Math accuracy: Normalizing quaternions
30 Aug 2017 · MathWhile investigating precision issues with ACL, I ran into two problems that I hadn’t seen documented elsewhere and that slightly surprised me.
Nicholas Frechette's Blog Raw bits
While investigating precision issues with ACL, I ran into two problems that I hadn’t seen documented elsewhere and that slightly surprised me.
This release marks an important milestone. It now supports a fully variable bit rate and it performs admirably so far. The numbers don’t lie. Without using any form of key reduction, we match the compression ratio of Unreal 4 (which uses a mix of linear key reduction with a form of variable quantization) and many more tricks will follow to push this even further. It is worth noting that this new variable bit rate algorithm is entirely different from the one I presented at the GDC 2017 and it should outperform it. In due time, more stats and graphs will be published to outline how the data looks across the whole dataset.
Over the years, I’ve had my fare share of discussions about animation compression and two things became obvious over time: we were all (re-)doing similar things and none of us had access to a state of art implementation to compare against. This lead to rampant speculation about which algorithm was superior or inferior. Having implemented a few algorithms in the past, I have finally decided to redo all that work once more and in the open this time. Say ‘Hello’ to the Animation Compression Library (ACL for short).
In modern video games, the 4x4 matrix multiplication is an important cornerstone. It is used for a very long list of things: moving individual character joints, physics simulation, rendering, etc. To generate a single video game image (and we typically generate between 25 and 60 per second), several thousand matrix multiplications will take place. Today, we will take an in-depth look at such a fundamental piece of code.
For a very long time now with every new gaming console generation came a new set of hardware. New hardware meant new constraints which meant we had to revisit old optimization tricks, rules, and popular beliefs.