Difference between revisions of "Hybrid GPU Shader Animations for Bones"

From JPCT
Jump to: navigation, search
(Created page with '== Overview == While working on an Android project, I found that jPCT+Bones for skeletal animations were becoming too CPU intensive on the UI thread. When adding twenty animated…')
 
(How It Works)
Line 14: Line 14:
  
 
'''Gains:''' Speed.  You will harness the GPU which is setup to handle floating point calculation quickly and is dedicated to such tasks.  As an example with my project: having 20 animated models dropped the frame-rate to 8-9 fps on my device.  Using this method, it went back up to the 60-62 fps cap.
 
'''Gains:''' Speed.  You will harness the GPU which is setup to handle floating point calculation quickly and is dedicated to such tasks.  As an example with my project: having 20 animated models dropped the frame-rate to 8-9 fps on my device.  Using this method, it went back up to the 60-62 fps cap.
 +
 
'''Losses:''' Software knowledge of the transformed pose.  Since the meshes of your models will not be animated on the software-side, it will have no knowledge of its transformed pose.  The polygons will be in the initial state, which will not give you an accurate bounding-box, ray-trace collisions, etc... of the current animated mesh pose as it will not actually be transformed on the software-side.
 
'''Losses:''' Software knowledge of the transformed pose.  Since the meshes of your models will not be animated on the software-side, it will have no knowledge of its transformed pose.  The polygons will be in the initial state, which will not give you an accurate bounding-box, ray-trace collisions, etc... of the current animated mesh pose as it will not actually be transformed on the software-side.
  
 
== Code ==
 
== Code ==

Revision as of 04:01, 25 June 2016

Overview

While working on an Android project, I found that jPCT+Bones for skeletal animations were becoming too CPU intensive on the UI thread. When adding twenty animated models on screen, I was noticing a considerable drop in the frame-rate. So I developed a hybrid solution of offloading calculations to the GPU for my project, which I will share with you now.

I have added notes and personal recommendations. This architecture may not be perfect for your project, so please alter and use to fit your needs.

Comments and suggestions are welcome as I will be adding a FAQ to the bottom of the page.

Requirements

JPCT (requires GL ES 2+ as it uses GLSL Shaders) Bones

How It Works

This implementation works by offloading all of the vertex calculations to the GPU. The bone matrix transforms for the poses are still performed on the CPU side.

Gains: Speed. You will harness the GPU which is setup to handle floating point calculation quickly and is dedicated to such tasks. As an example with my project: having 20 animated models dropped the frame-rate to 8-9 fps on my device. Using this method, it went back up to the 60-62 fps cap.

Losses: Software knowledge of the transformed pose. Since the meshes of your models will not be animated on the software-side, it will have no knowledge of its transformed pose. The polygons will be in the initial state, which will not give you an accurate bounding-box, ray-trace collisions, etc... of the current animated mesh pose as it will not actually be transformed on the software-side.

Code