async display

This commit is contained in:
2025-10-07 14:57:15 +02:00
parent c9f0f59630
commit 413e021e49
4 changed files with 81 additions and 112 deletions

View File

@@ -40,7 +40,6 @@ namespace DispTools {
reset_pixel(x, y);
}
}
static void draw_to_display() { SMD::draw(); }
// New simplified async pipeline wrappers
static void async_frame_start() { SMD::async_draw_wait(); } // call at frame start
static void async_frame_end() { SMD::async_draw_start(); } // call after rendering

View File

@@ -21,11 +21,9 @@ static constexpr size_t kLineBytes = DISP_WIDTH / 8;
static constexpr size_t kLineMultiSingle = (kLineBytes + 2);
static constexpr size_t kLineDataBytes = kLineMultiSingle * DISP_HEIGHT + 2;
DMA_ATTR extern uint8_t dma_buf[SMD::kLineDataBytes];
extern uint8_t dma_buf[SMD::kLineDataBytes];
void init();
void clear();
void draw();
// Simplified asynchronous frame pipeline:
// Usage pattern each frame:
// SMD::async_draw_wait(); // (start of frame) waits for previous transfer+clear & guarantees pixel area is zeroed
@@ -49,26 +47,19 @@ static void set_pixel(int x, int y, bool value) {
}
}
extern "C" void IRAM_ATTR s_spi_post_cb(spi_transaction_t* trans);
extern "C" void s_spi_post_cb(spi_transaction_t* trans);
static inline spi_device_interface_config_t _devcfg = {
.mode = 0, // SPI mode 0
.clock_speed_hz = 6 * 1000 * 1000, // Clock out at 10 MHz
.spics_io_num = SPI_DISP_CS, // CS pin
.flags = SPI_DEVICE_POSITIVE_CS,
.queue_size = 3,
.pre_cb = nullptr,
.post_cb = s_spi_post_cb,
.mode = 0, // SPI mode 0
.clock_speed_hz = 6 * 1000 * 1000, // Clock out at 10 MHz
.spics_io_num = SPI_DISP_CS, // CS pin
.flags = SPI_DEVICE_POSITIVE_CS,
.queue_size = 3,
.pre_cb = nullptr,
.post_cb = s_spi_post_cb,
};
extern spi_device_handle_t _spi;
extern bool _vcom;
extern bool _inFlight; // internal state; exposed only for rare diagnostics
extern spi_transaction_t _tx; // persistent transaction struct for async API
// Async memcpy engine handle & clear-in-progress flag
// Async memcpy driver handle (see esp_async_memcpy.h)
// Internal clearing flags (not part of user API anymore)
extern volatile bool _clearBusy;
extern volatile bool _clearRequested;
void ensure_clear_task(); // idempotent; called from init
void ensure_clear_task(); // idempotent; called from init
}; // namespace SMD
class SMDSurface : public Surface<SMDSurface, BwPixel>, public StandardEventQueue<SMDSurface> {

View File

@@ -171,8 +171,8 @@ struct PerfStats {
"step=%.3fms || renderAvg=%.3fms [brd=%.3f pcs=%.3f hud=%.3f ovl=%.3f bltQ=%.3f bltW=%.3f] "
"fps=%.1f\n",
(unsigned long) steps, (unsigned long) renders, (double) span / 1e6, avS(overlayUs), avS(inputUs),
avS(logicUs), avS(stepUs), avR(renderUs), avR(rBoardUs), avR(rPiecesUs), avR(rHUDUs), avR(rOverlayUs),
avR(rBlitQueueUs), avR(rBlitWaitUs), fps);
avS(logicUs), avS(stepUs), avR(renderUs), avR(rBoardUs), avR(rPiecesUs), avR(rHUDUs),
avR(rOverlayUs), avR(rBlitQueueUs), avR(rBlitWaitUs), fps);
overlayUs = inputUs = logicUs = renderUs = stepUs = 0;
rBoardUs = rPiecesUs = rHUDUs = rOverlayUs = rBlitQueueUs = rBlitWaitUs = 0;
steps = renders = 0;
@@ -408,10 +408,10 @@ public:
bool ghost, bool paused, bool gameOver, int fps, int avgFrameMs10) {
auto& ps = PerfStats::get();
auto tus = []() { return esp_timer_get_time(); };
uint64_t t0 = tus(); // frame start (no full clear; buffer already zeroed by async post-blit)
uint64_t t0 = tus(); // frame start (no full clear; buffer already zeroed by async post-blit)
drawBatteryOverlay();
rect(ox - 2, oy - 2, bw + 4, bh + 4, true);
uint64_t t2 = tus();
uint64_t t2 = tus();
for (int y = 0; y < cfg::BoardH; ++y)
for (int x = 0; x < cfg::BoardW; ++x) {
int v = b.get(x, y);
@@ -899,6 +899,8 @@ public:
uint64_t dur = (bwEnd - bwStart);
ps.rBlitWaitUs += dur; // actual DMA completion time
ps.lastRBlitWaitUs = dur;
} else {
PerfStats::get().lastRBlitWaitUs = 0;
}
const uint32_t oStart = clock.millis();
overlayTick(now);
@@ -1093,11 +1095,11 @@ public:
private:
// Power-aware overlay throttling
static constexpr uint32_t overlayIntervalMs = 500; // base interval (paused uses multiplier)
static constexpr uint32_t overlayIntervalMs = 1000; // base interval (paused uses multiplier)
static constexpr uint32_t maxIdleSleepMs = 120; // active state cap to keep input responsive
static constexpr uint32_t gameOverRestartDelayMs = 2000; // ms grace period before restart allowed
// Idle (paused/game over) timing tunables
static constexpr uint32_t idleLongCapMs = 2500; // longest deep sleep slice while idle (<= overlay 4s)
static constexpr uint32_t idleLongCapMs = 5000; // longest deep sleep slice while idle (<= overlay 4s)
static constexpr uint32_t idleShortPollMs = 50; // minimal periodic wake to check overlay refresh/input
static constexpr uint32_t idleActivePollMs = 40; // normal_ms provided to PowerHelper for responsiveness
static constexpr uint32_t activeNormalCapMs = 40; // cap for normal_ms during active play
@@ -1171,9 +1173,9 @@ private:
ps.renders++;
// Pure per-frame render metrics (does not include asynchronous blit wait from previous frame)
auto ms = [](uint64_t us) { return (double) us / 1000.0; };
printf("FRAME render=%.3fms [brd=%.3f pcs=%.3f hud=%.3f ovl=%.3f bltQ=%.3f]\n",
ms(ps.lastRenderUs), ms(ps.lastRBoardUs), ms(ps.lastRPiecesUs), ms(ps.lastRHUDUs),
ms(ps.lastROverlayUs), ms(ps.lastRBlitQueueUs));
printf("FRAME render=%.3fms [brd=%.3f pcs=%.3f hud=%.3f ovl=%.3f bltQ=%.3f]\n", ms(ps.lastRenderUs),
ms(ps.lastRBoardUs), ms(ps.lastRPiecesUs), ms(ps.lastRHUDUs), ms(ps.lastROverlayUs),
ms(ps.lastRBlitQueueUs));
}
dirty = false;
}
@@ -1340,9 +1342,9 @@ private:
extern "C" void app_main() {
// Configure dynamic frequency scaling & light sleep if enabled
#ifdef CONFIG_PM_ENABLE
// esp_pm_config_t pm_config = {
// .max_freq_mhz = CONFIG_ESP_DEFAULT_CPU_FREQ_MHZ, .min_freq_mhz = 16, .light_sleep_enable = true};
// ESP_ERROR_CHECK(esp_pm_configure(&pm_config));
esp_pm_config_t pm_config = {
.max_freq_mhz = CONFIG_ESP_DEFAULT_CPU_FREQ_MHZ, .min_freq_mhz = 16, .light_sleep_enable = true};
ESP_ERROR_CHECK(esp_pm_configure(&pm_config));
ESP_ERROR_CHECK(esp_sleep_enable_gpio_wakeup());
#endif

View File

@@ -5,68 +5,68 @@
#include <driver/gpio.h>
#include "disp_tools.hpp"
#include "driver/spi_master.h"
#include "esp_async_memcpy.h"
#include "freertos/FreeRTOS.h"
#include "freertos/semphr.h"
#include "freertos/task.h"
DMA_ATTR uint8_t SMD::dma_buf[SMD::kLineDataBytes]{};
spi_device_handle_t SMD::_spi;
bool SMD::_vcom = false;
bool SMD::_inFlight = false;
spi_transaction_t SMD::_tx{};
volatile bool SMD::_clearBusy = false;
volatile bool SMD::_clearRequested = false;
static TaskHandle_t s_clearTaskHandle = nullptr;
static SemaphoreHandle_t s_clearSem = nullptr;
static volatile bool s_spiDone = false;
DMA_ATTR uint8_t SMD::dma_buf[SMD::kLineDataBytes]{};
DMA_ATTR uint8_t dma_buf_template[SMD::kLineDataBytes]{};
spi_device_handle_t SMD::_spi;
static spi_transaction_t _tx{};
static bool _vcom = false;
volatile bool _inFlight = false;
static TaskHandle_t s_clearTaskHandle = nullptr;
static SemaphoreHandle_t s_clearReqSem = nullptr;
static SemaphoreHandle_t s_clearSem = nullptr;
static async_memcpy_config_t config = ASYNC_MEMCPY_DEFAULT_CONFIG();
// update the maximum data stream supported by underlying DMA engine
static async_memcpy_handle_t driver = NULL;
static unsigned char reverse_bits3(unsigned char b) { return (b * 0x0202020202ULL & 0x010884422010ULL) % 0x3ff; }
static bool IRAM_ATTR my_async_memcpy_cb(async_memcpy_handle_t mcp_hdl, async_memcpy_event_t* event, void* cb_args) {
BaseType_t high_task_wakeup = pdFALSE;
_inFlight = false;
xSemaphoreGiveFromISR(s_clearSem,
&high_task_wakeup); // high_task_wakeup set to pdTRUE if some high priority task unblocked
return high_task_wakeup == pdTRUE;
}
static void zero_framebuffer_payload() {
std::memset(SMD::dma_buf, 0xFF, SMD::kLineDataBytes);
for (uint8_t i = 0; i < DISP_HEIGHT; i++) {
SMD::dma_buf[SMD::kLineMultiSingle * i + 1] = reverse_bits3(i + 1);
}
ESP_ERROR_CHECK(esp_async_memcpy(driver, SMD::dma_buf, dma_buf_template, 12480, my_async_memcpy_cb, nullptr));
}
extern "C" void IRAM_ATTR s_spi_post_cb(spi_transaction_t* /*t*/) {
s_spiDone = true;
if (!SMD::_clearRequested)
SMD::_clearRequested = true;
if (s_clearSem) {
BaseType_t hpw = pdFALSE;
xSemaphoreGiveFromISR(s_clearSem, &hpw);
if (hpw)
portYIELD_FROM_ISR();
}
BaseType_t hpw = pdFALSE;
xSemaphoreGiveFromISR(s_clearReqSem, &hpw);
if (hpw)
portYIELD_FROM_ISR();
}
static void clear_task(void*) {
for (;;) {
if (xSemaphoreTake(s_clearSem, portMAX_DELAY) == pdTRUE) {
if (xSemaphoreTake(s_clearReqSem, portMAX_DELAY) == pdTRUE) {
printf("Started zeroing\n");
if (s_spiDone && SMD::_inFlight) {
spi_transaction_t* r = nullptr;
if (spi_device_get_trans_result(SMD::_spi, &r, 0) == ESP_OK) {
SMD::_inFlight = false;
}
s_spiDone = false;
}
if (SMD::_clearRequested && !SMD::_clearBusy && !SMD::_inFlight) {
printf("Started zeroing 2\n");
SMD::_clearBusy = true;
SMD::_clearRequested = false;
zero_framebuffer_payload();
SMD::_clearBusy = false;
printf("Zeroing done\n");
}
spi_transaction_t* r = nullptr;
ESP_ERROR_CHECK(spi_device_get_trans_result(SMD::_spi, &r, 0));
zero_framebuffer_payload();
// printf("Zeroing done\n");
}
}
}
void SMD::ensure_clear_task() {
if (!s_clearReqSem)
s_clearReqSem = xSemaphoreCreateBinary();
if (!s_clearSem)
s_clearSem = xSemaphoreCreateBinary();
xSemaphoreGive(s_clearSem);
if (!s_clearTaskHandle)
xTaskCreatePinnedToCore(clear_task, "fbclr", 1536, nullptr, tskIDLE_PRIORITY + 1, &s_clearTaskHandle, 0);
}
@@ -84,61 +84,38 @@ void SMD::init() {
dma_buf[2 + kLineMultiSingle * i + kLineBytes] = 0;
}
dma_buf[kLineDataBytes - 1] = 0;
}
void SMD::clear() {
std::array<uint8_t, 2> buf{};
buf[0] = 0b00100000;
spi_transaction_t t{};
t.tx_buffer = buf.data();
t.length = buf.size() * 8;
ESP_ERROR_CHECK(spi_device_transmit(_spi, &t));
}
void SMD::draw() {
_vcom = !_vcom;
_tx = {};
_tx.tx_buffer = dma_buf;
_tx.length = SMD::kLineDataBytes * 8;
dma_buf[0] = 0b10000000 | (_vcom << 6);
ESP_ERROR_CHECK(spi_device_transmit(_spi, &_tx));
for (int y = 0; y < DISP_HEIGHT; ++y)
for (int x = 0; x < DISP_WIDTH; ++x)
DispTools::set_pixel(x, y, false);
std::memcpy(dma_buf_template, dma_buf, sizeof(dma_buf_template));
ESP_ERROR_CHECK(esp_async_memcpy_install(&config, &driver)); // install driver with default DMA engine
}
bool SMD::async_draw_busy() { return _inFlight; }
void SMD::async_draw_start() {
if (_inFlight)
return;
assert(!_inFlight);
if (!xSemaphoreTake(s_clearSem, portMAX_DELAY))
assert(false);
_vcom = !_vcom;
_tx = {};
_tx.tx_buffer = dma_buf;
_tx.length = SMD::kLineDataBytes * 8;
dma_buf[0] = 0b10000000 | (_vcom << 6);
if (spi_device_queue_trans(_spi, &_tx, 0) == ESP_OK)
_inFlight = true;
_inFlight = true;
ESP_ERROR_CHECK(spi_device_queue_trans(_spi, &_tx, 0));
}
void SMD::async_draw_wait() {
if (!_inFlight)
if (uxSemaphoreGetCount(s_clearSem) || !_inFlight) {
assert((uxSemaphoreGetCount(s_clearSem) == 0) == _inFlight);
return;
// Wait until SPI DMA completion harvested and buffer cleared
while (_inFlight || _clearBusy) {
// If ISR fired but task hasn't processed result yet, try harvesting here
if (_inFlight && s_spiDone) {
spi_transaction_t* r = nullptr;
if (spi_device_get_trans_result(_spi, &r, 0) == ESP_OK) {
_inFlight = false;
s_spiDone = false;
// Trigger clear immediately if task not yet scheduled
if (!_clearRequested) {
_clearRequested = true;
if (s_clearSem)
xSemaphoreGive(s_clearSem);
}
}
}
vTaskDelay(1);
}
if (!xSemaphoreTake(s_clearSem, portMAX_DELAY))
assert(false);
if (!xSemaphoreGive(s_clearSem))
assert(false);
assert(!_inFlight);
}
// (clear_in_progress / wait_clear / request_clear removed from public API)