68881 VHDL implementation

[Kochise]

For what it's worth :

https://github.com/mattuna15/68881-fpga

[exxos]

Interesting!

[alexh]

It's very big in terms of FPGA utilisation so I don't think we'll see it alongside cores in MiSTer. I don't know if it's representative of the original silicon, I don't know how much RAM, registers, ALU elements there were in the original.

Very nice thing to have though

[Kochise]

There's a "LITE" version (see "fpu_lite_g => true"), the same as the 040 subset that needs the software emulation package for the missing instructions.

[alexh]

I'm not sure how useful the Lite version will be for non 040 implementations? At least for pre-compiled software?

IIUC 040 uses a different method of accessing the registers than 020 and 030? They use the co-processor bus and Line-F traps or something (this is not my area of expertise). There is a chance software compiled for those CPU/FPU combos are likely to expect the full instruction set? But who knows?

Nice thing to have definitely.

[Kochise]

Obviously is "saves" ~57-60% LUT, hence make it possible to fit smaller FPGA.

[alexh]

Like buses

https://github.com/harrowm/mh882

[Kochise]

If everyone is copying each other, now waiting for 68030 in VHDL.

No, 68060...

[alexh]

now waiting for 68030 in VHDL

Same guy as above has one in SystemVerilog.

https://github.com/harrowm/MH030

I don't see why you'd want VHDL. The world went (System) Verilog 15 years ago. And FPGA tools don't care. They mix and match modules/components in different languages no problem.

But if you needed to, you could use one of the tools available to convert to VHDL and use Formal Verification (Formality / YoSys) to prove it was identical.

[Kochise]

VHDL, Verilog, whatever, provided we can get a cycle exact Motorola CPU implementation that we can use to create new boards and accelerators...