ConcAugThread ideas

[yakra]

• Try removing log_mtx, replace << with + operators, and see how that works out.
• If that's not satisfactory, build a list of strings in each thread, splicing onto a list in the main thread when done, and finally output the file at once all threads finish.
• Try if (hs !=s) (This may also help the Python version) https://github.com/yakra/DataProcessing/blob/68315717a3802d7ab37a4c86496eae89826f042e/siteupdate/cplusplus/threads/ConcAugThread.cpp#L18-L19
  • Or alternately we could remove https://github.com/yakra/DataProcessing/blob/68315717a3802d7ab37a4c86496eae89826f042e/siteupdate/cplusplus/classes/TravelerList/TravelerList.cpp#L188
  • At 1st glance this looks better, as it avoids the IF & conditional branch.
  • OTOH, there may be more more competition for clin_mtx? All these operations are packed into a much smaller timeframe...
  • See what works out better in practice...

Similar: https://github.com/TravelMapping/DataProcessing/issues/281

[yakra]

• Try removing log_mtx, replace << with + operators, and see how that works out.

concurrencies.log gets mangled. Not as atomic as I thought.

---

• If that's not satisfactory, build a list of strings in each thread, splicing onto a list in the main thread when done, and finally output the file at once all threads finish.

Or rather, just keep an array of lists, then step through it & write each one. But anyway. This is a big help immediately. One thread is a bit slower now, due to constructing the list of strings rather than writing the log file directly. I could make up for this by interleaving some if (args.numthreads == 1) & else with the #ifdefs and using the same code for 1 thread as when threading is disabled, but don't really care; If running single-threaded for anything other than speed testing or debugging, one should disable threading at the preprocessor level anyway.

---

• Try if (hs !=s) (This may also help the Python version) https://github.com/yakra/DataProcessing/blob/68315717a3802d7ab37a4c86496eae89826f042e/siteupdate/cplusplus/threads/ConcAugThread.cpp#L18-L19

• Another moderate improvement for C++.
• [x] Python testing is underway.

---

• Or alternately we could remove https://github.com/yakra/DataProcessing/blob/68315717a3802d7ab37a4c86496eae89826f042e/siteupdate/cplusplus/classes/TravelerList/TravelerList.cpp#L188

Nope, not even attempted. Travelers would never get credited for non-concurrent segments.