x mask + scan, each a full-text pass) and still had two blind spots:
a 5-name list and single-line-definition-only detection. On Complementary-scale
packs (4.5MB of sources) that was ~68% of the compile phase.
The rename is now split by FAILURE LAYER, both halves sharing one name table
header so they cannot drift:
- A SPIR-V OpName pass in SanitizeAndOptimizeBinary covers the full ESSL 3.20
builtin set (~146 names). Renaming a function id is safe by construction:
builtin calls are GLSL.std.450 instructions and can never resolve to a user
OpFunction, overloads are distinct ids (a helper overload delegating to the
real builtin keeps working), dead preprocessor branches never reach SPIR-V,
and macro-expanded definitions are covered. ESSL 3.x is the only consumer
that forbids the redefinitions, and this pass runs before its transpile.
- A lexical pass covers only the 5 names whose exact-signature redefinitions
glslang's relaxed parse rejects outright (never producing SPIR-V for the
backstop): the historical fma/max3/min3/round/tanh. One TokenizeCode pass;
definition detection requires brace depth 0, a type-identifier previous
token that is neither a statement keyword nor a directive tail, and skips
files whose token-level braces do not balance (preprocessor-asymmetric
arms) - over-detection is unrecoverable, so every ambiguity falls through
to the backstop.
Measured on the compile phase (prefix-diff, 3-run medians, Espryt/NVIDIA):
complementary-reimagined 20.0s -> 5.5s, BSL 2.14s -> 1.85s. bliss (the pack
that ships from-scratch fma/tanh helpers) stays at SSIM 0.999962.
Tests: end-to-end ESSL assertions for the multiline-definition and
new-overload shapes, the three adversarial-review reproductions (statement-
keyword call under asymmetric braces, dead-#if compat shim, overload
delegating to the shadowed builtin), and a source-level assertion pinning
the lexical half specifically.
ES has neither glMultiDrawElements nor glMultiDrawElementsBaseVertex, so
both are emulated. DirectGLES had two ways of doing it - one
glMultiDrawElementsBaseVertexEXT where the driver has the extension
interaction, otherwise a per-draw loop. This adds the five MobileGlues
uses (gl/multidraw.cpp), so the ladder is now: one
glMultiDrawElementsBaseVertexEXT; one glMultiDrawElementsIndirectEXT
over a synthesized command buffer; one glDrawElementsIndirect per
command over that same buffer; the base-vertex replay; plain
glDrawElements over a CPU-rewritten index stream, for drivers with no
base-vertex draw at all; and a compute shader that flattens the whole
batch into one rebased index buffer drawn by a single glDrawElements.
They live in their own translation unit that owns the entry point
outright, preparation included - the compute tier has to dispatch BEFORE
PrepareForDraw, or it would have to unpick the program, storage-block
and index bindings the preparation just made, and a dispatch inside an
open transform-feedback span is not legal at all.
The auto ladder is ext -> basevertex -> multiindirect -> indirect ->
drawelements, which is NOT MobileGlues' order (it puts the indirect
tiers first). Measured on mc_sodium_multidraw, ns/op, median of three:
NVIDIA ES 3.2 basevertex 2500 vs multiindirect 5700 and indirect 5800;
Mesa llvmpipe ext 19300, basevertex 25200, multiindirect 27600,
drawelements 28700, indirect 31000. Ring-allocating the command staging
instead of respecifying per batch was tried first and moved the indirect
tiers by less than noise, so the cost is the indirect draw path itself,
not the upload; only a real multi-draw entry point beats replaying the
sub-draws. auto therefore resolves to basevertex on this box - byte for
byte the behaviour that shipped - and the new tiers are what a driver
with the ext interaction, or without base vertex at all, now gets.
compute is never chosen by auto (nor by MobileGlues'): it rewrites the
primitive stream rather than replaying it, and it measured slowest here.
Four places this deliberately does not follow MobileGlues, each a
correctness bug there. A rewritten stream is emitted as GL_UNSIGNED_INT
whatever came in, because GL adds baseVertex at full precision and
folding it into ushort indices wraps. The restart sentinel is carried
across a rebase unrebased, or an enabled primitive restart is lost. The
flattening tier declines strip/loop/fan modes, any sub-draw whose count
is not a whole number of primitives, and any batch at all while
primitive restart is enabled - a restart ends a primitive, so leftover
vertices would find a third vertex in the next sub-draw and become a
triangle GL never draws. And the indirect tiers decline client-memory
index arrays, which have no buffer to address.
gl_DrawID gets better rather than worse: the unrolled tiers now feed
each sub-draw its index (the spec's value, where the old loop left the
uniform untouched), and a program that actually reads it demotes the
batched tiers, which can only hold one value for the whole batch. The
per-batch cost is nil for the programs that do not read it.
Verified: the five DirectGLES retraces are byte-identical (md5) across
all six tiers on NVIDIA and on Mesa, each tier proven to have really
executed rather than silently demoted, via a per-tier announcement in
the log. Unit suite 421/421. The full retrace suite's five failures all
reproduce unchanged on a stashed tree, so none are new.
swung0x48
Swung0x48