00:15
stikonas_ has quit [Remote host closed the connection]
00:29
BenG83 has quit [Quit: Leaving]
02:02
megi has quit [Ping timeout: 240 seconds]
02:42
tgall_foo has joined #panfrost
02:45
tgall_foo has quit [Client Quit]
02:54
tgall_foo has joined #panfrost
03:25
Depau has joined #panfrost
04:48
chewitt has joined #panfrost
05:00
Lyude has quit [Quit: WeeChat 2.4]
05:01
Lyude has joined #panfrost
05:35
chewitt has quit [Quit: Adios!]
06:06
nerdboy has quit [Ping timeout: 265 seconds]
07:18
maccraft123 has joined #panfrost
07:29
guillaume_g has joined #panfrost
07:46
maccraft123 has quit [Ping timeout: 240 seconds]
07:48
<
tomeu >
alyssa: 0x1 for small fbs works just great :)
07:51
maccraft123 has joined #panfrost
07:54
guillaume_g has quit [Quit: Konversation terminated!]
07:57
guillaume_g has joined #panfrost
08:26
maccraft123 has quit [Quit: WeeChat 2.6]
08:28
maccraft123 has joined #panfrost
08:32
<
tomeu >
alyssa: looks like we need a mask of 0x82 for tiler jobs without a color fb
08:33
<
tomeu >
get a DATA_INVALID_FAULT otherwise
08:37
yann has quit [Ping timeout: 240 seconds]
08:40
maccraft123 has quit [Quit: WeeChat 2.6]
08:47
warpme_ has joined #panfrost
09:01
pH5 has joined #panfrost
09:40
davidlt has joined #panfrost
09:46
megi has joined #panfrost
09:56
yann has joined #panfrost
10:15
stikonas has joined #panfrost
10:36
maccraft123 has joined #panfrost
10:45
maccraft123 has quit [Quit: WeeChat 2.6]
10:50
Elpaulo has joined #panfrost
11:04
kaspter has quit [Remote host closed the connection]
11:04
kaspter has joined #panfrost
11:16
stikonas_ has joined #panfrost
11:16
stikonas has quit [Read error: Connection reset by peer]
11:26
stikonas_ is now known as stikonas
11:56
jernej has joined #panfrost
12:13
yann has quit [Ping timeout: 240 seconds]
12:27
yann has joined #panfrost
12:29
<
alyssa >
tomeu:
*checks notes*
12:29
<
alyssa >
tomeu: 0x82 correpsonds to 32x32 tiles
12:29
<
alyssa >
Sounds arbitrary :p
12:44
<
tomeu >
iirc, it was a 16x16 fb
12:45
<
alyssa >
Hrm indeed.
12:45
<
alyssa >
This is for depth-only jobs.
12:45
<
tomeu >
glmark2 -bshadow
12:51
<
alyssa >
tomeu: I mean I believe you it just seems arbitrary
12:51
<
alyssa >
Not that anything else about this GPU seems arbitrary at first blush.
12:51
<
alyssa >
tomeu: To be clear, though, that is a massive framebuffer
12:52
<
alyssa >
Maybe there's a limit on the number of tiles you can have on t720?
12:52
<
alyssa >
So if the framebuffer is too big, 16x16 isn't even an option?
12:52
<
alyssa >
but you mentioned it being like that for a 16x16 fb so uhm could you double check that?
12:53
<
alyssa >
Try passing --size 8x8 to glmark2-es2-wayland -bshadow
12:55
<
alyssa >
Note that (1920x1080/16x16) = (3840x2160/32x32) = 8100
12:55
<
alyssa >
That means the cutoff is somewhere between
12:56
<
alyssa >
And I know how much people like nice round numbers, so I would wager I guess 8192 is the max number of titles
12:57
<
alyssa >
You can confirm this by rendering (anything -- -bbuild work would) with respectively --size 1941x1080 and --size 1943x1080
12:57
<
alyssa >
lmk the diff
13:03
<
alyssa >
Disclaimer: not even compile tested.
13:04
<
alyssa >
(update: it does compile)
13:18
guillaume_g has quit [Quit: Konversation terminated!]
13:27
maccraft123 has joined #panfrost
13:30
<
tomeu >
alyssa: that one fixed 50 tests :)
13:30
<
tomeu >
will do the tiling experiments
13:40
guillaume_g has joined #panfrost
13:42
davidlt has quit [Ping timeout: 240 seconds]
13:44
<
tomeu >
alyssa: hmm, --size doesn't seem to take effect on -drm
13:44
<
tomeu >
and I don't have a wayland setup for the blob
13:45
<
tomeu >
I should be able to dump some different sizes with deqp-gles2 and the surfaceless backend
13:46
<
alyssa >
tomeu: Oh, yeah, -drm won't let you scale easily
13:46
<
alyssa >
but yes deqp would
13:46
<
alyssa >
tomeu: And ey nice :)
13:46
<
alyssa >
Should I send an MR then?
13:50
maccraft123 has quit [Quit: WeeChat 2.6]
13:52
kaspter has quit [Ping timeout: 265 seconds]
13:57
maccraft123 has joined #panfrost
14:01
<
tomeu >
alyssa: got the same mask (0x1c7) for both 1941 and 1943
14:05
<
alyssa >
tomeu: You need a scene with a ton of geometry
14:05
<
alyssa >
I'm not sure which in deqp would do that
14:05
<
alyssa >
glmark does it plenty but uhm
14:14
<
alyssa >
warning: major hax.
14:14
<
alyssa >
and then wait uhm
14:15
<
alyssa >
Yeah then you'll get a ton of geom in dEQP-GLES2.functional.read_pixels.*
14:15
<
alyssa >
which should get the masks where they need to be.
14:15
<
alyssa >
scientific method \o/
14:24
<
tomeu >
sounds good!
14:27
davidlt has joined #panfrost
14:33
<
tomeu >
alyssa: getting 0x41 in both cases (1941 and 1943)
14:49
davidlt has quit [Ping timeout: 246 seconds]
14:55
stikonas has quit [Remote host closed the connection]
14:57
stikonas has joined #panfrost
14:58
stikonas has quit [Remote host closed the connection]
15:00
stikonas has joined #panfrost
15:00
<
alyssa >
tomeu: Hrm! :V
15:02
<
alyssa >
I guess we know the limit isn't 8192 not then
15:02
<
alyssa >
if there is a limit
15:03
<
alyssa >
tomeu: What about like 3800x1080 and 3900x1080?
15:13
maccraft123 has quit [Quit: WeeChat 2.6]
15:18
<
tomeu >
alyssa: both 0x42
15:18
<
tomeu >
should I bisect to where it changes from 0x41 to 0x42?
15:20
<
tomeu >
or do you have a better guess?
15:29
<
tomeu >
alyssa: the blob starts using 0x42 at 2049x1080
15:41
<
tomeu >
well, using 0x42 seems to work quite well, and even gets -bterrain kind of working
15:41
<
tomeu >
at least, it's not getting the OUT_OF_MEMORY fault any more
15:42
<
tomeu >
no faults at all in a whole glmark2 run, awesome :)
15:53
<
alyssa >
>>> (2049*1080)/(16*16)
15:53
<
alyssa >
8644.21875
15:53
<
alyssa >
Sounds arbitrary :P
15:53
<
alyssa >
I thhink more likely is just that, well, 2048
15:53
adjtm has quit [Ping timeout: 250 seconds]
15:56
<
alyssa >
That is ... let's see, we have up to 2048x2048 with 16x16 tiles
15:56
<
alyssa >
In one dimension, even
15:57
<
alyssa >
up to 2048 with 16 tiles
15:57
<
alyssa >
but above 32 tiles
15:57
<
alyssa >
And I would predice if you did a framebuffer with 4097x1080 you would get a mask of 0x43, yes?
16:11
<
alyssa >
If so that suggests rules that in any given dimension, like,
16:12
<
alyssa >
the tiles must be no bigger than the framebuffer, and there must be no more than 128 tiles in each dimension
16:12
<
alyssa >
In other words, for the latter condition in a given dimension you have a calculation like:
16:13
<
alyssa >
>>> 2 ** math.ceil(math.log2(dimension) - math.log2(128))
16:13
<
alyssa >
^ To calculate the minimum tile size ... even without paying attention to the amount of geometry per tile, you can clamp the minimum size to that
16:13
<
alyssa >
(Implementation details left as an exercise to the curious reader.)
16:14
<
alyssa >
Also note that simplifies to:
16:14
<
alyssa >
>> 2 ** (math.ceil(math.log2(dimension)) - 7)
16:15
<
alyssa >
And we happen to have this wonderful little function called `util_logbase2_ceil` ;)
16:26
davidlt has joined #panfrost
16:40
yann has quit [Ping timeout: 265 seconds]
17:22
nerdboy has joined #panfrost
17:33
pH5 has quit [Quit: bye]
17:35
warpme_ has quit [Quit: Connection closed for inactivity]
17:40
guillaume_g has quit [Quit: Konversation terminated!]
17:57
maccraft123 has joined #panfrost
17:58
yann|work has joined #panfrost
18:00
davidlt has quit [Ping timeout: 240 seconds]
18:16
maccraft123 has quit [Quit: WeeChat 2.6]
18:21
maccraft123 has joined #panfrost
18:24
pH5 has joined #panfrost
19:47
davidlt has joined #panfrost
19:50
adjtm has joined #panfrost
20:23
maccraft123 has quit [Quit: WeeChat 2.6]
20:26
maccraft123 has joined #panfrost
20:51
robert_ancell has joined #panfrost
21:11
davidlt has quit [Ping timeout: 265 seconds]
21:15
maccraft123 has quit [Quit: WeeChat 2.6]
21:53
maccraft123 has joined #panfrost
22:07
maccraft123 has quit [Quit: WeeChat 2.6]
22:53
pH5 has quit [Remote host closed the connection]
23:00
stikonas has quit [Read error: Connection reset by peer]
23:01
stikonas has joined #panfrost