• adding statistics and validating approach
• cleaning up branch
• add "faster" path for majority of (bulk) transmissions
  
  Verified, if reuse of the optlen calculated for the last segment in a bulk tcp_output was "good enough", as that did address the IW10 off by one with Timestamps issue.
  
  Performed a few iperf3 tests across the public internet and various targets. Notable observations: not only in the packetdrill script, but also in real traffic the slight variance in timing (delay to send out a segment which would have been valid to send one call to tcp_output earlier) could be observed. In traffic testing this happend on sessions with higher loss rates, and not addressing this could negatively affect TCP performance.
  
  However, the vast majority of transmissions would work with the original "len >= t_maxseg" check, so keeping this in as first conditional as fast path.
  
  The "len+optlen" conditional was the initial approach, fixing the IW10 & TS off-by-one observation.
  
  Furthermore "len+TSoptLen" is doing a trinary check for the two currently supported flags, which would have TCP options always present in the header (2nd conditional to jump to send added in this Diff).
  
  Finally the "other reason" is a catch-all for all the other unrelated conditionals leading tcp_output to actually transmit a segment.
  
  Also found, that the trinary with -O2 will actually convert into branchless code (without compiler optimization, the emitted binary will have branches). However, the trinary improves readability over multiplying a boolean result with a constant (that would always result in branchless assembly but actually less readable with -O2).
  
  TCP connection count by state: 204365 len >= t_maxseg 22652 other reason to send 33 len+optlen >= t_maxseg 773 len+TSoptlen >= t_maxseg

Event: September 2020 Bugathon

• renaming macro to reduce collision risk