<daveshah>
Yes, although if you look at the original data it's more accurately undefined between 675ps and 1821ps after a clock
<daveshah>
But nextpnr should be taking this into account when calculating Fmax anyway
<meawoppl>
gotcha
<meawoppl>
I understand now that my `iverilog` treatment made the DATAOUT signal immediately valid
meawoppl has quit [Remote host closed the connection]
<lambda>
is there a way to read EDIF files back into yosys?
<daveshah>
No, I don't think there is an open source EDIF frontend
<lambda>
hm, that's unfortunate - seems like I found some bugs with EDIF output (or vivado at least understands it differently), reading it back and synthesizing with nextpnr would've been helpful for testing
<lambda>
any other format yosys can output and vivado can read?
<daveshah>
Verilog should work
<lambda>
right, sure
<lambda>
yeah, definitely multiple incompatibilities with EDIF even for a very simple design.
<daveshah>
Are you using write_edif
<daveshah>
You need to use write_edif -pvector bra
<lambda>
oh, that may have helped. but I think there were other issues as well, particularly vector ports defined as [7:0] being indexed as [0:7] in the EDIF, causing everything to be the wrong way around
<lambda>
(and differential ports not working at all because the positive and negative ports were also reversed)
<lambda>
hooray, my DVI output works when I synthesize with vivado - must be something with nextpnr then, I'm guessing the TMDS33 differential outputs.
<daveshah>
TMDS33 isn't supported
<daveshah>
I don't even think Xray has the bits for it
<lambda>
ah, fair enough.
<lambda>
I don't really need it anyway, it's just nice for demo purposes, and running a netlist through vivado is absolutely fine for those few times
<daveshah>
At low resolution using two LVCMOS33 outputs with inverted data (inverted at the OSERDES input) might work
<daveshah>
*inverted data on the negative side
<daveshah>
I've done this on the ECP5 in the past
<lambda>
not actually using a SERDES, which would make that even easier
<lambda>
(250MHz is still doable with a decent margin)
<lambda>
on the other hand, that added inverter might cause some delay between positive and negative
<lambda>
oh, just buffer it, duh. I think I'd rather do it properly or not at all though, and as long as nextpnr doesn't error because of the unsupported outputs it doesn't really matter