|
https://www.crowdsupply.com/sifive/hifive1 First RISC-V is silicon about to come out Admittedly it's only a not-very-impressive microcontroller at the moment (no on-chip flash because patents or some other such issue with the hard IP) but it's still an important milestone.
|
# ? Nov 30, 2016 16:05 |
|
|
# ? Jun 5, 2024 06:50 |
|
i thought risc-v was already in silicon in academia?
|
# ? Nov 30, 2016 19:23 |
|
also risc-v sucks b/c chisel is a trashfire
|
# ? Nov 30, 2016 19:23 |
|
lol the similar project links https://www.crowdsupply.com/eoma68/micro-desktop hello, old friend
|
# ? Nov 30, 2016 19:45 |
|
Bloody posted:also risc-v sucks b/c chisel is a trashfire as opposed to all those other fantastic hdls such as
|
# ? Nov 30, 2016 20:29 |
|
Sapozhnik posted:as opposed to all those other fantastic hdls such as yes, that is an exhaustive list
|
# ? Nov 30, 2016 20:30 |
|
Bloody posted:yes, that is an exhaustive list
|
# ? Nov 30, 2016 20:38 |
|
amazon has instances with fpga now
|
# ? Nov 30, 2016 22:57 |
|
Corla Plankun posted:amazon has instances with fpga now now you can run 68000 code in the cloud!
|
# ? Dec 1, 2016 00:48 |
|
Corla Plankun posted:amazon has instances with fpga now Corla Plankun posted:amazon has instances with fpga now so i had to look this up and didn't see an immediate answer are these vms multitenant? can pci-e even support multitenant securely where one of the tenants has complete control of a device on the bus? you can literally make malicious devices to try to escape the hardware protections of course you'd have to get your design to synthesize in the first place so we have that barrier of time
|
# ? Dec 1, 2016 14:33 |
|
Storysmith posted:
they're almost certainly not multitenant; lots of 'em just use the hypervisor to kick you out when you're done or if you run out of money
|
# ? Dec 1, 2016 14:44 |
|
I thought embedded development would be covered here but im now dealing with stuff like "AWT IoT Node.js SDK for Embedded Devices" I guess im in webdev now
|
# ? Dec 19, 2016 16:15 |
|
fun fact: you do not need a hardware fpu for javascript it's exactly as slow as you'd think
|
# ? Dec 19, 2016 16:21 |
|
last time i saw a javascript on the embedded thing it had a significant performance sensitivity to whitespace and comments in the source
|
# ? Dec 19, 2016 20:19 |
|
that's espruino it's almost as fast as ENIAC, but not quite. source of much amusement at work when it we learned of it.
|
# ? Dec 19, 2016 21:04 |
|
Bloody posted:last time i saw a javascript on the embedded thing it had a significant performance sensitivity to whitespace and comments in the source this is not a problem unique to embedded javascript
|
# ? Dec 19, 2016 21:05 |
|
hobbesmaster posted:fun fact: you do not need a hardware fpu for javascript but... but... doesn't js store all numbers as floats?
|
# ? Dec 19, 2016 21:08 |
|
you can do float ops just fine with integers. it justs takes thousands of cycles
|
# ? Dec 19, 2016 21:10 |
|
it's something the newbie programmers do all the time at work. write some formula using doubles then compile it for a Cortex M0 which not only has no FPU, but no divider and no count-leading-zero instruction. this kills the dog
|
# ? Dec 19, 2016 21:12 |
|
i actually had to spec out a dog wearable and they were insistent on it being a M4 or better
|
# ? Dec 19, 2016 21:16 |
|
JawnV6 posted:but... but... doesn't js store all numbers as floats? yes the poor arm926ej doesn't know what hit it
|
# ? Dec 19, 2016 21:20 |
|
http://www.trustedreviews.com/news/jawbone-up24-firmware-update-boosts-battery-life-to-14-days Software-implemented floating point is slow and bad. That firmware update switched from floating point to fixed point math (on an MSP430 without an FPU) and doubled the device's battery life.
|
# ? Dec 19, 2016 22:17 |
|
msp430s are insanely anemic. do they make any without the hw multiplier? cause if so that's gotta be the posest pos there is
|
# ? Dec 19, 2016 23:59 |
|
i will never understand what the "msp432" line is trying to appeal to
|
# ? Dec 20, 2016 00:03 |
|
aren't they just yet another cortex m4?
|
# ? Dec 20, 2016 00:06 |
|
looks like a m4 yeah probably marketing execs convinced they'll retain more msp430 customers who want to migrate to an arm by giving them the warm fuzzy feeling of a familiar product name
|
# ? Dec 20, 2016 00:31 |
|
Bloody posted:msp430s are insanely anemic. do they make any without the hw multiplier? cause if so that's gotta be the posest pos there is FRAM seems cool though? don't they use practically no power? (then again so do M0s)
|
# ? Dec 20, 2016 01:26 |
|
lol where did those crazy As come from yeah FRAM is cool but also the one time I used an fram part it was an absolute piece of poo poo so lol
|
# ? Dec 20, 2016 01:42 |
|
quote:before you slam it, fram it
|
# ? Dec 20, 2016 01:53 |
|
I am not a framer
|
# ? Dec 20, 2016 02:31 |
|
we're just testing out a fram drop in replacement for some eeprom chip and it's going well. blowing the minds of the guys responsible for the amazing page write implementation that updates a byte on a page, commits it, updates the next byte, commits that, etc. guess why we need an fram part?
|
# ? Dec 20, 2016 04:15 |
|
Bloody posted:msp430s are insanely anemic. do they make any without the hw multiplier? cause if so that's gotta be the posest pos there is the junior dev was working on something and tried to slow it down, like sending every 5th sample or something, but it ended up going extremely slow. root caused to his test loop, he'd written it as code:
BobHoward posted:msp430 customers who want to migrate to an arm by giving them the warm fuzzy feeling of a familiar product name
|
# ? Dec 21, 2016 00:16 |
|
can you believe we used to do floating point arithmetic on a 6502 running at 1.023 MHz?
|
# ? Dec 21, 2016 00:41 |
|
it always seems crazy that those old computers didn't always have an fpu coprocessor installed JawnV6 posted:yeah, i used a MSP430G2553, no hw multiplier do you use fizzbuzz in the interview process?
|
# ? Dec 21, 2016 00:56 |
|
hobbesmaster posted:do you use fizzbuzz in the interview process? JawnV6 fucked around with this message at 05:49 on Jan 6, 2017 |
# ? Dec 21, 2016 00:58 |
|
JawnV6 posted:root caused to his test loop, he'd written it as i remember a couple of years ago trying to implement a uint16->decimal conversion function for displaying a value on an LCD on a PIC16 iirc for 10-bit ADC values it was around 10x faster to just subtract 1000s, 100s etc. down to 0 than doing even a single modulo operation
|
# ? Dec 21, 2016 01:11 |
|
hobbesmaster posted:it always seems crazy that those old computers didn't always have an fpu coprocessor installed it's like how nowadays you expect every car to have air conditioning, even cheap models in the 80s it was a luxury
|
# ? Dec 21, 2016 02:58 |
|
Are there any EU individual bit janitors in here who would be willing to buy something from exp-tech.de and ship it to the United States? They are the only vendor I can find who has the Mojo FPGA servo shield in stock, but they won't take my drat money. Well, them and someone out of India who wants to charge far more than the part costs in shipping. Alternatively, if someone knows of a new Mojo servo shield (I'm not too excited about buying used) that can find its own way to the US without manual intervention, that would be cool, too. I'm happy to cover all purchase and shipping costs, obviously. PM me to set stuff up.
|
# ? Dec 28, 2016 17:46 |
|
Hey Harik, we found a cheaper chip for our embedded linux, it's the N32905. You don't have a problem making our GUI app with networking portions work on a 240mhz processor with 32MB RAM, right? Look, it even comes with a linux 2.6.35 BSP ready to go. But it is used in <list of cheap chinese poo poo>! We should put it in our $600 product! Just loving kill me. I would have targeted it at an RTOS if they wanted something that size to begin with, but the initial spec was for a tablet-class MCU. Now it's a library-heavy process expecting ~800-1ghz core and ~256mb ram. I used libraries because, hey, guess what, it's just me on this project and I'm not going to write the entire functionality solo. At least I convinced them to go into production with the current part instead of delaying everything 6 months while I redo it from the ground up. If they didn't go that path I would have simply resigned and let the next sucker clean up the mess. Now to find out how many libraries are expecting a modern kernel... It's not too awful, I'm only using 15MB of RAM and I hadn't ever looked into optimizing that since it was way under budget for a quarter gb part. 160 meg virt, but hopefully XIP will work for that. The much slower CPU is painful, since it's doing anti-aliased sprite scaling and fractional pixel placement without a GPU - I'll absolutely have to rewrite that to get more than 5-10 FPS on the UI. Venting aside, any tips on fast sprite work that doesn't look like complete poo poo? It looks good now, I don't think they'll accept a regression. Already done some of the lowest-hanging optimizations, like detecting nearly-aligned draw and turning it into memcpy(), pre-calculating alpha and supersampling fields etc, but I don't do tons of low-level graphics so I've probably missed something obvious.
|
# ? Jan 4, 2017 03:26 |
|
|
# ? Jun 5, 2024 06:50 |
|
at that point can't you get m4s that'll be more powerful?
|
# ? Jan 4, 2017 04:09 |