search

rust-cv / header-vec

Allows one to store a header struct and a vector all inline in the same memory on the heap and share weak versions for minimizing random lookups in data structures
MIT License
5 stars 2 forks source link

Add benchmarks against std vec #2

Closed makoConstruct closed 3 years ago

makoConstruct commented 3 years ago

(I'll do this now unless you have objections)

vadixidav commented 3 years ago

Go for it. My focus was mainly on the functionality initially, but, by all means, add any tests and benchmarks you want. I only benchmarked this crate indirectly using hgg, but a direct benchmark would be excellent. Thanks for contributing!

makoConstruct commented 3 years ago

resolved #3

Hm, so, to discuss the results, it's getting numbers like 

running 6 tests
test test_header_vec_create          ... bench:     246,006 ns/iter (+/- 1,133,418)
test test_header_vec_create_smaller  ... bench:       2,871 ns/iter (+/- 4,370)
test test_header_vec_read            ... bench:         507 ns/iter (+/- 41)
test test_regular_vec_create         ... bench:       3,258 ns/iter (+/- 274)
test test_regular_vec_create_smaller ... bench:         829 ns/iter (+/- 80)
test test_regular_vec_read           ... bench:         509 ns/iter (+/- 14)

Which is not the best. The test_header_vec_create bench's runtime varies between being a factor of 4 to 60 times slower than test_regular_vec_create. The extremeness of that variance could have something to do with bencher using too much memory and with my linux box being deeply stuck in swap right now, so I'm going to reboot and run it again.

makoConstruct commented 3 years ago

Okay I rebooted, results far less dire. header vec is about 4.3x slower.

running 6 tests
test test_header_vec_create          ... bench:      10,746 ns/iter (+/- 932)
test test_header_vec_create_smaller  ... bench:       2,562 ns/iter (+/- 55)
test test_header_vec_read            ... bench:         472 ns/iter (+/- 12)
test test_regular_vec_create         ... bench:       2,922 ns/iter (+/- 98)
test test_regular_vec_create_smaller ... bench:         792 ns/iter (+/- 44)
test test_regular_vec_read           ... bench:         470 ns/iter (+/- 12)
vadixidav commented 3 years ago

Yeah, I will open a tracking issue to see if we can get closer to Vec performance. That could help speed up hgg, and I would be pretty happy about that. Thanks again for the benchmarks and posting the results, that is so helpful.