BC7 encoding with Compressonator SDK prints "Err: requantized_image_err"

[ShukantPal]

If I use the Compressonator CLI, BC7 encoding works perfectly fine. However, I'm now using the Compressonator SDK with JNI bindings to Java. When I run CMP_ProcessTexture with BC7 encoding, the following is printed for almost 100K times.

Err: requantized_image_err

[ShukantPal]

Note that the output is perfectly fine and the destination texture gets saved successfully.

[NPCompress]

@ShukantPal This build setup is new and never tested with JNI. Interesting that this error is generated, its indicating that a large number of 4x4 processed blocks are not been re-quantized in (shake() call) for better quality.

You say "Note that the output is perfectly fine and the destination texture gets saved successfully." is the PSNR high when compared to the original file?

[ShukantPal]

Not sure, but the files did have lots of "errors" (solid color pixels that looked like they had the wrong value)

[NPCompress]

@ShukantPal That is the expected result for this type of err, something failed in processing the block in either the shake() or initial setting in BC7 options like errorThreshold value is not set to a valid range.

[ShukantPal]

I didn't set any errorThreshold value in my code. Are you talking about the kernel options?

[NPCompress]

@ShukantPal Yes, ref framework_example2.cpp for a BC7 example

[ShukantPal]

Will look into it - I would suggest to have sensible defaults in this case (in the constructor?)

[NPCompress]

@NPCompress It has defaults, but can be overwritten by user setting passed down in CMP_ProcessTexture. Will add additional parameter check for valid ranges. This however may not be the root cause of the issue reported and calls inside the code for BC7 shake() needs to be debugged using your setup environment.

Can you provide details as to how you have "Compressonator SDK with JNI bindings to Java" setup

[NPCompress]

Please review your JNI with latest release