Low Level Optimizations for GCN – Digital Dragons 2014 Slides

Finally back to Montreal after excellent Digital Dragons 2014 conference.

It was tons of fun and very relevant talks.

I was asked to deliver a rather ‘hardcore’ talk for programming track.

“Low Level Optimizations for GCN” – “Hacking the Next Generation” focuses on:

  • AMD GCN ISA performance characteristics
  • ALU bottlenecks
  • new ISA subset (vs. previous generation)
  • using specialized ISA functions to optimize several algorithms
  • manual interpolation and writing Geometry Shader alike shaders in Pixel Shaders
  • Major Axis problem analysis in context of normal compression
  • Fast Shader Transcendental functions using floating point math in Integer land

Next week I will post PC / HLSL header with Shader Fast Math library (shown and mentioned during the talk) as well as code samples for gradient search and algorithm analysis for Fast Shader Math.

Feel free to grab those slides here :

Low Level Optimizations for AMD GCN [pdf]

Low Level Optimizations for AMD GCN [pptx + notes]