fix(divan): support threaded walltime benchmarks

[not-matthias]

Changes in this PR:

• Prevent threaded Divan walltime benchmarks from concurrently writing to the process-wide instrument-hooks FIFOs (this alone doesn't fix the issue)
• Divan workers record monotonic bounds locally, then the coordinator emits one marker pair for the complete parallel sampling round. Benchmark bodies remain parallel.

As discussed, depends on CodSpeedHQ/instrument-hooks#30

[codspeed-hq]

Merging this PR will regress 15 benchmarks

⚠️ 59 benchmarks measured no execution time

Nothing ran under measurement, usually because the compiler removed the code under test. These results are not comparable, so they count as unchanged.

Preventing compiler optimizations

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 24 improved benchmarks
❌ 15 regressed benchmarks
✅ 552 untouched benchmarks
🆕 3 new benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	bench_array1[1]	1 ns	35 ns	-97.14%
❌	WallTime	iter_batched_large_input	5 ns	7 ns	-28.57%
❌	WallTime	find_highest_set_bit[255]	5 ns	6 ns	-16.67%
❌	WallTime	iter_batched_ref_large_input	5 ns	6 ns	-16.67%
❌	WallTime	graph_coloring[5]	1 µs	1.2 µs	-12.64%
❌	WallTime	recursive_memoized[HashMap<u64, u64>, 0]	16 ns	18 ns	-11.11%
❌	WallTime	iter_batched_per_iteration	43 ns	47 ns	-8.51%
❌	WallTime	n_queens_solver[4]	2.1 µs	2.3 µs	-8.47%
❌	WallTime	iter_with_setup	45 ns	48 ns	-6.25%
❌	Simulation	process_items_with_counter	1.8 µs	1.9 µs	-5.84%
❌	WallTime	init_array[42]	36 ns	38 ns	-5.26%
❌	WallTime	graph_coloring[6]	1.1 µs	1.2 µs	-5.11%
❌	Simulation	b	4.1 µs	4.3 µs	-3.8%
❌	Simulation	b	4.8 µs	4.9 µs	-3.34%
❌	Simulation	fibonacci_custom_measurement	1.6 µs	1.7 µs	-3.28%
⚡	WallTime	add_two_integers[(255, 255)]	20 ns	7 ns	×2.9
⚡	WallTime	add_two_integers[(42, 13)]	20 ns	7 ns	×2.9
⚡	Simulation	Iterative	177.1 ns	121.5 ns	+45.72%
⚡	Simulation	Iterative[20]	177.1 ns	121.5 ns	+45.72%
⚡	Simulation	Iterative[21]	178.4 ns	122.8 ns	+45.24%
...	...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing cod-3324-codspeed-runner-hangs-when-using-divans-threads-feature (5a9838f) with main (69f74b6)}

[codspeed-hq]

Unable to generate the flame graphs

The performance report has correctly been generated, but there was an internal error while generating the flame graphs for this run. We're working on fixing the issue. Feel free to contact us on Discord or at support@codspeed.io if the issue persists.

[avalanche-staging]

Congrats! CodSpeed is installed 🎉

🆕 622 new benchmarks were detected.

You will start to see performance impacts in the reports once the benchmarks are run from your default branch.

Detected benchmarks

• sum_fold (WallTime): 307 ns
• build_vec (WallTime): 216 ns
• knight_tour[5] (WallTime): 302.4 µs
• rem (WallTime): 6 ns
• generate_combinations[9] (WallTime): 6.1 µs
• permutations[3] (WallTime): 1.2 µs
• subset_sum[18] (WallTime): 4.7 µs
• n_queens_solver[5] (WallTime): 9.1 µs
• add_two_integers[(255, 255)] (WallTime): 19 ns
• subset_sum[10] (WallTime): 1.1 µs
• recursive[0] (WallTime): 1 ns
• hamiltonian_cycle[4] (WallTime): 857 ns
• recursive_memoized[HashMap<u64, u64>, 0] (WallTime): 18 ns
• add_two_integers[(65535, -65535)] (WallTime): 75 ns
• n_queens_solver[6] (WallTime): 21.6 µs
• find_highest_set_bit[42] (WallTime): 5 ns
• print_env_hello (WallTime): 4.9 µs
• recursive[5] (WallTime): 31 ns
• add_two_integers[(0, 0)] (WallTime): 5 ns
• iterative[0] (WallTime): 1 ns
• ...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

---

[greptile-apps]

Greptile Summary

The PR enables threaded Divan walltime benchmarks without concurrent writes to the process-wide instrument-hooks transport.

• Serializes all Linux instrument-hooks FFI transport operations across handles.
• Collects per-worker monotonic bounds and emits one aggregate marker envelope per parallel sampling round.
• Adds a four-worker native Divan benchmark to the walltime integration fixture.

Confidence Score: 5/5

The PR appears safe to merge, with no actionable correctness or security issues identified.

The transport is serialized without enclosing benchmark execution, worker samples are fully collected before aggregation, and the runner accepts timestamped markers independently of their delivery time.

Important Files Changed

Filename	Overview
crates/codspeed/src/instrument_hooks/mod.rs	Adds process-wide serialization around Linux instrument-hooks FFI operations and keeps each timestamp pair adjacent.
crates/divan_compat/divan_fork/src/bench/mod.rs	Moves marker transport out of worker threads and emits aggregate bounds after each parallel sample round.
crates/divan_compat/divan_fork/src/stats/sample.rs	Extends internal raw samples with the monotonic benchmark bounds required for coordinator aggregation.
crates/divan_compat/benches/thread_example.rs	Adds a four-worker benchmark on the native Divan path while retaining single-threaded simulation compatibility.

Sequence Diagram

sequenceDiagram
  participant C as Divan coordinator
  participant W as Worker threads
  participant H as InstrumentHooks
  participant R as Runner FIFO
  C->>W: Start parallel sampling round
  par Worker samples
    W->>W: Capture monotonic start
    W->>W: Run benchmark iterations
    W->>W: Capture monotonic end
  end
  W-->>C: Return samples and timestamp bounds
  C->>C: Select minimum start and maximum end
  C->>H: Add aggregate benchmark timestamps
  H->>H: Acquire process-wide transport lock
  H->>R: BenchmarkStart(minimum start)
  H->>R: BenchmarkEnd(maximum end)
  H->>H: Release transport lock

Loading

_{Reviews (1): Last reviewed commit: "fix(divan): support threaded walltime be..." | Re-trigger Greptile}

[GuillaumeLagrange]

olgtm, let me know your thoughts about the thread safety being insturment-hooks' responsibility, if you have compelling arguments I'll approve.

Although closing thoughts: how does the simulation instrument behave with the threads option?