From 82cd2c31fa6f01bb33dd354ae9b1041b141b22f2 Mon Sep 17 00:00:00 2001 From: Dominic Fischer Date: Thu, 3 Oct 2024 00:52:05 +0100 Subject: [PATCH] Add an example of a compressed DMA frame buffer --- esp-hal/src/dma/mod.rs | 13 ++++ examples/src/bin/lcd_i8080.rs | 139 ++++++++++++++++++++++++++++++++-- 2 files changed, 145 insertions(+), 7 deletions(-) diff --git a/esp-hal/src/dma/mod.rs b/esp-hal/src/dma/mod.rs index 64e052165f0..d3054833c51 100644 --- a/esp-hal/src/dma/mod.rs +++ b/esp-hal/src/dma/mod.rs @@ -2143,6 +2143,19 @@ pub struct Preparation { // burst_mode, alignment, check_owner, etc. } +impl Preparation { + /// Preparation for a bare minimum DMA transfer. + /// - Internal RAM only. + /// - Burst mode disabled. + /// - Maximum aligment. + pub fn minimal(start: *mut DmaDescriptor) -> Self { + Preparation { + start, + block_size: None, + } + } +} + /// [DmaTxBuffer] is a DMA descriptor + memory combo that can be used for /// transmitting data from a DMA channel to a peripheral's FIFO. /// diff --git a/examples/src/bin/lcd_i8080.rs b/examples/src/bin/lcd_i8080.rs index a87c1a2ea72..45925e70055 100644 --- a/examples/src/bin/lcd_i8080.rs +++ b/examples/src/bin/lcd_i8080.rs @@ -25,7 +25,16 @@ use esp_backtrace as _; use esp_hal::{ delay::Delay, - dma::{Dma, DmaChannel0, DmaPriority, DmaTxBuf}, + dma::{ + Dma, + DmaChannel0, + DmaDescriptor, + DmaPriority, + DmaTxBuf, + DmaTxBuffer, + Owner, + Preparation, + }, dma_tx_buffer, gpio::{Input, Io, Level, Output, Pull}, lcd_cam::{ @@ -36,6 +45,7 @@ use esp_hal::{ Blocking, }; use esp_println::println; +use static_cell::ConstStaticCell; #[entry] fn main() -> ! { @@ -206,13 +216,14 @@ fn main() -> ! { // Tearing Effect Line On bus.send(0x35, &[0]); - let width = 320u16; - let height = 480u16; + const WIDTH: usize = 320; + const HEIGHT: usize = 480; + { println!("Set addresses"); - let width_b = width.to_be_bytes(); - let height_b = height.to_be_bytes(); + let width_b = (WIDTH as u16).to_be_bytes(); + let height_b = (HEIGHT as u16).to_be_bytes(); bus.send(0x2A, &[0, 0, width_b[0], width_b[1]]); // CASET bus.send(0x2B, &[0, 0, height_b[0], height_b[1]]) // PASET } @@ -220,18 +231,19 @@ fn main() -> ! { println!("Drawing"); const RED: u16 = 0b00000_000000_11111; + const GREEN: u16 = 0b00000_111111_00000; const BLUE: u16 = 0b11111_000000_00000; backlight.set_high(); - let total_pixels = width as usize * height as usize; + let total_pixels = WIDTH * HEIGHT; let total_bytes = total_pixels * 2; let (mut i8080, mut dma_tx_buf) = bus.resources.take().unwrap(); dma_tx_buf.set_length(dma_tx_buf.capacity()); - for color in [RED, BLUE].iter().cycle() { + for color in [RED, BLUE].iter().cycle().take(4) { let color = color.to_be_bytes(); for chunk in dma_tx_buf.as_mut_slice().chunks_mut(2) { chunk.copy_from_slice(&color); @@ -268,7 +280,120 @@ fn main() -> ! { delay.delay_millis(1_000); } + // This compressed buffer works out to 160 DMA descriptors and a 1920 byte buffer, using up a + // total of 3,840 bytes (160 * (4 * 3) + 1920). This is much more space efficient than the + // 307,200 bytes (320 * 480 * 2) it would have taken to send out a full frame buffer at once. + // It is also more time/CPU efficient than sending out the frame in chunks, as you don't have + // to pay for the latency of starting each chunk, and the CPU has more time to do other things. + let mut compressed_buffer = { + const DESCRIPTOR_COUNT: usize = 160; + const BUFFER_SIZE: usize = 1920; + + static DESCRIPTORS: ConstStaticCell<[DmaDescriptor; DESCRIPTOR_COUNT]> = + ConstStaticCell::new([DmaDescriptor::EMPTY; DESCRIPTOR_COUNT]); + static SHARED_BUFFER: ConstStaticCell<[u8; BUFFER_SIZE]> = + ConstStaticCell::new([0; BUFFER_SIZE]); + + let descriptors = DESCRIPTORS.take(); + let shared_buffer = SHARED_BUFFER.take(); + + CompressedBuffer::new(descriptors, shared_buffer) + }; + + for color in [RED, GREEN, BLUE].iter().cycle() { + // This is very quick as it only sets 1920 bytes, but it can also be skipped by having one + // compressed buffer for each color. + compressed_buffer.set_color(*color); + + // Naive implementation of tear prevention. + { + // Wait for display to start refreshing. + while tear_effect.is_high() {} + // Wait for display to finish refreshing. + while tear_effect.is_low() {} + + // Now we have the maximum amount of time between each refresh available, for drawing. + } + + // Send an entire frame buffer in a single transfer. + (_, i8080, compressed_buffer) = i8080 + .send(0x2Cu8, 0, compressed_buffer) + .map_err(|e| e.0) + .unwrap() + .wait(); + + delay.delay_millis(300); + } + loop { delay.delay_millis(1_000); } } + +/// A simple implementation of a compressed solid frame buffer, using several descriptors to send +/// out the same buffer multiple times. +/// +/// To draw text or shapes in this, one can simply modify a small subset of the descriptors to +/// point to a different buffer. This will need some kind of allocation scheme. +struct CompressedBuffer { + descriptors: &'static mut [DmaDescriptor], + buffer: &'static mut [u8], +} + +impl CompressedBuffer { + fn new( + descriptors: &'static mut [DmaDescriptor], + buffer: &'static mut [u8], + ) -> CompressedBuffer { + assert!(buffer.len() < 4096); + assert!(buffer.len() > 0); + + let mut next = core::ptr::null_mut(); + for desc in descriptors.iter_mut().rev() { + // Setup the linked list + desc.next = next; + next = desc; + + // Use the shared buffer. + desc.buffer = buffer.as_mut_ptr(); + desc.set_size(buffer.len()); + desc.set_length(buffer.len()); + + // Optional but no harm in doing it. + desc.set_owner(Owner::Dma); + } + + CompressedBuffer { + descriptors, + buffer, + } + } + + fn set_color(&mut self, color: u16) { + for chunk in self.buffer.chunks_mut(2) { + chunk.copy_from_slice(&color.to_be_bytes()); + } + } +} + +unsafe impl DmaTxBuffer for CompressedBuffer { + type View = CompressedBufferView; + + fn prepare(&mut self) -> Preparation { + Preparation::minimal(self.descriptors.as_mut_ptr()) + } + + fn into_view(self) -> Self::View { + CompressedBufferView(self) + } + + fn from_view(view: Self::View) -> Self { + view.0 + } + + fn length(&self) -> usize { + unimplemented!() + } +} + +struct CompressedBufferView(CompressedBuffer);