I've got a fairly simple situation: I have a high speed producer of data in one thread that generates a buffer with [varying length] elements in it. Once this buffer is filled, I have a consumer that writes it to disk.
Right now, the producer thread, which needs to get back to producing ASAP, spins in place if the buffer hasn't been written by the consumer thread yet:
int volatile datatogo=0; // global with starting condition
while(datatogo != 0) // spin, buffer not yet written out
{
if (recflag == 0) return; // recording is terminated
}
// clipped code fills buffer, then:
datatogo = lengthoffill;
...and in the other thread, the buffer writer does this:
while(recflag)
{
if (datatogo)
{
if (m_File.write(sbuffer,datatogo) == -1)
{
recflag=0; // bail NOW
}
datatogo = 0; // buffer transferred
}
usleep(100); // 100 uS
}
The net effect of this is that the write-to-disk consumer gives up the CPU when it has completed a write, knowing that the producer will have to take some time to actually fill the buffer. With the consumer asleep when there is no data, the CPU is available to the producer. The consumer sleeps for 100 uS, checks for data, and goes back to sleep if there is none, so it doesn't do much in that state except sleep.
However, because the sleep time is arbitrary, this is unlikely to ever be optimum; even if I carefully tune it to work on my machine (8 cores, 3 GHz), it'll act differently on another machine. There will be times when data is ready to be written, and the consumer has just gone to sleep, tossing the entire 100 uS out the window, so to speak. I also have qualms about the resolution and reliability of such a small timing window -- and larger windows won't work.
So. There are multiple mechanisms in Qt to control access to things, which is essentially what I want to do. But I don't understand which one (if any) will do what I want to do, which is:
1) have the consumer sleep until the buffer is full, then write and go back to sleep, or until the job is stopped so it can close the file (it can check which when woken up)
2) while the producer sleeps only if, and while, the buffer is being written out, otherwise fills the buffer and goes back to producing its content.
I need as much CPU available as possible -- this is a software defined radio application, and there is data flying everywhere, multiple FFTs running, all manner of graphics hoo-ha happening, etc. Spin time is werry, werry bad.
Can some kind soul point me to the ideal Qt mechanism to use? I find the Qt docs on QMutex, QWaitCondition, QSemaphore to be... somewhat opaque.
