fix for "1 leaked semaphore objects to clean up at shutdown" on M1

[skurovec]

Implements fix by @Any-Winter-4079 referenced in #1016 (comment)

[Vargol]

Hi,
In the the second forward, has h been checked that it's on the mps device after its been derived from x ?
At work so can't check myself ATM.

[skurovec]

From my testing it looks like h inferring device from x:

"beautiful landscape" -s 15 -S 3972197351 -W 2048 -H 1024 -C 7.5 -A k_lms

Devices:  x mps:0 h mps:0
Devices:  x mps:0 h mps:0
Devices:  x mps:0 h mps:0
Devices:  x mps:0 h mps:0
Devices:  x mps:0 h mps:0
Devices:  x mps:0 h mps:0
Devices:  x mps:0 h mps:0
Devices:  x mps:0 h mps:0
Devices:  x mps:0 h mps:0
Devices:  x mps:0 h mps:0
Devices:  x mps:0 h mps:0
Devices:  x cpu h cpu
Devices:  x mps:0 h mps:0
Devices:  x mps:0 h mps:0

[Vargol]

Okay, should be get h back on the mps as soon as we can (assuming none of the other bits of processing break).
I'm sure it won't make too much difference.

I've also noticed that in #1016 up mention the images are not consistent at smaller sizes which concerns me a little, even though I know technically it should be running the same code. Does the new code get triggered at that size where it wasn't before ?

[skurovec]

I have consistent images pre and post patch. But different image than @Any-Winter-4079 with same seed

[skurovec]

"beautiful landscape" -s 50 -S 1962347883 -W 512 -H 512 -C 7.5 -A k_lms

development branch

branch with fix

[skurovec]

@Vargol Any-Winter-4079 just commented, that he copied wrong seed, so images ARE consistent

edit: typo

[Any-Winter-4079]

Yeah, my bad. It was late and I copied the wrong seed haha

@Vargol I think h infers the device, but we can make it explicit if you want.
h.device.type

mps
mps
mps
mps
mps
mps
mps
mps
mps
mps
cpu <- cpu
mps
mps

[Any-Winter-4079]

LGTM