sc_fifo reset

[Ramon Fried]

I'm wondering why there's no standard method to clear/reset a fifo.
Doing so activly, by looping and running:

while (nb_read(t))
 ;

has the unplesent effect of sending data_read_event().

What's the process of suggesting an addtional functionality to the spec ?

[David Black]

You only get a single read event.

If you need more, simply implement your own fifo by derivation or composition. You might also consider the tlm_fifo, which has slightly more functionality but still no reset per se.

You could also disable the readers for one delta cycle if you know their identities. Perhaps override data_read_event to capture that information:

// File: my_fifo.h
#pragma once

#if __cplusplus < 201402L // Might allow C++11
  #error "Requires C++14 or better"
#endif

#include <systemc>
#include <forward_list>
struct my_fifo : sc_core::sc_fifo
{
  my_fifo( int size = 16 ) : sc_fifo{ size } { sc_assert( size >= 0 ); }
  const sc_core::sc_event& data_read_event() override {
    observer.push_back( sc_core::sc_get_current_process_handle() );
    return sc_core::data_read_event();
  }
  void reset() { //< blocking
    // Disable observers
    for( auto& observer : observers ) observer.disable();
    // Empty the fifo
    while( nb_read() );
    // Exit this delta cycle
    wait( SC_ZERO_TIME );
    // Reenable observers
    for( auto& observer : observers ) observer.enable();
  }
private:
  std::forward_list<sc_core::sc_process_handle> observers{};
};

Above is untested, and you might want to add a my_fifo_if.h to support the reset functionality via a port. YMMV

[David Black]

Reflection: It might be better to store the observers in a std::unordered_set rather than the std::forward_list. This avoids the list getting duplicate entries and thus getting out of control (size). This would be the case if code dynamically waited on the event.

 

[Ramon Fried]

Thanks David !