As part of my work on openlane support in edalize/...
# caravel
k
As part of my work on openlane support in edalize/fusesoc, I'm adapting a fork of the Caravel repo (mpw-one-b tag at the moment) to use Fusesoc. A real world example to see what features we might want, and ideally me and Kindgren would like to do a shuttle submission with a wrapped SERV in the user project area built through Fusesoc commands. I'm having issues with building mgmt_core. I'll guess I'll get back with particular questions since I have a build going right now, but one thing I'm wondering is if build times up to around 24h sound like they are in a reasonable ballpark for this block? Using an i7-10750H here. It also seems to get completely stuck after reporting LVS mismatches, so I've disabled that entire step for now which is hardly ideal.
m
@Klas Nordmark my understanding is that macros included in caravel (mgmt_core, chip_io, digital_pll, SRAM, etc.) are to be treated as hard macros. It is not intended that these designs be remade using openlane. The openlane flow is used to rebuild the user_project_wrapper. Please correct me if I'm mistaken.
k
I understand that as the intended flow, but I would have thought rebuilding things yourself to be a possibility. Not of critical importance to me, I suppose, but a nice way to try a non-trivial use case, to see what kind of things we might need when generating tcl files and so on beyond the bare minimum.
m
my understanding is that we should be able to build all those blocks. But you may want to get @Tim Edwards to give you some pointers about what tags/configs of the tools were used to get the results
t
@Matt Venn, @Klas Nordmark: There is a "frozen" state tagged "mpw-one" (or something similar) on all the tools, with which it should be possible to exactly reproduce the Caravel build. I didn't build Caravel, though, so I don't know how long it takes.
m
Thanks Tim. Who did build it?
k
Good to know, kind of what I guessed to be the case - working with mpw-one-b tag on caravel and openlane, with the PDK fetched with make pdk in the openlane folder - I checked that the commit hashes were according to instructions in the repository as well. Well, I'll see how this build turns out in an hour or so.
t
@Ahmed Ghazy. But he just left AUC to pursue a Ph.D. in Europe, and I don't know how available he is. @Manar Abdelatty should be familiar with the flow and how long it takes to run Caravel through synthesis, place, and route.
m
I've had questions for him and he hasn't responded for weeks, so he is not really available, FYI.
t
That way my impression, too.