Skip to content

Compare256 function pointer#273

Draft
folkertdev wants to merge 2 commits into
mainfrom
compare256-function-pointer
Draft

Compare256 function pointer#273
folkertdev wants to merge 2 commits into
mainfrom
compare256-function-pointer

Conversation

@folkertdev

@folkertdev folkertdev commented Dec 24, 2024

Copy link
Copy Markdown
Member

when e.g. the avx2 target feature is not enabled at compile time, but the feature is available at runtime, this approach reduces branching. We still dispatch statically if the target feature is already enabled at compile time

@folkertdev
folkertdev marked this pull request as draft December 24, 2024 17:52
@folkertdev

folkertdev commented Dec 24, 2024

Copy link
Copy Markdown
Member Author

for level 1

Benchmark 1 (62 runs): target/release/examples/compress-baseline 1 rs silesia-small.tar
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          81.7ms ±  991us    80.3ms … 87.8ms          3 ( 5%)        0%
  peak_rss           26.7MB ± 53.8KB    26.6MB … 26.7MB         13 (21%)        0%
  cpu_cycles          305M  ± 3.65M      298M  …  328M           5 ( 8%)        0%
  instructions        661M  ±  278       661M  …  661M           0 ( 0%)        0%
  cache_references   19.8M  ±  185K     19.5M  … 20.8M           2 ( 3%)        0%
  cache_misses        432K  ± 86.9K      336K  …  783K           2 ( 3%)        0%
  branch_misses      2.94M  ± 8.26K     2.88M  … 2.96M           2 ( 3%)        0%
Benchmark 2 (63 runs): target/release/examples/blogpost-compress 1 rs silesia-small.tar
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          80.3ms ±  584us    78.7ms … 83.1ms          2 ( 3%)        ⚡-  1.6% ±  0.3%
  peak_rss           26.7MB ± 63.2KB    26.5MB … 26.7MB          0 ( 0%)          -  0.0% ±  0.1%
  cpu_cycles          299M  ± 2.38M      293M  …  312M           3 ( 5%)        ⚡-  2.0% ±  0.4%
  instructions        620M  ±  255       620M  …  620M           1 ( 2%)        ⚡-  6.2% ±  0.0%
  cache_references   19.7M  ±  146K     19.5M  … 20.2M           3 ( 5%)          -  0.8% ±  0.3%
  cache_misses        409K  ± 74.2K      311K  …  664K           1 ( 2%)          -  5.3% ±  6.6%
  branch_misses      2.95M  ± 15.7K     2.87M  … 2.96M           2 ( 3%)          +  0.6% ±  0.2%

@folkertdev folkertdev mentioned this pull request Dec 24, 2024
Merged
@brian-pane

brian-pane commented Dec 28, 2024 •
edited
Loading

Copy link
Copy Markdown
Contributor

This shows an improvement on x86_64 Intel as well (compiled without RUSTFLAGS="-Ctarget-cpu=native", in order to generate the new function-pointer code path).

Benchmark 1 (64 runs): ./blogpost-compress-baseline 1 rs silesia-small.tar
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          78.9ms ±  852us    77.6ms … 83.1ms          1 ( 2%)        0%
  peak_rss           26.6MB ± 60.6KB    26.5MB … 26.7MB          0 ( 0%)        0%
  cpu_cycles          307M  ±  536K      305M  …  308M           0 ( 0%)        0%
  instructions        661M  ±  290       661M  …  661M           0 ( 0%)        0%
  cache_references    265K  ± 3.68K      262K  …  285K           4 ( 6%)        0%
  cache_misses        231K  ± 6.74K      205K  …  237K           8 (13%)        0%
  branch_misses      2.93M  ± 3.84K     2.92M  … 2.94M           1 ( 2%)        0%
Benchmark 2 (67 runs): ./target/release/examples/blogpost-compress 1 rs silesia-small.tar
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          75.4ms ±  526us    74.6ms … 76.7ms          0 ( 0%)        ⚡-  4.3% ±  0.3%
  peak_rss           26.6MB ± 60.4KB    26.6MB … 26.7MB          0 ( 0%)          +  0.0% ±  0.1%
  cpu_cycles          292M  ±  545K      291M  …  294M           0 ( 0%)        ⚡-  4.6% ±  0.1%
  instructions        620M  ±  261       620M  …  620M           0 ( 0%)        ⚡-  6.2% ±  0.0%
  cache_references    266K  ± 5.65K      262K  …  298K           4 ( 6%)          +  0.2% ±  0.6%
  cache_misses        231K  ± 7.01K      203K  …  237K           5 ( 7%)          -  0.2% ±  1.0%
  branch_misses      3.03M  ± 4.91K     3.02M  … 3.04M           0 ( 0%)        💩+  3.6% ±  0.1%

@folkertdev
folkertdev force-pushed the compare256-function-pointer branch from 4e432fd to c2ce875 Compare January 20, 2025 09:38
@codecov

codecov Bot commented Jan 20, 2025 •
edited
Loading

Copy link
Copy Markdown

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines Coverage Δ
zlib-rs/src/deflate/algorithm/quick.rs 97.72% <100.00%> (-0.13%) ⬇️
zlib-rs/src/deflate/compare256.rs 98.31% <100.00%> (+0.30%) ⬆️

... and 2 files with indirect coverage changes

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
  • 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

@folkertdev
store function pointer for the correct compare256 in a static
076242a 
when e.g. the avx2 target feature is not enabled at compile time, but the feature is available at runtime, this approach reduces branching. We still dispatch statically if the target feature is already enabled at compile time
@folkertdev
simplify quick match matching
4c71540
@folkertdev
folkertdev force-pushed the compare256-function-pointer branch from c2ce875 to 4c71540 Compare January 20, 2025 13:32
@folkertdev
folkertdev force-pushed the main branch 2 times, most recently from 72c2c57 to af07d45 Compare February 6, 2025 11:06
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@folkertdev @brian-pane