feature(lvgl_port): Initial support for SIMD rendering in LVGL

    - Assembly source files for LVGL blend API integrated into lvgl_port
    - Initial assembly assembly implementation of:
        - ARGB8888 simple fill for esp32s3 and esp32
        - RGB565 simple fill for esp32
    - Functionality and benchmark test app

Author: peter-marcisovsky

.gitignore

@@ -7,3 +7,4 @@ dependencies.lock
 doxygen_output/**
 dist
 __pycache__
+gdbinit

components/esp_lvgl_port/CMakeLists.txt

@@ -76,6 +76,27 @@ if("usb_host_hid" IN_LIST build_components)
     list(APPEND ADD_LIBS idf::usb_host_hid)
 endif()
 
+# Include SIMD assembly source code for rendering, only for LVGL 9.x and only for esp32 and esp32s3
+if(lvgl_ver VERSION_GREATER_EQUAL "9.0.0")
+    if(CONFIG_IDF_TARGET_ESP32 OR CONFIG_IDF_TARGET_ESP32S3)
+        message(VERBOSE "Compiling SIMD")
+        if(CONFIG_IDF_TARGET_ESP32S3)
+            file(GLOB_RECURSE ASM_SRCS ${PORT_PATH}/simd/*_esp32s3.S)    # Select only esp32s3 related files
+        else()
+            file(GLOB_RECURSE ASM_SRCS ${PORT_PATH}/simd/*_esp32.S)      # Select only esp32 related files
+        endif()
+        list(APPEND ADD_SRCS ${ASM_SRCS})
+
+        # Include component libraries, so lvgl component would see lvgl_port includes
+        idf_component_get_property(lvgl_lib ${lvgl_name} COMPONENT_LIB)
+        target_include_directories(${lvgl_lib} PRIVATE "include")
+
+        # Force link .S files
+        set_property(TARGET ${COMPONENT_LIB} APPEND PROPERTY INTERFACE_LINK_LIBRARIES "-u lv_color_blend_to_argb8888_esp")
+        set_property(TARGET ${COMPONENT_LIB} APPEND PROPERTY INTERFACE_LINK_LIBRARIES "-u lv_color_blend_to_rgb565_esp")
+    endif()
+endif()
+
 # Here we create the real lvgl_port_lib
 add_library(lvgl_port_lib STATIC
     ${PORT_PATH}/esp_lvgl_port.c

components/esp_lvgl_port/include/esp_lvgl_port_lv_blend.h

@@ -0,0 +1,92 @@
+/*
+ * SPDX-FileCopyrightText: 2024 Espressif Systems (Shanghai) CO LTD
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+#pragma once
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "lvgl.h"
+
+/*********************
+ *      INCLUDES
+ *********************/
+
+#if !CONFIG_LV_DRAW_SW_ASM_CUSTOM
+#warning "esp_lvgl_port_lv_blend.h included, but CONFIG_LV_DRAW_SW_ASM_CUSTOM not set. Assembly rendering not used"
+#else
+
+/*********************
+ *      DEFINES
+ *********************/
+
+#ifndef LV_DRAW_SW_COLOR_BLEND_TO_ARGB8888
+#define LV_DRAW_SW_COLOR_BLEND_TO_ARGB8888(dsc) \
+    _lv_color_blend_to_argb8888_esp(dsc)
+#endif
+
+#ifndef LV_DRAW_SW_COLOR_BLEND_TO_RGB565
+#define LV_DRAW_SW_COLOR_BLEND_TO_RGB565(dsc) \
+    _lv_color_blend_to_rgb565_esp(dsc)
+#endif
+
+
+/**********************
+ *      TYPEDEFS
+ **********************/
+
+typedef struct {
+    uint32_t opa;
+    void *dst_buf;
+    uint32_t dst_w;
+    uint32_t dst_h;
+    uint32_t dst_stride;
+    const void *src_buf;
+    uint32_t src_stride;
+    const lv_opa_t *mask_buf;
+    uint32_t mask_stride;
+} asm_dsc_t;
+
+/**********************
+ * GLOBAL PROTOTYPES
+ **********************/
+
+extern int lv_color_blend_to_argb8888_esp(asm_dsc_t *asm_dsc);
+
+static inline lv_result_t _lv_color_blend_to_argb8888_esp(lv_draw_sw_blend_fill_dsc_t *dsc)
+{
+    asm_dsc_t asm_dsc = {
+        .dst_buf = dsc->dest_buf,
+        .dst_w = dsc->dest_w,
+        .dst_h = dsc->dest_h,
+        .dst_stride = dsc->dest_stride,
+        .src_buf = &dsc->color,
+    };
+
+    return lv_color_blend_to_argb8888_esp(&asm_dsc);
+}
+
+extern int lv_color_blend_to_rgb565_esp(asm_dsc_t *asm_dsc);
+
+static inline lv_result_t _lv_color_blend_to_rgb565_esp(lv_draw_sw_blend_fill_dsc_t *dsc)
+{
+    asm_dsc_t asm_dsc = {
+        .dst_buf = dsc->dest_buf,
+        .dst_w = dsc->dest_w,
+        .dst_h = dsc->dest_h,
+        .dst_stride = dsc->dest_stride,
+        .src_buf = &dsc->color,
+    };
+
+    return lv_color_blend_to_rgb565_esp(&asm_dsc);
+}
+
+#endif // CONFIG_LV_DRAW_SW_ASM_CUSTOM
+
+#ifdef __cplusplus
+} /*extern "C"*/
+#endif

components/esp_lvgl_port/src/lvgl9/simd/lv_color_blend_to_argb8888_esp32.S

@@ -0,0 +1,81 @@
+/*
+ * SPDX-FileCopyrightText: 2024 Espressif Systems (Shanghai) CO LTD
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+// This is LVGL ARGB8888 simple fill for ESP32 processor
+
+    .section .text
+    .align  4
+    .global lv_color_blend_to_argb8888_esp
+    .type   lv_color_blend_to_argb8888_esp,@function
+
+// The function implements the following C code:
+// void lv_color_blend_to_argb8888(_lv_draw_sw_blend_fill_dsc_t * dsc);
+
+// Input params
+//
+// dsc - a2
+
+// typedef struct {
+//     uint32_t opa;                l32i    0
+//     void * dst_buf;              l32i    4
+//     uint32_t dst_w;              l32i    8
+//     uint32_t dst_h;              l32i    12
+//     uint32_t dst_stride;         l32i    16
+//     const void * src_buf;        l32i    20
+//     uint32_t src_stride;         l32i    24
+//     const lv_opa_t * mask_buf;   l32i    28
+//     uint32_t mask_stride;        l32i    32
+// } asm_dsc_t;
+
+lv_color_blend_to_argb8888_esp:
+
+    entry   a1,    32
+
+    l32i.n   a3,    a2,    4                    // a3 - dest_buff
+    l32i.n   a4,    a2,    8                    // a4 - dest_w                in uint32_t
+    l32i.n   a5,    a2,    12                   // a5 - dest_h                in uint32_t
+    l32i.n   a6,    a2,    16                   // a6 - dest_stride           in bytes
+    l32i.n   a7,    a2,    20                   // a7 - src_buff (color)
+    l32i.n   a8,    a7,    0                    // a8 - color as value
+    slli     a11,   a4,    2                    // a11 - dest_w_bytes = sizeof(uint32_t) * dest_w
+
+    movi     a7,    0xff000000                  // oppactiy mask
+    or       a10,    a7,    a8                  // apply oppacity
+
+    srli    a9,    a4,   2                      // a9 - loop_len = dest_w / 4
+    sub     a6,    a6,   a11                    // dest_stride = dest_stride - dest_w_bytes
+
+    .outer_loop:
+
+        // Run main loop which sets 16 bytes in one loop run
+        loopnez a9, ._main_loop
+            s32i.n      a10,  a3,  0                    // save 32 bits from a10 to dest_buff a3
+            s32i.n      a10,  a3,  4                    // save 32 bits from a10 to dest_buff a3
+            s32i.n      a10,  a3,  8                    // save 32 bits from a10 to dest_buff a3
+            s32i.n      a10,  a3,  12                   // save 32 bits from a10 to dest_buff a3
+            addi.n      a3,   a3,  16                   // increment dest_buff pointer by 16 bytes
+        ._main_loop:
+
+        // Finish the remaining bytes out of the loop
+        // Check modulo 8 of the dest_w_bytes, if - then set 8 bytes
+        bbci a11, 3, _mod_8_check                       // branch if 2-nd bit of dest_w_bytes is clear
+            s32i.n      a10,  a3,  0                    // save 32 bits from a10 to dest_buff a3, offset 0 bytes
+            s32i.n      a10,  a3,  4                    // save 32 bits from a10 to dest_buff a3, offset 0 bytes
+            addi.n      a3,   a3,  8                    // increment dest_buff pointer by 8 bytes
+        _mod_8_check:
+
+        // Check modulo 4 of the dest_w_bytes, if - then set 4 bytes
+        bbci a11, 2, _mod_4_check                       // branch if 2-nd bit of dest_w_bytes is clear
+            s32i.n      a10,  a3,  0                    // save 32 bits from a10 to dest_buff a3, offset 0 bytes
+            addi.n      a3,   a3,  4                    // increment dest_buff pointer by 4 bytes
+        _mod_4_check:
+
+        add     a3,  a3,  a6                             // dest_buff + dest_stride
+        addi.n  a5,  a5,  -1                             // decrease the outer loop
+    bnez a5, .outer_loop
+
+    movi.n   a2, 1                                      // return LV_RESULT_OK = 1
+    retw.n                                              // return