Fix validation warnings #57
Commits on Sep 20, 2023
-
Add a test runner to avoid regressions
Only tested on Linux. Signed-off-by: Charlie Turner <[email protected]>
-
VK_LAYER_KHRONOS_validation is the more modern layer name. VK_EXT_DEBUG_REPORT_EXTENSION_NAME was mistakenly being requested as a layer, request the instance extension instead. Signed-off-by: Charlie Turner <[email protected]>
-
Fix vkGetDeviceQueue-queueFamilyIndex-00384 validation error
This happens in noPresent mode, since we aren't supposed to be retrieving queues from these families. Signed-off-by: Charlie Turner <[email protected]>
-
And use it to ignore the buffer profile information errors. These will become optional soon, and wiring them into the sample app is problematic, since the bitstream buffers are created before the video profile is known! Signed-off-by: Charlie Turner <[email protected]>
-
Forward selection of presentation to the decoder
Various initialization paths will need this information to become validation clean Signed-off-by: Charlie Turner <[email protected]>
-
Check for physical device features before using them
This is not complete, but silences some validation warnings. Before all features were being turned off! Signed-off-by: Charlie Turner <[email protected]>
-
Fix for update sequence count validation error
Needs better testing Signed-off-by: Charlie Turner <[email protected]>
-
Align bitstream offset and size as advertised by implementation
Signed-off-by: Charlie Turner <[email protected]>
-
Ignore DPB management validation errors
Need to figure out better how to handle these Signed-off-by: Charlie Turner <[email protected]>
-
Signed-off-by: Charlie Turner <[email protected]>
-
NFC: Rename get -> getVideoRelatedQueuesProperties
Signed-off-by: Charlie Turner <[email protected]>
-
NFC: AddOptinalDeviceExtension -> AddOptionalDeviceExtension
Signed-off-by: Charlie Turner <[email protected]>
-
NFC: Remove unnecessary nesting
Signed-off-by: Charlie Turner <[email protected]>
-
Signed-off-by: Charlie Turner <[email protected]>
-
Fix a sync2 validation warning
Signed-off-by: Charlie Turner <[email protected]>
-
NFC: Display the device being used
Signed-off-by: Charlie Turner <[email protected]>
-
Use concurrent sharing when the transfer queue != decode queue
Signed-off-by: Charlie Turner <[email protected]>
-
Always keep DECODE_DST bit on the output image
For linear mode on coincident implementations, this gives an validation error (although it feels OK to do...) Just make the layer happy, I suspect we should be using the image copying approach in general, and rather copy to a buffer. Signed-off-by: Charlie Turner <[email protected]>
-
Tidy up output and dpb usage variables (note about linear tiling)
Signed-off-by: Charlie Turner <[email protected]>
-
Ensure the timeline semaphore is selected
This is now unconditionally used, perhaps we should flag it off if not supported, so as to run on more devices. Signed-off-by: Charlie Turner <[email protected]>
-
Properly check supported formats and assert linear tiling is not supp…
…orted Demonstration for Tony, per our discussion in the weekly meeting. + assert(!outputImageSupportsLinearTiling); Please check if this is a driver bug. In general it might not be supported though, and we will need the ImageToBuffer fallback anyway. Signed-off-by: Charlie Turner <[email protected]>
-
Lets rely on CopyImageToBuffer, and tidy up the image handling code in the process. Signed-off-by: Charlie Turner <[email protected]>
-
Use CopyToBuffer in ConvertFrameToNV12
This is not proposed to merge, rather to help a discussion. Of course the code should be abstracted a bit, looking for early feedback. This also doesn't work for 10-bit out of the box. Need to do some bit manipulations of the packed 10-bit values to convert it properly to RAW YUV format (bottom 6 bits clear, rather than top 6) Before doing that, want to check if this is an acceptable approach in general, or whether we should enforce linear tiling for output images. Signed-off-by: Charlie Turner <[email protected]>
