alyssa changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - Logs https://freenode.irclog.whitequark.org/panfrost - <daniels> avoiding X is a huge feature
raster has joined #panfrost
alyssa has quit [Quit: leaving]
raster has quit [Quit: Gettin' stinky!]
stikonas has quit [Ping timeout: 260 seconds]
rando25892 has quit [Ping timeout: 260 seconds]
rando25892 has joined #panfrost
archetech has quit [Quit: Leaving]
archetech has joined #panfrost
vstehle has quit [Ping timeout: 256 seconds]
rak-zero has joined #panfrost
vstehle has joined #panfrost
vstehle has quit [Quit: WeeChat 2.9]
vstehle has joined #panfrost
archetech has quit [Quit: Leaving]
nlhowell has joined #panfrost
nlhowell has quit [Remote host closed the connection]
nlhowell has joined #panfrost
nlhowell has quit [Remote host closed the connection]
nlhowell has joined #panfrost
nlhowell has quit [Client Quit]
raster has joined #panfrost
<macc24> icecream95: thanks
warpme_ has joined #panfrost
icecream95 has quit [Ping timeout: 260 seconds]
<HdkR> robmur01: https://github.com/FEX-Emu/FEX/wiki/32Bit-x86-Woes#ioctl---54 How much would people hate me for exposing a new ioctl32 syscall?
nlhowell has joined #panfrost
<chewitt> all the screensaver/visualisations (shadertoy based) in Kodi are now working :)
<chewitt> this means I officially have nothing to nag about
<chewitt> huge kudos to the team!
stikonas has joined #panfrost
<robmur01> HdkR: a new syscall solely for the purpose of abusing the compat ABI that may not even be available? I'm gonna guess "not popular"...
<HdkR> Guess I might need to double check how `in_compat_syscall` works
<HdkR> But we do already have a whole class of syscalls on x86-64 to determine if the syscall is x86-64 or x32
<robmur01> x32 is a whole other mess though
<HdkR> I've heard that one nasty solution would be to have a custom memory allocator for my entire application and thieve that full 64bit address space, but leave the 32bit space alone for ioctls to be forced to stick memory in the low end
<HdkR> But that's a lot of work and I don't know what new problems it'll pose
<robmur01> I guess what you really want is the thread flag/address space/etc. bits of ILP32, but without the syscall mess that everyone hates, since presumably the code for marshalling syscalls between the emulation and the host *is* actually LP64
<HdkR> Right
<HdkR> A lot of the random state tracking needs to be emulated anyway, since x86->ARM64, just something as opaque as ioctl is hard
<macc24> HdkR: whats https://github.com/FEX-Emu/FEX
<HdkR> macc24: Userspace x86-64 (and soon 32bit x86) emulator for running Linux applications on an ARM64 host
<macc24> how fast is it?
<HdkR> Needs to be faster. Some games run full speed
<macc24> games?
<macc24> bookmarked
<HdkR> We're never going to receive native ports of most any game on Steam. Need an emulator that is worthwhile to run them on ARM
<macc24> yeah i already heard of box86 and tested it a bit
<HdkR> FEX does a lot of stuff like box86, but we are going to cover 32bit and 64bit x86
<HdkR> Just have additional pains since we are running 32bit apps in a 64bit environment
<macc24> doesn't linux support running 32bit binaries on 64bit host without issues?
<HdkR> Sure, if you're wanting to run x86 apps on x86-64 or AArch32 apps on ARMv8
<macc24> okay
<HdkR> But x86 on ARM64 is a different beast :P
<macc24> yeah
<HdkR> It's also not a good idea to support AArch32 as the host for FEX, we lose all the new ISA improvements
<macc24> b-but my 2015 laptop isn't arm64
<HdkR> Your 2015 laptop isn't the target audience :P
<HdkR> Emulating strong memory model x86 on ARMv8.1 is already a pain, don't want to make it too much worse
<HdkR> and....ugh, unaligned atomics
<macc24> uh-huh
<HdkR> Fun enough, ARMv8.6 solves a bunch of the strong memory model problems. Sadly we won't have ARMv8.6 until post-Hercules
<HdkR> Doesn't solve unaligned atomics that cross cachelines, but can't win them all
<macc24> post-what?
<HdkR> Hercules is Cortex-A789
<HdkR> lol, A78
* macc24 is waiting for arm cortex-a2137
<macc24> ok.... does FEX run quickly on cortex-a73?
<HdkR> and Apple's SoC cheats by having an MSR that enables global strong memory model for the process
<HdkR> ehhh depends on what you're running
<HdkR> It really wants A75 minimum
<HdkR> Almost all x86 instructions turn in to atomic loadstores
<HdkR> and you need ARMv8.1 for atomic memory ops to clean up a lot of them
<macc24> hmm
<macc24> so it wouldn't work on a73
<HdkR> It'll run on base ARMv8.0, but don't expect decent perf :P
<macc24> how fast will it run on just armv8?
<HdkR> Well it won't run supertuxkart full speed atm
<macc24> hmm
<macc24> did it use llvmpipe?
<HdkR> There's still work to go for getting it full speed
<HdkR> GL driver is unrelated
<HdkR> It'll use the native x86-64 libGL, or with thunks it'll use the host libGL
<macc24> hmm
<HdkR> I've tested freedreno, turnip, radeonsi, radv, and i965 through the emulated layer. Seemed fine
<robmur01> how about v8.0, but at 3.3GHz? :D
<HdkR> Someone else tested the Nvidia blob on the x86 host and it seemingly worked
<HdkR> pft
<macc24> robmur01: *beats 3.3GHz with 3.4GHz*
<HdkR> If it is 3.3Ghz and a Cortex-A53 then it'll be pretty bad
<macc24> HdkR: how about 2GHz a73?
<robmur01> well, a 3.3GHz A53 can probably perform as well as you like, because it's already in fantasy-land :P
<HdkR> hah
<robmur01> the eMAG, however, is real
<macc24> robmur01: whats eMAG
<HdkR> Anything in-order would have a rough time, I don't plan on having instruction scheduling at all
<macc24> well, Cortex A73 is out of order
<HdkR> Give it a couple of months and maybe I'll have a real "user experience" response
<HdkR> The plan is to have perf be within 25-50% slower than native as per https://github.com/FEX-Emu/FEX/blob/master/External/FEXCore/Readme.md
<HdkR> But that is also the expectation of atleast ARMv8.1
<macc24> hmmmm
<HdkR> My personal min-spec for CPU is still Cortex-A78 though. Anything less than that is just sad ;)
<macc24> do A78 SBCs even exist?
<HdkR> Nope
<HdkR> Samsung will have one in a couple months
<HdkR> Nvidia will have one late next year
<HdkR> QCom probably in a few months
Net147_ has joined #panfrost
Net147 has quit [*.net *.split]
urjaman has quit [*.net *.split]
klaxa has quit [*.net *.split]
<macc24> can't wait until there's SBC with it
urjaman has joined #panfrost
<HdkR> I'll definitely be buying a few Nvidia Orin boards
<macc24> isn't nvidia anti-open source?
<HdkR> Sure, I'll be annoyed every moment of using it. Nobody else making a SoC with 12 Cortex-A78 cores though
<macc24> i hope that there will be Orin SBC with PCI-E available
<HdkR> Should be, Xavier already did
<macc24> oh then maybe it can replace my x86 PC
<HdkR> Maybe. Carmel is a bit derpy CPU core
<macc24> well if it's faster than core 2 quad i will probaby buy it
<HdkR> At least it has core count. It's performance is all over the place depending on workload
<HdkR> Its*
<macc24> all over the place as in?
<HdkR> Worst case its performance can be around Cortex-A55 levels
<HdkR> But on average its more like...A76?
exit70[m] has quit [Ping timeout: 246 seconds]
<macc24> huh
clementp[m] has quit [Ping timeout: 240 seconds]
<HdkR> Derpy CPU
l-as has quit [Ping timeout: 268 seconds]
Ke has quit [Ping timeout: 268 seconds]
wiizzard has quit [Ping timeout: 268 seconds]
robmur01 has quit [Remote host closed the connection]
robmur01 has joined #panfrost
kinkinkijkin has joined #panfrost
exit70[m] has joined #panfrost
wiizzard has joined #panfrost
clementp[m] has joined #panfrost
archetech has joined #panfrost
davidlt has joined #panfrost
Ke has joined #panfrost
l-as has joined #panfrost
<macc24> kinkinkijkin: got a duet?
<kinkinkijkin> not yet, best buy had an issue
<macc24> there might be a debian derivative for duet Soon™
Elpaulo has joined #panfrost
italove has quit [Ping timeout: 240 seconds]
kinkinkijkin has quit [Remote host closed the connection]
raster has quit [Quit: Gettin' stinky!]
Net147_ is now known as Net147
rak-zero has quit [Ping timeout: 264 seconds]
uis has joined #panfrost
davidlt has quit [Ping timeout: 246 seconds]
raster has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
Green has quit [Ping timeout: 256 seconds]