I was just wondering if anyone else using cygwin has emailed you with concerns that OpenMP seems to not work for this homework when it comes to speeding it up. I can get it so the error rate is the same but it slows down a considerable amount no matter what I do for my pragmas.

Yes, I have heard that from a fair number of students that Cygwin sometimes does not behave as it should while using OpenMP. Just put it in your write-up and make a conservative guess or two about #pragmas that should speed up parallel operations.

UPDATE: You can try using Visual Studio. It may achieve better speedup than gcc under Cygwin. See here for guidance.

I ran into an issue parallelizing one of the for loops. In the main function, the second for loop both copies Temperature into Temperature_last and finds the maximum dt value. The problem with this is that this means that there is no time-effective way to run this loop in parallel: either the race condition on dt is avoided by running the loop serially, or the update of dt must be done in a critical block, which actually slows the loop down even more. So essentially, should this loop be done serially, limiting the potential speedup, or are we expected to change the structure of the actual C code to avoid the race condition?

You can make dt a reduction variable. I don’t know if that is slower or not but it is a correct way to solve this problem.

First, questions 6.3.1 and 6.5.1 both mention plotting values on a graph. How detailed is this graph supposed to be? Does it need to have actual data points or can we just show the general shape of the graph. Second, I am having trouble understanding question 6.4.3 and 6.4.4. It mentions content from chapter 4 and I believe it wants us to write code in MIPS? I looked over the section where this is covered in the textbook and I’m still a little unclear about what the question wants us to do.

Just a graph to show trends, not an exact graph. You can write code in MIPS or LLVM if you wish. Construct the code as described using loop unrolling and the newly-introduced send() / receive() functionality. Also, YOU DO NOT NEED TO SOLVE 6.4.1.

For this last homework, if I am not able to finish it by tomorrow night, can I use late days for this homework? I have not used a single late day yet, and I wonder when the deadline of this homework will be if I use all my late days.

Yes, you may use late days for this last homework. Please keep in mind, however, that the TAs and I have to grade your assignment and calculate and submit grades by a firm deadline so please take no longer than absolutely necessary.

Would the same also apply to the portion of 6.4.3 where we are asked to compute the number of cycles the system would take, as we do not have how long bne takes here either?

You can ignore the cycles taken by the bne instructions for this question.

Hi professor, I’m confused about problem 6.3.1 (I’m using late days). In the function, the variables low, mid, and high all are based off of previous iterations of the loop. Wouldn’t this mean the function cannot be correctly parallelized?

Loop dependencies aside, is there nothing you can do in parallel? While you are correct and your speedup will be fairly limited, you may be able to perform some statements in parallel (e.g., determine whether to stay in the while loop, solving for mid, etc.)