One pixel transformation for ILI948x

[tore-espressif]

ILI948x doesn't support 16bit color data in SPI mode, so we have to extrapolate from LVGL's RGB565 (16bit) to 18 bits.

Previously, the whole buffer was transformed, which required a lot of DMA capable RAM. This PR changes to transformation algorithm to process one pixel at a time.

Closes #85

[kisvegabor]

Hi,

I found a few things that might be worth to discuss.

[kisvegabor]

This way there is no maxima blue. E.g. 0b11111 << 3 is 0b11111000 instead of 0b11111111.

The original code converted to RGB666 and added the LSB twice.

[tore-espressif]

Thanks! I couldn't get me head around the old conversion function. Will fix

[kisvegabor]

Assuming that on UI there are a lot of adjacent same color (e.g. a large white background, a blue button, etc) calculating only if the current color is different from the previous else coping the previous result might cause significant speed up.

[C47D]

@tore-espressif Let me know if you want me to test this PR, I got this hardware and an ESP32 dev kit.

[SinglWolf]

I tried to apply your patch for my ILI9488 display.
The red color is gone. The picture, which should be in color, appears as monochrome on a blue background (appears on the screen very, very slowly).
On the watch, the colon, which should be blinking in time with the seconds, no longer blinks, but looks like a smooth slideshow from top to bottom. The numbers are updated very slowly too.
The only thing that made me happy was that the artifacts were gone.

Or have I done something wrong again?

[SinglWolf]

After a lot of searching on github, I stumbled upon repo https://github.com/anuprao/esp32_ili9488, which implements something similar for the ILI9488 display controller. Only the version of the LVGL is very old 5.3.0. Maybe this will come in handy for solving the problem?
Fragment of code:

uint8_t buffConv[320*3] = {};

// Used in buffered mode
void ili9488_flush(int32_t x1, int32_t y1, int32_t x2, int32_t y2, const lv_color_t * color_map)
{
	uint8_t data[4];
	
	uint8_t* pbuf;
	uint16_t i, len, chunk;
	//uint8_t r,g,b;
	
	lv_color16_t s;
	lv_color16_t* pColor = color_map;

	/*Column addresses*/
	ili9488_send_cmd(0x2A);
	data[0] = (x1 >> 8) & 0xFF;
	data[1] = x1 & 0xFF;
	data[2] = (x2 >> 8) & 0xFF;
	data[3] = x2 & 0xFF;
	ili9488_send_data(data, 4);

	/*Page addresses*/
	ili9488_send_cmd(0x2B);
	data[0] = (y1 >> 8) & 0xFF;
	data[1] = y1 & 0xFF;
	data[2] = (y2 >> 8) & 0xFF;
	data[3] = y2 & 0xFF;
	ili9488_send_data(data, 4);

	/*Memory write*/
	ili9488_send_cmd(0x2C);
	
	uint32_t size = (x2 - x1 + 1) * (y2 - y1 + 1);

	//Original code
	//ili9488_send_color((void*)color_map, size * 2);	/*Send the remaining data*/
	
	//ili9488_send_color((void*)color_map, size * 2);
	//ili9488_send_color(arrSample, 32);
	
	chunk = 0;
	len = size;
	while(len > 0)
	{
		chunk = (320 < len)? 320: len;
		
		pbuf = buffConv;
		for(i = 0; i<chunk; i++)
		{
			s = *pColor;
			
			*pbuf = s.red << 3;
			pbuf++;
			
			*pbuf = (s.green_h << 3 | s.green_l) << 2;
			pbuf++;
			
			*pbuf = s.blue << 3;
			pbuf++;
			
			pColor++;
		}
		
		ili9488_send_color(buffConv, chunk*3);
		
		len = len - chunk;
	}
	
	printf("size_flush = %d\r\n", size);
	
	//lv_flush_ready();

}

[tore-espressif]

Hello @SinglWolf ,
Thank you for your patience.
I ordered ILI9488 LCD controller a while ago, so I could test the performance, but the package got stuck at customs.
I will finalize this PR as soon as I get it.
Thanks!

[wreyford]

In my experience, this solution helps for the DMA issue, but renders the updates so slow, that the screen is not usable.
I have ordered extra PSRAM to try and solve this issue using hardware - still await that in the mail.

[tore-espressif]

Hello everyone,
I finally got my ILI9488 display and tested the performance. With the fix proposed in the PR it is really poor.
So if someone is interested in using ILI9488 (or any other that doesn't support RGB565 over SPI) I propose one of the following:

1. If you can, use different controller, that supports RGB565 over SPI
2. Use parallel interface (supports RGB565)
3. Use already implemented SW transformation (allocates lot of RAM)
4. Use hardware serial->parallel transformation. Example here: https://www.waveshare.com/w/upload/8/85/Pico-ResTouch-LCD-3.5_Sch.pdf