Arm-2D applies chromakeying to LV_IMG_CF_TRUE_COLOR images

[kjbracey]

Perform all steps below and tick them with [x]

• Read the FAQ
• Check the related part of the Documentation
• Update lvgl to the latest version
• Reproduce the issue in a Simulator

Describe the bug

The Arm-2D accelerator applies chromakeying even when not requested. Observed in 8.3.4, but issue still present by code inspection.

This section of code does not distinguish the values of cf.

lvgl/src/draw/arm2d/lv_gpu_arm2d.c

Lines 1078 to 1096 in b603fb9

	if((LV_COLOR_FORMAT_NATIVE_CHROMA_KEYED == cf) ||
	(LV_COLOR_FORMAT_NATIVE == cf)) {
	__ARM_2D_PREPARE_TRANS_AND_TARGET_REGION(
	arm_2d_tile_transform_with_opacity_prepare,
	&source_tile,
	source_center,
	ARM_2D_ANGLE((draw_dsc->angle / 10.0f)),
	draw_dsc->zoom / 256.0f,
	(color_int)lv_color_to_int(LV_COLOR_CHROMA_KEY),
	blend_dsc.opa);
	arm_2d_tile_transform(
	&target_tile,
	&target_region,
	NULL
	);
	is_accelerated = true;
	}

To Reproduce

    imageObj = lv_img_create(lv_scr_act());
    imageDesc.data = (const uint8_t *)lvgl_image;
    imageDesc.data_size = sizeof lvgl_image;
    imageDesc.header.always_zero = 0;
    imageDesc.header.reserved = 0;
    imageDesc.header.cf = LV_IMG_CF_TRUE_COLOR;
    imageDesc.header.w = 224;
    imageDesc.header.h = 224;
    lv_img_set_src(imageObj, &imageDesc);

With the image (test bars in my case) containing some pure green, with default config.

Expected behavior

No pixels should be transparent (unless indicated by alpha).

[kisvegabor]

@GorgonMeducer, could you take a look at it?

[GorgonMeducer]

OK, I will fix it.

@kjbracey Just one question, are you using Cortex-M55 or Cortex-M85 processors?

[GorgonMeducer]

The issue is fixed in cmsis-pack v8.3.7-dev

[kjbracey]

@kjbracey Just one question, are you using Cortex-M55 or Cortex-M85 processors?

M55.

Thanks for the quick fix!

[kisvegabor]

I've merged the PR.

[kjbracey]

As I asked on #4179, this looks like it's just disabling the acceleration for non-colorkey images - or does it end up accelerated via another path?

[GorgonMeducer]

@kjbracey Yes. It will use other basic acceleration.

So the whole acceleration strategy is:

First, use condition filters to find those specific scenarios that can be accelerated with dedicated arm-2d API directly,
After that, the left parts will go to the default acceleration routine which is also good.

[GorgonMeducer]

@kjbracey dedicated acceleration for transform-only-with-opacity is added: #4203
And as the code suggested,

lvgl/src/draw/arm2d/lv_gpu_arm2d.c

Lines 1313 to 1330 in 2b56e04

	#if ARM_2D_VERISON >= 10103
	else if (LV_IMG_CF_TRUE_COLOR == cf) {
	__ARM_2D_PREPARE_TRANS_AND_TARGET_REGION(
	arm_2d_tile_transform_only_with_opacity_prepare,
	&source_tile,
	source_center,
	ARM_2D_ANGLE((draw_dsc->angle / 10.0f)),
	draw_dsc->zoom / 256.0f,
	blend_dsc.opa);
	arm_2d_tile_transform(
	&target_tile,
	&target_region,
	NULL
	);
	is_accelerated = true;
	}
	#endif

you won't get the acceleration until the release of arm-2d v1.1.3.

[GorgonMeducer]

@kjbracey arm-2d v1.1.3 is released.
https://github.com/ARM-software/Arm-2D/releases/tag/v1.1.3

Please have a try and let me know whether the issue is solved or not.

[GorgonMeducer]

@kjbracey Any update? Or are you happy with this fix?

[kjbracey]

Sorry, I've not had a chance to try it yet. I will let you know when I do. (Looks good to me from inspection though).

[kjbracey]

Okay, tested 8.3.7 + 1.1.3 now, and it's fixed the display error in our app. But I've not profiled it to check performance, and I'm only testing that chromakey isn't applied - I have nothing attempting to use it. Thanks!

[GorgonMeducer]

@kjbracey Thank you for the confirmation. Don't worry about the performance part. I have done that locally. It is identical to the previous.