| lkcl | rush-job, for lauri to be able to do mp3 decode, i'm adding FP support to the ISACaller simulator | 17:44 |
|---|---|---|
| lkcl | i'm going to need help, esp. with unit tests | 17:55 |
| programmerjake | just finished the WebAssembly SIMD meeting, they were quite interested in SimpleV | 18:05 |
| lkcl | nice | 18:15 |
| lkcl | that's a good sign. | 18:15 |
| programmerjake | link to meeting minutes: https://docs.google.com/document/d/16pBUlJCzXz3fHwvHUS17WwmydALqEH2RanEJrVtsYtg/edit?usp=drivesdk | 18:20 |
| programmerjake | what's written is an approximation of what we said... | 18:21 |
| programmerjake | better link: https://github.com/WebAssembly/meetings/pull/781/files | 18:33 |
| lkcl | FL: what happens when you reduce VL to data that is out of bounds in registers | 18:39 |
| lkcl | his question's slightly unclear | 18:39 |
| lkcl | if it was this: | 18:39 |
| programmerjake | I think he meant: how do you prevent it from messing up registers past what the compiler allocated | 18:40 |
| lkcl | right. yeah | 18:41 |
| programmerjake | the minutes are a loose approximation of what was said | 18:41 |
| lkcl | then it's the compiler's job to know the allocation | 18:41 |
| lkcl | by way of issuing a setvl with a "maxvl=NN" argument | 18:41 |
| programmerjake | yup, I think I adequately explained that to him | 18:41 |
| lkcl | (which is always static - no way to set it dynamically from a register, this is very deliberate) | 18:42 |
| lkcl | and setvl *never* exceeds the current maxvl | 18:42 |
| lkcl | cool | 18:42 |
| lkcl | "If you run the configure vector instruction, it throws away all the stuff." | 18:42 |
| lkcl | we *do not* do that | 18:42 |
| lkcl | because it's on top of the standard regfiles | 18:43 |
| programmerjake | that's risc-v v's reconfigure vectors instruction | 18:43 |
| lkcl | yehyeh | 18:43 |
| lkcl | because they literally reconfigure the entire engine, internally | 18:43 |
| lkcl | i think there may be an option now to sit on top of the FP regfile | 18:44 |
| lkcl | btw well done not freaking them out with the different modes or with twin predication at an early stage :) | 19:04 |
| lkcl | i worked out a way to say what twin predication is: "back-to-back VREDUCE-VEXPAND, for any single-src single-dest arithmetic or memory operation" | 19:05 |
| lkcl | or better just "for any single-src single-dest operation" | 19:05 |
| programmerjake | I tried to reduce it to the features that require a fundamental rethink at the compiler-level, the rest of the stuff I'd consider more to be just new operations | 19:26 |
| programmerjake | my home file server won't boot after an automatic upgrade...will have to figure out what broke... fun! | 19:29 |
| lkcl | ngggh, now you know why i disable automatic updates... | 19:50 |
| lkcl | gaah found a long-standing bug that shouldn't have been in ISACaller (and wasn't for a long time) | 21:30 |
| programmerjake | it appears as though my HDDs are taking too long to come online, so the mount service is timing out, causing the system to go into emergency mode without a running ssh server...had to move it all to another room where I have a screen and keyboard | 21:47 |
| lkcl | add "bootdelay=something" in seconds to /etc/default/grub | 21:54 |
| lkcl | sorry, "rootdelay=NNN" | 21:55 |
| lkcl | https://www.linuxtopia.org/online_books/linux_kernel/kernel_configuration/re58.html | 21:55 |
| programmerjake | my root drive is a usb flash drive -- no issues there | 22:08 |
| programmerjake | I'm trying adding x-systemd.device-timeout=300s to fstab | 22:09 |
| programmerjake | it works! | 22:10 |
| lkcl | interesting choice of root boot media :) | 22:19 |
| lkcl | i mount /tmp, /var/tmp, and many other locations onto tmpfs when using usb flash | 22:19 |
| lkcl | otherwise you end up with massive wear on something that wasn't designed for it, and goodbye server | 22:20 |
| programmerjake | no tmpfs for me ... odroid xu4 with only 2GB ram | 22:28 |
| programmerjake | I'm using the flash drive because it's *waay* better than the sdcard it's designed to use, the sdcard would wear out in a year | 22:29 |
| programmerjake | I changed it to only have /boot on the sdcard | 22:30 |
| programmerjake | it's a 128GB flash drive, so it's less likely to quickly wear out from the few MB of temp files | 22:31 |
| programmerjake | besides...when it does finally wear out, that'll motivate me to finally get around to upgrade from ubuntu 16.04 | 22:33 |
| choozy | programmerjake, does it help to use f2fs? Or something like that? | 22:34 |
| programmerjake | idk, i'm using ext4 | 22:34 |
Generated by irclog2html.py 2.17.1 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!