but it is generally in that order of magnitude still
#uACPI - a portable and easy-to-integrate ACPI implementation
torpid ferry
fiery turtle
yeah sounds like bs
i wonder if you're supposed to first call some other method
linux does i think?
like display init
fiery turtle
_BQC returns zero btw
it reads from a hardcoded place in NVS
that it updates from BCM
so no wonder its 0
btw
the package it retruns just has every value from 0 to 100
Return (Package (0x67)
{
0x50,
0x32,
Zero,
One,
0x02,
0x03,
0x04,
0x05,
0x06,
0x07,
0x08,
0x09,
0x0A,
0x0B,
0x0C,
0x0D,
0x0E,
0x0F,
0x10,
0x11,
0x12,
0x13,
0x14,
0x15,
0x16,
0x17,
0x18,
0x19,
0x1A,
0x1B,
0x1C,
0x1D,
0x1E,
0x1F,
0x20,
0x21,
0x22,
0x23,
0x24,
0x25,
0x26,
0x27,
0x28,
0x29,
0x2A,
0x2B,
0x2C,
0x2D,
0x2E,
0x2F,
0x30,
0x31,
0x32,
0x33,
0x34,
0x35,
0x36,
0x37,
0x38,
0x39,
0x3A,
0x3B,
0x3C,
0x3D,
0x3E,
0x3F,
0x40,
0x41,
0x42,
0x43,
0x44,
0x45,
0x46,
0x47,
0x48,
0x49,
0x4A,
0x4B,
0x4C,
0x4D,
0x4E,
0x4F,
0x50,
0x51,
0x52,
0x53,
0x54,
0x55,
0x56,
0x57,
0x58,
0x59,
0x5A,
0x5B,
0x5C,
0x5D,
0x5E,
0x5F,
0x60,
0x61,
0x62,
0x63,
0x64
})
and if it detects linux it returns some other crap lol
fiery turtle
perhaps
calm latch
OBOS becomes DOS
north holly
maybe I need to call _DSS
north holly
seems like I need to pass 0x8000'0001
which tells it to enable the display, then commit the changes
fiery turtle
If (((Arg0 & 0xC0000000) == 0xC0000000))
{
CSTE = NSTE /* \NSTE */
}
your aml checks one bit
north holly
bruh 
calm latch
2 bits ¿
fiery turtle
hmm
fiery turtle
considering GFX0 doesnt even have a DSS
fiery turtle
why that specific DD
north holly
because that's where I get the brightness event
fiery turtle
ah
north holly
so I just did all my stuff there
fiery turtle
probably correct
left orbit
how long does run_tests.py --large take to run usually?
fiery turtle
like 5 minutes
north holly
use make
median crest
Add a big mutex around the logs 
fiery turtle
its python soo
left orbit
well in case it fails you could just append to a results array and print the errors at the end
fiery turtle
ye
left orbit
i'll look into making the runner parallel in a bit
fiery turtle
that'd be based
left orbit
i want to give sso a shot because im bored + where is pitust
fiery turtle
north holly
while you guys do that, I will be staring at AML
left orbit
i see that uacpi has no actual string type, everything seems to be a uacpi_buffer
so i guess it's a small buffer optimization
north holly
I hate AML
fiery turtle
who doesnt
north holly
acpi spec devs before starting work:
I think by now it's a good idea to just ditch this
since I can't figure out how to get it working
north holly
I'll get it working eventually™️
maybe the reason I wasn't able to get it to wokr
*work
was because I was reading the spec
fiery turtle
yup
https://elixir.bootlin.com/linux/v6.12.6/source/drivers/acpi/acpi_video.c#L679 maybe something to do with this?
linux calls this on init
north holly
perhaps
fiery turtle
idk it does kind of a lot
north holly
bus_DOS?
fiery turtle
https://elixir.bootlin.com/linux/v6.12.6/source/drivers/acpi/acpi_video.c#L679 m...
.
frank canopy
used to be exactly. now it's usually a fixed speed at around the same value so you can use it for timekeeping since cpu speed can vary
left orbit
damn i forgot how shit python multiprocessing is
fiery turtle
indeed
left orbit
i WILL make it work.
fiery turtle
based
left orbit
they already run in parallel i just need to fix passing results from workers back to the python runner
fiery turtle
fiery turtle
daaamn
north holly
@fiery turtle I have an update to obos' score:
[uACPI][INFO]: successfully loaded 1 AML blob, 1704 ops in 5ms (avg 323566/s)```free 200k?
left orbit
send iso + command line
median crest
<@353593948523986945> I have an update to obos' score:
```
[uACPI][INFO]: succes...
Big
north holly
send iso + command line
qemu-system-x86_64 obos.iso -M q35 -m 1G -accel kvm -cpu host -debugcon file:/dev/stdout
when it's done booting it suspends
left orbit
[ WARN ] Note: Framebuffer might die
didnt even get one in the first place 
HOW DO I QUIT
WTF
north holly
[ WARN ] Note: Framebuffer might die
that's because of suspending shenanigans
left orbit
[uACPI][INFO]: successfully loaded 1 AML blob, 1705 ops in 5ms (avg 314941/s) slower than urs wtf
north holly
HOW DO I QUIT
open a new tty and kill qemu
gentle peak
362128/s
north holly
qemu nographic
north holly
yeah that worked
left orbit
[uACPI][INFO]: successfully loaded 1 AML blob, 1705 ops in 3ms (avg 446980/s)
it varies wildly
i get 270k then 440k
north holly
weird
left orbit
315k, 270k, 320k, 309k
north holly
[uACPI][INFO]: successfully loaded 1 AML blob, 1705 ops in 3ms (avg 446980/s)
ok there's my new score
left orbit
lmaoo
gentle peak
the highest i got was 371k/s
left orbit
446k was the highest i got out of like 5 runs
ill try running it on windows under whpx
frank canopy
reminds me i should un-stub my mutex and event stuff since i have those implemented, its just that any timtout other than 0 or infinite is unimplemented atm
north holly
since a lot of spinlocks are taken, a lot of IRQL changes are made, which according to fadanoid, cr8 is pretty expensive
slim panther
[uACPI][INFO]: successfully loaded 1 AML blob, 1705 ops in 3ms (avg 499891/s)
north holly
`[uACPI][INFO]: successfully loaded 1 AML blob, 1705 ops in 3ms (avg 499891/s)`
ok new score 
north holly
let me fix it
slim panther
why do you need popcnt
slim panther
can that not be emulated
left orbit
technically everything can be emulated
north holly
i did +popcnt on the command line
or that
north holly
that's also what I got
left orbit
i can try on bare metal
north holly
when I did that
left orbit
but you'd need to get rid of suspend ☠️
north holly
i can try on bare metal
more ops=less score
left orbit
ah right
north holly
but you'd need to get rid of suspend ☠️
I assume you don't trust it's implementation
left orbit
no i just don't think i'd be able to read quickly enough
wouldn't it literally suspend my computer
left orbit
then i'd be able to read the score
north holly
(waiting on uDRM for that)
left orbit
i'm not rebooting now anyway because i am improving uacpi test runner
north holly
this one doesn't suspend
loud ice
okay after measuring sso isnt that big of a deal actually i think
north holly
`[uACPI][INFO]: successfully loaded 1 AML blob, 1705 ops in 3ms (avg 499891/s)`
this is my new score
loud ice
i should try rerunning the full uacpi benchmarks
fiery turtle
wait u managed to get it working?
loud ice
yeah
fiery turtle
Damn
loud ice
i also found out my measurements were crap
and i forgot about all the calloc'ed allocations lol
fiery turtle
Lmfao
loud ice
i'm getting 143k/s with tcg now (i also added a slightly better slab allocator to my shitty kernel)
frank canopy
my gut for handling strings like that has been a string table but the way zig makes that super easy relatively speaking has been skewing things a bit tbh
plus idk if string collisions are enough to be worth it at the scale above four-char ones where youd just use a u32
loud ice
most allocations still come from deep_copy_package though
and uacpi_execute_control_method
loud ice
and uacpi_execute_control_method
i think that's the "allocate literally any object" thing, though
cant do much about that
left orbit
@fiery turtle what do you use for formatting python?
fiery turtle
<@353593948523986945> what do you use for formatting python?
Nothing lol
Check the CI script for tools that do the linting
fiery turtle
cant do much about that
Yeah
fiery turtle
Probably
left orbit
upgraded the test runner's output a little bit
oh whoops
wow discord doesn't even show the full thing
but yeah it now prints the time each test takes and the total amount of time they all took
along with the progress on each test
the code ain't even that bad, i think it's ready for pr
i haven't really tested how it looks like when they fail but
the results are all printed sequentially and i basically stole the old code for printing that
shouldn't be too different
[28/65] references-4.aml:Call-by-value w/ Store() does NOT modify integers even through a reference OK in 0.01s
[29/65] references-10.aml:Recursively modify ArgX OK in 0.01s
[29/65] empty-objects.aml:Empty objects behave correctly FAILED in 0.01s
TEST OUTPUT FOR empty-objects.aml:Empty objects behave correctly:
stdout:
[uACPI][INFO] RSDP 0x0000783597103D90 00000024 v02 (uOEMID)
[uACPI][INFO] XSDT 0x0000504000001090 0000002C v02 (uTEST uTESTTBL)
[uACPI][INFO] DSDT 0x0000508000000220 00000055 v02 (uTEST TESTTABL)
[uACPI][INFO] FACP 0x00005120000004C0 00000114 v06 (uOEMID uTESTTBL)
[uACPI][INFO] FACS 0x0000506000000440 00000040
[uACPI][INFO] SSDT 0x0000000000000000 00000055 v01 (uTEST RUNRIDTB)
[uACPI][TRACE] load of 'DSDT' (OEM ID 'uTEST ' OEM Table ID 'TESTTABL')
[uACPI][TRACE] load of 'SSDT' (OEM ID 'uTEST ' OEM Table ID 'RUNRIDTB')
[uACPI][TRACE] [AML DEBUG] String => "TestRunner ID SSDT loaded!"
[uACPI][INFO] successfully loaded 2 AML blobs, 7 ops in 0ms (avg 33040/s)
[uACPI][TRACE] initialized GPE block _GPE[0->128], 0 AML handlers (IRQ 0)
[uACPI][TRACE] initialized GPE block _GPE[128->256], 0 AML handlers (IRQ 0)
[uACPI][TRACE] fixed event 0 enabled successfully
[uACPI][TRACE] activated all 'EmbeddedControl' opregions controlled by '\ ', 0 _REG() calls (0 errors)
got a GPE
[uACPI][TRACE] activated all 'PCI_Config' opregions controlled by '\ ', 0 _REG() calls (0 errors)
[uACPI][INFO] namespace initialization done in 0ms: 0 devices, 0 thermal zones
[uACPI][TRACE] _STA calls: 0 (0 errors), _INI calls: 0 (0 errors)
[uACPI][TRACE] [AML DEBUG] Uninitialized
[uACPI][TRACE] [AML DEBUG] Package @0000503000001bd0 (0000503000001c30) (0 elements)
[uACPI][TRACE] [AML DEBUG] Package @0000503000001f30 (0000503000001f90) (0 elements)
[uACPI][TRACE] fixed event 0 disabled successfully
stderr:
unexpected error: returned value '1' doesn't match expected '0'
[30/65] mutex-1.aml:Automatic mutex release OK in 0.01s
[31/65] indices-0.aml:Buffer Indices OK in 0.02s
[32/65] table-overrides.aml:Table overrides work OK in 0.01s
[33/65] store-copies-buffer.aml:Store Copies The Buffer OK in 0.01s
[34/65] eval-supports-plain-objects.aml:Eval supports plain objects OK in 0.02s```i made one of the tests fail on purpose and that's how it looks like, i think it's okay
[26/65] empty-objects.aml:Empty objects behave correctly FAILED in 0.01s
STDOUT FOR empty-objects.aml:Empty objects behave correctly:
[uACPI][INFO] RSDP 0x00007C5179F03D90 00000024 v02 (uOEMID)
[uACPI][INFO] XSDT 0x0000504000001090 0000002C v02 (uTEST uTESTTBL)
[uACPI][INFO] DSDT 0x0000508000000220 00000055 v02 (uTEST TESTTABL)
[uACPI][INFO] FACP 0x00005120000004C0 00000114 v06 (uOEMID uTESTTBL)
[uACPI][INFO] FACS 0x0000506000000440 00000040
[uACPI][INFO] SSDT 0x0000000000000000 00000055 v01 (uTEST RUNRIDTB)
[uACPI][TRACE] load of 'DSDT' (OEM ID 'uTEST ' OEM Table ID 'TESTTABL')
[uACPI][TRACE] load of 'SSDT' (OEM ID 'uTEST ' OEM Table ID 'RUNRIDTB')
[uACPI][TRACE] [AML DEBUG] String => "TestRunner ID SSDT loaded!"
[uACPI][INFO] successfully loaded 2 AML blobs, 7 ops in 0ms (avg 28497/s)
[uACPI][TRACE] initialized GPE block _GPE[0->128], 0 AML handlers (IRQ 0)
[uACPI][TRACE] initialized GPE block _GPE[128->256], 0 AML handlers (IRQ 0)
[uACPI][TRACE] fixed event 0 enabled successfully
[uACPI][TRACE] activated all 'EmbeddedControl' opregions controlled by '\ ', 0 _REG() calls (0 errors)
got a GPE
[uACPI][TRACE] activated all 'PCI_Config' opregions controlled by '\ ', 0 _REG() calls (0 errors)
[uACPI][INFO] namespace initialization done in 0ms: 0 devices, 0 thermal zones
[uACPI][TRACE] _STA calls: 0 (0 errors), _INI calls: 0 (0 errors)
[uACPI][TRACE] [AML DEBUG] Uninitialized
[uACPI][TRACE] [AML DEBUG] Package @0000503000001bd0 (0000503000001c30) (0 elements)
[uACPI][TRACE] [AML DEBUG] Package @0000503000001f30 (0000503000001f90) (0 elements)
[uACPI][TRACE] fixed event 0 disabled successfully
STDERR FOR empty-objects.aml:Empty objects behave correctly:
unexpected error: returned value '1' doesn't match expected '0'
[27/65] table-loading-0.aml:Load tables from buffers OK in 0.03s```fixed the indentation, let me pr this stuff
frank canopy
how badly will uacpi break if my mutex lock and event wait dont implement timeouts? a 0 timeout to just test is implemented but timing out a blocking wait is not
left orbit
i don't imagine it will break, just mutexes that should time out won't
i guess that is the definition of it breaking
frank canopy
lol
if itll just maybe wait longer than expected in some cases then i can probably just plug it in yeah
left orbit
"1 workflow awaiting approval" 
brb i gotta go to the store, pls run the ci when u can @fiery turtle
nvm not going yet
torpid root
how did you people improve the scores so much
north holly
how did you people improve the scores so much
uacpi update dropped
torpid root
nice
frank canopy
ill have to re-do my benchmarking too tbh
median crest
Once I finish cooking up the new locks I will update uacpi
gentle peak
well the performance improving update is currently still in a pr, not on master
fiery turtle
https://github.com/UltraOS/uACPI/pull/107
daaamn
really cool work
ill review in a bit
left orbit
sounds good
fiery turtle
where the f is the run ci button
rustic compass
i found a bug in uacpi, one case in status to string is missing
left orbit
If you are comfortable with running workflows on the pull request branch, return to the Conversation tab, and under "Workflow(s) awaiting approval", click Approve and run.
oh, i see that i's missing
that is really weird i guess github moment
fiery turtle
uhhh lmao
fiery turtle
i found a bug in uacpi, one case in status to string is missing
which one?
rustic compass
which one?
InvalidTableLength
rustic compass
status.h done, only 16 headers to go
fiery turtle
thanks for the report
fiery turtle
status.h done, only 16 headers to go
left orbit
<https://github.com/orgs/community/discussions/119459#discussioncomment-10228279...
no reply lol
maybe you can go there and approve it that way?
fiery turtle
that worked..
fiery turtle
how did u find it lol
fiery turtle
left orbit
i figured since the workflow was created it must be there
it just need to be approved
oh so the binary operator goes on the new line
fiery turtle
u have access to these right?
fiery turtle
nice
left orbit
single error
fiery turtle
yeah
left orbit
github kinda dumb
fiery turtle
and no button again
left orbit
great ui github
fiery turtle
indie website
left orbit
really went downhill after msft acquisition
fiery turtle
yup
left orbit
WHAT
BREAK AFTER BINARY OPERATOR BAD
BREAK BEFORE BINARY OPERATOR BAD
WHAT AM I SUPPOSED TO DO
fiery turtle
loool
fiery turtle
lemme see what its complaining about
left orbit
print(f"[{pass_count}/{len(cases)}] {case.name} "
+ f"{status_str} in {elapsed:.2f}s", flush=True)```this
- at the end of first line was bad
and this is bad too
fiery turtle
u dont need a plus there do u not?
left orbit
i need it because if i put it on one line it complains the line is too long
fiery turtle
>>> x = "hello" " world"
>>> x
'hello world'
pusehd
frank canopy
wtf python
left orbit
i like how the ones that failed are still running
frank canopy
this sort of nonsense is why I despise python lmao
left orbit
ok type checking fails, makes sense
flat badge
this sort of nonsense is why I despise python lmao
are you aware that C also does this? lol
rustic compass
@fiery turtle what happens if
uacpi_object_type uacpi_object_get_type(uacpi_object*);
gets a object with a type that is not present in the enum uacpi_object_type?
fiery turtle
thats not really possible
left orbit
ill install mypy locally and see if all works
frank canopy
are you aware that C also does this? lol
i dont use c lmao. ive not even used c++ since 2014ish either
fiery turtle
u can panic in this case if tahts appropriate in rust
frank canopy
did not actually know c also does that specific case though
flat badge
In any case, it's a feature in many langs to break up multi line strings
fiery turtle
its very useful in C
frank canopy
whats it good for in C? now im curious
frank canopy
ah yeah i can imagine its nice for macros
fiery turtle
e.g. make_str(arg0) "prefix" #arg0 "postfix"
left orbit
i should have tested it all locally anyway lol mb
frank canopy
make gh work for its living lmao
fiery turtle
lets see how long windows takes, it used to take 20 mins
fiery turtle
nice
fiery turtle
ubuntu took 6 minutes
left orbit
linux should see a bigger speed up with 4
oh 6 minutes is kinda slow
how long did it usually take?
fiery turtle
oh idk
left orbit
i should have added a print but oh well
fiery turtle
wait uh
Run python3 /Users/runner/work/uACPI/uACPI/tests/run_tests.py --bitness=64 --large
i dont see paralellism here
is it done automatically?
fiery turtle
ohh
left orbit
you can specify --parallelism N manually if you want
it does the large tests in 29.29s for me locally with parallelism=24
fiery turtle
oh yeah i didnt see default=os.cpu_count() or 1
left orbit
so hopefully it speeds up your local workflow too
fiery turtle
it definitely will
left orbit
i think they overprovision their vms
so while technically you get 4 cores you don't get 4x the cpu time you get with 1 core
but i'm also not sure if they give you 4 cores
fiery turtle
yeah probably
left orbit
they all do wtf
fiery turtle
i guess they're just not 100% of the cpu
left orbit
for me large tests finish in ~100s on 4 cores
SUMMARY: 359/359 in 131.25s (ALL PASS!) on ubuntu
for 32 bit runner
SUMMARY: 359/359 in 121.25s (ALL PASS!) on windows for 64 bit runner
fiery turtle
nice, I'll clone locally and give it a try, see if i can find nits perhaps (
)
left orbit
amazing
also run_tests builds with ninja if it can, gives a nice speed up on my machine
fiery turtle
i think i have ninja so it should pick that up automatically right
left orbit
yeah as long as you nuke tests/runner/build-*
because that is already configured with makefiles
fiery turtle
yeah
left orbit
SUMMARY: 359/359 in 112.19s (ALL PASS!) locally with parallelism=4
fiery turtle
those ci jobs also install dependencies and stuff
that could be taking some time too
frank canopy
yep
left orbit
oh yeah of course
fiery turtle
ye
left orbit
that 133.12s is from when it starts the first test to the end of the last test so it's purely compute bound
doesn't include compiling aml
could also be that fork+exec is slow-ish in a vm too, idk
fiery turtle
never a bad thing :)
indeed
fiery turtle
i'll look at sso in a bit, i still think it has potential
I think pitust has a working version?
although its private
left orbit
although its private <:meme:575445094589661215>
exactly
and i think it should make a noticeable difference
getting rid of all the allocations is big
and it should able to hold up to 14-15 bytes
left orbit
unfortunately no such big improvements for 32 bit builds, since strings are 8, not 16 bytes
fiery turtle
yeah
left orbit
but most targets are 64 bit anyway
left orbit
ok i go to the store and when i come back i do sso
fiery turtle
omg it detected the fact that i have ninja on windows and found some old ass garbage clang
instead of MSVC
is there any way to tell it to prefer msvc?
north holly
And you also need to do this before the first project statement
north holly
i want ninja to just prefer msvc if it exists
Ninja doesn't deal with setting the compiler afaik
fiery turtle
this already works correctly if u dont force -G ninja
fiery turtle
I think just not doing -G ninja if it detects os() == Windows is good enough
i want a solution file on windows anyway
not a ninja file
fiery turtle
I think just not doing -G ninja if it detects os() == Windows is good enough
czapek can we just maybe skip the ninja existance test on windows ig?
left orbit
i want ninja to just prefer msvc if it exists
its not ninja at fault but cmake
left orbit
czapek can we just maybe skip the ninja existance test on windows ig?
absolutely
fiery turtle
its not ninja at fault but cmake
yeye
fiery turtle
absolutely
amazing
left orbit
could you so that?
do*
im not home
and you can push to my branch
or merge and push to master if you want
fiery turtle
I tested on wsl and looks great
fiery turtle
ill be home in about 30 min
Sounds great
gentle peak
omg it detected the fact that i have ninja on windows and found some old ass gar...
afaik the compiler detection is completely separate from build system detection
left orbit
also cmake probably picks up clang from path
fiery turtle
I guess the default generator on windows just picks up MSVC
left orbit
yeah it should 100% given you want to test msvc
fiery turtle
Probably visual studio just does MSVC because its biased
left orbit
thats the default yeah
frank canopy
yep
cmake's default output is a solution file on windows, which then gets compiled by msbuild iirc? but using msvc
(i used to be the designated person who has windows for a cmake-based project which was meant to use mingw instead of msvc and fought with it for quite a time to get it all working just right)
hollow elm
set(CMAKE_CXX_COMPILER "cl")
what's a cmake toolchain file
frank canopy
yeah was about to say lol
you can use ninja with msvc on windows but youll need to do all the toolchain stuff manually
frank canopy
also the pci attach api is possibly gonna be real nice to have tbh
itll let me map things onto the same setup ive already got for my primary pci access functions
north holly
what's a cmake toolchain file
Just sets platform specific stuff
hollow elm
and the compiler to use etc
north holly
Ye
flat badge
pci attach api?
hollow elm
yeah so why shove that into the main cmake file
north holly
yeah so why shove that into the main cmake file
He was asking for a workaround
frank canopy
pci attach api?
in the same pr as making calloc optional; kinda like the io-map stuff but for pci stuff. most useful because it lets me attach the ECAM region pointer once instead of finding it again each time
because looping through the mcfg stuff each time is slow
fiery turtle
pci attach api?
flat badge
Ah i see
frank canopy
i may actually refactor my pci access functions to take a struct instead of a bunch of arguments because ive got those arguments as basically a struct stored in most of the places it gets accessed
either that or ill keep to working on the memory manager that i keep getting distracted from by all the uacpi things
fiery turtle
noo dont u want to spend countless hours doing random optimizations to get extra uacpi points
frank canopy
no but i do want to get more devices working
also i still need to figure out if the uacpi event thing is closer to a KEVENT or to a semaphore because i do not know linux apis and thus event_fd means nothing to me and the docs are as usual hard to read for linux things like that
north holly
noo dont u want to spend countless hours doing random optimizations to get extra...
Alas, we have been lead astray by thy, Oh Infy. Thou hast lead us to do countless hours of optimizing due to thy project
fiery turtle
also i still need to figure out if the uacpi event thing is closer to a KEVENT o...
whats kevent
gentle peak
also i still need to figure out if the uacpi event thing is closer to a KEVENT o...
it's just a semaphore
frank canopy
whats kevent
windows nt event object. can be either auto-reset or manual-reset
fiery turtle
like
frank canopy
the all caps is because its a type in the NT kernel so its always all caps
frank canopy
so its a semaphore then ok
fiery turtle
if its value 0 reads block
if its value is greater than 0, u decrement and instantly return
u can also increment this value
frank canopy
my brain operates entirely on windows terminology so i kept seeing event and getting hung up expecting the windows event object lol
winter orbit
the all caps is because its a type in the NT kernel so its always all caps
Microsoft 
fiery turtle
im not sure if kevent is similar
vast kestrel
In the windows world event is a bit different to a semaphore
fiery turtle
but sounds like it is
frank canopy
signaling a signaled event is a no-op for what windows calls events
vast kestrel
And given acpi comes from Microsoft it would make sense to be similar, even if it isn't in practice
frank canopy
it doesnt keep a counter of how many times its signalled
vast kestrel
^
fiery turtle
i got choccy milk
not bad
fiery turtle
it doesnt keep a counter of how many times its signalled
yeah thats just a binary semaphore
frank canopy
oh heck yeah choccy milk is good
fiery turtle
aka a mutex
fiery turtle
i think
frank canopy
aka a mutex
no because theres events that dont reset to unsignalled on wait
vast kestrel
And in general the wait can return with a different state (auto-reset vs manual-reset)
frank canopy
yeah
vast kestrel
AutoReset is basically using a mutex for an event
frank canopy
theres two kinds of events, one resets to unsignalled if you wait it and acts more like a mutex (though mutexes also exist and are separate on windows) and manual reset ones dont reset so waits on a signalled one are no-ops
yeah that
fiery turtle
linux eventfd is also the former example
frank canopy
the kernel also calls them synchronizationevent and notificationevent whereas windows usermode calls them autoresetevent and manualresetevent
fiery turtle
workflow waiting for approval
i hope it will auto-approve if u make other prs in the future
frank canopy
oh and the function to signal a windows event also has an option to not lower irql when it returns so you can do a non-preemptable signal-and-wait with it
left orbit
this is for new contributors only to prevent malicious prs i guess
or maybe to prevent people from spamming ci on popular repos
but not like they cant do it on their own public repos if they want to harm github/microsoft
tooootally hasn't been done already
fiery turtle
yeah idk its dumb
fiery turtle
<:pointandlaugh:978936313942245376>
aight ill 
once its done with ci
left orbit
lfg
flat badge
yeah thats just a binary semaphore
a binary semaphore can be used to implement a mutex but it's not exactly the same, right?
fiery turtle
is it?
left orbit
time to drink some choccy milk and give sso a go
flat badge
unlock -> lock is usually not defined for a mutex
fiery turtle
time to drink some choccy milk and give sso a go
lmao
left orbit
btw have you looked at zig? :D
fiery turtle
thats a good combo
left orbit
im curious
fiery turtle
btw have you looked at zig? :D
not yet but i should rn
i had some pr churn going
wrapping up for uacpi 1.0
winter orbit
which will come first uACPI 1.0 or dark finishing all 16 major device drivers 1607 implements in ACPI.sys 
fiery turtle
how many callbacks is that
fiery turtle
how big is one callback?
winter orbit
I haven’t counted all of them
winter orbit
how big is one callback?
Smol
frank canopy
which will come first uACPI 1.0 or dark finishing all 16 major device drivers 16...
now im curious what the 16 are
fiery turtle
im predicting at least a week for uacpi 1.0
winter orbit
let me post the wall
fiery turtle
im curious
btw i can move the template under ultra if u want lol
winter orbit
"ACPI0007"
"FixedButton"
"PNP0000"
"PNP0001"
"PNP0002"
"PNP0003"
"PNP0004"
"PNP0100"
"PNP0101"
"PNP0102"
"PNP0200"
"PNP0201"
"PNP0202"
"PNP0800"
"PNP0A05"
"PNP0A06"
"PNP0B00"
"PNP0C00"
"PNP0C01"
"PNP0C02"
"PNP0C04"
"PNP0C05"
"PNP0C09"
"PNP0C0B"
"PNP0C0C"
"PNP0C0D"
"PNP0C0E"
"PNP0C32"
"SNY5001"
"IBM0062"
"DockDevice
"ThermalZone"
"Processor"
there
Only 16 of these matter though
screw dockdevice
left orbit
btw i can move the template under ultra if u want lol
up to you, if you don't mind occasional zig breakages lol
before that happens tho i'd like a proper ultra protocol binding thing
fiery turtle
i mean i can make u the admin of the repo right
left orbit
yeah
fiery turtle
ah ok
winter orbit
does uACPI actually stand for Ultra
fiery turtle
it micro and ultra
fiery turtle
lmao
left orbit
uNIX when
winter orbit
uNT
winter orbit
GNU/uNIX
fiery turtle
GNu
winter orbit
that’s better actually
slim panther
ultra digital rights management
frank canopy
so processor, fixed button, all five types of interrupt controller, AT EISA and MCA timers, all three DMA controller types, AT-style speaker sound, generic acpi bus, generic acpi eio bus, AT RTC, "plug and play bios", system board, generic reserving resources, math coprocessor, apm bios, embedded controller, acpi fan, power button, lid, sleep button, HIDACPI button, dock, thermal zone, processor, and idk what sny5001 and ibm0062 are
idk why i have the devids file open always but i do
left orbit
ci about to be green 🔥
winter orbit
some of
Them are just empty IRP handlers yeah
winter orbit
Each one of those create a unique PDO though 
frank canopy
so when is pr 106 being merged into uacpi i want it
winter orbit
And MS shoved it all inside ACPI.sys
I thought windows 11 would be better but it’s architected exactly the same
fiery turtle
ci about to be green 🔥
frank canopy
that reminds me, is RTC guaranteed to be in acpi somewhere?
fiery turtle
thanks a lot, based work
left orbit
thanks 🎉
fiery turtle
that reminds me, is RTC guaranteed to be in acpi somewhere?
well idk about guaranteed, its firmware afterall
but usually yes
frank canopy
eh true
fiery turtle
and it even exposes methods to read it
left orbit
your contributor list on the main page is no longer a list, just a bunch of avatars
fiery turtle
so when is pr 106 being merged into uacpi i want it
later today
fiery turtle
- firmware dev
lmao
frank canopy
but usually yes
i just want to know if i need to bother with figuring a way to make drivers do a root device if they dont get one from enumeration for rtc lol
frank canopy
and it even exposes methods to read it
wait really? neat
fiery turtle
if a firmware dev tells u not to trust them u better listen
winter orbit
yeah we’re terrible people
vagrant hull
<:merged_pull:947239828045582416>
that's a good emote we should steal
fiery turtle
wait really? neat
yeah
fiery turtle
that's a good emote we should steal
how about 
and 
frank canopy
yeah
what are they called? went looking in my acpi dump and not finding anything really
winter orbit
ok I need to lock in I keep getting distracted by discord and our crappy kernel pushlock code
fiery turtle
what are they called? went looking in my acpi dump and not finding anything real...
ill try to look at the dumps that i have and let u know, idr off the top of my head
fiery turtle
ok I need to lock in <:oof:401357910476718090> I keep getting distracted by disc...
lmao same
left orbit
oh god
fiery turtle
huh?
left orbit
that will end terribly
winter orbit
yeah there’s something seriously wrong with the way our .. environment builds uACPI
When I build it as standard C it crashes MSVC
fiery turtle
bruh
frank canopy
ill try to look at the dumps that i have and let u know, idr off the top of my h...
looks like theres an abstraction device with HID ACPI000E that has the methods but neither of my machines have it afaik
left orbit
least broken msvc build env
fiery turtle
looks like theres an abstraction device with HID ACPI000E that has the methods b...
_GRT
flat badge
which should in theory work on any platform (= also on arm servers etc)
fiery turtle
OperationRegion (RTCM, SystemCMOS, Zero, 0x3F)
Field (RTCM, ByteAcc, Lock, Preserve)
{
SEC, 8,
Offset (0x02),
MIN, 8,
Offset (0x04),
HOR, 8,
Offset (0x07),
DAY, 8,
MON, 8,
YEAR, 8,
Offset (0x32),
CNTY, 8
}
Method (_GRT, 0, NotSerialized) // _GRT: Get Real Time
{
Local0 = Buffer (0x10){}
CreateWordField (Local0, Zero, Y)
CreateByteField (Local0, 0x02, M)
CreateByteField (Local0, 0x03, D)
CreateByteField (Local0, 0x04, H)
CreateByteField (Local0, 0x05, MIN1)
CreateByteField (Local0, 0x06, S)
CreateByteField (Local0, 0x07, V)
CreateWordField (Local0, 0x0A, TZ)
CreateByteField (Local0, 0x0C, DL)
FromBCD (YEAR, Local5)
FromBCD (CNTY, Local6)
Y = ((Local6 * 0x64) + Local5)
FromBCD (MON, M) /* \_SB_.AWAC._GRT.M___ */
FromBCD (DAY, D) /* \_SB_.AWAC._GRT.D___ */
FromBCD (HOR, H) /* \_SB_.AWAC._GRT.H___ */
FromBCD (MIN, MIN1) /* \_SB_.AWAC._GRT.MIN1 */
FromBCD (SEC, S) /* \_SB_.AWAC._GRT.S___ */
TZ = 0x07FF
DL = Zero
V = One
Return (Local0)
}
Example
frank canopy
instead ive just got a PNP0B00 device that defines a decode16 io region and optionally a single irq (if hpet is disabled)
fiery turtle
about 30% of machines have it
left orbit
how can i disassemble the tables from an acpidump?
fiery turtle
how can i disassemble the tables from an acpidump?
see pins
left orbit
thanks
frank canopy
and sometimes it tries to assemble the aml as if its asl because it gets confused in my experiencce
winter orbit
yeah there’s something seriously wrong with the way our .. environment builds uA...
This file specifically
Will just blow up MSVC
winter orbit
oh I’ll run the sync script
fiery turtle
`Method (_GRT, 0, Serialized) // _GRT: Get Real Time`
1/3 chance, nice lol
left orbit
it also does _SRT
winter orbit
clang was stripping a comma in a macro
This sounds like something clang would do
left orbit
i wonder how insane my motherboard's aml is
fiery turtle
look around see if u can find a 50 conditions while loop
winter orbit
Only 50?
fiery turtle
While ((GP50 || (GP51 || (GP52 || (GP53 || (GP54 || (
GP55 || (GP56 || (GP57 || (GP60 || (GP61 || (GP62 || (GP63 ||
(GP66 || (GP67 || (GP70 || (GP71 || (GP93 || (GP94 || (
GPAP || (GPBP || (GPCP || (GPDP || (GPEP || (GPFP || (GPGP ||
(GPHP || (FCDP || (FTVP || (FQSP || (GP96 || (GP97 || (
GPD0 || (GPD1 || (GPD2 || (GPD3 || (GPD4 || (BPFE || (TPBP ||
(ECDS || (B1ST || B2ST)))))))))))))))))))))))))))))))))))))))))
The worst thing I’ve seen was on an nvidia product
fiery turtle
rookie aml
winter orbit
```
While ((GP50 || (GP51 || (GP52 || (GP53 || (GP54 || (
GP55 || (GP56 || (...
could be worse
winter orbit
Imagine bios developers started testing against uACPI
left orbit
no weird while loops but```
While (One)
{
If ((!IBSY || (IERR == One)))
{
Break
}
If ((Local0 > Local1))
{
RPKG [Zero] = 0x03
Return (RPKG) /* \IPCS.RPKG */
}
Sleep (One)
Local0++
}```
a possible infinite loop
fiery turtle
Imagine bios developers started testing against uACPI
id hope it would the same as testing against latest nt
fiery turtle
nah anything without a _ prefix is private so to speak
winter orbit
id hope it would the same as testing against latest nt
here’s hoping
Win11 doesn’t have a lot of extra stuff
left orbit
ah makes sense
fiery turtle
it's a method called IPCS in scope \
usually u can figure out either by whos calling it or unstripped debug logs
fiery turtle
yeah
left orbit
i ain't reading all that
fiery turtle
they usually have a custom method called ADBG or similar
that writes to proprietary debug regs
hollow elm
does uacpi let you capture Debug writes somewhere?
fiery turtle
does uacpi let you capture `Debug` writes somewhere?
only via kernel log atm
as trace level
hollow elm
ah
fiery turtle
can't find anything matching *DBG
❯ grep -rni 'ADBG' -l | wc -l
568
seems like most blobs have it
fiery turtle
guess u got the unlucky stripped debug logs blob
hollow elm
ssdt1.dsl
26: Name (ADBG, Buffer (0x0100){})
also lol ```
8215: Debug = "Fn+ESC Pressed"
8222: Debug = "Fn+F1 Pressed"
8229: Debug = "Fn+F2 Pressed"
8238: Debug = "Fn+F3 Pressed"
8247: Debug = "Fn+F4 Pressed"
8254: Debug = "Fn+F6 Pressed"
8261: Debug = "Fn+B Pressed"
@fiery turtle what is a uacpi_data_view and what exactly does it do/represents?
rustic compass
the union confuses me why 2 pointer? isnt a slice ptr + len?
fiery turtle
because the data might be char data for strings, or binary data for data buffer
its a union
so its just for convenience
rustic compass
no i mean:
uacpi_u8 *bytes;
const uacpi_u8 *const_bytes;
because this is used as an in param or out param sometimes
so its either writable or readable data
depending on where its used
its just an anonymous data view
rustic compass
so the union contains 4 pointers?
fiery turtle
1 pointer with 4 names
left orbit
you said it yourself, it's a union
it's the same pointer but with a different type
for use in different scenarios
rustic compass
i got confused by the empty line
fiery turtle
are you familiar with C unions?
i think for rust u just want to store one pointer and have like as_something() ig
rustic compass
i know the concept of a union
fiery turtle
idk how rust does it
fiery turtle
untagged?
left orbit
or you can at least have repr(C) unions
that do the same thing
but muh ub unsafe and all that crap
idk
fiery turtle
i think for the purposes of rust having this data structure literally translated would be not super idiomatic
left orbit
not at all, no
frank canopy
INFO (uacpi): successfully loaded 1 AML blob, 1705 ops in 2ms (avg 689201/s) welp
this is still old branch but also
left orbit
it's slow as balls
uacpi does a lot of allocations that fit buckets 5-10
and it needs to create a lot of them
so every 30 or so allocations you end up creating a new bucket
and that is slow-ish
because it needs to insert it into the treap
and all that
fiery turtle
this is still old branch but also
is this the api-changes pr?
frank canopy
main branch
frank canopy
also the time in nanos computation hits a divide by 0 exception if i use tcg since i only use hypervisor leaves for calibration lmao
my pmm is singly linked lists of free pages with each next-pointer stored in the first bytes of the free page itself, using an hhdm to access it
im currently in the middle of redoing all of my memory manager stuff though
left orbit
that should be good enough
frank canopy
yeah im 100% sure the bottleneck is GPA
but also im not really going for records here
am a bit surprised though, this is lower perf then last time i tested it lmao
lemme try kvm on wsl2 instead of hyperv accel
frank canopy
on wsl2 with kvm accel i get 1669364/s, on windows with whpx accel im now getting 1006173/s and idk what changed
winter orbit
MSVC is NOT beating the insane allegations
fiery turtle
how is msvc subject to the same bug as clang on windows
winter orbit
that's how we build 32bit MSVC ros
fiery turtle
i wonder if its some libc bug
winter orbit
how is msvc subject to the same bug as clang on windows
it's insane
frank canopy
interestingly on kvm accel my actual device enumeration takes like 5-10 seconds to process but uacpi is ~75% faster to load aml compared to whpx/hyperv
whereas on whpx the device enumeration is fast enough that i dont notice it but uacpi is slower
fiery turtle
it's insane
and i compile with msvc all the time and it works fine, but my msvc is latest so thats why it works probably
winter orbit
nah this is latest we just do some weird stuff
fiery turtle
ah
frank canopy
either way on kvm accel i now get consistent 1.6-1.7mil ops/s
winter orbit
we actually have targets for clang, gcc and msvc at once
fiery turtle
either way on kvm accel i now get consistent 1.6-1.7mil ops/s
nice
left orbit
either way on kvm accel i now get consistent 1.6-1.7mil ops/s
wtf what did u do
fiery turtle
we actually have targets for clang, gcc and msvc at once
nice
left orbit
is interpreter.c the only place where string/buffer objects are created/used?
winter orbit
nice
we also support VS2017
but that's likely getting dropped
with this being imported
frank canopy
qemu-system-x86_64 -drive format=raw,file=drive.bin -accel kvm -d int,cpu_reset,guest_errors --no-reboot -M type=q35,smm=off,hpet=on -cpu qemu64,+invtsc,+pdpe1gb,+rdrand,+arat,+rdseed,+x2apic -m 4G -debugcon file:aaa.txt is my image and cli for anyone interested in comparing results
left orbit
amazing
fiery turtle
but that's likely getting dropped
will 2022 work?
winter orbit
yes
fiery turtle
ah ok based
winter orbit
this is 2022
winter orbit
2019 and 2022 are probably all we will support going forward
left orbit
speed
frank canopy
ok i guess my computer is just slow
left orbit
what cpu?
frank canopy
if others are getting double the perf
fiery turtle
INFO (uacpi): successfully loaded 1 AML blob, 1705 ops in 0ms (avg 3282507/s)
this is close to linux performance btw
but u have most stuff stubbed right
fiery turtle
yeah
once u implement those it will tank
not by a lot if u make a good mutex impl
frank canopy
and schedule/wait for work returns status unimplemented rn
fiery turtle
and schedule/wait for work returns status unimplemented rn
that stuff isnt used by init code luckily
left orbit
not used during init/ns load
frank canopy
yeah
left orbit
is interpreter.c the only place where string/buffer objects are created/used?
pong
it looks like it but i wanna make sure
fiery turtle
nope
left orbit
ah rip i gotta read more code then
fiery turtle
types.c via user api
left orbit
oh that is gonna be so fun
fiery turtle
create_object etc
fiery turtle
oh that is gonna be so fun
lmao
left orbit
are those objects opaque outside uacpi?
fiery turtle
yeah users cant mess with anything directly
left orbit
good
fiery turtle
there are getters for the string/buffer view
the struct u saw earlier
but its a view so its up to u what u set the pointer to
left orbit
alright then i gotta do it that way
fiery turtle
so far it works with O3 + LTO so no ub crimes have been commited by uacpi
at least not ones that would break instantly
frank canopy
actually yknow what, i literally have semaphore and mutex already, why not actually just plug the dang things into uacpi lmao
left orbit
// This is private API, but we have to use it for tests here.
extern "C" {
#include <uacpi/internal/resources.h>
}
makes sense
fiery turtle
trolled
winter orbit
fiery turtle
yea
fiery turtle
<:kongoudisgust:575445162705158156>
cmon its not that bad, this is a userspace cpp file that does resource conversion tests
fiery turtle
.data and .inline_data
makes sense
winter orbit
cmon its not that bad, this is a userspace cpp file that does resource conversio...
its ok
its not as bad as this
fiery turtle
lmao
fiery turtle
these comments remind me of pre-corporate linux
winter orbit
fiery turtle
now they arent even allowed to mention MS
frank canopy
welp i had to setup early table access to plug in sync stuff and now thats got a gp fault
fiery turtle
F
where does it happen
I forgor to rebase api-changes on top of iretq's changes and now i have to wait forever for ci 💀
frank canopy
i got interrupted before i could check the gp fault, checking now
rustic compass
@fiery turtle do apis that take uacpi_object ptr decrement its refcount when they drop it internally?
fiery turtle
like which
unless there's a comment saying otherwise the refcount is the same as it was pre-call
well and for uacpi_eval's arguments specifically, AML can take a RefOf(obj) internally and store it in some global variable
so thats something to be aware of
thats why uacpi doesnt allow stack-allocated objects
rustic compass
well rust likes its stack variables
frank canopy
ok why the heck is this a gp fault and not a page fault
fiery turtle
so when is pr 106 being merged into uacpi i want it
frank canopy
🎉 cool
fiery turtle
for anyone who wants free uacpi points u may want to update now :^)
(should be faster now because dynamic array grows by factors)
rustic compass
@fiery turtle
typedef uacpi_status (*uacpi_region_handler)
(uacpi_region_op op, uacpi_handle op_data);
typedef uacpi_status (*uacpi_notify_handler)
(uacpi_handle context, uacpi_namespace_node *node, uacpi_u64 value);
what the fuck is that?
fiery turtle
handlers for address space io and Notify requests
or do u mean as in what this means in C
rustic compass
yes
fiery turtle
it typedefs a function pointer type
typedef uacpi_status (*uacpi_region_handler)
(uacpi_region_op op, uacpi_handle op_data);
Function pointer type called uacpi_region_handler with signature uacpi_status foo(uacpi_region_op op, uacpi_handle op_data)
rustic compass
so basicly:
type Binop = fn(i32, i32) -> i32;
that
frank canopy
ok i feel kinda dumb now but at least i found my bug
turns out when i allocate something i should also initialize the struct contents
im down from 1.6->1.4 mil/s by adding mutex and semaphore stuff
fiery turtle
still pretty good
frank canopy
qemu-system-x86_64 -drive format=raw,file=drive.bin -accel kvm -d int,cpu_reset,guest_errors --no-reboot -M type=q35,smm=off,hpet=on -cpu qemu64,+invtsc,+pdpe1gb,+rdrand,+arat,+rdseed,+x2apic -m 4G -debugcon file:aaa.txt is my image and cli for anyone interested in comparing results (its the same cli)
thats still old branch btw
ive not pulled in new master
gentle peak
updated the leaderboard update message
fiery turtle
gentle peak
i should really add proper uacpi<->proxima error code conversion functions
rustic compass
your function pointer shenanigan is causing me real headache, i probably need rust macros and i never wrote one
fiery turtle
why is that
rustic compass
my current idea is that i create a Region/NotifyHandler function wrapper that wraps the kernel supplied handler and does safe type conversion, but for that i need compile time code expansion aka macro
gentle peak
`qemu-system-x86_64 -drive format=raw,file=drive.bin -accel kvm -d int,cpu_reset...
i get 2.2-2.3M/s with this cmdline and image
frank canopy
cool
fiery turtle
my current idea is that i create a Region/NotifyHandler function wrapper that wr...
couldnt u just use a wrapper in rust to safely dispatch to different "safe" handlers?
or i guess u want it to be free in terms of overhead
frank canopy
upgraded to newest uacpi master and ive got 2993260, 3185403, 3112091 as first three runs under kvm (seems pretty consistent around 2.9-3.2 mil)
my newest bin, same cli as before
rustic compass
couldnt u just use a wrapper in rust to safely dispatch to different "safe" hand...
yes but i assume you can register Handlers for all sort of things so i need to handle an arbitrary amount of handlers
so i need a macro like register_RegionHandler!(something, handler)
which expands to code that registers and wraps the handler
frank canopy
thats with mutexes and semaphores implemented not stubbed
fiery turtle
yes but i assume you can register Handlers for all sort of things so i need to h...
true
gentle peak
my newest bin, same cli as before
i get 4M with this
fiery turtle
upgraded to newest uacpi master and ive got 2993260, 3185403, 3112091 as first t...
so like 50% speedup by just upgrading
rustic compass
so i have a internal function signature that is uacpi one just in rust and a public one that the kernel uses
fiery turtle
i get 4M with this
what the fuck
frank canopy
apparently the zig gpa isnt as slow as id thought
again thats got full semaphores and mutexes
fiery turtle
thats like second place now
frank canopy
only sleep and stall are still stubbed
fiery turtle
but a lot of people still havent updated ig
frank canopy
(well and schedule/wait for work but thats not relevant to init benchmarking)
gentle peak
considering everything on the leaderboard rn is with unupdated uacpi, it'd probably be in between astral and menix
fiery turtle
fair
frank canopy
we'll find out when more people update ig
either way im callin this good for now lmao
fiery turtle
very very solid yes
gentle peak
i wonder what it'd take to get proxima to 7M
fiery turtle
SSO perhaps
dusky glade
using the nyaux allocator perhaps
rustic compass
what is sso?
dusky glade
left orbit
i am half way through interpreter.c
dusky glade
slowly starting out using the nyaux allocator
fiery turtle
what is sso?
storing small strings inside the pointer
left orbit
types.c and utilities.c done
fiery turtle
big
dusky glade
slowly starting out using the nyaux allocator
real meaning btw
left orbit
storing small strings inside the pointer
i also use 7 bytes out of the length
fiery turtle
daamn
left orbit
leaving 7 bits for length and 1 bit for the flag
fiery turtle
makes sense
calm latch
I forgot to dump the tables 😦
left orbit
and now instead of accessing buf->data, buf->text and buf->size you use thesec #define UACPI_INLINE_DATA_MAX_SIZE sizeof(void *) + sizeof(uacpi_size) - sizeof(uacpi_u8) #define UACPI_BUFFER_IS_INLINE(buffer) ((buffer)->inline_data.is_inline) #define UACPI_BUFFER_DATA(buffer) (UACPI_BUFFER_IS_INLINE(buffer) ? \ (buffer)->inline_data.inline_data : (buffer)->data.data) #define UACPI_BUFFER_TEXT(buffer) (UACPI_BUFFER_IS_INLINE(buffer) ? \ (buffer)->inline_data.inline_char_data : (buffer)->data.text) #define UACPI_BUFFER_SIZE(buffer) (UACPI_BUFFER_IS_INLINE(buffer) ? \ (buffer)->inline_data.size : (buffer)->data.size)
calm latch
I forgot to dump the tables 😦
On an insane PC which broke Limine at some point
median crest
I really wanna see how fast my new locks will make it, uncontended it will be 1 cas for lock and unlock
compared to the shitshow it was before
fiery turtle
I really wanna see how fast my new locks will make it, uncontended it will be 1 ...
im excited as well
especially with new uacpi
fiery turtle
and now instead of accessing buf->data, buf->text and buf->size you use these```...
makes sense
calm latch
I think pmOS is now faster than menix
fiery turtle
lmao
calm latch
I should've advertised RISC-V as well
I was the first RISC-V OS to support uACPI I think
left orbit
does aml concatenation always result in a buffer/string?
fiery turtle
yes
gentle peak
if i do everything i can think of to improve perf (including stuff like stubbing mutexes, compiling with -march=native -mtune=native, etc) i can get the current uacpi commit to 7,832,670/s
fiery turtle
insane
calm latch
I need to try uACPI with AVX512
gentle peak
oh right i should remove -mgeneral-regs-only and see what that does
frank canopy
oh right i should remove -mgeneral-regs-only and see what that does
might invalid opcode if you arent careful
left orbit
me when int + int = string
gentle peak
might invalid opcode if you arent careful
yeah i gotta move basic cr4 init to my asm entry
left orbit
or is concatenation done using some special operator
fiery turtle
does aml concatenation always result in a buffer/string?
if u look at opcodes.h for concatenate, it does a UACPI_PARSE_OP_OBJECT_ALLOC_TYPED, UACPI_OBJECT_BUFFER,
left orbit
i have no idea what that means
fiery turtle
me when int + int = string
what would u expect in this case
fiery turtle
concatenate is Concatenate(123, 321)