Marocchino devel

[stffrdhrn]

This is a WIP to get marocchino working in the or1k-tests suite with mor1kx-generic. You can see my repos for those. This compiles now, but still some work to do.

Comments:

• During this integration I found marocchino is almost a complete rewrite of mor1kx. Should we just separate it out to a new repo i.e. openrisc/marocchino? This will give it better exposure, and I am not so sure integrating it into mor1kx main will give it the credit it needs.
• The main issues I see with integrating it is:
  1. It uses a different wb interconnect (i.e. so there are new signals from WB to the pipeline)
  2. It doesnt use the toplevel mor1kx module

[bandvig]

@stffrdhrn

• During this integration I found marocchino is almost a complete rewrite of mor1kx.

You are right. MAROCCHINO is almost complete rewrite of CAPPUCCINO (if be more precise). I did that to exclude extra variability caused by large number of OPTION_* parameters and to separate my additional modifications (like pseudo CDC and ORFPX64A32) from other sources. I wished to prevent of exponential growing configuration-related “if”s.

Should we just separate it out to a new repo i.e. openrisc/marocchino? This will give it better exposure, and I am not so sure integrating it into mor1kx main will give it the credit it needs.

I haven’t got any objection. Perhaps we should ask other members of OpenRISC-team. Unfortunately, I’ve a problem with subscription for mailing list in LibreCores right now. No problem if you rise the topic in the mailing list till I restore my subscription. The only thing. I would prefer “openrisc/mor1kx_marocchino” name for such repo.

• The main issues I see with integrating it is:
  1. It uses a different wb interconnect (i.e. so there are new signals from WB to the pipeline)
  2. It doesnt use the toplevel mor1kx module

I had to redesign Wishbone bridges and their interface with implementing pseudo-CDC. I moved part of atomic and snoop support logic into data-bridge in Wishbone clock domain.

[stffrdhrn]

@bandvig

Should we just separate it out to a new repo i.e. openrisc/marocchino? This will give it better exposure, and I am not so sure integrating it into mor1kx main will give it the credit it needs.

I haven’t got any objection. Perhaps we should ask other members of OpenRISC-team. Unfortunately, I’ve a problem with subscription for mailing list in LibreCores right now. No problem if you rise the topic in the mailing list till I restore my subscription. The only thing. I would prefer “openrisc/mor1kx_marocchino” name for such repo.

Sure, since its derived from mor1kx that's fair. I do believe the alu is still shared. Copying/branching that should be fine though.

Let me forward this conversation to the list.

• The main issues I see with integrating it is:

  1. It uses a different wb interconnect (i.e. so there are new signals from WB to the pipeline)
  2. It doesnt use the toplevel mor1kx module

I had to redesign Wishbone bridges and their interface with implementing pseudo-CDC. I moved part of atomic and snoop support logic into data-bridge in Wishbone clock domain.

Yes, I noticed those signals. It seems reasonable and makes your WB interface more smart/efficient.

[bandvig]

@stffrdhrn

I would prefer “openrisc/mor1kx_marocchino” name for such repo.

Sure, since its derived from mor1kx that's fair.

Yep, and another minor reason. With such name the MAROCCHINO repo will be placed just before or after of “openrisc/mor1kx” in list of OpenRISC organization's repos :).

I do believe the alu is still shared. Copying/branching that should be fine though.

No, ALU isn’t shared. The mor1kx_execute_alu module (in mor1kx_execute_alu.v file) is shared across CAPPUCCINO, ESPRESSO an PRONTOESPRESSO. It includes circuits for divider, multiplier and all other operations. Opposite to that the mor1kx_execute_marocchino.v file includes three different modules:

• pipelined multiplier (module mor1kx_multiplier_marocchino). By the way it doesn’t generate overflow similar to “PIPELINED” OpenRISC’s multiplier

• serial divider (module mor1kx_divider_marocchino)

• and all other operations are placed into module mor1kx_exec_1clk_marocchino as these operations are 1-clock length

I know that good style is “one module per one file”. However, following that could be quite annoying sometimes :).

Let me forward this conversation to the list.

Sure.