When using VK_EXT_descriptor_heap, DLSS and DLSS-G are unable to read user-provided images. DLSS output is black. DLSS-G generates an image, but it's the same as if no motion vectors were provided. Debug visualizations are black for all user-provided images.
There are no errors and all images are still being created with the correct parameters, including the image view.
Binding a dummy descriptor set right before evaluating the DLSS feature seems to restore some funcionality, but the workaround doesn't work for DLSS-G since there is no evaluate call.
When using VK_EXT_descriptor_heap, DLSS and DLSS-G are unable to read user-provided images. DLSS output is black. DLSS-G generates an image, but it's the same as if no motion vectors were provided. Debug visualizations are black for all user-provided images.
There are no errors and all images are still being created with the correct parameters, including the image view.
Binding a dummy descriptor set right before evaluating the DLSS feature seems to restore some funcionality, but the workaround doesn't work for DLSS-G since there is no evaluate call.