Add measure_batch and adopt it for STIM measurement by david-pl · Pull Request #135 · QuEraComputing/ppvm

david-pl · 2026-06-22T09:28:10Z

Adds measure_batch to ppvm-tableau — measure a chosen set of qubit indices,
the analogue of the batched Clifford gates from #116. measure_all becomes the
0..n_qubits case (behavior unchanged, no regression), and the noiseless M/MZ
path in ppvm-stim now uses it, reusing the measurement scratch across targets
the same way #116 did for gates. Noisy readout and MR keep the per-qubit loop,
so RNG draws are unchanged.

Also adds a samply profiling harness (scripts/profile_measure_batch.sh) and a short findings doc. The cost splits into two regimes: a constant
~16µs O(n²) tableau floor (compute_decomposition + update_tableau), and the
case-a Vec↔HashMap round-trip (~35% at ~1000 coefficients). No optimization
applied yet — candidates are written up for follow-up.

Note: Rust-only — measure_all isn't exposed to Python, so no binding changes.

FYI, @rafaelha @Roger-luo we may want to consider the Python bindings for this in #134.

Add `LossyMeasureAll::measure_batch(&[usize])`, the indices-accepting analogue of the batched Clifford gates. `measure_all` becomes the `0..n_qubits` case; both share a private `measure_one_with_scratch` helper so the all-qubits path keeps its allocation-free loop. Also adds `measure_batch_with_scratch` for samplers, batch-vs-individual tests, and measure_batch cases in the measure-all / measure-scaling benches. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

The executor's M/MZ arm now calls `measure_batch` for noiseless readout, reusing the measurement scratch across targets like the batched gate paths. Noisy readout keeps the per-qubit `measure_noisy` loop so the interleaved measure/flip RNG draws are unchanged, and MR is untouched. Adds a grouped-vs-separate measurement characterization test. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Adds a samply-based profiling setup for measure_batch across three coefficient-count regimes (few ~1, mid ~100, large ~1000): - examples/profile_measure_batch.rs: quick-timing + sustained-loop driver - scripts/profile_measure_batch.sh: records via samply (no sudo on macOS) - scripts/samply_top.py: headless top-functions summary from the saved profile - benches/measure_batch_profiling.md: findings Profiling shows two regimes: a constant ~16us O(n^2) tableau-bound floor (compute_decomposition + update_tableau) dominating low-coefficient states, and the case-a Vec<->HashMap round-trip (~35% at 1024 coefficients) dominating coefficient-heavy ones. No measure_batch code change yet. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

github-actions · 2026-06-22T09:32:47Z

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-06-23 04:44 UTC

Copilot

Pull request overview

This PR adds a batched Z-basis measurement API to ppvm-tableau (measure_batch) and adopts it in the ppvm-stim executor for noiseless multi-target M/MZ, enabling reuse of measurement scratch across targets (analogous to the batched Clifford-gate work in #116). It also introduces profiling harness scripts and a short write-up capturing current performance findings.

Changes:

Add measure_batch / measure_batch_with_scratch in ppvm-tableau and refactor measure_all_with_scratch to share the same per-qubit helper.
Switch noiseless M/MZ in ppvm-stim to use measure_batch while preserving the per-qubit loop for noisy readout to keep RNG draw interleaving unchanged.
Add tests, benches, and profiling tooling/docs for measuring and analyzing measure_batch performance.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
scripts/samply_top.py	New utility to summarize samply profiles by self/inclusive time.
scripts/profile_measure_batch.sh	New profiling driver script for `measure_batch` across workload regimes.
crates/ppvm-tableau/tests/measure_batch.rs	Integration tests asserting `measure_batch` matches `measure`/`measure_all` behavior (including loss).
crates/ppvm-tableau/src/measure_all.rs	Introduces `measure_batch` in the measurement API and factors shared per-qubit logic.
crates/ppvm-tableau/examples/profile_measure_batch.rs	Profiling example executable for `measure_batch` cost regimes.
crates/ppvm-tableau/benches/measure-scaling.rs	Adds a benchmark variant measuring `measure_batch` over all qubits.
crates/ppvm-tableau/benches/measure-all.rs	Adds a `measure-batch-all` benchmark to check for regressions vs `measure_all`.
crates/ppvm-tableau/benches/measure_batch_profiling.md	Documents profiling results and follow-up optimization candidates.
crates/ppvm-stim/tests/executor.rs	Adds a seeded equivalence test ensuring grouped `M` matches separate `M` instructions.
crates/ppvm-stim/src/executor.rs	Uses `measure_batch` for noiseless `M`/`MZ` multi-target execution.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Roger-luo · 2026-06-22T14:41:03Z

@@ -0,0 +1,129 @@
+#!/usr/bin/env python3


What does this file do? If it renders something, maybe we can post the render result in this PR just for future reference.

Roger-luo · 2026-06-23T04:44:05Z

Superseded by #142, which carries this work forward onto the measure_many API that #134 established (folding the scratch-reusing implementation in as the tableau's LossyMeasure::measure_many override rather than a parallel measure_batch). Your commits are preserved there and you're credited as co-author. Closing in favor of #142.

…upersedes #135) (#142) Supersedes #135. Carries forward David's batched-measurement work (his original commits are preserved, and he's credited as co-author), reworked to land on the `measure_many` API that #134 already established rather than a parallel `measure_batch` name. ### Background: how this relates to #134 and #135 - **#134 (merged)** introduced `LossyMeasure::measure_many` as the canonical batched-measurement name — but only as a **default trait method** (a per-target `measure` loop that allocates a fresh `MeasureScratch` each call, so *no* scratch reuse) — and wired it through to Python. - **#135** added the *optimized* implementation (scratch reused across targets) plus the Stim executor adoption and a profiling harness, but under a second name (`measure_batch` on `LossyMeasureAll`) written before #134 settled the `*_many` convention. This PR folds #135's fast implementation in as the tableau's **override of `LossyMeasure::measure_many`**, so there's a single canonical name and the Python binding + Stim executor both pick up the speedup for free. ### Changes - **`ppvm-tableau/src/measure.rs`** — override `LossyMeasure::measure_many` for `GeneralizedTableau` with the scratch-reusing path (one `MeasureScratch` reused across the batch, amortizing the case-a HashMap / `b_entries` allocations and the cached odd-phase-destabilizer mask). - **`ppvm-tableau/src/measure_all.rs`** — drop `LossyMeasureAll::measure_batch`; rename `measure_batch_with_scratch` → `measure_many_with_scratch`. The shared `measure_one_with_scratch` helper now records `None` for lost qubits, so every batched path matches per-qubit `measure` on the **measurement record**, not just on returned outcomes (also closes a latent gap in `measure_all`). - **`ppvm-stim/src/executor.rs`** — noiseless `M`/`MZ` delegates to `measure_many`; noisy readout and `MR` keep the per-qubit loop so RNG draws are unchanged. - **No Python change needed** — `interface_tableau.rs` already calls `.measure_many(...)`, so the binding now routes through the fast override. - Renamed the tests / benches / profiling harness `*_batch*` → `*_many*`, and added a loss-aware record-equivalence test (`measure_many` vs sequential `measure`, including the measurement record). Profiling findings (`benches/measure_many_profiling.md`) are unchanged from #135: a constant ~16µs O(n²) tableau floor, plus a case-a `Vec`↔`HashMap` round-trip (~35% at ~1000 coefficients). No further optimization applied yet — candidates are written up for follow-up. ### Validation `cargo build --workspace`, `cargo clippy --workspace -- -D warnings`, `cargo test --workspace`, and `uv run --project ppvm-python --group dev pytest` (183 passed) — all green, plus the full prek hook suite. Co-authored-by: David Plankensteiner <dplankensteiner@quera.com> 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: David Plankensteiner <dplankensteiner@quera.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

david-pl and others added 3 commits June 22, 2026 10:29

Copilot AI review requested due to automatic review settings June 22, 2026 09:28

Copilot started reviewing on behalf of david-pl June 22, 2026 09:28

View session

Copilot AI reviewed Jun 22, 2026

View reviewed changes

Comment thread crates/ppvm-tableau/src/measure_all.rs

Comment thread scripts/samply_top.py

Comment thread scripts/profile_measure_batch.sh

Roger-luo reviewed Jun 22, 2026

View reviewed changes

Merge branch 'main' into david/measure-batch

8435b3b

Roger-luo mentioned this pull request Jun 23, 2026

perf(tableau): scratch-reusing measure_many, adopted for Stim M/MZ (supersedes #135) #142

Merged

Roger-luo closed this Jun 23, 2026

david-pl deleted the david/measure-batch branch July 20, 2026 15:19

Navigation Menu

Repository navigation

Add measure_batch and adopt it for STIM measurement - #135

Add measure_batch and adopt it for STIM measurement#135
david-pl wants to merge 4 commits into
mainfrom
david/measure-batch

david-pl commented Jun 22, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Jun 22, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Roger-luo Jun 22, 2026

Uh oh!

Roger-luo commented Jun 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

david-pl commented Jun 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Jun 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Roger-luo Jun 22, 2026

Choose a reason for hiding this comment

Uh oh!

Roger-luo commented Jun 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

david-pl commented Jun 22, 2026 •

edited

Loading

github-actions Bot commented Jun 22, 2026 •

edited

Loading