Marc's Public Blog - Arduino Hacking


All | Aquariums | Arduino | Btrfs | Cars | Cats | Clubbing | Computers | Diving | Dreamstate | Edc | Electronics | Exercising | Festivals | Flying | Halloween | Hbot | Hiking | Linux | Linuxha | Monuments | Museums | Oshkosh | Outings | Public | Rc | Sciencemuseums | Solar | Tfsf | Trips



Table of Content for arduino:

More pages: November 2024 June 2024 August 2023 June 2023 May 2023 April 2023 March 2022 January 2022 January 2021 December 2020 March 2020 January 2020 May 2019 April 2019 March 2019 January 2019 July 2018 May 2018 April 2018 January 2018 June 2017 April 2017 January 2017 February 2016 January 2015 September 2013 January 2012 December 2011 May 2011 January 2011



2020/03/16 Framebuffer::GFX, Choosing between its 3 2D APIs: FastLED XY(), NeoMatrix, and LEDMatrix, and detail of its many supported hardware backends
π 2020-03-16 01:01 in Arduino

Why Framebuffer::GFX?

After rewriting multiple related libraries, it became confusing to people how they fit together, Adafruit::NeoMatrix, FastLED::NeoMatrix, SmartMatrix::GFX, where does Framebuffer::GFX fit in?
Is Adafruit::GFX still there/needed?
How about LEDMatrix, isn't it a better 2D lib?
What hardware backends are supported?

How the modules/drivers fit together

I made this little graph to summarize everything:
  • the left shows the hardware drivers supported: FastLED, SmartMatrix, ILI9341, SSD1331, ST7735, rpi-rgb-led-matrix (for ArduinoOnPC on Raspberry Pi), FastLED_SDL (for ArduinoOnPC), and X11/Linux (also ArduinoOnPC)
  • The 2nd column is the list of drivers I wrote for all those hardware drivers
  • All those drivers I wrote inherit from FrameBuffer::GFX wihch stores the actual framebuffer in FastLED CRGB 24bpp Format
  • FrameBuffer::GFX implements (and replaces) Adafruit::NeoMatrix panel mapping, inherits from Adafruit::GFX while converting back and forth between GFX 16bpp RGB565 and 24bpp CRGB RGB888 (see color management in https://github.com/marcmerlin/Framebuffer_GFX).
  • LEDMatrix is an alternative to GFX that is mostly compatible, but offers its own extensions. I you wish to use it, I modified it so that it doesn't store the CRGB array within itself, but accepts an externally generated one (so that it can be shared with SmartMatrix if this is the ultimate backend, and allocated via malloc and not a global array, which is required on ESP32 for bigger sizes). More on whether you'd want to use it, later.
  • Low Level Drv|Glue Driver for FrameBuffer::GFX FastLED     - FastLED_NeoMatrix -------------\     FastLED CRGB Array SmartMatrix - SmartMatrix_GFX -----------------\   24bit FB storage       API Support                                               \  CRGB methods like AF:ILI9341-\                                   \ scale8/fadeToBlackBy AF:SSD1331 |--- FastLED_ArduinoGFX_TFT ----------|   | ,FastLED API AF:ST7735 /     FastLED_SPITFT_GFX (for AF)     |       |             / (XY 2D to 1D mapping) ArduinoGFX/  AF:Adafruit (ArduinoGFX is all in 1) |       |            
                                                     |       |          
    ArduinoOnPc-FastLED-GFX-LEDMatrix arduino         - FrameBuffer::GFX------ Adafruit::NeoMatrix + emulation for linux / Raspberry Pi:               |       |             \ Adafruit::GFX APIs ----------------------------------               /   Adafruit::GFX       \ rpi-rgb-led-matrix - FastLED_RPIRGBPanel_GFX ---/   LEDMatrix (optional)   `LEDMatrix API ArduinoOnPC X11/linux - FastLED_TFTWrapper_GFX
    FastLED_SDL (linux)   - FastLED_NeoMatrix   -/

    Which 2D API: FastLED XY(), Adafruit/FastLED::NeoMatrix, vs LEDMatrix

  • Old FastLED demos use an XY() function, to access leds[XY(x,y)] to do 2D to 1D mapping. Framebuffer::GFX supports this in a way that you can both read and write while supporting full rotation and tile mapping from NeoMatrix. This is honestly no reason to do this today, you would effectively rewrite the much better XY functionality that's in FastLED::NeoMatrix/Framebuffer::GFX. However, if you have code that uses an XY function, it's fully supported, simply use matrix->XY(x, y) to let NeoMatrix do the real mapping for you without having to worry about writing/rewriting the mapping function for each potential array.
  • The better and most commonly used API is Adafruit::GFX. Honestly I recommend it because of the sheer number of hardware devices supported by it. It has the downside of only being 16bit color, but if you use it in FrameBuffer::GFX, can you can use 24bit color with it. Its 2nd downside is that on its own, it has no framebuffer, so it is write only. Obviously with Framebuffer::GFX, it does have a framebuffer (which can be a downside on some slower hardware at higher resolutions, see TFT below). To learn more about how to use the GFX API, see https://learn.adafruit.com/adafruit-gfx-graphics-library?view=all
  • LEDMatrix is a better API with more fancy functions than GFX and native 24bpp color, but it's only meant to work on FastLED, and its matrix tiling support and rotation is pretty non trivial to use compared to NeoMatrix. I would only use it if you need its added functionality. On the plus side, Framebuffer::GFX allows you to run LEDMatrix code on any hardware backend, so you aren't limited there anymore. LEDMatrix pluses are
    • more primitives, including some flip/mirror screen options
    • Table_Mark_Estes is just that good, that it's worth having LEDMatrix for
    • LEDMatrix has Fancy Font Support. In my opinion it's more fancy than most people will need, and Adafruit::GFX built in front support is more than plenty, but it's there if you want it.
    • LEDSprites is nice if you need sprites, see my LEDSprites-Pacman demo

  • The SmartMatrix library has its own 2D API, it is quite fancy and includes layers, but it is not supported by FrameBuffer::GFX, so unless you're writing a SmartMatrix only project (which means you won't be able to upgrade to RGBPanels on rPI if you need a higher resolution), I recommend against using it. Technically, someone motivated could make the API work with FrameBuffer::GFX and allow rendering on another backend, but I haven't done that work and don't have the need to do so. If you'd like to contribute that, please do.
  • The good news is that if you don't know which one to choose Framebuffer::GFX allows you to run all 3 at the same time, thanks to the work done in neomatrix_config.h which while not required, I greatly recommend so that it'll setup everything for you and make it trivial to change from one hardware backend to another one.

    If you would like an example of each API, please go to https://github.com/marcmerlin/FastLED_NeoMatrix_SmartMatrix_LEDMatrix_GFX_Demos/ and check the demos in each per API directory.

    If you would like to read images from SPIFFS/FATFS (ESP8266/ESP32) or SDcard, see https://github.com/marcmerlin/AnimatedGIFs

    Supported hardware backends

    https://github.com/marcmerlin/FastLED_NeoMatrix/

    64x64 with Neopixels was 2 weeks of build work and 160 Amps at full power!
    64x64 with Neopixels was 2 weeks of build work and 160 Amps at full power!

    https://github.com/marcmerlin/SmartMatrix_GFX/

    FastLED::NeoMatrix in 32x32 and 24x32 vs SmartMatrix::GFX 96x64
    FastLED::NeoMatrix in 32x32 and 24x32 vs SmartMatrix::GFX 96x64

    https://github.com/marcmerlin/FastLED_ArduinoGFX_TFT (many TFTs including SSD1331, ILI9341, and ST7735 TFTs)

    This is now the library I recommend for all TFTs instead of the multiple libraries written by Adafruit (supported by FastLED_SPITFT_GFX), not quite with the same interface, and multiple drivers. I definitely recommend https://github.com/moononournation/Arduino_GFX instead for new work.

    Arduino_GFX supports running multiple TFTs at the same time if you have different CS pins:

    comparison of TFTs
    comparison of TFTs

    The demos below done on FastLED_SPITFT_GFX look exactly the same with FastLED_ArduinoGFX_TFT except for being faster with this driver.

    https://github.com/marcmerlin/FastLED_SPITFT_GFX (SSD1331, ILI9341, and ST7735 TFTs)

    Small SSD1331 in 96x64
    Small SSD1331 in 96x64

    The same exact resolution between RGBPanels and SSD1331
    The same exact resolution between RGBPanels and SSD1331

    after extra work to support PSRAM on ESP32, ILI9341 320x240 became possible on ESP32
    after extra work to support PSRAM on ESP32, ILI9341 320x240 became possible on ESP32

    almost the same resolution than my huge RGBPanel driven by rPi
    almost the same resolution than my huge RGBPanel driven by rPi

    https://github.com/marcmerlin/FastLED_RPIRGBPanel_GFX (Glue driver for https://github.com/marcmerlin/ArduinoOnPc-FastLED-GFX-LEDMatrix/ that allows displaying a matrix on RGBPanels using https://github.com/hzeller/rpi-rgb-led-matrix )

    My first rPi Display at 192x128
    My first rPi Display at 192x128

    Even bigger, 384x192
    Even bigger, 384x192

    https://github.com/marcmerlin/FastLED_TFTWrapper_GFX (Emulate a TFT screen on linux for https://github.com/marcmerlin/ArduinoOnPc-FastLED-GFX-LEDMatrix/ )

    This is probably the most useful driver I wrote out of all of them, the ability to write your code and debug on linux, without any hardware:


    Dealing with pushing bigger framebuffer sizes to TFTs like ILI9341

    For long the highest resolution target for arduino chips has been the ILI9341 TFTs. With a resolution of 320x240 over SPI, they push the limits of Framebuffer::GFX a little bit, because it's a lot of pixels to push. Unfortunately, the TFT seems to only support about 14 frames per second for a full refresh, which is needed with the framebuffer approach, and by the time you add required use of PSRAM on ESP32, which is slower than regular RAM but required because ESP32 does not have the required contiguous 224KB of RAM, frame refresh falls down to 8fps. Worse, still, once you add computation of data being sent, demos actually run closer to 5fps.
    This is far from ideal, but it's good enough or some uses still, and generally still cool that Framebuffer::GFX can be pushed so far on arduino-like chips. Using RGBPanels does not help there on arduino chips, because there is no arduino like chip that can run such a resolution on RGBPanels (Raspberry Pi can barely do it though, but that's also pushing the limits of the underlying hardware refresh capabilities).

    It this ends up being a problem, but you made the decision to stick to the Adafruit::GFX API, you always have the option to remove FrameBuffer::GFX and write directly to the TFT, without having to do full framebuffer refreshes.

    If you'd like numbers, I gathered as part of a test between SPI speed, raw TFT speed (empty frame push), loop to push data not from PSRAM, loop to push data from PSRAM. Actual speed in demos is still lower given that it takes time to generate high resolution frames to PSRAM (double PSRAM penalty, one to write, one to read), before they can be pushed.

    Arduino::GFX ILI9314
                        tft/gfx/bypass/copy
    40fhz, fps no PSRAM: 25/15/22/14       PSRAM: 25/11/21/8
    80fhz, fps no PSRAM: 42/19/33/18       PSRAM: 40/14/32/9 (Arduino_HWSPI)
    80fhz, fps no PSRAM: 53/21/38/20 Arduino_ESP32SPI
    80fhz, fps no PSRAM: 60/20/34/18 Arduino_ESP32SPI_DMA
    

    Adafruit ILI9314 (much slower) 80Mhz: TFT 40fps, NO PSRAM: 32fps, PSRAM show: 12fps

    Old Adafruit numbers: ST7735_128b160: 80Mhz: TFT153fps, NO PSRAM:104fps, PSRAM show: 45fps ST7735_128b160: 40Mhz: TFT 68fps, NO PSRAM: 56fps, PSRAM show: 32fps ST7735_128b160: 20Mhz: TFT 53fps, NO PSRAM: 45fps, PSRAM show: 29fps ST7735_128b128: 40Mhz: TFT117fps, NO PSRAM: 90fps, PSRAM show: 48fps

    Arduino::GFX is a lot faster (over 100fps) SSD1331: SWSPI: TFT 9fps, NO PSRAM: 9fps, PSRAM show: 8fps

    Now it's your turn, write your own 2D demos and contribute them back

    https://github.com/marcmerlin/FastLED_NeoMatrix_SmartMatrix_LEDMatrix_GFX_Demos would love your code, please contribute :)
    2020/03/13 RGB Panels, from 192x80, to 384x192, to 384x256 and maybe not much beyond
    π 2020-03-13 01:01 in Arduino
    This started with some 64x32 panels that Azerone nicely had amazon send me as they were returned by customers as not working. I also was not able to get them working with either, so I opened a bug with both libraries SmartMatrix or rpi-rgb-led-matrix. Thankfully some nice folks were on it, and eventually found a fix.

    victory, first time I got those new panels to light up
    victory, first time I got those new panels to light up

    I ended up getting 15 such panels from the nice folks at Azerone, because they were returns as other people couldn't use them due to lack of FM6126A support in the open source drivers at the time:

    if only I had proper mounting hardware, this would be a decent 192x160 array
    if only I had proper mounting hardware, this would be a decent 192x160 array

    Ok, after writing this blog, I ended up getting a recommendation for rails, which weren't built for this but worked well enough:

  • rails with holes mostly in usable places: https://www.homedepot.com/p/Everbilt-1-3-8-in-x-36-in-Zinc-Plated-Punch-Flat-Bar-800337/204325640
  • M3 screws and washers: https://www.amazon.com/gp/product/B07FCN64HV/ref=ppx_yo_dt_b_asin_title_o05_s00
  • you'll need something to cut the rails that are too long. I used a set of bolt cutters I had, some saws may work too, but the rails are pretty thick
  • holes don't all match, but enough of them do. Note that I put the panels in Z pattern
    holes don't all match, but enough of them do. Note that I put the panels in Z pattern

    adding power
    adding power

    on my first try, the metal cut the ribbon cable, so I removed the top loop to avoid having it go towards the metal
    on my first try, the metal cut the ribbon cable, so I removed the top loop to avoid having it go towards the metal

    Now, I had an expected problem that hzeller/rpi-rgb-led-matrix did not support panels stacked vertically, never mind if every panel was upside down. I ended up writing a new mapper that allowed both vertical stacking and zig-zag. After that, success was had, end result was about 300Hz for 31K pixels with the active-3 board:




    That said, while that bigger matrix with those 15 panels was 192x160, I had bigger plans. I got distracted due to other work, but then P2 panel appeared (2mm per pixel). The downsides of those panels is that the copper traces are so small that pixels come off easily, including during shipping:


    even pixels that don't pop off can be partially failed
    even pixels that don't pop off can be partially failed

    This was probably the first working demo on rPi of FM6126A at 128x128 with 128x64 P2 Panels from Azerone:

    In no time, I was able to make a display of 192x128 with 3x 64x128 panels and it was the same size as my existing 96x64 screen for my LED shirt:

    using the active-3 board with 3 chains of one panel each on rPi3
    using the active-3 board with 3 chains of one panel each on rPi3

    The first issue was that the newer 128x64 panels were FM6126A, which didn't work with default libraries as they need a special init. Thankfully this was added to rpi-rgb-led-matrix. Next issue was that refresh rate was going to suffer, and that's where the rPi solution is great: it allows for 3 channels in parallel thanks to the nice Electrodragon active-3 board.

    Chow He from Electrodragon was super nice, and sent me some boards with angled connectors (at my request) so that I could make the whole board more flat for my wearable application. I was really hopeful that I could put the connectors under the board for low profile after cutting the notch and moving it to the other side of the connector:


    However, wire routing was such that pins were in the wrong order and it was impossible without re-routing all the wires on the board, so I did the next best thing and put the connector on top of the board, which is still more flush than when it was pointing up:

    before (blue), not working (right), and after (left)
    before (blue), not working (right), and after (left)

    bottom is not possible, even after moving the notch
    bottom is not possible, even after moving the notch

    Still, 192x128 is a very nice display that happens to be the same size as my old 96x64 display using P4 panels, but since I ordered spare panels to get around the falling pixel problem, I had enough panels to make a 9x9 array of 384x192. I first tried with with rpi-rgb-panel's video viewer and it worked quite well:



    During that time, I did a lot of work on 2D APis for Arduino and lots of backends, and wrote the base class, FrameBuffer::GFX. In a nutshell, it allow talking to a lot of hardware displays, now both on Arduino and Raspberry Pi:

    Low Level Drv|Glue Driver for FrameBuffer::GFX
    FastLED     - FastLED_NeoMatrix  -------------\     FastLED CRGB Array 
    SmartMatrix - SmartMatrix_GFX -----------------\    24bit FB storage        API Support
    ILI9341 \                                       \   CRGB methods like
    SSD1331  |--- FastLED_SPITFT_GFX ----------------\  scale8/fadeToBlackBy   ,FastLED API
    ST7735  /                                         |        |              / (XY 2D to 1D mapping)
                                                      |        |             
    ArduinoOnPc-FastLED-GFX-LEDMatrix arduino - FrameBuffer::GFX------ Adafruit::NeoMatrix + emulation for linux / Raspberry Pi: | | \ Adafruit::GFX APIs ---------------------------------- / Adafruit::GFX \ rpi-rgb-led-matrix - FastLED_RPIRGBPanel_GFX ---/ LEDMatrix (optional) `LEDMatrix API ArduinoOnPC X11/linux - FastLED_TFTWrapper_GFX
    FastLED_SDL (linux) - FastLED_NeoMatrix -/

    Given that work, it was time to try my arduino code on this panel, using my freshly written FastLED_RPIRGBPanel_GFX on top of ArduinoOnPc-FastLED-GFX-LEDMatrix as explained my page Running FastLED, Adafruit::GFX, and LEDMatrix code on High Resolution RGBPanels with a Raspberry Pi:

    it's pretty big :)
    it's pretty big :)

    This first demo of simple demo code that scaled up pretty quickly, explaining some of the challenges of scaling past 256 pixels in any dimention and 64K pixels total which hits a FastLED uint16_t limit:

    This is an early demo of my shirt code running on a display that is so much bigger than what it was built for:

    After some work to find a couple of crash bugs that came from scaling up, I was able to run Mark Estes' fantastic demo code I was able to scale up to such big displays without too much trouble. See Table_Mark_Estes1 and 14.








    Table_Mark_Estes:

    Table_Mark_Estes14:

    Just put back an idea of scale between my first panel at 24x32, all the way up to 384x192. Yes, those are the exact same sprites with LEDSprites:

    While 384x192 is starting to push the physical refresh limits that are acceptable for a 3 parallel chain setup, I'm going to try 384x256 for fun once I can get the pixel mapping to be correct. Bigger than that will require multiple control boards and synchronization by some network:

    this was my first attempt, panel mapping didn't work, I'd have to write my own
    this was my first attempt, panel mapping didn't work, I'd have to write my own

    Interestingly, SmartMatrix adds panels in the vertical direction by default, while rpi-rgb-led-matrix adds them horizontally. This means I had to write a new Mapper for it: V-Mapper, which allowed me to make 3 parallel horizontal chains. The end result is not great as refresh rate is only 100Hz, but it works:

    success!
    success!

    that was a lot of work for a low resolution screen :)
    that was a lot of work for a low resolution screen :)

    video still plays
    video still plays


    New pictures of Table_Mark_Estes in the higher resolution:




    As above, thanks again to Chow He from Electrodragon for the great and cheap active-3 boards, and big thanks to Hongren Su from Azerone for selling me the panels I have used so far. You can find the Azerone store on amazon for panels you can buy and the 128x64 P2 Panels here.
    On the software side, many people to thank, but obviously I wouldn't have gotten started with Adafruit, Louis Beaudoin for SmartMatrix on arduino chips, and Henner Zeller for rpi-rgb-led-matrix of course.

    Getting Higher Speeds

    Pushing 256 lines deep (with 128 pixels wide for each chain) is pushing the limits of how much data you can push and still get an acceptable refresh rate. I had to lower some quality numbers to get a usable refresh rate:
    rgb_matrix::RGBMatrix::Options defaults;
    defaults.hardware_mapping = "regular"; // or e.g. "adafruit-hat"
    

    defaults.rows = 64; defaults.cols = 128; defaults.chain_length = 4; defaults.parallel = 3; // 50 is faster but makes the screen dimmer defaults.pwm_lsb_nanoseconds = 100; // lower color depth definitely helps refresh defaults.pwm_bits = 7; // Time dithering of lower bits // 2 changes speed from 400Hz (from 160Hz) // 1 is also a faster option that's a bit brigher defaults.pwm_dither_bits = 2; defaults.led_rgb_sequence = "RBG"; defaults.panel_type = "FM6126A"; defaults.pixel_mapper_config = "V-mapper";

    rgb_matrix::RuntimeOptions ropt; #ifdef RPI4 ropt.gpio_slowdown = 2; #else ropt.gpio_slowdown = 1; #endif

    Build

    This is a temporary frame I made for my 384x256 array, itself made out of 3 smaller 128x192 arrays I was going to use as portable battery powered frames for my outfit. Unfortunately I used one bolt that was a bit too long and damaged a line on one panel when I screwed it in. Doh :(


    And this is what it looks like:

    Nice plasma also from Mark Estes:

    Now, I have no excuse (well, still a lot of work actually) for not doing an even more fancy LED shirt display (read all about this silly hobby here)

    Note for myself, this site sells parts to put panels together: https://www.wiredwatts.com/build-a-matrix-kit


    More pages: November 2024 June 2024 August 2023 June 2023 May 2023 April 2023 March 2022 January 2022 January 2021 December 2020 March 2020 January 2020 May 2019 April 2019 March 2019 January 2019 July 2018 May 2018 April 2018 January 2018 June 2017 April 2017 January 2017 February 2016 January 2015 September 2013 January 2012 December 2011 May 2011 January 2011

    Contact Email