li9341: change to use the standard SPI

[sago35]

The following changes make the SPI faster, so you can now use the standard SPI.

tinygo-org/tinygo#1453

The FPS of examples/ili9341/pyportal_boing is as follows, with acceptable performance.

• feather-m4 + ili9341 320x240 spi
• org
  • 42 - 41 fps
• new
  • 42 - 40fps

However, FillRectangle(), which uses write16n(), is very slow.
For example, it is used in examples/ili9341/basic.
It is visibly slow.
The following is the waveform during FillRectangle(), the new one is very slow with a gap of less than 500ns.

org:
fillscreen (320x240) == 51.38ms

new:
fillscreen (320x240) == 380.70ms

[sago35]

The problem of slow write16n() is due to the multiple calls to SPI.
SPI.Tx() now waits for the SPI to finish sending/receiving at the end of SPI.Tx() to simplify the process.

SPI.TxN(w, r []byte, n int) to try it out, and the speed was the same as before the change.

I would like to add TxN() to the machine package (and tinygo-org/drivers.SPI interface), what do you think?
Please let me know if there is a better way.
Perhaps it would also be useful when using SPI with DMA.

func (spi SPI) TxN(w, r []byte, n int) error {
	// At this time, it can only transmit.
	for i := 0; i < len(w)*n; i++ {
		for !spi.Bus.INTFLAG.HasBits(sam.SERCOM_SPIM_INTFLAG_DRE) {
		}
		spi.Bus.DATA.Set(uint32(w[i%len(w)]))
	}
	for !spi.Bus.INTFLAG.HasBits(sam.SERCOM_SPIM_INTFLAG_TXC) {
	}

	// read to clear RXC register
	for spi.Bus.INTFLAG.HasBits(sam.SERCOM_SPIM_INTFLAG_RXC) {
		spi.Bus.DATA.Get()
	}

	return nil
}

[sago35]

This change is necessary.
When using the standard SPI, it is not possible to send while converting endianness.
The same problem will occur in the future when converting to DMA, so it is necessary to use byte-by-byte endianness-aware arrays.

[sago35]

SPI.Tx() every 64 bytes at most, with minimal slowdown
SPI.TxN() is not necessary.

[sago35]

Checked with feather-m4 (atsamd51) and xiao (atsamd21).

However, it does not work at 24MHz on xiao at this time.
This seems to be a problem with the implementation of machine_atsamd21.go.
I will try to fix it later.

It will probably work on other chips, such as nrf and teensy.
I'll check this part later as well.

[sago35]

I have confirmed that feather-m4, feather-nrf52840 and feather-stm32f405 work with exactly the same code by loading the following files.

// +build feather_m4 feather_m4_can feather_nrf52840 feather_stm32f405

package main

import (
	"machine"

	"tinygo.org/x/drivers/ili9341"
)

var (
	display = ili9341.NewSPI(
		machine.SPI0,
		machine.D10, // LCD_DC,
		machine.D11, // LCD_SS_PIN,
		machine.D12, // LCD_RESET,
	)

	backlight = machine.D13 // LCD_BACKLIGHT
)

func init() {
	machine.SPI0.Configure(machine.SPIConfig{
		SCK:       machine.SPI0_SCK_PIN,
		SDO:       machine.SPI0_SDO_PIN,
		SDI:       machine.SPI0_SDI_PIN,
		Frequency: 40000000,
	})
}

The fps is as follows.
It is possible to improve the speed a bit more by modifying the Frequency a bit for each board.

• feather-m4 (atsamd51)
  • 40 fps
• feather-nrf52840 (nrf52840)
  • 23 fps
• feather-stm32f405 (stm32f405)
  • 30 fps

The following picture shows it running on stm32f405.

[sago35]

However, according to #196, the spi driver for stm32f405 seems to have a speed of 68 - 88 fps.
#196 (comment)

I think this can be solved by using DMA on the SPI driver side.

[sago35]

The standard spi-driver is inevitably slow.
The following pictures are from stm32f405.

[sago35]

I've changed my approach to #274, so I'm closing this PR.