Tuesday, 2021-05-18

mepyahah I laughed hard when I read the blue flowers-ibm thing on Twitter xD10:56
mepyHave to go now. Greetings to everyone10:57
*** mepy <mepy!~mepy@> has left #libre-soc10:57
lkclcesar[m]1, thank you for reporting #648.  fixed and pushed, can you double-check?12:57
lkclNGI Forum 2021 is on, now12:57
lkclsorry, workshops https://2021.ngiforum.eu/workshops?track_id=1665612:58
lkclcesar[m]1, i finally fixed the bpermd ISACaller pseudocode13:22
cesar[m]1Yay! It was really annoying. Checking it right now.13:24
cesar[m]1It works! Everything is fine.14:27
lkclcesar[m]1, briiilliant17:28
lkcli don't know why i didn't check earlier, all it was, was "128" in an 8-bit space was being considered a negative number inside a loop-count17:29
* lkcl now focussing on SVP64 FP18:10
lkclsv.fadds works18:10
lkclsv.lfsx also works18:23
programmerjakelkcl: you can use simple-soft-float as a reference to test against...it should be 100% bit-accurate except it might give a different bit pattern for NaNs18:39
programmerjakeit supports all 4 ieee754 standard rounding modes...I haven't yet added support for round to nearest ties to odd or round to nearest ties to max magnitude yet (both of which are supported by just a few Power instructions iirc)18:42
lkclprogrammerjake, yeah this is where it gets interesting, we have to bootstrap up18:45
lkclssf-against-isacaller then18:45
lkclbut, TestIssuer with FP pipelines, when they were running at around 3 cycles per second last year? hoo-boy18:46
lkclright now though i am ennntiiirely skipping major unit testing18:47
lkclin order to get Lauri up and running as quickly as possible18:47
lkclhave to get back to the RADIX MMU as quickly as possible because that's the next major P.R. milestone18:48
lkcl(being able to run linux)18:48
lkclwe're maybe about 10-20 days out from being able to do that18:48
lkcl*if* i don't get pulled off to fire-fight other tasks18:49
lkclwe seriously, seriously need people to get helping with this project18:49
programmerjakeok. I'd at least run some spot-checks to ensure you didn't accidentally use the fp alus incorrectly or pick one of the broken ones.18:53
programmerjakesimple-soft-float is pretty easy to get set up and your going to get portable results that don't depend on your host cpu's fp support, and it supports non-roundeven rounding modes.18:56
programmerjakeI went to great effort to get the python `help()` text for simple-soft-float to be usable19:02
lkclreally appreciated: i am however very serious when i say that without additional help i simply cannot do everything, we are running out of time19:59
lkcli'm therefore adding absolute bare minimum unit tests20:01
lkcland then going to move off of this and back to RADIX MMU which has a much higher priority20:02
lkclif Lauri runs into difficulties i will have to fire-fight, drop everything, sort out the issue as quickly as possible and then get back to the RADIX MMU20:02
lkclthis is because i am having to do the tasks normally covered by at least 3 people.20:03
programmerjakethx for all the hard work!20:18
lkclmeeting 20mins22:39
lkclcesar[m]1, ^22:39
lkclprogrammerjake, ^22:39
lkcljn__, ^22:39
jn__will join in a bit22:40

Generated by irclog2html.py 2.17.1 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!