fmh_framebuffer is a framebuffer with an Avalon ST Video output. As such, It is compatible with the various components of the Intel Video and Image Processing Suite (VIP). It is capable of flipping its output horizontally or vertically, the combination of which can acheive 180 degree rotation.

It implements enough register level compatibility to be a drop-in replacement for the old Intel (Altera) Frame Reader when using the altvipfb Linux driver.

It works. It has been tested on a Cyclone V HPS system while clocked at 60 MHz. It was connected to RAM through a 64 bit AXI bus. Its video output sent 4 color channels of 8 bits per color to a VIP Clocked Video Output running at 800x480 resolution with a 60 Hz frame rate.

• All the colors of exactly one pixel are output in parallel on each beat of the video output stream. So, no support for sending the colors of a single pixel sequentially over multiple output beats, and no support for sending multiple pixels in parallel on a single output beat.
• 90 degree rotations are not supported.
• Progressive frames only, no interlaced.

The latest version of this core may be found at https://github.com/fmhess/fmh_framebuffer .

• Q: Why did you bother? The VIP suite already has framebuffer cores.
  • A: We needed a core capable of rotating its output by 180 degrees, which is a capability the VIP framebuffers do not provide.
• Q: Why didn't you make a dedicated rotator for VIP instead of an integrated framebuffer/rotator? The VIP suite is designed to consist of modular components which can be mixed and matched as the user desires.

  • A1: It is easier to write an simple integrated framebuffer/rotator than a pure rotator. A framebuffer only needs to read RAM and output a video stream. A pure rotator needs to do this as well, and additionally needs to write RAM and receive an input video stream.

  • A2: An integrated framebuffer/rotator is more efficient than a pure VIP rotator. The Avalon ST Video stream sends video frames in a prescribed order. The rows of pixels are sent starting from the top row and then working down. Thus, in order to perform a rotation or vertical flip, a pure VIP rotator needs to buffer an entire frame in RAM down to the last row before it can begin sending the first row of the rotated output frame.

    On the other hand, an integrated framebuffer/rotator gets its input from RAM which it can read in any order. Thus it only needs to cache a single row of the input frame at once in order to output a frame which is vertically or horizontally flipped, or 180 degree rotated. 90 degree rotations can be achieved efficiently while only needing to cache a few columns of the input frame at once.

Frank Mori Hess [email protected]