Use ESP32-S3 vector extensions for LUT processing and diffing (#281)

Adds optimized versions for 1bpp difference LUT lookup, highlevel framebuffer diffing, and output line masking.
Combined with 120MHz PSRAM (activate in the experimental options), we now get sub-second updates for a 1872x1404 display using epdiy V7.

CMakeLists.txt

@@ -7,10 +7,12 @@ set(app_sources "src/epdiy.c"
                 "src/output_i2s/rmt_pulse.c"
                 "src/output_i2s/i2s_data_bus.c"
                 "src/output_common/lut.c"
+                "src/output_common/lut.S"
                 "src/output_common/line_queue.c"
                 "src/output_common/render_context.c"
                 "src/font.c"
                 "src/displays.c"
+                "src/diff.S"
                 "src/board_specific.c"
                 "src/builtin_waveforms.c"
                 "src/highlevel.c"

examples/demo/main/CMakeLists.txt

@@ -1,3 +1,4 @@
 set(app_sources "main.c")
 
+set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 idf_component_register(SRCS ${app_sources} REQUIRES epdiy)

examples/demo/main/main.c

@@ -54,6 +54,11 @@ void idf_setup() {
         epd_rotated_display_height()
     );
 
+    // The display bus settings for V7 may be conservative, you can manually
+    // override the bus speed to tune for speed, i.e., if you set the PSRAM speed
+    // to 120MHz.
+    // epd_set_lcd_pixel_clock_MHz(17);
+
     heap_caps_print_heap_info(MALLOC_CAP_INTERNAL);
     heap_caps_print_heap_info(MALLOC_CAP_SPIRAM);
 }

examples/test/CMakeLists.txt

@@ -0,0 +1,11 @@
+# This is the project CMakeLists.txt file for the test subproject
+cmake_minimum_required(VERSION 3.16)
+
+# Add newly added components to one of these lines:
+# 1. Add here if the component is compatible with IDF >= v4.3
+set(EXTRA_COMPONENT_DIRS "../../")
+
+set(TEST_COMPONENTS "epdiy")
+
+include($ENV{IDF_PATH}/tools/cmake/project.cmake)
+project(epdiy_testrunner)

examples/test/main/CMakeLists.txt

@@ -0,0 +1,3 @@
+set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
+
+idf_component_register(SRCS "main.c" INCLUDE_DIRS ".")

examples/test/main/main.c

@@ -0,0 +1,19 @@
+#include <stdio.h>
+#include <unity.h>
+#include "unity_test_runner.h"
+
+static void print_banner(const char* text)
+{
+    printf("\n#### %s #####\n\n", text);
+}
+
+
+void app_main(void)
+{
+    print_banner("Running all the registered tests");
+    UNITY_BEGIN();
+    //unity_run_tests_by_tag("unit", false);
+    unity_run_all_tests();
+    UNITY_END();
+}
+

src/board/epd_board.c

@@ -25,7 +25,7 @@ void epd_set_board(const EpdBoardDefinition *board_definition) {
   if (epd_board == NULL) {
       epd_board = board_definition;
   } else {
-      ESP_LOGE("epdiy", "EPD board can only be set once!");
+      ESP_LOGW("epdiy", "EPD board can only be set once!");
   }
 }
 

src/board/epd_board_v6.c

@@ -1,3 +1,4 @@
+#include "driver/gpio.h"
 #include "epd_board.h"
 
 #include <esp_log.h>
@@ -144,8 +145,11 @@ static void epd_board_deinit() {
   vTaskDelay(500);
   pca9555_read_input(config_reg.port, 0);
   pca9555_read_input(config_reg.port, 1);
-  ESP_LOGI("epdiy", "going to sleep.");
   i2c_driver_delete(EPDIY_I2C_PORT);
+  gpio_isr_handler_remove(CFG_INTR);
+  gpio_uninstall_isr_service();
+  gpio_reset_pin(CFG_INTR);
+  gpio_reset_pin(V4_LATCH_ENABLE);
 }
 
 static void epd_board_set_ctrl(epd_ctrl_state_t *state, const epd_ctrl_state_t * const mask) {

src/board/epd_board_v7.c

@@ -92,22 +92,22 @@ static lcd_bus_config_t lcd_config = {
     .leh = LEH,
     .start_pulse = STH,
     .stv = STV,
-    .data_0 = D0,
-    .data_1 = D1,
-    .data_2 = D2,
-    .data_3 = D3,
-    .data_4 = D4,
-    .data_5 = D5,
-    .data_6 = D6,
-    .data_7 = D7,
-    .data_8 = D8,
-    .data_9 = D9,
-    .data_10 = D10,
-    .data_11 = D11,
-    .data_12 = D12,
-    .data_13 = D13,
-    .data_14 = D14,
-    .data_15 = D15,
+    .data[0] = D0,
+    .data[1] = D1,
+    .data[2] = D2,
+    .data[3] = D3,
+    .data[4] = D4,
+    .data[5] = D5,
+    .data[6] = D6,
+    .data[7] = D7,
+    .data[8] = D8,
+    .data[9] = D9,
+    .data[10] = D10,
+    .data[11] = D11,
+    .data[12] = D12,
+    .data[13] = D13,
+    .data[14] = D14,
+    .data[15] = D15,
 };
 
 static void epd_board_init(uint32_t epd_row_width) {
@@ -158,24 +158,27 @@ static void epd_board_init(uint32_t epd_row_width) {
 
 static void epd_board_deinit() {
 
+  epd_lcd_deinit();
+
   ESP_ERROR_CHECK(pca9555_set_config(config_reg.port, CFG_PIN_PWRGOOD | CFG_PIN_INT | CFG_PIN_VCOM_CTRL | CFG_PIN_PWRUP, 1));
 
   int tries = 0;
   while (!((pca9555_read_input(config_reg.port, 1) & 0xC0) == 0x80)) {
-    if (tries >= 500) {
+    if (tries >= 50) {
       ESP_LOGE("epdiy", "failed to shut down TPS65185!");
       break;
     }
     tries++;
     vTaskDelay(1);
-    printf("%X\n", pca9555_read_input(config_reg.port, 1));
   }
+
   // Not sure why we need this delay, but the TPS65185 seems to generate an interrupt after some time that needs to be cleared.
-  vTaskDelay(500);
+  vTaskDelay(50);
   pca9555_read_input(config_reg.port, 0);
   pca9555_read_input(config_reg.port, 1);
-  ESP_LOGI("epdiy", "going to sleep.");
   i2c_driver_delete(EPDIY_I2C_PORT);
+
+  gpio_uninstall_isr_service();
 }
 
 static void epd_board_set_ctrl(epd_ctrl_state_t *state, const epd_ctrl_state_t * const mask) {
@@ -214,8 +217,6 @@ static void epd_board_poweron(epd_ctrl_state_t *state) {
   while (!(pca9555_read_input(config_reg.port, 1) & CFG_PIN_PWRGOOD)) {
   }
 
-  printf("PG is up\n");
-
   ESP_ERROR_CHECK(tps_write_register(config_reg.port, TPS_REG_ENABLE, 0x3F));
 
   tps_set_vcom(config_reg.port, vcom);

src/board/pca9555.c

@@ -25,6 +25,9 @@ static esp_err_t i2c_master_read_slave(i2c_port_t i2c_num, uint8_t* data_rd, siz
         return ESP_OK;
     }
     i2c_cmd_handle_t cmd = i2c_cmd_link_create();
+    if (cmd == NULL) {
+        ESP_LOGE("epdiy", "insufficient memory for I2C transaction");
+    }
     i2c_master_start(cmd);
     i2c_master_write_byte(cmd, ( EPDIY_PCA9555_ADDR << 1 ) | I2C_MASTER_WRITE, true);
 	i2c_master_write_byte(cmd, reg, true);
@@ -37,6 +40,9 @@ static esp_err_t i2c_master_read_slave(i2c_port_t i2c_num, uint8_t* data_rd, siz
 	i2c_cmd_link_delete(cmd);
 
 	cmd = i2c_cmd_link_create();
+    if (cmd == NULL) {
+        ESP_LOGE("epdiy", "insufficient memory for I2C transaction");
+    }
     i2c_master_start(cmd);
 	i2c_master_write_byte(cmd, ( EPDIY_PCA9555_ADDR << 1 ) | I2C_MASTER_READ, true);
 	if (size > 1) {
@@ -57,6 +63,9 @@ static esp_err_t i2c_master_read_slave(i2c_port_t i2c_num, uint8_t* data_rd, siz
 static esp_err_t i2c_master_write_slave(i2c_port_t i2c_num, uint8_t ctrl,  uint8_t* data_wr, size_t size)
 {
     i2c_cmd_handle_t cmd = i2c_cmd_link_create();
+    if (cmd == NULL) {
+        ESP_LOGE("epdiy", "insufficient memory for I2C transaction");
+    }
     i2c_master_start(cmd);
     i2c_master_write_byte(cmd, ( EPDIY_PCA9555_ADDR  << 1 ) | I2C_MASTER_WRITE, true);
     i2c_master_write_byte(cmd, ctrl, true);

src/diff.S

@@ -0,0 +1,159 @@
+#include <xtensa/config/core-isa.h>
+#include <xtensa/config/core-matmap.h>
+#include "output_common/render_method.h"
+
+#ifdef RENDER_METHOD_LCD
+
+.text
+.align  4
+.global epd_interlace_4bpp_line_VE
+.type   epd_interlace_4bpp_line_VE,@function
+
+//        // CRASH AND BURN for debugging
+//        EE.MOVI.32.A q3, a2, 0
+//        EE.MOVI.32.A q3, a3, 1
+//        EE.MOVI.32.A q3, a4, 2
+//        EE.MOVI.32.A q3, a5, 3
+//        l8ui a10, a10, 0
+
+// bool interlace_line(
+//     const uint8_t *to,
+//     const uint8_t *from,
+//     uint8_t *col_dirtyness;
+//     uint8_t *interlaced,
+//     int fb_width
+// )
+epd_interlace_4bpp_line_VE: 
+// to   - a2
+// from   - a3
+// interlaced   - a4
+// col_dirtyness  - a5
+// fb_width      - a6
+
+    entry	a1, 32
+
+    // divide by 32 for loop count
+    srli a11, a6, 5
+
+    movi.n a10, 0xF0F0F0F0;
+    EE.MOVI.32.Q q6,a10,0
+    EE.MOVI.32.Q q6,a10,1
+    EE.MOVI.32.Q q6,a10,2
+    EE.MOVI.32.Q q6,a10,3
+
+    movi.n a10, 0x0F0F0F0F
+    EE.MOVI.32.Q q7,a10,0
+    EE.MOVI.32.Q q7,a10,1
+    EE.MOVI.32.Q q7,a10,2
+    EE.MOVI.32.Q q7,a10,3
+
+    // put 4 into shift amount
+    movi.n a10, 4
+    WSR.SAR a10
+
+    // "dirtyness" register
+    EE.ZERO.Q q5
+
+    // Instructions sometimes are in an unexpected order
+    // for best pipeline utilization
+    loopnez a11, .loop_end_difference
+
+        EE.VLD.128.IP q0, a2, 16
+        EE.VLD.128.IP q1, a3, 16
+
+        // load column dirtyness
+        EE.VLD.128.IP q3, a5, 0
+
+        // update dirtyness
+        EE.XORQ q4, q1, q0
+
+        // line dirtyness accumulator
+        EE.ORQ q5, q5, q4
+        // column dirtyness 
+        EE.ORQ q3, q3, q4
+
+        // store column dirtyness 
+        EE.VST.128.IP q3, a5, 16
+
+        // mask out every second value
+        EE.ANDQ q2, q0, q7
+        EE.ANDQ q0, q0, q6
+        EE.ANDQ q3, q1, q7
+        EE.ANDQ q1, q1, q6
+
+        // shift vectors to align
+        EE.VSL.32 q2, q2
+        EE.VSR.32 q1, q1
+
+        // the right shift sign-extends, 
+        // so we make sure the resulting shift is logical by masking again
+        EE.ANDQ q1, q1, q7
+
+        // Combine "from" and "to" nibble
+        EE.ORQ q2, q2, q3
+        EE.ORQ q0, q0, q1
+
+        // Zip masked out values together
+        EE.VZIP.8 q2, q0
+
+        // store interlaced buffer data
+        EE.VST.128.IP q2, a4, 16
+        EE.VST.128.IP q0, a4, 16
+
+.loop_end_difference:
+
+    EE.MOVI.32.A q5, a2, 0
+    EE.MOVI.32.A q5, a3, 1
+    EE.MOVI.32.A q5, a4, 2
+    EE.MOVI.32.A q5, a5, 3
+    or a2, a2, a3
+    or a2, a2, a4
+    or a2, a2, a5
+
+    //movi.n	a2, 1 // return "true"
+
+    // CRASH AND BURN for debugging
+    //EE.MOVI.32.A q5, a2, 0
+    //EE.MOVI.32.A q5, a3, 1
+    //EE.MOVI.32.A q5, a4, 2
+    //EE.MOVI.32.A q5, a5, 3
+    //movi.n a10, 0
+    //l8ui a10, a10, 0
+
+    retw.n
+
+
+.global epd_apply_line_mask_VE
+.type   epd_apply_line_mask_VE,@function
+
+// void epd_apply_line_mask_VE(
+//     uint8_t *line,
+//     const uint8_t *mask,
+//     int mask_len
+// )
+epd_apply_line_mask_VE: 
+// line   - a2
+// mask   - a3
+// mask_len   - a4
+
+    entry	a1, 32
+
+    // divide by 16 for loop count
+    srli a4, a4, 4
+
+    // Instructions sometimes are in an unexpected order
+    // for best pipeline utilization
+    loopnez a4, .loop_end_mask
+
+        EE.VLD.128.IP q0, a2, 0
+        EE.VLD.128.IP q1, a3, 16
+
+        EE.ANDQ q0, q0, q1
+
+        EE.VST.128.IP q0, a2, 16
+
+.loop_end_mask:
+
+    retw.n
+
+#endif

src/epd_highlevel.h

@@ -78,6 +78,8 @@ typedef struct {
   uint8_t* difference_fb;
   /// Tainted lines based on the last difference calculation.
   bool* dirty_lines;
+  /// Tainted column nibbles based on the last difference calculation.
+  uint8_t* dirty_columns;
   /// The waveform information to use.
   const EpdWaveform* waveform;
 } EpdiyHighlevelState;

src/epdiy.c

@@ -382,7 +382,7 @@ enum EpdDrawError epd_draw_image(EpdRect area, const uint8_t *data, const EpdWav
         .width = 0,
         .height = 0,
     };
-    return epd_draw_base(area, data, no_crop, EPD_MODE_DEFAULT, temperature, NULL, waveform);
+    return epd_draw_base(area, data, no_crop, EPD_MODE_DEFAULT, temperature, NULL, NULL, waveform);
 }
 
 void epd_set_rotation(enum EpdRotation rotation) {