Reworked TAA PR by HifiExperiments

[ksuprynowicz]

This one needs a lot of testing both in desktop mode and VR to make sure nothing got broken.
TAA quality is a lot nicet than in main branch.

[ksuprynowicz]

It looks like I broke antialiasing mode toggle while rebasing

[ksuprynowicz]

It looks like against skybox only top and right edge of the contour is antialiased properly.

[ksuprynowicz]

With this version some ghosting occurs when avatar is moving. I wonder how difficult it would be to add armature-related movement to velocity buffer?

[ksuprynowicz]

Sadly it also breaks bloom. Black square with four small white squares often appears in places where bloom occurs.

[HifiExperiments]

With this version some ghosting occurs when avatar is moving. I wonder how difficult it would be to add armature-related movement to velocity buffer?

yeah, the velocity buffer currently just uses the entity transform to calculate velocity so this breaks for skinned meshes. we'd need all information necessary to calculate skinning for the previous frame. the key on the shader side is this block here:

overte/libraries/render-utils/src/model.slv

Line 63 in 6e3586b

<@if HIFI_USE_DEFORMED or HIFI_USE_DEFORMEDDQ@>

we'd need an evalPrevMeshDeformer that does the same general thing but calculates prevPositionMS using the previous frame's buffers and uses that to get _prevPositionCS (instead of transformModelToWorldEyeClipPosAndPrevClipPos)

shouldn't be that hard but I forget how the cluster buffers work on the c++ side. probably need two and then ping pong between them

[HifiExperiments]

Sadly it also breaks bloom. Black square with four small white squares often appears in places where bloom occurs.

hmmm I don't remember this being an issue all those years ago...maybe renderdoc would help if you could catch this?