<Wanda[cis]>
apparently it's the thing that SB_MAC16 tiles originally come from; they weren't designed for Ultra*
<Wanda[cis]>
and they are... weird. the DSP columns contain 2 normal-seeming PLBs between every pair of DSP tiles.
<Wanda[cis]>
and the fucking carry chain goes through both PLBs and DSPs.
<Wanda[cis]>
apparently CO was originally just connected to the carry input of the next PLB in chain
<Wanda[cis]>
and when they made Ultra, they kind of moved DSPs to the side columns and routed CO via random IPconnect instead?
<Wanda[cis]>
probably that's why it's so weird
<Wanda[cis]>
also it has 16kbit RAM blocks instead of 4kbit. and hard TMDS receiver blocks.
<Wanda[cis]>
also a funny PLL variant that can accept differential clocks?
<Wanda[cis]>
also it's incredibly buggy in current icecube; seemingly reasonable netlists involving MX16K-specific features just crash placer or router with assertion failures
<Wanda[cis]>
I wonder if they broke that stuff when adding Ultra and hacking SB_MAC16 to work with the new connections; might try a much older icecube if I can find it
h_ro has quit [Ping timeout: 260 seconds]
gatecat[m] has joined #prjcombine
<gatecat[m]>
The Ultra IPConnect tiles are just logic tiles with the DSPs or whatever taking their inputs from the LUT inputs and routing their output into the LUTCascade input
<gatecat[m]>
Almost like they just patched the GDS or something, the LUT is still there
<gatecat[m]>
configured to feed from LUTCascade input to output by the tools
<Wanda[cis]>
<mupuf> "Wanda: PLL with differential..." <- PLL primitive that accepts a differential clock from two input pads
<Wanda[cis]>
likely because the PLLs are somehow used by the hard TMDS receiver macro, and TMDS involves a differential clock
<mupuf>
Oh, that would make sense indeed!
<whitequark[cis]>
<Wanda[cis]> "likely because the PLLs are..." <- TMDS transmits data at 7X the clock, doesn't it?
<Wanda[cis]>
yeah
<Wanda[cis]>
er
<Wanda[cis]>
hold on, no
<Wanda[cis]>
it's 8b10b encoded; it's 10x the clock
<Wanda[cis]>
FPD-link is the 7x one
<whitequark[cis]>
oh yeah wrong display standard
<whitequark[cis]>
well, samedifference
<Wanda[cis]>
... or 3.5x one actually
<whitequark[cis]>
* well, same difference
<whitequark[cis]>
3.5x?
<Wanda[cis]>
um
<Wanda[cis]>
no
<Wanda[cis]>
7x
<Wanda[cis]>
with funny duty cycle
<Wanda[cis]>
I got very confused about DDR
<whitequark[cis]>
a lot of people these days don't know much about the Deutsche Demokratische Republik
<Wanda[cis]>
<Wanda[cis]> "it's 8b10b encoded; it's 10x the..." <- and by "8b10b" I actually mean "an off-brand thing that's kinda like 8b10b but different"
<Wanda[cis]>
hmm
<Wanda[cis]>
I'd be very interested in obtaining the Silicon Blue ICE Technology Library PDF, version 2.4 or 2.5 specifically
<Wanda[cis]>
or maybe Lattice ICE Technology Library, not completely sure if it was before or after the acquisition
<whitequark[cis]>
found it
<whitequark[cis]>
oh wait, i found 2.3, not 2.4
<whitequark[cis]>
any idea which SiliconBlue family was called "Los Angeles"?
<Wanda[cis]>
I think just the base ice40 one
<Wanda[cis]>
LP/HX
<Wanda[cis]>
like, AFAICT it's just the codename for ice40