readme.txt

一个MPI程序

Dear KT:

> heatdis-MN0-pq is the standard send-recv version.
> heatdis-IMN0-pq is the non-blocking version.

PS.
I have writen 5-6 versions of this mpi-program. Through this course, I learned quite a few things^-^
At the very begining, I completely used my own way to implement the sequential version and corresponding mpi-program. However, after testing the sequential program, I found some sequential optimizing methods are needed as follows:
1) Array is no good. In your sequential progrm, the space is created by two-level pointers.
2) a > b is slower than b < a, especially when this is checked for 100000000(=3000*200*200) times.
3) assignment (=) operator is a very time-wasted sentence, especially when it is done for  100000000 times.
4) the performances of the following two versions are quite different.
for(k=0;k<10000;k++)
	for(I = 0;i<200;i++)
		for(j = 0; j<200;j++)
		{
			a +=d;
			b +=c;
		}

and 

for(k=0;k<10000;k++)
{
	for(i = 0;i<200;i++)
		for(j = 0; j<200;j++)
		{
			a +=d;
		}

	for(i = 0;i<200;i++)
		for(j = 0; j<200;j++)
		{
			b +=c;
		}
}

After that, I also found some probs in my parallel versions.
1) the order of the send and receive is very important, not only because of avoiding dead-lock, but also impoving performance.

Anyway, my final parallel program runs well.^_^

BTW, KT, I really appreciate your help!

sdi