Extending the video accelerator to add custom stream processors

This tutorial will cover:

• Adding a new stream processor to the Video Accelerator infrastructure.
• Exposing the new instruction to the user.

Adding a new stream processor

A stream processor has to be able to take a AXI4-Stream channel as input, process the data and output back to a AXI4-Stream channel as these are the only Input/Output options provided to it within the Video Accelerator infrastructure.

Let us assume that we wish to add

stream_processor processor(
   .aclk(aclk),
   .aresetn(aresetn),

   .in_ch(to_processor_ch),
   .out_ch(from_processor_ch)
);

Open the Accelerator System Verilog file: video_acc.sv. This snippet will be added at the end of the module.

However, it is missing the input/output. For that we need to declare the input and output channels:

// Make sure to add at the top of the file
nasti_stream_channel # (
   .DATA_WIDTH(DATA_WIDTH),
   .DEST_WIDTH(DEST_WIDTH)
) to_processor_ch(), from_processor_ch();

And add them to the mux and demux

nasti_stream_combiner #(
   .N_PORT(NR_FUN_UNITS + 1)
) glue(
   .slave(out_vein_ch),
   .master_0(routed_ch),
   .master_1(from_dct_ch),
   .master_2(from_idct_ch),
   .master_3(from_yuv422to444_ch),
   .master_4(from_yuv444toRGB_ch),
   .master_5(from_processor_ch),// Change a dummy_ch to your 'from' channel
   .master_6(dummy_ch),
   .master_7(dummy_ch)
);

nasti_stream_slicer #(
      .N_PORT(NR_FUN_UNITS + 1)
)  unglue(
   .master(in_vein_ch),
   .slave_0(output_buf_ch),
   .slave_1(to_dct_ch),
   .slave_2(to_idct_ch),
   .slave_3(to_yuv422to444_ch),
   .slave_4(to_yuv444toRGB_ch),
   .slave_5(to_processor_ch),// Change a dummy_ch to your 'to' channel
   .slave_6(dummy_ch),
   .slave_7(dummy_ch)
 );

Make a note of which master/slave you add your channels to as you will need this later.

Your accelerator will receive an 8-bit t_user. The format of these bits depends on yourself, but a common use is to have 1-bit as the “chain” bit. If the bit is high, you can set t_dest to a non-zero value which matches the number of stream processor you wish to chain to.

You need to increase NR_FUN_UNITS by one after wiring, and now your stream processor can be utilized via our SPA infrastructure.

Exposing the new instruction to the user

Bare metal

To expose the instruction to the user in bare metal we need to: Open videox.c and update

#define FUNC_MAX 3

to the value of NR_FUN_UNITS.

Then we need to open videox.h and add

#define FUNC_YOUR_OP 5

to allow easier use of the new operation.

Now you can use your accelerator using our provided videox.h interface.

Linux Driver

Open videox.c and update

#define NR_FUN_UNITS 3

to the value of NR_FUN_UNITS.