The RDNA4 offload campaign left three HIP probes and a driver sitting loose in
$HOME, where the next agent cannot find them and a stray `rm -rf` ends them. They
answer a recurring question on the R9700 box — is decode limited by the link or by
the engine's read path — so they move into scripts/probes/ with a driver, tests,
and a page.
Two things changed in the code, not just the address:
- The driver measures only after a guard that runs BEFORE anything is built:
gpu_busy_percent over PCIE_PROBE_MAX_BUSY, or any process holding /dev/kfd, is a
refusal (exit 3) rather than a measurement. A benchmark taken beside a live
endpoint is both disruptive and worthless, and silently unloading a production
service is not a tool's decision — --force is the human saying otherwise. The
refusal-before-build ordering is pinned by a test, because the sysfs and kfd
paths are overridable and a refusal is otherwise untestable off the card box.
- hip_smoke.hip and pcie_probe.hip called hipGetDeviceProperties, which SEGFAULTS
on this ROCm build even though hipGetDeviceCount works; only pcie_bw.hip avoided
it, and that is the one that produced the campaign's numbers. Device identity now
comes from the sysfs topology block the driver prints, so all three probes are
runnable on the build we ship. Verified here: refusal on a real busy APU (exit 3),
missing toolchain (exit 4), missing sources (exit 2). Compile-and-run still needs
the box with hipcc.
Refs #53
`ssh host 'cmd | tail -f log | sed -n /ready/q'` reported success on jobs that
never finished, three structural ways: the broker aborts the long connection, a
readiness pattern can match a log header printed before the work starts, and
`sed -n /x/q` exits the instant it matches — tearing down the very command it
was watching. The replacement runs the job detached on the host and writes
NAME.out, NAME.rc, then NAME.done LAST, so status is a cheap read of files over
a short-lived connection. That survives the connection, the restart, and the job
that restarts the endpoint the agent itself is running on.
Black-box tests pin the sentinel ordering (rc before done), that a finished-but-
failed job is DONE rc=N with the report still exiting 0, and that a name with no
state is never reported DONE.
Refs #53
Frank