Run the lavapipe Program 203 golden test with the same subgroup scratch, derived topology, and missing-barrier repairs as the iterationRP retrace matrix. This keeps the prerequisite integration job from failing before retrace jobs can start.
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.
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 benchmark job is the only one that brings a real GL context up - DriverBench
dlopens libEGL.so.1 and renders through it - but its apt list only asks for
libegl1, which is glvnd's dispatch layer and nothing more. The vendor library
behind it, libegl-mesa0, has been arriving as a Recommends of libegl1 rather
than because anything asked for it. That is too quiet a dependency for the one
job whose whole purpose is running a driver: a base image change, or
--no-install-recommends turning up anywhere upstream, would leave eglInitialize
with no vendor to dispatch to and fail the job for a reason nothing in the
workflow explains. Name it, next to libgl1-mesa-dri, which is listed for
exactly the same reason.
Verified with a full headless ctest -C Release -L benchmark - no $DISPLAY, no
$EGL_PLATFORM, mesa as the only EGL vendor: SanityBench, ProgramBench,
BufferBench and DriverBench all pass.
The fetch script tries git.hit.moe, then the repo.miawa.cn mirror, and only
then Git LFS, but it bailed out of the mirror loop on the first file no
mirror could serve and then pulled the whole case from GitHub. A case whose
mirrors served every file but one paid GitHub's LFS bandwidth for all of
them.
Collect the files that survived every mirror and every retry instead, and
scope the LFS fallback to just those, matching what the local macOS retrace
helper already does.
swung0x48
BZLZHH