A Performance and Energy Comparison of FPGAs, GPUs, and Multicores for Sliding-Window Applications 
Introduction
Case Study: Sliding Window
Sliding Window Applications
App 1: Sum of Absolute Differences (SAD)
App 2: 2D Convolution
App 3: Correntropy
Devices Targeted
FPGA Architecture
Window Generator
FPGA Architecture
FPGA Datapaths
FPGA Datapaths Cont.
GPU CUDA Framework
GPU CUDA Framework Cont.
GPU Implementations
CPU OpenCL Implementations
Experimental Setup
Application Case StudiesSum of Absolute Differences
Application Case Studies2D Convolution
Application Case StudiesCorrentropy
Speedup
Single Chip Implementations
Energy Comparison
Future Work
Conclusion

GPU Implementations

SAD: each thread computes SAD between kernel and the 4 windows in its Macro Block
2D Convolution: like SAD, but with multiply-accumulate
2D FFT Convolution: used CUFFT to implement frequency domain version
Correntropy: adds Gaussian lookup table to SAD, computes max values in parallel post processing