Stage updates to adopted GPU-resident buffers on the command timeline instead of overwriting memory that earlier draws still read. Wait for submitted copies as well as pending commands before CPU readback.
Commit coherent mapped writes before Android trace buffer copies, and retire buffer shadows on deletion so copies see current vertices and name reuse cannot leave dangling dirty-shadow entries.
Validated draw ordering, submitted-copy readback, traced buffer copies, and mapped-buffer deletion/name reuse on Redmi Adreno 830. Minecraft 26.3-rc-3 Magma passed movement, HUD, and menu checks and exited normally; the new trace parsed all 64,410 frames without warnings.
- A window with no Present divided by a faked 1 and printed the window TOTALS under
"bytes/f[...]": the *MultiDraw* slice (47 draws, no present) reported 1,404,550 bytes as a
PER-FRAME figure, a 47x overstatement of exactly the SEG_STAGE sizing input plan B section
8.2 / section 11 P0 asks this package to produce. FormatWindowLine now relabels the bracket
to "bytes[...]" and prints draws/f=n/a when the window holds no frame; acc/draw=n/a follows
the same rule, because "0.00" beside a non-zero acc= is the same lie. Pinned by
PipeStatsTest.SummaryLineSurvivesZeroFrames and the reworked SummaryLineSurvivesZeroDraws.
- Every per-frame and per-draw field now goes through one FormatFixed2 helper. draws/f read 1
for 26 draws over 14 frames (1.86) and buf read 97 for 1360 bytes (97.14): a systematic
downward truncation of up to a whole unit on the figures the package exists to produce.
Pinned by PipeStatsTest.PerFrameFieldsKeepTwoDecimals.
- FormatSummaryLine rewrote the window bases as a side effect of formatting, so any second
reader silently zeroed the next window. Split into a pure FormatWindowLine() and an explicit
AdvanceSummaryWindow(); EmitSummaryLine calls both. Pinned by
PipeStatsTest.FormattingTwiceDoesNotConsumeTheWindow.
- Init() called ResetForTesting(), against the header's own "not used by any shipping path".
Both now forward to an internal ResetCounters().
- TracyPlot published only the MISS half of each gate under the gate's unqualified name, so
the headline output channel of section 11 P0 carried no denominator. Two series per gate now
("...-hit" / "...-miss") from static literal arrays. The payload histogram stays unplotted
and says why: it is a run-total distribution over draws (section 4.5.7), not a per-frame
scalar, and it reaches the operator through the JSON dump.
- GetOrCreatePipeline's 15-call tally sat ABOVE the list-topology primitive-restart refusal,
so a declined draw added ten reads it never made - an OVER-count, which breaks the lower
bound contract every other tally keeps. Moved to immediately before the payload build, and
the enumeration reconciled with the constant: the excluded read is the sample-shading
capability, short-circuited by m_sampleRateShadingFeatureEnabled.
- Six real staging paths were uncounted while the inventory claimed coverage. The inventory
claim "DirectVulkan's own buffer staging ... has no second copy to count" was simply false.
Now wired: MultiDraw.cpp's UploadScratch/UploadScratchRing (indirect commands, the compute
tier's draw-info array, the rebased index stream - the class is passed in, so a new tier
cannot forget it); Managers.cpp's pool-recycle reseed and the VBO-backed Float64 narrowing;
VkBufferManager's eight host->device copies of buffer contents plus UploadTransient, the
single chokepoint for Magma's per-draw vertex/index/indirect staging; VkTextureManager's
packed staging slice, with the same box/rect SHAPE split Espryt already reported.
- New byte class stage-indirect-cmd, for draw PARAMETER bytes a backend synthesises and
stages. Kept out of stage-index-client because these are the population that becomes MGPipe
command-record payload (section 4.5.7), not resource bytes. A name addition, not a rename:
no recorded baseline is invalidated.
- stage-ubo-named moved past the zero-copy direct-bind decision in ResolveUniformBufferPayload
(a direct bind repacks nothing, so counting it there reported a copy that never happened),
and Magma's default-uniform-block image now feeds stage-ubo-global the way Espryt's does,
counted after the per-frame slice memo.
- The site inventory in PipeStats.cpp is rewritten to name every unwired path by file and
function. An inventory that overstates coverage is worse than a missing counter, because
the zero is then read as an answer.
- Evidence, lavapipe/llvmpipe, MOBILEGL_PIPE_STATS=1: the MultiDraw slice now prints
"window=0 draws/f=n/a bytes[buf=1404550 ...]"; the indirect tiers
(MOBILEGL_ESPRYT_MULTIDRAW_MODE=indirect|multiindirect) move icmd 0 -> 660; Magma's GuiBatch
line moves from buf=0 tex=0 ubog=0 to buf=685.71 tex=41.14 ubog=157.71 with tex[emit=9
box=9 rect=0 jobs=9]. Off-path A/B against the base tree with the env unset, 9 runs each of
a 40-scenario draw slice, sorted totals in ms: Espryt base 389..794 (median 399) vs branch
384..403 (median 394); Magma base 406..472 (median 410) vs branch 408..761 (median 414) -
the always-compiled guard is below this harness's noise on both backends.
- 1410 unit tests green (15 PipeStatsTest). integration-gpu 860/860, 860/860, 859/860; the one
failure is DirectVulkan.PointSizeDemotion...TheDemotionIsActuallyArmedWhenTheEnvironmentPins-
ItOn, a member of the load-dependent *IsActuallyArmed* flake family already present in the
untouched base tree, and it passes 8/8 standalone here. stdio gate and gen_pipe check green.
Every draw re-resolved its whole vertex binding array: for each enabled binding,
look up the buffer, acquire a slice from the buffer manager, apply the binding's
base offset, fill the VkBuffer and offset arrays, bind. In the Minecraft-shaped
benchmark the same few hundred vertex array objects cycle for the whole run and
each one's answer is stable, so UploadAndBindVertexBuffers was the single largest
cost in the backend at 7.9% of the render thread, with AcquireResidentSlice
another 3.8% underneath it.
The resolved array is now kept per vertex array object and revalidated instead of
rebuilt. Validation is two-tier. The vertex array's own configuration version
already invalidates its backend vertex-input state, so a changed attribute,
format, buffer or base offset yields a different state object - the memo compares
both that object's address and its hash, which mixes the bound buffers and the
whole layout. What that does not cover is the slice moving underneath an
unchanged configuration, so the buffer manager now carries a monotonic epoch that
every writer of slice-deciding state bumps: resident storage creation, respecify,
sub-data, flush of a mapped range, the promotion and demotion between streamed
and resident storage, each fresh arena allocation, and bulk release. The counter
is manager-wide and never reset, so a resource created at a recycled address
cannot reproduce a value some memo still holds.
The miss path was the thing to get right, because the previous attempt in this
area regressed the texture-upload and sampler-churn cases by 60-85%: it added a
verification pass that re-ran the resolution work it was trying to skip, so every
miss paid for it twice. Here a miss is one pointer-keyed lookup and a few stores,
and nothing else runs that the full path would not have run anyway.
ns per draw, DriverBench on a GTX 1660 SUPER: mc_ubo_range 924 -> 767,
mc_vanilla_draw 1346 -> 1227, mc_sampler_churn 1397 -> 1279,
mc_sodium_multidraw 3365 -> 3266. Magma is now 4.1x the native driver on the
per-draw uniform-range case, from 5.4x when this round started. No case
regressed on either backend.
Unit tests 421/421.
A per-draw CPU profile of a real Minecraft frame (perf on the render thread,
which sits at 100% of one core on both backends) said the deficit is translation
overhead, not the GPU, and named where it goes. This removes the largest items
it found, on both backends and in the shared frontend they both feed.
The single biggest one was not translation at all: IsBackendContextCurrentOnThisThread
called eglGetCurrentContext on every invocation, and glvnd answers that with a
getpid() fork check - a real syscall. The predicate sits two and three deep in
every draw (the deferred-release drain, the global-UBO ring availability check,
and the ring allocation), so it accounted for 16.3% of the render thread. EGL is
still the ground truth, but re-verifying it once per thread per frame catches an
external migration at the next frame boundary rather than the next call, which
recovers the same bookkeeping.
Texture uploads now carry a dirty region instead of a per-level flag. Minecraft
animates atlas sprites with 16x16 glTexSubImage2D calls into a 1024x512 atlas
and respecifies the lightmap every frame; a per-level flag turned each of those
into a full-level re-upload - about 3.6 MB a frame of texels nobody changed.
MipmapStorage accumulates the written box, Espryt uploads it with
UNPACK_ROW_LENGTH striding into the level shadow, and Magma stages just that box.
The box is a union, not a range list: repeated writes to one level widen it and
it degrades to exactly the old whole-level upload, which is the honest worst case.
glBufferData(NULL) is the orphaning idiom, and the backend was answering it by
uploading the stale CPU shadow - turning a rename the driver does for free into
a full synchronized upload. BufferObject now records that a NULL respecify leaves
the store undefined, and the upload is skipped until content is actually written.
The rest are smaller and of a kind: the deferred-release queue is probed without
taking its mutex, the UBO ring waits on the frame fence that frees the space it
needs instead of draining the whole pipeline with glFinish at the size cap, VAO
binds go through a shadow so a draw's second bind of the same object does not
reach the driver, the per-draw clean-texture probe short-circuits on the content
version before rebuilding shape info, glUniform drops byte-identical writes
(which otherwise dirty the whole UBO for the next draw), re-binding the texture
or VAO a slot already holds no longer bumps the generation counters a backend
fast path is keyed on, and the texture validators stopped taking shared_ptr by
value.
On Magma: descriptor-set reuse keeps four entries instead of one, because draws
alternating between two programs - the chunk/entity ping-pong - thrashed a single
slot into a full re-allocate and re-write every draw; a DynamicDraw buffer whose
contents survive two frame boundaries is promoted to resident storage instead of
being re-copied into the per-frame arena forever; and sampled-read barriers name
only the shader stages whose device feature is enabled, which also removes a
latent VUID violation (ALL_GRAPHICS names geometry and tessellation stages a
device need not have).
Measured with the Minecraft rig (render distance 32, p50 fps, same machine,
single sample each): vanilla 1.21.1 Espryt 10.8 -> 36.3 and Magma 31.3 -> 44.6;
26.2 snapshot Magma 114.5 -> 210.5. Fabric+Sodium moved inside noise on Magma
(854 -> 766) with the native baseline itself moving 838 -> 1031 between the two
sessions, so treat that cell as unresolved rather than a regression measured.
Unit tests 421/421. The CTS A/B was not run: these numbers and the test suite are
the whole of the evidence, and a conformance regression would not have been
caught here.
Five independent bits of per-draw and per-operation waste in the DirectVulkan
backend, all removing work whose answer was already known.
The per-draw descriptor walk iterated all 256 slots of bindingKinds to find the
one to eight bindings a real GL program declares, because that vector is sized to
the binding cap rather than to the program. Reflection now records the bindings it
actually assigned, and the draw path iterates that. It is built at the end of
ReflectLayout, not where bindingKinds is sized - at that point the vector is only
zero-initialised and the kinds are assigned further down, so a list built there
would be empty. It has to stay ascending: Vulkan consumes pDynamicOffsets in
binding order and the writer pushes them in iteration order, so an unordered list
would silently mis-pair dynamic offsets with their uniform blocks.
Descriptor pools were sized maxSets * the 256-binding cap, declaring 81,920
descriptors per pool and 245,760 across the frames in flight, for sets that hold
what shader reflection found. Sized from eight now; an outlier program is absorbed
by the VK_ERROR_OUT_OF_POOL_MEMORY path that already exists, which works because
pool sizes are aggregate budgets rather than per-set limits.
TrackLiveResource swept the whole live-buffer vector on every insert once it
passed 256 entries, and when the buffers are all live the sweep removes nothing
and the vector grows by one - so creating N live buffers cost about N^2/2
expired() checks. It sweeps on a doubling watermark now, with the same
reclamation semantics.
GenerateMipmap transitioned each destination level individually inside its loop,
but every generated level starts in the same layout and the loop only moves a
level out of TRANSFER_DST after writing it, so the whole range can be prepared in
one barrier - 3(N-1)+1 barrier commands become 2(N-1)+2. Each level is still
transitioned to TRANSFER_SRC before it is read, so the dependency between
consecutive levels is unchanged.
WaitForFrameSerial drained the entire graphics queue, as its own comment admitted.
Every submission records the frame serial it was made under, so it now waits on
the first fence at or past the requested serial. The narrow path deliberately does
not call NotifyDeviceIdle(): that claims every submission has retired, which is
only true after a real drain, so it stays on the fallback.
Verified with an 8213-case A/B (textures, buffers, queries, mipmaps, uniforms and
the whole direct_state_access suite): the Espryt failure list is identical, the
Magma failure list differs by one case, and both crash sets are unchanged on
Magma. That one case, buffer_storage.map_persistent_draw, does not reproduce in
isolation - running the buffer_storage group alone gives byte-identical results on
both builds (the same three failures, not including it), and it reports
NotSupported when run on its own. It is the same ordering-dependent behaviour this
suite shows elsewhere, and the three Espryt crash-set differences are the known
copy_image cluster moving chunk position. Flagging rather than hiding it.
direct_state_access stays at Espryt 370/371 and Magma 371/371; unit tests 421/421.
AcquirePersistentMap promises the storage it creates is never recreated, because
the frontend adopts it in place of the shadow and hands out pointers into it.
AcquireStreamedSlice broke that promise: its downgrade path releases the resident
storage unconditionally to avoid keeping a second stale copy, so binding such a
buffer as a vertex or index source freed the memory the application was still
pointing at.
It also fed that draw the wrong bytes. The streaming copy is uploaded from the
shadow, and a persistently mapped buffer can hold bytes the shadow never saw -- a
transform feedback capture writes straight into the resident storage. The next
capture into the same buffer then landed in freshly recreated storage while the
application kept reading the original, which is how the ping-pong in
transform_feedback.draw_xfb_feedbackk_test stalled after its first doubling.
Route a persistently mapped resource to the resident path instead, where its
single piece of storage is bound directly.
Reading a buffer a compute shader wrote gave zeros: the frontend shadow that MapBuffer
resolves against is only maintained by uploads, and Magma had no path back. Every
KHR-GL40.texture_gather case ends by dispatching a compute shader into an SSBO and
comparing the mapped result, so 66 of 75 failed on it.
Magma needs no readback: EnsureGpuResidentStorage - the same host-visible coherent
adoption the transform feedback capture already uses - makes the shadow BE the memory
the shader writes, so binding a buffer as a shader storage buffer now adopts it. What
coherence does not give is ordering: the writes are visible once they have happened,
and the CPU was reading before the dispatch had retired. The readback op therefore
submits the recorded work and waits.
That exposed a mistake in the frontend flag this rides on: MarkGpuWritten skipped
GPU-resident buffers, reasoning there was no shadow to refresh. True, but the wait is
still needed - "reconcile with the GPU write" is not always "copy it back", and which
of the two it is belongs to the backend. The flag now only says a write is outstanding;
DirectGLES's readback still skips its persistent-mapped buffers when copying.
KHR-GL40.texture_gather on Magma: 66 failures -> 19 (the rest are rectangle textures,
mipmap completeness and tessellation, all still to do). Espryt stays at 75/75.
Second stage of GL 3.0 transform feedback: captured draws now write real
data.
- Device setup enables the VK_EXT_transform_feedback feature when present
and loads the bind/begin/end entry points.
- Captured draws compile an XfbCapture program variant whose last
vertex-processing stage gets XfbBuffer/XfbStride/Offset decorations from
the program's resolved varyings (a new spirv-opt pass). A captured
gl_Position is mirrored into a dedicated output written before every
OpReturn - or before every OpEmitVertex in a geometry stage - ahead of
the position fixup, so the captured value is the shader's own pre-remap
position.
- DrawArrays/DrawElements wrap the draw in Begin/EndTransformFeedbackEXT;
a small counter buffer resumes the append position across draws within
one glBeginTransformFeedback (fresh Begin starts at the bound offsets).
- Capture targets are promoted to persistently-mapped host-coherent GPU
storage (persistent-map storage now also carries the transform feedback
usage), so MapBuffer/GetBufferSubData read the captured bytes after the
fence wait glEndTransformFeedback now performs.
- Draw-mode/feedback-mode validation defers to the geometry shader's
output primitive when one is present, and glGetBooleanv reports
GL_TRANSFORM_FEEDBACK_ACTIVE/PAUSED so dEQP's per-case state reset can
unwind an active capture.
KHR-GL33: transform_feedback capture_vertex_*/capture_geometry_*/
discard_*/draw_xfb and clip_distance.coverage now pass; queries
(PRIMITIVES_WRITTEN) and gl_ClipDistance capture remain.
Introduce a Mesa pipe_resource-style PipeResource that owns a GL buffer's bytes
and its backend GPU resource, abstracting WHERE the authoritative bytes live:
- Shadow mode (non-persistent buffers): a CPU Vector; the backend keeps its own
GPU copy in sync via BufferBackendOps, exactly as before.
- Persistent mode (coherent GL_MAP_PERSISTENT maps): the backend's host-visible,
COHERENT, persistently-mapped GPU memory is the single source of truth. The app
writes into it directly, every reader resolves against it, and NO per-write
backend transfer happens. The CPU shadow is released.
BufferObject no longer owns a raw shadow Vector; it holds a PipeResource and
exposes one accessor, MappedData(), that all readers go through. Every buffer-data
consumer (UBO payload, PBO texture upload, indirect draws, resident/streamed
uploads, both backends) was migrated from GetDataReadOnly()->data() to
MappedData(), so a persistent buffer's readers see GPU memory - not a stale
shadow. That stale-shadow inconsistency is what corrupted rendering (wrong UBOs ->
misplaced/"lost" vertices) in the first zero-copy attempt (625c8a6, reverted in
896cafc); routing every consumer through one accessor makes it structurally
impossible.
Backends provide the map via BufferBackendOps::AcquirePersistentMap:
- DirectVulkan: a HOST_VISIBLE|HOST_COHERENT (required, not just requested),
persistently mapped resident VkBuffer carrying every usage, seeded from the
shadow, never recreated; AcquireResidentSlice binds it directly.
- DirectGLES: EXT_buffer_storage immutable persistent+coherent glMapBufferRange,
falling back to the shadow when the extension is absent.
Fixes the ~7GB GpuMemory OOM + 100%-CPU/ANR running modern Blaze3D Minecraft on
both Magma and Espryt (per-draw whole-buffer re-upload of the coherent persistent
ring buffer), without the coherency/stale-read hazards of the reverted attempt.
BufferTest: zero-copy stress guard (15,360 draws -> 0 per-draw transfers, and every
reader resolves to GPU memory) + a shadow-fallback test. Host suite: 203/203 pass.
Device verification pending.
- Track every graphics-queue submission with a real fence: pooled fences
for mid-frame flushes, the frame slot's fence for Present and readback.
Completion advances a submit counter via vkGetFenceStatus polls,
slot-fence waits, and device-idle points, and raises the buffer-manager
serial floor from the frame serial each submission carried.
- GL sync objects now capture the submission index that will carry the
commands recorded so far; ClientWaitSync honors
GL_SYNC_FLUSH_COMMANDS_BIT with a mid-frame submit (gated on the index
still being unsubmitted so poll loops cannot split the render pass), and
blocking waits flush then vkWaitForFences with the caller timeout.
- FlushPendingCommands retires the submitted command buffer and restarts
recording on a fresh one; retired buffers are freed once the slot fence
is next waited, so an executing buffer is never reset.
- Rewind descriptor-set cursors exactly once per frame in Present (after
the slot-fence wait), plus after the synchronous readback drain,
replacing the ten lazy per-draw-path rewinds.
Verified: host tests 168/168, trace-replay 70/70.
Create 6 / Flywheel 1.0.6 now renders correctly with both flywheel:instancing
and flywheel:indirect on DirectGLES and DirectVulkan (verified in-game on
Adreno 830: waterwheels and cogwheels solid, animated, correct pairing, no
crashes across all four combinations).
- MG_State/MG_Impl: sync explicitly-ranged SSBO bindings of FLUSH_EXPLICIT
persistent maps to the backend before compute dispatches. Flywheel writes
its scatter-copy descriptors into the staging ring's persistent map and
never flushes that span (UB per spec, works on drivers whose maps alias
GPU-visible memory); our maps alias the CPU shadow, so the descriptors
never reached the GPU: the scatter compute copied nothing (GLES: empty
draw commands) or stale garbage (Vulkan: wild indirect commands ending in
VK_ERROR_DEVICE_LOST).
- MG_Impl/MG_Backend: real glFenceSync objects backed by backend fences
(GLES: native ES syncs guarded by context generation and owner thread;
Vulkan: buffer-manager frame serials), replacing always-signaled stubs
that let Flywheel reclaim staging memory the GPU still reads.
- MG_Backend/DirectGLES: compute dispatches now run the same per-program
resource sync as draws (uniform-block bindings and sampler units must be
re-established through the API because layout(binding) is stripped from
transpiled ESSL) and rebind texture units afterwards; the cull shader
used to read a stale _FlwFrameUniforms binding and the depth-pyramid
downsample sampled a stale unit-0 texture, zeroing the Hi-Z pyramid and
occlusion-culling all Flywheel geometry. Image uniforms are excluded from
glUniform1i (ES bakes their unit via layout(binding)); image-unit sync is
clamped to the device limit; eliminated/SSBO-classified uniform blocks
are skipped.
- MG_Backend/DirectGLES: gl_BaseInstance in native indirect draws reads the
GPU-written command buffer through an injected mg_IndirectParams SSBO
view addressed per draw instead of the zero CPU shadow; layout(binding)
is preserved for SSBO/image declarations (ES has no API rebinding for
them); the ES context ownership claim moved to a global atomic owner
thread with an EGL ground-truth check, and deferred buffer op state is
mutex-guarded, so ops cannot silently no-op after context migration.
- MG_Backend/DirectVulkan: new RebaseInstanceIndexPass rewrites vertex
InstanceIndex loads to (InstanceIndex - BaseInstance). glslang's relaxed
Vulkan mode aliases gl_InstanceID to InstanceIndex, which includes
firstInstance, but GL's gl_InstanceID is zero-based - draws with nonzero
baseInstance paired meshes with wrong instance data (cogwheel drawn as a
waterwheel, another wheel collapsed invisible). Gated on the
shaderDrawParameters device feature. Sampled-read barriers additionally
cover the compute stage (the Hi-Z downsample samples the depth
attachment from compute), and short uniform-buffer ranges keep the
existing zero-padding.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- Support cube map face uploads with cube-compatible images and per-face array layers
- Add uniform texel buffer descriptor support for samplerBuffer bindings
- Cache transient vertex/index buffer uploads per frame to avoid VMA allocation failures