<whitequark>
cr1901_modern: you could even use something like Souper
<whitequark>
i.e. use an SMT solver to automatically find *the* optimal sequence of instructions
<cr1901_modern>
Oh crap, that sounds awesome. And I see John Regehr's name on it!
<fseidel>
sorear: so A53 can dual issue 2 dependent adds?
<sorear>
yes
<sorear>
because it's designed to operate at clock cycles longer than 2 back-to-back simple ALU operations
<sorear>
so it just forwards the result sideways in the same pipe stage
<cr1901_modern>
Stupid idle thought: I wonder if that saves energy as well compared to computing the result twice, or if the detection circuitry for that offsets the power savings.
<implr>
obviously the mill is the future
<implr>
[mild irony]
<whitequark>
lol the mill
<fseidel>
neat
<cr1901_modern>
isn't that vaporware?
<fseidel>
oh yeah, I forgot mill was a thing, it's been coming soon for the last 10-15 years, right?
Miyu has quit [Ping timeout: 240 seconds]
<rqou>
heh, latest nurdrage is kinda neat
<rqou>
only problem is that the amount of sodium he got over 3 days of reacting is only worth like $30
<implr>
cr1901_modern: yes, but they are actually doing something - I know some people who responded to their 'work for us for free (except equity)' offer in the past ~year
<implr>
and they were working on a llvm port and similar stuff
<sorear>
have any of the mill patents expired yet
<implr>
I give mill less that 1% chance of becoming a major arch, but I'd still love to see them on the market, they had many interesting ideas
<sorear>
not that i'd want to use them
<implr>
sorear: just checked their page, it seems they haven't even filed them all yet
<implr>
>As of May 8, 2018, 14 of our patents have been granted.
<sorear>
so you realize intel is still filing patents?
<sorear>
"still filing patents" is not interesting for a non-dead arch
<q3k>
'filing patents' is not interesting
<implr>
meh, I phrased that badly
<implr>
I just meant to say that there's no way they're anywhere close to expiring if they just started filing recently
<implr>
I just want to see an itanium 2.0, even if it fails spectacularly it'll still be amusing
<whitequark>
if they're filing patents that makes it far less interesting
<sorear>
it's cute that they think they'll ever have the revenue to protect, yes
<fseidel>
any of you guys have something negative to say about the XC9500XL family, or is it a good choice?
<rqou>
PLAs > PALs
<whitequark>
why even bother with a CPLD?
<whitequark>
5V tolerance?
<rqou>
if you're a weirdo like my father and need the fast deterministic timing?
<cr1901_modern>
5v tolerance, and many families require external flash. Other than that, idk
<cr1901_modern>
Oh and "no need to wait to boot" :P?
<rqou>
coolrunner-ii has a boot time :P
<whitequark>
there are FPGAs with embedded flash
<whitequark>
i think uhh actel?
<cr1901_modern>
Right, mach is one of them
<fseidel>
I need 5V tolerance
<whitequark>
that makes sense
<rqou>
i don't think xc9500xl is 5v tolerant?
<cr1901_modern>
it is
<fseidel>
looks like it is from the datasheet
<sorear>
mach has a boot time too
<rqou>
hmm ok
<cr1901_modern>
right b/c it's an fpga.
<sorear>
but it's a load-a-row-at-a-time boot, not serial
<sorear>
few µs
<cr1901_modern>
I would prob use mach and eat the cost of voltage shifters (NON autosensing ones)
<fseidel>
why is that?
<rqou>
i'd use maxv :P
<fseidel>
I'm new to board design (this is my first PCB ever), so I was trying to reduce component count
<rqou>
maxv needs basically no external parts
<cr1901_modern>
fseidel: Oh, nothing more than personal preference
<whitequark>
i'd use ice40 and voltage shifters :P
<whitequark>
mostly because of toolchain availability
<whitequark>
autosensing voltage shifters do make sense... if you understand their issues
<fseidel>
I'm far more used to Altera's toolchain, so MAX V is attractive, but I didn't think it was 5V tolerant
<rqou>
well yeah, i need to get kinglerpar more working
<rqou>
you add series resistors or whatever
<fseidel>
I have an easy signal I can tap to use a non-autosensing voltage shifter
<whitequark>
rqou: or just use nextpnr
<whitequark>
oh right, kingler is analytical
<cr1901_modern>
whitequark: I thought your issue w/ autosense was "it's too easy to drive them in the opposite direction than intended"
<whitequark>
cr1901_modern: no
<whitequark>
well
<whitequark>
they're obviously the wrong choice for glasgow, but i had no choice for the non-bga revisions
<whitequark>
the ones with pullups are useless for anything that isn't i2c
<whitequark>
the ones without pullups have signal integrity issues and can't tolerate pullups, so you don't get i2c
<whitequark>
or can't use them to program a target that has e.g. pullups on CS pins or whatever
<whitequark>
or reset
<cr1901_modern>
and seeing that "fun" you had a few months back made me think "unless I'm using I2C, I'll eat the dir pin"
<whitequark>
fxmas can't be used for i2c
<cr1901_modern>
just a general comment about when I'd use them :)
<whitequark>
i'd use them in a case where i designed the entire board
<whitequark>
and have a bidi bus
<whitequark>
which is pretty rare, really
<fseidel>
anyway, I need to go catch a flight
<fseidel>
thanks for the help!
<cr1901_modern>
don't be a stranger :P
unixb0y has quit [Ping timeout: 240 seconds]
unixb0y has joined ##openfpga
* cr1901_modern
goes looking for his missing pair of socks that disappeared between starting this chat and now Always something exciting that happens when I actually chat in this room lol
<azonenberg_work>
fseidel, cr1901_modern: greenpak can run at VDD=5V
<azonenberg_work>
as long as yo dont need much logic :p
<balrog>
azonenberg_work: how much logic does it have that handles 5 volts?
<azonenberg_work>
balrog: the slg46620 has i think ten counters, a dozen ffs, two shift regs, 26 LUTs
<azonenberg_work>
It's 180nm TSMC
<cr1901_modern>
You can get up to 16 ffs if you're creative w/ shift regs
<azonenberg_work>
each shreg is 16 ffs
<azonenberg_work>
With two taps at bitstream-programmable (not runtime adjustable sadly) offsets, and an optional inverter on one of the taps
<cr1901_modern>
right, you use the two taps of each shreg, and you can get 4 ffs... _if_ you're creative :3
<cr1901_modern>
sadly not enough logic to make a UART :(
<cr1901_modern>
well in both dirs
<azonenberg_work>
eh
<azonenberg_work>
i have managed to make some degree of a uart
<azonenberg_work>
sending a fixed byte at least
<cr1901_modern>
I wanted both in one chip :P. Certainly I imagine sending is easy enough. Recving might be tougher
<rqou>
a uart with tx+rx plus some basic logic takes up ~95% of an xc2c32a
<fseidel>
I need a lot more 5V I/O than that
<rqou>
at least when i coded it
<azonenberg_work>
cr1901_modern: well pin count is an issue
<azonenberg_work>
i dont even think there are enough GPIOs
<fseidel>
need 32-bit address and data busses for 68030 interface, then 16-bit data and 24-bit address for connecting up to the original 68K bus
<cr1901_modern>
And a 64 channel logic analyzer :) (must be nice to have access to equipment :P)
<fseidel>
I've got a 136 channel logic analyzer on my desk that I got for $50 :-)
<fseidel>
HP 1660CS
<fseidel>
CRT needs some adjustment but it's a lovely scope
* cr1901_modern
doesn't have a scope. No room, no money for a good one
<fseidel>
I've heard that old HP equipment is generally the best way to get good stuff for cheap, but the issue is that they use CRTs and take up a lot of space
<cr1901_modern>
I'm sure if I scrounged around on CL or fleabay I could find something...
<balrog>
which is funny, because avr-gcc and avr-libc are mostly community work.
_whitelogger has joined ##openfpga
<qu1j0t3>
fseidel: +1
<awygle>
balrog: that's total garbage. Avr is in upstream gcc tho, and in llvm.
<awygle>
Err I think it's in gcc... I could be wrong.
azonenberg_work has quit [Ping timeout: 276 seconds]
azonenberg_work has joined ##openfpga
rohitksingh has joined ##openfpga
<awygle>
There's a fiber drop running into the network panel in my apartment, then into a silver box with an F-type RF output, which runs into my cable modem, which puts out RJ-45 to my router... Which has a fiber input port on it.
<awygle>
I sincerely doubt that just plugging the fiber in directly would work, but it *feels* like it should
<awygle>
Also it's cool my apartment has a somewhat legit network panel
<azonenberg_work>
awygle: it's probably GPON
<azonenberg_work>
not something sane like base-X
<azonenberg_work>
Is there a model number on the silver box?
<awygle>
CP8015U-02-00 Rev. D.2
<awygle>
"SDU RFoG CPE"
<awygle>
Oh wow RFoG is crazy
<awygle>
Or perhaps the opposite of crazy, it seems extremely brute force. "what if docsis, but over fiber"
rohitksingh has quit [Quit: Leaving.]
<azonenberg_work>
Lol
<azonenberg_work>
awygle: never understimate the amount of money and effort people will spend to allow them to continue using legacy gear
<azonenberg_work>
or legacy cable plants (docsis exists for this sole reason)
rohitksingh has joined ##openfpga
rohitksingh has quit [Quit: Leaving.]
rohitksingh has joined ##openfpga
knielsen has quit [Remote host closed the connection]
knielsen has joined ##openfpga
rohitksingh has quit [Quit: Leaving.]
<Prf_Jakob>
fseidel: I don't think you need to put the FPGA between the MC68030 and the MC68k bus.