add AVX512 support for filtering in place

[connortsui20]

Reorganizes some of the filter implementations and adds AVX512 filter implementation and benchmarks

[codecov[bot]]

Codecov Report

:x: Patch coverage is 75.83643% with 130 lines in your changes missing coverage. Please review. :white_check_mark: Project coverage is 85.45%. Comparing base (c75c8a3) to head (eae0511). :warning: Report is 78 commits behind head on develop.

Files with missing lines	Patch %	Lines
vortex-compute/src/filter/slice/simd_compress.rs	9.57%	85 Missing :warning:
vortex-compute/src/filter/slice/out/by_mask.rs	55.55%	28 Missing :warning:
vortex-compute/src/filter/slice/out/avx512.rs	91.07%	5 Missing :warning:
vortex-compute/src/filter/slice/out/by_bitview.rs	87.50%	5 Missing :warning:
vortex-compute/src/filter/slice/in_place/avx512.rs	92.30%	4 Missing :warning:
vortex-compute/src/filter/slice/out/mod.rs	98.38%	2 Missing :warning:
vortex-compute/src/filter/slice/in_place/mod.rs	99.08%	1 Missing :warning:

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

:rocket: New features to boost your workflow:

• :snowflake: Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• :package: JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[a10y]

Does GitHub give us CI runners with AVX512 support? My recollection from when I did the AVX2 take kernel is that they don't

[connortsui20]

not really sure how to deal with this, I think sometimes they have it, sometimes not?

[robert3005]

CI runs in AWS, you can get whatever machine you want, right now by default we use m7i/m7a which are sapphire rapids/zen4 which have AVX512. On benchmarks we run c6i which is ice-lake sp which also has avx512

[0ax1]

Heads up that we can't micro-benchmark avx512 with codspeed, as valgrind/callgrind does not support avx512. Saying you'll get numbers for avx2 if it is implemented:

      - name: Build benchmarks (shard 1)
        env:
          RUSTFLAGS: "-C target-feature=+avx2"
      ...

[connortsui20]

Given that were pivoting to batch I'm going to put a pin in this.

[gatesn]

Nooo I like this!

[codspeed-hq[bot]]

CodSpeed Performance Report

Merging #5399 will not alter performance

_{Comparing ct/avx512-filter (eae0511) with develop (c75c8a3)}

Summary

✅ 1478 untouched
🆕 56 new
⏩ 214 skipped[^skipped]

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
🆕	in_place_scalar[(1024, 0.0)]	N/A	4.6 µs	N/A
🆕	in_place_scalar[(1024, 0.1)]	N/A	6.2 µs	N/A
🆕	in_place_scalar[(1024, 0.25)]	N/A	7 µs	N/A
🆕	in_place_scalar[(1024, 0.5)]	N/A	7.6 µs	N/A
🆕	in_place_scalar[(1024, 0.75)]	N/A	8.2 µs	N/A
🆕	in_place_scalar[(1024, 0.9)]	N/A	8.5 µs	N/A
🆕	in_place_scalar[(1024, 1.0)]	N/A	8.8 µs	N/A
🆕	in_place_scalar[(131072, 0.0)]	N/A	553.9 µs	N/A
🆕	in_place_scalar[(131072, 0.1)]	N/A	784.8 µs	N/A
🆕	in_place_scalar[(131072, 0.25)]	N/A	866 µs	N/A
🆕	in_place_scalar[(131072, 0.5)]	N/A	943.8 µs	N/A
🆕	in_place_scalar[(131072, 0.75)]	N/A	1 ms	N/A
🆕	in_place_scalar[(131072, 0.9)]	N/A	1.1 ms	N/A
🆕	in_place_scalar[(131072, 1.0)]	N/A	1.1 ms	N/A
🆕	in_place_scalar[(16384, 0.0)]	N/A	69.5 µs	N/A
🆕	in_place_scalar[(16384, 0.1)]	N/A	97.8 µs	N/A
🆕	in_place_scalar[(16384, 0.25)]	N/A	108.1 µs	N/A
🆕	in_place_scalar[(16384, 0.5)]	N/A	117.7 µs	N/A
🆕	in_place_scalar[(16384, 0.75)]	N/A	126.8 µs	N/A
🆕	in_place_scalar[(16384, 0.9)]	N/A	132.2 µs	N/A
...	...	...	...	...

:information_source: Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks. [^skipped]: 214 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports.

[connortsui20]

@0ax1 is there a way for me to disable the benchmark just for codspeed?

[0ax1]

@0ax1 is there a way for me to disable the benchmark just for codspeed?

~~You shouldn't need to. We compile for AVX2, so the AVX512 version shouldn't be picked up.~~

~~Ah I see that we're running into failed to execute the benchmark process, exit code: 132 which I assume is picked based on runtime capabilities of the CPU?~~

Discussed to exclude the benchmark via: #[cfg(not(codspeed))]

[connortsui20]

hmm it seems I've done something wrong here then: https://github.com/vortex-data/vortex/actions/runs/19644046057/job/56254751182?pr=5399

[connortsui20]

I'm going to close this until we start doing e2e tests and we can actually measure how much faster things are.