Works for one cfg #271
Works for one cfg #271
Conversation
| input_queues iqs(isize); | ||
| for (size_t ind=0; ind<isize; ++ind) { | ||
| iqs[ind] = boost::any_cast<input_queue>(anyinqs[ind]); |
There was a problem hiding this comment.
We may don't need to change this. Will test later.
There was a problem hiding this comment.
Hi @HaiwangYu
This looks fine. Really just one comment which may be more an issue of my understanding than with the code changes.
aux/src/FrameSync.cxx
Outdated
| // push eos | ||
| for (size_t ind=0; ind<nin; ++ind) { | ||
| auto& iq = iqs[ind]; | ||
| if (iq.empty()) { | ||
| continue; | ||
| } | ||
| auto frame = iq.front(); | ||
| if (!frame) { | ||
| std::get<0>(oqs).push_back(nullptr); // forward EOS | ||
| iq.pop_front(); | ||
| return (*this)(iqs, oqs); | ||
| } | ||
| } | ||
|
|
||
| /// FIXEME: check all empty |
There was a problem hiding this comment.
This might be correct but I'm having trouble following if it is.
The "EOS sync" mechanism here is meant to cause an EOS nullptr to sit on each input queue until all input queues have nullptr at their fronts. That is, when neos equals the number of input queues. Then a single nullptr is sent out the output queue.
| get<1>(oqs).push_back(std::make_shared<EmptyFrame>(ds->ident())); | ||
| } | ||
| else { | ||
| log->debug("DeposOrBust: forward deposet"); |
There was a problem hiding this comment.
This log message would be more useful if it included some unique info. Eg, the number of depos in the set.
|
Hi @brettviren, I checked I noticed the |
|
Add some notes for documentation. |
This works for the cfg below, but could have missed stuff:
https://github.com/HaiwangYu/hydra-skip/blob/main/wct-sim-fans-hydra.jsonnet
ref:
https://indico.fnal.gov/event/60987/contributions/282811/