[io] Improve PCD read performance (more than 50%) by reusing istringstream

[kunaltyagi]

Single threaded performance increases by approximate 50%

#4333 (comment)

[SergioRAgostinho]

I'm not going to review this one. I have no progress on it.

[mvieth]

The changes look ok, but I feel that there has to be a nicer way to do all this. What about stod, stof, stoi, ...? Alternatively, maybe the boost:split in pcd_io.cpp can be removed and instead the stream used for the whole line?

[kunaltyagi]

What about stod, stof, stoi, ...?

AFAIK, these functions are locale dependent. In C, the locale is global and changing locale might affect downstream users during loading of a file.

Alternatively, maybe the boost:split in pcd_io.cpp can be removed and instead the stream used for the whole line?

I've marked that issue as help-wanted because I couldn't refactor it cleanly without touching quite a few lines. I too would like to see a cleaner implementation than this low hanging fruit.

[larshg]

I agree to @mvieth observations, but as you write @kunaltyagi it can be left for another day :)

50% performance increase, is also quite nice!

[mvieth]

This new functions lacks the documentation, especially regarding is. Also: "Uses aoti/atof to do the conversion." isn't really correct anymore.