Memory leaking or never deallocated about gpuimage2

[BradLarson]

Is this a Release or Debug build? When you run this under the Allocations instrument and mark generations, where do you see the accumulation?

[dgsmith]

I get this in debug and release builds. When I profile under release, looks like a lot of the memory is of the category Malloc 1.17 MiB, last few calls in the stack trace are:

swift_slowAlloc
specialized static UnsafeMutablePointer.alloc(Int) -> UnsafeMutablePointer<A>
specialized PictureOutput.cgImageFromFramebuffer(Framebuffer) -> CGImage
specialized PictureOutput.newFramebufferAvailable(Framebuffer, fromSourceIndex : UInt) -> ()

So it looks to be happening when the PictureOutput creates the CGImage, the memory just never gets released?

Let me know if I can provide any more info!

[BradLarson]

It's possible that the data provider release callback isn't being triggered, for some reason. Try putting a print statement within the dataProviderReleaseCallback() function at the very bottom of PictureOutput.swift and see if that ever gets triggered. It did in my tests, but maybe there's something about the extraction of JPEG-encoded data that I didn't account for.

[dgsmith]

Just tried that, it is actually getting called.

[dgsmith]

Should I just switch to using .png's for now?

Edit: I get the same error with .png's

[BradLarson]

I think I might have tracked down the source of this. Check out the latest commit, which should properly deallocate the memory being used by still images. The parameters were swapped in my C callback function and the correct memory was never being deallocated.

[dgsmith]

Yep, that fixed it. Thanks!

[dgsmith]

Hey Brad, I'm actually getting this exact same problem again with the latest updates. It seems that this time the dataProviderReleaseCallback(_:data:size:) isn't being called at all...

Still the same code as above, just writing the data to disk.

Thanks!

[spongemook]

Did you fix this problem? @dgsmith
I'm having the same problem.

[dgsmith]

Unfortunately, I think it requires a bit more in depth knowledge than I can provide haha. Since this time, the callback doesn't seem to be getting called...leading me to think the data is not being freed properly somewhere down the line? Perhaps some change to the Swift handling of Unsafe stuff has messed with this?