Key to Scalable Parallelism Regularity and Locality GPU Computing Forum Eight Algorithm Optimizations Techniques (so far) 1. 2. 3. 4. 5. 6. 7. 8. Scatter to Gather transformation Privatization Work granularity coarsening Data tiling/reuse Data.

Download Report

Transcript Key to Scalable Parallelism Regularity and Locality GPU Computing Forum Eight Algorithm Optimizations Techniques (so far) 1. 2. 3. 4. 5. 6. 7. 8. Scatter to Gather transformation Privatization Work granularity coarsening Data tiling/reuse Data. 

Key to Scalable Parallelism Regularity and Locality
GPU Computing Forum
Eight Algorithm Optimizations
Techniques (so far)
1.
2.
3.
4.
5.
6.
7.
8.
Scatter to Gather transformation
Privatization
Work granularity coarsening
Data tiling/reuse
Data layout and traversal ordering
Input data binning
Input compaction
Input extraction and regularization
http://courses.engr.illinois.edu/ece598/hk/
GPU Computing Forum
Currently a graduate-level practical algorithm course
“Orthogonal” to Traditional
Parallel Algorithms for Teaching
Granularity
Scatter to
Gather
✓
✓
✓
✓
Tiling Privatization Regularization Compaction Binning Data Layout Coarsening
✓
MRIGridding
✓
✓
CutCP
Histo
✓
Stencil
✓
✓
✓
✓
✓
✓
✓
LBM
✓
BFS
DMM
✓
✓
✓
✓
MRI-Q
✓
✓
SpMV
✓
✓
SAD
✓
✓
Tpacf
✓
FFT
✓ GPU Computing Forum
✓
✓
✓