1.7.430 alpha with new Super Structure module (Life replacement)

[cytan299]

OK, I've tried all the combo's "0", "1" in the cfg file; using gSwitch; using laptop display. No go, the problem remains The problem child is the Catalina MacBook Pro 16.

So, next, since I actually have 3 Macs (1 MacBook Pro 16 (2019 edition) running Catalina, 1 MacBook Pro (2012 edition) running Mojave and an iMac (2015 edition) running Mojave) I decided to try my Mojave iMac. And voila, no streaking problems at all and I didn't do anything special, i.e. cfg file etc.

iMac Mojave has the following specs:

track1.jpg (40.91 KiB) Viewed 6490 times

I'll try the same thing on my Mojave MacBook Pro (2012 edition) tonight. So, it begs the question whether it is a Catalina problem or a 5500 GPU problem.

cytan

[cytan299]

Hi Ivo,

One last data point with my MacBook Pro (2012 edition), it works But it takes forever to calculate when it is in the Super Structure module. Here's the About panel:

track3.jpg (31.24 KiB) Viewed 6478 times

So, I don't think that it's the slow process that's messing up ST on my MacBook Pro 16 because this MacBook Pro is not that fast Again, this MacBook Pro runs Mojave.

So do you have any examples of ST that works on a MacBook Pro 16 running Catalina?

cytan

[admin]

It seems that the GPU will necessarily be used when I use an external display. I *was* using an external display when I was ST. So, I think the easiest thing to do first is to use my MacBook Pro display and not my external monitor first to see if the problem goes away.

Did that help at all in combination with forcing macOS to use the integrated CPU for display duties, while using the discrete GPU for StarTools?

it takes forever to calculate when it is in the Super Structure module

Try a different preset as you enter the Super Structure module; there is a CPU component to Local Median Brightness Retention that is quite intensive. Using a different type of Brightess Retention will avoid that.

So do you have any examples of ST that works on a MacBook Pro 16 running Catalina?

Examples of Catalina yes (I was able to replicate the issue before on Catalina, but that issue went away after I granularised the processing), but I don't have further examples of Catalina on a MacBook Pro 16.

Do other GPU-intensive applications work OK?

Did you get the Recovery_Timeouts feature to work?

[admin]

I may have found an edge case that could trigger something along the behaviour we're seeing. It would need a specific sequence of events, with specific timing. Let me work on this. If it is indeed the problem, it would definitely make my top #5 of most interesting bugs of my career.

In the meantime, does the module yield correct results on the Macbok Pro 16 if you nudge the Airy Disk Radius parameter?

[cytan299]

It seems that the GPU will necessarily be used when I use an external display. I *was* using an external display when I was ST. So, I think the easiest thing to do first is to use my MacBook Pro display and not my external monitor first to see if the problem goes away.

Did that help at all in combination with forcing macOS to use the integrated CPU for display duties, while using the discrete GPU for StarTools?

I did all the tests using the laptop display. It wasn't connected to the external display. The combos are as follows:

1. With "0" set in opencldeviceindex.cfg . Same problem.
2. With "1" set in opencldeviceindex.cfg . Same problem.
3. Using gSwitch
   1. Integrated only. removed opencldeviceindex.cfg. Same problem.
   2. GPU only. removed opencldeviceindex.cfg. Same problem.

it takes forever to calculate when it is in the Super Structure module

Try a different preset as you enter the Super Structure module; there is a CPU component to Local Median Brightness Retention that is quite intensive. Using a different type of Brightess Retention will avoid that.

That's only on my 2012 MacBook Pro that took a really long time. I'll try your suggestion out if my MacBook Pro 16 doesn't give the right results and I have to use my 2012 MacBook Pro.

So do you have any examples of ST that works on a MacBook Pro 16 running Catalina?

Examples of Catalina yes (I was able to replicate the issue before on Catalina, but that issue went away after I granularised the processing), but I don't have further examples of Catalina on a MacBook Pro 16.

Do other GPU-intensive applications work OK?

I don't have any other GPU intensive apps

Did you get the Recovery_Timeouts feature to work?

Beg pardon: What is Recovery_Timeouts?

As another data point. Here's M31 that I only did Wipe->Color->Super Structure->Airy:

airy_50%.jpg (188.06 KiB) Viewed 6441 times

It's a lot different that the previous results. Looks like tiling and theres > 24 tiles which may correspond to work that is done by each of the 24 GPUs.

Also change airy disk radius to 60% gave the same thing:

airy_50%.jpg (188.06 KiB) Viewed 6441 times

cytan

[admin]

Another day, another version. @cytan299 can you see if 1.7.443 is any better?

[cytan299]

Another day, another version. @cytan299 can you see if 1.7.443 is any better?

Woohoo! It works!

Airy:

airy.jpg (186.19 KiB) Viewed 6325 times

Isolate:

isolate.jpg (456.42 KiB) Viewed 6325 times

What was this "Edge case" that you brought up and fixed?

Thanks for your hard work! This fix rocks!

cytan

[admin]

Woohoo! It works!
Thanks for your hard work! This fix rocks!

Awesome! Thank you for your patience and helping me test this.

The bug was an extremely specific timing issue that would only happen with a specific speed by which the CPU processes stuff and the specific speed by which the GPU processes stuff. Only when perfectly "tuned" would this happen. The bug was in a piece of code that handled the cancelling of the operations to restart the processing again. As a result it would sometimes keep the result of a cancelled operation and carry on. "Fun" bug

[cytan299]

Woohoo! It works!
Thanks for your hard work! This fix rocks!

Awesome! Thank you for your patience and helping me test this.

The bug was an extremely specific timing issue that would only happen with a specific speed by which the CPU processes stuff and the specific speed by which the GPU processes stuff. Only when perfectly "tuned" would this happen. The bug was in a piece of code that handled the cancelling of the operations to restart the processing again. As a result it would sometimes keep the result of a cancelled operation and carry on. "Fun" bug

That's a very interesting (or insidious) bug. I would've thought that it would've been a memory partitioning problem. That's a great catch and one for the books

cytan