Ongoing Optimisation Effort

Just a quick update to show that we (the developers) aren't just sitting around doing nothing. We've been very busy over the last few months trying to identify and improve the performance issues with Aqsis.

We've hit some major obstacles on the way, and have often found ourselves going two steps backwards instead of one forwards. The positives to be taken are that the changes we are making are improving the architecture, and making it feasible to get more performance out.

Changes in so far include...

• Dramatic changes to the occlusion culling code, it should be more efficient and cull more.
• Bucket overlap caching (from Beaker), that avoids resampling surfaces due to the expansion of buckets to accommodate filtering. Should result in a significantly reduced number of samples, especially for large filter widths.
• Caching of the jittered sample patterns, we now generate a large set of jittered patterns one time, and reuse them randomly.

Ongoing work includes...

• Refactoring the primvar storage (again Beaker).

Performance isn't improving at a dramatic rate at the moment, but we are making progress, and with these architectural changes in, I envisage more rapid improvements as we review and update the implementation details.

More to come....