clifford changed the topic of #yosys to: Yosys Open SYnthesis Suite: http://www.clifford.at/yosys/ -- Channel Logs: https://irclog.whitequark.org/yosys
dys has quit [Ping timeout: 265 seconds]
X-Scale has quit [Ping timeout: 265 seconds]
X-Scale has joined #yosys
kraiskil has quit [Ping timeout: 240 seconds]
X-Scale` has joined #yosys
X-Scale has quit [Ping timeout: 246 seconds]
X-Scale` is now known as X-Scale
dnotq has quit [Remote host closed the connection]
rohitksingh has joined #yosys
nrossi has joined #yosys
voxadam has quit [Quit: WeeChat 2.4]
voxadam has joined #yosys
citypw has joined #yosys
dys has joined #yosys
flaviusb has joined #yosys
lukego has joined #yosys
<lukego> Is there any Emacs mode that vaguely suits rtlil?
<ZirconiumX> Oh, hey lukego
<lukego> G'day ZirconiumX :)
<ZirconiumX> RTLIL is not really intended for human editing
<lukego> I'm mostly intending to read at the moment. I think a little syntax colouring would help to make the \src attributes less distracting for example.
<ZirconiumX> I mean, you can always write one :P
<lukego> Sure just curious first what other people are doing e.g. maybe some generic commands can do the job too. Just now I'm using 'grep -v' but that's clunky :)
kraiskil has joined #yosys
<lukego> Seems like the yosys package for nixpkgs doesn't pull in the runtime dependencies needed for 'yosys show'. I wonder if it should.
<ZirconiumX> cc emily, maybe?
<lukego> Good entry point for me into Yosys, checking for how it finds those commands.
<lukego> via PATH by the look of it. So I guess the nix package would need a wrapper script.
ZipCPU has quit [Excess Flood]
kraiskil has quit [Ping timeout: 240 seconds]
ZipCPU has joined #yosys
rohitksingh has quit [Remote host closed the connection]
kraiskil has joined #yosys
rohitksingh has joined #yosys
dys has quit [Ping timeout: 240 seconds]
jakobwenzel has quit [Quit: jakobwenzel]
jakobwenzel has joined #yosys
kraiskil has quit [Ping timeout: 265 seconds]
kraiskil has joined #yosys
rohitksingh has quit [Ping timeout: 245 seconds]
kraiskil has quit [Ping timeout: 240 seconds]
emeb has joined #yosys
<whitequark> daveshah: btw, have you ever considered porting some ecp5-only features of nextpnr back to ice40?
<daveshah> I think it's only out-of-context that is ecp5 specific
<daveshah> I will when I get round to it
<daveshah> (and I wanted to deal with a few other related things first)
<whitequark> oh alright, I somehow misremembered that there's more
<whitequark> related: what about making HeAP the default on iCE40?
<daveshah> Also on the TODO list (need to fix an edge case of a design with no regular IO, just oscillator and RGB primitives, causing a singular matrix)
<whitequark> gotcha, thanks!
<daveshah> I better get the iCE40 stuff into shape before the rumoured iCE28 or whatever is announced next month...
<whitequark> the what
<whitequark> i thought lattice gave up on that series
<daveshah> Oh it is only the spiritual successor
<daveshah> I don't think it will actually be called iCE28
<daveshah> But it will be a similar low power/feature set to the UltraPlus
<ZirconiumX> I just hope their product naming scheme is better than their primitive naming scheme
<daveshah> Primitive naming will doubtless be similar to ECP5
<whitequark> what about the CLBs?
<daveshah> At a guess roughly ECP5-like, although I've heard some possible further experimentation has gone on
<daveshah> The only thing they've officially said is still LUT4 based
<whitequark> ah hm
<whitequark> so they basically put the ice40 logic into trash?
<daveshah> Yeah definitely
dh73 has joined #yosys
<daveshah> By the UltraPlus you could tell they were really hacking at the designs with no real understanding of them
<daveshah> All the SiliconBlue people left soon after the merger, afaik
<daveshah> Indeed Radiant tries to make the iCE40 logic look more like ECP5, presumably to fit the Lattice CAD architecture (e.g. rotating the chip through 90degrees)
<ZirconiumX> So we're expecting this to be essentially a low-power ECP5-like?
<daveshah> Yes
<daveshah> I believe with iCE40UP-style large RAM blocks too (not sure if still single-ported)
citypw has quit [Ping timeout: 245 seconds]
<ZirconiumX> That's...actually interesting to look at
<ZirconiumX> "100% Focus on FPGA"
<ZirconiumX> But isn't MachXO3 a CPLD?
<daveshah> No, it's an FPGA
<daveshah> Anything with LUTs is an FPGA, imo
<ZirconiumX> Right, okay
<whitequark> oh, so you would call LUT-based arches "FPGA" and SOP-based "CPLD"? that's one way to do it
<whitequark> re SB people leaving: interesting. was their HDL/layout so impenetrable that the Lattice people couldn't figure it out?
<whitequark> or is it just hard?
<daveshah> My guess is that something like an FPGA is always going to be pretty nasty, because a lot is manually laid out rather than HDL
<daveshah> It was also doubtless rushed given the constraints on SiliconBlue
<daveshah> In fact I think SB would have gone bankrupt were it not for Lattice
<daveshah> To get you an idea of the bodgniness - the DSPs are interconnected through a hacked logic tile
<whitequark> I recall some of that I think
<daveshah> Said logic tile still has all the logic intact - but the tools always setting the LUT as a route through from cascade in to out
<daveshah> The DSP then taps of the LUT inputs and feeds its output into the logic tile cascade path
<daveshah> Where the LUT then feeds it back to fabric
<daveshah> If they knew what they were doing then they definitely wouldn't have left the redundant LUT in place adding delay but just kept the interconnect part of the tile
<whitequark> wait what
<whitequark> the ... LUTs are there
<whitequark> lmao
<daveshah> Yes
<whitequark> what the fuck
<whitequark> that's amazing
<daveshah> I think the FFs might be there too but I've never tested
<whitequark> this is webdev-level quality
<daveshah> The lm4k goes a different approach and connects its hard IP via complete IO tiles
<daveshah> Almost as if they laid two chips out next to each other and added some wires between the two
<whitequark> um
<mwk> they what
<daveshah> If they weren't in wafer level packaging I'd be putting them in an xray
<daveshah> But it must be one die, just laid out as two
<whitequark> can someone send me some so i can take die shots
<mwk> can... can you actually usefully program the LUT in any way?
<whitequark> at least they didn't use the logic tile FF as DSP output register
<daveshah> I think you could use the LUT just as a LUT if you were running low
<daveshah> Just without the cascade function
<whitequark> would be fun to make nextpnr do that
<daveshah> Yeah, I might have a play with that
<whitequark> hm, can't buy lm4k locally :/
<ZirconiumX> (could somebody expand lm4k for me?)
<mwk> ZirconiumX: one of the ice40 parts
<daveshah> lm1k and lm2k are the same dice btw
<daveshah> as lm4k
<whitequark> lp4k is a different device right
<ZirconiumX> Ah, I see
<daveshah> lp4k is rebranded lp8k
<daveshah> One of the classic parts
<whitequark> um
<whitequark> ice5lp4k?
<daveshah> Oh that's the iCE40 ultra
<whitequark> NOT CONFUSING AT ALL
<daveshah> idek where the 5 comes from
<daveshah> I don't think it's 5nm
<daveshah> And then they went back to 40 for ultralite and ultraplus
<whitequark> ultralite is um. lm4k?
<daveshah> No, ice40ul1k
<whitequark> what the hell
<whitequark> how many of these models do they -have-
<daveshah> lm4k is a really weird somewhat abandoned series between the lp/hx and the various ultra parts
<daveshah> It is not surprising its hard to get hold of
<daveshah> The official dev board is also ridiculously expensive for some reason
<whitequark> when did they rebrand UP as "low power AI"
<daveshah> It's their whole new strategy
<daveshah> "edge AI"
<daveshah> They've got some model compiler tool now too, sensAI
<whitequark> also what's up with having both diamond and radiant
<daveshah> Radiant is an attempt at a fresh start
<whitequark> ah, without neocad stuff
<whitequark> but... it can only do UP?
<daveshah> Oh no still neocad inside
<daveshah> At least to some extent
<whitequark> oh.
<daveshah> Just a new GUI and some supposed pnr improvements
<daveshah> Also some kind of IP integrator that doesn't quite work yet
<daveshah> Radiant will be the CAD tool for new archs going forward
<daveshah> The up5k is really just a proof of concept for it
<whitequark> oic
<ZirconiumX> Is Radiant a front-end to the old Diamond toolchain, or from-scratch?
<daveshah> Half and half
<daveshah> They've supposedly improved things a bit and changed the design database format
<daveshah> But there's still plenty of NeoCAD in thete
<daveshah> Unlike Vivado which was a near total from scratch job afaik
<whitequark> "1000 person years"
<dh73> I know two folks that worked at Xilinx when they started to develop Vivado. They said that, a lot of departments got really crazy deadlines, and lots of engineering burnout
dnotq has joined #yosys
kraiskil has joined #yosys
dys has joined #yosys
<daveshah> Oh another funny Radiant thing is that it is at least trying to replace some Tcl with Python (in particular the core generators) - but in true EDA style, Python 2.7
<daveshah> Yay for building using a dependency that will be obsolete by the time your tool is actually properly launched
<whitequark> lol
<whitequark> I'm actually not sure if I dislike Tcl that much
<whitequark> in particular, it's very lightweight, unlike Python
<whitequark> I would gladly use Tcl in nextpnr, because right now I assume I can't rely on any scripting capability at all
<daveshah> Yeah, Python makes things for stuff like describing FPGAs or more complex prototyping
<daveshah> It's a shame linking it is such a nuisance
<whitequark> do you think there's any chance nextpnr could gain tcl bindings?
<whitequark> I might volunteer to provide the implementation there
<whitequark> after some in-depth Vivado work I grew to appreciate its Tcl capabilities
<daveshah> I'd really rather not have to maintain two sets of bindings
<daveshah> Another possibility is micropython
<dnotq> Did you actually like Tcl though, or is the appreciation because it allowed you to do something you otherwise could not?
<daveshah> Although it wouldn't work out of the box with Boost because it doesn't implement the CPython API
<dnotq> Was Lua considered? It is designed to be embedded, is fast and lightweight. At work I have to maintain a code-base that binds Python and it is a PITA.
<ZirconiumX> The Lua C API is mildly terrifying
<ZirconiumX> setjmp/longjmp error handling
nrossi has quit [Quit: Connection closed for inactivity]
rektide_ has quit [Ping timeout: 250 seconds]
rektide has joined #yosys
dnotq has quit [Remote host closed the connection]
dnotq has joined #yosys
rohitksingh has joined #yosys
cr1901_modern has quit [Ping timeout: 240 seconds]
rohitksingh has quit [Ping timeout: 250 seconds]
rohitksingh has joined #yosys
cr1901_modern has joined #yosys
kraiskil has quit [Ping timeout: 265 seconds]
kraiskil has joined #yosys
tpb has quit [Remote host closed the connection]
tpb has joined #yosys