Forum Discussion

Ric2's avatar
Ric2
Icon for New Contributor rankNew Contributor
2 years ago

Deadlock while filling pipe for simulation

Hi, following best practice as descibed here we run into trouble with stalled writes. When you create a testbench for a oneAPI kernel that you intend to compile as an IP core, write all your data...
High-level Design Tools
Kevin_Xu_Intel's avatar
Kevin_Xu_Intel
Icon for New Contributor rankNew Contributor
2 years ago

Hi Ric,

Sorry this took so long. I think the root of the problem we encountered with all of our workaround that uses pthread is that multithreading isn't fully supported in FPGA flow with oneAPI as it is mention in this release note.

I would suggest performing pipe read and write in the same loop:

q.single_task<Threshold>(ThresholdKernel{});

    // Check that output pixels are below the threshold
    bool passed = true;
    for (int i = 0; i < (width * height); ++i) {
      // Write to input pipe
      bool start_of_packet = (i == 0);
      bool end_of_packet = (i == ((width * height) - 1));
      StreamingBeatT in_beat(i, start_of_packet, end_of_packet);
      InPixelPipe::write(q, in_beat);
      
      if(i%1000 == 0)
        std::cerr << "Input: " << i << "\t " << (width * height) << std::endl; 
      
      // Read from output pipe
      StreamingBeatT out_beat = OutPixelPipe::read(q);
      passed &= (out_beat.data <= kThreshold);
      if(i%1000 == 0)
        std::cerr << "Output: " << i << "\t " << out_beat.data << std::endl;
    }

Let me know if this could work for you,

Thanks,