“Using Intel (TM) Ct Technology for Data Parallel Computations in Your Graphics Pipeline” by Gilbert

Parallelism is highly important for the graphics pipeline, but hardware mechanisms for exploiting parallelism exist at many different levels. While many technologies exploit parallelism at the cluster level, this talk focuses on the node level and below, specifically core and vector parallelism. Areas of discussion include:

• Mechanisms for improving the efficiency of modular C++ code and increasing programmer productivity via abstraction.
• How to improve one’s ability to write efficient parallel code at the node level.
• A demo of workloads from seismic and medical imaging as well as Kirchhoff migration to show performance and productivity.