Support of R8G8B8 texture format ?

[hcomere]

Hello,

I do not see support for R8G8B8 texture format, is there a good reason ?

I am capturing images from a video that are in R8G8B8 pixel format, i would like to send it on GPU without conversion.

Regards,
Harold

[bkaradzic]

No GPU actually supports 24-bit RGB format. There is only legacy GL support for it, but otherwise unavailable on other graphics APIs.

[RichardGale]

Oh, did not know that. I always used that format as a nice saving over RGBA8888 for those full screen textures artists like. Is RGB565 our best option for non-alpha textures?

[bkaradzic]

I was just checking Vulkan headers and they have it too:
https://github.com/bkaradzic/bgfx/blob/master/3rdparty/khronos/vulkan/vulkan.h#L263

I might just add it, but D3D doesn't have it. Anyhow I have ability to convert between all uncompressed formats inside bgfx, just features is not hooked up for all formats, and I don't know when I'm going to do it.

[hcomere]

Ok, thanks for your answer.

I will convert my video image into RGBA8 until you add support for RGB8.
Will i receive an email when this will be ready ?

Regards,
Harold

[bkaradzic]

Yeah, I'll just post update here.

[hcomere]

A bit off-topic as i dont want to create an issue for this but ...

About this video-grabber, here what i am roughtly doing :

while video is not terminated
      grab next image from video
      update texture with image pixel buffer
      render one frame

That means the bgfx::memory i am creating to send pixel buffer to gpu is needed only one bgfx::frame().
But i have issues if i delete the memory just after the bgfx::frame, the doc stating :
" or you must make sure data is available for at least 2 bgfx::frame calls "

It makes not sense in my case to defer the memory deletion of one frame.
Why does bgfx requires memory to be alive for 2 frames ?

Regards,
Harold

[bkaradzic]

That's only requirement for bgfx::makeRef if you're not providing ReleaseFn callback. ReleaseFn callback in case you want to release it as soon as bgfx is done with it. The reason why there is 2 frames requirement is because bgfx is asynchronous and when you pass data, it gets processed at some later point (after 2 bgfx::frame calls it's guaranteed that data is processed).

There is also bgfx::alloc when you don't have your buffer, but want to allocate it for bgfx and not worry about it.

See here:
https://bkaradzic.github.io/bgfx/bgfx.html#_CPPv2N4bgfx6MemoryE

[bkaradzic]

Added: https://github.com/bkaradzic/bgfx/blob/master/include/bgfx/bgfx.h#L203

[hcomere]

Great, it works well for me !

Thank you