feat: add openssl-sha256 to benchmark

[jonathanpwang]

I added Sha256 from rust-crypto to Plonky3/Plonky3#413
Apparently the rust-crypto implementation even with NEON is perhaps 2x worse than the openssl one: RustCrypto/hashes#327 (comment)

Here I directly implement CryptographicHasher on the openssl sha256, let's see the benchmarks.

The openssl rust crate is just a FFI wrapper for their C library. I am hoping they use the ARMv8 SHA256 intrinsics: https://mijailovic.net/2018/06/06/sha256-armv8/

[jonathanpwang]

Inspired from seeing stwo using a custom simd blake2: https://github.com/starkware-libs/stwo/blob/dev/crates/prover/src/core/backend/simd/blake2s.rs

[github-actions]

Previous Benchmark

Benchmark for Predicate

Total Cells	Total Prove (ms)	Main Trace Gen (ms)	Perm Trace Gen (ms)	Calc Quotient Values (ms)	Rest of Prove (ms)
19_726_336	719.00	211.00	8.17	30.20	469.63

AIR metrics

Name	Rows	Cells	Prep Cols	Main Cols	Perm Cols
PageIndexScanInputAir	32_768	10_354_688	0	[97, 147]	[72]
PageIndexScanOutputAir	32_768	8_716_288	0	[97, 97]	[72]
RangeCheckerGateAir	65_536	655_360	0	[2]	[8]

Commit: 3dfdfb4
AWS Instance Type: r7g.8xlarge
Benchmark Workflow

New Benchmark

Benchmark for Predicate

Total Cells	Total Prove (ms)	Main Trace Gen (ms)	Perm Trace Gen (ms)	Calc Quotient Values (ms)	Rest of Prove (ms)
19_726_336	700.00	213.00	8.27	31.60	447.13

AIR metrics

Name	Rows	Cells	Prep Cols	Main Cols	Perm Cols
PageIndexScanInputAir	32_768	10_354_688	0	[97, 147]	[72]
PageIndexScanOutputAir	32_768	8_716_288	0	[97, 97]	[72]
RangeCheckerGateAir	65_536	655_360	0	[2]	[8]

Commit: 171c442
AWS Instance Type: r7g.8xlarge
Benchmark Workflow

[github-actions]

Previous Benchmark

Benchmark for ReadWrite

Total Cells	Total Prove (ms)	Main Trace Gen (ms)	Perm Trace Gen (ms)	Calc Quotient Values (ms)	Rest of Prove (ms)
38_764_544	1500.00	288.00	25.70	174.00	1012.30

AIR metrics

Name	Rows	Cells	Prep Cols	Main Cols	Perm Cols
PageReadAir	32_768	1_343_488	0	[33]	[8]
IndexedPageWriteAir	32_768	5_603_328	0	[33, 98]	[40]
PageOfflineChecker	131_072	29_753_344	0	[179]	[48]
RangeCheckerGateAir	65_536	655_360	0	[2]	[8]
ExecutionAir	32_768	1_409_024	0	[35]	[8]

Commit: 3dfdfb4
AWS Instance Type: r7g.8xlarge
Benchmark Workflow

New Benchmark

Benchmark for ReadWrite

Total Cells	Total Prove (ms)	Main Trace Gen (ms)	Perm Trace Gen (ms)	Calc Quotient Values (ms)	Rest of Prove (ms)
38_764_544	1530.00	297.00	26.30	173.00	1033.70

AIR metrics

Name	Rows	Cells	Prep Cols	Main Cols	Perm Cols
PageReadAir	32_768	1_343_488	0	[33]	[8]
IndexedPageWriteAir	32_768	5_603_328	0	[33, 98]	[40]
PageOfflineChecker	131_072	29_753_344	0	[179]	[48]
RangeCheckerGateAir	65_536	655_360	0	[2]	[8]
ExecutionAir	32_768	1_409_024	0	[35]	[8]

Commit: 171c442
AWS Instance Type: r7g.8xlarge
Benchmark Workflow

[jonathanpwang]

workflow for more comprehensive benches: https://github.com/axiom-crypto/afs-prototype/actions/runs/10311730058

results here: https://docs.google.com/spreadsheets/d/1wwhyGgFNGcapD2xk5_gixPFt8dlp722bRwwFHmaKzrE/edit?usp=sharing

Closing for now since there's no difference. I feel there is something slightly fishy about this, but I'd need to make some flamegraphs of the trace commitment to go deeper.