Honor bufferpool stride alignment when rendering overlays

Buffer pool may be allocated by hardware specific routines
that specify specific stride alignment. When rendering
overlays we must honor the specified constraints or risk
garbage being rendered downstream.

In the current specific use case the downstream element
is glupload, which NXP has modified to require a stride
that's 16*bpp aligned for their Vivante specific fastpath
to work. They also disabled all the standard fallback paths
for dmabufs assumed present, so unless upstream adheres
to the HW constrains garbage is rendered.

Change-Id: I2710446ddd4a9e23da5888c5a04f40833e28cbb5
1 file changed
tree: 747c456e97499ad2f1b3bcc055d195161e315453
  1. debian/
  2. edgetpuvision/
  3. .gitignore
  4. MANIFEST.in
  5. README.md
  6. setup.py
  7. stdeb.cfg
README.md

edgetpuvision

Python API to run inference on image data coming from the camera.

Build

python3 setup.py sdist python3 setup.py bdist python3 setup.py sdist_wheel

Debian pacakge

Install stdeb package by running apt-get install python3-stdeb or pip3 install stdeb. Then to generate debian folder run:

python3 setup.py --command-packages=stdeb.command debianize

To build debian pacakge run:

dpkg-buildpackage -b -rfakeroot -us -uc -tc