va: add VASurfaceAttribAlignmentSize

[ruijingdong]

Due to different HW implementation, the
surface could require a different alignment size
rather than the default ones, here introduces
a new VASurfaceAttribute, alignment size, which
contains two variables log2_width_alignment
and log2_height_alignment, each has 4 bits,
and the alignment needs to be left shifted
2**size from the application side.

The alignment is in the powers of 2 and
range in [20, ... 215] = [1, 2, 4, 8, ... 32768]

And this alignment should be met when creating
context as an add-on requirement. If not
implemented, the existing/default alignment
logic will be used.

[XinfengZhang]

I am wondering what's the usage of this surface alignments usage:
currently, when you call vaCreateSurfaces, you will get a surface with default alignment--- this default alignment should already met the alignment request. right?
in another words, when you query the alignment request, you get a value 16, then when you create surface (480x270), you create it as (480x272). is it the expected behavior.
but current design, when you create a 480x270, it will give a 480x272 to you by default, what's the difference of these two behaviors?

[ruijingdong]

I am wondering what's the usage of this surface alignments usage: currently, when you call vaCreateSurfaces, you will get a surface with default alignment--- this default alignment should already met the alignment request. right? in another words, when you query the alignment request, you get a value 16, then when you create surface (480x270), you create it as (480x272). is it the expected behavior. but current design, when you create a 480x270, it will give a 480x272 to you by default, what's the difference of these two behaviors?

The original issue has been discussed in
https://gitlab.freedesktop.org/mesa/mesa/-/issues/9196

This HEVC problem can be resolved by hacking below code in 
vaapi_encode_h265.c in FFmpeg.

from

ctx->surface_width  = FFALIGN(avctx->width, priv->min_cb_size);
ctx->surface_height = FFALIGN(avctx->height, priv->min_cb_size);

to

ctx->surface_width  = FFALIGN(avctx->width, 64);
ctx->surface_height = FFALIGN(avctx->height, 16);

That is due to the AMD HEVC encoder HW implementation requires 
the image_input surface dimension to be aligned with 64x16. And the surface_width 
and surface_height alignment seems nowhere to get. Considering this could be a 
general issue of different HW implementation constrains, we propose to have such
an interface so that this alignment information is more flexible to different
applications and HW implementations.

Thanks,
Ruijing

[XinfengZhang]

the change itself looks good to me, but do not touch the micro_version, thanks
it will be a part of 2.21.0, not 2.21.1

[ruijingdong]

Thanks for reviewing, anything I need to do from my side?