<courmisch>
though I don't suppose either Linux or BSD will cope well with vectors on only 1 of 2 cores
ldevulder has joined #riscv
notgull has quit [Ping timeout: 255 seconds]
notgull has joined #riscv
pecastro has joined #riscv
prabhakarlad has joined #riscv
danilogondolfo has joined #riscv
Jackneill has joined #riscv
mlw has quit [Ping timeout: 252 seconds]
<davidlt>
courmisch, you can always ignore the other core.
<gurki>
looks like he main focus is using the "kpu unit" whatever that is precisely anyway
<gurki>
video encoding at 20/40 fps ... isnt that super-uncommon Oo
<gurki>
now i wonder from which systems they repurposed this soc :3
<courmisch>
davidlt: yeah but I wonder what Linux does in the face of mismatched ISA extensions
<davidlt>
courmisch, not sure, most likely you would remove the extra extensions from other cores in DT.
<davidlt>
There will be more of this. StarFive next-gen SoC/IP also has different ISA between cores.
<conchuod>
If only some cores have the vector, and not others, it will be disabled.
<conchuod>
The DT should reflect the actual extensions each hart has.
<davidlt>
I bet at some point different ISA will get support.
aburgess has quit [Ping timeout: 246 seconds]
prabhakarlad has quit [Quit: Client closed]
raym has quit [Ping timeout: 255 seconds]
raym has joined #riscv
aburgess has joined #riscv
<gurki>
davidlt: it forced intel to ditch avx512 on desktop, im not certain about that one
<gurki>
however, i would be happy to see that
<davidlt>
gurki, I think (just an opinion) Intel is playing a long game. AVX10.X is happening. Thread Director is improving.
<davidlt>
I wouldn't be surprised that Intel plans to have cores with different capabilities (could be vector length, different ISA extensions).
<davidlt>
It's just a 10-year mission (slow burn).
<gurki>
as i said. i really hope it does :)
<davidlt>
But I am impressed in Zen 4 / Zen 4C, which is epic for AMD :)
raym has quit [Ping timeout: 246 seconds]
<davidlt>
IIUC it takes twice less space, just half of cache, and designed with denser low-power cells or whatever, but it's the same Zen 4 core design.
<gurki>
oh i had no intention of saying "look at great intel", just hinting at ppl hitting that wall before
<davidlt>
So it sounds cheaper compared to Intel doing different core designs (but they can pack more of these).
<davidlt>
Oh yeah, I still recall Exynos from Samsung with different ISA on the cores.
<davidlt>
And random segfaults (illegal instruction) is your thread was moved to a core that didn't implement some extension :)
<davidlt>
I kinda like 16 + 8, 32 + 8 (if that happens) designs.
<davidlt>
and AMD is kinda going that direction too with Zen 4 + Zen 4C mix in the laptops to begin with, but again ISA is the same there, and even the cores are kinda the same.
<courmisch>
on AArch64, mismatched extensions are reduced to lowest common denominator as well. Hotplug CPU are parked if they don't implement the boot-time lowest common denominator
<courmisch>
this is going to be a tough sell for Linux RISC-V to upstream