✧✦Catherine✦✧
#GlasgowInterfaceExplorer now has a high-performance SPI analyzer! it runs at up to 100 MHz (with some caveats)
as an example, I wrote this tiny script that extracts data from SD card SPI transactions and lets me reconstruct an image file that an embedded device is reading
full code at: https://github.com/GlasgowEmbedded/glasgow/pull/817
✧✦Catherine✦✧
to implement the SPI analyzer, I decided to clock the FPGA logic from the SPI clock directly instead of using the FPGA system clock (48 MHz), which would otherwise allow sampling SPI signals at up to 24 MHz; many real-world uses of SPI are above that (e.g. this SD card was clocked at 27 MHz, ECP5 configures at 65 MHz)
the FPGA logic runs at about 120 MHz, but since nextpnr doesn't support constraining input delays, you'll most likely start having issues somewhere below that frequency
JP
@whitequark dammit catherine stop tempting me to get a glasgow
((I will. eventually. when I have beaten sufficiently many other fires out of my life. and can finally allocate headspace to this.))
✧✦Catherine✦✧
@froztbyte I mean the idea is that owning one results in *fewer* fires, not *more*
JP
@whitequark it may sound like trolling/nonsense, but it isn't: just have a fucking pyre of life-on-fire 2020+ splash damage I'm _still_ working through
re the thing you shared in screenshot just now, and also over many others prior: I am consistently deeply impressed by how fucking _clean_ the glasgow abstractions are (given subject matter). I know what it takes to get there, and that y'all do... well, *well* fucking appreciated
✧✦Catherine✦✧
@froztbyte haha, I'm *incredibly* pleased that this gets noticed, because the amount of work that goes into getting abstractions that good is immense. literally years! I'm proud of the result though
JP
@whitequark oh yeah, absolutely! for one dimension of my fires
I have multidimensional fires
(is the signal number I have for you still valid? if yes, I will share some of the fires)
✧✦Catherine✦✧
@froztbyte hell, even just getting this one API call working took me like 3 or 4 experimental designs before I found one that actually managed to work at the full throughput without overflows or wasting FIFO space! (it uses COBS as a very low-overhead packet boundary encoding for SPI transactions where you inherently don't know how long they are)
JP
@whitequark nod! I don't know how many people would (notice). it requires a particular confluence of clue/attention that I think is fairly rare? but, seriously: everything I see about glasgow: chefskiss.jpg
✧✦Catherine✦✧
@froztbyte +44-something is valid. otherwise it's catherine.8051
✧✦Catherine✦✧
@froztbyte that makes me beam with happiness ^_^
i actually just finished a major refactor that's intended to bring glasgow more in line with the developments in amaranth
here is how the SPI analyzer is tested. this is all using public stable (not yet documented but soon) interfaces only:
JP
@whitequark ah yeah I still had the old! I'll hit you up on the new :)
won't spam tonight because I'm still dealing with other yaks but soon.gif
✧✦Catherine✦✧
@froztbyte it took a very significant effort to get to the point where it is genuinely easier to write end-to-end and/or unit tests for an applet than not write them, even for a one-off
✧✦Catherine✦✧
the caveats for "up to 100 MHz" boil down to, more or less, "you need to twist every signal wire with a ground wire, and ideally put an inline 100 ohm resistor, or else your data will be complete garbage"
✧✦Catherine✦✧
story continues here https://mastodon.social/@whitequark/114527204328155137
Joel Michael
@whitequark Also not across a breadboard. Now I’m wondering about using a CAT-6 cable with DuPont connectors crimped to each conductor on both ends…
GNU social JP is a social network, courtesy of GNU social JP管理人. It runs on GNU social, version 2.0.2-dev, available under the GNU Affero General Public License.
All GNU social JP content and data are available under the Creative Commons Attribution 3.0 license.