Defensively tear down DMA/PIO so it's in a known good state upon (soft)reset.
Issue was a race condition with DMA interrupts firing and not being achknowledged, leaving a stuck raised IRQ.
The blocking wait for DMA transactions also exacerbated this, turning a borken DMA interrupt and blank screen into an unrecoverable hardlock.
...not that the blank screen was recoverable without a soft reset anyway!
Switch from 12-bit to 10-bit gamma to fit RGB into a uint32_t. Simplifies PIO and halves the RAM usage for F/B buffer.
Switch "flip" to *literally* swap the front and back buffers, and then asyncronously DMA the new back buffer into the front ready for the next draw.
The ST7789's Tscycw (time between serial write clock cycles) is
16 ns. This can be found on page 44 of the datasheet I'm using:
https://www.waveshare.com/w/upload/a/ae/ST7789_Datasheet.pdf
(I do not know which manufacturer Pimoroni products use and if
their parts might be different. But it seems like this wouldn't
change.)
The existing code sets the SPI baud to 16 * 1000 * 1000. But baud
is Hz, not seconds. That 16 * 1000 * 1000 doesn't represent 16 ns.
It represents 16,000,000 Hz.
16 ns * (1 Hz / s) = 62,500,000 Hz.
This commit changes the baud from 16 * 1000 * 1000 to 62'500'000,
representing ~4x speed improvement in SPI and thus ~4x frame rate
improvement, since the display's frame rate is currently
SPI-limited.
A before & after video can be seen here:
https://www.youtube.com/watch?v=n2y19TCnATo
Note that also on page 44 of that datasheet Tscycr (the read speed)
is only 150 ns, not 16 ns. Right now, the Pimoroni code doesn't read
any values back from the ST7789 so it is safe to operate at the
higher speed.
Also note that the 16 * 1000 * 1000 is the requested baud. The actual
baud is the closest the Pico can get, which is 15,625,000.
The new requested baud of 62'500'000 results in an exact match.
ZodiusInfuser
Mike Bell
Mike Bell
Phil Howard
Philip Howard
Jonathan Williamson
Chris Blume