A open source Verilog implementation of DisplayPort protocol for FPGAs, released under the MIT License.
DisplayPort is quite a complex protocol. This is a minimal Verilog implementation in the Verilog language. Hopefully this will inspire others to improve on this.
This has now been tested using one or two lanes, and 800x600, 720p, 1080p and 2160p resolutions, but should work with four lanes too. YCC and 442 video are supported.
Note that this is still very alpha. It works for me on my hardware, but I don't expect it will work for you with a bit of effort.
Please feel free to send pull requests, and please make sure you add your name to the file headers. Also feel free to remove my headers for any new files you may add to the project - you deserve the credit not me!
Please make sure that where possible all files include the MIT License information.
My own test board is a Digilent Inc Nexys Video, using an Xilinx Artix 7 FPGA. However the most of the hardware specific parts are limited to the transcievers which can be replaced to support FPGAs from other vendors.
For a test display I have been using a ViewSonic VX2880ML, which is an older 4k monitor.
I will endevor to test with a few more monitors.
Resolution | Lanes | Colour Mode | Effective Pixel clock rate
-----------+-------+-------------+--------------
800x600 | 1 | RGB 444 | 40.00 MHz
800x600 | 2 | RGB 444 | 40.00 MHz
800x600 | 3 | RGB 444 | 40.00 MHz
1280x720 | 1 | RGB 444 | 74.25 MHz
1920x1080 | 2 | RGB 444 | 148.50 MHz
3240x2160 | 2 | YCC 422 | 165.00 MHz
There are in the src/test_streams directory. To change patterns, edit src/test_stream.v, switch the module name, and rebuild the file
These test streams are very crude, and could be greatly improved on.
DisplayPort have M and N values embedded in the data stream, which represent the ratio of the pixel clock to the lane symbol rate. For example 148.5MHz 1080p has a ratio of 11 to 20 of the 270MHz link speed. It also should embed the lowest 8 bits of the 'M counter' in the stream, to allow the sink to regenerate the pixel clock.
However 11:20 (or 22:40, or 2200:4000 or any other exact ratio) does not work but 0x4688:0x8000 (18,056:32768) does. I do not understand this. If somebody could explain this to me so I can document this I would be most greatful.
I suspect is has something to do with the ability for the source to down-spread the link speed, and the sink must be able to correct for this.