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.
The previous commit's fingerprint was pinned to one array's incidental
dimensions - workgroup exactly 32x16x1, element exactly vec2, length
exactly 32 - which is the auto-exposure reduction and nothing else. The
pack ships the same idiom twice:
- auto-exposure: 32x16 (512 invocations), shared vec2 prefixSumCache[32]
- RTW warp: 1024 invocations, shared float prefixSumCache[64]
so the warp kept writing 128 subgroups into 64 entries on an 8-lane
device and the retrace stayed bit-identically wrong (ssim 0.027902).
Key the fingerprint on the pack's idiom instead of one array's shape: a
workgroup array of 32-bit floats indexed by gl_SubgroupID, fed by a
subgroup scan, whose declared length is below ceil(invocations / native
width). Three properties keep that a targeted repair rather than a
general array resizer:
- the index must BE gl_SubgroupID (through OpCopyObject, a signedness
OpBitcast, or a spill whose every store is that id), so an index
masked or clamped into range is left alone;
- the >= 16-lane early-out is retained, so every module on the devices
the pack was written for passes through byte-identical;
- growth is certified against maxComputeSharedMemorySize using a
natural-alignment layout model, and declined outright when a
declaration cannot be sized, so a patched module can never fail
pipeline creation where the original would not have.
Verified against the shaders the CI trace actually contains: of the 14
compute modules in the fixture exactly these two change, the other
twelve are byte-identical, and all fourteen pass spirv-val. The
integration scenario grows a second case for the 1024-invocation shape;
both abort with heap corruption when the patch is disabled.
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.
The draw fast path still paid for its own proofs: the hottest single
load (20% of TrySetupDrawFastPath) was chasing the cold
VertexInputStateFactory heap entry just to answer "same vertex-input
layout?". That answer now comes from the frontend VAO's config-guarded
aux memo (layout hash + attribute masks), and a VAO-cycling stream with
a stable layout skips the pre-flight AND pipeline re-resolution
entirely. The VkProgramObject* is memoized on the snapshot behind a new
ProgramFactory cache-structure epoch (bumped on every insert/erase; use
is re-stamped so the idle sweep can never evict a live entry). A
render-state version move no longer forces the full path: the pipeline
value hash is refreshed in place and the 8-entry memo probed directly
(the GL_BLEND-toggle case).
The resolved-vertex-bindings memo now revalidates all-resident unmapped
entries ACROSS frames via per-binding slice epochs - minted from a
process-lifetime counter so a recycled address can never revalidate,
with every mutation path funnelled through BumpSliceEpoch - while
stamping each resource's GPU-use serial exactly as the skipped acquire
would, preserving the busy-tracking that glBufferSubData's
host-write-vs-staged-copy choice depends on. Resident index buffers get
the same treatment through an EBO slice memo.
The six-part dynamic-state tail (viewport/scissor/blend constants/depth
bias/line width/stencil) is gated behind one render-state-parameters
version + pass-geometry compare per command buffer. GetSlice is inlined;
SampledBindingsUnchanged walks only the program's declared bindings.
Quiet-box 6-round order-alternating A/B (on top of the frontend
VAO-bind commit): vanilla_draw -20.8% (790 -> 626 ns/op, 3.4x native to
2.5x), sampler_churn -28.2%, ubo_range -8.4%, state_toggle -3.8%;
tex_param's matrix flag (+10%) was adjudicated by an isolated
alternating re-run at +1.0% - position bias, not regression. Unit tests
421/421.
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.
glVertexAttribLFormat validated its arguments and then refused unconditionally
with "64-bit vertex attributes are not supported", so
direct_state_access.vertex_arrays_attribute_format failed every GL_DOUBLE
subcase on both backends - the format never landed, the draw fetched whatever
the attribute held before, and the captured values came back as reinterpreted
garbage.
The attribute is now real state. IsLong is its own bit rather than being
inferred from Float64, because glVertexAttribFormat(GL_DOUBLE) also reads
doubles - it just asks for them converted to float - so the type alone cannot
tell the two apart. It participates in the format comparison, so an L-format
call over a plain one still bumps the version, and glVertexAttribPointer clears
it inside the mutation block so the clear and the bump stay atomic.
GL_VERTEX_ATTRIB_ARRAY_LONG stops being hardcoded false, and the pname is now
accepted by the attribute queries at all.
Support is detected, never assumed. SupportsFloat64VertexAttributes comes from
VkPhysicalDeviceFeatures::shaderFloat64 on DirectVulkan and is false on
DirectGLES - not a driver question there and never will be, since ES has no
GL_DOUBLE vertex format and ESSL has no fp64 type to consume one with. A backend
without it declines in the entry point, with the GL error and a log line naming
the reason, rather than accepting state no draw could honour. Both cases get a
DriverPost row so the loss is named at startup instead of at draw setup.
On DirectVulkan the attribute deliberately does not use VK_FORMAT_R64*_SFLOAT:
those are optional and lavapipe advertises zero features for all four of them.
It is fetched as its 32-bit word pair (R32G32_UINT / R32G32B32A32_UINT) and
bitcast back to double in the shader by a new SPIR-V pass, which is bit-exact
and needs no format capability at all. The pass re-declares the input as uvec2 /
uvec4, demotes the original variable to a Private global and seeds it once at
the top of the entry point, so every existing load keeps its id and its double
type and no other instruction is rewritten. Both halves branch on nothing but
"is this attribute long", so they cannot disagree - and if the pass ever fails,
the assertion fires rather than letting a UINT format sit under a double input.
The pointer types are all created before any variable that names them and the
demoted variable is moved after them, since the types-and-variables section may
not forward-reference a type.
dvec3/dvec4 are declined rather than fetched wrong: six or eight uint32
components have no single VkFormat, and GL spreads such an input over two
attribute locations, which the location-per-index model here does not express.
Fixes vertex_arrays_attribute_format on Magma (369/371). On Espryt it stays
failing, now as a detected and explained decline rather than a blanket refusal.
The program cache is content-hash-shared across GL program names, so its key has
to cover everything that changes the modules it stores. The capture layout did
not: XfbCaptureDecoratePass bakes XfbBuffer/XfbStride/Offset into the SPIR-V from
the frontend's layout, none of which is in the SPIR-V being hashed.
Two programs with identical shaders and different glTransformFeedbackVaryings
therefore shared one entry, and the first one linked decided how both captured.
That is precisely what changing the buffer mode does -- the same varyings
recorded with GL_SEPARATE_ATTRIBS instead of GL_INTERLEAVED_ATTRIBS -- so the
separate-attribs pass of transform_feedback.draw_xfb_test replayed a capture that
was still interleaved into buffer 0.
Hash the captured varyings' names, buffer indices and offsets plus the per-buffer
strides, and only for a capturing compile, so no other program changes key.
Neither target API has GL_TEXTURE_RECTANGLE: ESSL has no rectangle sampler, and
Vulkan's SPIR-V environment does not allow Dim::Rect. Both emulate it on a plain 2D
texture, and the two differ in exactly one way - a rectangle lookup addresses texels
where a 2D one addresses [0,1].
That one difference now lives in one SPIR-V pass, so neither backend has to know about
it: every lookup taking normalized coordinates gets its coordinate divided by the size
the texture reports, and the image type is then rewritten to 2D. Magma had no rectangle
handling at all - it fed Dim::Rect straight to Vulkan, which read the texel coordinates
as normalized and sampled the edge, so all fifteen KHR-GL40.texture_gather.*-2drect
cases came back holding the clear colour.
This replaces the ESSL text rewrite that did the same divide for DirectGLES only. Doing
it in the module instead is both shorter and stricter: the pass resolves an operation's
image type through the sampled-image and pointer wrappers rather than matching a
sampler name in generated source, so it cannot be fooled by an expression where it
expected an identifier, and it needs no help from the frontend reflection to know which
samplers were rectangles.
Still declined, as before: the Dref *sample* forms, whose coordinate carries the compare
value in its last component, and the projective ones, where the divide would have to
happen after the perspective divide. texelFetch is deliberately untouched - integer
texel coordinates mean the same thing on both targets.
KHR-GL40.texture_gather: Magma 66 failures -> 2, 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.