Alignment with macros and using const-values for faster processing

[frehberg]

Not sure this is a useful patch, I tried to increase performance using macros, and using "const" values so that the compiler can do more static optimization regarding the alignment, but bench-tests showed no improvement. Please let me know if this patch is doing any good in your environment?

[frehberg]

Btw: I realized the Serializer/Deser. is using Result-Error-type around heavy struct Error-type,in contrast to a lightweight enum would do. This might cause lot of overhead for each return-statement. Did you check if CDR lib could use a more lightweight Result-Error-Type based on light-weight enum?

[hrektts]

In my environment, which is Ubuntu 18.04, modified one is partially slower than the original. Here are my results.

Original:

test lidar_point_msg                       ... bench:     918,800 ns/iter (+/- 30,570)
test lidar_point_msg_without_encapsulation ... bench:     914,525 ns/iter (+/- 30,187)
test string_msg                            ... bench:         234 ns/iter (+/- 6)
test string_msg_without_encapsulation      ... bench:         201 ns/iter (+/- 33)

Modified:

test lidar_point_msg                       ... bench:     933,000 ns/iter (+/- 26,260)
test lidar_point_msg_without_encapsulation ... bench:     935,650 ns/iter (+/- 48,097)
test string_msg                            ... bench:         227 ns/iter (+/- 5)
test string_msg_without_encapsulation      ... bench:         192 ns/iter (+/- 5)

I don't know whether the code below is evaluated as const or not:

const ALIGNMENT: usize = std::mem::size_of::<$T>();

[hrektts]

As for the error handling, I'll take benchmarks and modify it. Thank you for letting me know :)