<klys>
as a reminder I have the 8" floppy drives for it
<geist>
oooooh!
<klys>
for reals!! I haven't seen anything like this pop up in over a year.
<nikolar>
it's not even that expensive as far as old computers go
<heat>
linux kernel
<klys>
as a reminder rms wrote gcc on a pdp system
<geist>
the seller is fairly well known on the local retro computer club, since he's just up in Kenmore, WA
<klys>
there was an early 8in dec rx02 disk image floating around for the earliest known cp/m
<klys>
author g.kildal
<klys>
the chargen rom in mame doesn't appear to be quite correct for its emulated decmate ii, as the screen stays blank during emulation
<klys>
however I was able to dump the rx50 rom for write precomp
<pounce>
so i've been musing about a pretty general issue, and i'd like to hear your thoughts
<pounce>
modern cpus have tons of vector instructions and parallelism, but it's usually only utilized by a few hyper-optimized programs. since trying to figure out what CPU features a device has and route code paths to the correct instructions is a pita
<pounce>
what's a good os design solution for this that isn't "lets make everything bytecode"
<klys>
well most processors can't actually use avx512 well
<pounce>
this too
<pounce>
OSes commonly time instructions though. no reason this can't be tested
<klys>
what you would need is access to a temperature sensor or cpu thermal info as you proceed to test vector instructions like that
<pounce>
yeah fair, that's not really doable
<pounce>
i know vector instructions can't be run all the time but it still can speed up tons of common operations like copies
<klys>
qemu only adopted tcg bytecode in intermediate versions, so if you go back to qemu-1.0 for example, you can still virtualize without bytecode
<pounce>
yeah ik about tcg
<pounce>
i think it's less interesting though to try to vectorize machine code
<pounce>
and more interesting to provide a mechanism for programs to swap in more efficient code
<pounce>
the compiler that they use for their programming language can make better optimizations and vectorizations anyway
<kof673>
i don't know, either choose a language that meshes well with such things, or one that is deliberately a blank slate on such features :D
<pounce>
i don't think only jit'd languages should be able to generate efficient code
<kof673>
(my comment was for everything, nothing in particular)
FreeFull has quit [Ping timeout: 248 seconds]
FreeFull has joined #osdev
sham1 has quit [Server closed connection]
sham1 has joined #osdev
edr has quit [Quit: Leaving]
innegatives has quit [Quit: Connection closed for inactivity]
itrsea has quit [Remote host closed the connection]
itrsea has joined #osdev
c0co has quit [Quit: Ragequite]
_whitelogger has joined #osdev
_whitelogger has joined #osdev
chiselfuse has quit [Remote host closed the connection]