[AIE2/2P] Fix DONE latency at region end.

[SagarMaheshwari99]

No description provided.

[martien-de-jong]

nit: I hope no one needs to call this from outside. This one could go and instances could be private or inlined.

[SagarMaheshwari99]

made private.

[gbossu]

Can you document what that means?

[SagarMaheshwari99]

done.

[gbossu]

FYI @mludevid you might have done something similar when working on semaphores.

[andcarminati]

I think we can have a design where we need just one method to be reimplemented. Something like:

virtual std::optional<unsigned> getDoneLatency(...)

In this case, we can have a base implementation for getImplicitLatency that can be reimplemented by the target only if needed.

[martien-de-jong]

@andcarminati, The idea was to have a more generic name excluding 'Done'. It's basically an extra latency that can be given to the exit edge. But yes, we would only have one generic method that fills it in.

[andcarminati]

Hi, my idea is to avoid encode the latency knowledge separately. If we have a new done, we have two places to change.

[SagarMaheshwari99]

The aim was to encode all "special latencies" into one "getImplicitLatency", so that we don't touch AIEBaseSubTarget frequently. Is there another way other than having a separate function for Opcode to achieve all in one getImplicitLatency?

[andcarminati]

Just as a suggestion, I was thinking in something like:

// Return an optional latency if Opc is a done instruction.
std::optional<unsigned>AIE2InstrInfo::getDoneLatency(unsigned Opc) const {
 return (Opc == AIE2::DONE) ? 6 : std::nullopt
}

// This can be in the base class.
unsigned AIE2InstrInfo::getImplicitLatency(const MachineInstr &MI) const {
  if (auto OptLatency = getDoneLatency(MI.getOpcode()))
    return *OptLatency;

  return 0;
}

New targets can just implement getDoneLatency which will encode opcode + latency in the same place.

As I said, just a suggestion to have a single source of information here.