6517401f01
Every path-gated workflow begins with one small job -- `does this change touch
it`, or `which workspaces this change touches` for the standards build -- that
all of its real jobs declare in `needs`. All 55 of those gate jobs ran on
`linux-arm64`.
That is the one label the fleet cannot spare. Sampled while 267 runs were
queued, every waiting run had the same shape:
[waiting] linux-arm64 does this change touch it
[blocked] linux member · dataflow device, simulated
[blocked] macos member · dataflow device absent
237 runs, each holding a gate in the longest queue, and every real job blocked
behind it. Not only `linux` jobs: `macos`, `linux-hw`, `linux-arm64-hw` and
`macos-m3-ultra` jobs were blocked too. The measured effect was `linux-arm64`
263 deep against `linux` **zero**, with jetson-hub's 20 cores taking no job for
a full hour -- starved not for want of work but because the work could not be
unblocked. A gate waiting on the resource it is rationing is head-of-line
blocking, and it makes every other runner in the fleet idle on its behalf.
The gates move to `linux`. Nothing else changes: the diff is 55 `runs-on` lines
and no others, and every touched file still parses.
Why `linux` and not elsewhere. The gate is not free -- it compares
`BASE...HEAD`, which needs the merge base, so `fetch-depth: 0` is a full clone
rather than a ten-second script. `macos` is served by hosts on a Mac Studio
sitting at load 185; `linux` is a separate 20-core box that was measurably
empty. This moves a serialisation point to the least contended label; it does
not remove it, and the gate is still a single point of failure for its
workflow. The structural fix is an early-exit first step inside each job, which
costs one checkout per leg and is a larger change.
Follow-up, deliberately not bundled: jetson-hub runs at capacity 2 with 15 GB of
RAM, so 55 clones still serialise two at a time. Raising capacity helps gates
and hurts the heavy emulated-architecture jobs that share the label, so it wants
measuring on its own.
Co-Authored-By: Claude Opus 5 <[email protected]>
218 lines
15 KiB
YAML
218 lines
15 KiB
YAML
name: wai-quantum HIP member
|
|
|
|
# The fabric's HIP member: the CUDA member's kernel text, unchanged, compiled at
|
|
# run time by HIP's runtime compiler for each device's own architecture and
|
|
# certified bit-exact against the CPU reference's pinned corpus digests.
|
|
#
|
|
# - hip-hw (framework-128gb, its integrated GPU): the device tests must find
|
|
# exactly the declared device, agree with the CPU on the arithmetic probe and
|
|
# on random states (out-of-range runs included), reproduce all four pinned
|
|
# certificates (two gatesets, random-v1 and fabric-v1), and catch every
|
|
# kernel mutant; the matrix-instruction kernel, every product formed by the
|
|
# GPU's bf16 matrix instruction from offset-binary 8-bit limbs, must do all
|
|
# of the same under its own implementation id, with its own mutants caught;
|
|
# and the dense f64 state vector, run resident (the state on the device for
|
|
# the whole run), must come back with the CPU's bits on its pinned corpus,
|
|
# on the wider kernels' own corpus, at 16 and 20 qubits and on random
|
|
# circuits from |0...0> and from a given state, every state-vector mutant
|
|
# caught; and jobs placed on the fabric's members found there (the GPU by
|
|
# HIP and by OpenCL) must give the CPU's answers byte for byte, asked for by
|
|
# key and by `accelerator`, alone and with one state split over them, a
|
|
# state past a member refused, a job of wider gates placed, and auto
|
|
# placing a state on the GPU where the committed measurements say it is
|
|
# faster here, and on the CPU elsewhere.
|
|
# - hip-absent (a container with no GPU and no HIP libraries): the same build
|
|
# finds no member and says why, rather than failing; the device tests run over
|
|
# no member with the declaration `none`.
|
|
|
|
concurrency:
|
|
group: ${{ github.workflow }}-${{ github.ref }}
|
|
cancel-in-progress: true
|
|
|
|
on:
|
|
pull_request:
|
|
paths:
|
|
- 'wai/wai-quantum/**'
|
|
- 'rust-toolchain.toml'
|
|
- '.github/workflows/wai-quantum-hip.yml'
|
|
schedule:
|
|
- cron: '43 6 * * *'
|
|
|
|
jobs:
|
|
# Does this pull request's own diff touch the paths below? This forge matches a
|
|
# workflow's `paths:` against the push, so a merge of main into a branch fires the
|
|
# workflow whatever the pull request changes; every job here waits on this answer.
|
|
# A push to main, a schedule, or anything the script cannot determine answers
|
|
# `true` (scripts/pr-touches-paths.py).
|
|
relevant:
|
|
name: does this change touch it
|
|
runs-on: linux
|
|
outputs:
|
|
run: ${{ steps.gate.outputs.run }}
|
|
steps:
|
|
- uses: actions/checkout@v5
|
|
with:
|
|
fetch-depth: 0
|
|
- id: gate
|
|
run: |
|
|
set -euo pipefail
|
|
r=$(python3 scripts/pr-touches-paths.py .github/workflows/wai-quantum-hip.yml \
|
|
"${{ github.event.pull_request.base.sha }}" \
|
|
"${{ github.event.pull_request.head.sha }}")
|
|
echo "this change touches wai-quantum-hip: $r"
|
|
echo "run=$r" >> "$GITHUB_OUTPUT"
|
|
|
|
hip-hw:
|
|
name: member · HIP (framework-128gb)
|
|
needs: relevant
|
|
if: needs.relevant.outputs.run == 'true'
|
|
runs-on: linux-amd
|
|
timeout-minutes: 90
|
|
env:
|
|
CARGO_PROFILE_DEV_DEBUG: '0'
|
|
CARGO_BUILD_JOBS: '12'
|
|
steps:
|
|
- name: Fetch the source
|
|
shell: bash
|
|
run: |
|
|
set -euo pipefail
|
|
want="${{ github.event.pull_request.head.sha || github.sha }}"
|
|
rm -rf src && mkdir src && cd src
|
|
git init -q
|
|
# Fetch the commit itself, not the ref: a later commit on the ref that this
|
|
# workflow's paths do not match would otherwise be fetched instead. The
|
|
# host names its own route to the forge (WAI_FORGE_URL).
|
|
git fetch -q --depth 1 "${WAI_FORGE_URL:-http://100.64.224.0:3400}/Transaction-Science/open-standards.git" "$want"
|
|
git checkout -q FETCH_HEAD
|
|
got="$(git rev-parse HEAD)"
|
|
[ "$got" = "$want" ] || { echo "::error::fetched $got, not $want"; exit 1; }
|
|
- name: The member's unit tests
|
|
shell: bash
|
|
working-directory: src/wai/wai-quantum
|
|
run: |
|
|
set -euo pipefail
|
|
export PATH="$HOME/.cargo/bin:$PATH"
|
|
CARGO_TARGET_DIR="$HOME/hw-target/wai-quantum-hip" nice -n 10 cargo test --locked --features quantum_fabric_hip,quantum_sv --lib -- quantum_fabric_hip:: quantum_fabric::hip:: quantum_fabric::sv64:: 2>&1 | tee "$RUNNER_TEMP/unit.log"
|
|
grep -qE '^test result: ok\. [1-9][0-9]* passed; 0 failed' "$RUNNER_TEMP/unit.log" || { echo "::error::no unit test ran"; exit 1; }
|
|
- name: The member's device tests
|
|
shell: bash
|
|
working-directory: src/wai/wai-quantum
|
|
run: |
|
|
set -euo pipefail
|
|
export PATH="$HOME/.cargo/bin:$PATH"
|
|
# The devices this host has, declared here (or by the runner, so a GPU
|
|
# and its declaration change in one place).
|
|
key="${WAI_QUANTUM_AMD_HIP_EXPECT:-gpu-hip-gfx1151}"
|
|
report="$RUNNER_TEMP/hip-report.txt"; : > "$report"
|
|
WAI_QUANTUM_HIP_EXPECT="$key" WAI_QUANTUM_HIP_REPORT="$report" CARGO_TARGET_DIR="$HOME/hw-target/wai-quantum-hip" nice -n 10 \
|
|
cargo test --locked --release --features quantum_fabric_hip --test hip_member -- --test-threads=1 2>&1 | tee "$RUNNER_TEMP/device.log"
|
|
grep -qE '^test result: ok\. 9 passed; 0 failed' "$RUNNER_TEMP/device.log" || { echo "::error::not every device test ran and passed"; exit 1; }
|
|
cat "$report"
|
|
grep -qx "hip devices: as declared ($key)" "$report" || { echo "::error::the devices found are not the declared ones"; exit 1; }
|
|
n="$(grep -c "^hip certified: key=$key .* parity=bit-exact " "$report")"
|
|
[ "$n" = 4 ] || { echo "::error::$n of 4 certificates were earned"; exit 1; }
|
|
grep -qx "hip mutants: key=$key killed 6 of 6" "$report" || { echo "::error::a kernel mutant survived"; exit 1; }
|
|
- name: The matrix-instruction kernel, every product by the instruction
|
|
shell: bash
|
|
working-directory: src/wai/wai-quantum
|
|
run: |
|
|
set -euo pipefail
|
|
export PATH="$HOME/.cargo/bin:$PATH"
|
|
key="${WAI_QUANTUM_AMD_HIP_EXPECT:-gpu-hip-gfx1151}"
|
|
report="$RUNNER_TEMP/hip-wmma-report.txt"; : > "$report"
|
|
WAI_QUANTUM_HIP_EXPECT="$key" WAI_QUANTUM_HIP_REPORT="$report" CARGO_TARGET_DIR="$HOME/hw-target/wai-quantum-hip" nice -n 10 \
|
|
cargo test --locked --release --features quantum_fabric_hip --test hip_wmma -- --test-threads=1 2>&1 | tee "$RUNNER_TEMP/wmma.log"
|
|
grep -qE '^test result: ok\. 9 passed; 0 failed' "$RUNNER_TEMP/wmma.log" || { echo "::error::not every matrix-instruction test ran and passed"; exit 1; }
|
|
cat "$report"
|
|
grep -qx "hip wmma devices: as declared ($key)" "$report" || { echo "::error::the devices found are not the declared ones"; exit 1; }
|
|
grep -q "^hip wmma member: key=$key .* implementation=hip-wmma-limbs8/[0-9a-f]\{16\} " "$report" || { echo "::error::the device ran no matrix-instruction member"; exit 1; }
|
|
n="$(grep -c "^hip wmma certified: key=$key .* implementation=hip-wmma-limbs8/[0-9a-f]\{16\} parity=bit-exact " "$report")"
|
|
[ "$n" = 4 ] || { echo "::error::$n of 4 matrix-instruction certificates were earned"; exit 1; }
|
|
grep -qx "hip wmma mutants: key=$key killed 8 of 8" "$report" || { echo "::error::a matrix-instruction mutant survived"; exit 1; }
|
|
- name: The dense f64 state vector, bit for bit
|
|
shell: bash
|
|
working-directory: src/wai/wai-quantum
|
|
run: |
|
|
set -euo pipefail
|
|
export PATH="$HOME/.cargo/bin:$PATH"
|
|
key="${WAI_QUANTUM_AMD_HIP_EXPECT:-gpu-hip-gfx1151}"
|
|
report="$RUNNER_TEMP/hip-sv-report.txt"; : > "$report"
|
|
WAI_QUANTUM_HIP_EXPECT="$key" WAI_QUANTUM_HIP_REPORT="$report" CARGO_TARGET_DIR="$HOME/hw-target/wai-quantum-hip" nice -n 10 \
|
|
cargo test --locked --release --features quantum_fabric_hip,quantum_sv --test hip_sv -- --test-threads=1 2>&1 | tee "$RUNNER_TEMP/sv.log"
|
|
grep -qE '^test result: ok\. 4 passed; 0 failed' "$RUNNER_TEMP/sv.log" || { echo "::error::not every state-vector test ran and passed"; exit 1; }
|
|
cat "$report"
|
|
grep -q "^hip sv certified: key=$key .* parity=bit-exact cases=48 digest=b7d129d0593adf70dc164477664970c5c0b8c09f79f1a0573036bd60888dbcd1$" "$report" || { echo "::error::the state vector is not the CPU's bit for bit"; exit 1; }
|
|
grep -q "^hip sv wide certified: key=$key .* kernel=wai.quantum.sv/run-wide .* parity=bit-exact cases=120 digest=093614f2386ec5120ab6ccde5187a544518432751591535a6f58c7a44061ebaf$" "$report" || { echo "::error::the wider kernels are not the CPU's bit for bit"; exit 1; }
|
|
grep -qx "hip sv mutants: key=$key killed 4 of 4" "$report" || { echo "::error::a state-vector mutant survived"; exit 1; }
|
|
grep -qx "hip sv random: key=$key 140 resident runs on every kernel match the CPU bit for bit" "$report" || { echo "::error::a random circuit's resident run is not the CPU's"; exit 1; }
|
|
- name: Jobs placed on the fabric's members, and a state split over them
|
|
shell: bash
|
|
working-directory: src/wai/wai-quantum
|
|
run: |
|
|
set -euo pipefail
|
|
export PATH="$HOME/.cargo/bin:$PATH"
|
|
# The members placement finds here, sorted (the GPU by HIP and by OpenCL),
|
|
# declared here or by the runner; one state is split over them.
|
|
keys="${WAI_QUANTUM_AMD_FABRIC_KEYS:-gpu-hip-gfx1151,opencl-gpu-1002}"
|
|
report="$RUNNER_TEMP/fabric-report.txt"; : > "$report"
|
|
WAI_QUANTUM_FABRIC_EXPECT="$keys" WAI_QUANTUM_FABRIC_SPLIT="$keys" WAI_QUANTUM_FABRIC_PAULI_EXPECT=none WAI_QUANTUM_FABRIC_REPORT="$report" CARGO_TARGET_DIR="$HOME/hw-target/wai-quantum-hip" nice -n 10 \
|
|
cargo test --locked --release --features quantum_engine,quantum_fabric,quantum_fabric_hip,quantum_fabric_opencl --test fabric_placement -- --test-threads=1 2>&1 | tee "$RUNNER_TEMP/fabric.log"
|
|
grep -qE '^test result: ok\. 7 passed; 0 failed' "$RUNNER_TEMP/fabric.log" || { echo "::error::not every placement test ran and passed"; exit 1; }
|
|
cat "$report"
|
|
grep -qx "fabric members: as declared ($keys)" "$report" || { echo "::error::the members placement found are not the declared ones"; exit 1; }
|
|
for key in ${keys//,/ }; do
|
|
grep -q "^fabric placed: key=$key .* give the CPU's answers byte for byte$" "$report" || { echo "::error::$key: a placed job's answers are not the CPU's"; exit 1; }
|
|
grep -q "^fabric refused: key=$key " "$report" || { echo "::error::$key: a state past what it holds was not refused"; exit 1; }
|
|
grep -qx "fabric wide: key=$key placed at fusion 5, the CPU's answers byte for byte" "$report" || { echo "::error::$key: a job of wider gates did not run with the CPU's answers"; exit 1; }
|
|
done
|
|
grep -qx "fabric split: ${keys//,/+} over 14 and 19 qubits gives the CPU's answers byte for byte" "$report" || { echo "::error::the state split over the members is not the CPU's"; exit 1; }
|
|
# The measurements show the integrated GPU faster here from 18 qubits at
|
|
# fusion 2 and 3 (engine::fabric_measured): auto places the measured job
|
|
# on it, and keeps a shallower one on the CPU.
|
|
# (A runner whose processor is not the one measured, by its model or its
|
|
# thread count, matches no row, and auto keeps every state on its CPU.)
|
|
if [ "$(grep -c '^fabric auto host: .* measured=true$' "$report")" = 1 ]; then
|
|
grep -qE "^fabric auto: by measurement, 18 qubits at fusion 2 on (gpu-hip-gfx1151|opencl-gpu-1002), the CPU's answers byte for byte; a shallower job on the CPU$" "$report" || { echo "::error::auto did not follow the measurements"; exit 1; }
|
|
else
|
|
grep -qx "fabric auto: no member here measured faster; every state on the CPU, its result unnamed" "$report" || { echo "::error::auto placed a state on a host not measured"; exit 1; }
|
|
fi
|
|
grep -q "^fabric accelerator: .* the CPU's answers byte for byte" "$report" || { echo "::error::an accelerator request placed no state"; exit 1; }
|
|
# The pauli method's members are CUDA's alone: none here, and it says so.
|
|
grep -qx "fabric pauli: no member here certified for it; accelerator refused" "$report" || { echo "::error::a pauli job was placed with no member for it"; exit 1; }
|
|
|
|
hip-absent:
|
|
name: member · HIP absent (linux-arm64)
|
|
needs: relevant
|
|
if: needs.relevant.outputs.run == 'true'
|
|
runs-on: linux-arm64
|
|
timeout-minutes: 60
|
|
env:
|
|
WAI_QUANTUM_HIP_EXPECT: none
|
|
steps:
|
|
- uses: actions/checkout@v5
|
|
- name: Install Rust 1.98.0
|
|
shell: bash
|
|
run: |
|
|
export PATH="$HOME/.cargo/bin:$PATH"
|
|
command -v rustup >/dev/null 2>&1 || \
|
|
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y --profile minimal --default-toolchain 1.98.0
|
|
rustup toolchain install 1.98.0 --profile minimal
|
|
rustup default 1.98.0
|
|
echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
|
|
- name: No HIP here, and the member says so
|
|
shell: bash
|
|
working-directory: wai/wai-quantum
|
|
run: |
|
|
set -euo pipefail
|
|
report="$RUNNER_TEMP/hip-report.txt"; : > "$report"
|
|
WAI_QUANTUM_HIP_REPORT="$report" cargo test --locked --features quantum_fabric_hip --test hip_member -- --test-threads=1 2>&1 | tee "$RUNNER_TEMP/device.log"
|
|
grep -qE '^test result: ok\. 9 passed; 0 failed' "$RUNNER_TEMP/device.log" || { echo "::error::not every device test ran and passed"; exit 1; }
|
|
grep -qx 'hip devices: as declared (none)' "$report" || { echo "::error::the absent case was not established"; exit 1; }
|
|
WAI_QUANTUM_HIP_REPORT="$report" cargo test --locked --features quantum_fabric_hip --test hip_wmma -- --test-threads=1 2>&1 | tee "$RUNNER_TEMP/wmma.log"
|
|
grep -qE '^test result: ok\. 9 passed; 0 failed' "$RUNNER_TEMP/wmma.log" || { echo "::error::the matrix-instruction tests did not run over no device"; exit 1; }
|
|
grep -qx 'hip wmma devices: as declared (none)' "$report" || { echo "::error::the matrix-instruction absent case was not established"; exit 1; }
|
|
cargo test --locked --features quantum_fabric_hip,quantum_sv --test hip_sv -- --test-threads=1 2>&1 | tee "$RUNNER_TEMP/sv.log"
|
|
grep -qE '^test result: ok\. 4 passed; 0 failed' "$RUNNER_TEMP/sv.log" || { echo "::error::the state-vector tests did not run over no device"; exit 1; }
|
|
WAI_QUANTUM_FABRIC_EXPECT=none WAI_QUANTUM_FABRIC_REPORT="$report" cargo test --locked --features quantum_engine,quantum_fabric,quantum_fabric_hip --test fabric_placement -- --test-threads=1 2>&1 | tee "$RUNNER_TEMP/fabric.log"
|
|
grep -qE '^test result: ok\. 7 passed; 0 failed' "$RUNNER_TEMP/fabric.log" || { echo "::error::the placement tests did not run over no member"; exit 1; }
|
|
grep -qx 'fabric members: as declared (none)' "$report" || { echo "::error::the placement's absent case was not established"; exit 1; }
|