Displayio slow draw speeds

[raven_coda]

I'm doing some benchmarks and found the displayio module very slow. I'm hoping it is because I don't understand how to use the library efficiently. Using the following displayio I code I only get 3 FPS

Code: Select all

display_bus = displayio.FourWire(spi, command=board.GP20, chip_select=board.GP22, reset=board.GP21, baudrate=100000000)
display = adafruit_ili9341.ILI9341(display_bus, width=320, height=240, auto_refresh=False)

cb = displayio.Bitmap(320,240,1_)
cp = displayio.Palette(1)
bgs=displayio.TileGrid(cb, pixel_shader=cp, x=0, y=0)
group.append(bgs)

while True:
    cp[0] = 0xFF0000
    display.refresh()
    cp[0] = 0x0000FF
    display.refresh()

However, if I use the following code to send the data directly to the fourwire interface I get 21 FPS, 7x the speed.

Code: Select all

while True:
    display_bus.send(0x2A, b"\0\0\x01\x40")
    display_bus.send(0x2b, b"\0\0\0\xf0")
    display_bus.send(0x2c, b"\xFC\0"*240*320)
    display_bus.send(0x2A, b"\0\0\x01\x40")
    display_bus.send(0x2b, b"\0\0\0\xf0")
    display_bus.send(0x2c, b"\0\x3F"*240*320)

Is there a way I could make the displayio library more performant? For reference, I'm using the RP2040 which has more than enough RAM to hold a frame buffer in memory if that would address the issue.

[blakebr]

Tracking

[blakebr]

raven_coda,

Can you post your entire code.py file. I just got an ILI3941 and I would like to see how you made it work.

Thanks,
Bruce

[raven_coda]

Here is a more polished version

Code: Select all

import board
import displayio
import adafruit_ili9341
import time
from busio import SPI

direct = False  #change this to true to compare

displayio.release_displays()
spi = SPI(clock=board.GP18, MOSI=board.GP19, MISO=board.GP16)

display_bus = displayio.FourWire(spi, command=board.GP20, chip_select=board.GP22, reset=board.GP21,baudrate=125000000)
display = adafruit_ili9341.ILI9341(display_bus, width=320, height=240, auto_refresh=False)

group = displayio.Group()
display.show(group)

cb = displayio.Bitmap(320,240,1_)
cp = displayio.Palette(1)
cp[0] = 0x00FF00
bgs=displayio.TileGrid(cb, pixel_shader=cp, x=0, y=0)
group.append(bgs)

while True:
    start = time.monotonic()
    for i in range(10):
        if direct:
            display_bus.send(0x2A, b"\0\0\x01\x40")
            display_bus.send(0x2b, b"\0\0\0\xf0")
            display_bus.send(0x2c, b"\xFC\0"*240*320)
            display_bus.send(0x2A, b"\0\0\x01\x40")
            display_bus.send(0x2b, b"\0\0\0\xf0")
            display_bus.send(0x2c, b"\0\x3F"*240*320)
        else:
            display.refresh()
            cp[0] = 0x0000FF
            display.refresh()
            cp[0] = 0xFF0000
    span = time.monotonic()-start
    print(20 / span)

[blakebr]

Thank you!

[tannewt]

I'm doing some benchmarks and found the displayio module very slow.

Is there a way I could make the displayio library more performant? For reference, I'm using the RP2040 which has more than enough RAM to hold a frame buffer in memory if that would address the issue.

displayio isn't designed to be performant when updating the entire display. It trades speed for reduced memory use by computing pixels as need and in chunks. It also does dirty rectangle tracking for you to know where to update. Dirty rectangle tracking reduces data transferred to displays that have their own framebuffer. displayio is used across many microcontrollers so it isn't easy to rewrite for the rp2040.

What are you actually trying to do with displayio? Your benchmark is one of the worst cases. (Reading pixels from a file with OnDiskBitmap can also be very slow.)

[raven_coda]

I was just doing some benchmarks but the end goal is to implement a tile-based side scroller game. Since most of the screen would need updating I don't think the dirty rectangle optimization will work for my needs. Thanks for the reply though.

[tannewt]

For side scrolling I had intended on supporting the built in scroll ability of displays. However, it is only one direction usually and often it is the wrong axis.