Our partner have a question about i.MX6SDL vpu decoding.
When decode a 1024 byte jpeg file with vpudec for gstreamer, vpu can decode it correctly.
On the other hand, when decode a 1024 byte jpeg file with unit_test in L3.0.35_4.1.0, vpu cannot decode it and the output size will be "0".
In this case, vpudec is not used and libvpu in imx-vpu is used.
So we guess imx-vpu is the root cause of this issue.
Then, would you let me know how to resolve this issue when use libvpu?
Solved! Go to Solution.
I got additional information.
This issue occurs with the following condition.
Format = YCC420
Some file can be decoded, it depends on body size excluding header part.
If add 1 byte to the file that this issue occurs and it will be 1025 byte, decode was succeed.
I have tried sending the jpeg file, have you not yet received it?
If you have not received it, please tell me how to send the file as secret.
For your additional information, sometimes decoding was failed even if the size was 1025 bytes or Y42B format 1024 bytes image.
So would you investigate the condition when this problem occurs?
Will this issue cause partial decoding of JPG input as Corrupted MJPG decoding result with FSL 3.10.17 BSP on i.MX6Q VPU?
Does this issue exist in FSL 3.10.17 release as well? If so, does the patch above patch apply as well?
I am not sure. The patch was for 3.0.35 and should be in newer versions. I'll double check if it was included but from the description of the thread you referenced. I think that is more an mpeg problem than JPEG.
Can you provide a testcase to replicate the error?
Sorry for delay.
I got the command to reproduce the problem.
./mxc_vpu_test.out -D "-i dump_copy.jpeg -o dump_copy_out.raw -f 7 -c 1"
And I will send you the jpeg file by message to use the above command.
Sorry, I cannot upload this jpg file here because this file was made by our partner, and I have not allowed to upload it to public space.