00:21
jolan has joined #panfrost
00:32
_whitelogger has joined #panfrost
00:50
tlwoerner_ has quit [Ping timeout: 255 seconds]
00:56
<
alyssa >
Hm, RA bugs breaking a ton of looping-related tests
00:56
<
alyssa >
(This is connected to the hangs fwiw)
01:23
_whitelogger has joined #panfrost
01:25
TheCycoONE has joined #panfrost
02:20
<
HdkR >
I like the spelling mistake in the first header
02:23
<
HdkR >
I used to sit in on the openhours meetings. Sort of stopped
02:59
vstehle has quit [Ping timeout: 245 seconds]
04:53
_whitelogger has joined #panfrost
05:00
vstehle has joined #panfrost
05:18
LinguinePenguiny has joined #panfrost
06:04
fysa has quit [Ping timeout: 250 seconds]
06:06
fysa has joined #panfrost
06:27
LinguinePenguiny has quit [Ping timeout: 244 seconds]
07:24
stikonas has joined #panfrost
10:10
mateo` has quit [Ping timeout: 245 seconds]
10:24
mateo` has joined #panfrost
11:02
_whitelogger has joined #panfrost
13:01
<
paulk-leonov >
"What’s Panfros?" mhh
13:47
stikonas has quit [Remote host closed the connection]
13:58
TheCycoONE has quit [Ping timeout: 268 seconds]
14:30
TheCycoONE has joined #panfrost
15:35
stikonas has joined #panfrost
15:59
* alyssa
grumbles about register allocation
16:17
* alyssa
grumbles about loops
17:54
afaerber has joined #panfrost
17:54
afaerber has quit [Remote host closed the connection]
18:45
* alyssa
grumbles about IRC
18:46
* urjaman
loops about grumbles
18:50
* alyssa
grumbles about register allocation
18:52
stikonas_ has joined #panfrost
18:53
stikonas has quit [Ping timeout: 255 seconds]
19:06
* alyssa
grumbles about loops
19:08
* alyssa
grumbles about IRC
19:08
* urjaman
loops about grumbles
19:17
<
alyssa >
dEQP-GLES2.functional.shadrs.loops.* passing 625/625!!!!
19:17
<
urjaman >
sounds exciting.
19:17
<
alyssa >
urjaman: Loooops!
19:20
* alyssa
grumbles about register allocation, v2
19:30
<
alyssa >
Ach. imov precision thingamajig is back with a vengeance. Groan.
20:00
mifritscher has quit [Quit: Quit]
20:19
mifritscher has joined #panfrost
22:10
stikonas_ has quit [Remote host closed the connection]
22:11
BenG83 has joined #panfrost
22:14
* alyssa
pokes at int ops
22:14
<
alyssa >
gl_FrontFacing reveals much
22:15
BenG83 has quit [Client Quit]
22:18
jcdutton has joined #panfrost
23:18
<
alyssa >
Aaaaahaaa!!
23:19
<
alyssa >
Modifiers for ints mean, it would seem:
23:19
<
alyssa >
nothing: sign-extend
23:19
<
alyssa >
negate: use as-is
23:19
<
alyssa >
abs: zero-extend
23:20
<
alyssa >
HdkR: The next obvious question is what does -abs do :p
23:20
<
alyssa >
I'll concede I have no examples of that in my database
23:22
<
alyssa >
HdkR: Or, why does it matter which you choose if the input is the right size already (Actually, I think it doesn't -- Panfrost sets nothing [so sign extend] unconditionally rn..)
23:23
<
jcdutton >
alyssa, isn't abs = absolute sign. I.e -2 -> 2. 2 -> 2
23:23
<
HdkR >
It would only matter for <32bit types?
23:23
<
alyssa >
jcdutton: Sorry for my unconditionally sloppy notation
23:24
<
alyssa >
With floating point instructions, these bitsdo thr absolute sign / negation of the src
23:24
<
alyssa >
With int instructions, obviously that's not what's happening. Now we know what its
23:24
<
HdkR >
Modifiers named for the float instructions meaning different things when applied to int instructions :P
23:25
<
HdkR >
abs and negate as int modifiers would make the hardware pipelines a bit...painful
23:27
<
alyssa >
HdkR: OK, well, this is mystery #1 solved
23:27
<
alyssa >
Not the mystery I came for.
23:27
<
alyssa >
The more interesting question rn is why (u)intBitsToFloat sometimes needs to be preceded by a no-op fmov
23:32
<
alyssa >
But only sometimes!?
23:32
<
HdkR >
alyssa: no-op fmov on source or dest?
23:33
<
alyssa >
HdkR: Just, no-op
23:33
<
alyssa >
Rather than the program:
23:33
<
HdkR >
on completed unrelated registers even? :P
23:34
<
alyssa >
Rather than "imov r0, whatever"
23:34
<
alyssa >
We get "imov temp, whatever; fmov r0, temp"
23:34
<
alyssa >
We get the imov directly if all we're doing is just the intBitsToFloat call
23:36
<
alyssa >
[BTW, the 'abs' thing to zext also applies to scalar units]
23:36
<
alyssa >
If I do the intBitsToFloat and then just do some arithmetic, it's fine
23:37
<
HdkR >
Maybe the hardware is doing something dumb like type tracking on registers :P
23:38
<
alyssa >
I can do arithmetic before, too, and no temp
23:38
<
alyssa >
Starting to think I'm imagining things..?
23:42
<
alyssa >
Oh. No, nothing, I was imagining things after all.
23:45
<
alyssa >
What in the world
23:45
<
alyssa >
Maybe it just always needs to have something in the iadd slot
23:47
<
alyssa >
Yeah. Writeout quirk.
23:47
<
alyssa >
Well, case closed, I think.
23:47
<
alyssa >
Now with all of that in mind, maybe I can update the assembler and shake some meaning out of the imov woes.
23:56
jcdutton has quit [Quit: Leaving]