Lofty changed the topic of #prjmistral to: Project Mistral: Yosys (and hopefully nextpnr) on Cyclone FPGAs - https://github.com/ZirconiumX/mistral - logs: https://freenode.irclog.whitequark.org/prjmistral
Lofty has quit [Remote host closed the connection]
ZirconiumX has joined #prjmistral
ZirconiumX is now known as Lofty
<chipb> Lofty: I'm pretty far removed from Quartus development, let alone nitty-gritty device support details. that said, I think it'd be pretty close to "free" for some definition thereof.
<chipb> I doubt they would change it significantly if they could get away with it.
<chipb> less engineering resources needed to get it validated and out the door that way.
<Lofty> Mhm
<chipb> (hardly a surprise, I'm sure)
rombik_su has joined #prjmistral
peepsalot has quit [Ping timeout: 256 seconds]
peepsalot has joined #prjmistral
_whitelogger has joined #prjmistral
<Lofty> Morning (ish)
<daveshah> Afternoon :p
<Lofty> I woke up ten minutes ago, I'm allowed to call it morning
<Lofty> :P
<Lofty> rombik_su: how are the database dumps going?
<rombik_su> Lofty: In progress, need to dump SoCs and I will be done.
<Lofty> rombik_su: mind sending over what you've got?
<rombik_su> Sure, let me zip it. Do you want only .dmp, right?
<Lofty> Yep
<rombik_su> It's taking too much time for some reason
<rombik_su> Oh, maybe I shouldn't be running Quartus doing dumps in the background
<Lofty> Heh :P
<rombik_su> Maybe I should stop with buying FPGA e-waste from Ebay and invest in 1TB NVMe )
<rombik_su> Lofty: It's ridiculously long run. I'm at 13G now. I can rerun gzip with -9 and squeeze it down to 10G so I can upload it to rombik.su
peeps[zen] has joined #prjmistral
<Lofty> How about using pigz instead? :p
<rombik_su> I'll try it
peepsalot has quit [Ping timeout: 240 seconds]
<rombik_su> Lofty: It can't compress a directory, do you mind if I compress single files recursively and then tar it?
<Lofty> Tar it, then run pigz on the tar file
<Lofty> (gzip can't compress a directory either)
<Sarayan> Meh, use 7z :-)
<Sarayan> extremely efficient on the asm files
<Lofty> Sarayan: yeah, no.
<Lofty> We're optimising compression time here, not quality :P
<rombik_su> Lofty: Sure, but tar can invoke gzip. I don't think I have space for 250G tar file )
<Sarayan> 7z compresses the asm files to 2% of the original size
<Sarayan> and does not need an intermediate file
<Lofty> Sarayan: it's also single thread
<Sarayan> -mmt[N] : set number of CPU threads
<Sarayan> ah>
<Sarayan> ?
<Sarayan> (never tried that option though)
<Lofty> rombik_su: tar -l
<Lofty> (lowercase L)
<Lofty> ... I can't actually tell
<Lofty> ... it's -I
<Lofty> I hate the Latin alphabet so much.
<Lofty> Anyway
<Sarayan> it's all greek to you?
<Lofty> Υεσ, ιτ'σ αλλ Ηρεεκ το με.
<rombik_su> Yup, I was just looking at this )
<Lofty> The alternative is waiting forever for gzip
<rombik_su> It's already 3+ hrs since my 'Sure, let me zip it' message :p
<Lofty> :P
<rombik_su> Alright, it's doing something
peepsalot has joined #prjmistral
peeps[zen] has quit [Ping timeout: 260 seconds]
<rombik_su> Lofty: It's still >10G with pigz -9. I can split it to a 6G chunks and put it one-by-one to my server with md5/sha1
<Lofty> I think deflate just can't compress it well enough :P
<rombik_su> *one eternity later* Lofty, you can start with rombik.su/dmp_archive.tar.pigz_a.md5sum and rombik.su/dmp_archive.tar.pigz_aa
<Sarayan> how big is it?
<Sarayan> oh, 2G
<Sarayan> exactly 2G? It's cut in parts?
<rombik_su> Sarayan: Yes, it's a first chunk of 12G
<rombik_su> Cut with `split` command
<Sarayan> 12G? damn
* Sarayan stops the download
<Sarayan> I'll generate them myself if I need them :-)
<Sarayan> how big was the original?
* Sarayan is doing a random walk of the drivers
<Sarayan> gah, wrong channel
<rombik_su> Originally it was around 250G
<Sarayan> that's not a very good compression ratio
<Sarayan> should be 2-3G
<rombik_su> There's about 10% difference between default gzip compression and -9 in this particular case
<awygle> can somebody remind me the lineage of the two Cyclone 10 lines?
<awygle> as in what older chips are they clear descendents of
<rombik_su> Looks like C3 -> C4E -> C10LP
<rombik_su> C10GX apparently is made of A10GX dies, which in turn apparently based on one of the Stratix GX
<awygle> C10GX and A10GX are the same?
<rombik_su> C10GX and C10GX allegedly sharing same die, and C10GX is probably binned and have some of the resources locked out
<rombik_su> *and A10GX
<Lofty> I'm awake now
<Lofty> Sarayan: I can attempt to recompress it all if you really want.
<Lofty> Generating the routing is... probably not the best
<rombik_su> Lofty: Next one is http://rombik.su/dmp_archive.tar.pigz_ab
<rombik_su> Lofty: Any luck with dmp_archive.tar.pigz_aa?
<rombik_su> Can I delete it?
<Lofty> Literally just finished downloading, let me just verify it
<Lofty> rombik_su: got AA and it verifies; you can delete it
<rombik_su> Got it
<Lofty> rombik_su: got AB and verified it too
<Lofty> That can be deleted as well
<rombik_su> Lofty: http://rombik.su/dmp_archive.tar.pigz_ac is ready
<Lofty> Way ahead of you there :P
<rombik_su> Probably FTP upload for this hosting company is a low-priority, I'm getting solid 900KB/s
<Lofty> Ouch :P
<Lofty> rombik_su: AC received and verified
<rombik_su> Hmm, we should also try different PVT corners, since Quartuss will dump only selected
<rombik_su> *only selected corners
<Lofty> Mhm, but this is hopefully enough for now
<rombik_su> Still uploading AD, ETA 20m
<Sarayan> corners, that's temperatures?
<rombik_su> Process (fast/slow), voltage and temperature
<Sarayan> ok, syn will give you high temp, I think, fit will give you high, low and something that's either mediam or minimum or maximum, I'll have to check. asm won't give you anything more
<Lofty> rombik_su: AD received and verified
<Sarayan> so yo udon't need to redo asm and the gigantic dump that comes with it
<Sarayan> only map (as always when you change of model) and fit
<rombik_su> Sarayan: I've picked up random devices to dump with, some of them set 'extended temp range' (-40/+80C), some of them were 'commercial' (0/+50?C), so for different dies I got different PVT database dumps
<rombik_su> *some of them set in Quartus Device dialog ...
<rombik_su> B7 die is very stubborn, it sits in Assembler for 1.5 hrs now
<Lofty> That means it's pretty big :P
<Sarayan> windows or linux?
<Lofty> Linux IIRC
<rombik_su> Yup, Linux
<Sarayan> so you can see a file while it's being written
<Sarayan> just after the start of DYGR_ROUTE_ASM_INFO_BODY you have a hex number on the ptr.m_num_asm_nodes, that's the number of routing nodes
<Sarayan> then if you do tail on that file, you'll see the current number of ptr.m_route_asm_nodes[nnnn] it's on
<Sarayan> you'll know how far from the end you are then
<rombik_su> What's DYGR_ROUTE_ASM_INFO_BODY?
<Sarayan> the type of the structure that was streamed and which stores what bits to set for each connection
<rombik_su> Alright, it's done, almost 2hrs
<Sarayan> fwiw the biggest cyclone v, the gt300f, has 2847247 nodes giving a 88G file
<rombik_su> Gosh. Looks like B7 is the biggest one, 190K ALM.
<Lofty> How many nodes? :P
<Sarayan> how many alms is a lab?
<Lofty> 10.
<Sarayan> gt300 has 114K ALMs then
<Lofty> Which is 285K LEs.
<Lofty> Close enough to 300, I guess
<Sarayan> heh
<Lofty> (* 2.5)
<Sarayan> their numbers are 25 30 50 75 120 150 and 300
<Sarayan> oh, there's a 90 too
<Lofty> Well, models, but the dies are presumably smaller :P
<Lofty> Well, fewer, not smaller
<rombik_su> Lofty: http://rombik.su/dmp_archive.tar.pigz_ae is ready
<Lofty> Downloading
<Sarayan> there are 7 CV dies, e50, gt(x)25/75/250/300, sx50/120
<Lofty> Yep :P
<Sarayan> e25 is a e50 variant, se30 a sx50 variant, se90 a sx120 variant
<Lofty> Yep
<Sarayan> variant as in "pretend it has less LEs, and set one bit on the idcode"
<rombik_su> stonks /
<rombik_su> Lofty: http://rombik.su/dmp_archive.tar.pigz_af is ready
<Lofty> I have AE (sorry for the delay)
<Lofty> AF is downloading
<rombik_su> That's all, you can 'cat' them to the single file and get md5sum b8943da196edb98a0455773754e6d64a. Right now there's A3, A7 and D3 dies. I'm working on B3_ES and SOC_B5 atm.
_whitelogger has joined #prjmistral
<Lofty> rombik_su: how big is it unextracted, again? :P
<rombik_su> Lofty: ~250G
<rombik_su> Enjoy )
<Lofty> I need to free up some disk space then :P
* daveshah buys shares in storage companies
<Lofty> Ahahaha
<rombik_su> Whole Big EDA story is a Big Storage conspiracy
<daveshah> The dots are joining up...
<rombik_su> Alright, my devinfo is at 506G now
<rombik_su> I could use some computational storage right now tbh. Or any storage to that matter.
<daveshah> > The global datasphere will grow from 45 zettabytes in 2019 to 175 by 2025
<daveshah> Half of which is LLVM trees and EDA tools
<rombik_su> :D
<Lofty> daveshah: Can I tweet that
<daveshah> Sure
<Lofty> Now to wait for it to extract :P
<daveshah> You both need one of those gzip accelerator cards!
<rombik_su> Almost done with SOC_B5, one (B3_ES) to go. I think I should plug in another 1TB HDD for this.
<Lofty> Hahaha
<Lofty> daveshah: using an Arria V to accelerate dumping the Arria V databases /s
<rombik_su> At first we'll need to dump Arria I GX databases )
<rombik_su> Lofty: We can call it 'bootstrap'
<Lofty> Building an FPGA out of TTL logic to kickstart things :P
<rombik_su> That's a bit radical, let's start with FLEX and work our way up to Stratix 10
<Lofty> Okay, let's dump the Arria V DMF
<rombik_su> Lofty: delay management file?
<Lofty> I don't know what the acronym stands for, but it's the timings :P
<rombik_su> You can get it from the dumps or by some other means?
<Lofty> Via a script Sarayan wrote, I can get it from the .dmf.ddb
* Lofty resets the counter
<Lofty> Zero days since I've typoed "arriav" as "arriva" which is a bus company in the UK
<rombik_su> :D
<rombik_su> I'm gonna go, will upload rest of the goodness tomorrow (if needed).
<Lofty> Sure thing
<Lofty> Maybe the extraction will be done tomorrow too :P
rombik_su has quit [Ping timeout: 240 seconds]