Lecture 18

Learning objectives

After this class, you should be able to:

1. Explain the performance implications of atomic operations.
2. Use shared memory to reduce the overhead of atomic operations.

Reading assignment

1. UIUC Lecture 19.

Exercises and review questions

• Exercises and review questions on current lecture's material
  1. (i) Write code to perform reduction using atomic operations on global memory. (ii) Rewrite this code to use shared memory in order to reduce the overhead of atomic operations. (iii) Compare the performances of the above two with the reduction algorithm we had discussed earlier. Report your performance results on the discussion board.

  2. Write code with different threads performing atomic operations on different, but adjacent, locations in memory. Compare the performance with different data sizes. Report your performance results on the discussion board.

• Preparation for the next lecture
  1. How many lanes are there in the PCIe that connects a GPU to the host on gpu.cs.fsu.edu?

---

Last modified: 14 Mar 2013