✔ make one part of a RGMII MAC work by adding LUT4s to a design that do nothing
✔ make the other part of the RGMII MAC work by rearranging some of the wires to connect the exact same pins but with different wires adjacent to each other
on the picture it's 8 LUTs in a chain. but don't forget that by making it 8 LUTs I'm also getting at least a span4 worth of delay (maybe a span12 depending on the placement)
@whitequark I love how we're both sufficiently steeped in FPGA microarchitecture that "not all lut inputs are equally fast" is just something we can bring up in conversation as if it were perfectly normal.
Yosys actually models this (and at one point the inputs were reordered so that abc would pick the slightly faster path when mapping; abc assumes a particular ordering)
@whitequark haha, it's such a recognizable thing :D
do you find that it's the only thing that does what it does easily? I've tried a couple things and thus far haven't found anything else that really works
i don't know what's worse, that i understand signal integrity well enough to make this work, or that i continue to make assemblies like this despite understanding signal integrity enough to make them work
look at how incredibly crude it is! it's barely even doing anything. it's using a bunch of LUTs to delay the clocks. it's doing things with clock domains that aren't discussed in polite audience. it completely disregards the boundaries of Ethernet frames and just kind of shoves them into the other pipe. the PHY is connected to the MAC using wiring so shoddy half the time the bug was "oh, TXCTL fell out of the socket again". there's only one ground pin connected (the one paired with TXCLK)!
i wired this up and wrote half this code so high on progesterone i can only somewhat remember what i was doing and struggled with the list comprehension syntax at one point, and now Wanda is like "I have absolutely no reason to connect to WiFi, this works just fine"
@foone uh... so, what you do is you take More Progesterone until it makes you high.
however!
it might very well cause you to suddenly lose consciousness. so please do not do this in any situation where you might fall. because you could seriously hurt yourself. (i did once! six broken ribs and a scapula. though i was also on a scooter)
@ctrl@foone while i absolutely do not suggest that anyone else does it, we feel... fine? i've been doing this for quite a while, though not every day, just when i feel very bad
we have weird physiology though. like, i don't know of anybody else who took even a gram and stayed consciously
@foone@whitequark yeah, turns out it's metabolized to a lot of different neurosteroids which go on to cause a variety of effects, but a big one is allopregnanolone which, according to Wikipedia at least, is a positive allosteric modulator of the GABA-A receptor, not unlike benzodiazopine drugs!
@whitequark@foone that's interesting to hear. I've been reading about benzo receptor activity because my dad is highly recommending Passionflower as an auxiliary treatment for anxiety (he's been on it for a few months now and is tapering off his SSRI) and so I now have a bottle of it and I'm deciding whether to trial it. It contains a lot of flavinoids and alkaloids, but the main one seems to be apigenin, which binds to benzo receptors on GABA-A.
@manawyrm it feels like normal web browsing for the most part
re bufferbloat, buffer management with glasgow gets really hard due to the complex interplay between usb, fx2 bus, and python... i'm actually quite happy how it works with the default settings
@manawyrm@whitequark yeah if you get too many trans hackers together their cycles synchronize, except instead of anything biological it's just terrifying hacks
@manawyrm looking at iperf, you actually still see 1-2 ms latencies right up until the python process pegs a core, where it suddenly spikes to 100+ms. so unless you're downloading a copy of vivado it's fine
@whitequark yeah, that's classic buffer bloat behaviour and one of the biggest problems when building routers :)
when their cpu is exhausted, things also start to suck.
it's the typical "uploading a file kills my video call", which doesn't actually have to be a problem (these days).
@manawyrm glasgow actually has pushback support (not used in the snippet), unfortunately it's usually not as easy because USB effectively requires lots of huge buffers for throughput
@whitequark i don't think it's a problem which you should try to solve -- even the purpose-built USB NICs suck at this -- which is (one of the many reasons) why USB NICs suck so much in general :D
Add comment