Closed speles closed 1 year ago
I can confirm that with this patch the dEQP-VK.image.queue_transfer.*
consistently pass on Intel Geminilake.
Thanks!
Thanks, added to review process. It may take up to 2 weeks
Merged in 6054d27fb5c2d810cb97702a716c9aee0ebe2d5b
Thank you! Closing this PR now
This PR is related to a bug that was originally submitted as a mesa issue, but I believe it's actually a test bug. (similar to the one fixed in https://github.com/KhronosGroup/VK-GL-CTS/pull/372).
We allocate and map src/dst buffers, write data to src, and read it from dst. But we don't flush src buffer after writing or invalidate dst buffer before reading. So if we allocate those buffers from non-host-coherent memory, the data may be corrupted at both of those steps.