This has already been reported earlier (https://community.amd.com/thread/169828) for Catalyst 14.12 and was marked as resolved as the sample code I posted no longer reproduces the problem in Catalyst 15.7. However, if I slightly modify the sample (specifically, make the source image width not multiple of 4), the code no longer works correctly. I'm attaching a new sample.
Interestingly, if the region width is not a multiple of 4, everything is working fine.
In other words:
- copying 16x16 area from a 256x256 image - works well
- copying 17x16 area from a 256x256 image - works well
- copying 16x16 area from a 257x256 image - fails
- copying 17x16 area from a 257x256 image - works well
As before, the error is reproducible on a HD 7970, but with a newer driver - Catalyst 15.7.