How does labindex in spmd works?

Based on the spmd labindex coding provided by @Walter Roberson here, I wonder when do the following labindex (labindex 2, 3 and 4) execute their jobs? Is it based on clock cycle or any enable port?

Or labindex 2, 3 and 4 execute their jobs once they received data and in waiting (pause) condition when there is no data received?

framesize = 4096;
P = parpool(4);
spmd 
  if labindex == 1
    end 
  end
spmd 
  if labindex == 1 
    afr = dsp.AudioFileReader(Filename, 'SamplesPerFrame', framesize);
    while ~isDone(afr) 
      frame = step(afr);
      labSend(frame, 2);
    end
    labSend([], 2)
    release(afr) 
    elseif labindex == 2
      while true
        frame = labReceive(1);
        if isempty(frame) 
          labSend([], 3)
          break
        end
        Bres = BlockB(frame);
        labSend(Bres, 3);
      end
    elseif labindex == 3
       while true
         Bres = labReceive(2);
         if isempty(Bres) 
           labSend([], 4)
           break
         end
         Cres = BlockC(Bres) ;
         labSend(Cres, 4) 
      end
    else
      while true 
        Cres = labReceive(3);
        if isempty(Cres) 
           break 
        end 
        Dres = BlockD(Cres) ;
      end
    end

Answer 1

[Walter Roberson]

P = parpool(4); creates four processes. All four of them start executing the same block of code, with nearly the only difference between the processes being that if they call labIndex they will get different results from 1 to 4.

All four of the processes will begin executing at the same time, without waiting for anything. They will not wait for input.

As soon as any of the processes has nothing left to do, it will go into a "waiting to exit" state and will not participate in any further work except to help collect the outputs when the last of the labs exits. This applies even if another lab specifically does labSend() to it.

If your workers interact, then it is common to write the smpd block in terms of "for" or "while" loops. You can use either design pattern:

spmd
    for K = 1 : whatever
        if labIndex == 1
            something
        elseif labIndex == 2
            something else
        end
    end
end

or

spmd
    if labIndex == 1
        for K = 1 : whatever
            something
        end
    elseif labIndex == 2
        for K = 1 : whatever
            something else
        end
    end
end

The second of those is more common, as it permits the different labs to have different varieties of work.

The way to get a lab to wait for work is to have the lab execute labReceive() or labBarrier()

There is no clock and no enable port being used here. Each of the labs is a different process that runs as quickly as it can, and the different labs do not try to synchronize until you use labSend() / labReceive(0 or labBarrier(), or until they finish.

[Star Rats]

Thanks for the explanation @Walter!

[Star Rats]

The other thing, how many thread does each worker use? Is it possible to increase the number of computational threads as shown here in order to make the process run faster as what I ask in this post?

[Walter Roberson]

The default number of cores is 1 per worker, but I believe that can be increased using the techniques you have indicated.

[Star Rats]

Thanks @Walter. The Matlab magician!