RFC: bench: Add multi thread benchmarking #33740

1. 
   fjahr commented at 11:36 pm on October 29, 2025: contributor

   This adds a rough way to run specific benchmarks with different numbers of worker threads to see how these impact performance. This is useful for me in the batch validation context and for potential refactorings of checkqueue but I am not sure if this is useful in many other contexts so I am leaving this as a RFC draft for now to see if there is any interest in merging something like this.

   Example output:

    0$ build/bin/bench_bitcoin -filter=ConnectBlockAllSchnorr -min-time=1000 -scale-threads
    1Running benchmarks with worker threads from 1 to 15
    2
    3|            ns/block |             block/s |    err% |     total | benchmark
    4|--------------------:|--------------------:|--------:|----------:|:----------
    5|       43,118,187.50 |               23.19 |    0.1% |      0.99 | `ConnectBlockAllSchnorr_1_threads`
    6|       29,015,444.33 |               34.46 |    0.2% |      1.10 | `ConnectBlockAllSchnorr_2_threads`
    7|       22,241,437.50 |               44.96 |    0.1% |      1.09 | `ConnectBlockAllSchnorr_3_threads`
    8|       17,893,361.00 |               55.89 |    0.3% |      1.09 | `ConnectBlockAllSchnorr_4_threads`
    9|       15,105,819.50 |               66.20 |    0.2% |      1.09 | `ConnectBlockAllSchnorr_5_threads`
   10|       13,145,510.38 |               76.07 |    0.2% |      1.08 | `ConnectBlockAllSchnorr_6_threads`
   11|       11,646,614.62 |               85.86 |    0.2% |      1.06 | `ConnectBlockAllSchnorr_7_threads`
   12|       10,476,425.89 |               95.45 |    0.2% |      1.07 | `ConnectBlockAllSchnorr_8_threads`
   13|        9,547,125.00 |              104.74 |    0.2% |      1.08 | `ConnectBlockAllSchnorr_9_threads`
   14|        9,574,246.27 |              104.45 |    0.6% |      1.08 | `ConnectBlockAllSchnorr_10_threads`
   15|        9,557,633.40 |              104.63 |    1.1% |      1.08 | `ConnectBlockAllSchnorr_11_threads`
   16|        9,495,175.00 |              105.32 |    0.8% |      1.08 | `ConnectBlockAllSchnorr_12_threads`
   17|        9,297,012.50 |              107.56 |    0.3% |      1.07 | `ConnectBlockAllSchnorr_13_threads`
   18|        9,775,795.45 |              102.29 |    0.7% |      1.11 | `ConnectBlockAllSchnorr_14_threads`
   19|        9,776,140.18 |              102.29 |    0.9% |      1.06 | `ConnectBlockAllSchnorr_15_threads`
   

2. 
   DrahtBot commented at 11:36 pm on October 29, 2025: contributor

   The following sections might be updated with supplementary metadata relevant to reviewers and maintainers.

   Code Coverage & Benchmarks

   For details see: https://corecheck.dev/bitcoin/bitcoin/pulls/33740.

   Reviews

   See the guideline for information on the review process. A summary of reviews will appear here.

   Conflicts

   Reviewers, this pull request conflicts with the following ones:

   • #33300 (fuzz: compact block harness by Crypt-iQ)

   If you consider this pull request important, please also help to review the conflicting pull requests. Ideally, start with the one that should be merged first.

3. 
   fjahr commented at 11:44 pm on October 29, 2025: contributor

   I people could test the exact command as above it might be interesting to post the results here. For me at least, 13 worker threads are consistently delivering a better result than 15 worker threads. EDIT: After some time I started to see the best results with 14 and 15, so maybe this was nothing.
4. bench: Add step-wise multi thread benchmarking de25d1a855
5. fjahr force-pushed on Oct 29, 2025

Contributors
fjahr DrahtBot