feat(draw-sw): add simple Helium acceleration

[PeterBee97]

Description of the feature or fix

Similar to Neon, this patch added Helium acceleration to all blending functions with minimal code. Still, the assembly code will produce slightly different result (normally 1bit error) from C rendering results due to difference in implementation, but visually similar to the eye:

The argb8888 dst is also supported with parallel division by integer newton iteration method. Thanks to tail predication and scatter-gathering functions, the code is conciser than Neon version and should also be more efficient. However, speed is not yet tested as the development was done on QEMU.

Notes

• Update the Documentation if needed.
• Add Examples if relevant.
• Add Tests if applicable.
• If you added new options to lv_conf_template.h run lv_conf_internal_gen.py and update Kconfig.
• Run scripts/code-format.py (astyle needs to be installed) and follow the Code Conventions
• Mark the Pull request as Draft while you are working on the first version, and mark is as Ready when it's ready for review.
• When changes were requested, re-request review to notify the maintainers.
• Help us to review this Pull Request! Anyone can approve or request changes.

[PeterBee97]

Should be able to work together with arm2d. @GorgonMeducer please help review and test it! :)

[GorgonMeducer]

@PeterBee97 ALL LOOKS GOOD! (I haven't verified them on hardware yet. I will do it soon.)

Thank you.

Do you consider adding other Helium acceleration? For example, for transform, rotate screen etc.

By the way, would you please provide a macro option to only use arm-2d acceleration without the ASM version you added? (There is an option only to use your helium assembly version without arm-2d).

It is hard for me to compare the performance differences...

[PeterBee97]

Do you consider adding other Helium acceleration? For example, for transform, rotate screen etc.

I'll look into transforms later.

By the way, would you please provide a macro option to only use arm-2d acceleration without the ASM version you added? (There is an option only to use your helium assembly version without arm-2d).

I've moved arm2d header inclusion out of lv_blend_helium.h since arm2d already has its individual config LV_USE_DRAW_ARM2D_SYNC, so now the two accelerator can be toggled independently. Is that OK? @GorgonMeducer

[kisvegabor]

Amazing! However it's not entirely clear how to select between pure ASM and Arm2D. I guess you need to set LV_USE_DRAW_SW_ASM LV_DRAW_SW_ASM_HELIUM in both cases. And if you enable LV_USE_DRAW_ARM2D_SYNC Arm2D will be used, if it's disabled the plain ASM?

[PeterBee97]

Amazing! However it's not entirely clear how to select between pure ASM and Arm2D. I guess you need to set LV_USE_DRAW_SW_ASM LV_DRAW_SW_ASM_HELIUM in both cases. And if you enable LV_USE_DRAW_ARM2D_SYNC Arm2D will be used, if it's disabled the plain ASM?

That's what I previously thought, but since arm2d cannot override all of asm functions right now enabling LV_DRAW_SW_ASM_HELIUM means asm functions cannot be shutdown entirely (which is bad for Gabriel's test). Having the two options decoupled means we can have arm2d, or asm, or both (where asm is the backup). Or is there better idea to let arm2d work by itself? Need some help here. @kisvegabor

[GorgonMeducer]

@PeterBee97 @kisvegabor

That's what I previously thought, but since arm2d cannot override all of asm functions right now enabling LV_DRAW_SW_ASM_HELIUM means asm functions cannot be shutdown entirely (which is bad for Gabriel's test).

The original solution is the best one. I only need a macro option for testing purposes only. So please add an internal option but not a one used in lv_conf.h, etc.

Once you finished this part. I will send a PR to update cmsis-pack and add a small change to lv_conf_template.h for Helium feature auto-detection.

Amazing! However it's not entirely clear how to select between pure ASM and Arm2D.

No need to select. When LV_DRAW_SW_ASM_HELIUM and LV_USE_DRAW_ARM2D_SYNC are used, both native helium assembly and arm-2d works. When only LV_DRAW_SW_ASM_HELIUM is used, only the native helium assembly is used.

There is no need to allow ordinary users to disable the native helium assembly but keep the arm-2d part. I only need an option to disable the native helium assembly code for debugging purposes.

I am eager to see this PR getting merged.

[PeterBee97]

Added macro USE_NATIVE_ASSEMBLY to disable functions internally.

[kisvegabor]

Clear, looks good to me. We need to document these somewhere though.

I'll also do some tests today.

[kisvegabor]

I've just found something.

[kisvegabor]

It should come from a compiler define.

Please also change the name to LV_USE_NATIVE_ASSEMBLY

[PeterBee97]

Sure, moved to lv_conf.

[GorgonMeducer]

Looks good for me.

[kisvegabor]

At this point I'm not sure #define LV_USE_DRAW_SW_ASM LV_DRAW_SW_ASM_... is still required. As ASM implementation can overlap probably we should enable them differently.

Let's improve it in an other PR later.