Added Openssl feature

[nomick]

Changing the sha2 implementation to openssl may increase the speed tremendously. On a Cascade Lake Xeon we saw 60% increase in PUT performance. As openssl may not be a good choice in every situation, it is capsuled in a cargo feature. The unwraps may be not wanted, but avoiding them would change the function signature.

WARP before:

Operation: PUT. Concurrency: 20

• Average:216.98 MiB/s, 21.70 obj/s

Throughput, split into 111 x 1s:

• Fastest:244.7MiB/s, 24.47 obj/s
• 50% Median:217.6MiB/s, 21.76 obj/s
• Slowest:193.1MiB/s, 19.31 obj/s

WARP after:

Operation: PUT. Concurrency: 20

• Average:354.65 MiB/s, 35.46 obj/s

Throughput, split into 65 x 1s:

• Fastest:377.3MiB/s, 37.73 obj/s
• 50% Median:355.9MiB/s, 35.59 obj/s
• Slowest:324.6MiB/s, 32.46 obj/s

---

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[Nugine]

So the sha256 in openssl is much faster than pure rust sha2? It's suprising. We should add a simple benchmark to track this difference and report to RustCrypto/hashes/sha2.

[nomick]

Unfortunately. they are perfectly aware of this. See https://github.com/RustCrypto/hashes/issues/327

[St4NNi]

Was curious about this, I guess it really depends on your specific CPU. If your CPU supports the SHA-NI extension openssl and RustCrypto perform about the same, if not RustCrypto is up to 2x slower.

A quick test on our EPYC Milan (Zen3) and on my personal Zen4 machine showed, as expected, no measurable difference between RustCrypto and OpenSSL.