Trifton_ has quit [Read error: Connection reset by peer]
___nick___ has quit [Ping timeout: 256 seconds]
<conchuod>
prabhakarlad: welp, mock-up is sent :)
<conchuod>
Now to wait for Palmer or Arnd to say why it's a bad idea haha
<palmer>
Shouldn't take that long... ;)
<palmer>
ah, OK that makes sense
<palmer>
so probably more of a question for the other folks, I don't care that much where these end up
<conchuod>
As long as you don't hate that riscv_cache_maint_ops thing I introduced..
<palmer>
I think that might be controversial, but IMO it's fine until someone can demonstrate a performance issue
zjason`` has joined #riscv
zjason` has quit [Ping timeout: 264 seconds]
<conchuod>
The alternative is register an op for everything and use stringify() or w/e - but I'd rather get the platform working first.
<palmer>
I think the opposition was to having a table of functions for cache flushes, but that doesn't look like what this does
<palmer>
I haven't looked enough, though...
<conchuod>
Yah, people are welcome to scream if I ignored some history
BootLayer has quit [Quit: Leaving]
<palmer>
OK, so you're patching in the function call only for the non-zicbom flavors. That might be fine (though I'd go the other way, leave a function call in there and patch it out so you don't have to do the register save/restore)
mlaga97 has joined #riscv
<arnd>
I talked about this with prabhakarlad the other day, and suggested to do something like 'if (!ops->flush) zicbom_flush(); else ops->flush();' to have the branch predictor take care of the normal path. The cache management operations tend to already be so slow that any optimization beyond that is probably not measurable
<palmer>
I'm fine with that, I think we'd even be OK with a function for all ops (ie, even the zicbom stuff)
<palmer>
flushing to the point of coherence is going to be pretty slow, so I don't think it's a big deal to have a function call there
<conchuod>
I tried to change as little possible from what existed, prob have a different attitude to others, but I was trying to avoid wholesale changes that'd require testing the other platforms.
<conchuod>
With the intent of having the platform actually supported and then possibly cleaning up that alternative since there's gonna be several other implementations of that AFAICT
cousteau has quit [Quit: ♫ I can't forget the day I shot that network down ♫]
XYZ has joined #riscv
<conchuod>
arnd: if you wouldn't mind pointing that out against where it's relevant on the list that'd be neat. I never quite got what you were suggesting to do that instead of.
<bjdooks>
tbh, getting something going might be better than getting something perfect
Narrat has quit [Quit: They say a little knowledge is a dangerous thing, but it's not one half so bad as a lot of ignorance.]
KombuchaKip has quit [Quit: Leaving.]
wingsorc has joined #riscv
mechaniputer has joined #riscv
KombuchaKip has joined #riscv
pedja has quit [Quit: Leaving]
<conchuod>
Yah, to me it's nigh on a certainty that we'll be touching this again so having one more platform functional before restarting debate about how we do CMOs would be neat :)
Gravis_ has joined #riscv
Gravis has quit [Ping timeout: 246 seconds]
<bjdooks>
is it just the thead stuff at the moment, or does the visionfive also have the same issue
<conchuod>
heheheh
eroux has quit [Ping timeout: 252 seconds]
<conchuod>
renesas need it, we want it, the visionfive v1 wants it (but does anyone actually care enough to sort that out?) and I'm sure that there'll be more.
<bjdooks>
all the stuff i'm dealing with is coherent
<muurkha>
i wish i could say the same
<bjdooks>
urgh, has the kernel make system decided to ignore environment varaibles for some reason?
<bjdooks>
forgot export
<bjdooks>
grrrr
<muurkha>
heh
<bjdooks>
maybe time for bed
<muurkha>
sometimes if you're tired enough you can make negative progress on programming
<bjdooks>
for now, to get builds running i've done