09:53 <agg> daveshah: I'm back trying to get the ALU54 working, I've worked out a bunch of required connections and attributes and have a very minimal design that synthesises to a working MAC in Diamond but (with location constraints to use the same DSP tiles) doesn't work in nextpnr; if I manually copy all the MIB*_DSP* tile config from the working bitstream it still doesn't output anything which makes me wonder if

09:54 <agg> the ALU output to fabric in the CIB_DSP tile is wrong, any thoughts on what the best avenue of investigation might be?

09:54 <agg> happy to fire up the fuzzer and focus it on the ALU stuff or play with its python script if that's possibly the way to go, but I've not used any of the icestorm/trellis fuzzers before.

09:57 <daveshah> agg: hmm, it could be that the routing differs between different DSPs, which is not something that I've seen but the fuzzers aren't good at picking up

09:58 <daveshah> You could test that theory by picking a different set of tiles for the DSP interconnect fuzzer and seeing whether any of the pertinent routing has different results

09:58 <agg> using the fuzzer sounds more fun than doing that by hand at least

09:58 <daveshah> The DSPs are quite mysterious in general, unfortunately

09:59 <agg> incidentally is there an attribute like BEL=X9/Y22/MIB_DSP0 but for diamond?

09:59 <agg> I'd been letting diamond do what it wants and matching it in nextpnr

09:59 <agg> the diamond error strings file contains a lot of hints about the DSPs

09:59 <daveshah> It's a LOC attribute

09:59 <agg> https://i.imgur.com/p95bOqn.png

10:00 <daveshah> the bel names will differ though, you can find the naming Diamond uses from the floorplan view

10:01 <agg> thanks

10:02 <agg> do I need diamond 3.10 exactly for fuzzers or should 3.11 work?

10:04 <daveshah> 3.11 should work too