tgschultz_ has quit [Quit: Going offline, see ya! (www.adiirc.com)]
tgschultz has quit [Quit: Going offline, see ya! (www.adiirc.com)]
tgschultz has joined #haiku
tgschultz_ has joined #haiku
<julicenri>
x512[m]: Is your RADV work based on the Linux or OpenBSD driver?
<x512[m]>
RADV driver is basically unmodified driver from Mesa.
<julicenri>
So Haiku could potentially keep up with Linux in terms of Vulkan driver support?
<x512[m]>
I think yes.
<x512[m]>
Driver part that actually communicates with hardware (RadeonGfx) is written from scratch. On Linux it work is done by amdgpu kernel driver.
<x512[m]>
RadeonGfx run in userland.
<julicenri>
Do you think that it will be feasible in the near future to only port the Vulkan drivers and use Zink for GL -> VK, skipping the need to port GL drivers?
<x512[m]>
Yes, I am thinking to implement OpenGL over Vulkan over Zink or similar. OpenGL DRI drivers depend on Mesa infrastructure that may be troublesome.
<julicenri>
Are there any Haiku-specific patches that need to be upstreamed?
<x512[m]>
No for RADV if I remember correctly, but patches for Vulkan Intel driver will be likely needed.
<Not-5726>
[haikuports/haikuports] pulkomandy pushed 1 commit to master [+1/-1/±0] https://git.io/J1xvq
Maturion has quit [Remote host closed the connection]
Maturion has joined #haiku
humdinger has joined #haiku
<waddlesplash>
hmmmmm
<waddlesplash>
what if we implemented 32-bit compatibility mode on x86_64 purely in libroot
<waddlesplash>
by (ab)using ilp32 mode, and having a file with hooks to call x86_64 syscall entry
<x512[m]>
waddlesplash: It needs to run in separate CPU mode?
<nephele>
waddlesplash: that is basically the design wine (well, crossover) uses to run x86 binaries on amd64 macs (because the processor /does/ do x86 mode, just nowhere in the userland is that supported in libraries)
<waddlesplash>
right, however we can support it in our libraries
<waddlesplash>
so we need only translate the syscall hooks
<nephele>
I think it would be totally doable, yeah crossover also supports it in their libraries, just the OS itself doesn't, we don't have this disadvantage of course :)
<waddlesplash>
well! okay, that answers that question
<waddlesplash>
probably the offsets asm is broken
ClaudioM has quit [Quit: leaving]
<waddlesplash>
ok, the actual offsetof is 700, and in fact the generated header has this
<waddlesplash>
0x2b4==692, so we appear to be off by 8 bytes. how'd that happen?
<waddlesplash>
ah, I have a stale asm_offsets.h., that explains it
x10z has joined #haiku
<waddlesplash>
x512[m]: now I get the hang on rocket
<waddlesplash>
vm_soft_fault: va 0x7fe2a1609000 not covered by area in address space
jmairboeck has quit [Quit: Konversation terminated!]
<waddlesplash>
x512[m]: yes, it's indeed a stackoverflow
<x512[m]>
Maybe something similar to Clang with inserting memcpy call inside its implementation.
<waddlesplash>
probably
<waddlesplash>
however I have no idea what the symbols are
<x512[m]>
I often add KDL trigger in serious debug event function to investigate userland crashes on early boot steps.
<waddlesplash>
yep, that's what I did
<waddlesplash>
well, I injected it directly to VM, but yeah
<waddlesplash>
x512[m]: yep, it's memset
<Vidrep_64>
It looks like cdda2wav is no longer working on Haiku. I picked up where we left off with audio cloning on BurnItNow and now cdda2wav returns this error: "cdda2wav: Child reader sem request failed."
<Vidrep_64>
I also tried from CLI and it's the same
gouchi has quit [Remote host closed the connection]
bitigchi_2 has quit [Ping timeout: 480 seconds]
_Dario_ has quit [Quit: Vision[]: i've been blurred!]
<x512[m]>
Asteroids sample program run stable for several hours with hardware rendering. It can be made usable after architecture reorganizing.
<waddlesplash>
do you have a "gpu process mode" that supports multiple programs yet?
<waddlesplash>
this sounds only moderately useful if only one application can use 3D at once :-p
<x512[m]>
Not yet, that is planned "architecture reorganizing". I plan to use PortLink and port_id in shared area.
<waddlesplash>
ah
<waddlesplash>
PortLink sounds like what the basis for accelerants 2.0 should be :-p
JakeSays has joined #haiku
<x512[m]>
If Radeon display output is needed, it should use RadeonGfx GPU memory manager for framebuffer memory allocation that is currently not implemented.
<waddlesplash>
x512[m]: sounds like this should go inside the accelerant process then
<waddlesplash>
the primary accelerant can allocate radeongfx
<waddlesplash>
but then of course if the driver crashes, it also takes down app_server with it
<waddlesplash>
yeah, I guess a separate process for video drivers is really what we need
<x512[m]>
Yes. app_server already has protocol for setting mode available by BScreen.
<waddlesplash>
my plan for a while has been to move accelerants to a separate process, and get rid of accelerant cloning
<waddlesplash>
and then add something like PortLink for cloned accelerant operations to instead make requests to the server
<waddlesplash>
alternatively, accelerants can set a "cloned accelerant structure" if they really want to, but I imagine most won't
<x512[m]>
I also want to add common diagnostic protocol for GPU load, temperature, clock speed measure, buffers enumeration etc..
<waddlesplash>
ideally we would get a few people together for a meeting and draft a new accelerant API
<waddlesplash>
because we need one that handles: 1. multiple screens, 2. better accelerant information (like what you are describing), 3. paves the way for 3D acceleration, 4. potentially supports multiple accelerants at once (e.g. Intel+Radeon in your case)
<x512[m]>
Currently there are security flaw in many drivers implementation including radeon_hd that VRAM is globally mapped and can be accessed by everyone. RadeonGfx allocates internal structures there like page translation table. System can be easily hacked by writing to VRAM mapped area.
<waddlesplash>
yes, I've noticed
<waddlesplash>
another reason to get rid of BDirectWindow's primary implementation
<waddlesplash>
ok. let's see if this nvme_trim works
DKnoto has quit [Ping timeout: 480 seconds]
AlienSoldier has joined #haiku
DKnoto has joined #haiku
vdamewood has quit [Quit: Life beckons.]
<rennj>
r1b3 pkgman update today, go to select 1000 mame games.zip's opentracker utter failure isoimage copy paste fail
<rennj>
and mame fail,
<rennj>
heh aros-one no problem coping files from iso to mame roms dir, and mame worked with usb f310 gamepad
<rennj>
aros-one was kicking haiku's ass today..i could play games with my usb gamepad and copy 1000 files..
bitigchi_2 has joined #haiku
<waddlesplash>
what failed?
<rennj>
couldnt copy the files
<rennj>
select them alt-a
<rennj>
1017 files to be exact
<rennj>
aros-one no problem, meanwhile on haiku
<rennj>
no working mame
<rennj>
and just trying to copy files from isoimage to roms dir could not do with opentracker
<rennj>
alt-a select all
<rennj>
try to drag and drop on dir..no dice
<rennj>
thats after the 1hour long update
<waddlesplash>
yeah, what was the precise problem? do you have an error message?
<rennj>
r1beta3 : pkgman update
<rennj>
i told you
<rennj>
the opentracker fails
<rennj>
no error
<waddlesplash>
yes, but how does it fail. nothing happens?
<rennj>
just cant copy files
<rennj>
select all the files on iso..and cant copy them
<waddlesplash>
so, you hit copy and nothing happens?
<AlienSoldier>
waddlesplash i think the biggest hurdle is to allow appserver to open low resolution screen.
PopperBruda has joined #haiku
bitigchi_2 has quit [Ping timeout: 480 seconds]
<rennj>
what we got here is a failure to communicate
<rennj>
cool-hand-luke
<AlienSoldier>
rennj can you move big amount of files other than from that particular ISO?
<rennj>
rm -rf vm
<rennj>
it pissed me off
<waddlesplash>
well, GCC 11 builds now boot, the cvar code doesn't panic, and my working trees are almost entirely clean
<rennj>
pkgman update was painful enough from r1beta3 to 62 or whatever
<waddlesplash>
and the nvme trim code is now tested and merged. so, that was a pretty good day
<rennj>
then reboot it that failure
<rennj>
then reboot into that failure
<rennj>
where selecting 1000+files throws the filemanager into a freeze..
<waddlesplash>
oh, a freeze
<waddlesplash>
well, that's more actionable
<rennj>
same action in aros-one just worked...2 file panes..drag and drop
<rennj>
isoimage file pane, and roms dir file pane
<rennj>
alt-a/select all and gui drag and drop
<rennj>
not alt-c alt-v or what have you
<waddlesplash>
x512[m]: I think on Linux there's a way to force low-power mode for the GPU, so it would be interesting to see what the performance difference is then
<rennj>
well fglrx i had slider
<rennj>
amdgpu ive not looked at gui apps for powernow amd fu
x10z has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
<rennj>
ondemand performance p-state, c-state of cpu..the apu..graphics part i lost features going opensauce
<rennj>
laptop mode vs wall mode
<rennj>
fglrx closed source had options the amdgpu ive not seen yet
<waddlesplash>
up next, tomorrow: the Big GCC 11 Merge
<waddlesplash>
hopefully buildtools won't give me too hard a time
<rennj>
linux mint mame better smoke winblows10 mame...and aros mame...i want 850 games damnit...openbor master of the universe..aros openbor broken compared to win10 version.
<rennj>
openbor runs real pretty like on win10 vm
<rennj>
accelerated x11 for vmware is easy vblank_mode=0 glxgears
<rennj>
1200fps i know i got it
<rennj>
xorg log has info also.
<rennj>
karnov.zip
<rennj>
haha karnov arcade game
<AlienSoldier>
karnov is awsome
<rennj>
5035 frames in 5.0 seconds = 1006.797 FPS
<rennj>
thats with 2 vm's going
<rennj>
this vm has 20day uptime..almost 3weeks
<rennj>
aros-one was rocking..haiku was sucking...
<rennj>
aros-one-x86 version 1.6
<rennj>
2.7GB of software
<rennj>
plenty of broken software also
<rennj>
linux ftw!
<rennj>
bet linux-mint-mame better than win10-mame