64-bit attribute ("P") and Houdini Viewport problems

   1907   4   1
User Avatar
Member
13 posts
Joined: Oct. 2013
Offline
Hi,

in this video: https://www.youtube.com/watch?v=oxO8ps_5hbg [www.youtube.com] , Jeff
describes the new 64-bit attribute feature in Houdini (since H 18.0).

I tried to replicate this with the (in-)famous teapot and a transform of 100000 (translate x,y,z, 64 bit).
Calculation of point attributes (for transform & reverse transform) worked correctly.

However, the Houdini Viewport shows the teapot distorted at the (10^5, 10^5, 10^5) position, despite
the P Attributes being there correctly in 64 bit precision.

Is it possible, that the Houdini viewport (or camera) is currently unable to display geometry
with 64-bit "P" attributes correctly - or did I just miss some "hidden" 64-bit flag for the viewport ?

If this is not yet implemented, are there any hints on when this will be coming (eg. H 19.0) ?

- Thanks & Cheers

Edited by svenkr - Aug. 16, 2021 08:46:14

Attachments:
64bit_Distorted_View.png (386.1 KB)

User Avatar
Member
5040 posts
Joined: Feb. 2012
Offline
Hi,

The attribute can be 64bit but the GL viewport doesn't support 64bit attributes, that's why you can't see them in the viewport properly but their values will be properly preserved as 64bit as long as you stay within that precision.
Senior FX TD @ Industrial Light & Magic
Get to the NEXT level in Houdini & VEX with Pragmatic VEX! [www.pragmatic-vfx.com] https://lnk.bio/animatrix [lnk.bio]
User Avatar
Member
13 posts
Joined: Oct. 2013
Offline
Hi Animatrix,

thanks, that's what I feared
This means that one cannot currently display objects correctly in the Houdini viewport which are "far from origin",
even if they use 64-bit prec. points.

Now that SideFX provides 64bit attribute calculations in Houdini (nice feature - thanks @SideFX ), I hope that they
will implement correct display of 64bit point attributes in the viewport soon.

Technically this is possible, e.g. in Maya (which is using double precision) you can display such objects correctly
far from origin (within the limits of 64-bit precision).
User Avatar
Member
237 posts
Joined: Dec. 2016
Offline
Annyone know what the precision is in the vulkan vieuwport ?
User Avatar
Staff
5258 posts
Joined: July 2005
Offline
Vulkan is still 32b. GPUs have very limited 64b support. Using 64b math in the vertex shader would greatly affect performance (sometimes as bad as 32x slower).

However, computation of the object and camera matrices is done in 64b on the CPU, so if you model your object at the origin and then translate it 5M units away with the object transform, then put a camera close to it and look through it, precision will be maintained.
  • Quick Links