Video jitter

[thejpster]

I think I've worked out why there's visible jitter on the video - I think it's variable interrupt latency. From what I've read, it seems the interrupt latency can be affected by whatever code you happen to be running when the interrupt occurs. This is further exacerbated by the fact that at 80 MHz we must run the flash at 40 MHz and endure wait states. Some random thoughts follow:

• The latency of the scanline draw routine doesn't matter as we have a FIFO.
• We should consider moving the start of the scanline draw routine to the start of the front porch, to build up an extra 1us (16 or 24 pixels) in the buffer.
• It might be possible to fix the jitter in the SPI start routine by setting the timer to interrupt early, and then busy-waiting on a timer value.
• Just ahead of that busy-wait routine it may be necessary to read an 80 MHz timer value and conditionally jump to a routine with an appropriate number of nops to accurate set our phase relative to the divided 20 MHz pixel clock. In psuedo code:

let jumps = [ one_nop, two_nops, three_nops, four_nops ]
let t = read_timer();
function = jumps[t & 0x03];
function();

• Another approach is to run the SPI output constantly, outputting zero bytes when there's no signal. So once 50 bytes of visible data have been sent, we simply have to write another 16 bytes to get us across the sync pulse, and we're done! We then only have to start the SPI peripheral once (maybe once per frame, so we're not busy during the v-sync period), and make sure we never underflow! For reference, the SPI FIFO stores 8 words. If we wind it up to 16-bit words, we buffer twice as much (but make the sync-start routine take twice as long).

[thejpster]

Or we can set another timer isr which just nops a lot, at a priority just one lower. The jump into the start routine is then always from this isr.

[thejpster]

We could also put the routine in RAM.

[thejpster]

Or we can set another timer isr which just nops a lot, at a priority just one lower. The jump into the start routine is then always from this isr.

I fixed this in this way, but with a wfi instead of a nop. Putting routines in RAM helped, as did moving the vector table, but with this fix it runs fine with them in a flash, so I put them back.

See 4d1a9b7