feature(lvgl_port): RGB888 SIMD fill #480

pborcin · 2025-01-14T13:43:14Z

This MR adds SIMD support for LV_DRAW_SW_COLOR_BLEND_TO_RGB888

ESP-BSP Pull Request checklist

Note: For new BSPs create a PR with this link.

Version of modified component bumped
CI passing

Change description

esp32s3 SIMD assembly implementation of the RGB888 simple fill
esp32 assembly implementation of the RGB888 simple fill
Functionality and benchmark tests
Updated README with achieved benchmark results

peter-marcisovsky

@pborcin thanks for the changes. Nice work with the switch-case implementation. I have never tried that myself (did not have a proper use case) but this looks like a good one.

Sorry for the review comments regarding your style of commenting, but as you might have seen the assembly is very hard be read and to be navigated through and commenting makes it bit easier for us to know what is going on and which buffer are you aligning and where to are you jumping. Thank you.

I will also try the test app and let you know.

components/esp_lvgl_port/test_apps/simd/main/lv_blend/src/lv_draw_sw_blend_to_rgb888.c

components/esp_lvgl_port/test_apps/simd/README.md

components/esp_lvgl_port/test_apps/simd/main/lv_blend/src/lv_draw_sw_blend_to_rgb888.c

components/esp_lvgl_port/test_apps/simd/main/test_lv_fill_benchmark.c

components/esp_lvgl_port/test_apps/simd/main/test_lv_fill_functionality.c

components/esp_lvgl_port/src/lvgl9/simd/lv_color_blend_to_rgb888_esp32s3.S

peter-marcisovsky

Please fix the build errors I have described. Thank you.

peter-marcisovsky · 2025-01-30T09:12:49Z

components/esp_lvgl_port/test_apps/simd/main/test_lv_fill_functionality.c

+
+    // Canary bytes area must stay 0
+    TEST_ASSERT_EACH_EQUAL_UINT8_MESSAGE(0, (uint8_t *)test_case->buf.p_ansi + test_case->active_buf_len * test_case->data_type_size + CANARY_BYTES * test_case->data_type_size, CANARY_BYTES * test_case->data_type_size, test_msg_buf);
+    TEST_ASSERT_EACH_EQUAL_UINT8_MESSAGE(0, (uint8_t *)test_case->buf.p_asm + test_case->active_buf_len * test_case->data_type_size + CANARY_BYTES * test_case->data_type_size, CANARY_BYTES * test_case->data_type_size, test_msg_buf);


Can we simplify this a bit?

To something like this

Suggested change

TEST_ASSERT_EACH_EQUAL_UINT8_MESSAGE(0, (uint8_t *)test_case->buf.p_asm + test_case->active_buf_len * test_case->data_type_size + CANARY_BYTES * test_case->data_type_size, CANARY_BYTES * test_case->data_type_size, test_msg_buf);

TEST_ASSERT_EACH_EQUAL_UINT8_MESSAGE(0, (uint8_t *)test_case->buf.p_asm + (test_case->active_buf_len + CANARY_BYTES) * test_case->data_type_size, CANARY_BYTES * test_case->data_type_size, test_msg_buf);

Or like the other tests_eval_ functions above? Which use (total_buf_len - CANARY_BYTES) approach, they calculate the end Canary bytes area from the end of the total buffer. Whereas you are calculating it from beginning of the buffer.

Just to make it more readable.

peter-marcisovsky · 2025-01-30T09:15:31Z

components/esp_lvgl_port/test_apps/simd/main/test_lv_fill_functionality.c

+#endif
+
+    // Canary bytes area must stay 0
+    TEST_ASSERT_EACH_EQUAL_UINT8_MESSAGE(0, (uint8_t *)test_case->buf.p_ansi, CANARY_BYTES * test_case->data_type_size, test_msg_buf);


Just and idea here: you can create something like
static const canary_bytes_area = CANARY_BYTES * test_case->data_type_size and use this variable in tests asserts.

peter-marcisovsky · 2025-01-30T09:23:48Z

components/esp_lvgl_port/test_apps/simd/main/test_lv_fill_functionality.c

+        test_case->blend_api_func(&dsc_asm);    // Call the LVGL API with Assembly code
+        test_case->blend_api_func(&dsc_ansi);   // Call the LVGL API with ANSI code
+    } else if (test_case->blend_api_px_func != NULL) {
+        test_case->blend_api_px_func(&dsc_asm, 3);    // Call the LVGL API with Assembly code


Or something similar, just to know why do we have it here like this.

Suggested change

test_case->blend_api_px_func(&dsc_asm, 3); // Call the LVGL API with Assembly code

test_case->blend_api_px_func(&dsc_asm, 3); // Call the LVGL API with Assembly code with set pixel size

peter-marcisovsky · 2025-01-30T09:56:02Z

components/esp_lvgl_port/src/lvgl9/simd/lv_color_blend_to_rgb888_esp32.S

@@ -0,0 +1,105 @@
+/*
+ * SPDX-FileCopyrightText: 2024 Espressif Systems (Shanghai) CO LTD


Don't forget to update license years in relevant files.

peter-marcisovsky · 2025-01-30T10:29:09Z

components/esp_lvgl_port/test_apps/simd/main/lv_blend/src/lv_draw_sw_blend_to_rgb888.c

+    /*Simple fill*/
+    if (mask == NULL && opa >= LV_OPA_MAX) {
+        if (dsc->use_asm && dest_px_size == 3) {
+            LV_DRAW_SW_COLOR_BLEND_TO_RGB888(dsc);


I think, I have mentioned problem with this already.

The original, LVGL's LV_DRAW_SW_COLOR_BLEND_TO_RGB888 macro, looks like this

Thus, it takes 2 parameters dsc and dest_px_size. Whereas your macro takes only one parameter, dsc. And you have substituted the original macro with your implementation and changed the ANSI version of the blending funciton, which we are not suppose to do, to be compatible with the LVGL upstream.

Now, it you try to build the the any LVGL code with our assembly rendering enabled, the build will fail. Because of this changed macro.

Please, fix this issue and try to build some lvgl app with rendering enabled. I have prepared sdkconfig.ci.asm_render for this purpose. It is not run in any CI yet, so we shall check it manually for now.

components/lvgl__lvgl/src/draw/sw/blend/lv_draw_sw_blend_to_rgb888.c /home/peter/proj/esp-bsp/components/esp_lvgl_port/test_apps/lvgl_port/managed_components/lvgl__lvgl/src/draw/sw/blend/lv_draw_sw_blend_to_rgb888.c: In function 'lv_draw_sw_blend_color_to_rgb888': /home/peter/proj/esp-bsp/components/esp_lvgl_port/test_apps/lvgl_port/managed_components/lvgl__lvgl/src/draw/sw/blend/lv_draw_sw_blend_to_rgb888.c:153:83: error: macro "LV_DRAW_SW_COLOR_BLEND_TO_RGB888" passed 2 arguments, but takes just 1 153 | if(LV_RESULT_INVALID == LV_DRAW_SW_COLOR_BLEND_TO_RGB888(dsc, dest_px_size)) { | ^ In file included from /home/peter/proj/esp-bsp/components/esp_lvgl_port/test_apps/lvgl_port/managed_components/lvgl__lvgl/src/draw/sw/blend/lv_draw_sw_blend_to_rgb888.c:24: /home/peter/proj/esp-bsp/components/esp_lvgl_port/include/esp_lvgl_port_lv_blend.h:36:9: note: macro "LV_DRAW_SW_COLOR_BLEND_TO_RGB888" defined here 36 | #define LV_DRAW_SW_COLOR_BLEND_TO_RGB888(dsc) \ | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ /home/peter/proj/esp-bsp/components/esp_lvgl_port/test_apps/lvgl_port/managed_components/lvgl__lvgl/src/draw/sw/blend/lv_draw_sw_blend_to_rgb888.c:153:33: error: 'LV_DRAW_SW_COLOR_BLEND_TO_RGB888' undeclared (first use in this function) 153 | if(LV_RESULT_INVALID == LV_DRAW_SW_COLOR_BLEND_TO_RGB888(dsc, dest_px_size)) { | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ /home/peter/proj/esp-bsp/components/esp_lvgl_port/test_apps/lvgl_port/managed_components/lvgl__lvgl/src/draw/sw/blend/lv_draw_sw_blend_to_rgb888.c:153:33: note: each undeclared identifier is reported only once for each function it appears in

pborcin requested a review from peter-marcisovsky January 14, 2025 13:43

peter-marcisovsky reviewed Jan 15, 2025

View reviewed changes

pborcin force-pushed the rgb888_accel branch 2 times, most recently from 3da85ab to 234966a Compare January 22, 2025 10:29

pborcin force-pushed the rgb888_accel branch 4 times, most recently from e5983dc to 2d7a665 Compare January 28, 2025 10:40

feature(lvgl_port): RGB888 SIMD fill

0762ba6

pborcin force-pushed the rgb888_accel branch from 2d7a665 to 0762ba6 Compare January 29, 2025 10:13

peter-marcisovsky requested changes Jan 30, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feature(lvgl_port): RGB888 SIMD fill #480

feature(lvgl_port): RGB888 SIMD fill #480

pborcin commented Jan 14, 2025

peter-marcisovsky left a comment

peter-marcisovsky left a comment

peter-marcisovsky Jan 30, 2025

peter-marcisovsky Jan 30, 2025

peter-marcisovsky Jan 30, 2025

peter-marcisovsky Jan 30, 2025

peter-marcisovsky Jan 30, 2025

	TEST_ASSERT_EACH_EQUAL_UINT8_MESSAGE(0, (uint8_t )test_case->buf.p_asm + test_case->active_buf_len test_case->data_type_size + CANARY_BYTES * test_case->data_type_size, CANARY_BYTES * test_case->data_type_size, test_msg_buf);
	TEST_ASSERT_EACH_EQUAL_UINT8_MESSAGE(0, (uint8_t )test_case->buf.p_asm + (test_case->active_buf_len + CANARY_BYTES) test_case->data_type_size, CANARY_BYTES * test_case->data_type_size, test_msg_buf);

	test_case->blend_api_px_func(&dsc_asm, 3); // Call the LVGL API with Assembly code
	test_case->blend_api_px_func(&dsc_asm, 3); // Call the LVGL API with Assembly code with set pixel size

		@@ -0,0 +1,105 @@
		/*
		* SPDX-FileCopyrightText: 2024 Espressif Systems (Shanghai) CO LTD

feature(lvgl_port): RGB888 SIMD fill #480

Are you sure you want to change the base?

feature(lvgl_port): RGB888 SIMD fill #480

Conversation

pborcin commented Jan 14, 2025

ESP-BSP Pull Request checklist

Change description

peter-marcisovsky left a comment

Choose a reason for hiding this comment

peter-marcisovsky left a comment

Choose a reason for hiding this comment

peter-marcisovsky Jan 30, 2025

Choose a reason for hiding this comment

peter-marcisovsky Jan 30, 2025

Choose a reason for hiding this comment

peter-marcisovsky Jan 30, 2025

Choose a reason for hiding this comment

peter-marcisovsky Jan 30, 2025

Choose a reason for hiding this comment

peter-marcisovsky Jan 30, 2025

Choose a reason for hiding this comment