> I don’t think that recv_async_msg() blocks. So an infinite loop will chews up
> CPU.
I had a separate thread that only called recv_async_msg(). I have unused cores
on my machine, my reasoning was that I’d rather have a CPU busy looping than
risk Ls on my transmit because the hot loop Tx thread had to handle a bunch of
async messages in a row and create cascading issues.
I dug into the recv_async_msg() implementation a little. There are a few
implementations and I’m not that familiar with the UHD software architecture,
but the majority of them essentially wrap a call to
bounded_buffer::pop_with_timed_wait.
`UHD_INLINE bool pop_with_timed_wait(elem_type &elem, double timeout)`
` {`
` boost::mutex::scoped_lock lock(_mutex);`
` if (_buffer.empty())`
` {`
` if (not _empty_cond.timed_wait(lock, to_time_dur(timeout),`
` _not_empty_fcn))`
` {`
` return false;`
` }`
` }`
` this->pop_back(elem);`
` _full_cond.notify_one();`
` return true;`
` }`
It appears I was inadvertently adding contention over a mutex by putting this
call in a separate thread. It looks like this function is also relatively
lightweight as long as timeout is 0, so I’m going to try experimenting with
putting recv_async_msg() in my main Tx thread.
> I assume that redirection is not appropriate because your application uses
> stdout?
Correct. It doesn’t strictly need it, but I made the mistake of adding a debug
feature that the users have become accustomed to that I now have to maintain…
_______________________________________________
USRP-users mailing list -- [email protected]
To unsubscribe send an email to [email protected]
