64-bit

Last updated May 9, 2023 Edit Source

• Houdini

This is a write up of Jeff Lait’s masterclass about 64-bit Processing in Houdini.

Usually Houdini works with 32 bit data. To show the limitations you can transform some geo a couple million units away from the origin and then back. Values will get quantized and data will be lost, since those numbers go past floating point limitations. To fix this you can use the Attribute Cast node which specifies which attribute values should be calculated using 64 bit.

In the geometry spreadsheet as well as the mousewheel / inspection menu the attribute is displayed with a small 64 to indicate it’s precision. Also there is a intrinsic detail attribute called precision that defines the preferred precision to be used downstream.

To make vex process in 64-bit you have to change the precision option under the bindings tab.

There is no ‘double’-type in vex. It only depends on the precision chosen. If it’s set to 64-bit, everything will be 64 bits long (int, float, vector etc.). This also ensures that the code won’t need any changes when switching modes.

Since many old nodes were built to work with 32-bits SideFX introduced a global precision override to avoid having to change all precision settings. You can set this global preferred precision by using an Attribute Cast node again.

The mode is indicated by a small 64 badge next to the node.

New attributes will be created in the correct precision depending on the currently preferred one. (Works with wrangles and attribcreate nodes)

1. Geometry moves far away from the origin
2. You have a very large amount of geometry (think fur or feather grooms) -> anything above 2 billion points
3. opencl types get set differently depending on precision mode used

To make Vellum compute in 64-bit modes you just have to feed 64-bit geometry into the solver / constraint nodes. GPU preformance may suffer a lot doing this.

sources / further reading:

• 64-bit Processing in Houdini 18.0 - Jeff Lait (Houdini)