Encoding performance comparision

[dudycz]

Hello,
I have created some benchmark of PER codecs: rasn, hampi and asn1rs. If you are interested here are the results:

Codec	Encoding (µs)	Decoding (µs)
rasn	1658	199
asn1-codecs	132	49
asn1rs	70	68

Encoding and decoding nested SEQUENCE-OF (1000 elements, 5 bytes each)

Decoding speed of asn1-codecs / hampi is great, but in terms of encoding maybe there is something to look into (compared to asn1rs).
Repository with benchmark:
https://github.com/dudycz/asn1_codecs_bench

[gabhijit]

@dudycz : Thanks for reporting this! Will surely take a look at the encoding part and see where it can be improved!!

[gabhijit]

@dudycz : Really sorry for getting back too late on this.

Revisited this issue after nearly a year. When I tried to trace the codec for your particular usage. I realized the performance is limited by just one function in the codec encode_constrained_whole_number as this is the function that gets repeatedly called.

Also for another benchmark (NGAP_encode , NGAP_decode) which use other encode/decode functions the encode and decode performance is not that markedly different.

I was able to improve the performance of the encode path for the use-case you mentioned by about 10% by reserve ing the internal bit vector, at the moment I am not sure that's really the biggest bang for the buck.

Next will have to look at improving the above function (encode_unconstrained_whole_number) to get further benefits. Any ideas are welcome.

Right now I am not going to focus on that, but will keep this issue open with adding help wanted label for it.

[gabhijit]

With two changes 1. Inlining the append_bits function and 2. 'reserve'ing the bit vector, we see about 10% improvements in both the benchmarks that we tried, so overall this is worth it. Merging #141 .