Expose conv2d weight/bias preparation as ops

[LPanosTT]

Tickets

#13219, #12793

Problem description

Modelling convolutions in MLIR is difficult when you need to execute conv2d once in order to get the proper inputs for the next iteration.

What's changed

Moving weight and bias preparation to dedicated ops. The logic that prepared the ops on host already exists and is functional. The issue is that it is done within conv2d. Most of the code changes is copy/pasted.

Checklist

• Post commit CI passes
• Blackhole Post commit (if applicable)
• Model regression CI testing passes (if applicable)
• Device performance regression CI testing passes (if applicable)
• New/Existing tests provide coverage for changes

[LPanosTT]

@pavlejosipovic As for your question in the previous PR. If by c/p code you mean is the logic in that function copy/pasted then yes. All of the code is copy pasted since I'm just encapsulating existing logic in their own ops.

[pavlejosipovic]

@pavlejosipovic As for your question in the previous PR. If by c/p code you mean is the logic in that function copy/pasted then yes. All of the code is copy pasted since I'm just encapsulating existing logic in their own ops.

I meant this function is not used internally and we have same logic inside other functions (instead of using the function)?

[LPanosTT]

I meant this function is not used internally and we have same logic inside other functions (instead of using the function)?

This function does get used in prepare_conv2d_weights.cpp. I think the logic here is actually duplicated in the top level conv2d as well though. Let me clean this up some more.