1. Parsec Benchmark

2. Hybrid programs with mpi cuda ...

3. Start to look at MPI 3

4. - Traditional multi-core weak memory : non blocking data structure
    - Disjoint memory space (CPU, GPU): CUDA

5. Lock-free data structure benchmarks 
http://www.cse.iitk.ac.in/users/mainakc/lockfree.html

 http://htor.inf.ethz.ch/publications//img/hoefler-dsde-protocols.pdf
Look at Algorithm 2 : Nonblocking consensus - uses MPI-3 non-blocking collectives

6. Hybrid Examples from courses:

UW course:
   CSS 534: Parallel Programming in Grid and Cloud - Programming Tasks
HW 2 is a good example:
   courses.washington.edu/css534/prog/prog2.pdf

A Georgia Tech course:
   CSE 6230: HPC Tools and Apps. — CSE 6230: HPC Tools and Apps
and the relevant assignment:
   stumptown.cc.gt.atl.ga.us:8080/cse6230-hpcta-fa09/hw3.pdf

A course at cornell:
http://www.cac.cornell.edu/education/Training/Intro/Hybrid-090529.pdf

Another course
    ITCS 4145 Cluster Computing
and assignment 4:
    coitweb.uncc.edu/~abw/ITCS4145S13/Assignments/assign4S13.pdf