General Comments about Homework 3


    Problem 1
	Some people set the output to 0 when the flag was not set
	for a particular element of the input vector. If the flag
	is not set, don't add that value, but leave everything
	else alone. In principle, each node could set the flag of
	vector element i to a different value. v[i] should then be
	the sum of all the nodes where the flag was set.

	Some people collapsed the vector; i.e. returned a single
	number (in the first element of the vector). That is not
	correct. Simple add the individual vector elements (if the
	flag is set).


    Problem 2
	Wasn't a difficult problem for most people.


    Problem 3a: The program
	Some people let the C compiler allocated the matrix on the
	stack by reading N from the command line and declaring

	int matrix[N][N];

	This causes segmentation faults for large matrices. Large
	structures like that need to be allocated using malloc()
	in C.

	In the problem statement where I said

	    \[
		\sum_{j=0}^n \sum_{i=0}^n a_{ij}
	    \]
	    I.e., you are adding all the values in row $i$, before
	    adding the values in row $i+1$.

	I had it wrong. That's column-major order, since we are
	going through all elements of a column before adding the
	next column.

	Most students ignore my mistake and got it right.

	The elements of a 2-D array in C can be accessed using
	A[row][column]. Incrementing the column inside the inner
	loop, and the row in the outer loop produces best access
	performance since we are accessing adjacent locations in
	memory. That is called row-major access. (Some people called
	it column-major access.

	Column-major access is worse than row-major access.



    Problem 3b: The description
	Mostly OK.

    Problem 3c: The plot
	Some students did not see a big difference on hammer. Hammer
	seems to have a large cache and students did not let their
	matrix size grow large enough to really see the effect.