hw4_grading.txt

20/20 points: code compiles and runs

11/15 points: good code practices
- using (N+2)x(N+2) threads vs NxN (idle threads). 
- may need to use ".f" to ensure single precision operations are used (may slow timing)

15/15 points: discussion of parallelization strategy 

15/15 points: discussion of code verification
- should compare with N > 2...

14/15 points: provide throughput and bandwidth
- you need to try smaller numbers of threads per block (32x32 for Jacobi) to see good performance. 

15/15 points: roofline and computational discussion

Total: 95/100